Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * ring buffer based function tracer
4 *
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 *
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 *
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
14 */
15#include <linux/ring_buffer.h>
16#include <generated/utsrelease.h>
17#include <linux/stacktrace.h>
18#include <linux/writeback.h>
19#include <linux/kallsyms.h>
20#include <linux/security.h>
21#include <linux/seq_file.h>
22#include <linux/irqflags.h>
23#include <linux/debugfs.h>
24#include <linux/tracefs.h>
25#include <linux/pagemap.h>
26#include <linux/hardirq.h>
27#include <linux/linkage.h>
28#include <linux/uaccess.h>
29#include <linux/vmalloc.h>
30#include <linux/ftrace.h>
31#include <linux/module.h>
32#include <linux/percpu.h>
33#include <linux/splice.h>
34#include <linux/kdebug.h>
35#include <linux/string.h>
36#include <linux/mount.h>
37#include <linux/rwsem.h>
38#include <linux/slab.h>
39#include <linux/ctype.h>
40#include <linux/init.h>
41#include <linux/panic_notifier.h>
42#include <linux/kmemleak.h>
43#include <linux/poll.h>
44#include <linux/nmi.h>
45#include <linux/fs.h>
46#include <linux/trace.h>
47#include <linux/sched/clock.h>
48#include <linux/sched/rt.h>
49#include <linux/fsnotify.h>
50#include <linux/irq_work.h>
51#include <linux/workqueue.h>
52
53#include <asm/setup.h> /* COMMAND_LINE_SIZE */
54
55#include "trace.h"
56#include "trace_output.h"
57
58#ifdef CONFIG_FTRACE_STARTUP_TEST
59/*
60 * We need to change this state when a selftest is running.
61 * A selftest will lurk into the ring-buffer to count the
62 * entries inserted during the selftest although some concurrent
63 * insertions into the ring-buffer such as trace_printk could occurred
64 * at the same time, giving false positive or negative results.
65 */
66static bool __read_mostly tracing_selftest_running;
67
68/*
69 * If boot-time tracing including tracers/events via kernel cmdline
70 * is running, we do not want to run SELFTEST.
71 */
72bool __read_mostly tracing_selftest_disabled;
73
74void __init disable_tracing_selftest(const char *reason)
75{
76 if (!tracing_selftest_disabled) {
77 tracing_selftest_disabled = true;
78 pr_info("Ftrace startup test is disabled due to %s\n", reason);
79 }
80}
81#else
82#define tracing_selftest_running 0
83#define tracing_selftest_disabled 0
84#endif
85
86/* Pipe tracepoints to printk */
87static struct trace_iterator *tracepoint_print_iter;
88int tracepoint_printk;
89static bool tracepoint_printk_stop_on_boot __initdata;
90static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
91
92/* For tracers that don't implement custom flags */
93static struct tracer_opt dummy_tracer_opt[] = {
94 { }
95};
96
97static int
98dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
99{
100 return 0;
101}
102
103/*
104 * To prevent the comm cache from being overwritten when no
105 * tracing is active, only save the comm when a trace event
106 * occurred.
107 */
108static DEFINE_PER_CPU(bool, trace_taskinfo_save);
109
110/*
111 * Kill all tracing for good (never come back).
112 * It is initialized to 1 but will turn to zero if the initialization
113 * of the tracer is successful. But that is the only place that sets
114 * this back to zero.
115 */
116static int tracing_disabled = 1;
117
118cpumask_var_t __read_mostly tracing_buffer_mask;
119
120/*
121 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
122 *
123 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
124 * is set, then ftrace_dump is called. This will output the contents
125 * of the ftrace buffers to the console. This is very useful for
126 * capturing traces that lead to crashes and outputing it to a
127 * serial console.
128 *
129 * It is default off, but you can enable it with either specifying
130 * "ftrace_dump_on_oops" in the kernel command line, or setting
131 * /proc/sys/kernel/ftrace_dump_on_oops
132 * Set 1 if you want to dump buffers of all CPUs
133 * Set 2 if you want to dump the buffer of the CPU that triggered oops
134 */
135
136enum ftrace_dump_mode ftrace_dump_on_oops;
137
138/* When set, tracing will stop when a WARN*() is hit */
139int __disable_trace_on_warning;
140
141#ifdef CONFIG_TRACE_EVAL_MAP_FILE
142/* Map of enums to their values, for "eval_map" file */
143struct trace_eval_map_head {
144 struct module *mod;
145 unsigned long length;
146};
147
148union trace_eval_map_item;
149
150struct trace_eval_map_tail {
151 /*
152 * "end" is first and points to NULL as it must be different
153 * than "mod" or "eval_string"
154 */
155 union trace_eval_map_item *next;
156 const char *end; /* points to NULL */
157};
158
159static DEFINE_MUTEX(trace_eval_mutex);
160
161/*
162 * The trace_eval_maps are saved in an array with two extra elements,
163 * one at the beginning, and one at the end. The beginning item contains
164 * the count of the saved maps (head.length), and the module they
165 * belong to if not built in (head.mod). The ending item contains a
166 * pointer to the next array of saved eval_map items.
167 */
168union trace_eval_map_item {
169 struct trace_eval_map map;
170 struct trace_eval_map_head head;
171 struct trace_eval_map_tail tail;
172};
173
174static union trace_eval_map_item *trace_eval_maps;
175#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
176
177int tracing_set_tracer(struct trace_array *tr, const char *buf);
178static void ftrace_trace_userstack(struct trace_array *tr,
179 struct trace_buffer *buffer,
180 unsigned int trace_ctx);
181
182#define MAX_TRACER_SIZE 100
183static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
184static char *default_bootup_tracer;
185
186static bool allocate_snapshot;
187static bool snapshot_at_boot;
188
189static char boot_instance_info[COMMAND_LINE_SIZE] __initdata;
190static int boot_instance_index;
191
192static char boot_snapshot_info[COMMAND_LINE_SIZE] __initdata;
193static int boot_snapshot_index;
194
195static int __init set_cmdline_ftrace(char *str)
196{
197 strscpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
198 default_bootup_tracer = bootup_tracer_buf;
199 /* We are using ftrace early, expand it */
200 trace_set_ring_buffer_expanded(NULL);
201 return 1;
202}
203__setup("ftrace=", set_cmdline_ftrace);
204
205static int __init set_ftrace_dump_on_oops(char *str)
206{
207 if (*str++ != '=' || !*str || !strcmp("1", str)) {
208 ftrace_dump_on_oops = DUMP_ALL;
209 return 1;
210 }
211
212 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
213 ftrace_dump_on_oops = DUMP_ORIG;
214 return 1;
215 }
216
217 return 0;
218}
219__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
220
221static int __init stop_trace_on_warning(char *str)
222{
223 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
224 __disable_trace_on_warning = 1;
225 return 1;
226}
227__setup("traceoff_on_warning", stop_trace_on_warning);
228
229static int __init boot_alloc_snapshot(char *str)
230{
231 char *slot = boot_snapshot_info + boot_snapshot_index;
232 int left = sizeof(boot_snapshot_info) - boot_snapshot_index;
233 int ret;
234
235 if (str[0] == '=') {
236 str++;
237 if (strlen(str) >= left)
238 return -1;
239
240 ret = snprintf(slot, left, "%s\t", str);
241 boot_snapshot_index += ret;
242 } else {
243 allocate_snapshot = true;
244 /* We also need the main ring buffer expanded */
245 trace_set_ring_buffer_expanded(NULL);
246 }
247 return 1;
248}
249__setup("alloc_snapshot", boot_alloc_snapshot);
250
251
252static int __init boot_snapshot(char *str)
253{
254 snapshot_at_boot = true;
255 boot_alloc_snapshot(str);
256 return 1;
257}
258__setup("ftrace_boot_snapshot", boot_snapshot);
259
260
261static int __init boot_instance(char *str)
262{
263 char *slot = boot_instance_info + boot_instance_index;
264 int left = sizeof(boot_instance_info) - boot_instance_index;
265 int ret;
266
267 if (strlen(str) >= left)
268 return -1;
269
270 ret = snprintf(slot, left, "%s\t", str);
271 boot_instance_index += ret;
272
273 return 1;
274}
275__setup("trace_instance=", boot_instance);
276
277
278static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
279
280static int __init set_trace_boot_options(char *str)
281{
282 strscpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
283 return 1;
284}
285__setup("trace_options=", set_trace_boot_options);
286
287static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
288static char *trace_boot_clock __initdata;
289
290static int __init set_trace_boot_clock(char *str)
291{
292 strscpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
293 trace_boot_clock = trace_boot_clock_buf;
294 return 1;
295}
296__setup("trace_clock=", set_trace_boot_clock);
297
298static int __init set_tracepoint_printk(char *str)
299{
300 /* Ignore the "tp_printk_stop_on_boot" param */
301 if (*str == '_')
302 return 0;
303
304 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
305 tracepoint_printk = 1;
306 return 1;
307}
308__setup("tp_printk", set_tracepoint_printk);
309
310static int __init set_tracepoint_printk_stop(char *str)
311{
312 tracepoint_printk_stop_on_boot = true;
313 return 1;
314}
315__setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
316
317unsigned long long ns2usecs(u64 nsec)
318{
319 nsec += 500;
320 do_div(nsec, 1000);
321 return nsec;
322}
323
324static void
325trace_process_export(struct trace_export *export,
326 struct ring_buffer_event *event, int flag)
327{
328 struct trace_entry *entry;
329 unsigned int size = 0;
330
331 if (export->flags & flag) {
332 entry = ring_buffer_event_data(event);
333 size = ring_buffer_event_length(event);
334 export->write(export, entry, size);
335 }
336}
337
338static DEFINE_MUTEX(ftrace_export_lock);
339
340static struct trace_export __rcu *ftrace_exports_list __read_mostly;
341
342static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
343static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
344static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
345
346static inline void ftrace_exports_enable(struct trace_export *export)
347{
348 if (export->flags & TRACE_EXPORT_FUNCTION)
349 static_branch_inc(&trace_function_exports_enabled);
350
351 if (export->flags & TRACE_EXPORT_EVENT)
352 static_branch_inc(&trace_event_exports_enabled);
353
354 if (export->flags & TRACE_EXPORT_MARKER)
355 static_branch_inc(&trace_marker_exports_enabled);
356}
357
358static inline void ftrace_exports_disable(struct trace_export *export)
359{
360 if (export->flags & TRACE_EXPORT_FUNCTION)
361 static_branch_dec(&trace_function_exports_enabled);
362
363 if (export->flags & TRACE_EXPORT_EVENT)
364 static_branch_dec(&trace_event_exports_enabled);
365
366 if (export->flags & TRACE_EXPORT_MARKER)
367 static_branch_dec(&trace_marker_exports_enabled);
368}
369
370static void ftrace_exports(struct ring_buffer_event *event, int flag)
371{
372 struct trace_export *export;
373
374 preempt_disable_notrace();
375
376 export = rcu_dereference_raw_check(ftrace_exports_list);
377 while (export) {
378 trace_process_export(export, event, flag);
379 export = rcu_dereference_raw_check(export->next);
380 }
381
382 preempt_enable_notrace();
383}
384
385static inline void
386add_trace_export(struct trace_export **list, struct trace_export *export)
387{
388 rcu_assign_pointer(export->next, *list);
389 /*
390 * We are entering export into the list but another
391 * CPU might be walking that list. We need to make sure
392 * the export->next pointer is valid before another CPU sees
393 * the export pointer included into the list.
394 */
395 rcu_assign_pointer(*list, export);
396}
397
398static inline int
399rm_trace_export(struct trace_export **list, struct trace_export *export)
400{
401 struct trace_export **p;
402
403 for (p = list; *p != NULL; p = &(*p)->next)
404 if (*p == export)
405 break;
406
407 if (*p != export)
408 return -1;
409
410 rcu_assign_pointer(*p, (*p)->next);
411
412 return 0;
413}
414
415static inline void
416add_ftrace_export(struct trace_export **list, struct trace_export *export)
417{
418 ftrace_exports_enable(export);
419
420 add_trace_export(list, export);
421}
422
423static inline int
424rm_ftrace_export(struct trace_export **list, struct trace_export *export)
425{
426 int ret;
427
428 ret = rm_trace_export(list, export);
429 ftrace_exports_disable(export);
430
431 return ret;
432}
433
434int register_ftrace_export(struct trace_export *export)
435{
436 if (WARN_ON_ONCE(!export->write))
437 return -1;
438
439 mutex_lock(&ftrace_export_lock);
440
441 add_ftrace_export(&ftrace_exports_list, export);
442
443 mutex_unlock(&ftrace_export_lock);
444
445 return 0;
446}
447EXPORT_SYMBOL_GPL(register_ftrace_export);
448
449int unregister_ftrace_export(struct trace_export *export)
450{
451 int ret;
452
453 mutex_lock(&ftrace_export_lock);
454
455 ret = rm_ftrace_export(&ftrace_exports_list, export);
456
457 mutex_unlock(&ftrace_export_lock);
458
459 return ret;
460}
461EXPORT_SYMBOL_GPL(unregister_ftrace_export);
462
463/* trace_flags holds trace_options default values */
464#define TRACE_DEFAULT_FLAGS \
465 (FUNCTION_DEFAULT_FLAGS | \
466 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
467 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
468 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
469 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
470 TRACE_ITER_HASH_PTR)
471
472/* trace_options that are only supported by global_trace */
473#define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
474 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
475
476/* trace_flags that are default zero for instances */
477#define ZEROED_TRACE_FLAGS \
478 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
479
480/*
481 * The global_trace is the descriptor that holds the top-level tracing
482 * buffers for the live tracing.
483 */
484static struct trace_array global_trace = {
485 .trace_flags = TRACE_DEFAULT_FLAGS,
486};
487
488void trace_set_ring_buffer_expanded(struct trace_array *tr)
489{
490 if (!tr)
491 tr = &global_trace;
492 tr->ring_buffer_expanded = true;
493}
494
495LIST_HEAD(ftrace_trace_arrays);
496
497int trace_array_get(struct trace_array *this_tr)
498{
499 struct trace_array *tr;
500 int ret = -ENODEV;
501
502 mutex_lock(&trace_types_lock);
503 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
504 if (tr == this_tr) {
505 tr->ref++;
506 ret = 0;
507 break;
508 }
509 }
510 mutex_unlock(&trace_types_lock);
511
512 return ret;
513}
514
515static void __trace_array_put(struct trace_array *this_tr)
516{
517 WARN_ON(!this_tr->ref);
518 this_tr->ref--;
519}
520
521/**
522 * trace_array_put - Decrement the reference counter for this trace array.
523 * @this_tr : pointer to the trace array
524 *
525 * NOTE: Use this when we no longer need the trace array returned by
526 * trace_array_get_by_name(). This ensures the trace array can be later
527 * destroyed.
528 *
529 */
530void trace_array_put(struct trace_array *this_tr)
531{
532 if (!this_tr)
533 return;
534
535 mutex_lock(&trace_types_lock);
536 __trace_array_put(this_tr);
537 mutex_unlock(&trace_types_lock);
538}
539EXPORT_SYMBOL_GPL(trace_array_put);
540
541int tracing_check_open_get_tr(struct trace_array *tr)
542{
543 int ret;
544
545 ret = security_locked_down(LOCKDOWN_TRACEFS);
546 if (ret)
547 return ret;
548
549 if (tracing_disabled)
550 return -ENODEV;
551
552 if (tr && trace_array_get(tr) < 0)
553 return -ENODEV;
554
555 return 0;
556}
557
558int call_filter_check_discard(struct trace_event_call *call, void *rec,
559 struct trace_buffer *buffer,
560 struct ring_buffer_event *event)
561{
562 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
563 !filter_match_preds(call->filter, rec)) {
564 __trace_event_discard_commit(buffer, event);
565 return 1;
566 }
567
568 return 0;
569}
570
571/**
572 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
573 * @filtered_pids: The list of pids to check
574 * @search_pid: The PID to find in @filtered_pids
575 *
576 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
577 */
578bool
579trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
580{
581 return trace_pid_list_is_set(filtered_pids, search_pid);
582}
583
584/**
585 * trace_ignore_this_task - should a task be ignored for tracing
586 * @filtered_pids: The list of pids to check
587 * @filtered_no_pids: The list of pids not to be traced
588 * @task: The task that should be ignored if not filtered
589 *
590 * Checks if @task should be traced or not from @filtered_pids.
591 * Returns true if @task should *NOT* be traced.
592 * Returns false if @task should be traced.
593 */
594bool
595trace_ignore_this_task(struct trace_pid_list *filtered_pids,
596 struct trace_pid_list *filtered_no_pids,
597 struct task_struct *task)
598{
599 /*
600 * If filtered_no_pids is not empty, and the task's pid is listed
601 * in filtered_no_pids, then return true.
602 * Otherwise, if filtered_pids is empty, that means we can
603 * trace all tasks. If it has content, then only trace pids
604 * within filtered_pids.
605 */
606
607 return (filtered_pids &&
608 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
609 (filtered_no_pids &&
610 trace_find_filtered_pid(filtered_no_pids, task->pid));
611}
612
613/**
614 * trace_filter_add_remove_task - Add or remove a task from a pid_list
615 * @pid_list: The list to modify
616 * @self: The current task for fork or NULL for exit
617 * @task: The task to add or remove
618 *
619 * If adding a task, if @self is defined, the task is only added if @self
620 * is also included in @pid_list. This happens on fork and tasks should
621 * only be added when the parent is listed. If @self is NULL, then the
622 * @task pid will be removed from the list, which would happen on exit
623 * of a task.
624 */
625void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
626 struct task_struct *self,
627 struct task_struct *task)
628{
629 if (!pid_list)
630 return;
631
632 /* For forks, we only add if the forking task is listed */
633 if (self) {
634 if (!trace_find_filtered_pid(pid_list, self->pid))
635 return;
636 }
637
638 /* "self" is set for forks, and NULL for exits */
639 if (self)
640 trace_pid_list_set(pid_list, task->pid);
641 else
642 trace_pid_list_clear(pid_list, task->pid);
643}
644
645/**
646 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
647 * @pid_list: The pid list to show
648 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
649 * @pos: The position of the file
650 *
651 * This is used by the seq_file "next" operation to iterate the pids
652 * listed in a trace_pid_list structure.
653 *
654 * Returns the pid+1 as we want to display pid of zero, but NULL would
655 * stop the iteration.
656 */
657void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
658{
659 long pid = (unsigned long)v;
660 unsigned int next;
661
662 (*pos)++;
663
664 /* pid already is +1 of the actual previous bit */
665 if (trace_pid_list_next(pid_list, pid, &next) < 0)
666 return NULL;
667
668 pid = next;
669
670 /* Return pid + 1 to allow zero to be represented */
671 return (void *)(pid + 1);
672}
673
674/**
675 * trace_pid_start - Used for seq_file to start reading pid lists
676 * @pid_list: The pid list to show
677 * @pos: The position of the file
678 *
679 * This is used by seq_file "start" operation to start the iteration
680 * of listing pids.
681 *
682 * Returns the pid+1 as we want to display pid of zero, but NULL would
683 * stop the iteration.
684 */
685void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
686{
687 unsigned long pid;
688 unsigned int first;
689 loff_t l = 0;
690
691 if (trace_pid_list_first(pid_list, &first) < 0)
692 return NULL;
693
694 pid = first;
695
696 /* Return pid + 1 so that zero can be the exit value */
697 for (pid++; pid && l < *pos;
698 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
699 ;
700 return (void *)pid;
701}
702
703/**
704 * trace_pid_show - show the current pid in seq_file processing
705 * @m: The seq_file structure to write into
706 * @v: A void pointer of the pid (+1) value to display
707 *
708 * Can be directly used by seq_file operations to display the current
709 * pid value.
710 */
711int trace_pid_show(struct seq_file *m, void *v)
712{
713 unsigned long pid = (unsigned long)v - 1;
714
715 seq_printf(m, "%lu\n", pid);
716 return 0;
717}
718
719/* 128 should be much more than enough */
720#define PID_BUF_SIZE 127
721
722int trace_pid_write(struct trace_pid_list *filtered_pids,
723 struct trace_pid_list **new_pid_list,
724 const char __user *ubuf, size_t cnt)
725{
726 struct trace_pid_list *pid_list;
727 struct trace_parser parser;
728 unsigned long val;
729 int nr_pids = 0;
730 ssize_t read = 0;
731 ssize_t ret;
732 loff_t pos;
733 pid_t pid;
734
735 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
736 return -ENOMEM;
737
738 /*
739 * Always recreate a new array. The write is an all or nothing
740 * operation. Always create a new array when adding new pids by
741 * the user. If the operation fails, then the current list is
742 * not modified.
743 */
744 pid_list = trace_pid_list_alloc();
745 if (!pid_list) {
746 trace_parser_put(&parser);
747 return -ENOMEM;
748 }
749
750 if (filtered_pids) {
751 /* copy the current bits to the new max */
752 ret = trace_pid_list_first(filtered_pids, &pid);
753 while (!ret) {
754 trace_pid_list_set(pid_list, pid);
755 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
756 nr_pids++;
757 }
758 }
759
760 ret = 0;
761 while (cnt > 0) {
762
763 pos = 0;
764
765 ret = trace_get_user(&parser, ubuf, cnt, &pos);
766 if (ret < 0)
767 break;
768
769 read += ret;
770 ubuf += ret;
771 cnt -= ret;
772
773 if (!trace_parser_loaded(&parser))
774 break;
775
776 ret = -EINVAL;
777 if (kstrtoul(parser.buffer, 0, &val))
778 break;
779
780 pid = (pid_t)val;
781
782 if (trace_pid_list_set(pid_list, pid) < 0) {
783 ret = -1;
784 break;
785 }
786 nr_pids++;
787
788 trace_parser_clear(&parser);
789 ret = 0;
790 }
791 trace_parser_put(&parser);
792
793 if (ret < 0) {
794 trace_pid_list_free(pid_list);
795 return ret;
796 }
797
798 if (!nr_pids) {
799 /* Cleared the list of pids */
800 trace_pid_list_free(pid_list);
801 pid_list = NULL;
802 }
803
804 *new_pid_list = pid_list;
805
806 return read;
807}
808
809static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
810{
811 u64 ts;
812
813 /* Early boot up does not have a buffer yet */
814 if (!buf->buffer)
815 return trace_clock_local();
816
817 ts = ring_buffer_time_stamp(buf->buffer);
818 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
819
820 return ts;
821}
822
823u64 ftrace_now(int cpu)
824{
825 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
826}
827
828/**
829 * tracing_is_enabled - Show if global_trace has been enabled
830 *
831 * Shows if the global trace has been enabled or not. It uses the
832 * mirror flag "buffer_disabled" to be used in fast paths such as for
833 * the irqsoff tracer. But it may be inaccurate due to races. If you
834 * need to know the accurate state, use tracing_is_on() which is a little
835 * slower, but accurate.
836 */
837int tracing_is_enabled(void)
838{
839 /*
840 * For quick access (irqsoff uses this in fast path), just
841 * return the mirror variable of the state of the ring buffer.
842 * It's a little racy, but we don't really care.
843 */
844 smp_rmb();
845 return !global_trace.buffer_disabled;
846}
847
848/*
849 * trace_buf_size is the size in bytes that is allocated
850 * for a buffer. Note, the number of bytes is always rounded
851 * to page size.
852 *
853 * This number is purposely set to a low number of 16384.
854 * If the dump on oops happens, it will be much appreciated
855 * to not have to wait for all that output. Anyway this can be
856 * boot time and run time configurable.
857 */
858#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
859
860static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
861
862/* trace_types holds a link list of available tracers. */
863static struct tracer *trace_types __read_mostly;
864
865/*
866 * trace_types_lock is used to protect the trace_types list.
867 */
868DEFINE_MUTEX(trace_types_lock);
869
870/*
871 * serialize the access of the ring buffer
872 *
873 * ring buffer serializes readers, but it is low level protection.
874 * The validity of the events (which returns by ring_buffer_peek() ..etc)
875 * are not protected by ring buffer.
876 *
877 * The content of events may become garbage if we allow other process consumes
878 * these events concurrently:
879 * A) the page of the consumed events may become a normal page
880 * (not reader page) in ring buffer, and this page will be rewritten
881 * by events producer.
882 * B) The page of the consumed events may become a page for splice_read,
883 * and this page will be returned to system.
884 *
885 * These primitives allow multi process access to different cpu ring buffer
886 * concurrently.
887 *
888 * These primitives don't distinguish read-only and read-consume access.
889 * Multi read-only access are also serialized.
890 */
891
892#ifdef CONFIG_SMP
893static DECLARE_RWSEM(all_cpu_access_lock);
894static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
895
896static inline void trace_access_lock(int cpu)
897{
898 if (cpu == RING_BUFFER_ALL_CPUS) {
899 /* gain it for accessing the whole ring buffer. */
900 down_write(&all_cpu_access_lock);
901 } else {
902 /* gain it for accessing a cpu ring buffer. */
903
904 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
905 down_read(&all_cpu_access_lock);
906
907 /* Secondly block other access to this @cpu ring buffer. */
908 mutex_lock(&per_cpu(cpu_access_lock, cpu));
909 }
910}
911
912static inline void trace_access_unlock(int cpu)
913{
914 if (cpu == RING_BUFFER_ALL_CPUS) {
915 up_write(&all_cpu_access_lock);
916 } else {
917 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
918 up_read(&all_cpu_access_lock);
919 }
920}
921
922static inline void trace_access_lock_init(void)
923{
924 int cpu;
925
926 for_each_possible_cpu(cpu)
927 mutex_init(&per_cpu(cpu_access_lock, cpu));
928}
929
930#else
931
932static DEFINE_MUTEX(access_lock);
933
934static inline void trace_access_lock(int cpu)
935{
936 (void)cpu;
937 mutex_lock(&access_lock);
938}
939
940static inline void trace_access_unlock(int cpu)
941{
942 (void)cpu;
943 mutex_unlock(&access_lock);
944}
945
946static inline void trace_access_lock_init(void)
947{
948}
949
950#endif
951
952#ifdef CONFIG_STACKTRACE
953static void __ftrace_trace_stack(struct trace_buffer *buffer,
954 unsigned int trace_ctx,
955 int skip, struct pt_regs *regs);
956static inline void ftrace_trace_stack(struct trace_array *tr,
957 struct trace_buffer *buffer,
958 unsigned int trace_ctx,
959 int skip, struct pt_regs *regs);
960
961#else
962static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
963 unsigned int trace_ctx,
964 int skip, struct pt_regs *regs)
965{
966}
967static inline void ftrace_trace_stack(struct trace_array *tr,
968 struct trace_buffer *buffer,
969 unsigned long trace_ctx,
970 int skip, struct pt_regs *regs)
971{
972}
973
974#endif
975
976static __always_inline void
977trace_event_setup(struct ring_buffer_event *event,
978 int type, unsigned int trace_ctx)
979{
980 struct trace_entry *ent = ring_buffer_event_data(event);
981
982 tracing_generic_entry_update(ent, type, trace_ctx);
983}
984
985static __always_inline struct ring_buffer_event *
986__trace_buffer_lock_reserve(struct trace_buffer *buffer,
987 int type,
988 unsigned long len,
989 unsigned int trace_ctx)
990{
991 struct ring_buffer_event *event;
992
993 event = ring_buffer_lock_reserve(buffer, len);
994 if (event != NULL)
995 trace_event_setup(event, type, trace_ctx);
996
997 return event;
998}
999
1000void tracer_tracing_on(struct trace_array *tr)
1001{
1002 if (tr->array_buffer.buffer)
1003 ring_buffer_record_on(tr->array_buffer.buffer);
1004 /*
1005 * This flag is looked at when buffers haven't been allocated
1006 * yet, or by some tracers (like irqsoff), that just want to
1007 * know if the ring buffer has been disabled, but it can handle
1008 * races of where it gets disabled but we still do a record.
1009 * As the check is in the fast path of the tracers, it is more
1010 * important to be fast than accurate.
1011 */
1012 tr->buffer_disabled = 0;
1013 /* Make the flag seen by readers */
1014 smp_wmb();
1015}
1016
1017/**
1018 * tracing_on - enable tracing buffers
1019 *
1020 * This function enables tracing buffers that may have been
1021 * disabled with tracing_off.
1022 */
1023void tracing_on(void)
1024{
1025 tracer_tracing_on(&global_trace);
1026}
1027EXPORT_SYMBOL_GPL(tracing_on);
1028
1029
1030static __always_inline void
1031__buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
1032{
1033 __this_cpu_write(trace_taskinfo_save, true);
1034
1035 /* If this is the temp buffer, we need to commit fully */
1036 if (this_cpu_read(trace_buffered_event) == event) {
1037 /* Length is in event->array[0] */
1038 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1039 /* Release the temp buffer */
1040 this_cpu_dec(trace_buffered_event_cnt);
1041 /* ring_buffer_unlock_commit() enables preemption */
1042 preempt_enable_notrace();
1043 } else
1044 ring_buffer_unlock_commit(buffer);
1045}
1046
1047int __trace_array_puts(struct trace_array *tr, unsigned long ip,
1048 const char *str, int size)
1049{
1050 struct ring_buffer_event *event;
1051 struct trace_buffer *buffer;
1052 struct print_entry *entry;
1053 unsigned int trace_ctx;
1054 int alloc;
1055
1056 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1057 return 0;
1058
1059 if (unlikely(tracing_selftest_running && tr == &global_trace))
1060 return 0;
1061
1062 if (unlikely(tracing_disabled))
1063 return 0;
1064
1065 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1066
1067 trace_ctx = tracing_gen_ctx();
1068 buffer = tr->array_buffer.buffer;
1069 ring_buffer_nest_start(buffer);
1070 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1071 trace_ctx);
1072 if (!event) {
1073 size = 0;
1074 goto out;
1075 }
1076
1077 entry = ring_buffer_event_data(event);
1078 entry->ip = ip;
1079
1080 memcpy(&entry->buf, str, size);
1081
1082 /* Add a newline if necessary */
1083 if (entry->buf[size - 1] != '\n') {
1084 entry->buf[size] = '\n';
1085 entry->buf[size + 1] = '\0';
1086 } else
1087 entry->buf[size] = '\0';
1088
1089 __buffer_unlock_commit(buffer, event);
1090 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1091 out:
1092 ring_buffer_nest_end(buffer);
1093 return size;
1094}
1095EXPORT_SYMBOL_GPL(__trace_array_puts);
1096
1097/**
1098 * __trace_puts - write a constant string into the trace buffer.
1099 * @ip: The address of the caller
1100 * @str: The constant string to write
1101 * @size: The size of the string.
1102 */
1103int __trace_puts(unsigned long ip, const char *str, int size)
1104{
1105 return __trace_array_puts(&global_trace, ip, str, size);
1106}
1107EXPORT_SYMBOL_GPL(__trace_puts);
1108
1109/**
1110 * __trace_bputs - write the pointer to a constant string into trace buffer
1111 * @ip: The address of the caller
1112 * @str: The constant string to write to the buffer to
1113 */
1114int __trace_bputs(unsigned long ip, const char *str)
1115{
1116 struct ring_buffer_event *event;
1117 struct trace_buffer *buffer;
1118 struct bputs_entry *entry;
1119 unsigned int trace_ctx;
1120 int size = sizeof(struct bputs_entry);
1121 int ret = 0;
1122
1123 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1124 return 0;
1125
1126 if (unlikely(tracing_selftest_running || tracing_disabled))
1127 return 0;
1128
1129 trace_ctx = tracing_gen_ctx();
1130 buffer = global_trace.array_buffer.buffer;
1131
1132 ring_buffer_nest_start(buffer);
1133 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1134 trace_ctx);
1135 if (!event)
1136 goto out;
1137
1138 entry = ring_buffer_event_data(event);
1139 entry->ip = ip;
1140 entry->str = str;
1141
1142 __buffer_unlock_commit(buffer, event);
1143 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1144
1145 ret = 1;
1146 out:
1147 ring_buffer_nest_end(buffer);
1148 return ret;
1149}
1150EXPORT_SYMBOL_GPL(__trace_bputs);
1151
1152#ifdef CONFIG_TRACER_SNAPSHOT
1153static void tracing_snapshot_instance_cond(struct trace_array *tr,
1154 void *cond_data)
1155{
1156 struct tracer *tracer = tr->current_trace;
1157 unsigned long flags;
1158
1159 if (in_nmi()) {
1160 trace_array_puts(tr, "*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1161 trace_array_puts(tr, "*** snapshot is being ignored ***\n");
1162 return;
1163 }
1164
1165 if (!tr->allocated_snapshot) {
1166 trace_array_puts(tr, "*** SNAPSHOT NOT ALLOCATED ***\n");
1167 trace_array_puts(tr, "*** stopping trace here! ***\n");
1168 tracer_tracing_off(tr);
1169 return;
1170 }
1171
1172 /* Note, snapshot can not be used when the tracer uses it */
1173 if (tracer->use_max_tr) {
1174 trace_array_puts(tr, "*** LATENCY TRACER ACTIVE ***\n");
1175 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1176 return;
1177 }
1178
1179 local_irq_save(flags);
1180 update_max_tr(tr, current, smp_processor_id(), cond_data);
1181 local_irq_restore(flags);
1182}
1183
1184void tracing_snapshot_instance(struct trace_array *tr)
1185{
1186 tracing_snapshot_instance_cond(tr, NULL);
1187}
1188
1189/**
1190 * tracing_snapshot - take a snapshot of the current buffer.
1191 *
1192 * This causes a swap between the snapshot buffer and the current live
1193 * tracing buffer. You can use this to take snapshots of the live
1194 * trace when some condition is triggered, but continue to trace.
1195 *
1196 * Note, make sure to allocate the snapshot with either
1197 * a tracing_snapshot_alloc(), or by doing it manually
1198 * with: echo 1 > /sys/kernel/tracing/snapshot
1199 *
1200 * If the snapshot buffer is not allocated, it will stop tracing.
1201 * Basically making a permanent snapshot.
1202 */
1203void tracing_snapshot(void)
1204{
1205 struct trace_array *tr = &global_trace;
1206
1207 tracing_snapshot_instance(tr);
1208}
1209EXPORT_SYMBOL_GPL(tracing_snapshot);
1210
1211/**
1212 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1213 * @tr: The tracing instance to snapshot
1214 * @cond_data: The data to be tested conditionally, and possibly saved
1215 *
1216 * This is the same as tracing_snapshot() except that the snapshot is
1217 * conditional - the snapshot will only happen if the
1218 * cond_snapshot.update() implementation receiving the cond_data
1219 * returns true, which means that the trace array's cond_snapshot
1220 * update() operation used the cond_data to determine whether the
1221 * snapshot should be taken, and if it was, presumably saved it along
1222 * with the snapshot.
1223 */
1224void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1225{
1226 tracing_snapshot_instance_cond(tr, cond_data);
1227}
1228EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1229
1230/**
1231 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1232 * @tr: The tracing instance
1233 *
1234 * When the user enables a conditional snapshot using
1235 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1236 * with the snapshot. This accessor is used to retrieve it.
1237 *
1238 * Should not be called from cond_snapshot.update(), since it takes
1239 * the tr->max_lock lock, which the code calling
1240 * cond_snapshot.update() has already done.
1241 *
1242 * Returns the cond_data associated with the trace array's snapshot.
1243 */
1244void *tracing_cond_snapshot_data(struct trace_array *tr)
1245{
1246 void *cond_data = NULL;
1247
1248 local_irq_disable();
1249 arch_spin_lock(&tr->max_lock);
1250
1251 if (tr->cond_snapshot)
1252 cond_data = tr->cond_snapshot->cond_data;
1253
1254 arch_spin_unlock(&tr->max_lock);
1255 local_irq_enable();
1256
1257 return cond_data;
1258}
1259EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1260
1261static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1262 struct array_buffer *size_buf, int cpu_id);
1263static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1264
1265int tracing_alloc_snapshot_instance(struct trace_array *tr)
1266{
1267 int order;
1268 int ret;
1269
1270 if (!tr->allocated_snapshot) {
1271
1272 /* Make the snapshot buffer have the same order as main buffer */
1273 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
1274 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
1275 if (ret < 0)
1276 return ret;
1277
1278 /* allocate spare buffer */
1279 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1280 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1281 if (ret < 0)
1282 return ret;
1283
1284 tr->allocated_snapshot = true;
1285 }
1286
1287 return 0;
1288}
1289
1290static void free_snapshot(struct trace_array *tr)
1291{
1292 /*
1293 * We don't free the ring buffer. instead, resize it because
1294 * The max_tr ring buffer has some state (e.g. ring->clock) and
1295 * we want preserve it.
1296 */
1297 ring_buffer_subbuf_order_set(tr->max_buffer.buffer, 0);
1298 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1299 set_buffer_entries(&tr->max_buffer, 1);
1300 tracing_reset_online_cpus(&tr->max_buffer);
1301 tr->allocated_snapshot = false;
1302}
1303
1304/**
1305 * tracing_alloc_snapshot - allocate snapshot buffer.
1306 *
1307 * This only allocates the snapshot buffer if it isn't already
1308 * allocated - it doesn't also take a snapshot.
1309 *
1310 * This is meant to be used in cases where the snapshot buffer needs
1311 * to be set up for events that can't sleep but need to be able to
1312 * trigger a snapshot.
1313 */
1314int tracing_alloc_snapshot(void)
1315{
1316 struct trace_array *tr = &global_trace;
1317 int ret;
1318
1319 ret = tracing_alloc_snapshot_instance(tr);
1320 WARN_ON(ret < 0);
1321
1322 return ret;
1323}
1324EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1325
1326/**
1327 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1328 *
1329 * This is similar to tracing_snapshot(), but it will allocate the
1330 * snapshot buffer if it isn't already allocated. Use this only
1331 * where it is safe to sleep, as the allocation may sleep.
1332 *
1333 * This causes a swap between the snapshot buffer and the current live
1334 * tracing buffer. You can use this to take snapshots of the live
1335 * trace when some condition is triggered, but continue to trace.
1336 */
1337void tracing_snapshot_alloc(void)
1338{
1339 int ret;
1340
1341 ret = tracing_alloc_snapshot();
1342 if (ret < 0)
1343 return;
1344
1345 tracing_snapshot();
1346}
1347EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1348
1349/**
1350 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1351 * @tr: The tracing instance
1352 * @cond_data: User data to associate with the snapshot
1353 * @update: Implementation of the cond_snapshot update function
1354 *
1355 * Check whether the conditional snapshot for the given instance has
1356 * already been enabled, or if the current tracer is already using a
1357 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1358 * save the cond_data and update function inside.
1359 *
1360 * Returns 0 if successful, error otherwise.
1361 */
1362int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1363 cond_update_fn_t update)
1364{
1365 struct cond_snapshot *cond_snapshot;
1366 int ret = 0;
1367
1368 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1369 if (!cond_snapshot)
1370 return -ENOMEM;
1371
1372 cond_snapshot->cond_data = cond_data;
1373 cond_snapshot->update = update;
1374
1375 mutex_lock(&trace_types_lock);
1376
1377 ret = tracing_alloc_snapshot_instance(tr);
1378 if (ret)
1379 goto fail_unlock;
1380
1381 if (tr->current_trace->use_max_tr) {
1382 ret = -EBUSY;
1383 goto fail_unlock;
1384 }
1385
1386 /*
1387 * The cond_snapshot can only change to NULL without the
1388 * trace_types_lock. We don't care if we race with it going
1389 * to NULL, but we want to make sure that it's not set to
1390 * something other than NULL when we get here, which we can
1391 * do safely with only holding the trace_types_lock and not
1392 * having to take the max_lock.
1393 */
1394 if (tr->cond_snapshot) {
1395 ret = -EBUSY;
1396 goto fail_unlock;
1397 }
1398
1399 local_irq_disable();
1400 arch_spin_lock(&tr->max_lock);
1401 tr->cond_snapshot = cond_snapshot;
1402 arch_spin_unlock(&tr->max_lock);
1403 local_irq_enable();
1404
1405 mutex_unlock(&trace_types_lock);
1406
1407 return ret;
1408
1409 fail_unlock:
1410 mutex_unlock(&trace_types_lock);
1411 kfree(cond_snapshot);
1412 return ret;
1413}
1414EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1415
1416/**
1417 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1418 * @tr: The tracing instance
1419 *
1420 * Check whether the conditional snapshot for the given instance is
1421 * enabled; if so, free the cond_snapshot associated with it,
1422 * otherwise return -EINVAL.
1423 *
1424 * Returns 0 if successful, error otherwise.
1425 */
1426int tracing_snapshot_cond_disable(struct trace_array *tr)
1427{
1428 int ret = 0;
1429
1430 local_irq_disable();
1431 arch_spin_lock(&tr->max_lock);
1432
1433 if (!tr->cond_snapshot)
1434 ret = -EINVAL;
1435 else {
1436 kfree(tr->cond_snapshot);
1437 tr->cond_snapshot = NULL;
1438 }
1439
1440 arch_spin_unlock(&tr->max_lock);
1441 local_irq_enable();
1442
1443 return ret;
1444}
1445EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1446#else
1447void tracing_snapshot(void)
1448{
1449 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1450}
1451EXPORT_SYMBOL_GPL(tracing_snapshot);
1452void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1453{
1454 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1455}
1456EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1457int tracing_alloc_snapshot(void)
1458{
1459 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1460 return -ENODEV;
1461}
1462EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1463void tracing_snapshot_alloc(void)
1464{
1465 /* Give warning */
1466 tracing_snapshot();
1467}
1468EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1469void *tracing_cond_snapshot_data(struct trace_array *tr)
1470{
1471 return NULL;
1472}
1473EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1474int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1475{
1476 return -ENODEV;
1477}
1478EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1479int tracing_snapshot_cond_disable(struct trace_array *tr)
1480{
1481 return false;
1482}
1483EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1484#define free_snapshot(tr) do { } while (0)
1485#endif /* CONFIG_TRACER_SNAPSHOT */
1486
1487void tracer_tracing_off(struct trace_array *tr)
1488{
1489 if (tr->array_buffer.buffer)
1490 ring_buffer_record_off(tr->array_buffer.buffer);
1491 /*
1492 * This flag is looked at when buffers haven't been allocated
1493 * yet, or by some tracers (like irqsoff), that just want to
1494 * know if the ring buffer has been disabled, but it can handle
1495 * races of where it gets disabled but we still do a record.
1496 * As the check is in the fast path of the tracers, it is more
1497 * important to be fast than accurate.
1498 */
1499 tr->buffer_disabled = 1;
1500 /* Make the flag seen by readers */
1501 smp_wmb();
1502}
1503
1504/**
1505 * tracing_off - turn off tracing buffers
1506 *
1507 * This function stops the tracing buffers from recording data.
1508 * It does not disable any overhead the tracers themselves may
1509 * be causing. This function simply causes all recording to
1510 * the ring buffers to fail.
1511 */
1512void tracing_off(void)
1513{
1514 tracer_tracing_off(&global_trace);
1515}
1516EXPORT_SYMBOL_GPL(tracing_off);
1517
1518void disable_trace_on_warning(void)
1519{
1520 if (__disable_trace_on_warning) {
1521 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1522 "Disabling tracing due to warning\n");
1523 tracing_off();
1524 }
1525}
1526
1527/**
1528 * tracer_tracing_is_on - show real state of ring buffer enabled
1529 * @tr : the trace array to know if ring buffer is enabled
1530 *
1531 * Shows real state of the ring buffer if it is enabled or not.
1532 */
1533bool tracer_tracing_is_on(struct trace_array *tr)
1534{
1535 if (tr->array_buffer.buffer)
1536 return ring_buffer_record_is_set_on(tr->array_buffer.buffer);
1537 return !tr->buffer_disabled;
1538}
1539
1540/**
1541 * tracing_is_on - show state of ring buffers enabled
1542 */
1543int tracing_is_on(void)
1544{
1545 return tracer_tracing_is_on(&global_trace);
1546}
1547EXPORT_SYMBOL_GPL(tracing_is_on);
1548
1549static int __init set_buf_size(char *str)
1550{
1551 unsigned long buf_size;
1552
1553 if (!str)
1554 return 0;
1555 buf_size = memparse(str, &str);
1556 /*
1557 * nr_entries can not be zero and the startup
1558 * tests require some buffer space. Therefore
1559 * ensure we have at least 4096 bytes of buffer.
1560 */
1561 trace_buf_size = max(4096UL, buf_size);
1562 return 1;
1563}
1564__setup("trace_buf_size=", set_buf_size);
1565
1566static int __init set_tracing_thresh(char *str)
1567{
1568 unsigned long threshold;
1569 int ret;
1570
1571 if (!str)
1572 return 0;
1573 ret = kstrtoul(str, 0, &threshold);
1574 if (ret < 0)
1575 return 0;
1576 tracing_thresh = threshold * 1000;
1577 return 1;
1578}
1579__setup("tracing_thresh=", set_tracing_thresh);
1580
1581unsigned long nsecs_to_usecs(unsigned long nsecs)
1582{
1583 return nsecs / 1000;
1584}
1585
1586/*
1587 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1588 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1589 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1590 * of strings in the order that the evals (enum) were defined.
1591 */
1592#undef C
1593#define C(a, b) b
1594
1595/* These must match the bit positions in trace_iterator_flags */
1596static const char *trace_options[] = {
1597 TRACE_FLAGS
1598 NULL
1599};
1600
1601static struct {
1602 u64 (*func)(void);
1603 const char *name;
1604 int in_ns; /* is this clock in nanoseconds? */
1605} trace_clocks[] = {
1606 { trace_clock_local, "local", 1 },
1607 { trace_clock_global, "global", 1 },
1608 { trace_clock_counter, "counter", 0 },
1609 { trace_clock_jiffies, "uptime", 0 },
1610 { trace_clock, "perf", 1 },
1611 { ktime_get_mono_fast_ns, "mono", 1 },
1612 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1613 { ktime_get_boot_fast_ns, "boot", 1 },
1614 { ktime_get_tai_fast_ns, "tai", 1 },
1615 ARCH_TRACE_CLOCKS
1616};
1617
1618bool trace_clock_in_ns(struct trace_array *tr)
1619{
1620 if (trace_clocks[tr->clock_id].in_ns)
1621 return true;
1622
1623 return false;
1624}
1625
1626/*
1627 * trace_parser_get_init - gets the buffer for trace parser
1628 */
1629int trace_parser_get_init(struct trace_parser *parser, int size)
1630{
1631 memset(parser, 0, sizeof(*parser));
1632
1633 parser->buffer = kmalloc(size, GFP_KERNEL);
1634 if (!parser->buffer)
1635 return 1;
1636
1637 parser->size = size;
1638 return 0;
1639}
1640
1641/*
1642 * trace_parser_put - frees the buffer for trace parser
1643 */
1644void trace_parser_put(struct trace_parser *parser)
1645{
1646 kfree(parser->buffer);
1647 parser->buffer = NULL;
1648}
1649
1650/*
1651 * trace_get_user - reads the user input string separated by space
1652 * (matched by isspace(ch))
1653 *
1654 * For each string found the 'struct trace_parser' is updated,
1655 * and the function returns.
1656 *
1657 * Returns number of bytes read.
1658 *
1659 * See kernel/trace/trace.h for 'struct trace_parser' details.
1660 */
1661int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1662 size_t cnt, loff_t *ppos)
1663{
1664 char ch;
1665 size_t read = 0;
1666 ssize_t ret;
1667
1668 if (!*ppos)
1669 trace_parser_clear(parser);
1670
1671 ret = get_user(ch, ubuf++);
1672 if (ret)
1673 goto out;
1674
1675 read++;
1676 cnt--;
1677
1678 /*
1679 * The parser is not finished with the last write,
1680 * continue reading the user input without skipping spaces.
1681 */
1682 if (!parser->cont) {
1683 /* skip white space */
1684 while (cnt && isspace(ch)) {
1685 ret = get_user(ch, ubuf++);
1686 if (ret)
1687 goto out;
1688 read++;
1689 cnt--;
1690 }
1691
1692 parser->idx = 0;
1693
1694 /* only spaces were written */
1695 if (isspace(ch) || !ch) {
1696 *ppos += read;
1697 ret = read;
1698 goto out;
1699 }
1700 }
1701
1702 /* read the non-space input */
1703 while (cnt && !isspace(ch) && ch) {
1704 if (parser->idx < parser->size - 1)
1705 parser->buffer[parser->idx++] = ch;
1706 else {
1707 ret = -EINVAL;
1708 goto out;
1709 }
1710 ret = get_user(ch, ubuf++);
1711 if (ret)
1712 goto out;
1713 read++;
1714 cnt--;
1715 }
1716
1717 /* We either got finished input or we have to wait for another call. */
1718 if (isspace(ch) || !ch) {
1719 parser->buffer[parser->idx] = 0;
1720 parser->cont = false;
1721 } else if (parser->idx < parser->size - 1) {
1722 parser->cont = true;
1723 parser->buffer[parser->idx++] = ch;
1724 /* Make sure the parsed string always terminates with '\0'. */
1725 parser->buffer[parser->idx] = 0;
1726 } else {
1727 ret = -EINVAL;
1728 goto out;
1729 }
1730
1731 *ppos += read;
1732 ret = read;
1733
1734out:
1735 return ret;
1736}
1737
1738/* TODO add a seq_buf_to_buffer() */
1739static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1740{
1741 int len;
1742
1743 if (trace_seq_used(s) <= s->readpos)
1744 return -EBUSY;
1745
1746 len = trace_seq_used(s) - s->readpos;
1747 if (cnt > len)
1748 cnt = len;
1749 memcpy(buf, s->buffer + s->readpos, cnt);
1750
1751 s->readpos += cnt;
1752 return cnt;
1753}
1754
1755unsigned long __read_mostly tracing_thresh;
1756
1757#ifdef CONFIG_TRACER_MAX_TRACE
1758static const struct file_operations tracing_max_lat_fops;
1759
1760#ifdef LATENCY_FS_NOTIFY
1761
1762static struct workqueue_struct *fsnotify_wq;
1763
1764static void latency_fsnotify_workfn(struct work_struct *work)
1765{
1766 struct trace_array *tr = container_of(work, struct trace_array,
1767 fsnotify_work);
1768 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1769}
1770
1771static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1772{
1773 struct trace_array *tr = container_of(iwork, struct trace_array,
1774 fsnotify_irqwork);
1775 queue_work(fsnotify_wq, &tr->fsnotify_work);
1776}
1777
1778static void trace_create_maxlat_file(struct trace_array *tr,
1779 struct dentry *d_tracer)
1780{
1781 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1782 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1783 tr->d_max_latency = trace_create_file("tracing_max_latency",
1784 TRACE_MODE_WRITE,
1785 d_tracer, tr,
1786 &tracing_max_lat_fops);
1787}
1788
1789__init static int latency_fsnotify_init(void)
1790{
1791 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1792 WQ_UNBOUND | WQ_HIGHPRI, 0);
1793 if (!fsnotify_wq) {
1794 pr_err("Unable to allocate tr_max_lat_wq\n");
1795 return -ENOMEM;
1796 }
1797 return 0;
1798}
1799
1800late_initcall_sync(latency_fsnotify_init);
1801
1802void latency_fsnotify(struct trace_array *tr)
1803{
1804 if (!fsnotify_wq)
1805 return;
1806 /*
1807 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1808 * possible that we are called from __schedule() or do_idle(), which
1809 * could cause a deadlock.
1810 */
1811 irq_work_queue(&tr->fsnotify_irqwork);
1812}
1813
1814#else /* !LATENCY_FS_NOTIFY */
1815
1816#define trace_create_maxlat_file(tr, d_tracer) \
1817 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1818 d_tracer, tr, &tracing_max_lat_fops)
1819
1820#endif
1821
1822/*
1823 * Copy the new maximum trace into the separate maximum-trace
1824 * structure. (this way the maximum trace is permanently saved,
1825 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1826 */
1827static void
1828__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1829{
1830 struct array_buffer *trace_buf = &tr->array_buffer;
1831 struct array_buffer *max_buf = &tr->max_buffer;
1832 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1833 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1834
1835 max_buf->cpu = cpu;
1836 max_buf->time_start = data->preempt_timestamp;
1837
1838 max_data->saved_latency = tr->max_latency;
1839 max_data->critical_start = data->critical_start;
1840 max_data->critical_end = data->critical_end;
1841
1842 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1843 max_data->pid = tsk->pid;
1844 /*
1845 * If tsk == current, then use current_uid(), as that does not use
1846 * RCU. The irq tracer can be called out of RCU scope.
1847 */
1848 if (tsk == current)
1849 max_data->uid = current_uid();
1850 else
1851 max_data->uid = task_uid(tsk);
1852
1853 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1854 max_data->policy = tsk->policy;
1855 max_data->rt_priority = tsk->rt_priority;
1856
1857 /* record this tasks comm */
1858 tracing_record_cmdline(tsk);
1859 latency_fsnotify(tr);
1860}
1861
1862/**
1863 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1864 * @tr: tracer
1865 * @tsk: the task with the latency
1866 * @cpu: The cpu that initiated the trace.
1867 * @cond_data: User data associated with a conditional snapshot
1868 *
1869 * Flip the buffers between the @tr and the max_tr and record information
1870 * about which task was the cause of this latency.
1871 */
1872void
1873update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1874 void *cond_data)
1875{
1876 if (tr->stop_count)
1877 return;
1878
1879 WARN_ON_ONCE(!irqs_disabled());
1880
1881 if (!tr->allocated_snapshot) {
1882 /* Only the nop tracer should hit this when disabling */
1883 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1884 return;
1885 }
1886
1887 arch_spin_lock(&tr->max_lock);
1888
1889 /* Inherit the recordable setting from array_buffer */
1890 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1891 ring_buffer_record_on(tr->max_buffer.buffer);
1892 else
1893 ring_buffer_record_off(tr->max_buffer.buffer);
1894
1895#ifdef CONFIG_TRACER_SNAPSHOT
1896 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1897 arch_spin_unlock(&tr->max_lock);
1898 return;
1899 }
1900#endif
1901 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1902
1903 __update_max_tr(tr, tsk, cpu);
1904
1905 arch_spin_unlock(&tr->max_lock);
1906
1907 /* Any waiters on the old snapshot buffer need to wake up */
1908 ring_buffer_wake_waiters(tr->array_buffer.buffer, RING_BUFFER_ALL_CPUS);
1909}
1910
1911/**
1912 * update_max_tr_single - only copy one trace over, and reset the rest
1913 * @tr: tracer
1914 * @tsk: task with the latency
1915 * @cpu: the cpu of the buffer to copy.
1916 *
1917 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1918 */
1919void
1920update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1921{
1922 int ret;
1923
1924 if (tr->stop_count)
1925 return;
1926
1927 WARN_ON_ONCE(!irqs_disabled());
1928 if (!tr->allocated_snapshot) {
1929 /* Only the nop tracer should hit this when disabling */
1930 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1931 return;
1932 }
1933
1934 arch_spin_lock(&tr->max_lock);
1935
1936 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1937
1938 if (ret == -EBUSY) {
1939 /*
1940 * We failed to swap the buffer due to a commit taking
1941 * place on this CPU. We fail to record, but we reset
1942 * the max trace buffer (no one writes directly to it)
1943 * and flag that it failed.
1944 * Another reason is resize is in progress.
1945 */
1946 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1947 "Failed to swap buffers due to commit or resize in progress\n");
1948 }
1949
1950 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1951
1952 __update_max_tr(tr, tsk, cpu);
1953 arch_spin_unlock(&tr->max_lock);
1954}
1955
1956#endif /* CONFIG_TRACER_MAX_TRACE */
1957
1958static int wait_on_pipe(struct trace_iterator *iter, int full)
1959{
1960 int ret;
1961
1962 /* Iterators are static, they should be filled or empty */
1963 if (trace_buffer_iter(iter, iter->cpu_file))
1964 return 0;
1965
1966 ret = ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file, full);
1967
1968#ifdef CONFIG_TRACER_MAX_TRACE
1969 /*
1970 * Make sure this is still the snapshot buffer, as if a snapshot were
1971 * to happen, this would now be the main buffer.
1972 */
1973 if (iter->snapshot)
1974 iter->array_buffer = &iter->tr->max_buffer;
1975#endif
1976 return ret;
1977}
1978
1979#ifdef CONFIG_FTRACE_STARTUP_TEST
1980static bool selftests_can_run;
1981
1982struct trace_selftests {
1983 struct list_head list;
1984 struct tracer *type;
1985};
1986
1987static LIST_HEAD(postponed_selftests);
1988
1989static int save_selftest(struct tracer *type)
1990{
1991 struct trace_selftests *selftest;
1992
1993 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1994 if (!selftest)
1995 return -ENOMEM;
1996
1997 selftest->type = type;
1998 list_add(&selftest->list, &postponed_selftests);
1999 return 0;
2000}
2001
2002static int run_tracer_selftest(struct tracer *type)
2003{
2004 struct trace_array *tr = &global_trace;
2005 struct tracer *saved_tracer = tr->current_trace;
2006 int ret;
2007
2008 if (!type->selftest || tracing_selftest_disabled)
2009 return 0;
2010
2011 /*
2012 * If a tracer registers early in boot up (before scheduling is
2013 * initialized and such), then do not run its selftests yet.
2014 * Instead, run it a little later in the boot process.
2015 */
2016 if (!selftests_can_run)
2017 return save_selftest(type);
2018
2019 if (!tracing_is_on()) {
2020 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2021 type->name);
2022 return 0;
2023 }
2024
2025 /*
2026 * Run a selftest on this tracer.
2027 * Here we reset the trace buffer, and set the current
2028 * tracer to be this tracer. The tracer can then run some
2029 * internal tracing to verify that everything is in order.
2030 * If we fail, we do not register this tracer.
2031 */
2032 tracing_reset_online_cpus(&tr->array_buffer);
2033
2034 tr->current_trace = type;
2035
2036#ifdef CONFIG_TRACER_MAX_TRACE
2037 if (type->use_max_tr) {
2038 /* If we expanded the buffers, make sure the max is expanded too */
2039 if (tr->ring_buffer_expanded)
2040 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2041 RING_BUFFER_ALL_CPUS);
2042 tr->allocated_snapshot = true;
2043 }
2044#endif
2045
2046 /* the test is responsible for initializing and enabling */
2047 pr_info("Testing tracer %s: ", type->name);
2048 ret = type->selftest(type, tr);
2049 /* the test is responsible for resetting too */
2050 tr->current_trace = saved_tracer;
2051 if (ret) {
2052 printk(KERN_CONT "FAILED!\n");
2053 /* Add the warning after printing 'FAILED' */
2054 WARN_ON(1);
2055 return -1;
2056 }
2057 /* Only reset on passing, to avoid touching corrupted buffers */
2058 tracing_reset_online_cpus(&tr->array_buffer);
2059
2060#ifdef CONFIG_TRACER_MAX_TRACE
2061 if (type->use_max_tr) {
2062 tr->allocated_snapshot = false;
2063
2064 /* Shrink the max buffer again */
2065 if (tr->ring_buffer_expanded)
2066 ring_buffer_resize(tr->max_buffer.buffer, 1,
2067 RING_BUFFER_ALL_CPUS);
2068 }
2069#endif
2070
2071 printk(KERN_CONT "PASSED\n");
2072 return 0;
2073}
2074
2075static int do_run_tracer_selftest(struct tracer *type)
2076{
2077 int ret;
2078
2079 /*
2080 * Tests can take a long time, especially if they are run one after the
2081 * other, as does happen during bootup when all the tracers are
2082 * registered. This could cause the soft lockup watchdog to trigger.
2083 */
2084 cond_resched();
2085
2086 tracing_selftest_running = true;
2087 ret = run_tracer_selftest(type);
2088 tracing_selftest_running = false;
2089
2090 return ret;
2091}
2092
2093static __init int init_trace_selftests(void)
2094{
2095 struct trace_selftests *p, *n;
2096 struct tracer *t, **last;
2097 int ret;
2098
2099 selftests_can_run = true;
2100
2101 mutex_lock(&trace_types_lock);
2102
2103 if (list_empty(&postponed_selftests))
2104 goto out;
2105
2106 pr_info("Running postponed tracer tests:\n");
2107
2108 tracing_selftest_running = true;
2109 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2110 /* This loop can take minutes when sanitizers are enabled, so
2111 * lets make sure we allow RCU processing.
2112 */
2113 cond_resched();
2114 ret = run_tracer_selftest(p->type);
2115 /* If the test fails, then warn and remove from available_tracers */
2116 if (ret < 0) {
2117 WARN(1, "tracer: %s failed selftest, disabling\n",
2118 p->type->name);
2119 last = &trace_types;
2120 for (t = trace_types; t; t = t->next) {
2121 if (t == p->type) {
2122 *last = t->next;
2123 break;
2124 }
2125 last = &t->next;
2126 }
2127 }
2128 list_del(&p->list);
2129 kfree(p);
2130 }
2131 tracing_selftest_running = false;
2132
2133 out:
2134 mutex_unlock(&trace_types_lock);
2135
2136 return 0;
2137}
2138core_initcall(init_trace_selftests);
2139#else
2140static inline int run_tracer_selftest(struct tracer *type)
2141{
2142 return 0;
2143}
2144static inline int do_run_tracer_selftest(struct tracer *type)
2145{
2146 return 0;
2147}
2148#endif /* CONFIG_FTRACE_STARTUP_TEST */
2149
2150static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2151
2152static void __init apply_trace_boot_options(void);
2153
2154/**
2155 * register_tracer - register a tracer with the ftrace system.
2156 * @type: the plugin for the tracer
2157 *
2158 * Register a new plugin tracer.
2159 */
2160int __init register_tracer(struct tracer *type)
2161{
2162 struct tracer *t;
2163 int ret = 0;
2164
2165 if (!type->name) {
2166 pr_info("Tracer must have a name\n");
2167 return -1;
2168 }
2169
2170 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2171 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2172 return -1;
2173 }
2174
2175 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2176 pr_warn("Can not register tracer %s due to lockdown\n",
2177 type->name);
2178 return -EPERM;
2179 }
2180
2181 mutex_lock(&trace_types_lock);
2182
2183 for (t = trace_types; t; t = t->next) {
2184 if (strcmp(type->name, t->name) == 0) {
2185 /* already found */
2186 pr_info("Tracer %s already registered\n",
2187 type->name);
2188 ret = -1;
2189 goto out;
2190 }
2191 }
2192
2193 if (!type->set_flag)
2194 type->set_flag = &dummy_set_flag;
2195 if (!type->flags) {
2196 /*allocate a dummy tracer_flags*/
2197 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2198 if (!type->flags) {
2199 ret = -ENOMEM;
2200 goto out;
2201 }
2202 type->flags->val = 0;
2203 type->flags->opts = dummy_tracer_opt;
2204 } else
2205 if (!type->flags->opts)
2206 type->flags->opts = dummy_tracer_opt;
2207
2208 /* store the tracer for __set_tracer_option */
2209 type->flags->trace = type;
2210
2211 ret = do_run_tracer_selftest(type);
2212 if (ret < 0)
2213 goto out;
2214
2215 type->next = trace_types;
2216 trace_types = type;
2217 add_tracer_options(&global_trace, type);
2218
2219 out:
2220 mutex_unlock(&trace_types_lock);
2221
2222 if (ret || !default_bootup_tracer)
2223 goto out_unlock;
2224
2225 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2226 goto out_unlock;
2227
2228 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2229 /* Do we want this tracer to start on bootup? */
2230 tracing_set_tracer(&global_trace, type->name);
2231 default_bootup_tracer = NULL;
2232
2233 apply_trace_boot_options();
2234
2235 /* disable other selftests, since this will break it. */
2236 disable_tracing_selftest("running a tracer");
2237
2238 out_unlock:
2239 return ret;
2240}
2241
2242static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2243{
2244 struct trace_buffer *buffer = buf->buffer;
2245
2246 if (!buffer)
2247 return;
2248
2249 ring_buffer_record_disable(buffer);
2250
2251 /* Make sure all commits have finished */
2252 synchronize_rcu();
2253 ring_buffer_reset_cpu(buffer, cpu);
2254
2255 ring_buffer_record_enable(buffer);
2256}
2257
2258void tracing_reset_online_cpus(struct array_buffer *buf)
2259{
2260 struct trace_buffer *buffer = buf->buffer;
2261
2262 if (!buffer)
2263 return;
2264
2265 ring_buffer_record_disable(buffer);
2266
2267 /* Make sure all commits have finished */
2268 synchronize_rcu();
2269
2270 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2271
2272 ring_buffer_reset_online_cpus(buffer);
2273
2274 ring_buffer_record_enable(buffer);
2275}
2276
2277/* Must have trace_types_lock held */
2278void tracing_reset_all_online_cpus_unlocked(void)
2279{
2280 struct trace_array *tr;
2281
2282 lockdep_assert_held(&trace_types_lock);
2283
2284 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2285 if (!tr->clear_trace)
2286 continue;
2287 tr->clear_trace = false;
2288 tracing_reset_online_cpus(&tr->array_buffer);
2289#ifdef CONFIG_TRACER_MAX_TRACE
2290 tracing_reset_online_cpus(&tr->max_buffer);
2291#endif
2292 }
2293}
2294
2295void tracing_reset_all_online_cpus(void)
2296{
2297 mutex_lock(&trace_types_lock);
2298 tracing_reset_all_online_cpus_unlocked();
2299 mutex_unlock(&trace_types_lock);
2300}
2301
2302/*
2303 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2304 * is the tgid last observed corresponding to pid=i.
2305 */
2306static int *tgid_map;
2307
2308/* The maximum valid index into tgid_map. */
2309static size_t tgid_map_max;
2310
2311#define SAVED_CMDLINES_DEFAULT 128
2312#define NO_CMDLINE_MAP UINT_MAX
2313/*
2314 * Preemption must be disabled before acquiring trace_cmdline_lock.
2315 * The various trace_arrays' max_lock must be acquired in a context
2316 * where interrupt is disabled.
2317 */
2318static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2319struct saved_cmdlines_buffer {
2320 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2321 unsigned *map_cmdline_to_pid;
2322 unsigned cmdline_num;
2323 int cmdline_idx;
2324 char saved_cmdlines[];
2325};
2326static struct saved_cmdlines_buffer *savedcmd;
2327
2328static inline char *get_saved_cmdlines(int idx)
2329{
2330 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2331}
2332
2333static inline void set_cmdline(int idx, const char *cmdline)
2334{
2335 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2336}
2337
2338static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
2339{
2340 int order = get_order(sizeof(*s) + s->cmdline_num * TASK_COMM_LEN);
2341
2342 kfree(s->map_cmdline_to_pid);
2343 kmemleak_free(s);
2344 free_pages((unsigned long)s, order);
2345}
2346
2347static struct saved_cmdlines_buffer *allocate_cmdlines_buffer(unsigned int val)
2348{
2349 struct saved_cmdlines_buffer *s;
2350 struct page *page;
2351 int orig_size, size;
2352 int order;
2353
2354 /* Figure out how much is needed to hold the given number of cmdlines */
2355 orig_size = sizeof(*s) + val * TASK_COMM_LEN;
2356 order = get_order(orig_size);
2357 size = 1 << (order + PAGE_SHIFT);
2358 page = alloc_pages(GFP_KERNEL, order);
2359 if (!page)
2360 return NULL;
2361
2362 s = page_address(page);
2363 kmemleak_alloc(s, size, 1, GFP_KERNEL);
2364 memset(s, 0, sizeof(*s));
2365
2366 /* Round up to actual allocation */
2367 val = (size - sizeof(*s)) / TASK_COMM_LEN;
2368 s->cmdline_num = val;
2369
2370 s->map_cmdline_to_pid = kmalloc_array(val,
2371 sizeof(*s->map_cmdline_to_pid),
2372 GFP_KERNEL);
2373 if (!s->map_cmdline_to_pid) {
2374 free_saved_cmdlines_buffer(s);
2375 return NULL;
2376 }
2377
2378 s->cmdline_idx = 0;
2379 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2380 sizeof(s->map_pid_to_cmdline));
2381 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2382 val * sizeof(*s->map_cmdline_to_pid));
2383
2384 return s;
2385}
2386
2387static int trace_create_savedcmd(void)
2388{
2389 savedcmd = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT);
2390
2391 return savedcmd ? 0 : -ENOMEM;
2392}
2393
2394int is_tracing_stopped(void)
2395{
2396 return global_trace.stop_count;
2397}
2398
2399static void tracing_start_tr(struct trace_array *tr)
2400{
2401 struct trace_buffer *buffer;
2402 unsigned long flags;
2403
2404 if (tracing_disabled)
2405 return;
2406
2407 raw_spin_lock_irqsave(&tr->start_lock, flags);
2408 if (--tr->stop_count) {
2409 if (WARN_ON_ONCE(tr->stop_count < 0)) {
2410 /* Someone screwed up their debugging */
2411 tr->stop_count = 0;
2412 }
2413 goto out;
2414 }
2415
2416 /* Prevent the buffers from switching */
2417 arch_spin_lock(&tr->max_lock);
2418
2419 buffer = tr->array_buffer.buffer;
2420 if (buffer)
2421 ring_buffer_record_enable(buffer);
2422
2423#ifdef CONFIG_TRACER_MAX_TRACE
2424 buffer = tr->max_buffer.buffer;
2425 if (buffer)
2426 ring_buffer_record_enable(buffer);
2427#endif
2428
2429 arch_spin_unlock(&tr->max_lock);
2430
2431 out:
2432 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2433}
2434
2435/**
2436 * tracing_start - quick start of the tracer
2437 *
2438 * If tracing is enabled but was stopped by tracing_stop,
2439 * this will start the tracer back up.
2440 */
2441void tracing_start(void)
2442
2443{
2444 return tracing_start_tr(&global_trace);
2445}
2446
2447static void tracing_stop_tr(struct trace_array *tr)
2448{
2449 struct trace_buffer *buffer;
2450 unsigned long flags;
2451
2452 raw_spin_lock_irqsave(&tr->start_lock, flags);
2453 if (tr->stop_count++)
2454 goto out;
2455
2456 /* Prevent the buffers from switching */
2457 arch_spin_lock(&tr->max_lock);
2458
2459 buffer = tr->array_buffer.buffer;
2460 if (buffer)
2461 ring_buffer_record_disable(buffer);
2462
2463#ifdef CONFIG_TRACER_MAX_TRACE
2464 buffer = tr->max_buffer.buffer;
2465 if (buffer)
2466 ring_buffer_record_disable(buffer);
2467#endif
2468
2469 arch_spin_unlock(&tr->max_lock);
2470
2471 out:
2472 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2473}
2474
2475/**
2476 * tracing_stop - quick stop of the tracer
2477 *
2478 * Light weight way to stop tracing. Use in conjunction with
2479 * tracing_start.
2480 */
2481void tracing_stop(void)
2482{
2483 return tracing_stop_tr(&global_trace);
2484}
2485
2486static int trace_save_cmdline(struct task_struct *tsk)
2487{
2488 unsigned tpid, idx;
2489
2490 /* treat recording of idle task as a success */
2491 if (!tsk->pid)
2492 return 1;
2493
2494 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2495
2496 /*
2497 * It's not the end of the world if we don't get
2498 * the lock, but we also don't want to spin
2499 * nor do we want to disable interrupts,
2500 * so if we miss here, then better luck next time.
2501 *
2502 * This is called within the scheduler and wake up, so interrupts
2503 * had better been disabled and run queue lock been held.
2504 */
2505 lockdep_assert_preemption_disabled();
2506 if (!arch_spin_trylock(&trace_cmdline_lock))
2507 return 0;
2508
2509 idx = savedcmd->map_pid_to_cmdline[tpid];
2510 if (idx == NO_CMDLINE_MAP) {
2511 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2512
2513 savedcmd->map_pid_to_cmdline[tpid] = idx;
2514 savedcmd->cmdline_idx = idx;
2515 }
2516
2517 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2518 set_cmdline(idx, tsk->comm);
2519
2520 arch_spin_unlock(&trace_cmdline_lock);
2521
2522 return 1;
2523}
2524
2525static void __trace_find_cmdline(int pid, char comm[])
2526{
2527 unsigned map;
2528 int tpid;
2529
2530 if (!pid) {
2531 strcpy(comm, "<idle>");
2532 return;
2533 }
2534
2535 if (WARN_ON_ONCE(pid < 0)) {
2536 strcpy(comm, "<XXX>");
2537 return;
2538 }
2539
2540 tpid = pid & (PID_MAX_DEFAULT - 1);
2541 map = savedcmd->map_pid_to_cmdline[tpid];
2542 if (map != NO_CMDLINE_MAP) {
2543 tpid = savedcmd->map_cmdline_to_pid[map];
2544 if (tpid == pid) {
2545 strscpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2546 return;
2547 }
2548 }
2549 strcpy(comm, "<...>");
2550}
2551
2552void trace_find_cmdline(int pid, char comm[])
2553{
2554 preempt_disable();
2555 arch_spin_lock(&trace_cmdline_lock);
2556
2557 __trace_find_cmdline(pid, comm);
2558
2559 arch_spin_unlock(&trace_cmdline_lock);
2560 preempt_enable();
2561}
2562
2563static int *trace_find_tgid_ptr(int pid)
2564{
2565 /*
2566 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2567 * if we observe a non-NULL tgid_map then we also observe the correct
2568 * tgid_map_max.
2569 */
2570 int *map = smp_load_acquire(&tgid_map);
2571
2572 if (unlikely(!map || pid > tgid_map_max))
2573 return NULL;
2574
2575 return &map[pid];
2576}
2577
2578int trace_find_tgid(int pid)
2579{
2580 int *ptr = trace_find_tgid_ptr(pid);
2581
2582 return ptr ? *ptr : 0;
2583}
2584
2585static int trace_save_tgid(struct task_struct *tsk)
2586{
2587 int *ptr;
2588
2589 /* treat recording of idle task as a success */
2590 if (!tsk->pid)
2591 return 1;
2592
2593 ptr = trace_find_tgid_ptr(tsk->pid);
2594 if (!ptr)
2595 return 0;
2596
2597 *ptr = tsk->tgid;
2598 return 1;
2599}
2600
2601static bool tracing_record_taskinfo_skip(int flags)
2602{
2603 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2604 return true;
2605 if (!__this_cpu_read(trace_taskinfo_save))
2606 return true;
2607 return false;
2608}
2609
2610/**
2611 * tracing_record_taskinfo - record the task info of a task
2612 *
2613 * @task: task to record
2614 * @flags: TRACE_RECORD_CMDLINE for recording comm
2615 * TRACE_RECORD_TGID for recording tgid
2616 */
2617void tracing_record_taskinfo(struct task_struct *task, int flags)
2618{
2619 bool done;
2620
2621 if (tracing_record_taskinfo_skip(flags))
2622 return;
2623
2624 /*
2625 * Record as much task information as possible. If some fail, continue
2626 * to try to record the others.
2627 */
2628 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2629 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2630
2631 /* If recording any information failed, retry again soon. */
2632 if (!done)
2633 return;
2634
2635 __this_cpu_write(trace_taskinfo_save, false);
2636}
2637
2638/**
2639 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2640 *
2641 * @prev: previous task during sched_switch
2642 * @next: next task during sched_switch
2643 * @flags: TRACE_RECORD_CMDLINE for recording comm
2644 * TRACE_RECORD_TGID for recording tgid
2645 */
2646void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2647 struct task_struct *next, int flags)
2648{
2649 bool done;
2650
2651 if (tracing_record_taskinfo_skip(flags))
2652 return;
2653
2654 /*
2655 * Record as much task information as possible. If some fail, continue
2656 * to try to record the others.
2657 */
2658 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2659 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2660 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2661 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2662
2663 /* If recording any information failed, retry again soon. */
2664 if (!done)
2665 return;
2666
2667 __this_cpu_write(trace_taskinfo_save, false);
2668}
2669
2670/* Helpers to record a specific task information */
2671void tracing_record_cmdline(struct task_struct *task)
2672{
2673 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2674}
2675
2676void tracing_record_tgid(struct task_struct *task)
2677{
2678 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2679}
2680
2681/*
2682 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2683 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2684 * simplifies those functions and keeps them in sync.
2685 */
2686enum print_line_t trace_handle_return(struct trace_seq *s)
2687{
2688 return trace_seq_has_overflowed(s) ?
2689 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2690}
2691EXPORT_SYMBOL_GPL(trace_handle_return);
2692
2693static unsigned short migration_disable_value(void)
2694{
2695#if defined(CONFIG_SMP)
2696 return current->migration_disabled;
2697#else
2698 return 0;
2699#endif
2700}
2701
2702unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2703{
2704 unsigned int trace_flags = irqs_status;
2705 unsigned int pc;
2706
2707 pc = preempt_count();
2708
2709 if (pc & NMI_MASK)
2710 trace_flags |= TRACE_FLAG_NMI;
2711 if (pc & HARDIRQ_MASK)
2712 trace_flags |= TRACE_FLAG_HARDIRQ;
2713 if (in_serving_softirq())
2714 trace_flags |= TRACE_FLAG_SOFTIRQ;
2715 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2716 trace_flags |= TRACE_FLAG_BH_OFF;
2717
2718 if (tif_need_resched())
2719 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2720 if (test_preempt_need_resched())
2721 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2722 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2723 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2724}
2725
2726struct ring_buffer_event *
2727trace_buffer_lock_reserve(struct trace_buffer *buffer,
2728 int type,
2729 unsigned long len,
2730 unsigned int trace_ctx)
2731{
2732 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2733}
2734
2735DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2736DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2737static int trace_buffered_event_ref;
2738
2739/**
2740 * trace_buffered_event_enable - enable buffering events
2741 *
2742 * When events are being filtered, it is quicker to use a temporary
2743 * buffer to write the event data into if there's a likely chance
2744 * that it will not be committed. The discard of the ring buffer
2745 * is not as fast as committing, and is much slower than copying
2746 * a commit.
2747 *
2748 * When an event is to be filtered, allocate per cpu buffers to
2749 * write the event data into, and if the event is filtered and discarded
2750 * it is simply dropped, otherwise, the entire data is to be committed
2751 * in one shot.
2752 */
2753void trace_buffered_event_enable(void)
2754{
2755 struct ring_buffer_event *event;
2756 struct page *page;
2757 int cpu;
2758
2759 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2760
2761 if (trace_buffered_event_ref++)
2762 return;
2763
2764 for_each_tracing_cpu(cpu) {
2765 page = alloc_pages_node(cpu_to_node(cpu),
2766 GFP_KERNEL | __GFP_NORETRY, 0);
2767 /* This is just an optimization and can handle failures */
2768 if (!page) {
2769 pr_err("Failed to allocate event buffer\n");
2770 break;
2771 }
2772
2773 event = page_address(page);
2774 memset(event, 0, sizeof(*event));
2775
2776 per_cpu(trace_buffered_event, cpu) = event;
2777
2778 preempt_disable();
2779 if (cpu == smp_processor_id() &&
2780 __this_cpu_read(trace_buffered_event) !=
2781 per_cpu(trace_buffered_event, cpu))
2782 WARN_ON_ONCE(1);
2783 preempt_enable();
2784 }
2785}
2786
2787static void enable_trace_buffered_event(void *data)
2788{
2789 /* Probably not needed, but do it anyway */
2790 smp_rmb();
2791 this_cpu_dec(trace_buffered_event_cnt);
2792}
2793
2794static void disable_trace_buffered_event(void *data)
2795{
2796 this_cpu_inc(trace_buffered_event_cnt);
2797}
2798
2799/**
2800 * trace_buffered_event_disable - disable buffering events
2801 *
2802 * When a filter is removed, it is faster to not use the buffered
2803 * events, and to commit directly into the ring buffer. Free up
2804 * the temp buffers when there are no more users. This requires
2805 * special synchronization with current events.
2806 */
2807void trace_buffered_event_disable(void)
2808{
2809 int cpu;
2810
2811 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2812
2813 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2814 return;
2815
2816 if (--trace_buffered_event_ref)
2817 return;
2818
2819 /* For each CPU, set the buffer as used. */
2820 on_each_cpu_mask(tracing_buffer_mask, disable_trace_buffered_event,
2821 NULL, true);
2822
2823 /* Wait for all current users to finish */
2824 synchronize_rcu();
2825
2826 for_each_tracing_cpu(cpu) {
2827 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2828 per_cpu(trace_buffered_event, cpu) = NULL;
2829 }
2830
2831 /*
2832 * Wait for all CPUs that potentially started checking if they can use
2833 * their event buffer only after the previous synchronize_rcu() call and
2834 * they still read a valid pointer from trace_buffered_event. It must be
2835 * ensured they don't see cleared trace_buffered_event_cnt else they
2836 * could wrongly decide to use the pointed-to buffer which is now freed.
2837 */
2838 synchronize_rcu();
2839
2840 /* For each CPU, relinquish the buffer */
2841 on_each_cpu_mask(tracing_buffer_mask, enable_trace_buffered_event, NULL,
2842 true);
2843}
2844
2845static struct trace_buffer *temp_buffer;
2846
2847struct ring_buffer_event *
2848trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2849 struct trace_event_file *trace_file,
2850 int type, unsigned long len,
2851 unsigned int trace_ctx)
2852{
2853 struct ring_buffer_event *entry;
2854 struct trace_array *tr = trace_file->tr;
2855 int val;
2856
2857 *current_rb = tr->array_buffer.buffer;
2858
2859 if (!tr->no_filter_buffering_ref &&
2860 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2861 preempt_disable_notrace();
2862 /*
2863 * Filtering is on, so try to use the per cpu buffer first.
2864 * This buffer will simulate a ring_buffer_event,
2865 * where the type_len is zero and the array[0] will
2866 * hold the full length.
2867 * (see include/linux/ring-buffer.h for details on
2868 * how the ring_buffer_event is structured).
2869 *
2870 * Using a temp buffer during filtering and copying it
2871 * on a matched filter is quicker than writing directly
2872 * into the ring buffer and then discarding it when
2873 * it doesn't match. That is because the discard
2874 * requires several atomic operations to get right.
2875 * Copying on match and doing nothing on a failed match
2876 * is still quicker than no copy on match, but having
2877 * to discard out of the ring buffer on a failed match.
2878 */
2879 if ((entry = __this_cpu_read(trace_buffered_event))) {
2880 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2881
2882 val = this_cpu_inc_return(trace_buffered_event_cnt);
2883
2884 /*
2885 * Preemption is disabled, but interrupts and NMIs
2886 * can still come in now. If that happens after
2887 * the above increment, then it will have to go
2888 * back to the old method of allocating the event
2889 * on the ring buffer, and if the filter fails, it
2890 * will have to call ring_buffer_discard_commit()
2891 * to remove it.
2892 *
2893 * Need to also check the unlikely case that the
2894 * length is bigger than the temp buffer size.
2895 * If that happens, then the reserve is pretty much
2896 * guaranteed to fail, as the ring buffer currently
2897 * only allows events less than a page. But that may
2898 * change in the future, so let the ring buffer reserve
2899 * handle the failure in that case.
2900 */
2901 if (val == 1 && likely(len <= max_len)) {
2902 trace_event_setup(entry, type, trace_ctx);
2903 entry->array[0] = len;
2904 /* Return with preemption disabled */
2905 return entry;
2906 }
2907 this_cpu_dec(trace_buffered_event_cnt);
2908 }
2909 /* __trace_buffer_lock_reserve() disables preemption */
2910 preempt_enable_notrace();
2911 }
2912
2913 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2914 trace_ctx);
2915 /*
2916 * If tracing is off, but we have triggers enabled
2917 * we still need to look at the event data. Use the temp_buffer
2918 * to store the trace event for the trigger to use. It's recursive
2919 * safe and will not be recorded anywhere.
2920 */
2921 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2922 *current_rb = temp_buffer;
2923 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2924 trace_ctx);
2925 }
2926 return entry;
2927}
2928EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2929
2930static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2931static DEFINE_MUTEX(tracepoint_printk_mutex);
2932
2933static void output_printk(struct trace_event_buffer *fbuffer)
2934{
2935 struct trace_event_call *event_call;
2936 struct trace_event_file *file;
2937 struct trace_event *event;
2938 unsigned long flags;
2939 struct trace_iterator *iter = tracepoint_print_iter;
2940
2941 /* We should never get here if iter is NULL */
2942 if (WARN_ON_ONCE(!iter))
2943 return;
2944
2945 event_call = fbuffer->trace_file->event_call;
2946 if (!event_call || !event_call->event.funcs ||
2947 !event_call->event.funcs->trace)
2948 return;
2949
2950 file = fbuffer->trace_file;
2951 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2952 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2953 !filter_match_preds(file->filter, fbuffer->entry)))
2954 return;
2955
2956 event = &fbuffer->trace_file->event_call->event;
2957
2958 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2959 trace_seq_init(&iter->seq);
2960 iter->ent = fbuffer->entry;
2961 event_call->event.funcs->trace(iter, 0, event);
2962 trace_seq_putc(&iter->seq, 0);
2963 printk("%s", iter->seq.buffer);
2964
2965 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2966}
2967
2968int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2969 void *buffer, size_t *lenp,
2970 loff_t *ppos)
2971{
2972 int save_tracepoint_printk;
2973 int ret;
2974
2975 mutex_lock(&tracepoint_printk_mutex);
2976 save_tracepoint_printk = tracepoint_printk;
2977
2978 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2979
2980 /*
2981 * This will force exiting early, as tracepoint_printk
2982 * is always zero when tracepoint_printk_iter is not allocated
2983 */
2984 if (!tracepoint_print_iter)
2985 tracepoint_printk = 0;
2986
2987 if (save_tracepoint_printk == tracepoint_printk)
2988 goto out;
2989
2990 if (tracepoint_printk)
2991 static_key_enable(&tracepoint_printk_key.key);
2992 else
2993 static_key_disable(&tracepoint_printk_key.key);
2994
2995 out:
2996 mutex_unlock(&tracepoint_printk_mutex);
2997
2998 return ret;
2999}
3000
3001void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
3002{
3003 enum event_trigger_type tt = ETT_NONE;
3004 struct trace_event_file *file = fbuffer->trace_file;
3005
3006 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
3007 fbuffer->entry, &tt))
3008 goto discard;
3009
3010 if (static_key_false(&tracepoint_printk_key.key))
3011 output_printk(fbuffer);
3012
3013 if (static_branch_unlikely(&trace_event_exports_enabled))
3014 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
3015
3016 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
3017 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
3018
3019discard:
3020 if (tt)
3021 event_triggers_post_call(file, tt);
3022
3023}
3024EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
3025
3026/*
3027 * Skip 3:
3028 *
3029 * trace_buffer_unlock_commit_regs()
3030 * trace_event_buffer_commit()
3031 * trace_event_raw_event_xxx()
3032 */
3033# define STACK_SKIP 3
3034
3035void trace_buffer_unlock_commit_regs(struct trace_array *tr,
3036 struct trace_buffer *buffer,
3037 struct ring_buffer_event *event,
3038 unsigned int trace_ctx,
3039 struct pt_regs *regs)
3040{
3041 __buffer_unlock_commit(buffer, event);
3042
3043 /*
3044 * If regs is not set, then skip the necessary functions.
3045 * Note, we can still get here via blktrace, wakeup tracer
3046 * and mmiotrace, but that's ok if they lose a function or
3047 * two. They are not that meaningful.
3048 */
3049 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
3050 ftrace_trace_userstack(tr, buffer, trace_ctx);
3051}
3052
3053/*
3054 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
3055 */
3056void
3057trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
3058 struct ring_buffer_event *event)
3059{
3060 __buffer_unlock_commit(buffer, event);
3061}
3062
3063void
3064trace_function(struct trace_array *tr, unsigned long ip, unsigned long
3065 parent_ip, unsigned int trace_ctx)
3066{
3067 struct trace_event_call *call = &event_function;
3068 struct trace_buffer *buffer = tr->array_buffer.buffer;
3069 struct ring_buffer_event *event;
3070 struct ftrace_entry *entry;
3071
3072 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
3073 trace_ctx);
3074 if (!event)
3075 return;
3076 entry = ring_buffer_event_data(event);
3077 entry->ip = ip;
3078 entry->parent_ip = parent_ip;
3079
3080 if (!call_filter_check_discard(call, entry, buffer, event)) {
3081 if (static_branch_unlikely(&trace_function_exports_enabled))
3082 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
3083 __buffer_unlock_commit(buffer, event);
3084 }
3085}
3086
3087#ifdef CONFIG_STACKTRACE
3088
3089/* Allow 4 levels of nesting: normal, softirq, irq, NMI */
3090#define FTRACE_KSTACK_NESTING 4
3091
3092#define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3093
3094struct ftrace_stack {
3095 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3096};
3097
3098
3099struct ftrace_stacks {
3100 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3101};
3102
3103static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3104static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3105
3106static void __ftrace_trace_stack(struct trace_buffer *buffer,
3107 unsigned int trace_ctx,
3108 int skip, struct pt_regs *regs)
3109{
3110 struct trace_event_call *call = &event_kernel_stack;
3111 struct ring_buffer_event *event;
3112 unsigned int size, nr_entries;
3113 struct ftrace_stack *fstack;
3114 struct stack_entry *entry;
3115 int stackidx;
3116
3117 /*
3118 * Add one, for this function and the call to save_stack_trace()
3119 * If regs is set, then these functions will not be in the way.
3120 */
3121#ifndef CONFIG_UNWINDER_ORC
3122 if (!regs)
3123 skip++;
3124#endif
3125
3126 preempt_disable_notrace();
3127
3128 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3129
3130 /* This should never happen. If it does, yell once and skip */
3131 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3132 goto out;
3133
3134 /*
3135 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3136 * interrupt will either see the value pre increment or post
3137 * increment. If the interrupt happens pre increment it will have
3138 * restored the counter when it returns. We just need a barrier to
3139 * keep gcc from moving things around.
3140 */
3141 barrier();
3142
3143 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3144 size = ARRAY_SIZE(fstack->calls);
3145
3146 if (regs) {
3147 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3148 size, skip);
3149 } else {
3150 nr_entries = stack_trace_save(fstack->calls, size, skip);
3151 }
3152
3153 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3154 struct_size(entry, caller, nr_entries),
3155 trace_ctx);
3156 if (!event)
3157 goto out;
3158 entry = ring_buffer_event_data(event);
3159
3160 entry->size = nr_entries;
3161 memcpy(&entry->caller, fstack->calls,
3162 flex_array_size(entry, caller, nr_entries));
3163
3164 if (!call_filter_check_discard(call, entry, buffer, event))
3165 __buffer_unlock_commit(buffer, event);
3166
3167 out:
3168 /* Again, don't let gcc optimize things here */
3169 barrier();
3170 __this_cpu_dec(ftrace_stack_reserve);
3171 preempt_enable_notrace();
3172
3173}
3174
3175static inline void ftrace_trace_stack(struct trace_array *tr,
3176 struct trace_buffer *buffer,
3177 unsigned int trace_ctx,
3178 int skip, struct pt_regs *regs)
3179{
3180 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3181 return;
3182
3183 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3184}
3185
3186void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3187 int skip)
3188{
3189 struct trace_buffer *buffer = tr->array_buffer.buffer;
3190
3191 if (rcu_is_watching()) {
3192 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3193 return;
3194 }
3195
3196 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3197 return;
3198
3199 /*
3200 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3201 * but if the above rcu_is_watching() failed, then the NMI
3202 * triggered someplace critical, and ct_irq_enter() should
3203 * not be called from NMI.
3204 */
3205 if (unlikely(in_nmi()))
3206 return;
3207
3208 ct_irq_enter_irqson();
3209 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3210 ct_irq_exit_irqson();
3211}
3212
3213/**
3214 * trace_dump_stack - record a stack back trace in the trace buffer
3215 * @skip: Number of functions to skip (helper handlers)
3216 */
3217void trace_dump_stack(int skip)
3218{
3219 if (tracing_disabled || tracing_selftest_running)
3220 return;
3221
3222#ifndef CONFIG_UNWINDER_ORC
3223 /* Skip 1 to skip this function. */
3224 skip++;
3225#endif
3226 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3227 tracing_gen_ctx(), skip, NULL);
3228}
3229EXPORT_SYMBOL_GPL(trace_dump_stack);
3230
3231#ifdef CONFIG_USER_STACKTRACE_SUPPORT
3232static DEFINE_PER_CPU(int, user_stack_count);
3233
3234static void
3235ftrace_trace_userstack(struct trace_array *tr,
3236 struct trace_buffer *buffer, unsigned int trace_ctx)
3237{
3238 struct trace_event_call *call = &event_user_stack;
3239 struct ring_buffer_event *event;
3240 struct userstack_entry *entry;
3241
3242 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3243 return;
3244
3245 /*
3246 * NMIs can not handle page faults, even with fix ups.
3247 * The save user stack can (and often does) fault.
3248 */
3249 if (unlikely(in_nmi()))
3250 return;
3251
3252 /*
3253 * prevent recursion, since the user stack tracing may
3254 * trigger other kernel events.
3255 */
3256 preempt_disable();
3257 if (__this_cpu_read(user_stack_count))
3258 goto out;
3259
3260 __this_cpu_inc(user_stack_count);
3261
3262 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3263 sizeof(*entry), trace_ctx);
3264 if (!event)
3265 goto out_drop_count;
3266 entry = ring_buffer_event_data(event);
3267
3268 entry->tgid = current->tgid;
3269 memset(&entry->caller, 0, sizeof(entry->caller));
3270
3271 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3272 if (!call_filter_check_discard(call, entry, buffer, event))
3273 __buffer_unlock_commit(buffer, event);
3274
3275 out_drop_count:
3276 __this_cpu_dec(user_stack_count);
3277 out:
3278 preempt_enable();
3279}
3280#else /* CONFIG_USER_STACKTRACE_SUPPORT */
3281static void ftrace_trace_userstack(struct trace_array *tr,
3282 struct trace_buffer *buffer,
3283 unsigned int trace_ctx)
3284{
3285}
3286#endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3287
3288#endif /* CONFIG_STACKTRACE */
3289
3290static inline void
3291func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3292 unsigned long long delta)
3293{
3294 entry->bottom_delta_ts = delta & U32_MAX;
3295 entry->top_delta_ts = (delta >> 32);
3296}
3297
3298void trace_last_func_repeats(struct trace_array *tr,
3299 struct trace_func_repeats *last_info,
3300 unsigned int trace_ctx)
3301{
3302 struct trace_buffer *buffer = tr->array_buffer.buffer;
3303 struct func_repeats_entry *entry;
3304 struct ring_buffer_event *event;
3305 u64 delta;
3306
3307 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3308 sizeof(*entry), trace_ctx);
3309 if (!event)
3310 return;
3311
3312 delta = ring_buffer_event_time_stamp(buffer, event) -
3313 last_info->ts_last_call;
3314
3315 entry = ring_buffer_event_data(event);
3316 entry->ip = last_info->ip;
3317 entry->parent_ip = last_info->parent_ip;
3318 entry->count = last_info->count;
3319 func_repeats_set_delta_ts(entry, delta);
3320
3321 __buffer_unlock_commit(buffer, event);
3322}
3323
3324/* created for use with alloc_percpu */
3325struct trace_buffer_struct {
3326 int nesting;
3327 char buffer[4][TRACE_BUF_SIZE];
3328};
3329
3330static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3331
3332/*
3333 * This allows for lockless recording. If we're nested too deeply, then
3334 * this returns NULL.
3335 */
3336static char *get_trace_buf(void)
3337{
3338 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3339
3340 if (!trace_percpu_buffer || buffer->nesting >= 4)
3341 return NULL;
3342
3343 buffer->nesting++;
3344
3345 /* Interrupts must see nesting incremented before we use the buffer */
3346 barrier();
3347 return &buffer->buffer[buffer->nesting - 1][0];
3348}
3349
3350static void put_trace_buf(void)
3351{
3352 /* Don't let the decrement of nesting leak before this */
3353 barrier();
3354 this_cpu_dec(trace_percpu_buffer->nesting);
3355}
3356
3357static int alloc_percpu_trace_buffer(void)
3358{
3359 struct trace_buffer_struct __percpu *buffers;
3360
3361 if (trace_percpu_buffer)
3362 return 0;
3363
3364 buffers = alloc_percpu(struct trace_buffer_struct);
3365 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3366 return -ENOMEM;
3367
3368 trace_percpu_buffer = buffers;
3369 return 0;
3370}
3371
3372static int buffers_allocated;
3373
3374void trace_printk_init_buffers(void)
3375{
3376 if (buffers_allocated)
3377 return;
3378
3379 if (alloc_percpu_trace_buffer())
3380 return;
3381
3382 /* trace_printk() is for debug use only. Don't use it in production. */
3383
3384 pr_warn("\n");
3385 pr_warn("**********************************************************\n");
3386 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3387 pr_warn("** **\n");
3388 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3389 pr_warn("** **\n");
3390 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3391 pr_warn("** unsafe for production use. **\n");
3392 pr_warn("** **\n");
3393 pr_warn("** If you see this message and you are not debugging **\n");
3394 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3395 pr_warn("** **\n");
3396 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3397 pr_warn("**********************************************************\n");
3398
3399 /* Expand the buffers to set size */
3400 tracing_update_buffers(&global_trace);
3401
3402 buffers_allocated = 1;
3403
3404 /*
3405 * trace_printk_init_buffers() can be called by modules.
3406 * If that happens, then we need to start cmdline recording
3407 * directly here. If the global_trace.buffer is already
3408 * allocated here, then this was called by module code.
3409 */
3410 if (global_trace.array_buffer.buffer)
3411 tracing_start_cmdline_record();
3412}
3413EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3414
3415void trace_printk_start_comm(void)
3416{
3417 /* Start tracing comms if trace printk is set */
3418 if (!buffers_allocated)
3419 return;
3420 tracing_start_cmdline_record();
3421}
3422
3423static void trace_printk_start_stop_comm(int enabled)
3424{
3425 if (!buffers_allocated)
3426 return;
3427
3428 if (enabled)
3429 tracing_start_cmdline_record();
3430 else
3431 tracing_stop_cmdline_record();
3432}
3433
3434/**
3435 * trace_vbprintk - write binary msg to tracing buffer
3436 * @ip: The address of the caller
3437 * @fmt: The string format to write to the buffer
3438 * @args: Arguments for @fmt
3439 */
3440int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3441{
3442 struct trace_event_call *call = &event_bprint;
3443 struct ring_buffer_event *event;
3444 struct trace_buffer *buffer;
3445 struct trace_array *tr = &global_trace;
3446 struct bprint_entry *entry;
3447 unsigned int trace_ctx;
3448 char *tbuffer;
3449 int len = 0, size;
3450
3451 if (unlikely(tracing_selftest_running || tracing_disabled))
3452 return 0;
3453
3454 /* Don't pollute graph traces with trace_vprintk internals */
3455 pause_graph_tracing();
3456
3457 trace_ctx = tracing_gen_ctx();
3458 preempt_disable_notrace();
3459
3460 tbuffer = get_trace_buf();
3461 if (!tbuffer) {
3462 len = 0;
3463 goto out_nobuffer;
3464 }
3465
3466 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3467
3468 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3469 goto out_put;
3470
3471 size = sizeof(*entry) + sizeof(u32) * len;
3472 buffer = tr->array_buffer.buffer;
3473 ring_buffer_nest_start(buffer);
3474 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3475 trace_ctx);
3476 if (!event)
3477 goto out;
3478 entry = ring_buffer_event_data(event);
3479 entry->ip = ip;
3480 entry->fmt = fmt;
3481
3482 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3483 if (!call_filter_check_discard(call, entry, buffer, event)) {
3484 __buffer_unlock_commit(buffer, event);
3485 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3486 }
3487
3488out:
3489 ring_buffer_nest_end(buffer);
3490out_put:
3491 put_trace_buf();
3492
3493out_nobuffer:
3494 preempt_enable_notrace();
3495 unpause_graph_tracing();
3496
3497 return len;
3498}
3499EXPORT_SYMBOL_GPL(trace_vbprintk);
3500
3501__printf(3, 0)
3502static int
3503__trace_array_vprintk(struct trace_buffer *buffer,
3504 unsigned long ip, const char *fmt, va_list args)
3505{
3506 struct trace_event_call *call = &event_print;
3507 struct ring_buffer_event *event;
3508 int len = 0, size;
3509 struct print_entry *entry;
3510 unsigned int trace_ctx;
3511 char *tbuffer;
3512
3513 if (tracing_disabled)
3514 return 0;
3515
3516 /* Don't pollute graph traces with trace_vprintk internals */
3517 pause_graph_tracing();
3518
3519 trace_ctx = tracing_gen_ctx();
3520 preempt_disable_notrace();
3521
3522
3523 tbuffer = get_trace_buf();
3524 if (!tbuffer) {
3525 len = 0;
3526 goto out_nobuffer;
3527 }
3528
3529 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3530
3531 size = sizeof(*entry) + len + 1;
3532 ring_buffer_nest_start(buffer);
3533 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3534 trace_ctx);
3535 if (!event)
3536 goto out;
3537 entry = ring_buffer_event_data(event);
3538 entry->ip = ip;
3539
3540 memcpy(&entry->buf, tbuffer, len + 1);
3541 if (!call_filter_check_discard(call, entry, buffer, event)) {
3542 __buffer_unlock_commit(buffer, event);
3543 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3544 }
3545
3546out:
3547 ring_buffer_nest_end(buffer);
3548 put_trace_buf();
3549
3550out_nobuffer:
3551 preempt_enable_notrace();
3552 unpause_graph_tracing();
3553
3554 return len;
3555}
3556
3557__printf(3, 0)
3558int trace_array_vprintk(struct trace_array *tr,
3559 unsigned long ip, const char *fmt, va_list args)
3560{
3561 if (tracing_selftest_running && tr == &global_trace)
3562 return 0;
3563
3564 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3565}
3566
3567/**
3568 * trace_array_printk - Print a message to a specific instance
3569 * @tr: The instance trace_array descriptor
3570 * @ip: The instruction pointer that this is called from.
3571 * @fmt: The format to print (printf format)
3572 *
3573 * If a subsystem sets up its own instance, they have the right to
3574 * printk strings into their tracing instance buffer using this
3575 * function. Note, this function will not write into the top level
3576 * buffer (use trace_printk() for that), as writing into the top level
3577 * buffer should only have events that can be individually disabled.
3578 * trace_printk() is only used for debugging a kernel, and should not
3579 * be ever incorporated in normal use.
3580 *
3581 * trace_array_printk() can be used, as it will not add noise to the
3582 * top level tracing buffer.
3583 *
3584 * Note, trace_array_init_printk() must be called on @tr before this
3585 * can be used.
3586 */
3587__printf(3, 0)
3588int trace_array_printk(struct trace_array *tr,
3589 unsigned long ip, const char *fmt, ...)
3590{
3591 int ret;
3592 va_list ap;
3593
3594 if (!tr)
3595 return -ENOENT;
3596
3597 /* This is only allowed for created instances */
3598 if (tr == &global_trace)
3599 return 0;
3600
3601 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3602 return 0;
3603
3604 va_start(ap, fmt);
3605 ret = trace_array_vprintk(tr, ip, fmt, ap);
3606 va_end(ap);
3607 return ret;
3608}
3609EXPORT_SYMBOL_GPL(trace_array_printk);
3610
3611/**
3612 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3613 * @tr: The trace array to initialize the buffers for
3614 *
3615 * As trace_array_printk() only writes into instances, they are OK to
3616 * have in the kernel (unlike trace_printk()). This needs to be called
3617 * before trace_array_printk() can be used on a trace_array.
3618 */
3619int trace_array_init_printk(struct trace_array *tr)
3620{
3621 if (!tr)
3622 return -ENOENT;
3623
3624 /* This is only allowed for created instances */
3625 if (tr == &global_trace)
3626 return -EINVAL;
3627
3628 return alloc_percpu_trace_buffer();
3629}
3630EXPORT_SYMBOL_GPL(trace_array_init_printk);
3631
3632__printf(3, 4)
3633int trace_array_printk_buf(struct trace_buffer *buffer,
3634 unsigned long ip, const char *fmt, ...)
3635{
3636 int ret;
3637 va_list ap;
3638
3639 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3640 return 0;
3641
3642 va_start(ap, fmt);
3643 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3644 va_end(ap);
3645 return ret;
3646}
3647
3648__printf(2, 0)
3649int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3650{
3651 return trace_array_vprintk(&global_trace, ip, fmt, args);
3652}
3653EXPORT_SYMBOL_GPL(trace_vprintk);
3654
3655static void trace_iterator_increment(struct trace_iterator *iter)
3656{
3657 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3658
3659 iter->idx++;
3660 if (buf_iter)
3661 ring_buffer_iter_advance(buf_iter);
3662}
3663
3664static struct trace_entry *
3665peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3666 unsigned long *lost_events)
3667{
3668 struct ring_buffer_event *event;
3669 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3670
3671 if (buf_iter) {
3672 event = ring_buffer_iter_peek(buf_iter, ts);
3673 if (lost_events)
3674 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3675 (unsigned long)-1 : 0;
3676 } else {
3677 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3678 lost_events);
3679 }
3680
3681 if (event) {
3682 iter->ent_size = ring_buffer_event_length(event);
3683 return ring_buffer_event_data(event);
3684 }
3685 iter->ent_size = 0;
3686 return NULL;
3687}
3688
3689static struct trace_entry *
3690__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3691 unsigned long *missing_events, u64 *ent_ts)
3692{
3693 struct trace_buffer *buffer = iter->array_buffer->buffer;
3694 struct trace_entry *ent, *next = NULL;
3695 unsigned long lost_events = 0, next_lost = 0;
3696 int cpu_file = iter->cpu_file;
3697 u64 next_ts = 0, ts;
3698 int next_cpu = -1;
3699 int next_size = 0;
3700 int cpu;
3701
3702 /*
3703 * If we are in a per_cpu trace file, don't bother by iterating over
3704 * all cpu and peek directly.
3705 */
3706 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3707 if (ring_buffer_empty_cpu(buffer, cpu_file))
3708 return NULL;
3709 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3710 if (ent_cpu)
3711 *ent_cpu = cpu_file;
3712
3713 return ent;
3714 }
3715
3716 for_each_tracing_cpu(cpu) {
3717
3718 if (ring_buffer_empty_cpu(buffer, cpu))
3719 continue;
3720
3721 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3722
3723 /*
3724 * Pick the entry with the smallest timestamp:
3725 */
3726 if (ent && (!next || ts < next_ts)) {
3727 next = ent;
3728 next_cpu = cpu;
3729 next_ts = ts;
3730 next_lost = lost_events;
3731 next_size = iter->ent_size;
3732 }
3733 }
3734
3735 iter->ent_size = next_size;
3736
3737 if (ent_cpu)
3738 *ent_cpu = next_cpu;
3739
3740 if (ent_ts)
3741 *ent_ts = next_ts;
3742
3743 if (missing_events)
3744 *missing_events = next_lost;
3745
3746 return next;
3747}
3748
3749#define STATIC_FMT_BUF_SIZE 128
3750static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3751
3752char *trace_iter_expand_format(struct trace_iterator *iter)
3753{
3754 char *tmp;
3755
3756 /*
3757 * iter->tr is NULL when used with tp_printk, which makes
3758 * this get called where it is not safe to call krealloc().
3759 */
3760 if (!iter->tr || iter->fmt == static_fmt_buf)
3761 return NULL;
3762
3763 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3764 GFP_KERNEL);
3765 if (tmp) {
3766 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3767 iter->fmt = tmp;
3768 }
3769
3770 return tmp;
3771}
3772
3773/* Returns true if the string is safe to dereference from an event */
3774static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3775 bool star, int len)
3776{
3777 unsigned long addr = (unsigned long)str;
3778 struct trace_event *trace_event;
3779 struct trace_event_call *event;
3780
3781 /* Ignore strings with no length */
3782 if (star && !len)
3783 return true;
3784
3785 /* OK if part of the event data */
3786 if ((addr >= (unsigned long)iter->ent) &&
3787 (addr < (unsigned long)iter->ent + iter->ent_size))
3788 return true;
3789
3790 /* OK if part of the temp seq buffer */
3791 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3792 (addr < (unsigned long)iter->tmp_seq.buffer + TRACE_SEQ_BUFFER_SIZE))
3793 return true;
3794
3795 /* Core rodata can not be freed */
3796 if (is_kernel_rodata(addr))
3797 return true;
3798
3799 if (trace_is_tracepoint_string(str))
3800 return true;
3801
3802 /*
3803 * Now this could be a module event, referencing core module
3804 * data, which is OK.
3805 */
3806 if (!iter->ent)
3807 return false;
3808
3809 trace_event = ftrace_find_event(iter->ent->type);
3810 if (!trace_event)
3811 return false;
3812
3813 event = container_of(trace_event, struct trace_event_call, event);
3814 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3815 return false;
3816
3817 /* Would rather have rodata, but this will suffice */
3818 if (within_module_core(addr, event->module))
3819 return true;
3820
3821 return false;
3822}
3823
3824static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3825
3826static int test_can_verify_check(const char *fmt, ...)
3827{
3828 char buf[16];
3829 va_list ap;
3830 int ret;
3831
3832 /*
3833 * The verifier is dependent on vsnprintf() modifies the va_list
3834 * passed to it, where it is sent as a reference. Some architectures
3835 * (like x86_32) passes it by value, which means that vsnprintf()
3836 * does not modify the va_list passed to it, and the verifier
3837 * would then need to be able to understand all the values that
3838 * vsnprintf can use. If it is passed by value, then the verifier
3839 * is disabled.
3840 */
3841 va_start(ap, fmt);
3842 vsnprintf(buf, 16, "%d", ap);
3843 ret = va_arg(ap, int);
3844 va_end(ap);
3845
3846 return ret;
3847}
3848
3849static void test_can_verify(void)
3850{
3851 if (!test_can_verify_check("%d %d", 0, 1)) {
3852 pr_info("trace event string verifier disabled\n");
3853 static_branch_inc(&trace_no_verify);
3854 }
3855}
3856
3857/**
3858 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3859 * @iter: The iterator that holds the seq buffer and the event being printed
3860 * @fmt: The format used to print the event
3861 * @ap: The va_list holding the data to print from @fmt.
3862 *
3863 * This writes the data into the @iter->seq buffer using the data from
3864 * @fmt and @ap. If the format has a %s, then the source of the string
3865 * is examined to make sure it is safe to print, otherwise it will
3866 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3867 * pointer.
3868 */
3869void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3870 va_list ap)
3871{
3872 const char *p = fmt;
3873 const char *str;
3874 int i, j;
3875
3876 if (WARN_ON_ONCE(!fmt))
3877 return;
3878
3879 if (static_branch_unlikely(&trace_no_verify))
3880 goto print;
3881
3882 /* Don't bother checking when doing a ftrace_dump() */
3883 if (iter->fmt == static_fmt_buf)
3884 goto print;
3885
3886 while (*p) {
3887 bool star = false;
3888 int len = 0;
3889
3890 j = 0;
3891
3892 /* We only care about %s and variants */
3893 for (i = 0; p[i]; i++) {
3894 if (i + 1 >= iter->fmt_size) {
3895 /*
3896 * If we can't expand the copy buffer,
3897 * just print it.
3898 */
3899 if (!trace_iter_expand_format(iter))
3900 goto print;
3901 }
3902
3903 if (p[i] == '\\' && p[i+1]) {
3904 i++;
3905 continue;
3906 }
3907 if (p[i] == '%') {
3908 /* Need to test cases like %08.*s */
3909 for (j = 1; p[i+j]; j++) {
3910 if (isdigit(p[i+j]) ||
3911 p[i+j] == '.')
3912 continue;
3913 if (p[i+j] == '*') {
3914 star = true;
3915 continue;
3916 }
3917 break;
3918 }
3919 if (p[i+j] == 's')
3920 break;
3921 star = false;
3922 }
3923 j = 0;
3924 }
3925 /* If no %s found then just print normally */
3926 if (!p[i])
3927 break;
3928
3929 /* Copy up to the %s, and print that */
3930 strncpy(iter->fmt, p, i);
3931 iter->fmt[i] = '\0';
3932 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3933
3934 /*
3935 * If iter->seq is full, the above call no longer guarantees
3936 * that ap is in sync with fmt processing, and further calls
3937 * to va_arg() can return wrong positional arguments.
3938 *
3939 * Ensure that ap is no longer used in this case.
3940 */
3941 if (iter->seq.full) {
3942 p = "";
3943 break;
3944 }
3945
3946 if (star)
3947 len = va_arg(ap, int);
3948
3949 /* The ap now points to the string data of the %s */
3950 str = va_arg(ap, const char *);
3951
3952 /*
3953 * If you hit this warning, it is likely that the
3954 * trace event in question used %s on a string that
3955 * was saved at the time of the event, but may not be
3956 * around when the trace is read. Use __string(),
3957 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3958 * instead. See samples/trace_events/trace-events-sample.h
3959 * for reference.
3960 */
3961 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3962 "fmt: '%s' current_buffer: '%s'",
3963 fmt, seq_buf_str(&iter->seq.seq))) {
3964 int ret;
3965
3966 /* Try to safely read the string */
3967 if (star) {
3968 if (len + 1 > iter->fmt_size)
3969 len = iter->fmt_size - 1;
3970 if (len < 0)
3971 len = 0;
3972 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3973 iter->fmt[len] = 0;
3974 star = false;
3975 } else {
3976 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3977 iter->fmt_size);
3978 }
3979 if (ret < 0)
3980 trace_seq_printf(&iter->seq, "(0x%px)", str);
3981 else
3982 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3983 str, iter->fmt);
3984 str = "[UNSAFE-MEMORY]";
3985 strcpy(iter->fmt, "%s");
3986 } else {
3987 strncpy(iter->fmt, p + i, j + 1);
3988 iter->fmt[j+1] = '\0';
3989 }
3990 if (star)
3991 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3992 else
3993 trace_seq_printf(&iter->seq, iter->fmt, str);
3994
3995 p += i + j + 1;
3996 }
3997 print:
3998 if (*p)
3999 trace_seq_vprintf(&iter->seq, p, ap);
4000}
4001
4002const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
4003{
4004 const char *p, *new_fmt;
4005 char *q;
4006
4007 if (WARN_ON_ONCE(!fmt))
4008 return fmt;
4009
4010 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
4011 return fmt;
4012
4013 p = fmt;
4014 new_fmt = q = iter->fmt;
4015 while (*p) {
4016 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
4017 if (!trace_iter_expand_format(iter))
4018 return fmt;
4019
4020 q += iter->fmt - new_fmt;
4021 new_fmt = iter->fmt;
4022 }
4023
4024 *q++ = *p++;
4025
4026 /* Replace %p with %px */
4027 if (p[-1] == '%') {
4028 if (p[0] == '%') {
4029 *q++ = *p++;
4030 } else if (p[0] == 'p' && !isalnum(p[1])) {
4031 *q++ = *p++;
4032 *q++ = 'x';
4033 }
4034 }
4035 }
4036 *q = '\0';
4037
4038 return new_fmt;
4039}
4040
4041#define STATIC_TEMP_BUF_SIZE 128
4042static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
4043
4044/* Find the next real entry, without updating the iterator itself */
4045struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
4046 int *ent_cpu, u64 *ent_ts)
4047{
4048 /* __find_next_entry will reset ent_size */
4049 int ent_size = iter->ent_size;
4050 struct trace_entry *entry;
4051
4052 /*
4053 * If called from ftrace_dump(), then the iter->temp buffer
4054 * will be the static_temp_buf and not created from kmalloc.
4055 * If the entry size is greater than the buffer, we can
4056 * not save it. Just return NULL in that case. This is only
4057 * used to add markers when two consecutive events' time
4058 * stamps have a large delta. See trace_print_lat_context()
4059 */
4060 if (iter->temp == static_temp_buf &&
4061 STATIC_TEMP_BUF_SIZE < ent_size)
4062 return NULL;
4063
4064 /*
4065 * The __find_next_entry() may call peek_next_entry(), which may
4066 * call ring_buffer_peek() that may make the contents of iter->ent
4067 * undefined. Need to copy iter->ent now.
4068 */
4069 if (iter->ent && iter->ent != iter->temp) {
4070 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4071 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4072 void *temp;
4073 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4074 if (!temp)
4075 return NULL;
4076 kfree(iter->temp);
4077 iter->temp = temp;
4078 iter->temp_size = iter->ent_size;
4079 }
4080 memcpy(iter->temp, iter->ent, iter->ent_size);
4081 iter->ent = iter->temp;
4082 }
4083 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4084 /* Put back the original ent_size */
4085 iter->ent_size = ent_size;
4086
4087 return entry;
4088}
4089
4090/* Find the next real entry, and increment the iterator to the next entry */
4091void *trace_find_next_entry_inc(struct trace_iterator *iter)
4092{
4093 iter->ent = __find_next_entry(iter, &iter->cpu,
4094 &iter->lost_events, &iter->ts);
4095
4096 if (iter->ent)
4097 trace_iterator_increment(iter);
4098
4099 return iter->ent ? iter : NULL;
4100}
4101
4102static void trace_consume(struct trace_iterator *iter)
4103{
4104 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4105 &iter->lost_events);
4106}
4107
4108static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4109{
4110 struct trace_iterator *iter = m->private;
4111 int i = (int)*pos;
4112 void *ent;
4113
4114 WARN_ON_ONCE(iter->leftover);
4115
4116 (*pos)++;
4117
4118 /* can't go backwards */
4119 if (iter->idx > i)
4120 return NULL;
4121
4122 if (iter->idx < 0)
4123 ent = trace_find_next_entry_inc(iter);
4124 else
4125 ent = iter;
4126
4127 while (ent && iter->idx < i)
4128 ent = trace_find_next_entry_inc(iter);
4129
4130 iter->pos = *pos;
4131
4132 return ent;
4133}
4134
4135void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4136{
4137 struct ring_buffer_iter *buf_iter;
4138 unsigned long entries = 0;
4139 u64 ts;
4140
4141 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4142
4143 buf_iter = trace_buffer_iter(iter, cpu);
4144 if (!buf_iter)
4145 return;
4146
4147 ring_buffer_iter_reset(buf_iter);
4148
4149 /*
4150 * We could have the case with the max latency tracers
4151 * that a reset never took place on a cpu. This is evident
4152 * by the timestamp being before the start of the buffer.
4153 */
4154 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4155 if (ts >= iter->array_buffer->time_start)
4156 break;
4157 entries++;
4158 ring_buffer_iter_advance(buf_iter);
4159 }
4160
4161 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4162}
4163
4164/*
4165 * The current tracer is copied to avoid a global locking
4166 * all around.
4167 */
4168static void *s_start(struct seq_file *m, loff_t *pos)
4169{
4170 struct trace_iterator *iter = m->private;
4171 struct trace_array *tr = iter->tr;
4172 int cpu_file = iter->cpu_file;
4173 void *p = NULL;
4174 loff_t l = 0;
4175 int cpu;
4176
4177 mutex_lock(&trace_types_lock);
4178 if (unlikely(tr->current_trace != iter->trace)) {
4179 /* Close iter->trace before switching to the new current tracer */
4180 if (iter->trace->close)
4181 iter->trace->close(iter);
4182 iter->trace = tr->current_trace;
4183 /* Reopen the new current tracer */
4184 if (iter->trace->open)
4185 iter->trace->open(iter);
4186 }
4187 mutex_unlock(&trace_types_lock);
4188
4189#ifdef CONFIG_TRACER_MAX_TRACE
4190 if (iter->snapshot && iter->trace->use_max_tr)
4191 return ERR_PTR(-EBUSY);
4192#endif
4193
4194 if (*pos != iter->pos) {
4195 iter->ent = NULL;
4196 iter->cpu = 0;
4197 iter->idx = -1;
4198
4199 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4200 for_each_tracing_cpu(cpu)
4201 tracing_iter_reset(iter, cpu);
4202 } else
4203 tracing_iter_reset(iter, cpu_file);
4204
4205 iter->leftover = 0;
4206 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4207 ;
4208
4209 } else {
4210 /*
4211 * If we overflowed the seq_file before, then we want
4212 * to just reuse the trace_seq buffer again.
4213 */
4214 if (iter->leftover)
4215 p = iter;
4216 else {
4217 l = *pos - 1;
4218 p = s_next(m, p, &l);
4219 }
4220 }
4221
4222 trace_event_read_lock();
4223 trace_access_lock(cpu_file);
4224 return p;
4225}
4226
4227static void s_stop(struct seq_file *m, void *p)
4228{
4229 struct trace_iterator *iter = m->private;
4230
4231#ifdef CONFIG_TRACER_MAX_TRACE
4232 if (iter->snapshot && iter->trace->use_max_tr)
4233 return;
4234#endif
4235
4236 trace_access_unlock(iter->cpu_file);
4237 trace_event_read_unlock();
4238}
4239
4240static void
4241get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4242 unsigned long *entries, int cpu)
4243{
4244 unsigned long count;
4245
4246 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4247 /*
4248 * If this buffer has skipped entries, then we hold all
4249 * entries for the trace and we need to ignore the
4250 * ones before the time stamp.
4251 */
4252 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4253 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4254 /* total is the same as the entries */
4255 *total = count;
4256 } else
4257 *total = count +
4258 ring_buffer_overrun_cpu(buf->buffer, cpu);
4259 *entries = count;
4260}
4261
4262static void
4263get_total_entries(struct array_buffer *buf,
4264 unsigned long *total, unsigned long *entries)
4265{
4266 unsigned long t, e;
4267 int cpu;
4268
4269 *total = 0;
4270 *entries = 0;
4271
4272 for_each_tracing_cpu(cpu) {
4273 get_total_entries_cpu(buf, &t, &e, cpu);
4274 *total += t;
4275 *entries += e;
4276 }
4277}
4278
4279unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4280{
4281 unsigned long total, entries;
4282
4283 if (!tr)
4284 tr = &global_trace;
4285
4286 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4287
4288 return entries;
4289}
4290
4291unsigned long trace_total_entries(struct trace_array *tr)
4292{
4293 unsigned long total, entries;
4294
4295 if (!tr)
4296 tr = &global_trace;
4297
4298 get_total_entries(&tr->array_buffer, &total, &entries);
4299
4300 return entries;
4301}
4302
4303static void print_lat_help_header(struct seq_file *m)
4304{
4305 seq_puts(m, "# _------=> CPU# \n"
4306 "# / _-----=> irqs-off/BH-disabled\n"
4307 "# | / _----=> need-resched \n"
4308 "# || / _---=> hardirq/softirq \n"
4309 "# ||| / _--=> preempt-depth \n"
4310 "# |||| / _-=> migrate-disable \n"
4311 "# ||||| / delay \n"
4312 "# cmd pid |||||| time | caller \n"
4313 "# \\ / |||||| \\ | / \n");
4314}
4315
4316static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4317{
4318 unsigned long total;
4319 unsigned long entries;
4320
4321 get_total_entries(buf, &total, &entries);
4322 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4323 entries, total, num_online_cpus());
4324 seq_puts(m, "#\n");
4325}
4326
4327static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4328 unsigned int flags)
4329{
4330 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4331
4332 print_event_info(buf, m);
4333
4334 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4335 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4336}
4337
4338static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4339 unsigned int flags)
4340{
4341 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4342 static const char space[] = " ";
4343 int prec = tgid ? 12 : 2;
4344
4345 print_event_info(buf, m);
4346
4347 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4348 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4349 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4350 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4351 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4352 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4353 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4354 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4355}
4356
4357void
4358print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4359{
4360 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4361 struct array_buffer *buf = iter->array_buffer;
4362 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4363 struct tracer *type = iter->trace;
4364 unsigned long entries;
4365 unsigned long total;
4366 const char *name = type->name;
4367
4368 get_total_entries(buf, &total, &entries);
4369
4370 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4371 name, UTS_RELEASE);
4372 seq_puts(m, "# -----------------------------------"
4373 "---------------------------------\n");
4374 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4375 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4376 nsecs_to_usecs(data->saved_latency),
4377 entries,
4378 total,
4379 buf->cpu,
4380 preempt_model_none() ? "server" :
4381 preempt_model_voluntary() ? "desktop" :
4382 preempt_model_full() ? "preempt" :
4383 preempt_model_rt() ? "preempt_rt" :
4384 "unknown",
4385 /* These are reserved for later use */
4386 0, 0, 0, 0);
4387#ifdef CONFIG_SMP
4388 seq_printf(m, " #P:%d)\n", num_online_cpus());
4389#else
4390 seq_puts(m, ")\n");
4391#endif
4392 seq_puts(m, "# -----------------\n");
4393 seq_printf(m, "# | task: %.16s-%d "
4394 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4395 data->comm, data->pid,
4396 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4397 data->policy, data->rt_priority);
4398 seq_puts(m, "# -----------------\n");
4399
4400 if (data->critical_start) {
4401 seq_puts(m, "# => started at: ");
4402 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4403 trace_print_seq(m, &iter->seq);
4404 seq_puts(m, "\n# => ended at: ");
4405 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4406 trace_print_seq(m, &iter->seq);
4407 seq_puts(m, "\n#\n");
4408 }
4409
4410 seq_puts(m, "#\n");
4411}
4412
4413static void test_cpu_buff_start(struct trace_iterator *iter)
4414{
4415 struct trace_seq *s = &iter->seq;
4416 struct trace_array *tr = iter->tr;
4417
4418 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4419 return;
4420
4421 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4422 return;
4423
4424 if (cpumask_available(iter->started) &&
4425 cpumask_test_cpu(iter->cpu, iter->started))
4426 return;
4427
4428 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4429 return;
4430
4431 if (cpumask_available(iter->started))
4432 cpumask_set_cpu(iter->cpu, iter->started);
4433
4434 /* Don't print started cpu buffer for the first entry of the trace */
4435 if (iter->idx > 1)
4436 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4437 iter->cpu);
4438}
4439
4440static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4441{
4442 struct trace_array *tr = iter->tr;
4443 struct trace_seq *s = &iter->seq;
4444 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4445 struct trace_entry *entry;
4446 struct trace_event *event;
4447
4448 entry = iter->ent;
4449
4450 test_cpu_buff_start(iter);
4451
4452 event = ftrace_find_event(entry->type);
4453
4454 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4455 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4456 trace_print_lat_context(iter);
4457 else
4458 trace_print_context(iter);
4459 }
4460
4461 if (trace_seq_has_overflowed(s))
4462 return TRACE_TYPE_PARTIAL_LINE;
4463
4464 if (event) {
4465 if (tr->trace_flags & TRACE_ITER_FIELDS)
4466 return print_event_fields(iter, event);
4467 return event->funcs->trace(iter, sym_flags, event);
4468 }
4469
4470 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4471
4472 return trace_handle_return(s);
4473}
4474
4475static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4476{
4477 struct trace_array *tr = iter->tr;
4478 struct trace_seq *s = &iter->seq;
4479 struct trace_entry *entry;
4480 struct trace_event *event;
4481
4482 entry = iter->ent;
4483
4484 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4485 trace_seq_printf(s, "%d %d %llu ",
4486 entry->pid, iter->cpu, iter->ts);
4487
4488 if (trace_seq_has_overflowed(s))
4489 return TRACE_TYPE_PARTIAL_LINE;
4490
4491 event = ftrace_find_event(entry->type);
4492 if (event)
4493 return event->funcs->raw(iter, 0, event);
4494
4495 trace_seq_printf(s, "%d ?\n", entry->type);
4496
4497 return trace_handle_return(s);
4498}
4499
4500static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4501{
4502 struct trace_array *tr = iter->tr;
4503 struct trace_seq *s = &iter->seq;
4504 unsigned char newline = '\n';
4505 struct trace_entry *entry;
4506 struct trace_event *event;
4507
4508 entry = iter->ent;
4509
4510 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4511 SEQ_PUT_HEX_FIELD(s, entry->pid);
4512 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4513 SEQ_PUT_HEX_FIELD(s, iter->ts);
4514 if (trace_seq_has_overflowed(s))
4515 return TRACE_TYPE_PARTIAL_LINE;
4516 }
4517
4518 event = ftrace_find_event(entry->type);
4519 if (event) {
4520 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4521 if (ret != TRACE_TYPE_HANDLED)
4522 return ret;
4523 }
4524
4525 SEQ_PUT_FIELD(s, newline);
4526
4527 return trace_handle_return(s);
4528}
4529
4530static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4531{
4532 struct trace_array *tr = iter->tr;
4533 struct trace_seq *s = &iter->seq;
4534 struct trace_entry *entry;
4535 struct trace_event *event;
4536
4537 entry = iter->ent;
4538
4539 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4540 SEQ_PUT_FIELD(s, entry->pid);
4541 SEQ_PUT_FIELD(s, iter->cpu);
4542 SEQ_PUT_FIELD(s, iter->ts);
4543 if (trace_seq_has_overflowed(s))
4544 return TRACE_TYPE_PARTIAL_LINE;
4545 }
4546
4547 event = ftrace_find_event(entry->type);
4548 return event ? event->funcs->binary(iter, 0, event) :
4549 TRACE_TYPE_HANDLED;
4550}
4551
4552int trace_empty(struct trace_iterator *iter)
4553{
4554 struct ring_buffer_iter *buf_iter;
4555 int cpu;
4556
4557 /* If we are looking at one CPU buffer, only check that one */
4558 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4559 cpu = iter->cpu_file;
4560 buf_iter = trace_buffer_iter(iter, cpu);
4561 if (buf_iter) {
4562 if (!ring_buffer_iter_empty(buf_iter))
4563 return 0;
4564 } else {
4565 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4566 return 0;
4567 }
4568 return 1;
4569 }
4570
4571 for_each_tracing_cpu(cpu) {
4572 buf_iter = trace_buffer_iter(iter, cpu);
4573 if (buf_iter) {
4574 if (!ring_buffer_iter_empty(buf_iter))
4575 return 0;
4576 } else {
4577 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4578 return 0;
4579 }
4580 }
4581
4582 return 1;
4583}
4584
4585/* Called with trace_event_read_lock() held. */
4586enum print_line_t print_trace_line(struct trace_iterator *iter)
4587{
4588 struct trace_array *tr = iter->tr;
4589 unsigned long trace_flags = tr->trace_flags;
4590 enum print_line_t ret;
4591
4592 if (iter->lost_events) {
4593 if (iter->lost_events == (unsigned long)-1)
4594 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4595 iter->cpu);
4596 else
4597 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4598 iter->cpu, iter->lost_events);
4599 if (trace_seq_has_overflowed(&iter->seq))
4600 return TRACE_TYPE_PARTIAL_LINE;
4601 }
4602
4603 if (iter->trace && iter->trace->print_line) {
4604 ret = iter->trace->print_line(iter);
4605 if (ret != TRACE_TYPE_UNHANDLED)
4606 return ret;
4607 }
4608
4609 if (iter->ent->type == TRACE_BPUTS &&
4610 trace_flags & TRACE_ITER_PRINTK &&
4611 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4612 return trace_print_bputs_msg_only(iter);
4613
4614 if (iter->ent->type == TRACE_BPRINT &&
4615 trace_flags & TRACE_ITER_PRINTK &&
4616 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4617 return trace_print_bprintk_msg_only(iter);
4618
4619 if (iter->ent->type == TRACE_PRINT &&
4620 trace_flags & TRACE_ITER_PRINTK &&
4621 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4622 return trace_print_printk_msg_only(iter);
4623
4624 if (trace_flags & TRACE_ITER_BIN)
4625 return print_bin_fmt(iter);
4626
4627 if (trace_flags & TRACE_ITER_HEX)
4628 return print_hex_fmt(iter);
4629
4630 if (trace_flags & TRACE_ITER_RAW)
4631 return print_raw_fmt(iter);
4632
4633 return print_trace_fmt(iter);
4634}
4635
4636void trace_latency_header(struct seq_file *m)
4637{
4638 struct trace_iterator *iter = m->private;
4639 struct trace_array *tr = iter->tr;
4640
4641 /* print nothing if the buffers are empty */
4642 if (trace_empty(iter))
4643 return;
4644
4645 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4646 print_trace_header(m, iter);
4647
4648 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4649 print_lat_help_header(m);
4650}
4651
4652void trace_default_header(struct seq_file *m)
4653{
4654 struct trace_iterator *iter = m->private;
4655 struct trace_array *tr = iter->tr;
4656 unsigned long trace_flags = tr->trace_flags;
4657
4658 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4659 return;
4660
4661 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4662 /* print nothing if the buffers are empty */
4663 if (trace_empty(iter))
4664 return;
4665 print_trace_header(m, iter);
4666 if (!(trace_flags & TRACE_ITER_VERBOSE))
4667 print_lat_help_header(m);
4668 } else {
4669 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4670 if (trace_flags & TRACE_ITER_IRQ_INFO)
4671 print_func_help_header_irq(iter->array_buffer,
4672 m, trace_flags);
4673 else
4674 print_func_help_header(iter->array_buffer, m,
4675 trace_flags);
4676 }
4677 }
4678}
4679
4680static void test_ftrace_alive(struct seq_file *m)
4681{
4682 if (!ftrace_is_dead())
4683 return;
4684 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4685 "# MAY BE MISSING FUNCTION EVENTS\n");
4686}
4687
4688#ifdef CONFIG_TRACER_MAX_TRACE
4689static void show_snapshot_main_help(struct seq_file *m)
4690{
4691 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4692 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4693 "# Takes a snapshot of the main buffer.\n"
4694 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4695 "# (Doesn't have to be '2' works with any number that\n"
4696 "# is not a '0' or '1')\n");
4697}
4698
4699static void show_snapshot_percpu_help(struct seq_file *m)
4700{
4701 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4702#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4703 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4704 "# Takes a snapshot of the main buffer for this cpu.\n");
4705#else
4706 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4707 "# Must use main snapshot file to allocate.\n");
4708#endif
4709 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4710 "# (Doesn't have to be '2' works with any number that\n"
4711 "# is not a '0' or '1')\n");
4712}
4713
4714static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4715{
4716 if (iter->tr->allocated_snapshot)
4717 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4718 else
4719 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4720
4721 seq_puts(m, "# Snapshot commands:\n");
4722 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4723 show_snapshot_main_help(m);
4724 else
4725 show_snapshot_percpu_help(m);
4726}
4727#else
4728/* Should never be called */
4729static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4730#endif
4731
4732static int s_show(struct seq_file *m, void *v)
4733{
4734 struct trace_iterator *iter = v;
4735 int ret;
4736
4737 if (iter->ent == NULL) {
4738 if (iter->tr) {
4739 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4740 seq_puts(m, "#\n");
4741 test_ftrace_alive(m);
4742 }
4743 if (iter->snapshot && trace_empty(iter))
4744 print_snapshot_help(m, iter);
4745 else if (iter->trace && iter->trace->print_header)
4746 iter->trace->print_header(m);
4747 else
4748 trace_default_header(m);
4749
4750 } else if (iter->leftover) {
4751 /*
4752 * If we filled the seq_file buffer earlier, we
4753 * want to just show it now.
4754 */
4755 ret = trace_print_seq(m, &iter->seq);
4756
4757 /* ret should this time be zero, but you never know */
4758 iter->leftover = ret;
4759
4760 } else {
4761 ret = print_trace_line(iter);
4762 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4763 iter->seq.full = 0;
4764 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
4765 }
4766 ret = trace_print_seq(m, &iter->seq);
4767 /*
4768 * If we overflow the seq_file buffer, then it will
4769 * ask us for this data again at start up.
4770 * Use that instead.
4771 * ret is 0 if seq_file write succeeded.
4772 * -1 otherwise.
4773 */
4774 iter->leftover = ret;
4775 }
4776
4777 return 0;
4778}
4779
4780/*
4781 * Should be used after trace_array_get(), trace_types_lock
4782 * ensures that i_cdev was already initialized.
4783 */
4784static inline int tracing_get_cpu(struct inode *inode)
4785{
4786 if (inode->i_cdev) /* See trace_create_cpu_file() */
4787 return (long)inode->i_cdev - 1;
4788 return RING_BUFFER_ALL_CPUS;
4789}
4790
4791static const struct seq_operations tracer_seq_ops = {
4792 .start = s_start,
4793 .next = s_next,
4794 .stop = s_stop,
4795 .show = s_show,
4796};
4797
4798/*
4799 * Note, as iter itself can be allocated and freed in different
4800 * ways, this function is only used to free its content, and not
4801 * the iterator itself. The only requirement to all the allocations
4802 * is that it must zero all fields (kzalloc), as freeing works with
4803 * ethier allocated content or NULL.
4804 */
4805static void free_trace_iter_content(struct trace_iterator *iter)
4806{
4807 /* The fmt is either NULL, allocated or points to static_fmt_buf */
4808 if (iter->fmt != static_fmt_buf)
4809 kfree(iter->fmt);
4810
4811 kfree(iter->temp);
4812 kfree(iter->buffer_iter);
4813 mutex_destroy(&iter->mutex);
4814 free_cpumask_var(iter->started);
4815}
4816
4817static struct trace_iterator *
4818__tracing_open(struct inode *inode, struct file *file, bool snapshot)
4819{
4820 struct trace_array *tr = inode->i_private;
4821 struct trace_iterator *iter;
4822 int cpu;
4823
4824 if (tracing_disabled)
4825 return ERR_PTR(-ENODEV);
4826
4827 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4828 if (!iter)
4829 return ERR_PTR(-ENOMEM);
4830
4831 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4832 GFP_KERNEL);
4833 if (!iter->buffer_iter)
4834 goto release;
4835
4836 /*
4837 * trace_find_next_entry() may need to save off iter->ent.
4838 * It will place it into the iter->temp buffer. As most
4839 * events are less than 128, allocate a buffer of that size.
4840 * If one is greater, then trace_find_next_entry() will
4841 * allocate a new buffer to adjust for the bigger iter->ent.
4842 * It's not critical if it fails to get allocated here.
4843 */
4844 iter->temp = kmalloc(128, GFP_KERNEL);
4845 if (iter->temp)
4846 iter->temp_size = 128;
4847
4848 /*
4849 * trace_event_printf() may need to modify given format
4850 * string to replace %p with %px so that it shows real address
4851 * instead of hash value. However, that is only for the event
4852 * tracing, other tracer may not need. Defer the allocation
4853 * until it is needed.
4854 */
4855 iter->fmt = NULL;
4856 iter->fmt_size = 0;
4857
4858 mutex_lock(&trace_types_lock);
4859 iter->trace = tr->current_trace;
4860
4861 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4862 goto fail;
4863
4864 iter->tr = tr;
4865
4866#ifdef CONFIG_TRACER_MAX_TRACE
4867 /* Currently only the top directory has a snapshot */
4868 if (tr->current_trace->print_max || snapshot)
4869 iter->array_buffer = &tr->max_buffer;
4870 else
4871#endif
4872 iter->array_buffer = &tr->array_buffer;
4873 iter->snapshot = snapshot;
4874 iter->pos = -1;
4875 iter->cpu_file = tracing_get_cpu(inode);
4876 mutex_init(&iter->mutex);
4877
4878 /* Notify the tracer early; before we stop tracing. */
4879 if (iter->trace->open)
4880 iter->trace->open(iter);
4881
4882 /* Annotate start of buffers if we had overruns */
4883 if (ring_buffer_overruns(iter->array_buffer->buffer))
4884 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4885
4886 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4887 if (trace_clocks[tr->clock_id].in_ns)
4888 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4889
4890 /*
4891 * If pause-on-trace is enabled, then stop the trace while
4892 * dumping, unless this is the "snapshot" file
4893 */
4894 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4895 tracing_stop_tr(tr);
4896
4897 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4898 for_each_tracing_cpu(cpu) {
4899 iter->buffer_iter[cpu] =
4900 ring_buffer_read_prepare(iter->array_buffer->buffer,
4901 cpu, GFP_KERNEL);
4902 }
4903 ring_buffer_read_prepare_sync();
4904 for_each_tracing_cpu(cpu) {
4905 ring_buffer_read_start(iter->buffer_iter[cpu]);
4906 tracing_iter_reset(iter, cpu);
4907 }
4908 } else {
4909 cpu = iter->cpu_file;
4910 iter->buffer_iter[cpu] =
4911 ring_buffer_read_prepare(iter->array_buffer->buffer,
4912 cpu, GFP_KERNEL);
4913 ring_buffer_read_prepare_sync();
4914 ring_buffer_read_start(iter->buffer_iter[cpu]);
4915 tracing_iter_reset(iter, cpu);
4916 }
4917
4918 mutex_unlock(&trace_types_lock);
4919
4920 return iter;
4921
4922 fail:
4923 mutex_unlock(&trace_types_lock);
4924 free_trace_iter_content(iter);
4925release:
4926 seq_release_private(inode, file);
4927 return ERR_PTR(-ENOMEM);
4928}
4929
4930int tracing_open_generic(struct inode *inode, struct file *filp)
4931{
4932 int ret;
4933
4934 ret = tracing_check_open_get_tr(NULL);
4935 if (ret)
4936 return ret;
4937
4938 filp->private_data = inode->i_private;
4939 return 0;
4940}
4941
4942bool tracing_is_disabled(void)
4943{
4944 return (tracing_disabled) ? true: false;
4945}
4946
4947/*
4948 * Open and update trace_array ref count.
4949 * Must have the current trace_array passed to it.
4950 */
4951int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4952{
4953 struct trace_array *tr = inode->i_private;
4954 int ret;
4955
4956 ret = tracing_check_open_get_tr(tr);
4957 if (ret)
4958 return ret;
4959
4960 filp->private_data = inode->i_private;
4961
4962 return 0;
4963}
4964
4965/*
4966 * The private pointer of the inode is the trace_event_file.
4967 * Update the tr ref count associated to it.
4968 */
4969int tracing_open_file_tr(struct inode *inode, struct file *filp)
4970{
4971 struct trace_event_file *file = inode->i_private;
4972 int ret;
4973
4974 ret = tracing_check_open_get_tr(file->tr);
4975 if (ret)
4976 return ret;
4977
4978 mutex_lock(&event_mutex);
4979
4980 /* Fail if the file is marked for removal */
4981 if (file->flags & EVENT_FILE_FL_FREED) {
4982 trace_array_put(file->tr);
4983 ret = -ENODEV;
4984 } else {
4985 event_file_get(file);
4986 }
4987
4988 mutex_unlock(&event_mutex);
4989 if (ret)
4990 return ret;
4991
4992 filp->private_data = inode->i_private;
4993
4994 return 0;
4995}
4996
4997int tracing_release_file_tr(struct inode *inode, struct file *filp)
4998{
4999 struct trace_event_file *file = inode->i_private;
5000
5001 trace_array_put(file->tr);
5002 event_file_put(file);
5003
5004 return 0;
5005}
5006
5007int tracing_single_release_file_tr(struct inode *inode, struct file *filp)
5008{
5009 tracing_release_file_tr(inode, filp);
5010 return single_release(inode, filp);
5011}
5012
5013static int tracing_mark_open(struct inode *inode, struct file *filp)
5014{
5015 stream_open(inode, filp);
5016 return tracing_open_generic_tr(inode, filp);
5017}
5018
5019static int tracing_release(struct inode *inode, struct file *file)
5020{
5021 struct trace_array *tr = inode->i_private;
5022 struct seq_file *m = file->private_data;
5023 struct trace_iterator *iter;
5024 int cpu;
5025
5026 if (!(file->f_mode & FMODE_READ)) {
5027 trace_array_put(tr);
5028 return 0;
5029 }
5030
5031 /* Writes do not use seq_file */
5032 iter = m->private;
5033 mutex_lock(&trace_types_lock);
5034
5035 for_each_tracing_cpu(cpu) {
5036 if (iter->buffer_iter[cpu])
5037 ring_buffer_read_finish(iter->buffer_iter[cpu]);
5038 }
5039
5040 if (iter->trace && iter->trace->close)
5041 iter->trace->close(iter);
5042
5043 if (!iter->snapshot && tr->stop_count)
5044 /* reenable tracing if it was previously enabled */
5045 tracing_start_tr(tr);
5046
5047 __trace_array_put(tr);
5048
5049 mutex_unlock(&trace_types_lock);
5050
5051 free_trace_iter_content(iter);
5052 seq_release_private(inode, file);
5053
5054 return 0;
5055}
5056
5057int tracing_release_generic_tr(struct inode *inode, struct file *file)
5058{
5059 struct trace_array *tr = inode->i_private;
5060
5061 trace_array_put(tr);
5062 return 0;
5063}
5064
5065static int tracing_single_release_tr(struct inode *inode, struct file *file)
5066{
5067 struct trace_array *tr = inode->i_private;
5068
5069 trace_array_put(tr);
5070
5071 return single_release(inode, file);
5072}
5073
5074static int tracing_open(struct inode *inode, struct file *file)
5075{
5076 struct trace_array *tr = inode->i_private;
5077 struct trace_iterator *iter;
5078 int ret;
5079
5080 ret = tracing_check_open_get_tr(tr);
5081 if (ret)
5082 return ret;
5083
5084 /* If this file was open for write, then erase contents */
5085 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
5086 int cpu = tracing_get_cpu(inode);
5087 struct array_buffer *trace_buf = &tr->array_buffer;
5088
5089#ifdef CONFIG_TRACER_MAX_TRACE
5090 if (tr->current_trace->print_max)
5091 trace_buf = &tr->max_buffer;
5092#endif
5093
5094 if (cpu == RING_BUFFER_ALL_CPUS)
5095 tracing_reset_online_cpus(trace_buf);
5096 else
5097 tracing_reset_cpu(trace_buf, cpu);
5098 }
5099
5100 if (file->f_mode & FMODE_READ) {
5101 iter = __tracing_open(inode, file, false);
5102 if (IS_ERR(iter))
5103 ret = PTR_ERR(iter);
5104 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
5105 iter->iter_flags |= TRACE_FILE_LAT_FMT;
5106 }
5107
5108 if (ret < 0)
5109 trace_array_put(tr);
5110
5111 return ret;
5112}
5113
5114/*
5115 * Some tracers are not suitable for instance buffers.
5116 * A tracer is always available for the global array (toplevel)
5117 * or if it explicitly states that it is.
5118 */
5119static bool
5120trace_ok_for_array(struct tracer *t, struct trace_array *tr)
5121{
5122 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5123}
5124
5125/* Find the next tracer that this trace array may use */
5126static struct tracer *
5127get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5128{
5129 while (t && !trace_ok_for_array(t, tr))
5130 t = t->next;
5131
5132 return t;
5133}
5134
5135static void *
5136t_next(struct seq_file *m, void *v, loff_t *pos)
5137{
5138 struct trace_array *tr = m->private;
5139 struct tracer *t = v;
5140
5141 (*pos)++;
5142
5143 if (t)
5144 t = get_tracer_for_array(tr, t->next);
5145
5146 return t;
5147}
5148
5149static void *t_start(struct seq_file *m, loff_t *pos)
5150{
5151 struct trace_array *tr = m->private;
5152 struct tracer *t;
5153 loff_t l = 0;
5154
5155 mutex_lock(&trace_types_lock);
5156
5157 t = get_tracer_for_array(tr, trace_types);
5158 for (; t && l < *pos; t = t_next(m, t, &l))
5159 ;
5160
5161 return t;
5162}
5163
5164static void t_stop(struct seq_file *m, void *p)
5165{
5166 mutex_unlock(&trace_types_lock);
5167}
5168
5169static int t_show(struct seq_file *m, void *v)
5170{
5171 struct tracer *t = v;
5172
5173 if (!t)
5174 return 0;
5175
5176 seq_puts(m, t->name);
5177 if (t->next)
5178 seq_putc(m, ' ');
5179 else
5180 seq_putc(m, '\n');
5181
5182 return 0;
5183}
5184
5185static const struct seq_operations show_traces_seq_ops = {
5186 .start = t_start,
5187 .next = t_next,
5188 .stop = t_stop,
5189 .show = t_show,
5190};
5191
5192static int show_traces_open(struct inode *inode, struct file *file)
5193{
5194 struct trace_array *tr = inode->i_private;
5195 struct seq_file *m;
5196 int ret;
5197
5198 ret = tracing_check_open_get_tr(tr);
5199 if (ret)
5200 return ret;
5201
5202 ret = seq_open(file, &show_traces_seq_ops);
5203 if (ret) {
5204 trace_array_put(tr);
5205 return ret;
5206 }
5207
5208 m = file->private_data;
5209 m->private = tr;
5210
5211 return 0;
5212}
5213
5214static int show_traces_release(struct inode *inode, struct file *file)
5215{
5216 struct trace_array *tr = inode->i_private;
5217
5218 trace_array_put(tr);
5219 return seq_release(inode, file);
5220}
5221
5222static ssize_t
5223tracing_write_stub(struct file *filp, const char __user *ubuf,
5224 size_t count, loff_t *ppos)
5225{
5226 return count;
5227}
5228
5229loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5230{
5231 int ret;
5232
5233 if (file->f_mode & FMODE_READ)
5234 ret = seq_lseek(file, offset, whence);
5235 else
5236 file->f_pos = ret = 0;
5237
5238 return ret;
5239}
5240
5241static const struct file_operations tracing_fops = {
5242 .open = tracing_open,
5243 .read = seq_read,
5244 .read_iter = seq_read_iter,
5245 .splice_read = copy_splice_read,
5246 .write = tracing_write_stub,
5247 .llseek = tracing_lseek,
5248 .release = tracing_release,
5249};
5250
5251static const struct file_operations show_traces_fops = {
5252 .open = show_traces_open,
5253 .read = seq_read,
5254 .llseek = seq_lseek,
5255 .release = show_traces_release,
5256};
5257
5258static ssize_t
5259tracing_cpumask_read(struct file *filp, char __user *ubuf,
5260 size_t count, loff_t *ppos)
5261{
5262 struct trace_array *tr = file_inode(filp)->i_private;
5263 char *mask_str;
5264 int len;
5265
5266 len = snprintf(NULL, 0, "%*pb\n",
5267 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5268 mask_str = kmalloc(len, GFP_KERNEL);
5269 if (!mask_str)
5270 return -ENOMEM;
5271
5272 len = snprintf(mask_str, len, "%*pb\n",
5273 cpumask_pr_args(tr->tracing_cpumask));
5274 if (len >= count) {
5275 count = -EINVAL;
5276 goto out_err;
5277 }
5278 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5279
5280out_err:
5281 kfree(mask_str);
5282
5283 return count;
5284}
5285
5286int tracing_set_cpumask(struct trace_array *tr,
5287 cpumask_var_t tracing_cpumask_new)
5288{
5289 int cpu;
5290
5291 if (!tr)
5292 return -EINVAL;
5293
5294 local_irq_disable();
5295 arch_spin_lock(&tr->max_lock);
5296 for_each_tracing_cpu(cpu) {
5297 /*
5298 * Increase/decrease the disabled counter if we are
5299 * about to flip a bit in the cpumask:
5300 */
5301 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5302 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5303 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5304 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5305#ifdef CONFIG_TRACER_MAX_TRACE
5306 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5307#endif
5308 }
5309 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5310 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5311 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5312 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5313#ifdef CONFIG_TRACER_MAX_TRACE
5314 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5315#endif
5316 }
5317 }
5318 arch_spin_unlock(&tr->max_lock);
5319 local_irq_enable();
5320
5321 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5322
5323 return 0;
5324}
5325
5326static ssize_t
5327tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5328 size_t count, loff_t *ppos)
5329{
5330 struct trace_array *tr = file_inode(filp)->i_private;
5331 cpumask_var_t tracing_cpumask_new;
5332 int err;
5333
5334 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5335 return -ENOMEM;
5336
5337 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5338 if (err)
5339 goto err_free;
5340
5341 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5342 if (err)
5343 goto err_free;
5344
5345 free_cpumask_var(tracing_cpumask_new);
5346
5347 return count;
5348
5349err_free:
5350 free_cpumask_var(tracing_cpumask_new);
5351
5352 return err;
5353}
5354
5355static const struct file_operations tracing_cpumask_fops = {
5356 .open = tracing_open_generic_tr,
5357 .read = tracing_cpumask_read,
5358 .write = tracing_cpumask_write,
5359 .release = tracing_release_generic_tr,
5360 .llseek = generic_file_llseek,
5361};
5362
5363static int tracing_trace_options_show(struct seq_file *m, void *v)
5364{
5365 struct tracer_opt *trace_opts;
5366 struct trace_array *tr = m->private;
5367 u32 tracer_flags;
5368 int i;
5369
5370 mutex_lock(&trace_types_lock);
5371 tracer_flags = tr->current_trace->flags->val;
5372 trace_opts = tr->current_trace->flags->opts;
5373
5374 for (i = 0; trace_options[i]; i++) {
5375 if (tr->trace_flags & (1 << i))
5376 seq_printf(m, "%s\n", trace_options[i]);
5377 else
5378 seq_printf(m, "no%s\n", trace_options[i]);
5379 }
5380
5381 for (i = 0; trace_opts[i].name; i++) {
5382 if (tracer_flags & trace_opts[i].bit)
5383 seq_printf(m, "%s\n", trace_opts[i].name);
5384 else
5385 seq_printf(m, "no%s\n", trace_opts[i].name);
5386 }
5387 mutex_unlock(&trace_types_lock);
5388
5389 return 0;
5390}
5391
5392static int __set_tracer_option(struct trace_array *tr,
5393 struct tracer_flags *tracer_flags,
5394 struct tracer_opt *opts, int neg)
5395{
5396 struct tracer *trace = tracer_flags->trace;
5397 int ret;
5398
5399 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5400 if (ret)
5401 return ret;
5402
5403 if (neg)
5404 tracer_flags->val &= ~opts->bit;
5405 else
5406 tracer_flags->val |= opts->bit;
5407 return 0;
5408}
5409
5410/* Try to assign a tracer specific option */
5411static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5412{
5413 struct tracer *trace = tr->current_trace;
5414 struct tracer_flags *tracer_flags = trace->flags;
5415 struct tracer_opt *opts = NULL;
5416 int i;
5417
5418 for (i = 0; tracer_flags->opts[i].name; i++) {
5419 opts = &tracer_flags->opts[i];
5420
5421 if (strcmp(cmp, opts->name) == 0)
5422 return __set_tracer_option(tr, trace->flags, opts, neg);
5423 }
5424
5425 return -EINVAL;
5426}
5427
5428/* Some tracers require overwrite to stay enabled */
5429int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5430{
5431 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5432 return -1;
5433
5434 return 0;
5435}
5436
5437int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5438{
5439 int *map;
5440
5441 if ((mask == TRACE_ITER_RECORD_TGID) ||
5442 (mask == TRACE_ITER_RECORD_CMD))
5443 lockdep_assert_held(&event_mutex);
5444
5445 /* do nothing if flag is already set */
5446 if (!!(tr->trace_flags & mask) == !!enabled)
5447 return 0;
5448
5449 /* Give the tracer a chance to approve the change */
5450 if (tr->current_trace->flag_changed)
5451 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5452 return -EINVAL;
5453
5454 if (enabled)
5455 tr->trace_flags |= mask;
5456 else
5457 tr->trace_flags &= ~mask;
5458
5459 if (mask == TRACE_ITER_RECORD_CMD)
5460 trace_event_enable_cmd_record(enabled);
5461
5462 if (mask == TRACE_ITER_RECORD_TGID) {
5463 if (!tgid_map) {
5464 tgid_map_max = pid_max;
5465 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5466 GFP_KERNEL);
5467
5468 /*
5469 * Pairs with smp_load_acquire() in
5470 * trace_find_tgid_ptr() to ensure that if it observes
5471 * the tgid_map we just allocated then it also observes
5472 * the corresponding tgid_map_max value.
5473 */
5474 smp_store_release(&tgid_map, map);
5475 }
5476 if (!tgid_map) {
5477 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5478 return -ENOMEM;
5479 }
5480
5481 trace_event_enable_tgid_record(enabled);
5482 }
5483
5484 if (mask == TRACE_ITER_EVENT_FORK)
5485 trace_event_follow_fork(tr, enabled);
5486
5487 if (mask == TRACE_ITER_FUNC_FORK)
5488 ftrace_pid_follow_fork(tr, enabled);
5489
5490 if (mask == TRACE_ITER_OVERWRITE) {
5491 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5492#ifdef CONFIG_TRACER_MAX_TRACE
5493 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5494#endif
5495 }
5496
5497 if (mask == TRACE_ITER_PRINTK) {
5498 trace_printk_start_stop_comm(enabled);
5499 trace_printk_control(enabled);
5500 }
5501
5502 return 0;
5503}
5504
5505int trace_set_options(struct trace_array *tr, char *option)
5506{
5507 char *cmp;
5508 int neg = 0;
5509 int ret;
5510 size_t orig_len = strlen(option);
5511 int len;
5512
5513 cmp = strstrip(option);
5514
5515 len = str_has_prefix(cmp, "no");
5516 if (len)
5517 neg = 1;
5518
5519 cmp += len;
5520
5521 mutex_lock(&event_mutex);
5522 mutex_lock(&trace_types_lock);
5523
5524 ret = match_string(trace_options, -1, cmp);
5525 /* If no option could be set, test the specific tracer options */
5526 if (ret < 0)
5527 ret = set_tracer_option(tr, cmp, neg);
5528 else
5529 ret = set_tracer_flag(tr, 1 << ret, !neg);
5530
5531 mutex_unlock(&trace_types_lock);
5532 mutex_unlock(&event_mutex);
5533
5534 /*
5535 * If the first trailing whitespace is replaced with '\0' by strstrip,
5536 * turn it back into a space.
5537 */
5538 if (orig_len > strlen(option))
5539 option[strlen(option)] = ' ';
5540
5541 return ret;
5542}
5543
5544static void __init apply_trace_boot_options(void)
5545{
5546 char *buf = trace_boot_options_buf;
5547 char *option;
5548
5549 while (true) {
5550 option = strsep(&buf, ",");
5551
5552 if (!option)
5553 break;
5554
5555 if (*option)
5556 trace_set_options(&global_trace, option);
5557
5558 /* Put back the comma to allow this to be called again */
5559 if (buf)
5560 *(buf - 1) = ',';
5561 }
5562}
5563
5564static ssize_t
5565tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5566 size_t cnt, loff_t *ppos)
5567{
5568 struct seq_file *m = filp->private_data;
5569 struct trace_array *tr = m->private;
5570 char buf[64];
5571 int ret;
5572
5573 if (cnt >= sizeof(buf))
5574 return -EINVAL;
5575
5576 if (copy_from_user(buf, ubuf, cnt))
5577 return -EFAULT;
5578
5579 buf[cnt] = 0;
5580
5581 ret = trace_set_options(tr, buf);
5582 if (ret < 0)
5583 return ret;
5584
5585 *ppos += cnt;
5586
5587 return cnt;
5588}
5589
5590static int tracing_trace_options_open(struct inode *inode, struct file *file)
5591{
5592 struct trace_array *tr = inode->i_private;
5593 int ret;
5594
5595 ret = tracing_check_open_get_tr(tr);
5596 if (ret)
5597 return ret;
5598
5599 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5600 if (ret < 0)
5601 trace_array_put(tr);
5602
5603 return ret;
5604}
5605
5606static const struct file_operations tracing_iter_fops = {
5607 .open = tracing_trace_options_open,
5608 .read = seq_read,
5609 .llseek = seq_lseek,
5610 .release = tracing_single_release_tr,
5611 .write = tracing_trace_options_write,
5612};
5613
5614static const char readme_msg[] =
5615 "tracing mini-HOWTO:\n\n"
5616 "# echo 0 > tracing_on : quick way to disable tracing\n"
5617 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5618 " Important files:\n"
5619 " trace\t\t\t- The static contents of the buffer\n"
5620 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5621 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5622 " current_tracer\t- function and latency tracers\n"
5623 " available_tracers\t- list of configured tracers for current_tracer\n"
5624 " error_log\t- error log for failed commands (that support it)\n"
5625 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5626 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5627 " trace_clock\t\t- change the clock used to order events\n"
5628 " local: Per cpu clock but may not be synced across CPUs\n"
5629 " global: Synced across CPUs but slows tracing down.\n"
5630 " counter: Not a clock, but just an increment\n"
5631 " uptime: Jiffy counter from time of boot\n"
5632 " perf: Same clock that perf events use\n"
5633#ifdef CONFIG_X86_64
5634 " x86-tsc: TSC cycle counter\n"
5635#endif
5636 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5637 " delta: Delta difference against a buffer-wide timestamp\n"
5638 " absolute: Absolute (standalone) timestamp\n"
5639 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5640 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5641 " tracing_cpumask\t- Limit which CPUs to trace\n"
5642 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5643 "\t\t\t Remove sub-buffer with rmdir\n"
5644 " trace_options\t\t- Set format or modify how tracing happens\n"
5645 "\t\t\t Disable an option by prefixing 'no' to the\n"
5646 "\t\t\t option name\n"
5647 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5648#ifdef CONFIG_DYNAMIC_FTRACE
5649 "\n available_filter_functions - list of functions that can be filtered on\n"
5650 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5651 "\t\t\t functions\n"
5652 "\t accepts: func_full_name or glob-matching-pattern\n"
5653 "\t modules: Can select a group via module\n"
5654 "\t Format: :mod:<module-name>\n"
5655 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5656 "\t triggers: a command to perform when function is hit\n"
5657 "\t Format: <function>:<trigger>[:count]\n"
5658 "\t trigger: traceon, traceoff\n"
5659 "\t\t enable_event:<system>:<event>\n"
5660 "\t\t disable_event:<system>:<event>\n"
5661#ifdef CONFIG_STACKTRACE
5662 "\t\t stacktrace\n"
5663#endif
5664#ifdef CONFIG_TRACER_SNAPSHOT
5665 "\t\t snapshot\n"
5666#endif
5667 "\t\t dump\n"
5668 "\t\t cpudump\n"
5669 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5670 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5671 "\t The first one will disable tracing every time do_fault is hit\n"
5672 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5673 "\t The first time do trap is hit and it disables tracing, the\n"
5674 "\t counter will decrement to 2. If tracing is already disabled,\n"
5675 "\t the counter will not decrement. It only decrements when the\n"
5676 "\t trigger did work\n"
5677 "\t To remove trigger without count:\n"
5678 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5679 "\t To remove trigger with a count:\n"
5680 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5681 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5682 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5683 "\t modules: Can select a group via module command :mod:\n"
5684 "\t Does not accept triggers\n"
5685#endif /* CONFIG_DYNAMIC_FTRACE */
5686#ifdef CONFIG_FUNCTION_TRACER
5687 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5688 "\t\t (function)\n"
5689 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5690 "\t\t (function)\n"
5691#endif
5692#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5693 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5694 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5695 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5696#endif
5697#ifdef CONFIG_TRACER_SNAPSHOT
5698 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5699 "\t\t\t snapshot buffer. Read the contents for more\n"
5700 "\t\t\t information\n"
5701#endif
5702#ifdef CONFIG_STACK_TRACER
5703 " stack_trace\t\t- Shows the max stack trace when active\n"
5704 " stack_max_size\t- Shows current max stack size that was traced\n"
5705 "\t\t\t Write into this file to reset the max size (trigger a\n"
5706 "\t\t\t new trace)\n"
5707#ifdef CONFIG_DYNAMIC_FTRACE
5708 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5709 "\t\t\t traces\n"
5710#endif
5711#endif /* CONFIG_STACK_TRACER */
5712#ifdef CONFIG_DYNAMIC_EVENTS
5713 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5714 "\t\t\t Write into this file to define/undefine new trace events.\n"
5715#endif
5716#ifdef CONFIG_KPROBE_EVENTS
5717 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5718 "\t\t\t Write into this file to define/undefine new trace events.\n"
5719#endif
5720#ifdef CONFIG_UPROBE_EVENTS
5721 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5722 "\t\t\t Write into this file to define/undefine new trace events.\n"
5723#endif
5724#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5725 defined(CONFIG_FPROBE_EVENTS)
5726 "\t accepts: event-definitions (one definition per line)\n"
5727#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5728 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5729 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5730#endif
5731#ifdef CONFIG_FPROBE_EVENTS
5732 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5733 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5734#endif
5735#ifdef CONFIG_HIST_TRIGGERS
5736 "\t s:[synthetic/]<event> <field> [<field>]\n"
5737#endif
5738 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5739 "\t -:[<group>/][<event>]\n"
5740#ifdef CONFIG_KPROBE_EVENTS
5741 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5742 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5743#endif
5744#ifdef CONFIG_UPROBE_EVENTS
5745 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5746#endif
5747 "\t args: <name>=fetcharg[:type]\n"
5748 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5749#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5750#ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5751 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5752 "\t <argname>[->field[->field|.field...]],\n"
5753#else
5754 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5755#endif
5756#else
5757 "\t $stack<index>, $stack, $retval, $comm,\n"
5758#endif
5759 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5760 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5761 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5762 "\t symstr, <type>\\[<array-size>\\]\n"
5763#ifdef CONFIG_HIST_TRIGGERS
5764 "\t field: <stype> <name>;\n"
5765 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5766 "\t [unsigned] char/int/long\n"
5767#endif
5768 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5769 "\t of the <attached-group>/<attached-event>.\n"
5770#endif
5771 " events/\t\t- Directory containing all trace event subsystems:\n"
5772 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5773 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5774 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5775 "\t\t\t events\n"
5776 " filter\t\t- If set, only events passing filter are traced\n"
5777 " events/<system>/<event>/\t- Directory containing control files for\n"
5778 "\t\t\t <event>:\n"
5779 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5780 " filter\t\t- If set, only events passing filter are traced\n"
5781 " trigger\t\t- If set, a command to perform when event is hit\n"
5782 "\t Format: <trigger>[:count][if <filter>]\n"
5783 "\t trigger: traceon, traceoff\n"
5784 "\t enable_event:<system>:<event>\n"
5785 "\t disable_event:<system>:<event>\n"
5786#ifdef CONFIG_HIST_TRIGGERS
5787 "\t enable_hist:<system>:<event>\n"
5788 "\t disable_hist:<system>:<event>\n"
5789#endif
5790#ifdef CONFIG_STACKTRACE
5791 "\t\t stacktrace\n"
5792#endif
5793#ifdef CONFIG_TRACER_SNAPSHOT
5794 "\t\t snapshot\n"
5795#endif
5796#ifdef CONFIG_HIST_TRIGGERS
5797 "\t\t hist (see below)\n"
5798#endif
5799 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5800 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5801 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5802 "\t events/block/block_unplug/trigger\n"
5803 "\t The first disables tracing every time block_unplug is hit.\n"
5804 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5805 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5806 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5807 "\t Like function triggers, the counter is only decremented if it\n"
5808 "\t enabled or disabled tracing.\n"
5809 "\t To remove a trigger without a count:\n"
5810 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5811 "\t To remove a trigger with a count:\n"
5812 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5813 "\t Filters can be ignored when removing a trigger.\n"
5814#ifdef CONFIG_HIST_TRIGGERS
5815 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5816 "\t Format: hist:keys=<field1[,field2,...]>\n"
5817 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5818 "\t [:values=<field1[,field2,...]>]\n"
5819 "\t [:sort=<field1[,field2,...]>]\n"
5820 "\t [:size=#entries]\n"
5821 "\t [:pause][:continue][:clear]\n"
5822 "\t [:name=histname1]\n"
5823 "\t [:nohitcount]\n"
5824 "\t [:<handler>.<action>]\n"
5825 "\t [if <filter>]\n\n"
5826 "\t Note, special fields can be used as well:\n"
5827 "\t common_timestamp - to record current timestamp\n"
5828 "\t common_cpu - to record the CPU the event happened on\n"
5829 "\n"
5830 "\t A hist trigger variable can be:\n"
5831 "\t - a reference to a field e.g. x=current_timestamp,\n"
5832 "\t - a reference to another variable e.g. y=$x,\n"
5833 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5834 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5835 "\n"
5836 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5837 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5838 "\t variable reference, field or numeric literal.\n"
5839 "\n"
5840 "\t When a matching event is hit, an entry is added to a hash\n"
5841 "\t table using the key(s) and value(s) named, and the value of a\n"
5842 "\t sum called 'hitcount' is incremented. Keys and values\n"
5843 "\t correspond to fields in the event's format description. Keys\n"
5844 "\t can be any field, or the special string 'common_stacktrace'.\n"
5845 "\t Compound keys consisting of up to two fields can be specified\n"
5846 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5847 "\t fields. Sort keys consisting of up to two fields can be\n"
5848 "\t specified using the 'sort' keyword. The sort direction can\n"
5849 "\t be modified by appending '.descending' or '.ascending' to a\n"
5850 "\t sort field. The 'size' parameter can be used to specify more\n"
5851 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5852 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5853 "\t its histogram data will be shared with other triggers of the\n"
5854 "\t same name, and trigger hits will update this common data.\n\n"
5855 "\t Reading the 'hist' file for the event will dump the hash\n"
5856 "\t table in its entirety to stdout. If there are multiple hist\n"
5857 "\t triggers attached to an event, there will be a table for each\n"
5858 "\t trigger in the output. The table displayed for a named\n"
5859 "\t trigger will be the same as any other instance having the\n"
5860 "\t same name. The default format used to display a given field\n"
5861 "\t can be modified by appending any of the following modifiers\n"
5862 "\t to the field name, as applicable:\n\n"
5863 "\t .hex display a number as a hex value\n"
5864 "\t .sym display an address as a symbol\n"
5865 "\t .sym-offset display an address as a symbol and offset\n"
5866 "\t .execname display a common_pid as a program name\n"
5867 "\t .syscall display a syscall id as a syscall name\n"
5868 "\t .log2 display log2 value rather than raw number\n"
5869 "\t .buckets=size display values in groups of size rather than raw number\n"
5870 "\t .usecs display a common_timestamp in microseconds\n"
5871 "\t .percent display a number of percentage value\n"
5872 "\t .graph display a bar-graph of a value\n\n"
5873 "\t The 'pause' parameter can be used to pause an existing hist\n"
5874 "\t trigger or to start a hist trigger but not log any events\n"
5875 "\t until told to do so. 'continue' can be used to start or\n"
5876 "\t restart a paused hist trigger.\n\n"
5877 "\t The 'clear' parameter will clear the contents of a running\n"
5878 "\t hist trigger and leave its current paused/active state\n"
5879 "\t unchanged.\n\n"
5880 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5881 "\t raw hitcount in the histogram.\n\n"
5882 "\t The enable_hist and disable_hist triggers can be used to\n"
5883 "\t have one event conditionally start and stop another event's\n"
5884 "\t already-attached hist trigger. The syntax is analogous to\n"
5885 "\t the enable_event and disable_event triggers.\n\n"
5886 "\t Hist trigger handlers and actions are executed whenever a\n"
5887 "\t a histogram entry is added or updated. They take the form:\n\n"
5888 "\t <handler>.<action>\n\n"
5889 "\t The available handlers are:\n\n"
5890 "\t onmatch(matching.event) - invoke on addition or update\n"
5891 "\t onmax(var) - invoke if var exceeds current max\n"
5892 "\t onchange(var) - invoke action if var changes\n\n"
5893 "\t The available actions are:\n\n"
5894 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5895 "\t save(field,...) - save current event fields\n"
5896#ifdef CONFIG_TRACER_SNAPSHOT
5897 "\t snapshot() - snapshot the trace buffer\n\n"
5898#endif
5899#ifdef CONFIG_SYNTH_EVENTS
5900 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5901 "\t Write into this file to define/undefine new synthetic events.\n"
5902 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5903#endif
5904#endif
5905;
5906
5907static ssize_t
5908tracing_readme_read(struct file *filp, char __user *ubuf,
5909 size_t cnt, loff_t *ppos)
5910{
5911 return simple_read_from_buffer(ubuf, cnt, ppos,
5912 readme_msg, strlen(readme_msg));
5913}
5914
5915static const struct file_operations tracing_readme_fops = {
5916 .open = tracing_open_generic,
5917 .read = tracing_readme_read,
5918 .llseek = generic_file_llseek,
5919};
5920
5921static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5922{
5923 int pid = ++(*pos);
5924
5925 return trace_find_tgid_ptr(pid);
5926}
5927
5928static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5929{
5930 int pid = *pos;
5931
5932 return trace_find_tgid_ptr(pid);
5933}
5934
5935static void saved_tgids_stop(struct seq_file *m, void *v)
5936{
5937}
5938
5939static int saved_tgids_show(struct seq_file *m, void *v)
5940{
5941 int *entry = (int *)v;
5942 int pid = entry - tgid_map;
5943 int tgid = *entry;
5944
5945 if (tgid == 0)
5946 return SEQ_SKIP;
5947
5948 seq_printf(m, "%d %d\n", pid, tgid);
5949 return 0;
5950}
5951
5952static const struct seq_operations tracing_saved_tgids_seq_ops = {
5953 .start = saved_tgids_start,
5954 .stop = saved_tgids_stop,
5955 .next = saved_tgids_next,
5956 .show = saved_tgids_show,
5957};
5958
5959static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5960{
5961 int ret;
5962
5963 ret = tracing_check_open_get_tr(NULL);
5964 if (ret)
5965 return ret;
5966
5967 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5968}
5969
5970
5971static const struct file_operations tracing_saved_tgids_fops = {
5972 .open = tracing_saved_tgids_open,
5973 .read = seq_read,
5974 .llseek = seq_lseek,
5975 .release = seq_release,
5976};
5977
5978static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5979{
5980 unsigned int *ptr = v;
5981
5982 if (*pos || m->count)
5983 ptr++;
5984
5985 (*pos)++;
5986
5987 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5988 ptr++) {
5989 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5990 continue;
5991
5992 return ptr;
5993 }
5994
5995 return NULL;
5996}
5997
5998static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5999{
6000 void *v;
6001 loff_t l = 0;
6002
6003 preempt_disable();
6004 arch_spin_lock(&trace_cmdline_lock);
6005
6006 v = &savedcmd->map_cmdline_to_pid[0];
6007 while (l <= *pos) {
6008 v = saved_cmdlines_next(m, v, &l);
6009 if (!v)
6010 return NULL;
6011 }
6012
6013 return v;
6014}
6015
6016static void saved_cmdlines_stop(struct seq_file *m, void *v)
6017{
6018 arch_spin_unlock(&trace_cmdline_lock);
6019 preempt_enable();
6020}
6021
6022static int saved_cmdlines_show(struct seq_file *m, void *v)
6023{
6024 char buf[TASK_COMM_LEN];
6025 unsigned int *pid = v;
6026
6027 __trace_find_cmdline(*pid, buf);
6028 seq_printf(m, "%d %s\n", *pid, buf);
6029 return 0;
6030}
6031
6032static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
6033 .start = saved_cmdlines_start,
6034 .next = saved_cmdlines_next,
6035 .stop = saved_cmdlines_stop,
6036 .show = saved_cmdlines_show,
6037};
6038
6039static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
6040{
6041 int ret;
6042
6043 ret = tracing_check_open_get_tr(NULL);
6044 if (ret)
6045 return ret;
6046
6047 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
6048}
6049
6050static const struct file_operations tracing_saved_cmdlines_fops = {
6051 .open = tracing_saved_cmdlines_open,
6052 .read = seq_read,
6053 .llseek = seq_lseek,
6054 .release = seq_release,
6055};
6056
6057static ssize_t
6058tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
6059 size_t cnt, loff_t *ppos)
6060{
6061 char buf[64];
6062 int r;
6063
6064 preempt_disable();
6065 arch_spin_lock(&trace_cmdline_lock);
6066 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
6067 arch_spin_unlock(&trace_cmdline_lock);
6068 preempt_enable();
6069
6070 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6071}
6072
6073static int tracing_resize_saved_cmdlines(unsigned int val)
6074{
6075 struct saved_cmdlines_buffer *s, *savedcmd_temp;
6076
6077 s = allocate_cmdlines_buffer(val);
6078 if (!s)
6079 return -ENOMEM;
6080
6081 preempt_disable();
6082 arch_spin_lock(&trace_cmdline_lock);
6083 savedcmd_temp = savedcmd;
6084 savedcmd = s;
6085 arch_spin_unlock(&trace_cmdline_lock);
6086 preempt_enable();
6087 free_saved_cmdlines_buffer(savedcmd_temp);
6088
6089 return 0;
6090}
6091
6092static ssize_t
6093tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
6094 size_t cnt, loff_t *ppos)
6095{
6096 unsigned long val;
6097 int ret;
6098
6099 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6100 if (ret)
6101 return ret;
6102
6103 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
6104 if (!val || val > PID_MAX_DEFAULT)
6105 return -EINVAL;
6106
6107 ret = tracing_resize_saved_cmdlines((unsigned int)val);
6108 if (ret < 0)
6109 return ret;
6110
6111 *ppos += cnt;
6112
6113 return cnt;
6114}
6115
6116static const struct file_operations tracing_saved_cmdlines_size_fops = {
6117 .open = tracing_open_generic,
6118 .read = tracing_saved_cmdlines_size_read,
6119 .write = tracing_saved_cmdlines_size_write,
6120};
6121
6122#ifdef CONFIG_TRACE_EVAL_MAP_FILE
6123static union trace_eval_map_item *
6124update_eval_map(union trace_eval_map_item *ptr)
6125{
6126 if (!ptr->map.eval_string) {
6127 if (ptr->tail.next) {
6128 ptr = ptr->tail.next;
6129 /* Set ptr to the next real item (skip head) */
6130 ptr++;
6131 } else
6132 return NULL;
6133 }
6134 return ptr;
6135}
6136
6137static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6138{
6139 union trace_eval_map_item *ptr = v;
6140
6141 /*
6142 * Paranoid! If ptr points to end, we don't want to increment past it.
6143 * This really should never happen.
6144 */
6145 (*pos)++;
6146 ptr = update_eval_map(ptr);
6147 if (WARN_ON_ONCE(!ptr))
6148 return NULL;
6149
6150 ptr++;
6151 ptr = update_eval_map(ptr);
6152
6153 return ptr;
6154}
6155
6156static void *eval_map_start(struct seq_file *m, loff_t *pos)
6157{
6158 union trace_eval_map_item *v;
6159 loff_t l = 0;
6160
6161 mutex_lock(&trace_eval_mutex);
6162
6163 v = trace_eval_maps;
6164 if (v)
6165 v++;
6166
6167 while (v && l < *pos) {
6168 v = eval_map_next(m, v, &l);
6169 }
6170
6171 return v;
6172}
6173
6174static void eval_map_stop(struct seq_file *m, void *v)
6175{
6176 mutex_unlock(&trace_eval_mutex);
6177}
6178
6179static int eval_map_show(struct seq_file *m, void *v)
6180{
6181 union trace_eval_map_item *ptr = v;
6182
6183 seq_printf(m, "%s %ld (%s)\n",
6184 ptr->map.eval_string, ptr->map.eval_value,
6185 ptr->map.system);
6186
6187 return 0;
6188}
6189
6190static const struct seq_operations tracing_eval_map_seq_ops = {
6191 .start = eval_map_start,
6192 .next = eval_map_next,
6193 .stop = eval_map_stop,
6194 .show = eval_map_show,
6195};
6196
6197static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6198{
6199 int ret;
6200
6201 ret = tracing_check_open_get_tr(NULL);
6202 if (ret)
6203 return ret;
6204
6205 return seq_open(filp, &tracing_eval_map_seq_ops);
6206}
6207
6208static const struct file_operations tracing_eval_map_fops = {
6209 .open = tracing_eval_map_open,
6210 .read = seq_read,
6211 .llseek = seq_lseek,
6212 .release = seq_release,
6213};
6214
6215static inline union trace_eval_map_item *
6216trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6217{
6218 /* Return tail of array given the head */
6219 return ptr + ptr->head.length + 1;
6220}
6221
6222static void
6223trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6224 int len)
6225{
6226 struct trace_eval_map **stop;
6227 struct trace_eval_map **map;
6228 union trace_eval_map_item *map_array;
6229 union trace_eval_map_item *ptr;
6230
6231 stop = start + len;
6232
6233 /*
6234 * The trace_eval_maps contains the map plus a head and tail item,
6235 * where the head holds the module and length of array, and the
6236 * tail holds a pointer to the next list.
6237 */
6238 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6239 if (!map_array) {
6240 pr_warn("Unable to allocate trace eval mapping\n");
6241 return;
6242 }
6243
6244 mutex_lock(&trace_eval_mutex);
6245
6246 if (!trace_eval_maps)
6247 trace_eval_maps = map_array;
6248 else {
6249 ptr = trace_eval_maps;
6250 for (;;) {
6251 ptr = trace_eval_jmp_to_tail(ptr);
6252 if (!ptr->tail.next)
6253 break;
6254 ptr = ptr->tail.next;
6255
6256 }
6257 ptr->tail.next = map_array;
6258 }
6259 map_array->head.mod = mod;
6260 map_array->head.length = len;
6261 map_array++;
6262
6263 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6264 map_array->map = **map;
6265 map_array++;
6266 }
6267 memset(map_array, 0, sizeof(*map_array));
6268
6269 mutex_unlock(&trace_eval_mutex);
6270}
6271
6272static void trace_create_eval_file(struct dentry *d_tracer)
6273{
6274 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6275 NULL, &tracing_eval_map_fops);
6276}
6277
6278#else /* CONFIG_TRACE_EVAL_MAP_FILE */
6279static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6280static inline void trace_insert_eval_map_file(struct module *mod,
6281 struct trace_eval_map **start, int len) { }
6282#endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6283
6284static void trace_insert_eval_map(struct module *mod,
6285 struct trace_eval_map **start, int len)
6286{
6287 struct trace_eval_map **map;
6288
6289 if (len <= 0)
6290 return;
6291
6292 map = start;
6293
6294 trace_event_eval_update(map, len);
6295
6296 trace_insert_eval_map_file(mod, start, len);
6297}
6298
6299static ssize_t
6300tracing_set_trace_read(struct file *filp, char __user *ubuf,
6301 size_t cnt, loff_t *ppos)
6302{
6303 struct trace_array *tr = filp->private_data;
6304 char buf[MAX_TRACER_SIZE+2];
6305 int r;
6306
6307 mutex_lock(&trace_types_lock);
6308 r = sprintf(buf, "%s\n", tr->current_trace->name);
6309 mutex_unlock(&trace_types_lock);
6310
6311 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6312}
6313
6314int tracer_init(struct tracer *t, struct trace_array *tr)
6315{
6316 tracing_reset_online_cpus(&tr->array_buffer);
6317 return t->init(tr);
6318}
6319
6320static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6321{
6322 int cpu;
6323
6324 for_each_tracing_cpu(cpu)
6325 per_cpu_ptr(buf->data, cpu)->entries = val;
6326}
6327
6328static void update_buffer_entries(struct array_buffer *buf, int cpu)
6329{
6330 if (cpu == RING_BUFFER_ALL_CPUS) {
6331 set_buffer_entries(buf, ring_buffer_size(buf->buffer, 0));
6332 } else {
6333 per_cpu_ptr(buf->data, cpu)->entries = ring_buffer_size(buf->buffer, cpu);
6334 }
6335}
6336
6337#ifdef CONFIG_TRACER_MAX_TRACE
6338/* resize @tr's buffer to the size of @size_tr's entries */
6339static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6340 struct array_buffer *size_buf, int cpu_id)
6341{
6342 int cpu, ret = 0;
6343
6344 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6345 for_each_tracing_cpu(cpu) {
6346 ret = ring_buffer_resize(trace_buf->buffer,
6347 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6348 if (ret < 0)
6349 break;
6350 per_cpu_ptr(trace_buf->data, cpu)->entries =
6351 per_cpu_ptr(size_buf->data, cpu)->entries;
6352 }
6353 } else {
6354 ret = ring_buffer_resize(trace_buf->buffer,
6355 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6356 if (ret == 0)
6357 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6358 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6359 }
6360
6361 return ret;
6362}
6363#endif /* CONFIG_TRACER_MAX_TRACE */
6364
6365static int __tracing_resize_ring_buffer(struct trace_array *tr,
6366 unsigned long size, int cpu)
6367{
6368 int ret;
6369
6370 /*
6371 * If kernel or user changes the size of the ring buffer
6372 * we use the size that was given, and we can forget about
6373 * expanding it later.
6374 */
6375 trace_set_ring_buffer_expanded(tr);
6376
6377 /* May be called before buffers are initialized */
6378 if (!tr->array_buffer.buffer)
6379 return 0;
6380
6381 /* Do not allow tracing while resizing ring buffer */
6382 tracing_stop_tr(tr);
6383
6384 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6385 if (ret < 0)
6386 goto out_start;
6387
6388#ifdef CONFIG_TRACER_MAX_TRACE
6389 if (!tr->allocated_snapshot)
6390 goto out;
6391
6392 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6393 if (ret < 0) {
6394 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6395 &tr->array_buffer, cpu);
6396 if (r < 0) {
6397 /*
6398 * AARGH! We are left with different
6399 * size max buffer!!!!
6400 * The max buffer is our "snapshot" buffer.
6401 * When a tracer needs a snapshot (one of the
6402 * latency tracers), it swaps the max buffer
6403 * with the saved snap shot. We succeeded to
6404 * update the size of the main buffer, but failed to
6405 * update the size of the max buffer. But when we tried
6406 * to reset the main buffer to the original size, we
6407 * failed there too. This is very unlikely to
6408 * happen, but if it does, warn and kill all
6409 * tracing.
6410 */
6411 WARN_ON(1);
6412 tracing_disabled = 1;
6413 }
6414 goto out_start;
6415 }
6416
6417 update_buffer_entries(&tr->max_buffer, cpu);
6418
6419 out:
6420#endif /* CONFIG_TRACER_MAX_TRACE */
6421
6422 update_buffer_entries(&tr->array_buffer, cpu);
6423 out_start:
6424 tracing_start_tr(tr);
6425 return ret;
6426}
6427
6428ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6429 unsigned long size, int cpu_id)
6430{
6431 int ret;
6432
6433 mutex_lock(&trace_types_lock);
6434
6435 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6436 /* make sure, this cpu is enabled in the mask */
6437 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6438 ret = -EINVAL;
6439 goto out;
6440 }
6441 }
6442
6443 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6444 if (ret < 0)
6445 ret = -ENOMEM;
6446
6447out:
6448 mutex_unlock(&trace_types_lock);
6449
6450 return ret;
6451}
6452
6453
6454/**
6455 * tracing_update_buffers - used by tracing facility to expand ring buffers
6456 * @tr: The tracing instance
6457 *
6458 * To save on memory when the tracing is never used on a system with it
6459 * configured in. The ring buffers are set to a minimum size. But once
6460 * a user starts to use the tracing facility, then they need to grow
6461 * to their default size.
6462 *
6463 * This function is to be called when a tracer is about to be used.
6464 */
6465int tracing_update_buffers(struct trace_array *tr)
6466{
6467 int ret = 0;
6468
6469 mutex_lock(&trace_types_lock);
6470 if (!tr->ring_buffer_expanded)
6471 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6472 RING_BUFFER_ALL_CPUS);
6473 mutex_unlock(&trace_types_lock);
6474
6475 return ret;
6476}
6477
6478struct trace_option_dentry;
6479
6480static void
6481create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6482
6483/*
6484 * Used to clear out the tracer before deletion of an instance.
6485 * Must have trace_types_lock held.
6486 */
6487static void tracing_set_nop(struct trace_array *tr)
6488{
6489 if (tr->current_trace == &nop_trace)
6490 return;
6491
6492 tr->current_trace->enabled--;
6493
6494 if (tr->current_trace->reset)
6495 tr->current_trace->reset(tr);
6496
6497 tr->current_trace = &nop_trace;
6498}
6499
6500static bool tracer_options_updated;
6501
6502static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6503{
6504 /* Only enable if the directory has been created already. */
6505 if (!tr->dir)
6506 return;
6507
6508 /* Only create trace option files after update_tracer_options finish */
6509 if (!tracer_options_updated)
6510 return;
6511
6512 create_trace_option_files(tr, t);
6513}
6514
6515int tracing_set_tracer(struct trace_array *tr, const char *buf)
6516{
6517 struct tracer *t;
6518#ifdef CONFIG_TRACER_MAX_TRACE
6519 bool had_max_tr;
6520#endif
6521 int ret = 0;
6522
6523 mutex_lock(&trace_types_lock);
6524
6525 if (!tr->ring_buffer_expanded) {
6526 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6527 RING_BUFFER_ALL_CPUS);
6528 if (ret < 0)
6529 goto out;
6530 ret = 0;
6531 }
6532
6533 for (t = trace_types; t; t = t->next) {
6534 if (strcmp(t->name, buf) == 0)
6535 break;
6536 }
6537 if (!t) {
6538 ret = -EINVAL;
6539 goto out;
6540 }
6541 if (t == tr->current_trace)
6542 goto out;
6543
6544#ifdef CONFIG_TRACER_SNAPSHOT
6545 if (t->use_max_tr) {
6546 local_irq_disable();
6547 arch_spin_lock(&tr->max_lock);
6548 if (tr->cond_snapshot)
6549 ret = -EBUSY;
6550 arch_spin_unlock(&tr->max_lock);
6551 local_irq_enable();
6552 if (ret)
6553 goto out;
6554 }
6555#endif
6556 /* Some tracers won't work on kernel command line */
6557 if (system_state < SYSTEM_RUNNING && t->noboot) {
6558 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6559 t->name);
6560 goto out;
6561 }
6562
6563 /* Some tracers are only allowed for the top level buffer */
6564 if (!trace_ok_for_array(t, tr)) {
6565 ret = -EINVAL;
6566 goto out;
6567 }
6568
6569 /* If trace pipe files are being read, we can't change the tracer */
6570 if (tr->trace_ref) {
6571 ret = -EBUSY;
6572 goto out;
6573 }
6574
6575 trace_branch_disable();
6576
6577 tr->current_trace->enabled--;
6578
6579 if (tr->current_trace->reset)
6580 tr->current_trace->reset(tr);
6581
6582#ifdef CONFIG_TRACER_MAX_TRACE
6583 had_max_tr = tr->current_trace->use_max_tr;
6584
6585 /* Current trace needs to be nop_trace before synchronize_rcu */
6586 tr->current_trace = &nop_trace;
6587
6588 if (had_max_tr && !t->use_max_tr) {
6589 /*
6590 * We need to make sure that the update_max_tr sees that
6591 * current_trace changed to nop_trace to keep it from
6592 * swapping the buffers after we resize it.
6593 * The update_max_tr is called from interrupts disabled
6594 * so a synchronized_sched() is sufficient.
6595 */
6596 synchronize_rcu();
6597 free_snapshot(tr);
6598 }
6599
6600 if (t->use_max_tr && !tr->allocated_snapshot) {
6601 ret = tracing_alloc_snapshot_instance(tr);
6602 if (ret < 0)
6603 goto out;
6604 }
6605#else
6606 tr->current_trace = &nop_trace;
6607#endif
6608
6609 if (t->init) {
6610 ret = tracer_init(t, tr);
6611 if (ret)
6612 goto out;
6613 }
6614
6615 tr->current_trace = t;
6616 tr->current_trace->enabled++;
6617 trace_branch_enable(tr);
6618 out:
6619 mutex_unlock(&trace_types_lock);
6620
6621 return ret;
6622}
6623
6624static ssize_t
6625tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6626 size_t cnt, loff_t *ppos)
6627{
6628 struct trace_array *tr = filp->private_data;
6629 char buf[MAX_TRACER_SIZE+1];
6630 char *name;
6631 size_t ret;
6632 int err;
6633
6634 ret = cnt;
6635
6636 if (cnt > MAX_TRACER_SIZE)
6637 cnt = MAX_TRACER_SIZE;
6638
6639 if (copy_from_user(buf, ubuf, cnt))
6640 return -EFAULT;
6641
6642 buf[cnt] = 0;
6643
6644 name = strim(buf);
6645
6646 err = tracing_set_tracer(tr, name);
6647 if (err)
6648 return err;
6649
6650 *ppos += ret;
6651
6652 return ret;
6653}
6654
6655static ssize_t
6656tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6657 size_t cnt, loff_t *ppos)
6658{
6659 char buf[64];
6660 int r;
6661
6662 r = snprintf(buf, sizeof(buf), "%ld\n",
6663 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6664 if (r > sizeof(buf))
6665 r = sizeof(buf);
6666 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6667}
6668
6669static ssize_t
6670tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6671 size_t cnt, loff_t *ppos)
6672{
6673 unsigned long val;
6674 int ret;
6675
6676 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6677 if (ret)
6678 return ret;
6679
6680 *ptr = val * 1000;
6681
6682 return cnt;
6683}
6684
6685static ssize_t
6686tracing_thresh_read(struct file *filp, char __user *ubuf,
6687 size_t cnt, loff_t *ppos)
6688{
6689 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6690}
6691
6692static ssize_t
6693tracing_thresh_write(struct file *filp, const char __user *ubuf,
6694 size_t cnt, loff_t *ppos)
6695{
6696 struct trace_array *tr = filp->private_data;
6697 int ret;
6698
6699 mutex_lock(&trace_types_lock);
6700 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6701 if (ret < 0)
6702 goto out;
6703
6704 if (tr->current_trace->update_thresh) {
6705 ret = tr->current_trace->update_thresh(tr);
6706 if (ret < 0)
6707 goto out;
6708 }
6709
6710 ret = cnt;
6711out:
6712 mutex_unlock(&trace_types_lock);
6713
6714 return ret;
6715}
6716
6717#ifdef CONFIG_TRACER_MAX_TRACE
6718
6719static ssize_t
6720tracing_max_lat_read(struct file *filp, char __user *ubuf,
6721 size_t cnt, loff_t *ppos)
6722{
6723 struct trace_array *tr = filp->private_data;
6724
6725 return tracing_nsecs_read(&tr->max_latency, ubuf, cnt, ppos);
6726}
6727
6728static ssize_t
6729tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6730 size_t cnt, loff_t *ppos)
6731{
6732 struct trace_array *tr = filp->private_data;
6733
6734 return tracing_nsecs_write(&tr->max_latency, ubuf, cnt, ppos);
6735}
6736
6737#endif
6738
6739static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6740{
6741 if (cpu == RING_BUFFER_ALL_CPUS) {
6742 if (cpumask_empty(tr->pipe_cpumask)) {
6743 cpumask_setall(tr->pipe_cpumask);
6744 return 0;
6745 }
6746 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6747 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6748 return 0;
6749 }
6750 return -EBUSY;
6751}
6752
6753static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6754{
6755 if (cpu == RING_BUFFER_ALL_CPUS) {
6756 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6757 cpumask_clear(tr->pipe_cpumask);
6758 } else {
6759 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6760 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6761 }
6762}
6763
6764static int tracing_open_pipe(struct inode *inode, struct file *filp)
6765{
6766 struct trace_array *tr = inode->i_private;
6767 struct trace_iterator *iter;
6768 int cpu;
6769 int ret;
6770
6771 ret = tracing_check_open_get_tr(tr);
6772 if (ret)
6773 return ret;
6774
6775 mutex_lock(&trace_types_lock);
6776 cpu = tracing_get_cpu(inode);
6777 ret = open_pipe_on_cpu(tr, cpu);
6778 if (ret)
6779 goto fail_pipe_on_cpu;
6780
6781 /* create a buffer to store the information to pass to userspace */
6782 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6783 if (!iter) {
6784 ret = -ENOMEM;
6785 goto fail_alloc_iter;
6786 }
6787
6788 trace_seq_init(&iter->seq);
6789 iter->trace = tr->current_trace;
6790
6791 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6792 ret = -ENOMEM;
6793 goto fail;
6794 }
6795
6796 /* trace pipe does not show start of buffer */
6797 cpumask_setall(iter->started);
6798
6799 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6800 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6801
6802 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6803 if (trace_clocks[tr->clock_id].in_ns)
6804 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6805
6806 iter->tr = tr;
6807 iter->array_buffer = &tr->array_buffer;
6808 iter->cpu_file = cpu;
6809 mutex_init(&iter->mutex);
6810 filp->private_data = iter;
6811
6812 if (iter->trace->pipe_open)
6813 iter->trace->pipe_open(iter);
6814
6815 nonseekable_open(inode, filp);
6816
6817 tr->trace_ref++;
6818
6819 mutex_unlock(&trace_types_lock);
6820 return ret;
6821
6822fail:
6823 kfree(iter);
6824fail_alloc_iter:
6825 close_pipe_on_cpu(tr, cpu);
6826fail_pipe_on_cpu:
6827 __trace_array_put(tr);
6828 mutex_unlock(&trace_types_lock);
6829 return ret;
6830}
6831
6832static int tracing_release_pipe(struct inode *inode, struct file *file)
6833{
6834 struct trace_iterator *iter = file->private_data;
6835 struct trace_array *tr = inode->i_private;
6836
6837 mutex_lock(&trace_types_lock);
6838
6839 tr->trace_ref--;
6840
6841 if (iter->trace->pipe_close)
6842 iter->trace->pipe_close(iter);
6843 close_pipe_on_cpu(tr, iter->cpu_file);
6844 mutex_unlock(&trace_types_lock);
6845
6846 free_trace_iter_content(iter);
6847 kfree(iter);
6848
6849 trace_array_put(tr);
6850
6851 return 0;
6852}
6853
6854static __poll_t
6855trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6856{
6857 struct trace_array *tr = iter->tr;
6858
6859 /* Iterators are static, they should be filled or empty */
6860 if (trace_buffer_iter(iter, iter->cpu_file))
6861 return EPOLLIN | EPOLLRDNORM;
6862
6863 if (tr->trace_flags & TRACE_ITER_BLOCK)
6864 /*
6865 * Always select as readable when in blocking mode
6866 */
6867 return EPOLLIN | EPOLLRDNORM;
6868 else
6869 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6870 filp, poll_table, iter->tr->buffer_percent);
6871}
6872
6873static __poll_t
6874tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6875{
6876 struct trace_iterator *iter = filp->private_data;
6877
6878 return trace_poll(iter, filp, poll_table);
6879}
6880
6881/* Must be called with iter->mutex held. */
6882static int tracing_wait_pipe(struct file *filp)
6883{
6884 struct trace_iterator *iter = filp->private_data;
6885 int ret;
6886
6887 while (trace_empty(iter)) {
6888
6889 if ((filp->f_flags & O_NONBLOCK)) {
6890 return -EAGAIN;
6891 }
6892
6893 /*
6894 * We block until we read something and tracing is disabled.
6895 * We still block if tracing is disabled, but we have never
6896 * read anything. This allows a user to cat this file, and
6897 * then enable tracing. But after we have read something,
6898 * we give an EOF when tracing is again disabled.
6899 *
6900 * iter->pos will be 0 if we haven't read anything.
6901 */
6902 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6903 break;
6904
6905 mutex_unlock(&iter->mutex);
6906
6907 ret = wait_on_pipe(iter, 0);
6908
6909 mutex_lock(&iter->mutex);
6910
6911 if (ret)
6912 return ret;
6913 }
6914
6915 return 1;
6916}
6917
6918/*
6919 * Consumer reader.
6920 */
6921static ssize_t
6922tracing_read_pipe(struct file *filp, char __user *ubuf,
6923 size_t cnt, loff_t *ppos)
6924{
6925 struct trace_iterator *iter = filp->private_data;
6926 ssize_t sret;
6927
6928 /*
6929 * Avoid more than one consumer on a single file descriptor
6930 * This is just a matter of traces coherency, the ring buffer itself
6931 * is protected.
6932 */
6933 mutex_lock(&iter->mutex);
6934
6935 /* return any leftover data */
6936 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6937 if (sret != -EBUSY)
6938 goto out;
6939
6940 trace_seq_init(&iter->seq);
6941
6942 if (iter->trace->read) {
6943 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6944 if (sret)
6945 goto out;
6946 }
6947
6948waitagain:
6949 sret = tracing_wait_pipe(filp);
6950 if (sret <= 0)
6951 goto out;
6952
6953 /* stop when tracing is finished */
6954 if (trace_empty(iter)) {
6955 sret = 0;
6956 goto out;
6957 }
6958
6959 if (cnt >= TRACE_SEQ_BUFFER_SIZE)
6960 cnt = TRACE_SEQ_BUFFER_SIZE - 1;
6961
6962 /* reset all but tr, trace, and overruns */
6963 trace_iterator_reset(iter);
6964 cpumask_clear(iter->started);
6965 trace_seq_init(&iter->seq);
6966
6967 trace_event_read_lock();
6968 trace_access_lock(iter->cpu_file);
6969 while (trace_find_next_entry_inc(iter) != NULL) {
6970 enum print_line_t ret;
6971 int save_len = iter->seq.seq.len;
6972
6973 ret = print_trace_line(iter);
6974 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6975 /*
6976 * If one print_trace_line() fills entire trace_seq in one shot,
6977 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6978 * In this case, we need to consume it, otherwise, loop will peek
6979 * this event next time, resulting in an infinite loop.
6980 */
6981 if (save_len == 0) {
6982 iter->seq.full = 0;
6983 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6984 trace_consume(iter);
6985 break;
6986 }
6987
6988 /* In other cases, don't print partial lines */
6989 iter->seq.seq.len = save_len;
6990 break;
6991 }
6992 if (ret != TRACE_TYPE_NO_CONSUME)
6993 trace_consume(iter);
6994
6995 if (trace_seq_used(&iter->seq) >= cnt)
6996 break;
6997
6998 /*
6999 * Setting the full flag means we reached the trace_seq buffer
7000 * size and we should leave by partial output condition above.
7001 * One of the trace_seq_* functions is not used properly.
7002 */
7003 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
7004 iter->ent->type);
7005 }
7006 trace_access_unlock(iter->cpu_file);
7007 trace_event_read_unlock();
7008
7009 /* Now copy what we have to the user */
7010 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
7011 if (iter->seq.readpos >= trace_seq_used(&iter->seq))
7012 trace_seq_init(&iter->seq);
7013
7014 /*
7015 * If there was nothing to send to user, in spite of consuming trace
7016 * entries, go back to wait for more entries.
7017 */
7018 if (sret == -EBUSY)
7019 goto waitagain;
7020
7021out:
7022 mutex_unlock(&iter->mutex);
7023
7024 return sret;
7025}
7026
7027static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
7028 unsigned int idx)
7029{
7030 __free_page(spd->pages[idx]);
7031}
7032
7033static size_t
7034tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
7035{
7036 size_t count;
7037 int save_len;
7038 int ret;
7039
7040 /* Seq buffer is page-sized, exactly what we need. */
7041 for (;;) {
7042 save_len = iter->seq.seq.len;
7043 ret = print_trace_line(iter);
7044
7045 if (trace_seq_has_overflowed(&iter->seq)) {
7046 iter->seq.seq.len = save_len;
7047 break;
7048 }
7049
7050 /*
7051 * This should not be hit, because it should only
7052 * be set if the iter->seq overflowed. But check it
7053 * anyway to be safe.
7054 */
7055 if (ret == TRACE_TYPE_PARTIAL_LINE) {
7056 iter->seq.seq.len = save_len;
7057 break;
7058 }
7059
7060 count = trace_seq_used(&iter->seq) - save_len;
7061 if (rem < count) {
7062 rem = 0;
7063 iter->seq.seq.len = save_len;
7064 break;
7065 }
7066
7067 if (ret != TRACE_TYPE_NO_CONSUME)
7068 trace_consume(iter);
7069 rem -= count;
7070 if (!trace_find_next_entry_inc(iter)) {
7071 rem = 0;
7072 iter->ent = NULL;
7073 break;
7074 }
7075 }
7076
7077 return rem;
7078}
7079
7080static ssize_t tracing_splice_read_pipe(struct file *filp,
7081 loff_t *ppos,
7082 struct pipe_inode_info *pipe,
7083 size_t len,
7084 unsigned int flags)
7085{
7086 struct page *pages_def[PIPE_DEF_BUFFERS];
7087 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7088 struct trace_iterator *iter = filp->private_data;
7089 struct splice_pipe_desc spd = {
7090 .pages = pages_def,
7091 .partial = partial_def,
7092 .nr_pages = 0, /* This gets updated below. */
7093 .nr_pages_max = PIPE_DEF_BUFFERS,
7094 .ops = &default_pipe_buf_ops,
7095 .spd_release = tracing_spd_release_pipe,
7096 };
7097 ssize_t ret;
7098 size_t rem;
7099 unsigned int i;
7100
7101 if (splice_grow_spd(pipe, &spd))
7102 return -ENOMEM;
7103
7104 mutex_lock(&iter->mutex);
7105
7106 if (iter->trace->splice_read) {
7107 ret = iter->trace->splice_read(iter, filp,
7108 ppos, pipe, len, flags);
7109 if (ret)
7110 goto out_err;
7111 }
7112
7113 ret = tracing_wait_pipe(filp);
7114 if (ret <= 0)
7115 goto out_err;
7116
7117 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
7118 ret = -EFAULT;
7119 goto out_err;
7120 }
7121
7122 trace_event_read_lock();
7123 trace_access_lock(iter->cpu_file);
7124
7125 /* Fill as many pages as possible. */
7126 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
7127 spd.pages[i] = alloc_page(GFP_KERNEL);
7128 if (!spd.pages[i])
7129 break;
7130
7131 rem = tracing_fill_pipe_page(rem, iter);
7132
7133 /* Copy the data into the page, so we can start over. */
7134 ret = trace_seq_to_buffer(&iter->seq,
7135 page_address(spd.pages[i]),
7136 trace_seq_used(&iter->seq));
7137 if (ret < 0) {
7138 __free_page(spd.pages[i]);
7139 break;
7140 }
7141 spd.partial[i].offset = 0;
7142 spd.partial[i].len = trace_seq_used(&iter->seq);
7143
7144 trace_seq_init(&iter->seq);
7145 }
7146
7147 trace_access_unlock(iter->cpu_file);
7148 trace_event_read_unlock();
7149 mutex_unlock(&iter->mutex);
7150
7151 spd.nr_pages = i;
7152
7153 if (i)
7154 ret = splice_to_pipe(pipe, &spd);
7155 else
7156 ret = 0;
7157out:
7158 splice_shrink_spd(&spd);
7159 return ret;
7160
7161out_err:
7162 mutex_unlock(&iter->mutex);
7163 goto out;
7164}
7165
7166static ssize_t
7167tracing_entries_read(struct file *filp, char __user *ubuf,
7168 size_t cnt, loff_t *ppos)
7169{
7170 struct inode *inode = file_inode(filp);
7171 struct trace_array *tr = inode->i_private;
7172 int cpu = tracing_get_cpu(inode);
7173 char buf[64];
7174 int r = 0;
7175 ssize_t ret;
7176
7177 mutex_lock(&trace_types_lock);
7178
7179 if (cpu == RING_BUFFER_ALL_CPUS) {
7180 int cpu, buf_size_same;
7181 unsigned long size;
7182
7183 size = 0;
7184 buf_size_same = 1;
7185 /* check if all cpu sizes are same */
7186 for_each_tracing_cpu(cpu) {
7187 /* fill in the size from first enabled cpu */
7188 if (size == 0)
7189 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7190 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7191 buf_size_same = 0;
7192 break;
7193 }
7194 }
7195
7196 if (buf_size_same) {
7197 if (!tr->ring_buffer_expanded)
7198 r = sprintf(buf, "%lu (expanded: %lu)\n",
7199 size >> 10,
7200 trace_buf_size >> 10);
7201 else
7202 r = sprintf(buf, "%lu\n", size >> 10);
7203 } else
7204 r = sprintf(buf, "X\n");
7205 } else
7206 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7207
7208 mutex_unlock(&trace_types_lock);
7209
7210 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7211 return ret;
7212}
7213
7214static ssize_t
7215tracing_entries_write(struct file *filp, const char __user *ubuf,
7216 size_t cnt, loff_t *ppos)
7217{
7218 struct inode *inode = file_inode(filp);
7219 struct trace_array *tr = inode->i_private;
7220 unsigned long val;
7221 int ret;
7222
7223 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7224 if (ret)
7225 return ret;
7226
7227 /* must have at least 1 entry */
7228 if (!val)
7229 return -EINVAL;
7230
7231 /* value is in KB */
7232 val <<= 10;
7233 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7234 if (ret < 0)
7235 return ret;
7236
7237 *ppos += cnt;
7238
7239 return cnt;
7240}
7241
7242static ssize_t
7243tracing_total_entries_read(struct file *filp, char __user *ubuf,
7244 size_t cnt, loff_t *ppos)
7245{
7246 struct trace_array *tr = filp->private_data;
7247 char buf[64];
7248 int r, cpu;
7249 unsigned long size = 0, expanded_size = 0;
7250
7251 mutex_lock(&trace_types_lock);
7252 for_each_tracing_cpu(cpu) {
7253 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7254 if (!tr->ring_buffer_expanded)
7255 expanded_size += trace_buf_size >> 10;
7256 }
7257 if (tr->ring_buffer_expanded)
7258 r = sprintf(buf, "%lu\n", size);
7259 else
7260 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7261 mutex_unlock(&trace_types_lock);
7262
7263 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7264}
7265
7266static ssize_t
7267tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7268 size_t cnt, loff_t *ppos)
7269{
7270 /*
7271 * There is no need to read what the user has written, this function
7272 * is just to make sure that there is no error when "echo" is used
7273 */
7274
7275 *ppos += cnt;
7276
7277 return cnt;
7278}
7279
7280static int
7281tracing_free_buffer_release(struct inode *inode, struct file *filp)
7282{
7283 struct trace_array *tr = inode->i_private;
7284
7285 /* disable tracing ? */
7286 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7287 tracer_tracing_off(tr);
7288 /* resize the ring buffer to 0 */
7289 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7290
7291 trace_array_put(tr);
7292
7293 return 0;
7294}
7295
7296#define TRACE_MARKER_MAX_SIZE 4096
7297
7298static ssize_t
7299tracing_mark_write(struct file *filp, const char __user *ubuf,
7300 size_t cnt, loff_t *fpos)
7301{
7302 struct trace_array *tr = filp->private_data;
7303 struct ring_buffer_event *event;
7304 enum event_trigger_type tt = ETT_NONE;
7305 struct trace_buffer *buffer;
7306 struct print_entry *entry;
7307 int meta_size;
7308 ssize_t written;
7309 size_t size;
7310 int len;
7311
7312/* Used in tracing_mark_raw_write() as well */
7313#define FAULTED_STR "<faulted>"
7314#define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7315
7316 if (tracing_disabled)
7317 return -EINVAL;
7318
7319 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7320 return -EINVAL;
7321
7322 if ((ssize_t)cnt < 0)
7323 return -EINVAL;
7324
7325 if (cnt > TRACE_MARKER_MAX_SIZE)
7326 cnt = TRACE_MARKER_MAX_SIZE;
7327
7328 meta_size = sizeof(*entry) + 2; /* add '\0' and possible '\n' */
7329 again:
7330 size = cnt + meta_size;
7331
7332 /* If less than "<faulted>", then make sure we can still add that */
7333 if (cnt < FAULTED_SIZE)
7334 size += FAULTED_SIZE - cnt;
7335
7336 buffer = tr->array_buffer.buffer;
7337 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7338 tracing_gen_ctx());
7339 if (unlikely(!event)) {
7340 /*
7341 * If the size was greater than what was allowed, then
7342 * make it smaller and try again.
7343 */
7344 if (size > ring_buffer_max_event_size(buffer)) {
7345 /* cnt < FAULTED size should never be bigger than max */
7346 if (WARN_ON_ONCE(cnt < FAULTED_SIZE))
7347 return -EBADF;
7348 cnt = ring_buffer_max_event_size(buffer) - meta_size;
7349 /* The above should only happen once */
7350 if (WARN_ON_ONCE(cnt + meta_size == size))
7351 return -EBADF;
7352 goto again;
7353 }
7354
7355 /* Ring buffer disabled, return as if not open for write */
7356 return -EBADF;
7357 }
7358
7359 entry = ring_buffer_event_data(event);
7360 entry->ip = _THIS_IP_;
7361
7362 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7363 if (len) {
7364 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7365 cnt = FAULTED_SIZE;
7366 written = -EFAULT;
7367 } else
7368 written = cnt;
7369
7370 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7371 /* do not add \n before testing triggers, but add \0 */
7372 entry->buf[cnt] = '\0';
7373 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7374 }
7375
7376 if (entry->buf[cnt - 1] != '\n') {
7377 entry->buf[cnt] = '\n';
7378 entry->buf[cnt + 1] = '\0';
7379 } else
7380 entry->buf[cnt] = '\0';
7381
7382 if (static_branch_unlikely(&trace_marker_exports_enabled))
7383 ftrace_exports(event, TRACE_EXPORT_MARKER);
7384 __buffer_unlock_commit(buffer, event);
7385
7386 if (tt)
7387 event_triggers_post_call(tr->trace_marker_file, tt);
7388
7389 return written;
7390}
7391
7392static ssize_t
7393tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7394 size_t cnt, loff_t *fpos)
7395{
7396 struct trace_array *tr = filp->private_data;
7397 struct ring_buffer_event *event;
7398 struct trace_buffer *buffer;
7399 struct raw_data_entry *entry;
7400 ssize_t written;
7401 int size;
7402 int len;
7403
7404#define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7405
7406 if (tracing_disabled)
7407 return -EINVAL;
7408
7409 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7410 return -EINVAL;
7411
7412 /* The marker must at least have a tag id */
7413 if (cnt < sizeof(unsigned int))
7414 return -EINVAL;
7415
7416 size = sizeof(*entry) + cnt;
7417 if (cnt < FAULT_SIZE_ID)
7418 size += FAULT_SIZE_ID - cnt;
7419
7420 buffer = tr->array_buffer.buffer;
7421
7422 if (size > ring_buffer_max_event_size(buffer))
7423 return -EINVAL;
7424
7425 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7426 tracing_gen_ctx());
7427 if (!event)
7428 /* Ring buffer disabled, return as if not open for write */
7429 return -EBADF;
7430
7431 entry = ring_buffer_event_data(event);
7432
7433 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7434 if (len) {
7435 entry->id = -1;
7436 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7437 written = -EFAULT;
7438 } else
7439 written = cnt;
7440
7441 __buffer_unlock_commit(buffer, event);
7442
7443 return written;
7444}
7445
7446static int tracing_clock_show(struct seq_file *m, void *v)
7447{
7448 struct trace_array *tr = m->private;
7449 int i;
7450
7451 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7452 seq_printf(m,
7453 "%s%s%s%s", i ? " " : "",
7454 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7455 i == tr->clock_id ? "]" : "");
7456 seq_putc(m, '\n');
7457
7458 return 0;
7459}
7460
7461int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7462{
7463 int i;
7464
7465 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7466 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7467 break;
7468 }
7469 if (i == ARRAY_SIZE(trace_clocks))
7470 return -EINVAL;
7471
7472 mutex_lock(&trace_types_lock);
7473
7474 tr->clock_id = i;
7475
7476 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7477
7478 /*
7479 * New clock may not be consistent with the previous clock.
7480 * Reset the buffer so that it doesn't have incomparable timestamps.
7481 */
7482 tracing_reset_online_cpus(&tr->array_buffer);
7483
7484#ifdef CONFIG_TRACER_MAX_TRACE
7485 if (tr->max_buffer.buffer)
7486 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7487 tracing_reset_online_cpus(&tr->max_buffer);
7488#endif
7489
7490 mutex_unlock(&trace_types_lock);
7491
7492 return 0;
7493}
7494
7495static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7496 size_t cnt, loff_t *fpos)
7497{
7498 struct seq_file *m = filp->private_data;
7499 struct trace_array *tr = m->private;
7500 char buf[64];
7501 const char *clockstr;
7502 int ret;
7503
7504 if (cnt >= sizeof(buf))
7505 return -EINVAL;
7506
7507 if (copy_from_user(buf, ubuf, cnt))
7508 return -EFAULT;
7509
7510 buf[cnt] = 0;
7511
7512 clockstr = strstrip(buf);
7513
7514 ret = tracing_set_clock(tr, clockstr);
7515 if (ret)
7516 return ret;
7517
7518 *fpos += cnt;
7519
7520 return cnt;
7521}
7522
7523static int tracing_clock_open(struct inode *inode, struct file *file)
7524{
7525 struct trace_array *tr = inode->i_private;
7526 int ret;
7527
7528 ret = tracing_check_open_get_tr(tr);
7529 if (ret)
7530 return ret;
7531
7532 ret = single_open(file, tracing_clock_show, inode->i_private);
7533 if (ret < 0)
7534 trace_array_put(tr);
7535
7536 return ret;
7537}
7538
7539static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7540{
7541 struct trace_array *tr = m->private;
7542
7543 mutex_lock(&trace_types_lock);
7544
7545 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7546 seq_puts(m, "delta [absolute]\n");
7547 else
7548 seq_puts(m, "[delta] absolute\n");
7549
7550 mutex_unlock(&trace_types_lock);
7551
7552 return 0;
7553}
7554
7555static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7556{
7557 struct trace_array *tr = inode->i_private;
7558 int ret;
7559
7560 ret = tracing_check_open_get_tr(tr);
7561 if (ret)
7562 return ret;
7563
7564 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7565 if (ret < 0)
7566 trace_array_put(tr);
7567
7568 return ret;
7569}
7570
7571u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7572{
7573 if (rbe == this_cpu_read(trace_buffered_event))
7574 return ring_buffer_time_stamp(buffer);
7575
7576 return ring_buffer_event_time_stamp(buffer, rbe);
7577}
7578
7579/*
7580 * Set or disable using the per CPU trace_buffer_event when possible.
7581 */
7582int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7583{
7584 int ret = 0;
7585
7586 mutex_lock(&trace_types_lock);
7587
7588 if (set && tr->no_filter_buffering_ref++)
7589 goto out;
7590
7591 if (!set) {
7592 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7593 ret = -EINVAL;
7594 goto out;
7595 }
7596
7597 --tr->no_filter_buffering_ref;
7598 }
7599 out:
7600 mutex_unlock(&trace_types_lock);
7601
7602 return ret;
7603}
7604
7605struct ftrace_buffer_info {
7606 struct trace_iterator iter;
7607 void *spare;
7608 unsigned int spare_cpu;
7609 unsigned int spare_size;
7610 unsigned int read;
7611};
7612
7613#ifdef CONFIG_TRACER_SNAPSHOT
7614static int tracing_snapshot_open(struct inode *inode, struct file *file)
7615{
7616 struct trace_array *tr = inode->i_private;
7617 struct trace_iterator *iter;
7618 struct seq_file *m;
7619 int ret;
7620
7621 ret = tracing_check_open_get_tr(tr);
7622 if (ret)
7623 return ret;
7624
7625 if (file->f_mode & FMODE_READ) {
7626 iter = __tracing_open(inode, file, true);
7627 if (IS_ERR(iter))
7628 ret = PTR_ERR(iter);
7629 } else {
7630 /* Writes still need the seq_file to hold the private data */
7631 ret = -ENOMEM;
7632 m = kzalloc(sizeof(*m), GFP_KERNEL);
7633 if (!m)
7634 goto out;
7635 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7636 if (!iter) {
7637 kfree(m);
7638 goto out;
7639 }
7640 ret = 0;
7641
7642 iter->tr = tr;
7643 iter->array_buffer = &tr->max_buffer;
7644 iter->cpu_file = tracing_get_cpu(inode);
7645 m->private = iter;
7646 file->private_data = m;
7647 }
7648out:
7649 if (ret < 0)
7650 trace_array_put(tr);
7651
7652 return ret;
7653}
7654
7655static void tracing_swap_cpu_buffer(void *tr)
7656{
7657 update_max_tr_single((struct trace_array *)tr, current, smp_processor_id());
7658}
7659
7660static ssize_t
7661tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7662 loff_t *ppos)
7663{
7664 struct seq_file *m = filp->private_data;
7665 struct trace_iterator *iter = m->private;
7666 struct trace_array *tr = iter->tr;
7667 unsigned long val;
7668 int ret;
7669
7670 ret = tracing_update_buffers(tr);
7671 if (ret < 0)
7672 return ret;
7673
7674 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7675 if (ret)
7676 return ret;
7677
7678 mutex_lock(&trace_types_lock);
7679
7680 if (tr->current_trace->use_max_tr) {
7681 ret = -EBUSY;
7682 goto out;
7683 }
7684
7685 local_irq_disable();
7686 arch_spin_lock(&tr->max_lock);
7687 if (tr->cond_snapshot)
7688 ret = -EBUSY;
7689 arch_spin_unlock(&tr->max_lock);
7690 local_irq_enable();
7691 if (ret)
7692 goto out;
7693
7694 switch (val) {
7695 case 0:
7696 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7697 ret = -EINVAL;
7698 break;
7699 }
7700 if (tr->allocated_snapshot)
7701 free_snapshot(tr);
7702 break;
7703 case 1:
7704/* Only allow per-cpu swap if the ring buffer supports it */
7705#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7706 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7707 ret = -EINVAL;
7708 break;
7709 }
7710#endif
7711 if (tr->allocated_snapshot)
7712 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7713 &tr->array_buffer, iter->cpu_file);
7714 else
7715 ret = tracing_alloc_snapshot_instance(tr);
7716 if (ret < 0)
7717 break;
7718 /* Now, we're going to swap */
7719 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
7720 local_irq_disable();
7721 update_max_tr(tr, current, smp_processor_id(), NULL);
7722 local_irq_enable();
7723 } else {
7724 smp_call_function_single(iter->cpu_file, tracing_swap_cpu_buffer,
7725 (void *)tr, 1);
7726 }
7727 break;
7728 default:
7729 if (tr->allocated_snapshot) {
7730 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7731 tracing_reset_online_cpus(&tr->max_buffer);
7732 else
7733 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7734 }
7735 break;
7736 }
7737
7738 if (ret >= 0) {
7739 *ppos += cnt;
7740 ret = cnt;
7741 }
7742out:
7743 mutex_unlock(&trace_types_lock);
7744 return ret;
7745}
7746
7747static int tracing_snapshot_release(struct inode *inode, struct file *file)
7748{
7749 struct seq_file *m = file->private_data;
7750 int ret;
7751
7752 ret = tracing_release(inode, file);
7753
7754 if (file->f_mode & FMODE_READ)
7755 return ret;
7756
7757 /* If write only, the seq_file is just a stub */
7758 if (m)
7759 kfree(m->private);
7760 kfree(m);
7761
7762 return 0;
7763}
7764
7765static int tracing_buffers_open(struct inode *inode, struct file *filp);
7766static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7767 size_t count, loff_t *ppos);
7768static int tracing_buffers_release(struct inode *inode, struct file *file);
7769static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7770 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7771
7772static int snapshot_raw_open(struct inode *inode, struct file *filp)
7773{
7774 struct ftrace_buffer_info *info;
7775 int ret;
7776
7777 /* The following checks for tracefs lockdown */
7778 ret = tracing_buffers_open(inode, filp);
7779 if (ret < 0)
7780 return ret;
7781
7782 info = filp->private_data;
7783
7784 if (info->iter.trace->use_max_tr) {
7785 tracing_buffers_release(inode, filp);
7786 return -EBUSY;
7787 }
7788
7789 info->iter.snapshot = true;
7790 info->iter.array_buffer = &info->iter.tr->max_buffer;
7791
7792 return ret;
7793}
7794
7795#endif /* CONFIG_TRACER_SNAPSHOT */
7796
7797
7798static const struct file_operations tracing_thresh_fops = {
7799 .open = tracing_open_generic,
7800 .read = tracing_thresh_read,
7801 .write = tracing_thresh_write,
7802 .llseek = generic_file_llseek,
7803};
7804
7805#ifdef CONFIG_TRACER_MAX_TRACE
7806static const struct file_operations tracing_max_lat_fops = {
7807 .open = tracing_open_generic_tr,
7808 .read = tracing_max_lat_read,
7809 .write = tracing_max_lat_write,
7810 .llseek = generic_file_llseek,
7811 .release = tracing_release_generic_tr,
7812};
7813#endif
7814
7815static const struct file_operations set_tracer_fops = {
7816 .open = tracing_open_generic_tr,
7817 .read = tracing_set_trace_read,
7818 .write = tracing_set_trace_write,
7819 .llseek = generic_file_llseek,
7820 .release = tracing_release_generic_tr,
7821};
7822
7823static const struct file_operations tracing_pipe_fops = {
7824 .open = tracing_open_pipe,
7825 .poll = tracing_poll_pipe,
7826 .read = tracing_read_pipe,
7827 .splice_read = tracing_splice_read_pipe,
7828 .release = tracing_release_pipe,
7829 .llseek = no_llseek,
7830};
7831
7832static const struct file_operations tracing_entries_fops = {
7833 .open = tracing_open_generic_tr,
7834 .read = tracing_entries_read,
7835 .write = tracing_entries_write,
7836 .llseek = generic_file_llseek,
7837 .release = tracing_release_generic_tr,
7838};
7839
7840static const struct file_operations tracing_total_entries_fops = {
7841 .open = tracing_open_generic_tr,
7842 .read = tracing_total_entries_read,
7843 .llseek = generic_file_llseek,
7844 .release = tracing_release_generic_tr,
7845};
7846
7847static const struct file_operations tracing_free_buffer_fops = {
7848 .open = tracing_open_generic_tr,
7849 .write = tracing_free_buffer_write,
7850 .release = tracing_free_buffer_release,
7851};
7852
7853static const struct file_operations tracing_mark_fops = {
7854 .open = tracing_mark_open,
7855 .write = tracing_mark_write,
7856 .release = tracing_release_generic_tr,
7857};
7858
7859static const struct file_operations tracing_mark_raw_fops = {
7860 .open = tracing_mark_open,
7861 .write = tracing_mark_raw_write,
7862 .release = tracing_release_generic_tr,
7863};
7864
7865static const struct file_operations trace_clock_fops = {
7866 .open = tracing_clock_open,
7867 .read = seq_read,
7868 .llseek = seq_lseek,
7869 .release = tracing_single_release_tr,
7870 .write = tracing_clock_write,
7871};
7872
7873static const struct file_operations trace_time_stamp_mode_fops = {
7874 .open = tracing_time_stamp_mode_open,
7875 .read = seq_read,
7876 .llseek = seq_lseek,
7877 .release = tracing_single_release_tr,
7878};
7879
7880#ifdef CONFIG_TRACER_SNAPSHOT
7881static const struct file_operations snapshot_fops = {
7882 .open = tracing_snapshot_open,
7883 .read = seq_read,
7884 .write = tracing_snapshot_write,
7885 .llseek = tracing_lseek,
7886 .release = tracing_snapshot_release,
7887};
7888
7889static const struct file_operations snapshot_raw_fops = {
7890 .open = snapshot_raw_open,
7891 .read = tracing_buffers_read,
7892 .release = tracing_buffers_release,
7893 .splice_read = tracing_buffers_splice_read,
7894 .llseek = no_llseek,
7895};
7896
7897#endif /* CONFIG_TRACER_SNAPSHOT */
7898
7899/*
7900 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7901 * @filp: The active open file structure
7902 * @ubuf: The userspace provided buffer to read value into
7903 * @cnt: The maximum number of bytes to read
7904 * @ppos: The current "file" position
7905 *
7906 * This function implements the write interface for a struct trace_min_max_param.
7907 * The filp->private_data must point to a trace_min_max_param structure that
7908 * defines where to write the value, the min and the max acceptable values,
7909 * and a lock to protect the write.
7910 */
7911static ssize_t
7912trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7913{
7914 struct trace_min_max_param *param = filp->private_data;
7915 u64 val;
7916 int err;
7917
7918 if (!param)
7919 return -EFAULT;
7920
7921 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7922 if (err)
7923 return err;
7924
7925 if (param->lock)
7926 mutex_lock(param->lock);
7927
7928 if (param->min && val < *param->min)
7929 err = -EINVAL;
7930
7931 if (param->max && val > *param->max)
7932 err = -EINVAL;
7933
7934 if (!err)
7935 *param->val = val;
7936
7937 if (param->lock)
7938 mutex_unlock(param->lock);
7939
7940 if (err)
7941 return err;
7942
7943 return cnt;
7944}
7945
7946/*
7947 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7948 * @filp: The active open file structure
7949 * @ubuf: The userspace provided buffer to read value into
7950 * @cnt: The maximum number of bytes to read
7951 * @ppos: The current "file" position
7952 *
7953 * This function implements the read interface for a struct trace_min_max_param.
7954 * The filp->private_data must point to a trace_min_max_param struct with valid
7955 * data.
7956 */
7957static ssize_t
7958trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7959{
7960 struct trace_min_max_param *param = filp->private_data;
7961 char buf[U64_STR_SIZE];
7962 int len;
7963 u64 val;
7964
7965 if (!param)
7966 return -EFAULT;
7967
7968 val = *param->val;
7969
7970 if (cnt > sizeof(buf))
7971 cnt = sizeof(buf);
7972
7973 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7974
7975 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7976}
7977
7978const struct file_operations trace_min_max_fops = {
7979 .open = tracing_open_generic,
7980 .read = trace_min_max_read,
7981 .write = trace_min_max_write,
7982};
7983
7984#define TRACING_LOG_ERRS_MAX 8
7985#define TRACING_LOG_LOC_MAX 128
7986
7987#define CMD_PREFIX " Command: "
7988
7989struct err_info {
7990 const char **errs; /* ptr to loc-specific array of err strings */
7991 u8 type; /* index into errs -> specific err string */
7992 u16 pos; /* caret position */
7993 u64 ts;
7994};
7995
7996struct tracing_log_err {
7997 struct list_head list;
7998 struct err_info info;
7999 char loc[TRACING_LOG_LOC_MAX]; /* err location */
8000 char *cmd; /* what caused err */
8001};
8002
8003static DEFINE_MUTEX(tracing_err_log_lock);
8004
8005static struct tracing_log_err *alloc_tracing_log_err(int len)
8006{
8007 struct tracing_log_err *err;
8008
8009 err = kzalloc(sizeof(*err), GFP_KERNEL);
8010 if (!err)
8011 return ERR_PTR(-ENOMEM);
8012
8013 err->cmd = kzalloc(len, GFP_KERNEL);
8014 if (!err->cmd) {
8015 kfree(err);
8016 return ERR_PTR(-ENOMEM);
8017 }
8018
8019 return err;
8020}
8021
8022static void free_tracing_log_err(struct tracing_log_err *err)
8023{
8024 kfree(err->cmd);
8025 kfree(err);
8026}
8027
8028static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
8029 int len)
8030{
8031 struct tracing_log_err *err;
8032 char *cmd;
8033
8034 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
8035 err = alloc_tracing_log_err(len);
8036 if (PTR_ERR(err) != -ENOMEM)
8037 tr->n_err_log_entries++;
8038
8039 return err;
8040 }
8041 cmd = kzalloc(len, GFP_KERNEL);
8042 if (!cmd)
8043 return ERR_PTR(-ENOMEM);
8044 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
8045 kfree(err->cmd);
8046 err->cmd = cmd;
8047 list_del(&err->list);
8048
8049 return err;
8050}
8051
8052/**
8053 * err_pos - find the position of a string within a command for error careting
8054 * @cmd: The tracing command that caused the error
8055 * @str: The string to position the caret at within @cmd
8056 *
8057 * Finds the position of the first occurrence of @str within @cmd. The
8058 * return value can be passed to tracing_log_err() for caret placement
8059 * within @cmd.
8060 *
8061 * Returns the index within @cmd of the first occurrence of @str or 0
8062 * if @str was not found.
8063 */
8064unsigned int err_pos(char *cmd, const char *str)
8065{
8066 char *found;
8067
8068 if (WARN_ON(!strlen(cmd)))
8069 return 0;
8070
8071 found = strstr(cmd, str);
8072 if (found)
8073 return found - cmd;
8074
8075 return 0;
8076}
8077
8078/**
8079 * tracing_log_err - write an error to the tracing error log
8080 * @tr: The associated trace array for the error (NULL for top level array)
8081 * @loc: A string describing where the error occurred
8082 * @cmd: The tracing command that caused the error
8083 * @errs: The array of loc-specific static error strings
8084 * @type: The index into errs[], which produces the specific static err string
8085 * @pos: The position the caret should be placed in the cmd
8086 *
8087 * Writes an error into tracing/error_log of the form:
8088 *
8089 * <loc>: error: <text>
8090 * Command: <cmd>
8091 * ^
8092 *
8093 * tracing/error_log is a small log file containing the last
8094 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
8095 * unless there has been a tracing error, and the error log can be
8096 * cleared and have its memory freed by writing the empty string in
8097 * truncation mode to it i.e. echo > tracing/error_log.
8098 *
8099 * NOTE: the @errs array along with the @type param are used to
8100 * produce a static error string - this string is not copied and saved
8101 * when the error is logged - only a pointer to it is saved. See
8102 * existing callers for examples of how static strings are typically
8103 * defined for use with tracing_log_err().
8104 */
8105void tracing_log_err(struct trace_array *tr,
8106 const char *loc, const char *cmd,
8107 const char **errs, u8 type, u16 pos)
8108{
8109 struct tracing_log_err *err;
8110 int len = 0;
8111
8112 if (!tr)
8113 tr = &global_trace;
8114
8115 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
8116
8117 mutex_lock(&tracing_err_log_lock);
8118 err = get_tracing_log_err(tr, len);
8119 if (PTR_ERR(err) == -ENOMEM) {
8120 mutex_unlock(&tracing_err_log_lock);
8121 return;
8122 }
8123
8124 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
8125 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
8126
8127 err->info.errs = errs;
8128 err->info.type = type;
8129 err->info.pos = pos;
8130 err->info.ts = local_clock();
8131
8132 list_add_tail(&err->list, &tr->err_log);
8133 mutex_unlock(&tracing_err_log_lock);
8134}
8135
8136static void clear_tracing_err_log(struct trace_array *tr)
8137{
8138 struct tracing_log_err *err, *next;
8139
8140 mutex_lock(&tracing_err_log_lock);
8141 list_for_each_entry_safe(err, next, &tr->err_log, list) {
8142 list_del(&err->list);
8143 free_tracing_log_err(err);
8144 }
8145
8146 tr->n_err_log_entries = 0;
8147 mutex_unlock(&tracing_err_log_lock);
8148}
8149
8150static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
8151{
8152 struct trace_array *tr = m->private;
8153
8154 mutex_lock(&tracing_err_log_lock);
8155
8156 return seq_list_start(&tr->err_log, *pos);
8157}
8158
8159static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
8160{
8161 struct trace_array *tr = m->private;
8162
8163 return seq_list_next(v, &tr->err_log, pos);
8164}
8165
8166static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
8167{
8168 mutex_unlock(&tracing_err_log_lock);
8169}
8170
8171static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
8172{
8173 u16 i;
8174
8175 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
8176 seq_putc(m, ' ');
8177 for (i = 0; i < pos; i++)
8178 seq_putc(m, ' ');
8179 seq_puts(m, "^\n");
8180}
8181
8182static int tracing_err_log_seq_show(struct seq_file *m, void *v)
8183{
8184 struct tracing_log_err *err = v;
8185
8186 if (err) {
8187 const char *err_text = err->info.errs[err->info.type];
8188 u64 sec = err->info.ts;
8189 u32 nsec;
8190
8191 nsec = do_div(sec, NSEC_PER_SEC);
8192 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
8193 err->loc, err_text);
8194 seq_printf(m, "%s", err->cmd);
8195 tracing_err_log_show_pos(m, err->info.pos);
8196 }
8197
8198 return 0;
8199}
8200
8201static const struct seq_operations tracing_err_log_seq_ops = {
8202 .start = tracing_err_log_seq_start,
8203 .next = tracing_err_log_seq_next,
8204 .stop = tracing_err_log_seq_stop,
8205 .show = tracing_err_log_seq_show
8206};
8207
8208static int tracing_err_log_open(struct inode *inode, struct file *file)
8209{
8210 struct trace_array *tr = inode->i_private;
8211 int ret = 0;
8212
8213 ret = tracing_check_open_get_tr(tr);
8214 if (ret)
8215 return ret;
8216
8217 /* If this file was opened for write, then erase contents */
8218 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8219 clear_tracing_err_log(tr);
8220
8221 if (file->f_mode & FMODE_READ) {
8222 ret = seq_open(file, &tracing_err_log_seq_ops);
8223 if (!ret) {
8224 struct seq_file *m = file->private_data;
8225 m->private = tr;
8226 } else {
8227 trace_array_put(tr);
8228 }
8229 }
8230 return ret;
8231}
8232
8233static ssize_t tracing_err_log_write(struct file *file,
8234 const char __user *buffer,
8235 size_t count, loff_t *ppos)
8236{
8237 return count;
8238}
8239
8240static int tracing_err_log_release(struct inode *inode, struct file *file)
8241{
8242 struct trace_array *tr = inode->i_private;
8243
8244 trace_array_put(tr);
8245
8246 if (file->f_mode & FMODE_READ)
8247 seq_release(inode, file);
8248
8249 return 0;
8250}
8251
8252static const struct file_operations tracing_err_log_fops = {
8253 .open = tracing_err_log_open,
8254 .write = tracing_err_log_write,
8255 .read = seq_read,
8256 .llseek = tracing_lseek,
8257 .release = tracing_err_log_release,
8258};
8259
8260static int tracing_buffers_open(struct inode *inode, struct file *filp)
8261{
8262 struct trace_array *tr = inode->i_private;
8263 struct ftrace_buffer_info *info;
8264 int ret;
8265
8266 ret = tracing_check_open_get_tr(tr);
8267 if (ret)
8268 return ret;
8269
8270 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8271 if (!info) {
8272 trace_array_put(tr);
8273 return -ENOMEM;
8274 }
8275
8276 mutex_lock(&trace_types_lock);
8277
8278 info->iter.tr = tr;
8279 info->iter.cpu_file = tracing_get_cpu(inode);
8280 info->iter.trace = tr->current_trace;
8281 info->iter.array_buffer = &tr->array_buffer;
8282 info->spare = NULL;
8283 /* Force reading ring buffer for first read */
8284 info->read = (unsigned int)-1;
8285
8286 filp->private_data = info;
8287
8288 tr->trace_ref++;
8289
8290 mutex_unlock(&trace_types_lock);
8291
8292 ret = nonseekable_open(inode, filp);
8293 if (ret < 0)
8294 trace_array_put(tr);
8295
8296 return ret;
8297}
8298
8299static __poll_t
8300tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8301{
8302 struct ftrace_buffer_info *info = filp->private_data;
8303 struct trace_iterator *iter = &info->iter;
8304
8305 return trace_poll(iter, filp, poll_table);
8306}
8307
8308static ssize_t
8309tracing_buffers_read(struct file *filp, char __user *ubuf,
8310 size_t count, loff_t *ppos)
8311{
8312 struct ftrace_buffer_info *info = filp->private_data;
8313 struct trace_iterator *iter = &info->iter;
8314 void *trace_data;
8315 int page_size;
8316 ssize_t ret = 0;
8317 ssize_t size;
8318
8319 if (!count)
8320 return 0;
8321
8322#ifdef CONFIG_TRACER_MAX_TRACE
8323 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8324 return -EBUSY;
8325#endif
8326
8327 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8328
8329 /* Make sure the spare matches the current sub buffer size */
8330 if (info->spare) {
8331 if (page_size != info->spare_size) {
8332 ring_buffer_free_read_page(iter->array_buffer->buffer,
8333 info->spare_cpu, info->spare);
8334 info->spare = NULL;
8335 }
8336 }
8337
8338 if (!info->spare) {
8339 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8340 iter->cpu_file);
8341 if (IS_ERR(info->spare)) {
8342 ret = PTR_ERR(info->spare);
8343 info->spare = NULL;
8344 } else {
8345 info->spare_cpu = iter->cpu_file;
8346 info->spare_size = page_size;
8347 }
8348 }
8349 if (!info->spare)
8350 return ret;
8351
8352 /* Do we have previous read data to read? */
8353 if (info->read < page_size)
8354 goto read;
8355
8356 again:
8357 trace_access_lock(iter->cpu_file);
8358 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8359 info->spare,
8360 count,
8361 iter->cpu_file, 0);
8362 trace_access_unlock(iter->cpu_file);
8363
8364 if (ret < 0) {
8365 if (trace_empty(iter)) {
8366 if ((filp->f_flags & O_NONBLOCK))
8367 return -EAGAIN;
8368
8369 ret = wait_on_pipe(iter, 0);
8370 if (ret)
8371 return ret;
8372
8373 goto again;
8374 }
8375 return 0;
8376 }
8377
8378 info->read = 0;
8379 read:
8380 size = page_size - info->read;
8381 if (size > count)
8382 size = count;
8383 trace_data = ring_buffer_read_page_data(info->spare);
8384 ret = copy_to_user(ubuf, trace_data + info->read, size);
8385 if (ret == size)
8386 return -EFAULT;
8387
8388 size -= ret;
8389
8390 *ppos += size;
8391 info->read += size;
8392
8393 return size;
8394}
8395
8396static int tracing_buffers_flush(struct file *file, fl_owner_t id)
8397{
8398 struct ftrace_buffer_info *info = file->private_data;
8399 struct trace_iterator *iter = &info->iter;
8400
8401 iter->wait_index++;
8402 /* Make sure the waiters see the new wait_index */
8403 smp_wmb();
8404
8405 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8406
8407 return 0;
8408}
8409
8410static int tracing_buffers_release(struct inode *inode, struct file *file)
8411{
8412 struct ftrace_buffer_info *info = file->private_data;
8413 struct trace_iterator *iter = &info->iter;
8414
8415 mutex_lock(&trace_types_lock);
8416
8417 iter->tr->trace_ref--;
8418
8419 __trace_array_put(iter->tr);
8420
8421 if (info->spare)
8422 ring_buffer_free_read_page(iter->array_buffer->buffer,
8423 info->spare_cpu, info->spare);
8424 kvfree(info);
8425
8426 mutex_unlock(&trace_types_lock);
8427
8428 return 0;
8429}
8430
8431struct buffer_ref {
8432 struct trace_buffer *buffer;
8433 void *page;
8434 int cpu;
8435 refcount_t refcount;
8436};
8437
8438static void buffer_ref_release(struct buffer_ref *ref)
8439{
8440 if (!refcount_dec_and_test(&ref->refcount))
8441 return;
8442 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8443 kfree(ref);
8444}
8445
8446static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8447 struct pipe_buffer *buf)
8448{
8449 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8450
8451 buffer_ref_release(ref);
8452 buf->private = 0;
8453}
8454
8455static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8456 struct pipe_buffer *buf)
8457{
8458 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8459
8460 if (refcount_read(&ref->refcount) > INT_MAX/2)
8461 return false;
8462
8463 refcount_inc(&ref->refcount);
8464 return true;
8465}
8466
8467/* Pipe buffer operations for a buffer. */
8468static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8469 .release = buffer_pipe_buf_release,
8470 .get = buffer_pipe_buf_get,
8471};
8472
8473/*
8474 * Callback from splice_to_pipe(), if we need to release some pages
8475 * at the end of the spd in case we error'ed out in filling the pipe.
8476 */
8477static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8478{
8479 struct buffer_ref *ref =
8480 (struct buffer_ref *)spd->partial[i].private;
8481
8482 buffer_ref_release(ref);
8483 spd->partial[i].private = 0;
8484}
8485
8486static ssize_t
8487tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8488 struct pipe_inode_info *pipe, size_t len,
8489 unsigned int flags)
8490{
8491 struct ftrace_buffer_info *info = file->private_data;
8492 struct trace_iterator *iter = &info->iter;
8493 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8494 struct page *pages_def[PIPE_DEF_BUFFERS];
8495 struct splice_pipe_desc spd = {
8496 .pages = pages_def,
8497 .partial = partial_def,
8498 .nr_pages_max = PIPE_DEF_BUFFERS,
8499 .ops = &buffer_pipe_buf_ops,
8500 .spd_release = buffer_spd_release,
8501 };
8502 struct buffer_ref *ref;
8503 int page_size;
8504 int entries, i;
8505 ssize_t ret = 0;
8506
8507#ifdef CONFIG_TRACER_MAX_TRACE
8508 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8509 return -EBUSY;
8510#endif
8511
8512 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8513 if (*ppos & (page_size - 1))
8514 return -EINVAL;
8515
8516 if (len & (page_size - 1)) {
8517 if (len < page_size)
8518 return -EINVAL;
8519 len &= (~(page_size - 1));
8520 }
8521
8522 if (splice_grow_spd(pipe, &spd))
8523 return -ENOMEM;
8524
8525 again:
8526 trace_access_lock(iter->cpu_file);
8527 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8528
8529 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= page_size) {
8530 struct page *page;
8531 int r;
8532
8533 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8534 if (!ref) {
8535 ret = -ENOMEM;
8536 break;
8537 }
8538
8539 refcount_set(&ref->refcount, 1);
8540 ref->buffer = iter->array_buffer->buffer;
8541 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8542 if (IS_ERR(ref->page)) {
8543 ret = PTR_ERR(ref->page);
8544 ref->page = NULL;
8545 kfree(ref);
8546 break;
8547 }
8548 ref->cpu = iter->cpu_file;
8549
8550 r = ring_buffer_read_page(ref->buffer, ref->page,
8551 len, iter->cpu_file, 1);
8552 if (r < 0) {
8553 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8554 ref->page);
8555 kfree(ref);
8556 break;
8557 }
8558
8559 page = virt_to_page(ring_buffer_read_page_data(ref->page));
8560
8561 spd.pages[i] = page;
8562 spd.partial[i].len = page_size;
8563 spd.partial[i].offset = 0;
8564 spd.partial[i].private = (unsigned long)ref;
8565 spd.nr_pages++;
8566 *ppos += page_size;
8567
8568 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8569 }
8570
8571 trace_access_unlock(iter->cpu_file);
8572 spd.nr_pages = i;
8573
8574 /* did we read anything? */
8575 if (!spd.nr_pages) {
8576 long wait_index;
8577
8578 if (ret)
8579 goto out;
8580
8581 ret = -EAGAIN;
8582 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8583 goto out;
8584
8585 wait_index = READ_ONCE(iter->wait_index);
8586
8587 ret = wait_on_pipe(iter, iter->snapshot ? 0 : iter->tr->buffer_percent);
8588 if (ret)
8589 goto out;
8590
8591 /* No need to wait after waking up when tracing is off */
8592 if (!tracer_tracing_is_on(iter->tr))
8593 goto out;
8594
8595 /* Make sure we see the new wait_index */
8596 smp_rmb();
8597 if (wait_index != iter->wait_index)
8598 goto out;
8599
8600 goto again;
8601 }
8602
8603 ret = splice_to_pipe(pipe, &spd);
8604out:
8605 splice_shrink_spd(&spd);
8606
8607 return ret;
8608}
8609
8610/* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8611static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8612{
8613 struct ftrace_buffer_info *info = file->private_data;
8614 struct trace_iterator *iter = &info->iter;
8615
8616 if (cmd)
8617 return -ENOIOCTLCMD;
8618
8619 mutex_lock(&trace_types_lock);
8620
8621 iter->wait_index++;
8622 /* Make sure the waiters see the new wait_index */
8623 smp_wmb();
8624
8625 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8626
8627 mutex_unlock(&trace_types_lock);
8628 return 0;
8629}
8630
8631static const struct file_operations tracing_buffers_fops = {
8632 .open = tracing_buffers_open,
8633 .read = tracing_buffers_read,
8634 .poll = tracing_buffers_poll,
8635 .release = tracing_buffers_release,
8636 .flush = tracing_buffers_flush,
8637 .splice_read = tracing_buffers_splice_read,
8638 .unlocked_ioctl = tracing_buffers_ioctl,
8639 .llseek = no_llseek,
8640};
8641
8642static ssize_t
8643tracing_stats_read(struct file *filp, char __user *ubuf,
8644 size_t count, loff_t *ppos)
8645{
8646 struct inode *inode = file_inode(filp);
8647 struct trace_array *tr = inode->i_private;
8648 struct array_buffer *trace_buf = &tr->array_buffer;
8649 int cpu = tracing_get_cpu(inode);
8650 struct trace_seq *s;
8651 unsigned long cnt;
8652 unsigned long long t;
8653 unsigned long usec_rem;
8654
8655 s = kmalloc(sizeof(*s), GFP_KERNEL);
8656 if (!s)
8657 return -ENOMEM;
8658
8659 trace_seq_init(s);
8660
8661 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8662 trace_seq_printf(s, "entries: %ld\n", cnt);
8663
8664 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8665 trace_seq_printf(s, "overrun: %ld\n", cnt);
8666
8667 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8668 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8669
8670 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8671 trace_seq_printf(s, "bytes: %ld\n", cnt);
8672
8673 if (trace_clocks[tr->clock_id].in_ns) {
8674 /* local or global for trace_clock */
8675 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8676 usec_rem = do_div(t, USEC_PER_SEC);
8677 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8678 t, usec_rem);
8679
8680 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8681 usec_rem = do_div(t, USEC_PER_SEC);
8682 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8683 } else {
8684 /* counter or tsc mode for trace_clock */
8685 trace_seq_printf(s, "oldest event ts: %llu\n",
8686 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8687
8688 trace_seq_printf(s, "now ts: %llu\n",
8689 ring_buffer_time_stamp(trace_buf->buffer));
8690 }
8691
8692 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8693 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8694
8695 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8696 trace_seq_printf(s, "read events: %ld\n", cnt);
8697
8698 count = simple_read_from_buffer(ubuf, count, ppos,
8699 s->buffer, trace_seq_used(s));
8700
8701 kfree(s);
8702
8703 return count;
8704}
8705
8706static const struct file_operations tracing_stats_fops = {
8707 .open = tracing_open_generic_tr,
8708 .read = tracing_stats_read,
8709 .llseek = generic_file_llseek,
8710 .release = tracing_release_generic_tr,
8711};
8712
8713#ifdef CONFIG_DYNAMIC_FTRACE
8714
8715static ssize_t
8716tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8717 size_t cnt, loff_t *ppos)
8718{
8719 ssize_t ret;
8720 char *buf;
8721 int r;
8722
8723 /* 256 should be plenty to hold the amount needed */
8724 buf = kmalloc(256, GFP_KERNEL);
8725 if (!buf)
8726 return -ENOMEM;
8727
8728 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8729 ftrace_update_tot_cnt,
8730 ftrace_number_of_pages,
8731 ftrace_number_of_groups);
8732
8733 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8734 kfree(buf);
8735 return ret;
8736}
8737
8738static const struct file_operations tracing_dyn_info_fops = {
8739 .open = tracing_open_generic,
8740 .read = tracing_read_dyn_info,
8741 .llseek = generic_file_llseek,
8742};
8743#endif /* CONFIG_DYNAMIC_FTRACE */
8744
8745#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8746static void
8747ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8748 struct trace_array *tr, struct ftrace_probe_ops *ops,
8749 void *data)
8750{
8751 tracing_snapshot_instance(tr);
8752}
8753
8754static void
8755ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8756 struct trace_array *tr, struct ftrace_probe_ops *ops,
8757 void *data)
8758{
8759 struct ftrace_func_mapper *mapper = data;
8760 long *count = NULL;
8761
8762 if (mapper)
8763 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8764
8765 if (count) {
8766
8767 if (*count <= 0)
8768 return;
8769
8770 (*count)--;
8771 }
8772
8773 tracing_snapshot_instance(tr);
8774}
8775
8776static int
8777ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8778 struct ftrace_probe_ops *ops, void *data)
8779{
8780 struct ftrace_func_mapper *mapper = data;
8781 long *count = NULL;
8782
8783 seq_printf(m, "%ps:", (void *)ip);
8784
8785 seq_puts(m, "snapshot");
8786
8787 if (mapper)
8788 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8789
8790 if (count)
8791 seq_printf(m, ":count=%ld\n", *count);
8792 else
8793 seq_puts(m, ":unlimited\n");
8794
8795 return 0;
8796}
8797
8798static int
8799ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8800 unsigned long ip, void *init_data, void **data)
8801{
8802 struct ftrace_func_mapper *mapper = *data;
8803
8804 if (!mapper) {
8805 mapper = allocate_ftrace_func_mapper();
8806 if (!mapper)
8807 return -ENOMEM;
8808 *data = mapper;
8809 }
8810
8811 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8812}
8813
8814static void
8815ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8816 unsigned long ip, void *data)
8817{
8818 struct ftrace_func_mapper *mapper = data;
8819
8820 if (!ip) {
8821 if (!mapper)
8822 return;
8823 free_ftrace_func_mapper(mapper, NULL);
8824 return;
8825 }
8826
8827 ftrace_func_mapper_remove_ip(mapper, ip);
8828}
8829
8830static struct ftrace_probe_ops snapshot_probe_ops = {
8831 .func = ftrace_snapshot,
8832 .print = ftrace_snapshot_print,
8833};
8834
8835static struct ftrace_probe_ops snapshot_count_probe_ops = {
8836 .func = ftrace_count_snapshot,
8837 .print = ftrace_snapshot_print,
8838 .init = ftrace_snapshot_init,
8839 .free = ftrace_snapshot_free,
8840};
8841
8842static int
8843ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8844 char *glob, char *cmd, char *param, int enable)
8845{
8846 struct ftrace_probe_ops *ops;
8847 void *count = (void *)-1;
8848 char *number;
8849 int ret;
8850
8851 if (!tr)
8852 return -ENODEV;
8853
8854 /* hash funcs only work with set_ftrace_filter */
8855 if (!enable)
8856 return -EINVAL;
8857
8858 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8859
8860 if (glob[0] == '!')
8861 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8862
8863 if (!param)
8864 goto out_reg;
8865
8866 number = strsep(¶m, ":");
8867
8868 if (!strlen(number))
8869 goto out_reg;
8870
8871 /*
8872 * We use the callback data field (which is a pointer)
8873 * as our counter.
8874 */
8875 ret = kstrtoul(number, 0, (unsigned long *)&count);
8876 if (ret)
8877 return ret;
8878
8879 out_reg:
8880 ret = tracing_alloc_snapshot_instance(tr);
8881 if (ret < 0)
8882 goto out;
8883
8884 ret = register_ftrace_function_probe(glob, tr, ops, count);
8885
8886 out:
8887 return ret < 0 ? ret : 0;
8888}
8889
8890static struct ftrace_func_command ftrace_snapshot_cmd = {
8891 .name = "snapshot",
8892 .func = ftrace_trace_snapshot_callback,
8893};
8894
8895static __init int register_snapshot_cmd(void)
8896{
8897 return register_ftrace_command(&ftrace_snapshot_cmd);
8898}
8899#else
8900static inline __init int register_snapshot_cmd(void) { return 0; }
8901#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8902
8903static struct dentry *tracing_get_dentry(struct trace_array *tr)
8904{
8905 if (WARN_ON(!tr->dir))
8906 return ERR_PTR(-ENODEV);
8907
8908 /* Top directory uses NULL as the parent */
8909 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8910 return NULL;
8911
8912 /* All sub buffers have a descriptor */
8913 return tr->dir;
8914}
8915
8916static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8917{
8918 struct dentry *d_tracer;
8919
8920 if (tr->percpu_dir)
8921 return tr->percpu_dir;
8922
8923 d_tracer = tracing_get_dentry(tr);
8924 if (IS_ERR(d_tracer))
8925 return NULL;
8926
8927 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8928
8929 MEM_FAIL(!tr->percpu_dir,
8930 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8931
8932 return tr->percpu_dir;
8933}
8934
8935static struct dentry *
8936trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8937 void *data, long cpu, const struct file_operations *fops)
8938{
8939 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8940
8941 if (ret) /* See tracing_get_cpu() */
8942 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8943 return ret;
8944}
8945
8946static void
8947tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8948{
8949 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8950 struct dentry *d_cpu;
8951 char cpu_dir[30]; /* 30 characters should be more than enough */
8952
8953 if (!d_percpu)
8954 return;
8955
8956 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8957 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8958 if (!d_cpu) {
8959 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8960 return;
8961 }
8962
8963 /* per cpu trace_pipe */
8964 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8965 tr, cpu, &tracing_pipe_fops);
8966
8967 /* per cpu trace */
8968 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8969 tr, cpu, &tracing_fops);
8970
8971 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8972 tr, cpu, &tracing_buffers_fops);
8973
8974 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8975 tr, cpu, &tracing_stats_fops);
8976
8977 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8978 tr, cpu, &tracing_entries_fops);
8979
8980#ifdef CONFIG_TRACER_SNAPSHOT
8981 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8982 tr, cpu, &snapshot_fops);
8983
8984 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8985 tr, cpu, &snapshot_raw_fops);
8986#endif
8987}
8988
8989#ifdef CONFIG_FTRACE_SELFTEST
8990/* Let selftest have access to static functions in this file */
8991#include "trace_selftest.c"
8992#endif
8993
8994static ssize_t
8995trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8996 loff_t *ppos)
8997{
8998 struct trace_option_dentry *topt = filp->private_data;
8999 char *buf;
9000
9001 if (topt->flags->val & topt->opt->bit)
9002 buf = "1\n";
9003 else
9004 buf = "0\n";
9005
9006 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9007}
9008
9009static ssize_t
9010trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
9011 loff_t *ppos)
9012{
9013 struct trace_option_dentry *topt = filp->private_data;
9014 unsigned long val;
9015 int ret;
9016
9017 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9018 if (ret)
9019 return ret;
9020
9021 if (val != 0 && val != 1)
9022 return -EINVAL;
9023
9024 if (!!(topt->flags->val & topt->opt->bit) != val) {
9025 mutex_lock(&trace_types_lock);
9026 ret = __set_tracer_option(topt->tr, topt->flags,
9027 topt->opt, !val);
9028 mutex_unlock(&trace_types_lock);
9029 if (ret)
9030 return ret;
9031 }
9032
9033 *ppos += cnt;
9034
9035 return cnt;
9036}
9037
9038static int tracing_open_options(struct inode *inode, struct file *filp)
9039{
9040 struct trace_option_dentry *topt = inode->i_private;
9041 int ret;
9042
9043 ret = tracing_check_open_get_tr(topt->tr);
9044 if (ret)
9045 return ret;
9046
9047 filp->private_data = inode->i_private;
9048 return 0;
9049}
9050
9051static int tracing_release_options(struct inode *inode, struct file *file)
9052{
9053 struct trace_option_dentry *topt = file->private_data;
9054
9055 trace_array_put(topt->tr);
9056 return 0;
9057}
9058
9059static const struct file_operations trace_options_fops = {
9060 .open = tracing_open_options,
9061 .read = trace_options_read,
9062 .write = trace_options_write,
9063 .llseek = generic_file_llseek,
9064 .release = tracing_release_options,
9065};
9066
9067/*
9068 * In order to pass in both the trace_array descriptor as well as the index
9069 * to the flag that the trace option file represents, the trace_array
9070 * has a character array of trace_flags_index[], which holds the index
9071 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
9072 * The address of this character array is passed to the flag option file
9073 * read/write callbacks.
9074 *
9075 * In order to extract both the index and the trace_array descriptor,
9076 * get_tr_index() uses the following algorithm.
9077 *
9078 * idx = *ptr;
9079 *
9080 * As the pointer itself contains the address of the index (remember
9081 * index[1] == 1).
9082 *
9083 * Then to get the trace_array descriptor, by subtracting that index
9084 * from the ptr, we get to the start of the index itself.
9085 *
9086 * ptr - idx == &index[0]
9087 *
9088 * Then a simple container_of() from that pointer gets us to the
9089 * trace_array descriptor.
9090 */
9091static void get_tr_index(void *data, struct trace_array **ptr,
9092 unsigned int *pindex)
9093{
9094 *pindex = *(unsigned char *)data;
9095
9096 *ptr = container_of(data - *pindex, struct trace_array,
9097 trace_flags_index);
9098}
9099
9100static ssize_t
9101trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
9102 loff_t *ppos)
9103{
9104 void *tr_index = filp->private_data;
9105 struct trace_array *tr;
9106 unsigned int index;
9107 char *buf;
9108
9109 get_tr_index(tr_index, &tr, &index);
9110
9111 if (tr->trace_flags & (1 << index))
9112 buf = "1\n";
9113 else
9114 buf = "0\n";
9115
9116 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9117}
9118
9119static ssize_t
9120trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
9121 loff_t *ppos)
9122{
9123 void *tr_index = filp->private_data;
9124 struct trace_array *tr;
9125 unsigned int index;
9126 unsigned long val;
9127 int ret;
9128
9129 get_tr_index(tr_index, &tr, &index);
9130
9131 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9132 if (ret)
9133 return ret;
9134
9135 if (val != 0 && val != 1)
9136 return -EINVAL;
9137
9138 mutex_lock(&event_mutex);
9139 mutex_lock(&trace_types_lock);
9140 ret = set_tracer_flag(tr, 1 << index, val);
9141 mutex_unlock(&trace_types_lock);
9142 mutex_unlock(&event_mutex);
9143
9144 if (ret < 0)
9145 return ret;
9146
9147 *ppos += cnt;
9148
9149 return cnt;
9150}
9151
9152static const struct file_operations trace_options_core_fops = {
9153 .open = tracing_open_generic,
9154 .read = trace_options_core_read,
9155 .write = trace_options_core_write,
9156 .llseek = generic_file_llseek,
9157};
9158
9159struct dentry *trace_create_file(const char *name,
9160 umode_t mode,
9161 struct dentry *parent,
9162 void *data,
9163 const struct file_operations *fops)
9164{
9165 struct dentry *ret;
9166
9167 ret = tracefs_create_file(name, mode, parent, data, fops);
9168 if (!ret)
9169 pr_warn("Could not create tracefs '%s' entry\n", name);
9170
9171 return ret;
9172}
9173
9174
9175static struct dentry *trace_options_init_dentry(struct trace_array *tr)
9176{
9177 struct dentry *d_tracer;
9178
9179 if (tr->options)
9180 return tr->options;
9181
9182 d_tracer = tracing_get_dentry(tr);
9183 if (IS_ERR(d_tracer))
9184 return NULL;
9185
9186 tr->options = tracefs_create_dir("options", d_tracer);
9187 if (!tr->options) {
9188 pr_warn("Could not create tracefs directory 'options'\n");
9189 return NULL;
9190 }
9191
9192 return tr->options;
9193}
9194
9195static void
9196create_trace_option_file(struct trace_array *tr,
9197 struct trace_option_dentry *topt,
9198 struct tracer_flags *flags,
9199 struct tracer_opt *opt)
9200{
9201 struct dentry *t_options;
9202
9203 t_options = trace_options_init_dentry(tr);
9204 if (!t_options)
9205 return;
9206
9207 topt->flags = flags;
9208 topt->opt = opt;
9209 topt->tr = tr;
9210
9211 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
9212 t_options, topt, &trace_options_fops);
9213
9214}
9215
9216static void
9217create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
9218{
9219 struct trace_option_dentry *topts;
9220 struct trace_options *tr_topts;
9221 struct tracer_flags *flags;
9222 struct tracer_opt *opts;
9223 int cnt;
9224 int i;
9225
9226 if (!tracer)
9227 return;
9228
9229 flags = tracer->flags;
9230
9231 if (!flags || !flags->opts)
9232 return;
9233
9234 /*
9235 * If this is an instance, only create flags for tracers
9236 * the instance may have.
9237 */
9238 if (!trace_ok_for_array(tracer, tr))
9239 return;
9240
9241 for (i = 0; i < tr->nr_topts; i++) {
9242 /* Make sure there's no duplicate flags. */
9243 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
9244 return;
9245 }
9246
9247 opts = flags->opts;
9248
9249 for (cnt = 0; opts[cnt].name; cnt++)
9250 ;
9251
9252 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
9253 if (!topts)
9254 return;
9255
9256 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9257 GFP_KERNEL);
9258 if (!tr_topts) {
9259 kfree(topts);
9260 return;
9261 }
9262
9263 tr->topts = tr_topts;
9264 tr->topts[tr->nr_topts].tracer = tracer;
9265 tr->topts[tr->nr_topts].topts = topts;
9266 tr->nr_topts++;
9267
9268 for (cnt = 0; opts[cnt].name; cnt++) {
9269 create_trace_option_file(tr, &topts[cnt], flags,
9270 &opts[cnt]);
9271 MEM_FAIL(topts[cnt].entry == NULL,
9272 "Failed to create trace option: %s",
9273 opts[cnt].name);
9274 }
9275}
9276
9277static struct dentry *
9278create_trace_option_core_file(struct trace_array *tr,
9279 const char *option, long index)
9280{
9281 struct dentry *t_options;
9282
9283 t_options = trace_options_init_dentry(tr);
9284 if (!t_options)
9285 return NULL;
9286
9287 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9288 (void *)&tr->trace_flags_index[index],
9289 &trace_options_core_fops);
9290}
9291
9292static void create_trace_options_dir(struct trace_array *tr)
9293{
9294 struct dentry *t_options;
9295 bool top_level = tr == &global_trace;
9296 int i;
9297
9298 t_options = trace_options_init_dentry(tr);
9299 if (!t_options)
9300 return;
9301
9302 for (i = 0; trace_options[i]; i++) {
9303 if (top_level ||
9304 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9305 create_trace_option_core_file(tr, trace_options[i], i);
9306 }
9307}
9308
9309static ssize_t
9310rb_simple_read(struct file *filp, char __user *ubuf,
9311 size_t cnt, loff_t *ppos)
9312{
9313 struct trace_array *tr = filp->private_data;
9314 char buf[64];
9315 int r;
9316
9317 r = tracer_tracing_is_on(tr);
9318 r = sprintf(buf, "%d\n", r);
9319
9320 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9321}
9322
9323static ssize_t
9324rb_simple_write(struct file *filp, const char __user *ubuf,
9325 size_t cnt, loff_t *ppos)
9326{
9327 struct trace_array *tr = filp->private_data;
9328 struct trace_buffer *buffer = tr->array_buffer.buffer;
9329 unsigned long val;
9330 int ret;
9331
9332 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9333 if (ret)
9334 return ret;
9335
9336 if (buffer) {
9337 mutex_lock(&trace_types_lock);
9338 if (!!val == tracer_tracing_is_on(tr)) {
9339 val = 0; /* do nothing */
9340 } else if (val) {
9341 tracer_tracing_on(tr);
9342 if (tr->current_trace->start)
9343 tr->current_trace->start(tr);
9344 } else {
9345 tracer_tracing_off(tr);
9346 if (tr->current_trace->stop)
9347 tr->current_trace->stop(tr);
9348 /* Wake up any waiters */
9349 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9350 }
9351 mutex_unlock(&trace_types_lock);
9352 }
9353
9354 (*ppos)++;
9355
9356 return cnt;
9357}
9358
9359static const struct file_operations rb_simple_fops = {
9360 .open = tracing_open_generic_tr,
9361 .read = rb_simple_read,
9362 .write = rb_simple_write,
9363 .release = tracing_release_generic_tr,
9364 .llseek = default_llseek,
9365};
9366
9367static ssize_t
9368buffer_percent_read(struct file *filp, char __user *ubuf,
9369 size_t cnt, loff_t *ppos)
9370{
9371 struct trace_array *tr = filp->private_data;
9372 char buf[64];
9373 int r;
9374
9375 r = tr->buffer_percent;
9376 r = sprintf(buf, "%d\n", r);
9377
9378 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9379}
9380
9381static ssize_t
9382buffer_percent_write(struct file *filp, const char __user *ubuf,
9383 size_t cnt, loff_t *ppos)
9384{
9385 struct trace_array *tr = filp->private_data;
9386 unsigned long val;
9387 int ret;
9388
9389 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9390 if (ret)
9391 return ret;
9392
9393 if (val > 100)
9394 return -EINVAL;
9395
9396 tr->buffer_percent = val;
9397
9398 (*ppos)++;
9399
9400 return cnt;
9401}
9402
9403static const struct file_operations buffer_percent_fops = {
9404 .open = tracing_open_generic_tr,
9405 .read = buffer_percent_read,
9406 .write = buffer_percent_write,
9407 .release = tracing_release_generic_tr,
9408 .llseek = default_llseek,
9409};
9410
9411static ssize_t
9412buffer_subbuf_size_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
9413{
9414 struct trace_array *tr = filp->private_data;
9415 size_t size;
9416 char buf[64];
9417 int order;
9418 int r;
9419
9420 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9421 size = (PAGE_SIZE << order) / 1024;
9422
9423 r = sprintf(buf, "%zd\n", size);
9424
9425 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9426}
9427
9428static ssize_t
9429buffer_subbuf_size_write(struct file *filp, const char __user *ubuf,
9430 size_t cnt, loff_t *ppos)
9431{
9432 struct trace_array *tr = filp->private_data;
9433 unsigned long val;
9434 int old_order;
9435 int order;
9436 int pages;
9437 int ret;
9438
9439 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9440 if (ret)
9441 return ret;
9442
9443 val *= 1024; /* value passed in is in KB */
9444
9445 pages = DIV_ROUND_UP(val, PAGE_SIZE);
9446 order = fls(pages - 1);
9447
9448 /* limit between 1 and 128 system pages */
9449 if (order < 0 || order > 7)
9450 return -EINVAL;
9451
9452 /* Do not allow tracing while changing the order of the ring buffer */
9453 tracing_stop_tr(tr);
9454
9455 old_order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9456 if (old_order == order)
9457 goto out;
9458
9459 ret = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, order);
9460 if (ret)
9461 goto out;
9462
9463#ifdef CONFIG_TRACER_MAX_TRACE
9464
9465 if (!tr->allocated_snapshot)
9466 goto out_max;
9467
9468 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
9469 if (ret) {
9470 /* Put back the old order */
9471 cnt = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, old_order);
9472 if (WARN_ON_ONCE(cnt)) {
9473 /*
9474 * AARGH! We are left with different orders!
9475 * The max buffer is our "snapshot" buffer.
9476 * When a tracer needs a snapshot (one of the
9477 * latency tracers), it swaps the max buffer
9478 * with the saved snap shot. We succeeded to
9479 * update the order of the main buffer, but failed to
9480 * update the order of the max buffer. But when we tried
9481 * to reset the main buffer to the original size, we
9482 * failed there too. This is very unlikely to
9483 * happen, but if it does, warn and kill all
9484 * tracing.
9485 */
9486 tracing_disabled = 1;
9487 }
9488 goto out;
9489 }
9490 out_max:
9491#endif
9492 (*ppos)++;
9493 out:
9494 if (ret)
9495 cnt = ret;
9496 tracing_start_tr(tr);
9497 return cnt;
9498}
9499
9500static const struct file_operations buffer_subbuf_size_fops = {
9501 .open = tracing_open_generic_tr,
9502 .read = buffer_subbuf_size_read,
9503 .write = buffer_subbuf_size_write,
9504 .release = tracing_release_generic_tr,
9505 .llseek = default_llseek,
9506};
9507
9508static struct dentry *trace_instance_dir;
9509
9510static void
9511init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9512
9513static int
9514allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9515{
9516 enum ring_buffer_flags rb_flags;
9517
9518 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9519
9520 buf->tr = tr;
9521
9522 buf->buffer = ring_buffer_alloc(size, rb_flags);
9523 if (!buf->buffer)
9524 return -ENOMEM;
9525
9526 buf->data = alloc_percpu(struct trace_array_cpu);
9527 if (!buf->data) {
9528 ring_buffer_free(buf->buffer);
9529 buf->buffer = NULL;
9530 return -ENOMEM;
9531 }
9532
9533 /* Allocate the first page for all buffers */
9534 set_buffer_entries(&tr->array_buffer,
9535 ring_buffer_size(tr->array_buffer.buffer, 0));
9536
9537 return 0;
9538}
9539
9540static void free_trace_buffer(struct array_buffer *buf)
9541{
9542 if (buf->buffer) {
9543 ring_buffer_free(buf->buffer);
9544 buf->buffer = NULL;
9545 free_percpu(buf->data);
9546 buf->data = NULL;
9547 }
9548}
9549
9550static int allocate_trace_buffers(struct trace_array *tr, int size)
9551{
9552 int ret;
9553
9554 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9555 if (ret)
9556 return ret;
9557
9558#ifdef CONFIG_TRACER_MAX_TRACE
9559 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9560 allocate_snapshot ? size : 1);
9561 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9562 free_trace_buffer(&tr->array_buffer);
9563 return -ENOMEM;
9564 }
9565 tr->allocated_snapshot = allocate_snapshot;
9566
9567 allocate_snapshot = false;
9568#endif
9569
9570 return 0;
9571}
9572
9573static void free_trace_buffers(struct trace_array *tr)
9574{
9575 if (!tr)
9576 return;
9577
9578 free_trace_buffer(&tr->array_buffer);
9579
9580#ifdef CONFIG_TRACER_MAX_TRACE
9581 free_trace_buffer(&tr->max_buffer);
9582#endif
9583}
9584
9585static void init_trace_flags_index(struct trace_array *tr)
9586{
9587 int i;
9588
9589 /* Used by the trace options files */
9590 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9591 tr->trace_flags_index[i] = i;
9592}
9593
9594static void __update_tracer_options(struct trace_array *tr)
9595{
9596 struct tracer *t;
9597
9598 for (t = trace_types; t; t = t->next)
9599 add_tracer_options(tr, t);
9600}
9601
9602static void update_tracer_options(struct trace_array *tr)
9603{
9604 mutex_lock(&trace_types_lock);
9605 tracer_options_updated = true;
9606 __update_tracer_options(tr);
9607 mutex_unlock(&trace_types_lock);
9608}
9609
9610/* Must have trace_types_lock held */
9611struct trace_array *trace_array_find(const char *instance)
9612{
9613 struct trace_array *tr, *found = NULL;
9614
9615 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9616 if (tr->name && strcmp(tr->name, instance) == 0) {
9617 found = tr;
9618 break;
9619 }
9620 }
9621
9622 return found;
9623}
9624
9625struct trace_array *trace_array_find_get(const char *instance)
9626{
9627 struct trace_array *tr;
9628
9629 mutex_lock(&trace_types_lock);
9630 tr = trace_array_find(instance);
9631 if (tr)
9632 tr->ref++;
9633 mutex_unlock(&trace_types_lock);
9634
9635 return tr;
9636}
9637
9638static int trace_array_create_dir(struct trace_array *tr)
9639{
9640 int ret;
9641
9642 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9643 if (!tr->dir)
9644 return -EINVAL;
9645
9646 ret = event_trace_add_tracer(tr->dir, tr);
9647 if (ret) {
9648 tracefs_remove(tr->dir);
9649 return ret;
9650 }
9651
9652 init_tracer_tracefs(tr, tr->dir);
9653 __update_tracer_options(tr);
9654
9655 return ret;
9656}
9657
9658static struct trace_array *
9659trace_array_create_systems(const char *name, const char *systems)
9660{
9661 struct trace_array *tr;
9662 int ret;
9663
9664 ret = -ENOMEM;
9665 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9666 if (!tr)
9667 return ERR_PTR(ret);
9668
9669 tr->name = kstrdup(name, GFP_KERNEL);
9670 if (!tr->name)
9671 goto out_free_tr;
9672
9673 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9674 goto out_free_tr;
9675
9676 if (!zalloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9677 goto out_free_tr;
9678
9679 if (systems) {
9680 tr->system_names = kstrdup_const(systems, GFP_KERNEL);
9681 if (!tr->system_names)
9682 goto out_free_tr;
9683 }
9684
9685 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9686
9687 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9688
9689 raw_spin_lock_init(&tr->start_lock);
9690
9691 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9692
9693 tr->current_trace = &nop_trace;
9694
9695 INIT_LIST_HEAD(&tr->systems);
9696 INIT_LIST_HEAD(&tr->events);
9697 INIT_LIST_HEAD(&tr->hist_vars);
9698 INIT_LIST_HEAD(&tr->err_log);
9699
9700 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9701 goto out_free_tr;
9702
9703 /* The ring buffer is defaultly expanded */
9704 trace_set_ring_buffer_expanded(tr);
9705
9706 if (ftrace_allocate_ftrace_ops(tr) < 0)
9707 goto out_free_tr;
9708
9709 ftrace_init_trace_array(tr);
9710
9711 init_trace_flags_index(tr);
9712
9713 if (trace_instance_dir) {
9714 ret = trace_array_create_dir(tr);
9715 if (ret)
9716 goto out_free_tr;
9717 } else
9718 __trace_early_add_events(tr);
9719
9720 list_add(&tr->list, &ftrace_trace_arrays);
9721
9722 tr->ref++;
9723
9724 return tr;
9725
9726 out_free_tr:
9727 ftrace_free_ftrace_ops(tr);
9728 free_trace_buffers(tr);
9729 free_cpumask_var(tr->pipe_cpumask);
9730 free_cpumask_var(tr->tracing_cpumask);
9731 kfree_const(tr->system_names);
9732 kfree(tr->name);
9733 kfree(tr);
9734
9735 return ERR_PTR(ret);
9736}
9737
9738static struct trace_array *trace_array_create(const char *name)
9739{
9740 return trace_array_create_systems(name, NULL);
9741}
9742
9743static int instance_mkdir(const char *name)
9744{
9745 struct trace_array *tr;
9746 int ret;
9747
9748 mutex_lock(&event_mutex);
9749 mutex_lock(&trace_types_lock);
9750
9751 ret = -EEXIST;
9752 if (trace_array_find(name))
9753 goto out_unlock;
9754
9755 tr = trace_array_create(name);
9756
9757 ret = PTR_ERR_OR_ZERO(tr);
9758
9759out_unlock:
9760 mutex_unlock(&trace_types_lock);
9761 mutex_unlock(&event_mutex);
9762 return ret;
9763}
9764
9765/**
9766 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9767 * @name: The name of the trace array to be looked up/created.
9768 * @systems: A list of systems to create event directories for (NULL for all)
9769 *
9770 * Returns pointer to trace array with given name.
9771 * NULL, if it cannot be created.
9772 *
9773 * NOTE: This function increments the reference counter associated with the
9774 * trace array returned. This makes sure it cannot be freed while in use.
9775 * Use trace_array_put() once the trace array is no longer needed.
9776 * If the trace_array is to be freed, trace_array_destroy() needs to
9777 * be called after the trace_array_put(), or simply let user space delete
9778 * it from the tracefs instances directory. But until the
9779 * trace_array_put() is called, user space can not delete it.
9780 *
9781 */
9782struct trace_array *trace_array_get_by_name(const char *name, const char *systems)
9783{
9784 struct trace_array *tr;
9785
9786 mutex_lock(&event_mutex);
9787 mutex_lock(&trace_types_lock);
9788
9789 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9790 if (tr->name && strcmp(tr->name, name) == 0)
9791 goto out_unlock;
9792 }
9793
9794 tr = trace_array_create_systems(name, systems);
9795
9796 if (IS_ERR(tr))
9797 tr = NULL;
9798out_unlock:
9799 if (tr)
9800 tr->ref++;
9801
9802 mutex_unlock(&trace_types_lock);
9803 mutex_unlock(&event_mutex);
9804 return tr;
9805}
9806EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9807
9808static int __remove_instance(struct trace_array *tr)
9809{
9810 int i;
9811
9812 /* Reference counter for a newly created trace array = 1. */
9813 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9814 return -EBUSY;
9815
9816 list_del(&tr->list);
9817
9818 /* Disable all the flags that were enabled coming in */
9819 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9820 if ((1 << i) & ZEROED_TRACE_FLAGS)
9821 set_tracer_flag(tr, 1 << i, 0);
9822 }
9823
9824 tracing_set_nop(tr);
9825 clear_ftrace_function_probes(tr);
9826 event_trace_del_tracer(tr);
9827 ftrace_clear_pids(tr);
9828 ftrace_destroy_function_files(tr);
9829 tracefs_remove(tr->dir);
9830 free_percpu(tr->last_func_repeats);
9831 free_trace_buffers(tr);
9832 clear_tracing_err_log(tr);
9833
9834 for (i = 0; i < tr->nr_topts; i++) {
9835 kfree(tr->topts[i].topts);
9836 }
9837 kfree(tr->topts);
9838
9839 free_cpumask_var(tr->pipe_cpumask);
9840 free_cpumask_var(tr->tracing_cpumask);
9841 kfree_const(tr->system_names);
9842 kfree(tr->name);
9843 kfree(tr);
9844
9845 return 0;
9846}
9847
9848int trace_array_destroy(struct trace_array *this_tr)
9849{
9850 struct trace_array *tr;
9851 int ret;
9852
9853 if (!this_tr)
9854 return -EINVAL;
9855
9856 mutex_lock(&event_mutex);
9857 mutex_lock(&trace_types_lock);
9858
9859 ret = -ENODEV;
9860
9861 /* Making sure trace array exists before destroying it. */
9862 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9863 if (tr == this_tr) {
9864 ret = __remove_instance(tr);
9865 break;
9866 }
9867 }
9868
9869 mutex_unlock(&trace_types_lock);
9870 mutex_unlock(&event_mutex);
9871
9872 return ret;
9873}
9874EXPORT_SYMBOL_GPL(trace_array_destroy);
9875
9876static int instance_rmdir(const char *name)
9877{
9878 struct trace_array *tr;
9879 int ret;
9880
9881 mutex_lock(&event_mutex);
9882 mutex_lock(&trace_types_lock);
9883
9884 ret = -ENODEV;
9885 tr = trace_array_find(name);
9886 if (tr)
9887 ret = __remove_instance(tr);
9888
9889 mutex_unlock(&trace_types_lock);
9890 mutex_unlock(&event_mutex);
9891
9892 return ret;
9893}
9894
9895static __init void create_trace_instances(struct dentry *d_tracer)
9896{
9897 struct trace_array *tr;
9898
9899 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9900 instance_mkdir,
9901 instance_rmdir);
9902 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9903 return;
9904
9905 mutex_lock(&event_mutex);
9906 mutex_lock(&trace_types_lock);
9907
9908 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9909 if (!tr->name)
9910 continue;
9911 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9912 "Failed to create instance directory\n"))
9913 break;
9914 }
9915
9916 mutex_unlock(&trace_types_lock);
9917 mutex_unlock(&event_mutex);
9918}
9919
9920static void
9921init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9922{
9923 int cpu;
9924
9925 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9926 tr, &show_traces_fops);
9927
9928 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9929 tr, &set_tracer_fops);
9930
9931 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9932 tr, &tracing_cpumask_fops);
9933
9934 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9935 tr, &tracing_iter_fops);
9936
9937 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9938 tr, &tracing_fops);
9939
9940 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9941 tr, &tracing_pipe_fops);
9942
9943 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9944 tr, &tracing_entries_fops);
9945
9946 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9947 tr, &tracing_total_entries_fops);
9948
9949 trace_create_file("free_buffer", 0200, d_tracer,
9950 tr, &tracing_free_buffer_fops);
9951
9952 trace_create_file("trace_marker", 0220, d_tracer,
9953 tr, &tracing_mark_fops);
9954
9955 tr->trace_marker_file = __find_event_file(tr, "ftrace", "print");
9956
9957 trace_create_file("trace_marker_raw", 0220, d_tracer,
9958 tr, &tracing_mark_raw_fops);
9959
9960 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9961 &trace_clock_fops);
9962
9963 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9964 tr, &rb_simple_fops);
9965
9966 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9967 &trace_time_stamp_mode_fops);
9968
9969 tr->buffer_percent = 50;
9970
9971 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9972 tr, &buffer_percent_fops);
9973
9974 trace_create_file("buffer_subbuf_size_kb", TRACE_MODE_WRITE, d_tracer,
9975 tr, &buffer_subbuf_size_fops);
9976
9977 create_trace_options_dir(tr);
9978
9979#ifdef CONFIG_TRACER_MAX_TRACE
9980 trace_create_maxlat_file(tr, d_tracer);
9981#endif
9982
9983 if (ftrace_create_function_files(tr, d_tracer))
9984 MEM_FAIL(1, "Could not allocate function filter files");
9985
9986#ifdef CONFIG_TRACER_SNAPSHOT
9987 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9988 tr, &snapshot_fops);
9989#endif
9990
9991 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9992 tr, &tracing_err_log_fops);
9993
9994 for_each_tracing_cpu(cpu)
9995 tracing_init_tracefs_percpu(tr, cpu);
9996
9997 ftrace_init_tracefs(tr, d_tracer);
9998}
9999
10000static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
10001{
10002 struct vfsmount *mnt;
10003 struct file_system_type *type;
10004
10005 /*
10006 * To maintain backward compatibility for tools that mount
10007 * debugfs to get to the tracing facility, tracefs is automatically
10008 * mounted to the debugfs/tracing directory.
10009 */
10010 type = get_fs_type("tracefs");
10011 if (!type)
10012 return NULL;
10013 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
10014 put_filesystem(type);
10015 if (IS_ERR(mnt))
10016 return NULL;
10017 mntget(mnt);
10018
10019 return mnt;
10020}
10021
10022/**
10023 * tracing_init_dentry - initialize top level trace array
10024 *
10025 * This is called when creating files or directories in the tracing
10026 * directory. It is called via fs_initcall() by any of the boot up code
10027 * and expects to return the dentry of the top level tracing directory.
10028 */
10029int tracing_init_dentry(void)
10030{
10031 struct trace_array *tr = &global_trace;
10032
10033 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10034 pr_warn("Tracing disabled due to lockdown\n");
10035 return -EPERM;
10036 }
10037
10038 /* The top level trace array uses NULL as parent */
10039 if (tr->dir)
10040 return 0;
10041
10042 if (WARN_ON(!tracefs_initialized()))
10043 return -ENODEV;
10044
10045 /*
10046 * As there may still be users that expect the tracing
10047 * files to exist in debugfs/tracing, we must automount
10048 * the tracefs file system there, so older tools still
10049 * work with the newer kernel.
10050 */
10051 tr->dir = debugfs_create_automount("tracing", NULL,
10052 trace_automount, NULL);
10053
10054 return 0;
10055}
10056
10057extern struct trace_eval_map *__start_ftrace_eval_maps[];
10058extern struct trace_eval_map *__stop_ftrace_eval_maps[];
10059
10060static struct workqueue_struct *eval_map_wq __initdata;
10061static struct work_struct eval_map_work __initdata;
10062static struct work_struct tracerfs_init_work __initdata;
10063
10064static void __init eval_map_work_func(struct work_struct *work)
10065{
10066 int len;
10067
10068 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
10069 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
10070}
10071
10072static int __init trace_eval_init(void)
10073{
10074 INIT_WORK(&eval_map_work, eval_map_work_func);
10075
10076 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
10077 if (!eval_map_wq) {
10078 pr_err("Unable to allocate eval_map_wq\n");
10079 /* Do work here */
10080 eval_map_work_func(&eval_map_work);
10081 return -ENOMEM;
10082 }
10083
10084 queue_work(eval_map_wq, &eval_map_work);
10085 return 0;
10086}
10087
10088subsys_initcall(trace_eval_init);
10089
10090static int __init trace_eval_sync(void)
10091{
10092 /* Make sure the eval map updates are finished */
10093 if (eval_map_wq)
10094 destroy_workqueue(eval_map_wq);
10095 return 0;
10096}
10097
10098late_initcall_sync(trace_eval_sync);
10099
10100
10101#ifdef CONFIG_MODULES
10102static void trace_module_add_evals(struct module *mod)
10103{
10104 if (!mod->num_trace_evals)
10105 return;
10106
10107 /*
10108 * Modules with bad taint do not have events created, do
10109 * not bother with enums either.
10110 */
10111 if (trace_module_has_bad_taint(mod))
10112 return;
10113
10114 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
10115}
10116
10117#ifdef CONFIG_TRACE_EVAL_MAP_FILE
10118static void trace_module_remove_evals(struct module *mod)
10119{
10120 union trace_eval_map_item *map;
10121 union trace_eval_map_item **last = &trace_eval_maps;
10122
10123 if (!mod->num_trace_evals)
10124 return;
10125
10126 mutex_lock(&trace_eval_mutex);
10127
10128 map = trace_eval_maps;
10129
10130 while (map) {
10131 if (map->head.mod == mod)
10132 break;
10133 map = trace_eval_jmp_to_tail(map);
10134 last = &map->tail.next;
10135 map = map->tail.next;
10136 }
10137 if (!map)
10138 goto out;
10139
10140 *last = trace_eval_jmp_to_tail(map)->tail.next;
10141 kfree(map);
10142 out:
10143 mutex_unlock(&trace_eval_mutex);
10144}
10145#else
10146static inline void trace_module_remove_evals(struct module *mod) { }
10147#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
10148
10149static int trace_module_notify(struct notifier_block *self,
10150 unsigned long val, void *data)
10151{
10152 struct module *mod = data;
10153
10154 switch (val) {
10155 case MODULE_STATE_COMING:
10156 trace_module_add_evals(mod);
10157 break;
10158 case MODULE_STATE_GOING:
10159 trace_module_remove_evals(mod);
10160 break;
10161 }
10162
10163 return NOTIFY_OK;
10164}
10165
10166static struct notifier_block trace_module_nb = {
10167 .notifier_call = trace_module_notify,
10168 .priority = 0,
10169};
10170#endif /* CONFIG_MODULES */
10171
10172static __init void tracer_init_tracefs_work_func(struct work_struct *work)
10173{
10174
10175 event_trace_init();
10176
10177 init_tracer_tracefs(&global_trace, NULL);
10178 ftrace_init_tracefs_toplevel(&global_trace, NULL);
10179
10180 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
10181 &global_trace, &tracing_thresh_fops);
10182
10183 trace_create_file("README", TRACE_MODE_READ, NULL,
10184 NULL, &tracing_readme_fops);
10185
10186 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
10187 NULL, &tracing_saved_cmdlines_fops);
10188
10189 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
10190 NULL, &tracing_saved_cmdlines_size_fops);
10191
10192 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
10193 NULL, &tracing_saved_tgids_fops);
10194
10195 trace_create_eval_file(NULL);
10196
10197#ifdef CONFIG_MODULES
10198 register_module_notifier(&trace_module_nb);
10199#endif
10200
10201#ifdef CONFIG_DYNAMIC_FTRACE
10202 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
10203 NULL, &tracing_dyn_info_fops);
10204#endif
10205
10206 create_trace_instances(NULL);
10207
10208 update_tracer_options(&global_trace);
10209}
10210
10211static __init int tracer_init_tracefs(void)
10212{
10213 int ret;
10214
10215 trace_access_lock_init();
10216
10217 ret = tracing_init_dentry();
10218 if (ret)
10219 return 0;
10220
10221 if (eval_map_wq) {
10222 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
10223 queue_work(eval_map_wq, &tracerfs_init_work);
10224 } else {
10225 tracer_init_tracefs_work_func(NULL);
10226 }
10227
10228 rv_init_interface();
10229
10230 return 0;
10231}
10232
10233fs_initcall(tracer_init_tracefs);
10234
10235static int trace_die_panic_handler(struct notifier_block *self,
10236 unsigned long ev, void *unused);
10237
10238static struct notifier_block trace_panic_notifier = {
10239 .notifier_call = trace_die_panic_handler,
10240 .priority = INT_MAX - 1,
10241};
10242
10243static struct notifier_block trace_die_notifier = {
10244 .notifier_call = trace_die_panic_handler,
10245 .priority = INT_MAX - 1,
10246};
10247
10248/*
10249 * The idea is to execute the following die/panic callback early, in order
10250 * to avoid showing irrelevant information in the trace (like other panic
10251 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
10252 * warnings get disabled (to prevent potential log flooding).
10253 */
10254static int trace_die_panic_handler(struct notifier_block *self,
10255 unsigned long ev, void *unused)
10256{
10257 if (!ftrace_dump_on_oops)
10258 return NOTIFY_DONE;
10259
10260 /* The die notifier requires DIE_OOPS to trigger */
10261 if (self == &trace_die_notifier && ev != DIE_OOPS)
10262 return NOTIFY_DONE;
10263
10264 ftrace_dump(ftrace_dump_on_oops);
10265
10266 return NOTIFY_DONE;
10267}
10268
10269/*
10270 * printk is set to max of 1024, we really don't need it that big.
10271 * Nothing should be printing 1000 characters anyway.
10272 */
10273#define TRACE_MAX_PRINT 1000
10274
10275/*
10276 * Define here KERN_TRACE so that we have one place to modify
10277 * it if we decide to change what log level the ftrace dump
10278 * should be at.
10279 */
10280#define KERN_TRACE KERN_EMERG
10281
10282void
10283trace_printk_seq(struct trace_seq *s)
10284{
10285 /* Probably should print a warning here. */
10286 if (s->seq.len >= TRACE_MAX_PRINT)
10287 s->seq.len = TRACE_MAX_PRINT;
10288
10289 /*
10290 * More paranoid code. Although the buffer size is set to
10291 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
10292 * an extra layer of protection.
10293 */
10294 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
10295 s->seq.len = s->seq.size - 1;
10296
10297 /* should be zero ended, but we are paranoid. */
10298 s->buffer[s->seq.len] = 0;
10299
10300 printk(KERN_TRACE "%s", s->buffer);
10301
10302 trace_seq_init(s);
10303}
10304
10305void trace_init_global_iter(struct trace_iterator *iter)
10306{
10307 iter->tr = &global_trace;
10308 iter->trace = iter->tr->current_trace;
10309 iter->cpu_file = RING_BUFFER_ALL_CPUS;
10310 iter->array_buffer = &global_trace.array_buffer;
10311
10312 if (iter->trace && iter->trace->open)
10313 iter->trace->open(iter);
10314
10315 /* Annotate start of buffers if we had overruns */
10316 if (ring_buffer_overruns(iter->array_buffer->buffer))
10317 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10318
10319 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10320 if (trace_clocks[iter->tr->clock_id].in_ns)
10321 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10322
10323 /* Can not use kmalloc for iter.temp and iter.fmt */
10324 iter->temp = static_temp_buf;
10325 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10326 iter->fmt = static_fmt_buf;
10327 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10328}
10329
10330void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10331{
10332 /* use static because iter can be a bit big for the stack */
10333 static struct trace_iterator iter;
10334 static atomic_t dump_running;
10335 struct trace_array *tr = &global_trace;
10336 unsigned int old_userobj;
10337 unsigned long flags;
10338 int cnt = 0, cpu;
10339
10340 /* Only allow one dump user at a time. */
10341 if (atomic_inc_return(&dump_running) != 1) {
10342 atomic_dec(&dump_running);
10343 return;
10344 }
10345
10346 /*
10347 * Always turn off tracing when we dump.
10348 * We don't need to show trace output of what happens
10349 * between multiple crashes.
10350 *
10351 * If the user does a sysrq-z, then they can re-enable
10352 * tracing with echo 1 > tracing_on.
10353 */
10354 tracing_off();
10355
10356 local_irq_save(flags);
10357
10358 /* Simulate the iterator */
10359 trace_init_global_iter(&iter);
10360
10361 for_each_tracing_cpu(cpu) {
10362 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10363 }
10364
10365 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10366
10367 /* don't look at user memory in panic mode */
10368 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10369
10370 switch (oops_dump_mode) {
10371 case DUMP_ALL:
10372 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10373 break;
10374 case DUMP_ORIG:
10375 iter.cpu_file = raw_smp_processor_id();
10376 break;
10377 case DUMP_NONE:
10378 goto out_enable;
10379 default:
10380 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10381 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10382 }
10383
10384 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10385
10386 /* Did function tracer already get disabled? */
10387 if (ftrace_is_dead()) {
10388 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10389 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10390 }
10391
10392 /*
10393 * We need to stop all tracing on all CPUS to read
10394 * the next buffer. This is a bit expensive, but is
10395 * not done often. We fill all what we can read,
10396 * and then release the locks again.
10397 */
10398
10399 while (!trace_empty(&iter)) {
10400
10401 if (!cnt)
10402 printk(KERN_TRACE "---------------------------------\n");
10403
10404 cnt++;
10405
10406 trace_iterator_reset(&iter);
10407 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10408
10409 if (trace_find_next_entry_inc(&iter) != NULL) {
10410 int ret;
10411
10412 ret = print_trace_line(&iter);
10413 if (ret != TRACE_TYPE_NO_CONSUME)
10414 trace_consume(&iter);
10415 }
10416 touch_nmi_watchdog();
10417
10418 trace_printk_seq(&iter.seq);
10419 }
10420
10421 if (!cnt)
10422 printk(KERN_TRACE " (ftrace buffer empty)\n");
10423 else
10424 printk(KERN_TRACE "---------------------------------\n");
10425
10426 out_enable:
10427 tr->trace_flags |= old_userobj;
10428
10429 for_each_tracing_cpu(cpu) {
10430 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10431 }
10432 atomic_dec(&dump_running);
10433 local_irq_restore(flags);
10434}
10435EXPORT_SYMBOL_GPL(ftrace_dump);
10436
10437#define WRITE_BUFSIZE 4096
10438
10439ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10440 size_t count, loff_t *ppos,
10441 int (*createfn)(const char *))
10442{
10443 char *kbuf, *buf, *tmp;
10444 int ret = 0;
10445 size_t done = 0;
10446 size_t size;
10447
10448 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10449 if (!kbuf)
10450 return -ENOMEM;
10451
10452 while (done < count) {
10453 size = count - done;
10454
10455 if (size >= WRITE_BUFSIZE)
10456 size = WRITE_BUFSIZE - 1;
10457
10458 if (copy_from_user(kbuf, buffer + done, size)) {
10459 ret = -EFAULT;
10460 goto out;
10461 }
10462 kbuf[size] = '\0';
10463 buf = kbuf;
10464 do {
10465 tmp = strchr(buf, '\n');
10466 if (tmp) {
10467 *tmp = '\0';
10468 size = tmp - buf + 1;
10469 } else {
10470 size = strlen(buf);
10471 if (done + size < count) {
10472 if (buf != kbuf)
10473 break;
10474 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10475 pr_warn("Line length is too long: Should be less than %d\n",
10476 WRITE_BUFSIZE - 2);
10477 ret = -EINVAL;
10478 goto out;
10479 }
10480 }
10481 done += size;
10482
10483 /* Remove comments */
10484 tmp = strchr(buf, '#');
10485
10486 if (tmp)
10487 *tmp = '\0';
10488
10489 ret = createfn(buf);
10490 if (ret)
10491 goto out;
10492 buf += size;
10493
10494 } while (done < count);
10495 }
10496 ret = done;
10497
10498out:
10499 kfree(kbuf);
10500
10501 return ret;
10502}
10503
10504#ifdef CONFIG_TRACER_MAX_TRACE
10505__init static bool tr_needs_alloc_snapshot(const char *name)
10506{
10507 char *test;
10508 int len = strlen(name);
10509 bool ret;
10510
10511 if (!boot_snapshot_index)
10512 return false;
10513
10514 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10515 boot_snapshot_info[len] == '\t')
10516 return true;
10517
10518 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10519 if (!test)
10520 return false;
10521
10522 sprintf(test, "\t%s\t", name);
10523 ret = strstr(boot_snapshot_info, test) == NULL;
10524 kfree(test);
10525 return ret;
10526}
10527
10528__init static void do_allocate_snapshot(const char *name)
10529{
10530 if (!tr_needs_alloc_snapshot(name))
10531 return;
10532
10533 /*
10534 * When allocate_snapshot is set, the next call to
10535 * allocate_trace_buffers() (called by trace_array_get_by_name())
10536 * will allocate the snapshot buffer. That will alse clear
10537 * this flag.
10538 */
10539 allocate_snapshot = true;
10540}
10541#else
10542static inline void do_allocate_snapshot(const char *name) { }
10543#endif
10544
10545__init static void enable_instances(void)
10546{
10547 struct trace_array *tr;
10548 char *curr_str;
10549 char *str;
10550 char *tok;
10551
10552 /* A tab is always appended */
10553 boot_instance_info[boot_instance_index - 1] = '\0';
10554 str = boot_instance_info;
10555
10556 while ((curr_str = strsep(&str, "\t"))) {
10557
10558 tok = strsep(&curr_str, ",");
10559
10560 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10561 do_allocate_snapshot(tok);
10562
10563 tr = trace_array_get_by_name(tok, NULL);
10564 if (!tr) {
10565 pr_warn("Failed to create instance buffer %s\n", curr_str);
10566 continue;
10567 }
10568 /* Allow user space to delete it */
10569 trace_array_put(tr);
10570
10571 while ((tok = strsep(&curr_str, ","))) {
10572 early_enable_events(tr, tok, true);
10573 }
10574 }
10575}
10576
10577__init static int tracer_alloc_buffers(void)
10578{
10579 int ring_buf_size;
10580 int ret = -ENOMEM;
10581
10582
10583 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10584 pr_warn("Tracing disabled due to lockdown\n");
10585 return -EPERM;
10586 }
10587
10588 /*
10589 * Make sure we don't accidentally add more trace options
10590 * than we have bits for.
10591 */
10592 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10593
10594 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10595 goto out;
10596
10597 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10598 goto out_free_buffer_mask;
10599
10600 /* Only allocate trace_printk buffers if a trace_printk exists */
10601 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10602 /* Must be called before global_trace.buffer is allocated */
10603 trace_printk_init_buffers();
10604
10605 /* To save memory, keep the ring buffer size to its minimum */
10606 if (global_trace.ring_buffer_expanded)
10607 ring_buf_size = trace_buf_size;
10608 else
10609 ring_buf_size = 1;
10610
10611 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10612 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10613
10614 raw_spin_lock_init(&global_trace.start_lock);
10615
10616 /*
10617 * The prepare callbacks allocates some memory for the ring buffer. We
10618 * don't free the buffer if the CPU goes down. If we were to free
10619 * the buffer, then the user would lose any trace that was in the
10620 * buffer. The memory will be removed once the "instance" is removed.
10621 */
10622 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10623 "trace/RB:prepare", trace_rb_cpu_prepare,
10624 NULL);
10625 if (ret < 0)
10626 goto out_free_cpumask;
10627 /* Used for event triggers */
10628 ret = -ENOMEM;
10629 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10630 if (!temp_buffer)
10631 goto out_rm_hp_state;
10632
10633 if (trace_create_savedcmd() < 0)
10634 goto out_free_temp_buffer;
10635
10636 if (!zalloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10637 goto out_free_savedcmd;
10638
10639 /* TODO: make the number of buffers hot pluggable with CPUS */
10640 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10641 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10642 goto out_free_pipe_cpumask;
10643 }
10644 if (global_trace.buffer_disabled)
10645 tracing_off();
10646
10647 if (trace_boot_clock) {
10648 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10649 if (ret < 0)
10650 pr_warn("Trace clock %s not defined, going back to default\n",
10651 trace_boot_clock);
10652 }
10653
10654 /*
10655 * register_tracer() might reference current_trace, so it
10656 * needs to be set before we register anything. This is
10657 * just a bootstrap of current_trace anyway.
10658 */
10659 global_trace.current_trace = &nop_trace;
10660
10661 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10662
10663 ftrace_init_global_array_ops(&global_trace);
10664
10665 init_trace_flags_index(&global_trace);
10666
10667 register_tracer(&nop_trace);
10668
10669 /* Function tracing may start here (via kernel command line) */
10670 init_function_trace();
10671
10672 /* All seems OK, enable tracing */
10673 tracing_disabled = 0;
10674
10675 atomic_notifier_chain_register(&panic_notifier_list,
10676 &trace_panic_notifier);
10677
10678 register_die_notifier(&trace_die_notifier);
10679
10680 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10681
10682 INIT_LIST_HEAD(&global_trace.systems);
10683 INIT_LIST_HEAD(&global_trace.events);
10684 INIT_LIST_HEAD(&global_trace.hist_vars);
10685 INIT_LIST_HEAD(&global_trace.err_log);
10686 list_add(&global_trace.list, &ftrace_trace_arrays);
10687
10688 apply_trace_boot_options();
10689
10690 register_snapshot_cmd();
10691
10692 test_can_verify();
10693
10694 return 0;
10695
10696out_free_pipe_cpumask:
10697 free_cpumask_var(global_trace.pipe_cpumask);
10698out_free_savedcmd:
10699 free_saved_cmdlines_buffer(savedcmd);
10700out_free_temp_buffer:
10701 ring_buffer_free(temp_buffer);
10702out_rm_hp_state:
10703 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10704out_free_cpumask:
10705 free_cpumask_var(global_trace.tracing_cpumask);
10706out_free_buffer_mask:
10707 free_cpumask_var(tracing_buffer_mask);
10708out:
10709 return ret;
10710}
10711
10712void __init ftrace_boot_snapshot(void)
10713{
10714#ifdef CONFIG_TRACER_MAX_TRACE
10715 struct trace_array *tr;
10716
10717 if (!snapshot_at_boot)
10718 return;
10719
10720 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10721 if (!tr->allocated_snapshot)
10722 continue;
10723
10724 tracing_snapshot_instance(tr);
10725 trace_array_puts(tr, "** Boot snapshot taken **\n");
10726 }
10727#endif
10728}
10729
10730void __init early_trace_init(void)
10731{
10732 if (tracepoint_printk) {
10733 tracepoint_print_iter =
10734 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10735 if (MEM_FAIL(!tracepoint_print_iter,
10736 "Failed to allocate trace iterator\n"))
10737 tracepoint_printk = 0;
10738 else
10739 static_key_enable(&tracepoint_printk_key.key);
10740 }
10741 tracer_alloc_buffers();
10742
10743 init_events();
10744}
10745
10746void __init trace_init(void)
10747{
10748 trace_event_init();
10749
10750 if (boot_instance_index)
10751 enable_instances();
10752}
10753
10754__init static void clear_boot_tracer(void)
10755{
10756 /*
10757 * The default tracer at boot buffer is an init section.
10758 * This function is called in lateinit. If we did not
10759 * find the boot tracer, then clear it out, to prevent
10760 * later registration from accessing the buffer that is
10761 * about to be freed.
10762 */
10763 if (!default_bootup_tracer)
10764 return;
10765
10766 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10767 default_bootup_tracer);
10768 default_bootup_tracer = NULL;
10769}
10770
10771#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10772__init static void tracing_set_default_clock(void)
10773{
10774 /* sched_clock_stable() is determined in late_initcall */
10775 if (!trace_boot_clock && !sched_clock_stable()) {
10776 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10777 pr_warn("Can not set tracing clock due to lockdown\n");
10778 return;
10779 }
10780
10781 printk(KERN_WARNING
10782 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10783 "If you want to keep using the local clock, then add:\n"
10784 " \"trace_clock=local\"\n"
10785 "on the kernel command line\n");
10786 tracing_set_clock(&global_trace, "global");
10787 }
10788}
10789#else
10790static inline void tracing_set_default_clock(void) { }
10791#endif
10792
10793__init static int late_trace_init(void)
10794{
10795 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10796 static_key_disable(&tracepoint_printk_key.key);
10797 tracepoint_printk = 0;
10798 }
10799
10800 tracing_set_default_clock();
10801 clear_boot_tracer();
10802 return 0;
10803}
10804
10805late_initcall_sync(late_trace_init);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * ring buffer based function tracer
4 *
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 *
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 *
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
14 */
15#include <linux/ring_buffer.h>
16#include <generated/utsrelease.h>
17#include <linux/stacktrace.h>
18#include <linux/writeback.h>
19#include <linux/kallsyms.h>
20#include <linux/security.h>
21#include <linux/seq_file.h>
22#include <linux/notifier.h>
23#include <linux/irqflags.h>
24#include <linux/debugfs.h>
25#include <linux/tracefs.h>
26#include <linux/pagemap.h>
27#include <linux/hardirq.h>
28#include <linux/linkage.h>
29#include <linux/uaccess.h>
30#include <linux/vmalloc.h>
31#include <linux/ftrace.h>
32#include <linux/module.h>
33#include <linux/percpu.h>
34#include <linux/splice.h>
35#include <linux/kdebug.h>
36#include <linux/string.h>
37#include <linux/mount.h>
38#include <linux/rwsem.h>
39#include <linux/slab.h>
40#include <linux/ctype.h>
41#include <linux/init.h>
42#include <linux/poll.h>
43#include <linux/nmi.h>
44#include <linux/fs.h>
45#include <linux/trace.h>
46#include <linux/sched/clock.h>
47#include <linux/sched/rt.h>
48#include <linux/fsnotify.h>
49#include <linux/irq_work.h>
50#include <linux/workqueue.h>
51
52#include "trace.h"
53#include "trace_output.h"
54
55/*
56 * On boot up, the ring buffer is set to the minimum size, so that
57 * we do not waste memory on systems that are not using tracing.
58 */
59bool ring_buffer_expanded;
60
61/*
62 * We need to change this state when a selftest is running.
63 * A selftest will lurk into the ring-buffer to count the
64 * entries inserted during the selftest although some concurrent
65 * insertions into the ring-buffer such as trace_printk could occurred
66 * at the same time, giving false positive or negative results.
67 */
68static bool __read_mostly tracing_selftest_running;
69
70/*
71 * If a tracer is running, we do not want to run SELFTEST.
72 */
73bool __read_mostly tracing_selftest_disabled;
74
75/* Pipe tracepoints to printk */
76struct trace_iterator *tracepoint_print_iter;
77int tracepoint_printk;
78static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
79
80/* For tracers that don't implement custom flags */
81static struct tracer_opt dummy_tracer_opt[] = {
82 { }
83};
84
85static int
86dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
87{
88 return 0;
89}
90
91/*
92 * To prevent the comm cache from being overwritten when no
93 * tracing is active, only save the comm when a trace event
94 * occurred.
95 */
96static DEFINE_PER_CPU(bool, trace_taskinfo_save);
97
98/*
99 * Kill all tracing for good (never come back).
100 * It is initialized to 1 but will turn to zero if the initialization
101 * of the tracer is successful. But that is the only place that sets
102 * this back to zero.
103 */
104static int tracing_disabled = 1;
105
106cpumask_var_t __read_mostly tracing_buffer_mask;
107
108/*
109 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
110 *
111 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
112 * is set, then ftrace_dump is called. This will output the contents
113 * of the ftrace buffers to the console. This is very useful for
114 * capturing traces that lead to crashes and outputing it to a
115 * serial console.
116 *
117 * It is default off, but you can enable it with either specifying
118 * "ftrace_dump_on_oops" in the kernel command line, or setting
119 * /proc/sys/kernel/ftrace_dump_on_oops
120 * Set 1 if you want to dump buffers of all CPUs
121 * Set 2 if you want to dump the buffer of the CPU that triggered oops
122 */
123
124enum ftrace_dump_mode ftrace_dump_on_oops;
125
126/* When set, tracing will stop when a WARN*() is hit */
127int __disable_trace_on_warning;
128
129#ifdef CONFIG_TRACE_EVAL_MAP_FILE
130/* Map of enums to their values, for "eval_map" file */
131struct trace_eval_map_head {
132 struct module *mod;
133 unsigned long length;
134};
135
136union trace_eval_map_item;
137
138struct trace_eval_map_tail {
139 /*
140 * "end" is first and points to NULL as it must be different
141 * than "mod" or "eval_string"
142 */
143 union trace_eval_map_item *next;
144 const char *end; /* points to NULL */
145};
146
147static DEFINE_MUTEX(trace_eval_mutex);
148
149/*
150 * The trace_eval_maps are saved in an array with two extra elements,
151 * one at the beginning, and one at the end. The beginning item contains
152 * the count of the saved maps (head.length), and the module they
153 * belong to if not built in (head.mod). The ending item contains a
154 * pointer to the next array of saved eval_map items.
155 */
156union trace_eval_map_item {
157 struct trace_eval_map map;
158 struct trace_eval_map_head head;
159 struct trace_eval_map_tail tail;
160};
161
162static union trace_eval_map_item *trace_eval_maps;
163#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
164
165int tracing_set_tracer(struct trace_array *tr, const char *buf);
166static void ftrace_trace_userstack(struct trace_buffer *buffer,
167 unsigned long flags, int pc);
168
169#define MAX_TRACER_SIZE 100
170static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
171static char *default_bootup_tracer;
172
173static bool allocate_snapshot;
174
175static int __init set_cmdline_ftrace(char *str)
176{
177 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
178 default_bootup_tracer = bootup_tracer_buf;
179 /* We are using ftrace early, expand it */
180 ring_buffer_expanded = true;
181 return 1;
182}
183__setup("ftrace=", set_cmdline_ftrace);
184
185static int __init set_ftrace_dump_on_oops(char *str)
186{
187 if (*str++ != '=' || !*str) {
188 ftrace_dump_on_oops = DUMP_ALL;
189 return 1;
190 }
191
192 if (!strcmp("orig_cpu", str)) {
193 ftrace_dump_on_oops = DUMP_ORIG;
194 return 1;
195 }
196
197 return 0;
198}
199__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
200
201static int __init stop_trace_on_warning(char *str)
202{
203 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
204 __disable_trace_on_warning = 1;
205 return 1;
206}
207__setup("traceoff_on_warning", stop_trace_on_warning);
208
209static int __init boot_alloc_snapshot(char *str)
210{
211 allocate_snapshot = true;
212 /* We also need the main ring buffer expanded */
213 ring_buffer_expanded = true;
214 return 1;
215}
216__setup("alloc_snapshot", boot_alloc_snapshot);
217
218
219static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
220
221static int __init set_trace_boot_options(char *str)
222{
223 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
224 return 0;
225}
226__setup("trace_options=", set_trace_boot_options);
227
228static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
229static char *trace_boot_clock __initdata;
230
231static int __init set_trace_boot_clock(char *str)
232{
233 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
234 trace_boot_clock = trace_boot_clock_buf;
235 return 0;
236}
237__setup("trace_clock=", set_trace_boot_clock);
238
239static int __init set_tracepoint_printk(char *str)
240{
241 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
242 tracepoint_printk = 1;
243 return 1;
244}
245__setup("tp_printk", set_tracepoint_printk);
246
247unsigned long long ns2usecs(u64 nsec)
248{
249 nsec += 500;
250 do_div(nsec, 1000);
251 return nsec;
252}
253
254/* trace_flags holds trace_options default values */
255#define TRACE_DEFAULT_FLAGS \
256 (FUNCTION_DEFAULT_FLAGS | \
257 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
258 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
259 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
260 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS)
261
262/* trace_options that are only supported by global_trace */
263#define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
264 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
265
266/* trace_flags that are default zero for instances */
267#define ZEROED_TRACE_FLAGS \
268 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
269
270/*
271 * The global_trace is the descriptor that holds the top-level tracing
272 * buffers for the live tracing.
273 */
274static struct trace_array global_trace = {
275 .trace_flags = TRACE_DEFAULT_FLAGS,
276};
277
278LIST_HEAD(ftrace_trace_arrays);
279
280int trace_array_get(struct trace_array *this_tr)
281{
282 struct trace_array *tr;
283 int ret = -ENODEV;
284
285 mutex_lock(&trace_types_lock);
286 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
287 if (tr == this_tr) {
288 tr->ref++;
289 ret = 0;
290 break;
291 }
292 }
293 mutex_unlock(&trace_types_lock);
294
295 return ret;
296}
297
298static void __trace_array_put(struct trace_array *this_tr)
299{
300 WARN_ON(!this_tr->ref);
301 this_tr->ref--;
302}
303
304/**
305 * trace_array_put - Decrement the reference counter for this trace array.
306 *
307 * NOTE: Use this when we no longer need the trace array returned by
308 * trace_array_get_by_name(). This ensures the trace array can be later
309 * destroyed.
310 *
311 */
312void trace_array_put(struct trace_array *this_tr)
313{
314 if (!this_tr)
315 return;
316
317 mutex_lock(&trace_types_lock);
318 __trace_array_put(this_tr);
319 mutex_unlock(&trace_types_lock);
320}
321EXPORT_SYMBOL_GPL(trace_array_put);
322
323int tracing_check_open_get_tr(struct trace_array *tr)
324{
325 int ret;
326
327 ret = security_locked_down(LOCKDOWN_TRACEFS);
328 if (ret)
329 return ret;
330
331 if (tracing_disabled)
332 return -ENODEV;
333
334 if (tr && trace_array_get(tr) < 0)
335 return -ENODEV;
336
337 return 0;
338}
339
340int call_filter_check_discard(struct trace_event_call *call, void *rec,
341 struct trace_buffer *buffer,
342 struct ring_buffer_event *event)
343{
344 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
345 !filter_match_preds(call->filter, rec)) {
346 __trace_event_discard_commit(buffer, event);
347 return 1;
348 }
349
350 return 0;
351}
352
353void trace_free_pid_list(struct trace_pid_list *pid_list)
354{
355 vfree(pid_list->pids);
356 kfree(pid_list);
357}
358
359/**
360 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
361 * @filtered_pids: The list of pids to check
362 * @search_pid: The PID to find in @filtered_pids
363 *
364 * Returns true if @search_pid is fonud in @filtered_pids, and false otherwis.
365 */
366bool
367trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
368{
369 /*
370 * If pid_max changed after filtered_pids was created, we
371 * by default ignore all pids greater than the previous pid_max.
372 */
373 if (search_pid >= filtered_pids->pid_max)
374 return false;
375
376 return test_bit(search_pid, filtered_pids->pids);
377}
378
379/**
380 * trace_ignore_this_task - should a task be ignored for tracing
381 * @filtered_pids: The list of pids to check
382 * @task: The task that should be ignored if not filtered
383 *
384 * Checks if @task should be traced or not from @filtered_pids.
385 * Returns true if @task should *NOT* be traced.
386 * Returns false if @task should be traced.
387 */
388bool
389trace_ignore_this_task(struct trace_pid_list *filtered_pids,
390 struct trace_pid_list *filtered_no_pids,
391 struct task_struct *task)
392{
393 /*
394 * If filterd_no_pids is not empty, and the task's pid is listed
395 * in filtered_no_pids, then return true.
396 * Otherwise, if filtered_pids is empty, that means we can
397 * trace all tasks. If it has content, then only trace pids
398 * within filtered_pids.
399 */
400
401 return (filtered_pids &&
402 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
403 (filtered_no_pids &&
404 trace_find_filtered_pid(filtered_no_pids, task->pid));
405}
406
407/**
408 * trace_filter_add_remove_task - Add or remove a task from a pid_list
409 * @pid_list: The list to modify
410 * @self: The current task for fork or NULL for exit
411 * @task: The task to add or remove
412 *
413 * If adding a task, if @self is defined, the task is only added if @self
414 * is also included in @pid_list. This happens on fork and tasks should
415 * only be added when the parent is listed. If @self is NULL, then the
416 * @task pid will be removed from the list, which would happen on exit
417 * of a task.
418 */
419void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
420 struct task_struct *self,
421 struct task_struct *task)
422{
423 if (!pid_list)
424 return;
425
426 /* For forks, we only add if the forking task is listed */
427 if (self) {
428 if (!trace_find_filtered_pid(pid_list, self->pid))
429 return;
430 }
431
432 /* Sorry, but we don't support pid_max changing after setting */
433 if (task->pid >= pid_list->pid_max)
434 return;
435
436 /* "self" is set for forks, and NULL for exits */
437 if (self)
438 set_bit(task->pid, pid_list->pids);
439 else
440 clear_bit(task->pid, pid_list->pids);
441}
442
443/**
444 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
445 * @pid_list: The pid list to show
446 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
447 * @pos: The position of the file
448 *
449 * This is used by the seq_file "next" operation to iterate the pids
450 * listed in a trace_pid_list structure.
451 *
452 * Returns the pid+1 as we want to display pid of zero, but NULL would
453 * stop the iteration.
454 */
455void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
456{
457 unsigned long pid = (unsigned long)v;
458
459 (*pos)++;
460
461 /* pid already is +1 of the actual prevous bit */
462 pid = find_next_bit(pid_list->pids, pid_list->pid_max, pid);
463
464 /* Return pid + 1 to allow zero to be represented */
465 if (pid < pid_list->pid_max)
466 return (void *)(pid + 1);
467
468 return NULL;
469}
470
471/**
472 * trace_pid_start - Used for seq_file to start reading pid lists
473 * @pid_list: The pid list to show
474 * @pos: The position of the file
475 *
476 * This is used by seq_file "start" operation to start the iteration
477 * of listing pids.
478 *
479 * Returns the pid+1 as we want to display pid of zero, but NULL would
480 * stop the iteration.
481 */
482void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
483{
484 unsigned long pid;
485 loff_t l = 0;
486
487 pid = find_first_bit(pid_list->pids, pid_list->pid_max);
488 if (pid >= pid_list->pid_max)
489 return NULL;
490
491 /* Return pid + 1 so that zero can be the exit value */
492 for (pid++; pid && l < *pos;
493 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
494 ;
495 return (void *)pid;
496}
497
498/**
499 * trace_pid_show - show the current pid in seq_file processing
500 * @m: The seq_file structure to write into
501 * @v: A void pointer of the pid (+1) value to display
502 *
503 * Can be directly used by seq_file operations to display the current
504 * pid value.
505 */
506int trace_pid_show(struct seq_file *m, void *v)
507{
508 unsigned long pid = (unsigned long)v - 1;
509
510 seq_printf(m, "%lu\n", pid);
511 return 0;
512}
513
514/* 128 should be much more than enough */
515#define PID_BUF_SIZE 127
516
517int trace_pid_write(struct trace_pid_list *filtered_pids,
518 struct trace_pid_list **new_pid_list,
519 const char __user *ubuf, size_t cnt)
520{
521 struct trace_pid_list *pid_list;
522 struct trace_parser parser;
523 unsigned long val;
524 int nr_pids = 0;
525 ssize_t read = 0;
526 ssize_t ret = 0;
527 loff_t pos;
528 pid_t pid;
529
530 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
531 return -ENOMEM;
532
533 /*
534 * Always recreate a new array. The write is an all or nothing
535 * operation. Always create a new array when adding new pids by
536 * the user. If the operation fails, then the current list is
537 * not modified.
538 */
539 pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
540 if (!pid_list) {
541 trace_parser_put(&parser);
542 return -ENOMEM;
543 }
544
545 pid_list->pid_max = READ_ONCE(pid_max);
546
547 /* Only truncating will shrink pid_max */
548 if (filtered_pids && filtered_pids->pid_max > pid_list->pid_max)
549 pid_list->pid_max = filtered_pids->pid_max;
550
551 pid_list->pids = vzalloc((pid_list->pid_max + 7) >> 3);
552 if (!pid_list->pids) {
553 trace_parser_put(&parser);
554 kfree(pid_list);
555 return -ENOMEM;
556 }
557
558 if (filtered_pids) {
559 /* copy the current bits to the new max */
560 for_each_set_bit(pid, filtered_pids->pids,
561 filtered_pids->pid_max) {
562 set_bit(pid, pid_list->pids);
563 nr_pids++;
564 }
565 }
566
567 while (cnt > 0) {
568
569 pos = 0;
570
571 ret = trace_get_user(&parser, ubuf, cnt, &pos);
572 if (ret < 0 || !trace_parser_loaded(&parser))
573 break;
574
575 read += ret;
576 ubuf += ret;
577 cnt -= ret;
578
579 ret = -EINVAL;
580 if (kstrtoul(parser.buffer, 0, &val))
581 break;
582 if (val >= pid_list->pid_max)
583 break;
584
585 pid = (pid_t)val;
586
587 set_bit(pid, pid_list->pids);
588 nr_pids++;
589
590 trace_parser_clear(&parser);
591 ret = 0;
592 }
593 trace_parser_put(&parser);
594
595 if (ret < 0) {
596 trace_free_pid_list(pid_list);
597 return ret;
598 }
599
600 if (!nr_pids) {
601 /* Cleared the list of pids */
602 trace_free_pid_list(pid_list);
603 read = ret;
604 pid_list = NULL;
605 }
606
607 *new_pid_list = pid_list;
608
609 return read;
610}
611
612static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
613{
614 u64 ts;
615
616 /* Early boot up does not have a buffer yet */
617 if (!buf->buffer)
618 return trace_clock_local();
619
620 ts = ring_buffer_time_stamp(buf->buffer, cpu);
621 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
622
623 return ts;
624}
625
626u64 ftrace_now(int cpu)
627{
628 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
629}
630
631/**
632 * tracing_is_enabled - Show if global_trace has been disabled
633 *
634 * Shows if the global trace has been enabled or not. It uses the
635 * mirror flag "buffer_disabled" to be used in fast paths such as for
636 * the irqsoff tracer. But it may be inaccurate due to races. If you
637 * need to know the accurate state, use tracing_is_on() which is a little
638 * slower, but accurate.
639 */
640int tracing_is_enabled(void)
641{
642 /*
643 * For quick access (irqsoff uses this in fast path), just
644 * return the mirror variable of the state of the ring buffer.
645 * It's a little racy, but we don't really care.
646 */
647 smp_rmb();
648 return !global_trace.buffer_disabled;
649}
650
651/*
652 * trace_buf_size is the size in bytes that is allocated
653 * for a buffer. Note, the number of bytes is always rounded
654 * to page size.
655 *
656 * This number is purposely set to a low number of 16384.
657 * If the dump on oops happens, it will be much appreciated
658 * to not have to wait for all that output. Anyway this can be
659 * boot time and run time configurable.
660 */
661#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
662
663static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
664
665/* trace_types holds a link list of available tracers. */
666static struct tracer *trace_types __read_mostly;
667
668/*
669 * trace_types_lock is used to protect the trace_types list.
670 */
671DEFINE_MUTEX(trace_types_lock);
672
673/*
674 * serialize the access of the ring buffer
675 *
676 * ring buffer serializes readers, but it is low level protection.
677 * The validity of the events (which returns by ring_buffer_peek() ..etc)
678 * are not protected by ring buffer.
679 *
680 * The content of events may become garbage if we allow other process consumes
681 * these events concurrently:
682 * A) the page of the consumed events may become a normal page
683 * (not reader page) in ring buffer, and this page will be rewrited
684 * by events producer.
685 * B) The page of the consumed events may become a page for splice_read,
686 * and this page will be returned to system.
687 *
688 * These primitives allow multi process access to different cpu ring buffer
689 * concurrently.
690 *
691 * These primitives don't distinguish read-only and read-consume access.
692 * Multi read-only access are also serialized.
693 */
694
695#ifdef CONFIG_SMP
696static DECLARE_RWSEM(all_cpu_access_lock);
697static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
698
699static inline void trace_access_lock(int cpu)
700{
701 if (cpu == RING_BUFFER_ALL_CPUS) {
702 /* gain it for accessing the whole ring buffer. */
703 down_write(&all_cpu_access_lock);
704 } else {
705 /* gain it for accessing a cpu ring buffer. */
706
707 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
708 down_read(&all_cpu_access_lock);
709
710 /* Secondly block other access to this @cpu ring buffer. */
711 mutex_lock(&per_cpu(cpu_access_lock, cpu));
712 }
713}
714
715static inline void trace_access_unlock(int cpu)
716{
717 if (cpu == RING_BUFFER_ALL_CPUS) {
718 up_write(&all_cpu_access_lock);
719 } else {
720 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
721 up_read(&all_cpu_access_lock);
722 }
723}
724
725static inline void trace_access_lock_init(void)
726{
727 int cpu;
728
729 for_each_possible_cpu(cpu)
730 mutex_init(&per_cpu(cpu_access_lock, cpu));
731}
732
733#else
734
735static DEFINE_MUTEX(access_lock);
736
737static inline void trace_access_lock(int cpu)
738{
739 (void)cpu;
740 mutex_lock(&access_lock);
741}
742
743static inline void trace_access_unlock(int cpu)
744{
745 (void)cpu;
746 mutex_unlock(&access_lock);
747}
748
749static inline void trace_access_lock_init(void)
750{
751}
752
753#endif
754
755#ifdef CONFIG_STACKTRACE
756static void __ftrace_trace_stack(struct trace_buffer *buffer,
757 unsigned long flags,
758 int skip, int pc, struct pt_regs *regs);
759static inline void ftrace_trace_stack(struct trace_array *tr,
760 struct trace_buffer *buffer,
761 unsigned long flags,
762 int skip, int pc, struct pt_regs *regs);
763
764#else
765static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
766 unsigned long flags,
767 int skip, int pc, struct pt_regs *regs)
768{
769}
770static inline void ftrace_trace_stack(struct trace_array *tr,
771 struct trace_buffer *buffer,
772 unsigned long flags,
773 int skip, int pc, struct pt_regs *regs)
774{
775}
776
777#endif
778
779static __always_inline void
780trace_event_setup(struct ring_buffer_event *event,
781 int type, unsigned long flags, int pc)
782{
783 struct trace_entry *ent = ring_buffer_event_data(event);
784
785 tracing_generic_entry_update(ent, type, flags, pc);
786}
787
788static __always_inline struct ring_buffer_event *
789__trace_buffer_lock_reserve(struct trace_buffer *buffer,
790 int type,
791 unsigned long len,
792 unsigned long flags, int pc)
793{
794 struct ring_buffer_event *event;
795
796 event = ring_buffer_lock_reserve(buffer, len);
797 if (event != NULL)
798 trace_event_setup(event, type, flags, pc);
799
800 return event;
801}
802
803void tracer_tracing_on(struct trace_array *tr)
804{
805 if (tr->array_buffer.buffer)
806 ring_buffer_record_on(tr->array_buffer.buffer);
807 /*
808 * This flag is looked at when buffers haven't been allocated
809 * yet, or by some tracers (like irqsoff), that just want to
810 * know if the ring buffer has been disabled, but it can handle
811 * races of where it gets disabled but we still do a record.
812 * As the check is in the fast path of the tracers, it is more
813 * important to be fast than accurate.
814 */
815 tr->buffer_disabled = 0;
816 /* Make the flag seen by readers */
817 smp_wmb();
818}
819
820/**
821 * tracing_on - enable tracing buffers
822 *
823 * This function enables tracing buffers that may have been
824 * disabled with tracing_off.
825 */
826void tracing_on(void)
827{
828 tracer_tracing_on(&global_trace);
829}
830EXPORT_SYMBOL_GPL(tracing_on);
831
832
833static __always_inline void
834__buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
835{
836 __this_cpu_write(trace_taskinfo_save, true);
837
838 /* If this is the temp buffer, we need to commit fully */
839 if (this_cpu_read(trace_buffered_event) == event) {
840 /* Length is in event->array[0] */
841 ring_buffer_write(buffer, event->array[0], &event->array[1]);
842 /* Release the temp buffer */
843 this_cpu_dec(trace_buffered_event_cnt);
844 } else
845 ring_buffer_unlock_commit(buffer, event);
846}
847
848/**
849 * __trace_puts - write a constant string into the trace buffer.
850 * @ip: The address of the caller
851 * @str: The constant string to write
852 * @size: The size of the string.
853 */
854int __trace_puts(unsigned long ip, const char *str, int size)
855{
856 struct ring_buffer_event *event;
857 struct trace_buffer *buffer;
858 struct print_entry *entry;
859 unsigned long irq_flags;
860 int alloc;
861 int pc;
862
863 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
864 return 0;
865
866 pc = preempt_count();
867
868 if (unlikely(tracing_selftest_running || tracing_disabled))
869 return 0;
870
871 alloc = sizeof(*entry) + size + 2; /* possible \n added */
872
873 local_save_flags(irq_flags);
874 buffer = global_trace.array_buffer.buffer;
875 ring_buffer_nest_start(buffer);
876 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
877 irq_flags, pc);
878 if (!event) {
879 size = 0;
880 goto out;
881 }
882
883 entry = ring_buffer_event_data(event);
884 entry->ip = ip;
885
886 memcpy(&entry->buf, str, size);
887
888 /* Add a newline if necessary */
889 if (entry->buf[size - 1] != '\n') {
890 entry->buf[size] = '\n';
891 entry->buf[size + 1] = '\0';
892 } else
893 entry->buf[size] = '\0';
894
895 __buffer_unlock_commit(buffer, event);
896 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
897 out:
898 ring_buffer_nest_end(buffer);
899 return size;
900}
901EXPORT_SYMBOL_GPL(__trace_puts);
902
903/**
904 * __trace_bputs - write the pointer to a constant string into trace buffer
905 * @ip: The address of the caller
906 * @str: The constant string to write to the buffer to
907 */
908int __trace_bputs(unsigned long ip, const char *str)
909{
910 struct ring_buffer_event *event;
911 struct trace_buffer *buffer;
912 struct bputs_entry *entry;
913 unsigned long irq_flags;
914 int size = sizeof(struct bputs_entry);
915 int ret = 0;
916 int pc;
917
918 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
919 return 0;
920
921 pc = preempt_count();
922
923 if (unlikely(tracing_selftest_running || tracing_disabled))
924 return 0;
925
926 local_save_flags(irq_flags);
927 buffer = global_trace.array_buffer.buffer;
928
929 ring_buffer_nest_start(buffer);
930 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
931 irq_flags, pc);
932 if (!event)
933 goto out;
934
935 entry = ring_buffer_event_data(event);
936 entry->ip = ip;
937 entry->str = str;
938
939 __buffer_unlock_commit(buffer, event);
940 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
941
942 ret = 1;
943 out:
944 ring_buffer_nest_end(buffer);
945 return ret;
946}
947EXPORT_SYMBOL_GPL(__trace_bputs);
948
949#ifdef CONFIG_TRACER_SNAPSHOT
950static void tracing_snapshot_instance_cond(struct trace_array *tr,
951 void *cond_data)
952{
953 struct tracer *tracer = tr->current_trace;
954 unsigned long flags;
955
956 if (in_nmi()) {
957 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
958 internal_trace_puts("*** snapshot is being ignored ***\n");
959 return;
960 }
961
962 if (!tr->allocated_snapshot) {
963 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
964 internal_trace_puts("*** stopping trace here! ***\n");
965 tracing_off();
966 return;
967 }
968
969 /* Note, snapshot can not be used when the tracer uses it */
970 if (tracer->use_max_tr) {
971 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
972 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
973 return;
974 }
975
976 local_irq_save(flags);
977 update_max_tr(tr, current, smp_processor_id(), cond_data);
978 local_irq_restore(flags);
979}
980
981void tracing_snapshot_instance(struct trace_array *tr)
982{
983 tracing_snapshot_instance_cond(tr, NULL);
984}
985
986/**
987 * tracing_snapshot - take a snapshot of the current buffer.
988 *
989 * This causes a swap between the snapshot buffer and the current live
990 * tracing buffer. You can use this to take snapshots of the live
991 * trace when some condition is triggered, but continue to trace.
992 *
993 * Note, make sure to allocate the snapshot with either
994 * a tracing_snapshot_alloc(), or by doing it manually
995 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
996 *
997 * If the snapshot buffer is not allocated, it will stop tracing.
998 * Basically making a permanent snapshot.
999 */
1000void tracing_snapshot(void)
1001{
1002 struct trace_array *tr = &global_trace;
1003
1004 tracing_snapshot_instance(tr);
1005}
1006EXPORT_SYMBOL_GPL(tracing_snapshot);
1007
1008/**
1009 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1010 * @tr: The tracing instance to snapshot
1011 * @cond_data: The data to be tested conditionally, and possibly saved
1012 *
1013 * This is the same as tracing_snapshot() except that the snapshot is
1014 * conditional - the snapshot will only happen if the
1015 * cond_snapshot.update() implementation receiving the cond_data
1016 * returns true, which means that the trace array's cond_snapshot
1017 * update() operation used the cond_data to determine whether the
1018 * snapshot should be taken, and if it was, presumably saved it along
1019 * with the snapshot.
1020 */
1021void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1022{
1023 tracing_snapshot_instance_cond(tr, cond_data);
1024}
1025EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1026
1027/**
1028 * tracing_snapshot_cond_data - get the user data associated with a snapshot
1029 * @tr: The tracing instance
1030 *
1031 * When the user enables a conditional snapshot using
1032 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1033 * with the snapshot. This accessor is used to retrieve it.
1034 *
1035 * Should not be called from cond_snapshot.update(), since it takes
1036 * the tr->max_lock lock, which the code calling
1037 * cond_snapshot.update() has already done.
1038 *
1039 * Returns the cond_data associated with the trace array's snapshot.
1040 */
1041void *tracing_cond_snapshot_data(struct trace_array *tr)
1042{
1043 void *cond_data = NULL;
1044
1045 arch_spin_lock(&tr->max_lock);
1046
1047 if (tr->cond_snapshot)
1048 cond_data = tr->cond_snapshot->cond_data;
1049
1050 arch_spin_unlock(&tr->max_lock);
1051
1052 return cond_data;
1053}
1054EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1055
1056static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1057 struct array_buffer *size_buf, int cpu_id);
1058static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1059
1060int tracing_alloc_snapshot_instance(struct trace_array *tr)
1061{
1062 int ret;
1063
1064 if (!tr->allocated_snapshot) {
1065
1066 /* allocate spare buffer */
1067 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1068 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1069 if (ret < 0)
1070 return ret;
1071
1072 tr->allocated_snapshot = true;
1073 }
1074
1075 return 0;
1076}
1077
1078static void free_snapshot(struct trace_array *tr)
1079{
1080 /*
1081 * We don't free the ring buffer. instead, resize it because
1082 * The max_tr ring buffer has some state (e.g. ring->clock) and
1083 * we want preserve it.
1084 */
1085 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1086 set_buffer_entries(&tr->max_buffer, 1);
1087 tracing_reset_online_cpus(&tr->max_buffer);
1088 tr->allocated_snapshot = false;
1089}
1090
1091/**
1092 * tracing_alloc_snapshot - allocate snapshot buffer.
1093 *
1094 * This only allocates the snapshot buffer if it isn't already
1095 * allocated - it doesn't also take a snapshot.
1096 *
1097 * This is meant to be used in cases where the snapshot buffer needs
1098 * to be set up for events that can't sleep but need to be able to
1099 * trigger a snapshot.
1100 */
1101int tracing_alloc_snapshot(void)
1102{
1103 struct trace_array *tr = &global_trace;
1104 int ret;
1105
1106 ret = tracing_alloc_snapshot_instance(tr);
1107 WARN_ON(ret < 0);
1108
1109 return ret;
1110}
1111EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1112
1113/**
1114 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1115 *
1116 * This is similar to tracing_snapshot(), but it will allocate the
1117 * snapshot buffer if it isn't already allocated. Use this only
1118 * where it is safe to sleep, as the allocation may sleep.
1119 *
1120 * This causes a swap between the snapshot buffer and the current live
1121 * tracing buffer. You can use this to take snapshots of the live
1122 * trace when some condition is triggered, but continue to trace.
1123 */
1124void tracing_snapshot_alloc(void)
1125{
1126 int ret;
1127
1128 ret = tracing_alloc_snapshot();
1129 if (ret < 0)
1130 return;
1131
1132 tracing_snapshot();
1133}
1134EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1135
1136/**
1137 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1138 * @tr: The tracing instance
1139 * @cond_data: User data to associate with the snapshot
1140 * @update: Implementation of the cond_snapshot update function
1141 *
1142 * Check whether the conditional snapshot for the given instance has
1143 * already been enabled, or if the current tracer is already using a
1144 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1145 * save the cond_data and update function inside.
1146 *
1147 * Returns 0 if successful, error otherwise.
1148 */
1149int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1150 cond_update_fn_t update)
1151{
1152 struct cond_snapshot *cond_snapshot;
1153 int ret = 0;
1154
1155 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1156 if (!cond_snapshot)
1157 return -ENOMEM;
1158
1159 cond_snapshot->cond_data = cond_data;
1160 cond_snapshot->update = update;
1161
1162 mutex_lock(&trace_types_lock);
1163
1164 ret = tracing_alloc_snapshot_instance(tr);
1165 if (ret)
1166 goto fail_unlock;
1167
1168 if (tr->current_trace->use_max_tr) {
1169 ret = -EBUSY;
1170 goto fail_unlock;
1171 }
1172
1173 /*
1174 * The cond_snapshot can only change to NULL without the
1175 * trace_types_lock. We don't care if we race with it going
1176 * to NULL, but we want to make sure that it's not set to
1177 * something other than NULL when we get here, which we can
1178 * do safely with only holding the trace_types_lock and not
1179 * having to take the max_lock.
1180 */
1181 if (tr->cond_snapshot) {
1182 ret = -EBUSY;
1183 goto fail_unlock;
1184 }
1185
1186 arch_spin_lock(&tr->max_lock);
1187 tr->cond_snapshot = cond_snapshot;
1188 arch_spin_unlock(&tr->max_lock);
1189
1190 mutex_unlock(&trace_types_lock);
1191
1192 return ret;
1193
1194 fail_unlock:
1195 mutex_unlock(&trace_types_lock);
1196 kfree(cond_snapshot);
1197 return ret;
1198}
1199EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1200
1201/**
1202 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1203 * @tr: The tracing instance
1204 *
1205 * Check whether the conditional snapshot for the given instance is
1206 * enabled; if so, free the cond_snapshot associated with it,
1207 * otherwise return -EINVAL.
1208 *
1209 * Returns 0 if successful, error otherwise.
1210 */
1211int tracing_snapshot_cond_disable(struct trace_array *tr)
1212{
1213 int ret = 0;
1214
1215 arch_spin_lock(&tr->max_lock);
1216
1217 if (!tr->cond_snapshot)
1218 ret = -EINVAL;
1219 else {
1220 kfree(tr->cond_snapshot);
1221 tr->cond_snapshot = NULL;
1222 }
1223
1224 arch_spin_unlock(&tr->max_lock);
1225
1226 return ret;
1227}
1228EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1229#else
1230void tracing_snapshot(void)
1231{
1232 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1233}
1234EXPORT_SYMBOL_GPL(tracing_snapshot);
1235void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1236{
1237 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1238}
1239EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1240int tracing_alloc_snapshot(void)
1241{
1242 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1243 return -ENODEV;
1244}
1245EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1246void tracing_snapshot_alloc(void)
1247{
1248 /* Give warning */
1249 tracing_snapshot();
1250}
1251EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1252void *tracing_cond_snapshot_data(struct trace_array *tr)
1253{
1254 return NULL;
1255}
1256EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1257int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1258{
1259 return -ENODEV;
1260}
1261EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1262int tracing_snapshot_cond_disable(struct trace_array *tr)
1263{
1264 return false;
1265}
1266EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1267#endif /* CONFIG_TRACER_SNAPSHOT */
1268
1269void tracer_tracing_off(struct trace_array *tr)
1270{
1271 if (tr->array_buffer.buffer)
1272 ring_buffer_record_off(tr->array_buffer.buffer);
1273 /*
1274 * This flag is looked at when buffers haven't been allocated
1275 * yet, or by some tracers (like irqsoff), that just want to
1276 * know if the ring buffer has been disabled, but it can handle
1277 * races of where it gets disabled but we still do a record.
1278 * As the check is in the fast path of the tracers, it is more
1279 * important to be fast than accurate.
1280 */
1281 tr->buffer_disabled = 1;
1282 /* Make the flag seen by readers */
1283 smp_wmb();
1284}
1285
1286/**
1287 * tracing_off - turn off tracing buffers
1288 *
1289 * This function stops the tracing buffers from recording data.
1290 * It does not disable any overhead the tracers themselves may
1291 * be causing. This function simply causes all recording to
1292 * the ring buffers to fail.
1293 */
1294void tracing_off(void)
1295{
1296 tracer_tracing_off(&global_trace);
1297}
1298EXPORT_SYMBOL_GPL(tracing_off);
1299
1300void disable_trace_on_warning(void)
1301{
1302 if (__disable_trace_on_warning) {
1303 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1304 "Disabling tracing due to warning\n");
1305 tracing_off();
1306 }
1307}
1308
1309/**
1310 * tracer_tracing_is_on - show real state of ring buffer enabled
1311 * @tr : the trace array to know if ring buffer is enabled
1312 *
1313 * Shows real state of the ring buffer if it is enabled or not.
1314 */
1315bool tracer_tracing_is_on(struct trace_array *tr)
1316{
1317 if (tr->array_buffer.buffer)
1318 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1319 return !tr->buffer_disabled;
1320}
1321
1322/**
1323 * tracing_is_on - show state of ring buffers enabled
1324 */
1325int tracing_is_on(void)
1326{
1327 return tracer_tracing_is_on(&global_trace);
1328}
1329EXPORT_SYMBOL_GPL(tracing_is_on);
1330
1331static int __init set_buf_size(char *str)
1332{
1333 unsigned long buf_size;
1334
1335 if (!str)
1336 return 0;
1337 buf_size = memparse(str, &str);
1338 /* nr_entries can not be zero */
1339 if (buf_size == 0)
1340 return 0;
1341 trace_buf_size = buf_size;
1342 return 1;
1343}
1344__setup("trace_buf_size=", set_buf_size);
1345
1346static int __init set_tracing_thresh(char *str)
1347{
1348 unsigned long threshold;
1349 int ret;
1350
1351 if (!str)
1352 return 0;
1353 ret = kstrtoul(str, 0, &threshold);
1354 if (ret < 0)
1355 return 0;
1356 tracing_thresh = threshold * 1000;
1357 return 1;
1358}
1359__setup("tracing_thresh=", set_tracing_thresh);
1360
1361unsigned long nsecs_to_usecs(unsigned long nsecs)
1362{
1363 return nsecs / 1000;
1364}
1365
1366/*
1367 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1368 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1369 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1370 * of strings in the order that the evals (enum) were defined.
1371 */
1372#undef C
1373#define C(a, b) b
1374
1375/* These must match the bit postions in trace_iterator_flags */
1376static const char *trace_options[] = {
1377 TRACE_FLAGS
1378 NULL
1379};
1380
1381static struct {
1382 u64 (*func)(void);
1383 const char *name;
1384 int in_ns; /* is this clock in nanoseconds? */
1385} trace_clocks[] = {
1386 { trace_clock_local, "local", 1 },
1387 { trace_clock_global, "global", 1 },
1388 { trace_clock_counter, "counter", 0 },
1389 { trace_clock_jiffies, "uptime", 0 },
1390 { trace_clock, "perf", 1 },
1391 { ktime_get_mono_fast_ns, "mono", 1 },
1392 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1393 { ktime_get_boot_fast_ns, "boot", 1 },
1394 ARCH_TRACE_CLOCKS
1395};
1396
1397bool trace_clock_in_ns(struct trace_array *tr)
1398{
1399 if (trace_clocks[tr->clock_id].in_ns)
1400 return true;
1401
1402 return false;
1403}
1404
1405/*
1406 * trace_parser_get_init - gets the buffer for trace parser
1407 */
1408int trace_parser_get_init(struct trace_parser *parser, int size)
1409{
1410 memset(parser, 0, sizeof(*parser));
1411
1412 parser->buffer = kmalloc(size, GFP_KERNEL);
1413 if (!parser->buffer)
1414 return 1;
1415
1416 parser->size = size;
1417 return 0;
1418}
1419
1420/*
1421 * trace_parser_put - frees the buffer for trace parser
1422 */
1423void trace_parser_put(struct trace_parser *parser)
1424{
1425 kfree(parser->buffer);
1426 parser->buffer = NULL;
1427}
1428
1429/*
1430 * trace_get_user - reads the user input string separated by space
1431 * (matched by isspace(ch))
1432 *
1433 * For each string found the 'struct trace_parser' is updated,
1434 * and the function returns.
1435 *
1436 * Returns number of bytes read.
1437 *
1438 * See kernel/trace/trace.h for 'struct trace_parser' details.
1439 */
1440int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1441 size_t cnt, loff_t *ppos)
1442{
1443 char ch;
1444 size_t read = 0;
1445 ssize_t ret;
1446
1447 if (!*ppos)
1448 trace_parser_clear(parser);
1449
1450 ret = get_user(ch, ubuf++);
1451 if (ret)
1452 goto out;
1453
1454 read++;
1455 cnt--;
1456
1457 /*
1458 * The parser is not finished with the last write,
1459 * continue reading the user input without skipping spaces.
1460 */
1461 if (!parser->cont) {
1462 /* skip white space */
1463 while (cnt && isspace(ch)) {
1464 ret = get_user(ch, ubuf++);
1465 if (ret)
1466 goto out;
1467 read++;
1468 cnt--;
1469 }
1470
1471 parser->idx = 0;
1472
1473 /* only spaces were written */
1474 if (isspace(ch) || !ch) {
1475 *ppos += read;
1476 ret = read;
1477 goto out;
1478 }
1479 }
1480
1481 /* read the non-space input */
1482 while (cnt && !isspace(ch) && ch) {
1483 if (parser->idx < parser->size - 1)
1484 parser->buffer[parser->idx++] = ch;
1485 else {
1486 ret = -EINVAL;
1487 goto out;
1488 }
1489 ret = get_user(ch, ubuf++);
1490 if (ret)
1491 goto out;
1492 read++;
1493 cnt--;
1494 }
1495
1496 /* We either got finished input or we have to wait for another call. */
1497 if (isspace(ch) || !ch) {
1498 parser->buffer[parser->idx] = 0;
1499 parser->cont = false;
1500 } else if (parser->idx < parser->size - 1) {
1501 parser->cont = true;
1502 parser->buffer[parser->idx++] = ch;
1503 /* Make sure the parsed string always terminates with '\0'. */
1504 parser->buffer[parser->idx] = 0;
1505 } else {
1506 ret = -EINVAL;
1507 goto out;
1508 }
1509
1510 *ppos += read;
1511 ret = read;
1512
1513out:
1514 return ret;
1515}
1516
1517/* TODO add a seq_buf_to_buffer() */
1518static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1519{
1520 int len;
1521
1522 if (trace_seq_used(s) <= s->seq.readpos)
1523 return -EBUSY;
1524
1525 len = trace_seq_used(s) - s->seq.readpos;
1526 if (cnt > len)
1527 cnt = len;
1528 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1529
1530 s->seq.readpos += cnt;
1531 return cnt;
1532}
1533
1534unsigned long __read_mostly tracing_thresh;
1535static const struct file_operations tracing_max_lat_fops;
1536
1537#if (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1538 defined(CONFIG_FSNOTIFY)
1539
1540static struct workqueue_struct *fsnotify_wq;
1541
1542static void latency_fsnotify_workfn(struct work_struct *work)
1543{
1544 struct trace_array *tr = container_of(work, struct trace_array,
1545 fsnotify_work);
1546 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1547}
1548
1549static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1550{
1551 struct trace_array *tr = container_of(iwork, struct trace_array,
1552 fsnotify_irqwork);
1553 queue_work(fsnotify_wq, &tr->fsnotify_work);
1554}
1555
1556static void trace_create_maxlat_file(struct trace_array *tr,
1557 struct dentry *d_tracer)
1558{
1559 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1560 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1561 tr->d_max_latency = trace_create_file("tracing_max_latency", 0644,
1562 d_tracer, &tr->max_latency,
1563 &tracing_max_lat_fops);
1564}
1565
1566__init static int latency_fsnotify_init(void)
1567{
1568 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1569 WQ_UNBOUND | WQ_HIGHPRI, 0);
1570 if (!fsnotify_wq) {
1571 pr_err("Unable to allocate tr_max_lat_wq\n");
1572 return -ENOMEM;
1573 }
1574 return 0;
1575}
1576
1577late_initcall_sync(latency_fsnotify_init);
1578
1579void latency_fsnotify(struct trace_array *tr)
1580{
1581 if (!fsnotify_wq)
1582 return;
1583 /*
1584 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1585 * possible that we are called from __schedule() or do_idle(), which
1586 * could cause a deadlock.
1587 */
1588 irq_work_queue(&tr->fsnotify_irqwork);
1589}
1590
1591/*
1592 * (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1593 * defined(CONFIG_FSNOTIFY)
1594 */
1595#else
1596
1597#define trace_create_maxlat_file(tr, d_tracer) \
1598 trace_create_file("tracing_max_latency", 0644, d_tracer, \
1599 &tr->max_latency, &tracing_max_lat_fops)
1600
1601#endif
1602
1603#ifdef CONFIG_TRACER_MAX_TRACE
1604/*
1605 * Copy the new maximum trace into the separate maximum-trace
1606 * structure. (this way the maximum trace is permanently saved,
1607 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1608 */
1609static void
1610__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1611{
1612 struct array_buffer *trace_buf = &tr->array_buffer;
1613 struct array_buffer *max_buf = &tr->max_buffer;
1614 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1615 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1616
1617 max_buf->cpu = cpu;
1618 max_buf->time_start = data->preempt_timestamp;
1619
1620 max_data->saved_latency = tr->max_latency;
1621 max_data->critical_start = data->critical_start;
1622 max_data->critical_end = data->critical_end;
1623
1624 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1625 max_data->pid = tsk->pid;
1626 /*
1627 * If tsk == current, then use current_uid(), as that does not use
1628 * RCU. The irq tracer can be called out of RCU scope.
1629 */
1630 if (tsk == current)
1631 max_data->uid = current_uid();
1632 else
1633 max_data->uid = task_uid(tsk);
1634
1635 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1636 max_data->policy = tsk->policy;
1637 max_data->rt_priority = tsk->rt_priority;
1638
1639 /* record this tasks comm */
1640 tracing_record_cmdline(tsk);
1641 latency_fsnotify(tr);
1642}
1643
1644/**
1645 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1646 * @tr: tracer
1647 * @tsk: the task with the latency
1648 * @cpu: The cpu that initiated the trace.
1649 * @cond_data: User data associated with a conditional snapshot
1650 *
1651 * Flip the buffers between the @tr and the max_tr and record information
1652 * about which task was the cause of this latency.
1653 */
1654void
1655update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1656 void *cond_data)
1657{
1658 if (tr->stop_count)
1659 return;
1660
1661 WARN_ON_ONCE(!irqs_disabled());
1662
1663 if (!tr->allocated_snapshot) {
1664 /* Only the nop tracer should hit this when disabling */
1665 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1666 return;
1667 }
1668
1669 arch_spin_lock(&tr->max_lock);
1670
1671 /* Inherit the recordable setting from array_buffer */
1672 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1673 ring_buffer_record_on(tr->max_buffer.buffer);
1674 else
1675 ring_buffer_record_off(tr->max_buffer.buffer);
1676
1677#ifdef CONFIG_TRACER_SNAPSHOT
1678 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data))
1679 goto out_unlock;
1680#endif
1681 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1682
1683 __update_max_tr(tr, tsk, cpu);
1684
1685 out_unlock:
1686 arch_spin_unlock(&tr->max_lock);
1687}
1688
1689/**
1690 * update_max_tr_single - only copy one trace over, and reset the rest
1691 * @tr: tracer
1692 * @tsk: task with the latency
1693 * @cpu: the cpu of the buffer to copy.
1694 *
1695 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1696 */
1697void
1698update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1699{
1700 int ret;
1701
1702 if (tr->stop_count)
1703 return;
1704
1705 WARN_ON_ONCE(!irqs_disabled());
1706 if (!tr->allocated_snapshot) {
1707 /* Only the nop tracer should hit this when disabling */
1708 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1709 return;
1710 }
1711
1712 arch_spin_lock(&tr->max_lock);
1713
1714 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1715
1716 if (ret == -EBUSY) {
1717 /*
1718 * We failed to swap the buffer due to a commit taking
1719 * place on this CPU. We fail to record, but we reset
1720 * the max trace buffer (no one writes directly to it)
1721 * and flag that it failed.
1722 */
1723 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1724 "Failed to swap buffers due to commit in progress\n");
1725 }
1726
1727 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1728
1729 __update_max_tr(tr, tsk, cpu);
1730 arch_spin_unlock(&tr->max_lock);
1731}
1732#endif /* CONFIG_TRACER_MAX_TRACE */
1733
1734static int wait_on_pipe(struct trace_iterator *iter, int full)
1735{
1736 /* Iterators are static, they should be filled or empty */
1737 if (trace_buffer_iter(iter, iter->cpu_file))
1738 return 0;
1739
1740 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1741 full);
1742}
1743
1744#ifdef CONFIG_FTRACE_STARTUP_TEST
1745static bool selftests_can_run;
1746
1747struct trace_selftests {
1748 struct list_head list;
1749 struct tracer *type;
1750};
1751
1752static LIST_HEAD(postponed_selftests);
1753
1754static int save_selftest(struct tracer *type)
1755{
1756 struct trace_selftests *selftest;
1757
1758 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1759 if (!selftest)
1760 return -ENOMEM;
1761
1762 selftest->type = type;
1763 list_add(&selftest->list, &postponed_selftests);
1764 return 0;
1765}
1766
1767static int run_tracer_selftest(struct tracer *type)
1768{
1769 struct trace_array *tr = &global_trace;
1770 struct tracer *saved_tracer = tr->current_trace;
1771 int ret;
1772
1773 if (!type->selftest || tracing_selftest_disabled)
1774 return 0;
1775
1776 /*
1777 * If a tracer registers early in boot up (before scheduling is
1778 * initialized and such), then do not run its selftests yet.
1779 * Instead, run it a little later in the boot process.
1780 */
1781 if (!selftests_can_run)
1782 return save_selftest(type);
1783
1784 /*
1785 * Run a selftest on this tracer.
1786 * Here we reset the trace buffer, and set the current
1787 * tracer to be this tracer. The tracer can then run some
1788 * internal tracing to verify that everything is in order.
1789 * If we fail, we do not register this tracer.
1790 */
1791 tracing_reset_online_cpus(&tr->array_buffer);
1792
1793 tr->current_trace = type;
1794
1795#ifdef CONFIG_TRACER_MAX_TRACE
1796 if (type->use_max_tr) {
1797 /* If we expanded the buffers, make sure the max is expanded too */
1798 if (ring_buffer_expanded)
1799 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1800 RING_BUFFER_ALL_CPUS);
1801 tr->allocated_snapshot = true;
1802 }
1803#endif
1804
1805 /* the test is responsible for initializing and enabling */
1806 pr_info("Testing tracer %s: ", type->name);
1807 ret = type->selftest(type, tr);
1808 /* the test is responsible for resetting too */
1809 tr->current_trace = saved_tracer;
1810 if (ret) {
1811 printk(KERN_CONT "FAILED!\n");
1812 /* Add the warning after printing 'FAILED' */
1813 WARN_ON(1);
1814 return -1;
1815 }
1816 /* Only reset on passing, to avoid touching corrupted buffers */
1817 tracing_reset_online_cpus(&tr->array_buffer);
1818
1819#ifdef CONFIG_TRACER_MAX_TRACE
1820 if (type->use_max_tr) {
1821 tr->allocated_snapshot = false;
1822
1823 /* Shrink the max buffer again */
1824 if (ring_buffer_expanded)
1825 ring_buffer_resize(tr->max_buffer.buffer, 1,
1826 RING_BUFFER_ALL_CPUS);
1827 }
1828#endif
1829
1830 printk(KERN_CONT "PASSED\n");
1831 return 0;
1832}
1833
1834static __init int init_trace_selftests(void)
1835{
1836 struct trace_selftests *p, *n;
1837 struct tracer *t, **last;
1838 int ret;
1839
1840 selftests_can_run = true;
1841
1842 mutex_lock(&trace_types_lock);
1843
1844 if (list_empty(&postponed_selftests))
1845 goto out;
1846
1847 pr_info("Running postponed tracer tests:\n");
1848
1849 tracing_selftest_running = true;
1850 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
1851 /* This loop can take minutes when sanitizers are enabled, so
1852 * lets make sure we allow RCU processing.
1853 */
1854 cond_resched();
1855 ret = run_tracer_selftest(p->type);
1856 /* If the test fails, then warn and remove from available_tracers */
1857 if (ret < 0) {
1858 WARN(1, "tracer: %s failed selftest, disabling\n",
1859 p->type->name);
1860 last = &trace_types;
1861 for (t = trace_types; t; t = t->next) {
1862 if (t == p->type) {
1863 *last = t->next;
1864 break;
1865 }
1866 last = &t->next;
1867 }
1868 }
1869 list_del(&p->list);
1870 kfree(p);
1871 }
1872 tracing_selftest_running = false;
1873
1874 out:
1875 mutex_unlock(&trace_types_lock);
1876
1877 return 0;
1878}
1879core_initcall(init_trace_selftests);
1880#else
1881static inline int run_tracer_selftest(struct tracer *type)
1882{
1883 return 0;
1884}
1885#endif /* CONFIG_FTRACE_STARTUP_TEST */
1886
1887static void add_tracer_options(struct trace_array *tr, struct tracer *t);
1888
1889static void __init apply_trace_boot_options(void);
1890
1891/**
1892 * register_tracer - register a tracer with the ftrace system.
1893 * @type: the plugin for the tracer
1894 *
1895 * Register a new plugin tracer.
1896 */
1897int __init register_tracer(struct tracer *type)
1898{
1899 struct tracer *t;
1900 int ret = 0;
1901
1902 if (!type->name) {
1903 pr_info("Tracer must have a name\n");
1904 return -1;
1905 }
1906
1907 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1908 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1909 return -1;
1910 }
1911
1912 if (security_locked_down(LOCKDOWN_TRACEFS)) {
1913 pr_warn("Can not register tracer %s due to lockdown\n",
1914 type->name);
1915 return -EPERM;
1916 }
1917
1918 mutex_lock(&trace_types_lock);
1919
1920 tracing_selftest_running = true;
1921
1922 for (t = trace_types; t; t = t->next) {
1923 if (strcmp(type->name, t->name) == 0) {
1924 /* already found */
1925 pr_info("Tracer %s already registered\n",
1926 type->name);
1927 ret = -1;
1928 goto out;
1929 }
1930 }
1931
1932 if (!type->set_flag)
1933 type->set_flag = &dummy_set_flag;
1934 if (!type->flags) {
1935 /*allocate a dummy tracer_flags*/
1936 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
1937 if (!type->flags) {
1938 ret = -ENOMEM;
1939 goto out;
1940 }
1941 type->flags->val = 0;
1942 type->flags->opts = dummy_tracer_opt;
1943 } else
1944 if (!type->flags->opts)
1945 type->flags->opts = dummy_tracer_opt;
1946
1947 /* store the tracer for __set_tracer_option */
1948 type->flags->trace = type;
1949
1950 ret = run_tracer_selftest(type);
1951 if (ret < 0)
1952 goto out;
1953
1954 type->next = trace_types;
1955 trace_types = type;
1956 add_tracer_options(&global_trace, type);
1957
1958 out:
1959 tracing_selftest_running = false;
1960 mutex_unlock(&trace_types_lock);
1961
1962 if (ret || !default_bootup_tracer)
1963 goto out_unlock;
1964
1965 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1966 goto out_unlock;
1967
1968 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1969 /* Do we want this tracer to start on bootup? */
1970 tracing_set_tracer(&global_trace, type->name);
1971 default_bootup_tracer = NULL;
1972
1973 apply_trace_boot_options();
1974
1975 /* disable other selftests, since this will break it. */
1976 tracing_selftest_disabled = true;
1977#ifdef CONFIG_FTRACE_STARTUP_TEST
1978 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1979 type->name);
1980#endif
1981
1982 out_unlock:
1983 return ret;
1984}
1985
1986static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
1987{
1988 struct trace_buffer *buffer = buf->buffer;
1989
1990 if (!buffer)
1991 return;
1992
1993 ring_buffer_record_disable(buffer);
1994
1995 /* Make sure all commits have finished */
1996 synchronize_rcu();
1997 ring_buffer_reset_cpu(buffer, cpu);
1998
1999 ring_buffer_record_enable(buffer);
2000}
2001
2002void tracing_reset_online_cpus(struct array_buffer *buf)
2003{
2004 struct trace_buffer *buffer = buf->buffer;
2005
2006 if (!buffer)
2007 return;
2008
2009 ring_buffer_record_disable(buffer);
2010
2011 /* Make sure all commits have finished */
2012 synchronize_rcu();
2013
2014 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2015
2016 ring_buffer_reset_online_cpus(buffer);
2017
2018 ring_buffer_record_enable(buffer);
2019}
2020
2021/* Must have trace_types_lock held */
2022void tracing_reset_all_online_cpus(void)
2023{
2024 struct trace_array *tr;
2025
2026 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2027 if (!tr->clear_trace)
2028 continue;
2029 tr->clear_trace = false;
2030 tracing_reset_online_cpus(&tr->array_buffer);
2031#ifdef CONFIG_TRACER_MAX_TRACE
2032 tracing_reset_online_cpus(&tr->max_buffer);
2033#endif
2034 }
2035}
2036
2037static int *tgid_map;
2038
2039#define SAVED_CMDLINES_DEFAULT 128
2040#define NO_CMDLINE_MAP UINT_MAX
2041static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2042struct saved_cmdlines_buffer {
2043 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2044 unsigned *map_cmdline_to_pid;
2045 unsigned cmdline_num;
2046 int cmdline_idx;
2047 char *saved_cmdlines;
2048};
2049static struct saved_cmdlines_buffer *savedcmd;
2050
2051/* temporary disable recording */
2052static atomic_t trace_record_taskinfo_disabled __read_mostly;
2053
2054static inline char *get_saved_cmdlines(int idx)
2055{
2056 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2057}
2058
2059static inline void set_cmdline(int idx, const char *cmdline)
2060{
2061 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2062}
2063
2064static int allocate_cmdlines_buffer(unsigned int val,
2065 struct saved_cmdlines_buffer *s)
2066{
2067 s->map_cmdline_to_pid = kmalloc_array(val,
2068 sizeof(*s->map_cmdline_to_pid),
2069 GFP_KERNEL);
2070 if (!s->map_cmdline_to_pid)
2071 return -ENOMEM;
2072
2073 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2074 if (!s->saved_cmdlines) {
2075 kfree(s->map_cmdline_to_pid);
2076 return -ENOMEM;
2077 }
2078
2079 s->cmdline_idx = 0;
2080 s->cmdline_num = val;
2081 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2082 sizeof(s->map_pid_to_cmdline));
2083 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2084 val * sizeof(*s->map_cmdline_to_pid));
2085
2086 return 0;
2087}
2088
2089static int trace_create_savedcmd(void)
2090{
2091 int ret;
2092
2093 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2094 if (!savedcmd)
2095 return -ENOMEM;
2096
2097 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2098 if (ret < 0) {
2099 kfree(savedcmd);
2100 savedcmd = NULL;
2101 return -ENOMEM;
2102 }
2103
2104 return 0;
2105}
2106
2107int is_tracing_stopped(void)
2108{
2109 return global_trace.stop_count;
2110}
2111
2112/**
2113 * tracing_start - quick start of the tracer
2114 *
2115 * If tracing is enabled but was stopped by tracing_stop,
2116 * this will start the tracer back up.
2117 */
2118void tracing_start(void)
2119{
2120 struct trace_buffer *buffer;
2121 unsigned long flags;
2122
2123 if (tracing_disabled)
2124 return;
2125
2126 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2127 if (--global_trace.stop_count) {
2128 if (global_trace.stop_count < 0) {
2129 /* Someone screwed up their debugging */
2130 WARN_ON_ONCE(1);
2131 global_trace.stop_count = 0;
2132 }
2133 goto out;
2134 }
2135
2136 /* Prevent the buffers from switching */
2137 arch_spin_lock(&global_trace.max_lock);
2138
2139 buffer = global_trace.array_buffer.buffer;
2140 if (buffer)
2141 ring_buffer_record_enable(buffer);
2142
2143#ifdef CONFIG_TRACER_MAX_TRACE
2144 buffer = global_trace.max_buffer.buffer;
2145 if (buffer)
2146 ring_buffer_record_enable(buffer);
2147#endif
2148
2149 arch_spin_unlock(&global_trace.max_lock);
2150
2151 out:
2152 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2153}
2154
2155static void tracing_start_tr(struct trace_array *tr)
2156{
2157 struct trace_buffer *buffer;
2158 unsigned long flags;
2159
2160 if (tracing_disabled)
2161 return;
2162
2163 /* If global, we need to also start the max tracer */
2164 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2165 return tracing_start();
2166
2167 raw_spin_lock_irqsave(&tr->start_lock, flags);
2168
2169 if (--tr->stop_count) {
2170 if (tr->stop_count < 0) {
2171 /* Someone screwed up their debugging */
2172 WARN_ON_ONCE(1);
2173 tr->stop_count = 0;
2174 }
2175 goto out;
2176 }
2177
2178 buffer = tr->array_buffer.buffer;
2179 if (buffer)
2180 ring_buffer_record_enable(buffer);
2181
2182 out:
2183 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2184}
2185
2186/**
2187 * tracing_stop - quick stop of the tracer
2188 *
2189 * Light weight way to stop tracing. Use in conjunction with
2190 * tracing_start.
2191 */
2192void tracing_stop(void)
2193{
2194 struct trace_buffer *buffer;
2195 unsigned long flags;
2196
2197 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2198 if (global_trace.stop_count++)
2199 goto out;
2200
2201 /* Prevent the buffers from switching */
2202 arch_spin_lock(&global_trace.max_lock);
2203
2204 buffer = global_trace.array_buffer.buffer;
2205 if (buffer)
2206 ring_buffer_record_disable(buffer);
2207
2208#ifdef CONFIG_TRACER_MAX_TRACE
2209 buffer = global_trace.max_buffer.buffer;
2210 if (buffer)
2211 ring_buffer_record_disable(buffer);
2212#endif
2213
2214 arch_spin_unlock(&global_trace.max_lock);
2215
2216 out:
2217 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2218}
2219
2220static void tracing_stop_tr(struct trace_array *tr)
2221{
2222 struct trace_buffer *buffer;
2223 unsigned long flags;
2224
2225 /* If global, we need to also stop the max tracer */
2226 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2227 return tracing_stop();
2228
2229 raw_spin_lock_irqsave(&tr->start_lock, flags);
2230 if (tr->stop_count++)
2231 goto out;
2232
2233 buffer = tr->array_buffer.buffer;
2234 if (buffer)
2235 ring_buffer_record_disable(buffer);
2236
2237 out:
2238 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2239}
2240
2241static int trace_save_cmdline(struct task_struct *tsk)
2242{
2243 unsigned pid, idx;
2244
2245 /* treat recording of idle task as a success */
2246 if (!tsk->pid)
2247 return 1;
2248
2249 if (unlikely(tsk->pid > PID_MAX_DEFAULT))
2250 return 0;
2251
2252 /*
2253 * It's not the end of the world if we don't get
2254 * the lock, but we also don't want to spin
2255 * nor do we want to disable interrupts,
2256 * so if we miss here, then better luck next time.
2257 */
2258 if (!arch_spin_trylock(&trace_cmdline_lock))
2259 return 0;
2260
2261 idx = savedcmd->map_pid_to_cmdline[tsk->pid];
2262 if (idx == NO_CMDLINE_MAP) {
2263 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2264
2265 /*
2266 * Check whether the cmdline buffer at idx has a pid
2267 * mapped. We are going to overwrite that entry so we
2268 * need to clear the map_pid_to_cmdline. Otherwise we
2269 * would read the new comm for the old pid.
2270 */
2271 pid = savedcmd->map_cmdline_to_pid[idx];
2272 if (pid != NO_CMDLINE_MAP)
2273 savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
2274
2275 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2276 savedcmd->map_pid_to_cmdline[tsk->pid] = idx;
2277
2278 savedcmd->cmdline_idx = idx;
2279 }
2280
2281 set_cmdline(idx, tsk->comm);
2282
2283 arch_spin_unlock(&trace_cmdline_lock);
2284
2285 return 1;
2286}
2287
2288static void __trace_find_cmdline(int pid, char comm[])
2289{
2290 unsigned map;
2291
2292 if (!pid) {
2293 strcpy(comm, "<idle>");
2294 return;
2295 }
2296
2297 if (WARN_ON_ONCE(pid < 0)) {
2298 strcpy(comm, "<XXX>");
2299 return;
2300 }
2301
2302 if (pid > PID_MAX_DEFAULT) {
2303 strcpy(comm, "<...>");
2304 return;
2305 }
2306
2307 map = savedcmd->map_pid_to_cmdline[pid];
2308 if (map != NO_CMDLINE_MAP)
2309 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2310 else
2311 strcpy(comm, "<...>");
2312}
2313
2314void trace_find_cmdline(int pid, char comm[])
2315{
2316 preempt_disable();
2317 arch_spin_lock(&trace_cmdline_lock);
2318
2319 __trace_find_cmdline(pid, comm);
2320
2321 arch_spin_unlock(&trace_cmdline_lock);
2322 preempt_enable();
2323}
2324
2325int trace_find_tgid(int pid)
2326{
2327 if (unlikely(!tgid_map || !pid || pid > PID_MAX_DEFAULT))
2328 return 0;
2329
2330 return tgid_map[pid];
2331}
2332
2333static int trace_save_tgid(struct task_struct *tsk)
2334{
2335 /* treat recording of idle task as a success */
2336 if (!tsk->pid)
2337 return 1;
2338
2339 if (unlikely(!tgid_map || tsk->pid > PID_MAX_DEFAULT))
2340 return 0;
2341
2342 tgid_map[tsk->pid] = tsk->tgid;
2343 return 1;
2344}
2345
2346static bool tracing_record_taskinfo_skip(int flags)
2347{
2348 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2349 return true;
2350 if (atomic_read(&trace_record_taskinfo_disabled) || !tracing_is_on())
2351 return true;
2352 if (!__this_cpu_read(trace_taskinfo_save))
2353 return true;
2354 return false;
2355}
2356
2357/**
2358 * tracing_record_taskinfo - record the task info of a task
2359 *
2360 * @task: task to record
2361 * @flags: TRACE_RECORD_CMDLINE for recording comm
2362 * TRACE_RECORD_TGID for recording tgid
2363 */
2364void tracing_record_taskinfo(struct task_struct *task, int flags)
2365{
2366 bool done;
2367
2368 if (tracing_record_taskinfo_skip(flags))
2369 return;
2370
2371 /*
2372 * Record as much task information as possible. If some fail, continue
2373 * to try to record the others.
2374 */
2375 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2376 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2377
2378 /* If recording any information failed, retry again soon. */
2379 if (!done)
2380 return;
2381
2382 __this_cpu_write(trace_taskinfo_save, false);
2383}
2384
2385/**
2386 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2387 *
2388 * @prev: previous task during sched_switch
2389 * @next: next task during sched_switch
2390 * @flags: TRACE_RECORD_CMDLINE for recording comm
2391 * TRACE_RECORD_TGID for recording tgid
2392 */
2393void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2394 struct task_struct *next, int flags)
2395{
2396 bool done;
2397
2398 if (tracing_record_taskinfo_skip(flags))
2399 return;
2400
2401 /*
2402 * Record as much task information as possible. If some fail, continue
2403 * to try to record the others.
2404 */
2405 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2406 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2407 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2408 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2409
2410 /* If recording any information failed, retry again soon. */
2411 if (!done)
2412 return;
2413
2414 __this_cpu_write(trace_taskinfo_save, false);
2415}
2416
2417/* Helpers to record a specific task information */
2418void tracing_record_cmdline(struct task_struct *task)
2419{
2420 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2421}
2422
2423void tracing_record_tgid(struct task_struct *task)
2424{
2425 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2426}
2427
2428/*
2429 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2430 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2431 * simplifies those functions and keeps them in sync.
2432 */
2433enum print_line_t trace_handle_return(struct trace_seq *s)
2434{
2435 return trace_seq_has_overflowed(s) ?
2436 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2437}
2438EXPORT_SYMBOL_GPL(trace_handle_return);
2439
2440void
2441tracing_generic_entry_update(struct trace_entry *entry, unsigned short type,
2442 unsigned long flags, int pc)
2443{
2444 struct task_struct *tsk = current;
2445
2446 entry->preempt_count = pc & 0xff;
2447 entry->pid = (tsk) ? tsk->pid : 0;
2448 entry->type = type;
2449 entry->flags =
2450#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
2451 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
2452#else
2453 TRACE_FLAG_IRQS_NOSUPPORT |
2454#endif
2455 ((pc & NMI_MASK ) ? TRACE_FLAG_NMI : 0) |
2456 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
2457 ((pc & SOFTIRQ_OFFSET) ? TRACE_FLAG_SOFTIRQ : 0) |
2458 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
2459 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
2460}
2461EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
2462
2463struct ring_buffer_event *
2464trace_buffer_lock_reserve(struct trace_buffer *buffer,
2465 int type,
2466 unsigned long len,
2467 unsigned long flags, int pc)
2468{
2469 return __trace_buffer_lock_reserve(buffer, type, len, flags, pc);
2470}
2471
2472DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2473DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2474static int trace_buffered_event_ref;
2475
2476/**
2477 * trace_buffered_event_enable - enable buffering events
2478 *
2479 * When events are being filtered, it is quicker to use a temporary
2480 * buffer to write the event data into if there's a likely chance
2481 * that it will not be committed. The discard of the ring buffer
2482 * is not as fast as committing, and is much slower than copying
2483 * a commit.
2484 *
2485 * When an event is to be filtered, allocate per cpu buffers to
2486 * write the event data into, and if the event is filtered and discarded
2487 * it is simply dropped, otherwise, the entire data is to be committed
2488 * in one shot.
2489 */
2490void trace_buffered_event_enable(void)
2491{
2492 struct ring_buffer_event *event;
2493 struct page *page;
2494 int cpu;
2495
2496 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2497
2498 if (trace_buffered_event_ref++)
2499 return;
2500
2501 for_each_tracing_cpu(cpu) {
2502 page = alloc_pages_node(cpu_to_node(cpu),
2503 GFP_KERNEL | __GFP_NORETRY, 0);
2504 if (!page)
2505 goto failed;
2506
2507 event = page_address(page);
2508 memset(event, 0, sizeof(*event));
2509
2510 per_cpu(trace_buffered_event, cpu) = event;
2511
2512 preempt_disable();
2513 if (cpu == smp_processor_id() &&
2514 this_cpu_read(trace_buffered_event) !=
2515 per_cpu(trace_buffered_event, cpu))
2516 WARN_ON_ONCE(1);
2517 preempt_enable();
2518 }
2519
2520 return;
2521 failed:
2522 trace_buffered_event_disable();
2523}
2524
2525static void enable_trace_buffered_event(void *data)
2526{
2527 /* Probably not needed, but do it anyway */
2528 smp_rmb();
2529 this_cpu_dec(trace_buffered_event_cnt);
2530}
2531
2532static void disable_trace_buffered_event(void *data)
2533{
2534 this_cpu_inc(trace_buffered_event_cnt);
2535}
2536
2537/**
2538 * trace_buffered_event_disable - disable buffering events
2539 *
2540 * When a filter is removed, it is faster to not use the buffered
2541 * events, and to commit directly into the ring buffer. Free up
2542 * the temp buffers when there are no more users. This requires
2543 * special synchronization with current events.
2544 */
2545void trace_buffered_event_disable(void)
2546{
2547 int cpu;
2548
2549 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2550
2551 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2552 return;
2553
2554 if (--trace_buffered_event_ref)
2555 return;
2556
2557 preempt_disable();
2558 /* For each CPU, set the buffer as used. */
2559 smp_call_function_many(tracing_buffer_mask,
2560 disable_trace_buffered_event, NULL, 1);
2561 preempt_enable();
2562
2563 /* Wait for all current users to finish */
2564 synchronize_rcu();
2565
2566 for_each_tracing_cpu(cpu) {
2567 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2568 per_cpu(trace_buffered_event, cpu) = NULL;
2569 }
2570 /*
2571 * Make sure trace_buffered_event is NULL before clearing
2572 * trace_buffered_event_cnt.
2573 */
2574 smp_wmb();
2575
2576 preempt_disable();
2577 /* Do the work on each cpu */
2578 smp_call_function_many(tracing_buffer_mask,
2579 enable_trace_buffered_event, NULL, 1);
2580 preempt_enable();
2581}
2582
2583static struct trace_buffer *temp_buffer;
2584
2585struct ring_buffer_event *
2586trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2587 struct trace_event_file *trace_file,
2588 int type, unsigned long len,
2589 unsigned long flags, int pc)
2590{
2591 struct ring_buffer_event *entry;
2592 int val;
2593
2594 *current_rb = trace_file->tr->array_buffer.buffer;
2595
2596 if (!ring_buffer_time_stamp_abs(*current_rb) && (trace_file->flags &
2597 (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED)) &&
2598 (entry = this_cpu_read(trace_buffered_event))) {
2599 /* Try to use the per cpu buffer first */
2600 val = this_cpu_inc_return(trace_buffered_event_cnt);
2601 if (val == 1) {
2602 trace_event_setup(entry, type, flags, pc);
2603 entry->array[0] = len;
2604 return entry;
2605 }
2606 this_cpu_dec(trace_buffered_event_cnt);
2607 }
2608
2609 entry = __trace_buffer_lock_reserve(*current_rb,
2610 type, len, flags, pc);
2611 /*
2612 * If tracing is off, but we have triggers enabled
2613 * we still need to look at the event data. Use the temp_buffer
2614 * to store the trace event for the tigger to use. It's recusive
2615 * safe and will not be recorded anywhere.
2616 */
2617 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2618 *current_rb = temp_buffer;
2619 entry = __trace_buffer_lock_reserve(*current_rb,
2620 type, len, flags, pc);
2621 }
2622 return entry;
2623}
2624EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2625
2626static DEFINE_SPINLOCK(tracepoint_iter_lock);
2627static DEFINE_MUTEX(tracepoint_printk_mutex);
2628
2629static void output_printk(struct trace_event_buffer *fbuffer)
2630{
2631 struct trace_event_call *event_call;
2632 struct trace_event_file *file;
2633 struct trace_event *event;
2634 unsigned long flags;
2635 struct trace_iterator *iter = tracepoint_print_iter;
2636
2637 /* We should never get here if iter is NULL */
2638 if (WARN_ON_ONCE(!iter))
2639 return;
2640
2641 event_call = fbuffer->trace_file->event_call;
2642 if (!event_call || !event_call->event.funcs ||
2643 !event_call->event.funcs->trace)
2644 return;
2645
2646 file = fbuffer->trace_file;
2647 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2648 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2649 !filter_match_preds(file->filter, fbuffer->entry)))
2650 return;
2651
2652 event = &fbuffer->trace_file->event_call->event;
2653
2654 spin_lock_irqsave(&tracepoint_iter_lock, flags);
2655 trace_seq_init(&iter->seq);
2656 iter->ent = fbuffer->entry;
2657 event_call->event.funcs->trace(iter, 0, event);
2658 trace_seq_putc(&iter->seq, 0);
2659 printk("%s", iter->seq.buffer);
2660
2661 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2662}
2663
2664int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2665 void *buffer, size_t *lenp,
2666 loff_t *ppos)
2667{
2668 int save_tracepoint_printk;
2669 int ret;
2670
2671 mutex_lock(&tracepoint_printk_mutex);
2672 save_tracepoint_printk = tracepoint_printk;
2673
2674 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2675
2676 /*
2677 * This will force exiting early, as tracepoint_printk
2678 * is always zero when tracepoint_printk_iter is not allocated
2679 */
2680 if (!tracepoint_print_iter)
2681 tracepoint_printk = 0;
2682
2683 if (save_tracepoint_printk == tracepoint_printk)
2684 goto out;
2685
2686 if (tracepoint_printk)
2687 static_key_enable(&tracepoint_printk_key.key);
2688 else
2689 static_key_disable(&tracepoint_printk_key.key);
2690
2691 out:
2692 mutex_unlock(&tracepoint_printk_mutex);
2693
2694 return ret;
2695}
2696
2697void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2698{
2699 if (static_key_false(&tracepoint_printk_key.key))
2700 output_printk(fbuffer);
2701
2702 event_trigger_unlock_commit_regs(fbuffer->trace_file, fbuffer->buffer,
2703 fbuffer->event, fbuffer->entry,
2704 fbuffer->flags, fbuffer->pc, fbuffer->regs);
2705}
2706EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2707
2708/*
2709 * Skip 3:
2710 *
2711 * trace_buffer_unlock_commit_regs()
2712 * trace_event_buffer_commit()
2713 * trace_event_raw_event_xxx()
2714 */
2715# define STACK_SKIP 3
2716
2717void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2718 struct trace_buffer *buffer,
2719 struct ring_buffer_event *event,
2720 unsigned long flags, int pc,
2721 struct pt_regs *regs)
2722{
2723 __buffer_unlock_commit(buffer, event);
2724
2725 /*
2726 * If regs is not set, then skip the necessary functions.
2727 * Note, we can still get here via blktrace, wakeup tracer
2728 * and mmiotrace, but that's ok if they lose a function or
2729 * two. They are not that meaningful.
2730 */
2731 ftrace_trace_stack(tr, buffer, flags, regs ? 0 : STACK_SKIP, pc, regs);
2732 ftrace_trace_userstack(buffer, flags, pc);
2733}
2734
2735/*
2736 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2737 */
2738void
2739trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2740 struct ring_buffer_event *event)
2741{
2742 __buffer_unlock_commit(buffer, event);
2743}
2744
2745static void
2746trace_process_export(struct trace_export *export,
2747 struct ring_buffer_event *event)
2748{
2749 struct trace_entry *entry;
2750 unsigned int size = 0;
2751
2752 entry = ring_buffer_event_data(event);
2753 size = ring_buffer_event_length(event);
2754 export->write(export, entry, size);
2755}
2756
2757static DEFINE_MUTEX(ftrace_export_lock);
2758
2759static struct trace_export __rcu *ftrace_exports_list __read_mostly;
2760
2761static DEFINE_STATIC_KEY_FALSE(ftrace_exports_enabled);
2762
2763static inline void ftrace_exports_enable(void)
2764{
2765 static_branch_enable(&ftrace_exports_enabled);
2766}
2767
2768static inline void ftrace_exports_disable(void)
2769{
2770 static_branch_disable(&ftrace_exports_enabled);
2771}
2772
2773static void ftrace_exports(struct ring_buffer_event *event)
2774{
2775 struct trace_export *export;
2776
2777 preempt_disable_notrace();
2778
2779 export = rcu_dereference_raw_check(ftrace_exports_list);
2780 while (export) {
2781 trace_process_export(export, event);
2782 export = rcu_dereference_raw_check(export->next);
2783 }
2784
2785 preempt_enable_notrace();
2786}
2787
2788static inline void
2789add_trace_export(struct trace_export **list, struct trace_export *export)
2790{
2791 rcu_assign_pointer(export->next, *list);
2792 /*
2793 * We are entering export into the list but another
2794 * CPU might be walking that list. We need to make sure
2795 * the export->next pointer is valid before another CPU sees
2796 * the export pointer included into the list.
2797 */
2798 rcu_assign_pointer(*list, export);
2799}
2800
2801static inline int
2802rm_trace_export(struct trace_export **list, struct trace_export *export)
2803{
2804 struct trace_export **p;
2805
2806 for (p = list; *p != NULL; p = &(*p)->next)
2807 if (*p == export)
2808 break;
2809
2810 if (*p != export)
2811 return -1;
2812
2813 rcu_assign_pointer(*p, (*p)->next);
2814
2815 return 0;
2816}
2817
2818static inline void
2819add_ftrace_export(struct trace_export **list, struct trace_export *export)
2820{
2821 if (*list == NULL)
2822 ftrace_exports_enable();
2823
2824 add_trace_export(list, export);
2825}
2826
2827static inline int
2828rm_ftrace_export(struct trace_export **list, struct trace_export *export)
2829{
2830 int ret;
2831
2832 ret = rm_trace_export(list, export);
2833 if (*list == NULL)
2834 ftrace_exports_disable();
2835
2836 return ret;
2837}
2838
2839int register_ftrace_export(struct trace_export *export)
2840{
2841 if (WARN_ON_ONCE(!export->write))
2842 return -1;
2843
2844 mutex_lock(&ftrace_export_lock);
2845
2846 add_ftrace_export(&ftrace_exports_list, export);
2847
2848 mutex_unlock(&ftrace_export_lock);
2849
2850 return 0;
2851}
2852EXPORT_SYMBOL_GPL(register_ftrace_export);
2853
2854int unregister_ftrace_export(struct trace_export *export)
2855{
2856 int ret;
2857
2858 mutex_lock(&ftrace_export_lock);
2859
2860 ret = rm_ftrace_export(&ftrace_exports_list, export);
2861
2862 mutex_unlock(&ftrace_export_lock);
2863
2864 return ret;
2865}
2866EXPORT_SYMBOL_GPL(unregister_ftrace_export);
2867
2868void
2869trace_function(struct trace_array *tr,
2870 unsigned long ip, unsigned long parent_ip, unsigned long flags,
2871 int pc)
2872{
2873 struct trace_event_call *call = &event_function;
2874 struct trace_buffer *buffer = tr->array_buffer.buffer;
2875 struct ring_buffer_event *event;
2876 struct ftrace_entry *entry;
2877
2878 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2879 flags, pc);
2880 if (!event)
2881 return;
2882 entry = ring_buffer_event_data(event);
2883 entry->ip = ip;
2884 entry->parent_ip = parent_ip;
2885
2886 if (!call_filter_check_discard(call, entry, buffer, event)) {
2887 if (static_branch_unlikely(&ftrace_exports_enabled))
2888 ftrace_exports(event);
2889 __buffer_unlock_commit(buffer, event);
2890 }
2891}
2892
2893#ifdef CONFIG_STACKTRACE
2894
2895/* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2896#define FTRACE_KSTACK_NESTING 4
2897
2898#define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
2899
2900struct ftrace_stack {
2901 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2902};
2903
2904
2905struct ftrace_stacks {
2906 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
2907};
2908
2909static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
2910static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2911
2912static void __ftrace_trace_stack(struct trace_buffer *buffer,
2913 unsigned long flags,
2914 int skip, int pc, struct pt_regs *regs)
2915{
2916 struct trace_event_call *call = &event_kernel_stack;
2917 struct ring_buffer_event *event;
2918 unsigned int size, nr_entries;
2919 struct ftrace_stack *fstack;
2920 struct stack_entry *entry;
2921 int stackidx;
2922
2923 /*
2924 * Add one, for this function and the call to save_stack_trace()
2925 * If regs is set, then these functions will not be in the way.
2926 */
2927#ifndef CONFIG_UNWINDER_ORC
2928 if (!regs)
2929 skip++;
2930#endif
2931
2932 preempt_disable_notrace();
2933
2934 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
2935
2936 /* This should never happen. If it does, yell once and skip */
2937 if (WARN_ON_ONCE(stackidx > FTRACE_KSTACK_NESTING))
2938 goto out;
2939
2940 /*
2941 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
2942 * interrupt will either see the value pre increment or post
2943 * increment. If the interrupt happens pre increment it will have
2944 * restored the counter when it returns. We just need a barrier to
2945 * keep gcc from moving things around.
2946 */
2947 barrier();
2948
2949 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
2950 size = ARRAY_SIZE(fstack->calls);
2951
2952 if (regs) {
2953 nr_entries = stack_trace_save_regs(regs, fstack->calls,
2954 size, skip);
2955 } else {
2956 nr_entries = stack_trace_save(fstack->calls, size, skip);
2957 }
2958
2959 size = nr_entries * sizeof(unsigned long);
2960 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2961 sizeof(*entry) + size, flags, pc);
2962 if (!event)
2963 goto out;
2964 entry = ring_buffer_event_data(event);
2965
2966 memcpy(&entry->caller, fstack->calls, size);
2967 entry->size = nr_entries;
2968
2969 if (!call_filter_check_discard(call, entry, buffer, event))
2970 __buffer_unlock_commit(buffer, event);
2971
2972 out:
2973 /* Again, don't let gcc optimize things here */
2974 barrier();
2975 __this_cpu_dec(ftrace_stack_reserve);
2976 preempt_enable_notrace();
2977
2978}
2979
2980static inline void ftrace_trace_stack(struct trace_array *tr,
2981 struct trace_buffer *buffer,
2982 unsigned long flags,
2983 int skip, int pc, struct pt_regs *regs)
2984{
2985 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
2986 return;
2987
2988 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
2989}
2990
2991void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
2992 int pc)
2993{
2994 struct trace_buffer *buffer = tr->array_buffer.buffer;
2995
2996 if (rcu_is_watching()) {
2997 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
2998 return;
2999 }
3000
3001 /*
3002 * When an NMI triggers, RCU is enabled via rcu_nmi_enter(),
3003 * but if the above rcu_is_watching() failed, then the NMI
3004 * triggered someplace critical, and rcu_irq_enter() should
3005 * not be called from NMI.
3006 */
3007 if (unlikely(in_nmi()))
3008 return;
3009
3010 rcu_irq_enter_irqson();
3011 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
3012 rcu_irq_exit_irqson();
3013}
3014
3015/**
3016 * trace_dump_stack - record a stack back trace in the trace buffer
3017 * @skip: Number of functions to skip (helper handlers)
3018 */
3019void trace_dump_stack(int skip)
3020{
3021 unsigned long flags;
3022
3023 if (tracing_disabled || tracing_selftest_running)
3024 return;
3025
3026 local_save_flags(flags);
3027
3028#ifndef CONFIG_UNWINDER_ORC
3029 /* Skip 1 to skip this function. */
3030 skip++;
3031#endif
3032 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3033 flags, skip, preempt_count(), NULL);
3034}
3035EXPORT_SYMBOL_GPL(trace_dump_stack);
3036
3037#ifdef CONFIG_USER_STACKTRACE_SUPPORT
3038static DEFINE_PER_CPU(int, user_stack_count);
3039
3040static void
3041ftrace_trace_userstack(struct trace_buffer *buffer, unsigned long flags, int pc)
3042{
3043 struct trace_event_call *call = &event_user_stack;
3044 struct ring_buffer_event *event;
3045 struct userstack_entry *entry;
3046
3047 if (!(global_trace.trace_flags & TRACE_ITER_USERSTACKTRACE))
3048 return;
3049
3050 /*
3051 * NMIs can not handle page faults, even with fix ups.
3052 * The save user stack can (and often does) fault.
3053 */
3054 if (unlikely(in_nmi()))
3055 return;
3056
3057 /*
3058 * prevent recursion, since the user stack tracing may
3059 * trigger other kernel events.
3060 */
3061 preempt_disable();
3062 if (__this_cpu_read(user_stack_count))
3063 goto out;
3064
3065 __this_cpu_inc(user_stack_count);
3066
3067 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3068 sizeof(*entry), flags, pc);
3069 if (!event)
3070 goto out_drop_count;
3071 entry = ring_buffer_event_data(event);
3072
3073 entry->tgid = current->tgid;
3074 memset(&entry->caller, 0, sizeof(entry->caller));
3075
3076 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3077 if (!call_filter_check_discard(call, entry, buffer, event))
3078 __buffer_unlock_commit(buffer, event);
3079
3080 out_drop_count:
3081 __this_cpu_dec(user_stack_count);
3082 out:
3083 preempt_enable();
3084}
3085#else /* CONFIG_USER_STACKTRACE_SUPPORT */
3086static void ftrace_trace_userstack(struct trace_buffer *buffer,
3087 unsigned long flags, int pc)
3088{
3089}
3090#endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3091
3092#endif /* CONFIG_STACKTRACE */
3093
3094/* created for use with alloc_percpu */
3095struct trace_buffer_struct {
3096 int nesting;
3097 char buffer[4][TRACE_BUF_SIZE];
3098};
3099
3100static struct trace_buffer_struct *trace_percpu_buffer;
3101
3102/*
3103 * Thise allows for lockless recording. If we're nested too deeply, then
3104 * this returns NULL.
3105 */
3106static char *get_trace_buf(void)
3107{
3108 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3109
3110 if (!buffer || buffer->nesting >= 4)
3111 return NULL;
3112
3113 buffer->nesting++;
3114
3115 /* Interrupts must see nesting incremented before we use the buffer */
3116 barrier();
3117 return &buffer->buffer[buffer->nesting][0];
3118}
3119
3120static void put_trace_buf(void)
3121{
3122 /* Don't let the decrement of nesting leak before this */
3123 barrier();
3124 this_cpu_dec(trace_percpu_buffer->nesting);
3125}
3126
3127static int alloc_percpu_trace_buffer(void)
3128{
3129 struct trace_buffer_struct *buffers;
3130
3131 if (trace_percpu_buffer)
3132 return 0;
3133
3134 buffers = alloc_percpu(struct trace_buffer_struct);
3135 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3136 return -ENOMEM;
3137
3138 trace_percpu_buffer = buffers;
3139 return 0;
3140}
3141
3142static int buffers_allocated;
3143
3144void trace_printk_init_buffers(void)
3145{
3146 if (buffers_allocated)
3147 return;
3148
3149 if (alloc_percpu_trace_buffer())
3150 return;
3151
3152 /* trace_printk() is for debug use only. Don't use it in production. */
3153
3154 pr_warn("\n");
3155 pr_warn("**********************************************************\n");
3156 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3157 pr_warn("** **\n");
3158 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3159 pr_warn("** **\n");
3160 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3161 pr_warn("** unsafe for production use. **\n");
3162 pr_warn("** **\n");
3163 pr_warn("** If you see this message and you are not debugging **\n");
3164 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3165 pr_warn("** **\n");
3166 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3167 pr_warn("**********************************************************\n");
3168
3169 /* Expand the buffers to set size */
3170 tracing_update_buffers();
3171
3172 buffers_allocated = 1;
3173
3174 /*
3175 * trace_printk_init_buffers() can be called by modules.
3176 * If that happens, then we need to start cmdline recording
3177 * directly here. If the global_trace.buffer is already
3178 * allocated here, then this was called by module code.
3179 */
3180 if (global_trace.array_buffer.buffer)
3181 tracing_start_cmdline_record();
3182}
3183EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3184
3185void trace_printk_start_comm(void)
3186{
3187 /* Start tracing comms if trace printk is set */
3188 if (!buffers_allocated)
3189 return;
3190 tracing_start_cmdline_record();
3191}
3192
3193static void trace_printk_start_stop_comm(int enabled)
3194{
3195 if (!buffers_allocated)
3196 return;
3197
3198 if (enabled)
3199 tracing_start_cmdline_record();
3200 else
3201 tracing_stop_cmdline_record();
3202}
3203
3204/**
3205 * trace_vbprintk - write binary msg to tracing buffer
3206 * @ip: The address of the caller
3207 * @fmt: The string format to write to the buffer
3208 * @args: Arguments for @fmt
3209 */
3210int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3211{
3212 struct trace_event_call *call = &event_bprint;
3213 struct ring_buffer_event *event;
3214 struct trace_buffer *buffer;
3215 struct trace_array *tr = &global_trace;
3216 struct bprint_entry *entry;
3217 unsigned long flags;
3218 char *tbuffer;
3219 int len = 0, size, pc;
3220
3221 if (unlikely(tracing_selftest_running || tracing_disabled))
3222 return 0;
3223
3224 /* Don't pollute graph traces with trace_vprintk internals */
3225 pause_graph_tracing();
3226
3227 pc = preempt_count();
3228 preempt_disable_notrace();
3229
3230 tbuffer = get_trace_buf();
3231 if (!tbuffer) {
3232 len = 0;
3233 goto out_nobuffer;
3234 }
3235
3236 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3237
3238 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3239 goto out_put;
3240
3241 local_save_flags(flags);
3242 size = sizeof(*entry) + sizeof(u32) * len;
3243 buffer = tr->array_buffer.buffer;
3244 ring_buffer_nest_start(buffer);
3245 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3246 flags, pc);
3247 if (!event)
3248 goto out;
3249 entry = ring_buffer_event_data(event);
3250 entry->ip = ip;
3251 entry->fmt = fmt;
3252
3253 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3254 if (!call_filter_check_discard(call, entry, buffer, event)) {
3255 __buffer_unlock_commit(buffer, event);
3256 ftrace_trace_stack(tr, buffer, flags, 6, pc, NULL);
3257 }
3258
3259out:
3260 ring_buffer_nest_end(buffer);
3261out_put:
3262 put_trace_buf();
3263
3264out_nobuffer:
3265 preempt_enable_notrace();
3266 unpause_graph_tracing();
3267
3268 return len;
3269}
3270EXPORT_SYMBOL_GPL(trace_vbprintk);
3271
3272__printf(3, 0)
3273static int
3274__trace_array_vprintk(struct trace_buffer *buffer,
3275 unsigned long ip, const char *fmt, va_list args)
3276{
3277 struct trace_event_call *call = &event_print;
3278 struct ring_buffer_event *event;
3279 int len = 0, size, pc;
3280 struct print_entry *entry;
3281 unsigned long flags;
3282 char *tbuffer;
3283
3284 if (tracing_disabled || tracing_selftest_running)
3285 return 0;
3286
3287 /* Don't pollute graph traces with trace_vprintk internals */
3288 pause_graph_tracing();
3289
3290 pc = preempt_count();
3291 preempt_disable_notrace();
3292
3293
3294 tbuffer = get_trace_buf();
3295 if (!tbuffer) {
3296 len = 0;
3297 goto out_nobuffer;
3298 }
3299
3300 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3301
3302 local_save_flags(flags);
3303 size = sizeof(*entry) + len + 1;
3304 ring_buffer_nest_start(buffer);
3305 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3306 flags, pc);
3307 if (!event)
3308 goto out;
3309 entry = ring_buffer_event_data(event);
3310 entry->ip = ip;
3311
3312 memcpy(&entry->buf, tbuffer, len + 1);
3313 if (!call_filter_check_discard(call, entry, buffer, event)) {
3314 __buffer_unlock_commit(buffer, event);
3315 ftrace_trace_stack(&global_trace, buffer, flags, 6, pc, NULL);
3316 }
3317
3318out:
3319 ring_buffer_nest_end(buffer);
3320 put_trace_buf();
3321
3322out_nobuffer:
3323 preempt_enable_notrace();
3324 unpause_graph_tracing();
3325
3326 return len;
3327}
3328
3329__printf(3, 0)
3330int trace_array_vprintk(struct trace_array *tr,
3331 unsigned long ip, const char *fmt, va_list args)
3332{
3333 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3334}
3335
3336/**
3337 * trace_array_printk - Print a message to a specific instance
3338 * @tr: The instance trace_array descriptor
3339 * @ip: The instruction pointer that this is called from.
3340 * @fmt: The format to print (printf format)
3341 *
3342 * If a subsystem sets up its own instance, they have the right to
3343 * printk strings into their tracing instance buffer using this
3344 * function. Note, this function will not write into the top level
3345 * buffer (use trace_printk() for that), as writing into the top level
3346 * buffer should only have events that can be individually disabled.
3347 * trace_printk() is only used for debugging a kernel, and should not
3348 * be ever encorporated in normal use.
3349 *
3350 * trace_array_printk() can be used, as it will not add noise to the
3351 * top level tracing buffer.
3352 *
3353 * Note, trace_array_init_printk() must be called on @tr before this
3354 * can be used.
3355 */
3356__printf(3, 0)
3357int trace_array_printk(struct trace_array *tr,
3358 unsigned long ip, const char *fmt, ...)
3359{
3360 int ret;
3361 va_list ap;
3362
3363 if (!tr)
3364 return -ENOENT;
3365
3366 /* This is only allowed for created instances */
3367 if (tr == &global_trace)
3368 return 0;
3369
3370 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3371 return 0;
3372
3373 va_start(ap, fmt);
3374 ret = trace_array_vprintk(tr, ip, fmt, ap);
3375 va_end(ap);
3376 return ret;
3377}
3378EXPORT_SYMBOL_GPL(trace_array_printk);
3379
3380/**
3381 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3382 * @tr: The trace array to initialize the buffers for
3383 *
3384 * As trace_array_printk() only writes into instances, they are OK to
3385 * have in the kernel (unlike trace_printk()). This needs to be called
3386 * before trace_array_printk() can be used on a trace_array.
3387 */
3388int trace_array_init_printk(struct trace_array *tr)
3389{
3390 if (!tr)
3391 return -ENOENT;
3392
3393 /* This is only allowed for created instances */
3394 if (tr == &global_trace)
3395 return -EINVAL;
3396
3397 return alloc_percpu_trace_buffer();
3398}
3399EXPORT_SYMBOL_GPL(trace_array_init_printk);
3400
3401__printf(3, 4)
3402int trace_array_printk_buf(struct trace_buffer *buffer,
3403 unsigned long ip, const char *fmt, ...)
3404{
3405 int ret;
3406 va_list ap;
3407
3408 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3409 return 0;
3410
3411 va_start(ap, fmt);
3412 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3413 va_end(ap);
3414 return ret;
3415}
3416
3417__printf(2, 0)
3418int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3419{
3420 return trace_array_vprintk(&global_trace, ip, fmt, args);
3421}
3422EXPORT_SYMBOL_GPL(trace_vprintk);
3423
3424static void trace_iterator_increment(struct trace_iterator *iter)
3425{
3426 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3427
3428 iter->idx++;
3429 if (buf_iter)
3430 ring_buffer_iter_advance(buf_iter);
3431}
3432
3433static struct trace_entry *
3434peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3435 unsigned long *lost_events)
3436{
3437 struct ring_buffer_event *event;
3438 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3439
3440 if (buf_iter) {
3441 event = ring_buffer_iter_peek(buf_iter, ts);
3442 if (lost_events)
3443 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3444 (unsigned long)-1 : 0;
3445 } else {
3446 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3447 lost_events);
3448 }
3449
3450 if (event) {
3451 iter->ent_size = ring_buffer_event_length(event);
3452 return ring_buffer_event_data(event);
3453 }
3454 iter->ent_size = 0;
3455 return NULL;
3456}
3457
3458static struct trace_entry *
3459__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3460 unsigned long *missing_events, u64 *ent_ts)
3461{
3462 struct trace_buffer *buffer = iter->array_buffer->buffer;
3463 struct trace_entry *ent, *next = NULL;
3464 unsigned long lost_events = 0, next_lost = 0;
3465 int cpu_file = iter->cpu_file;
3466 u64 next_ts = 0, ts;
3467 int next_cpu = -1;
3468 int next_size = 0;
3469 int cpu;
3470
3471 /*
3472 * If we are in a per_cpu trace file, don't bother by iterating over
3473 * all cpu and peek directly.
3474 */
3475 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3476 if (ring_buffer_empty_cpu(buffer, cpu_file))
3477 return NULL;
3478 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3479 if (ent_cpu)
3480 *ent_cpu = cpu_file;
3481
3482 return ent;
3483 }
3484
3485 for_each_tracing_cpu(cpu) {
3486
3487 if (ring_buffer_empty_cpu(buffer, cpu))
3488 continue;
3489
3490 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3491
3492 /*
3493 * Pick the entry with the smallest timestamp:
3494 */
3495 if (ent && (!next || ts < next_ts)) {
3496 next = ent;
3497 next_cpu = cpu;
3498 next_ts = ts;
3499 next_lost = lost_events;
3500 next_size = iter->ent_size;
3501 }
3502 }
3503
3504 iter->ent_size = next_size;
3505
3506 if (ent_cpu)
3507 *ent_cpu = next_cpu;
3508
3509 if (ent_ts)
3510 *ent_ts = next_ts;
3511
3512 if (missing_events)
3513 *missing_events = next_lost;
3514
3515 return next;
3516}
3517
3518#define STATIC_TEMP_BUF_SIZE 128
3519static char static_temp_buf[STATIC_TEMP_BUF_SIZE];
3520
3521/* Find the next real entry, without updating the iterator itself */
3522struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3523 int *ent_cpu, u64 *ent_ts)
3524{
3525 /* __find_next_entry will reset ent_size */
3526 int ent_size = iter->ent_size;
3527 struct trace_entry *entry;
3528
3529 /*
3530 * If called from ftrace_dump(), then the iter->temp buffer
3531 * will be the static_temp_buf and not created from kmalloc.
3532 * If the entry size is greater than the buffer, we can
3533 * not save it. Just return NULL in that case. This is only
3534 * used to add markers when two consecutive events' time
3535 * stamps have a large delta. See trace_print_lat_context()
3536 */
3537 if (iter->temp == static_temp_buf &&
3538 STATIC_TEMP_BUF_SIZE < ent_size)
3539 return NULL;
3540
3541 /*
3542 * The __find_next_entry() may call peek_next_entry(), which may
3543 * call ring_buffer_peek() that may make the contents of iter->ent
3544 * undefined. Need to copy iter->ent now.
3545 */
3546 if (iter->ent && iter->ent != iter->temp) {
3547 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3548 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3549 void *temp;
3550 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3551 if (!temp)
3552 return NULL;
3553 kfree(iter->temp);
3554 iter->temp = temp;
3555 iter->temp_size = iter->ent_size;
3556 }
3557 memcpy(iter->temp, iter->ent, iter->ent_size);
3558 iter->ent = iter->temp;
3559 }
3560 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3561 /* Put back the original ent_size */
3562 iter->ent_size = ent_size;
3563
3564 return entry;
3565}
3566
3567/* Find the next real entry, and increment the iterator to the next entry */
3568void *trace_find_next_entry_inc(struct trace_iterator *iter)
3569{
3570 iter->ent = __find_next_entry(iter, &iter->cpu,
3571 &iter->lost_events, &iter->ts);
3572
3573 if (iter->ent)
3574 trace_iterator_increment(iter);
3575
3576 return iter->ent ? iter : NULL;
3577}
3578
3579static void trace_consume(struct trace_iterator *iter)
3580{
3581 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3582 &iter->lost_events);
3583}
3584
3585static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3586{
3587 struct trace_iterator *iter = m->private;
3588 int i = (int)*pos;
3589 void *ent;
3590
3591 WARN_ON_ONCE(iter->leftover);
3592
3593 (*pos)++;
3594
3595 /* can't go backwards */
3596 if (iter->idx > i)
3597 return NULL;
3598
3599 if (iter->idx < 0)
3600 ent = trace_find_next_entry_inc(iter);
3601 else
3602 ent = iter;
3603
3604 while (ent && iter->idx < i)
3605 ent = trace_find_next_entry_inc(iter);
3606
3607 iter->pos = *pos;
3608
3609 return ent;
3610}
3611
3612void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3613{
3614 struct ring_buffer_iter *buf_iter;
3615 unsigned long entries = 0;
3616 u64 ts;
3617
3618 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
3619
3620 buf_iter = trace_buffer_iter(iter, cpu);
3621 if (!buf_iter)
3622 return;
3623
3624 ring_buffer_iter_reset(buf_iter);
3625
3626 /*
3627 * We could have the case with the max latency tracers
3628 * that a reset never took place on a cpu. This is evident
3629 * by the timestamp being before the start of the buffer.
3630 */
3631 while (ring_buffer_iter_peek(buf_iter, &ts)) {
3632 if (ts >= iter->array_buffer->time_start)
3633 break;
3634 entries++;
3635 ring_buffer_iter_advance(buf_iter);
3636 }
3637
3638 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
3639}
3640
3641/*
3642 * The current tracer is copied to avoid a global locking
3643 * all around.
3644 */
3645static void *s_start(struct seq_file *m, loff_t *pos)
3646{
3647 struct trace_iterator *iter = m->private;
3648 struct trace_array *tr = iter->tr;
3649 int cpu_file = iter->cpu_file;
3650 void *p = NULL;
3651 loff_t l = 0;
3652 int cpu;
3653
3654 /*
3655 * copy the tracer to avoid using a global lock all around.
3656 * iter->trace is a copy of current_trace, the pointer to the
3657 * name may be used instead of a strcmp(), as iter->trace->name
3658 * will point to the same string as current_trace->name.
3659 */
3660 mutex_lock(&trace_types_lock);
3661 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
3662 *iter->trace = *tr->current_trace;
3663 mutex_unlock(&trace_types_lock);
3664
3665#ifdef CONFIG_TRACER_MAX_TRACE
3666 if (iter->snapshot && iter->trace->use_max_tr)
3667 return ERR_PTR(-EBUSY);
3668#endif
3669
3670 if (!iter->snapshot)
3671 atomic_inc(&trace_record_taskinfo_disabled);
3672
3673 if (*pos != iter->pos) {
3674 iter->ent = NULL;
3675 iter->cpu = 0;
3676 iter->idx = -1;
3677
3678 if (cpu_file == RING_BUFFER_ALL_CPUS) {
3679 for_each_tracing_cpu(cpu)
3680 tracing_iter_reset(iter, cpu);
3681 } else
3682 tracing_iter_reset(iter, cpu_file);
3683
3684 iter->leftover = 0;
3685 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
3686 ;
3687
3688 } else {
3689 /*
3690 * If we overflowed the seq_file before, then we want
3691 * to just reuse the trace_seq buffer again.
3692 */
3693 if (iter->leftover)
3694 p = iter;
3695 else {
3696 l = *pos - 1;
3697 p = s_next(m, p, &l);
3698 }
3699 }
3700
3701 trace_event_read_lock();
3702 trace_access_lock(cpu_file);
3703 return p;
3704}
3705
3706static void s_stop(struct seq_file *m, void *p)
3707{
3708 struct trace_iterator *iter = m->private;
3709
3710#ifdef CONFIG_TRACER_MAX_TRACE
3711 if (iter->snapshot && iter->trace->use_max_tr)
3712 return;
3713#endif
3714
3715 if (!iter->snapshot)
3716 atomic_dec(&trace_record_taskinfo_disabled);
3717
3718 trace_access_unlock(iter->cpu_file);
3719 trace_event_read_unlock();
3720}
3721
3722static void
3723get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
3724 unsigned long *entries, int cpu)
3725{
3726 unsigned long count;
3727
3728 count = ring_buffer_entries_cpu(buf->buffer, cpu);
3729 /*
3730 * If this buffer has skipped entries, then we hold all
3731 * entries for the trace and we need to ignore the
3732 * ones before the time stamp.
3733 */
3734 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
3735 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
3736 /* total is the same as the entries */
3737 *total = count;
3738 } else
3739 *total = count +
3740 ring_buffer_overrun_cpu(buf->buffer, cpu);
3741 *entries = count;
3742}
3743
3744static void
3745get_total_entries(struct array_buffer *buf,
3746 unsigned long *total, unsigned long *entries)
3747{
3748 unsigned long t, e;
3749 int cpu;
3750
3751 *total = 0;
3752 *entries = 0;
3753
3754 for_each_tracing_cpu(cpu) {
3755 get_total_entries_cpu(buf, &t, &e, cpu);
3756 *total += t;
3757 *entries += e;
3758 }
3759}
3760
3761unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
3762{
3763 unsigned long total, entries;
3764
3765 if (!tr)
3766 tr = &global_trace;
3767
3768 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
3769
3770 return entries;
3771}
3772
3773unsigned long trace_total_entries(struct trace_array *tr)
3774{
3775 unsigned long total, entries;
3776
3777 if (!tr)
3778 tr = &global_trace;
3779
3780 get_total_entries(&tr->array_buffer, &total, &entries);
3781
3782 return entries;
3783}
3784
3785static void print_lat_help_header(struct seq_file *m)
3786{
3787 seq_puts(m, "# _------=> CPU# \n"
3788 "# / _-----=> irqs-off \n"
3789 "# | / _----=> need-resched \n"
3790 "# || / _---=> hardirq/softirq \n"
3791 "# ||| / _--=> preempt-depth \n"
3792 "# |||| / delay \n"
3793 "# cmd pid ||||| time | caller \n"
3794 "# \\ / ||||| \\ | / \n");
3795}
3796
3797static void print_event_info(struct array_buffer *buf, struct seq_file *m)
3798{
3799 unsigned long total;
3800 unsigned long entries;
3801
3802 get_total_entries(buf, &total, &entries);
3803 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
3804 entries, total, num_online_cpus());
3805 seq_puts(m, "#\n");
3806}
3807
3808static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
3809 unsigned int flags)
3810{
3811 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3812
3813 print_event_info(buf, m);
3814
3815 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
3816 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
3817}
3818
3819static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
3820 unsigned int flags)
3821{
3822 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3823 const char *space = " ";
3824 int prec = tgid ? 12 : 2;
3825
3826 print_event_info(buf, m);
3827
3828 seq_printf(m, "# %.*s _-----=> irqs-off\n", prec, space);
3829 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
3830 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
3831 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
3832 seq_printf(m, "# %.*s||| / delay\n", prec, space);
3833 seq_printf(m, "# TASK-PID %.*s CPU# |||| TIMESTAMP FUNCTION\n", prec, " TGID ");
3834 seq_printf(m, "# | | %.*s | |||| | |\n", prec, " | ");
3835}
3836
3837void
3838print_trace_header(struct seq_file *m, struct trace_iterator *iter)
3839{
3840 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
3841 struct array_buffer *buf = iter->array_buffer;
3842 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
3843 struct tracer *type = iter->trace;
3844 unsigned long entries;
3845 unsigned long total;
3846 const char *name = "preemption";
3847
3848 name = type->name;
3849
3850 get_total_entries(buf, &total, &entries);
3851
3852 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
3853 name, UTS_RELEASE);
3854 seq_puts(m, "# -----------------------------------"
3855 "---------------------------------\n");
3856 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
3857 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
3858 nsecs_to_usecs(data->saved_latency),
3859 entries,
3860 total,
3861 buf->cpu,
3862#if defined(CONFIG_PREEMPT_NONE)
3863 "server",
3864#elif defined(CONFIG_PREEMPT_VOLUNTARY)
3865 "desktop",
3866#elif defined(CONFIG_PREEMPT)
3867 "preempt",
3868#elif defined(CONFIG_PREEMPT_RT)
3869 "preempt_rt",
3870#else
3871 "unknown",
3872#endif
3873 /* These are reserved for later use */
3874 0, 0, 0, 0);
3875#ifdef CONFIG_SMP
3876 seq_printf(m, " #P:%d)\n", num_online_cpus());
3877#else
3878 seq_puts(m, ")\n");
3879#endif
3880 seq_puts(m, "# -----------------\n");
3881 seq_printf(m, "# | task: %.16s-%d "
3882 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
3883 data->comm, data->pid,
3884 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
3885 data->policy, data->rt_priority);
3886 seq_puts(m, "# -----------------\n");
3887
3888 if (data->critical_start) {
3889 seq_puts(m, "# => started at: ");
3890 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
3891 trace_print_seq(m, &iter->seq);
3892 seq_puts(m, "\n# => ended at: ");
3893 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
3894 trace_print_seq(m, &iter->seq);
3895 seq_puts(m, "\n#\n");
3896 }
3897
3898 seq_puts(m, "#\n");
3899}
3900
3901static void test_cpu_buff_start(struct trace_iterator *iter)
3902{
3903 struct trace_seq *s = &iter->seq;
3904 struct trace_array *tr = iter->tr;
3905
3906 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
3907 return;
3908
3909 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
3910 return;
3911
3912 if (cpumask_available(iter->started) &&
3913 cpumask_test_cpu(iter->cpu, iter->started))
3914 return;
3915
3916 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
3917 return;
3918
3919 if (cpumask_available(iter->started))
3920 cpumask_set_cpu(iter->cpu, iter->started);
3921
3922 /* Don't print started cpu buffer for the first entry of the trace */
3923 if (iter->idx > 1)
3924 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
3925 iter->cpu);
3926}
3927
3928static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
3929{
3930 struct trace_array *tr = iter->tr;
3931 struct trace_seq *s = &iter->seq;
3932 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
3933 struct trace_entry *entry;
3934 struct trace_event *event;
3935
3936 entry = iter->ent;
3937
3938 test_cpu_buff_start(iter);
3939
3940 event = ftrace_find_event(entry->type);
3941
3942 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3943 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
3944 trace_print_lat_context(iter);
3945 else
3946 trace_print_context(iter);
3947 }
3948
3949 if (trace_seq_has_overflowed(s))
3950 return TRACE_TYPE_PARTIAL_LINE;
3951
3952 if (event)
3953 return event->funcs->trace(iter, sym_flags, event);
3954
3955 trace_seq_printf(s, "Unknown type %d\n", entry->type);
3956
3957 return trace_handle_return(s);
3958}
3959
3960static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
3961{
3962 struct trace_array *tr = iter->tr;
3963 struct trace_seq *s = &iter->seq;
3964 struct trace_entry *entry;
3965 struct trace_event *event;
3966
3967 entry = iter->ent;
3968
3969 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
3970 trace_seq_printf(s, "%d %d %llu ",
3971 entry->pid, iter->cpu, iter->ts);
3972
3973 if (trace_seq_has_overflowed(s))
3974 return TRACE_TYPE_PARTIAL_LINE;
3975
3976 event = ftrace_find_event(entry->type);
3977 if (event)
3978 return event->funcs->raw(iter, 0, event);
3979
3980 trace_seq_printf(s, "%d ?\n", entry->type);
3981
3982 return trace_handle_return(s);
3983}
3984
3985static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
3986{
3987 struct trace_array *tr = iter->tr;
3988 struct trace_seq *s = &iter->seq;
3989 unsigned char newline = '\n';
3990 struct trace_entry *entry;
3991 struct trace_event *event;
3992
3993 entry = iter->ent;
3994
3995 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3996 SEQ_PUT_HEX_FIELD(s, entry->pid);
3997 SEQ_PUT_HEX_FIELD(s, iter->cpu);
3998 SEQ_PUT_HEX_FIELD(s, iter->ts);
3999 if (trace_seq_has_overflowed(s))
4000 return TRACE_TYPE_PARTIAL_LINE;
4001 }
4002
4003 event = ftrace_find_event(entry->type);
4004 if (event) {
4005 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4006 if (ret != TRACE_TYPE_HANDLED)
4007 return ret;
4008 }
4009
4010 SEQ_PUT_FIELD(s, newline);
4011
4012 return trace_handle_return(s);
4013}
4014
4015static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4016{
4017 struct trace_array *tr = iter->tr;
4018 struct trace_seq *s = &iter->seq;
4019 struct trace_entry *entry;
4020 struct trace_event *event;
4021
4022 entry = iter->ent;
4023
4024 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4025 SEQ_PUT_FIELD(s, entry->pid);
4026 SEQ_PUT_FIELD(s, iter->cpu);
4027 SEQ_PUT_FIELD(s, iter->ts);
4028 if (trace_seq_has_overflowed(s))
4029 return TRACE_TYPE_PARTIAL_LINE;
4030 }
4031
4032 event = ftrace_find_event(entry->type);
4033 return event ? event->funcs->binary(iter, 0, event) :
4034 TRACE_TYPE_HANDLED;
4035}
4036
4037int trace_empty(struct trace_iterator *iter)
4038{
4039 struct ring_buffer_iter *buf_iter;
4040 int cpu;
4041
4042 /* If we are looking at one CPU buffer, only check that one */
4043 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4044 cpu = iter->cpu_file;
4045 buf_iter = trace_buffer_iter(iter, cpu);
4046 if (buf_iter) {
4047 if (!ring_buffer_iter_empty(buf_iter))
4048 return 0;
4049 } else {
4050 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4051 return 0;
4052 }
4053 return 1;
4054 }
4055
4056 for_each_tracing_cpu(cpu) {
4057 buf_iter = trace_buffer_iter(iter, cpu);
4058 if (buf_iter) {
4059 if (!ring_buffer_iter_empty(buf_iter))
4060 return 0;
4061 } else {
4062 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4063 return 0;
4064 }
4065 }
4066
4067 return 1;
4068}
4069
4070/* Called with trace_event_read_lock() held. */
4071enum print_line_t print_trace_line(struct trace_iterator *iter)
4072{
4073 struct trace_array *tr = iter->tr;
4074 unsigned long trace_flags = tr->trace_flags;
4075 enum print_line_t ret;
4076
4077 if (iter->lost_events) {
4078 if (iter->lost_events == (unsigned long)-1)
4079 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4080 iter->cpu);
4081 else
4082 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4083 iter->cpu, iter->lost_events);
4084 if (trace_seq_has_overflowed(&iter->seq))
4085 return TRACE_TYPE_PARTIAL_LINE;
4086 }
4087
4088 if (iter->trace && iter->trace->print_line) {
4089 ret = iter->trace->print_line(iter);
4090 if (ret != TRACE_TYPE_UNHANDLED)
4091 return ret;
4092 }
4093
4094 if (iter->ent->type == TRACE_BPUTS &&
4095 trace_flags & TRACE_ITER_PRINTK &&
4096 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4097 return trace_print_bputs_msg_only(iter);
4098
4099 if (iter->ent->type == TRACE_BPRINT &&
4100 trace_flags & TRACE_ITER_PRINTK &&
4101 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4102 return trace_print_bprintk_msg_only(iter);
4103
4104 if (iter->ent->type == TRACE_PRINT &&
4105 trace_flags & TRACE_ITER_PRINTK &&
4106 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4107 return trace_print_printk_msg_only(iter);
4108
4109 if (trace_flags & TRACE_ITER_BIN)
4110 return print_bin_fmt(iter);
4111
4112 if (trace_flags & TRACE_ITER_HEX)
4113 return print_hex_fmt(iter);
4114
4115 if (trace_flags & TRACE_ITER_RAW)
4116 return print_raw_fmt(iter);
4117
4118 return print_trace_fmt(iter);
4119}
4120
4121void trace_latency_header(struct seq_file *m)
4122{
4123 struct trace_iterator *iter = m->private;
4124 struct trace_array *tr = iter->tr;
4125
4126 /* print nothing if the buffers are empty */
4127 if (trace_empty(iter))
4128 return;
4129
4130 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4131 print_trace_header(m, iter);
4132
4133 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4134 print_lat_help_header(m);
4135}
4136
4137void trace_default_header(struct seq_file *m)
4138{
4139 struct trace_iterator *iter = m->private;
4140 struct trace_array *tr = iter->tr;
4141 unsigned long trace_flags = tr->trace_flags;
4142
4143 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4144 return;
4145
4146 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4147 /* print nothing if the buffers are empty */
4148 if (trace_empty(iter))
4149 return;
4150 print_trace_header(m, iter);
4151 if (!(trace_flags & TRACE_ITER_VERBOSE))
4152 print_lat_help_header(m);
4153 } else {
4154 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4155 if (trace_flags & TRACE_ITER_IRQ_INFO)
4156 print_func_help_header_irq(iter->array_buffer,
4157 m, trace_flags);
4158 else
4159 print_func_help_header(iter->array_buffer, m,
4160 trace_flags);
4161 }
4162 }
4163}
4164
4165static void test_ftrace_alive(struct seq_file *m)
4166{
4167 if (!ftrace_is_dead())
4168 return;
4169 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4170 "# MAY BE MISSING FUNCTION EVENTS\n");
4171}
4172
4173#ifdef CONFIG_TRACER_MAX_TRACE
4174static void show_snapshot_main_help(struct seq_file *m)
4175{
4176 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4177 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4178 "# Takes a snapshot of the main buffer.\n"
4179 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4180 "# (Doesn't have to be '2' works with any number that\n"
4181 "# is not a '0' or '1')\n");
4182}
4183
4184static void show_snapshot_percpu_help(struct seq_file *m)
4185{
4186 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4187#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4188 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4189 "# Takes a snapshot of the main buffer for this cpu.\n");
4190#else
4191 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4192 "# Must use main snapshot file to allocate.\n");
4193#endif
4194 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4195 "# (Doesn't have to be '2' works with any number that\n"
4196 "# is not a '0' or '1')\n");
4197}
4198
4199static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4200{
4201 if (iter->tr->allocated_snapshot)
4202 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4203 else
4204 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4205
4206 seq_puts(m, "# Snapshot commands:\n");
4207 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4208 show_snapshot_main_help(m);
4209 else
4210 show_snapshot_percpu_help(m);
4211}
4212#else
4213/* Should never be called */
4214static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4215#endif
4216
4217static int s_show(struct seq_file *m, void *v)
4218{
4219 struct trace_iterator *iter = v;
4220 int ret;
4221
4222 if (iter->ent == NULL) {
4223 if (iter->tr) {
4224 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4225 seq_puts(m, "#\n");
4226 test_ftrace_alive(m);
4227 }
4228 if (iter->snapshot && trace_empty(iter))
4229 print_snapshot_help(m, iter);
4230 else if (iter->trace && iter->trace->print_header)
4231 iter->trace->print_header(m);
4232 else
4233 trace_default_header(m);
4234
4235 } else if (iter->leftover) {
4236 /*
4237 * If we filled the seq_file buffer earlier, we
4238 * want to just show it now.
4239 */
4240 ret = trace_print_seq(m, &iter->seq);
4241
4242 /* ret should this time be zero, but you never know */
4243 iter->leftover = ret;
4244
4245 } else {
4246 print_trace_line(iter);
4247 ret = trace_print_seq(m, &iter->seq);
4248 /*
4249 * If we overflow the seq_file buffer, then it will
4250 * ask us for this data again at start up.
4251 * Use that instead.
4252 * ret is 0 if seq_file write succeeded.
4253 * -1 otherwise.
4254 */
4255 iter->leftover = ret;
4256 }
4257
4258 return 0;
4259}
4260
4261/*
4262 * Should be used after trace_array_get(), trace_types_lock
4263 * ensures that i_cdev was already initialized.
4264 */
4265static inline int tracing_get_cpu(struct inode *inode)
4266{
4267 if (inode->i_cdev) /* See trace_create_cpu_file() */
4268 return (long)inode->i_cdev - 1;
4269 return RING_BUFFER_ALL_CPUS;
4270}
4271
4272static const struct seq_operations tracer_seq_ops = {
4273 .start = s_start,
4274 .next = s_next,
4275 .stop = s_stop,
4276 .show = s_show,
4277};
4278
4279static struct trace_iterator *
4280__tracing_open(struct inode *inode, struct file *file, bool snapshot)
4281{
4282 struct trace_array *tr = inode->i_private;
4283 struct trace_iterator *iter;
4284 int cpu;
4285
4286 if (tracing_disabled)
4287 return ERR_PTR(-ENODEV);
4288
4289 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4290 if (!iter)
4291 return ERR_PTR(-ENOMEM);
4292
4293 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4294 GFP_KERNEL);
4295 if (!iter->buffer_iter)
4296 goto release;
4297
4298 /*
4299 * trace_find_next_entry() may need to save off iter->ent.
4300 * It will place it into the iter->temp buffer. As most
4301 * events are less than 128, allocate a buffer of that size.
4302 * If one is greater, then trace_find_next_entry() will
4303 * allocate a new buffer to adjust for the bigger iter->ent.
4304 * It's not critical if it fails to get allocated here.
4305 */
4306 iter->temp = kmalloc(128, GFP_KERNEL);
4307 if (iter->temp)
4308 iter->temp_size = 128;
4309
4310 /*
4311 * We make a copy of the current tracer to avoid concurrent
4312 * changes on it while we are reading.
4313 */
4314 mutex_lock(&trace_types_lock);
4315 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4316 if (!iter->trace)
4317 goto fail;
4318
4319 *iter->trace = *tr->current_trace;
4320
4321 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4322 goto fail;
4323
4324 iter->tr = tr;
4325
4326#ifdef CONFIG_TRACER_MAX_TRACE
4327 /* Currently only the top directory has a snapshot */
4328 if (tr->current_trace->print_max || snapshot)
4329 iter->array_buffer = &tr->max_buffer;
4330 else
4331#endif
4332 iter->array_buffer = &tr->array_buffer;
4333 iter->snapshot = snapshot;
4334 iter->pos = -1;
4335 iter->cpu_file = tracing_get_cpu(inode);
4336 mutex_init(&iter->mutex);
4337
4338 /* Notify the tracer early; before we stop tracing. */
4339 if (iter->trace->open)
4340 iter->trace->open(iter);
4341
4342 /* Annotate start of buffers if we had overruns */
4343 if (ring_buffer_overruns(iter->array_buffer->buffer))
4344 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4345
4346 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4347 if (trace_clocks[tr->clock_id].in_ns)
4348 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4349
4350 /*
4351 * If pause-on-trace is enabled, then stop the trace while
4352 * dumping, unless this is the "snapshot" file
4353 */
4354 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4355 tracing_stop_tr(tr);
4356
4357 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4358 for_each_tracing_cpu(cpu) {
4359 iter->buffer_iter[cpu] =
4360 ring_buffer_read_prepare(iter->array_buffer->buffer,
4361 cpu, GFP_KERNEL);
4362 }
4363 ring_buffer_read_prepare_sync();
4364 for_each_tracing_cpu(cpu) {
4365 ring_buffer_read_start(iter->buffer_iter[cpu]);
4366 tracing_iter_reset(iter, cpu);
4367 }
4368 } else {
4369 cpu = iter->cpu_file;
4370 iter->buffer_iter[cpu] =
4371 ring_buffer_read_prepare(iter->array_buffer->buffer,
4372 cpu, GFP_KERNEL);
4373 ring_buffer_read_prepare_sync();
4374 ring_buffer_read_start(iter->buffer_iter[cpu]);
4375 tracing_iter_reset(iter, cpu);
4376 }
4377
4378 mutex_unlock(&trace_types_lock);
4379
4380 return iter;
4381
4382 fail:
4383 mutex_unlock(&trace_types_lock);
4384 kfree(iter->trace);
4385 kfree(iter->temp);
4386 kfree(iter->buffer_iter);
4387release:
4388 seq_release_private(inode, file);
4389 return ERR_PTR(-ENOMEM);
4390}
4391
4392int tracing_open_generic(struct inode *inode, struct file *filp)
4393{
4394 int ret;
4395
4396 ret = tracing_check_open_get_tr(NULL);
4397 if (ret)
4398 return ret;
4399
4400 filp->private_data = inode->i_private;
4401 return 0;
4402}
4403
4404bool tracing_is_disabled(void)
4405{
4406 return (tracing_disabled) ? true: false;
4407}
4408
4409/*
4410 * Open and update trace_array ref count.
4411 * Must have the current trace_array passed to it.
4412 */
4413int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4414{
4415 struct trace_array *tr = inode->i_private;
4416 int ret;
4417
4418 ret = tracing_check_open_get_tr(tr);
4419 if (ret)
4420 return ret;
4421
4422 filp->private_data = inode->i_private;
4423
4424 return 0;
4425}
4426
4427static int tracing_release(struct inode *inode, struct file *file)
4428{
4429 struct trace_array *tr = inode->i_private;
4430 struct seq_file *m = file->private_data;
4431 struct trace_iterator *iter;
4432 int cpu;
4433
4434 if (!(file->f_mode & FMODE_READ)) {
4435 trace_array_put(tr);
4436 return 0;
4437 }
4438
4439 /* Writes do not use seq_file */
4440 iter = m->private;
4441 mutex_lock(&trace_types_lock);
4442
4443 for_each_tracing_cpu(cpu) {
4444 if (iter->buffer_iter[cpu])
4445 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4446 }
4447
4448 if (iter->trace && iter->trace->close)
4449 iter->trace->close(iter);
4450
4451 if (!iter->snapshot && tr->stop_count)
4452 /* reenable tracing if it was previously enabled */
4453 tracing_start_tr(tr);
4454
4455 __trace_array_put(tr);
4456
4457 mutex_unlock(&trace_types_lock);
4458
4459 mutex_destroy(&iter->mutex);
4460 free_cpumask_var(iter->started);
4461 kfree(iter->temp);
4462 kfree(iter->trace);
4463 kfree(iter->buffer_iter);
4464 seq_release_private(inode, file);
4465
4466 return 0;
4467}
4468
4469static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4470{
4471 struct trace_array *tr = inode->i_private;
4472
4473 trace_array_put(tr);
4474 return 0;
4475}
4476
4477static int tracing_single_release_tr(struct inode *inode, struct file *file)
4478{
4479 struct trace_array *tr = inode->i_private;
4480
4481 trace_array_put(tr);
4482
4483 return single_release(inode, file);
4484}
4485
4486static int tracing_open(struct inode *inode, struct file *file)
4487{
4488 struct trace_array *tr = inode->i_private;
4489 struct trace_iterator *iter;
4490 int ret;
4491
4492 ret = tracing_check_open_get_tr(tr);
4493 if (ret)
4494 return ret;
4495
4496 /* If this file was open for write, then erase contents */
4497 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4498 int cpu = tracing_get_cpu(inode);
4499 struct array_buffer *trace_buf = &tr->array_buffer;
4500
4501#ifdef CONFIG_TRACER_MAX_TRACE
4502 if (tr->current_trace->print_max)
4503 trace_buf = &tr->max_buffer;
4504#endif
4505
4506 if (cpu == RING_BUFFER_ALL_CPUS)
4507 tracing_reset_online_cpus(trace_buf);
4508 else
4509 tracing_reset_cpu(trace_buf, cpu);
4510 }
4511
4512 if (file->f_mode & FMODE_READ) {
4513 iter = __tracing_open(inode, file, false);
4514 if (IS_ERR(iter))
4515 ret = PTR_ERR(iter);
4516 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4517 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4518 }
4519
4520 if (ret < 0)
4521 trace_array_put(tr);
4522
4523 return ret;
4524}
4525
4526/*
4527 * Some tracers are not suitable for instance buffers.
4528 * A tracer is always available for the global array (toplevel)
4529 * or if it explicitly states that it is.
4530 */
4531static bool
4532trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4533{
4534 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4535}
4536
4537/* Find the next tracer that this trace array may use */
4538static struct tracer *
4539get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4540{
4541 while (t && !trace_ok_for_array(t, tr))
4542 t = t->next;
4543
4544 return t;
4545}
4546
4547static void *
4548t_next(struct seq_file *m, void *v, loff_t *pos)
4549{
4550 struct trace_array *tr = m->private;
4551 struct tracer *t = v;
4552
4553 (*pos)++;
4554
4555 if (t)
4556 t = get_tracer_for_array(tr, t->next);
4557
4558 return t;
4559}
4560
4561static void *t_start(struct seq_file *m, loff_t *pos)
4562{
4563 struct trace_array *tr = m->private;
4564 struct tracer *t;
4565 loff_t l = 0;
4566
4567 mutex_lock(&trace_types_lock);
4568
4569 t = get_tracer_for_array(tr, trace_types);
4570 for (; t && l < *pos; t = t_next(m, t, &l))
4571 ;
4572
4573 return t;
4574}
4575
4576static void t_stop(struct seq_file *m, void *p)
4577{
4578 mutex_unlock(&trace_types_lock);
4579}
4580
4581static int t_show(struct seq_file *m, void *v)
4582{
4583 struct tracer *t = v;
4584
4585 if (!t)
4586 return 0;
4587
4588 seq_puts(m, t->name);
4589 if (t->next)
4590 seq_putc(m, ' ');
4591 else
4592 seq_putc(m, '\n');
4593
4594 return 0;
4595}
4596
4597static const struct seq_operations show_traces_seq_ops = {
4598 .start = t_start,
4599 .next = t_next,
4600 .stop = t_stop,
4601 .show = t_show,
4602};
4603
4604static int show_traces_open(struct inode *inode, struct file *file)
4605{
4606 struct trace_array *tr = inode->i_private;
4607 struct seq_file *m;
4608 int ret;
4609
4610 ret = tracing_check_open_get_tr(tr);
4611 if (ret)
4612 return ret;
4613
4614 ret = seq_open(file, &show_traces_seq_ops);
4615 if (ret) {
4616 trace_array_put(tr);
4617 return ret;
4618 }
4619
4620 m = file->private_data;
4621 m->private = tr;
4622
4623 return 0;
4624}
4625
4626static int show_traces_release(struct inode *inode, struct file *file)
4627{
4628 struct trace_array *tr = inode->i_private;
4629
4630 trace_array_put(tr);
4631 return seq_release(inode, file);
4632}
4633
4634static ssize_t
4635tracing_write_stub(struct file *filp, const char __user *ubuf,
4636 size_t count, loff_t *ppos)
4637{
4638 return count;
4639}
4640
4641loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
4642{
4643 int ret;
4644
4645 if (file->f_mode & FMODE_READ)
4646 ret = seq_lseek(file, offset, whence);
4647 else
4648 file->f_pos = ret = 0;
4649
4650 return ret;
4651}
4652
4653static const struct file_operations tracing_fops = {
4654 .open = tracing_open,
4655 .read = seq_read,
4656 .write = tracing_write_stub,
4657 .llseek = tracing_lseek,
4658 .release = tracing_release,
4659};
4660
4661static const struct file_operations show_traces_fops = {
4662 .open = show_traces_open,
4663 .read = seq_read,
4664 .llseek = seq_lseek,
4665 .release = show_traces_release,
4666};
4667
4668static ssize_t
4669tracing_cpumask_read(struct file *filp, char __user *ubuf,
4670 size_t count, loff_t *ppos)
4671{
4672 struct trace_array *tr = file_inode(filp)->i_private;
4673 char *mask_str;
4674 int len;
4675
4676 len = snprintf(NULL, 0, "%*pb\n",
4677 cpumask_pr_args(tr->tracing_cpumask)) + 1;
4678 mask_str = kmalloc(len, GFP_KERNEL);
4679 if (!mask_str)
4680 return -ENOMEM;
4681
4682 len = snprintf(mask_str, len, "%*pb\n",
4683 cpumask_pr_args(tr->tracing_cpumask));
4684 if (len >= count) {
4685 count = -EINVAL;
4686 goto out_err;
4687 }
4688 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
4689
4690out_err:
4691 kfree(mask_str);
4692
4693 return count;
4694}
4695
4696int tracing_set_cpumask(struct trace_array *tr,
4697 cpumask_var_t tracing_cpumask_new)
4698{
4699 int cpu;
4700
4701 if (!tr)
4702 return -EINVAL;
4703
4704 local_irq_disable();
4705 arch_spin_lock(&tr->max_lock);
4706 for_each_tracing_cpu(cpu) {
4707 /*
4708 * Increase/decrease the disabled counter if we are
4709 * about to flip a bit in the cpumask:
4710 */
4711 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4712 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4713 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
4714 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
4715 }
4716 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4717 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4718 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
4719 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
4720 }
4721 }
4722 arch_spin_unlock(&tr->max_lock);
4723 local_irq_enable();
4724
4725 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
4726
4727 return 0;
4728}
4729
4730static ssize_t
4731tracing_cpumask_write(struct file *filp, const char __user *ubuf,
4732 size_t count, loff_t *ppos)
4733{
4734 struct trace_array *tr = file_inode(filp)->i_private;
4735 cpumask_var_t tracing_cpumask_new;
4736 int err;
4737
4738 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
4739 return -ENOMEM;
4740
4741 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
4742 if (err)
4743 goto err_free;
4744
4745 err = tracing_set_cpumask(tr, tracing_cpumask_new);
4746 if (err)
4747 goto err_free;
4748
4749 free_cpumask_var(tracing_cpumask_new);
4750
4751 return count;
4752
4753err_free:
4754 free_cpumask_var(tracing_cpumask_new);
4755
4756 return err;
4757}
4758
4759static const struct file_operations tracing_cpumask_fops = {
4760 .open = tracing_open_generic_tr,
4761 .read = tracing_cpumask_read,
4762 .write = tracing_cpumask_write,
4763 .release = tracing_release_generic_tr,
4764 .llseek = generic_file_llseek,
4765};
4766
4767static int tracing_trace_options_show(struct seq_file *m, void *v)
4768{
4769 struct tracer_opt *trace_opts;
4770 struct trace_array *tr = m->private;
4771 u32 tracer_flags;
4772 int i;
4773
4774 mutex_lock(&trace_types_lock);
4775 tracer_flags = tr->current_trace->flags->val;
4776 trace_opts = tr->current_trace->flags->opts;
4777
4778 for (i = 0; trace_options[i]; i++) {
4779 if (tr->trace_flags & (1 << i))
4780 seq_printf(m, "%s\n", trace_options[i]);
4781 else
4782 seq_printf(m, "no%s\n", trace_options[i]);
4783 }
4784
4785 for (i = 0; trace_opts[i].name; i++) {
4786 if (tracer_flags & trace_opts[i].bit)
4787 seq_printf(m, "%s\n", trace_opts[i].name);
4788 else
4789 seq_printf(m, "no%s\n", trace_opts[i].name);
4790 }
4791 mutex_unlock(&trace_types_lock);
4792
4793 return 0;
4794}
4795
4796static int __set_tracer_option(struct trace_array *tr,
4797 struct tracer_flags *tracer_flags,
4798 struct tracer_opt *opts, int neg)
4799{
4800 struct tracer *trace = tracer_flags->trace;
4801 int ret;
4802
4803 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
4804 if (ret)
4805 return ret;
4806
4807 if (neg)
4808 tracer_flags->val &= ~opts->bit;
4809 else
4810 tracer_flags->val |= opts->bit;
4811 return 0;
4812}
4813
4814/* Try to assign a tracer specific option */
4815static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
4816{
4817 struct tracer *trace = tr->current_trace;
4818 struct tracer_flags *tracer_flags = trace->flags;
4819 struct tracer_opt *opts = NULL;
4820 int i;
4821
4822 for (i = 0; tracer_flags->opts[i].name; i++) {
4823 opts = &tracer_flags->opts[i];
4824
4825 if (strcmp(cmp, opts->name) == 0)
4826 return __set_tracer_option(tr, trace->flags, opts, neg);
4827 }
4828
4829 return -EINVAL;
4830}
4831
4832/* Some tracers require overwrite to stay enabled */
4833int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
4834{
4835 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
4836 return -1;
4837
4838 return 0;
4839}
4840
4841int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
4842{
4843 if ((mask == TRACE_ITER_RECORD_TGID) ||
4844 (mask == TRACE_ITER_RECORD_CMD))
4845 lockdep_assert_held(&event_mutex);
4846
4847 /* do nothing if flag is already set */
4848 if (!!(tr->trace_flags & mask) == !!enabled)
4849 return 0;
4850
4851 /* Give the tracer a chance to approve the change */
4852 if (tr->current_trace->flag_changed)
4853 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
4854 return -EINVAL;
4855
4856 if (enabled)
4857 tr->trace_flags |= mask;
4858 else
4859 tr->trace_flags &= ~mask;
4860
4861 if (mask == TRACE_ITER_RECORD_CMD)
4862 trace_event_enable_cmd_record(enabled);
4863
4864 if (mask == TRACE_ITER_RECORD_TGID) {
4865 if (!tgid_map)
4866 tgid_map = kvcalloc(PID_MAX_DEFAULT + 1,
4867 sizeof(*tgid_map),
4868 GFP_KERNEL);
4869 if (!tgid_map) {
4870 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
4871 return -ENOMEM;
4872 }
4873
4874 trace_event_enable_tgid_record(enabled);
4875 }
4876
4877 if (mask == TRACE_ITER_EVENT_FORK)
4878 trace_event_follow_fork(tr, enabled);
4879
4880 if (mask == TRACE_ITER_FUNC_FORK)
4881 ftrace_pid_follow_fork(tr, enabled);
4882
4883 if (mask == TRACE_ITER_OVERWRITE) {
4884 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
4885#ifdef CONFIG_TRACER_MAX_TRACE
4886 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
4887#endif
4888 }
4889
4890 if (mask == TRACE_ITER_PRINTK) {
4891 trace_printk_start_stop_comm(enabled);
4892 trace_printk_control(enabled);
4893 }
4894
4895 return 0;
4896}
4897
4898int trace_set_options(struct trace_array *tr, char *option)
4899{
4900 char *cmp;
4901 int neg = 0;
4902 int ret;
4903 size_t orig_len = strlen(option);
4904 int len;
4905
4906 cmp = strstrip(option);
4907
4908 len = str_has_prefix(cmp, "no");
4909 if (len)
4910 neg = 1;
4911
4912 cmp += len;
4913
4914 mutex_lock(&event_mutex);
4915 mutex_lock(&trace_types_lock);
4916
4917 ret = match_string(trace_options, -1, cmp);
4918 /* If no option could be set, test the specific tracer options */
4919 if (ret < 0)
4920 ret = set_tracer_option(tr, cmp, neg);
4921 else
4922 ret = set_tracer_flag(tr, 1 << ret, !neg);
4923
4924 mutex_unlock(&trace_types_lock);
4925 mutex_unlock(&event_mutex);
4926
4927 /*
4928 * If the first trailing whitespace is replaced with '\0' by strstrip,
4929 * turn it back into a space.
4930 */
4931 if (orig_len > strlen(option))
4932 option[strlen(option)] = ' ';
4933
4934 return ret;
4935}
4936
4937static void __init apply_trace_boot_options(void)
4938{
4939 char *buf = trace_boot_options_buf;
4940 char *option;
4941
4942 while (true) {
4943 option = strsep(&buf, ",");
4944
4945 if (!option)
4946 break;
4947
4948 if (*option)
4949 trace_set_options(&global_trace, option);
4950
4951 /* Put back the comma to allow this to be called again */
4952 if (buf)
4953 *(buf - 1) = ',';
4954 }
4955}
4956
4957static ssize_t
4958tracing_trace_options_write(struct file *filp, const char __user *ubuf,
4959 size_t cnt, loff_t *ppos)
4960{
4961 struct seq_file *m = filp->private_data;
4962 struct trace_array *tr = m->private;
4963 char buf[64];
4964 int ret;
4965
4966 if (cnt >= sizeof(buf))
4967 return -EINVAL;
4968
4969 if (copy_from_user(buf, ubuf, cnt))
4970 return -EFAULT;
4971
4972 buf[cnt] = 0;
4973
4974 ret = trace_set_options(tr, buf);
4975 if (ret < 0)
4976 return ret;
4977
4978 *ppos += cnt;
4979
4980 return cnt;
4981}
4982
4983static int tracing_trace_options_open(struct inode *inode, struct file *file)
4984{
4985 struct trace_array *tr = inode->i_private;
4986 int ret;
4987
4988 ret = tracing_check_open_get_tr(tr);
4989 if (ret)
4990 return ret;
4991
4992 ret = single_open(file, tracing_trace_options_show, inode->i_private);
4993 if (ret < 0)
4994 trace_array_put(tr);
4995
4996 return ret;
4997}
4998
4999static const struct file_operations tracing_iter_fops = {
5000 .open = tracing_trace_options_open,
5001 .read = seq_read,
5002 .llseek = seq_lseek,
5003 .release = tracing_single_release_tr,
5004 .write = tracing_trace_options_write,
5005};
5006
5007static const char readme_msg[] =
5008 "tracing mini-HOWTO:\n\n"
5009 "# echo 0 > tracing_on : quick way to disable tracing\n"
5010 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5011 " Important files:\n"
5012 " trace\t\t\t- The static contents of the buffer\n"
5013 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5014 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5015 " current_tracer\t- function and latency tracers\n"
5016 " available_tracers\t- list of configured tracers for current_tracer\n"
5017 " error_log\t- error log for failed commands (that support it)\n"
5018 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5019 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5020 " trace_clock\t\t-change the clock used to order events\n"
5021 " local: Per cpu clock but may not be synced across CPUs\n"
5022 " global: Synced across CPUs but slows tracing down.\n"
5023 " counter: Not a clock, but just an increment\n"
5024 " uptime: Jiffy counter from time of boot\n"
5025 " perf: Same clock that perf events use\n"
5026#ifdef CONFIG_X86_64
5027 " x86-tsc: TSC cycle counter\n"
5028#endif
5029 "\n timestamp_mode\t-view the mode used to timestamp events\n"
5030 " delta: Delta difference against a buffer-wide timestamp\n"
5031 " absolute: Absolute (standalone) timestamp\n"
5032 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5033 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5034 " tracing_cpumask\t- Limit which CPUs to trace\n"
5035 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5036 "\t\t\t Remove sub-buffer with rmdir\n"
5037 " trace_options\t\t- Set format or modify how tracing happens\n"
5038 "\t\t\t Disable an option by prefixing 'no' to the\n"
5039 "\t\t\t option name\n"
5040 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5041#ifdef CONFIG_DYNAMIC_FTRACE
5042 "\n available_filter_functions - list of functions that can be filtered on\n"
5043 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5044 "\t\t\t functions\n"
5045 "\t accepts: func_full_name or glob-matching-pattern\n"
5046 "\t modules: Can select a group via module\n"
5047 "\t Format: :mod:<module-name>\n"
5048 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5049 "\t triggers: a command to perform when function is hit\n"
5050 "\t Format: <function>:<trigger>[:count]\n"
5051 "\t trigger: traceon, traceoff\n"
5052 "\t\t enable_event:<system>:<event>\n"
5053 "\t\t disable_event:<system>:<event>\n"
5054#ifdef CONFIG_STACKTRACE
5055 "\t\t stacktrace\n"
5056#endif
5057#ifdef CONFIG_TRACER_SNAPSHOT
5058 "\t\t snapshot\n"
5059#endif
5060 "\t\t dump\n"
5061 "\t\t cpudump\n"
5062 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5063 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5064 "\t The first one will disable tracing every time do_fault is hit\n"
5065 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5066 "\t The first time do trap is hit and it disables tracing, the\n"
5067 "\t counter will decrement to 2. If tracing is already disabled,\n"
5068 "\t the counter will not decrement. It only decrements when the\n"
5069 "\t trigger did work\n"
5070 "\t To remove trigger without count:\n"
5071 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5072 "\t To remove trigger with a count:\n"
5073 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5074 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5075 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5076 "\t modules: Can select a group via module command :mod:\n"
5077 "\t Does not accept triggers\n"
5078#endif /* CONFIG_DYNAMIC_FTRACE */
5079#ifdef CONFIG_FUNCTION_TRACER
5080 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5081 "\t\t (function)\n"
5082 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5083 "\t\t (function)\n"
5084#endif
5085#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5086 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5087 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5088 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5089#endif
5090#ifdef CONFIG_TRACER_SNAPSHOT
5091 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5092 "\t\t\t snapshot buffer. Read the contents for more\n"
5093 "\t\t\t information\n"
5094#endif
5095#ifdef CONFIG_STACK_TRACER
5096 " stack_trace\t\t- Shows the max stack trace when active\n"
5097 " stack_max_size\t- Shows current max stack size that was traced\n"
5098 "\t\t\t Write into this file to reset the max size (trigger a\n"
5099 "\t\t\t new trace)\n"
5100#ifdef CONFIG_DYNAMIC_FTRACE
5101 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5102 "\t\t\t traces\n"
5103#endif
5104#endif /* CONFIG_STACK_TRACER */
5105#ifdef CONFIG_DYNAMIC_EVENTS
5106 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5107 "\t\t\t Write into this file to define/undefine new trace events.\n"
5108#endif
5109#ifdef CONFIG_KPROBE_EVENTS
5110 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5111 "\t\t\t Write into this file to define/undefine new trace events.\n"
5112#endif
5113#ifdef CONFIG_UPROBE_EVENTS
5114 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5115 "\t\t\t Write into this file to define/undefine new trace events.\n"
5116#endif
5117#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5118 "\t accepts: event-definitions (one definition per line)\n"
5119 "\t Format: p[:[<group>/]<event>] <place> [<args>]\n"
5120 "\t r[maxactive][:[<group>/]<event>] <place> [<args>]\n"
5121#ifdef CONFIG_HIST_TRIGGERS
5122 "\t s:[synthetic/]<event> <field> [<field>]\n"
5123#endif
5124 "\t -:[<group>/]<event>\n"
5125#ifdef CONFIG_KPROBE_EVENTS
5126 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5127 "place (kretprobe): [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5128#endif
5129#ifdef CONFIG_UPROBE_EVENTS
5130 " place (uprobe): <path>:<offset>[(ref_ctr_offset)]\n"
5131#endif
5132 "\t args: <name>=fetcharg[:type]\n"
5133 "\t fetcharg: %<register>, @<address>, @<symbol>[+|-<offset>],\n"
5134#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5135 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5136#else
5137 "\t $stack<index>, $stack, $retval, $comm,\n"
5138#endif
5139 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5140 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string, symbol,\n"
5141 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5142 "\t <type>\\[<array-size>\\]\n"
5143#ifdef CONFIG_HIST_TRIGGERS
5144 "\t field: <stype> <name>;\n"
5145 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5146 "\t [unsigned] char/int/long\n"
5147#endif
5148#endif
5149 " events/\t\t- Directory containing all trace event subsystems:\n"
5150 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5151 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5152 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5153 "\t\t\t events\n"
5154 " filter\t\t- If set, only events passing filter are traced\n"
5155 " events/<system>/<event>/\t- Directory containing control files for\n"
5156 "\t\t\t <event>:\n"
5157 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5158 " filter\t\t- If set, only events passing filter are traced\n"
5159 " trigger\t\t- If set, a command to perform when event is hit\n"
5160 "\t Format: <trigger>[:count][if <filter>]\n"
5161 "\t trigger: traceon, traceoff\n"
5162 "\t enable_event:<system>:<event>\n"
5163 "\t disable_event:<system>:<event>\n"
5164#ifdef CONFIG_HIST_TRIGGERS
5165 "\t enable_hist:<system>:<event>\n"
5166 "\t disable_hist:<system>:<event>\n"
5167#endif
5168#ifdef CONFIG_STACKTRACE
5169 "\t\t stacktrace\n"
5170#endif
5171#ifdef CONFIG_TRACER_SNAPSHOT
5172 "\t\t snapshot\n"
5173#endif
5174#ifdef CONFIG_HIST_TRIGGERS
5175 "\t\t hist (see below)\n"
5176#endif
5177 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5178 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5179 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5180 "\t events/block/block_unplug/trigger\n"
5181 "\t The first disables tracing every time block_unplug is hit.\n"
5182 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5183 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5184 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5185 "\t Like function triggers, the counter is only decremented if it\n"
5186 "\t enabled or disabled tracing.\n"
5187 "\t To remove a trigger without a count:\n"
5188 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5189 "\t To remove a trigger with a count:\n"
5190 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5191 "\t Filters can be ignored when removing a trigger.\n"
5192#ifdef CONFIG_HIST_TRIGGERS
5193 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5194 "\t Format: hist:keys=<field1[,field2,...]>\n"
5195 "\t [:values=<field1[,field2,...]>]\n"
5196 "\t [:sort=<field1[,field2,...]>]\n"
5197 "\t [:size=#entries]\n"
5198 "\t [:pause][:continue][:clear]\n"
5199 "\t [:name=histname1]\n"
5200 "\t [:<handler>.<action>]\n"
5201 "\t [if <filter>]\n\n"
5202 "\t When a matching event is hit, an entry is added to a hash\n"
5203 "\t table using the key(s) and value(s) named, and the value of a\n"
5204 "\t sum called 'hitcount' is incremented. Keys and values\n"
5205 "\t correspond to fields in the event's format description. Keys\n"
5206 "\t can be any field, or the special string 'stacktrace'.\n"
5207 "\t Compound keys consisting of up to two fields can be specified\n"
5208 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5209 "\t fields. Sort keys consisting of up to two fields can be\n"
5210 "\t specified using the 'sort' keyword. The sort direction can\n"
5211 "\t be modified by appending '.descending' or '.ascending' to a\n"
5212 "\t sort field. The 'size' parameter can be used to specify more\n"
5213 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5214 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5215 "\t its histogram data will be shared with other triggers of the\n"
5216 "\t same name, and trigger hits will update this common data.\n\n"
5217 "\t Reading the 'hist' file for the event will dump the hash\n"
5218 "\t table in its entirety to stdout. If there are multiple hist\n"
5219 "\t triggers attached to an event, there will be a table for each\n"
5220 "\t trigger in the output. The table displayed for a named\n"
5221 "\t trigger will be the same as any other instance having the\n"
5222 "\t same name. The default format used to display a given field\n"
5223 "\t can be modified by appending any of the following modifiers\n"
5224 "\t to the field name, as applicable:\n\n"
5225 "\t .hex display a number as a hex value\n"
5226 "\t .sym display an address as a symbol\n"
5227 "\t .sym-offset display an address as a symbol and offset\n"
5228 "\t .execname display a common_pid as a program name\n"
5229 "\t .syscall display a syscall id as a syscall name\n"
5230 "\t .log2 display log2 value rather than raw number\n"
5231 "\t .usecs display a common_timestamp in microseconds\n\n"
5232 "\t The 'pause' parameter can be used to pause an existing hist\n"
5233 "\t trigger or to start a hist trigger but not log any events\n"
5234 "\t until told to do so. 'continue' can be used to start or\n"
5235 "\t restart a paused hist trigger.\n\n"
5236 "\t The 'clear' parameter will clear the contents of a running\n"
5237 "\t hist trigger and leave its current paused/active state\n"
5238 "\t unchanged.\n\n"
5239 "\t The enable_hist and disable_hist triggers can be used to\n"
5240 "\t have one event conditionally start and stop another event's\n"
5241 "\t already-attached hist trigger. The syntax is analogous to\n"
5242 "\t the enable_event and disable_event triggers.\n\n"
5243 "\t Hist trigger handlers and actions are executed whenever a\n"
5244 "\t a histogram entry is added or updated. They take the form:\n\n"
5245 "\t <handler>.<action>\n\n"
5246 "\t The available handlers are:\n\n"
5247 "\t onmatch(matching.event) - invoke on addition or update\n"
5248 "\t onmax(var) - invoke if var exceeds current max\n"
5249 "\t onchange(var) - invoke action if var changes\n\n"
5250 "\t The available actions are:\n\n"
5251 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5252 "\t save(field,...) - save current event fields\n"
5253#ifdef CONFIG_TRACER_SNAPSHOT
5254 "\t snapshot() - snapshot the trace buffer\n"
5255#endif
5256#endif
5257;
5258
5259static ssize_t
5260tracing_readme_read(struct file *filp, char __user *ubuf,
5261 size_t cnt, loff_t *ppos)
5262{
5263 return simple_read_from_buffer(ubuf, cnt, ppos,
5264 readme_msg, strlen(readme_msg));
5265}
5266
5267static const struct file_operations tracing_readme_fops = {
5268 .open = tracing_open_generic,
5269 .read = tracing_readme_read,
5270 .llseek = generic_file_llseek,
5271};
5272
5273static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5274{
5275 int *ptr = v;
5276
5277 if (*pos || m->count)
5278 ptr++;
5279
5280 (*pos)++;
5281
5282 for (; ptr <= &tgid_map[PID_MAX_DEFAULT]; ptr++) {
5283 if (trace_find_tgid(*ptr))
5284 return ptr;
5285 }
5286
5287 return NULL;
5288}
5289
5290static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5291{
5292 void *v;
5293 loff_t l = 0;
5294
5295 if (!tgid_map)
5296 return NULL;
5297
5298 v = &tgid_map[0];
5299 while (l <= *pos) {
5300 v = saved_tgids_next(m, v, &l);
5301 if (!v)
5302 return NULL;
5303 }
5304
5305 return v;
5306}
5307
5308static void saved_tgids_stop(struct seq_file *m, void *v)
5309{
5310}
5311
5312static int saved_tgids_show(struct seq_file *m, void *v)
5313{
5314 int pid = (int *)v - tgid_map;
5315
5316 seq_printf(m, "%d %d\n", pid, trace_find_tgid(pid));
5317 return 0;
5318}
5319
5320static const struct seq_operations tracing_saved_tgids_seq_ops = {
5321 .start = saved_tgids_start,
5322 .stop = saved_tgids_stop,
5323 .next = saved_tgids_next,
5324 .show = saved_tgids_show,
5325};
5326
5327static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5328{
5329 int ret;
5330
5331 ret = tracing_check_open_get_tr(NULL);
5332 if (ret)
5333 return ret;
5334
5335 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5336}
5337
5338
5339static const struct file_operations tracing_saved_tgids_fops = {
5340 .open = tracing_saved_tgids_open,
5341 .read = seq_read,
5342 .llseek = seq_lseek,
5343 .release = seq_release,
5344};
5345
5346static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5347{
5348 unsigned int *ptr = v;
5349
5350 if (*pos || m->count)
5351 ptr++;
5352
5353 (*pos)++;
5354
5355 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5356 ptr++) {
5357 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5358 continue;
5359
5360 return ptr;
5361 }
5362
5363 return NULL;
5364}
5365
5366static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5367{
5368 void *v;
5369 loff_t l = 0;
5370
5371 preempt_disable();
5372 arch_spin_lock(&trace_cmdline_lock);
5373
5374 v = &savedcmd->map_cmdline_to_pid[0];
5375 while (l <= *pos) {
5376 v = saved_cmdlines_next(m, v, &l);
5377 if (!v)
5378 return NULL;
5379 }
5380
5381 return v;
5382}
5383
5384static void saved_cmdlines_stop(struct seq_file *m, void *v)
5385{
5386 arch_spin_unlock(&trace_cmdline_lock);
5387 preempt_enable();
5388}
5389
5390static int saved_cmdlines_show(struct seq_file *m, void *v)
5391{
5392 char buf[TASK_COMM_LEN];
5393 unsigned int *pid = v;
5394
5395 __trace_find_cmdline(*pid, buf);
5396 seq_printf(m, "%d %s\n", *pid, buf);
5397 return 0;
5398}
5399
5400static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5401 .start = saved_cmdlines_start,
5402 .next = saved_cmdlines_next,
5403 .stop = saved_cmdlines_stop,
5404 .show = saved_cmdlines_show,
5405};
5406
5407static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5408{
5409 int ret;
5410
5411 ret = tracing_check_open_get_tr(NULL);
5412 if (ret)
5413 return ret;
5414
5415 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5416}
5417
5418static const struct file_operations tracing_saved_cmdlines_fops = {
5419 .open = tracing_saved_cmdlines_open,
5420 .read = seq_read,
5421 .llseek = seq_lseek,
5422 .release = seq_release,
5423};
5424
5425static ssize_t
5426tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5427 size_t cnt, loff_t *ppos)
5428{
5429 char buf[64];
5430 int r;
5431
5432 arch_spin_lock(&trace_cmdline_lock);
5433 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5434 arch_spin_unlock(&trace_cmdline_lock);
5435
5436 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5437}
5438
5439static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5440{
5441 kfree(s->saved_cmdlines);
5442 kfree(s->map_cmdline_to_pid);
5443 kfree(s);
5444}
5445
5446static int tracing_resize_saved_cmdlines(unsigned int val)
5447{
5448 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5449
5450 s = kmalloc(sizeof(*s), GFP_KERNEL);
5451 if (!s)
5452 return -ENOMEM;
5453
5454 if (allocate_cmdlines_buffer(val, s) < 0) {
5455 kfree(s);
5456 return -ENOMEM;
5457 }
5458
5459 arch_spin_lock(&trace_cmdline_lock);
5460 savedcmd_temp = savedcmd;
5461 savedcmd = s;
5462 arch_spin_unlock(&trace_cmdline_lock);
5463 free_saved_cmdlines_buffer(savedcmd_temp);
5464
5465 return 0;
5466}
5467
5468static ssize_t
5469tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
5470 size_t cnt, loff_t *ppos)
5471{
5472 unsigned long val;
5473 int ret;
5474
5475 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5476 if (ret)
5477 return ret;
5478
5479 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
5480 if (!val || val > PID_MAX_DEFAULT)
5481 return -EINVAL;
5482
5483 ret = tracing_resize_saved_cmdlines((unsigned int)val);
5484 if (ret < 0)
5485 return ret;
5486
5487 *ppos += cnt;
5488
5489 return cnt;
5490}
5491
5492static const struct file_operations tracing_saved_cmdlines_size_fops = {
5493 .open = tracing_open_generic,
5494 .read = tracing_saved_cmdlines_size_read,
5495 .write = tracing_saved_cmdlines_size_write,
5496};
5497
5498#ifdef CONFIG_TRACE_EVAL_MAP_FILE
5499static union trace_eval_map_item *
5500update_eval_map(union trace_eval_map_item *ptr)
5501{
5502 if (!ptr->map.eval_string) {
5503 if (ptr->tail.next) {
5504 ptr = ptr->tail.next;
5505 /* Set ptr to the next real item (skip head) */
5506 ptr++;
5507 } else
5508 return NULL;
5509 }
5510 return ptr;
5511}
5512
5513static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5514{
5515 union trace_eval_map_item *ptr = v;
5516
5517 /*
5518 * Paranoid! If ptr points to end, we don't want to increment past it.
5519 * This really should never happen.
5520 */
5521 (*pos)++;
5522 ptr = update_eval_map(ptr);
5523 if (WARN_ON_ONCE(!ptr))
5524 return NULL;
5525
5526 ptr++;
5527 ptr = update_eval_map(ptr);
5528
5529 return ptr;
5530}
5531
5532static void *eval_map_start(struct seq_file *m, loff_t *pos)
5533{
5534 union trace_eval_map_item *v;
5535 loff_t l = 0;
5536
5537 mutex_lock(&trace_eval_mutex);
5538
5539 v = trace_eval_maps;
5540 if (v)
5541 v++;
5542
5543 while (v && l < *pos) {
5544 v = eval_map_next(m, v, &l);
5545 }
5546
5547 return v;
5548}
5549
5550static void eval_map_stop(struct seq_file *m, void *v)
5551{
5552 mutex_unlock(&trace_eval_mutex);
5553}
5554
5555static int eval_map_show(struct seq_file *m, void *v)
5556{
5557 union trace_eval_map_item *ptr = v;
5558
5559 seq_printf(m, "%s %ld (%s)\n",
5560 ptr->map.eval_string, ptr->map.eval_value,
5561 ptr->map.system);
5562
5563 return 0;
5564}
5565
5566static const struct seq_operations tracing_eval_map_seq_ops = {
5567 .start = eval_map_start,
5568 .next = eval_map_next,
5569 .stop = eval_map_stop,
5570 .show = eval_map_show,
5571};
5572
5573static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5574{
5575 int ret;
5576
5577 ret = tracing_check_open_get_tr(NULL);
5578 if (ret)
5579 return ret;
5580
5581 return seq_open(filp, &tracing_eval_map_seq_ops);
5582}
5583
5584static const struct file_operations tracing_eval_map_fops = {
5585 .open = tracing_eval_map_open,
5586 .read = seq_read,
5587 .llseek = seq_lseek,
5588 .release = seq_release,
5589};
5590
5591static inline union trace_eval_map_item *
5592trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
5593{
5594 /* Return tail of array given the head */
5595 return ptr + ptr->head.length + 1;
5596}
5597
5598static void
5599trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
5600 int len)
5601{
5602 struct trace_eval_map **stop;
5603 struct trace_eval_map **map;
5604 union trace_eval_map_item *map_array;
5605 union trace_eval_map_item *ptr;
5606
5607 stop = start + len;
5608
5609 /*
5610 * The trace_eval_maps contains the map plus a head and tail item,
5611 * where the head holds the module and length of array, and the
5612 * tail holds a pointer to the next list.
5613 */
5614 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
5615 if (!map_array) {
5616 pr_warn("Unable to allocate trace eval mapping\n");
5617 return;
5618 }
5619
5620 mutex_lock(&trace_eval_mutex);
5621
5622 if (!trace_eval_maps)
5623 trace_eval_maps = map_array;
5624 else {
5625 ptr = trace_eval_maps;
5626 for (;;) {
5627 ptr = trace_eval_jmp_to_tail(ptr);
5628 if (!ptr->tail.next)
5629 break;
5630 ptr = ptr->tail.next;
5631
5632 }
5633 ptr->tail.next = map_array;
5634 }
5635 map_array->head.mod = mod;
5636 map_array->head.length = len;
5637 map_array++;
5638
5639 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
5640 map_array->map = **map;
5641 map_array++;
5642 }
5643 memset(map_array, 0, sizeof(*map_array));
5644
5645 mutex_unlock(&trace_eval_mutex);
5646}
5647
5648static void trace_create_eval_file(struct dentry *d_tracer)
5649{
5650 trace_create_file("eval_map", 0444, d_tracer,
5651 NULL, &tracing_eval_map_fops);
5652}
5653
5654#else /* CONFIG_TRACE_EVAL_MAP_FILE */
5655static inline void trace_create_eval_file(struct dentry *d_tracer) { }
5656static inline void trace_insert_eval_map_file(struct module *mod,
5657 struct trace_eval_map **start, int len) { }
5658#endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
5659
5660static void trace_insert_eval_map(struct module *mod,
5661 struct trace_eval_map **start, int len)
5662{
5663 struct trace_eval_map **map;
5664
5665 if (len <= 0)
5666 return;
5667
5668 map = start;
5669
5670 trace_event_eval_update(map, len);
5671
5672 trace_insert_eval_map_file(mod, start, len);
5673}
5674
5675static ssize_t
5676tracing_set_trace_read(struct file *filp, char __user *ubuf,
5677 size_t cnt, loff_t *ppos)
5678{
5679 struct trace_array *tr = filp->private_data;
5680 char buf[MAX_TRACER_SIZE+2];
5681 int r;
5682
5683 mutex_lock(&trace_types_lock);
5684 r = sprintf(buf, "%s\n", tr->current_trace->name);
5685 mutex_unlock(&trace_types_lock);
5686
5687 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5688}
5689
5690int tracer_init(struct tracer *t, struct trace_array *tr)
5691{
5692 tracing_reset_online_cpus(&tr->array_buffer);
5693 return t->init(tr);
5694}
5695
5696static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
5697{
5698 int cpu;
5699
5700 for_each_tracing_cpu(cpu)
5701 per_cpu_ptr(buf->data, cpu)->entries = val;
5702}
5703
5704#ifdef CONFIG_TRACER_MAX_TRACE
5705/* resize @tr's buffer to the size of @size_tr's entries */
5706static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
5707 struct array_buffer *size_buf, int cpu_id)
5708{
5709 int cpu, ret = 0;
5710
5711 if (cpu_id == RING_BUFFER_ALL_CPUS) {
5712 for_each_tracing_cpu(cpu) {
5713 ret = ring_buffer_resize(trace_buf->buffer,
5714 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
5715 if (ret < 0)
5716 break;
5717 per_cpu_ptr(trace_buf->data, cpu)->entries =
5718 per_cpu_ptr(size_buf->data, cpu)->entries;
5719 }
5720 } else {
5721 ret = ring_buffer_resize(trace_buf->buffer,
5722 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
5723 if (ret == 0)
5724 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
5725 per_cpu_ptr(size_buf->data, cpu_id)->entries;
5726 }
5727
5728 return ret;
5729}
5730#endif /* CONFIG_TRACER_MAX_TRACE */
5731
5732static int __tracing_resize_ring_buffer(struct trace_array *tr,
5733 unsigned long size, int cpu)
5734{
5735 int ret;
5736
5737 /*
5738 * If kernel or user changes the size of the ring buffer
5739 * we use the size that was given, and we can forget about
5740 * expanding it later.
5741 */
5742 ring_buffer_expanded = true;
5743
5744 /* May be called before buffers are initialized */
5745 if (!tr->array_buffer.buffer)
5746 return 0;
5747
5748 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
5749 if (ret < 0)
5750 return ret;
5751
5752#ifdef CONFIG_TRACER_MAX_TRACE
5753 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
5754 !tr->current_trace->use_max_tr)
5755 goto out;
5756
5757 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
5758 if (ret < 0) {
5759 int r = resize_buffer_duplicate_size(&tr->array_buffer,
5760 &tr->array_buffer, cpu);
5761 if (r < 0) {
5762 /*
5763 * AARGH! We are left with different
5764 * size max buffer!!!!
5765 * The max buffer is our "snapshot" buffer.
5766 * When a tracer needs a snapshot (one of the
5767 * latency tracers), it swaps the max buffer
5768 * with the saved snap shot. We succeeded to
5769 * update the size of the main buffer, but failed to
5770 * update the size of the max buffer. But when we tried
5771 * to reset the main buffer to the original size, we
5772 * failed there too. This is very unlikely to
5773 * happen, but if it does, warn and kill all
5774 * tracing.
5775 */
5776 WARN_ON(1);
5777 tracing_disabled = 1;
5778 }
5779 return ret;
5780 }
5781
5782 if (cpu == RING_BUFFER_ALL_CPUS)
5783 set_buffer_entries(&tr->max_buffer, size);
5784 else
5785 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
5786
5787 out:
5788#endif /* CONFIG_TRACER_MAX_TRACE */
5789
5790 if (cpu == RING_BUFFER_ALL_CPUS)
5791 set_buffer_entries(&tr->array_buffer, size);
5792 else
5793 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
5794
5795 return ret;
5796}
5797
5798ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
5799 unsigned long size, int cpu_id)
5800{
5801 int ret = size;
5802
5803 mutex_lock(&trace_types_lock);
5804
5805 if (cpu_id != RING_BUFFER_ALL_CPUS) {
5806 /* make sure, this cpu is enabled in the mask */
5807 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
5808 ret = -EINVAL;
5809 goto out;
5810 }
5811 }
5812
5813 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
5814 if (ret < 0)
5815 ret = -ENOMEM;
5816
5817out:
5818 mutex_unlock(&trace_types_lock);
5819
5820 return ret;
5821}
5822
5823
5824/**
5825 * tracing_update_buffers - used by tracing facility to expand ring buffers
5826 *
5827 * To save on memory when the tracing is never used on a system with it
5828 * configured in. The ring buffers are set to a minimum size. But once
5829 * a user starts to use the tracing facility, then they need to grow
5830 * to their default size.
5831 *
5832 * This function is to be called when a tracer is about to be used.
5833 */
5834int tracing_update_buffers(void)
5835{
5836 int ret = 0;
5837
5838 mutex_lock(&trace_types_lock);
5839 if (!ring_buffer_expanded)
5840 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
5841 RING_BUFFER_ALL_CPUS);
5842 mutex_unlock(&trace_types_lock);
5843
5844 return ret;
5845}
5846
5847struct trace_option_dentry;
5848
5849static void
5850create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
5851
5852/*
5853 * Used to clear out the tracer before deletion of an instance.
5854 * Must have trace_types_lock held.
5855 */
5856static void tracing_set_nop(struct trace_array *tr)
5857{
5858 if (tr->current_trace == &nop_trace)
5859 return;
5860
5861 tr->current_trace->enabled--;
5862
5863 if (tr->current_trace->reset)
5864 tr->current_trace->reset(tr);
5865
5866 tr->current_trace = &nop_trace;
5867}
5868
5869static void add_tracer_options(struct trace_array *tr, struct tracer *t)
5870{
5871 /* Only enable if the directory has been created already. */
5872 if (!tr->dir)
5873 return;
5874
5875 create_trace_option_files(tr, t);
5876}
5877
5878int tracing_set_tracer(struct trace_array *tr, const char *buf)
5879{
5880 struct tracer *t;
5881#ifdef CONFIG_TRACER_MAX_TRACE
5882 bool had_max_tr;
5883#endif
5884 int ret = 0;
5885
5886 mutex_lock(&trace_types_lock);
5887
5888 if (!ring_buffer_expanded) {
5889 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
5890 RING_BUFFER_ALL_CPUS);
5891 if (ret < 0)
5892 goto out;
5893 ret = 0;
5894 }
5895
5896 for (t = trace_types; t; t = t->next) {
5897 if (strcmp(t->name, buf) == 0)
5898 break;
5899 }
5900 if (!t) {
5901 ret = -EINVAL;
5902 goto out;
5903 }
5904 if (t == tr->current_trace)
5905 goto out;
5906
5907#ifdef CONFIG_TRACER_SNAPSHOT
5908 if (t->use_max_tr) {
5909 arch_spin_lock(&tr->max_lock);
5910 if (tr->cond_snapshot)
5911 ret = -EBUSY;
5912 arch_spin_unlock(&tr->max_lock);
5913 if (ret)
5914 goto out;
5915 }
5916#endif
5917 /* Some tracers won't work on kernel command line */
5918 if (system_state < SYSTEM_RUNNING && t->noboot) {
5919 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
5920 t->name);
5921 goto out;
5922 }
5923
5924 /* Some tracers are only allowed for the top level buffer */
5925 if (!trace_ok_for_array(t, tr)) {
5926 ret = -EINVAL;
5927 goto out;
5928 }
5929
5930 /* If trace pipe files are being read, we can't change the tracer */
5931 if (tr->trace_ref) {
5932 ret = -EBUSY;
5933 goto out;
5934 }
5935
5936 trace_branch_disable();
5937
5938 tr->current_trace->enabled--;
5939
5940 if (tr->current_trace->reset)
5941 tr->current_trace->reset(tr);
5942
5943 /* Current trace needs to be nop_trace before synchronize_rcu */
5944 tr->current_trace = &nop_trace;
5945
5946#ifdef CONFIG_TRACER_MAX_TRACE
5947 had_max_tr = tr->allocated_snapshot;
5948
5949 if (had_max_tr && !t->use_max_tr) {
5950 /*
5951 * We need to make sure that the update_max_tr sees that
5952 * current_trace changed to nop_trace to keep it from
5953 * swapping the buffers after we resize it.
5954 * The update_max_tr is called from interrupts disabled
5955 * so a synchronized_sched() is sufficient.
5956 */
5957 synchronize_rcu();
5958 free_snapshot(tr);
5959 }
5960#endif
5961
5962#ifdef CONFIG_TRACER_MAX_TRACE
5963 if (t->use_max_tr && !had_max_tr) {
5964 ret = tracing_alloc_snapshot_instance(tr);
5965 if (ret < 0)
5966 goto out;
5967 }
5968#endif
5969
5970 if (t->init) {
5971 ret = tracer_init(t, tr);
5972 if (ret)
5973 goto out;
5974 }
5975
5976 tr->current_trace = t;
5977 tr->current_trace->enabled++;
5978 trace_branch_enable(tr);
5979 out:
5980 mutex_unlock(&trace_types_lock);
5981
5982 return ret;
5983}
5984
5985static ssize_t
5986tracing_set_trace_write(struct file *filp, const char __user *ubuf,
5987 size_t cnt, loff_t *ppos)
5988{
5989 struct trace_array *tr = filp->private_data;
5990 char buf[MAX_TRACER_SIZE+1];
5991 int i;
5992 size_t ret;
5993 int err;
5994
5995 ret = cnt;
5996
5997 if (cnt > MAX_TRACER_SIZE)
5998 cnt = MAX_TRACER_SIZE;
5999
6000 if (copy_from_user(buf, ubuf, cnt))
6001 return -EFAULT;
6002
6003 buf[cnt] = 0;
6004
6005 /* strip ending whitespace. */
6006 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
6007 buf[i] = 0;
6008
6009 err = tracing_set_tracer(tr, buf);
6010 if (err)
6011 return err;
6012
6013 *ppos += ret;
6014
6015 return ret;
6016}
6017
6018static ssize_t
6019tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6020 size_t cnt, loff_t *ppos)
6021{
6022 char buf[64];
6023 int r;
6024
6025 r = snprintf(buf, sizeof(buf), "%ld\n",
6026 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6027 if (r > sizeof(buf))
6028 r = sizeof(buf);
6029 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6030}
6031
6032static ssize_t
6033tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6034 size_t cnt, loff_t *ppos)
6035{
6036 unsigned long val;
6037 int ret;
6038
6039 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6040 if (ret)
6041 return ret;
6042
6043 *ptr = val * 1000;
6044
6045 return cnt;
6046}
6047
6048static ssize_t
6049tracing_thresh_read(struct file *filp, char __user *ubuf,
6050 size_t cnt, loff_t *ppos)
6051{
6052 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6053}
6054
6055static ssize_t
6056tracing_thresh_write(struct file *filp, const char __user *ubuf,
6057 size_t cnt, loff_t *ppos)
6058{
6059 struct trace_array *tr = filp->private_data;
6060 int ret;
6061
6062 mutex_lock(&trace_types_lock);
6063 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6064 if (ret < 0)
6065 goto out;
6066
6067 if (tr->current_trace->update_thresh) {
6068 ret = tr->current_trace->update_thresh(tr);
6069 if (ret < 0)
6070 goto out;
6071 }
6072
6073 ret = cnt;
6074out:
6075 mutex_unlock(&trace_types_lock);
6076
6077 return ret;
6078}
6079
6080#if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
6081
6082static ssize_t
6083tracing_max_lat_read(struct file *filp, char __user *ubuf,
6084 size_t cnt, loff_t *ppos)
6085{
6086 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6087}
6088
6089static ssize_t
6090tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6091 size_t cnt, loff_t *ppos)
6092{
6093 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6094}
6095
6096#endif
6097
6098static int tracing_open_pipe(struct inode *inode, struct file *filp)
6099{
6100 struct trace_array *tr = inode->i_private;
6101 struct trace_iterator *iter;
6102 int ret;
6103
6104 ret = tracing_check_open_get_tr(tr);
6105 if (ret)
6106 return ret;
6107
6108 mutex_lock(&trace_types_lock);
6109
6110 /* create a buffer to store the information to pass to userspace */
6111 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6112 if (!iter) {
6113 ret = -ENOMEM;
6114 __trace_array_put(tr);
6115 goto out;
6116 }
6117
6118 trace_seq_init(&iter->seq);
6119 iter->trace = tr->current_trace;
6120
6121 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6122 ret = -ENOMEM;
6123 goto fail;
6124 }
6125
6126 /* trace pipe does not show start of buffer */
6127 cpumask_setall(iter->started);
6128
6129 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6130 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6131
6132 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6133 if (trace_clocks[tr->clock_id].in_ns)
6134 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6135
6136 iter->tr = tr;
6137 iter->array_buffer = &tr->array_buffer;
6138 iter->cpu_file = tracing_get_cpu(inode);
6139 mutex_init(&iter->mutex);
6140 filp->private_data = iter;
6141
6142 if (iter->trace->pipe_open)
6143 iter->trace->pipe_open(iter);
6144
6145 nonseekable_open(inode, filp);
6146
6147 tr->trace_ref++;
6148out:
6149 mutex_unlock(&trace_types_lock);
6150 return ret;
6151
6152fail:
6153 kfree(iter);
6154 __trace_array_put(tr);
6155 mutex_unlock(&trace_types_lock);
6156 return ret;
6157}
6158
6159static int tracing_release_pipe(struct inode *inode, struct file *file)
6160{
6161 struct trace_iterator *iter = file->private_data;
6162 struct trace_array *tr = inode->i_private;
6163
6164 mutex_lock(&trace_types_lock);
6165
6166 tr->trace_ref--;
6167
6168 if (iter->trace->pipe_close)
6169 iter->trace->pipe_close(iter);
6170
6171 mutex_unlock(&trace_types_lock);
6172
6173 free_cpumask_var(iter->started);
6174 mutex_destroy(&iter->mutex);
6175 kfree(iter);
6176
6177 trace_array_put(tr);
6178
6179 return 0;
6180}
6181
6182static __poll_t
6183trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6184{
6185 struct trace_array *tr = iter->tr;
6186
6187 /* Iterators are static, they should be filled or empty */
6188 if (trace_buffer_iter(iter, iter->cpu_file))
6189 return EPOLLIN | EPOLLRDNORM;
6190
6191 if (tr->trace_flags & TRACE_ITER_BLOCK)
6192 /*
6193 * Always select as readable when in blocking mode
6194 */
6195 return EPOLLIN | EPOLLRDNORM;
6196 else
6197 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6198 filp, poll_table);
6199}
6200
6201static __poll_t
6202tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6203{
6204 struct trace_iterator *iter = filp->private_data;
6205
6206 return trace_poll(iter, filp, poll_table);
6207}
6208
6209/* Must be called with iter->mutex held. */
6210static int tracing_wait_pipe(struct file *filp)
6211{
6212 struct trace_iterator *iter = filp->private_data;
6213 int ret;
6214
6215 while (trace_empty(iter)) {
6216
6217 if ((filp->f_flags & O_NONBLOCK)) {
6218 return -EAGAIN;
6219 }
6220
6221 /*
6222 * We block until we read something and tracing is disabled.
6223 * We still block if tracing is disabled, but we have never
6224 * read anything. This allows a user to cat this file, and
6225 * then enable tracing. But after we have read something,
6226 * we give an EOF when tracing is again disabled.
6227 *
6228 * iter->pos will be 0 if we haven't read anything.
6229 */
6230 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6231 break;
6232
6233 mutex_unlock(&iter->mutex);
6234
6235 ret = wait_on_pipe(iter, 0);
6236
6237 mutex_lock(&iter->mutex);
6238
6239 if (ret)
6240 return ret;
6241 }
6242
6243 return 1;
6244}
6245
6246/*
6247 * Consumer reader.
6248 */
6249static ssize_t
6250tracing_read_pipe(struct file *filp, char __user *ubuf,
6251 size_t cnt, loff_t *ppos)
6252{
6253 struct trace_iterator *iter = filp->private_data;
6254 ssize_t sret;
6255
6256 /*
6257 * Avoid more than one consumer on a single file descriptor
6258 * This is just a matter of traces coherency, the ring buffer itself
6259 * is protected.
6260 */
6261 mutex_lock(&iter->mutex);
6262
6263 /* return any leftover data */
6264 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6265 if (sret != -EBUSY)
6266 goto out;
6267
6268 trace_seq_init(&iter->seq);
6269
6270 if (iter->trace->read) {
6271 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6272 if (sret)
6273 goto out;
6274 }
6275
6276waitagain:
6277 sret = tracing_wait_pipe(filp);
6278 if (sret <= 0)
6279 goto out;
6280
6281 /* stop when tracing is finished */
6282 if (trace_empty(iter)) {
6283 sret = 0;
6284 goto out;
6285 }
6286
6287 if (cnt >= PAGE_SIZE)
6288 cnt = PAGE_SIZE - 1;
6289
6290 /* reset all but tr, trace, and overruns */
6291 memset(&iter->seq, 0,
6292 sizeof(struct trace_iterator) -
6293 offsetof(struct trace_iterator, seq));
6294 cpumask_clear(iter->started);
6295 trace_seq_init(&iter->seq);
6296 iter->pos = -1;
6297
6298 trace_event_read_lock();
6299 trace_access_lock(iter->cpu_file);
6300 while (trace_find_next_entry_inc(iter) != NULL) {
6301 enum print_line_t ret;
6302 int save_len = iter->seq.seq.len;
6303
6304 ret = print_trace_line(iter);
6305 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6306 /* don't print partial lines */
6307 iter->seq.seq.len = save_len;
6308 break;
6309 }
6310 if (ret != TRACE_TYPE_NO_CONSUME)
6311 trace_consume(iter);
6312
6313 if (trace_seq_used(&iter->seq) >= cnt)
6314 break;
6315
6316 /*
6317 * Setting the full flag means we reached the trace_seq buffer
6318 * size and we should leave by partial output condition above.
6319 * One of the trace_seq_* functions is not used properly.
6320 */
6321 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6322 iter->ent->type);
6323 }
6324 trace_access_unlock(iter->cpu_file);
6325 trace_event_read_unlock();
6326
6327 /* Now copy what we have to the user */
6328 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6329 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6330 trace_seq_init(&iter->seq);
6331
6332 /*
6333 * If there was nothing to send to user, in spite of consuming trace
6334 * entries, go back to wait for more entries.
6335 */
6336 if (sret == -EBUSY)
6337 goto waitagain;
6338
6339out:
6340 mutex_unlock(&iter->mutex);
6341
6342 return sret;
6343}
6344
6345static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6346 unsigned int idx)
6347{
6348 __free_page(spd->pages[idx]);
6349}
6350
6351static size_t
6352tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6353{
6354 size_t count;
6355 int save_len;
6356 int ret;
6357
6358 /* Seq buffer is page-sized, exactly what we need. */
6359 for (;;) {
6360 save_len = iter->seq.seq.len;
6361 ret = print_trace_line(iter);
6362
6363 if (trace_seq_has_overflowed(&iter->seq)) {
6364 iter->seq.seq.len = save_len;
6365 break;
6366 }
6367
6368 /*
6369 * This should not be hit, because it should only
6370 * be set if the iter->seq overflowed. But check it
6371 * anyway to be safe.
6372 */
6373 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6374 iter->seq.seq.len = save_len;
6375 break;
6376 }
6377
6378 count = trace_seq_used(&iter->seq) - save_len;
6379 if (rem < count) {
6380 rem = 0;
6381 iter->seq.seq.len = save_len;
6382 break;
6383 }
6384
6385 if (ret != TRACE_TYPE_NO_CONSUME)
6386 trace_consume(iter);
6387 rem -= count;
6388 if (!trace_find_next_entry_inc(iter)) {
6389 rem = 0;
6390 iter->ent = NULL;
6391 break;
6392 }
6393 }
6394
6395 return rem;
6396}
6397
6398static ssize_t tracing_splice_read_pipe(struct file *filp,
6399 loff_t *ppos,
6400 struct pipe_inode_info *pipe,
6401 size_t len,
6402 unsigned int flags)
6403{
6404 struct page *pages_def[PIPE_DEF_BUFFERS];
6405 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6406 struct trace_iterator *iter = filp->private_data;
6407 struct splice_pipe_desc spd = {
6408 .pages = pages_def,
6409 .partial = partial_def,
6410 .nr_pages = 0, /* This gets updated below. */
6411 .nr_pages_max = PIPE_DEF_BUFFERS,
6412 .ops = &default_pipe_buf_ops,
6413 .spd_release = tracing_spd_release_pipe,
6414 };
6415 ssize_t ret;
6416 size_t rem;
6417 unsigned int i;
6418
6419 if (splice_grow_spd(pipe, &spd))
6420 return -ENOMEM;
6421
6422 mutex_lock(&iter->mutex);
6423
6424 if (iter->trace->splice_read) {
6425 ret = iter->trace->splice_read(iter, filp,
6426 ppos, pipe, len, flags);
6427 if (ret)
6428 goto out_err;
6429 }
6430
6431 ret = tracing_wait_pipe(filp);
6432 if (ret <= 0)
6433 goto out_err;
6434
6435 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6436 ret = -EFAULT;
6437 goto out_err;
6438 }
6439
6440 trace_event_read_lock();
6441 trace_access_lock(iter->cpu_file);
6442
6443 /* Fill as many pages as possible. */
6444 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6445 spd.pages[i] = alloc_page(GFP_KERNEL);
6446 if (!spd.pages[i])
6447 break;
6448
6449 rem = tracing_fill_pipe_page(rem, iter);
6450
6451 /* Copy the data into the page, so we can start over. */
6452 ret = trace_seq_to_buffer(&iter->seq,
6453 page_address(spd.pages[i]),
6454 trace_seq_used(&iter->seq));
6455 if (ret < 0) {
6456 __free_page(spd.pages[i]);
6457 break;
6458 }
6459 spd.partial[i].offset = 0;
6460 spd.partial[i].len = trace_seq_used(&iter->seq);
6461
6462 trace_seq_init(&iter->seq);
6463 }
6464
6465 trace_access_unlock(iter->cpu_file);
6466 trace_event_read_unlock();
6467 mutex_unlock(&iter->mutex);
6468
6469 spd.nr_pages = i;
6470
6471 if (i)
6472 ret = splice_to_pipe(pipe, &spd);
6473 else
6474 ret = 0;
6475out:
6476 splice_shrink_spd(&spd);
6477 return ret;
6478
6479out_err:
6480 mutex_unlock(&iter->mutex);
6481 goto out;
6482}
6483
6484static ssize_t
6485tracing_entries_read(struct file *filp, char __user *ubuf,
6486 size_t cnt, loff_t *ppos)
6487{
6488 struct inode *inode = file_inode(filp);
6489 struct trace_array *tr = inode->i_private;
6490 int cpu = tracing_get_cpu(inode);
6491 char buf[64];
6492 int r = 0;
6493 ssize_t ret;
6494
6495 mutex_lock(&trace_types_lock);
6496
6497 if (cpu == RING_BUFFER_ALL_CPUS) {
6498 int cpu, buf_size_same;
6499 unsigned long size;
6500
6501 size = 0;
6502 buf_size_same = 1;
6503 /* check if all cpu sizes are same */
6504 for_each_tracing_cpu(cpu) {
6505 /* fill in the size from first enabled cpu */
6506 if (size == 0)
6507 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6508 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6509 buf_size_same = 0;
6510 break;
6511 }
6512 }
6513
6514 if (buf_size_same) {
6515 if (!ring_buffer_expanded)
6516 r = sprintf(buf, "%lu (expanded: %lu)\n",
6517 size >> 10,
6518 trace_buf_size >> 10);
6519 else
6520 r = sprintf(buf, "%lu\n", size >> 10);
6521 } else
6522 r = sprintf(buf, "X\n");
6523 } else
6524 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6525
6526 mutex_unlock(&trace_types_lock);
6527
6528 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6529 return ret;
6530}
6531
6532static ssize_t
6533tracing_entries_write(struct file *filp, const char __user *ubuf,
6534 size_t cnt, loff_t *ppos)
6535{
6536 struct inode *inode = file_inode(filp);
6537 struct trace_array *tr = inode->i_private;
6538 unsigned long val;
6539 int ret;
6540
6541 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6542 if (ret)
6543 return ret;
6544
6545 /* must have at least 1 entry */
6546 if (!val)
6547 return -EINVAL;
6548
6549 /* value is in KB */
6550 val <<= 10;
6551 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6552 if (ret < 0)
6553 return ret;
6554
6555 *ppos += cnt;
6556
6557 return cnt;
6558}
6559
6560static ssize_t
6561tracing_total_entries_read(struct file *filp, char __user *ubuf,
6562 size_t cnt, loff_t *ppos)
6563{
6564 struct trace_array *tr = filp->private_data;
6565 char buf[64];
6566 int r, cpu;
6567 unsigned long size = 0, expanded_size = 0;
6568
6569 mutex_lock(&trace_types_lock);
6570 for_each_tracing_cpu(cpu) {
6571 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6572 if (!ring_buffer_expanded)
6573 expanded_size += trace_buf_size >> 10;
6574 }
6575 if (ring_buffer_expanded)
6576 r = sprintf(buf, "%lu\n", size);
6577 else
6578 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6579 mutex_unlock(&trace_types_lock);
6580
6581 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6582}
6583
6584static ssize_t
6585tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
6586 size_t cnt, loff_t *ppos)
6587{
6588 /*
6589 * There is no need to read what the user has written, this function
6590 * is just to make sure that there is no error when "echo" is used
6591 */
6592
6593 *ppos += cnt;
6594
6595 return cnt;
6596}
6597
6598static int
6599tracing_free_buffer_release(struct inode *inode, struct file *filp)
6600{
6601 struct trace_array *tr = inode->i_private;
6602
6603 /* disable tracing ? */
6604 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
6605 tracer_tracing_off(tr);
6606 /* resize the ring buffer to 0 */
6607 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
6608
6609 trace_array_put(tr);
6610
6611 return 0;
6612}
6613
6614static ssize_t
6615tracing_mark_write(struct file *filp, const char __user *ubuf,
6616 size_t cnt, loff_t *fpos)
6617{
6618 struct trace_array *tr = filp->private_data;
6619 struct ring_buffer_event *event;
6620 enum event_trigger_type tt = ETT_NONE;
6621 struct trace_buffer *buffer;
6622 struct print_entry *entry;
6623 unsigned long irq_flags;
6624 ssize_t written;
6625 int size;
6626 int len;
6627
6628/* Used in tracing_mark_raw_write() as well */
6629#define FAULTED_STR "<faulted>"
6630#define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
6631
6632 if (tracing_disabled)
6633 return -EINVAL;
6634
6635 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6636 return -EINVAL;
6637
6638 if (cnt > TRACE_BUF_SIZE)
6639 cnt = TRACE_BUF_SIZE;
6640
6641 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6642
6643 local_save_flags(irq_flags);
6644 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
6645
6646 /* If less than "<faulted>", then make sure we can still add that */
6647 if (cnt < FAULTED_SIZE)
6648 size += FAULTED_SIZE - cnt;
6649
6650 buffer = tr->array_buffer.buffer;
6651 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
6652 irq_flags, preempt_count());
6653 if (unlikely(!event))
6654 /* Ring buffer disabled, return as if not open for write */
6655 return -EBADF;
6656
6657 entry = ring_buffer_event_data(event);
6658 entry->ip = _THIS_IP_;
6659
6660 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
6661 if (len) {
6662 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6663 cnt = FAULTED_SIZE;
6664 written = -EFAULT;
6665 } else
6666 written = cnt;
6667 len = cnt;
6668
6669 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
6670 /* do not add \n before testing triggers, but add \0 */
6671 entry->buf[cnt] = '\0';
6672 tt = event_triggers_call(tr->trace_marker_file, entry, event);
6673 }
6674
6675 if (entry->buf[cnt - 1] != '\n') {
6676 entry->buf[cnt] = '\n';
6677 entry->buf[cnt + 1] = '\0';
6678 } else
6679 entry->buf[cnt] = '\0';
6680
6681 __buffer_unlock_commit(buffer, event);
6682
6683 if (tt)
6684 event_triggers_post_call(tr->trace_marker_file, tt);
6685
6686 if (written > 0)
6687 *fpos += written;
6688
6689 return written;
6690}
6691
6692/* Limit it for now to 3K (including tag) */
6693#define RAW_DATA_MAX_SIZE (1024*3)
6694
6695static ssize_t
6696tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
6697 size_t cnt, loff_t *fpos)
6698{
6699 struct trace_array *tr = filp->private_data;
6700 struct ring_buffer_event *event;
6701 struct trace_buffer *buffer;
6702 struct raw_data_entry *entry;
6703 unsigned long irq_flags;
6704 ssize_t written;
6705 int size;
6706 int len;
6707
6708#define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
6709
6710 if (tracing_disabled)
6711 return -EINVAL;
6712
6713 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6714 return -EINVAL;
6715
6716 /* The marker must at least have a tag id */
6717 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
6718 return -EINVAL;
6719
6720 if (cnt > TRACE_BUF_SIZE)
6721 cnt = TRACE_BUF_SIZE;
6722
6723 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6724
6725 local_save_flags(irq_flags);
6726 size = sizeof(*entry) + cnt;
6727 if (cnt < FAULT_SIZE_ID)
6728 size += FAULT_SIZE_ID - cnt;
6729
6730 buffer = tr->array_buffer.buffer;
6731 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
6732 irq_flags, preempt_count());
6733 if (!event)
6734 /* Ring buffer disabled, return as if not open for write */
6735 return -EBADF;
6736
6737 entry = ring_buffer_event_data(event);
6738
6739 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
6740 if (len) {
6741 entry->id = -1;
6742 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6743 written = -EFAULT;
6744 } else
6745 written = cnt;
6746
6747 __buffer_unlock_commit(buffer, event);
6748
6749 if (written > 0)
6750 *fpos += written;
6751
6752 return written;
6753}
6754
6755static int tracing_clock_show(struct seq_file *m, void *v)
6756{
6757 struct trace_array *tr = m->private;
6758 int i;
6759
6760 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
6761 seq_printf(m,
6762 "%s%s%s%s", i ? " " : "",
6763 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
6764 i == tr->clock_id ? "]" : "");
6765 seq_putc(m, '\n');
6766
6767 return 0;
6768}
6769
6770int tracing_set_clock(struct trace_array *tr, const char *clockstr)
6771{
6772 int i;
6773
6774 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
6775 if (strcmp(trace_clocks[i].name, clockstr) == 0)
6776 break;
6777 }
6778 if (i == ARRAY_SIZE(trace_clocks))
6779 return -EINVAL;
6780
6781 mutex_lock(&trace_types_lock);
6782
6783 tr->clock_id = i;
6784
6785 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
6786
6787 /*
6788 * New clock may not be consistent with the previous clock.
6789 * Reset the buffer so that it doesn't have incomparable timestamps.
6790 */
6791 tracing_reset_online_cpus(&tr->array_buffer);
6792
6793#ifdef CONFIG_TRACER_MAX_TRACE
6794 if (tr->max_buffer.buffer)
6795 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
6796 tracing_reset_online_cpus(&tr->max_buffer);
6797#endif
6798
6799 mutex_unlock(&trace_types_lock);
6800
6801 return 0;
6802}
6803
6804static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
6805 size_t cnt, loff_t *fpos)
6806{
6807 struct seq_file *m = filp->private_data;
6808 struct trace_array *tr = m->private;
6809 char buf[64];
6810 const char *clockstr;
6811 int ret;
6812
6813 if (cnt >= sizeof(buf))
6814 return -EINVAL;
6815
6816 if (copy_from_user(buf, ubuf, cnt))
6817 return -EFAULT;
6818
6819 buf[cnt] = 0;
6820
6821 clockstr = strstrip(buf);
6822
6823 ret = tracing_set_clock(tr, clockstr);
6824 if (ret)
6825 return ret;
6826
6827 *fpos += cnt;
6828
6829 return cnt;
6830}
6831
6832static int tracing_clock_open(struct inode *inode, struct file *file)
6833{
6834 struct trace_array *tr = inode->i_private;
6835 int ret;
6836
6837 ret = tracing_check_open_get_tr(tr);
6838 if (ret)
6839 return ret;
6840
6841 ret = single_open(file, tracing_clock_show, inode->i_private);
6842 if (ret < 0)
6843 trace_array_put(tr);
6844
6845 return ret;
6846}
6847
6848static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
6849{
6850 struct trace_array *tr = m->private;
6851
6852 mutex_lock(&trace_types_lock);
6853
6854 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
6855 seq_puts(m, "delta [absolute]\n");
6856 else
6857 seq_puts(m, "[delta] absolute\n");
6858
6859 mutex_unlock(&trace_types_lock);
6860
6861 return 0;
6862}
6863
6864static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
6865{
6866 struct trace_array *tr = inode->i_private;
6867 int ret;
6868
6869 ret = tracing_check_open_get_tr(tr);
6870 if (ret)
6871 return ret;
6872
6873 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
6874 if (ret < 0)
6875 trace_array_put(tr);
6876
6877 return ret;
6878}
6879
6880int tracing_set_time_stamp_abs(struct trace_array *tr, bool abs)
6881{
6882 int ret = 0;
6883
6884 mutex_lock(&trace_types_lock);
6885
6886 if (abs && tr->time_stamp_abs_ref++)
6887 goto out;
6888
6889 if (!abs) {
6890 if (WARN_ON_ONCE(!tr->time_stamp_abs_ref)) {
6891 ret = -EINVAL;
6892 goto out;
6893 }
6894
6895 if (--tr->time_stamp_abs_ref)
6896 goto out;
6897 }
6898
6899 ring_buffer_set_time_stamp_abs(tr->array_buffer.buffer, abs);
6900
6901#ifdef CONFIG_TRACER_MAX_TRACE
6902 if (tr->max_buffer.buffer)
6903 ring_buffer_set_time_stamp_abs(tr->max_buffer.buffer, abs);
6904#endif
6905 out:
6906 mutex_unlock(&trace_types_lock);
6907
6908 return ret;
6909}
6910
6911struct ftrace_buffer_info {
6912 struct trace_iterator iter;
6913 void *spare;
6914 unsigned int spare_cpu;
6915 unsigned int read;
6916};
6917
6918#ifdef CONFIG_TRACER_SNAPSHOT
6919static int tracing_snapshot_open(struct inode *inode, struct file *file)
6920{
6921 struct trace_array *tr = inode->i_private;
6922 struct trace_iterator *iter;
6923 struct seq_file *m;
6924 int ret;
6925
6926 ret = tracing_check_open_get_tr(tr);
6927 if (ret)
6928 return ret;
6929
6930 if (file->f_mode & FMODE_READ) {
6931 iter = __tracing_open(inode, file, true);
6932 if (IS_ERR(iter))
6933 ret = PTR_ERR(iter);
6934 } else {
6935 /* Writes still need the seq_file to hold the private data */
6936 ret = -ENOMEM;
6937 m = kzalloc(sizeof(*m), GFP_KERNEL);
6938 if (!m)
6939 goto out;
6940 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6941 if (!iter) {
6942 kfree(m);
6943 goto out;
6944 }
6945 ret = 0;
6946
6947 iter->tr = tr;
6948 iter->array_buffer = &tr->max_buffer;
6949 iter->cpu_file = tracing_get_cpu(inode);
6950 m->private = iter;
6951 file->private_data = m;
6952 }
6953out:
6954 if (ret < 0)
6955 trace_array_put(tr);
6956
6957 return ret;
6958}
6959
6960static ssize_t
6961tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
6962 loff_t *ppos)
6963{
6964 struct seq_file *m = filp->private_data;
6965 struct trace_iterator *iter = m->private;
6966 struct trace_array *tr = iter->tr;
6967 unsigned long val;
6968 int ret;
6969
6970 ret = tracing_update_buffers();
6971 if (ret < 0)
6972 return ret;
6973
6974 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6975 if (ret)
6976 return ret;
6977
6978 mutex_lock(&trace_types_lock);
6979
6980 if (tr->current_trace->use_max_tr) {
6981 ret = -EBUSY;
6982 goto out;
6983 }
6984
6985 arch_spin_lock(&tr->max_lock);
6986 if (tr->cond_snapshot)
6987 ret = -EBUSY;
6988 arch_spin_unlock(&tr->max_lock);
6989 if (ret)
6990 goto out;
6991
6992 switch (val) {
6993 case 0:
6994 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
6995 ret = -EINVAL;
6996 break;
6997 }
6998 if (tr->allocated_snapshot)
6999 free_snapshot(tr);
7000 break;
7001 case 1:
7002/* Only allow per-cpu swap if the ring buffer supports it */
7003#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7004 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7005 ret = -EINVAL;
7006 break;
7007 }
7008#endif
7009 if (tr->allocated_snapshot)
7010 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7011 &tr->array_buffer, iter->cpu_file);
7012 else
7013 ret = tracing_alloc_snapshot_instance(tr);
7014 if (ret < 0)
7015 break;
7016 local_irq_disable();
7017 /* Now, we're going to swap */
7018 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7019 update_max_tr(tr, current, smp_processor_id(), NULL);
7020 else
7021 update_max_tr_single(tr, current, iter->cpu_file);
7022 local_irq_enable();
7023 break;
7024 default:
7025 if (tr->allocated_snapshot) {
7026 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7027 tracing_reset_online_cpus(&tr->max_buffer);
7028 else
7029 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7030 }
7031 break;
7032 }
7033
7034 if (ret >= 0) {
7035 *ppos += cnt;
7036 ret = cnt;
7037 }
7038out:
7039 mutex_unlock(&trace_types_lock);
7040 return ret;
7041}
7042
7043static int tracing_snapshot_release(struct inode *inode, struct file *file)
7044{
7045 struct seq_file *m = file->private_data;
7046 int ret;
7047
7048 ret = tracing_release(inode, file);
7049
7050 if (file->f_mode & FMODE_READ)
7051 return ret;
7052
7053 /* If write only, the seq_file is just a stub */
7054 if (m)
7055 kfree(m->private);
7056 kfree(m);
7057
7058 return 0;
7059}
7060
7061static int tracing_buffers_open(struct inode *inode, struct file *filp);
7062static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7063 size_t count, loff_t *ppos);
7064static int tracing_buffers_release(struct inode *inode, struct file *file);
7065static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7066 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7067
7068static int snapshot_raw_open(struct inode *inode, struct file *filp)
7069{
7070 struct ftrace_buffer_info *info;
7071 int ret;
7072
7073 /* The following checks for tracefs lockdown */
7074 ret = tracing_buffers_open(inode, filp);
7075 if (ret < 0)
7076 return ret;
7077
7078 info = filp->private_data;
7079
7080 if (info->iter.trace->use_max_tr) {
7081 tracing_buffers_release(inode, filp);
7082 return -EBUSY;
7083 }
7084
7085 info->iter.snapshot = true;
7086 info->iter.array_buffer = &info->iter.tr->max_buffer;
7087
7088 return ret;
7089}
7090
7091#endif /* CONFIG_TRACER_SNAPSHOT */
7092
7093
7094static const struct file_operations tracing_thresh_fops = {
7095 .open = tracing_open_generic,
7096 .read = tracing_thresh_read,
7097 .write = tracing_thresh_write,
7098 .llseek = generic_file_llseek,
7099};
7100
7101#if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
7102static const struct file_operations tracing_max_lat_fops = {
7103 .open = tracing_open_generic,
7104 .read = tracing_max_lat_read,
7105 .write = tracing_max_lat_write,
7106 .llseek = generic_file_llseek,
7107};
7108#endif
7109
7110static const struct file_operations set_tracer_fops = {
7111 .open = tracing_open_generic,
7112 .read = tracing_set_trace_read,
7113 .write = tracing_set_trace_write,
7114 .llseek = generic_file_llseek,
7115};
7116
7117static const struct file_operations tracing_pipe_fops = {
7118 .open = tracing_open_pipe,
7119 .poll = tracing_poll_pipe,
7120 .read = tracing_read_pipe,
7121 .splice_read = tracing_splice_read_pipe,
7122 .release = tracing_release_pipe,
7123 .llseek = no_llseek,
7124};
7125
7126static const struct file_operations tracing_entries_fops = {
7127 .open = tracing_open_generic_tr,
7128 .read = tracing_entries_read,
7129 .write = tracing_entries_write,
7130 .llseek = generic_file_llseek,
7131 .release = tracing_release_generic_tr,
7132};
7133
7134static const struct file_operations tracing_total_entries_fops = {
7135 .open = tracing_open_generic_tr,
7136 .read = tracing_total_entries_read,
7137 .llseek = generic_file_llseek,
7138 .release = tracing_release_generic_tr,
7139};
7140
7141static const struct file_operations tracing_free_buffer_fops = {
7142 .open = tracing_open_generic_tr,
7143 .write = tracing_free_buffer_write,
7144 .release = tracing_free_buffer_release,
7145};
7146
7147static const struct file_operations tracing_mark_fops = {
7148 .open = tracing_open_generic_tr,
7149 .write = tracing_mark_write,
7150 .llseek = generic_file_llseek,
7151 .release = tracing_release_generic_tr,
7152};
7153
7154static const struct file_operations tracing_mark_raw_fops = {
7155 .open = tracing_open_generic_tr,
7156 .write = tracing_mark_raw_write,
7157 .llseek = generic_file_llseek,
7158 .release = tracing_release_generic_tr,
7159};
7160
7161static const struct file_operations trace_clock_fops = {
7162 .open = tracing_clock_open,
7163 .read = seq_read,
7164 .llseek = seq_lseek,
7165 .release = tracing_single_release_tr,
7166 .write = tracing_clock_write,
7167};
7168
7169static const struct file_operations trace_time_stamp_mode_fops = {
7170 .open = tracing_time_stamp_mode_open,
7171 .read = seq_read,
7172 .llseek = seq_lseek,
7173 .release = tracing_single_release_tr,
7174};
7175
7176#ifdef CONFIG_TRACER_SNAPSHOT
7177static const struct file_operations snapshot_fops = {
7178 .open = tracing_snapshot_open,
7179 .read = seq_read,
7180 .write = tracing_snapshot_write,
7181 .llseek = tracing_lseek,
7182 .release = tracing_snapshot_release,
7183};
7184
7185static const struct file_operations snapshot_raw_fops = {
7186 .open = snapshot_raw_open,
7187 .read = tracing_buffers_read,
7188 .release = tracing_buffers_release,
7189 .splice_read = tracing_buffers_splice_read,
7190 .llseek = no_llseek,
7191};
7192
7193#endif /* CONFIG_TRACER_SNAPSHOT */
7194
7195#define TRACING_LOG_ERRS_MAX 8
7196#define TRACING_LOG_LOC_MAX 128
7197
7198#define CMD_PREFIX " Command: "
7199
7200struct err_info {
7201 const char **errs; /* ptr to loc-specific array of err strings */
7202 u8 type; /* index into errs -> specific err string */
7203 u8 pos; /* MAX_FILTER_STR_VAL = 256 */
7204 u64 ts;
7205};
7206
7207struct tracing_log_err {
7208 struct list_head list;
7209 struct err_info info;
7210 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7211 char cmd[MAX_FILTER_STR_VAL]; /* what caused err */
7212};
7213
7214static DEFINE_MUTEX(tracing_err_log_lock);
7215
7216static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr)
7217{
7218 struct tracing_log_err *err;
7219
7220 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7221 err = kzalloc(sizeof(*err), GFP_KERNEL);
7222 if (!err)
7223 err = ERR_PTR(-ENOMEM);
7224 tr->n_err_log_entries++;
7225
7226 return err;
7227 }
7228
7229 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7230 list_del(&err->list);
7231
7232 return err;
7233}
7234
7235/**
7236 * err_pos - find the position of a string within a command for error careting
7237 * @cmd: The tracing command that caused the error
7238 * @str: The string to position the caret at within @cmd
7239 *
7240 * Finds the position of the first occurence of @str within @cmd. The
7241 * return value can be passed to tracing_log_err() for caret placement
7242 * within @cmd.
7243 *
7244 * Returns the index within @cmd of the first occurence of @str or 0
7245 * if @str was not found.
7246 */
7247unsigned int err_pos(char *cmd, const char *str)
7248{
7249 char *found;
7250
7251 if (WARN_ON(!strlen(cmd)))
7252 return 0;
7253
7254 found = strstr(cmd, str);
7255 if (found)
7256 return found - cmd;
7257
7258 return 0;
7259}
7260
7261/**
7262 * tracing_log_err - write an error to the tracing error log
7263 * @tr: The associated trace array for the error (NULL for top level array)
7264 * @loc: A string describing where the error occurred
7265 * @cmd: The tracing command that caused the error
7266 * @errs: The array of loc-specific static error strings
7267 * @type: The index into errs[], which produces the specific static err string
7268 * @pos: The position the caret should be placed in the cmd
7269 *
7270 * Writes an error into tracing/error_log of the form:
7271 *
7272 * <loc>: error: <text>
7273 * Command: <cmd>
7274 * ^
7275 *
7276 * tracing/error_log is a small log file containing the last
7277 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7278 * unless there has been a tracing error, and the error log can be
7279 * cleared and have its memory freed by writing the empty string in
7280 * truncation mode to it i.e. echo > tracing/error_log.
7281 *
7282 * NOTE: the @errs array along with the @type param are used to
7283 * produce a static error string - this string is not copied and saved
7284 * when the error is logged - only a pointer to it is saved. See
7285 * existing callers for examples of how static strings are typically
7286 * defined for use with tracing_log_err().
7287 */
7288void tracing_log_err(struct trace_array *tr,
7289 const char *loc, const char *cmd,
7290 const char **errs, u8 type, u8 pos)
7291{
7292 struct tracing_log_err *err;
7293
7294 if (!tr)
7295 tr = &global_trace;
7296
7297 mutex_lock(&tracing_err_log_lock);
7298 err = get_tracing_log_err(tr);
7299 if (PTR_ERR(err) == -ENOMEM) {
7300 mutex_unlock(&tracing_err_log_lock);
7301 return;
7302 }
7303
7304 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7305 snprintf(err->cmd, MAX_FILTER_STR_VAL,"\n" CMD_PREFIX "%s\n", cmd);
7306
7307 err->info.errs = errs;
7308 err->info.type = type;
7309 err->info.pos = pos;
7310 err->info.ts = local_clock();
7311
7312 list_add_tail(&err->list, &tr->err_log);
7313 mutex_unlock(&tracing_err_log_lock);
7314}
7315
7316static void clear_tracing_err_log(struct trace_array *tr)
7317{
7318 struct tracing_log_err *err, *next;
7319
7320 mutex_lock(&tracing_err_log_lock);
7321 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7322 list_del(&err->list);
7323 kfree(err);
7324 }
7325
7326 tr->n_err_log_entries = 0;
7327 mutex_unlock(&tracing_err_log_lock);
7328}
7329
7330static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7331{
7332 struct trace_array *tr = m->private;
7333
7334 mutex_lock(&tracing_err_log_lock);
7335
7336 return seq_list_start(&tr->err_log, *pos);
7337}
7338
7339static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7340{
7341 struct trace_array *tr = m->private;
7342
7343 return seq_list_next(v, &tr->err_log, pos);
7344}
7345
7346static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7347{
7348 mutex_unlock(&tracing_err_log_lock);
7349}
7350
7351static void tracing_err_log_show_pos(struct seq_file *m, u8 pos)
7352{
7353 u8 i;
7354
7355 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7356 seq_putc(m, ' ');
7357 for (i = 0; i < pos; i++)
7358 seq_putc(m, ' ');
7359 seq_puts(m, "^\n");
7360}
7361
7362static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7363{
7364 struct tracing_log_err *err = v;
7365
7366 if (err) {
7367 const char *err_text = err->info.errs[err->info.type];
7368 u64 sec = err->info.ts;
7369 u32 nsec;
7370
7371 nsec = do_div(sec, NSEC_PER_SEC);
7372 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7373 err->loc, err_text);
7374 seq_printf(m, "%s", err->cmd);
7375 tracing_err_log_show_pos(m, err->info.pos);
7376 }
7377
7378 return 0;
7379}
7380
7381static const struct seq_operations tracing_err_log_seq_ops = {
7382 .start = tracing_err_log_seq_start,
7383 .next = tracing_err_log_seq_next,
7384 .stop = tracing_err_log_seq_stop,
7385 .show = tracing_err_log_seq_show
7386};
7387
7388static int tracing_err_log_open(struct inode *inode, struct file *file)
7389{
7390 struct trace_array *tr = inode->i_private;
7391 int ret = 0;
7392
7393 ret = tracing_check_open_get_tr(tr);
7394 if (ret)
7395 return ret;
7396
7397 /* If this file was opened for write, then erase contents */
7398 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7399 clear_tracing_err_log(tr);
7400
7401 if (file->f_mode & FMODE_READ) {
7402 ret = seq_open(file, &tracing_err_log_seq_ops);
7403 if (!ret) {
7404 struct seq_file *m = file->private_data;
7405 m->private = tr;
7406 } else {
7407 trace_array_put(tr);
7408 }
7409 }
7410 return ret;
7411}
7412
7413static ssize_t tracing_err_log_write(struct file *file,
7414 const char __user *buffer,
7415 size_t count, loff_t *ppos)
7416{
7417 return count;
7418}
7419
7420static int tracing_err_log_release(struct inode *inode, struct file *file)
7421{
7422 struct trace_array *tr = inode->i_private;
7423
7424 trace_array_put(tr);
7425
7426 if (file->f_mode & FMODE_READ)
7427 seq_release(inode, file);
7428
7429 return 0;
7430}
7431
7432static const struct file_operations tracing_err_log_fops = {
7433 .open = tracing_err_log_open,
7434 .write = tracing_err_log_write,
7435 .read = seq_read,
7436 .llseek = seq_lseek,
7437 .release = tracing_err_log_release,
7438};
7439
7440static int tracing_buffers_open(struct inode *inode, struct file *filp)
7441{
7442 struct trace_array *tr = inode->i_private;
7443 struct ftrace_buffer_info *info;
7444 int ret;
7445
7446 ret = tracing_check_open_get_tr(tr);
7447 if (ret)
7448 return ret;
7449
7450 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7451 if (!info) {
7452 trace_array_put(tr);
7453 return -ENOMEM;
7454 }
7455
7456 mutex_lock(&trace_types_lock);
7457
7458 info->iter.tr = tr;
7459 info->iter.cpu_file = tracing_get_cpu(inode);
7460 info->iter.trace = tr->current_trace;
7461 info->iter.array_buffer = &tr->array_buffer;
7462 info->spare = NULL;
7463 /* Force reading ring buffer for first read */
7464 info->read = (unsigned int)-1;
7465
7466 filp->private_data = info;
7467
7468 tr->trace_ref++;
7469
7470 mutex_unlock(&trace_types_lock);
7471
7472 ret = nonseekable_open(inode, filp);
7473 if (ret < 0)
7474 trace_array_put(tr);
7475
7476 return ret;
7477}
7478
7479static __poll_t
7480tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7481{
7482 struct ftrace_buffer_info *info = filp->private_data;
7483 struct trace_iterator *iter = &info->iter;
7484
7485 return trace_poll(iter, filp, poll_table);
7486}
7487
7488static ssize_t
7489tracing_buffers_read(struct file *filp, char __user *ubuf,
7490 size_t count, loff_t *ppos)
7491{
7492 struct ftrace_buffer_info *info = filp->private_data;
7493 struct trace_iterator *iter = &info->iter;
7494 ssize_t ret = 0;
7495 ssize_t size;
7496
7497 if (!count)
7498 return 0;
7499
7500#ifdef CONFIG_TRACER_MAX_TRACE
7501 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7502 return -EBUSY;
7503#endif
7504
7505 if (!info->spare) {
7506 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
7507 iter->cpu_file);
7508 if (IS_ERR(info->spare)) {
7509 ret = PTR_ERR(info->spare);
7510 info->spare = NULL;
7511 } else {
7512 info->spare_cpu = iter->cpu_file;
7513 }
7514 }
7515 if (!info->spare)
7516 return ret;
7517
7518 /* Do we have previous read data to read? */
7519 if (info->read < PAGE_SIZE)
7520 goto read;
7521
7522 again:
7523 trace_access_lock(iter->cpu_file);
7524 ret = ring_buffer_read_page(iter->array_buffer->buffer,
7525 &info->spare,
7526 count,
7527 iter->cpu_file, 0);
7528 trace_access_unlock(iter->cpu_file);
7529
7530 if (ret < 0) {
7531 if (trace_empty(iter)) {
7532 if ((filp->f_flags & O_NONBLOCK))
7533 return -EAGAIN;
7534
7535 ret = wait_on_pipe(iter, 0);
7536 if (ret)
7537 return ret;
7538
7539 goto again;
7540 }
7541 return 0;
7542 }
7543
7544 info->read = 0;
7545 read:
7546 size = PAGE_SIZE - info->read;
7547 if (size > count)
7548 size = count;
7549
7550 ret = copy_to_user(ubuf, info->spare + info->read, size);
7551 if (ret == size)
7552 return -EFAULT;
7553
7554 size -= ret;
7555
7556 *ppos += size;
7557 info->read += size;
7558
7559 return size;
7560}
7561
7562static int tracing_buffers_release(struct inode *inode, struct file *file)
7563{
7564 struct ftrace_buffer_info *info = file->private_data;
7565 struct trace_iterator *iter = &info->iter;
7566
7567 mutex_lock(&trace_types_lock);
7568
7569 iter->tr->trace_ref--;
7570
7571 __trace_array_put(iter->tr);
7572
7573 if (info->spare)
7574 ring_buffer_free_read_page(iter->array_buffer->buffer,
7575 info->spare_cpu, info->spare);
7576 kvfree(info);
7577
7578 mutex_unlock(&trace_types_lock);
7579
7580 return 0;
7581}
7582
7583struct buffer_ref {
7584 struct trace_buffer *buffer;
7585 void *page;
7586 int cpu;
7587 refcount_t refcount;
7588};
7589
7590static void buffer_ref_release(struct buffer_ref *ref)
7591{
7592 if (!refcount_dec_and_test(&ref->refcount))
7593 return;
7594 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
7595 kfree(ref);
7596}
7597
7598static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
7599 struct pipe_buffer *buf)
7600{
7601 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
7602
7603 buffer_ref_release(ref);
7604 buf->private = 0;
7605}
7606
7607static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
7608 struct pipe_buffer *buf)
7609{
7610 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
7611
7612 if (refcount_read(&ref->refcount) > INT_MAX/2)
7613 return false;
7614
7615 refcount_inc(&ref->refcount);
7616 return true;
7617}
7618
7619/* Pipe buffer operations for a buffer. */
7620static const struct pipe_buf_operations buffer_pipe_buf_ops = {
7621 .release = buffer_pipe_buf_release,
7622 .get = buffer_pipe_buf_get,
7623};
7624
7625/*
7626 * Callback from splice_to_pipe(), if we need to release some pages
7627 * at the end of the spd in case we error'ed out in filling the pipe.
7628 */
7629static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
7630{
7631 struct buffer_ref *ref =
7632 (struct buffer_ref *)spd->partial[i].private;
7633
7634 buffer_ref_release(ref);
7635 spd->partial[i].private = 0;
7636}
7637
7638static ssize_t
7639tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7640 struct pipe_inode_info *pipe, size_t len,
7641 unsigned int flags)
7642{
7643 struct ftrace_buffer_info *info = file->private_data;
7644 struct trace_iterator *iter = &info->iter;
7645 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7646 struct page *pages_def[PIPE_DEF_BUFFERS];
7647 struct splice_pipe_desc spd = {
7648 .pages = pages_def,
7649 .partial = partial_def,
7650 .nr_pages_max = PIPE_DEF_BUFFERS,
7651 .ops = &buffer_pipe_buf_ops,
7652 .spd_release = buffer_spd_release,
7653 };
7654 struct buffer_ref *ref;
7655 int entries, i;
7656 ssize_t ret = 0;
7657
7658#ifdef CONFIG_TRACER_MAX_TRACE
7659 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7660 return -EBUSY;
7661#endif
7662
7663 if (*ppos & (PAGE_SIZE - 1))
7664 return -EINVAL;
7665
7666 if (len & (PAGE_SIZE - 1)) {
7667 if (len < PAGE_SIZE)
7668 return -EINVAL;
7669 len &= PAGE_MASK;
7670 }
7671
7672 if (splice_grow_spd(pipe, &spd))
7673 return -ENOMEM;
7674
7675 again:
7676 trace_access_lock(iter->cpu_file);
7677 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
7678
7679 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
7680 struct page *page;
7681 int r;
7682
7683 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
7684 if (!ref) {
7685 ret = -ENOMEM;
7686 break;
7687 }
7688
7689 refcount_set(&ref->refcount, 1);
7690 ref->buffer = iter->array_buffer->buffer;
7691 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
7692 if (IS_ERR(ref->page)) {
7693 ret = PTR_ERR(ref->page);
7694 ref->page = NULL;
7695 kfree(ref);
7696 break;
7697 }
7698 ref->cpu = iter->cpu_file;
7699
7700 r = ring_buffer_read_page(ref->buffer, &ref->page,
7701 len, iter->cpu_file, 1);
7702 if (r < 0) {
7703 ring_buffer_free_read_page(ref->buffer, ref->cpu,
7704 ref->page);
7705 kfree(ref);
7706 break;
7707 }
7708
7709 page = virt_to_page(ref->page);
7710
7711 spd.pages[i] = page;
7712 spd.partial[i].len = PAGE_SIZE;
7713 spd.partial[i].offset = 0;
7714 spd.partial[i].private = (unsigned long)ref;
7715 spd.nr_pages++;
7716 *ppos += PAGE_SIZE;
7717
7718 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
7719 }
7720
7721 trace_access_unlock(iter->cpu_file);
7722 spd.nr_pages = i;
7723
7724 /* did we read anything? */
7725 if (!spd.nr_pages) {
7726 if (ret)
7727 goto out;
7728
7729 ret = -EAGAIN;
7730 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
7731 goto out;
7732
7733 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
7734 if (ret)
7735 goto out;
7736
7737 goto again;
7738 }
7739
7740 ret = splice_to_pipe(pipe, &spd);
7741out:
7742 splice_shrink_spd(&spd);
7743
7744 return ret;
7745}
7746
7747static const struct file_operations tracing_buffers_fops = {
7748 .open = tracing_buffers_open,
7749 .read = tracing_buffers_read,
7750 .poll = tracing_buffers_poll,
7751 .release = tracing_buffers_release,
7752 .splice_read = tracing_buffers_splice_read,
7753 .llseek = no_llseek,
7754};
7755
7756static ssize_t
7757tracing_stats_read(struct file *filp, char __user *ubuf,
7758 size_t count, loff_t *ppos)
7759{
7760 struct inode *inode = file_inode(filp);
7761 struct trace_array *tr = inode->i_private;
7762 struct array_buffer *trace_buf = &tr->array_buffer;
7763 int cpu = tracing_get_cpu(inode);
7764 struct trace_seq *s;
7765 unsigned long cnt;
7766 unsigned long long t;
7767 unsigned long usec_rem;
7768
7769 s = kmalloc(sizeof(*s), GFP_KERNEL);
7770 if (!s)
7771 return -ENOMEM;
7772
7773 trace_seq_init(s);
7774
7775 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
7776 trace_seq_printf(s, "entries: %ld\n", cnt);
7777
7778 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
7779 trace_seq_printf(s, "overrun: %ld\n", cnt);
7780
7781 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
7782 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
7783
7784 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
7785 trace_seq_printf(s, "bytes: %ld\n", cnt);
7786
7787 if (trace_clocks[tr->clock_id].in_ns) {
7788 /* local or global for trace_clock */
7789 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
7790 usec_rem = do_div(t, USEC_PER_SEC);
7791 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
7792 t, usec_rem);
7793
7794 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
7795 usec_rem = do_div(t, USEC_PER_SEC);
7796 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
7797 } else {
7798 /* counter or tsc mode for trace_clock */
7799 trace_seq_printf(s, "oldest event ts: %llu\n",
7800 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
7801
7802 trace_seq_printf(s, "now ts: %llu\n",
7803 ring_buffer_time_stamp(trace_buf->buffer, cpu));
7804 }
7805
7806 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
7807 trace_seq_printf(s, "dropped events: %ld\n", cnt);
7808
7809 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
7810 trace_seq_printf(s, "read events: %ld\n", cnt);
7811
7812 count = simple_read_from_buffer(ubuf, count, ppos,
7813 s->buffer, trace_seq_used(s));
7814
7815 kfree(s);
7816
7817 return count;
7818}
7819
7820static const struct file_operations tracing_stats_fops = {
7821 .open = tracing_open_generic_tr,
7822 .read = tracing_stats_read,
7823 .llseek = generic_file_llseek,
7824 .release = tracing_release_generic_tr,
7825};
7826
7827#ifdef CONFIG_DYNAMIC_FTRACE
7828
7829static ssize_t
7830tracing_read_dyn_info(struct file *filp, char __user *ubuf,
7831 size_t cnt, loff_t *ppos)
7832{
7833 ssize_t ret;
7834 char *buf;
7835 int r;
7836
7837 /* 256 should be plenty to hold the amount needed */
7838 buf = kmalloc(256, GFP_KERNEL);
7839 if (!buf)
7840 return -ENOMEM;
7841
7842 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
7843 ftrace_update_tot_cnt,
7844 ftrace_number_of_pages,
7845 ftrace_number_of_groups);
7846
7847 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7848 kfree(buf);
7849 return ret;
7850}
7851
7852static const struct file_operations tracing_dyn_info_fops = {
7853 .open = tracing_open_generic,
7854 .read = tracing_read_dyn_info,
7855 .llseek = generic_file_llseek,
7856};
7857#endif /* CONFIG_DYNAMIC_FTRACE */
7858
7859#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
7860static void
7861ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
7862 struct trace_array *tr, struct ftrace_probe_ops *ops,
7863 void *data)
7864{
7865 tracing_snapshot_instance(tr);
7866}
7867
7868static void
7869ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
7870 struct trace_array *tr, struct ftrace_probe_ops *ops,
7871 void *data)
7872{
7873 struct ftrace_func_mapper *mapper = data;
7874 long *count = NULL;
7875
7876 if (mapper)
7877 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7878
7879 if (count) {
7880
7881 if (*count <= 0)
7882 return;
7883
7884 (*count)--;
7885 }
7886
7887 tracing_snapshot_instance(tr);
7888}
7889
7890static int
7891ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
7892 struct ftrace_probe_ops *ops, void *data)
7893{
7894 struct ftrace_func_mapper *mapper = data;
7895 long *count = NULL;
7896
7897 seq_printf(m, "%ps:", (void *)ip);
7898
7899 seq_puts(m, "snapshot");
7900
7901 if (mapper)
7902 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7903
7904 if (count)
7905 seq_printf(m, ":count=%ld\n", *count);
7906 else
7907 seq_puts(m, ":unlimited\n");
7908
7909 return 0;
7910}
7911
7912static int
7913ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
7914 unsigned long ip, void *init_data, void **data)
7915{
7916 struct ftrace_func_mapper *mapper = *data;
7917
7918 if (!mapper) {
7919 mapper = allocate_ftrace_func_mapper();
7920 if (!mapper)
7921 return -ENOMEM;
7922 *data = mapper;
7923 }
7924
7925 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
7926}
7927
7928static void
7929ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
7930 unsigned long ip, void *data)
7931{
7932 struct ftrace_func_mapper *mapper = data;
7933
7934 if (!ip) {
7935 if (!mapper)
7936 return;
7937 free_ftrace_func_mapper(mapper, NULL);
7938 return;
7939 }
7940
7941 ftrace_func_mapper_remove_ip(mapper, ip);
7942}
7943
7944static struct ftrace_probe_ops snapshot_probe_ops = {
7945 .func = ftrace_snapshot,
7946 .print = ftrace_snapshot_print,
7947};
7948
7949static struct ftrace_probe_ops snapshot_count_probe_ops = {
7950 .func = ftrace_count_snapshot,
7951 .print = ftrace_snapshot_print,
7952 .init = ftrace_snapshot_init,
7953 .free = ftrace_snapshot_free,
7954};
7955
7956static int
7957ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
7958 char *glob, char *cmd, char *param, int enable)
7959{
7960 struct ftrace_probe_ops *ops;
7961 void *count = (void *)-1;
7962 char *number;
7963 int ret;
7964
7965 if (!tr)
7966 return -ENODEV;
7967
7968 /* hash funcs only work with set_ftrace_filter */
7969 if (!enable)
7970 return -EINVAL;
7971
7972 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
7973
7974 if (glob[0] == '!')
7975 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
7976
7977 if (!param)
7978 goto out_reg;
7979
7980 number = strsep(¶m, ":");
7981
7982 if (!strlen(number))
7983 goto out_reg;
7984
7985 /*
7986 * We use the callback data field (which is a pointer)
7987 * as our counter.
7988 */
7989 ret = kstrtoul(number, 0, (unsigned long *)&count);
7990 if (ret)
7991 return ret;
7992
7993 out_reg:
7994 ret = tracing_alloc_snapshot_instance(tr);
7995 if (ret < 0)
7996 goto out;
7997
7998 ret = register_ftrace_function_probe(glob, tr, ops, count);
7999
8000 out:
8001 return ret < 0 ? ret : 0;
8002}
8003
8004static struct ftrace_func_command ftrace_snapshot_cmd = {
8005 .name = "snapshot",
8006 .func = ftrace_trace_snapshot_callback,
8007};
8008
8009static __init int register_snapshot_cmd(void)
8010{
8011 return register_ftrace_command(&ftrace_snapshot_cmd);
8012}
8013#else
8014static inline __init int register_snapshot_cmd(void) { return 0; }
8015#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8016
8017static struct dentry *tracing_get_dentry(struct trace_array *tr)
8018{
8019 if (WARN_ON(!tr->dir))
8020 return ERR_PTR(-ENODEV);
8021
8022 /* Top directory uses NULL as the parent */
8023 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8024 return NULL;
8025
8026 /* All sub buffers have a descriptor */
8027 return tr->dir;
8028}
8029
8030static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8031{
8032 struct dentry *d_tracer;
8033
8034 if (tr->percpu_dir)
8035 return tr->percpu_dir;
8036
8037 d_tracer = tracing_get_dentry(tr);
8038 if (IS_ERR(d_tracer))
8039 return NULL;
8040
8041 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8042
8043 MEM_FAIL(!tr->percpu_dir,
8044 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8045
8046 return tr->percpu_dir;
8047}
8048
8049static struct dentry *
8050trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8051 void *data, long cpu, const struct file_operations *fops)
8052{
8053 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8054
8055 if (ret) /* See tracing_get_cpu() */
8056 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8057 return ret;
8058}
8059
8060static void
8061tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8062{
8063 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8064 struct dentry *d_cpu;
8065 char cpu_dir[30]; /* 30 characters should be more than enough */
8066
8067 if (!d_percpu)
8068 return;
8069
8070 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8071 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8072 if (!d_cpu) {
8073 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8074 return;
8075 }
8076
8077 /* per cpu trace_pipe */
8078 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
8079 tr, cpu, &tracing_pipe_fops);
8080
8081 /* per cpu trace */
8082 trace_create_cpu_file("trace", 0644, d_cpu,
8083 tr, cpu, &tracing_fops);
8084
8085 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
8086 tr, cpu, &tracing_buffers_fops);
8087
8088 trace_create_cpu_file("stats", 0444, d_cpu,
8089 tr, cpu, &tracing_stats_fops);
8090
8091 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
8092 tr, cpu, &tracing_entries_fops);
8093
8094#ifdef CONFIG_TRACER_SNAPSHOT
8095 trace_create_cpu_file("snapshot", 0644, d_cpu,
8096 tr, cpu, &snapshot_fops);
8097
8098 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
8099 tr, cpu, &snapshot_raw_fops);
8100#endif
8101}
8102
8103#ifdef CONFIG_FTRACE_SELFTEST
8104/* Let selftest have access to static functions in this file */
8105#include "trace_selftest.c"
8106#endif
8107
8108static ssize_t
8109trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8110 loff_t *ppos)
8111{
8112 struct trace_option_dentry *topt = filp->private_data;
8113 char *buf;
8114
8115 if (topt->flags->val & topt->opt->bit)
8116 buf = "1\n";
8117 else
8118 buf = "0\n";
8119
8120 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8121}
8122
8123static ssize_t
8124trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8125 loff_t *ppos)
8126{
8127 struct trace_option_dentry *topt = filp->private_data;
8128 unsigned long val;
8129 int ret;
8130
8131 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8132 if (ret)
8133 return ret;
8134
8135 if (val != 0 && val != 1)
8136 return -EINVAL;
8137
8138 if (!!(topt->flags->val & topt->opt->bit) != val) {
8139 mutex_lock(&trace_types_lock);
8140 ret = __set_tracer_option(topt->tr, topt->flags,
8141 topt->opt, !val);
8142 mutex_unlock(&trace_types_lock);
8143 if (ret)
8144 return ret;
8145 }
8146
8147 *ppos += cnt;
8148
8149 return cnt;
8150}
8151
8152
8153static const struct file_operations trace_options_fops = {
8154 .open = tracing_open_generic,
8155 .read = trace_options_read,
8156 .write = trace_options_write,
8157 .llseek = generic_file_llseek,
8158};
8159
8160/*
8161 * In order to pass in both the trace_array descriptor as well as the index
8162 * to the flag that the trace option file represents, the trace_array
8163 * has a character array of trace_flags_index[], which holds the index
8164 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8165 * The address of this character array is passed to the flag option file
8166 * read/write callbacks.
8167 *
8168 * In order to extract both the index and the trace_array descriptor,
8169 * get_tr_index() uses the following algorithm.
8170 *
8171 * idx = *ptr;
8172 *
8173 * As the pointer itself contains the address of the index (remember
8174 * index[1] == 1).
8175 *
8176 * Then to get the trace_array descriptor, by subtracting that index
8177 * from the ptr, we get to the start of the index itself.
8178 *
8179 * ptr - idx == &index[0]
8180 *
8181 * Then a simple container_of() from that pointer gets us to the
8182 * trace_array descriptor.
8183 */
8184static void get_tr_index(void *data, struct trace_array **ptr,
8185 unsigned int *pindex)
8186{
8187 *pindex = *(unsigned char *)data;
8188
8189 *ptr = container_of(data - *pindex, struct trace_array,
8190 trace_flags_index);
8191}
8192
8193static ssize_t
8194trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8195 loff_t *ppos)
8196{
8197 void *tr_index = filp->private_data;
8198 struct trace_array *tr;
8199 unsigned int index;
8200 char *buf;
8201
8202 get_tr_index(tr_index, &tr, &index);
8203
8204 if (tr->trace_flags & (1 << index))
8205 buf = "1\n";
8206 else
8207 buf = "0\n";
8208
8209 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8210}
8211
8212static ssize_t
8213trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8214 loff_t *ppos)
8215{
8216 void *tr_index = filp->private_data;
8217 struct trace_array *tr;
8218 unsigned int index;
8219 unsigned long val;
8220 int ret;
8221
8222 get_tr_index(tr_index, &tr, &index);
8223
8224 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8225 if (ret)
8226 return ret;
8227
8228 if (val != 0 && val != 1)
8229 return -EINVAL;
8230
8231 mutex_lock(&event_mutex);
8232 mutex_lock(&trace_types_lock);
8233 ret = set_tracer_flag(tr, 1 << index, val);
8234 mutex_unlock(&trace_types_lock);
8235 mutex_unlock(&event_mutex);
8236
8237 if (ret < 0)
8238 return ret;
8239
8240 *ppos += cnt;
8241
8242 return cnt;
8243}
8244
8245static const struct file_operations trace_options_core_fops = {
8246 .open = tracing_open_generic,
8247 .read = trace_options_core_read,
8248 .write = trace_options_core_write,
8249 .llseek = generic_file_llseek,
8250};
8251
8252struct dentry *trace_create_file(const char *name,
8253 umode_t mode,
8254 struct dentry *parent,
8255 void *data,
8256 const struct file_operations *fops)
8257{
8258 struct dentry *ret;
8259
8260 ret = tracefs_create_file(name, mode, parent, data, fops);
8261 if (!ret)
8262 pr_warn("Could not create tracefs '%s' entry\n", name);
8263
8264 return ret;
8265}
8266
8267
8268static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8269{
8270 struct dentry *d_tracer;
8271
8272 if (tr->options)
8273 return tr->options;
8274
8275 d_tracer = tracing_get_dentry(tr);
8276 if (IS_ERR(d_tracer))
8277 return NULL;
8278
8279 tr->options = tracefs_create_dir("options", d_tracer);
8280 if (!tr->options) {
8281 pr_warn("Could not create tracefs directory 'options'\n");
8282 return NULL;
8283 }
8284
8285 return tr->options;
8286}
8287
8288static void
8289create_trace_option_file(struct trace_array *tr,
8290 struct trace_option_dentry *topt,
8291 struct tracer_flags *flags,
8292 struct tracer_opt *opt)
8293{
8294 struct dentry *t_options;
8295
8296 t_options = trace_options_init_dentry(tr);
8297 if (!t_options)
8298 return;
8299
8300 topt->flags = flags;
8301 topt->opt = opt;
8302 topt->tr = tr;
8303
8304 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
8305 &trace_options_fops);
8306
8307}
8308
8309static void
8310create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8311{
8312 struct trace_option_dentry *topts;
8313 struct trace_options *tr_topts;
8314 struct tracer_flags *flags;
8315 struct tracer_opt *opts;
8316 int cnt;
8317 int i;
8318
8319 if (!tracer)
8320 return;
8321
8322 flags = tracer->flags;
8323
8324 if (!flags || !flags->opts)
8325 return;
8326
8327 /*
8328 * If this is an instance, only create flags for tracers
8329 * the instance may have.
8330 */
8331 if (!trace_ok_for_array(tracer, tr))
8332 return;
8333
8334 for (i = 0; i < tr->nr_topts; i++) {
8335 /* Make sure there's no duplicate flags. */
8336 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8337 return;
8338 }
8339
8340 opts = flags->opts;
8341
8342 for (cnt = 0; opts[cnt].name; cnt++)
8343 ;
8344
8345 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8346 if (!topts)
8347 return;
8348
8349 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8350 GFP_KERNEL);
8351 if (!tr_topts) {
8352 kfree(topts);
8353 return;
8354 }
8355
8356 tr->topts = tr_topts;
8357 tr->topts[tr->nr_topts].tracer = tracer;
8358 tr->topts[tr->nr_topts].topts = topts;
8359 tr->nr_topts++;
8360
8361 for (cnt = 0; opts[cnt].name; cnt++) {
8362 create_trace_option_file(tr, &topts[cnt], flags,
8363 &opts[cnt]);
8364 MEM_FAIL(topts[cnt].entry == NULL,
8365 "Failed to create trace option: %s",
8366 opts[cnt].name);
8367 }
8368}
8369
8370static struct dentry *
8371create_trace_option_core_file(struct trace_array *tr,
8372 const char *option, long index)
8373{
8374 struct dentry *t_options;
8375
8376 t_options = trace_options_init_dentry(tr);
8377 if (!t_options)
8378 return NULL;
8379
8380 return trace_create_file(option, 0644, t_options,
8381 (void *)&tr->trace_flags_index[index],
8382 &trace_options_core_fops);
8383}
8384
8385static void create_trace_options_dir(struct trace_array *tr)
8386{
8387 struct dentry *t_options;
8388 bool top_level = tr == &global_trace;
8389 int i;
8390
8391 t_options = trace_options_init_dentry(tr);
8392 if (!t_options)
8393 return;
8394
8395 for (i = 0; trace_options[i]; i++) {
8396 if (top_level ||
8397 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8398 create_trace_option_core_file(tr, trace_options[i], i);
8399 }
8400}
8401
8402static ssize_t
8403rb_simple_read(struct file *filp, char __user *ubuf,
8404 size_t cnt, loff_t *ppos)
8405{
8406 struct trace_array *tr = filp->private_data;
8407 char buf[64];
8408 int r;
8409
8410 r = tracer_tracing_is_on(tr);
8411 r = sprintf(buf, "%d\n", r);
8412
8413 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8414}
8415
8416static ssize_t
8417rb_simple_write(struct file *filp, const char __user *ubuf,
8418 size_t cnt, loff_t *ppos)
8419{
8420 struct trace_array *tr = filp->private_data;
8421 struct trace_buffer *buffer = tr->array_buffer.buffer;
8422 unsigned long val;
8423 int ret;
8424
8425 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8426 if (ret)
8427 return ret;
8428
8429 if (buffer) {
8430 mutex_lock(&trace_types_lock);
8431 if (!!val == tracer_tracing_is_on(tr)) {
8432 val = 0; /* do nothing */
8433 } else if (val) {
8434 tracer_tracing_on(tr);
8435 if (tr->current_trace->start)
8436 tr->current_trace->start(tr);
8437 } else {
8438 tracer_tracing_off(tr);
8439 if (tr->current_trace->stop)
8440 tr->current_trace->stop(tr);
8441 }
8442 mutex_unlock(&trace_types_lock);
8443 }
8444
8445 (*ppos)++;
8446
8447 return cnt;
8448}
8449
8450static const struct file_operations rb_simple_fops = {
8451 .open = tracing_open_generic_tr,
8452 .read = rb_simple_read,
8453 .write = rb_simple_write,
8454 .release = tracing_release_generic_tr,
8455 .llseek = default_llseek,
8456};
8457
8458static ssize_t
8459buffer_percent_read(struct file *filp, char __user *ubuf,
8460 size_t cnt, loff_t *ppos)
8461{
8462 struct trace_array *tr = filp->private_data;
8463 char buf[64];
8464 int r;
8465
8466 r = tr->buffer_percent;
8467 r = sprintf(buf, "%d\n", r);
8468
8469 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8470}
8471
8472static ssize_t
8473buffer_percent_write(struct file *filp, const char __user *ubuf,
8474 size_t cnt, loff_t *ppos)
8475{
8476 struct trace_array *tr = filp->private_data;
8477 unsigned long val;
8478 int ret;
8479
8480 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8481 if (ret)
8482 return ret;
8483
8484 if (val > 100)
8485 return -EINVAL;
8486
8487 if (!val)
8488 val = 1;
8489
8490 tr->buffer_percent = val;
8491
8492 (*ppos)++;
8493
8494 return cnt;
8495}
8496
8497static const struct file_operations buffer_percent_fops = {
8498 .open = tracing_open_generic_tr,
8499 .read = buffer_percent_read,
8500 .write = buffer_percent_write,
8501 .release = tracing_release_generic_tr,
8502 .llseek = default_llseek,
8503};
8504
8505static struct dentry *trace_instance_dir;
8506
8507static void
8508init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
8509
8510static int
8511allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
8512{
8513 enum ring_buffer_flags rb_flags;
8514
8515 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
8516
8517 buf->tr = tr;
8518
8519 buf->buffer = ring_buffer_alloc(size, rb_flags);
8520 if (!buf->buffer)
8521 return -ENOMEM;
8522
8523 buf->data = alloc_percpu(struct trace_array_cpu);
8524 if (!buf->data) {
8525 ring_buffer_free(buf->buffer);
8526 buf->buffer = NULL;
8527 return -ENOMEM;
8528 }
8529
8530 /* Allocate the first page for all buffers */
8531 set_buffer_entries(&tr->array_buffer,
8532 ring_buffer_size(tr->array_buffer.buffer, 0));
8533
8534 return 0;
8535}
8536
8537static int allocate_trace_buffers(struct trace_array *tr, int size)
8538{
8539 int ret;
8540
8541 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
8542 if (ret)
8543 return ret;
8544
8545#ifdef CONFIG_TRACER_MAX_TRACE
8546 ret = allocate_trace_buffer(tr, &tr->max_buffer,
8547 allocate_snapshot ? size : 1);
8548 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
8549 ring_buffer_free(tr->array_buffer.buffer);
8550 tr->array_buffer.buffer = NULL;
8551 free_percpu(tr->array_buffer.data);
8552 tr->array_buffer.data = NULL;
8553 return -ENOMEM;
8554 }
8555 tr->allocated_snapshot = allocate_snapshot;
8556
8557 /*
8558 * Only the top level trace array gets its snapshot allocated
8559 * from the kernel command line.
8560 */
8561 allocate_snapshot = false;
8562#endif
8563
8564 return 0;
8565}
8566
8567static void free_trace_buffer(struct array_buffer *buf)
8568{
8569 if (buf->buffer) {
8570 ring_buffer_free(buf->buffer);
8571 buf->buffer = NULL;
8572 free_percpu(buf->data);
8573 buf->data = NULL;
8574 }
8575}
8576
8577static void free_trace_buffers(struct trace_array *tr)
8578{
8579 if (!tr)
8580 return;
8581
8582 free_trace_buffer(&tr->array_buffer);
8583
8584#ifdef CONFIG_TRACER_MAX_TRACE
8585 free_trace_buffer(&tr->max_buffer);
8586#endif
8587}
8588
8589static void init_trace_flags_index(struct trace_array *tr)
8590{
8591 int i;
8592
8593 /* Used by the trace options files */
8594 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
8595 tr->trace_flags_index[i] = i;
8596}
8597
8598static void __update_tracer_options(struct trace_array *tr)
8599{
8600 struct tracer *t;
8601
8602 for (t = trace_types; t; t = t->next)
8603 add_tracer_options(tr, t);
8604}
8605
8606static void update_tracer_options(struct trace_array *tr)
8607{
8608 mutex_lock(&trace_types_lock);
8609 __update_tracer_options(tr);
8610 mutex_unlock(&trace_types_lock);
8611}
8612
8613/* Must have trace_types_lock held */
8614struct trace_array *trace_array_find(const char *instance)
8615{
8616 struct trace_array *tr, *found = NULL;
8617
8618 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8619 if (tr->name && strcmp(tr->name, instance) == 0) {
8620 found = tr;
8621 break;
8622 }
8623 }
8624
8625 return found;
8626}
8627
8628struct trace_array *trace_array_find_get(const char *instance)
8629{
8630 struct trace_array *tr;
8631
8632 mutex_lock(&trace_types_lock);
8633 tr = trace_array_find(instance);
8634 if (tr)
8635 tr->ref++;
8636 mutex_unlock(&trace_types_lock);
8637
8638 return tr;
8639}
8640
8641static struct trace_array *trace_array_create(const char *name)
8642{
8643 struct trace_array *tr;
8644 int ret;
8645
8646 ret = -ENOMEM;
8647 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
8648 if (!tr)
8649 return ERR_PTR(ret);
8650
8651 tr->name = kstrdup(name, GFP_KERNEL);
8652 if (!tr->name)
8653 goto out_free_tr;
8654
8655 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
8656 goto out_free_tr;
8657
8658 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
8659
8660 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
8661
8662 raw_spin_lock_init(&tr->start_lock);
8663
8664 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
8665
8666 tr->current_trace = &nop_trace;
8667
8668 INIT_LIST_HEAD(&tr->systems);
8669 INIT_LIST_HEAD(&tr->events);
8670 INIT_LIST_HEAD(&tr->hist_vars);
8671 INIT_LIST_HEAD(&tr->err_log);
8672
8673 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
8674 goto out_free_tr;
8675
8676 tr->dir = tracefs_create_dir(name, trace_instance_dir);
8677 if (!tr->dir)
8678 goto out_free_tr;
8679
8680 ret = event_trace_add_tracer(tr->dir, tr);
8681 if (ret) {
8682 tracefs_remove(tr->dir);
8683 goto out_free_tr;
8684 }
8685
8686 ftrace_init_trace_array(tr);
8687
8688 init_tracer_tracefs(tr, tr->dir);
8689 init_trace_flags_index(tr);
8690 __update_tracer_options(tr);
8691
8692 list_add(&tr->list, &ftrace_trace_arrays);
8693
8694 tr->ref++;
8695
8696
8697 return tr;
8698
8699 out_free_tr:
8700 free_trace_buffers(tr);
8701 free_cpumask_var(tr->tracing_cpumask);
8702 kfree(tr->name);
8703 kfree(tr);
8704
8705 return ERR_PTR(ret);
8706}
8707
8708static int instance_mkdir(const char *name)
8709{
8710 struct trace_array *tr;
8711 int ret;
8712
8713 mutex_lock(&event_mutex);
8714 mutex_lock(&trace_types_lock);
8715
8716 ret = -EEXIST;
8717 if (trace_array_find(name))
8718 goto out_unlock;
8719
8720 tr = trace_array_create(name);
8721
8722 ret = PTR_ERR_OR_ZERO(tr);
8723
8724out_unlock:
8725 mutex_unlock(&trace_types_lock);
8726 mutex_unlock(&event_mutex);
8727 return ret;
8728}
8729
8730/**
8731 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
8732 * @name: The name of the trace array to be looked up/created.
8733 *
8734 * Returns pointer to trace array with given name.
8735 * NULL, if it cannot be created.
8736 *
8737 * NOTE: This function increments the reference counter associated with the
8738 * trace array returned. This makes sure it cannot be freed while in use.
8739 * Use trace_array_put() once the trace array is no longer needed.
8740 * If the trace_array is to be freed, trace_array_destroy() needs to
8741 * be called after the trace_array_put(), or simply let user space delete
8742 * it from the tracefs instances directory. But until the
8743 * trace_array_put() is called, user space can not delete it.
8744 *
8745 */
8746struct trace_array *trace_array_get_by_name(const char *name)
8747{
8748 struct trace_array *tr;
8749
8750 mutex_lock(&event_mutex);
8751 mutex_lock(&trace_types_lock);
8752
8753 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8754 if (tr->name && strcmp(tr->name, name) == 0)
8755 goto out_unlock;
8756 }
8757
8758 tr = trace_array_create(name);
8759
8760 if (IS_ERR(tr))
8761 tr = NULL;
8762out_unlock:
8763 if (tr)
8764 tr->ref++;
8765
8766 mutex_unlock(&trace_types_lock);
8767 mutex_unlock(&event_mutex);
8768 return tr;
8769}
8770EXPORT_SYMBOL_GPL(trace_array_get_by_name);
8771
8772static int __remove_instance(struct trace_array *tr)
8773{
8774 int i;
8775
8776 /* Reference counter for a newly created trace array = 1. */
8777 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
8778 return -EBUSY;
8779
8780 list_del(&tr->list);
8781
8782 /* Disable all the flags that were enabled coming in */
8783 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
8784 if ((1 << i) & ZEROED_TRACE_FLAGS)
8785 set_tracer_flag(tr, 1 << i, 0);
8786 }
8787
8788 tracing_set_nop(tr);
8789 clear_ftrace_function_probes(tr);
8790 event_trace_del_tracer(tr);
8791 ftrace_clear_pids(tr);
8792 ftrace_destroy_function_files(tr);
8793 tracefs_remove(tr->dir);
8794 free_trace_buffers(tr);
8795
8796 for (i = 0; i < tr->nr_topts; i++) {
8797 kfree(tr->topts[i].topts);
8798 }
8799 kfree(tr->topts);
8800
8801 free_cpumask_var(tr->tracing_cpumask);
8802 kfree(tr->name);
8803 kfree(tr);
8804 tr = NULL;
8805
8806 return 0;
8807}
8808
8809int trace_array_destroy(struct trace_array *this_tr)
8810{
8811 struct trace_array *tr;
8812 int ret;
8813
8814 if (!this_tr)
8815 return -EINVAL;
8816
8817 mutex_lock(&event_mutex);
8818 mutex_lock(&trace_types_lock);
8819
8820 ret = -ENODEV;
8821
8822 /* Making sure trace array exists before destroying it. */
8823 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8824 if (tr == this_tr) {
8825 ret = __remove_instance(tr);
8826 break;
8827 }
8828 }
8829
8830 mutex_unlock(&trace_types_lock);
8831 mutex_unlock(&event_mutex);
8832
8833 return ret;
8834}
8835EXPORT_SYMBOL_GPL(trace_array_destroy);
8836
8837static int instance_rmdir(const char *name)
8838{
8839 struct trace_array *tr;
8840 int ret;
8841
8842 mutex_lock(&event_mutex);
8843 mutex_lock(&trace_types_lock);
8844
8845 ret = -ENODEV;
8846 tr = trace_array_find(name);
8847 if (tr)
8848 ret = __remove_instance(tr);
8849
8850 mutex_unlock(&trace_types_lock);
8851 mutex_unlock(&event_mutex);
8852
8853 return ret;
8854}
8855
8856static __init void create_trace_instances(struct dentry *d_tracer)
8857{
8858 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
8859 instance_mkdir,
8860 instance_rmdir);
8861 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
8862 return;
8863}
8864
8865static void
8866init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
8867{
8868 struct trace_event_file *file;
8869 int cpu;
8870
8871 trace_create_file("available_tracers", 0444, d_tracer,
8872 tr, &show_traces_fops);
8873
8874 trace_create_file("current_tracer", 0644, d_tracer,
8875 tr, &set_tracer_fops);
8876
8877 trace_create_file("tracing_cpumask", 0644, d_tracer,
8878 tr, &tracing_cpumask_fops);
8879
8880 trace_create_file("trace_options", 0644, d_tracer,
8881 tr, &tracing_iter_fops);
8882
8883 trace_create_file("trace", 0644, d_tracer,
8884 tr, &tracing_fops);
8885
8886 trace_create_file("trace_pipe", 0444, d_tracer,
8887 tr, &tracing_pipe_fops);
8888
8889 trace_create_file("buffer_size_kb", 0644, d_tracer,
8890 tr, &tracing_entries_fops);
8891
8892 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
8893 tr, &tracing_total_entries_fops);
8894
8895 trace_create_file("free_buffer", 0200, d_tracer,
8896 tr, &tracing_free_buffer_fops);
8897
8898 trace_create_file("trace_marker", 0220, d_tracer,
8899 tr, &tracing_mark_fops);
8900
8901 file = __find_event_file(tr, "ftrace", "print");
8902 if (file && file->dir)
8903 trace_create_file("trigger", 0644, file->dir, file,
8904 &event_trigger_fops);
8905 tr->trace_marker_file = file;
8906
8907 trace_create_file("trace_marker_raw", 0220, d_tracer,
8908 tr, &tracing_mark_raw_fops);
8909
8910 trace_create_file("trace_clock", 0644, d_tracer, tr,
8911 &trace_clock_fops);
8912
8913 trace_create_file("tracing_on", 0644, d_tracer,
8914 tr, &rb_simple_fops);
8915
8916 trace_create_file("timestamp_mode", 0444, d_tracer, tr,
8917 &trace_time_stamp_mode_fops);
8918
8919 tr->buffer_percent = 50;
8920
8921 trace_create_file("buffer_percent", 0444, d_tracer,
8922 tr, &buffer_percent_fops);
8923
8924 create_trace_options_dir(tr);
8925
8926#if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
8927 trace_create_maxlat_file(tr, d_tracer);
8928#endif
8929
8930 if (ftrace_create_function_files(tr, d_tracer))
8931 MEM_FAIL(1, "Could not allocate function filter files");
8932
8933#ifdef CONFIG_TRACER_SNAPSHOT
8934 trace_create_file("snapshot", 0644, d_tracer,
8935 tr, &snapshot_fops);
8936#endif
8937
8938 trace_create_file("error_log", 0644, d_tracer,
8939 tr, &tracing_err_log_fops);
8940
8941 for_each_tracing_cpu(cpu)
8942 tracing_init_tracefs_percpu(tr, cpu);
8943
8944 ftrace_init_tracefs(tr, d_tracer);
8945}
8946
8947static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
8948{
8949 struct vfsmount *mnt;
8950 struct file_system_type *type;
8951
8952 /*
8953 * To maintain backward compatibility for tools that mount
8954 * debugfs to get to the tracing facility, tracefs is automatically
8955 * mounted to the debugfs/tracing directory.
8956 */
8957 type = get_fs_type("tracefs");
8958 if (!type)
8959 return NULL;
8960 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
8961 put_filesystem(type);
8962 if (IS_ERR(mnt))
8963 return NULL;
8964 mntget(mnt);
8965
8966 return mnt;
8967}
8968
8969/**
8970 * tracing_init_dentry - initialize top level trace array
8971 *
8972 * This is called when creating files or directories in the tracing
8973 * directory. It is called via fs_initcall() by any of the boot up code
8974 * and expects to return the dentry of the top level tracing directory.
8975 */
8976struct dentry *tracing_init_dentry(void)
8977{
8978 struct trace_array *tr = &global_trace;
8979
8980 if (security_locked_down(LOCKDOWN_TRACEFS)) {
8981 pr_warn("Tracing disabled due to lockdown\n");
8982 return ERR_PTR(-EPERM);
8983 }
8984
8985 /* The top level trace array uses NULL as parent */
8986 if (tr->dir)
8987 return NULL;
8988
8989 if (WARN_ON(!tracefs_initialized()))
8990 return ERR_PTR(-ENODEV);
8991
8992 /*
8993 * As there may still be users that expect the tracing
8994 * files to exist in debugfs/tracing, we must automount
8995 * the tracefs file system there, so older tools still
8996 * work with the newer kerenl.
8997 */
8998 tr->dir = debugfs_create_automount("tracing", NULL,
8999 trace_automount, NULL);
9000
9001 return NULL;
9002}
9003
9004extern struct trace_eval_map *__start_ftrace_eval_maps[];
9005extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9006
9007static void __init trace_eval_init(void)
9008{
9009 int len;
9010
9011 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9012 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9013}
9014
9015#ifdef CONFIG_MODULES
9016static void trace_module_add_evals(struct module *mod)
9017{
9018 if (!mod->num_trace_evals)
9019 return;
9020
9021 /*
9022 * Modules with bad taint do not have events created, do
9023 * not bother with enums either.
9024 */
9025 if (trace_module_has_bad_taint(mod))
9026 return;
9027
9028 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9029}
9030
9031#ifdef CONFIG_TRACE_EVAL_MAP_FILE
9032static void trace_module_remove_evals(struct module *mod)
9033{
9034 union trace_eval_map_item *map;
9035 union trace_eval_map_item **last = &trace_eval_maps;
9036
9037 if (!mod->num_trace_evals)
9038 return;
9039
9040 mutex_lock(&trace_eval_mutex);
9041
9042 map = trace_eval_maps;
9043
9044 while (map) {
9045 if (map->head.mod == mod)
9046 break;
9047 map = trace_eval_jmp_to_tail(map);
9048 last = &map->tail.next;
9049 map = map->tail.next;
9050 }
9051 if (!map)
9052 goto out;
9053
9054 *last = trace_eval_jmp_to_tail(map)->tail.next;
9055 kfree(map);
9056 out:
9057 mutex_unlock(&trace_eval_mutex);
9058}
9059#else
9060static inline void trace_module_remove_evals(struct module *mod) { }
9061#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9062
9063static int trace_module_notify(struct notifier_block *self,
9064 unsigned long val, void *data)
9065{
9066 struct module *mod = data;
9067
9068 switch (val) {
9069 case MODULE_STATE_COMING:
9070 trace_module_add_evals(mod);
9071 break;
9072 case MODULE_STATE_GOING:
9073 trace_module_remove_evals(mod);
9074 break;
9075 }
9076
9077 return 0;
9078}
9079
9080static struct notifier_block trace_module_nb = {
9081 .notifier_call = trace_module_notify,
9082 .priority = 0,
9083};
9084#endif /* CONFIG_MODULES */
9085
9086static __init int tracer_init_tracefs(void)
9087{
9088 struct dentry *d_tracer;
9089
9090 trace_access_lock_init();
9091
9092 d_tracer = tracing_init_dentry();
9093 if (IS_ERR(d_tracer))
9094 return 0;
9095
9096 event_trace_init();
9097
9098 init_tracer_tracefs(&global_trace, d_tracer);
9099 ftrace_init_tracefs_toplevel(&global_trace, d_tracer);
9100
9101 trace_create_file("tracing_thresh", 0644, d_tracer,
9102 &global_trace, &tracing_thresh_fops);
9103
9104 trace_create_file("README", 0444, d_tracer,
9105 NULL, &tracing_readme_fops);
9106
9107 trace_create_file("saved_cmdlines", 0444, d_tracer,
9108 NULL, &tracing_saved_cmdlines_fops);
9109
9110 trace_create_file("saved_cmdlines_size", 0644, d_tracer,
9111 NULL, &tracing_saved_cmdlines_size_fops);
9112
9113 trace_create_file("saved_tgids", 0444, d_tracer,
9114 NULL, &tracing_saved_tgids_fops);
9115
9116 trace_eval_init();
9117
9118 trace_create_eval_file(d_tracer);
9119
9120#ifdef CONFIG_MODULES
9121 register_module_notifier(&trace_module_nb);
9122#endif
9123
9124#ifdef CONFIG_DYNAMIC_FTRACE
9125 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
9126 NULL, &tracing_dyn_info_fops);
9127#endif
9128
9129 create_trace_instances(d_tracer);
9130
9131 update_tracer_options(&global_trace);
9132
9133 return 0;
9134}
9135
9136static int trace_panic_handler(struct notifier_block *this,
9137 unsigned long event, void *unused)
9138{
9139 if (ftrace_dump_on_oops)
9140 ftrace_dump(ftrace_dump_on_oops);
9141 return NOTIFY_OK;
9142}
9143
9144static struct notifier_block trace_panic_notifier = {
9145 .notifier_call = trace_panic_handler,
9146 .next = NULL,
9147 .priority = 150 /* priority: INT_MAX >= x >= 0 */
9148};
9149
9150static int trace_die_handler(struct notifier_block *self,
9151 unsigned long val,
9152 void *data)
9153{
9154 switch (val) {
9155 case DIE_OOPS:
9156 if (ftrace_dump_on_oops)
9157 ftrace_dump(ftrace_dump_on_oops);
9158 break;
9159 default:
9160 break;
9161 }
9162 return NOTIFY_OK;
9163}
9164
9165static struct notifier_block trace_die_notifier = {
9166 .notifier_call = trace_die_handler,
9167 .priority = 200
9168};
9169
9170/*
9171 * printk is set to max of 1024, we really don't need it that big.
9172 * Nothing should be printing 1000 characters anyway.
9173 */
9174#define TRACE_MAX_PRINT 1000
9175
9176/*
9177 * Define here KERN_TRACE so that we have one place to modify
9178 * it if we decide to change what log level the ftrace dump
9179 * should be at.
9180 */
9181#define KERN_TRACE KERN_EMERG
9182
9183void
9184trace_printk_seq(struct trace_seq *s)
9185{
9186 /* Probably should print a warning here. */
9187 if (s->seq.len >= TRACE_MAX_PRINT)
9188 s->seq.len = TRACE_MAX_PRINT;
9189
9190 /*
9191 * More paranoid code. Although the buffer size is set to
9192 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9193 * an extra layer of protection.
9194 */
9195 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9196 s->seq.len = s->seq.size - 1;
9197
9198 /* should be zero ended, but we are paranoid. */
9199 s->buffer[s->seq.len] = 0;
9200
9201 printk(KERN_TRACE "%s", s->buffer);
9202
9203 trace_seq_init(s);
9204}
9205
9206void trace_init_global_iter(struct trace_iterator *iter)
9207{
9208 iter->tr = &global_trace;
9209 iter->trace = iter->tr->current_trace;
9210 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9211 iter->array_buffer = &global_trace.array_buffer;
9212
9213 if (iter->trace && iter->trace->open)
9214 iter->trace->open(iter);
9215
9216 /* Annotate start of buffers if we had overruns */
9217 if (ring_buffer_overruns(iter->array_buffer->buffer))
9218 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9219
9220 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9221 if (trace_clocks[iter->tr->clock_id].in_ns)
9222 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9223}
9224
9225void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
9226{
9227 /* use static because iter can be a bit big for the stack */
9228 static struct trace_iterator iter;
9229 static atomic_t dump_running;
9230 struct trace_array *tr = &global_trace;
9231 unsigned int old_userobj;
9232 unsigned long flags;
9233 int cnt = 0, cpu;
9234
9235 /* Only allow one dump user at a time. */
9236 if (atomic_inc_return(&dump_running) != 1) {
9237 atomic_dec(&dump_running);
9238 return;
9239 }
9240
9241 /*
9242 * Always turn off tracing when we dump.
9243 * We don't need to show trace output of what happens
9244 * between multiple crashes.
9245 *
9246 * If the user does a sysrq-z, then they can re-enable
9247 * tracing with echo 1 > tracing_on.
9248 */
9249 tracing_off();
9250
9251 local_irq_save(flags);
9252 printk_nmi_direct_enter();
9253
9254 /* Simulate the iterator */
9255 trace_init_global_iter(&iter);
9256 /* Can not use kmalloc for iter.temp */
9257 iter.temp = static_temp_buf;
9258 iter.temp_size = STATIC_TEMP_BUF_SIZE;
9259
9260 for_each_tracing_cpu(cpu) {
9261 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9262 }
9263
9264 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
9265
9266 /* don't look at user memory in panic mode */
9267 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
9268
9269 switch (oops_dump_mode) {
9270 case DUMP_ALL:
9271 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9272 break;
9273 case DUMP_ORIG:
9274 iter.cpu_file = raw_smp_processor_id();
9275 break;
9276 case DUMP_NONE:
9277 goto out_enable;
9278 default:
9279 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
9280 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9281 }
9282
9283 printk(KERN_TRACE "Dumping ftrace buffer:\n");
9284
9285 /* Did function tracer already get disabled? */
9286 if (ftrace_is_dead()) {
9287 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
9288 printk("# MAY BE MISSING FUNCTION EVENTS\n");
9289 }
9290
9291 /*
9292 * We need to stop all tracing on all CPUS to read the
9293 * the next buffer. This is a bit expensive, but is
9294 * not done often. We fill all what we can read,
9295 * and then release the locks again.
9296 */
9297
9298 while (!trace_empty(&iter)) {
9299
9300 if (!cnt)
9301 printk(KERN_TRACE "---------------------------------\n");
9302
9303 cnt++;
9304
9305 trace_iterator_reset(&iter);
9306 iter.iter_flags |= TRACE_FILE_LAT_FMT;
9307
9308 if (trace_find_next_entry_inc(&iter) != NULL) {
9309 int ret;
9310
9311 ret = print_trace_line(&iter);
9312 if (ret != TRACE_TYPE_NO_CONSUME)
9313 trace_consume(&iter);
9314 }
9315 touch_nmi_watchdog();
9316
9317 trace_printk_seq(&iter.seq);
9318 }
9319
9320 if (!cnt)
9321 printk(KERN_TRACE " (ftrace buffer empty)\n");
9322 else
9323 printk(KERN_TRACE "---------------------------------\n");
9324
9325 out_enable:
9326 tr->trace_flags |= old_userobj;
9327
9328 for_each_tracing_cpu(cpu) {
9329 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9330 }
9331 atomic_dec(&dump_running);
9332 printk_nmi_direct_exit();
9333 local_irq_restore(flags);
9334}
9335EXPORT_SYMBOL_GPL(ftrace_dump);
9336
9337int trace_run_command(const char *buf, int (*createfn)(int, char **))
9338{
9339 char **argv;
9340 int argc, ret;
9341
9342 argc = 0;
9343 ret = 0;
9344 argv = argv_split(GFP_KERNEL, buf, &argc);
9345 if (!argv)
9346 return -ENOMEM;
9347
9348 if (argc)
9349 ret = createfn(argc, argv);
9350
9351 argv_free(argv);
9352
9353 return ret;
9354}
9355
9356#define WRITE_BUFSIZE 4096
9357
9358ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
9359 size_t count, loff_t *ppos,
9360 int (*createfn)(int, char **))
9361{
9362 char *kbuf, *buf, *tmp;
9363 int ret = 0;
9364 size_t done = 0;
9365 size_t size;
9366
9367 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
9368 if (!kbuf)
9369 return -ENOMEM;
9370
9371 while (done < count) {
9372 size = count - done;
9373
9374 if (size >= WRITE_BUFSIZE)
9375 size = WRITE_BUFSIZE - 1;
9376
9377 if (copy_from_user(kbuf, buffer + done, size)) {
9378 ret = -EFAULT;
9379 goto out;
9380 }
9381 kbuf[size] = '\0';
9382 buf = kbuf;
9383 do {
9384 tmp = strchr(buf, '\n');
9385 if (tmp) {
9386 *tmp = '\0';
9387 size = tmp - buf + 1;
9388 } else {
9389 size = strlen(buf);
9390 if (done + size < count) {
9391 if (buf != kbuf)
9392 break;
9393 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
9394 pr_warn("Line length is too long: Should be less than %d\n",
9395 WRITE_BUFSIZE - 2);
9396 ret = -EINVAL;
9397 goto out;
9398 }
9399 }
9400 done += size;
9401
9402 /* Remove comments */
9403 tmp = strchr(buf, '#');
9404
9405 if (tmp)
9406 *tmp = '\0';
9407
9408 ret = trace_run_command(buf, createfn);
9409 if (ret)
9410 goto out;
9411 buf += size;
9412
9413 } while (done < count);
9414 }
9415 ret = done;
9416
9417out:
9418 kfree(kbuf);
9419
9420 return ret;
9421}
9422
9423__init static int tracer_alloc_buffers(void)
9424{
9425 int ring_buf_size;
9426 int ret = -ENOMEM;
9427
9428
9429 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9430 pr_warn("Tracing disabled due to lockdown\n");
9431 return -EPERM;
9432 }
9433
9434 /*
9435 * Make sure we don't accidently add more trace options
9436 * than we have bits for.
9437 */
9438 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
9439
9440 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
9441 goto out;
9442
9443 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
9444 goto out_free_buffer_mask;
9445
9446 /* Only allocate trace_printk buffers if a trace_printk exists */
9447 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
9448 /* Must be called before global_trace.buffer is allocated */
9449 trace_printk_init_buffers();
9450
9451 /* To save memory, keep the ring buffer size to its minimum */
9452 if (ring_buffer_expanded)
9453 ring_buf_size = trace_buf_size;
9454 else
9455 ring_buf_size = 1;
9456
9457 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
9458 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
9459
9460 raw_spin_lock_init(&global_trace.start_lock);
9461
9462 /*
9463 * The prepare callbacks allocates some memory for the ring buffer. We
9464 * don't free the buffer if the if the CPU goes down. If we were to free
9465 * the buffer, then the user would lose any trace that was in the
9466 * buffer. The memory will be removed once the "instance" is removed.
9467 */
9468 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
9469 "trace/RB:preapre", trace_rb_cpu_prepare,
9470 NULL);
9471 if (ret < 0)
9472 goto out_free_cpumask;
9473 /* Used for event triggers */
9474 ret = -ENOMEM;
9475 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
9476 if (!temp_buffer)
9477 goto out_rm_hp_state;
9478
9479 if (trace_create_savedcmd() < 0)
9480 goto out_free_temp_buffer;
9481
9482 /* TODO: make the number of buffers hot pluggable with CPUS */
9483 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
9484 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
9485 goto out_free_savedcmd;
9486 }
9487
9488 if (global_trace.buffer_disabled)
9489 tracing_off();
9490
9491 if (trace_boot_clock) {
9492 ret = tracing_set_clock(&global_trace, trace_boot_clock);
9493 if (ret < 0)
9494 pr_warn("Trace clock %s not defined, going back to default\n",
9495 trace_boot_clock);
9496 }
9497
9498 /*
9499 * register_tracer() might reference current_trace, so it
9500 * needs to be set before we register anything. This is
9501 * just a bootstrap of current_trace anyway.
9502 */
9503 global_trace.current_trace = &nop_trace;
9504
9505 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9506
9507 ftrace_init_global_array_ops(&global_trace);
9508
9509 init_trace_flags_index(&global_trace);
9510
9511 register_tracer(&nop_trace);
9512
9513 /* Function tracing may start here (via kernel command line) */
9514 init_function_trace();
9515
9516 /* All seems OK, enable tracing */
9517 tracing_disabled = 0;
9518
9519 atomic_notifier_chain_register(&panic_notifier_list,
9520 &trace_panic_notifier);
9521
9522 register_die_notifier(&trace_die_notifier);
9523
9524 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
9525
9526 INIT_LIST_HEAD(&global_trace.systems);
9527 INIT_LIST_HEAD(&global_trace.events);
9528 INIT_LIST_HEAD(&global_trace.hist_vars);
9529 INIT_LIST_HEAD(&global_trace.err_log);
9530 list_add(&global_trace.list, &ftrace_trace_arrays);
9531
9532 apply_trace_boot_options();
9533
9534 register_snapshot_cmd();
9535
9536 return 0;
9537
9538out_free_savedcmd:
9539 free_saved_cmdlines_buffer(savedcmd);
9540out_free_temp_buffer:
9541 ring_buffer_free(temp_buffer);
9542out_rm_hp_state:
9543 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
9544out_free_cpumask:
9545 free_cpumask_var(global_trace.tracing_cpumask);
9546out_free_buffer_mask:
9547 free_cpumask_var(tracing_buffer_mask);
9548out:
9549 return ret;
9550}
9551
9552void __init early_trace_init(void)
9553{
9554 if (tracepoint_printk) {
9555 tracepoint_print_iter =
9556 kmalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
9557 if (MEM_FAIL(!tracepoint_print_iter,
9558 "Failed to allocate trace iterator\n"))
9559 tracepoint_printk = 0;
9560 else
9561 static_key_enable(&tracepoint_printk_key.key);
9562 }
9563 tracer_alloc_buffers();
9564}
9565
9566void __init trace_init(void)
9567{
9568 trace_event_init();
9569}
9570
9571__init static int clear_boot_tracer(void)
9572{
9573 /*
9574 * The default tracer at boot buffer is an init section.
9575 * This function is called in lateinit. If we did not
9576 * find the boot tracer, then clear it out, to prevent
9577 * later registration from accessing the buffer that is
9578 * about to be freed.
9579 */
9580 if (!default_bootup_tracer)
9581 return 0;
9582
9583 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
9584 default_bootup_tracer);
9585 default_bootup_tracer = NULL;
9586
9587 return 0;
9588}
9589
9590fs_initcall(tracer_init_tracefs);
9591late_initcall_sync(clear_boot_tracer);
9592
9593#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
9594__init static int tracing_set_default_clock(void)
9595{
9596 /* sched_clock_stable() is determined in late_initcall */
9597 if (!trace_boot_clock && !sched_clock_stable()) {
9598 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9599 pr_warn("Can not set tracing clock due to lockdown\n");
9600 return -EPERM;
9601 }
9602
9603 printk(KERN_WARNING
9604 "Unstable clock detected, switching default tracing clock to \"global\"\n"
9605 "If you want to keep using the local clock, then add:\n"
9606 " \"trace_clock=local\"\n"
9607 "on the kernel command line\n");
9608 tracing_set_clock(&global_trace, "global");
9609 }
9610
9611 return 0;
9612}
9613late_initcall_sync(tracing_set_default_clock);
9614#endif