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/*
2 * ring buffer based function tracer
3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6 *
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
9 *
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
13 */
14#include <linux/ring_buffer.h>
15#include <generated/utsrelease.h>
16#include <linux/stacktrace.h>
17#include <linux/writeback.h>
18#include <linux/kallsyms.h>
19#include <linux/seq_file.h>
20#include <linux/notifier.h>
21#include <linux/irqflags.h>
22#include <linux/debugfs.h>
23#include <linux/pagemap.h>
24#include <linux/hardirq.h>
25#include <linux/linkage.h>
26#include <linux/uaccess.h>
27#include <linux/kprobes.h>
28#include <linux/ftrace.h>
29#include <linux/module.h>
30#include <linux/percpu.h>
31#include <linux/splice.h>
32#include <linux/kdebug.h>
33#include <linux/string.h>
34#include <linux/rwsem.h>
35#include <linux/slab.h>
36#include <linux/ctype.h>
37#include <linux/init.h>
38#include <linux/poll.h>
39#include <linux/fs.h>
40
41#include "trace.h"
42#include "trace_output.h"
43
44/*
45 * On boot up, the ring buffer is set to the minimum size, so that
46 * we do not waste memory on systems that are not using tracing.
47 */
48int ring_buffer_expanded;
49
50/*
51 * We need to change this state when a selftest is running.
52 * A selftest will lurk into the ring-buffer to count the
53 * entries inserted during the selftest although some concurrent
54 * insertions into the ring-buffer such as trace_printk could occurred
55 * at the same time, giving false positive or negative results.
56 */
57static bool __read_mostly tracing_selftest_running;
58
59/*
60 * If a tracer is running, we do not want to run SELFTEST.
61 */
62bool __read_mostly tracing_selftest_disabled;
63
64/* For tracers that don't implement custom flags */
65static struct tracer_opt dummy_tracer_opt[] = {
66 { }
67};
68
69static struct tracer_flags dummy_tracer_flags = {
70 .val = 0,
71 .opts = dummy_tracer_opt
72};
73
74static int dummy_set_flag(u32 old_flags, u32 bit, int set)
75{
76 return 0;
77}
78
79/*
80 * Kill all tracing for good (never come back).
81 * It is initialized to 1 but will turn to zero if the initialization
82 * of the tracer is successful. But that is the only place that sets
83 * this back to zero.
84 */
85static int tracing_disabled = 1;
86
87DEFINE_PER_CPU(int, ftrace_cpu_disabled);
88
89static inline void ftrace_disable_cpu(void)
90{
91 preempt_disable();
92 __this_cpu_inc(ftrace_cpu_disabled);
93}
94
95static inline void ftrace_enable_cpu(void)
96{
97 __this_cpu_dec(ftrace_cpu_disabled);
98 preempt_enable();
99}
100
101cpumask_var_t __read_mostly tracing_buffer_mask;
102
103/*
104 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
105 *
106 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
107 * is set, then ftrace_dump is called. This will output the contents
108 * of the ftrace buffers to the console. This is very useful for
109 * capturing traces that lead to crashes and outputing it to a
110 * serial console.
111 *
112 * It is default off, but you can enable it with either specifying
113 * "ftrace_dump_on_oops" in the kernel command line, or setting
114 * /proc/sys/kernel/ftrace_dump_on_oops
115 * Set 1 if you want to dump buffers of all CPUs
116 * Set 2 if you want to dump the buffer of the CPU that triggered oops
117 */
118
119enum ftrace_dump_mode ftrace_dump_on_oops;
120
121static int tracing_set_tracer(const char *buf);
122
123#define MAX_TRACER_SIZE 100
124static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
125static char *default_bootup_tracer;
126
127static int __init set_cmdline_ftrace(char *str)
128{
129 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
130 default_bootup_tracer = bootup_tracer_buf;
131 /* We are using ftrace early, expand it */
132 ring_buffer_expanded = 1;
133 return 1;
134}
135__setup("ftrace=", set_cmdline_ftrace);
136
137static int __init set_ftrace_dump_on_oops(char *str)
138{
139 if (*str++ != '=' || !*str) {
140 ftrace_dump_on_oops = DUMP_ALL;
141 return 1;
142 }
143
144 if (!strcmp("orig_cpu", str)) {
145 ftrace_dump_on_oops = DUMP_ORIG;
146 return 1;
147 }
148
149 return 0;
150}
151__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
152
153unsigned long long ns2usecs(cycle_t nsec)
154{
155 nsec += 500;
156 do_div(nsec, 1000);
157 return nsec;
158}
159
160/*
161 * The global_trace is the descriptor that holds the tracing
162 * buffers for the live tracing. For each CPU, it contains
163 * a link list of pages that will store trace entries. The
164 * page descriptor of the pages in the memory is used to hold
165 * the link list by linking the lru item in the page descriptor
166 * to each of the pages in the buffer per CPU.
167 *
168 * For each active CPU there is a data field that holds the
169 * pages for the buffer for that CPU. Each CPU has the same number
170 * of pages allocated for its buffer.
171 */
172static struct trace_array global_trace;
173
174static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
175
176int filter_current_check_discard(struct ring_buffer *buffer,
177 struct ftrace_event_call *call, void *rec,
178 struct ring_buffer_event *event)
179{
180 return filter_check_discard(call, rec, buffer, event);
181}
182EXPORT_SYMBOL_GPL(filter_current_check_discard);
183
184cycle_t ftrace_now(int cpu)
185{
186 u64 ts;
187
188 /* Early boot up does not have a buffer yet */
189 if (!global_trace.buffer)
190 return trace_clock_local();
191
192 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
193 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
194
195 return ts;
196}
197
198/*
199 * The max_tr is used to snapshot the global_trace when a maximum
200 * latency is reached. Some tracers will use this to store a maximum
201 * trace while it continues examining live traces.
202 *
203 * The buffers for the max_tr are set up the same as the global_trace.
204 * When a snapshot is taken, the link list of the max_tr is swapped
205 * with the link list of the global_trace and the buffers are reset for
206 * the global_trace so the tracing can continue.
207 */
208static struct trace_array max_tr;
209
210static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
211
212/* tracer_enabled is used to toggle activation of a tracer */
213static int tracer_enabled = 1;
214
215/**
216 * tracing_is_enabled - return tracer_enabled status
217 *
218 * This function is used by other tracers to know the status
219 * of the tracer_enabled flag. Tracers may use this function
220 * to know if it should enable their features when starting
221 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
222 */
223int tracing_is_enabled(void)
224{
225 return tracer_enabled;
226}
227
228/*
229 * trace_buf_size is the size in bytes that is allocated
230 * for a buffer. Note, the number of bytes is always rounded
231 * to page size.
232 *
233 * This number is purposely set to a low number of 16384.
234 * If the dump on oops happens, it will be much appreciated
235 * to not have to wait for all that output. Anyway this can be
236 * boot time and run time configurable.
237 */
238#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
239
240static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
241
242/* trace_types holds a link list of available tracers. */
243static struct tracer *trace_types __read_mostly;
244
245/* current_trace points to the tracer that is currently active */
246static struct tracer *current_trace __read_mostly;
247
248/*
249 * trace_types_lock is used to protect the trace_types list.
250 */
251static DEFINE_MUTEX(trace_types_lock);
252
253/*
254 * serialize the access of the ring buffer
255 *
256 * ring buffer serializes readers, but it is low level protection.
257 * The validity of the events (which returns by ring_buffer_peek() ..etc)
258 * are not protected by ring buffer.
259 *
260 * The content of events may become garbage if we allow other process consumes
261 * these events concurrently:
262 * A) the page of the consumed events may become a normal page
263 * (not reader page) in ring buffer, and this page will be rewrited
264 * by events producer.
265 * B) The page of the consumed events may become a page for splice_read,
266 * and this page will be returned to system.
267 *
268 * These primitives allow multi process access to different cpu ring buffer
269 * concurrently.
270 *
271 * These primitives don't distinguish read-only and read-consume access.
272 * Multi read-only access are also serialized.
273 */
274
275#ifdef CONFIG_SMP
276static DECLARE_RWSEM(all_cpu_access_lock);
277static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
278
279static inline void trace_access_lock(int cpu)
280{
281 if (cpu == TRACE_PIPE_ALL_CPU) {
282 /* gain it for accessing the whole ring buffer. */
283 down_write(&all_cpu_access_lock);
284 } else {
285 /* gain it for accessing a cpu ring buffer. */
286
287 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
288 down_read(&all_cpu_access_lock);
289
290 /* Secondly block other access to this @cpu ring buffer. */
291 mutex_lock(&per_cpu(cpu_access_lock, cpu));
292 }
293}
294
295static inline void trace_access_unlock(int cpu)
296{
297 if (cpu == TRACE_PIPE_ALL_CPU) {
298 up_write(&all_cpu_access_lock);
299 } else {
300 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
301 up_read(&all_cpu_access_lock);
302 }
303}
304
305static inline void trace_access_lock_init(void)
306{
307 int cpu;
308
309 for_each_possible_cpu(cpu)
310 mutex_init(&per_cpu(cpu_access_lock, cpu));
311}
312
313#else
314
315static DEFINE_MUTEX(access_lock);
316
317static inline void trace_access_lock(int cpu)
318{
319 (void)cpu;
320 mutex_lock(&access_lock);
321}
322
323static inline void trace_access_unlock(int cpu)
324{
325 (void)cpu;
326 mutex_unlock(&access_lock);
327}
328
329static inline void trace_access_lock_init(void)
330{
331}
332
333#endif
334
335/* trace_wait is a waitqueue for tasks blocked on trace_poll */
336static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
337
338/* trace_flags holds trace_options default values */
339unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
340 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
341 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE;
342
343static int trace_stop_count;
344static DEFINE_SPINLOCK(tracing_start_lock);
345
346static void wakeup_work_handler(struct work_struct *work)
347{
348 wake_up(&trace_wait);
349}
350
351static DECLARE_DELAYED_WORK(wakeup_work, wakeup_work_handler);
352
353/**
354 * trace_wake_up - wake up tasks waiting for trace input
355 *
356 * Schedules a delayed work to wake up any task that is blocked on the
357 * trace_wait queue. These is used with trace_poll for tasks polling the
358 * trace.
359 */
360void trace_wake_up(void)
361{
362 const unsigned long delay = msecs_to_jiffies(2);
363
364 if (trace_flags & TRACE_ITER_BLOCK)
365 return;
366 schedule_delayed_work(&wakeup_work, delay);
367}
368
369static int __init set_buf_size(char *str)
370{
371 unsigned long buf_size;
372
373 if (!str)
374 return 0;
375 buf_size = memparse(str, &str);
376 /* nr_entries can not be zero */
377 if (buf_size == 0)
378 return 0;
379 trace_buf_size = buf_size;
380 return 1;
381}
382__setup("trace_buf_size=", set_buf_size);
383
384static int __init set_tracing_thresh(char *str)
385{
386 unsigned long threshhold;
387 int ret;
388
389 if (!str)
390 return 0;
391 ret = strict_strtoul(str, 0, &threshhold);
392 if (ret < 0)
393 return 0;
394 tracing_thresh = threshhold * 1000;
395 return 1;
396}
397__setup("tracing_thresh=", set_tracing_thresh);
398
399unsigned long nsecs_to_usecs(unsigned long nsecs)
400{
401 return nsecs / 1000;
402}
403
404/* These must match the bit postions in trace_iterator_flags */
405static const char *trace_options[] = {
406 "print-parent",
407 "sym-offset",
408 "sym-addr",
409 "verbose",
410 "raw",
411 "hex",
412 "bin",
413 "block",
414 "stacktrace",
415 "trace_printk",
416 "ftrace_preempt",
417 "branch",
418 "annotate",
419 "userstacktrace",
420 "sym-userobj",
421 "printk-msg-only",
422 "context-info",
423 "latency-format",
424 "sleep-time",
425 "graph-time",
426 "record-cmd",
427 "overwrite",
428 "disable_on_free",
429 NULL
430};
431
432static struct {
433 u64 (*func)(void);
434 const char *name;
435} trace_clocks[] = {
436 { trace_clock_local, "local" },
437 { trace_clock_global, "global" },
438};
439
440int trace_clock_id;
441
442/*
443 * trace_parser_get_init - gets the buffer for trace parser
444 */
445int trace_parser_get_init(struct trace_parser *parser, int size)
446{
447 memset(parser, 0, sizeof(*parser));
448
449 parser->buffer = kmalloc(size, GFP_KERNEL);
450 if (!parser->buffer)
451 return 1;
452
453 parser->size = size;
454 return 0;
455}
456
457/*
458 * trace_parser_put - frees the buffer for trace parser
459 */
460void trace_parser_put(struct trace_parser *parser)
461{
462 kfree(parser->buffer);
463}
464
465/*
466 * trace_get_user - reads the user input string separated by space
467 * (matched by isspace(ch))
468 *
469 * For each string found the 'struct trace_parser' is updated,
470 * and the function returns.
471 *
472 * Returns number of bytes read.
473 *
474 * See kernel/trace/trace.h for 'struct trace_parser' details.
475 */
476int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
477 size_t cnt, loff_t *ppos)
478{
479 char ch;
480 size_t read = 0;
481 ssize_t ret;
482
483 if (!*ppos)
484 trace_parser_clear(parser);
485
486 ret = get_user(ch, ubuf++);
487 if (ret)
488 goto out;
489
490 read++;
491 cnt--;
492
493 /*
494 * The parser is not finished with the last write,
495 * continue reading the user input without skipping spaces.
496 */
497 if (!parser->cont) {
498 /* skip white space */
499 while (cnt && isspace(ch)) {
500 ret = get_user(ch, ubuf++);
501 if (ret)
502 goto out;
503 read++;
504 cnt--;
505 }
506
507 /* only spaces were written */
508 if (isspace(ch)) {
509 *ppos += read;
510 ret = read;
511 goto out;
512 }
513
514 parser->idx = 0;
515 }
516
517 /* read the non-space input */
518 while (cnt && !isspace(ch)) {
519 if (parser->idx < parser->size - 1)
520 parser->buffer[parser->idx++] = ch;
521 else {
522 ret = -EINVAL;
523 goto out;
524 }
525 ret = get_user(ch, ubuf++);
526 if (ret)
527 goto out;
528 read++;
529 cnt--;
530 }
531
532 /* We either got finished input or we have to wait for another call. */
533 if (isspace(ch)) {
534 parser->buffer[parser->idx] = 0;
535 parser->cont = false;
536 } else {
537 parser->cont = true;
538 parser->buffer[parser->idx++] = ch;
539 }
540
541 *ppos += read;
542 ret = read;
543
544out:
545 return ret;
546}
547
548ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
549{
550 int len;
551 int ret;
552
553 if (!cnt)
554 return 0;
555
556 if (s->len <= s->readpos)
557 return -EBUSY;
558
559 len = s->len - s->readpos;
560 if (cnt > len)
561 cnt = len;
562 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
563 if (ret == cnt)
564 return -EFAULT;
565
566 cnt -= ret;
567
568 s->readpos += cnt;
569 return cnt;
570}
571
572static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
573{
574 int len;
575 void *ret;
576
577 if (s->len <= s->readpos)
578 return -EBUSY;
579
580 len = s->len - s->readpos;
581 if (cnt > len)
582 cnt = len;
583 ret = memcpy(buf, s->buffer + s->readpos, cnt);
584 if (!ret)
585 return -EFAULT;
586
587 s->readpos += cnt;
588 return cnt;
589}
590
591/*
592 * ftrace_max_lock is used to protect the swapping of buffers
593 * when taking a max snapshot. The buffers themselves are
594 * protected by per_cpu spinlocks. But the action of the swap
595 * needs its own lock.
596 *
597 * This is defined as a arch_spinlock_t in order to help
598 * with performance when lockdep debugging is enabled.
599 *
600 * It is also used in other places outside the update_max_tr
601 * so it needs to be defined outside of the
602 * CONFIG_TRACER_MAX_TRACE.
603 */
604static arch_spinlock_t ftrace_max_lock =
605 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
606
607unsigned long __read_mostly tracing_thresh;
608
609#ifdef CONFIG_TRACER_MAX_TRACE
610unsigned long __read_mostly tracing_max_latency;
611
612/*
613 * Copy the new maximum trace into the separate maximum-trace
614 * structure. (this way the maximum trace is permanently saved,
615 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
616 */
617static void
618__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
619{
620 struct trace_array_cpu *data = tr->data[cpu];
621 struct trace_array_cpu *max_data;
622
623 max_tr.cpu = cpu;
624 max_tr.time_start = data->preempt_timestamp;
625
626 max_data = max_tr.data[cpu];
627 max_data->saved_latency = tracing_max_latency;
628 max_data->critical_start = data->critical_start;
629 max_data->critical_end = data->critical_end;
630
631 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
632 max_data->pid = tsk->pid;
633 max_data->uid = task_uid(tsk);
634 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
635 max_data->policy = tsk->policy;
636 max_data->rt_priority = tsk->rt_priority;
637
638 /* record this tasks comm */
639 tracing_record_cmdline(tsk);
640}
641
642/**
643 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
644 * @tr: tracer
645 * @tsk: the task with the latency
646 * @cpu: The cpu that initiated the trace.
647 *
648 * Flip the buffers between the @tr and the max_tr and record information
649 * about which task was the cause of this latency.
650 */
651void
652update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
653{
654 struct ring_buffer *buf = tr->buffer;
655
656 if (trace_stop_count)
657 return;
658
659 WARN_ON_ONCE(!irqs_disabled());
660 if (!current_trace->use_max_tr) {
661 WARN_ON_ONCE(1);
662 return;
663 }
664 arch_spin_lock(&ftrace_max_lock);
665
666 tr->buffer = max_tr.buffer;
667 max_tr.buffer = buf;
668
669 __update_max_tr(tr, tsk, cpu);
670 arch_spin_unlock(&ftrace_max_lock);
671}
672
673/**
674 * update_max_tr_single - only copy one trace over, and reset the rest
675 * @tr - tracer
676 * @tsk - task with the latency
677 * @cpu - the cpu of the buffer to copy.
678 *
679 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
680 */
681void
682update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
683{
684 int ret;
685
686 if (trace_stop_count)
687 return;
688
689 WARN_ON_ONCE(!irqs_disabled());
690 if (!current_trace->use_max_tr) {
691 WARN_ON_ONCE(1);
692 return;
693 }
694
695 arch_spin_lock(&ftrace_max_lock);
696
697 ftrace_disable_cpu();
698
699 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
700
701 if (ret == -EBUSY) {
702 /*
703 * We failed to swap the buffer due to a commit taking
704 * place on this CPU. We fail to record, but we reset
705 * the max trace buffer (no one writes directly to it)
706 * and flag that it failed.
707 */
708 trace_array_printk(&max_tr, _THIS_IP_,
709 "Failed to swap buffers due to commit in progress\n");
710 }
711
712 ftrace_enable_cpu();
713
714 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
715
716 __update_max_tr(tr, tsk, cpu);
717 arch_spin_unlock(&ftrace_max_lock);
718}
719#endif /* CONFIG_TRACER_MAX_TRACE */
720
721/**
722 * register_tracer - register a tracer with the ftrace system.
723 * @type - the plugin for the tracer
724 *
725 * Register a new plugin tracer.
726 */
727int register_tracer(struct tracer *type)
728__releases(kernel_lock)
729__acquires(kernel_lock)
730{
731 struct tracer *t;
732 int ret = 0;
733
734 if (!type->name) {
735 pr_info("Tracer must have a name\n");
736 return -1;
737 }
738
739 if (strlen(type->name) >= MAX_TRACER_SIZE) {
740 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
741 return -1;
742 }
743
744 mutex_lock(&trace_types_lock);
745
746 tracing_selftest_running = true;
747
748 for (t = trace_types; t; t = t->next) {
749 if (strcmp(type->name, t->name) == 0) {
750 /* already found */
751 pr_info("Tracer %s already registered\n",
752 type->name);
753 ret = -1;
754 goto out;
755 }
756 }
757
758 if (!type->set_flag)
759 type->set_flag = &dummy_set_flag;
760 if (!type->flags)
761 type->flags = &dummy_tracer_flags;
762 else
763 if (!type->flags->opts)
764 type->flags->opts = dummy_tracer_opt;
765 if (!type->wait_pipe)
766 type->wait_pipe = default_wait_pipe;
767
768
769#ifdef CONFIG_FTRACE_STARTUP_TEST
770 if (type->selftest && !tracing_selftest_disabled) {
771 struct tracer *saved_tracer = current_trace;
772 struct trace_array *tr = &global_trace;
773
774 /*
775 * Run a selftest on this tracer.
776 * Here we reset the trace buffer, and set the current
777 * tracer to be this tracer. The tracer can then run some
778 * internal tracing to verify that everything is in order.
779 * If we fail, we do not register this tracer.
780 */
781 tracing_reset_online_cpus(tr);
782
783 current_trace = type;
784
785 /* If we expanded the buffers, make sure the max is expanded too */
786 if (ring_buffer_expanded && type->use_max_tr)
787 ring_buffer_resize(max_tr.buffer, trace_buf_size);
788
789 /* the test is responsible for initializing and enabling */
790 pr_info("Testing tracer %s: ", type->name);
791 ret = type->selftest(type, tr);
792 /* the test is responsible for resetting too */
793 current_trace = saved_tracer;
794 if (ret) {
795 printk(KERN_CONT "FAILED!\n");
796 goto out;
797 }
798 /* Only reset on passing, to avoid touching corrupted buffers */
799 tracing_reset_online_cpus(tr);
800
801 /* Shrink the max buffer again */
802 if (ring_buffer_expanded && type->use_max_tr)
803 ring_buffer_resize(max_tr.buffer, 1);
804
805 printk(KERN_CONT "PASSED\n");
806 }
807#endif
808
809 type->next = trace_types;
810 trace_types = type;
811
812 out:
813 tracing_selftest_running = false;
814 mutex_unlock(&trace_types_lock);
815
816 if (ret || !default_bootup_tracer)
817 goto out_unlock;
818
819 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
820 goto out_unlock;
821
822 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
823 /* Do we want this tracer to start on bootup? */
824 tracing_set_tracer(type->name);
825 default_bootup_tracer = NULL;
826 /* disable other selftests, since this will break it. */
827 tracing_selftest_disabled = 1;
828#ifdef CONFIG_FTRACE_STARTUP_TEST
829 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
830 type->name);
831#endif
832
833 out_unlock:
834 return ret;
835}
836
837void unregister_tracer(struct tracer *type)
838{
839 struct tracer **t;
840
841 mutex_lock(&trace_types_lock);
842 for (t = &trace_types; *t; t = &(*t)->next) {
843 if (*t == type)
844 goto found;
845 }
846 pr_info("Tracer %s not registered\n", type->name);
847 goto out;
848
849 found:
850 *t = (*t)->next;
851
852 if (type == current_trace && tracer_enabled) {
853 tracer_enabled = 0;
854 tracing_stop();
855 if (current_trace->stop)
856 current_trace->stop(&global_trace);
857 current_trace = &nop_trace;
858 }
859out:
860 mutex_unlock(&trace_types_lock);
861}
862
863static void __tracing_reset(struct ring_buffer *buffer, int cpu)
864{
865 ftrace_disable_cpu();
866 ring_buffer_reset_cpu(buffer, cpu);
867 ftrace_enable_cpu();
868}
869
870void tracing_reset(struct trace_array *tr, int cpu)
871{
872 struct ring_buffer *buffer = tr->buffer;
873
874 ring_buffer_record_disable(buffer);
875
876 /* Make sure all commits have finished */
877 synchronize_sched();
878 __tracing_reset(buffer, cpu);
879
880 ring_buffer_record_enable(buffer);
881}
882
883void tracing_reset_online_cpus(struct trace_array *tr)
884{
885 struct ring_buffer *buffer = tr->buffer;
886 int cpu;
887
888 ring_buffer_record_disable(buffer);
889
890 /* Make sure all commits have finished */
891 synchronize_sched();
892
893 tr->time_start = ftrace_now(tr->cpu);
894
895 for_each_online_cpu(cpu)
896 __tracing_reset(buffer, cpu);
897
898 ring_buffer_record_enable(buffer);
899}
900
901void tracing_reset_current(int cpu)
902{
903 tracing_reset(&global_trace, cpu);
904}
905
906void tracing_reset_current_online_cpus(void)
907{
908 tracing_reset_online_cpus(&global_trace);
909}
910
911#define SAVED_CMDLINES 128
912#define NO_CMDLINE_MAP UINT_MAX
913static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
914static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
915static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
916static int cmdline_idx;
917static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
918
919/* temporary disable recording */
920static atomic_t trace_record_cmdline_disabled __read_mostly;
921
922static void trace_init_cmdlines(void)
923{
924 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
925 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
926 cmdline_idx = 0;
927}
928
929int is_tracing_stopped(void)
930{
931 return trace_stop_count;
932}
933
934/**
935 * ftrace_off_permanent - disable all ftrace code permanently
936 *
937 * This should only be called when a serious anomally has
938 * been detected. This will turn off the function tracing,
939 * ring buffers, and other tracing utilites. It takes no
940 * locks and can be called from any context.
941 */
942void ftrace_off_permanent(void)
943{
944 tracing_disabled = 1;
945 ftrace_stop();
946 tracing_off_permanent();
947}
948
949/**
950 * tracing_start - quick start of the tracer
951 *
952 * If tracing is enabled but was stopped by tracing_stop,
953 * this will start the tracer back up.
954 */
955void tracing_start(void)
956{
957 struct ring_buffer *buffer;
958 unsigned long flags;
959
960 if (tracing_disabled)
961 return;
962
963 spin_lock_irqsave(&tracing_start_lock, flags);
964 if (--trace_stop_count) {
965 if (trace_stop_count < 0) {
966 /* Someone screwed up their debugging */
967 WARN_ON_ONCE(1);
968 trace_stop_count = 0;
969 }
970 goto out;
971 }
972
973 /* Prevent the buffers from switching */
974 arch_spin_lock(&ftrace_max_lock);
975
976 buffer = global_trace.buffer;
977 if (buffer)
978 ring_buffer_record_enable(buffer);
979
980 buffer = max_tr.buffer;
981 if (buffer)
982 ring_buffer_record_enable(buffer);
983
984 arch_spin_unlock(&ftrace_max_lock);
985
986 ftrace_start();
987 out:
988 spin_unlock_irqrestore(&tracing_start_lock, flags);
989}
990
991/**
992 * tracing_stop - quick stop of the tracer
993 *
994 * Light weight way to stop tracing. Use in conjunction with
995 * tracing_start.
996 */
997void tracing_stop(void)
998{
999 struct ring_buffer *buffer;
1000 unsigned long flags;
1001
1002 ftrace_stop();
1003 spin_lock_irqsave(&tracing_start_lock, flags);
1004 if (trace_stop_count++)
1005 goto out;
1006
1007 /* Prevent the buffers from switching */
1008 arch_spin_lock(&ftrace_max_lock);
1009
1010 buffer = global_trace.buffer;
1011 if (buffer)
1012 ring_buffer_record_disable(buffer);
1013
1014 buffer = max_tr.buffer;
1015 if (buffer)
1016 ring_buffer_record_disable(buffer);
1017
1018 arch_spin_unlock(&ftrace_max_lock);
1019
1020 out:
1021 spin_unlock_irqrestore(&tracing_start_lock, flags);
1022}
1023
1024void trace_stop_cmdline_recording(void);
1025
1026static void trace_save_cmdline(struct task_struct *tsk)
1027{
1028 unsigned pid, idx;
1029
1030 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1031 return;
1032
1033 /*
1034 * It's not the end of the world if we don't get
1035 * the lock, but we also don't want to spin
1036 * nor do we want to disable interrupts,
1037 * so if we miss here, then better luck next time.
1038 */
1039 if (!arch_spin_trylock(&trace_cmdline_lock))
1040 return;
1041
1042 idx = map_pid_to_cmdline[tsk->pid];
1043 if (idx == NO_CMDLINE_MAP) {
1044 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1045
1046 /*
1047 * Check whether the cmdline buffer at idx has a pid
1048 * mapped. We are going to overwrite that entry so we
1049 * need to clear the map_pid_to_cmdline. Otherwise we
1050 * would read the new comm for the old pid.
1051 */
1052 pid = map_cmdline_to_pid[idx];
1053 if (pid != NO_CMDLINE_MAP)
1054 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1055
1056 map_cmdline_to_pid[idx] = tsk->pid;
1057 map_pid_to_cmdline[tsk->pid] = idx;
1058
1059 cmdline_idx = idx;
1060 }
1061
1062 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1063
1064 arch_spin_unlock(&trace_cmdline_lock);
1065}
1066
1067void trace_find_cmdline(int pid, char comm[])
1068{
1069 unsigned map;
1070
1071 if (!pid) {
1072 strcpy(comm, "<idle>");
1073 return;
1074 }
1075
1076 if (WARN_ON_ONCE(pid < 0)) {
1077 strcpy(comm, "<XXX>");
1078 return;
1079 }
1080
1081 if (pid > PID_MAX_DEFAULT) {
1082 strcpy(comm, "<...>");
1083 return;
1084 }
1085
1086 preempt_disable();
1087 arch_spin_lock(&trace_cmdline_lock);
1088 map = map_pid_to_cmdline[pid];
1089 if (map != NO_CMDLINE_MAP)
1090 strcpy(comm, saved_cmdlines[map]);
1091 else
1092 strcpy(comm, "<...>");
1093
1094 arch_spin_unlock(&trace_cmdline_lock);
1095 preempt_enable();
1096}
1097
1098void tracing_record_cmdline(struct task_struct *tsk)
1099{
1100 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1101 !tracing_is_on())
1102 return;
1103
1104 trace_save_cmdline(tsk);
1105}
1106
1107void
1108tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1109 int pc)
1110{
1111 struct task_struct *tsk = current;
1112
1113 entry->preempt_count = pc & 0xff;
1114 entry->pid = (tsk) ? tsk->pid : 0;
1115 entry->padding = 0;
1116 entry->flags =
1117#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1118 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1119#else
1120 TRACE_FLAG_IRQS_NOSUPPORT |
1121#endif
1122 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1123 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1124 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1125}
1126EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1127
1128struct ring_buffer_event *
1129trace_buffer_lock_reserve(struct ring_buffer *buffer,
1130 int type,
1131 unsigned long len,
1132 unsigned long flags, int pc)
1133{
1134 struct ring_buffer_event *event;
1135
1136 event = ring_buffer_lock_reserve(buffer, len);
1137 if (event != NULL) {
1138 struct trace_entry *ent = ring_buffer_event_data(event);
1139
1140 tracing_generic_entry_update(ent, flags, pc);
1141 ent->type = type;
1142 }
1143
1144 return event;
1145}
1146
1147static inline void
1148__trace_buffer_unlock_commit(struct ring_buffer *buffer,
1149 struct ring_buffer_event *event,
1150 unsigned long flags, int pc,
1151 int wake)
1152{
1153 ring_buffer_unlock_commit(buffer, event);
1154
1155 ftrace_trace_stack(buffer, flags, 6, pc);
1156 ftrace_trace_userstack(buffer, flags, pc);
1157
1158 if (wake)
1159 trace_wake_up();
1160}
1161
1162void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1163 struct ring_buffer_event *event,
1164 unsigned long flags, int pc)
1165{
1166 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1167}
1168
1169struct ring_buffer_event *
1170trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1171 int type, unsigned long len,
1172 unsigned long flags, int pc)
1173{
1174 *current_rb = global_trace.buffer;
1175 return trace_buffer_lock_reserve(*current_rb,
1176 type, len, flags, pc);
1177}
1178EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1179
1180void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1181 struct ring_buffer_event *event,
1182 unsigned long flags, int pc)
1183{
1184 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1185}
1186EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1187
1188void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1189 struct ring_buffer_event *event,
1190 unsigned long flags, int pc)
1191{
1192 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1193}
1194EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1195
1196void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1197 struct ring_buffer_event *event,
1198 unsigned long flags, int pc,
1199 struct pt_regs *regs)
1200{
1201 ring_buffer_unlock_commit(buffer, event);
1202
1203 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1204 ftrace_trace_userstack(buffer, flags, pc);
1205}
1206EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs);
1207
1208void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1209 struct ring_buffer_event *event)
1210{
1211 ring_buffer_discard_commit(buffer, event);
1212}
1213EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1214
1215void
1216trace_function(struct trace_array *tr,
1217 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1218 int pc)
1219{
1220 struct ftrace_event_call *call = &event_function;
1221 struct ring_buffer *buffer = tr->buffer;
1222 struct ring_buffer_event *event;
1223 struct ftrace_entry *entry;
1224
1225 /* If we are reading the ring buffer, don't trace */
1226 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1227 return;
1228
1229 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1230 flags, pc);
1231 if (!event)
1232 return;
1233 entry = ring_buffer_event_data(event);
1234 entry->ip = ip;
1235 entry->parent_ip = parent_ip;
1236
1237 if (!filter_check_discard(call, entry, buffer, event))
1238 ring_buffer_unlock_commit(buffer, event);
1239}
1240
1241void
1242ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1243 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1244 int pc)
1245{
1246 if (likely(!atomic_read(&data->disabled)))
1247 trace_function(tr, ip, parent_ip, flags, pc);
1248}
1249
1250#ifdef CONFIG_STACKTRACE
1251
1252#define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1253struct ftrace_stack {
1254 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1255};
1256
1257static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1258static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1259
1260static void __ftrace_trace_stack(struct ring_buffer *buffer,
1261 unsigned long flags,
1262 int skip, int pc, struct pt_regs *regs)
1263{
1264 struct ftrace_event_call *call = &event_kernel_stack;
1265 struct ring_buffer_event *event;
1266 struct stack_entry *entry;
1267 struct stack_trace trace;
1268 int use_stack;
1269 int size = FTRACE_STACK_ENTRIES;
1270
1271 trace.nr_entries = 0;
1272 trace.skip = skip;
1273
1274 /*
1275 * Since events can happen in NMIs there's no safe way to
1276 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1277 * or NMI comes in, it will just have to use the default
1278 * FTRACE_STACK_SIZE.
1279 */
1280 preempt_disable_notrace();
1281
1282 use_stack = ++__get_cpu_var(ftrace_stack_reserve);
1283 /*
1284 * We don't need any atomic variables, just a barrier.
1285 * If an interrupt comes in, we don't care, because it would
1286 * have exited and put the counter back to what we want.
1287 * We just need a barrier to keep gcc from moving things
1288 * around.
1289 */
1290 barrier();
1291 if (use_stack == 1) {
1292 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1293 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1294
1295 if (regs)
1296 save_stack_trace_regs(regs, &trace);
1297 else
1298 save_stack_trace(&trace);
1299
1300 if (trace.nr_entries > size)
1301 size = trace.nr_entries;
1302 } else
1303 /* From now on, use_stack is a boolean */
1304 use_stack = 0;
1305
1306 size *= sizeof(unsigned long);
1307
1308 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1309 sizeof(*entry) + size, flags, pc);
1310 if (!event)
1311 goto out;
1312 entry = ring_buffer_event_data(event);
1313
1314 memset(&entry->caller, 0, size);
1315
1316 if (use_stack)
1317 memcpy(&entry->caller, trace.entries,
1318 trace.nr_entries * sizeof(unsigned long));
1319 else {
1320 trace.max_entries = FTRACE_STACK_ENTRIES;
1321 trace.entries = entry->caller;
1322 if (regs)
1323 save_stack_trace_regs(regs, &trace);
1324 else
1325 save_stack_trace(&trace);
1326 }
1327
1328 entry->size = trace.nr_entries;
1329
1330 if (!filter_check_discard(call, entry, buffer, event))
1331 ring_buffer_unlock_commit(buffer, event);
1332
1333 out:
1334 /* Again, don't let gcc optimize things here */
1335 barrier();
1336 __get_cpu_var(ftrace_stack_reserve)--;
1337 preempt_enable_notrace();
1338
1339}
1340
1341void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1342 int skip, int pc, struct pt_regs *regs)
1343{
1344 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1345 return;
1346
1347 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1348}
1349
1350void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1351 int skip, int pc)
1352{
1353 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1354 return;
1355
1356 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1357}
1358
1359void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1360 int pc)
1361{
1362 __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1363}
1364
1365/**
1366 * trace_dump_stack - record a stack back trace in the trace buffer
1367 */
1368void trace_dump_stack(void)
1369{
1370 unsigned long flags;
1371
1372 if (tracing_disabled || tracing_selftest_running)
1373 return;
1374
1375 local_save_flags(flags);
1376
1377 /* skipping 3 traces, seems to get us at the caller of this function */
1378 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1379}
1380
1381static DEFINE_PER_CPU(int, user_stack_count);
1382
1383void
1384ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1385{
1386 struct ftrace_event_call *call = &event_user_stack;
1387 struct ring_buffer_event *event;
1388 struct userstack_entry *entry;
1389 struct stack_trace trace;
1390
1391 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1392 return;
1393
1394 /*
1395 * NMIs can not handle page faults, even with fix ups.
1396 * The save user stack can (and often does) fault.
1397 */
1398 if (unlikely(in_nmi()))
1399 return;
1400
1401 /*
1402 * prevent recursion, since the user stack tracing may
1403 * trigger other kernel events.
1404 */
1405 preempt_disable();
1406 if (__this_cpu_read(user_stack_count))
1407 goto out;
1408
1409 __this_cpu_inc(user_stack_count);
1410
1411 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1412 sizeof(*entry), flags, pc);
1413 if (!event)
1414 goto out_drop_count;
1415 entry = ring_buffer_event_data(event);
1416
1417 entry->tgid = current->tgid;
1418 memset(&entry->caller, 0, sizeof(entry->caller));
1419
1420 trace.nr_entries = 0;
1421 trace.max_entries = FTRACE_STACK_ENTRIES;
1422 trace.skip = 0;
1423 trace.entries = entry->caller;
1424
1425 save_stack_trace_user(&trace);
1426 if (!filter_check_discard(call, entry, buffer, event))
1427 ring_buffer_unlock_commit(buffer, event);
1428
1429 out_drop_count:
1430 __this_cpu_dec(user_stack_count);
1431 out:
1432 preempt_enable();
1433}
1434
1435#ifdef UNUSED
1436static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1437{
1438 ftrace_trace_userstack(tr, flags, preempt_count());
1439}
1440#endif /* UNUSED */
1441
1442#endif /* CONFIG_STACKTRACE */
1443
1444/**
1445 * trace_vbprintk - write binary msg to tracing buffer
1446 *
1447 */
1448int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1449{
1450 static arch_spinlock_t trace_buf_lock =
1451 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1452 static u32 trace_buf[TRACE_BUF_SIZE];
1453
1454 struct ftrace_event_call *call = &event_bprint;
1455 struct ring_buffer_event *event;
1456 struct ring_buffer *buffer;
1457 struct trace_array *tr = &global_trace;
1458 struct trace_array_cpu *data;
1459 struct bprint_entry *entry;
1460 unsigned long flags;
1461 int disable;
1462 int cpu, len = 0, size, pc;
1463
1464 if (unlikely(tracing_selftest_running || tracing_disabled))
1465 return 0;
1466
1467 /* Don't pollute graph traces with trace_vprintk internals */
1468 pause_graph_tracing();
1469
1470 pc = preempt_count();
1471 preempt_disable_notrace();
1472 cpu = raw_smp_processor_id();
1473 data = tr->data[cpu];
1474
1475 disable = atomic_inc_return(&data->disabled);
1476 if (unlikely(disable != 1))
1477 goto out;
1478
1479 /* Lockdep uses trace_printk for lock tracing */
1480 local_irq_save(flags);
1481 arch_spin_lock(&trace_buf_lock);
1482 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1483
1484 if (len > TRACE_BUF_SIZE || len < 0)
1485 goto out_unlock;
1486
1487 size = sizeof(*entry) + sizeof(u32) * len;
1488 buffer = tr->buffer;
1489 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1490 flags, pc);
1491 if (!event)
1492 goto out_unlock;
1493 entry = ring_buffer_event_data(event);
1494 entry->ip = ip;
1495 entry->fmt = fmt;
1496
1497 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1498 if (!filter_check_discard(call, entry, buffer, event)) {
1499 ring_buffer_unlock_commit(buffer, event);
1500 ftrace_trace_stack(buffer, flags, 6, pc);
1501 }
1502
1503out_unlock:
1504 arch_spin_unlock(&trace_buf_lock);
1505 local_irq_restore(flags);
1506
1507out:
1508 atomic_dec_return(&data->disabled);
1509 preempt_enable_notrace();
1510 unpause_graph_tracing();
1511
1512 return len;
1513}
1514EXPORT_SYMBOL_GPL(trace_vbprintk);
1515
1516int trace_array_printk(struct trace_array *tr,
1517 unsigned long ip, const char *fmt, ...)
1518{
1519 int ret;
1520 va_list ap;
1521
1522 if (!(trace_flags & TRACE_ITER_PRINTK))
1523 return 0;
1524
1525 va_start(ap, fmt);
1526 ret = trace_array_vprintk(tr, ip, fmt, ap);
1527 va_end(ap);
1528 return ret;
1529}
1530
1531int trace_array_vprintk(struct trace_array *tr,
1532 unsigned long ip, const char *fmt, va_list args)
1533{
1534 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1535 static char trace_buf[TRACE_BUF_SIZE];
1536
1537 struct ftrace_event_call *call = &event_print;
1538 struct ring_buffer_event *event;
1539 struct ring_buffer *buffer;
1540 struct trace_array_cpu *data;
1541 int cpu, len = 0, size, pc;
1542 struct print_entry *entry;
1543 unsigned long irq_flags;
1544 int disable;
1545
1546 if (tracing_disabled || tracing_selftest_running)
1547 return 0;
1548
1549 pc = preempt_count();
1550 preempt_disable_notrace();
1551 cpu = raw_smp_processor_id();
1552 data = tr->data[cpu];
1553
1554 disable = atomic_inc_return(&data->disabled);
1555 if (unlikely(disable != 1))
1556 goto out;
1557
1558 pause_graph_tracing();
1559 raw_local_irq_save(irq_flags);
1560 arch_spin_lock(&trace_buf_lock);
1561 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1562
1563 size = sizeof(*entry) + len + 1;
1564 buffer = tr->buffer;
1565 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1566 irq_flags, pc);
1567 if (!event)
1568 goto out_unlock;
1569 entry = ring_buffer_event_data(event);
1570 entry->ip = ip;
1571
1572 memcpy(&entry->buf, trace_buf, len);
1573 entry->buf[len] = '\0';
1574 if (!filter_check_discard(call, entry, buffer, event)) {
1575 ring_buffer_unlock_commit(buffer, event);
1576 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1577 }
1578
1579 out_unlock:
1580 arch_spin_unlock(&trace_buf_lock);
1581 raw_local_irq_restore(irq_flags);
1582 unpause_graph_tracing();
1583 out:
1584 atomic_dec_return(&data->disabled);
1585 preempt_enable_notrace();
1586
1587 return len;
1588}
1589
1590int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1591{
1592 return trace_array_vprintk(&global_trace, ip, fmt, args);
1593}
1594EXPORT_SYMBOL_GPL(trace_vprintk);
1595
1596static void trace_iterator_increment(struct trace_iterator *iter)
1597{
1598 /* Don't allow ftrace to trace into the ring buffers */
1599 ftrace_disable_cpu();
1600
1601 iter->idx++;
1602 if (iter->buffer_iter[iter->cpu])
1603 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1604
1605 ftrace_enable_cpu();
1606}
1607
1608static struct trace_entry *
1609peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1610 unsigned long *lost_events)
1611{
1612 struct ring_buffer_event *event;
1613 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1614
1615 /* Don't allow ftrace to trace into the ring buffers */
1616 ftrace_disable_cpu();
1617
1618 if (buf_iter)
1619 event = ring_buffer_iter_peek(buf_iter, ts);
1620 else
1621 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1622 lost_events);
1623
1624 ftrace_enable_cpu();
1625
1626 if (event) {
1627 iter->ent_size = ring_buffer_event_length(event);
1628 return ring_buffer_event_data(event);
1629 }
1630 iter->ent_size = 0;
1631 return NULL;
1632}
1633
1634static struct trace_entry *
1635__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1636 unsigned long *missing_events, u64 *ent_ts)
1637{
1638 struct ring_buffer *buffer = iter->tr->buffer;
1639 struct trace_entry *ent, *next = NULL;
1640 unsigned long lost_events = 0, next_lost = 0;
1641 int cpu_file = iter->cpu_file;
1642 u64 next_ts = 0, ts;
1643 int next_cpu = -1;
1644 int cpu;
1645
1646 /*
1647 * If we are in a per_cpu trace file, don't bother by iterating over
1648 * all cpu and peek directly.
1649 */
1650 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1651 if (ring_buffer_empty_cpu(buffer, cpu_file))
1652 return NULL;
1653 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1654 if (ent_cpu)
1655 *ent_cpu = cpu_file;
1656
1657 return ent;
1658 }
1659
1660 for_each_tracing_cpu(cpu) {
1661
1662 if (ring_buffer_empty_cpu(buffer, cpu))
1663 continue;
1664
1665 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1666
1667 /*
1668 * Pick the entry with the smallest timestamp:
1669 */
1670 if (ent && (!next || ts < next_ts)) {
1671 next = ent;
1672 next_cpu = cpu;
1673 next_ts = ts;
1674 next_lost = lost_events;
1675 }
1676 }
1677
1678 if (ent_cpu)
1679 *ent_cpu = next_cpu;
1680
1681 if (ent_ts)
1682 *ent_ts = next_ts;
1683
1684 if (missing_events)
1685 *missing_events = next_lost;
1686
1687 return next;
1688}
1689
1690/* Find the next real entry, without updating the iterator itself */
1691struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1692 int *ent_cpu, u64 *ent_ts)
1693{
1694 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1695}
1696
1697/* Find the next real entry, and increment the iterator to the next entry */
1698void *trace_find_next_entry_inc(struct trace_iterator *iter)
1699{
1700 iter->ent = __find_next_entry(iter, &iter->cpu,
1701 &iter->lost_events, &iter->ts);
1702
1703 if (iter->ent)
1704 trace_iterator_increment(iter);
1705
1706 return iter->ent ? iter : NULL;
1707}
1708
1709static void trace_consume(struct trace_iterator *iter)
1710{
1711 /* Don't allow ftrace to trace into the ring buffers */
1712 ftrace_disable_cpu();
1713 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1714 &iter->lost_events);
1715 ftrace_enable_cpu();
1716}
1717
1718static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1719{
1720 struct trace_iterator *iter = m->private;
1721 int i = (int)*pos;
1722 void *ent;
1723
1724 WARN_ON_ONCE(iter->leftover);
1725
1726 (*pos)++;
1727
1728 /* can't go backwards */
1729 if (iter->idx > i)
1730 return NULL;
1731
1732 if (iter->idx < 0)
1733 ent = trace_find_next_entry_inc(iter);
1734 else
1735 ent = iter;
1736
1737 while (ent && iter->idx < i)
1738 ent = trace_find_next_entry_inc(iter);
1739
1740 iter->pos = *pos;
1741
1742 return ent;
1743}
1744
1745void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1746{
1747 struct trace_array *tr = iter->tr;
1748 struct ring_buffer_event *event;
1749 struct ring_buffer_iter *buf_iter;
1750 unsigned long entries = 0;
1751 u64 ts;
1752
1753 tr->data[cpu]->skipped_entries = 0;
1754
1755 if (!iter->buffer_iter[cpu])
1756 return;
1757
1758 buf_iter = iter->buffer_iter[cpu];
1759 ring_buffer_iter_reset(buf_iter);
1760
1761 /*
1762 * We could have the case with the max latency tracers
1763 * that a reset never took place on a cpu. This is evident
1764 * by the timestamp being before the start of the buffer.
1765 */
1766 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1767 if (ts >= iter->tr->time_start)
1768 break;
1769 entries++;
1770 ring_buffer_read(buf_iter, NULL);
1771 }
1772
1773 tr->data[cpu]->skipped_entries = entries;
1774}
1775
1776/*
1777 * The current tracer is copied to avoid a global locking
1778 * all around.
1779 */
1780static void *s_start(struct seq_file *m, loff_t *pos)
1781{
1782 struct trace_iterator *iter = m->private;
1783 static struct tracer *old_tracer;
1784 int cpu_file = iter->cpu_file;
1785 void *p = NULL;
1786 loff_t l = 0;
1787 int cpu;
1788
1789 /* copy the tracer to avoid using a global lock all around */
1790 mutex_lock(&trace_types_lock);
1791 if (unlikely(old_tracer != current_trace && current_trace)) {
1792 old_tracer = current_trace;
1793 *iter->trace = *current_trace;
1794 }
1795 mutex_unlock(&trace_types_lock);
1796
1797 atomic_inc(&trace_record_cmdline_disabled);
1798
1799 if (*pos != iter->pos) {
1800 iter->ent = NULL;
1801 iter->cpu = 0;
1802 iter->idx = -1;
1803
1804 ftrace_disable_cpu();
1805
1806 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1807 for_each_tracing_cpu(cpu)
1808 tracing_iter_reset(iter, cpu);
1809 } else
1810 tracing_iter_reset(iter, cpu_file);
1811
1812 ftrace_enable_cpu();
1813
1814 iter->leftover = 0;
1815 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1816 ;
1817
1818 } else {
1819 /*
1820 * If we overflowed the seq_file before, then we want
1821 * to just reuse the trace_seq buffer again.
1822 */
1823 if (iter->leftover)
1824 p = iter;
1825 else {
1826 l = *pos - 1;
1827 p = s_next(m, p, &l);
1828 }
1829 }
1830
1831 trace_event_read_lock();
1832 trace_access_lock(cpu_file);
1833 return p;
1834}
1835
1836static void s_stop(struct seq_file *m, void *p)
1837{
1838 struct trace_iterator *iter = m->private;
1839
1840 atomic_dec(&trace_record_cmdline_disabled);
1841 trace_access_unlock(iter->cpu_file);
1842 trace_event_read_unlock();
1843}
1844
1845static void print_lat_help_header(struct seq_file *m)
1846{
1847 seq_puts(m, "# _------=> CPU# \n");
1848 seq_puts(m, "# / _-----=> irqs-off \n");
1849 seq_puts(m, "# | / _----=> need-resched \n");
1850 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1851 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1852 seq_puts(m, "# |||| / delay \n");
1853 seq_puts(m, "# cmd pid ||||| time | caller \n");
1854 seq_puts(m, "# \\ / ||||| \\ | / \n");
1855}
1856
1857static void print_func_help_header(struct seq_file *m)
1858{
1859 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1860 seq_puts(m, "# | | | | |\n");
1861}
1862
1863
1864void
1865print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1866{
1867 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1868 struct trace_array *tr = iter->tr;
1869 struct trace_array_cpu *data = tr->data[tr->cpu];
1870 struct tracer *type = current_trace;
1871 unsigned long entries = 0;
1872 unsigned long total = 0;
1873 unsigned long count;
1874 const char *name = "preemption";
1875 int cpu;
1876
1877 if (type)
1878 name = type->name;
1879
1880
1881 for_each_tracing_cpu(cpu) {
1882 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1883 /*
1884 * If this buffer has skipped entries, then we hold all
1885 * entries for the trace and we need to ignore the
1886 * ones before the time stamp.
1887 */
1888 if (tr->data[cpu]->skipped_entries) {
1889 count -= tr->data[cpu]->skipped_entries;
1890 /* total is the same as the entries */
1891 total += count;
1892 } else
1893 total += count +
1894 ring_buffer_overrun_cpu(tr->buffer, cpu);
1895 entries += count;
1896 }
1897
1898 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1899 name, UTS_RELEASE);
1900 seq_puts(m, "# -----------------------------------"
1901 "---------------------------------\n");
1902 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1903 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1904 nsecs_to_usecs(data->saved_latency),
1905 entries,
1906 total,
1907 tr->cpu,
1908#if defined(CONFIG_PREEMPT_NONE)
1909 "server",
1910#elif defined(CONFIG_PREEMPT_VOLUNTARY)
1911 "desktop",
1912#elif defined(CONFIG_PREEMPT)
1913 "preempt",
1914#else
1915 "unknown",
1916#endif
1917 /* These are reserved for later use */
1918 0, 0, 0, 0);
1919#ifdef CONFIG_SMP
1920 seq_printf(m, " #P:%d)\n", num_online_cpus());
1921#else
1922 seq_puts(m, ")\n");
1923#endif
1924 seq_puts(m, "# -----------------\n");
1925 seq_printf(m, "# | task: %.16s-%d "
1926 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1927 data->comm, data->pid, data->uid, data->nice,
1928 data->policy, data->rt_priority);
1929 seq_puts(m, "# -----------------\n");
1930
1931 if (data->critical_start) {
1932 seq_puts(m, "# => started at: ");
1933 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1934 trace_print_seq(m, &iter->seq);
1935 seq_puts(m, "\n# => ended at: ");
1936 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1937 trace_print_seq(m, &iter->seq);
1938 seq_puts(m, "\n#\n");
1939 }
1940
1941 seq_puts(m, "#\n");
1942}
1943
1944static void test_cpu_buff_start(struct trace_iterator *iter)
1945{
1946 struct trace_seq *s = &iter->seq;
1947
1948 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1949 return;
1950
1951 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1952 return;
1953
1954 if (cpumask_test_cpu(iter->cpu, iter->started))
1955 return;
1956
1957 if (iter->tr->data[iter->cpu]->skipped_entries)
1958 return;
1959
1960 cpumask_set_cpu(iter->cpu, iter->started);
1961
1962 /* Don't print started cpu buffer for the first entry of the trace */
1963 if (iter->idx > 1)
1964 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1965 iter->cpu);
1966}
1967
1968static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1969{
1970 struct trace_seq *s = &iter->seq;
1971 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1972 struct trace_entry *entry;
1973 struct trace_event *event;
1974
1975 entry = iter->ent;
1976
1977 test_cpu_buff_start(iter);
1978
1979 event = ftrace_find_event(entry->type);
1980
1981 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1982 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1983 if (!trace_print_lat_context(iter))
1984 goto partial;
1985 } else {
1986 if (!trace_print_context(iter))
1987 goto partial;
1988 }
1989 }
1990
1991 if (event)
1992 return event->funcs->trace(iter, sym_flags, event);
1993
1994 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1995 goto partial;
1996
1997 return TRACE_TYPE_HANDLED;
1998partial:
1999 return TRACE_TYPE_PARTIAL_LINE;
2000}
2001
2002static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2003{
2004 struct trace_seq *s = &iter->seq;
2005 struct trace_entry *entry;
2006 struct trace_event *event;
2007
2008 entry = iter->ent;
2009
2010 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2011 if (!trace_seq_printf(s, "%d %d %llu ",
2012 entry->pid, iter->cpu, iter->ts))
2013 goto partial;
2014 }
2015
2016 event = ftrace_find_event(entry->type);
2017 if (event)
2018 return event->funcs->raw(iter, 0, event);
2019
2020 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2021 goto partial;
2022
2023 return TRACE_TYPE_HANDLED;
2024partial:
2025 return TRACE_TYPE_PARTIAL_LINE;
2026}
2027
2028static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2029{
2030 struct trace_seq *s = &iter->seq;
2031 unsigned char newline = '\n';
2032 struct trace_entry *entry;
2033 struct trace_event *event;
2034
2035 entry = iter->ent;
2036
2037 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2038 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2039 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2040 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2041 }
2042
2043 event = ftrace_find_event(entry->type);
2044 if (event) {
2045 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2046 if (ret != TRACE_TYPE_HANDLED)
2047 return ret;
2048 }
2049
2050 SEQ_PUT_FIELD_RET(s, newline);
2051
2052 return TRACE_TYPE_HANDLED;
2053}
2054
2055static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2056{
2057 struct trace_seq *s = &iter->seq;
2058 struct trace_entry *entry;
2059 struct trace_event *event;
2060
2061 entry = iter->ent;
2062
2063 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2064 SEQ_PUT_FIELD_RET(s, entry->pid);
2065 SEQ_PUT_FIELD_RET(s, iter->cpu);
2066 SEQ_PUT_FIELD_RET(s, iter->ts);
2067 }
2068
2069 event = ftrace_find_event(entry->type);
2070 return event ? event->funcs->binary(iter, 0, event) :
2071 TRACE_TYPE_HANDLED;
2072}
2073
2074int trace_empty(struct trace_iterator *iter)
2075{
2076 int cpu;
2077
2078 /* If we are looking at one CPU buffer, only check that one */
2079 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2080 cpu = iter->cpu_file;
2081 if (iter->buffer_iter[cpu]) {
2082 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2083 return 0;
2084 } else {
2085 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2086 return 0;
2087 }
2088 return 1;
2089 }
2090
2091 for_each_tracing_cpu(cpu) {
2092 if (iter->buffer_iter[cpu]) {
2093 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2094 return 0;
2095 } else {
2096 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2097 return 0;
2098 }
2099 }
2100
2101 return 1;
2102}
2103
2104/* Called with trace_event_read_lock() held. */
2105enum print_line_t print_trace_line(struct trace_iterator *iter)
2106{
2107 enum print_line_t ret;
2108
2109 if (iter->lost_events &&
2110 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2111 iter->cpu, iter->lost_events))
2112 return TRACE_TYPE_PARTIAL_LINE;
2113
2114 if (iter->trace && iter->trace->print_line) {
2115 ret = iter->trace->print_line(iter);
2116 if (ret != TRACE_TYPE_UNHANDLED)
2117 return ret;
2118 }
2119
2120 if (iter->ent->type == TRACE_BPRINT &&
2121 trace_flags & TRACE_ITER_PRINTK &&
2122 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2123 return trace_print_bprintk_msg_only(iter);
2124
2125 if (iter->ent->type == TRACE_PRINT &&
2126 trace_flags & TRACE_ITER_PRINTK &&
2127 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2128 return trace_print_printk_msg_only(iter);
2129
2130 if (trace_flags & TRACE_ITER_BIN)
2131 return print_bin_fmt(iter);
2132
2133 if (trace_flags & TRACE_ITER_HEX)
2134 return print_hex_fmt(iter);
2135
2136 if (trace_flags & TRACE_ITER_RAW)
2137 return print_raw_fmt(iter);
2138
2139 return print_trace_fmt(iter);
2140}
2141
2142void trace_default_header(struct seq_file *m)
2143{
2144 struct trace_iterator *iter = m->private;
2145
2146 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2147 return;
2148
2149 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2150 /* print nothing if the buffers are empty */
2151 if (trace_empty(iter))
2152 return;
2153 print_trace_header(m, iter);
2154 if (!(trace_flags & TRACE_ITER_VERBOSE))
2155 print_lat_help_header(m);
2156 } else {
2157 if (!(trace_flags & TRACE_ITER_VERBOSE))
2158 print_func_help_header(m);
2159 }
2160}
2161
2162static int s_show(struct seq_file *m, void *v)
2163{
2164 struct trace_iterator *iter = v;
2165 int ret;
2166
2167 if (iter->ent == NULL) {
2168 if (iter->tr) {
2169 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2170 seq_puts(m, "#\n");
2171 }
2172 if (iter->trace && iter->trace->print_header)
2173 iter->trace->print_header(m);
2174 else
2175 trace_default_header(m);
2176
2177 } else if (iter->leftover) {
2178 /*
2179 * If we filled the seq_file buffer earlier, we
2180 * want to just show it now.
2181 */
2182 ret = trace_print_seq(m, &iter->seq);
2183
2184 /* ret should this time be zero, but you never know */
2185 iter->leftover = ret;
2186
2187 } else {
2188 print_trace_line(iter);
2189 ret = trace_print_seq(m, &iter->seq);
2190 /*
2191 * If we overflow the seq_file buffer, then it will
2192 * ask us for this data again at start up.
2193 * Use that instead.
2194 * ret is 0 if seq_file write succeeded.
2195 * -1 otherwise.
2196 */
2197 iter->leftover = ret;
2198 }
2199
2200 return 0;
2201}
2202
2203static const struct seq_operations tracer_seq_ops = {
2204 .start = s_start,
2205 .next = s_next,
2206 .stop = s_stop,
2207 .show = s_show,
2208};
2209
2210static struct trace_iterator *
2211__tracing_open(struct inode *inode, struct file *file)
2212{
2213 long cpu_file = (long) inode->i_private;
2214 void *fail_ret = ERR_PTR(-ENOMEM);
2215 struct trace_iterator *iter;
2216 struct seq_file *m;
2217 int cpu, ret;
2218
2219 if (tracing_disabled)
2220 return ERR_PTR(-ENODEV);
2221
2222 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2223 if (!iter)
2224 return ERR_PTR(-ENOMEM);
2225
2226 /*
2227 * We make a copy of the current tracer to avoid concurrent
2228 * changes on it while we are reading.
2229 */
2230 mutex_lock(&trace_types_lock);
2231 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2232 if (!iter->trace)
2233 goto fail;
2234
2235 if (current_trace)
2236 *iter->trace = *current_trace;
2237
2238 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2239 goto fail;
2240
2241 if (current_trace && current_trace->print_max)
2242 iter->tr = &max_tr;
2243 else
2244 iter->tr = &global_trace;
2245 iter->pos = -1;
2246 mutex_init(&iter->mutex);
2247 iter->cpu_file = cpu_file;
2248
2249 /* Notify the tracer early; before we stop tracing. */
2250 if (iter->trace && iter->trace->open)
2251 iter->trace->open(iter);
2252
2253 /* Annotate start of buffers if we had overruns */
2254 if (ring_buffer_overruns(iter->tr->buffer))
2255 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2256
2257 /* stop the trace while dumping */
2258 tracing_stop();
2259
2260 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2261 for_each_tracing_cpu(cpu) {
2262 iter->buffer_iter[cpu] =
2263 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2264 }
2265 ring_buffer_read_prepare_sync();
2266 for_each_tracing_cpu(cpu) {
2267 ring_buffer_read_start(iter->buffer_iter[cpu]);
2268 tracing_iter_reset(iter, cpu);
2269 }
2270 } else {
2271 cpu = iter->cpu_file;
2272 iter->buffer_iter[cpu] =
2273 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2274 ring_buffer_read_prepare_sync();
2275 ring_buffer_read_start(iter->buffer_iter[cpu]);
2276 tracing_iter_reset(iter, cpu);
2277 }
2278
2279 ret = seq_open(file, &tracer_seq_ops);
2280 if (ret < 0) {
2281 fail_ret = ERR_PTR(ret);
2282 goto fail_buffer;
2283 }
2284
2285 m = file->private_data;
2286 m->private = iter;
2287
2288 mutex_unlock(&trace_types_lock);
2289
2290 return iter;
2291
2292 fail_buffer:
2293 for_each_tracing_cpu(cpu) {
2294 if (iter->buffer_iter[cpu])
2295 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2296 }
2297 free_cpumask_var(iter->started);
2298 tracing_start();
2299 fail:
2300 mutex_unlock(&trace_types_lock);
2301 kfree(iter->trace);
2302 kfree(iter);
2303
2304 return fail_ret;
2305}
2306
2307int tracing_open_generic(struct inode *inode, struct file *filp)
2308{
2309 if (tracing_disabled)
2310 return -ENODEV;
2311
2312 filp->private_data = inode->i_private;
2313 return 0;
2314}
2315
2316static int tracing_release(struct inode *inode, struct file *file)
2317{
2318 struct seq_file *m = file->private_data;
2319 struct trace_iterator *iter;
2320 int cpu;
2321
2322 if (!(file->f_mode & FMODE_READ))
2323 return 0;
2324
2325 iter = m->private;
2326
2327 mutex_lock(&trace_types_lock);
2328 for_each_tracing_cpu(cpu) {
2329 if (iter->buffer_iter[cpu])
2330 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2331 }
2332
2333 if (iter->trace && iter->trace->close)
2334 iter->trace->close(iter);
2335
2336 /* reenable tracing if it was previously enabled */
2337 tracing_start();
2338 mutex_unlock(&trace_types_lock);
2339
2340 seq_release(inode, file);
2341 mutex_destroy(&iter->mutex);
2342 free_cpumask_var(iter->started);
2343 kfree(iter->trace);
2344 kfree(iter);
2345 return 0;
2346}
2347
2348static int tracing_open(struct inode *inode, struct file *file)
2349{
2350 struct trace_iterator *iter;
2351 int ret = 0;
2352
2353 /* If this file was open for write, then erase contents */
2354 if ((file->f_mode & FMODE_WRITE) &&
2355 (file->f_flags & O_TRUNC)) {
2356 long cpu = (long) inode->i_private;
2357
2358 if (cpu == TRACE_PIPE_ALL_CPU)
2359 tracing_reset_online_cpus(&global_trace);
2360 else
2361 tracing_reset(&global_trace, cpu);
2362 }
2363
2364 if (file->f_mode & FMODE_READ) {
2365 iter = __tracing_open(inode, file);
2366 if (IS_ERR(iter))
2367 ret = PTR_ERR(iter);
2368 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2369 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2370 }
2371 return ret;
2372}
2373
2374static void *
2375t_next(struct seq_file *m, void *v, loff_t *pos)
2376{
2377 struct tracer *t = v;
2378
2379 (*pos)++;
2380
2381 if (t)
2382 t = t->next;
2383
2384 return t;
2385}
2386
2387static void *t_start(struct seq_file *m, loff_t *pos)
2388{
2389 struct tracer *t;
2390 loff_t l = 0;
2391
2392 mutex_lock(&trace_types_lock);
2393 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2394 ;
2395
2396 return t;
2397}
2398
2399static void t_stop(struct seq_file *m, void *p)
2400{
2401 mutex_unlock(&trace_types_lock);
2402}
2403
2404static int t_show(struct seq_file *m, void *v)
2405{
2406 struct tracer *t = v;
2407
2408 if (!t)
2409 return 0;
2410
2411 seq_printf(m, "%s", t->name);
2412 if (t->next)
2413 seq_putc(m, ' ');
2414 else
2415 seq_putc(m, '\n');
2416
2417 return 0;
2418}
2419
2420static const struct seq_operations show_traces_seq_ops = {
2421 .start = t_start,
2422 .next = t_next,
2423 .stop = t_stop,
2424 .show = t_show,
2425};
2426
2427static int show_traces_open(struct inode *inode, struct file *file)
2428{
2429 if (tracing_disabled)
2430 return -ENODEV;
2431
2432 return seq_open(file, &show_traces_seq_ops);
2433}
2434
2435static ssize_t
2436tracing_write_stub(struct file *filp, const char __user *ubuf,
2437 size_t count, loff_t *ppos)
2438{
2439 return count;
2440}
2441
2442static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2443{
2444 if (file->f_mode & FMODE_READ)
2445 return seq_lseek(file, offset, origin);
2446 else
2447 return 0;
2448}
2449
2450static const struct file_operations tracing_fops = {
2451 .open = tracing_open,
2452 .read = seq_read,
2453 .write = tracing_write_stub,
2454 .llseek = tracing_seek,
2455 .release = tracing_release,
2456};
2457
2458static const struct file_operations show_traces_fops = {
2459 .open = show_traces_open,
2460 .read = seq_read,
2461 .release = seq_release,
2462 .llseek = seq_lseek,
2463};
2464
2465/*
2466 * Only trace on a CPU if the bitmask is set:
2467 */
2468static cpumask_var_t tracing_cpumask;
2469
2470/*
2471 * The tracer itself will not take this lock, but still we want
2472 * to provide a consistent cpumask to user-space:
2473 */
2474static DEFINE_MUTEX(tracing_cpumask_update_lock);
2475
2476/*
2477 * Temporary storage for the character representation of the
2478 * CPU bitmask (and one more byte for the newline):
2479 */
2480static char mask_str[NR_CPUS + 1];
2481
2482static ssize_t
2483tracing_cpumask_read(struct file *filp, char __user *ubuf,
2484 size_t count, loff_t *ppos)
2485{
2486 int len;
2487
2488 mutex_lock(&tracing_cpumask_update_lock);
2489
2490 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2491 if (count - len < 2) {
2492 count = -EINVAL;
2493 goto out_err;
2494 }
2495 len += sprintf(mask_str + len, "\n");
2496 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2497
2498out_err:
2499 mutex_unlock(&tracing_cpumask_update_lock);
2500
2501 return count;
2502}
2503
2504static ssize_t
2505tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2506 size_t count, loff_t *ppos)
2507{
2508 int err, cpu;
2509 cpumask_var_t tracing_cpumask_new;
2510
2511 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2512 return -ENOMEM;
2513
2514 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2515 if (err)
2516 goto err_unlock;
2517
2518 mutex_lock(&tracing_cpumask_update_lock);
2519
2520 local_irq_disable();
2521 arch_spin_lock(&ftrace_max_lock);
2522 for_each_tracing_cpu(cpu) {
2523 /*
2524 * Increase/decrease the disabled counter if we are
2525 * about to flip a bit in the cpumask:
2526 */
2527 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2528 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2529 atomic_inc(&global_trace.data[cpu]->disabled);
2530 }
2531 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2532 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2533 atomic_dec(&global_trace.data[cpu]->disabled);
2534 }
2535 }
2536 arch_spin_unlock(&ftrace_max_lock);
2537 local_irq_enable();
2538
2539 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2540
2541 mutex_unlock(&tracing_cpumask_update_lock);
2542 free_cpumask_var(tracing_cpumask_new);
2543
2544 return count;
2545
2546err_unlock:
2547 free_cpumask_var(tracing_cpumask_new);
2548
2549 return err;
2550}
2551
2552static const struct file_operations tracing_cpumask_fops = {
2553 .open = tracing_open_generic,
2554 .read = tracing_cpumask_read,
2555 .write = tracing_cpumask_write,
2556 .llseek = generic_file_llseek,
2557};
2558
2559static int tracing_trace_options_show(struct seq_file *m, void *v)
2560{
2561 struct tracer_opt *trace_opts;
2562 u32 tracer_flags;
2563 int i;
2564
2565 mutex_lock(&trace_types_lock);
2566 tracer_flags = current_trace->flags->val;
2567 trace_opts = current_trace->flags->opts;
2568
2569 for (i = 0; trace_options[i]; i++) {
2570 if (trace_flags & (1 << i))
2571 seq_printf(m, "%s\n", trace_options[i]);
2572 else
2573 seq_printf(m, "no%s\n", trace_options[i]);
2574 }
2575
2576 for (i = 0; trace_opts[i].name; i++) {
2577 if (tracer_flags & trace_opts[i].bit)
2578 seq_printf(m, "%s\n", trace_opts[i].name);
2579 else
2580 seq_printf(m, "no%s\n", trace_opts[i].name);
2581 }
2582 mutex_unlock(&trace_types_lock);
2583
2584 return 0;
2585}
2586
2587static int __set_tracer_option(struct tracer *trace,
2588 struct tracer_flags *tracer_flags,
2589 struct tracer_opt *opts, int neg)
2590{
2591 int ret;
2592
2593 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2594 if (ret)
2595 return ret;
2596
2597 if (neg)
2598 tracer_flags->val &= ~opts->bit;
2599 else
2600 tracer_flags->val |= opts->bit;
2601 return 0;
2602}
2603
2604/* Try to assign a tracer specific option */
2605static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2606{
2607 struct tracer_flags *tracer_flags = trace->flags;
2608 struct tracer_opt *opts = NULL;
2609 int i;
2610
2611 for (i = 0; tracer_flags->opts[i].name; i++) {
2612 opts = &tracer_flags->opts[i];
2613
2614 if (strcmp(cmp, opts->name) == 0)
2615 return __set_tracer_option(trace, trace->flags,
2616 opts, neg);
2617 }
2618
2619 return -EINVAL;
2620}
2621
2622static void set_tracer_flags(unsigned int mask, int enabled)
2623{
2624 /* do nothing if flag is already set */
2625 if (!!(trace_flags & mask) == !!enabled)
2626 return;
2627
2628 if (enabled)
2629 trace_flags |= mask;
2630 else
2631 trace_flags &= ~mask;
2632
2633 if (mask == TRACE_ITER_RECORD_CMD)
2634 trace_event_enable_cmd_record(enabled);
2635
2636 if (mask == TRACE_ITER_OVERWRITE)
2637 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2638}
2639
2640static ssize_t
2641tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2642 size_t cnt, loff_t *ppos)
2643{
2644 char buf[64];
2645 char *cmp;
2646 int neg = 0;
2647 int ret;
2648 int i;
2649
2650 if (cnt >= sizeof(buf))
2651 return -EINVAL;
2652
2653 if (copy_from_user(&buf, ubuf, cnt))
2654 return -EFAULT;
2655
2656 buf[cnt] = 0;
2657 cmp = strstrip(buf);
2658
2659 if (strncmp(cmp, "no", 2) == 0) {
2660 neg = 1;
2661 cmp += 2;
2662 }
2663
2664 for (i = 0; trace_options[i]; i++) {
2665 if (strcmp(cmp, trace_options[i]) == 0) {
2666 set_tracer_flags(1 << i, !neg);
2667 break;
2668 }
2669 }
2670
2671 /* If no option could be set, test the specific tracer options */
2672 if (!trace_options[i]) {
2673 mutex_lock(&trace_types_lock);
2674 ret = set_tracer_option(current_trace, cmp, neg);
2675 mutex_unlock(&trace_types_lock);
2676 if (ret)
2677 return ret;
2678 }
2679
2680 *ppos += cnt;
2681
2682 return cnt;
2683}
2684
2685static int tracing_trace_options_open(struct inode *inode, struct file *file)
2686{
2687 if (tracing_disabled)
2688 return -ENODEV;
2689 return single_open(file, tracing_trace_options_show, NULL);
2690}
2691
2692static const struct file_operations tracing_iter_fops = {
2693 .open = tracing_trace_options_open,
2694 .read = seq_read,
2695 .llseek = seq_lseek,
2696 .release = single_release,
2697 .write = tracing_trace_options_write,
2698};
2699
2700static const char readme_msg[] =
2701 "tracing mini-HOWTO:\n\n"
2702 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2703 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2704 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2705 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2706 "nop\n"
2707 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2708 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2709 "sched_switch\n"
2710 "# cat /sys/kernel/debug/tracing/trace_options\n"
2711 "noprint-parent nosym-offset nosym-addr noverbose\n"
2712 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2713 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2714 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2715 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2716;
2717
2718static ssize_t
2719tracing_readme_read(struct file *filp, char __user *ubuf,
2720 size_t cnt, loff_t *ppos)
2721{
2722 return simple_read_from_buffer(ubuf, cnt, ppos,
2723 readme_msg, strlen(readme_msg));
2724}
2725
2726static const struct file_operations tracing_readme_fops = {
2727 .open = tracing_open_generic,
2728 .read = tracing_readme_read,
2729 .llseek = generic_file_llseek,
2730};
2731
2732static ssize_t
2733tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2734 size_t cnt, loff_t *ppos)
2735{
2736 char *buf_comm;
2737 char *file_buf;
2738 char *buf;
2739 int len = 0;
2740 int pid;
2741 int i;
2742
2743 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2744 if (!file_buf)
2745 return -ENOMEM;
2746
2747 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2748 if (!buf_comm) {
2749 kfree(file_buf);
2750 return -ENOMEM;
2751 }
2752
2753 buf = file_buf;
2754
2755 for (i = 0; i < SAVED_CMDLINES; i++) {
2756 int r;
2757
2758 pid = map_cmdline_to_pid[i];
2759 if (pid == -1 || pid == NO_CMDLINE_MAP)
2760 continue;
2761
2762 trace_find_cmdline(pid, buf_comm);
2763 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2764 buf += r;
2765 len += r;
2766 }
2767
2768 len = simple_read_from_buffer(ubuf, cnt, ppos,
2769 file_buf, len);
2770
2771 kfree(file_buf);
2772 kfree(buf_comm);
2773
2774 return len;
2775}
2776
2777static const struct file_operations tracing_saved_cmdlines_fops = {
2778 .open = tracing_open_generic,
2779 .read = tracing_saved_cmdlines_read,
2780 .llseek = generic_file_llseek,
2781};
2782
2783static ssize_t
2784tracing_ctrl_read(struct file *filp, char __user *ubuf,
2785 size_t cnt, loff_t *ppos)
2786{
2787 char buf[64];
2788 int r;
2789
2790 r = sprintf(buf, "%u\n", tracer_enabled);
2791 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2792}
2793
2794static ssize_t
2795tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2796 size_t cnt, loff_t *ppos)
2797{
2798 struct trace_array *tr = filp->private_data;
2799 unsigned long val;
2800 int ret;
2801
2802 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
2803 if (ret)
2804 return ret;
2805
2806 val = !!val;
2807
2808 mutex_lock(&trace_types_lock);
2809 if (tracer_enabled ^ val) {
2810
2811 /* Only need to warn if this is used to change the state */
2812 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2813
2814 if (val) {
2815 tracer_enabled = 1;
2816 if (current_trace->start)
2817 current_trace->start(tr);
2818 tracing_start();
2819 } else {
2820 tracer_enabled = 0;
2821 tracing_stop();
2822 if (current_trace->stop)
2823 current_trace->stop(tr);
2824 }
2825 }
2826 mutex_unlock(&trace_types_lock);
2827
2828 *ppos += cnt;
2829
2830 return cnt;
2831}
2832
2833static ssize_t
2834tracing_set_trace_read(struct file *filp, char __user *ubuf,
2835 size_t cnt, loff_t *ppos)
2836{
2837 char buf[MAX_TRACER_SIZE+2];
2838 int r;
2839
2840 mutex_lock(&trace_types_lock);
2841 if (current_trace)
2842 r = sprintf(buf, "%s\n", current_trace->name);
2843 else
2844 r = sprintf(buf, "\n");
2845 mutex_unlock(&trace_types_lock);
2846
2847 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2848}
2849
2850int tracer_init(struct tracer *t, struct trace_array *tr)
2851{
2852 tracing_reset_online_cpus(tr);
2853 return t->init(tr);
2854}
2855
2856static int __tracing_resize_ring_buffer(unsigned long size)
2857{
2858 int ret;
2859
2860 /*
2861 * If kernel or user changes the size of the ring buffer
2862 * we use the size that was given, and we can forget about
2863 * expanding it later.
2864 */
2865 ring_buffer_expanded = 1;
2866
2867 ret = ring_buffer_resize(global_trace.buffer, size);
2868 if (ret < 0)
2869 return ret;
2870
2871 if (!current_trace->use_max_tr)
2872 goto out;
2873
2874 ret = ring_buffer_resize(max_tr.buffer, size);
2875 if (ret < 0) {
2876 int r;
2877
2878 r = ring_buffer_resize(global_trace.buffer,
2879 global_trace.entries);
2880 if (r < 0) {
2881 /*
2882 * AARGH! We are left with different
2883 * size max buffer!!!!
2884 * The max buffer is our "snapshot" buffer.
2885 * When a tracer needs a snapshot (one of the
2886 * latency tracers), it swaps the max buffer
2887 * with the saved snap shot. We succeeded to
2888 * update the size of the main buffer, but failed to
2889 * update the size of the max buffer. But when we tried
2890 * to reset the main buffer to the original size, we
2891 * failed there too. This is very unlikely to
2892 * happen, but if it does, warn and kill all
2893 * tracing.
2894 */
2895 WARN_ON(1);
2896 tracing_disabled = 1;
2897 }
2898 return ret;
2899 }
2900
2901 max_tr.entries = size;
2902 out:
2903 global_trace.entries = size;
2904
2905 return ret;
2906}
2907
2908static ssize_t tracing_resize_ring_buffer(unsigned long size)
2909{
2910 int cpu, ret = size;
2911
2912 mutex_lock(&trace_types_lock);
2913
2914 tracing_stop();
2915
2916 /* disable all cpu buffers */
2917 for_each_tracing_cpu(cpu) {
2918 if (global_trace.data[cpu])
2919 atomic_inc(&global_trace.data[cpu]->disabled);
2920 if (max_tr.data[cpu])
2921 atomic_inc(&max_tr.data[cpu]->disabled);
2922 }
2923
2924 if (size != global_trace.entries)
2925 ret = __tracing_resize_ring_buffer(size);
2926
2927 if (ret < 0)
2928 ret = -ENOMEM;
2929
2930 for_each_tracing_cpu(cpu) {
2931 if (global_trace.data[cpu])
2932 atomic_dec(&global_trace.data[cpu]->disabled);
2933 if (max_tr.data[cpu])
2934 atomic_dec(&max_tr.data[cpu]->disabled);
2935 }
2936
2937 tracing_start();
2938 mutex_unlock(&trace_types_lock);
2939
2940 return ret;
2941}
2942
2943
2944/**
2945 * tracing_update_buffers - used by tracing facility to expand ring buffers
2946 *
2947 * To save on memory when the tracing is never used on a system with it
2948 * configured in. The ring buffers are set to a minimum size. But once
2949 * a user starts to use the tracing facility, then they need to grow
2950 * to their default size.
2951 *
2952 * This function is to be called when a tracer is about to be used.
2953 */
2954int tracing_update_buffers(void)
2955{
2956 int ret = 0;
2957
2958 mutex_lock(&trace_types_lock);
2959 if (!ring_buffer_expanded)
2960 ret = __tracing_resize_ring_buffer(trace_buf_size);
2961 mutex_unlock(&trace_types_lock);
2962
2963 return ret;
2964}
2965
2966struct trace_option_dentry;
2967
2968static struct trace_option_dentry *
2969create_trace_option_files(struct tracer *tracer);
2970
2971static void
2972destroy_trace_option_files(struct trace_option_dentry *topts);
2973
2974static int tracing_set_tracer(const char *buf)
2975{
2976 static struct trace_option_dentry *topts;
2977 struct trace_array *tr = &global_trace;
2978 struct tracer *t;
2979 int ret = 0;
2980
2981 mutex_lock(&trace_types_lock);
2982
2983 if (!ring_buffer_expanded) {
2984 ret = __tracing_resize_ring_buffer(trace_buf_size);
2985 if (ret < 0)
2986 goto out;
2987 ret = 0;
2988 }
2989
2990 for (t = trace_types; t; t = t->next) {
2991 if (strcmp(t->name, buf) == 0)
2992 break;
2993 }
2994 if (!t) {
2995 ret = -EINVAL;
2996 goto out;
2997 }
2998 if (t == current_trace)
2999 goto out;
3000
3001 trace_branch_disable();
3002 if (current_trace && current_trace->reset)
3003 current_trace->reset(tr);
3004 if (current_trace && current_trace->use_max_tr) {
3005 /*
3006 * We don't free the ring buffer. instead, resize it because
3007 * The max_tr ring buffer has some state (e.g. ring->clock) and
3008 * we want preserve it.
3009 */
3010 ring_buffer_resize(max_tr.buffer, 1);
3011 max_tr.entries = 1;
3012 }
3013 destroy_trace_option_files(topts);
3014
3015 current_trace = t;
3016
3017 topts = create_trace_option_files(current_trace);
3018 if (current_trace->use_max_tr) {
3019 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
3020 if (ret < 0)
3021 goto out;
3022 max_tr.entries = global_trace.entries;
3023 }
3024
3025 if (t->init) {
3026 ret = tracer_init(t, tr);
3027 if (ret)
3028 goto out;
3029 }
3030
3031 trace_branch_enable(tr);
3032 out:
3033 mutex_unlock(&trace_types_lock);
3034
3035 return ret;
3036}
3037
3038static ssize_t
3039tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3040 size_t cnt, loff_t *ppos)
3041{
3042 char buf[MAX_TRACER_SIZE+1];
3043 int i;
3044 size_t ret;
3045 int err;
3046
3047 ret = cnt;
3048
3049 if (cnt > MAX_TRACER_SIZE)
3050 cnt = MAX_TRACER_SIZE;
3051
3052 if (copy_from_user(&buf, ubuf, cnt))
3053 return -EFAULT;
3054
3055 buf[cnt] = 0;
3056
3057 /* strip ending whitespace. */
3058 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3059 buf[i] = 0;
3060
3061 err = tracing_set_tracer(buf);
3062 if (err)
3063 return err;
3064
3065 *ppos += ret;
3066
3067 return ret;
3068}
3069
3070static ssize_t
3071tracing_max_lat_read(struct file *filp, char __user *ubuf,
3072 size_t cnt, loff_t *ppos)
3073{
3074 unsigned long *ptr = filp->private_data;
3075 char buf[64];
3076 int r;
3077
3078 r = snprintf(buf, sizeof(buf), "%ld\n",
3079 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3080 if (r > sizeof(buf))
3081 r = sizeof(buf);
3082 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3083}
3084
3085static ssize_t
3086tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3087 size_t cnt, loff_t *ppos)
3088{
3089 unsigned long *ptr = filp->private_data;
3090 unsigned long val;
3091 int ret;
3092
3093 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3094 if (ret)
3095 return ret;
3096
3097 *ptr = val * 1000;
3098
3099 return cnt;
3100}
3101
3102static int tracing_open_pipe(struct inode *inode, struct file *filp)
3103{
3104 long cpu_file = (long) inode->i_private;
3105 struct trace_iterator *iter;
3106 int ret = 0;
3107
3108 if (tracing_disabled)
3109 return -ENODEV;
3110
3111 mutex_lock(&trace_types_lock);
3112
3113 /* create a buffer to store the information to pass to userspace */
3114 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3115 if (!iter) {
3116 ret = -ENOMEM;
3117 goto out;
3118 }
3119
3120 /*
3121 * We make a copy of the current tracer to avoid concurrent
3122 * changes on it while we are reading.
3123 */
3124 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3125 if (!iter->trace) {
3126 ret = -ENOMEM;
3127 goto fail;
3128 }
3129 if (current_trace)
3130 *iter->trace = *current_trace;
3131
3132 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3133 ret = -ENOMEM;
3134 goto fail;
3135 }
3136
3137 /* trace pipe does not show start of buffer */
3138 cpumask_setall(iter->started);
3139
3140 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3141 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3142
3143 iter->cpu_file = cpu_file;
3144 iter->tr = &global_trace;
3145 mutex_init(&iter->mutex);
3146 filp->private_data = iter;
3147
3148 if (iter->trace->pipe_open)
3149 iter->trace->pipe_open(iter);
3150
3151 nonseekable_open(inode, filp);
3152out:
3153 mutex_unlock(&trace_types_lock);
3154 return ret;
3155
3156fail:
3157 kfree(iter->trace);
3158 kfree(iter);
3159 mutex_unlock(&trace_types_lock);
3160 return ret;
3161}
3162
3163static int tracing_release_pipe(struct inode *inode, struct file *file)
3164{
3165 struct trace_iterator *iter = file->private_data;
3166
3167 mutex_lock(&trace_types_lock);
3168
3169 if (iter->trace->pipe_close)
3170 iter->trace->pipe_close(iter);
3171
3172 mutex_unlock(&trace_types_lock);
3173
3174 free_cpumask_var(iter->started);
3175 mutex_destroy(&iter->mutex);
3176 kfree(iter->trace);
3177 kfree(iter);
3178
3179 return 0;
3180}
3181
3182static unsigned int
3183tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3184{
3185 struct trace_iterator *iter = filp->private_data;
3186
3187 if (trace_flags & TRACE_ITER_BLOCK) {
3188 /*
3189 * Always select as readable when in blocking mode
3190 */
3191 return POLLIN | POLLRDNORM;
3192 } else {
3193 if (!trace_empty(iter))
3194 return POLLIN | POLLRDNORM;
3195 poll_wait(filp, &trace_wait, poll_table);
3196 if (!trace_empty(iter))
3197 return POLLIN | POLLRDNORM;
3198
3199 return 0;
3200 }
3201}
3202
3203
3204void default_wait_pipe(struct trace_iterator *iter)
3205{
3206 DEFINE_WAIT(wait);
3207
3208 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3209
3210 if (trace_empty(iter))
3211 schedule();
3212
3213 finish_wait(&trace_wait, &wait);
3214}
3215
3216/*
3217 * This is a make-shift waitqueue.
3218 * A tracer might use this callback on some rare cases:
3219 *
3220 * 1) the current tracer might hold the runqueue lock when it wakes up
3221 * a reader, hence a deadlock (sched, function, and function graph tracers)
3222 * 2) the function tracers, trace all functions, we don't want
3223 * the overhead of calling wake_up and friends
3224 * (and tracing them too)
3225 *
3226 * Anyway, this is really very primitive wakeup.
3227 */
3228void poll_wait_pipe(struct trace_iterator *iter)
3229{
3230 set_current_state(TASK_INTERRUPTIBLE);
3231 /* sleep for 100 msecs, and try again. */
3232 schedule_timeout(HZ / 10);
3233}
3234
3235/* Must be called with trace_types_lock mutex held. */
3236static int tracing_wait_pipe(struct file *filp)
3237{
3238 struct trace_iterator *iter = filp->private_data;
3239
3240 while (trace_empty(iter)) {
3241
3242 if ((filp->f_flags & O_NONBLOCK)) {
3243 return -EAGAIN;
3244 }
3245
3246 mutex_unlock(&iter->mutex);
3247
3248 iter->trace->wait_pipe(iter);
3249
3250 mutex_lock(&iter->mutex);
3251
3252 if (signal_pending(current))
3253 return -EINTR;
3254
3255 /*
3256 * We block until we read something and tracing is disabled.
3257 * We still block if tracing is disabled, but we have never
3258 * read anything. This allows a user to cat this file, and
3259 * then enable tracing. But after we have read something,
3260 * we give an EOF when tracing is again disabled.
3261 *
3262 * iter->pos will be 0 if we haven't read anything.
3263 */
3264 if (!tracer_enabled && iter->pos)
3265 break;
3266 }
3267
3268 return 1;
3269}
3270
3271/*
3272 * Consumer reader.
3273 */
3274static ssize_t
3275tracing_read_pipe(struct file *filp, char __user *ubuf,
3276 size_t cnt, loff_t *ppos)
3277{
3278 struct trace_iterator *iter = filp->private_data;
3279 static struct tracer *old_tracer;
3280 ssize_t sret;
3281
3282 /* return any leftover data */
3283 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3284 if (sret != -EBUSY)
3285 return sret;
3286
3287 trace_seq_init(&iter->seq);
3288
3289 /* copy the tracer to avoid using a global lock all around */
3290 mutex_lock(&trace_types_lock);
3291 if (unlikely(old_tracer != current_trace && current_trace)) {
3292 old_tracer = current_trace;
3293 *iter->trace = *current_trace;
3294 }
3295 mutex_unlock(&trace_types_lock);
3296
3297 /*
3298 * Avoid more than one consumer on a single file descriptor
3299 * This is just a matter of traces coherency, the ring buffer itself
3300 * is protected.
3301 */
3302 mutex_lock(&iter->mutex);
3303 if (iter->trace->read) {
3304 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3305 if (sret)
3306 goto out;
3307 }
3308
3309waitagain:
3310 sret = tracing_wait_pipe(filp);
3311 if (sret <= 0)
3312 goto out;
3313
3314 /* stop when tracing is finished */
3315 if (trace_empty(iter)) {
3316 sret = 0;
3317 goto out;
3318 }
3319
3320 if (cnt >= PAGE_SIZE)
3321 cnt = PAGE_SIZE - 1;
3322
3323 /* reset all but tr, trace, and overruns */
3324 memset(&iter->seq, 0,
3325 sizeof(struct trace_iterator) -
3326 offsetof(struct trace_iterator, seq));
3327 iter->pos = -1;
3328
3329 trace_event_read_lock();
3330 trace_access_lock(iter->cpu_file);
3331 while (trace_find_next_entry_inc(iter) != NULL) {
3332 enum print_line_t ret;
3333 int len = iter->seq.len;
3334
3335 ret = print_trace_line(iter);
3336 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3337 /* don't print partial lines */
3338 iter->seq.len = len;
3339 break;
3340 }
3341 if (ret != TRACE_TYPE_NO_CONSUME)
3342 trace_consume(iter);
3343
3344 if (iter->seq.len >= cnt)
3345 break;
3346
3347 /*
3348 * Setting the full flag means we reached the trace_seq buffer
3349 * size and we should leave by partial output condition above.
3350 * One of the trace_seq_* functions is not used properly.
3351 */
3352 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3353 iter->ent->type);
3354 }
3355 trace_access_unlock(iter->cpu_file);
3356 trace_event_read_unlock();
3357
3358 /* Now copy what we have to the user */
3359 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3360 if (iter->seq.readpos >= iter->seq.len)
3361 trace_seq_init(&iter->seq);
3362
3363 /*
3364 * If there was nothing to send to user, in spite of consuming trace
3365 * entries, go back to wait for more entries.
3366 */
3367 if (sret == -EBUSY)
3368 goto waitagain;
3369
3370out:
3371 mutex_unlock(&iter->mutex);
3372
3373 return sret;
3374}
3375
3376static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3377 struct pipe_buffer *buf)
3378{
3379 __free_page(buf->page);
3380}
3381
3382static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3383 unsigned int idx)
3384{
3385 __free_page(spd->pages[idx]);
3386}
3387
3388static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3389 .can_merge = 0,
3390 .map = generic_pipe_buf_map,
3391 .unmap = generic_pipe_buf_unmap,
3392 .confirm = generic_pipe_buf_confirm,
3393 .release = tracing_pipe_buf_release,
3394 .steal = generic_pipe_buf_steal,
3395 .get = generic_pipe_buf_get,
3396};
3397
3398static size_t
3399tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3400{
3401 size_t count;
3402 int ret;
3403
3404 /* Seq buffer is page-sized, exactly what we need. */
3405 for (;;) {
3406 count = iter->seq.len;
3407 ret = print_trace_line(iter);
3408 count = iter->seq.len - count;
3409 if (rem < count) {
3410 rem = 0;
3411 iter->seq.len -= count;
3412 break;
3413 }
3414 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3415 iter->seq.len -= count;
3416 break;
3417 }
3418
3419 if (ret != TRACE_TYPE_NO_CONSUME)
3420 trace_consume(iter);
3421 rem -= count;
3422 if (!trace_find_next_entry_inc(iter)) {
3423 rem = 0;
3424 iter->ent = NULL;
3425 break;
3426 }
3427 }
3428
3429 return rem;
3430}
3431
3432static ssize_t tracing_splice_read_pipe(struct file *filp,
3433 loff_t *ppos,
3434 struct pipe_inode_info *pipe,
3435 size_t len,
3436 unsigned int flags)
3437{
3438 struct page *pages_def[PIPE_DEF_BUFFERS];
3439 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3440 struct trace_iterator *iter = filp->private_data;
3441 struct splice_pipe_desc spd = {
3442 .pages = pages_def,
3443 .partial = partial_def,
3444 .nr_pages = 0, /* This gets updated below. */
3445 .flags = flags,
3446 .ops = &tracing_pipe_buf_ops,
3447 .spd_release = tracing_spd_release_pipe,
3448 };
3449 static struct tracer *old_tracer;
3450 ssize_t ret;
3451 size_t rem;
3452 unsigned int i;
3453
3454 if (splice_grow_spd(pipe, &spd))
3455 return -ENOMEM;
3456
3457 /* copy the tracer to avoid using a global lock all around */
3458 mutex_lock(&trace_types_lock);
3459 if (unlikely(old_tracer != current_trace && current_trace)) {
3460 old_tracer = current_trace;
3461 *iter->trace = *current_trace;
3462 }
3463 mutex_unlock(&trace_types_lock);
3464
3465 mutex_lock(&iter->mutex);
3466
3467 if (iter->trace->splice_read) {
3468 ret = iter->trace->splice_read(iter, filp,
3469 ppos, pipe, len, flags);
3470 if (ret)
3471 goto out_err;
3472 }
3473
3474 ret = tracing_wait_pipe(filp);
3475 if (ret <= 0)
3476 goto out_err;
3477
3478 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3479 ret = -EFAULT;
3480 goto out_err;
3481 }
3482
3483 trace_event_read_lock();
3484 trace_access_lock(iter->cpu_file);
3485
3486 /* Fill as many pages as possible. */
3487 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3488 spd.pages[i] = alloc_page(GFP_KERNEL);
3489 if (!spd.pages[i])
3490 break;
3491
3492 rem = tracing_fill_pipe_page(rem, iter);
3493
3494 /* Copy the data into the page, so we can start over. */
3495 ret = trace_seq_to_buffer(&iter->seq,
3496 page_address(spd.pages[i]),
3497 iter->seq.len);
3498 if (ret < 0) {
3499 __free_page(spd.pages[i]);
3500 break;
3501 }
3502 spd.partial[i].offset = 0;
3503 spd.partial[i].len = iter->seq.len;
3504
3505 trace_seq_init(&iter->seq);
3506 }
3507
3508 trace_access_unlock(iter->cpu_file);
3509 trace_event_read_unlock();
3510 mutex_unlock(&iter->mutex);
3511
3512 spd.nr_pages = i;
3513
3514 ret = splice_to_pipe(pipe, &spd);
3515out:
3516 splice_shrink_spd(pipe, &spd);
3517 return ret;
3518
3519out_err:
3520 mutex_unlock(&iter->mutex);
3521 goto out;
3522}
3523
3524static ssize_t
3525tracing_entries_read(struct file *filp, char __user *ubuf,
3526 size_t cnt, loff_t *ppos)
3527{
3528 struct trace_array *tr = filp->private_data;
3529 char buf[96];
3530 int r;
3531
3532 mutex_lock(&trace_types_lock);
3533 if (!ring_buffer_expanded)
3534 r = sprintf(buf, "%lu (expanded: %lu)\n",
3535 tr->entries >> 10,
3536 trace_buf_size >> 10);
3537 else
3538 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3539 mutex_unlock(&trace_types_lock);
3540
3541 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3542}
3543
3544static ssize_t
3545tracing_entries_write(struct file *filp, const char __user *ubuf,
3546 size_t cnt, loff_t *ppos)
3547{
3548 unsigned long val;
3549 int ret;
3550
3551 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3552 if (ret)
3553 return ret;
3554
3555 /* must have at least 1 entry */
3556 if (!val)
3557 return -EINVAL;
3558
3559 /* value is in KB */
3560 val <<= 10;
3561
3562 ret = tracing_resize_ring_buffer(val);
3563 if (ret < 0)
3564 return ret;
3565
3566 *ppos += cnt;
3567
3568 return cnt;
3569}
3570
3571static ssize_t
3572tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3573 size_t cnt, loff_t *ppos)
3574{
3575 /*
3576 * There is no need to read what the user has written, this function
3577 * is just to make sure that there is no error when "echo" is used
3578 */
3579
3580 *ppos += cnt;
3581
3582 return cnt;
3583}
3584
3585static int
3586tracing_free_buffer_release(struct inode *inode, struct file *filp)
3587{
3588 /* disable tracing ? */
3589 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3590 tracing_off();
3591 /* resize the ring buffer to 0 */
3592 tracing_resize_ring_buffer(0);
3593
3594 return 0;
3595}
3596
3597static int mark_printk(const char *fmt, ...)
3598{
3599 int ret;
3600 va_list args;
3601 va_start(args, fmt);
3602 ret = trace_vprintk(0, fmt, args);
3603 va_end(args);
3604 return ret;
3605}
3606
3607static ssize_t
3608tracing_mark_write(struct file *filp, const char __user *ubuf,
3609 size_t cnt, loff_t *fpos)
3610{
3611 char *buf;
3612 size_t written;
3613
3614 if (tracing_disabled)
3615 return -EINVAL;
3616
3617 if (cnt > TRACE_BUF_SIZE)
3618 cnt = TRACE_BUF_SIZE;
3619
3620 buf = kmalloc(cnt + 2, GFP_KERNEL);
3621 if (buf == NULL)
3622 return -ENOMEM;
3623
3624 if (copy_from_user(buf, ubuf, cnt)) {
3625 kfree(buf);
3626 return -EFAULT;
3627 }
3628 if (buf[cnt-1] != '\n') {
3629 buf[cnt] = '\n';
3630 buf[cnt+1] = '\0';
3631 } else
3632 buf[cnt] = '\0';
3633
3634 written = mark_printk("%s", buf);
3635 kfree(buf);
3636 *fpos += written;
3637
3638 /* don't tell userspace we wrote more - it might confuse them */
3639 if (written > cnt)
3640 written = cnt;
3641
3642 return written;
3643}
3644
3645static int tracing_clock_show(struct seq_file *m, void *v)
3646{
3647 int i;
3648
3649 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3650 seq_printf(m,
3651 "%s%s%s%s", i ? " " : "",
3652 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3653 i == trace_clock_id ? "]" : "");
3654 seq_putc(m, '\n');
3655
3656 return 0;
3657}
3658
3659static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3660 size_t cnt, loff_t *fpos)
3661{
3662 char buf[64];
3663 const char *clockstr;
3664 int i;
3665
3666 if (cnt >= sizeof(buf))
3667 return -EINVAL;
3668
3669 if (copy_from_user(&buf, ubuf, cnt))
3670 return -EFAULT;
3671
3672 buf[cnt] = 0;
3673
3674 clockstr = strstrip(buf);
3675
3676 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3677 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3678 break;
3679 }
3680 if (i == ARRAY_SIZE(trace_clocks))
3681 return -EINVAL;
3682
3683 trace_clock_id = i;
3684
3685 mutex_lock(&trace_types_lock);
3686
3687 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3688 if (max_tr.buffer)
3689 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3690
3691 mutex_unlock(&trace_types_lock);
3692
3693 *fpos += cnt;
3694
3695 return cnt;
3696}
3697
3698static int tracing_clock_open(struct inode *inode, struct file *file)
3699{
3700 if (tracing_disabled)
3701 return -ENODEV;
3702 return single_open(file, tracing_clock_show, NULL);
3703}
3704
3705static const struct file_operations tracing_max_lat_fops = {
3706 .open = tracing_open_generic,
3707 .read = tracing_max_lat_read,
3708 .write = tracing_max_lat_write,
3709 .llseek = generic_file_llseek,
3710};
3711
3712static const struct file_operations tracing_ctrl_fops = {
3713 .open = tracing_open_generic,
3714 .read = tracing_ctrl_read,
3715 .write = tracing_ctrl_write,
3716 .llseek = generic_file_llseek,
3717};
3718
3719static const struct file_operations set_tracer_fops = {
3720 .open = tracing_open_generic,
3721 .read = tracing_set_trace_read,
3722 .write = tracing_set_trace_write,
3723 .llseek = generic_file_llseek,
3724};
3725
3726static const struct file_operations tracing_pipe_fops = {
3727 .open = tracing_open_pipe,
3728 .poll = tracing_poll_pipe,
3729 .read = tracing_read_pipe,
3730 .splice_read = tracing_splice_read_pipe,
3731 .release = tracing_release_pipe,
3732 .llseek = no_llseek,
3733};
3734
3735static const struct file_operations tracing_entries_fops = {
3736 .open = tracing_open_generic,
3737 .read = tracing_entries_read,
3738 .write = tracing_entries_write,
3739 .llseek = generic_file_llseek,
3740};
3741
3742static const struct file_operations tracing_free_buffer_fops = {
3743 .write = tracing_free_buffer_write,
3744 .release = tracing_free_buffer_release,
3745};
3746
3747static const struct file_operations tracing_mark_fops = {
3748 .open = tracing_open_generic,
3749 .write = tracing_mark_write,
3750 .llseek = generic_file_llseek,
3751};
3752
3753static const struct file_operations trace_clock_fops = {
3754 .open = tracing_clock_open,
3755 .read = seq_read,
3756 .llseek = seq_lseek,
3757 .release = single_release,
3758 .write = tracing_clock_write,
3759};
3760
3761struct ftrace_buffer_info {
3762 struct trace_array *tr;
3763 void *spare;
3764 int cpu;
3765 unsigned int read;
3766};
3767
3768static int tracing_buffers_open(struct inode *inode, struct file *filp)
3769{
3770 int cpu = (int)(long)inode->i_private;
3771 struct ftrace_buffer_info *info;
3772
3773 if (tracing_disabled)
3774 return -ENODEV;
3775
3776 info = kzalloc(sizeof(*info), GFP_KERNEL);
3777 if (!info)
3778 return -ENOMEM;
3779
3780 info->tr = &global_trace;
3781 info->cpu = cpu;
3782 info->spare = NULL;
3783 /* Force reading ring buffer for first read */
3784 info->read = (unsigned int)-1;
3785
3786 filp->private_data = info;
3787
3788 return nonseekable_open(inode, filp);
3789}
3790
3791static ssize_t
3792tracing_buffers_read(struct file *filp, char __user *ubuf,
3793 size_t count, loff_t *ppos)
3794{
3795 struct ftrace_buffer_info *info = filp->private_data;
3796 ssize_t ret;
3797 size_t size;
3798
3799 if (!count)
3800 return 0;
3801
3802 if (!info->spare)
3803 info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
3804 if (!info->spare)
3805 return -ENOMEM;
3806
3807 /* Do we have previous read data to read? */
3808 if (info->read < PAGE_SIZE)
3809 goto read;
3810
3811 info->read = 0;
3812
3813 trace_access_lock(info->cpu);
3814 ret = ring_buffer_read_page(info->tr->buffer,
3815 &info->spare,
3816 count,
3817 info->cpu, 0);
3818 trace_access_unlock(info->cpu);
3819 if (ret < 0)
3820 return 0;
3821
3822read:
3823 size = PAGE_SIZE - info->read;
3824 if (size > count)
3825 size = count;
3826
3827 ret = copy_to_user(ubuf, info->spare + info->read, size);
3828 if (ret == size)
3829 return -EFAULT;
3830 size -= ret;
3831
3832 *ppos += size;
3833 info->read += size;
3834
3835 return size;
3836}
3837
3838static int tracing_buffers_release(struct inode *inode, struct file *file)
3839{
3840 struct ftrace_buffer_info *info = file->private_data;
3841
3842 if (info->spare)
3843 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3844 kfree(info);
3845
3846 return 0;
3847}
3848
3849struct buffer_ref {
3850 struct ring_buffer *buffer;
3851 void *page;
3852 int ref;
3853};
3854
3855static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3856 struct pipe_buffer *buf)
3857{
3858 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3859
3860 if (--ref->ref)
3861 return;
3862
3863 ring_buffer_free_read_page(ref->buffer, ref->page);
3864 kfree(ref);
3865 buf->private = 0;
3866}
3867
3868static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3869 struct pipe_buffer *buf)
3870{
3871 return 1;
3872}
3873
3874static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3875 struct pipe_buffer *buf)
3876{
3877 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3878
3879 ref->ref++;
3880}
3881
3882/* Pipe buffer operations for a buffer. */
3883static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3884 .can_merge = 0,
3885 .map = generic_pipe_buf_map,
3886 .unmap = generic_pipe_buf_unmap,
3887 .confirm = generic_pipe_buf_confirm,
3888 .release = buffer_pipe_buf_release,
3889 .steal = buffer_pipe_buf_steal,
3890 .get = buffer_pipe_buf_get,
3891};
3892
3893/*
3894 * Callback from splice_to_pipe(), if we need to release some pages
3895 * at the end of the spd in case we error'ed out in filling the pipe.
3896 */
3897static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3898{
3899 struct buffer_ref *ref =
3900 (struct buffer_ref *)spd->partial[i].private;
3901
3902 if (--ref->ref)
3903 return;
3904
3905 ring_buffer_free_read_page(ref->buffer, ref->page);
3906 kfree(ref);
3907 spd->partial[i].private = 0;
3908}
3909
3910static ssize_t
3911tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3912 struct pipe_inode_info *pipe, size_t len,
3913 unsigned int flags)
3914{
3915 struct ftrace_buffer_info *info = file->private_data;
3916 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3917 struct page *pages_def[PIPE_DEF_BUFFERS];
3918 struct splice_pipe_desc spd = {
3919 .pages = pages_def,
3920 .partial = partial_def,
3921 .flags = flags,
3922 .ops = &buffer_pipe_buf_ops,
3923 .spd_release = buffer_spd_release,
3924 };
3925 struct buffer_ref *ref;
3926 int entries, size, i;
3927 size_t ret;
3928
3929 if (splice_grow_spd(pipe, &spd))
3930 return -ENOMEM;
3931
3932 if (*ppos & (PAGE_SIZE - 1)) {
3933 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3934 ret = -EINVAL;
3935 goto out;
3936 }
3937
3938 if (len & (PAGE_SIZE - 1)) {
3939 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3940 if (len < PAGE_SIZE) {
3941 ret = -EINVAL;
3942 goto out;
3943 }
3944 len &= PAGE_MASK;
3945 }
3946
3947 trace_access_lock(info->cpu);
3948 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3949
3950 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3951 struct page *page;
3952 int r;
3953
3954 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3955 if (!ref)
3956 break;
3957
3958 ref->ref = 1;
3959 ref->buffer = info->tr->buffer;
3960 ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
3961 if (!ref->page) {
3962 kfree(ref);
3963 break;
3964 }
3965
3966 r = ring_buffer_read_page(ref->buffer, &ref->page,
3967 len, info->cpu, 1);
3968 if (r < 0) {
3969 ring_buffer_free_read_page(ref->buffer, ref->page);
3970 kfree(ref);
3971 break;
3972 }
3973
3974 /*
3975 * zero out any left over data, this is going to
3976 * user land.
3977 */
3978 size = ring_buffer_page_len(ref->page);
3979 if (size < PAGE_SIZE)
3980 memset(ref->page + size, 0, PAGE_SIZE - size);
3981
3982 page = virt_to_page(ref->page);
3983
3984 spd.pages[i] = page;
3985 spd.partial[i].len = PAGE_SIZE;
3986 spd.partial[i].offset = 0;
3987 spd.partial[i].private = (unsigned long)ref;
3988 spd.nr_pages++;
3989 *ppos += PAGE_SIZE;
3990
3991 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3992 }
3993
3994 trace_access_unlock(info->cpu);
3995 spd.nr_pages = i;
3996
3997 /* did we read anything? */
3998 if (!spd.nr_pages) {
3999 if (flags & SPLICE_F_NONBLOCK)
4000 ret = -EAGAIN;
4001 else
4002 ret = 0;
4003 /* TODO: block */
4004 goto out;
4005 }
4006
4007 ret = splice_to_pipe(pipe, &spd);
4008 splice_shrink_spd(pipe, &spd);
4009out:
4010 return ret;
4011}
4012
4013static const struct file_operations tracing_buffers_fops = {
4014 .open = tracing_buffers_open,
4015 .read = tracing_buffers_read,
4016 .release = tracing_buffers_release,
4017 .splice_read = tracing_buffers_splice_read,
4018 .llseek = no_llseek,
4019};
4020
4021static ssize_t
4022tracing_stats_read(struct file *filp, char __user *ubuf,
4023 size_t count, loff_t *ppos)
4024{
4025 unsigned long cpu = (unsigned long)filp->private_data;
4026 struct trace_array *tr = &global_trace;
4027 struct trace_seq *s;
4028 unsigned long cnt;
4029
4030 s = kmalloc(sizeof(*s), GFP_KERNEL);
4031 if (!s)
4032 return -ENOMEM;
4033
4034 trace_seq_init(s);
4035
4036 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4037 trace_seq_printf(s, "entries: %ld\n", cnt);
4038
4039 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4040 trace_seq_printf(s, "overrun: %ld\n", cnt);
4041
4042 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4043 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4044
4045 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4046
4047 kfree(s);
4048
4049 return count;
4050}
4051
4052static const struct file_operations tracing_stats_fops = {
4053 .open = tracing_open_generic,
4054 .read = tracing_stats_read,
4055 .llseek = generic_file_llseek,
4056};
4057
4058#ifdef CONFIG_DYNAMIC_FTRACE
4059
4060int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4061{
4062 return 0;
4063}
4064
4065static ssize_t
4066tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4067 size_t cnt, loff_t *ppos)
4068{
4069 static char ftrace_dyn_info_buffer[1024];
4070 static DEFINE_MUTEX(dyn_info_mutex);
4071 unsigned long *p = filp->private_data;
4072 char *buf = ftrace_dyn_info_buffer;
4073 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4074 int r;
4075
4076 mutex_lock(&dyn_info_mutex);
4077 r = sprintf(buf, "%ld ", *p);
4078
4079 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4080 buf[r++] = '\n';
4081
4082 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4083
4084 mutex_unlock(&dyn_info_mutex);
4085
4086 return r;
4087}
4088
4089static const struct file_operations tracing_dyn_info_fops = {
4090 .open = tracing_open_generic,
4091 .read = tracing_read_dyn_info,
4092 .llseek = generic_file_llseek,
4093};
4094#endif
4095
4096static struct dentry *d_tracer;
4097
4098struct dentry *tracing_init_dentry(void)
4099{
4100 static int once;
4101
4102 if (d_tracer)
4103 return d_tracer;
4104
4105 if (!debugfs_initialized())
4106 return NULL;
4107
4108 d_tracer = debugfs_create_dir("tracing", NULL);
4109
4110 if (!d_tracer && !once) {
4111 once = 1;
4112 pr_warning("Could not create debugfs directory 'tracing'\n");
4113 return NULL;
4114 }
4115
4116 return d_tracer;
4117}
4118
4119static struct dentry *d_percpu;
4120
4121struct dentry *tracing_dentry_percpu(void)
4122{
4123 static int once;
4124 struct dentry *d_tracer;
4125
4126 if (d_percpu)
4127 return d_percpu;
4128
4129 d_tracer = tracing_init_dentry();
4130
4131 if (!d_tracer)
4132 return NULL;
4133
4134 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4135
4136 if (!d_percpu && !once) {
4137 once = 1;
4138 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4139 return NULL;
4140 }
4141
4142 return d_percpu;
4143}
4144
4145static void tracing_init_debugfs_percpu(long cpu)
4146{
4147 struct dentry *d_percpu = tracing_dentry_percpu();
4148 struct dentry *d_cpu;
4149 char cpu_dir[30]; /* 30 characters should be more than enough */
4150
4151 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4152 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4153 if (!d_cpu) {
4154 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4155 return;
4156 }
4157
4158 /* per cpu trace_pipe */
4159 trace_create_file("trace_pipe", 0444, d_cpu,
4160 (void *) cpu, &tracing_pipe_fops);
4161
4162 /* per cpu trace */
4163 trace_create_file("trace", 0644, d_cpu,
4164 (void *) cpu, &tracing_fops);
4165
4166 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4167 (void *) cpu, &tracing_buffers_fops);
4168
4169 trace_create_file("stats", 0444, d_cpu,
4170 (void *) cpu, &tracing_stats_fops);
4171}
4172
4173#ifdef CONFIG_FTRACE_SELFTEST
4174/* Let selftest have access to static functions in this file */
4175#include "trace_selftest.c"
4176#endif
4177
4178struct trace_option_dentry {
4179 struct tracer_opt *opt;
4180 struct tracer_flags *flags;
4181 struct dentry *entry;
4182};
4183
4184static ssize_t
4185trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4186 loff_t *ppos)
4187{
4188 struct trace_option_dentry *topt = filp->private_data;
4189 char *buf;
4190
4191 if (topt->flags->val & topt->opt->bit)
4192 buf = "1\n";
4193 else
4194 buf = "0\n";
4195
4196 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4197}
4198
4199static ssize_t
4200trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4201 loff_t *ppos)
4202{
4203 struct trace_option_dentry *topt = filp->private_data;
4204 unsigned long val;
4205 int ret;
4206
4207 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4208 if (ret)
4209 return ret;
4210
4211 if (val != 0 && val != 1)
4212 return -EINVAL;
4213
4214 if (!!(topt->flags->val & topt->opt->bit) != val) {
4215 mutex_lock(&trace_types_lock);
4216 ret = __set_tracer_option(current_trace, topt->flags,
4217 topt->opt, !val);
4218 mutex_unlock(&trace_types_lock);
4219 if (ret)
4220 return ret;
4221 }
4222
4223 *ppos += cnt;
4224
4225 return cnt;
4226}
4227
4228
4229static const struct file_operations trace_options_fops = {
4230 .open = tracing_open_generic,
4231 .read = trace_options_read,
4232 .write = trace_options_write,
4233 .llseek = generic_file_llseek,
4234};
4235
4236static ssize_t
4237trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4238 loff_t *ppos)
4239{
4240 long index = (long)filp->private_data;
4241 char *buf;
4242
4243 if (trace_flags & (1 << index))
4244 buf = "1\n";
4245 else
4246 buf = "0\n";
4247
4248 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4249}
4250
4251static ssize_t
4252trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4253 loff_t *ppos)
4254{
4255 long index = (long)filp->private_data;
4256 unsigned long val;
4257 int ret;
4258
4259 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4260 if (ret)
4261 return ret;
4262
4263 if (val != 0 && val != 1)
4264 return -EINVAL;
4265 set_tracer_flags(1 << index, val);
4266
4267 *ppos += cnt;
4268
4269 return cnt;
4270}
4271
4272static const struct file_operations trace_options_core_fops = {
4273 .open = tracing_open_generic,
4274 .read = trace_options_core_read,
4275 .write = trace_options_core_write,
4276 .llseek = generic_file_llseek,
4277};
4278
4279struct dentry *trace_create_file(const char *name,
4280 mode_t mode,
4281 struct dentry *parent,
4282 void *data,
4283 const struct file_operations *fops)
4284{
4285 struct dentry *ret;
4286
4287 ret = debugfs_create_file(name, mode, parent, data, fops);
4288 if (!ret)
4289 pr_warning("Could not create debugfs '%s' entry\n", name);
4290
4291 return ret;
4292}
4293
4294
4295static struct dentry *trace_options_init_dentry(void)
4296{
4297 struct dentry *d_tracer;
4298 static struct dentry *t_options;
4299
4300 if (t_options)
4301 return t_options;
4302
4303 d_tracer = tracing_init_dentry();
4304 if (!d_tracer)
4305 return NULL;
4306
4307 t_options = debugfs_create_dir("options", d_tracer);
4308 if (!t_options) {
4309 pr_warning("Could not create debugfs directory 'options'\n");
4310 return NULL;
4311 }
4312
4313 return t_options;
4314}
4315
4316static void
4317create_trace_option_file(struct trace_option_dentry *topt,
4318 struct tracer_flags *flags,
4319 struct tracer_opt *opt)
4320{
4321 struct dentry *t_options;
4322
4323 t_options = trace_options_init_dentry();
4324 if (!t_options)
4325 return;
4326
4327 topt->flags = flags;
4328 topt->opt = opt;
4329
4330 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4331 &trace_options_fops);
4332
4333}
4334
4335static struct trace_option_dentry *
4336create_trace_option_files(struct tracer *tracer)
4337{
4338 struct trace_option_dentry *topts;
4339 struct tracer_flags *flags;
4340 struct tracer_opt *opts;
4341 int cnt;
4342
4343 if (!tracer)
4344 return NULL;
4345
4346 flags = tracer->flags;
4347
4348 if (!flags || !flags->opts)
4349 return NULL;
4350
4351 opts = flags->opts;
4352
4353 for (cnt = 0; opts[cnt].name; cnt++)
4354 ;
4355
4356 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4357 if (!topts)
4358 return NULL;
4359
4360 for (cnt = 0; opts[cnt].name; cnt++)
4361 create_trace_option_file(&topts[cnt], flags,
4362 &opts[cnt]);
4363
4364 return topts;
4365}
4366
4367static void
4368destroy_trace_option_files(struct trace_option_dentry *topts)
4369{
4370 int cnt;
4371
4372 if (!topts)
4373 return;
4374
4375 for (cnt = 0; topts[cnt].opt; cnt++) {
4376 if (topts[cnt].entry)
4377 debugfs_remove(topts[cnt].entry);
4378 }
4379
4380 kfree(topts);
4381}
4382
4383static struct dentry *
4384create_trace_option_core_file(const char *option, long index)
4385{
4386 struct dentry *t_options;
4387
4388 t_options = trace_options_init_dentry();
4389 if (!t_options)
4390 return NULL;
4391
4392 return trace_create_file(option, 0644, t_options, (void *)index,
4393 &trace_options_core_fops);
4394}
4395
4396static __init void create_trace_options_dir(void)
4397{
4398 struct dentry *t_options;
4399 int i;
4400
4401 t_options = trace_options_init_dentry();
4402 if (!t_options)
4403 return;
4404
4405 for (i = 0; trace_options[i]; i++)
4406 create_trace_option_core_file(trace_options[i], i);
4407}
4408
4409static __init int tracer_init_debugfs(void)
4410{
4411 struct dentry *d_tracer;
4412 int cpu;
4413
4414 trace_access_lock_init();
4415
4416 d_tracer = tracing_init_dentry();
4417
4418 trace_create_file("tracing_enabled", 0644, d_tracer,
4419 &global_trace, &tracing_ctrl_fops);
4420
4421 trace_create_file("trace_options", 0644, d_tracer,
4422 NULL, &tracing_iter_fops);
4423
4424 trace_create_file("tracing_cpumask", 0644, d_tracer,
4425 NULL, &tracing_cpumask_fops);
4426
4427 trace_create_file("trace", 0644, d_tracer,
4428 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4429
4430 trace_create_file("available_tracers", 0444, d_tracer,
4431 &global_trace, &show_traces_fops);
4432
4433 trace_create_file("current_tracer", 0644, d_tracer,
4434 &global_trace, &set_tracer_fops);
4435
4436#ifdef CONFIG_TRACER_MAX_TRACE
4437 trace_create_file("tracing_max_latency", 0644, d_tracer,
4438 &tracing_max_latency, &tracing_max_lat_fops);
4439#endif
4440
4441 trace_create_file("tracing_thresh", 0644, d_tracer,
4442 &tracing_thresh, &tracing_max_lat_fops);
4443
4444 trace_create_file("README", 0444, d_tracer,
4445 NULL, &tracing_readme_fops);
4446
4447 trace_create_file("trace_pipe", 0444, d_tracer,
4448 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4449
4450 trace_create_file("buffer_size_kb", 0644, d_tracer,
4451 &global_trace, &tracing_entries_fops);
4452
4453 trace_create_file("free_buffer", 0644, d_tracer,
4454 &global_trace, &tracing_free_buffer_fops);
4455
4456 trace_create_file("trace_marker", 0220, d_tracer,
4457 NULL, &tracing_mark_fops);
4458
4459 trace_create_file("saved_cmdlines", 0444, d_tracer,
4460 NULL, &tracing_saved_cmdlines_fops);
4461
4462 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4463 &trace_clock_fops);
4464
4465#ifdef CONFIG_DYNAMIC_FTRACE
4466 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4467 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4468#endif
4469
4470 create_trace_options_dir();
4471
4472 for_each_tracing_cpu(cpu)
4473 tracing_init_debugfs_percpu(cpu);
4474
4475 return 0;
4476}
4477
4478static int trace_panic_handler(struct notifier_block *this,
4479 unsigned long event, void *unused)
4480{
4481 if (ftrace_dump_on_oops)
4482 ftrace_dump(ftrace_dump_on_oops);
4483 return NOTIFY_OK;
4484}
4485
4486static struct notifier_block trace_panic_notifier = {
4487 .notifier_call = trace_panic_handler,
4488 .next = NULL,
4489 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4490};
4491
4492static int trace_die_handler(struct notifier_block *self,
4493 unsigned long val,
4494 void *data)
4495{
4496 switch (val) {
4497 case DIE_OOPS:
4498 if (ftrace_dump_on_oops)
4499 ftrace_dump(ftrace_dump_on_oops);
4500 break;
4501 default:
4502 break;
4503 }
4504 return NOTIFY_OK;
4505}
4506
4507static struct notifier_block trace_die_notifier = {
4508 .notifier_call = trace_die_handler,
4509 .priority = 200
4510};
4511
4512/*
4513 * printk is set to max of 1024, we really don't need it that big.
4514 * Nothing should be printing 1000 characters anyway.
4515 */
4516#define TRACE_MAX_PRINT 1000
4517
4518/*
4519 * Define here KERN_TRACE so that we have one place to modify
4520 * it if we decide to change what log level the ftrace dump
4521 * should be at.
4522 */
4523#define KERN_TRACE KERN_EMERG
4524
4525void
4526trace_printk_seq(struct trace_seq *s)
4527{
4528 /* Probably should print a warning here. */
4529 if (s->len >= 1000)
4530 s->len = 1000;
4531
4532 /* should be zero ended, but we are paranoid. */
4533 s->buffer[s->len] = 0;
4534
4535 printk(KERN_TRACE "%s", s->buffer);
4536
4537 trace_seq_init(s);
4538}
4539
4540void trace_init_global_iter(struct trace_iterator *iter)
4541{
4542 iter->tr = &global_trace;
4543 iter->trace = current_trace;
4544 iter->cpu_file = TRACE_PIPE_ALL_CPU;
4545}
4546
4547static void
4548__ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4549{
4550 static arch_spinlock_t ftrace_dump_lock =
4551 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4552 /* use static because iter can be a bit big for the stack */
4553 static struct trace_iterator iter;
4554 unsigned int old_userobj;
4555 static int dump_ran;
4556 unsigned long flags;
4557 int cnt = 0, cpu;
4558
4559 /* only one dump */
4560 local_irq_save(flags);
4561 arch_spin_lock(&ftrace_dump_lock);
4562 if (dump_ran)
4563 goto out;
4564
4565 dump_ran = 1;
4566
4567 tracing_off();
4568
4569 if (disable_tracing)
4570 ftrace_kill();
4571
4572 trace_init_global_iter(&iter);
4573
4574 for_each_tracing_cpu(cpu) {
4575 atomic_inc(&iter.tr->data[cpu]->disabled);
4576 }
4577
4578 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4579
4580 /* don't look at user memory in panic mode */
4581 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4582
4583 /* Simulate the iterator */
4584 iter.tr = &global_trace;
4585 iter.trace = current_trace;
4586
4587 switch (oops_dump_mode) {
4588 case DUMP_ALL:
4589 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4590 break;
4591 case DUMP_ORIG:
4592 iter.cpu_file = raw_smp_processor_id();
4593 break;
4594 case DUMP_NONE:
4595 goto out_enable;
4596 default:
4597 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4598 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4599 }
4600
4601 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4602
4603 /*
4604 * We need to stop all tracing on all CPUS to read the
4605 * the next buffer. This is a bit expensive, but is
4606 * not done often. We fill all what we can read,
4607 * and then release the locks again.
4608 */
4609
4610 while (!trace_empty(&iter)) {
4611
4612 if (!cnt)
4613 printk(KERN_TRACE "---------------------------------\n");
4614
4615 cnt++;
4616
4617 /* reset all but tr, trace, and overruns */
4618 memset(&iter.seq, 0,
4619 sizeof(struct trace_iterator) -
4620 offsetof(struct trace_iterator, seq));
4621 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4622 iter.pos = -1;
4623
4624 if (trace_find_next_entry_inc(&iter) != NULL) {
4625 int ret;
4626
4627 ret = print_trace_line(&iter);
4628 if (ret != TRACE_TYPE_NO_CONSUME)
4629 trace_consume(&iter);
4630 }
4631
4632 trace_printk_seq(&iter.seq);
4633 }
4634
4635 if (!cnt)
4636 printk(KERN_TRACE " (ftrace buffer empty)\n");
4637 else
4638 printk(KERN_TRACE "---------------------------------\n");
4639
4640 out_enable:
4641 /* Re-enable tracing if requested */
4642 if (!disable_tracing) {
4643 trace_flags |= old_userobj;
4644
4645 for_each_tracing_cpu(cpu) {
4646 atomic_dec(&iter.tr->data[cpu]->disabled);
4647 }
4648 tracing_on();
4649 }
4650
4651 out:
4652 arch_spin_unlock(&ftrace_dump_lock);
4653 local_irq_restore(flags);
4654}
4655
4656/* By default: disable tracing after the dump */
4657void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4658{
4659 __ftrace_dump(true, oops_dump_mode);
4660}
4661
4662__init static int tracer_alloc_buffers(void)
4663{
4664 int ring_buf_size;
4665 enum ring_buffer_flags rb_flags;
4666 int i;
4667 int ret = -ENOMEM;
4668
4669
4670 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4671 goto out;
4672
4673 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4674 goto out_free_buffer_mask;
4675
4676 /* To save memory, keep the ring buffer size to its minimum */
4677 if (ring_buffer_expanded)
4678 ring_buf_size = trace_buf_size;
4679 else
4680 ring_buf_size = 1;
4681
4682 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
4683
4684 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4685 cpumask_copy(tracing_cpumask, cpu_all_mask);
4686
4687 /* TODO: make the number of buffers hot pluggable with CPUS */
4688 global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
4689 if (!global_trace.buffer) {
4690 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4691 WARN_ON(1);
4692 goto out_free_cpumask;
4693 }
4694 global_trace.entries = ring_buffer_size(global_trace.buffer);
4695
4696
4697#ifdef CONFIG_TRACER_MAX_TRACE
4698 max_tr.buffer = ring_buffer_alloc(1, rb_flags);
4699 if (!max_tr.buffer) {
4700 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4701 WARN_ON(1);
4702 ring_buffer_free(global_trace.buffer);
4703 goto out_free_cpumask;
4704 }
4705 max_tr.entries = 1;
4706#endif
4707
4708 /* Allocate the first page for all buffers */
4709 for_each_tracing_cpu(i) {
4710 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4711 max_tr.data[i] = &per_cpu(max_tr_data, i);
4712 }
4713
4714 trace_init_cmdlines();
4715
4716 register_tracer(&nop_trace);
4717 current_trace = &nop_trace;
4718 /* All seems OK, enable tracing */
4719 tracing_disabled = 0;
4720
4721 atomic_notifier_chain_register(&panic_notifier_list,
4722 &trace_panic_notifier);
4723
4724 register_die_notifier(&trace_die_notifier);
4725
4726 return 0;
4727
4728out_free_cpumask:
4729 free_cpumask_var(tracing_cpumask);
4730out_free_buffer_mask:
4731 free_cpumask_var(tracing_buffer_mask);
4732out:
4733 return ret;
4734}
4735
4736__init static int clear_boot_tracer(void)
4737{
4738 /*
4739 * The default tracer at boot buffer is an init section.
4740 * This function is called in lateinit. If we did not
4741 * find the boot tracer, then clear it out, to prevent
4742 * later registration from accessing the buffer that is
4743 * about to be freed.
4744 */
4745 if (!default_bootup_tracer)
4746 return 0;
4747
4748 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4749 default_bootup_tracer);
4750 default_bootup_tracer = NULL;
4751
4752 return 0;
4753}
4754
4755early_initcall(tracer_alloc_buffers);
4756fs_initcall(tracer_init_debugfs);
4757late_initcall(clear_boot_tracer);