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/notifier.h>
23#include <linux/irqflags.h>
24#include <linux/debugfs.h>
25#include <linux/tracefs.h>
26#include <linux/pagemap.h>
27#include <linux/hardirq.h>
28#include <linux/linkage.h>
29#include <linux/uaccess.h>
30#include <linux/vmalloc.h>
31#include <linux/ftrace.h>
32#include <linux/module.h>
33#include <linux/percpu.h>
34#include <linux/splice.h>
35#include <linux/kdebug.h>
36#include <linux/string.h>
37#include <linux/mount.h>
38#include <linux/rwsem.h>
39#include <linux/slab.h>
40#include <linux/ctype.h>
41#include <linux/init.h>
42#include <linux/panic_notifier.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 "trace.h"
54#include "trace_output.h"
55
56/*
57 * On boot up, the ring buffer is set to the minimum size, so that
58 * we do not waste memory on systems that are not using tracing.
59 */
60bool ring_buffer_expanded;
61
62/*
63 * We need to change this state when a selftest is running.
64 * A selftest will lurk into the ring-buffer to count the
65 * entries inserted during the selftest although some concurrent
66 * insertions into the ring-buffer such as trace_printk could occurred
67 * at the same time, giving false positive or negative results.
68 */
69static bool __read_mostly tracing_selftest_running;
70
71/*
72 * If boot-time tracing including tracers/events via kernel cmdline
73 * is running, we do not want to run SELFTEST.
74 */
75bool __read_mostly tracing_selftest_disabled;
76
77#ifdef CONFIG_FTRACE_STARTUP_TEST
78void __init disable_tracing_selftest(const char *reason)
79{
80 if (!tracing_selftest_disabled) {
81 tracing_selftest_disabled = true;
82 pr_info("Ftrace startup test is disabled due to %s\n", reason);
83 }
84}
85#endif
86
87/* Pipe tracepoints to printk */
88struct trace_iterator *tracepoint_print_iter;
89int tracepoint_printk;
90static bool tracepoint_printk_stop_on_boot __initdata;
91static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
92
93/* For tracers that don't implement custom flags */
94static struct tracer_opt dummy_tracer_opt[] = {
95 { }
96};
97
98static int
99dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
100{
101 return 0;
102}
103
104/*
105 * To prevent the comm cache from being overwritten when no
106 * tracing is active, only save the comm when a trace event
107 * occurred.
108 */
109static DEFINE_PER_CPU(bool, trace_taskinfo_save);
110
111/*
112 * Kill all tracing for good (never come back).
113 * It is initialized to 1 but will turn to zero if the initialization
114 * of the tracer is successful. But that is the only place that sets
115 * this back to zero.
116 */
117static int tracing_disabled = 1;
118
119cpumask_var_t __read_mostly tracing_buffer_mask;
120
121/*
122 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
123 *
124 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
125 * is set, then ftrace_dump is called. This will output the contents
126 * of the ftrace buffers to the console. This is very useful for
127 * capturing traces that lead to crashes and outputing it to a
128 * serial console.
129 *
130 * It is default off, but you can enable it with either specifying
131 * "ftrace_dump_on_oops" in the kernel command line, or setting
132 * /proc/sys/kernel/ftrace_dump_on_oops
133 * Set 1 if you want to dump buffers of all CPUs
134 * Set 2 if you want to dump the buffer of the CPU that triggered oops
135 */
136
137enum ftrace_dump_mode ftrace_dump_on_oops;
138
139/* When set, tracing will stop when a WARN*() is hit */
140int __disable_trace_on_warning;
141
142#ifdef CONFIG_TRACE_EVAL_MAP_FILE
143/* Map of enums to their values, for "eval_map" file */
144struct trace_eval_map_head {
145 struct module *mod;
146 unsigned long length;
147};
148
149union trace_eval_map_item;
150
151struct trace_eval_map_tail {
152 /*
153 * "end" is first and points to NULL as it must be different
154 * than "mod" or "eval_string"
155 */
156 union trace_eval_map_item *next;
157 const char *end; /* points to NULL */
158};
159
160static DEFINE_MUTEX(trace_eval_mutex);
161
162/*
163 * The trace_eval_maps are saved in an array with two extra elements,
164 * one at the beginning, and one at the end. The beginning item contains
165 * the count of the saved maps (head.length), and the module they
166 * belong to if not built in (head.mod). The ending item contains a
167 * pointer to the next array of saved eval_map items.
168 */
169union trace_eval_map_item {
170 struct trace_eval_map map;
171 struct trace_eval_map_head head;
172 struct trace_eval_map_tail tail;
173};
174
175static union trace_eval_map_item *trace_eval_maps;
176#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
177
178int tracing_set_tracer(struct trace_array *tr, const char *buf);
179static void ftrace_trace_userstack(struct trace_array *tr,
180 struct trace_buffer *buffer,
181 unsigned int trace_ctx);
182
183#define MAX_TRACER_SIZE 100
184static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
185static char *default_bootup_tracer;
186
187static bool allocate_snapshot;
188
189static int __init set_cmdline_ftrace(char *str)
190{
191 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
192 default_bootup_tracer = bootup_tracer_buf;
193 /* We are using ftrace early, expand it */
194 ring_buffer_expanded = true;
195 return 1;
196}
197__setup("ftrace=", set_cmdline_ftrace);
198
199static int __init set_ftrace_dump_on_oops(char *str)
200{
201 if (*str++ != '=' || !*str || !strcmp("1", str)) {
202 ftrace_dump_on_oops = DUMP_ALL;
203 return 1;
204 }
205
206 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
207 ftrace_dump_on_oops = DUMP_ORIG;
208 return 1;
209 }
210
211 return 0;
212}
213__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
214
215static int __init stop_trace_on_warning(char *str)
216{
217 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
218 __disable_trace_on_warning = 1;
219 return 1;
220}
221__setup("traceoff_on_warning", stop_trace_on_warning);
222
223static int __init boot_alloc_snapshot(char *str)
224{
225 allocate_snapshot = true;
226 /* We also need the main ring buffer expanded */
227 ring_buffer_expanded = true;
228 return 1;
229}
230__setup("alloc_snapshot", boot_alloc_snapshot);
231
232
233static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
234
235static int __init set_trace_boot_options(char *str)
236{
237 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
238 return 0;
239}
240__setup("trace_options=", set_trace_boot_options);
241
242static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
243static char *trace_boot_clock __initdata;
244
245static int __init set_trace_boot_clock(char *str)
246{
247 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
248 trace_boot_clock = trace_boot_clock_buf;
249 return 0;
250}
251__setup("trace_clock=", set_trace_boot_clock);
252
253static int __init set_tracepoint_printk(char *str)
254{
255 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
256 tracepoint_printk = 1;
257 return 1;
258}
259__setup("tp_printk", set_tracepoint_printk);
260
261static int __init set_tracepoint_printk_stop(char *str)
262{
263 tracepoint_printk_stop_on_boot = true;
264 return 1;
265}
266__setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
267
268unsigned long long ns2usecs(u64 nsec)
269{
270 nsec += 500;
271 do_div(nsec, 1000);
272 return nsec;
273}
274
275static void
276trace_process_export(struct trace_export *export,
277 struct ring_buffer_event *event, int flag)
278{
279 struct trace_entry *entry;
280 unsigned int size = 0;
281
282 if (export->flags & flag) {
283 entry = ring_buffer_event_data(event);
284 size = ring_buffer_event_length(event);
285 export->write(export, entry, size);
286 }
287}
288
289static DEFINE_MUTEX(ftrace_export_lock);
290
291static struct trace_export __rcu *ftrace_exports_list __read_mostly;
292
293static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
294static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
295static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
296
297static inline void ftrace_exports_enable(struct trace_export *export)
298{
299 if (export->flags & TRACE_EXPORT_FUNCTION)
300 static_branch_inc(&trace_function_exports_enabled);
301
302 if (export->flags & TRACE_EXPORT_EVENT)
303 static_branch_inc(&trace_event_exports_enabled);
304
305 if (export->flags & TRACE_EXPORT_MARKER)
306 static_branch_inc(&trace_marker_exports_enabled);
307}
308
309static inline void ftrace_exports_disable(struct trace_export *export)
310{
311 if (export->flags & TRACE_EXPORT_FUNCTION)
312 static_branch_dec(&trace_function_exports_enabled);
313
314 if (export->flags & TRACE_EXPORT_EVENT)
315 static_branch_dec(&trace_event_exports_enabled);
316
317 if (export->flags & TRACE_EXPORT_MARKER)
318 static_branch_dec(&trace_marker_exports_enabled);
319}
320
321static void ftrace_exports(struct ring_buffer_event *event, int flag)
322{
323 struct trace_export *export;
324
325 preempt_disable_notrace();
326
327 export = rcu_dereference_raw_check(ftrace_exports_list);
328 while (export) {
329 trace_process_export(export, event, flag);
330 export = rcu_dereference_raw_check(export->next);
331 }
332
333 preempt_enable_notrace();
334}
335
336static inline void
337add_trace_export(struct trace_export **list, struct trace_export *export)
338{
339 rcu_assign_pointer(export->next, *list);
340 /*
341 * We are entering export into the list but another
342 * CPU might be walking that list. We need to make sure
343 * the export->next pointer is valid before another CPU sees
344 * the export pointer included into the list.
345 */
346 rcu_assign_pointer(*list, export);
347}
348
349static inline int
350rm_trace_export(struct trace_export **list, struct trace_export *export)
351{
352 struct trace_export **p;
353
354 for (p = list; *p != NULL; p = &(*p)->next)
355 if (*p == export)
356 break;
357
358 if (*p != export)
359 return -1;
360
361 rcu_assign_pointer(*p, (*p)->next);
362
363 return 0;
364}
365
366static inline void
367add_ftrace_export(struct trace_export **list, struct trace_export *export)
368{
369 ftrace_exports_enable(export);
370
371 add_trace_export(list, export);
372}
373
374static inline int
375rm_ftrace_export(struct trace_export **list, struct trace_export *export)
376{
377 int ret;
378
379 ret = rm_trace_export(list, export);
380 ftrace_exports_disable(export);
381
382 return ret;
383}
384
385int register_ftrace_export(struct trace_export *export)
386{
387 if (WARN_ON_ONCE(!export->write))
388 return -1;
389
390 mutex_lock(&ftrace_export_lock);
391
392 add_ftrace_export(&ftrace_exports_list, export);
393
394 mutex_unlock(&ftrace_export_lock);
395
396 return 0;
397}
398EXPORT_SYMBOL_GPL(register_ftrace_export);
399
400int unregister_ftrace_export(struct trace_export *export)
401{
402 int ret;
403
404 mutex_lock(&ftrace_export_lock);
405
406 ret = rm_ftrace_export(&ftrace_exports_list, export);
407
408 mutex_unlock(&ftrace_export_lock);
409
410 return ret;
411}
412EXPORT_SYMBOL_GPL(unregister_ftrace_export);
413
414/* trace_flags holds trace_options default values */
415#define TRACE_DEFAULT_FLAGS \
416 (FUNCTION_DEFAULT_FLAGS | \
417 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
418 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
419 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
420 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
421 TRACE_ITER_HASH_PTR)
422
423/* trace_options that are only supported by global_trace */
424#define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
425 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
426
427/* trace_flags that are default zero for instances */
428#define ZEROED_TRACE_FLAGS \
429 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
430
431/*
432 * The global_trace is the descriptor that holds the top-level tracing
433 * buffers for the live tracing.
434 */
435static struct trace_array global_trace = {
436 .trace_flags = TRACE_DEFAULT_FLAGS,
437};
438
439LIST_HEAD(ftrace_trace_arrays);
440
441int trace_array_get(struct trace_array *this_tr)
442{
443 struct trace_array *tr;
444 int ret = -ENODEV;
445
446 mutex_lock(&trace_types_lock);
447 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
448 if (tr == this_tr) {
449 tr->ref++;
450 ret = 0;
451 break;
452 }
453 }
454 mutex_unlock(&trace_types_lock);
455
456 return ret;
457}
458
459static void __trace_array_put(struct trace_array *this_tr)
460{
461 WARN_ON(!this_tr->ref);
462 this_tr->ref--;
463}
464
465/**
466 * trace_array_put - Decrement the reference counter for this trace array.
467 * @this_tr : pointer to the trace array
468 *
469 * NOTE: Use this when we no longer need the trace array returned by
470 * trace_array_get_by_name(). This ensures the trace array can be later
471 * destroyed.
472 *
473 */
474void trace_array_put(struct trace_array *this_tr)
475{
476 if (!this_tr)
477 return;
478
479 mutex_lock(&trace_types_lock);
480 __trace_array_put(this_tr);
481 mutex_unlock(&trace_types_lock);
482}
483EXPORT_SYMBOL_GPL(trace_array_put);
484
485int tracing_check_open_get_tr(struct trace_array *tr)
486{
487 int ret;
488
489 ret = security_locked_down(LOCKDOWN_TRACEFS);
490 if (ret)
491 return ret;
492
493 if (tracing_disabled)
494 return -ENODEV;
495
496 if (tr && trace_array_get(tr) < 0)
497 return -ENODEV;
498
499 return 0;
500}
501
502int call_filter_check_discard(struct trace_event_call *call, void *rec,
503 struct trace_buffer *buffer,
504 struct ring_buffer_event *event)
505{
506 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
507 !filter_match_preds(call->filter, rec)) {
508 __trace_event_discard_commit(buffer, event);
509 return 1;
510 }
511
512 return 0;
513}
514
515void trace_free_pid_list(struct trace_pid_list *pid_list)
516{
517 vfree(pid_list->pids);
518 kfree(pid_list);
519}
520
521/**
522 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
523 * @filtered_pids: The list of pids to check
524 * @search_pid: The PID to find in @filtered_pids
525 *
526 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
527 */
528bool
529trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
530{
531 /*
532 * If pid_max changed after filtered_pids was created, we
533 * by default ignore all pids greater than the previous pid_max.
534 */
535 if (search_pid >= filtered_pids->pid_max)
536 return false;
537
538 return test_bit(search_pid, filtered_pids->pids);
539}
540
541/**
542 * trace_ignore_this_task - should a task be ignored for tracing
543 * @filtered_pids: The list of pids to check
544 * @filtered_no_pids: The list of pids not to be traced
545 * @task: The task that should be ignored if not filtered
546 *
547 * Checks if @task should be traced or not from @filtered_pids.
548 * Returns true if @task should *NOT* be traced.
549 * Returns false if @task should be traced.
550 */
551bool
552trace_ignore_this_task(struct trace_pid_list *filtered_pids,
553 struct trace_pid_list *filtered_no_pids,
554 struct task_struct *task)
555{
556 /*
557 * If filtered_no_pids is not empty, and the task's pid is listed
558 * in filtered_no_pids, then return true.
559 * Otherwise, if filtered_pids is empty, that means we can
560 * trace all tasks. If it has content, then only trace pids
561 * within filtered_pids.
562 */
563
564 return (filtered_pids &&
565 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
566 (filtered_no_pids &&
567 trace_find_filtered_pid(filtered_no_pids, task->pid));
568}
569
570/**
571 * trace_filter_add_remove_task - Add or remove a task from a pid_list
572 * @pid_list: The list to modify
573 * @self: The current task for fork or NULL for exit
574 * @task: The task to add or remove
575 *
576 * If adding a task, if @self is defined, the task is only added if @self
577 * is also included in @pid_list. This happens on fork and tasks should
578 * only be added when the parent is listed. If @self is NULL, then the
579 * @task pid will be removed from the list, which would happen on exit
580 * of a task.
581 */
582void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
583 struct task_struct *self,
584 struct task_struct *task)
585{
586 if (!pid_list)
587 return;
588
589 /* For forks, we only add if the forking task is listed */
590 if (self) {
591 if (!trace_find_filtered_pid(pid_list, self->pid))
592 return;
593 }
594
595 /* Sorry, but we don't support pid_max changing after setting */
596 if (task->pid >= pid_list->pid_max)
597 return;
598
599 /* "self" is set for forks, and NULL for exits */
600 if (self)
601 set_bit(task->pid, pid_list->pids);
602 else
603 clear_bit(task->pid, pid_list->pids);
604}
605
606/**
607 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
608 * @pid_list: The pid list to show
609 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
610 * @pos: The position of the file
611 *
612 * This is used by the seq_file "next" operation to iterate the pids
613 * listed in a trace_pid_list structure.
614 *
615 * Returns the pid+1 as we want to display pid of zero, but NULL would
616 * stop the iteration.
617 */
618void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
619{
620 unsigned long pid = (unsigned long)v;
621
622 (*pos)++;
623
624 /* pid already is +1 of the actual previous bit */
625 pid = find_next_bit(pid_list->pids, pid_list->pid_max, pid);
626
627 /* Return pid + 1 to allow zero to be represented */
628 if (pid < pid_list->pid_max)
629 return (void *)(pid + 1);
630
631 return NULL;
632}
633
634/**
635 * trace_pid_start - Used for seq_file to start reading pid lists
636 * @pid_list: The pid list to show
637 * @pos: The position of the file
638 *
639 * This is used by seq_file "start" operation to start the iteration
640 * of listing pids.
641 *
642 * Returns the pid+1 as we want to display pid of zero, but NULL would
643 * stop the iteration.
644 */
645void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
646{
647 unsigned long pid;
648 loff_t l = 0;
649
650 pid = find_first_bit(pid_list->pids, pid_list->pid_max);
651 if (pid >= pid_list->pid_max)
652 return NULL;
653
654 /* Return pid + 1 so that zero can be the exit value */
655 for (pid++; pid && l < *pos;
656 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
657 ;
658 return (void *)pid;
659}
660
661/**
662 * trace_pid_show - show the current pid in seq_file processing
663 * @m: The seq_file structure to write into
664 * @v: A void pointer of the pid (+1) value to display
665 *
666 * Can be directly used by seq_file operations to display the current
667 * pid value.
668 */
669int trace_pid_show(struct seq_file *m, void *v)
670{
671 unsigned long pid = (unsigned long)v - 1;
672
673 seq_printf(m, "%lu\n", pid);
674 return 0;
675}
676
677/* 128 should be much more than enough */
678#define PID_BUF_SIZE 127
679
680int trace_pid_write(struct trace_pid_list *filtered_pids,
681 struct trace_pid_list **new_pid_list,
682 const char __user *ubuf, size_t cnt)
683{
684 struct trace_pid_list *pid_list;
685 struct trace_parser parser;
686 unsigned long val;
687 int nr_pids = 0;
688 ssize_t read = 0;
689 ssize_t ret = 0;
690 loff_t pos;
691 pid_t pid;
692
693 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
694 return -ENOMEM;
695
696 /*
697 * Always recreate a new array. The write is an all or nothing
698 * operation. Always create a new array when adding new pids by
699 * the user. If the operation fails, then the current list is
700 * not modified.
701 */
702 pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
703 if (!pid_list) {
704 trace_parser_put(&parser);
705 return -ENOMEM;
706 }
707
708 pid_list->pid_max = READ_ONCE(pid_max);
709
710 /* Only truncating will shrink pid_max */
711 if (filtered_pids && filtered_pids->pid_max > pid_list->pid_max)
712 pid_list->pid_max = filtered_pids->pid_max;
713
714 pid_list->pids = vzalloc((pid_list->pid_max + 7) >> 3);
715 if (!pid_list->pids) {
716 trace_parser_put(&parser);
717 kfree(pid_list);
718 return -ENOMEM;
719 }
720
721 if (filtered_pids) {
722 /* copy the current bits to the new max */
723 for_each_set_bit(pid, filtered_pids->pids,
724 filtered_pids->pid_max) {
725 set_bit(pid, pid_list->pids);
726 nr_pids++;
727 }
728 }
729
730 while (cnt > 0) {
731
732 pos = 0;
733
734 ret = trace_get_user(&parser, ubuf, cnt, &pos);
735 if (ret < 0 || !trace_parser_loaded(&parser))
736 break;
737
738 read += ret;
739 ubuf += ret;
740 cnt -= ret;
741
742 ret = -EINVAL;
743 if (kstrtoul(parser.buffer, 0, &val))
744 break;
745 if (val >= pid_list->pid_max)
746 break;
747
748 pid = (pid_t)val;
749
750 set_bit(pid, pid_list->pids);
751 nr_pids++;
752
753 trace_parser_clear(&parser);
754 ret = 0;
755 }
756 trace_parser_put(&parser);
757
758 if (ret < 0) {
759 trace_free_pid_list(pid_list);
760 return ret;
761 }
762
763 if (!nr_pids) {
764 /* Cleared the list of pids */
765 trace_free_pid_list(pid_list);
766 read = ret;
767 pid_list = NULL;
768 }
769
770 *new_pid_list = pid_list;
771
772 return read;
773}
774
775static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
776{
777 u64 ts;
778
779 /* Early boot up does not have a buffer yet */
780 if (!buf->buffer)
781 return trace_clock_local();
782
783 ts = ring_buffer_time_stamp(buf->buffer);
784 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
785
786 return ts;
787}
788
789u64 ftrace_now(int cpu)
790{
791 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
792}
793
794/**
795 * tracing_is_enabled - Show if global_trace has been enabled
796 *
797 * Shows if the global trace has been enabled or not. It uses the
798 * mirror flag "buffer_disabled" to be used in fast paths such as for
799 * the irqsoff tracer. But it may be inaccurate due to races. If you
800 * need to know the accurate state, use tracing_is_on() which is a little
801 * slower, but accurate.
802 */
803int tracing_is_enabled(void)
804{
805 /*
806 * For quick access (irqsoff uses this in fast path), just
807 * return the mirror variable of the state of the ring buffer.
808 * It's a little racy, but we don't really care.
809 */
810 smp_rmb();
811 return !global_trace.buffer_disabled;
812}
813
814/*
815 * trace_buf_size is the size in bytes that is allocated
816 * for a buffer. Note, the number of bytes is always rounded
817 * to page size.
818 *
819 * This number is purposely set to a low number of 16384.
820 * If the dump on oops happens, it will be much appreciated
821 * to not have to wait for all that output. Anyway this can be
822 * boot time and run time configurable.
823 */
824#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
825
826static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
827
828/* trace_types holds a link list of available tracers. */
829static struct tracer *trace_types __read_mostly;
830
831/*
832 * trace_types_lock is used to protect the trace_types list.
833 */
834DEFINE_MUTEX(trace_types_lock);
835
836/*
837 * serialize the access of the ring buffer
838 *
839 * ring buffer serializes readers, but it is low level protection.
840 * The validity of the events (which returns by ring_buffer_peek() ..etc)
841 * are not protected by ring buffer.
842 *
843 * The content of events may become garbage if we allow other process consumes
844 * these events concurrently:
845 * A) the page of the consumed events may become a normal page
846 * (not reader page) in ring buffer, and this page will be rewritten
847 * by events producer.
848 * B) The page of the consumed events may become a page for splice_read,
849 * and this page will be returned to system.
850 *
851 * These primitives allow multi process access to different cpu ring buffer
852 * concurrently.
853 *
854 * These primitives don't distinguish read-only and read-consume access.
855 * Multi read-only access are also serialized.
856 */
857
858#ifdef CONFIG_SMP
859static DECLARE_RWSEM(all_cpu_access_lock);
860static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
861
862static inline void trace_access_lock(int cpu)
863{
864 if (cpu == RING_BUFFER_ALL_CPUS) {
865 /* gain it for accessing the whole ring buffer. */
866 down_write(&all_cpu_access_lock);
867 } else {
868 /* gain it for accessing a cpu ring buffer. */
869
870 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
871 down_read(&all_cpu_access_lock);
872
873 /* Secondly block other access to this @cpu ring buffer. */
874 mutex_lock(&per_cpu(cpu_access_lock, cpu));
875 }
876}
877
878static inline void trace_access_unlock(int cpu)
879{
880 if (cpu == RING_BUFFER_ALL_CPUS) {
881 up_write(&all_cpu_access_lock);
882 } else {
883 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
884 up_read(&all_cpu_access_lock);
885 }
886}
887
888static inline void trace_access_lock_init(void)
889{
890 int cpu;
891
892 for_each_possible_cpu(cpu)
893 mutex_init(&per_cpu(cpu_access_lock, cpu));
894}
895
896#else
897
898static DEFINE_MUTEX(access_lock);
899
900static inline void trace_access_lock(int cpu)
901{
902 (void)cpu;
903 mutex_lock(&access_lock);
904}
905
906static inline void trace_access_unlock(int cpu)
907{
908 (void)cpu;
909 mutex_unlock(&access_lock);
910}
911
912static inline void trace_access_lock_init(void)
913{
914}
915
916#endif
917
918#ifdef CONFIG_STACKTRACE
919static void __ftrace_trace_stack(struct trace_buffer *buffer,
920 unsigned int trace_ctx,
921 int skip, struct pt_regs *regs);
922static inline void ftrace_trace_stack(struct trace_array *tr,
923 struct trace_buffer *buffer,
924 unsigned int trace_ctx,
925 int skip, struct pt_regs *regs);
926
927#else
928static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
929 unsigned int trace_ctx,
930 int skip, struct pt_regs *regs)
931{
932}
933static inline void ftrace_trace_stack(struct trace_array *tr,
934 struct trace_buffer *buffer,
935 unsigned long trace_ctx,
936 int skip, struct pt_regs *regs)
937{
938}
939
940#endif
941
942static __always_inline void
943trace_event_setup(struct ring_buffer_event *event,
944 int type, unsigned int trace_ctx)
945{
946 struct trace_entry *ent = ring_buffer_event_data(event);
947
948 tracing_generic_entry_update(ent, type, trace_ctx);
949}
950
951static __always_inline struct ring_buffer_event *
952__trace_buffer_lock_reserve(struct trace_buffer *buffer,
953 int type,
954 unsigned long len,
955 unsigned int trace_ctx)
956{
957 struct ring_buffer_event *event;
958
959 event = ring_buffer_lock_reserve(buffer, len);
960 if (event != NULL)
961 trace_event_setup(event, type, trace_ctx);
962
963 return event;
964}
965
966void tracer_tracing_on(struct trace_array *tr)
967{
968 if (tr->array_buffer.buffer)
969 ring_buffer_record_on(tr->array_buffer.buffer);
970 /*
971 * This flag is looked at when buffers haven't been allocated
972 * yet, or by some tracers (like irqsoff), that just want to
973 * know if the ring buffer has been disabled, but it can handle
974 * races of where it gets disabled but we still do a record.
975 * As the check is in the fast path of the tracers, it is more
976 * important to be fast than accurate.
977 */
978 tr->buffer_disabled = 0;
979 /* Make the flag seen by readers */
980 smp_wmb();
981}
982
983/**
984 * tracing_on - enable tracing buffers
985 *
986 * This function enables tracing buffers that may have been
987 * disabled with tracing_off.
988 */
989void tracing_on(void)
990{
991 tracer_tracing_on(&global_trace);
992}
993EXPORT_SYMBOL_GPL(tracing_on);
994
995
996static __always_inline void
997__buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
998{
999 __this_cpu_write(trace_taskinfo_save, true);
1000
1001 /* If this is the temp buffer, we need to commit fully */
1002 if (this_cpu_read(trace_buffered_event) == event) {
1003 /* Length is in event->array[0] */
1004 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1005 /* Release the temp buffer */
1006 this_cpu_dec(trace_buffered_event_cnt);
1007 } else
1008 ring_buffer_unlock_commit(buffer, event);
1009}
1010
1011/**
1012 * __trace_puts - write a constant string into the trace buffer.
1013 * @ip: The address of the caller
1014 * @str: The constant string to write
1015 * @size: The size of the string.
1016 */
1017int __trace_puts(unsigned long ip, const char *str, int size)
1018{
1019 struct ring_buffer_event *event;
1020 struct trace_buffer *buffer;
1021 struct print_entry *entry;
1022 unsigned int trace_ctx;
1023 int alloc;
1024
1025 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1026 return 0;
1027
1028 if (unlikely(tracing_selftest_running || tracing_disabled))
1029 return 0;
1030
1031 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1032
1033 trace_ctx = tracing_gen_ctx();
1034 buffer = global_trace.array_buffer.buffer;
1035 ring_buffer_nest_start(buffer);
1036 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1037 trace_ctx);
1038 if (!event) {
1039 size = 0;
1040 goto out;
1041 }
1042
1043 entry = ring_buffer_event_data(event);
1044 entry->ip = ip;
1045
1046 memcpy(&entry->buf, str, size);
1047
1048 /* Add a newline if necessary */
1049 if (entry->buf[size - 1] != '\n') {
1050 entry->buf[size] = '\n';
1051 entry->buf[size + 1] = '\0';
1052 } else
1053 entry->buf[size] = '\0';
1054
1055 __buffer_unlock_commit(buffer, event);
1056 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1057 out:
1058 ring_buffer_nest_end(buffer);
1059 return size;
1060}
1061EXPORT_SYMBOL_GPL(__trace_puts);
1062
1063/**
1064 * __trace_bputs - write the pointer to a constant string into trace buffer
1065 * @ip: The address of the caller
1066 * @str: The constant string to write to the buffer to
1067 */
1068int __trace_bputs(unsigned long ip, const char *str)
1069{
1070 struct ring_buffer_event *event;
1071 struct trace_buffer *buffer;
1072 struct bputs_entry *entry;
1073 unsigned int trace_ctx;
1074 int size = sizeof(struct bputs_entry);
1075 int ret = 0;
1076
1077 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1078 return 0;
1079
1080 if (unlikely(tracing_selftest_running || tracing_disabled))
1081 return 0;
1082
1083 trace_ctx = tracing_gen_ctx();
1084 buffer = global_trace.array_buffer.buffer;
1085
1086 ring_buffer_nest_start(buffer);
1087 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1088 trace_ctx);
1089 if (!event)
1090 goto out;
1091
1092 entry = ring_buffer_event_data(event);
1093 entry->ip = ip;
1094 entry->str = str;
1095
1096 __buffer_unlock_commit(buffer, event);
1097 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1098
1099 ret = 1;
1100 out:
1101 ring_buffer_nest_end(buffer);
1102 return ret;
1103}
1104EXPORT_SYMBOL_GPL(__trace_bputs);
1105
1106#ifdef CONFIG_TRACER_SNAPSHOT
1107static void tracing_snapshot_instance_cond(struct trace_array *tr,
1108 void *cond_data)
1109{
1110 struct tracer *tracer = tr->current_trace;
1111 unsigned long flags;
1112
1113 if (in_nmi()) {
1114 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1115 internal_trace_puts("*** snapshot is being ignored ***\n");
1116 return;
1117 }
1118
1119 if (!tr->allocated_snapshot) {
1120 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
1121 internal_trace_puts("*** stopping trace here! ***\n");
1122 tracing_off();
1123 return;
1124 }
1125
1126 /* Note, snapshot can not be used when the tracer uses it */
1127 if (tracer->use_max_tr) {
1128 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
1129 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
1130 return;
1131 }
1132
1133 local_irq_save(flags);
1134 update_max_tr(tr, current, smp_processor_id(), cond_data);
1135 local_irq_restore(flags);
1136}
1137
1138void tracing_snapshot_instance(struct trace_array *tr)
1139{
1140 tracing_snapshot_instance_cond(tr, NULL);
1141}
1142
1143/**
1144 * tracing_snapshot - take a snapshot of the current buffer.
1145 *
1146 * This causes a swap between the snapshot buffer and the current live
1147 * tracing buffer. You can use this to take snapshots of the live
1148 * trace when some condition is triggered, but continue to trace.
1149 *
1150 * Note, make sure to allocate the snapshot with either
1151 * a tracing_snapshot_alloc(), or by doing it manually
1152 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
1153 *
1154 * If the snapshot buffer is not allocated, it will stop tracing.
1155 * Basically making a permanent snapshot.
1156 */
1157void tracing_snapshot(void)
1158{
1159 struct trace_array *tr = &global_trace;
1160
1161 tracing_snapshot_instance(tr);
1162}
1163EXPORT_SYMBOL_GPL(tracing_snapshot);
1164
1165/**
1166 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1167 * @tr: The tracing instance to snapshot
1168 * @cond_data: The data to be tested conditionally, and possibly saved
1169 *
1170 * This is the same as tracing_snapshot() except that the snapshot is
1171 * conditional - the snapshot will only happen if the
1172 * cond_snapshot.update() implementation receiving the cond_data
1173 * returns true, which means that the trace array's cond_snapshot
1174 * update() operation used the cond_data to determine whether the
1175 * snapshot should be taken, and if it was, presumably saved it along
1176 * with the snapshot.
1177 */
1178void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1179{
1180 tracing_snapshot_instance_cond(tr, cond_data);
1181}
1182EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1183
1184/**
1185 * tracing_snapshot_cond_data - get the user data associated with a snapshot
1186 * @tr: The tracing instance
1187 *
1188 * When the user enables a conditional snapshot using
1189 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1190 * with the snapshot. This accessor is used to retrieve it.
1191 *
1192 * Should not be called from cond_snapshot.update(), since it takes
1193 * the tr->max_lock lock, which the code calling
1194 * cond_snapshot.update() has already done.
1195 *
1196 * Returns the cond_data associated with the trace array's snapshot.
1197 */
1198void *tracing_cond_snapshot_data(struct trace_array *tr)
1199{
1200 void *cond_data = NULL;
1201
1202 arch_spin_lock(&tr->max_lock);
1203
1204 if (tr->cond_snapshot)
1205 cond_data = tr->cond_snapshot->cond_data;
1206
1207 arch_spin_unlock(&tr->max_lock);
1208
1209 return cond_data;
1210}
1211EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1212
1213static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1214 struct array_buffer *size_buf, int cpu_id);
1215static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1216
1217int tracing_alloc_snapshot_instance(struct trace_array *tr)
1218{
1219 int ret;
1220
1221 if (!tr->allocated_snapshot) {
1222
1223 /* allocate spare buffer */
1224 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1225 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1226 if (ret < 0)
1227 return ret;
1228
1229 tr->allocated_snapshot = true;
1230 }
1231
1232 return 0;
1233}
1234
1235static void free_snapshot(struct trace_array *tr)
1236{
1237 /*
1238 * We don't free the ring buffer. instead, resize it because
1239 * The max_tr ring buffer has some state (e.g. ring->clock) and
1240 * we want preserve it.
1241 */
1242 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1243 set_buffer_entries(&tr->max_buffer, 1);
1244 tracing_reset_online_cpus(&tr->max_buffer);
1245 tr->allocated_snapshot = false;
1246}
1247
1248/**
1249 * tracing_alloc_snapshot - allocate snapshot buffer.
1250 *
1251 * This only allocates the snapshot buffer if it isn't already
1252 * allocated - it doesn't also take a snapshot.
1253 *
1254 * This is meant to be used in cases where the snapshot buffer needs
1255 * to be set up for events that can't sleep but need to be able to
1256 * trigger a snapshot.
1257 */
1258int tracing_alloc_snapshot(void)
1259{
1260 struct trace_array *tr = &global_trace;
1261 int ret;
1262
1263 ret = tracing_alloc_snapshot_instance(tr);
1264 WARN_ON(ret < 0);
1265
1266 return ret;
1267}
1268EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1269
1270/**
1271 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1272 *
1273 * This is similar to tracing_snapshot(), but it will allocate the
1274 * snapshot buffer if it isn't already allocated. Use this only
1275 * where it is safe to sleep, as the allocation may sleep.
1276 *
1277 * This causes a swap between the snapshot buffer and the current live
1278 * tracing buffer. You can use this to take snapshots of the live
1279 * trace when some condition is triggered, but continue to trace.
1280 */
1281void tracing_snapshot_alloc(void)
1282{
1283 int ret;
1284
1285 ret = tracing_alloc_snapshot();
1286 if (ret < 0)
1287 return;
1288
1289 tracing_snapshot();
1290}
1291EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1292
1293/**
1294 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1295 * @tr: The tracing instance
1296 * @cond_data: User data to associate with the snapshot
1297 * @update: Implementation of the cond_snapshot update function
1298 *
1299 * Check whether the conditional snapshot for the given instance has
1300 * already been enabled, or if the current tracer is already using a
1301 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1302 * save the cond_data and update function inside.
1303 *
1304 * Returns 0 if successful, error otherwise.
1305 */
1306int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1307 cond_update_fn_t update)
1308{
1309 struct cond_snapshot *cond_snapshot;
1310 int ret = 0;
1311
1312 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1313 if (!cond_snapshot)
1314 return -ENOMEM;
1315
1316 cond_snapshot->cond_data = cond_data;
1317 cond_snapshot->update = update;
1318
1319 mutex_lock(&trace_types_lock);
1320
1321 ret = tracing_alloc_snapshot_instance(tr);
1322 if (ret)
1323 goto fail_unlock;
1324
1325 if (tr->current_trace->use_max_tr) {
1326 ret = -EBUSY;
1327 goto fail_unlock;
1328 }
1329
1330 /*
1331 * The cond_snapshot can only change to NULL without the
1332 * trace_types_lock. We don't care if we race with it going
1333 * to NULL, but we want to make sure that it's not set to
1334 * something other than NULL when we get here, which we can
1335 * do safely with only holding the trace_types_lock and not
1336 * having to take the max_lock.
1337 */
1338 if (tr->cond_snapshot) {
1339 ret = -EBUSY;
1340 goto fail_unlock;
1341 }
1342
1343 arch_spin_lock(&tr->max_lock);
1344 tr->cond_snapshot = cond_snapshot;
1345 arch_spin_unlock(&tr->max_lock);
1346
1347 mutex_unlock(&trace_types_lock);
1348
1349 return ret;
1350
1351 fail_unlock:
1352 mutex_unlock(&trace_types_lock);
1353 kfree(cond_snapshot);
1354 return ret;
1355}
1356EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1357
1358/**
1359 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1360 * @tr: The tracing instance
1361 *
1362 * Check whether the conditional snapshot for the given instance is
1363 * enabled; if so, free the cond_snapshot associated with it,
1364 * otherwise return -EINVAL.
1365 *
1366 * Returns 0 if successful, error otherwise.
1367 */
1368int tracing_snapshot_cond_disable(struct trace_array *tr)
1369{
1370 int ret = 0;
1371
1372 arch_spin_lock(&tr->max_lock);
1373
1374 if (!tr->cond_snapshot)
1375 ret = -EINVAL;
1376 else {
1377 kfree(tr->cond_snapshot);
1378 tr->cond_snapshot = NULL;
1379 }
1380
1381 arch_spin_unlock(&tr->max_lock);
1382
1383 return ret;
1384}
1385EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1386#else
1387void tracing_snapshot(void)
1388{
1389 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1390}
1391EXPORT_SYMBOL_GPL(tracing_snapshot);
1392void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1393{
1394 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1395}
1396EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1397int tracing_alloc_snapshot(void)
1398{
1399 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1400 return -ENODEV;
1401}
1402EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1403void tracing_snapshot_alloc(void)
1404{
1405 /* Give warning */
1406 tracing_snapshot();
1407}
1408EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1409void *tracing_cond_snapshot_data(struct trace_array *tr)
1410{
1411 return NULL;
1412}
1413EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1414int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1415{
1416 return -ENODEV;
1417}
1418EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1419int tracing_snapshot_cond_disable(struct trace_array *tr)
1420{
1421 return false;
1422}
1423EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1424#endif /* CONFIG_TRACER_SNAPSHOT */
1425
1426void tracer_tracing_off(struct trace_array *tr)
1427{
1428 if (tr->array_buffer.buffer)
1429 ring_buffer_record_off(tr->array_buffer.buffer);
1430 /*
1431 * This flag is looked at when buffers haven't been allocated
1432 * yet, or by some tracers (like irqsoff), that just want to
1433 * know if the ring buffer has been disabled, but it can handle
1434 * races of where it gets disabled but we still do a record.
1435 * As the check is in the fast path of the tracers, it is more
1436 * important to be fast than accurate.
1437 */
1438 tr->buffer_disabled = 1;
1439 /* Make the flag seen by readers */
1440 smp_wmb();
1441}
1442
1443/**
1444 * tracing_off - turn off tracing buffers
1445 *
1446 * This function stops the tracing buffers from recording data.
1447 * It does not disable any overhead the tracers themselves may
1448 * be causing. This function simply causes all recording to
1449 * the ring buffers to fail.
1450 */
1451void tracing_off(void)
1452{
1453 tracer_tracing_off(&global_trace);
1454}
1455EXPORT_SYMBOL_GPL(tracing_off);
1456
1457void disable_trace_on_warning(void)
1458{
1459 if (__disable_trace_on_warning) {
1460 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1461 "Disabling tracing due to warning\n");
1462 tracing_off();
1463 }
1464}
1465
1466/**
1467 * tracer_tracing_is_on - show real state of ring buffer enabled
1468 * @tr : the trace array to know if ring buffer is enabled
1469 *
1470 * Shows real state of the ring buffer if it is enabled or not.
1471 */
1472bool tracer_tracing_is_on(struct trace_array *tr)
1473{
1474 if (tr->array_buffer.buffer)
1475 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1476 return !tr->buffer_disabled;
1477}
1478
1479/**
1480 * tracing_is_on - show state of ring buffers enabled
1481 */
1482int tracing_is_on(void)
1483{
1484 return tracer_tracing_is_on(&global_trace);
1485}
1486EXPORT_SYMBOL_GPL(tracing_is_on);
1487
1488static int __init set_buf_size(char *str)
1489{
1490 unsigned long buf_size;
1491
1492 if (!str)
1493 return 0;
1494 buf_size = memparse(str, &str);
1495 /* nr_entries can not be zero */
1496 if (buf_size == 0)
1497 return 0;
1498 trace_buf_size = buf_size;
1499 return 1;
1500}
1501__setup("trace_buf_size=", set_buf_size);
1502
1503static int __init set_tracing_thresh(char *str)
1504{
1505 unsigned long threshold;
1506 int ret;
1507
1508 if (!str)
1509 return 0;
1510 ret = kstrtoul(str, 0, &threshold);
1511 if (ret < 0)
1512 return 0;
1513 tracing_thresh = threshold * 1000;
1514 return 1;
1515}
1516__setup("tracing_thresh=", set_tracing_thresh);
1517
1518unsigned long nsecs_to_usecs(unsigned long nsecs)
1519{
1520 return nsecs / 1000;
1521}
1522
1523/*
1524 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1525 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1526 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1527 * of strings in the order that the evals (enum) were defined.
1528 */
1529#undef C
1530#define C(a, b) b
1531
1532/* These must match the bit positions in trace_iterator_flags */
1533static const char *trace_options[] = {
1534 TRACE_FLAGS
1535 NULL
1536};
1537
1538static struct {
1539 u64 (*func)(void);
1540 const char *name;
1541 int in_ns; /* is this clock in nanoseconds? */
1542} trace_clocks[] = {
1543 { trace_clock_local, "local", 1 },
1544 { trace_clock_global, "global", 1 },
1545 { trace_clock_counter, "counter", 0 },
1546 { trace_clock_jiffies, "uptime", 0 },
1547 { trace_clock, "perf", 1 },
1548 { ktime_get_mono_fast_ns, "mono", 1 },
1549 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1550 { ktime_get_boot_fast_ns, "boot", 1 },
1551 ARCH_TRACE_CLOCKS
1552};
1553
1554bool trace_clock_in_ns(struct trace_array *tr)
1555{
1556 if (trace_clocks[tr->clock_id].in_ns)
1557 return true;
1558
1559 return false;
1560}
1561
1562/*
1563 * trace_parser_get_init - gets the buffer for trace parser
1564 */
1565int trace_parser_get_init(struct trace_parser *parser, int size)
1566{
1567 memset(parser, 0, sizeof(*parser));
1568
1569 parser->buffer = kmalloc(size, GFP_KERNEL);
1570 if (!parser->buffer)
1571 return 1;
1572
1573 parser->size = size;
1574 return 0;
1575}
1576
1577/*
1578 * trace_parser_put - frees the buffer for trace parser
1579 */
1580void trace_parser_put(struct trace_parser *parser)
1581{
1582 kfree(parser->buffer);
1583 parser->buffer = NULL;
1584}
1585
1586/*
1587 * trace_get_user - reads the user input string separated by space
1588 * (matched by isspace(ch))
1589 *
1590 * For each string found the 'struct trace_parser' is updated,
1591 * and the function returns.
1592 *
1593 * Returns number of bytes read.
1594 *
1595 * See kernel/trace/trace.h for 'struct trace_parser' details.
1596 */
1597int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1598 size_t cnt, loff_t *ppos)
1599{
1600 char ch;
1601 size_t read = 0;
1602 ssize_t ret;
1603
1604 if (!*ppos)
1605 trace_parser_clear(parser);
1606
1607 ret = get_user(ch, ubuf++);
1608 if (ret)
1609 goto out;
1610
1611 read++;
1612 cnt--;
1613
1614 /*
1615 * The parser is not finished with the last write,
1616 * continue reading the user input without skipping spaces.
1617 */
1618 if (!parser->cont) {
1619 /* skip white space */
1620 while (cnt && isspace(ch)) {
1621 ret = get_user(ch, ubuf++);
1622 if (ret)
1623 goto out;
1624 read++;
1625 cnt--;
1626 }
1627
1628 parser->idx = 0;
1629
1630 /* only spaces were written */
1631 if (isspace(ch) || !ch) {
1632 *ppos += read;
1633 ret = read;
1634 goto out;
1635 }
1636 }
1637
1638 /* read the non-space input */
1639 while (cnt && !isspace(ch) && ch) {
1640 if (parser->idx < parser->size - 1)
1641 parser->buffer[parser->idx++] = ch;
1642 else {
1643 ret = -EINVAL;
1644 goto out;
1645 }
1646 ret = get_user(ch, ubuf++);
1647 if (ret)
1648 goto out;
1649 read++;
1650 cnt--;
1651 }
1652
1653 /* We either got finished input or we have to wait for another call. */
1654 if (isspace(ch) || !ch) {
1655 parser->buffer[parser->idx] = 0;
1656 parser->cont = false;
1657 } else if (parser->idx < parser->size - 1) {
1658 parser->cont = true;
1659 parser->buffer[parser->idx++] = ch;
1660 /* Make sure the parsed string always terminates with '\0'. */
1661 parser->buffer[parser->idx] = 0;
1662 } else {
1663 ret = -EINVAL;
1664 goto out;
1665 }
1666
1667 *ppos += read;
1668 ret = read;
1669
1670out:
1671 return ret;
1672}
1673
1674/* TODO add a seq_buf_to_buffer() */
1675static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1676{
1677 int len;
1678
1679 if (trace_seq_used(s) <= s->seq.readpos)
1680 return -EBUSY;
1681
1682 len = trace_seq_used(s) - s->seq.readpos;
1683 if (cnt > len)
1684 cnt = len;
1685 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1686
1687 s->seq.readpos += cnt;
1688 return cnt;
1689}
1690
1691unsigned long __read_mostly tracing_thresh;
1692static const struct file_operations tracing_max_lat_fops;
1693
1694#ifdef LATENCY_FS_NOTIFY
1695
1696static struct workqueue_struct *fsnotify_wq;
1697
1698static void latency_fsnotify_workfn(struct work_struct *work)
1699{
1700 struct trace_array *tr = container_of(work, struct trace_array,
1701 fsnotify_work);
1702 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1703}
1704
1705static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1706{
1707 struct trace_array *tr = container_of(iwork, struct trace_array,
1708 fsnotify_irqwork);
1709 queue_work(fsnotify_wq, &tr->fsnotify_work);
1710}
1711
1712static void trace_create_maxlat_file(struct trace_array *tr,
1713 struct dentry *d_tracer)
1714{
1715 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1716 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1717 tr->d_max_latency = trace_create_file("tracing_max_latency", 0644,
1718 d_tracer, &tr->max_latency,
1719 &tracing_max_lat_fops);
1720}
1721
1722__init static int latency_fsnotify_init(void)
1723{
1724 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1725 WQ_UNBOUND | WQ_HIGHPRI, 0);
1726 if (!fsnotify_wq) {
1727 pr_err("Unable to allocate tr_max_lat_wq\n");
1728 return -ENOMEM;
1729 }
1730 return 0;
1731}
1732
1733late_initcall_sync(latency_fsnotify_init);
1734
1735void latency_fsnotify(struct trace_array *tr)
1736{
1737 if (!fsnotify_wq)
1738 return;
1739 /*
1740 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1741 * possible that we are called from __schedule() or do_idle(), which
1742 * could cause a deadlock.
1743 */
1744 irq_work_queue(&tr->fsnotify_irqwork);
1745}
1746
1747#elif defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER) \
1748 || defined(CONFIG_OSNOISE_TRACER)
1749
1750#define trace_create_maxlat_file(tr, d_tracer) \
1751 trace_create_file("tracing_max_latency", 0644, d_tracer, \
1752 &tr->max_latency, &tracing_max_lat_fops)
1753
1754#else
1755#define trace_create_maxlat_file(tr, d_tracer) do { } while (0)
1756#endif
1757
1758#ifdef CONFIG_TRACER_MAX_TRACE
1759/*
1760 * Copy the new maximum trace into the separate maximum-trace
1761 * structure. (this way the maximum trace is permanently saved,
1762 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1763 */
1764static void
1765__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1766{
1767 struct array_buffer *trace_buf = &tr->array_buffer;
1768 struct array_buffer *max_buf = &tr->max_buffer;
1769 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1770 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1771
1772 max_buf->cpu = cpu;
1773 max_buf->time_start = data->preempt_timestamp;
1774
1775 max_data->saved_latency = tr->max_latency;
1776 max_data->critical_start = data->critical_start;
1777 max_data->critical_end = data->critical_end;
1778
1779 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1780 max_data->pid = tsk->pid;
1781 /*
1782 * If tsk == current, then use current_uid(), as that does not use
1783 * RCU. The irq tracer can be called out of RCU scope.
1784 */
1785 if (tsk == current)
1786 max_data->uid = current_uid();
1787 else
1788 max_data->uid = task_uid(tsk);
1789
1790 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1791 max_data->policy = tsk->policy;
1792 max_data->rt_priority = tsk->rt_priority;
1793
1794 /* record this tasks comm */
1795 tracing_record_cmdline(tsk);
1796 latency_fsnotify(tr);
1797}
1798
1799/**
1800 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1801 * @tr: tracer
1802 * @tsk: the task with the latency
1803 * @cpu: The cpu that initiated the trace.
1804 * @cond_data: User data associated with a conditional snapshot
1805 *
1806 * Flip the buffers between the @tr and the max_tr and record information
1807 * about which task was the cause of this latency.
1808 */
1809void
1810update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1811 void *cond_data)
1812{
1813 if (tr->stop_count)
1814 return;
1815
1816 WARN_ON_ONCE(!irqs_disabled());
1817
1818 if (!tr->allocated_snapshot) {
1819 /* Only the nop tracer should hit this when disabling */
1820 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1821 return;
1822 }
1823
1824 arch_spin_lock(&tr->max_lock);
1825
1826 /* Inherit the recordable setting from array_buffer */
1827 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1828 ring_buffer_record_on(tr->max_buffer.buffer);
1829 else
1830 ring_buffer_record_off(tr->max_buffer.buffer);
1831
1832#ifdef CONFIG_TRACER_SNAPSHOT
1833 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data))
1834 goto out_unlock;
1835#endif
1836 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1837
1838 __update_max_tr(tr, tsk, cpu);
1839
1840 out_unlock:
1841 arch_spin_unlock(&tr->max_lock);
1842}
1843
1844/**
1845 * update_max_tr_single - only copy one trace over, and reset the rest
1846 * @tr: tracer
1847 * @tsk: task with the latency
1848 * @cpu: the cpu of the buffer to copy.
1849 *
1850 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1851 */
1852void
1853update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1854{
1855 int ret;
1856
1857 if (tr->stop_count)
1858 return;
1859
1860 WARN_ON_ONCE(!irqs_disabled());
1861 if (!tr->allocated_snapshot) {
1862 /* Only the nop tracer should hit this when disabling */
1863 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1864 return;
1865 }
1866
1867 arch_spin_lock(&tr->max_lock);
1868
1869 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1870
1871 if (ret == -EBUSY) {
1872 /*
1873 * We failed to swap the buffer due to a commit taking
1874 * place on this CPU. We fail to record, but we reset
1875 * the max trace buffer (no one writes directly to it)
1876 * and flag that it failed.
1877 */
1878 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1879 "Failed to swap buffers due to commit in progress\n");
1880 }
1881
1882 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1883
1884 __update_max_tr(tr, tsk, cpu);
1885 arch_spin_unlock(&tr->max_lock);
1886}
1887#endif /* CONFIG_TRACER_MAX_TRACE */
1888
1889static int wait_on_pipe(struct trace_iterator *iter, int full)
1890{
1891 /* Iterators are static, they should be filled or empty */
1892 if (trace_buffer_iter(iter, iter->cpu_file))
1893 return 0;
1894
1895 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1896 full);
1897}
1898
1899#ifdef CONFIG_FTRACE_STARTUP_TEST
1900static bool selftests_can_run;
1901
1902struct trace_selftests {
1903 struct list_head list;
1904 struct tracer *type;
1905};
1906
1907static LIST_HEAD(postponed_selftests);
1908
1909static int save_selftest(struct tracer *type)
1910{
1911 struct trace_selftests *selftest;
1912
1913 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1914 if (!selftest)
1915 return -ENOMEM;
1916
1917 selftest->type = type;
1918 list_add(&selftest->list, &postponed_selftests);
1919 return 0;
1920}
1921
1922static int run_tracer_selftest(struct tracer *type)
1923{
1924 struct trace_array *tr = &global_trace;
1925 struct tracer *saved_tracer = tr->current_trace;
1926 int ret;
1927
1928 if (!type->selftest || tracing_selftest_disabled)
1929 return 0;
1930
1931 /*
1932 * If a tracer registers early in boot up (before scheduling is
1933 * initialized and such), then do not run its selftests yet.
1934 * Instead, run it a little later in the boot process.
1935 */
1936 if (!selftests_can_run)
1937 return save_selftest(type);
1938
1939 if (!tracing_is_on()) {
1940 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
1941 type->name);
1942 return 0;
1943 }
1944
1945 /*
1946 * Run a selftest on this tracer.
1947 * Here we reset the trace buffer, and set the current
1948 * tracer to be this tracer. The tracer can then run some
1949 * internal tracing to verify that everything is in order.
1950 * If we fail, we do not register this tracer.
1951 */
1952 tracing_reset_online_cpus(&tr->array_buffer);
1953
1954 tr->current_trace = type;
1955
1956#ifdef CONFIG_TRACER_MAX_TRACE
1957 if (type->use_max_tr) {
1958 /* If we expanded the buffers, make sure the max is expanded too */
1959 if (ring_buffer_expanded)
1960 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1961 RING_BUFFER_ALL_CPUS);
1962 tr->allocated_snapshot = true;
1963 }
1964#endif
1965
1966 /* the test is responsible for initializing and enabling */
1967 pr_info("Testing tracer %s: ", type->name);
1968 ret = type->selftest(type, tr);
1969 /* the test is responsible for resetting too */
1970 tr->current_trace = saved_tracer;
1971 if (ret) {
1972 printk(KERN_CONT "FAILED!\n");
1973 /* Add the warning after printing 'FAILED' */
1974 WARN_ON(1);
1975 return -1;
1976 }
1977 /* Only reset on passing, to avoid touching corrupted buffers */
1978 tracing_reset_online_cpus(&tr->array_buffer);
1979
1980#ifdef CONFIG_TRACER_MAX_TRACE
1981 if (type->use_max_tr) {
1982 tr->allocated_snapshot = false;
1983
1984 /* Shrink the max buffer again */
1985 if (ring_buffer_expanded)
1986 ring_buffer_resize(tr->max_buffer.buffer, 1,
1987 RING_BUFFER_ALL_CPUS);
1988 }
1989#endif
1990
1991 printk(KERN_CONT "PASSED\n");
1992 return 0;
1993}
1994
1995static __init int init_trace_selftests(void)
1996{
1997 struct trace_selftests *p, *n;
1998 struct tracer *t, **last;
1999 int ret;
2000
2001 selftests_can_run = true;
2002
2003 mutex_lock(&trace_types_lock);
2004
2005 if (list_empty(&postponed_selftests))
2006 goto out;
2007
2008 pr_info("Running postponed tracer tests:\n");
2009
2010 tracing_selftest_running = true;
2011 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2012 /* This loop can take minutes when sanitizers are enabled, so
2013 * lets make sure we allow RCU processing.
2014 */
2015 cond_resched();
2016 ret = run_tracer_selftest(p->type);
2017 /* If the test fails, then warn and remove from available_tracers */
2018 if (ret < 0) {
2019 WARN(1, "tracer: %s failed selftest, disabling\n",
2020 p->type->name);
2021 last = &trace_types;
2022 for (t = trace_types; t; t = t->next) {
2023 if (t == p->type) {
2024 *last = t->next;
2025 break;
2026 }
2027 last = &t->next;
2028 }
2029 }
2030 list_del(&p->list);
2031 kfree(p);
2032 }
2033 tracing_selftest_running = false;
2034
2035 out:
2036 mutex_unlock(&trace_types_lock);
2037
2038 return 0;
2039}
2040core_initcall(init_trace_selftests);
2041#else
2042static inline int run_tracer_selftest(struct tracer *type)
2043{
2044 return 0;
2045}
2046#endif /* CONFIG_FTRACE_STARTUP_TEST */
2047
2048static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2049
2050static void __init apply_trace_boot_options(void);
2051
2052/**
2053 * register_tracer - register a tracer with the ftrace system.
2054 * @type: the plugin for the tracer
2055 *
2056 * Register a new plugin tracer.
2057 */
2058int __init register_tracer(struct tracer *type)
2059{
2060 struct tracer *t;
2061 int ret = 0;
2062
2063 if (!type->name) {
2064 pr_info("Tracer must have a name\n");
2065 return -1;
2066 }
2067
2068 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2069 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2070 return -1;
2071 }
2072
2073 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2074 pr_warn("Can not register tracer %s due to lockdown\n",
2075 type->name);
2076 return -EPERM;
2077 }
2078
2079 mutex_lock(&trace_types_lock);
2080
2081 tracing_selftest_running = true;
2082
2083 for (t = trace_types; t; t = t->next) {
2084 if (strcmp(type->name, t->name) == 0) {
2085 /* already found */
2086 pr_info("Tracer %s already registered\n",
2087 type->name);
2088 ret = -1;
2089 goto out;
2090 }
2091 }
2092
2093 if (!type->set_flag)
2094 type->set_flag = &dummy_set_flag;
2095 if (!type->flags) {
2096 /*allocate a dummy tracer_flags*/
2097 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2098 if (!type->flags) {
2099 ret = -ENOMEM;
2100 goto out;
2101 }
2102 type->flags->val = 0;
2103 type->flags->opts = dummy_tracer_opt;
2104 } else
2105 if (!type->flags->opts)
2106 type->flags->opts = dummy_tracer_opt;
2107
2108 /* store the tracer for __set_tracer_option */
2109 type->flags->trace = type;
2110
2111 ret = run_tracer_selftest(type);
2112 if (ret < 0)
2113 goto out;
2114
2115 type->next = trace_types;
2116 trace_types = type;
2117 add_tracer_options(&global_trace, type);
2118
2119 out:
2120 tracing_selftest_running = false;
2121 mutex_unlock(&trace_types_lock);
2122
2123 if (ret || !default_bootup_tracer)
2124 goto out_unlock;
2125
2126 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2127 goto out_unlock;
2128
2129 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2130 /* Do we want this tracer to start on bootup? */
2131 tracing_set_tracer(&global_trace, type->name);
2132 default_bootup_tracer = NULL;
2133
2134 apply_trace_boot_options();
2135
2136 /* disable other selftests, since this will break it. */
2137 disable_tracing_selftest("running a tracer");
2138
2139 out_unlock:
2140 return ret;
2141}
2142
2143static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2144{
2145 struct trace_buffer *buffer = buf->buffer;
2146
2147 if (!buffer)
2148 return;
2149
2150 ring_buffer_record_disable(buffer);
2151
2152 /* Make sure all commits have finished */
2153 synchronize_rcu();
2154 ring_buffer_reset_cpu(buffer, cpu);
2155
2156 ring_buffer_record_enable(buffer);
2157}
2158
2159void tracing_reset_online_cpus(struct array_buffer *buf)
2160{
2161 struct trace_buffer *buffer = buf->buffer;
2162
2163 if (!buffer)
2164 return;
2165
2166 ring_buffer_record_disable(buffer);
2167
2168 /* Make sure all commits have finished */
2169 synchronize_rcu();
2170
2171 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2172
2173 ring_buffer_reset_online_cpus(buffer);
2174
2175 ring_buffer_record_enable(buffer);
2176}
2177
2178/* Must have trace_types_lock held */
2179void tracing_reset_all_online_cpus(void)
2180{
2181 struct trace_array *tr;
2182
2183 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2184 if (!tr->clear_trace)
2185 continue;
2186 tr->clear_trace = false;
2187 tracing_reset_online_cpus(&tr->array_buffer);
2188#ifdef CONFIG_TRACER_MAX_TRACE
2189 tracing_reset_online_cpus(&tr->max_buffer);
2190#endif
2191 }
2192}
2193
2194/*
2195 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2196 * is the tgid last observed corresponding to pid=i.
2197 */
2198static int *tgid_map;
2199
2200/* The maximum valid index into tgid_map. */
2201static size_t tgid_map_max;
2202
2203#define SAVED_CMDLINES_DEFAULT 128
2204#define NO_CMDLINE_MAP UINT_MAX
2205static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2206struct saved_cmdlines_buffer {
2207 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2208 unsigned *map_cmdline_to_pid;
2209 unsigned cmdline_num;
2210 int cmdline_idx;
2211 char *saved_cmdlines;
2212};
2213static struct saved_cmdlines_buffer *savedcmd;
2214
2215static inline char *get_saved_cmdlines(int idx)
2216{
2217 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2218}
2219
2220static inline void set_cmdline(int idx, const char *cmdline)
2221{
2222 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2223}
2224
2225static int allocate_cmdlines_buffer(unsigned int val,
2226 struct saved_cmdlines_buffer *s)
2227{
2228 s->map_cmdline_to_pid = kmalloc_array(val,
2229 sizeof(*s->map_cmdline_to_pid),
2230 GFP_KERNEL);
2231 if (!s->map_cmdline_to_pid)
2232 return -ENOMEM;
2233
2234 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2235 if (!s->saved_cmdlines) {
2236 kfree(s->map_cmdline_to_pid);
2237 return -ENOMEM;
2238 }
2239
2240 s->cmdline_idx = 0;
2241 s->cmdline_num = val;
2242 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2243 sizeof(s->map_pid_to_cmdline));
2244 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2245 val * sizeof(*s->map_cmdline_to_pid));
2246
2247 return 0;
2248}
2249
2250static int trace_create_savedcmd(void)
2251{
2252 int ret;
2253
2254 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2255 if (!savedcmd)
2256 return -ENOMEM;
2257
2258 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2259 if (ret < 0) {
2260 kfree(savedcmd);
2261 savedcmd = NULL;
2262 return -ENOMEM;
2263 }
2264
2265 return 0;
2266}
2267
2268int is_tracing_stopped(void)
2269{
2270 return global_trace.stop_count;
2271}
2272
2273/**
2274 * tracing_start - quick start of the tracer
2275 *
2276 * If tracing is enabled but was stopped by tracing_stop,
2277 * this will start the tracer back up.
2278 */
2279void tracing_start(void)
2280{
2281 struct trace_buffer *buffer;
2282 unsigned long flags;
2283
2284 if (tracing_disabled)
2285 return;
2286
2287 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2288 if (--global_trace.stop_count) {
2289 if (global_trace.stop_count < 0) {
2290 /* Someone screwed up their debugging */
2291 WARN_ON_ONCE(1);
2292 global_trace.stop_count = 0;
2293 }
2294 goto out;
2295 }
2296
2297 /* Prevent the buffers from switching */
2298 arch_spin_lock(&global_trace.max_lock);
2299
2300 buffer = global_trace.array_buffer.buffer;
2301 if (buffer)
2302 ring_buffer_record_enable(buffer);
2303
2304#ifdef CONFIG_TRACER_MAX_TRACE
2305 buffer = global_trace.max_buffer.buffer;
2306 if (buffer)
2307 ring_buffer_record_enable(buffer);
2308#endif
2309
2310 arch_spin_unlock(&global_trace.max_lock);
2311
2312 out:
2313 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2314}
2315
2316static void tracing_start_tr(struct trace_array *tr)
2317{
2318 struct trace_buffer *buffer;
2319 unsigned long flags;
2320
2321 if (tracing_disabled)
2322 return;
2323
2324 /* If global, we need to also start the max tracer */
2325 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2326 return tracing_start();
2327
2328 raw_spin_lock_irqsave(&tr->start_lock, flags);
2329
2330 if (--tr->stop_count) {
2331 if (tr->stop_count < 0) {
2332 /* Someone screwed up their debugging */
2333 WARN_ON_ONCE(1);
2334 tr->stop_count = 0;
2335 }
2336 goto out;
2337 }
2338
2339 buffer = tr->array_buffer.buffer;
2340 if (buffer)
2341 ring_buffer_record_enable(buffer);
2342
2343 out:
2344 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2345}
2346
2347/**
2348 * tracing_stop - quick stop of the tracer
2349 *
2350 * Light weight way to stop tracing. Use in conjunction with
2351 * tracing_start.
2352 */
2353void tracing_stop(void)
2354{
2355 struct trace_buffer *buffer;
2356 unsigned long flags;
2357
2358 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2359 if (global_trace.stop_count++)
2360 goto out;
2361
2362 /* Prevent the buffers from switching */
2363 arch_spin_lock(&global_trace.max_lock);
2364
2365 buffer = global_trace.array_buffer.buffer;
2366 if (buffer)
2367 ring_buffer_record_disable(buffer);
2368
2369#ifdef CONFIG_TRACER_MAX_TRACE
2370 buffer = global_trace.max_buffer.buffer;
2371 if (buffer)
2372 ring_buffer_record_disable(buffer);
2373#endif
2374
2375 arch_spin_unlock(&global_trace.max_lock);
2376
2377 out:
2378 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2379}
2380
2381static void tracing_stop_tr(struct trace_array *tr)
2382{
2383 struct trace_buffer *buffer;
2384 unsigned long flags;
2385
2386 /* If global, we need to also stop the max tracer */
2387 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2388 return tracing_stop();
2389
2390 raw_spin_lock_irqsave(&tr->start_lock, flags);
2391 if (tr->stop_count++)
2392 goto out;
2393
2394 buffer = tr->array_buffer.buffer;
2395 if (buffer)
2396 ring_buffer_record_disable(buffer);
2397
2398 out:
2399 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2400}
2401
2402static int trace_save_cmdline(struct task_struct *tsk)
2403{
2404 unsigned tpid, idx;
2405
2406 /* treat recording of idle task as a success */
2407 if (!tsk->pid)
2408 return 1;
2409
2410 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2411
2412 /*
2413 * It's not the end of the world if we don't get
2414 * the lock, but we also don't want to spin
2415 * nor do we want to disable interrupts,
2416 * so if we miss here, then better luck next time.
2417 */
2418 if (!arch_spin_trylock(&trace_cmdline_lock))
2419 return 0;
2420
2421 idx = savedcmd->map_pid_to_cmdline[tpid];
2422 if (idx == NO_CMDLINE_MAP) {
2423 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2424
2425 savedcmd->map_pid_to_cmdline[tpid] = idx;
2426 savedcmd->cmdline_idx = idx;
2427 }
2428
2429 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2430 set_cmdline(idx, tsk->comm);
2431
2432 arch_spin_unlock(&trace_cmdline_lock);
2433
2434 return 1;
2435}
2436
2437static void __trace_find_cmdline(int pid, char comm[])
2438{
2439 unsigned map;
2440 int tpid;
2441
2442 if (!pid) {
2443 strcpy(comm, "<idle>");
2444 return;
2445 }
2446
2447 if (WARN_ON_ONCE(pid < 0)) {
2448 strcpy(comm, "<XXX>");
2449 return;
2450 }
2451
2452 tpid = pid & (PID_MAX_DEFAULT - 1);
2453 map = savedcmd->map_pid_to_cmdline[tpid];
2454 if (map != NO_CMDLINE_MAP) {
2455 tpid = savedcmd->map_cmdline_to_pid[map];
2456 if (tpid == pid) {
2457 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2458 return;
2459 }
2460 }
2461 strcpy(comm, "<...>");
2462}
2463
2464void trace_find_cmdline(int pid, char comm[])
2465{
2466 preempt_disable();
2467 arch_spin_lock(&trace_cmdline_lock);
2468
2469 __trace_find_cmdline(pid, comm);
2470
2471 arch_spin_unlock(&trace_cmdline_lock);
2472 preempt_enable();
2473}
2474
2475static int *trace_find_tgid_ptr(int pid)
2476{
2477 /*
2478 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2479 * if we observe a non-NULL tgid_map then we also observe the correct
2480 * tgid_map_max.
2481 */
2482 int *map = smp_load_acquire(&tgid_map);
2483
2484 if (unlikely(!map || pid > tgid_map_max))
2485 return NULL;
2486
2487 return &map[pid];
2488}
2489
2490int trace_find_tgid(int pid)
2491{
2492 int *ptr = trace_find_tgid_ptr(pid);
2493
2494 return ptr ? *ptr : 0;
2495}
2496
2497static int trace_save_tgid(struct task_struct *tsk)
2498{
2499 int *ptr;
2500
2501 /* treat recording of idle task as a success */
2502 if (!tsk->pid)
2503 return 1;
2504
2505 ptr = trace_find_tgid_ptr(tsk->pid);
2506 if (!ptr)
2507 return 0;
2508
2509 *ptr = tsk->tgid;
2510 return 1;
2511}
2512
2513static bool tracing_record_taskinfo_skip(int flags)
2514{
2515 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2516 return true;
2517 if (!__this_cpu_read(trace_taskinfo_save))
2518 return true;
2519 return false;
2520}
2521
2522/**
2523 * tracing_record_taskinfo - record the task info of a task
2524 *
2525 * @task: task to record
2526 * @flags: TRACE_RECORD_CMDLINE for recording comm
2527 * TRACE_RECORD_TGID for recording tgid
2528 */
2529void tracing_record_taskinfo(struct task_struct *task, int flags)
2530{
2531 bool done;
2532
2533 if (tracing_record_taskinfo_skip(flags))
2534 return;
2535
2536 /*
2537 * Record as much task information as possible. If some fail, continue
2538 * to try to record the others.
2539 */
2540 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2541 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2542
2543 /* If recording any information failed, retry again soon. */
2544 if (!done)
2545 return;
2546
2547 __this_cpu_write(trace_taskinfo_save, false);
2548}
2549
2550/**
2551 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2552 *
2553 * @prev: previous task during sched_switch
2554 * @next: next task during sched_switch
2555 * @flags: TRACE_RECORD_CMDLINE for recording comm
2556 * TRACE_RECORD_TGID for recording tgid
2557 */
2558void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2559 struct task_struct *next, int flags)
2560{
2561 bool done;
2562
2563 if (tracing_record_taskinfo_skip(flags))
2564 return;
2565
2566 /*
2567 * Record as much task information as possible. If some fail, continue
2568 * to try to record the others.
2569 */
2570 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2571 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2572 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2573 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2574
2575 /* If recording any information failed, retry again soon. */
2576 if (!done)
2577 return;
2578
2579 __this_cpu_write(trace_taskinfo_save, false);
2580}
2581
2582/* Helpers to record a specific task information */
2583void tracing_record_cmdline(struct task_struct *task)
2584{
2585 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2586}
2587
2588void tracing_record_tgid(struct task_struct *task)
2589{
2590 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2591}
2592
2593/*
2594 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2595 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2596 * simplifies those functions and keeps them in sync.
2597 */
2598enum print_line_t trace_handle_return(struct trace_seq *s)
2599{
2600 return trace_seq_has_overflowed(s) ?
2601 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2602}
2603EXPORT_SYMBOL_GPL(trace_handle_return);
2604
2605unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2606{
2607 unsigned int trace_flags = irqs_status;
2608 unsigned int pc;
2609
2610 pc = preempt_count();
2611
2612 if (pc & NMI_MASK)
2613 trace_flags |= TRACE_FLAG_NMI;
2614 if (pc & HARDIRQ_MASK)
2615 trace_flags |= TRACE_FLAG_HARDIRQ;
2616 if (in_serving_softirq())
2617 trace_flags |= TRACE_FLAG_SOFTIRQ;
2618
2619 if (tif_need_resched())
2620 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2621 if (test_preempt_need_resched())
2622 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2623 return (trace_flags << 16) | (pc & 0xff);
2624}
2625
2626struct ring_buffer_event *
2627trace_buffer_lock_reserve(struct trace_buffer *buffer,
2628 int type,
2629 unsigned long len,
2630 unsigned int trace_ctx)
2631{
2632 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2633}
2634
2635DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2636DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2637static int trace_buffered_event_ref;
2638
2639/**
2640 * trace_buffered_event_enable - enable buffering events
2641 *
2642 * When events are being filtered, it is quicker to use a temporary
2643 * buffer to write the event data into if there's a likely chance
2644 * that it will not be committed. The discard of the ring buffer
2645 * is not as fast as committing, and is much slower than copying
2646 * a commit.
2647 *
2648 * When an event is to be filtered, allocate per cpu buffers to
2649 * write the event data into, and if the event is filtered and discarded
2650 * it is simply dropped, otherwise, the entire data is to be committed
2651 * in one shot.
2652 */
2653void trace_buffered_event_enable(void)
2654{
2655 struct ring_buffer_event *event;
2656 struct page *page;
2657 int cpu;
2658
2659 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2660
2661 if (trace_buffered_event_ref++)
2662 return;
2663
2664 for_each_tracing_cpu(cpu) {
2665 page = alloc_pages_node(cpu_to_node(cpu),
2666 GFP_KERNEL | __GFP_NORETRY, 0);
2667 if (!page)
2668 goto failed;
2669
2670 event = page_address(page);
2671 memset(event, 0, sizeof(*event));
2672
2673 per_cpu(trace_buffered_event, cpu) = event;
2674
2675 preempt_disable();
2676 if (cpu == smp_processor_id() &&
2677 __this_cpu_read(trace_buffered_event) !=
2678 per_cpu(trace_buffered_event, cpu))
2679 WARN_ON_ONCE(1);
2680 preempt_enable();
2681 }
2682
2683 return;
2684 failed:
2685 trace_buffered_event_disable();
2686}
2687
2688static void enable_trace_buffered_event(void *data)
2689{
2690 /* Probably not needed, but do it anyway */
2691 smp_rmb();
2692 this_cpu_dec(trace_buffered_event_cnt);
2693}
2694
2695static void disable_trace_buffered_event(void *data)
2696{
2697 this_cpu_inc(trace_buffered_event_cnt);
2698}
2699
2700/**
2701 * trace_buffered_event_disable - disable buffering events
2702 *
2703 * When a filter is removed, it is faster to not use the buffered
2704 * events, and to commit directly into the ring buffer. Free up
2705 * the temp buffers when there are no more users. This requires
2706 * special synchronization with current events.
2707 */
2708void trace_buffered_event_disable(void)
2709{
2710 int cpu;
2711
2712 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2713
2714 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2715 return;
2716
2717 if (--trace_buffered_event_ref)
2718 return;
2719
2720 preempt_disable();
2721 /* For each CPU, set the buffer as used. */
2722 smp_call_function_many(tracing_buffer_mask,
2723 disable_trace_buffered_event, NULL, 1);
2724 preempt_enable();
2725
2726 /* Wait for all current users to finish */
2727 synchronize_rcu();
2728
2729 for_each_tracing_cpu(cpu) {
2730 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2731 per_cpu(trace_buffered_event, cpu) = NULL;
2732 }
2733 /*
2734 * Make sure trace_buffered_event is NULL before clearing
2735 * trace_buffered_event_cnt.
2736 */
2737 smp_wmb();
2738
2739 preempt_disable();
2740 /* Do the work on each cpu */
2741 smp_call_function_many(tracing_buffer_mask,
2742 enable_trace_buffered_event, NULL, 1);
2743 preempt_enable();
2744}
2745
2746static struct trace_buffer *temp_buffer;
2747
2748struct ring_buffer_event *
2749trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2750 struct trace_event_file *trace_file,
2751 int type, unsigned long len,
2752 unsigned int trace_ctx)
2753{
2754 struct ring_buffer_event *entry;
2755 struct trace_array *tr = trace_file->tr;
2756 int val;
2757
2758 *current_rb = tr->array_buffer.buffer;
2759
2760 if (!tr->no_filter_buffering_ref &&
2761 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED)) &&
2762 (entry = this_cpu_read(trace_buffered_event))) {
2763 /*
2764 * Filtering is on, so try to use the per cpu buffer first.
2765 * This buffer will simulate a ring_buffer_event,
2766 * where the type_len is zero and the array[0] will
2767 * hold the full length.
2768 * (see include/linux/ring-buffer.h for details on
2769 * how the ring_buffer_event is structured).
2770 *
2771 * Using a temp buffer during filtering and copying it
2772 * on a matched filter is quicker than writing directly
2773 * into the ring buffer and then discarding it when
2774 * it doesn't match. That is because the discard
2775 * requires several atomic operations to get right.
2776 * Copying on match and doing nothing on a failed match
2777 * is still quicker than no copy on match, but having
2778 * to discard out of the ring buffer on a failed match.
2779 */
2780 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2781
2782 val = this_cpu_inc_return(trace_buffered_event_cnt);
2783
2784 /*
2785 * Preemption is disabled, but interrupts and NMIs
2786 * can still come in now. If that happens after
2787 * the above increment, then it will have to go
2788 * back to the old method of allocating the event
2789 * on the ring buffer, and if the filter fails, it
2790 * will have to call ring_buffer_discard_commit()
2791 * to remove it.
2792 *
2793 * Need to also check the unlikely case that the
2794 * length is bigger than the temp buffer size.
2795 * If that happens, then the reserve is pretty much
2796 * guaranteed to fail, as the ring buffer currently
2797 * only allows events less than a page. But that may
2798 * change in the future, so let the ring buffer reserve
2799 * handle the failure in that case.
2800 */
2801 if (val == 1 && likely(len <= max_len)) {
2802 trace_event_setup(entry, type, trace_ctx);
2803 entry->array[0] = len;
2804 return entry;
2805 }
2806 this_cpu_dec(trace_buffered_event_cnt);
2807 }
2808
2809 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2810 trace_ctx);
2811 /*
2812 * If tracing is off, but we have triggers enabled
2813 * we still need to look at the event data. Use the temp_buffer
2814 * to store the trace event for the trigger to use. It's recursive
2815 * safe and will not be recorded anywhere.
2816 */
2817 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2818 *current_rb = temp_buffer;
2819 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2820 trace_ctx);
2821 }
2822 return entry;
2823}
2824EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2825
2826static DEFINE_SPINLOCK(tracepoint_iter_lock);
2827static DEFINE_MUTEX(tracepoint_printk_mutex);
2828
2829static void output_printk(struct trace_event_buffer *fbuffer)
2830{
2831 struct trace_event_call *event_call;
2832 struct trace_event_file *file;
2833 struct trace_event *event;
2834 unsigned long flags;
2835 struct trace_iterator *iter = tracepoint_print_iter;
2836
2837 /* We should never get here if iter is NULL */
2838 if (WARN_ON_ONCE(!iter))
2839 return;
2840
2841 event_call = fbuffer->trace_file->event_call;
2842 if (!event_call || !event_call->event.funcs ||
2843 !event_call->event.funcs->trace)
2844 return;
2845
2846 file = fbuffer->trace_file;
2847 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2848 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2849 !filter_match_preds(file->filter, fbuffer->entry)))
2850 return;
2851
2852 event = &fbuffer->trace_file->event_call->event;
2853
2854 spin_lock_irqsave(&tracepoint_iter_lock, flags);
2855 trace_seq_init(&iter->seq);
2856 iter->ent = fbuffer->entry;
2857 event_call->event.funcs->trace(iter, 0, event);
2858 trace_seq_putc(&iter->seq, 0);
2859 printk("%s", iter->seq.buffer);
2860
2861 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2862}
2863
2864int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2865 void *buffer, size_t *lenp,
2866 loff_t *ppos)
2867{
2868 int save_tracepoint_printk;
2869 int ret;
2870
2871 mutex_lock(&tracepoint_printk_mutex);
2872 save_tracepoint_printk = tracepoint_printk;
2873
2874 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2875
2876 /*
2877 * This will force exiting early, as tracepoint_printk
2878 * is always zero when tracepoint_printk_iter is not allocated
2879 */
2880 if (!tracepoint_print_iter)
2881 tracepoint_printk = 0;
2882
2883 if (save_tracepoint_printk == tracepoint_printk)
2884 goto out;
2885
2886 if (tracepoint_printk)
2887 static_key_enable(&tracepoint_printk_key.key);
2888 else
2889 static_key_disable(&tracepoint_printk_key.key);
2890
2891 out:
2892 mutex_unlock(&tracepoint_printk_mutex);
2893
2894 return ret;
2895}
2896
2897void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2898{
2899 enum event_trigger_type tt = ETT_NONE;
2900 struct trace_event_file *file = fbuffer->trace_file;
2901
2902 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2903 fbuffer->entry, &tt))
2904 goto discard;
2905
2906 if (static_key_false(&tracepoint_printk_key.key))
2907 output_printk(fbuffer);
2908
2909 if (static_branch_unlikely(&trace_event_exports_enabled))
2910 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2911
2912 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2913 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
2914
2915discard:
2916 if (tt)
2917 event_triggers_post_call(file, tt);
2918
2919}
2920EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2921
2922/*
2923 * Skip 3:
2924 *
2925 * trace_buffer_unlock_commit_regs()
2926 * trace_event_buffer_commit()
2927 * trace_event_raw_event_xxx()
2928 */
2929# define STACK_SKIP 3
2930
2931void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2932 struct trace_buffer *buffer,
2933 struct ring_buffer_event *event,
2934 unsigned int trace_ctx,
2935 struct pt_regs *regs)
2936{
2937 __buffer_unlock_commit(buffer, event);
2938
2939 /*
2940 * If regs is not set, then skip the necessary functions.
2941 * Note, we can still get here via blktrace, wakeup tracer
2942 * and mmiotrace, but that's ok if they lose a function or
2943 * two. They are not that meaningful.
2944 */
2945 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
2946 ftrace_trace_userstack(tr, buffer, trace_ctx);
2947}
2948
2949/*
2950 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2951 */
2952void
2953trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2954 struct ring_buffer_event *event)
2955{
2956 __buffer_unlock_commit(buffer, event);
2957}
2958
2959void
2960trace_function(struct trace_array *tr, unsigned long ip, unsigned long
2961 parent_ip, unsigned int trace_ctx)
2962{
2963 struct trace_event_call *call = &event_function;
2964 struct trace_buffer *buffer = tr->array_buffer.buffer;
2965 struct ring_buffer_event *event;
2966 struct ftrace_entry *entry;
2967
2968 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2969 trace_ctx);
2970 if (!event)
2971 return;
2972 entry = ring_buffer_event_data(event);
2973 entry->ip = ip;
2974 entry->parent_ip = parent_ip;
2975
2976 if (!call_filter_check_discard(call, entry, buffer, event)) {
2977 if (static_branch_unlikely(&trace_function_exports_enabled))
2978 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
2979 __buffer_unlock_commit(buffer, event);
2980 }
2981}
2982
2983#ifdef CONFIG_STACKTRACE
2984
2985/* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2986#define FTRACE_KSTACK_NESTING 4
2987
2988#define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
2989
2990struct ftrace_stack {
2991 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2992};
2993
2994
2995struct ftrace_stacks {
2996 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
2997};
2998
2999static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3000static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3001
3002static void __ftrace_trace_stack(struct trace_buffer *buffer,
3003 unsigned int trace_ctx,
3004 int skip, struct pt_regs *regs)
3005{
3006 struct trace_event_call *call = &event_kernel_stack;
3007 struct ring_buffer_event *event;
3008 unsigned int size, nr_entries;
3009 struct ftrace_stack *fstack;
3010 struct stack_entry *entry;
3011 int stackidx;
3012
3013 /*
3014 * Add one, for this function and the call to save_stack_trace()
3015 * If regs is set, then these functions will not be in the way.
3016 */
3017#ifndef CONFIG_UNWINDER_ORC
3018 if (!regs)
3019 skip++;
3020#endif
3021
3022 preempt_disable_notrace();
3023
3024 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3025
3026 /* This should never happen. If it does, yell once and skip */
3027 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3028 goto out;
3029
3030 /*
3031 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3032 * interrupt will either see the value pre increment or post
3033 * increment. If the interrupt happens pre increment it will have
3034 * restored the counter when it returns. We just need a barrier to
3035 * keep gcc from moving things around.
3036 */
3037 barrier();
3038
3039 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3040 size = ARRAY_SIZE(fstack->calls);
3041
3042 if (regs) {
3043 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3044 size, skip);
3045 } else {
3046 nr_entries = stack_trace_save(fstack->calls, size, skip);
3047 }
3048
3049 size = nr_entries * sizeof(unsigned long);
3050 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3051 (sizeof(*entry) - sizeof(entry->caller)) + size,
3052 trace_ctx);
3053 if (!event)
3054 goto out;
3055 entry = ring_buffer_event_data(event);
3056
3057 memcpy(&entry->caller, fstack->calls, size);
3058 entry->size = nr_entries;
3059
3060 if (!call_filter_check_discard(call, entry, buffer, event))
3061 __buffer_unlock_commit(buffer, event);
3062
3063 out:
3064 /* Again, don't let gcc optimize things here */
3065 barrier();
3066 __this_cpu_dec(ftrace_stack_reserve);
3067 preempt_enable_notrace();
3068
3069}
3070
3071static inline void ftrace_trace_stack(struct trace_array *tr,
3072 struct trace_buffer *buffer,
3073 unsigned int trace_ctx,
3074 int skip, struct pt_regs *regs)
3075{
3076 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3077 return;
3078
3079 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3080}
3081
3082void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3083 int skip)
3084{
3085 struct trace_buffer *buffer = tr->array_buffer.buffer;
3086
3087 if (rcu_is_watching()) {
3088 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3089 return;
3090 }
3091
3092 /*
3093 * When an NMI triggers, RCU is enabled via rcu_nmi_enter(),
3094 * but if the above rcu_is_watching() failed, then the NMI
3095 * triggered someplace critical, and rcu_irq_enter() should
3096 * not be called from NMI.
3097 */
3098 if (unlikely(in_nmi()))
3099 return;
3100
3101 rcu_irq_enter_irqson();
3102 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3103 rcu_irq_exit_irqson();
3104}
3105
3106/**
3107 * trace_dump_stack - record a stack back trace in the trace buffer
3108 * @skip: Number of functions to skip (helper handlers)
3109 */
3110void trace_dump_stack(int skip)
3111{
3112 if (tracing_disabled || tracing_selftest_running)
3113 return;
3114
3115#ifndef CONFIG_UNWINDER_ORC
3116 /* Skip 1 to skip this function. */
3117 skip++;
3118#endif
3119 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3120 tracing_gen_ctx(), skip, NULL);
3121}
3122EXPORT_SYMBOL_GPL(trace_dump_stack);
3123
3124#ifdef CONFIG_USER_STACKTRACE_SUPPORT
3125static DEFINE_PER_CPU(int, user_stack_count);
3126
3127static void
3128ftrace_trace_userstack(struct trace_array *tr,
3129 struct trace_buffer *buffer, unsigned int trace_ctx)
3130{
3131 struct trace_event_call *call = &event_user_stack;
3132 struct ring_buffer_event *event;
3133 struct userstack_entry *entry;
3134
3135 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3136 return;
3137
3138 /*
3139 * NMIs can not handle page faults, even with fix ups.
3140 * The save user stack can (and often does) fault.
3141 */
3142 if (unlikely(in_nmi()))
3143 return;
3144
3145 /*
3146 * prevent recursion, since the user stack tracing may
3147 * trigger other kernel events.
3148 */
3149 preempt_disable();
3150 if (__this_cpu_read(user_stack_count))
3151 goto out;
3152
3153 __this_cpu_inc(user_stack_count);
3154
3155 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3156 sizeof(*entry), trace_ctx);
3157 if (!event)
3158 goto out_drop_count;
3159 entry = ring_buffer_event_data(event);
3160
3161 entry->tgid = current->tgid;
3162 memset(&entry->caller, 0, sizeof(entry->caller));
3163
3164 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3165 if (!call_filter_check_discard(call, entry, buffer, event))
3166 __buffer_unlock_commit(buffer, event);
3167
3168 out_drop_count:
3169 __this_cpu_dec(user_stack_count);
3170 out:
3171 preempt_enable();
3172}
3173#else /* CONFIG_USER_STACKTRACE_SUPPORT */
3174static void ftrace_trace_userstack(struct trace_array *tr,
3175 struct trace_buffer *buffer,
3176 unsigned int trace_ctx)
3177{
3178}
3179#endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3180
3181#endif /* CONFIG_STACKTRACE */
3182
3183static inline void
3184func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3185 unsigned long long delta)
3186{
3187 entry->bottom_delta_ts = delta & U32_MAX;
3188 entry->top_delta_ts = (delta >> 32);
3189}
3190
3191void trace_last_func_repeats(struct trace_array *tr,
3192 struct trace_func_repeats *last_info,
3193 unsigned int trace_ctx)
3194{
3195 struct trace_buffer *buffer = tr->array_buffer.buffer;
3196 struct func_repeats_entry *entry;
3197 struct ring_buffer_event *event;
3198 u64 delta;
3199
3200 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3201 sizeof(*entry), trace_ctx);
3202 if (!event)
3203 return;
3204
3205 delta = ring_buffer_event_time_stamp(buffer, event) -
3206 last_info->ts_last_call;
3207
3208 entry = ring_buffer_event_data(event);
3209 entry->ip = last_info->ip;
3210 entry->parent_ip = last_info->parent_ip;
3211 entry->count = last_info->count;
3212 func_repeats_set_delta_ts(entry, delta);
3213
3214 __buffer_unlock_commit(buffer, event);
3215}
3216
3217/* created for use with alloc_percpu */
3218struct trace_buffer_struct {
3219 int nesting;
3220 char buffer[4][TRACE_BUF_SIZE];
3221};
3222
3223static struct trace_buffer_struct *trace_percpu_buffer;
3224
3225/*
3226 * This allows for lockless recording. If we're nested too deeply, then
3227 * this returns NULL.
3228 */
3229static char *get_trace_buf(void)
3230{
3231 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3232
3233 if (!buffer || buffer->nesting >= 4)
3234 return NULL;
3235
3236 buffer->nesting++;
3237
3238 /* Interrupts must see nesting incremented before we use the buffer */
3239 barrier();
3240 return &buffer->buffer[buffer->nesting - 1][0];
3241}
3242
3243static void put_trace_buf(void)
3244{
3245 /* Don't let the decrement of nesting leak before this */
3246 barrier();
3247 this_cpu_dec(trace_percpu_buffer->nesting);
3248}
3249
3250static int alloc_percpu_trace_buffer(void)
3251{
3252 struct trace_buffer_struct *buffers;
3253
3254 if (trace_percpu_buffer)
3255 return 0;
3256
3257 buffers = alloc_percpu(struct trace_buffer_struct);
3258 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3259 return -ENOMEM;
3260
3261 trace_percpu_buffer = buffers;
3262 return 0;
3263}
3264
3265static int buffers_allocated;
3266
3267void trace_printk_init_buffers(void)
3268{
3269 if (buffers_allocated)
3270 return;
3271
3272 if (alloc_percpu_trace_buffer())
3273 return;
3274
3275 /* trace_printk() is for debug use only. Don't use it in production. */
3276
3277 pr_warn("\n");
3278 pr_warn("**********************************************************\n");
3279 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3280 pr_warn("** **\n");
3281 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3282 pr_warn("** **\n");
3283 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3284 pr_warn("** unsafe for production use. **\n");
3285 pr_warn("** **\n");
3286 pr_warn("** If you see this message and you are not debugging **\n");
3287 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3288 pr_warn("** **\n");
3289 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3290 pr_warn("**********************************************************\n");
3291
3292 /* Expand the buffers to set size */
3293 tracing_update_buffers();
3294
3295 buffers_allocated = 1;
3296
3297 /*
3298 * trace_printk_init_buffers() can be called by modules.
3299 * If that happens, then we need to start cmdline recording
3300 * directly here. If the global_trace.buffer is already
3301 * allocated here, then this was called by module code.
3302 */
3303 if (global_trace.array_buffer.buffer)
3304 tracing_start_cmdline_record();
3305}
3306EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3307
3308void trace_printk_start_comm(void)
3309{
3310 /* Start tracing comms if trace printk is set */
3311 if (!buffers_allocated)
3312 return;
3313 tracing_start_cmdline_record();
3314}
3315
3316static void trace_printk_start_stop_comm(int enabled)
3317{
3318 if (!buffers_allocated)
3319 return;
3320
3321 if (enabled)
3322 tracing_start_cmdline_record();
3323 else
3324 tracing_stop_cmdline_record();
3325}
3326
3327/**
3328 * trace_vbprintk - write binary msg to tracing buffer
3329 * @ip: The address of the caller
3330 * @fmt: The string format to write to the buffer
3331 * @args: Arguments for @fmt
3332 */
3333int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3334{
3335 struct trace_event_call *call = &event_bprint;
3336 struct ring_buffer_event *event;
3337 struct trace_buffer *buffer;
3338 struct trace_array *tr = &global_trace;
3339 struct bprint_entry *entry;
3340 unsigned int trace_ctx;
3341 char *tbuffer;
3342 int len = 0, size;
3343
3344 if (unlikely(tracing_selftest_running || tracing_disabled))
3345 return 0;
3346
3347 /* Don't pollute graph traces with trace_vprintk internals */
3348 pause_graph_tracing();
3349
3350 trace_ctx = tracing_gen_ctx();
3351 preempt_disable_notrace();
3352
3353 tbuffer = get_trace_buf();
3354 if (!tbuffer) {
3355 len = 0;
3356 goto out_nobuffer;
3357 }
3358
3359 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3360
3361 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3362 goto out_put;
3363
3364 size = sizeof(*entry) + sizeof(u32) * len;
3365 buffer = tr->array_buffer.buffer;
3366 ring_buffer_nest_start(buffer);
3367 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3368 trace_ctx);
3369 if (!event)
3370 goto out;
3371 entry = ring_buffer_event_data(event);
3372 entry->ip = ip;
3373 entry->fmt = fmt;
3374
3375 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3376 if (!call_filter_check_discard(call, entry, buffer, event)) {
3377 __buffer_unlock_commit(buffer, event);
3378 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3379 }
3380
3381out:
3382 ring_buffer_nest_end(buffer);
3383out_put:
3384 put_trace_buf();
3385
3386out_nobuffer:
3387 preempt_enable_notrace();
3388 unpause_graph_tracing();
3389
3390 return len;
3391}
3392EXPORT_SYMBOL_GPL(trace_vbprintk);
3393
3394__printf(3, 0)
3395static int
3396__trace_array_vprintk(struct trace_buffer *buffer,
3397 unsigned long ip, const char *fmt, va_list args)
3398{
3399 struct trace_event_call *call = &event_print;
3400 struct ring_buffer_event *event;
3401 int len = 0, size;
3402 struct print_entry *entry;
3403 unsigned int trace_ctx;
3404 char *tbuffer;
3405
3406 if (tracing_disabled || tracing_selftest_running)
3407 return 0;
3408
3409 /* Don't pollute graph traces with trace_vprintk internals */
3410 pause_graph_tracing();
3411
3412 trace_ctx = tracing_gen_ctx();
3413 preempt_disable_notrace();
3414
3415
3416 tbuffer = get_trace_buf();
3417 if (!tbuffer) {
3418 len = 0;
3419 goto out_nobuffer;
3420 }
3421
3422 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3423
3424 size = sizeof(*entry) + len + 1;
3425 ring_buffer_nest_start(buffer);
3426 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3427 trace_ctx);
3428 if (!event)
3429 goto out;
3430 entry = ring_buffer_event_data(event);
3431 entry->ip = ip;
3432
3433 memcpy(&entry->buf, tbuffer, len + 1);
3434 if (!call_filter_check_discard(call, entry, buffer, event)) {
3435 __buffer_unlock_commit(buffer, event);
3436 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3437 }
3438
3439out:
3440 ring_buffer_nest_end(buffer);
3441 put_trace_buf();
3442
3443out_nobuffer:
3444 preempt_enable_notrace();
3445 unpause_graph_tracing();
3446
3447 return len;
3448}
3449
3450__printf(3, 0)
3451int trace_array_vprintk(struct trace_array *tr,
3452 unsigned long ip, const char *fmt, va_list args)
3453{
3454 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3455}
3456
3457/**
3458 * trace_array_printk - Print a message to a specific instance
3459 * @tr: The instance trace_array descriptor
3460 * @ip: The instruction pointer that this is called from.
3461 * @fmt: The format to print (printf format)
3462 *
3463 * If a subsystem sets up its own instance, they have the right to
3464 * printk strings into their tracing instance buffer using this
3465 * function. Note, this function will not write into the top level
3466 * buffer (use trace_printk() for that), as writing into the top level
3467 * buffer should only have events that can be individually disabled.
3468 * trace_printk() is only used for debugging a kernel, and should not
3469 * be ever incorporated in normal use.
3470 *
3471 * trace_array_printk() can be used, as it will not add noise to the
3472 * top level tracing buffer.
3473 *
3474 * Note, trace_array_init_printk() must be called on @tr before this
3475 * can be used.
3476 */
3477__printf(3, 0)
3478int trace_array_printk(struct trace_array *tr,
3479 unsigned long ip, const char *fmt, ...)
3480{
3481 int ret;
3482 va_list ap;
3483
3484 if (!tr)
3485 return -ENOENT;
3486
3487 /* This is only allowed for created instances */
3488 if (tr == &global_trace)
3489 return 0;
3490
3491 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3492 return 0;
3493
3494 va_start(ap, fmt);
3495 ret = trace_array_vprintk(tr, ip, fmt, ap);
3496 va_end(ap);
3497 return ret;
3498}
3499EXPORT_SYMBOL_GPL(trace_array_printk);
3500
3501/**
3502 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3503 * @tr: The trace array to initialize the buffers for
3504 *
3505 * As trace_array_printk() only writes into instances, they are OK to
3506 * have in the kernel (unlike trace_printk()). This needs to be called
3507 * before trace_array_printk() can be used on a trace_array.
3508 */
3509int trace_array_init_printk(struct trace_array *tr)
3510{
3511 if (!tr)
3512 return -ENOENT;
3513
3514 /* This is only allowed for created instances */
3515 if (tr == &global_trace)
3516 return -EINVAL;
3517
3518 return alloc_percpu_trace_buffer();
3519}
3520EXPORT_SYMBOL_GPL(trace_array_init_printk);
3521
3522__printf(3, 4)
3523int trace_array_printk_buf(struct trace_buffer *buffer,
3524 unsigned long ip, const char *fmt, ...)
3525{
3526 int ret;
3527 va_list ap;
3528
3529 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3530 return 0;
3531
3532 va_start(ap, fmt);
3533 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3534 va_end(ap);
3535 return ret;
3536}
3537
3538__printf(2, 0)
3539int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3540{
3541 return trace_array_vprintk(&global_trace, ip, fmt, args);
3542}
3543EXPORT_SYMBOL_GPL(trace_vprintk);
3544
3545static void trace_iterator_increment(struct trace_iterator *iter)
3546{
3547 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3548
3549 iter->idx++;
3550 if (buf_iter)
3551 ring_buffer_iter_advance(buf_iter);
3552}
3553
3554static struct trace_entry *
3555peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3556 unsigned long *lost_events)
3557{
3558 struct ring_buffer_event *event;
3559 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3560
3561 if (buf_iter) {
3562 event = ring_buffer_iter_peek(buf_iter, ts);
3563 if (lost_events)
3564 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3565 (unsigned long)-1 : 0;
3566 } else {
3567 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3568 lost_events);
3569 }
3570
3571 if (event) {
3572 iter->ent_size = ring_buffer_event_length(event);
3573 return ring_buffer_event_data(event);
3574 }
3575 iter->ent_size = 0;
3576 return NULL;
3577}
3578
3579static struct trace_entry *
3580__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3581 unsigned long *missing_events, u64 *ent_ts)
3582{
3583 struct trace_buffer *buffer = iter->array_buffer->buffer;
3584 struct trace_entry *ent, *next = NULL;
3585 unsigned long lost_events = 0, next_lost = 0;
3586 int cpu_file = iter->cpu_file;
3587 u64 next_ts = 0, ts;
3588 int next_cpu = -1;
3589 int next_size = 0;
3590 int cpu;
3591
3592 /*
3593 * If we are in a per_cpu trace file, don't bother by iterating over
3594 * all cpu and peek directly.
3595 */
3596 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3597 if (ring_buffer_empty_cpu(buffer, cpu_file))
3598 return NULL;
3599 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3600 if (ent_cpu)
3601 *ent_cpu = cpu_file;
3602
3603 return ent;
3604 }
3605
3606 for_each_tracing_cpu(cpu) {
3607
3608 if (ring_buffer_empty_cpu(buffer, cpu))
3609 continue;
3610
3611 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3612
3613 /*
3614 * Pick the entry with the smallest timestamp:
3615 */
3616 if (ent && (!next || ts < next_ts)) {
3617 next = ent;
3618 next_cpu = cpu;
3619 next_ts = ts;
3620 next_lost = lost_events;
3621 next_size = iter->ent_size;
3622 }
3623 }
3624
3625 iter->ent_size = next_size;
3626
3627 if (ent_cpu)
3628 *ent_cpu = next_cpu;
3629
3630 if (ent_ts)
3631 *ent_ts = next_ts;
3632
3633 if (missing_events)
3634 *missing_events = next_lost;
3635
3636 return next;
3637}
3638
3639#define STATIC_FMT_BUF_SIZE 128
3640static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3641
3642static char *trace_iter_expand_format(struct trace_iterator *iter)
3643{
3644 char *tmp;
3645
3646 /*
3647 * iter->tr is NULL when used with tp_printk, which makes
3648 * this get called where it is not safe to call krealloc().
3649 */
3650 if (!iter->tr || iter->fmt == static_fmt_buf)
3651 return NULL;
3652
3653 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3654 GFP_KERNEL);
3655 if (tmp) {
3656 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3657 iter->fmt = tmp;
3658 }
3659
3660 return tmp;
3661}
3662
3663/* Returns true if the string is safe to dereference from an event */
3664static bool trace_safe_str(struct trace_iterator *iter, const char *str)
3665{
3666 unsigned long addr = (unsigned long)str;
3667 struct trace_event *trace_event;
3668 struct trace_event_call *event;
3669
3670 /* OK if part of the event data */
3671 if ((addr >= (unsigned long)iter->ent) &&
3672 (addr < (unsigned long)iter->ent + iter->ent_size))
3673 return true;
3674
3675 /* OK if part of the temp seq buffer */
3676 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3677 (addr < (unsigned long)iter->tmp_seq.buffer + PAGE_SIZE))
3678 return true;
3679
3680 /* Core rodata can not be freed */
3681 if (is_kernel_rodata(addr))
3682 return true;
3683
3684 if (trace_is_tracepoint_string(str))
3685 return true;
3686
3687 /*
3688 * Now this could be a module event, referencing core module
3689 * data, which is OK.
3690 */
3691 if (!iter->ent)
3692 return false;
3693
3694 trace_event = ftrace_find_event(iter->ent->type);
3695 if (!trace_event)
3696 return false;
3697
3698 event = container_of(trace_event, struct trace_event_call, event);
3699 if (!event->mod)
3700 return false;
3701
3702 /* Would rather have rodata, but this will suffice */
3703 if (within_module_core(addr, event->mod))
3704 return true;
3705
3706 return false;
3707}
3708
3709static const char *show_buffer(struct trace_seq *s)
3710{
3711 struct seq_buf *seq = &s->seq;
3712
3713 seq_buf_terminate(seq);
3714
3715 return seq->buffer;
3716}
3717
3718static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3719
3720static int test_can_verify_check(const char *fmt, ...)
3721{
3722 char buf[16];
3723 va_list ap;
3724 int ret;
3725
3726 /*
3727 * The verifier is dependent on vsnprintf() modifies the va_list
3728 * passed to it, where it is sent as a reference. Some architectures
3729 * (like x86_32) passes it by value, which means that vsnprintf()
3730 * does not modify the va_list passed to it, and the verifier
3731 * would then need to be able to understand all the values that
3732 * vsnprintf can use. If it is passed by value, then the verifier
3733 * is disabled.
3734 */
3735 va_start(ap, fmt);
3736 vsnprintf(buf, 16, "%d", ap);
3737 ret = va_arg(ap, int);
3738 va_end(ap);
3739
3740 return ret;
3741}
3742
3743static void test_can_verify(void)
3744{
3745 if (!test_can_verify_check("%d %d", 0, 1)) {
3746 pr_info("trace event string verifier disabled\n");
3747 static_branch_inc(&trace_no_verify);
3748 }
3749}
3750
3751/**
3752 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3753 * @iter: The iterator that holds the seq buffer and the event being printed
3754 * @fmt: The format used to print the event
3755 * @ap: The va_list holding the data to print from @fmt.
3756 *
3757 * This writes the data into the @iter->seq buffer using the data from
3758 * @fmt and @ap. If the format has a %s, then the source of the string
3759 * is examined to make sure it is safe to print, otherwise it will
3760 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3761 * pointer.
3762 */
3763void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3764 va_list ap)
3765{
3766 const char *p = fmt;
3767 const char *str;
3768 int i, j;
3769
3770 if (WARN_ON_ONCE(!fmt))
3771 return;
3772
3773 if (static_branch_unlikely(&trace_no_verify))
3774 goto print;
3775
3776 /* Don't bother checking when doing a ftrace_dump() */
3777 if (iter->fmt == static_fmt_buf)
3778 goto print;
3779
3780 while (*p) {
3781 bool star = false;
3782 int len = 0;
3783
3784 j = 0;
3785
3786 /* We only care about %s and variants */
3787 for (i = 0; p[i]; i++) {
3788 if (i + 1 >= iter->fmt_size) {
3789 /*
3790 * If we can't expand the copy buffer,
3791 * just print it.
3792 */
3793 if (!trace_iter_expand_format(iter))
3794 goto print;
3795 }
3796
3797 if (p[i] == '\\' && p[i+1]) {
3798 i++;
3799 continue;
3800 }
3801 if (p[i] == '%') {
3802 /* Need to test cases like %08.*s */
3803 for (j = 1; p[i+j]; j++) {
3804 if (isdigit(p[i+j]) ||
3805 p[i+j] == '.')
3806 continue;
3807 if (p[i+j] == '*') {
3808 star = true;
3809 continue;
3810 }
3811 break;
3812 }
3813 if (p[i+j] == 's')
3814 break;
3815 star = false;
3816 }
3817 j = 0;
3818 }
3819 /* If no %s found then just print normally */
3820 if (!p[i])
3821 break;
3822
3823 /* Copy up to the %s, and print that */
3824 strncpy(iter->fmt, p, i);
3825 iter->fmt[i] = '\0';
3826 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3827
3828 if (star)
3829 len = va_arg(ap, int);
3830
3831 /* The ap now points to the string data of the %s */
3832 str = va_arg(ap, const char *);
3833
3834 /*
3835 * If you hit this warning, it is likely that the
3836 * trace event in question used %s on a string that
3837 * was saved at the time of the event, but may not be
3838 * around when the trace is read. Use __string(),
3839 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3840 * instead. See samples/trace_events/trace-events-sample.h
3841 * for reference.
3842 */
3843 if (WARN_ONCE(!trace_safe_str(iter, str),
3844 "fmt: '%s' current_buffer: '%s'",
3845 fmt, show_buffer(&iter->seq))) {
3846 int ret;
3847
3848 /* Try to safely read the string */
3849 if (star) {
3850 if (len + 1 > iter->fmt_size)
3851 len = iter->fmt_size - 1;
3852 if (len < 0)
3853 len = 0;
3854 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3855 iter->fmt[len] = 0;
3856 star = false;
3857 } else {
3858 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3859 iter->fmt_size);
3860 }
3861 if (ret < 0)
3862 trace_seq_printf(&iter->seq, "(0x%px)", str);
3863 else
3864 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3865 str, iter->fmt);
3866 str = "[UNSAFE-MEMORY]";
3867 strcpy(iter->fmt, "%s");
3868 } else {
3869 strncpy(iter->fmt, p + i, j + 1);
3870 iter->fmt[j+1] = '\0';
3871 }
3872 if (star)
3873 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3874 else
3875 trace_seq_printf(&iter->seq, iter->fmt, str);
3876
3877 p += i + j + 1;
3878 }
3879 print:
3880 if (*p)
3881 trace_seq_vprintf(&iter->seq, p, ap);
3882}
3883
3884const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3885{
3886 const char *p, *new_fmt;
3887 char *q;
3888
3889 if (WARN_ON_ONCE(!fmt))
3890 return fmt;
3891
3892 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
3893 return fmt;
3894
3895 p = fmt;
3896 new_fmt = q = iter->fmt;
3897 while (*p) {
3898 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
3899 if (!trace_iter_expand_format(iter))
3900 return fmt;
3901
3902 q += iter->fmt - new_fmt;
3903 new_fmt = iter->fmt;
3904 }
3905
3906 *q++ = *p++;
3907
3908 /* Replace %p with %px */
3909 if (p[-1] == '%') {
3910 if (p[0] == '%') {
3911 *q++ = *p++;
3912 } else if (p[0] == 'p' && !isalnum(p[1])) {
3913 *q++ = *p++;
3914 *q++ = 'x';
3915 }
3916 }
3917 }
3918 *q = '\0';
3919
3920 return new_fmt;
3921}
3922
3923#define STATIC_TEMP_BUF_SIZE 128
3924static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3925
3926/* Find the next real entry, without updating the iterator itself */
3927struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3928 int *ent_cpu, u64 *ent_ts)
3929{
3930 /* __find_next_entry will reset ent_size */
3931 int ent_size = iter->ent_size;
3932 struct trace_entry *entry;
3933
3934 /*
3935 * If called from ftrace_dump(), then the iter->temp buffer
3936 * will be the static_temp_buf and not created from kmalloc.
3937 * If the entry size is greater than the buffer, we can
3938 * not save it. Just return NULL in that case. This is only
3939 * used to add markers when two consecutive events' time
3940 * stamps have a large delta. See trace_print_lat_context()
3941 */
3942 if (iter->temp == static_temp_buf &&
3943 STATIC_TEMP_BUF_SIZE < ent_size)
3944 return NULL;
3945
3946 /*
3947 * The __find_next_entry() may call peek_next_entry(), which may
3948 * call ring_buffer_peek() that may make the contents of iter->ent
3949 * undefined. Need to copy iter->ent now.
3950 */
3951 if (iter->ent && iter->ent != iter->temp) {
3952 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3953 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3954 void *temp;
3955 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3956 if (!temp)
3957 return NULL;
3958 kfree(iter->temp);
3959 iter->temp = temp;
3960 iter->temp_size = iter->ent_size;
3961 }
3962 memcpy(iter->temp, iter->ent, iter->ent_size);
3963 iter->ent = iter->temp;
3964 }
3965 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3966 /* Put back the original ent_size */
3967 iter->ent_size = ent_size;
3968
3969 return entry;
3970}
3971
3972/* Find the next real entry, and increment the iterator to the next entry */
3973void *trace_find_next_entry_inc(struct trace_iterator *iter)
3974{
3975 iter->ent = __find_next_entry(iter, &iter->cpu,
3976 &iter->lost_events, &iter->ts);
3977
3978 if (iter->ent)
3979 trace_iterator_increment(iter);
3980
3981 return iter->ent ? iter : NULL;
3982}
3983
3984static void trace_consume(struct trace_iterator *iter)
3985{
3986 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3987 &iter->lost_events);
3988}
3989
3990static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3991{
3992 struct trace_iterator *iter = m->private;
3993 int i = (int)*pos;
3994 void *ent;
3995
3996 WARN_ON_ONCE(iter->leftover);
3997
3998 (*pos)++;
3999
4000 /* can't go backwards */
4001 if (iter->idx > i)
4002 return NULL;
4003
4004 if (iter->idx < 0)
4005 ent = trace_find_next_entry_inc(iter);
4006 else
4007 ent = iter;
4008
4009 while (ent && iter->idx < i)
4010 ent = trace_find_next_entry_inc(iter);
4011
4012 iter->pos = *pos;
4013
4014 return ent;
4015}
4016
4017void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4018{
4019 struct ring_buffer_iter *buf_iter;
4020 unsigned long entries = 0;
4021 u64 ts;
4022
4023 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4024
4025 buf_iter = trace_buffer_iter(iter, cpu);
4026 if (!buf_iter)
4027 return;
4028
4029 ring_buffer_iter_reset(buf_iter);
4030
4031 /*
4032 * We could have the case with the max latency tracers
4033 * that a reset never took place on a cpu. This is evident
4034 * by the timestamp being before the start of the buffer.
4035 */
4036 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4037 if (ts >= iter->array_buffer->time_start)
4038 break;
4039 entries++;
4040 ring_buffer_iter_advance(buf_iter);
4041 }
4042
4043 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4044}
4045
4046/*
4047 * The current tracer is copied to avoid a global locking
4048 * all around.
4049 */
4050static void *s_start(struct seq_file *m, loff_t *pos)
4051{
4052 struct trace_iterator *iter = m->private;
4053 struct trace_array *tr = iter->tr;
4054 int cpu_file = iter->cpu_file;
4055 void *p = NULL;
4056 loff_t l = 0;
4057 int cpu;
4058
4059 /*
4060 * copy the tracer to avoid using a global lock all around.
4061 * iter->trace is a copy of current_trace, the pointer to the
4062 * name may be used instead of a strcmp(), as iter->trace->name
4063 * will point to the same string as current_trace->name.
4064 */
4065 mutex_lock(&trace_types_lock);
4066 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
4067 *iter->trace = *tr->current_trace;
4068 mutex_unlock(&trace_types_lock);
4069
4070#ifdef CONFIG_TRACER_MAX_TRACE
4071 if (iter->snapshot && iter->trace->use_max_tr)
4072 return ERR_PTR(-EBUSY);
4073#endif
4074
4075 if (*pos != iter->pos) {
4076 iter->ent = NULL;
4077 iter->cpu = 0;
4078 iter->idx = -1;
4079
4080 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4081 for_each_tracing_cpu(cpu)
4082 tracing_iter_reset(iter, cpu);
4083 } else
4084 tracing_iter_reset(iter, cpu_file);
4085
4086 iter->leftover = 0;
4087 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4088 ;
4089
4090 } else {
4091 /*
4092 * If we overflowed the seq_file before, then we want
4093 * to just reuse the trace_seq buffer again.
4094 */
4095 if (iter->leftover)
4096 p = iter;
4097 else {
4098 l = *pos - 1;
4099 p = s_next(m, p, &l);
4100 }
4101 }
4102
4103 trace_event_read_lock();
4104 trace_access_lock(cpu_file);
4105 return p;
4106}
4107
4108static void s_stop(struct seq_file *m, void *p)
4109{
4110 struct trace_iterator *iter = m->private;
4111
4112#ifdef CONFIG_TRACER_MAX_TRACE
4113 if (iter->snapshot && iter->trace->use_max_tr)
4114 return;
4115#endif
4116
4117 trace_access_unlock(iter->cpu_file);
4118 trace_event_read_unlock();
4119}
4120
4121static void
4122get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4123 unsigned long *entries, int cpu)
4124{
4125 unsigned long count;
4126
4127 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4128 /*
4129 * If this buffer has skipped entries, then we hold all
4130 * entries for the trace and we need to ignore the
4131 * ones before the time stamp.
4132 */
4133 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4134 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4135 /* total is the same as the entries */
4136 *total = count;
4137 } else
4138 *total = count +
4139 ring_buffer_overrun_cpu(buf->buffer, cpu);
4140 *entries = count;
4141}
4142
4143static void
4144get_total_entries(struct array_buffer *buf,
4145 unsigned long *total, unsigned long *entries)
4146{
4147 unsigned long t, e;
4148 int cpu;
4149
4150 *total = 0;
4151 *entries = 0;
4152
4153 for_each_tracing_cpu(cpu) {
4154 get_total_entries_cpu(buf, &t, &e, cpu);
4155 *total += t;
4156 *entries += e;
4157 }
4158}
4159
4160unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4161{
4162 unsigned long total, entries;
4163
4164 if (!tr)
4165 tr = &global_trace;
4166
4167 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4168
4169 return entries;
4170}
4171
4172unsigned long trace_total_entries(struct trace_array *tr)
4173{
4174 unsigned long total, entries;
4175
4176 if (!tr)
4177 tr = &global_trace;
4178
4179 get_total_entries(&tr->array_buffer, &total, &entries);
4180
4181 return entries;
4182}
4183
4184static void print_lat_help_header(struct seq_file *m)
4185{
4186 seq_puts(m, "# _------=> CPU# \n"
4187 "# / _-----=> irqs-off \n"
4188 "# | / _----=> need-resched \n"
4189 "# || / _---=> hardirq/softirq \n"
4190 "# ||| / _--=> preempt-depth \n"
4191 "# |||| / delay \n"
4192 "# cmd pid ||||| time | caller \n"
4193 "# \\ / ||||| \\ | / \n");
4194}
4195
4196static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4197{
4198 unsigned long total;
4199 unsigned long entries;
4200
4201 get_total_entries(buf, &total, &entries);
4202 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4203 entries, total, num_online_cpus());
4204 seq_puts(m, "#\n");
4205}
4206
4207static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4208 unsigned int flags)
4209{
4210 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4211
4212 print_event_info(buf, m);
4213
4214 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4215 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4216}
4217
4218static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4219 unsigned int flags)
4220{
4221 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4222 const char *space = " ";
4223 int prec = tgid ? 12 : 2;
4224
4225 print_event_info(buf, m);
4226
4227 seq_printf(m, "# %.*s _-----=> irqs-off\n", prec, space);
4228 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4229 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4230 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4231 seq_printf(m, "# %.*s||| / delay\n", prec, space);
4232 seq_printf(m, "# TASK-PID %.*s CPU# |||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4233 seq_printf(m, "# | | %.*s | |||| | |\n", prec, " | ");
4234}
4235
4236void
4237print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4238{
4239 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4240 struct array_buffer *buf = iter->array_buffer;
4241 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4242 struct tracer *type = iter->trace;
4243 unsigned long entries;
4244 unsigned long total;
4245 const char *name = "preemption";
4246
4247 name = type->name;
4248
4249 get_total_entries(buf, &total, &entries);
4250
4251 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4252 name, UTS_RELEASE);
4253 seq_puts(m, "# -----------------------------------"
4254 "---------------------------------\n");
4255 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4256 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4257 nsecs_to_usecs(data->saved_latency),
4258 entries,
4259 total,
4260 buf->cpu,
4261#if defined(CONFIG_PREEMPT_NONE)
4262 "server",
4263#elif defined(CONFIG_PREEMPT_VOLUNTARY)
4264 "desktop",
4265#elif defined(CONFIG_PREEMPT)
4266 "preempt",
4267#elif defined(CONFIG_PREEMPT_RT)
4268 "preempt_rt",
4269#else
4270 "unknown",
4271#endif
4272 /* These are reserved for later use */
4273 0, 0, 0, 0);
4274#ifdef CONFIG_SMP
4275 seq_printf(m, " #P:%d)\n", num_online_cpus());
4276#else
4277 seq_puts(m, ")\n");
4278#endif
4279 seq_puts(m, "# -----------------\n");
4280 seq_printf(m, "# | task: %.16s-%d "
4281 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4282 data->comm, data->pid,
4283 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4284 data->policy, data->rt_priority);
4285 seq_puts(m, "# -----------------\n");
4286
4287 if (data->critical_start) {
4288 seq_puts(m, "# => started at: ");
4289 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4290 trace_print_seq(m, &iter->seq);
4291 seq_puts(m, "\n# => ended at: ");
4292 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4293 trace_print_seq(m, &iter->seq);
4294 seq_puts(m, "\n#\n");
4295 }
4296
4297 seq_puts(m, "#\n");
4298}
4299
4300static void test_cpu_buff_start(struct trace_iterator *iter)
4301{
4302 struct trace_seq *s = &iter->seq;
4303 struct trace_array *tr = iter->tr;
4304
4305 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4306 return;
4307
4308 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4309 return;
4310
4311 if (cpumask_available(iter->started) &&
4312 cpumask_test_cpu(iter->cpu, iter->started))
4313 return;
4314
4315 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4316 return;
4317
4318 if (cpumask_available(iter->started))
4319 cpumask_set_cpu(iter->cpu, iter->started);
4320
4321 /* Don't print started cpu buffer for the first entry of the trace */
4322 if (iter->idx > 1)
4323 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4324 iter->cpu);
4325}
4326
4327static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4328{
4329 struct trace_array *tr = iter->tr;
4330 struct trace_seq *s = &iter->seq;
4331 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4332 struct trace_entry *entry;
4333 struct trace_event *event;
4334
4335 entry = iter->ent;
4336
4337 test_cpu_buff_start(iter);
4338
4339 event = ftrace_find_event(entry->type);
4340
4341 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4342 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4343 trace_print_lat_context(iter);
4344 else
4345 trace_print_context(iter);
4346 }
4347
4348 if (trace_seq_has_overflowed(s))
4349 return TRACE_TYPE_PARTIAL_LINE;
4350
4351 if (event)
4352 return event->funcs->trace(iter, sym_flags, event);
4353
4354 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4355
4356 return trace_handle_return(s);
4357}
4358
4359static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4360{
4361 struct trace_array *tr = iter->tr;
4362 struct trace_seq *s = &iter->seq;
4363 struct trace_entry *entry;
4364 struct trace_event *event;
4365
4366 entry = iter->ent;
4367
4368 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4369 trace_seq_printf(s, "%d %d %llu ",
4370 entry->pid, iter->cpu, iter->ts);
4371
4372 if (trace_seq_has_overflowed(s))
4373 return TRACE_TYPE_PARTIAL_LINE;
4374
4375 event = ftrace_find_event(entry->type);
4376 if (event)
4377 return event->funcs->raw(iter, 0, event);
4378
4379 trace_seq_printf(s, "%d ?\n", entry->type);
4380
4381 return trace_handle_return(s);
4382}
4383
4384static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4385{
4386 struct trace_array *tr = iter->tr;
4387 struct trace_seq *s = &iter->seq;
4388 unsigned char newline = '\n';
4389 struct trace_entry *entry;
4390 struct trace_event *event;
4391
4392 entry = iter->ent;
4393
4394 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4395 SEQ_PUT_HEX_FIELD(s, entry->pid);
4396 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4397 SEQ_PUT_HEX_FIELD(s, iter->ts);
4398 if (trace_seq_has_overflowed(s))
4399 return TRACE_TYPE_PARTIAL_LINE;
4400 }
4401
4402 event = ftrace_find_event(entry->type);
4403 if (event) {
4404 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4405 if (ret != TRACE_TYPE_HANDLED)
4406 return ret;
4407 }
4408
4409 SEQ_PUT_FIELD(s, newline);
4410
4411 return trace_handle_return(s);
4412}
4413
4414static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4415{
4416 struct trace_array *tr = iter->tr;
4417 struct trace_seq *s = &iter->seq;
4418 struct trace_entry *entry;
4419 struct trace_event *event;
4420
4421 entry = iter->ent;
4422
4423 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4424 SEQ_PUT_FIELD(s, entry->pid);
4425 SEQ_PUT_FIELD(s, iter->cpu);
4426 SEQ_PUT_FIELD(s, iter->ts);
4427 if (trace_seq_has_overflowed(s))
4428 return TRACE_TYPE_PARTIAL_LINE;
4429 }
4430
4431 event = ftrace_find_event(entry->type);
4432 return event ? event->funcs->binary(iter, 0, event) :
4433 TRACE_TYPE_HANDLED;
4434}
4435
4436int trace_empty(struct trace_iterator *iter)
4437{
4438 struct ring_buffer_iter *buf_iter;
4439 int cpu;
4440
4441 /* If we are looking at one CPU buffer, only check that one */
4442 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4443 cpu = iter->cpu_file;
4444 buf_iter = trace_buffer_iter(iter, cpu);
4445 if (buf_iter) {
4446 if (!ring_buffer_iter_empty(buf_iter))
4447 return 0;
4448 } else {
4449 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4450 return 0;
4451 }
4452 return 1;
4453 }
4454
4455 for_each_tracing_cpu(cpu) {
4456 buf_iter = trace_buffer_iter(iter, cpu);
4457 if (buf_iter) {
4458 if (!ring_buffer_iter_empty(buf_iter))
4459 return 0;
4460 } else {
4461 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4462 return 0;
4463 }
4464 }
4465
4466 return 1;
4467}
4468
4469/* Called with trace_event_read_lock() held. */
4470enum print_line_t print_trace_line(struct trace_iterator *iter)
4471{
4472 struct trace_array *tr = iter->tr;
4473 unsigned long trace_flags = tr->trace_flags;
4474 enum print_line_t ret;
4475
4476 if (iter->lost_events) {
4477 if (iter->lost_events == (unsigned long)-1)
4478 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4479 iter->cpu);
4480 else
4481 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4482 iter->cpu, iter->lost_events);
4483 if (trace_seq_has_overflowed(&iter->seq))
4484 return TRACE_TYPE_PARTIAL_LINE;
4485 }
4486
4487 if (iter->trace && iter->trace->print_line) {
4488 ret = iter->trace->print_line(iter);
4489 if (ret != TRACE_TYPE_UNHANDLED)
4490 return ret;
4491 }
4492
4493 if (iter->ent->type == TRACE_BPUTS &&
4494 trace_flags & TRACE_ITER_PRINTK &&
4495 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4496 return trace_print_bputs_msg_only(iter);
4497
4498 if (iter->ent->type == TRACE_BPRINT &&
4499 trace_flags & TRACE_ITER_PRINTK &&
4500 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4501 return trace_print_bprintk_msg_only(iter);
4502
4503 if (iter->ent->type == TRACE_PRINT &&
4504 trace_flags & TRACE_ITER_PRINTK &&
4505 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4506 return trace_print_printk_msg_only(iter);
4507
4508 if (trace_flags & TRACE_ITER_BIN)
4509 return print_bin_fmt(iter);
4510
4511 if (trace_flags & TRACE_ITER_HEX)
4512 return print_hex_fmt(iter);
4513
4514 if (trace_flags & TRACE_ITER_RAW)
4515 return print_raw_fmt(iter);
4516
4517 return print_trace_fmt(iter);
4518}
4519
4520void trace_latency_header(struct seq_file *m)
4521{
4522 struct trace_iterator *iter = m->private;
4523 struct trace_array *tr = iter->tr;
4524
4525 /* print nothing if the buffers are empty */
4526 if (trace_empty(iter))
4527 return;
4528
4529 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4530 print_trace_header(m, iter);
4531
4532 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4533 print_lat_help_header(m);
4534}
4535
4536void trace_default_header(struct seq_file *m)
4537{
4538 struct trace_iterator *iter = m->private;
4539 struct trace_array *tr = iter->tr;
4540 unsigned long trace_flags = tr->trace_flags;
4541
4542 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4543 return;
4544
4545 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4546 /* print nothing if the buffers are empty */
4547 if (trace_empty(iter))
4548 return;
4549 print_trace_header(m, iter);
4550 if (!(trace_flags & TRACE_ITER_VERBOSE))
4551 print_lat_help_header(m);
4552 } else {
4553 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4554 if (trace_flags & TRACE_ITER_IRQ_INFO)
4555 print_func_help_header_irq(iter->array_buffer,
4556 m, trace_flags);
4557 else
4558 print_func_help_header(iter->array_buffer, m,
4559 trace_flags);
4560 }
4561 }
4562}
4563
4564static void test_ftrace_alive(struct seq_file *m)
4565{
4566 if (!ftrace_is_dead())
4567 return;
4568 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4569 "# MAY BE MISSING FUNCTION EVENTS\n");
4570}
4571
4572#ifdef CONFIG_TRACER_MAX_TRACE
4573static void show_snapshot_main_help(struct seq_file *m)
4574{
4575 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4576 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4577 "# Takes a snapshot of the main buffer.\n"
4578 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4579 "# (Doesn't have to be '2' works with any number that\n"
4580 "# is not a '0' or '1')\n");
4581}
4582
4583static void show_snapshot_percpu_help(struct seq_file *m)
4584{
4585 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4586#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4587 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4588 "# Takes a snapshot of the main buffer for this cpu.\n");
4589#else
4590 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4591 "# Must use main snapshot file to allocate.\n");
4592#endif
4593 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4594 "# (Doesn't have to be '2' works with any number that\n"
4595 "# is not a '0' or '1')\n");
4596}
4597
4598static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4599{
4600 if (iter->tr->allocated_snapshot)
4601 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4602 else
4603 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4604
4605 seq_puts(m, "# Snapshot commands:\n");
4606 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4607 show_snapshot_main_help(m);
4608 else
4609 show_snapshot_percpu_help(m);
4610}
4611#else
4612/* Should never be called */
4613static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4614#endif
4615
4616static int s_show(struct seq_file *m, void *v)
4617{
4618 struct trace_iterator *iter = v;
4619 int ret;
4620
4621 if (iter->ent == NULL) {
4622 if (iter->tr) {
4623 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4624 seq_puts(m, "#\n");
4625 test_ftrace_alive(m);
4626 }
4627 if (iter->snapshot && trace_empty(iter))
4628 print_snapshot_help(m, iter);
4629 else if (iter->trace && iter->trace->print_header)
4630 iter->trace->print_header(m);
4631 else
4632 trace_default_header(m);
4633
4634 } else if (iter->leftover) {
4635 /*
4636 * If we filled the seq_file buffer earlier, we
4637 * want to just show it now.
4638 */
4639 ret = trace_print_seq(m, &iter->seq);
4640
4641 /* ret should this time be zero, but you never know */
4642 iter->leftover = ret;
4643
4644 } else {
4645 print_trace_line(iter);
4646 ret = trace_print_seq(m, &iter->seq);
4647 /*
4648 * If we overflow the seq_file buffer, then it will
4649 * ask us for this data again at start up.
4650 * Use that instead.
4651 * ret is 0 if seq_file write succeeded.
4652 * -1 otherwise.
4653 */
4654 iter->leftover = ret;
4655 }
4656
4657 return 0;
4658}
4659
4660/*
4661 * Should be used after trace_array_get(), trace_types_lock
4662 * ensures that i_cdev was already initialized.
4663 */
4664static inline int tracing_get_cpu(struct inode *inode)
4665{
4666 if (inode->i_cdev) /* See trace_create_cpu_file() */
4667 return (long)inode->i_cdev - 1;
4668 return RING_BUFFER_ALL_CPUS;
4669}
4670
4671static const struct seq_operations tracer_seq_ops = {
4672 .start = s_start,
4673 .next = s_next,
4674 .stop = s_stop,
4675 .show = s_show,
4676};
4677
4678static struct trace_iterator *
4679__tracing_open(struct inode *inode, struct file *file, bool snapshot)
4680{
4681 struct trace_array *tr = inode->i_private;
4682 struct trace_iterator *iter;
4683 int cpu;
4684
4685 if (tracing_disabled)
4686 return ERR_PTR(-ENODEV);
4687
4688 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4689 if (!iter)
4690 return ERR_PTR(-ENOMEM);
4691
4692 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4693 GFP_KERNEL);
4694 if (!iter->buffer_iter)
4695 goto release;
4696
4697 /*
4698 * trace_find_next_entry() may need to save off iter->ent.
4699 * It will place it into the iter->temp buffer. As most
4700 * events are less than 128, allocate a buffer of that size.
4701 * If one is greater, then trace_find_next_entry() will
4702 * allocate a new buffer to adjust for the bigger iter->ent.
4703 * It's not critical if it fails to get allocated here.
4704 */
4705 iter->temp = kmalloc(128, GFP_KERNEL);
4706 if (iter->temp)
4707 iter->temp_size = 128;
4708
4709 /*
4710 * trace_event_printf() may need to modify given format
4711 * string to replace %p with %px so that it shows real address
4712 * instead of hash value. However, that is only for the event
4713 * tracing, other tracer may not need. Defer the allocation
4714 * until it is needed.
4715 */
4716 iter->fmt = NULL;
4717 iter->fmt_size = 0;
4718
4719 /*
4720 * We make a copy of the current tracer to avoid concurrent
4721 * changes on it while we are reading.
4722 */
4723 mutex_lock(&trace_types_lock);
4724 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4725 if (!iter->trace)
4726 goto fail;
4727
4728 *iter->trace = *tr->current_trace;
4729
4730 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4731 goto fail;
4732
4733 iter->tr = tr;
4734
4735#ifdef CONFIG_TRACER_MAX_TRACE
4736 /* Currently only the top directory has a snapshot */
4737 if (tr->current_trace->print_max || snapshot)
4738 iter->array_buffer = &tr->max_buffer;
4739 else
4740#endif
4741 iter->array_buffer = &tr->array_buffer;
4742 iter->snapshot = snapshot;
4743 iter->pos = -1;
4744 iter->cpu_file = tracing_get_cpu(inode);
4745 mutex_init(&iter->mutex);
4746
4747 /* Notify the tracer early; before we stop tracing. */
4748 if (iter->trace->open)
4749 iter->trace->open(iter);
4750
4751 /* Annotate start of buffers if we had overruns */
4752 if (ring_buffer_overruns(iter->array_buffer->buffer))
4753 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4754
4755 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4756 if (trace_clocks[tr->clock_id].in_ns)
4757 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4758
4759 /*
4760 * If pause-on-trace is enabled, then stop the trace while
4761 * dumping, unless this is the "snapshot" file
4762 */
4763 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4764 tracing_stop_tr(tr);
4765
4766 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4767 for_each_tracing_cpu(cpu) {
4768 iter->buffer_iter[cpu] =
4769 ring_buffer_read_prepare(iter->array_buffer->buffer,
4770 cpu, GFP_KERNEL);
4771 }
4772 ring_buffer_read_prepare_sync();
4773 for_each_tracing_cpu(cpu) {
4774 ring_buffer_read_start(iter->buffer_iter[cpu]);
4775 tracing_iter_reset(iter, cpu);
4776 }
4777 } else {
4778 cpu = iter->cpu_file;
4779 iter->buffer_iter[cpu] =
4780 ring_buffer_read_prepare(iter->array_buffer->buffer,
4781 cpu, GFP_KERNEL);
4782 ring_buffer_read_prepare_sync();
4783 ring_buffer_read_start(iter->buffer_iter[cpu]);
4784 tracing_iter_reset(iter, cpu);
4785 }
4786
4787 mutex_unlock(&trace_types_lock);
4788
4789 return iter;
4790
4791 fail:
4792 mutex_unlock(&trace_types_lock);
4793 kfree(iter->trace);
4794 kfree(iter->temp);
4795 kfree(iter->buffer_iter);
4796release:
4797 seq_release_private(inode, file);
4798 return ERR_PTR(-ENOMEM);
4799}
4800
4801int tracing_open_generic(struct inode *inode, struct file *filp)
4802{
4803 int ret;
4804
4805 ret = tracing_check_open_get_tr(NULL);
4806 if (ret)
4807 return ret;
4808
4809 filp->private_data = inode->i_private;
4810 return 0;
4811}
4812
4813bool tracing_is_disabled(void)
4814{
4815 return (tracing_disabled) ? true: false;
4816}
4817
4818/*
4819 * Open and update trace_array ref count.
4820 * Must have the current trace_array passed to it.
4821 */
4822int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4823{
4824 struct trace_array *tr = inode->i_private;
4825 int ret;
4826
4827 ret = tracing_check_open_get_tr(tr);
4828 if (ret)
4829 return ret;
4830
4831 filp->private_data = inode->i_private;
4832
4833 return 0;
4834}
4835
4836static int tracing_release(struct inode *inode, struct file *file)
4837{
4838 struct trace_array *tr = inode->i_private;
4839 struct seq_file *m = file->private_data;
4840 struct trace_iterator *iter;
4841 int cpu;
4842
4843 if (!(file->f_mode & FMODE_READ)) {
4844 trace_array_put(tr);
4845 return 0;
4846 }
4847
4848 /* Writes do not use seq_file */
4849 iter = m->private;
4850 mutex_lock(&trace_types_lock);
4851
4852 for_each_tracing_cpu(cpu) {
4853 if (iter->buffer_iter[cpu])
4854 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4855 }
4856
4857 if (iter->trace && iter->trace->close)
4858 iter->trace->close(iter);
4859
4860 if (!iter->snapshot && tr->stop_count)
4861 /* reenable tracing if it was previously enabled */
4862 tracing_start_tr(tr);
4863
4864 __trace_array_put(tr);
4865
4866 mutex_unlock(&trace_types_lock);
4867
4868 mutex_destroy(&iter->mutex);
4869 free_cpumask_var(iter->started);
4870 kfree(iter->fmt);
4871 kfree(iter->temp);
4872 kfree(iter->trace);
4873 kfree(iter->buffer_iter);
4874 seq_release_private(inode, file);
4875
4876 return 0;
4877}
4878
4879static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4880{
4881 struct trace_array *tr = inode->i_private;
4882
4883 trace_array_put(tr);
4884 return 0;
4885}
4886
4887static int tracing_single_release_tr(struct inode *inode, struct file *file)
4888{
4889 struct trace_array *tr = inode->i_private;
4890
4891 trace_array_put(tr);
4892
4893 return single_release(inode, file);
4894}
4895
4896static int tracing_open(struct inode *inode, struct file *file)
4897{
4898 struct trace_array *tr = inode->i_private;
4899 struct trace_iterator *iter;
4900 int ret;
4901
4902 ret = tracing_check_open_get_tr(tr);
4903 if (ret)
4904 return ret;
4905
4906 /* If this file was open for write, then erase contents */
4907 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4908 int cpu = tracing_get_cpu(inode);
4909 struct array_buffer *trace_buf = &tr->array_buffer;
4910
4911#ifdef CONFIG_TRACER_MAX_TRACE
4912 if (tr->current_trace->print_max)
4913 trace_buf = &tr->max_buffer;
4914#endif
4915
4916 if (cpu == RING_BUFFER_ALL_CPUS)
4917 tracing_reset_online_cpus(trace_buf);
4918 else
4919 tracing_reset_cpu(trace_buf, cpu);
4920 }
4921
4922 if (file->f_mode & FMODE_READ) {
4923 iter = __tracing_open(inode, file, false);
4924 if (IS_ERR(iter))
4925 ret = PTR_ERR(iter);
4926 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4927 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4928 }
4929
4930 if (ret < 0)
4931 trace_array_put(tr);
4932
4933 return ret;
4934}
4935
4936/*
4937 * Some tracers are not suitable for instance buffers.
4938 * A tracer is always available for the global array (toplevel)
4939 * or if it explicitly states that it is.
4940 */
4941static bool
4942trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4943{
4944 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4945}
4946
4947/* Find the next tracer that this trace array may use */
4948static struct tracer *
4949get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4950{
4951 while (t && !trace_ok_for_array(t, tr))
4952 t = t->next;
4953
4954 return t;
4955}
4956
4957static void *
4958t_next(struct seq_file *m, void *v, loff_t *pos)
4959{
4960 struct trace_array *tr = m->private;
4961 struct tracer *t = v;
4962
4963 (*pos)++;
4964
4965 if (t)
4966 t = get_tracer_for_array(tr, t->next);
4967
4968 return t;
4969}
4970
4971static void *t_start(struct seq_file *m, loff_t *pos)
4972{
4973 struct trace_array *tr = m->private;
4974 struct tracer *t;
4975 loff_t l = 0;
4976
4977 mutex_lock(&trace_types_lock);
4978
4979 t = get_tracer_for_array(tr, trace_types);
4980 for (; t && l < *pos; t = t_next(m, t, &l))
4981 ;
4982
4983 return t;
4984}
4985
4986static void t_stop(struct seq_file *m, void *p)
4987{
4988 mutex_unlock(&trace_types_lock);
4989}
4990
4991static int t_show(struct seq_file *m, void *v)
4992{
4993 struct tracer *t = v;
4994
4995 if (!t)
4996 return 0;
4997
4998 seq_puts(m, t->name);
4999 if (t->next)
5000 seq_putc(m, ' ');
5001 else
5002 seq_putc(m, '\n');
5003
5004 return 0;
5005}
5006
5007static const struct seq_operations show_traces_seq_ops = {
5008 .start = t_start,
5009 .next = t_next,
5010 .stop = t_stop,
5011 .show = t_show,
5012};
5013
5014static int show_traces_open(struct inode *inode, struct file *file)
5015{
5016 struct trace_array *tr = inode->i_private;
5017 struct seq_file *m;
5018 int ret;
5019
5020 ret = tracing_check_open_get_tr(tr);
5021 if (ret)
5022 return ret;
5023
5024 ret = seq_open(file, &show_traces_seq_ops);
5025 if (ret) {
5026 trace_array_put(tr);
5027 return ret;
5028 }
5029
5030 m = file->private_data;
5031 m->private = tr;
5032
5033 return 0;
5034}
5035
5036static int show_traces_release(struct inode *inode, struct file *file)
5037{
5038 struct trace_array *tr = inode->i_private;
5039
5040 trace_array_put(tr);
5041 return seq_release(inode, file);
5042}
5043
5044static ssize_t
5045tracing_write_stub(struct file *filp, const char __user *ubuf,
5046 size_t count, loff_t *ppos)
5047{
5048 return count;
5049}
5050
5051loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5052{
5053 int ret;
5054
5055 if (file->f_mode & FMODE_READ)
5056 ret = seq_lseek(file, offset, whence);
5057 else
5058 file->f_pos = ret = 0;
5059
5060 return ret;
5061}
5062
5063static const struct file_operations tracing_fops = {
5064 .open = tracing_open,
5065 .read = seq_read,
5066 .write = tracing_write_stub,
5067 .llseek = tracing_lseek,
5068 .release = tracing_release,
5069};
5070
5071static const struct file_operations show_traces_fops = {
5072 .open = show_traces_open,
5073 .read = seq_read,
5074 .llseek = seq_lseek,
5075 .release = show_traces_release,
5076};
5077
5078static ssize_t
5079tracing_cpumask_read(struct file *filp, char __user *ubuf,
5080 size_t count, loff_t *ppos)
5081{
5082 struct trace_array *tr = file_inode(filp)->i_private;
5083 char *mask_str;
5084 int len;
5085
5086 len = snprintf(NULL, 0, "%*pb\n",
5087 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5088 mask_str = kmalloc(len, GFP_KERNEL);
5089 if (!mask_str)
5090 return -ENOMEM;
5091
5092 len = snprintf(mask_str, len, "%*pb\n",
5093 cpumask_pr_args(tr->tracing_cpumask));
5094 if (len >= count) {
5095 count = -EINVAL;
5096 goto out_err;
5097 }
5098 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5099
5100out_err:
5101 kfree(mask_str);
5102
5103 return count;
5104}
5105
5106int tracing_set_cpumask(struct trace_array *tr,
5107 cpumask_var_t tracing_cpumask_new)
5108{
5109 int cpu;
5110
5111 if (!tr)
5112 return -EINVAL;
5113
5114 local_irq_disable();
5115 arch_spin_lock(&tr->max_lock);
5116 for_each_tracing_cpu(cpu) {
5117 /*
5118 * Increase/decrease the disabled counter if we are
5119 * about to flip a bit in the cpumask:
5120 */
5121 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5122 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5123 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5124 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5125 }
5126 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5127 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5128 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5129 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5130 }
5131 }
5132 arch_spin_unlock(&tr->max_lock);
5133 local_irq_enable();
5134
5135 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5136
5137 return 0;
5138}
5139
5140static ssize_t
5141tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5142 size_t count, loff_t *ppos)
5143{
5144 struct trace_array *tr = file_inode(filp)->i_private;
5145 cpumask_var_t tracing_cpumask_new;
5146 int err;
5147
5148 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5149 return -ENOMEM;
5150
5151 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5152 if (err)
5153 goto err_free;
5154
5155 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5156 if (err)
5157 goto err_free;
5158
5159 free_cpumask_var(tracing_cpumask_new);
5160
5161 return count;
5162
5163err_free:
5164 free_cpumask_var(tracing_cpumask_new);
5165
5166 return err;
5167}
5168
5169static const struct file_operations tracing_cpumask_fops = {
5170 .open = tracing_open_generic_tr,
5171 .read = tracing_cpumask_read,
5172 .write = tracing_cpumask_write,
5173 .release = tracing_release_generic_tr,
5174 .llseek = generic_file_llseek,
5175};
5176
5177static int tracing_trace_options_show(struct seq_file *m, void *v)
5178{
5179 struct tracer_opt *trace_opts;
5180 struct trace_array *tr = m->private;
5181 u32 tracer_flags;
5182 int i;
5183
5184 mutex_lock(&trace_types_lock);
5185 tracer_flags = tr->current_trace->flags->val;
5186 trace_opts = tr->current_trace->flags->opts;
5187
5188 for (i = 0; trace_options[i]; i++) {
5189 if (tr->trace_flags & (1 << i))
5190 seq_printf(m, "%s\n", trace_options[i]);
5191 else
5192 seq_printf(m, "no%s\n", trace_options[i]);
5193 }
5194
5195 for (i = 0; trace_opts[i].name; i++) {
5196 if (tracer_flags & trace_opts[i].bit)
5197 seq_printf(m, "%s\n", trace_opts[i].name);
5198 else
5199 seq_printf(m, "no%s\n", trace_opts[i].name);
5200 }
5201 mutex_unlock(&trace_types_lock);
5202
5203 return 0;
5204}
5205
5206static int __set_tracer_option(struct trace_array *tr,
5207 struct tracer_flags *tracer_flags,
5208 struct tracer_opt *opts, int neg)
5209{
5210 struct tracer *trace = tracer_flags->trace;
5211 int ret;
5212
5213 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5214 if (ret)
5215 return ret;
5216
5217 if (neg)
5218 tracer_flags->val &= ~opts->bit;
5219 else
5220 tracer_flags->val |= opts->bit;
5221 return 0;
5222}
5223
5224/* Try to assign a tracer specific option */
5225static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5226{
5227 struct tracer *trace = tr->current_trace;
5228 struct tracer_flags *tracer_flags = trace->flags;
5229 struct tracer_opt *opts = NULL;
5230 int i;
5231
5232 for (i = 0; tracer_flags->opts[i].name; i++) {
5233 opts = &tracer_flags->opts[i];
5234
5235 if (strcmp(cmp, opts->name) == 0)
5236 return __set_tracer_option(tr, trace->flags, opts, neg);
5237 }
5238
5239 return -EINVAL;
5240}
5241
5242/* Some tracers require overwrite to stay enabled */
5243int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5244{
5245 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5246 return -1;
5247
5248 return 0;
5249}
5250
5251int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5252{
5253 int *map;
5254
5255 if ((mask == TRACE_ITER_RECORD_TGID) ||
5256 (mask == TRACE_ITER_RECORD_CMD))
5257 lockdep_assert_held(&event_mutex);
5258
5259 /* do nothing if flag is already set */
5260 if (!!(tr->trace_flags & mask) == !!enabled)
5261 return 0;
5262
5263 /* Give the tracer a chance to approve the change */
5264 if (tr->current_trace->flag_changed)
5265 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5266 return -EINVAL;
5267
5268 if (enabled)
5269 tr->trace_flags |= mask;
5270 else
5271 tr->trace_flags &= ~mask;
5272
5273 if (mask == TRACE_ITER_RECORD_CMD)
5274 trace_event_enable_cmd_record(enabled);
5275
5276 if (mask == TRACE_ITER_RECORD_TGID) {
5277 if (!tgid_map) {
5278 tgid_map_max = pid_max;
5279 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5280 GFP_KERNEL);
5281
5282 /*
5283 * Pairs with smp_load_acquire() in
5284 * trace_find_tgid_ptr() to ensure that if it observes
5285 * the tgid_map we just allocated then it also observes
5286 * the corresponding tgid_map_max value.
5287 */
5288 smp_store_release(&tgid_map, map);
5289 }
5290 if (!tgid_map) {
5291 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5292 return -ENOMEM;
5293 }
5294
5295 trace_event_enable_tgid_record(enabled);
5296 }
5297
5298 if (mask == TRACE_ITER_EVENT_FORK)
5299 trace_event_follow_fork(tr, enabled);
5300
5301 if (mask == TRACE_ITER_FUNC_FORK)
5302 ftrace_pid_follow_fork(tr, enabled);
5303
5304 if (mask == TRACE_ITER_OVERWRITE) {
5305 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5306#ifdef CONFIG_TRACER_MAX_TRACE
5307 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5308#endif
5309 }
5310
5311 if (mask == TRACE_ITER_PRINTK) {
5312 trace_printk_start_stop_comm(enabled);
5313 trace_printk_control(enabled);
5314 }
5315
5316 return 0;
5317}
5318
5319int trace_set_options(struct trace_array *tr, char *option)
5320{
5321 char *cmp;
5322 int neg = 0;
5323 int ret;
5324 size_t orig_len = strlen(option);
5325 int len;
5326
5327 cmp = strstrip(option);
5328
5329 len = str_has_prefix(cmp, "no");
5330 if (len)
5331 neg = 1;
5332
5333 cmp += len;
5334
5335 mutex_lock(&event_mutex);
5336 mutex_lock(&trace_types_lock);
5337
5338 ret = match_string(trace_options, -1, cmp);
5339 /* If no option could be set, test the specific tracer options */
5340 if (ret < 0)
5341 ret = set_tracer_option(tr, cmp, neg);
5342 else
5343 ret = set_tracer_flag(tr, 1 << ret, !neg);
5344
5345 mutex_unlock(&trace_types_lock);
5346 mutex_unlock(&event_mutex);
5347
5348 /*
5349 * If the first trailing whitespace is replaced with '\0' by strstrip,
5350 * turn it back into a space.
5351 */
5352 if (orig_len > strlen(option))
5353 option[strlen(option)] = ' ';
5354
5355 return ret;
5356}
5357
5358static void __init apply_trace_boot_options(void)
5359{
5360 char *buf = trace_boot_options_buf;
5361 char *option;
5362
5363 while (true) {
5364 option = strsep(&buf, ",");
5365
5366 if (!option)
5367 break;
5368
5369 if (*option)
5370 trace_set_options(&global_trace, option);
5371
5372 /* Put back the comma to allow this to be called again */
5373 if (buf)
5374 *(buf - 1) = ',';
5375 }
5376}
5377
5378static ssize_t
5379tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5380 size_t cnt, loff_t *ppos)
5381{
5382 struct seq_file *m = filp->private_data;
5383 struct trace_array *tr = m->private;
5384 char buf[64];
5385 int ret;
5386
5387 if (cnt >= sizeof(buf))
5388 return -EINVAL;
5389
5390 if (copy_from_user(buf, ubuf, cnt))
5391 return -EFAULT;
5392
5393 buf[cnt] = 0;
5394
5395 ret = trace_set_options(tr, buf);
5396 if (ret < 0)
5397 return ret;
5398
5399 *ppos += cnt;
5400
5401 return cnt;
5402}
5403
5404static int tracing_trace_options_open(struct inode *inode, struct file *file)
5405{
5406 struct trace_array *tr = inode->i_private;
5407 int ret;
5408
5409 ret = tracing_check_open_get_tr(tr);
5410 if (ret)
5411 return ret;
5412
5413 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5414 if (ret < 0)
5415 trace_array_put(tr);
5416
5417 return ret;
5418}
5419
5420static const struct file_operations tracing_iter_fops = {
5421 .open = tracing_trace_options_open,
5422 .read = seq_read,
5423 .llseek = seq_lseek,
5424 .release = tracing_single_release_tr,
5425 .write = tracing_trace_options_write,
5426};
5427
5428static const char readme_msg[] =
5429 "tracing mini-HOWTO:\n\n"
5430 "# echo 0 > tracing_on : quick way to disable tracing\n"
5431 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5432 " Important files:\n"
5433 " trace\t\t\t- The static contents of the buffer\n"
5434 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5435 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5436 " current_tracer\t- function and latency tracers\n"
5437 " available_tracers\t- list of configured tracers for current_tracer\n"
5438 " error_log\t- error log for failed commands (that support it)\n"
5439 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5440 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5441 " trace_clock\t\t-change the clock used to order events\n"
5442 " local: Per cpu clock but may not be synced across CPUs\n"
5443 " global: Synced across CPUs but slows tracing down.\n"
5444 " counter: Not a clock, but just an increment\n"
5445 " uptime: Jiffy counter from time of boot\n"
5446 " perf: Same clock that perf events use\n"
5447#ifdef CONFIG_X86_64
5448 " x86-tsc: TSC cycle counter\n"
5449#endif
5450 "\n timestamp_mode\t-view the mode used to timestamp events\n"
5451 " delta: Delta difference against a buffer-wide timestamp\n"
5452 " absolute: Absolute (standalone) timestamp\n"
5453 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5454 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5455 " tracing_cpumask\t- Limit which CPUs to trace\n"
5456 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5457 "\t\t\t Remove sub-buffer with rmdir\n"
5458 " trace_options\t\t- Set format or modify how tracing happens\n"
5459 "\t\t\t Disable an option by prefixing 'no' to the\n"
5460 "\t\t\t option name\n"
5461 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5462#ifdef CONFIG_DYNAMIC_FTRACE
5463 "\n available_filter_functions - list of functions that can be filtered on\n"
5464 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5465 "\t\t\t functions\n"
5466 "\t accepts: func_full_name or glob-matching-pattern\n"
5467 "\t modules: Can select a group via module\n"
5468 "\t Format: :mod:<module-name>\n"
5469 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5470 "\t triggers: a command to perform when function is hit\n"
5471 "\t Format: <function>:<trigger>[:count]\n"
5472 "\t trigger: traceon, traceoff\n"
5473 "\t\t enable_event:<system>:<event>\n"
5474 "\t\t disable_event:<system>:<event>\n"
5475#ifdef CONFIG_STACKTRACE
5476 "\t\t stacktrace\n"
5477#endif
5478#ifdef CONFIG_TRACER_SNAPSHOT
5479 "\t\t snapshot\n"
5480#endif
5481 "\t\t dump\n"
5482 "\t\t cpudump\n"
5483 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5484 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5485 "\t The first one will disable tracing every time do_fault is hit\n"
5486 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5487 "\t The first time do trap is hit and it disables tracing, the\n"
5488 "\t counter will decrement to 2. If tracing is already disabled,\n"
5489 "\t the counter will not decrement. It only decrements when the\n"
5490 "\t trigger did work\n"
5491 "\t To remove trigger without count:\n"
5492 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5493 "\t To remove trigger with a count:\n"
5494 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5495 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5496 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5497 "\t modules: Can select a group via module command :mod:\n"
5498 "\t Does not accept triggers\n"
5499#endif /* CONFIG_DYNAMIC_FTRACE */
5500#ifdef CONFIG_FUNCTION_TRACER
5501 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5502 "\t\t (function)\n"
5503 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5504 "\t\t (function)\n"
5505#endif
5506#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5507 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5508 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5509 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5510#endif
5511#ifdef CONFIG_TRACER_SNAPSHOT
5512 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5513 "\t\t\t snapshot buffer. Read the contents for more\n"
5514 "\t\t\t information\n"
5515#endif
5516#ifdef CONFIG_STACK_TRACER
5517 " stack_trace\t\t- Shows the max stack trace when active\n"
5518 " stack_max_size\t- Shows current max stack size that was traced\n"
5519 "\t\t\t Write into this file to reset the max size (trigger a\n"
5520 "\t\t\t new trace)\n"
5521#ifdef CONFIG_DYNAMIC_FTRACE
5522 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5523 "\t\t\t traces\n"
5524#endif
5525#endif /* CONFIG_STACK_TRACER */
5526#ifdef CONFIG_DYNAMIC_EVENTS
5527 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5528 "\t\t\t Write into this file to define/undefine new trace events.\n"
5529#endif
5530#ifdef CONFIG_KPROBE_EVENTS
5531 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5532 "\t\t\t Write into this file to define/undefine new trace events.\n"
5533#endif
5534#ifdef CONFIG_UPROBE_EVENTS
5535 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5536 "\t\t\t Write into this file to define/undefine new trace events.\n"
5537#endif
5538#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5539 "\t accepts: event-definitions (one definition per line)\n"
5540 "\t Format: p[:[<group>/]<event>] <place> [<args>]\n"
5541 "\t r[maxactive][:[<group>/]<event>] <place> [<args>]\n"
5542#ifdef CONFIG_HIST_TRIGGERS
5543 "\t s:[synthetic/]<event> <field> [<field>]\n"
5544#endif
5545 "\t -:[<group>/]<event>\n"
5546#ifdef CONFIG_KPROBE_EVENTS
5547 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5548 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5549#endif
5550#ifdef CONFIG_UPROBE_EVENTS
5551 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5552#endif
5553 "\t args: <name>=fetcharg[:type]\n"
5554 "\t fetcharg: %<register>, @<address>, @<symbol>[+|-<offset>],\n"
5555#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5556 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5557#else
5558 "\t $stack<index>, $stack, $retval, $comm,\n"
5559#endif
5560 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5561 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string, symbol,\n"
5562 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5563 "\t <type>\\[<array-size>\\]\n"
5564#ifdef CONFIG_HIST_TRIGGERS
5565 "\t field: <stype> <name>;\n"
5566 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5567 "\t [unsigned] char/int/long\n"
5568#endif
5569#endif
5570 " events/\t\t- Directory containing all trace event subsystems:\n"
5571 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5572 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5573 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5574 "\t\t\t events\n"
5575 " filter\t\t- If set, only events passing filter are traced\n"
5576 " events/<system>/<event>/\t- Directory containing control files for\n"
5577 "\t\t\t <event>:\n"
5578 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5579 " filter\t\t- If set, only events passing filter are traced\n"
5580 " trigger\t\t- If set, a command to perform when event is hit\n"
5581 "\t Format: <trigger>[:count][if <filter>]\n"
5582 "\t trigger: traceon, traceoff\n"
5583 "\t enable_event:<system>:<event>\n"
5584 "\t disable_event:<system>:<event>\n"
5585#ifdef CONFIG_HIST_TRIGGERS
5586 "\t enable_hist:<system>:<event>\n"
5587 "\t disable_hist:<system>:<event>\n"
5588#endif
5589#ifdef CONFIG_STACKTRACE
5590 "\t\t stacktrace\n"
5591#endif
5592#ifdef CONFIG_TRACER_SNAPSHOT
5593 "\t\t snapshot\n"
5594#endif
5595#ifdef CONFIG_HIST_TRIGGERS
5596 "\t\t hist (see below)\n"
5597#endif
5598 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5599 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5600 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5601 "\t events/block/block_unplug/trigger\n"
5602 "\t The first disables tracing every time block_unplug is hit.\n"
5603 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5604 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5605 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5606 "\t Like function triggers, the counter is only decremented if it\n"
5607 "\t enabled or disabled tracing.\n"
5608 "\t To remove a trigger without a count:\n"
5609 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5610 "\t To remove a trigger with a count:\n"
5611 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5612 "\t Filters can be ignored when removing a trigger.\n"
5613#ifdef CONFIG_HIST_TRIGGERS
5614 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5615 "\t Format: hist:keys=<field1[,field2,...]>\n"
5616 "\t [:values=<field1[,field2,...]>]\n"
5617 "\t [:sort=<field1[,field2,...]>]\n"
5618 "\t [:size=#entries]\n"
5619 "\t [:pause][:continue][:clear]\n"
5620 "\t [:name=histname1]\n"
5621 "\t [:<handler>.<action>]\n"
5622 "\t [if <filter>]\n\n"
5623 "\t Note, special fields can be used as well:\n"
5624 "\t common_timestamp - to record current timestamp\n"
5625 "\t common_cpu - to record the CPU the event happened on\n"
5626 "\n"
5627 "\t When a matching event is hit, an entry is added to a hash\n"
5628 "\t table using the key(s) and value(s) named, and the value of a\n"
5629 "\t sum called 'hitcount' is incremented. Keys and values\n"
5630 "\t correspond to fields in the event's format description. Keys\n"
5631 "\t can be any field, or the special string 'stacktrace'.\n"
5632 "\t Compound keys consisting of up to two fields can be specified\n"
5633 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5634 "\t fields. Sort keys consisting of up to two fields can be\n"
5635 "\t specified using the 'sort' keyword. The sort direction can\n"
5636 "\t be modified by appending '.descending' or '.ascending' to a\n"
5637 "\t sort field. The 'size' parameter can be used to specify more\n"
5638 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5639 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5640 "\t its histogram data will be shared with other triggers of the\n"
5641 "\t same name, and trigger hits will update this common data.\n\n"
5642 "\t Reading the 'hist' file for the event will dump the hash\n"
5643 "\t table in its entirety to stdout. If there are multiple hist\n"
5644 "\t triggers attached to an event, there will be a table for each\n"
5645 "\t trigger in the output. The table displayed for a named\n"
5646 "\t trigger will be the same as any other instance having the\n"
5647 "\t same name. The default format used to display a given field\n"
5648 "\t can be modified by appending any of the following modifiers\n"
5649 "\t to the field name, as applicable:\n\n"
5650 "\t .hex display a number as a hex value\n"
5651 "\t .sym display an address as a symbol\n"
5652 "\t .sym-offset display an address as a symbol and offset\n"
5653 "\t .execname display a common_pid as a program name\n"
5654 "\t .syscall display a syscall id as a syscall name\n"
5655 "\t .log2 display log2 value rather than raw number\n"
5656 "\t .usecs display a common_timestamp in microseconds\n\n"
5657 "\t The 'pause' parameter can be used to pause an existing hist\n"
5658 "\t trigger or to start a hist trigger but not log any events\n"
5659 "\t until told to do so. 'continue' can be used to start or\n"
5660 "\t restart a paused hist trigger.\n\n"
5661 "\t The 'clear' parameter will clear the contents of a running\n"
5662 "\t hist trigger and leave its current paused/active state\n"
5663 "\t unchanged.\n\n"
5664 "\t The enable_hist and disable_hist triggers can be used to\n"
5665 "\t have one event conditionally start and stop another event's\n"
5666 "\t already-attached hist trigger. The syntax is analogous to\n"
5667 "\t the enable_event and disable_event triggers.\n\n"
5668 "\t Hist trigger handlers and actions are executed whenever a\n"
5669 "\t a histogram entry is added or updated. They take the form:\n\n"
5670 "\t <handler>.<action>\n\n"
5671 "\t The available handlers are:\n\n"
5672 "\t onmatch(matching.event) - invoke on addition or update\n"
5673 "\t onmax(var) - invoke if var exceeds current max\n"
5674 "\t onchange(var) - invoke action if var changes\n\n"
5675 "\t The available actions are:\n\n"
5676 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5677 "\t save(field,...) - save current event fields\n"
5678#ifdef CONFIG_TRACER_SNAPSHOT
5679 "\t snapshot() - snapshot the trace buffer\n\n"
5680#endif
5681#ifdef CONFIG_SYNTH_EVENTS
5682 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5683 "\t Write into this file to define/undefine new synthetic events.\n"
5684 "\t example: echo 'myevent u64 lat; char name[]' >> synthetic_events\n"
5685#endif
5686#endif
5687;
5688
5689static ssize_t
5690tracing_readme_read(struct file *filp, char __user *ubuf,
5691 size_t cnt, loff_t *ppos)
5692{
5693 return simple_read_from_buffer(ubuf, cnt, ppos,
5694 readme_msg, strlen(readme_msg));
5695}
5696
5697static const struct file_operations tracing_readme_fops = {
5698 .open = tracing_open_generic,
5699 .read = tracing_readme_read,
5700 .llseek = generic_file_llseek,
5701};
5702
5703static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5704{
5705 int pid = ++(*pos);
5706
5707 return trace_find_tgid_ptr(pid);
5708}
5709
5710static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5711{
5712 int pid = *pos;
5713
5714 return trace_find_tgid_ptr(pid);
5715}
5716
5717static void saved_tgids_stop(struct seq_file *m, void *v)
5718{
5719}
5720
5721static int saved_tgids_show(struct seq_file *m, void *v)
5722{
5723 int *entry = (int *)v;
5724 int pid = entry - tgid_map;
5725 int tgid = *entry;
5726
5727 if (tgid == 0)
5728 return SEQ_SKIP;
5729
5730 seq_printf(m, "%d %d\n", pid, tgid);
5731 return 0;
5732}
5733
5734static const struct seq_operations tracing_saved_tgids_seq_ops = {
5735 .start = saved_tgids_start,
5736 .stop = saved_tgids_stop,
5737 .next = saved_tgids_next,
5738 .show = saved_tgids_show,
5739};
5740
5741static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5742{
5743 int ret;
5744
5745 ret = tracing_check_open_get_tr(NULL);
5746 if (ret)
5747 return ret;
5748
5749 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5750}
5751
5752
5753static const struct file_operations tracing_saved_tgids_fops = {
5754 .open = tracing_saved_tgids_open,
5755 .read = seq_read,
5756 .llseek = seq_lseek,
5757 .release = seq_release,
5758};
5759
5760static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5761{
5762 unsigned int *ptr = v;
5763
5764 if (*pos || m->count)
5765 ptr++;
5766
5767 (*pos)++;
5768
5769 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5770 ptr++) {
5771 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5772 continue;
5773
5774 return ptr;
5775 }
5776
5777 return NULL;
5778}
5779
5780static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5781{
5782 void *v;
5783 loff_t l = 0;
5784
5785 preempt_disable();
5786 arch_spin_lock(&trace_cmdline_lock);
5787
5788 v = &savedcmd->map_cmdline_to_pid[0];
5789 while (l <= *pos) {
5790 v = saved_cmdlines_next(m, v, &l);
5791 if (!v)
5792 return NULL;
5793 }
5794
5795 return v;
5796}
5797
5798static void saved_cmdlines_stop(struct seq_file *m, void *v)
5799{
5800 arch_spin_unlock(&trace_cmdline_lock);
5801 preempt_enable();
5802}
5803
5804static int saved_cmdlines_show(struct seq_file *m, void *v)
5805{
5806 char buf[TASK_COMM_LEN];
5807 unsigned int *pid = v;
5808
5809 __trace_find_cmdline(*pid, buf);
5810 seq_printf(m, "%d %s\n", *pid, buf);
5811 return 0;
5812}
5813
5814static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5815 .start = saved_cmdlines_start,
5816 .next = saved_cmdlines_next,
5817 .stop = saved_cmdlines_stop,
5818 .show = saved_cmdlines_show,
5819};
5820
5821static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5822{
5823 int ret;
5824
5825 ret = tracing_check_open_get_tr(NULL);
5826 if (ret)
5827 return ret;
5828
5829 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5830}
5831
5832static const struct file_operations tracing_saved_cmdlines_fops = {
5833 .open = tracing_saved_cmdlines_open,
5834 .read = seq_read,
5835 .llseek = seq_lseek,
5836 .release = seq_release,
5837};
5838
5839static ssize_t
5840tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5841 size_t cnt, loff_t *ppos)
5842{
5843 char buf[64];
5844 int r;
5845
5846 arch_spin_lock(&trace_cmdline_lock);
5847 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5848 arch_spin_unlock(&trace_cmdline_lock);
5849
5850 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5851}
5852
5853static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5854{
5855 kfree(s->saved_cmdlines);
5856 kfree(s->map_cmdline_to_pid);
5857 kfree(s);
5858}
5859
5860static int tracing_resize_saved_cmdlines(unsigned int val)
5861{
5862 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5863
5864 s = kmalloc(sizeof(*s), GFP_KERNEL);
5865 if (!s)
5866 return -ENOMEM;
5867
5868 if (allocate_cmdlines_buffer(val, s) < 0) {
5869 kfree(s);
5870 return -ENOMEM;
5871 }
5872
5873 arch_spin_lock(&trace_cmdline_lock);
5874 savedcmd_temp = savedcmd;
5875 savedcmd = s;
5876 arch_spin_unlock(&trace_cmdline_lock);
5877 free_saved_cmdlines_buffer(savedcmd_temp);
5878
5879 return 0;
5880}
5881
5882static ssize_t
5883tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
5884 size_t cnt, loff_t *ppos)
5885{
5886 unsigned long val;
5887 int ret;
5888
5889 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5890 if (ret)
5891 return ret;
5892
5893 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
5894 if (!val || val > PID_MAX_DEFAULT)
5895 return -EINVAL;
5896
5897 ret = tracing_resize_saved_cmdlines((unsigned int)val);
5898 if (ret < 0)
5899 return ret;
5900
5901 *ppos += cnt;
5902
5903 return cnt;
5904}
5905
5906static const struct file_operations tracing_saved_cmdlines_size_fops = {
5907 .open = tracing_open_generic,
5908 .read = tracing_saved_cmdlines_size_read,
5909 .write = tracing_saved_cmdlines_size_write,
5910};
5911
5912#ifdef CONFIG_TRACE_EVAL_MAP_FILE
5913static union trace_eval_map_item *
5914update_eval_map(union trace_eval_map_item *ptr)
5915{
5916 if (!ptr->map.eval_string) {
5917 if (ptr->tail.next) {
5918 ptr = ptr->tail.next;
5919 /* Set ptr to the next real item (skip head) */
5920 ptr++;
5921 } else
5922 return NULL;
5923 }
5924 return ptr;
5925}
5926
5927static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5928{
5929 union trace_eval_map_item *ptr = v;
5930
5931 /*
5932 * Paranoid! If ptr points to end, we don't want to increment past it.
5933 * This really should never happen.
5934 */
5935 (*pos)++;
5936 ptr = update_eval_map(ptr);
5937 if (WARN_ON_ONCE(!ptr))
5938 return NULL;
5939
5940 ptr++;
5941 ptr = update_eval_map(ptr);
5942
5943 return ptr;
5944}
5945
5946static void *eval_map_start(struct seq_file *m, loff_t *pos)
5947{
5948 union trace_eval_map_item *v;
5949 loff_t l = 0;
5950
5951 mutex_lock(&trace_eval_mutex);
5952
5953 v = trace_eval_maps;
5954 if (v)
5955 v++;
5956
5957 while (v && l < *pos) {
5958 v = eval_map_next(m, v, &l);
5959 }
5960
5961 return v;
5962}
5963
5964static void eval_map_stop(struct seq_file *m, void *v)
5965{
5966 mutex_unlock(&trace_eval_mutex);
5967}
5968
5969static int eval_map_show(struct seq_file *m, void *v)
5970{
5971 union trace_eval_map_item *ptr = v;
5972
5973 seq_printf(m, "%s %ld (%s)\n",
5974 ptr->map.eval_string, ptr->map.eval_value,
5975 ptr->map.system);
5976
5977 return 0;
5978}
5979
5980static const struct seq_operations tracing_eval_map_seq_ops = {
5981 .start = eval_map_start,
5982 .next = eval_map_next,
5983 .stop = eval_map_stop,
5984 .show = eval_map_show,
5985};
5986
5987static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5988{
5989 int ret;
5990
5991 ret = tracing_check_open_get_tr(NULL);
5992 if (ret)
5993 return ret;
5994
5995 return seq_open(filp, &tracing_eval_map_seq_ops);
5996}
5997
5998static const struct file_operations tracing_eval_map_fops = {
5999 .open = tracing_eval_map_open,
6000 .read = seq_read,
6001 .llseek = seq_lseek,
6002 .release = seq_release,
6003};
6004
6005static inline union trace_eval_map_item *
6006trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6007{
6008 /* Return tail of array given the head */
6009 return ptr + ptr->head.length + 1;
6010}
6011
6012static void
6013trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6014 int len)
6015{
6016 struct trace_eval_map **stop;
6017 struct trace_eval_map **map;
6018 union trace_eval_map_item *map_array;
6019 union trace_eval_map_item *ptr;
6020
6021 stop = start + len;
6022
6023 /*
6024 * The trace_eval_maps contains the map plus a head and tail item,
6025 * where the head holds the module and length of array, and the
6026 * tail holds a pointer to the next list.
6027 */
6028 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6029 if (!map_array) {
6030 pr_warn("Unable to allocate trace eval mapping\n");
6031 return;
6032 }
6033
6034 mutex_lock(&trace_eval_mutex);
6035
6036 if (!trace_eval_maps)
6037 trace_eval_maps = map_array;
6038 else {
6039 ptr = trace_eval_maps;
6040 for (;;) {
6041 ptr = trace_eval_jmp_to_tail(ptr);
6042 if (!ptr->tail.next)
6043 break;
6044 ptr = ptr->tail.next;
6045
6046 }
6047 ptr->tail.next = map_array;
6048 }
6049 map_array->head.mod = mod;
6050 map_array->head.length = len;
6051 map_array++;
6052
6053 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6054 map_array->map = **map;
6055 map_array++;
6056 }
6057 memset(map_array, 0, sizeof(*map_array));
6058
6059 mutex_unlock(&trace_eval_mutex);
6060}
6061
6062static void trace_create_eval_file(struct dentry *d_tracer)
6063{
6064 trace_create_file("eval_map", 0444, d_tracer,
6065 NULL, &tracing_eval_map_fops);
6066}
6067
6068#else /* CONFIG_TRACE_EVAL_MAP_FILE */
6069static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6070static inline void trace_insert_eval_map_file(struct module *mod,
6071 struct trace_eval_map **start, int len) { }
6072#endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6073
6074static void trace_insert_eval_map(struct module *mod,
6075 struct trace_eval_map **start, int len)
6076{
6077 struct trace_eval_map **map;
6078
6079 if (len <= 0)
6080 return;
6081
6082 map = start;
6083
6084 trace_event_eval_update(map, len);
6085
6086 trace_insert_eval_map_file(mod, start, len);
6087}
6088
6089static ssize_t
6090tracing_set_trace_read(struct file *filp, char __user *ubuf,
6091 size_t cnt, loff_t *ppos)
6092{
6093 struct trace_array *tr = filp->private_data;
6094 char buf[MAX_TRACER_SIZE+2];
6095 int r;
6096
6097 mutex_lock(&trace_types_lock);
6098 r = sprintf(buf, "%s\n", tr->current_trace->name);
6099 mutex_unlock(&trace_types_lock);
6100
6101 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6102}
6103
6104int tracer_init(struct tracer *t, struct trace_array *tr)
6105{
6106 tracing_reset_online_cpus(&tr->array_buffer);
6107 return t->init(tr);
6108}
6109
6110static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6111{
6112 int cpu;
6113
6114 for_each_tracing_cpu(cpu)
6115 per_cpu_ptr(buf->data, cpu)->entries = val;
6116}
6117
6118#ifdef CONFIG_TRACER_MAX_TRACE
6119/* resize @tr's buffer to the size of @size_tr's entries */
6120static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6121 struct array_buffer *size_buf, int cpu_id)
6122{
6123 int cpu, ret = 0;
6124
6125 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6126 for_each_tracing_cpu(cpu) {
6127 ret = ring_buffer_resize(trace_buf->buffer,
6128 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6129 if (ret < 0)
6130 break;
6131 per_cpu_ptr(trace_buf->data, cpu)->entries =
6132 per_cpu_ptr(size_buf->data, cpu)->entries;
6133 }
6134 } else {
6135 ret = ring_buffer_resize(trace_buf->buffer,
6136 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6137 if (ret == 0)
6138 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6139 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6140 }
6141
6142 return ret;
6143}
6144#endif /* CONFIG_TRACER_MAX_TRACE */
6145
6146static int __tracing_resize_ring_buffer(struct trace_array *tr,
6147 unsigned long size, int cpu)
6148{
6149 int ret;
6150
6151 /*
6152 * If kernel or user changes the size of the ring buffer
6153 * we use the size that was given, and we can forget about
6154 * expanding it later.
6155 */
6156 ring_buffer_expanded = true;
6157
6158 /* May be called before buffers are initialized */
6159 if (!tr->array_buffer.buffer)
6160 return 0;
6161
6162 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6163 if (ret < 0)
6164 return ret;
6165
6166#ifdef CONFIG_TRACER_MAX_TRACE
6167 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
6168 !tr->current_trace->use_max_tr)
6169 goto out;
6170
6171 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6172 if (ret < 0) {
6173 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6174 &tr->array_buffer, cpu);
6175 if (r < 0) {
6176 /*
6177 * AARGH! We are left with different
6178 * size max buffer!!!!
6179 * The max buffer is our "snapshot" buffer.
6180 * When a tracer needs a snapshot (one of the
6181 * latency tracers), it swaps the max buffer
6182 * with the saved snap shot. We succeeded to
6183 * update the size of the main buffer, but failed to
6184 * update the size of the max buffer. But when we tried
6185 * to reset the main buffer to the original size, we
6186 * failed there too. This is very unlikely to
6187 * happen, but if it does, warn and kill all
6188 * tracing.
6189 */
6190 WARN_ON(1);
6191 tracing_disabled = 1;
6192 }
6193 return ret;
6194 }
6195
6196 if (cpu == RING_BUFFER_ALL_CPUS)
6197 set_buffer_entries(&tr->max_buffer, size);
6198 else
6199 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
6200
6201 out:
6202#endif /* CONFIG_TRACER_MAX_TRACE */
6203
6204 if (cpu == RING_BUFFER_ALL_CPUS)
6205 set_buffer_entries(&tr->array_buffer, size);
6206 else
6207 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
6208
6209 return ret;
6210}
6211
6212ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6213 unsigned long size, int cpu_id)
6214{
6215 int ret;
6216
6217 mutex_lock(&trace_types_lock);
6218
6219 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6220 /* make sure, this cpu is enabled in the mask */
6221 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6222 ret = -EINVAL;
6223 goto out;
6224 }
6225 }
6226
6227 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6228 if (ret < 0)
6229 ret = -ENOMEM;
6230
6231out:
6232 mutex_unlock(&trace_types_lock);
6233
6234 return ret;
6235}
6236
6237
6238/**
6239 * tracing_update_buffers - used by tracing facility to expand ring buffers
6240 *
6241 * To save on memory when the tracing is never used on a system with it
6242 * configured in. The ring buffers are set to a minimum size. But once
6243 * a user starts to use the tracing facility, then they need to grow
6244 * to their default size.
6245 *
6246 * This function is to be called when a tracer is about to be used.
6247 */
6248int tracing_update_buffers(void)
6249{
6250 int ret = 0;
6251
6252 mutex_lock(&trace_types_lock);
6253 if (!ring_buffer_expanded)
6254 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
6255 RING_BUFFER_ALL_CPUS);
6256 mutex_unlock(&trace_types_lock);
6257
6258 return ret;
6259}
6260
6261struct trace_option_dentry;
6262
6263static void
6264create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6265
6266/*
6267 * Used to clear out the tracer before deletion of an instance.
6268 * Must have trace_types_lock held.
6269 */
6270static void tracing_set_nop(struct trace_array *tr)
6271{
6272 if (tr->current_trace == &nop_trace)
6273 return;
6274
6275 tr->current_trace->enabled--;
6276
6277 if (tr->current_trace->reset)
6278 tr->current_trace->reset(tr);
6279
6280 tr->current_trace = &nop_trace;
6281}
6282
6283static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6284{
6285 /* Only enable if the directory has been created already. */
6286 if (!tr->dir)
6287 return;
6288
6289 create_trace_option_files(tr, t);
6290}
6291
6292int tracing_set_tracer(struct trace_array *tr, const char *buf)
6293{
6294 struct tracer *t;
6295#ifdef CONFIG_TRACER_MAX_TRACE
6296 bool had_max_tr;
6297#endif
6298 int ret = 0;
6299
6300 mutex_lock(&trace_types_lock);
6301
6302 if (!ring_buffer_expanded) {
6303 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6304 RING_BUFFER_ALL_CPUS);
6305 if (ret < 0)
6306 goto out;
6307 ret = 0;
6308 }
6309
6310 for (t = trace_types; t; t = t->next) {
6311 if (strcmp(t->name, buf) == 0)
6312 break;
6313 }
6314 if (!t) {
6315 ret = -EINVAL;
6316 goto out;
6317 }
6318 if (t == tr->current_trace)
6319 goto out;
6320
6321#ifdef CONFIG_TRACER_SNAPSHOT
6322 if (t->use_max_tr) {
6323 arch_spin_lock(&tr->max_lock);
6324 if (tr->cond_snapshot)
6325 ret = -EBUSY;
6326 arch_spin_unlock(&tr->max_lock);
6327 if (ret)
6328 goto out;
6329 }
6330#endif
6331 /* Some tracers won't work on kernel command line */
6332 if (system_state < SYSTEM_RUNNING && t->noboot) {
6333 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6334 t->name);
6335 goto out;
6336 }
6337
6338 /* Some tracers are only allowed for the top level buffer */
6339 if (!trace_ok_for_array(t, tr)) {
6340 ret = -EINVAL;
6341 goto out;
6342 }
6343
6344 /* If trace pipe files are being read, we can't change the tracer */
6345 if (tr->trace_ref) {
6346 ret = -EBUSY;
6347 goto out;
6348 }
6349
6350 trace_branch_disable();
6351
6352 tr->current_trace->enabled--;
6353
6354 if (tr->current_trace->reset)
6355 tr->current_trace->reset(tr);
6356
6357 /* Current trace needs to be nop_trace before synchronize_rcu */
6358 tr->current_trace = &nop_trace;
6359
6360#ifdef CONFIG_TRACER_MAX_TRACE
6361 had_max_tr = tr->allocated_snapshot;
6362
6363 if (had_max_tr && !t->use_max_tr) {
6364 /*
6365 * We need to make sure that the update_max_tr sees that
6366 * current_trace changed to nop_trace to keep it from
6367 * swapping the buffers after we resize it.
6368 * The update_max_tr is called from interrupts disabled
6369 * so a synchronized_sched() is sufficient.
6370 */
6371 synchronize_rcu();
6372 free_snapshot(tr);
6373 }
6374#endif
6375
6376#ifdef CONFIG_TRACER_MAX_TRACE
6377 if (t->use_max_tr && !had_max_tr) {
6378 ret = tracing_alloc_snapshot_instance(tr);
6379 if (ret < 0)
6380 goto out;
6381 }
6382#endif
6383
6384 if (t->init) {
6385 ret = tracer_init(t, tr);
6386 if (ret)
6387 goto out;
6388 }
6389
6390 tr->current_trace = t;
6391 tr->current_trace->enabled++;
6392 trace_branch_enable(tr);
6393 out:
6394 mutex_unlock(&trace_types_lock);
6395
6396 return ret;
6397}
6398
6399static ssize_t
6400tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6401 size_t cnt, loff_t *ppos)
6402{
6403 struct trace_array *tr = filp->private_data;
6404 char buf[MAX_TRACER_SIZE+1];
6405 int i;
6406 size_t ret;
6407 int err;
6408
6409 ret = cnt;
6410
6411 if (cnt > MAX_TRACER_SIZE)
6412 cnt = MAX_TRACER_SIZE;
6413
6414 if (copy_from_user(buf, ubuf, cnt))
6415 return -EFAULT;
6416
6417 buf[cnt] = 0;
6418
6419 /* strip ending whitespace. */
6420 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
6421 buf[i] = 0;
6422
6423 err = tracing_set_tracer(tr, buf);
6424 if (err)
6425 return err;
6426
6427 *ppos += ret;
6428
6429 return ret;
6430}
6431
6432static ssize_t
6433tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6434 size_t cnt, loff_t *ppos)
6435{
6436 char buf[64];
6437 int r;
6438
6439 r = snprintf(buf, sizeof(buf), "%ld\n",
6440 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6441 if (r > sizeof(buf))
6442 r = sizeof(buf);
6443 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6444}
6445
6446static ssize_t
6447tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6448 size_t cnt, loff_t *ppos)
6449{
6450 unsigned long val;
6451 int ret;
6452
6453 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6454 if (ret)
6455 return ret;
6456
6457 *ptr = val * 1000;
6458
6459 return cnt;
6460}
6461
6462static ssize_t
6463tracing_thresh_read(struct file *filp, char __user *ubuf,
6464 size_t cnt, loff_t *ppos)
6465{
6466 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6467}
6468
6469static ssize_t
6470tracing_thresh_write(struct file *filp, const char __user *ubuf,
6471 size_t cnt, loff_t *ppos)
6472{
6473 struct trace_array *tr = filp->private_data;
6474 int ret;
6475
6476 mutex_lock(&trace_types_lock);
6477 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6478 if (ret < 0)
6479 goto out;
6480
6481 if (tr->current_trace->update_thresh) {
6482 ret = tr->current_trace->update_thresh(tr);
6483 if (ret < 0)
6484 goto out;
6485 }
6486
6487 ret = cnt;
6488out:
6489 mutex_unlock(&trace_types_lock);
6490
6491 return ret;
6492}
6493
6494#if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
6495
6496static ssize_t
6497tracing_max_lat_read(struct file *filp, char __user *ubuf,
6498 size_t cnt, loff_t *ppos)
6499{
6500 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6501}
6502
6503static ssize_t
6504tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6505 size_t cnt, loff_t *ppos)
6506{
6507 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6508}
6509
6510#endif
6511
6512static int tracing_open_pipe(struct inode *inode, struct file *filp)
6513{
6514 struct trace_array *tr = inode->i_private;
6515 struct trace_iterator *iter;
6516 int ret;
6517
6518 ret = tracing_check_open_get_tr(tr);
6519 if (ret)
6520 return ret;
6521
6522 mutex_lock(&trace_types_lock);
6523
6524 /* create a buffer to store the information to pass to userspace */
6525 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6526 if (!iter) {
6527 ret = -ENOMEM;
6528 __trace_array_put(tr);
6529 goto out;
6530 }
6531
6532 trace_seq_init(&iter->seq);
6533 iter->trace = tr->current_trace;
6534
6535 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6536 ret = -ENOMEM;
6537 goto fail;
6538 }
6539
6540 /* trace pipe does not show start of buffer */
6541 cpumask_setall(iter->started);
6542
6543 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6544 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6545
6546 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6547 if (trace_clocks[tr->clock_id].in_ns)
6548 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6549
6550 iter->tr = tr;
6551 iter->array_buffer = &tr->array_buffer;
6552 iter->cpu_file = tracing_get_cpu(inode);
6553 mutex_init(&iter->mutex);
6554 filp->private_data = iter;
6555
6556 if (iter->trace->pipe_open)
6557 iter->trace->pipe_open(iter);
6558
6559 nonseekable_open(inode, filp);
6560
6561 tr->trace_ref++;
6562out:
6563 mutex_unlock(&trace_types_lock);
6564 return ret;
6565
6566fail:
6567 kfree(iter);
6568 __trace_array_put(tr);
6569 mutex_unlock(&trace_types_lock);
6570 return ret;
6571}
6572
6573static int tracing_release_pipe(struct inode *inode, struct file *file)
6574{
6575 struct trace_iterator *iter = file->private_data;
6576 struct trace_array *tr = inode->i_private;
6577
6578 mutex_lock(&trace_types_lock);
6579
6580 tr->trace_ref--;
6581
6582 if (iter->trace->pipe_close)
6583 iter->trace->pipe_close(iter);
6584
6585 mutex_unlock(&trace_types_lock);
6586
6587 free_cpumask_var(iter->started);
6588 mutex_destroy(&iter->mutex);
6589 kfree(iter);
6590
6591 trace_array_put(tr);
6592
6593 return 0;
6594}
6595
6596static __poll_t
6597trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6598{
6599 struct trace_array *tr = iter->tr;
6600
6601 /* Iterators are static, they should be filled or empty */
6602 if (trace_buffer_iter(iter, iter->cpu_file))
6603 return EPOLLIN | EPOLLRDNORM;
6604
6605 if (tr->trace_flags & TRACE_ITER_BLOCK)
6606 /*
6607 * Always select as readable when in blocking mode
6608 */
6609 return EPOLLIN | EPOLLRDNORM;
6610 else
6611 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6612 filp, poll_table);
6613}
6614
6615static __poll_t
6616tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6617{
6618 struct trace_iterator *iter = filp->private_data;
6619
6620 return trace_poll(iter, filp, poll_table);
6621}
6622
6623/* Must be called with iter->mutex held. */
6624static int tracing_wait_pipe(struct file *filp)
6625{
6626 struct trace_iterator *iter = filp->private_data;
6627 int ret;
6628
6629 while (trace_empty(iter)) {
6630
6631 if ((filp->f_flags & O_NONBLOCK)) {
6632 return -EAGAIN;
6633 }
6634
6635 /*
6636 * We block until we read something and tracing is disabled.
6637 * We still block if tracing is disabled, but we have never
6638 * read anything. This allows a user to cat this file, and
6639 * then enable tracing. But after we have read something,
6640 * we give an EOF when tracing is again disabled.
6641 *
6642 * iter->pos will be 0 if we haven't read anything.
6643 */
6644 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6645 break;
6646
6647 mutex_unlock(&iter->mutex);
6648
6649 ret = wait_on_pipe(iter, 0);
6650
6651 mutex_lock(&iter->mutex);
6652
6653 if (ret)
6654 return ret;
6655 }
6656
6657 return 1;
6658}
6659
6660/*
6661 * Consumer reader.
6662 */
6663static ssize_t
6664tracing_read_pipe(struct file *filp, char __user *ubuf,
6665 size_t cnt, loff_t *ppos)
6666{
6667 struct trace_iterator *iter = filp->private_data;
6668 ssize_t sret;
6669
6670 /*
6671 * Avoid more than one consumer on a single file descriptor
6672 * This is just a matter of traces coherency, the ring buffer itself
6673 * is protected.
6674 */
6675 mutex_lock(&iter->mutex);
6676
6677 /* return any leftover data */
6678 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6679 if (sret != -EBUSY)
6680 goto out;
6681
6682 trace_seq_init(&iter->seq);
6683
6684 if (iter->trace->read) {
6685 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6686 if (sret)
6687 goto out;
6688 }
6689
6690waitagain:
6691 sret = tracing_wait_pipe(filp);
6692 if (sret <= 0)
6693 goto out;
6694
6695 /* stop when tracing is finished */
6696 if (trace_empty(iter)) {
6697 sret = 0;
6698 goto out;
6699 }
6700
6701 if (cnt >= PAGE_SIZE)
6702 cnt = PAGE_SIZE - 1;
6703
6704 /* reset all but tr, trace, and overruns */
6705 memset(&iter->seq, 0,
6706 sizeof(struct trace_iterator) -
6707 offsetof(struct trace_iterator, seq));
6708 cpumask_clear(iter->started);
6709 trace_seq_init(&iter->seq);
6710 iter->pos = -1;
6711
6712 trace_event_read_lock();
6713 trace_access_lock(iter->cpu_file);
6714 while (trace_find_next_entry_inc(iter) != NULL) {
6715 enum print_line_t ret;
6716 int save_len = iter->seq.seq.len;
6717
6718 ret = print_trace_line(iter);
6719 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6720 /* don't print partial lines */
6721 iter->seq.seq.len = save_len;
6722 break;
6723 }
6724 if (ret != TRACE_TYPE_NO_CONSUME)
6725 trace_consume(iter);
6726
6727 if (trace_seq_used(&iter->seq) >= cnt)
6728 break;
6729
6730 /*
6731 * Setting the full flag means we reached the trace_seq buffer
6732 * size and we should leave by partial output condition above.
6733 * One of the trace_seq_* functions is not used properly.
6734 */
6735 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6736 iter->ent->type);
6737 }
6738 trace_access_unlock(iter->cpu_file);
6739 trace_event_read_unlock();
6740
6741 /* Now copy what we have to the user */
6742 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6743 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6744 trace_seq_init(&iter->seq);
6745
6746 /*
6747 * If there was nothing to send to user, in spite of consuming trace
6748 * entries, go back to wait for more entries.
6749 */
6750 if (sret == -EBUSY)
6751 goto waitagain;
6752
6753out:
6754 mutex_unlock(&iter->mutex);
6755
6756 return sret;
6757}
6758
6759static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6760 unsigned int idx)
6761{
6762 __free_page(spd->pages[idx]);
6763}
6764
6765static size_t
6766tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6767{
6768 size_t count;
6769 int save_len;
6770 int ret;
6771
6772 /* Seq buffer is page-sized, exactly what we need. */
6773 for (;;) {
6774 save_len = iter->seq.seq.len;
6775 ret = print_trace_line(iter);
6776
6777 if (trace_seq_has_overflowed(&iter->seq)) {
6778 iter->seq.seq.len = save_len;
6779 break;
6780 }
6781
6782 /*
6783 * This should not be hit, because it should only
6784 * be set if the iter->seq overflowed. But check it
6785 * anyway to be safe.
6786 */
6787 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6788 iter->seq.seq.len = save_len;
6789 break;
6790 }
6791
6792 count = trace_seq_used(&iter->seq) - save_len;
6793 if (rem < count) {
6794 rem = 0;
6795 iter->seq.seq.len = save_len;
6796 break;
6797 }
6798
6799 if (ret != TRACE_TYPE_NO_CONSUME)
6800 trace_consume(iter);
6801 rem -= count;
6802 if (!trace_find_next_entry_inc(iter)) {
6803 rem = 0;
6804 iter->ent = NULL;
6805 break;
6806 }
6807 }
6808
6809 return rem;
6810}
6811
6812static ssize_t tracing_splice_read_pipe(struct file *filp,
6813 loff_t *ppos,
6814 struct pipe_inode_info *pipe,
6815 size_t len,
6816 unsigned int flags)
6817{
6818 struct page *pages_def[PIPE_DEF_BUFFERS];
6819 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6820 struct trace_iterator *iter = filp->private_data;
6821 struct splice_pipe_desc spd = {
6822 .pages = pages_def,
6823 .partial = partial_def,
6824 .nr_pages = 0, /* This gets updated below. */
6825 .nr_pages_max = PIPE_DEF_BUFFERS,
6826 .ops = &default_pipe_buf_ops,
6827 .spd_release = tracing_spd_release_pipe,
6828 };
6829 ssize_t ret;
6830 size_t rem;
6831 unsigned int i;
6832
6833 if (splice_grow_spd(pipe, &spd))
6834 return -ENOMEM;
6835
6836 mutex_lock(&iter->mutex);
6837
6838 if (iter->trace->splice_read) {
6839 ret = iter->trace->splice_read(iter, filp,
6840 ppos, pipe, len, flags);
6841 if (ret)
6842 goto out_err;
6843 }
6844
6845 ret = tracing_wait_pipe(filp);
6846 if (ret <= 0)
6847 goto out_err;
6848
6849 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6850 ret = -EFAULT;
6851 goto out_err;
6852 }
6853
6854 trace_event_read_lock();
6855 trace_access_lock(iter->cpu_file);
6856
6857 /* Fill as many pages as possible. */
6858 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6859 spd.pages[i] = alloc_page(GFP_KERNEL);
6860 if (!spd.pages[i])
6861 break;
6862
6863 rem = tracing_fill_pipe_page(rem, iter);
6864
6865 /* Copy the data into the page, so we can start over. */
6866 ret = trace_seq_to_buffer(&iter->seq,
6867 page_address(spd.pages[i]),
6868 trace_seq_used(&iter->seq));
6869 if (ret < 0) {
6870 __free_page(spd.pages[i]);
6871 break;
6872 }
6873 spd.partial[i].offset = 0;
6874 spd.partial[i].len = trace_seq_used(&iter->seq);
6875
6876 trace_seq_init(&iter->seq);
6877 }
6878
6879 trace_access_unlock(iter->cpu_file);
6880 trace_event_read_unlock();
6881 mutex_unlock(&iter->mutex);
6882
6883 spd.nr_pages = i;
6884
6885 if (i)
6886 ret = splice_to_pipe(pipe, &spd);
6887 else
6888 ret = 0;
6889out:
6890 splice_shrink_spd(&spd);
6891 return ret;
6892
6893out_err:
6894 mutex_unlock(&iter->mutex);
6895 goto out;
6896}
6897
6898static ssize_t
6899tracing_entries_read(struct file *filp, char __user *ubuf,
6900 size_t cnt, loff_t *ppos)
6901{
6902 struct inode *inode = file_inode(filp);
6903 struct trace_array *tr = inode->i_private;
6904 int cpu = tracing_get_cpu(inode);
6905 char buf[64];
6906 int r = 0;
6907 ssize_t ret;
6908
6909 mutex_lock(&trace_types_lock);
6910
6911 if (cpu == RING_BUFFER_ALL_CPUS) {
6912 int cpu, buf_size_same;
6913 unsigned long size;
6914
6915 size = 0;
6916 buf_size_same = 1;
6917 /* check if all cpu sizes are same */
6918 for_each_tracing_cpu(cpu) {
6919 /* fill in the size from first enabled cpu */
6920 if (size == 0)
6921 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6922 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6923 buf_size_same = 0;
6924 break;
6925 }
6926 }
6927
6928 if (buf_size_same) {
6929 if (!ring_buffer_expanded)
6930 r = sprintf(buf, "%lu (expanded: %lu)\n",
6931 size >> 10,
6932 trace_buf_size >> 10);
6933 else
6934 r = sprintf(buf, "%lu\n", size >> 10);
6935 } else
6936 r = sprintf(buf, "X\n");
6937 } else
6938 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6939
6940 mutex_unlock(&trace_types_lock);
6941
6942 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6943 return ret;
6944}
6945
6946static ssize_t
6947tracing_entries_write(struct file *filp, const char __user *ubuf,
6948 size_t cnt, loff_t *ppos)
6949{
6950 struct inode *inode = file_inode(filp);
6951 struct trace_array *tr = inode->i_private;
6952 unsigned long val;
6953 int ret;
6954
6955 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6956 if (ret)
6957 return ret;
6958
6959 /* must have at least 1 entry */
6960 if (!val)
6961 return -EINVAL;
6962
6963 /* value is in KB */
6964 val <<= 10;
6965 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6966 if (ret < 0)
6967 return ret;
6968
6969 *ppos += cnt;
6970
6971 return cnt;
6972}
6973
6974static ssize_t
6975tracing_total_entries_read(struct file *filp, char __user *ubuf,
6976 size_t cnt, loff_t *ppos)
6977{
6978 struct trace_array *tr = filp->private_data;
6979 char buf[64];
6980 int r, cpu;
6981 unsigned long size = 0, expanded_size = 0;
6982
6983 mutex_lock(&trace_types_lock);
6984 for_each_tracing_cpu(cpu) {
6985 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6986 if (!ring_buffer_expanded)
6987 expanded_size += trace_buf_size >> 10;
6988 }
6989 if (ring_buffer_expanded)
6990 r = sprintf(buf, "%lu\n", size);
6991 else
6992 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6993 mutex_unlock(&trace_types_lock);
6994
6995 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6996}
6997
6998static ssize_t
6999tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7000 size_t cnt, loff_t *ppos)
7001{
7002 /*
7003 * There is no need to read what the user has written, this function
7004 * is just to make sure that there is no error when "echo" is used
7005 */
7006
7007 *ppos += cnt;
7008
7009 return cnt;
7010}
7011
7012static int
7013tracing_free_buffer_release(struct inode *inode, struct file *filp)
7014{
7015 struct trace_array *tr = inode->i_private;
7016
7017 /* disable tracing ? */
7018 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7019 tracer_tracing_off(tr);
7020 /* resize the ring buffer to 0 */
7021 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7022
7023 trace_array_put(tr);
7024
7025 return 0;
7026}
7027
7028static ssize_t
7029tracing_mark_write(struct file *filp, const char __user *ubuf,
7030 size_t cnt, loff_t *fpos)
7031{
7032 struct trace_array *tr = filp->private_data;
7033 struct ring_buffer_event *event;
7034 enum event_trigger_type tt = ETT_NONE;
7035 struct trace_buffer *buffer;
7036 struct print_entry *entry;
7037 ssize_t written;
7038 int size;
7039 int len;
7040
7041/* Used in tracing_mark_raw_write() as well */
7042#define FAULTED_STR "<faulted>"
7043#define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7044
7045 if (tracing_disabled)
7046 return -EINVAL;
7047
7048 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7049 return -EINVAL;
7050
7051 if (cnt > TRACE_BUF_SIZE)
7052 cnt = TRACE_BUF_SIZE;
7053
7054 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7055
7056 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
7057
7058 /* If less than "<faulted>", then make sure we can still add that */
7059 if (cnt < FAULTED_SIZE)
7060 size += FAULTED_SIZE - cnt;
7061
7062 buffer = tr->array_buffer.buffer;
7063 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7064 tracing_gen_ctx());
7065 if (unlikely(!event))
7066 /* Ring buffer disabled, return as if not open for write */
7067 return -EBADF;
7068
7069 entry = ring_buffer_event_data(event);
7070 entry->ip = _THIS_IP_;
7071
7072 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7073 if (len) {
7074 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7075 cnt = FAULTED_SIZE;
7076 written = -EFAULT;
7077 } else
7078 written = cnt;
7079
7080 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7081 /* do not add \n before testing triggers, but add \0 */
7082 entry->buf[cnt] = '\0';
7083 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7084 }
7085
7086 if (entry->buf[cnt - 1] != '\n') {
7087 entry->buf[cnt] = '\n';
7088 entry->buf[cnt + 1] = '\0';
7089 } else
7090 entry->buf[cnt] = '\0';
7091
7092 if (static_branch_unlikely(&trace_marker_exports_enabled))
7093 ftrace_exports(event, TRACE_EXPORT_MARKER);
7094 __buffer_unlock_commit(buffer, event);
7095
7096 if (tt)
7097 event_triggers_post_call(tr->trace_marker_file, tt);
7098
7099 if (written > 0)
7100 *fpos += written;
7101
7102 return written;
7103}
7104
7105/* Limit it for now to 3K (including tag) */
7106#define RAW_DATA_MAX_SIZE (1024*3)
7107
7108static ssize_t
7109tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7110 size_t cnt, loff_t *fpos)
7111{
7112 struct trace_array *tr = filp->private_data;
7113 struct ring_buffer_event *event;
7114 struct trace_buffer *buffer;
7115 struct raw_data_entry *entry;
7116 ssize_t written;
7117 int size;
7118 int len;
7119
7120#define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7121
7122 if (tracing_disabled)
7123 return -EINVAL;
7124
7125 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7126 return -EINVAL;
7127
7128 /* The marker must at least have a tag id */
7129 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
7130 return -EINVAL;
7131
7132 if (cnt > TRACE_BUF_SIZE)
7133 cnt = TRACE_BUF_SIZE;
7134
7135 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7136
7137 size = sizeof(*entry) + cnt;
7138 if (cnt < FAULT_SIZE_ID)
7139 size += FAULT_SIZE_ID - cnt;
7140
7141 buffer = tr->array_buffer.buffer;
7142 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7143 tracing_gen_ctx());
7144 if (!event)
7145 /* Ring buffer disabled, return as if not open for write */
7146 return -EBADF;
7147
7148 entry = ring_buffer_event_data(event);
7149
7150 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7151 if (len) {
7152 entry->id = -1;
7153 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7154 written = -EFAULT;
7155 } else
7156 written = cnt;
7157
7158 __buffer_unlock_commit(buffer, event);
7159
7160 if (written > 0)
7161 *fpos += written;
7162
7163 return written;
7164}
7165
7166static int tracing_clock_show(struct seq_file *m, void *v)
7167{
7168 struct trace_array *tr = m->private;
7169 int i;
7170
7171 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7172 seq_printf(m,
7173 "%s%s%s%s", i ? " " : "",
7174 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7175 i == tr->clock_id ? "]" : "");
7176 seq_putc(m, '\n');
7177
7178 return 0;
7179}
7180
7181int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7182{
7183 int i;
7184
7185 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7186 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7187 break;
7188 }
7189 if (i == ARRAY_SIZE(trace_clocks))
7190 return -EINVAL;
7191
7192 mutex_lock(&trace_types_lock);
7193
7194 tr->clock_id = i;
7195
7196 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7197
7198 /*
7199 * New clock may not be consistent with the previous clock.
7200 * Reset the buffer so that it doesn't have incomparable timestamps.
7201 */
7202 tracing_reset_online_cpus(&tr->array_buffer);
7203
7204#ifdef CONFIG_TRACER_MAX_TRACE
7205 if (tr->max_buffer.buffer)
7206 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7207 tracing_reset_online_cpus(&tr->max_buffer);
7208#endif
7209
7210 mutex_unlock(&trace_types_lock);
7211
7212 return 0;
7213}
7214
7215static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7216 size_t cnt, loff_t *fpos)
7217{
7218 struct seq_file *m = filp->private_data;
7219 struct trace_array *tr = m->private;
7220 char buf[64];
7221 const char *clockstr;
7222 int ret;
7223
7224 if (cnt >= sizeof(buf))
7225 return -EINVAL;
7226
7227 if (copy_from_user(buf, ubuf, cnt))
7228 return -EFAULT;
7229
7230 buf[cnt] = 0;
7231
7232 clockstr = strstrip(buf);
7233
7234 ret = tracing_set_clock(tr, clockstr);
7235 if (ret)
7236 return ret;
7237
7238 *fpos += cnt;
7239
7240 return cnt;
7241}
7242
7243static int tracing_clock_open(struct inode *inode, struct file *file)
7244{
7245 struct trace_array *tr = inode->i_private;
7246 int ret;
7247
7248 ret = tracing_check_open_get_tr(tr);
7249 if (ret)
7250 return ret;
7251
7252 ret = single_open(file, tracing_clock_show, inode->i_private);
7253 if (ret < 0)
7254 trace_array_put(tr);
7255
7256 return ret;
7257}
7258
7259static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7260{
7261 struct trace_array *tr = m->private;
7262
7263 mutex_lock(&trace_types_lock);
7264
7265 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7266 seq_puts(m, "delta [absolute]\n");
7267 else
7268 seq_puts(m, "[delta] absolute\n");
7269
7270 mutex_unlock(&trace_types_lock);
7271
7272 return 0;
7273}
7274
7275static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7276{
7277 struct trace_array *tr = inode->i_private;
7278 int ret;
7279
7280 ret = tracing_check_open_get_tr(tr);
7281 if (ret)
7282 return ret;
7283
7284 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7285 if (ret < 0)
7286 trace_array_put(tr);
7287
7288 return ret;
7289}
7290
7291u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7292{
7293 if (rbe == this_cpu_read(trace_buffered_event))
7294 return ring_buffer_time_stamp(buffer);
7295
7296 return ring_buffer_event_time_stamp(buffer, rbe);
7297}
7298
7299/*
7300 * Set or disable using the per CPU trace_buffer_event when possible.
7301 */
7302int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7303{
7304 int ret = 0;
7305
7306 mutex_lock(&trace_types_lock);
7307
7308 if (set && tr->no_filter_buffering_ref++)
7309 goto out;
7310
7311 if (!set) {
7312 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7313 ret = -EINVAL;
7314 goto out;
7315 }
7316
7317 --tr->no_filter_buffering_ref;
7318 }
7319 out:
7320 mutex_unlock(&trace_types_lock);
7321
7322 return ret;
7323}
7324
7325struct ftrace_buffer_info {
7326 struct trace_iterator iter;
7327 void *spare;
7328 unsigned int spare_cpu;
7329 unsigned int read;
7330};
7331
7332#ifdef CONFIG_TRACER_SNAPSHOT
7333static int tracing_snapshot_open(struct inode *inode, struct file *file)
7334{
7335 struct trace_array *tr = inode->i_private;
7336 struct trace_iterator *iter;
7337 struct seq_file *m;
7338 int ret;
7339
7340 ret = tracing_check_open_get_tr(tr);
7341 if (ret)
7342 return ret;
7343
7344 if (file->f_mode & FMODE_READ) {
7345 iter = __tracing_open(inode, file, true);
7346 if (IS_ERR(iter))
7347 ret = PTR_ERR(iter);
7348 } else {
7349 /* Writes still need the seq_file to hold the private data */
7350 ret = -ENOMEM;
7351 m = kzalloc(sizeof(*m), GFP_KERNEL);
7352 if (!m)
7353 goto out;
7354 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7355 if (!iter) {
7356 kfree(m);
7357 goto out;
7358 }
7359 ret = 0;
7360
7361 iter->tr = tr;
7362 iter->array_buffer = &tr->max_buffer;
7363 iter->cpu_file = tracing_get_cpu(inode);
7364 m->private = iter;
7365 file->private_data = m;
7366 }
7367out:
7368 if (ret < 0)
7369 trace_array_put(tr);
7370
7371 return ret;
7372}
7373
7374static ssize_t
7375tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7376 loff_t *ppos)
7377{
7378 struct seq_file *m = filp->private_data;
7379 struct trace_iterator *iter = m->private;
7380 struct trace_array *tr = iter->tr;
7381 unsigned long val;
7382 int ret;
7383
7384 ret = tracing_update_buffers();
7385 if (ret < 0)
7386 return ret;
7387
7388 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7389 if (ret)
7390 return ret;
7391
7392 mutex_lock(&trace_types_lock);
7393
7394 if (tr->current_trace->use_max_tr) {
7395 ret = -EBUSY;
7396 goto out;
7397 }
7398
7399 arch_spin_lock(&tr->max_lock);
7400 if (tr->cond_snapshot)
7401 ret = -EBUSY;
7402 arch_spin_unlock(&tr->max_lock);
7403 if (ret)
7404 goto out;
7405
7406 switch (val) {
7407 case 0:
7408 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7409 ret = -EINVAL;
7410 break;
7411 }
7412 if (tr->allocated_snapshot)
7413 free_snapshot(tr);
7414 break;
7415 case 1:
7416/* Only allow per-cpu swap if the ring buffer supports it */
7417#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7418 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7419 ret = -EINVAL;
7420 break;
7421 }
7422#endif
7423 if (tr->allocated_snapshot)
7424 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7425 &tr->array_buffer, iter->cpu_file);
7426 else
7427 ret = tracing_alloc_snapshot_instance(tr);
7428 if (ret < 0)
7429 break;
7430 local_irq_disable();
7431 /* Now, we're going to swap */
7432 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7433 update_max_tr(tr, current, smp_processor_id(), NULL);
7434 else
7435 update_max_tr_single(tr, current, iter->cpu_file);
7436 local_irq_enable();
7437 break;
7438 default:
7439 if (tr->allocated_snapshot) {
7440 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7441 tracing_reset_online_cpus(&tr->max_buffer);
7442 else
7443 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7444 }
7445 break;
7446 }
7447
7448 if (ret >= 0) {
7449 *ppos += cnt;
7450 ret = cnt;
7451 }
7452out:
7453 mutex_unlock(&trace_types_lock);
7454 return ret;
7455}
7456
7457static int tracing_snapshot_release(struct inode *inode, struct file *file)
7458{
7459 struct seq_file *m = file->private_data;
7460 int ret;
7461
7462 ret = tracing_release(inode, file);
7463
7464 if (file->f_mode & FMODE_READ)
7465 return ret;
7466
7467 /* If write only, the seq_file is just a stub */
7468 if (m)
7469 kfree(m->private);
7470 kfree(m);
7471
7472 return 0;
7473}
7474
7475static int tracing_buffers_open(struct inode *inode, struct file *filp);
7476static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7477 size_t count, loff_t *ppos);
7478static int tracing_buffers_release(struct inode *inode, struct file *file);
7479static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7480 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7481
7482static int snapshot_raw_open(struct inode *inode, struct file *filp)
7483{
7484 struct ftrace_buffer_info *info;
7485 int ret;
7486
7487 /* The following checks for tracefs lockdown */
7488 ret = tracing_buffers_open(inode, filp);
7489 if (ret < 0)
7490 return ret;
7491
7492 info = filp->private_data;
7493
7494 if (info->iter.trace->use_max_tr) {
7495 tracing_buffers_release(inode, filp);
7496 return -EBUSY;
7497 }
7498
7499 info->iter.snapshot = true;
7500 info->iter.array_buffer = &info->iter.tr->max_buffer;
7501
7502 return ret;
7503}
7504
7505#endif /* CONFIG_TRACER_SNAPSHOT */
7506
7507
7508static const struct file_operations tracing_thresh_fops = {
7509 .open = tracing_open_generic,
7510 .read = tracing_thresh_read,
7511 .write = tracing_thresh_write,
7512 .llseek = generic_file_llseek,
7513};
7514
7515#if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
7516static const struct file_operations tracing_max_lat_fops = {
7517 .open = tracing_open_generic,
7518 .read = tracing_max_lat_read,
7519 .write = tracing_max_lat_write,
7520 .llseek = generic_file_llseek,
7521};
7522#endif
7523
7524static const struct file_operations set_tracer_fops = {
7525 .open = tracing_open_generic,
7526 .read = tracing_set_trace_read,
7527 .write = tracing_set_trace_write,
7528 .llseek = generic_file_llseek,
7529};
7530
7531static const struct file_operations tracing_pipe_fops = {
7532 .open = tracing_open_pipe,
7533 .poll = tracing_poll_pipe,
7534 .read = tracing_read_pipe,
7535 .splice_read = tracing_splice_read_pipe,
7536 .release = tracing_release_pipe,
7537 .llseek = no_llseek,
7538};
7539
7540static const struct file_operations tracing_entries_fops = {
7541 .open = tracing_open_generic_tr,
7542 .read = tracing_entries_read,
7543 .write = tracing_entries_write,
7544 .llseek = generic_file_llseek,
7545 .release = tracing_release_generic_tr,
7546};
7547
7548static const struct file_operations tracing_total_entries_fops = {
7549 .open = tracing_open_generic_tr,
7550 .read = tracing_total_entries_read,
7551 .llseek = generic_file_llseek,
7552 .release = tracing_release_generic_tr,
7553};
7554
7555static const struct file_operations tracing_free_buffer_fops = {
7556 .open = tracing_open_generic_tr,
7557 .write = tracing_free_buffer_write,
7558 .release = tracing_free_buffer_release,
7559};
7560
7561static const struct file_operations tracing_mark_fops = {
7562 .open = tracing_open_generic_tr,
7563 .write = tracing_mark_write,
7564 .llseek = generic_file_llseek,
7565 .release = tracing_release_generic_tr,
7566};
7567
7568static const struct file_operations tracing_mark_raw_fops = {
7569 .open = tracing_open_generic_tr,
7570 .write = tracing_mark_raw_write,
7571 .llseek = generic_file_llseek,
7572 .release = tracing_release_generic_tr,
7573};
7574
7575static const struct file_operations trace_clock_fops = {
7576 .open = tracing_clock_open,
7577 .read = seq_read,
7578 .llseek = seq_lseek,
7579 .release = tracing_single_release_tr,
7580 .write = tracing_clock_write,
7581};
7582
7583static const struct file_operations trace_time_stamp_mode_fops = {
7584 .open = tracing_time_stamp_mode_open,
7585 .read = seq_read,
7586 .llseek = seq_lseek,
7587 .release = tracing_single_release_tr,
7588};
7589
7590#ifdef CONFIG_TRACER_SNAPSHOT
7591static const struct file_operations snapshot_fops = {
7592 .open = tracing_snapshot_open,
7593 .read = seq_read,
7594 .write = tracing_snapshot_write,
7595 .llseek = tracing_lseek,
7596 .release = tracing_snapshot_release,
7597};
7598
7599static const struct file_operations snapshot_raw_fops = {
7600 .open = snapshot_raw_open,
7601 .read = tracing_buffers_read,
7602 .release = tracing_buffers_release,
7603 .splice_read = tracing_buffers_splice_read,
7604 .llseek = no_llseek,
7605};
7606
7607#endif /* CONFIG_TRACER_SNAPSHOT */
7608
7609/*
7610 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7611 * @filp: The active open file structure
7612 * @ubuf: The userspace provided buffer to read value into
7613 * @cnt: The maximum number of bytes to read
7614 * @ppos: The current "file" position
7615 *
7616 * This function implements the write interface for a struct trace_min_max_param.
7617 * The filp->private_data must point to a trace_min_max_param structure that
7618 * defines where to write the value, the min and the max acceptable values,
7619 * and a lock to protect the write.
7620 */
7621static ssize_t
7622trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7623{
7624 struct trace_min_max_param *param = filp->private_data;
7625 u64 val;
7626 int err;
7627
7628 if (!param)
7629 return -EFAULT;
7630
7631 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7632 if (err)
7633 return err;
7634
7635 if (param->lock)
7636 mutex_lock(param->lock);
7637
7638 if (param->min && val < *param->min)
7639 err = -EINVAL;
7640
7641 if (param->max && val > *param->max)
7642 err = -EINVAL;
7643
7644 if (!err)
7645 *param->val = val;
7646
7647 if (param->lock)
7648 mutex_unlock(param->lock);
7649
7650 if (err)
7651 return err;
7652
7653 return cnt;
7654}
7655
7656/*
7657 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7658 * @filp: The active open file structure
7659 * @ubuf: The userspace provided buffer to read value into
7660 * @cnt: The maximum number of bytes to read
7661 * @ppos: The current "file" position
7662 *
7663 * This function implements the read interface for a struct trace_min_max_param.
7664 * The filp->private_data must point to a trace_min_max_param struct with valid
7665 * data.
7666 */
7667static ssize_t
7668trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7669{
7670 struct trace_min_max_param *param = filp->private_data;
7671 char buf[U64_STR_SIZE];
7672 int len;
7673 u64 val;
7674
7675 if (!param)
7676 return -EFAULT;
7677
7678 val = *param->val;
7679
7680 if (cnt > sizeof(buf))
7681 cnt = sizeof(buf);
7682
7683 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7684
7685 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7686}
7687
7688const struct file_operations trace_min_max_fops = {
7689 .open = tracing_open_generic,
7690 .read = trace_min_max_read,
7691 .write = trace_min_max_write,
7692};
7693
7694#define TRACING_LOG_ERRS_MAX 8
7695#define TRACING_LOG_LOC_MAX 128
7696
7697#define CMD_PREFIX " Command: "
7698
7699struct err_info {
7700 const char **errs; /* ptr to loc-specific array of err strings */
7701 u8 type; /* index into errs -> specific err string */
7702 u8 pos; /* MAX_FILTER_STR_VAL = 256 */
7703 u64 ts;
7704};
7705
7706struct tracing_log_err {
7707 struct list_head list;
7708 struct err_info info;
7709 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7710 char cmd[MAX_FILTER_STR_VAL]; /* what caused err */
7711};
7712
7713static DEFINE_MUTEX(tracing_err_log_lock);
7714
7715static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr)
7716{
7717 struct tracing_log_err *err;
7718
7719 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7720 err = kzalloc(sizeof(*err), GFP_KERNEL);
7721 if (!err)
7722 err = ERR_PTR(-ENOMEM);
7723 tr->n_err_log_entries++;
7724
7725 return err;
7726 }
7727
7728 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7729 list_del(&err->list);
7730
7731 return err;
7732}
7733
7734/**
7735 * err_pos - find the position of a string within a command for error careting
7736 * @cmd: The tracing command that caused the error
7737 * @str: The string to position the caret at within @cmd
7738 *
7739 * Finds the position of the first occurrence of @str within @cmd. The
7740 * return value can be passed to tracing_log_err() for caret placement
7741 * within @cmd.
7742 *
7743 * Returns the index within @cmd of the first occurrence of @str or 0
7744 * if @str was not found.
7745 */
7746unsigned int err_pos(char *cmd, const char *str)
7747{
7748 char *found;
7749
7750 if (WARN_ON(!strlen(cmd)))
7751 return 0;
7752
7753 found = strstr(cmd, str);
7754 if (found)
7755 return found - cmd;
7756
7757 return 0;
7758}
7759
7760/**
7761 * tracing_log_err - write an error to the tracing error log
7762 * @tr: The associated trace array for the error (NULL for top level array)
7763 * @loc: A string describing where the error occurred
7764 * @cmd: The tracing command that caused the error
7765 * @errs: The array of loc-specific static error strings
7766 * @type: The index into errs[], which produces the specific static err string
7767 * @pos: The position the caret should be placed in the cmd
7768 *
7769 * Writes an error into tracing/error_log of the form:
7770 *
7771 * <loc>: error: <text>
7772 * Command: <cmd>
7773 * ^
7774 *
7775 * tracing/error_log is a small log file containing the last
7776 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7777 * unless there has been a tracing error, and the error log can be
7778 * cleared and have its memory freed by writing the empty string in
7779 * truncation mode to it i.e. echo > tracing/error_log.
7780 *
7781 * NOTE: the @errs array along with the @type param are used to
7782 * produce a static error string - this string is not copied and saved
7783 * when the error is logged - only a pointer to it is saved. See
7784 * existing callers for examples of how static strings are typically
7785 * defined for use with tracing_log_err().
7786 */
7787void tracing_log_err(struct trace_array *tr,
7788 const char *loc, const char *cmd,
7789 const char **errs, u8 type, u8 pos)
7790{
7791 struct tracing_log_err *err;
7792
7793 if (!tr)
7794 tr = &global_trace;
7795
7796 mutex_lock(&tracing_err_log_lock);
7797 err = get_tracing_log_err(tr);
7798 if (PTR_ERR(err) == -ENOMEM) {
7799 mutex_unlock(&tracing_err_log_lock);
7800 return;
7801 }
7802
7803 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7804 snprintf(err->cmd, MAX_FILTER_STR_VAL,"\n" CMD_PREFIX "%s\n", cmd);
7805
7806 err->info.errs = errs;
7807 err->info.type = type;
7808 err->info.pos = pos;
7809 err->info.ts = local_clock();
7810
7811 list_add_tail(&err->list, &tr->err_log);
7812 mutex_unlock(&tracing_err_log_lock);
7813}
7814
7815static void clear_tracing_err_log(struct trace_array *tr)
7816{
7817 struct tracing_log_err *err, *next;
7818
7819 mutex_lock(&tracing_err_log_lock);
7820 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7821 list_del(&err->list);
7822 kfree(err);
7823 }
7824
7825 tr->n_err_log_entries = 0;
7826 mutex_unlock(&tracing_err_log_lock);
7827}
7828
7829static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7830{
7831 struct trace_array *tr = m->private;
7832
7833 mutex_lock(&tracing_err_log_lock);
7834
7835 return seq_list_start(&tr->err_log, *pos);
7836}
7837
7838static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7839{
7840 struct trace_array *tr = m->private;
7841
7842 return seq_list_next(v, &tr->err_log, pos);
7843}
7844
7845static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7846{
7847 mutex_unlock(&tracing_err_log_lock);
7848}
7849
7850static void tracing_err_log_show_pos(struct seq_file *m, u8 pos)
7851{
7852 u8 i;
7853
7854 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7855 seq_putc(m, ' ');
7856 for (i = 0; i < pos; i++)
7857 seq_putc(m, ' ');
7858 seq_puts(m, "^\n");
7859}
7860
7861static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7862{
7863 struct tracing_log_err *err = v;
7864
7865 if (err) {
7866 const char *err_text = err->info.errs[err->info.type];
7867 u64 sec = err->info.ts;
7868 u32 nsec;
7869
7870 nsec = do_div(sec, NSEC_PER_SEC);
7871 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7872 err->loc, err_text);
7873 seq_printf(m, "%s", err->cmd);
7874 tracing_err_log_show_pos(m, err->info.pos);
7875 }
7876
7877 return 0;
7878}
7879
7880static const struct seq_operations tracing_err_log_seq_ops = {
7881 .start = tracing_err_log_seq_start,
7882 .next = tracing_err_log_seq_next,
7883 .stop = tracing_err_log_seq_stop,
7884 .show = tracing_err_log_seq_show
7885};
7886
7887static int tracing_err_log_open(struct inode *inode, struct file *file)
7888{
7889 struct trace_array *tr = inode->i_private;
7890 int ret = 0;
7891
7892 ret = tracing_check_open_get_tr(tr);
7893 if (ret)
7894 return ret;
7895
7896 /* If this file was opened for write, then erase contents */
7897 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7898 clear_tracing_err_log(tr);
7899
7900 if (file->f_mode & FMODE_READ) {
7901 ret = seq_open(file, &tracing_err_log_seq_ops);
7902 if (!ret) {
7903 struct seq_file *m = file->private_data;
7904 m->private = tr;
7905 } else {
7906 trace_array_put(tr);
7907 }
7908 }
7909 return ret;
7910}
7911
7912static ssize_t tracing_err_log_write(struct file *file,
7913 const char __user *buffer,
7914 size_t count, loff_t *ppos)
7915{
7916 return count;
7917}
7918
7919static int tracing_err_log_release(struct inode *inode, struct file *file)
7920{
7921 struct trace_array *tr = inode->i_private;
7922
7923 trace_array_put(tr);
7924
7925 if (file->f_mode & FMODE_READ)
7926 seq_release(inode, file);
7927
7928 return 0;
7929}
7930
7931static const struct file_operations tracing_err_log_fops = {
7932 .open = tracing_err_log_open,
7933 .write = tracing_err_log_write,
7934 .read = seq_read,
7935 .llseek = seq_lseek,
7936 .release = tracing_err_log_release,
7937};
7938
7939static int tracing_buffers_open(struct inode *inode, struct file *filp)
7940{
7941 struct trace_array *tr = inode->i_private;
7942 struct ftrace_buffer_info *info;
7943 int ret;
7944
7945 ret = tracing_check_open_get_tr(tr);
7946 if (ret)
7947 return ret;
7948
7949 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7950 if (!info) {
7951 trace_array_put(tr);
7952 return -ENOMEM;
7953 }
7954
7955 mutex_lock(&trace_types_lock);
7956
7957 info->iter.tr = tr;
7958 info->iter.cpu_file = tracing_get_cpu(inode);
7959 info->iter.trace = tr->current_trace;
7960 info->iter.array_buffer = &tr->array_buffer;
7961 info->spare = NULL;
7962 /* Force reading ring buffer for first read */
7963 info->read = (unsigned int)-1;
7964
7965 filp->private_data = info;
7966
7967 tr->trace_ref++;
7968
7969 mutex_unlock(&trace_types_lock);
7970
7971 ret = nonseekable_open(inode, filp);
7972 if (ret < 0)
7973 trace_array_put(tr);
7974
7975 return ret;
7976}
7977
7978static __poll_t
7979tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7980{
7981 struct ftrace_buffer_info *info = filp->private_data;
7982 struct trace_iterator *iter = &info->iter;
7983
7984 return trace_poll(iter, filp, poll_table);
7985}
7986
7987static ssize_t
7988tracing_buffers_read(struct file *filp, char __user *ubuf,
7989 size_t count, loff_t *ppos)
7990{
7991 struct ftrace_buffer_info *info = filp->private_data;
7992 struct trace_iterator *iter = &info->iter;
7993 ssize_t ret = 0;
7994 ssize_t size;
7995
7996 if (!count)
7997 return 0;
7998
7999#ifdef CONFIG_TRACER_MAX_TRACE
8000 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8001 return -EBUSY;
8002#endif
8003
8004 if (!info->spare) {
8005 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8006 iter->cpu_file);
8007 if (IS_ERR(info->spare)) {
8008 ret = PTR_ERR(info->spare);
8009 info->spare = NULL;
8010 } else {
8011 info->spare_cpu = iter->cpu_file;
8012 }
8013 }
8014 if (!info->spare)
8015 return ret;
8016
8017 /* Do we have previous read data to read? */
8018 if (info->read < PAGE_SIZE)
8019 goto read;
8020
8021 again:
8022 trace_access_lock(iter->cpu_file);
8023 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8024 &info->spare,
8025 count,
8026 iter->cpu_file, 0);
8027 trace_access_unlock(iter->cpu_file);
8028
8029 if (ret < 0) {
8030 if (trace_empty(iter)) {
8031 if ((filp->f_flags & O_NONBLOCK))
8032 return -EAGAIN;
8033
8034 ret = wait_on_pipe(iter, 0);
8035 if (ret)
8036 return ret;
8037
8038 goto again;
8039 }
8040 return 0;
8041 }
8042
8043 info->read = 0;
8044 read:
8045 size = PAGE_SIZE - info->read;
8046 if (size > count)
8047 size = count;
8048
8049 ret = copy_to_user(ubuf, info->spare + info->read, size);
8050 if (ret == size)
8051 return -EFAULT;
8052
8053 size -= ret;
8054
8055 *ppos += size;
8056 info->read += size;
8057
8058 return size;
8059}
8060
8061static int tracing_buffers_release(struct inode *inode, struct file *file)
8062{
8063 struct ftrace_buffer_info *info = file->private_data;
8064 struct trace_iterator *iter = &info->iter;
8065
8066 mutex_lock(&trace_types_lock);
8067
8068 iter->tr->trace_ref--;
8069
8070 __trace_array_put(iter->tr);
8071
8072 if (info->spare)
8073 ring_buffer_free_read_page(iter->array_buffer->buffer,
8074 info->spare_cpu, info->spare);
8075 kvfree(info);
8076
8077 mutex_unlock(&trace_types_lock);
8078
8079 return 0;
8080}
8081
8082struct buffer_ref {
8083 struct trace_buffer *buffer;
8084 void *page;
8085 int cpu;
8086 refcount_t refcount;
8087};
8088
8089static void buffer_ref_release(struct buffer_ref *ref)
8090{
8091 if (!refcount_dec_and_test(&ref->refcount))
8092 return;
8093 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8094 kfree(ref);
8095}
8096
8097static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8098 struct pipe_buffer *buf)
8099{
8100 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8101
8102 buffer_ref_release(ref);
8103 buf->private = 0;
8104}
8105
8106static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8107 struct pipe_buffer *buf)
8108{
8109 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8110
8111 if (refcount_read(&ref->refcount) > INT_MAX/2)
8112 return false;
8113
8114 refcount_inc(&ref->refcount);
8115 return true;
8116}
8117
8118/* Pipe buffer operations for a buffer. */
8119static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8120 .release = buffer_pipe_buf_release,
8121 .get = buffer_pipe_buf_get,
8122};
8123
8124/*
8125 * Callback from splice_to_pipe(), if we need to release some pages
8126 * at the end of the spd in case we error'ed out in filling the pipe.
8127 */
8128static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8129{
8130 struct buffer_ref *ref =
8131 (struct buffer_ref *)spd->partial[i].private;
8132
8133 buffer_ref_release(ref);
8134 spd->partial[i].private = 0;
8135}
8136
8137static ssize_t
8138tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8139 struct pipe_inode_info *pipe, size_t len,
8140 unsigned int flags)
8141{
8142 struct ftrace_buffer_info *info = file->private_data;
8143 struct trace_iterator *iter = &info->iter;
8144 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8145 struct page *pages_def[PIPE_DEF_BUFFERS];
8146 struct splice_pipe_desc spd = {
8147 .pages = pages_def,
8148 .partial = partial_def,
8149 .nr_pages_max = PIPE_DEF_BUFFERS,
8150 .ops = &buffer_pipe_buf_ops,
8151 .spd_release = buffer_spd_release,
8152 };
8153 struct buffer_ref *ref;
8154 int entries, i;
8155 ssize_t ret = 0;
8156
8157#ifdef CONFIG_TRACER_MAX_TRACE
8158 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8159 return -EBUSY;
8160#endif
8161
8162 if (*ppos & (PAGE_SIZE - 1))
8163 return -EINVAL;
8164
8165 if (len & (PAGE_SIZE - 1)) {
8166 if (len < PAGE_SIZE)
8167 return -EINVAL;
8168 len &= PAGE_MASK;
8169 }
8170
8171 if (splice_grow_spd(pipe, &spd))
8172 return -ENOMEM;
8173
8174 again:
8175 trace_access_lock(iter->cpu_file);
8176 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8177
8178 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
8179 struct page *page;
8180 int r;
8181
8182 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8183 if (!ref) {
8184 ret = -ENOMEM;
8185 break;
8186 }
8187
8188 refcount_set(&ref->refcount, 1);
8189 ref->buffer = iter->array_buffer->buffer;
8190 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8191 if (IS_ERR(ref->page)) {
8192 ret = PTR_ERR(ref->page);
8193 ref->page = NULL;
8194 kfree(ref);
8195 break;
8196 }
8197 ref->cpu = iter->cpu_file;
8198
8199 r = ring_buffer_read_page(ref->buffer, &ref->page,
8200 len, iter->cpu_file, 1);
8201 if (r < 0) {
8202 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8203 ref->page);
8204 kfree(ref);
8205 break;
8206 }
8207
8208 page = virt_to_page(ref->page);
8209
8210 spd.pages[i] = page;
8211 spd.partial[i].len = PAGE_SIZE;
8212 spd.partial[i].offset = 0;
8213 spd.partial[i].private = (unsigned long)ref;
8214 spd.nr_pages++;
8215 *ppos += PAGE_SIZE;
8216
8217 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8218 }
8219
8220 trace_access_unlock(iter->cpu_file);
8221 spd.nr_pages = i;
8222
8223 /* did we read anything? */
8224 if (!spd.nr_pages) {
8225 if (ret)
8226 goto out;
8227
8228 ret = -EAGAIN;
8229 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8230 goto out;
8231
8232 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
8233 if (ret)
8234 goto out;
8235
8236 goto again;
8237 }
8238
8239 ret = splice_to_pipe(pipe, &spd);
8240out:
8241 splice_shrink_spd(&spd);
8242
8243 return ret;
8244}
8245
8246static const struct file_operations tracing_buffers_fops = {
8247 .open = tracing_buffers_open,
8248 .read = tracing_buffers_read,
8249 .poll = tracing_buffers_poll,
8250 .release = tracing_buffers_release,
8251 .splice_read = tracing_buffers_splice_read,
8252 .llseek = no_llseek,
8253};
8254
8255static ssize_t
8256tracing_stats_read(struct file *filp, char __user *ubuf,
8257 size_t count, loff_t *ppos)
8258{
8259 struct inode *inode = file_inode(filp);
8260 struct trace_array *tr = inode->i_private;
8261 struct array_buffer *trace_buf = &tr->array_buffer;
8262 int cpu = tracing_get_cpu(inode);
8263 struct trace_seq *s;
8264 unsigned long cnt;
8265 unsigned long long t;
8266 unsigned long usec_rem;
8267
8268 s = kmalloc(sizeof(*s), GFP_KERNEL);
8269 if (!s)
8270 return -ENOMEM;
8271
8272 trace_seq_init(s);
8273
8274 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8275 trace_seq_printf(s, "entries: %ld\n", cnt);
8276
8277 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8278 trace_seq_printf(s, "overrun: %ld\n", cnt);
8279
8280 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8281 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8282
8283 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8284 trace_seq_printf(s, "bytes: %ld\n", cnt);
8285
8286 if (trace_clocks[tr->clock_id].in_ns) {
8287 /* local or global for trace_clock */
8288 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8289 usec_rem = do_div(t, USEC_PER_SEC);
8290 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8291 t, usec_rem);
8292
8293 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8294 usec_rem = do_div(t, USEC_PER_SEC);
8295 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8296 } else {
8297 /* counter or tsc mode for trace_clock */
8298 trace_seq_printf(s, "oldest event ts: %llu\n",
8299 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8300
8301 trace_seq_printf(s, "now ts: %llu\n",
8302 ring_buffer_time_stamp(trace_buf->buffer));
8303 }
8304
8305 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8306 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8307
8308 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8309 trace_seq_printf(s, "read events: %ld\n", cnt);
8310
8311 count = simple_read_from_buffer(ubuf, count, ppos,
8312 s->buffer, trace_seq_used(s));
8313
8314 kfree(s);
8315
8316 return count;
8317}
8318
8319static const struct file_operations tracing_stats_fops = {
8320 .open = tracing_open_generic_tr,
8321 .read = tracing_stats_read,
8322 .llseek = generic_file_llseek,
8323 .release = tracing_release_generic_tr,
8324};
8325
8326#ifdef CONFIG_DYNAMIC_FTRACE
8327
8328static ssize_t
8329tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8330 size_t cnt, loff_t *ppos)
8331{
8332 ssize_t ret;
8333 char *buf;
8334 int r;
8335
8336 /* 256 should be plenty to hold the amount needed */
8337 buf = kmalloc(256, GFP_KERNEL);
8338 if (!buf)
8339 return -ENOMEM;
8340
8341 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8342 ftrace_update_tot_cnt,
8343 ftrace_number_of_pages,
8344 ftrace_number_of_groups);
8345
8346 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8347 kfree(buf);
8348 return ret;
8349}
8350
8351static const struct file_operations tracing_dyn_info_fops = {
8352 .open = tracing_open_generic,
8353 .read = tracing_read_dyn_info,
8354 .llseek = generic_file_llseek,
8355};
8356#endif /* CONFIG_DYNAMIC_FTRACE */
8357
8358#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8359static void
8360ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8361 struct trace_array *tr, struct ftrace_probe_ops *ops,
8362 void *data)
8363{
8364 tracing_snapshot_instance(tr);
8365}
8366
8367static void
8368ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8369 struct trace_array *tr, struct ftrace_probe_ops *ops,
8370 void *data)
8371{
8372 struct ftrace_func_mapper *mapper = data;
8373 long *count = NULL;
8374
8375 if (mapper)
8376 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8377
8378 if (count) {
8379
8380 if (*count <= 0)
8381 return;
8382
8383 (*count)--;
8384 }
8385
8386 tracing_snapshot_instance(tr);
8387}
8388
8389static int
8390ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8391 struct ftrace_probe_ops *ops, void *data)
8392{
8393 struct ftrace_func_mapper *mapper = data;
8394 long *count = NULL;
8395
8396 seq_printf(m, "%ps:", (void *)ip);
8397
8398 seq_puts(m, "snapshot");
8399
8400 if (mapper)
8401 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8402
8403 if (count)
8404 seq_printf(m, ":count=%ld\n", *count);
8405 else
8406 seq_puts(m, ":unlimited\n");
8407
8408 return 0;
8409}
8410
8411static int
8412ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8413 unsigned long ip, void *init_data, void **data)
8414{
8415 struct ftrace_func_mapper *mapper = *data;
8416
8417 if (!mapper) {
8418 mapper = allocate_ftrace_func_mapper();
8419 if (!mapper)
8420 return -ENOMEM;
8421 *data = mapper;
8422 }
8423
8424 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8425}
8426
8427static void
8428ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8429 unsigned long ip, void *data)
8430{
8431 struct ftrace_func_mapper *mapper = data;
8432
8433 if (!ip) {
8434 if (!mapper)
8435 return;
8436 free_ftrace_func_mapper(mapper, NULL);
8437 return;
8438 }
8439
8440 ftrace_func_mapper_remove_ip(mapper, ip);
8441}
8442
8443static struct ftrace_probe_ops snapshot_probe_ops = {
8444 .func = ftrace_snapshot,
8445 .print = ftrace_snapshot_print,
8446};
8447
8448static struct ftrace_probe_ops snapshot_count_probe_ops = {
8449 .func = ftrace_count_snapshot,
8450 .print = ftrace_snapshot_print,
8451 .init = ftrace_snapshot_init,
8452 .free = ftrace_snapshot_free,
8453};
8454
8455static int
8456ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8457 char *glob, char *cmd, char *param, int enable)
8458{
8459 struct ftrace_probe_ops *ops;
8460 void *count = (void *)-1;
8461 char *number;
8462 int ret;
8463
8464 if (!tr)
8465 return -ENODEV;
8466
8467 /* hash funcs only work with set_ftrace_filter */
8468 if (!enable)
8469 return -EINVAL;
8470
8471 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8472
8473 if (glob[0] == '!')
8474 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8475
8476 if (!param)
8477 goto out_reg;
8478
8479 number = strsep(¶m, ":");
8480
8481 if (!strlen(number))
8482 goto out_reg;
8483
8484 /*
8485 * We use the callback data field (which is a pointer)
8486 * as our counter.
8487 */
8488 ret = kstrtoul(number, 0, (unsigned long *)&count);
8489 if (ret)
8490 return ret;
8491
8492 out_reg:
8493 ret = tracing_alloc_snapshot_instance(tr);
8494 if (ret < 0)
8495 goto out;
8496
8497 ret = register_ftrace_function_probe(glob, tr, ops, count);
8498
8499 out:
8500 return ret < 0 ? ret : 0;
8501}
8502
8503static struct ftrace_func_command ftrace_snapshot_cmd = {
8504 .name = "snapshot",
8505 .func = ftrace_trace_snapshot_callback,
8506};
8507
8508static __init int register_snapshot_cmd(void)
8509{
8510 return register_ftrace_command(&ftrace_snapshot_cmd);
8511}
8512#else
8513static inline __init int register_snapshot_cmd(void) { return 0; }
8514#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8515
8516static struct dentry *tracing_get_dentry(struct trace_array *tr)
8517{
8518 if (WARN_ON(!tr->dir))
8519 return ERR_PTR(-ENODEV);
8520
8521 /* Top directory uses NULL as the parent */
8522 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8523 return NULL;
8524
8525 /* All sub buffers have a descriptor */
8526 return tr->dir;
8527}
8528
8529static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8530{
8531 struct dentry *d_tracer;
8532
8533 if (tr->percpu_dir)
8534 return tr->percpu_dir;
8535
8536 d_tracer = tracing_get_dentry(tr);
8537 if (IS_ERR(d_tracer))
8538 return NULL;
8539
8540 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8541
8542 MEM_FAIL(!tr->percpu_dir,
8543 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8544
8545 return tr->percpu_dir;
8546}
8547
8548static struct dentry *
8549trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8550 void *data, long cpu, const struct file_operations *fops)
8551{
8552 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8553
8554 if (ret) /* See tracing_get_cpu() */
8555 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8556 return ret;
8557}
8558
8559static void
8560tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8561{
8562 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8563 struct dentry *d_cpu;
8564 char cpu_dir[30]; /* 30 characters should be more than enough */
8565
8566 if (!d_percpu)
8567 return;
8568
8569 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8570 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8571 if (!d_cpu) {
8572 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8573 return;
8574 }
8575
8576 /* per cpu trace_pipe */
8577 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
8578 tr, cpu, &tracing_pipe_fops);
8579
8580 /* per cpu trace */
8581 trace_create_cpu_file("trace", 0644, d_cpu,
8582 tr, cpu, &tracing_fops);
8583
8584 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
8585 tr, cpu, &tracing_buffers_fops);
8586
8587 trace_create_cpu_file("stats", 0444, d_cpu,
8588 tr, cpu, &tracing_stats_fops);
8589
8590 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
8591 tr, cpu, &tracing_entries_fops);
8592
8593#ifdef CONFIG_TRACER_SNAPSHOT
8594 trace_create_cpu_file("snapshot", 0644, d_cpu,
8595 tr, cpu, &snapshot_fops);
8596
8597 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
8598 tr, cpu, &snapshot_raw_fops);
8599#endif
8600}
8601
8602#ifdef CONFIG_FTRACE_SELFTEST
8603/* Let selftest have access to static functions in this file */
8604#include "trace_selftest.c"
8605#endif
8606
8607static ssize_t
8608trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8609 loff_t *ppos)
8610{
8611 struct trace_option_dentry *topt = filp->private_data;
8612 char *buf;
8613
8614 if (topt->flags->val & topt->opt->bit)
8615 buf = "1\n";
8616 else
8617 buf = "0\n";
8618
8619 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8620}
8621
8622static ssize_t
8623trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8624 loff_t *ppos)
8625{
8626 struct trace_option_dentry *topt = filp->private_data;
8627 unsigned long val;
8628 int ret;
8629
8630 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8631 if (ret)
8632 return ret;
8633
8634 if (val != 0 && val != 1)
8635 return -EINVAL;
8636
8637 if (!!(topt->flags->val & topt->opt->bit) != val) {
8638 mutex_lock(&trace_types_lock);
8639 ret = __set_tracer_option(topt->tr, topt->flags,
8640 topt->opt, !val);
8641 mutex_unlock(&trace_types_lock);
8642 if (ret)
8643 return ret;
8644 }
8645
8646 *ppos += cnt;
8647
8648 return cnt;
8649}
8650
8651
8652static const struct file_operations trace_options_fops = {
8653 .open = tracing_open_generic,
8654 .read = trace_options_read,
8655 .write = trace_options_write,
8656 .llseek = generic_file_llseek,
8657};
8658
8659/*
8660 * In order to pass in both the trace_array descriptor as well as the index
8661 * to the flag that the trace option file represents, the trace_array
8662 * has a character array of trace_flags_index[], which holds the index
8663 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8664 * The address of this character array is passed to the flag option file
8665 * read/write callbacks.
8666 *
8667 * In order to extract both the index and the trace_array descriptor,
8668 * get_tr_index() uses the following algorithm.
8669 *
8670 * idx = *ptr;
8671 *
8672 * As the pointer itself contains the address of the index (remember
8673 * index[1] == 1).
8674 *
8675 * Then to get the trace_array descriptor, by subtracting that index
8676 * from the ptr, we get to the start of the index itself.
8677 *
8678 * ptr - idx == &index[0]
8679 *
8680 * Then a simple container_of() from that pointer gets us to the
8681 * trace_array descriptor.
8682 */
8683static void get_tr_index(void *data, struct trace_array **ptr,
8684 unsigned int *pindex)
8685{
8686 *pindex = *(unsigned char *)data;
8687
8688 *ptr = container_of(data - *pindex, struct trace_array,
8689 trace_flags_index);
8690}
8691
8692static ssize_t
8693trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8694 loff_t *ppos)
8695{
8696 void *tr_index = filp->private_data;
8697 struct trace_array *tr;
8698 unsigned int index;
8699 char *buf;
8700
8701 get_tr_index(tr_index, &tr, &index);
8702
8703 if (tr->trace_flags & (1 << index))
8704 buf = "1\n";
8705 else
8706 buf = "0\n";
8707
8708 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8709}
8710
8711static ssize_t
8712trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8713 loff_t *ppos)
8714{
8715 void *tr_index = filp->private_data;
8716 struct trace_array *tr;
8717 unsigned int index;
8718 unsigned long val;
8719 int ret;
8720
8721 get_tr_index(tr_index, &tr, &index);
8722
8723 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8724 if (ret)
8725 return ret;
8726
8727 if (val != 0 && val != 1)
8728 return -EINVAL;
8729
8730 mutex_lock(&event_mutex);
8731 mutex_lock(&trace_types_lock);
8732 ret = set_tracer_flag(tr, 1 << index, val);
8733 mutex_unlock(&trace_types_lock);
8734 mutex_unlock(&event_mutex);
8735
8736 if (ret < 0)
8737 return ret;
8738
8739 *ppos += cnt;
8740
8741 return cnt;
8742}
8743
8744static const struct file_operations trace_options_core_fops = {
8745 .open = tracing_open_generic,
8746 .read = trace_options_core_read,
8747 .write = trace_options_core_write,
8748 .llseek = generic_file_llseek,
8749};
8750
8751struct dentry *trace_create_file(const char *name,
8752 umode_t mode,
8753 struct dentry *parent,
8754 void *data,
8755 const struct file_operations *fops)
8756{
8757 struct dentry *ret;
8758
8759 ret = tracefs_create_file(name, mode, parent, data, fops);
8760 if (!ret)
8761 pr_warn("Could not create tracefs '%s' entry\n", name);
8762
8763 return ret;
8764}
8765
8766
8767static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8768{
8769 struct dentry *d_tracer;
8770
8771 if (tr->options)
8772 return tr->options;
8773
8774 d_tracer = tracing_get_dentry(tr);
8775 if (IS_ERR(d_tracer))
8776 return NULL;
8777
8778 tr->options = tracefs_create_dir("options", d_tracer);
8779 if (!tr->options) {
8780 pr_warn("Could not create tracefs directory 'options'\n");
8781 return NULL;
8782 }
8783
8784 return tr->options;
8785}
8786
8787static void
8788create_trace_option_file(struct trace_array *tr,
8789 struct trace_option_dentry *topt,
8790 struct tracer_flags *flags,
8791 struct tracer_opt *opt)
8792{
8793 struct dentry *t_options;
8794
8795 t_options = trace_options_init_dentry(tr);
8796 if (!t_options)
8797 return;
8798
8799 topt->flags = flags;
8800 topt->opt = opt;
8801 topt->tr = tr;
8802
8803 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
8804 &trace_options_fops);
8805
8806}
8807
8808static void
8809create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8810{
8811 struct trace_option_dentry *topts;
8812 struct trace_options *tr_topts;
8813 struct tracer_flags *flags;
8814 struct tracer_opt *opts;
8815 int cnt;
8816 int i;
8817
8818 if (!tracer)
8819 return;
8820
8821 flags = tracer->flags;
8822
8823 if (!flags || !flags->opts)
8824 return;
8825
8826 /*
8827 * If this is an instance, only create flags for tracers
8828 * the instance may have.
8829 */
8830 if (!trace_ok_for_array(tracer, tr))
8831 return;
8832
8833 for (i = 0; i < tr->nr_topts; i++) {
8834 /* Make sure there's no duplicate flags. */
8835 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8836 return;
8837 }
8838
8839 opts = flags->opts;
8840
8841 for (cnt = 0; opts[cnt].name; cnt++)
8842 ;
8843
8844 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8845 if (!topts)
8846 return;
8847
8848 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8849 GFP_KERNEL);
8850 if (!tr_topts) {
8851 kfree(topts);
8852 return;
8853 }
8854
8855 tr->topts = tr_topts;
8856 tr->topts[tr->nr_topts].tracer = tracer;
8857 tr->topts[tr->nr_topts].topts = topts;
8858 tr->nr_topts++;
8859
8860 for (cnt = 0; opts[cnt].name; cnt++) {
8861 create_trace_option_file(tr, &topts[cnt], flags,
8862 &opts[cnt]);
8863 MEM_FAIL(topts[cnt].entry == NULL,
8864 "Failed to create trace option: %s",
8865 opts[cnt].name);
8866 }
8867}
8868
8869static struct dentry *
8870create_trace_option_core_file(struct trace_array *tr,
8871 const char *option, long index)
8872{
8873 struct dentry *t_options;
8874
8875 t_options = trace_options_init_dentry(tr);
8876 if (!t_options)
8877 return NULL;
8878
8879 return trace_create_file(option, 0644, t_options,
8880 (void *)&tr->trace_flags_index[index],
8881 &trace_options_core_fops);
8882}
8883
8884static void create_trace_options_dir(struct trace_array *tr)
8885{
8886 struct dentry *t_options;
8887 bool top_level = tr == &global_trace;
8888 int i;
8889
8890 t_options = trace_options_init_dentry(tr);
8891 if (!t_options)
8892 return;
8893
8894 for (i = 0; trace_options[i]; i++) {
8895 if (top_level ||
8896 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8897 create_trace_option_core_file(tr, trace_options[i], i);
8898 }
8899}
8900
8901static ssize_t
8902rb_simple_read(struct file *filp, char __user *ubuf,
8903 size_t cnt, loff_t *ppos)
8904{
8905 struct trace_array *tr = filp->private_data;
8906 char buf[64];
8907 int r;
8908
8909 r = tracer_tracing_is_on(tr);
8910 r = sprintf(buf, "%d\n", r);
8911
8912 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8913}
8914
8915static ssize_t
8916rb_simple_write(struct file *filp, const char __user *ubuf,
8917 size_t cnt, loff_t *ppos)
8918{
8919 struct trace_array *tr = filp->private_data;
8920 struct trace_buffer *buffer = tr->array_buffer.buffer;
8921 unsigned long val;
8922 int ret;
8923
8924 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8925 if (ret)
8926 return ret;
8927
8928 if (buffer) {
8929 mutex_lock(&trace_types_lock);
8930 if (!!val == tracer_tracing_is_on(tr)) {
8931 val = 0; /* do nothing */
8932 } else if (val) {
8933 tracer_tracing_on(tr);
8934 if (tr->current_trace->start)
8935 tr->current_trace->start(tr);
8936 } else {
8937 tracer_tracing_off(tr);
8938 if (tr->current_trace->stop)
8939 tr->current_trace->stop(tr);
8940 }
8941 mutex_unlock(&trace_types_lock);
8942 }
8943
8944 (*ppos)++;
8945
8946 return cnt;
8947}
8948
8949static const struct file_operations rb_simple_fops = {
8950 .open = tracing_open_generic_tr,
8951 .read = rb_simple_read,
8952 .write = rb_simple_write,
8953 .release = tracing_release_generic_tr,
8954 .llseek = default_llseek,
8955};
8956
8957static ssize_t
8958buffer_percent_read(struct file *filp, char __user *ubuf,
8959 size_t cnt, loff_t *ppos)
8960{
8961 struct trace_array *tr = filp->private_data;
8962 char buf[64];
8963 int r;
8964
8965 r = tr->buffer_percent;
8966 r = sprintf(buf, "%d\n", r);
8967
8968 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8969}
8970
8971static ssize_t
8972buffer_percent_write(struct file *filp, const char __user *ubuf,
8973 size_t cnt, loff_t *ppos)
8974{
8975 struct trace_array *tr = filp->private_data;
8976 unsigned long val;
8977 int ret;
8978
8979 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8980 if (ret)
8981 return ret;
8982
8983 if (val > 100)
8984 return -EINVAL;
8985
8986 if (!val)
8987 val = 1;
8988
8989 tr->buffer_percent = val;
8990
8991 (*ppos)++;
8992
8993 return cnt;
8994}
8995
8996static const struct file_operations buffer_percent_fops = {
8997 .open = tracing_open_generic_tr,
8998 .read = buffer_percent_read,
8999 .write = buffer_percent_write,
9000 .release = tracing_release_generic_tr,
9001 .llseek = default_llseek,
9002};
9003
9004static struct dentry *trace_instance_dir;
9005
9006static void
9007init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9008
9009static int
9010allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9011{
9012 enum ring_buffer_flags rb_flags;
9013
9014 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9015
9016 buf->tr = tr;
9017
9018 buf->buffer = ring_buffer_alloc(size, rb_flags);
9019 if (!buf->buffer)
9020 return -ENOMEM;
9021
9022 buf->data = alloc_percpu(struct trace_array_cpu);
9023 if (!buf->data) {
9024 ring_buffer_free(buf->buffer);
9025 buf->buffer = NULL;
9026 return -ENOMEM;
9027 }
9028
9029 /* Allocate the first page for all buffers */
9030 set_buffer_entries(&tr->array_buffer,
9031 ring_buffer_size(tr->array_buffer.buffer, 0));
9032
9033 return 0;
9034}
9035
9036static int allocate_trace_buffers(struct trace_array *tr, int size)
9037{
9038 int ret;
9039
9040 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9041 if (ret)
9042 return ret;
9043
9044#ifdef CONFIG_TRACER_MAX_TRACE
9045 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9046 allocate_snapshot ? size : 1);
9047 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9048 ring_buffer_free(tr->array_buffer.buffer);
9049 tr->array_buffer.buffer = NULL;
9050 free_percpu(tr->array_buffer.data);
9051 tr->array_buffer.data = NULL;
9052 return -ENOMEM;
9053 }
9054 tr->allocated_snapshot = allocate_snapshot;
9055
9056 /*
9057 * Only the top level trace array gets its snapshot allocated
9058 * from the kernel command line.
9059 */
9060 allocate_snapshot = false;
9061#endif
9062
9063 return 0;
9064}
9065
9066static void free_trace_buffer(struct array_buffer *buf)
9067{
9068 if (buf->buffer) {
9069 ring_buffer_free(buf->buffer);
9070 buf->buffer = NULL;
9071 free_percpu(buf->data);
9072 buf->data = NULL;
9073 }
9074}
9075
9076static void free_trace_buffers(struct trace_array *tr)
9077{
9078 if (!tr)
9079 return;
9080
9081 free_trace_buffer(&tr->array_buffer);
9082
9083#ifdef CONFIG_TRACER_MAX_TRACE
9084 free_trace_buffer(&tr->max_buffer);
9085#endif
9086}
9087
9088static void init_trace_flags_index(struct trace_array *tr)
9089{
9090 int i;
9091
9092 /* Used by the trace options files */
9093 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9094 tr->trace_flags_index[i] = i;
9095}
9096
9097static void __update_tracer_options(struct trace_array *tr)
9098{
9099 struct tracer *t;
9100
9101 for (t = trace_types; t; t = t->next)
9102 add_tracer_options(tr, t);
9103}
9104
9105static void update_tracer_options(struct trace_array *tr)
9106{
9107 mutex_lock(&trace_types_lock);
9108 __update_tracer_options(tr);
9109 mutex_unlock(&trace_types_lock);
9110}
9111
9112/* Must have trace_types_lock held */
9113struct trace_array *trace_array_find(const char *instance)
9114{
9115 struct trace_array *tr, *found = NULL;
9116
9117 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9118 if (tr->name && strcmp(tr->name, instance) == 0) {
9119 found = tr;
9120 break;
9121 }
9122 }
9123
9124 return found;
9125}
9126
9127struct trace_array *trace_array_find_get(const char *instance)
9128{
9129 struct trace_array *tr;
9130
9131 mutex_lock(&trace_types_lock);
9132 tr = trace_array_find(instance);
9133 if (tr)
9134 tr->ref++;
9135 mutex_unlock(&trace_types_lock);
9136
9137 return tr;
9138}
9139
9140static int trace_array_create_dir(struct trace_array *tr)
9141{
9142 int ret;
9143
9144 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9145 if (!tr->dir)
9146 return -EINVAL;
9147
9148 ret = event_trace_add_tracer(tr->dir, tr);
9149 if (ret) {
9150 tracefs_remove(tr->dir);
9151 return ret;
9152 }
9153
9154 init_tracer_tracefs(tr, tr->dir);
9155 __update_tracer_options(tr);
9156
9157 return ret;
9158}
9159
9160static struct trace_array *trace_array_create(const char *name)
9161{
9162 struct trace_array *tr;
9163 int ret;
9164
9165 ret = -ENOMEM;
9166 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9167 if (!tr)
9168 return ERR_PTR(ret);
9169
9170 tr->name = kstrdup(name, GFP_KERNEL);
9171 if (!tr->name)
9172 goto out_free_tr;
9173
9174 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9175 goto out_free_tr;
9176
9177 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9178
9179 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9180
9181 raw_spin_lock_init(&tr->start_lock);
9182
9183 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9184
9185 tr->current_trace = &nop_trace;
9186
9187 INIT_LIST_HEAD(&tr->systems);
9188 INIT_LIST_HEAD(&tr->events);
9189 INIT_LIST_HEAD(&tr->hist_vars);
9190 INIT_LIST_HEAD(&tr->err_log);
9191
9192 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9193 goto out_free_tr;
9194
9195 if (ftrace_allocate_ftrace_ops(tr) < 0)
9196 goto out_free_tr;
9197
9198 ftrace_init_trace_array(tr);
9199
9200 init_trace_flags_index(tr);
9201
9202 if (trace_instance_dir) {
9203 ret = trace_array_create_dir(tr);
9204 if (ret)
9205 goto out_free_tr;
9206 } else
9207 __trace_early_add_events(tr);
9208
9209 list_add(&tr->list, &ftrace_trace_arrays);
9210
9211 tr->ref++;
9212
9213 return tr;
9214
9215 out_free_tr:
9216 ftrace_free_ftrace_ops(tr);
9217 free_trace_buffers(tr);
9218 free_cpumask_var(tr->tracing_cpumask);
9219 kfree(tr->name);
9220 kfree(tr);
9221
9222 return ERR_PTR(ret);
9223}
9224
9225static int instance_mkdir(const char *name)
9226{
9227 struct trace_array *tr;
9228 int ret;
9229
9230 mutex_lock(&event_mutex);
9231 mutex_lock(&trace_types_lock);
9232
9233 ret = -EEXIST;
9234 if (trace_array_find(name))
9235 goto out_unlock;
9236
9237 tr = trace_array_create(name);
9238
9239 ret = PTR_ERR_OR_ZERO(tr);
9240
9241out_unlock:
9242 mutex_unlock(&trace_types_lock);
9243 mutex_unlock(&event_mutex);
9244 return ret;
9245}
9246
9247/**
9248 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9249 * @name: The name of the trace array to be looked up/created.
9250 *
9251 * Returns pointer to trace array with given name.
9252 * NULL, if it cannot be created.
9253 *
9254 * NOTE: This function increments the reference counter associated with the
9255 * trace array returned. This makes sure it cannot be freed while in use.
9256 * Use trace_array_put() once the trace array is no longer needed.
9257 * If the trace_array is to be freed, trace_array_destroy() needs to
9258 * be called after the trace_array_put(), or simply let user space delete
9259 * it from the tracefs instances directory. But until the
9260 * trace_array_put() is called, user space can not delete it.
9261 *
9262 */
9263struct trace_array *trace_array_get_by_name(const char *name)
9264{
9265 struct trace_array *tr;
9266
9267 mutex_lock(&event_mutex);
9268 mutex_lock(&trace_types_lock);
9269
9270 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9271 if (tr->name && strcmp(tr->name, name) == 0)
9272 goto out_unlock;
9273 }
9274
9275 tr = trace_array_create(name);
9276
9277 if (IS_ERR(tr))
9278 tr = NULL;
9279out_unlock:
9280 if (tr)
9281 tr->ref++;
9282
9283 mutex_unlock(&trace_types_lock);
9284 mutex_unlock(&event_mutex);
9285 return tr;
9286}
9287EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9288
9289static int __remove_instance(struct trace_array *tr)
9290{
9291 int i;
9292
9293 /* Reference counter for a newly created trace array = 1. */
9294 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9295 return -EBUSY;
9296
9297 list_del(&tr->list);
9298
9299 /* Disable all the flags that were enabled coming in */
9300 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9301 if ((1 << i) & ZEROED_TRACE_FLAGS)
9302 set_tracer_flag(tr, 1 << i, 0);
9303 }
9304
9305 tracing_set_nop(tr);
9306 clear_ftrace_function_probes(tr);
9307 event_trace_del_tracer(tr);
9308 ftrace_clear_pids(tr);
9309 ftrace_destroy_function_files(tr);
9310 tracefs_remove(tr->dir);
9311 free_percpu(tr->last_func_repeats);
9312 free_trace_buffers(tr);
9313
9314 for (i = 0; i < tr->nr_topts; i++) {
9315 kfree(tr->topts[i].topts);
9316 }
9317 kfree(tr->topts);
9318
9319 free_cpumask_var(tr->tracing_cpumask);
9320 kfree(tr->name);
9321 kfree(tr);
9322
9323 return 0;
9324}
9325
9326int trace_array_destroy(struct trace_array *this_tr)
9327{
9328 struct trace_array *tr;
9329 int ret;
9330
9331 if (!this_tr)
9332 return -EINVAL;
9333
9334 mutex_lock(&event_mutex);
9335 mutex_lock(&trace_types_lock);
9336
9337 ret = -ENODEV;
9338
9339 /* Making sure trace array exists before destroying it. */
9340 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9341 if (tr == this_tr) {
9342 ret = __remove_instance(tr);
9343 break;
9344 }
9345 }
9346
9347 mutex_unlock(&trace_types_lock);
9348 mutex_unlock(&event_mutex);
9349
9350 return ret;
9351}
9352EXPORT_SYMBOL_GPL(trace_array_destroy);
9353
9354static int instance_rmdir(const char *name)
9355{
9356 struct trace_array *tr;
9357 int ret;
9358
9359 mutex_lock(&event_mutex);
9360 mutex_lock(&trace_types_lock);
9361
9362 ret = -ENODEV;
9363 tr = trace_array_find(name);
9364 if (tr)
9365 ret = __remove_instance(tr);
9366
9367 mutex_unlock(&trace_types_lock);
9368 mutex_unlock(&event_mutex);
9369
9370 return ret;
9371}
9372
9373static __init void create_trace_instances(struct dentry *d_tracer)
9374{
9375 struct trace_array *tr;
9376
9377 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9378 instance_mkdir,
9379 instance_rmdir);
9380 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9381 return;
9382
9383 mutex_lock(&event_mutex);
9384 mutex_lock(&trace_types_lock);
9385
9386 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9387 if (!tr->name)
9388 continue;
9389 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9390 "Failed to create instance directory\n"))
9391 break;
9392 }
9393
9394 mutex_unlock(&trace_types_lock);
9395 mutex_unlock(&event_mutex);
9396}
9397
9398static void
9399init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9400{
9401 struct trace_event_file *file;
9402 int cpu;
9403
9404 trace_create_file("available_tracers", 0444, d_tracer,
9405 tr, &show_traces_fops);
9406
9407 trace_create_file("current_tracer", 0644, d_tracer,
9408 tr, &set_tracer_fops);
9409
9410 trace_create_file("tracing_cpumask", 0644, d_tracer,
9411 tr, &tracing_cpumask_fops);
9412
9413 trace_create_file("trace_options", 0644, d_tracer,
9414 tr, &tracing_iter_fops);
9415
9416 trace_create_file("trace", 0644, d_tracer,
9417 tr, &tracing_fops);
9418
9419 trace_create_file("trace_pipe", 0444, d_tracer,
9420 tr, &tracing_pipe_fops);
9421
9422 trace_create_file("buffer_size_kb", 0644, d_tracer,
9423 tr, &tracing_entries_fops);
9424
9425 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
9426 tr, &tracing_total_entries_fops);
9427
9428 trace_create_file("free_buffer", 0200, d_tracer,
9429 tr, &tracing_free_buffer_fops);
9430
9431 trace_create_file("trace_marker", 0220, d_tracer,
9432 tr, &tracing_mark_fops);
9433
9434 file = __find_event_file(tr, "ftrace", "print");
9435 if (file && file->dir)
9436 trace_create_file("trigger", 0644, file->dir, file,
9437 &event_trigger_fops);
9438 tr->trace_marker_file = file;
9439
9440 trace_create_file("trace_marker_raw", 0220, d_tracer,
9441 tr, &tracing_mark_raw_fops);
9442
9443 trace_create_file("trace_clock", 0644, d_tracer, tr,
9444 &trace_clock_fops);
9445
9446 trace_create_file("tracing_on", 0644, d_tracer,
9447 tr, &rb_simple_fops);
9448
9449 trace_create_file("timestamp_mode", 0444, d_tracer, tr,
9450 &trace_time_stamp_mode_fops);
9451
9452 tr->buffer_percent = 50;
9453
9454 trace_create_file("buffer_percent", 0444, d_tracer,
9455 tr, &buffer_percent_fops);
9456
9457 create_trace_options_dir(tr);
9458
9459 trace_create_maxlat_file(tr, d_tracer);
9460
9461 if (ftrace_create_function_files(tr, d_tracer))
9462 MEM_FAIL(1, "Could not allocate function filter files");
9463
9464#ifdef CONFIG_TRACER_SNAPSHOT
9465 trace_create_file("snapshot", 0644, d_tracer,
9466 tr, &snapshot_fops);
9467#endif
9468
9469 trace_create_file("error_log", 0644, d_tracer,
9470 tr, &tracing_err_log_fops);
9471
9472 for_each_tracing_cpu(cpu)
9473 tracing_init_tracefs_percpu(tr, cpu);
9474
9475 ftrace_init_tracefs(tr, d_tracer);
9476}
9477
9478static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9479{
9480 struct vfsmount *mnt;
9481 struct file_system_type *type;
9482
9483 /*
9484 * To maintain backward compatibility for tools that mount
9485 * debugfs to get to the tracing facility, tracefs is automatically
9486 * mounted to the debugfs/tracing directory.
9487 */
9488 type = get_fs_type("tracefs");
9489 if (!type)
9490 return NULL;
9491 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9492 put_filesystem(type);
9493 if (IS_ERR(mnt))
9494 return NULL;
9495 mntget(mnt);
9496
9497 return mnt;
9498}
9499
9500/**
9501 * tracing_init_dentry - initialize top level trace array
9502 *
9503 * This is called when creating files or directories in the tracing
9504 * directory. It is called via fs_initcall() by any of the boot up code
9505 * and expects to return the dentry of the top level tracing directory.
9506 */
9507int tracing_init_dentry(void)
9508{
9509 struct trace_array *tr = &global_trace;
9510
9511 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9512 pr_warn("Tracing disabled due to lockdown\n");
9513 return -EPERM;
9514 }
9515
9516 /* The top level trace array uses NULL as parent */
9517 if (tr->dir)
9518 return 0;
9519
9520 if (WARN_ON(!tracefs_initialized()))
9521 return -ENODEV;
9522
9523 /*
9524 * As there may still be users that expect the tracing
9525 * files to exist in debugfs/tracing, we must automount
9526 * the tracefs file system there, so older tools still
9527 * work with the newer kernel.
9528 */
9529 tr->dir = debugfs_create_automount("tracing", NULL,
9530 trace_automount, NULL);
9531
9532 return 0;
9533}
9534
9535extern struct trace_eval_map *__start_ftrace_eval_maps[];
9536extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9537
9538static struct workqueue_struct *eval_map_wq __initdata;
9539static struct work_struct eval_map_work __initdata;
9540
9541static void __init eval_map_work_func(struct work_struct *work)
9542{
9543 int len;
9544
9545 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9546 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9547}
9548
9549static int __init trace_eval_init(void)
9550{
9551 INIT_WORK(&eval_map_work, eval_map_work_func);
9552
9553 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9554 if (!eval_map_wq) {
9555 pr_err("Unable to allocate eval_map_wq\n");
9556 /* Do work here */
9557 eval_map_work_func(&eval_map_work);
9558 return -ENOMEM;
9559 }
9560
9561 queue_work(eval_map_wq, &eval_map_work);
9562 return 0;
9563}
9564
9565static int __init trace_eval_sync(void)
9566{
9567 /* Make sure the eval map updates are finished */
9568 if (eval_map_wq)
9569 destroy_workqueue(eval_map_wq);
9570 return 0;
9571}
9572
9573late_initcall_sync(trace_eval_sync);
9574
9575
9576#ifdef CONFIG_MODULES
9577static void trace_module_add_evals(struct module *mod)
9578{
9579 if (!mod->num_trace_evals)
9580 return;
9581
9582 /*
9583 * Modules with bad taint do not have events created, do
9584 * not bother with enums either.
9585 */
9586 if (trace_module_has_bad_taint(mod))
9587 return;
9588
9589 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9590}
9591
9592#ifdef CONFIG_TRACE_EVAL_MAP_FILE
9593static void trace_module_remove_evals(struct module *mod)
9594{
9595 union trace_eval_map_item *map;
9596 union trace_eval_map_item **last = &trace_eval_maps;
9597
9598 if (!mod->num_trace_evals)
9599 return;
9600
9601 mutex_lock(&trace_eval_mutex);
9602
9603 map = trace_eval_maps;
9604
9605 while (map) {
9606 if (map->head.mod == mod)
9607 break;
9608 map = trace_eval_jmp_to_tail(map);
9609 last = &map->tail.next;
9610 map = map->tail.next;
9611 }
9612 if (!map)
9613 goto out;
9614
9615 *last = trace_eval_jmp_to_tail(map)->tail.next;
9616 kfree(map);
9617 out:
9618 mutex_unlock(&trace_eval_mutex);
9619}
9620#else
9621static inline void trace_module_remove_evals(struct module *mod) { }
9622#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9623
9624static int trace_module_notify(struct notifier_block *self,
9625 unsigned long val, void *data)
9626{
9627 struct module *mod = data;
9628
9629 switch (val) {
9630 case MODULE_STATE_COMING:
9631 trace_module_add_evals(mod);
9632 break;
9633 case MODULE_STATE_GOING:
9634 trace_module_remove_evals(mod);
9635 break;
9636 }
9637
9638 return NOTIFY_OK;
9639}
9640
9641static struct notifier_block trace_module_nb = {
9642 .notifier_call = trace_module_notify,
9643 .priority = 0,
9644};
9645#endif /* CONFIG_MODULES */
9646
9647static __init int tracer_init_tracefs(void)
9648{
9649 int ret;
9650
9651 trace_access_lock_init();
9652
9653 ret = tracing_init_dentry();
9654 if (ret)
9655 return 0;
9656
9657 event_trace_init();
9658
9659 init_tracer_tracefs(&global_trace, NULL);
9660 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9661
9662 trace_create_file("tracing_thresh", 0644, NULL,
9663 &global_trace, &tracing_thresh_fops);
9664
9665 trace_create_file("README", 0444, NULL,
9666 NULL, &tracing_readme_fops);
9667
9668 trace_create_file("saved_cmdlines", 0444, NULL,
9669 NULL, &tracing_saved_cmdlines_fops);
9670
9671 trace_create_file("saved_cmdlines_size", 0644, NULL,
9672 NULL, &tracing_saved_cmdlines_size_fops);
9673
9674 trace_create_file("saved_tgids", 0444, NULL,
9675 NULL, &tracing_saved_tgids_fops);
9676
9677 trace_eval_init();
9678
9679 trace_create_eval_file(NULL);
9680
9681#ifdef CONFIG_MODULES
9682 register_module_notifier(&trace_module_nb);
9683#endif
9684
9685#ifdef CONFIG_DYNAMIC_FTRACE
9686 trace_create_file("dyn_ftrace_total_info", 0444, NULL,
9687 NULL, &tracing_dyn_info_fops);
9688#endif
9689
9690 create_trace_instances(NULL);
9691
9692 update_tracer_options(&global_trace);
9693
9694 return 0;
9695}
9696
9697fs_initcall(tracer_init_tracefs);
9698
9699static int trace_panic_handler(struct notifier_block *this,
9700 unsigned long event, void *unused)
9701{
9702 if (ftrace_dump_on_oops)
9703 ftrace_dump(ftrace_dump_on_oops);
9704 return NOTIFY_OK;
9705}
9706
9707static struct notifier_block trace_panic_notifier = {
9708 .notifier_call = trace_panic_handler,
9709 .next = NULL,
9710 .priority = 150 /* priority: INT_MAX >= x >= 0 */
9711};
9712
9713static int trace_die_handler(struct notifier_block *self,
9714 unsigned long val,
9715 void *data)
9716{
9717 switch (val) {
9718 case DIE_OOPS:
9719 if (ftrace_dump_on_oops)
9720 ftrace_dump(ftrace_dump_on_oops);
9721 break;
9722 default:
9723 break;
9724 }
9725 return NOTIFY_OK;
9726}
9727
9728static struct notifier_block trace_die_notifier = {
9729 .notifier_call = trace_die_handler,
9730 .priority = 200
9731};
9732
9733/*
9734 * printk is set to max of 1024, we really don't need it that big.
9735 * Nothing should be printing 1000 characters anyway.
9736 */
9737#define TRACE_MAX_PRINT 1000
9738
9739/*
9740 * Define here KERN_TRACE so that we have one place to modify
9741 * it if we decide to change what log level the ftrace dump
9742 * should be at.
9743 */
9744#define KERN_TRACE KERN_EMERG
9745
9746void
9747trace_printk_seq(struct trace_seq *s)
9748{
9749 /* Probably should print a warning here. */
9750 if (s->seq.len >= TRACE_MAX_PRINT)
9751 s->seq.len = TRACE_MAX_PRINT;
9752
9753 /*
9754 * More paranoid code. Although the buffer size is set to
9755 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9756 * an extra layer of protection.
9757 */
9758 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9759 s->seq.len = s->seq.size - 1;
9760
9761 /* should be zero ended, but we are paranoid. */
9762 s->buffer[s->seq.len] = 0;
9763
9764 printk(KERN_TRACE "%s", s->buffer);
9765
9766 trace_seq_init(s);
9767}
9768
9769void trace_init_global_iter(struct trace_iterator *iter)
9770{
9771 iter->tr = &global_trace;
9772 iter->trace = iter->tr->current_trace;
9773 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9774 iter->array_buffer = &global_trace.array_buffer;
9775
9776 if (iter->trace && iter->trace->open)
9777 iter->trace->open(iter);
9778
9779 /* Annotate start of buffers if we had overruns */
9780 if (ring_buffer_overruns(iter->array_buffer->buffer))
9781 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9782
9783 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9784 if (trace_clocks[iter->tr->clock_id].in_ns)
9785 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9786}
9787
9788void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
9789{
9790 /* use static because iter can be a bit big for the stack */
9791 static struct trace_iterator iter;
9792 static atomic_t dump_running;
9793 struct trace_array *tr = &global_trace;
9794 unsigned int old_userobj;
9795 unsigned long flags;
9796 int cnt = 0, cpu;
9797
9798 /* Only allow one dump user at a time. */
9799 if (atomic_inc_return(&dump_running) != 1) {
9800 atomic_dec(&dump_running);
9801 return;
9802 }
9803
9804 /*
9805 * Always turn off tracing when we dump.
9806 * We don't need to show trace output of what happens
9807 * between multiple crashes.
9808 *
9809 * If the user does a sysrq-z, then they can re-enable
9810 * tracing with echo 1 > tracing_on.
9811 */
9812 tracing_off();
9813
9814 local_irq_save(flags);
9815 printk_nmi_direct_enter();
9816
9817 /* Simulate the iterator */
9818 trace_init_global_iter(&iter);
9819 /* Can not use kmalloc for iter.temp and iter.fmt */
9820 iter.temp = static_temp_buf;
9821 iter.temp_size = STATIC_TEMP_BUF_SIZE;
9822 iter.fmt = static_fmt_buf;
9823 iter.fmt_size = STATIC_FMT_BUF_SIZE;
9824
9825 for_each_tracing_cpu(cpu) {
9826 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9827 }
9828
9829 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
9830
9831 /* don't look at user memory in panic mode */
9832 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
9833
9834 switch (oops_dump_mode) {
9835 case DUMP_ALL:
9836 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9837 break;
9838 case DUMP_ORIG:
9839 iter.cpu_file = raw_smp_processor_id();
9840 break;
9841 case DUMP_NONE:
9842 goto out_enable;
9843 default:
9844 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
9845 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9846 }
9847
9848 printk(KERN_TRACE "Dumping ftrace buffer:\n");
9849
9850 /* Did function tracer already get disabled? */
9851 if (ftrace_is_dead()) {
9852 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
9853 printk("# MAY BE MISSING FUNCTION EVENTS\n");
9854 }
9855
9856 /*
9857 * We need to stop all tracing on all CPUS to read
9858 * the next buffer. This is a bit expensive, but is
9859 * not done often. We fill all what we can read,
9860 * and then release the locks again.
9861 */
9862
9863 while (!trace_empty(&iter)) {
9864
9865 if (!cnt)
9866 printk(KERN_TRACE "---------------------------------\n");
9867
9868 cnt++;
9869
9870 trace_iterator_reset(&iter);
9871 iter.iter_flags |= TRACE_FILE_LAT_FMT;
9872
9873 if (trace_find_next_entry_inc(&iter) != NULL) {
9874 int ret;
9875
9876 ret = print_trace_line(&iter);
9877 if (ret != TRACE_TYPE_NO_CONSUME)
9878 trace_consume(&iter);
9879 }
9880 touch_nmi_watchdog();
9881
9882 trace_printk_seq(&iter.seq);
9883 }
9884
9885 if (!cnt)
9886 printk(KERN_TRACE " (ftrace buffer empty)\n");
9887 else
9888 printk(KERN_TRACE "---------------------------------\n");
9889
9890 out_enable:
9891 tr->trace_flags |= old_userobj;
9892
9893 for_each_tracing_cpu(cpu) {
9894 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9895 }
9896 atomic_dec(&dump_running);
9897 printk_nmi_direct_exit();
9898 local_irq_restore(flags);
9899}
9900EXPORT_SYMBOL_GPL(ftrace_dump);
9901
9902#define WRITE_BUFSIZE 4096
9903
9904ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
9905 size_t count, loff_t *ppos,
9906 int (*createfn)(const char *))
9907{
9908 char *kbuf, *buf, *tmp;
9909 int ret = 0;
9910 size_t done = 0;
9911 size_t size;
9912
9913 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
9914 if (!kbuf)
9915 return -ENOMEM;
9916
9917 while (done < count) {
9918 size = count - done;
9919
9920 if (size >= WRITE_BUFSIZE)
9921 size = WRITE_BUFSIZE - 1;
9922
9923 if (copy_from_user(kbuf, buffer + done, size)) {
9924 ret = -EFAULT;
9925 goto out;
9926 }
9927 kbuf[size] = '\0';
9928 buf = kbuf;
9929 do {
9930 tmp = strchr(buf, '\n');
9931 if (tmp) {
9932 *tmp = '\0';
9933 size = tmp - buf + 1;
9934 } else {
9935 size = strlen(buf);
9936 if (done + size < count) {
9937 if (buf != kbuf)
9938 break;
9939 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
9940 pr_warn("Line length is too long: Should be less than %d\n",
9941 WRITE_BUFSIZE - 2);
9942 ret = -EINVAL;
9943 goto out;
9944 }
9945 }
9946 done += size;
9947
9948 /* Remove comments */
9949 tmp = strchr(buf, '#');
9950
9951 if (tmp)
9952 *tmp = '\0';
9953
9954 ret = createfn(buf);
9955 if (ret)
9956 goto out;
9957 buf += size;
9958
9959 } while (done < count);
9960 }
9961 ret = done;
9962
9963out:
9964 kfree(kbuf);
9965
9966 return ret;
9967}
9968
9969__init static int tracer_alloc_buffers(void)
9970{
9971 int ring_buf_size;
9972 int ret = -ENOMEM;
9973
9974
9975 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9976 pr_warn("Tracing disabled due to lockdown\n");
9977 return -EPERM;
9978 }
9979
9980 /*
9981 * Make sure we don't accidentally add more trace options
9982 * than we have bits for.
9983 */
9984 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
9985
9986 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
9987 goto out;
9988
9989 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
9990 goto out_free_buffer_mask;
9991
9992 /* Only allocate trace_printk buffers if a trace_printk exists */
9993 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
9994 /* Must be called before global_trace.buffer is allocated */
9995 trace_printk_init_buffers();
9996
9997 /* To save memory, keep the ring buffer size to its minimum */
9998 if (ring_buffer_expanded)
9999 ring_buf_size = trace_buf_size;
10000 else
10001 ring_buf_size = 1;
10002
10003 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10004 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10005
10006 raw_spin_lock_init(&global_trace.start_lock);
10007
10008 /*
10009 * The prepare callbacks allocates some memory for the ring buffer. We
10010 * don't free the buffer if the CPU goes down. If we were to free
10011 * the buffer, then the user would lose any trace that was in the
10012 * buffer. The memory will be removed once the "instance" is removed.
10013 */
10014 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10015 "trace/RB:preapre", trace_rb_cpu_prepare,
10016 NULL);
10017 if (ret < 0)
10018 goto out_free_cpumask;
10019 /* Used for event triggers */
10020 ret = -ENOMEM;
10021 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10022 if (!temp_buffer)
10023 goto out_rm_hp_state;
10024
10025 if (trace_create_savedcmd() < 0)
10026 goto out_free_temp_buffer;
10027
10028 /* TODO: make the number of buffers hot pluggable with CPUS */
10029 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10030 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10031 goto out_free_savedcmd;
10032 }
10033
10034 if (global_trace.buffer_disabled)
10035 tracing_off();
10036
10037 if (trace_boot_clock) {
10038 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10039 if (ret < 0)
10040 pr_warn("Trace clock %s not defined, going back to default\n",
10041 trace_boot_clock);
10042 }
10043
10044 /*
10045 * register_tracer() might reference current_trace, so it
10046 * needs to be set before we register anything. This is
10047 * just a bootstrap of current_trace anyway.
10048 */
10049 global_trace.current_trace = &nop_trace;
10050
10051 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10052
10053 ftrace_init_global_array_ops(&global_trace);
10054
10055 init_trace_flags_index(&global_trace);
10056
10057 register_tracer(&nop_trace);
10058
10059 /* Function tracing may start here (via kernel command line) */
10060 init_function_trace();
10061
10062 /* All seems OK, enable tracing */
10063 tracing_disabled = 0;
10064
10065 atomic_notifier_chain_register(&panic_notifier_list,
10066 &trace_panic_notifier);
10067
10068 register_die_notifier(&trace_die_notifier);
10069
10070 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10071
10072 INIT_LIST_HEAD(&global_trace.systems);
10073 INIT_LIST_HEAD(&global_trace.events);
10074 INIT_LIST_HEAD(&global_trace.hist_vars);
10075 INIT_LIST_HEAD(&global_trace.err_log);
10076 list_add(&global_trace.list, &ftrace_trace_arrays);
10077
10078 apply_trace_boot_options();
10079
10080 register_snapshot_cmd();
10081
10082 test_can_verify();
10083
10084 return 0;
10085
10086out_free_savedcmd:
10087 free_saved_cmdlines_buffer(savedcmd);
10088out_free_temp_buffer:
10089 ring_buffer_free(temp_buffer);
10090out_rm_hp_state:
10091 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10092out_free_cpumask:
10093 free_cpumask_var(global_trace.tracing_cpumask);
10094out_free_buffer_mask:
10095 free_cpumask_var(tracing_buffer_mask);
10096out:
10097 return ret;
10098}
10099
10100void __init early_trace_init(void)
10101{
10102 if (tracepoint_printk) {
10103 tracepoint_print_iter =
10104 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10105 if (MEM_FAIL(!tracepoint_print_iter,
10106 "Failed to allocate trace iterator\n"))
10107 tracepoint_printk = 0;
10108 else
10109 static_key_enable(&tracepoint_printk_key.key);
10110 }
10111 tracer_alloc_buffers();
10112}
10113
10114void __init trace_init(void)
10115{
10116 trace_event_init();
10117}
10118
10119__init static void clear_boot_tracer(void)
10120{
10121 /*
10122 * The default tracer at boot buffer is an init section.
10123 * This function is called in lateinit. If we did not
10124 * find the boot tracer, then clear it out, to prevent
10125 * later registration from accessing the buffer that is
10126 * about to be freed.
10127 */
10128 if (!default_bootup_tracer)
10129 return;
10130
10131 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10132 default_bootup_tracer);
10133 default_bootup_tracer = NULL;
10134}
10135
10136#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10137__init static void tracing_set_default_clock(void)
10138{
10139 /* sched_clock_stable() is determined in late_initcall */
10140 if (!trace_boot_clock && !sched_clock_stable()) {
10141 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10142 pr_warn("Can not set tracing clock due to lockdown\n");
10143 return;
10144 }
10145
10146 printk(KERN_WARNING
10147 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10148 "If you want to keep using the local clock, then add:\n"
10149 " \"trace_clock=local\"\n"
10150 "on the kernel command line\n");
10151 tracing_set_clock(&global_trace, "global");
10152 }
10153}
10154#else
10155static inline void tracing_set_default_clock(void) { }
10156#endif
10157
10158__init static int late_trace_init(void)
10159{
10160 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10161 static_key_disable(&tracepoint_printk_key.key);
10162 tracepoint_printk = 0;
10163 }
10164
10165 tracing_set_default_clock();
10166 clear_boot_tracer();
10167 return 0;
10168}
10169
10170late_initcall_sync(late_trace_init);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * ring buffer based function tracer
4 *
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 *
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 *
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
14 */
15#include <linux/ring_buffer.h>
16#include <generated/utsrelease.h>
17#include <linux/stacktrace.h>
18#include <linux/writeback.h>
19#include <linux/kallsyms.h>
20#include <linux/security.h>
21#include <linux/seq_file.h>
22#include <linux/notifier.h>
23#include <linux/irqflags.h>
24#include <linux/debugfs.h>
25#include <linux/tracefs.h>
26#include <linux/pagemap.h>
27#include <linux/hardirq.h>
28#include <linux/linkage.h>
29#include <linux/uaccess.h>
30#include <linux/vmalloc.h>
31#include <linux/ftrace.h>
32#include <linux/module.h>
33#include <linux/percpu.h>
34#include <linux/splice.h>
35#include <linux/kdebug.h>
36#include <linux/string.h>
37#include <linux/mount.h>
38#include <linux/rwsem.h>
39#include <linux/slab.h>
40#include <linux/ctype.h>
41#include <linux/init.h>
42#include <linux/poll.h>
43#include <linux/nmi.h>
44#include <linux/fs.h>
45#include <linux/trace.h>
46#include <linux/sched/clock.h>
47#include <linux/sched/rt.h>
48#include <linux/fsnotify.h>
49#include <linux/irq_work.h>
50#include <linux/workqueue.h>
51
52#include "trace.h"
53#include "trace_output.h"
54
55/*
56 * On boot up, the ring buffer is set to the minimum size, so that
57 * we do not waste memory on systems that are not using tracing.
58 */
59bool ring_buffer_expanded;
60
61/*
62 * We need to change this state when a selftest is running.
63 * A selftest will lurk into the ring-buffer to count the
64 * entries inserted during the selftest although some concurrent
65 * insertions into the ring-buffer such as trace_printk could occurred
66 * at the same time, giving false positive or negative results.
67 */
68static bool __read_mostly tracing_selftest_running;
69
70/*
71 * If a tracer is running, we do not want to run SELFTEST.
72 */
73bool __read_mostly tracing_selftest_disabled;
74
75/* Pipe tracepoints to printk */
76struct trace_iterator *tracepoint_print_iter;
77int tracepoint_printk;
78static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
79
80/* For tracers that don't implement custom flags */
81static struct tracer_opt dummy_tracer_opt[] = {
82 { }
83};
84
85static int
86dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
87{
88 return 0;
89}
90
91/*
92 * To prevent the comm cache from being overwritten when no
93 * tracing is active, only save the comm when a trace event
94 * occurred.
95 */
96static DEFINE_PER_CPU(bool, trace_taskinfo_save);
97
98/*
99 * Kill all tracing for good (never come back).
100 * It is initialized to 1 but will turn to zero if the initialization
101 * of the tracer is successful. But that is the only place that sets
102 * this back to zero.
103 */
104static int tracing_disabled = 1;
105
106cpumask_var_t __read_mostly tracing_buffer_mask;
107
108/*
109 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
110 *
111 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
112 * is set, then ftrace_dump is called. This will output the contents
113 * of the ftrace buffers to the console. This is very useful for
114 * capturing traces that lead to crashes and outputing it to a
115 * serial console.
116 *
117 * It is default off, but you can enable it with either specifying
118 * "ftrace_dump_on_oops" in the kernel command line, or setting
119 * /proc/sys/kernel/ftrace_dump_on_oops
120 * Set 1 if you want to dump buffers of all CPUs
121 * Set 2 if you want to dump the buffer of the CPU that triggered oops
122 */
123
124enum ftrace_dump_mode ftrace_dump_on_oops;
125
126/* When set, tracing will stop when a WARN*() is hit */
127int __disable_trace_on_warning;
128
129#ifdef CONFIG_TRACE_EVAL_MAP_FILE
130/* Map of enums to their values, for "eval_map" file */
131struct trace_eval_map_head {
132 struct module *mod;
133 unsigned long length;
134};
135
136union trace_eval_map_item;
137
138struct trace_eval_map_tail {
139 /*
140 * "end" is first and points to NULL as it must be different
141 * than "mod" or "eval_string"
142 */
143 union trace_eval_map_item *next;
144 const char *end; /* points to NULL */
145};
146
147static DEFINE_MUTEX(trace_eval_mutex);
148
149/*
150 * The trace_eval_maps are saved in an array with two extra elements,
151 * one at the beginning, and one at the end. The beginning item contains
152 * the count of the saved maps (head.length), and the module they
153 * belong to if not built in (head.mod). The ending item contains a
154 * pointer to the next array of saved eval_map items.
155 */
156union trace_eval_map_item {
157 struct trace_eval_map map;
158 struct trace_eval_map_head head;
159 struct trace_eval_map_tail tail;
160};
161
162static union trace_eval_map_item *trace_eval_maps;
163#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
164
165int tracing_set_tracer(struct trace_array *tr, const char *buf);
166static void ftrace_trace_userstack(struct trace_buffer *buffer,
167 unsigned long flags, int pc);
168
169#define MAX_TRACER_SIZE 100
170static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
171static char *default_bootup_tracer;
172
173static bool allocate_snapshot;
174
175static int __init set_cmdline_ftrace(char *str)
176{
177 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
178 default_bootup_tracer = bootup_tracer_buf;
179 /* We are using ftrace early, expand it */
180 ring_buffer_expanded = true;
181 return 1;
182}
183__setup("ftrace=", set_cmdline_ftrace);
184
185static int __init set_ftrace_dump_on_oops(char *str)
186{
187 if (*str++ != '=' || !*str) {
188 ftrace_dump_on_oops = DUMP_ALL;
189 return 1;
190 }
191
192 if (!strcmp("orig_cpu", str)) {
193 ftrace_dump_on_oops = DUMP_ORIG;
194 return 1;
195 }
196
197 return 0;
198}
199__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
200
201static int __init stop_trace_on_warning(char *str)
202{
203 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
204 __disable_trace_on_warning = 1;
205 return 1;
206}
207__setup("traceoff_on_warning", stop_trace_on_warning);
208
209static int __init boot_alloc_snapshot(char *str)
210{
211 allocate_snapshot = true;
212 /* We also need the main ring buffer expanded */
213 ring_buffer_expanded = true;
214 return 1;
215}
216__setup("alloc_snapshot", boot_alloc_snapshot);
217
218
219static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
220
221static int __init set_trace_boot_options(char *str)
222{
223 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
224 return 0;
225}
226__setup("trace_options=", set_trace_boot_options);
227
228static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
229static char *trace_boot_clock __initdata;
230
231static int __init set_trace_boot_clock(char *str)
232{
233 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
234 trace_boot_clock = trace_boot_clock_buf;
235 return 0;
236}
237__setup("trace_clock=", set_trace_boot_clock);
238
239static int __init set_tracepoint_printk(char *str)
240{
241 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
242 tracepoint_printk = 1;
243 return 1;
244}
245__setup("tp_printk", set_tracepoint_printk);
246
247unsigned long long ns2usecs(u64 nsec)
248{
249 nsec += 500;
250 do_div(nsec, 1000);
251 return nsec;
252}
253
254/* trace_flags holds trace_options default values */
255#define TRACE_DEFAULT_FLAGS \
256 (FUNCTION_DEFAULT_FLAGS | \
257 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
258 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
259 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
260 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS)
261
262/* trace_options that are only supported by global_trace */
263#define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
264 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
265
266/* trace_flags that are default zero for instances */
267#define ZEROED_TRACE_FLAGS \
268 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
269
270/*
271 * The global_trace is the descriptor that holds the top-level tracing
272 * buffers for the live tracing.
273 */
274static struct trace_array global_trace = {
275 .trace_flags = TRACE_DEFAULT_FLAGS,
276};
277
278LIST_HEAD(ftrace_trace_arrays);
279
280int trace_array_get(struct trace_array *this_tr)
281{
282 struct trace_array *tr;
283 int ret = -ENODEV;
284
285 mutex_lock(&trace_types_lock);
286 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
287 if (tr == this_tr) {
288 tr->ref++;
289 ret = 0;
290 break;
291 }
292 }
293 mutex_unlock(&trace_types_lock);
294
295 return ret;
296}
297
298static void __trace_array_put(struct trace_array *this_tr)
299{
300 WARN_ON(!this_tr->ref);
301 this_tr->ref--;
302}
303
304/**
305 * trace_array_put - Decrement the reference counter for this trace array.
306 *
307 * NOTE: Use this when we no longer need the trace array returned by
308 * trace_array_get_by_name(). This ensures the trace array can be later
309 * destroyed.
310 *
311 */
312void trace_array_put(struct trace_array *this_tr)
313{
314 if (!this_tr)
315 return;
316
317 mutex_lock(&trace_types_lock);
318 __trace_array_put(this_tr);
319 mutex_unlock(&trace_types_lock);
320}
321EXPORT_SYMBOL_GPL(trace_array_put);
322
323int tracing_check_open_get_tr(struct trace_array *tr)
324{
325 int ret;
326
327 ret = security_locked_down(LOCKDOWN_TRACEFS);
328 if (ret)
329 return ret;
330
331 if (tracing_disabled)
332 return -ENODEV;
333
334 if (tr && trace_array_get(tr) < 0)
335 return -ENODEV;
336
337 return 0;
338}
339
340int call_filter_check_discard(struct trace_event_call *call, void *rec,
341 struct trace_buffer *buffer,
342 struct ring_buffer_event *event)
343{
344 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
345 !filter_match_preds(call->filter, rec)) {
346 __trace_event_discard_commit(buffer, event);
347 return 1;
348 }
349
350 return 0;
351}
352
353void trace_free_pid_list(struct trace_pid_list *pid_list)
354{
355 vfree(pid_list->pids);
356 kfree(pid_list);
357}
358
359/**
360 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
361 * @filtered_pids: The list of pids to check
362 * @search_pid: The PID to find in @filtered_pids
363 *
364 * Returns true if @search_pid is fonud in @filtered_pids, and false otherwis.
365 */
366bool
367trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
368{
369 /*
370 * If pid_max changed after filtered_pids was created, we
371 * by default ignore all pids greater than the previous pid_max.
372 */
373 if (search_pid >= filtered_pids->pid_max)
374 return false;
375
376 return test_bit(search_pid, filtered_pids->pids);
377}
378
379/**
380 * trace_ignore_this_task - should a task be ignored for tracing
381 * @filtered_pids: The list of pids to check
382 * @task: The task that should be ignored if not filtered
383 *
384 * Checks if @task should be traced or not from @filtered_pids.
385 * Returns true if @task should *NOT* be traced.
386 * Returns false if @task should be traced.
387 */
388bool
389trace_ignore_this_task(struct trace_pid_list *filtered_pids,
390 struct trace_pid_list *filtered_no_pids,
391 struct task_struct *task)
392{
393 /*
394 * If filterd_no_pids is not empty, and the task's pid is listed
395 * in filtered_no_pids, then return true.
396 * Otherwise, if filtered_pids is empty, that means we can
397 * trace all tasks. If it has content, then only trace pids
398 * within filtered_pids.
399 */
400
401 return (filtered_pids &&
402 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
403 (filtered_no_pids &&
404 trace_find_filtered_pid(filtered_no_pids, task->pid));
405}
406
407/**
408 * trace_filter_add_remove_task - Add or remove a task from a pid_list
409 * @pid_list: The list to modify
410 * @self: The current task for fork or NULL for exit
411 * @task: The task to add or remove
412 *
413 * If adding a task, if @self is defined, the task is only added if @self
414 * is also included in @pid_list. This happens on fork and tasks should
415 * only be added when the parent is listed. If @self is NULL, then the
416 * @task pid will be removed from the list, which would happen on exit
417 * of a task.
418 */
419void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
420 struct task_struct *self,
421 struct task_struct *task)
422{
423 if (!pid_list)
424 return;
425
426 /* For forks, we only add if the forking task is listed */
427 if (self) {
428 if (!trace_find_filtered_pid(pid_list, self->pid))
429 return;
430 }
431
432 /* Sorry, but we don't support pid_max changing after setting */
433 if (task->pid >= pid_list->pid_max)
434 return;
435
436 /* "self" is set for forks, and NULL for exits */
437 if (self)
438 set_bit(task->pid, pid_list->pids);
439 else
440 clear_bit(task->pid, pid_list->pids);
441}
442
443/**
444 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
445 * @pid_list: The pid list to show
446 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
447 * @pos: The position of the file
448 *
449 * This is used by the seq_file "next" operation to iterate the pids
450 * listed in a trace_pid_list structure.
451 *
452 * Returns the pid+1 as we want to display pid of zero, but NULL would
453 * stop the iteration.
454 */
455void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
456{
457 unsigned long pid = (unsigned long)v;
458
459 (*pos)++;
460
461 /* pid already is +1 of the actual prevous bit */
462 pid = find_next_bit(pid_list->pids, pid_list->pid_max, pid);
463
464 /* Return pid + 1 to allow zero to be represented */
465 if (pid < pid_list->pid_max)
466 return (void *)(pid + 1);
467
468 return NULL;
469}
470
471/**
472 * trace_pid_start - Used for seq_file to start reading pid lists
473 * @pid_list: The pid list to show
474 * @pos: The position of the file
475 *
476 * This is used by seq_file "start" operation to start the iteration
477 * of listing pids.
478 *
479 * Returns the pid+1 as we want to display pid of zero, but NULL would
480 * stop the iteration.
481 */
482void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
483{
484 unsigned long pid;
485 loff_t l = 0;
486
487 pid = find_first_bit(pid_list->pids, pid_list->pid_max);
488 if (pid >= pid_list->pid_max)
489 return NULL;
490
491 /* Return pid + 1 so that zero can be the exit value */
492 for (pid++; pid && l < *pos;
493 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
494 ;
495 return (void *)pid;
496}
497
498/**
499 * trace_pid_show - show the current pid in seq_file processing
500 * @m: The seq_file structure to write into
501 * @v: A void pointer of the pid (+1) value to display
502 *
503 * Can be directly used by seq_file operations to display the current
504 * pid value.
505 */
506int trace_pid_show(struct seq_file *m, void *v)
507{
508 unsigned long pid = (unsigned long)v - 1;
509
510 seq_printf(m, "%lu\n", pid);
511 return 0;
512}
513
514/* 128 should be much more than enough */
515#define PID_BUF_SIZE 127
516
517int trace_pid_write(struct trace_pid_list *filtered_pids,
518 struct trace_pid_list **new_pid_list,
519 const char __user *ubuf, size_t cnt)
520{
521 struct trace_pid_list *pid_list;
522 struct trace_parser parser;
523 unsigned long val;
524 int nr_pids = 0;
525 ssize_t read = 0;
526 ssize_t ret = 0;
527 loff_t pos;
528 pid_t pid;
529
530 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
531 return -ENOMEM;
532
533 /*
534 * Always recreate a new array. The write is an all or nothing
535 * operation. Always create a new array when adding new pids by
536 * the user. If the operation fails, then the current list is
537 * not modified.
538 */
539 pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
540 if (!pid_list) {
541 trace_parser_put(&parser);
542 return -ENOMEM;
543 }
544
545 pid_list->pid_max = READ_ONCE(pid_max);
546
547 /* Only truncating will shrink pid_max */
548 if (filtered_pids && filtered_pids->pid_max > pid_list->pid_max)
549 pid_list->pid_max = filtered_pids->pid_max;
550
551 pid_list->pids = vzalloc((pid_list->pid_max + 7) >> 3);
552 if (!pid_list->pids) {
553 trace_parser_put(&parser);
554 kfree(pid_list);
555 return -ENOMEM;
556 }
557
558 if (filtered_pids) {
559 /* copy the current bits to the new max */
560 for_each_set_bit(pid, filtered_pids->pids,
561 filtered_pids->pid_max) {
562 set_bit(pid, pid_list->pids);
563 nr_pids++;
564 }
565 }
566
567 while (cnt > 0) {
568
569 pos = 0;
570
571 ret = trace_get_user(&parser, ubuf, cnt, &pos);
572 if (ret < 0 || !trace_parser_loaded(&parser))
573 break;
574
575 read += ret;
576 ubuf += ret;
577 cnt -= ret;
578
579 ret = -EINVAL;
580 if (kstrtoul(parser.buffer, 0, &val))
581 break;
582 if (val >= pid_list->pid_max)
583 break;
584
585 pid = (pid_t)val;
586
587 set_bit(pid, pid_list->pids);
588 nr_pids++;
589
590 trace_parser_clear(&parser);
591 ret = 0;
592 }
593 trace_parser_put(&parser);
594
595 if (ret < 0) {
596 trace_free_pid_list(pid_list);
597 return ret;
598 }
599
600 if (!nr_pids) {
601 /* Cleared the list of pids */
602 trace_free_pid_list(pid_list);
603 read = ret;
604 pid_list = NULL;
605 }
606
607 *new_pid_list = pid_list;
608
609 return read;
610}
611
612static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
613{
614 u64 ts;
615
616 /* Early boot up does not have a buffer yet */
617 if (!buf->buffer)
618 return trace_clock_local();
619
620 ts = ring_buffer_time_stamp(buf->buffer, cpu);
621 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
622
623 return ts;
624}
625
626u64 ftrace_now(int cpu)
627{
628 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
629}
630
631/**
632 * tracing_is_enabled - Show if global_trace has been disabled
633 *
634 * Shows if the global trace has been enabled or not. It uses the
635 * mirror flag "buffer_disabled" to be used in fast paths such as for
636 * the irqsoff tracer. But it may be inaccurate due to races. If you
637 * need to know the accurate state, use tracing_is_on() which is a little
638 * slower, but accurate.
639 */
640int tracing_is_enabled(void)
641{
642 /*
643 * For quick access (irqsoff uses this in fast path), just
644 * return the mirror variable of the state of the ring buffer.
645 * It's a little racy, but we don't really care.
646 */
647 smp_rmb();
648 return !global_trace.buffer_disabled;
649}
650
651/*
652 * trace_buf_size is the size in bytes that is allocated
653 * for a buffer. Note, the number of bytes is always rounded
654 * to page size.
655 *
656 * This number is purposely set to a low number of 16384.
657 * If the dump on oops happens, it will be much appreciated
658 * to not have to wait for all that output. Anyway this can be
659 * boot time and run time configurable.
660 */
661#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
662
663static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
664
665/* trace_types holds a link list of available tracers. */
666static struct tracer *trace_types __read_mostly;
667
668/*
669 * trace_types_lock is used to protect the trace_types list.
670 */
671DEFINE_MUTEX(trace_types_lock);
672
673/*
674 * serialize the access of the ring buffer
675 *
676 * ring buffer serializes readers, but it is low level protection.
677 * The validity of the events (which returns by ring_buffer_peek() ..etc)
678 * are not protected by ring buffer.
679 *
680 * The content of events may become garbage if we allow other process consumes
681 * these events concurrently:
682 * A) the page of the consumed events may become a normal page
683 * (not reader page) in ring buffer, and this page will be rewrited
684 * by events producer.
685 * B) The page of the consumed events may become a page for splice_read,
686 * and this page will be returned to system.
687 *
688 * These primitives allow multi process access to different cpu ring buffer
689 * concurrently.
690 *
691 * These primitives don't distinguish read-only and read-consume access.
692 * Multi read-only access are also serialized.
693 */
694
695#ifdef CONFIG_SMP
696static DECLARE_RWSEM(all_cpu_access_lock);
697static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
698
699static inline void trace_access_lock(int cpu)
700{
701 if (cpu == RING_BUFFER_ALL_CPUS) {
702 /* gain it for accessing the whole ring buffer. */
703 down_write(&all_cpu_access_lock);
704 } else {
705 /* gain it for accessing a cpu ring buffer. */
706
707 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
708 down_read(&all_cpu_access_lock);
709
710 /* Secondly block other access to this @cpu ring buffer. */
711 mutex_lock(&per_cpu(cpu_access_lock, cpu));
712 }
713}
714
715static inline void trace_access_unlock(int cpu)
716{
717 if (cpu == RING_BUFFER_ALL_CPUS) {
718 up_write(&all_cpu_access_lock);
719 } else {
720 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
721 up_read(&all_cpu_access_lock);
722 }
723}
724
725static inline void trace_access_lock_init(void)
726{
727 int cpu;
728
729 for_each_possible_cpu(cpu)
730 mutex_init(&per_cpu(cpu_access_lock, cpu));
731}
732
733#else
734
735static DEFINE_MUTEX(access_lock);
736
737static inline void trace_access_lock(int cpu)
738{
739 (void)cpu;
740 mutex_lock(&access_lock);
741}
742
743static inline void trace_access_unlock(int cpu)
744{
745 (void)cpu;
746 mutex_unlock(&access_lock);
747}
748
749static inline void trace_access_lock_init(void)
750{
751}
752
753#endif
754
755#ifdef CONFIG_STACKTRACE
756static void __ftrace_trace_stack(struct trace_buffer *buffer,
757 unsigned long flags,
758 int skip, int pc, struct pt_regs *regs);
759static inline void ftrace_trace_stack(struct trace_array *tr,
760 struct trace_buffer *buffer,
761 unsigned long flags,
762 int skip, int pc, struct pt_regs *regs);
763
764#else
765static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
766 unsigned long flags,
767 int skip, int pc, struct pt_regs *regs)
768{
769}
770static inline void ftrace_trace_stack(struct trace_array *tr,
771 struct trace_buffer *buffer,
772 unsigned long flags,
773 int skip, int pc, struct pt_regs *regs)
774{
775}
776
777#endif
778
779static __always_inline void
780trace_event_setup(struct ring_buffer_event *event,
781 int type, unsigned long flags, int pc)
782{
783 struct trace_entry *ent = ring_buffer_event_data(event);
784
785 tracing_generic_entry_update(ent, type, flags, pc);
786}
787
788static __always_inline struct ring_buffer_event *
789__trace_buffer_lock_reserve(struct trace_buffer *buffer,
790 int type,
791 unsigned long len,
792 unsigned long flags, int pc)
793{
794 struct ring_buffer_event *event;
795
796 event = ring_buffer_lock_reserve(buffer, len);
797 if (event != NULL)
798 trace_event_setup(event, type, flags, pc);
799
800 return event;
801}
802
803void tracer_tracing_on(struct trace_array *tr)
804{
805 if (tr->array_buffer.buffer)
806 ring_buffer_record_on(tr->array_buffer.buffer);
807 /*
808 * This flag is looked at when buffers haven't been allocated
809 * yet, or by some tracers (like irqsoff), that just want to
810 * know if the ring buffer has been disabled, but it can handle
811 * races of where it gets disabled but we still do a record.
812 * As the check is in the fast path of the tracers, it is more
813 * important to be fast than accurate.
814 */
815 tr->buffer_disabled = 0;
816 /* Make the flag seen by readers */
817 smp_wmb();
818}
819
820/**
821 * tracing_on - enable tracing buffers
822 *
823 * This function enables tracing buffers that may have been
824 * disabled with tracing_off.
825 */
826void tracing_on(void)
827{
828 tracer_tracing_on(&global_trace);
829}
830EXPORT_SYMBOL_GPL(tracing_on);
831
832
833static __always_inline void
834__buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
835{
836 __this_cpu_write(trace_taskinfo_save, true);
837
838 /* If this is the temp buffer, we need to commit fully */
839 if (this_cpu_read(trace_buffered_event) == event) {
840 /* Length is in event->array[0] */
841 ring_buffer_write(buffer, event->array[0], &event->array[1]);
842 /* Release the temp buffer */
843 this_cpu_dec(trace_buffered_event_cnt);
844 } else
845 ring_buffer_unlock_commit(buffer, event);
846}
847
848/**
849 * __trace_puts - write a constant string into the trace buffer.
850 * @ip: The address of the caller
851 * @str: The constant string to write
852 * @size: The size of the string.
853 */
854int __trace_puts(unsigned long ip, const char *str, int size)
855{
856 struct ring_buffer_event *event;
857 struct trace_buffer *buffer;
858 struct print_entry *entry;
859 unsigned long irq_flags;
860 int alloc;
861 int pc;
862
863 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
864 return 0;
865
866 pc = preempt_count();
867
868 if (unlikely(tracing_selftest_running || tracing_disabled))
869 return 0;
870
871 alloc = sizeof(*entry) + size + 2; /* possible \n added */
872
873 local_save_flags(irq_flags);
874 buffer = global_trace.array_buffer.buffer;
875 ring_buffer_nest_start(buffer);
876 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
877 irq_flags, pc);
878 if (!event) {
879 size = 0;
880 goto out;
881 }
882
883 entry = ring_buffer_event_data(event);
884 entry->ip = ip;
885
886 memcpy(&entry->buf, str, size);
887
888 /* Add a newline if necessary */
889 if (entry->buf[size - 1] != '\n') {
890 entry->buf[size] = '\n';
891 entry->buf[size + 1] = '\0';
892 } else
893 entry->buf[size] = '\0';
894
895 __buffer_unlock_commit(buffer, event);
896 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
897 out:
898 ring_buffer_nest_end(buffer);
899 return size;
900}
901EXPORT_SYMBOL_GPL(__trace_puts);
902
903/**
904 * __trace_bputs - write the pointer to a constant string into trace buffer
905 * @ip: The address of the caller
906 * @str: The constant string to write to the buffer to
907 */
908int __trace_bputs(unsigned long ip, const char *str)
909{
910 struct ring_buffer_event *event;
911 struct trace_buffer *buffer;
912 struct bputs_entry *entry;
913 unsigned long irq_flags;
914 int size = sizeof(struct bputs_entry);
915 int ret = 0;
916 int pc;
917
918 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
919 return 0;
920
921 pc = preempt_count();
922
923 if (unlikely(tracing_selftest_running || tracing_disabled))
924 return 0;
925
926 local_save_flags(irq_flags);
927 buffer = global_trace.array_buffer.buffer;
928
929 ring_buffer_nest_start(buffer);
930 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
931 irq_flags, pc);
932 if (!event)
933 goto out;
934
935 entry = ring_buffer_event_data(event);
936 entry->ip = ip;
937 entry->str = str;
938
939 __buffer_unlock_commit(buffer, event);
940 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
941
942 ret = 1;
943 out:
944 ring_buffer_nest_end(buffer);
945 return ret;
946}
947EXPORT_SYMBOL_GPL(__trace_bputs);
948
949#ifdef CONFIG_TRACER_SNAPSHOT
950static void tracing_snapshot_instance_cond(struct trace_array *tr,
951 void *cond_data)
952{
953 struct tracer *tracer = tr->current_trace;
954 unsigned long flags;
955
956 if (in_nmi()) {
957 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
958 internal_trace_puts("*** snapshot is being ignored ***\n");
959 return;
960 }
961
962 if (!tr->allocated_snapshot) {
963 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
964 internal_trace_puts("*** stopping trace here! ***\n");
965 tracing_off();
966 return;
967 }
968
969 /* Note, snapshot can not be used when the tracer uses it */
970 if (tracer->use_max_tr) {
971 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
972 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
973 return;
974 }
975
976 local_irq_save(flags);
977 update_max_tr(tr, current, smp_processor_id(), cond_data);
978 local_irq_restore(flags);
979}
980
981void tracing_snapshot_instance(struct trace_array *tr)
982{
983 tracing_snapshot_instance_cond(tr, NULL);
984}
985
986/**
987 * tracing_snapshot - take a snapshot of the current buffer.
988 *
989 * This causes a swap between the snapshot buffer and the current live
990 * tracing buffer. You can use this to take snapshots of the live
991 * trace when some condition is triggered, but continue to trace.
992 *
993 * Note, make sure to allocate the snapshot with either
994 * a tracing_snapshot_alloc(), or by doing it manually
995 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
996 *
997 * If the snapshot buffer is not allocated, it will stop tracing.
998 * Basically making a permanent snapshot.
999 */
1000void tracing_snapshot(void)
1001{
1002 struct trace_array *tr = &global_trace;
1003
1004 tracing_snapshot_instance(tr);
1005}
1006EXPORT_SYMBOL_GPL(tracing_snapshot);
1007
1008/**
1009 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1010 * @tr: The tracing instance to snapshot
1011 * @cond_data: The data to be tested conditionally, and possibly saved
1012 *
1013 * This is the same as tracing_snapshot() except that the snapshot is
1014 * conditional - the snapshot will only happen if the
1015 * cond_snapshot.update() implementation receiving the cond_data
1016 * returns true, which means that the trace array's cond_snapshot
1017 * update() operation used the cond_data to determine whether the
1018 * snapshot should be taken, and if it was, presumably saved it along
1019 * with the snapshot.
1020 */
1021void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1022{
1023 tracing_snapshot_instance_cond(tr, cond_data);
1024}
1025EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1026
1027/**
1028 * tracing_snapshot_cond_data - get the user data associated with a snapshot
1029 * @tr: The tracing instance
1030 *
1031 * When the user enables a conditional snapshot using
1032 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1033 * with the snapshot. This accessor is used to retrieve it.
1034 *
1035 * Should not be called from cond_snapshot.update(), since it takes
1036 * the tr->max_lock lock, which the code calling
1037 * cond_snapshot.update() has already done.
1038 *
1039 * Returns the cond_data associated with the trace array's snapshot.
1040 */
1041void *tracing_cond_snapshot_data(struct trace_array *tr)
1042{
1043 void *cond_data = NULL;
1044
1045 arch_spin_lock(&tr->max_lock);
1046
1047 if (tr->cond_snapshot)
1048 cond_data = tr->cond_snapshot->cond_data;
1049
1050 arch_spin_unlock(&tr->max_lock);
1051
1052 return cond_data;
1053}
1054EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1055
1056static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1057 struct array_buffer *size_buf, int cpu_id);
1058static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1059
1060int tracing_alloc_snapshot_instance(struct trace_array *tr)
1061{
1062 int ret;
1063
1064 if (!tr->allocated_snapshot) {
1065
1066 /* allocate spare buffer */
1067 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1068 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1069 if (ret < 0)
1070 return ret;
1071
1072 tr->allocated_snapshot = true;
1073 }
1074
1075 return 0;
1076}
1077
1078static void free_snapshot(struct trace_array *tr)
1079{
1080 /*
1081 * We don't free the ring buffer. instead, resize it because
1082 * The max_tr ring buffer has some state (e.g. ring->clock) and
1083 * we want preserve it.
1084 */
1085 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1086 set_buffer_entries(&tr->max_buffer, 1);
1087 tracing_reset_online_cpus(&tr->max_buffer);
1088 tr->allocated_snapshot = false;
1089}
1090
1091/**
1092 * tracing_alloc_snapshot - allocate snapshot buffer.
1093 *
1094 * This only allocates the snapshot buffer if it isn't already
1095 * allocated - it doesn't also take a snapshot.
1096 *
1097 * This is meant to be used in cases where the snapshot buffer needs
1098 * to be set up for events that can't sleep but need to be able to
1099 * trigger a snapshot.
1100 */
1101int tracing_alloc_snapshot(void)
1102{
1103 struct trace_array *tr = &global_trace;
1104 int ret;
1105
1106 ret = tracing_alloc_snapshot_instance(tr);
1107 WARN_ON(ret < 0);
1108
1109 return ret;
1110}
1111EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1112
1113/**
1114 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1115 *
1116 * This is similar to tracing_snapshot(), but it will allocate the
1117 * snapshot buffer if it isn't already allocated. Use this only
1118 * where it is safe to sleep, as the allocation may sleep.
1119 *
1120 * This causes a swap between the snapshot buffer and the current live
1121 * tracing buffer. You can use this to take snapshots of the live
1122 * trace when some condition is triggered, but continue to trace.
1123 */
1124void tracing_snapshot_alloc(void)
1125{
1126 int ret;
1127
1128 ret = tracing_alloc_snapshot();
1129 if (ret < 0)
1130 return;
1131
1132 tracing_snapshot();
1133}
1134EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1135
1136/**
1137 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1138 * @tr: The tracing instance
1139 * @cond_data: User data to associate with the snapshot
1140 * @update: Implementation of the cond_snapshot update function
1141 *
1142 * Check whether the conditional snapshot for the given instance has
1143 * already been enabled, or if the current tracer is already using a
1144 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1145 * save the cond_data and update function inside.
1146 *
1147 * Returns 0 if successful, error otherwise.
1148 */
1149int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1150 cond_update_fn_t update)
1151{
1152 struct cond_snapshot *cond_snapshot;
1153 int ret = 0;
1154
1155 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1156 if (!cond_snapshot)
1157 return -ENOMEM;
1158
1159 cond_snapshot->cond_data = cond_data;
1160 cond_snapshot->update = update;
1161
1162 mutex_lock(&trace_types_lock);
1163
1164 ret = tracing_alloc_snapshot_instance(tr);
1165 if (ret)
1166 goto fail_unlock;
1167
1168 if (tr->current_trace->use_max_tr) {
1169 ret = -EBUSY;
1170 goto fail_unlock;
1171 }
1172
1173 /*
1174 * The cond_snapshot can only change to NULL without the
1175 * trace_types_lock. We don't care if we race with it going
1176 * to NULL, but we want to make sure that it's not set to
1177 * something other than NULL when we get here, which we can
1178 * do safely with only holding the trace_types_lock and not
1179 * having to take the max_lock.
1180 */
1181 if (tr->cond_snapshot) {
1182 ret = -EBUSY;
1183 goto fail_unlock;
1184 }
1185
1186 arch_spin_lock(&tr->max_lock);
1187 tr->cond_snapshot = cond_snapshot;
1188 arch_spin_unlock(&tr->max_lock);
1189
1190 mutex_unlock(&trace_types_lock);
1191
1192 return ret;
1193
1194 fail_unlock:
1195 mutex_unlock(&trace_types_lock);
1196 kfree(cond_snapshot);
1197 return ret;
1198}
1199EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1200
1201/**
1202 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1203 * @tr: The tracing instance
1204 *
1205 * Check whether the conditional snapshot for the given instance is
1206 * enabled; if so, free the cond_snapshot associated with it,
1207 * otherwise return -EINVAL.
1208 *
1209 * Returns 0 if successful, error otherwise.
1210 */
1211int tracing_snapshot_cond_disable(struct trace_array *tr)
1212{
1213 int ret = 0;
1214
1215 arch_spin_lock(&tr->max_lock);
1216
1217 if (!tr->cond_snapshot)
1218 ret = -EINVAL;
1219 else {
1220 kfree(tr->cond_snapshot);
1221 tr->cond_snapshot = NULL;
1222 }
1223
1224 arch_spin_unlock(&tr->max_lock);
1225
1226 return ret;
1227}
1228EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1229#else
1230void tracing_snapshot(void)
1231{
1232 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1233}
1234EXPORT_SYMBOL_GPL(tracing_snapshot);
1235void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1236{
1237 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1238}
1239EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1240int tracing_alloc_snapshot(void)
1241{
1242 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1243 return -ENODEV;
1244}
1245EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1246void tracing_snapshot_alloc(void)
1247{
1248 /* Give warning */
1249 tracing_snapshot();
1250}
1251EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1252void *tracing_cond_snapshot_data(struct trace_array *tr)
1253{
1254 return NULL;
1255}
1256EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1257int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1258{
1259 return -ENODEV;
1260}
1261EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1262int tracing_snapshot_cond_disable(struct trace_array *tr)
1263{
1264 return false;
1265}
1266EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1267#endif /* CONFIG_TRACER_SNAPSHOT */
1268
1269void tracer_tracing_off(struct trace_array *tr)
1270{
1271 if (tr->array_buffer.buffer)
1272 ring_buffer_record_off(tr->array_buffer.buffer);
1273 /*
1274 * This flag is looked at when buffers haven't been allocated
1275 * yet, or by some tracers (like irqsoff), that just want to
1276 * know if the ring buffer has been disabled, but it can handle
1277 * races of where it gets disabled but we still do a record.
1278 * As the check is in the fast path of the tracers, it is more
1279 * important to be fast than accurate.
1280 */
1281 tr->buffer_disabled = 1;
1282 /* Make the flag seen by readers */
1283 smp_wmb();
1284}
1285
1286/**
1287 * tracing_off - turn off tracing buffers
1288 *
1289 * This function stops the tracing buffers from recording data.
1290 * It does not disable any overhead the tracers themselves may
1291 * be causing. This function simply causes all recording to
1292 * the ring buffers to fail.
1293 */
1294void tracing_off(void)
1295{
1296 tracer_tracing_off(&global_trace);
1297}
1298EXPORT_SYMBOL_GPL(tracing_off);
1299
1300void disable_trace_on_warning(void)
1301{
1302 if (__disable_trace_on_warning) {
1303 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1304 "Disabling tracing due to warning\n");
1305 tracing_off();
1306 }
1307}
1308
1309/**
1310 * tracer_tracing_is_on - show real state of ring buffer enabled
1311 * @tr : the trace array to know if ring buffer is enabled
1312 *
1313 * Shows real state of the ring buffer if it is enabled or not.
1314 */
1315bool tracer_tracing_is_on(struct trace_array *tr)
1316{
1317 if (tr->array_buffer.buffer)
1318 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1319 return !tr->buffer_disabled;
1320}
1321
1322/**
1323 * tracing_is_on - show state of ring buffers enabled
1324 */
1325int tracing_is_on(void)
1326{
1327 return tracer_tracing_is_on(&global_trace);
1328}
1329EXPORT_SYMBOL_GPL(tracing_is_on);
1330
1331static int __init set_buf_size(char *str)
1332{
1333 unsigned long buf_size;
1334
1335 if (!str)
1336 return 0;
1337 buf_size = memparse(str, &str);
1338 /* nr_entries can not be zero */
1339 if (buf_size == 0)
1340 return 0;
1341 trace_buf_size = buf_size;
1342 return 1;
1343}
1344__setup("trace_buf_size=", set_buf_size);
1345
1346static int __init set_tracing_thresh(char *str)
1347{
1348 unsigned long threshold;
1349 int ret;
1350
1351 if (!str)
1352 return 0;
1353 ret = kstrtoul(str, 0, &threshold);
1354 if (ret < 0)
1355 return 0;
1356 tracing_thresh = threshold * 1000;
1357 return 1;
1358}
1359__setup("tracing_thresh=", set_tracing_thresh);
1360
1361unsigned long nsecs_to_usecs(unsigned long nsecs)
1362{
1363 return nsecs / 1000;
1364}
1365
1366/*
1367 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1368 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1369 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1370 * of strings in the order that the evals (enum) were defined.
1371 */
1372#undef C
1373#define C(a, b) b
1374
1375/* These must match the bit postions in trace_iterator_flags */
1376static const char *trace_options[] = {
1377 TRACE_FLAGS
1378 NULL
1379};
1380
1381static struct {
1382 u64 (*func)(void);
1383 const char *name;
1384 int in_ns; /* is this clock in nanoseconds? */
1385} trace_clocks[] = {
1386 { trace_clock_local, "local", 1 },
1387 { trace_clock_global, "global", 1 },
1388 { trace_clock_counter, "counter", 0 },
1389 { trace_clock_jiffies, "uptime", 0 },
1390 { trace_clock, "perf", 1 },
1391 { ktime_get_mono_fast_ns, "mono", 1 },
1392 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1393 { ktime_get_boot_fast_ns, "boot", 1 },
1394 ARCH_TRACE_CLOCKS
1395};
1396
1397bool trace_clock_in_ns(struct trace_array *tr)
1398{
1399 if (trace_clocks[tr->clock_id].in_ns)
1400 return true;
1401
1402 return false;
1403}
1404
1405/*
1406 * trace_parser_get_init - gets the buffer for trace parser
1407 */
1408int trace_parser_get_init(struct trace_parser *parser, int size)
1409{
1410 memset(parser, 0, sizeof(*parser));
1411
1412 parser->buffer = kmalloc(size, GFP_KERNEL);
1413 if (!parser->buffer)
1414 return 1;
1415
1416 parser->size = size;
1417 return 0;
1418}
1419
1420/*
1421 * trace_parser_put - frees the buffer for trace parser
1422 */
1423void trace_parser_put(struct trace_parser *parser)
1424{
1425 kfree(parser->buffer);
1426 parser->buffer = NULL;
1427}
1428
1429/*
1430 * trace_get_user - reads the user input string separated by space
1431 * (matched by isspace(ch))
1432 *
1433 * For each string found the 'struct trace_parser' is updated,
1434 * and the function returns.
1435 *
1436 * Returns number of bytes read.
1437 *
1438 * See kernel/trace/trace.h for 'struct trace_parser' details.
1439 */
1440int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1441 size_t cnt, loff_t *ppos)
1442{
1443 char ch;
1444 size_t read = 0;
1445 ssize_t ret;
1446
1447 if (!*ppos)
1448 trace_parser_clear(parser);
1449
1450 ret = get_user(ch, ubuf++);
1451 if (ret)
1452 goto out;
1453
1454 read++;
1455 cnt--;
1456
1457 /*
1458 * The parser is not finished with the last write,
1459 * continue reading the user input without skipping spaces.
1460 */
1461 if (!parser->cont) {
1462 /* skip white space */
1463 while (cnt && isspace(ch)) {
1464 ret = get_user(ch, ubuf++);
1465 if (ret)
1466 goto out;
1467 read++;
1468 cnt--;
1469 }
1470
1471 parser->idx = 0;
1472
1473 /* only spaces were written */
1474 if (isspace(ch) || !ch) {
1475 *ppos += read;
1476 ret = read;
1477 goto out;
1478 }
1479 }
1480
1481 /* read the non-space input */
1482 while (cnt && !isspace(ch) && ch) {
1483 if (parser->idx < parser->size - 1)
1484 parser->buffer[parser->idx++] = ch;
1485 else {
1486 ret = -EINVAL;
1487 goto out;
1488 }
1489 ret = get_user(ch, ubuf++);
1490 if (ret)
1491 goto out;
1492 read++;
1493 cnt--;
1494 }
1495
1496 /* We either got finished input or we have to wait for another call. */
1497 if (isspace(ch) || !ch) {
1498 parser->buffer[parser->idx] = 0;
1499 parser->cont = false;
1500 } else if (parser->idx < parser->size - 1) {
1501 parser->cont = true;
1502 parser->buffer[parser->idx++] = ch;
1503 /* Make sure the parsed string always terminates with '\0'. */
1504 parser->buffer[parser->idx] = 0;
1505 } else {
1506 ret = -EINVAL;
1507 goto out;
1508 }
1509
1510 *ppos += read;
1511 ret = read;
1512
1513out:
1514 return ret;
1515}
1516
1517/* TODO add a seq_buf_to_buffer() */
1518static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1519{
1520 int len;
1521
1522 if (trace_seq_used(s) <= s->seq.readpos)
1523 return -EBUSY;
1524
1525 len = trace_seq_used(s) - s->seq.readpos;
1526 if (cnt > len)
1527 cnt = len;
1528 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1529
1530 s->seq.readpos += cnt;
1531 return cnt;
1532}
1533
1534unsigned long __read_mostly tracing_thresh;
1535static const struct file_operations tracing_max_lat_fops;
1536
1537#if (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1538 defined(CONFIG_FSNOTIFY)
1539
1540static struct workqueue_struct *fsnotify_wq;
1541
1542static void latency_fsnotify_workfn(struct work_struct *work)
1543{
1544 struct trace_array *tr = container_of(work, struct trace_array,
1545 fsnotify_work);
1546 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1547}
1548
1549static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1550{
1551 struct trace_array *tr = container_of(iwork, struct trace_array,
1552 fsnotify_irqwork);
1553 queue_work(fsnotify_wq, &tr->fsnotify_work);
1554}
1555
1556static void trace_create_maxlat_file(struct trace_array *tr,
1557 struct dentry *d_tracer)
1558{
1559 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1560 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1561 tr->d_max_latency = trace_create_file("tracing_max_latency", 0644,
1562 d_tracer, &tr->max_latency,
1563 &tracing_max_lat_fops);
1564}
1565
1566__init static int latency_fsnotify_init(void)
1567{
1568 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1569 WQ_UNBOUND | WQ_HIGHPRI, 0);
1570 if (!fsnotify_wq) {
1571 pr_err("Unable to allocate tr_max_lat_wq\n");
1572 return -ENOMEM;
1573 }
1574 return 0;
1575}
1576
1577late_initcall_sync(latency_fsnotify_init);
1578
1579void latency_fsnotify(struct trace_array *tr)
1580{
1581 if (!fsnotify_wq)
1582 return;
1583 /*
1584 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1585 * possible that we are called from __schedule() or do_idle(), which
1586 * could cause a deadlock.
1587 */
1588 irq_work_queue(&tr->fsnotify_irqwork);
1589}
1590
1591/*
1592 * (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1593 * defined(CONFIG_FSNOTIFY)
1594 */
1595#else
1596
1597#define trace_create_maxlat_file(tr, d_tracer) \
1598 trace_create_file("tracing_max_latency", 0644, d_tracer, \
1599 &tr->max_latency, &tracing_max_lat_fops)
1600
1601#endif
1602
1603#ifdef CONFIG_TRACER_MAX_TRACE
1604/*
1605 * Copy the new maximum trace into the separate maximum-trace
1606 * structure. (this way the maximum trace is permanently saved,
1607 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1608 */
1609static void
1610__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1611{
1612 struct array_buffer *trace_buf = &tr->array_buffer;
1613 struct array_buffer *max_buf = &tr->max_buffer;
1614 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1615 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1616
1617 max_buf->cpu = cpu;
1618 max_buf->time_start = data->preempt_timestamp;
1619
1620 max_data->saved_latency = tr->max_latency;
1621 max_data->critical_start = data->critical_start;
1622 max_data->critical_end = data->critical_end;
1623
1624 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1625 max_data->pid = tsk->pid;
1626 /*
1627 * If tsk == current, then use current_uid(), as that does not use
1628 * RCU. The irq tracer can be called out of RCU scope.
1629 */
1630 if (tsk == current)
1631 max_data->uid = current_uid();
1632 else
1633 max_data->uid = task_uid(tsk);
1634
1635 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1636 max_data->policy = tsk->policy;
1637 max_data->rt_priority = tsk->rt_priority;
1638
1639 /* record this tasks comm */
1640 tracing_record_cmdline(tsk);
1641 latency_fsnotify(tr);
1642}
1643
1644/**
1645 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1646 * @tr: tracer
1647 * @tsk: the task with the latency
1648 * @cpu: The cpu that initiated the trace.
1649 * @cond_data: User data associated with a conditional snapshot
1650 *
1651 * Flip the buffers between the @tr and the max_tr and record information
1652 * about which task was the cause of this latency.
1653 */
1654void
1655update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1656 void *cond_data)
1657{
1658 if (tr->stop_count)
1659 return;
1660
1661 WARN_ON_ONCE(!irqs_disabled());
1662
1663 if (!tr->allocated_snapshot) {
1664 /* Only the nop tracer should hit this when disabling */
1665 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1666 return;
1667 }
1668
1669 arch_spin_lock(&tr->max_lock);
1670
1671 /* Inherit the recordable setting from array_buffer */
1672 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1673 ring_buffer_record_on(tr->max_buffer.buffer);
1674 else
1675 ring_buffer_record_off(tr->max_buffer.buffer);
1676
1677#ifdef CONFIG_TRACER_SNAPSHOT
1678 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data))
1679 goto out_unlock;
1680#endif
1681 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1682
1683 __update_max_tr(tr, tsk, cpu);
1684
1685 out_unlock:
1686 arch_spin_unlock(&tr->max_lock);
1687}
1688
1689/**
1690 * update_max_tr_single - only copy one trace over, and reset the rest
1691 * @tr: tracer
1692 * @tsk: task with the latency
1693 * @cpu: the cpu of the buffer to copy.
1694 *
1695 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1696 */
1697void
1698update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1699{
1700 int ret;
1701
1702 if (tr->stop_count)
1703 return;
1704
1705 WARN_ON_ONCE(!irqs_disabled());
1706 if (!tr->allocated_snapshot) {
1707 /* Only the nop tracer should hit this when disabling */
1708 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1709 return;
1710 }
1711
1712 arch_spin_lock(&tr->max_lock);
1713
1714 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1715
1716 if (ret == -EBUSY) {
1717 /*
1718 * We failed to swap the buffer due to a commit taking
1719 * place on this CPU. We fail to record, but we reset
1720 * the max trace buffer (no one writes directly to it)
1721 * and flag that it failed.
1722 */
1723 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1724 "Failed to swap buffers due to commit in progress\n");
1725 }
1726
1727 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1728
1729 __update_max_tr(tr, tsk, cpu);
1730 arch_spin_unlock(&tr->max_lock);
1731}
1732#endif /* CONFIG_TRACER_MAX_TRACE */
1733
1734static int wait_on_pipe(struct trace_iterator *iter, int full)
1735{
1736 /* Iterators are static, they should be filled or empty */
1737 if (trace_buffer_iter(iter, iter->cpu_file))
1738 return 0;
1739
1740 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1741 full);
1742}
1743
1744#ifdef CONFIG_FTRACE_STARTUP_TEST
1745static bool selftests_can_run;
1746
1747struct trace_selftests {
1748 struct list_head list;
1749 struct tracer *type;
1750};
1751
1752static LIST_HEAD(postponed_selftests);
1753
1754static int save_selftest(struct tracer *type)
1755{
1756 struct trace_selftests *selftest;
1757
1758 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1759 if (!selftest)
1760 return -ENOMEM;
1761
1762 selftest->type = type;
1763 list_add(&selftest->list, &postponed_selftests);
1764 return 0;
1765}
1766
1767static int run_tracer_selftest(struct tracer *type)
1768{
1769 struct trace_array *tr = &global_trace;
1770 struct tracer *saved_tracer = tr->current_trace;
1771 int ret;
1772
1773 if (!type->selftest || tracing_selftest_disabled)
1774 return 0;
1775
1776 /*
1777 * If a tracer registers early in boot up (before scheduling is
1778 * initialized and such), then do not run its selftests yet.
1779 * Instead, run it a little later in the boot process.
1780 */
1781 if (!selftests_can_run)
1782 return save_selftest(type);
1783
1784 /*
1785 * Run a selftest on this tracer.
1786 * Here we reset the trace buffer, and set the current
1787 * tracer to be this tracer. The tracer can then run some
1788 * internal tracing to verify that everything is in order.
1789 * If we fail, we do not register this tracer.
1790 */
1791 tracing_reset_online_cpus(&tr->array_buffer);
1792
1793 tr->current_trace = type;
1794
1795#ifdef CONFIG_TRACER_MAX_TRACE
1796 if (type->use_max_tr) {
1797 /* If we expanded the buffers, make sure the max is expanded too */
1798 if (ring_buffer_expanded)
1799 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1800 RING_BUFFER_ALL_CPUS);
1801 tr->allocated_snapshot = true;
1802 }
1803#endif
1804
1805 /* the test is responsible for initializing and enabling */
1806 pr_info("Testing tracer %s: ", type->name);
1807 ret = type->selftest(type, tr);
1808 /* the test is responsible for resetting too */
1809 tr->current_trace = saved_tracer;
1810 if (ret) {
1811 printk(KERN_CONT "FAILED!\n");
1812 /* Add the warning after printing 'FAILED' */
1813 WARN_ON(1);
1814 return -1;
1815 }
1816 /* Only reset on passing, to avoid touching corrupted buffers */
1817 tracing_reset_online_cpus(&tr->array_buffer);
1818
1819#ifdef CONFIG_TRACER_MAX_TRACE
1820 if (type->use_max_tr) {
1821 tr->allocated_snapshot = false;
1822
1823 /* Shrink the max buffer again */
1824 if (ring_buffer_expanded)
1825 ring_buffer_resize(tr->max_buffer.buffer, 1,
1826 RING_BUFFER_ALL_CPUS);
1827 }
1828#endif
1829
1830 printk(KERN_CONT "PASSED\n");
1831 return 0;
1832}
1833
1834static __init int init_trace_selftests(void)
1835{
1836 struct trace_selftests *p, *n;
1837 struct tracer *t, **last;
1838 int ret;
1839
1840 selftests_can_run = true;
1841
1842 mutex_lock(&trace_types_lock);
1843
1844 if (list_empty(&postponed_selftests))
1845 goto out;
1846
1847 pr_info("Running postponed tracer tests:\n");
1848
1849 tracing_selftest_running = true;
1850 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
1851 /* This loop can take minutes when sanitizers are enabled, so
1852 * lets make sure we allow RCU processing.
1853 */
1854 cond_resched();
1855 ret = run_tracer_selftest(p->type);
1856 /* If the test fails, then warn and remove from available_tracers */
1857 if (ret < 0) {
1858 WARN(1, "tracer: %s failed selftest, disabling\n",
1859 p->type->name);
1860 last = &trace_types;
1861 for (t = trace_types; t; t = t->next) {
1862 if (t == p->type) {
1863 *last = t->next;
1864 break;
1865 }
1866 last = &t->next;
1867 }
1868 }
1869 list_del(&p->list);
1870 kfree(p);
1871 }
1872 tracing_selftest_running = false;
1873
1874 out:
1875 mutex_unlock(&trace_types_lock);
1876
1877 return 0;
1878}
1879core_initcall(init_trace_selftests);
1880#else
1881static inline int run_tracer_selftest(struct tracer *type)
1882{
1883 return 0;
1884}
1885#endif /* CONFIG_FTRACE_STARTUP_TEST */
1886
1887static void add_tracer_options(struct trace_array *tr, struct tracer *t);
1888
1889static void __init apply_trace_boot_options(void);
1890
1891/**
1892 * register_tracer - register a tracer with the ftrace system.
1893 * @type: the plugin for the tracer
1894 *
1895 * Register a new plugin tracer.
1896 */
1897int __init register_tracer(struct tracer *type)
1898{
1899 struct tracer *t;
1900 int ret = 0;
1901
1902 if (!type->name) {
1903 pr_info("Tracer must have a name\n");
1904 return -1;
1905 }
1906
1907 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1908 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1909 return -1;
1910 }
1911
1912 if (security_locked_down(LOCKDOWN_TRACEFS)) {
1913 pr_warn("Can not register tracer %s due to lockdown\n",
1914 type->name);
1915 return -EPERM;
1916 }
1917
1918 mutex_lock(&trace_types_lock);
1919
1920 tracing_selftest_running = true;
1921
1922 for (t = trace_types; t; t = t->next) {
1923 if (strcmp(type->name, t->name) == 0) {
1924 /* already found */
1925 pr_info("Tracer %s already registered\n",
1926 type->name);
1927 ret = -1;
1928 goto out;
1929 }
1930 }
1931
1932 if (!type->set_flag)
1933 type->set_flag = &dummy_set_flag;
1934 if (!type->flags) {
1935 /*allocate a dummy tracer_flags*/
1936 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
1937 if (!type->flags) {
1938 ret = -ENOMEM;
1939 goto out;
1940 }
1941 type->flags->val = 0;
1942 type->flags->opts = dummy_tracer_opt;
1943 } else
1944 if (!type->flags->opts)
1945 type->flags->opts = dummy_tracer_opt;
1946
1947 /* store the tracer for __set_tracer_option */
1948 type->flags->trace = type;
1949
1950 ret = run_tracer_selftest(type);
1951 if (ret < 0)
1952 goto out;
1953
1954 type->next = trace_types;
1955 trace_types = type;
1956 add_tracer_options(&global_trace, type);
1957
1958 out:
1959 tracing_selftest_running = false;
1960 mutex_unlock(&trace_types_lock);
1961
1962 if (ret || !default_bootup_tracer)
1963 goto out_unlock;
1964
1965 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1966 goto out_unlock;
1967
1968 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1969 /* Do we want this tracer to start on bootup? */
1970 tracing_set_tracer(&global_trace, type->name);
1971 default_bootup_tracer = NULL;
1972
1973 apply_trace_boot_options();
1974
1975 /* disable other selftests, since this will break it. */
1976 tracing_selftest_disabled = true;
1977#ifdef CONFIG_FTRACE_STARTUP_TEST
1978 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1979 type->name);
1980#endif
1981
1982 out_unlock:
1983 return ret;
1984}
1985
1986static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
1987{
1988 struct trace_buffer *buffer = buf->buffer;
1989
1990 if (!buffer)
1991 return;
1992
1993 ring_buffer_record_disable(buffer);
1994
1995 /* Make sure all commits have finished */
1996 synchronize_rcu();
1997 ring_buffer_reset_cpu(buffer, cpu);
1998
1999 ring_buffer_record_enable(buffer);
2000}
2001
2002void tracing_reset_online_cpus(struct array_buffer *buf)
2003{
2004 struct trace_buffer *buffer = buf->buffer;
2005
2006 if (!buffer)
2007 return;
2008
2009 ring_buffer_record_disable(buffer);
2010
2011 /* Make sure all commits have finished */
2012 synchronize_rcu();
2013
2014 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2015
2016 ring_buffer_reset_online_cpus(buffer);
2017
2018 ring_buffer_record_enable(buffer);
2019}
2020
2021/* Must have trace_types_lock held */
2022void tracing_reset_all_online_cpus(void)
2023{
2024 struct trace_array *tr;
2025
2026 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2027 if (!tr->clear_trace)
2028 continue;
2029 tr->clear_trace = false;
2030 tracing_reset_online_cpus(&tr->array_buffer);
2031#ifdef CONFIG_TRACER_MAX_TRACE
2032 tracing_reset_online_cpus(&tr->max_buffer);
2033#endif
2034 }
2035}
2036
2037static int *tgid_map;
2038
2039#define SAVED_CMDLINES_DEFAULT 128
2040#define NO_CMDLINE_MAP UINT_MAX
2041static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2042struct saved_cmdlines_buffer {
2043 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2044 unsigned *map_cmdline_to_pid;
2045 unsigned cmdline_num;
2046 int cmdline_idx;
2047 char *saved_cmdlines;
2048};
2049static struct saved_cmdlines_buffer *savedcmd;
2050
2051/* temporary disable recording */
2052static atomic_t trace_record_taskinfo_disabled __read_mostly;
2053
2054static inline char *get_saved_cmdlines(int idx)
2055{
2056 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2057}
2058
2059static inline void set_cmdline(int idx, const char *cmdline)
2060{
2061 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2062}
2063
2064static int allocate_cmdlines_buffer(unsigned int val,
2065 struct saved_cmdlines_buffer *s)
2066{
2067 s->map_cmdline_to_pid = kmalloc_array(val,
2068 sizeof(*s->map_cmdline_to_pid),
2069 GFP_KERNEL);
2070 if (!s->map_cmdline_to_pid)
2071 return -ENOMEM;
2072
2073 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2074 if (!s->saved_cmdlines) {
2075 kfree(s->map_cmdline_to_pid);
2076 return -ENOMEM;
2077 }
2078
2079 s->cmdline_idx = 0;
2080 s->cmdline_num = val;
2081 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2082 sizeof(s->map_pid_to_cmdline));
2083 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2084 val * sizeof(*s->map_cmdline_to_pid));
2085
2086 return 0;
2087}
2088
2089static int trace_create_savedcmd(void)
2090{
2091 int ret;
2092
2093 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2094 if (!savedcmd)
2095 return -ENOMEM;
2096
2097 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2098 if (ret < 0) {
2099 kfree(savedcmd);
2100 savedcmd = NULL;
2101 return -ENOMEM;
2102 }
2103
2104 return 0;
2105}
2106
2107int is_tracing_stopped(void)
2108{
2109 return global_trace.stop_count;
2110}
2111
2112/**
2113 * tracing_start - quick start of the tracer
2114 *
2115 * If tracing is enabled but was stopped by tracing_stop,
2116 * this will start the tracer back up.
2117 */
2118void tracing_start(void)
2119{
2120 struct trace_buffer *buffer;
2121 unsigned long flags;
2122
2123 if (tracing_disabled)
2124 return;
2125
2126 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2127 if (--global_trace.stop_count) {
2128 if (global_trace.stop_count < 0) {
2129 /* Someone screwed up their debugging */
2130 WARN_ON_ONCE(1);
2131 global_trace.stop_count = 0;
2132 }
2133 goto out;
2134 }
2135
2136 /* Prevent the buffers from switching */
2137 arch_spin_lock(&global_trace.max_lock);
2138
2139 buffer = global_trace.array_buffer.buffer;
2140 if (buffer)
2141 ring_buffer_record_enable(buffer);
2142
2143#ifdef CONFIG_TRACER_MAX_TRACE
2144 buffer = global_trace.max_buffer.buffer;
2145 if (buffer)
2146 ring_buffer_record_enable(buffer);
2147#endif
2148
2149 arch_spin_unlock(&global_trace.max_lock);
2150
2151 out:
2152 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2153}
2154
2155static void tracing_start_tr(struct trace_array *tr)
2156{
2157 struct trace_buffer *buffer;
2158 unsigned long flags;
2159
2160 if (tracing_disabled)
2161 return;
2162
2163 /* If global, we need to also start the max tracer */
2164 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2165 return tracing_start();
2166
2167 raw_spin_lock_irqsave(&tr->start_lock, flags);
2168
2169 if (--tr->stop_count) {
2170 if (tr->stop_count < 0) {
2171 /* Someone screwed up their debugging */
2172 WARN_ON_ONCE(1);
2173 tr->stop_count = 0;
2174 }
2175 goto out;
2176 }
2177
2178 buffer = tr->array_buffer.buffer;
2179 if (buffer)
2180 ring_buffer_record_enable(buffer);
2181
2182 out:
2183 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2184}
2185
2186/**
2187 * tracing_stop - quick stop of the tracer
2188 *
2189 * Light weight way to stop tracing. Use in conjunction with
2190 * tracing_start.
2191 */
2192void tracing_stop(void)
2193{
2194 struct trace_buffer *buffer;
2195 unsigned long flags;
2196
2197 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2198 if (global_trace.stop_count++)
2199 goto out;
2200
2201 /* Prevent the buffers from switching */
2202 arch_spin_lock(&global_trace.max_lock);
2203
2204 buffer = global_trace.array_buffer.buffer;
2205 if (buffer)
2206 ring_buffer_record_disable(buffer);
2207
2208#ifdef CONFIG_TRACER_MAX_TRACE
2209 buffer = global_trace.max_buffer.buffer;
2210 if (buffer)
2211 ring_buffer_record_disable(buffer);
2212#endif
2213
2214 arch_spin_unlock(&global_trace.max_lock);
2215
2216 out:
2217 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2218}
2219
2220static void tracing_stop_tr(struct trace_array *tr)
2221{
2222 struct trace_buffer *buffer;
2223 unsigned long flags;
2224
2225 /* If global, we need to also stop the max tracer */
2226 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2227 return tracing_stop();
2228
2229 raw_spin_lock_irqsave(&tr->start_lock, flags);
2230 if (tr->stop_count++)
2231 goto out;
2232
2233 buffer = tr->array_buffer.buffer;
2234 if (buffer)
2235 ring_buffer_record_disable(buffer);
2236
2237 out:
2238 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2239}
2240
2241static int trace_save_cmdline(struct task_struct *tsk)
2242{
2243 unsigned pid, idx;
2244
2245 /* treat recording of idle task as a success */
2246 if (!tsk->pid)
2247 return 1;
2248
2249 if (unlikely(tsk->pid > PID_MAX_DEFAULT))
2250 return 0;
2251
2252 /*
2253 * It's not the end of the world if we don't get
2254 * the lock, but we also don't want to spin
2255 * nor do we want to disable interrupts,
2256 * so if we miss here, then better luck next time.
2257 */
2258 if (!arch_spin_trylock(&trace_cmdline_lock))
2259 return 0;
2260
2261 idx = savedcmd->map_pid_to_cmdline[tsk->pid];
2262 if (idx == NO_CMDLINE_MAP) {
2263 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2264
2265 /*
2266 * Check whether the cmdline buffer at idx has a pid
2267 * mapped. We are going to overwrite that entry so we
2268 * need to clear the map_pid_to_cmdline. Otherwise we
2269 * would read the new comm for the old pid.
2270 */
2271 pid = savedcmd->map_cmdline_to_pid[idx];
2272 if (pid != NO_CMDLINE_MAP)
2273 savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
2274
2275 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2276 savedcmd->map_pid_to_cmdline[tsk->pid] = idx;
2277
2278 savedcmd->cmdline_idx = idx;
2279 }
2280
2281 set_cmdline(idx, tsk->comm);
2282
2283 arch_spin_unlock(&trace_cmdline_lock);
2284
2285 return 1;
2286}
2287
2288static void __trace_find_cmdline(int pid, char comm[])
2289{
2290 unsigned map;
2291
2292 if (!pid) {
2293 strcpy(comm, "<idle>");
2294 return;
2295 }
2296
2297 if (WARN_ON_ONCE(pid < 0)) {
2298 strcpy(comm, "<XXX>");
2299 return;
2300 }
2301
2302 if (pid > PID_MAX_DEFAULT) {
2303 strcpy(comm, "<...>");
2304 return;
2305 }
2306
2307 map = savedcmd->map_pid_to_cmdline[pid];
2308 if (map != NO_CMDLINE_MAP)
2309 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2310 else
2311 strcpy(comm, "<...>");
2312}
2313
2314void trace_find_cmdline(int pid, char comm[])
2315{
2316 preempt_disable();
2317 arch_spin_lock(&trace_cmdline_lock);
2318
2319 __trace_find_cmdline(pid, comm);
2320
2321 arch_spin_unlock(&trace_cmdline_lock);
2322 preempt_enable();
2323}
2324
2325int trace_find_tgid(int pid)
2326{
2327 if (unlikely(!tgid_map || !pid || pid > PID_MAX_DEFAULT))
2328 return 0;
2329
2330 return tgid_map[pid];
2331}
2332
2333static int trace_save_tgid(struct task_struct *tsk)
2334{
2335 /* treat recording of idle task as a success */
2336 if (!tsk->pid)
2337 return 1;
2338
2339 if (unlikely(!tgid_map || tsk->pid > PID_MAX_DEFAULT))
2340 return 0;
2341
2342 tgid_map[tsk->pid] = tsk->tgid;
2343 return 1;
2344}
2345
2346static bool tracing_record_taskinfo_skip(int flags)
2347{
2348 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2349 return true;
2350 if (atomic_read(&trace_record_taskinfo_disabled) || !tracing_is_on())
2351 return true;
2352 if (!__this_cpu_read(trace_taskinfo_save))
2353 return true;
2354 return false;
2355}
2356
2357/**
2358 * tracing_record_taskinfo - record the task info of a task
2359 *
2360 * @task: task to record
2361 * @flags: TRACE_RECORD_CMDLINE for recording comm
2362 * TRACE_RECORD_TGID for recording tgid
2363 */
2364void tracing_record_taskinfo(struct task_struct *task, int flags)
2365{
2366 bool done;
2367
2368 if (tracing_record_taskinfo_skip(flags))
2369 return;
2370
2371 /*
2372 * Record as much task information as possible. If some fail, continue
2373 * to try to record the others.
2374 */
2375 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2376 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2377
2378 /* If recording any information failed, retry again soon. */
2379 if (!done)
2380 return;
2381
2382 __this_cpu_write(trace_taskinfo_save, false);
2383}
2384
2385/**
2386 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2387 *
2388 * @prev: previous task during sched_switch
2389 * @next: next task during sched_switch
2390 * @flags: TRACE_RECORD_CMDLINE for recording comm
2391 * TRACE_RECORD_TGID for recording tgid
2392 */
2393void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2394 struct task_struct *next, int flags)
2395{
2396 bool done;
2397
2398 if (tracing_record_taskinfo_skip(flags))
2399 return;
2400
2401 /*
2402 * Record as much task information as possible. If some fail, continue
2403 * to try to record the others.
2404 */
2405 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2406 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2407 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2408 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2409
2410 /* If recording any information failed, retry again soon. */
2411 if (!done)
2412 return;
2413
2414 __this_cpu_write(trace_taskinfo_save, false);
2415}
2416
2417/* Helpers to record a specific task information */
2418void tracing_record_cmdline(struct task_struct *task)
2419{
2420 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2421}
2422
2423void tracing_record_tgid(struct task_struct *task)
2424{
2425 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2426}
2427
2428/*
2429 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2430 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2431 * simplifies those functions and keeps them in sync.
2432 */
2433enum print_line_t trace_handle_return(struct trace_seq *s)
2434{
2435 return trace_seq_has_overflowed(s) ?
2436 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2437}
2438EXPORT_SYMBOL_GPL(trace_handle_return);
2439
2440void
2441tracing_generic_entry_update(struct trace_entry *entry, unsigned short type,
2442 unsigned long flags, int pc)
2443{
2444 struct task_struct *tsk = current;
2445
2446 entry->preempt_count = pc & 0xff;
2447 entry->pid = (tsk) ? tsk->pid : 0;
2448 entry->type = type;
2449 entry->flags =
2450#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
2451 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
2452#else
2453 TRACE_FLAG_IRQS_NOSUPPORT |
2454#endif
2455 ((pc & NMI_MASK ) ? TRACE_FLAG_NMI : 0) |
2456 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
2457 ((pc & SOFTIRQ_OFFSET) ? TRACE_FLAG_SOFTIRQ : 0) |
2458 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
2459 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
2460}
2461EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
2462
2463struct ring_buffer_event *
2464trace_buffer_lock_reserve(struct trace_buffer *buffer,
2465 int type,
2466 unsigned long len,
2467 unsigned long flags, int pc)
2468{
2469 return __trace_buffer_lock_reserve(buffer, type, len, flags, pc);
2470}
2471
2472DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2473DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2474static int trace_buffered_event_ref;
2475
2476/**
2477 * trace_buffered_event_enable - enable buffering events
2478 *
2479 * When events are being filtered, it is quicker to use a temporary
2480 * buffer to write the event data into if there's a likely chance
2481 * that it will not be committed. The discard of the ring buffer
2482 * is not as fast as committing, and is much slower than copying
2483 * a commit.
2484 *
2485 * When an event is to be filtered, allocate per cpu buffers to
2486 * write the event data into, and if the event is filtered and discarded
2487 * it is simply dropped, otherwise, the entire data is to be committed
2488 * in one shot.
2489 */
2490void trace_buffered_event_enable(void)
2491{
2492 struct ring_buffer_event *event;
2493 struct page *page;
2494 int cpu;
2495
2496 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2497
2498 if (trace_buffered_event_ref++)
2499 return;
2500
2501 for_each_tracing_cpu(cpu) {
2502 page = alloc_pages_node(cpu_to_node(cpu),
2503 GFP_KERNEL | __GFP_NORETRY, 0);
2504 if (!page)
2505 goto failed;
2506
2507 event = page_address(page);
2508 memset(event, 0, sizeof(*event));
2509
2510 per_cpu(trace_buffered_event, cpu) = event;
2511
2512 preempt_disable();
2513 if (cpu == smp_processor_id() &&
2514 this_cpu_read(trace_buffered_event) !=
2515 per_cpu(trace_buffered_event, cpu))
2516 WARN_ON_ONCE(1);
2517 preempt_enable();
2518 }
2519
2520 return;
2521 failed:
2522 trace_buffered_event_disable();
2523}
2524
2525static void enable_trace_buffered_event(void *data)
2526{
2527 /* Probably not needed, but do it anyway */
2528 smp_rmb();
2529 this_cpu_dec(trace_buffered_event_cnt);
2530}
2531
2532static void disable_trace_buffered_event(void *data)
2533{
2534 this_cpu_inc(trace_buffered_event_cnt);
2535}
2536
2537/**
2538 * trace_buffered_event_disable - disable buffering events
2539 *
2540 * When a filter is removed, it is faster to not use the buffered
2541 * events, and to commit directly into the ring buffer. Free up
2542 * the temp buffers when there are no more users. This requires
2543 * special synchronization with current events.
2544 */
2545void trace_buffered_event_disable(void)
2546{
2547 int cpu;
2548
2549 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2550
2551 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2552 return;
2553
2554 if (--trace_buffered_event_ref)
2555 return;
2556
2557 preempt_disable();
2558 /* For each CPU, set the buffer as used. */
2559 smp_call_function_many(tracing_buffer_mask,
2560 disable_trace_buffered_event, NULL, 1);
2561 preempt_enable();
2562
2563 /* Wait for all current users to finish */
2564 synchronize_rcu();
2565
2566 for_each_tracing_cpu(cpu) {
2567 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2568 per_cpu(trace_buffered_event, cpu) = NULL;
2569 }
2570 /*
2571 * Make sure trace_buffered_event is NULL before clearing
2572 * trace_buffered_event_cnt.
2573 */
2574 smp_wmb();
2575
2576 preempt_disable();
2577 /* Do the work on each cpu */
2578 smp_call_function_many(tracing_buffer_mask,
2579 enable_trace_buffered_event, NULL, 1);
2580 preempt_enable();
2581}
2582
2583static struct trace_buffer *temp_buffer;
2584
2585struct ring_buffer_event *
2586trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2587 struct trace_event_file *trace_file,
2588 int type, unsigned long len,
2589 unsigned long flags, int pc)
2590{
2591 struct ring_buffer_event *entry;
2592 int val;
2593
2594 *current_rb = trace_file->tr->array_buffer.buffer;
2595
2596 if (!ring_buffer_time_stamp_abs(*current_rb) && (trace_file->flags &
2597 (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED)) &&
2598 (entry = this_cpu_read(trace_buffered_event))) {
2599 /* Try to use the per cpu buffer first */
2600 val = this_cpu_inc_return(trace_buffered_event_cnt);
2601 if (val == 1) {
2602 trace_event_setup(entry, type, flags, pc);
2603 entry->array[0] = len;
2604 return entry;
2605 }
2606 this_cpu_dec(trace_buffered_event_cnt);
2607 }
2608
2609 entry = __trace_buffer_lock_reserve(*current_rb,
2610 type, len, flags, pc);
2611 /*
2612 * If tracing is off, but we have triggers enabled
2613 * we still need to look at the event data. Use the temp_buffer
2614 * to store the trace event for the tigger to use. It's recusive
2615 * safe and will not be recorded anywhere.
2616 */
2617 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2618 *current_rb = temp_buffer;
2619 entry = __trace_buffer_lock_reserve(*current_rb,
2620 type, len, flags, pc);
2621 }
2622 return entry;
2623}
2624EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2625
2626static DEFINE_SPINLOCK(tracepoint_iter_lock);
2627static DEFINE_MUTEX(tracepoint_printk_mutex);
2628
2629static void output_printk(struct trace_event_buffer *fbuffer)
2630{
2631 struct trace_event_call *event_call;
2632 struct trace_event_file *file;
2633 struct trace_event *event;
2634 unsigned long flags;
2635 struct trace_iterator *iter = tracepoint_print_iter;
2636
2637 /* We should never get here if iter is NULL */
2638 if (WARN_ON_ONCE(!iter))
2639 return;
2640
2641 event_call = fbuffer->trace_file->event_call;
2642 if (!event_call || !event_call->event.funcs ||
2643 !event_call->event.funcs->trace)
2644 return;
2645
2646 file = fbuffer->trace_file;
2647 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2648 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2649 !filter_match_preds(file->filter, fbuffer->entry)))
2650 return;
2651
2652 event = &fbuffer->trace_file->event_call->event;
2653
2654 spin_lock_irqsave(&tracepoint_iter_lock, flags);
2655 trace_seq_init(&iter->seq);
2656 iter->ent = fbuffer->entry;
2657 event_call->event.funcs->trace(iter, 0, event);
2658 trace_seq_putc(&iter->seq, 0);
2659 printk("%s", iter->seq.buffer);
2660
2661 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2662}
2663
2664int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2665 void *buffer, size_t *lenp,
2666 loff_t *ppos)
2667{
2668 int save_tracepoint_printk;
2669 int ret;
2670
2671 mutex_lock(&tracepoint_printk_mutex);
2672 save_tracepoint_printk = tracepoint_printk;
2673
2674 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2675
2676 /*
2677 * This will force exiting early, as tracepoint_printk
2678 * is always zero when tracepoint_printk_iter is not allocated
2679 */
2680 if (!tracepoint_print_iter)
2681 tracepoint_printk = 0;
2682
2683 if (save_tracepoint_printk == tracepoint_printk)
2684 goto out;
2685
2686 if (tracepoint_printk)
2687 static_key_enable(&tracepoint_printk_key.key);
2688 else
2689 static_key_disable(&tracepoint_printk_key.key);
2690
2691 out:
2692 mutex_unlock(&tracepoint_printk_mutex);
2693
2694 return ret;
2695}
2696
2697void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2698{
2699 if (static_key_false(&tracepoint_printk_key.key))
2700 output_printk(fbuffer);
2701
2702 event_trigger_unlock_commit_regs(fbuffer->trace_file, fbuffer->buffer,
2703 fbuffer->event, fbuffer->entry,
2704 fbuffer->flags, fbuffer->pc, fbuffer->regs);
2705}
2706EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2707
2708/*
2709 * Skip 3:
2710 *
2711 * trace_buffer_unlock_commit_regs()
2712 * trace_event_buffer_commit()
2713 * trace_event_raw_event_xxx()
2714 */
2715# define STACK_SKIP 3
2716
2717void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2718 struct trace_buffer *buffer,
2719 struct ring_buffer_event *event,
2720 unsigned long flags, int pc,
2721 struct pt_regs *regs)
2722{
2723 __buffer_unlock_commit(buffer, event);
2724
2725 /*
2726 * If regs is not set, then skip the necessary functions.
2727 * Note, we can still get here via blktrace, wakeup tracer
2728 * and mmiotrace, but that's ok if they lose a function or
2729 * two. They are not that meaningful.
2730 */
2731 ftrace_trace_stack(tr, buffer, flags, regs ? 0 : STACK_SKIP, pc, regs);
2732 ftrace_trace_userstack(buffer, flags, pc);
2733}
2734
2735/*
2736 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2737 */
2738void
2739trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2740 struct ring_buffer_event *event)
2741{
2742 __buffer_unlock_commit(buffer, event);
2743}
2744
2745static void
2746trace_process_export(struct trace_export *export,
2747 struct ring_buffer_event *event)
2748{
2749 struct trace_entry *entry;
2750 unsigned int size = 0;
2751
2752 entry = ring_buffer_event_data(event);
2753 size = ring_buffer_event_length(event);
2754 export->write(export, entry, size);
2755}
2756
2757static DEFINE_MUTEX(ftrace_export_lock);
2758
2759static struct trace_export __rcu *ftrace_exports_list __read_mostly;
2760
2761static DEFINE_STATIC_KEY_FALSE(ftrace_exports_enabled);
2762
2763static inline void ftrace_exports_enable(void)
2764{
2765 static_branch_enable(&ftrace_exports_enabled);
2766}
2767
2768static inline void ftrace_exports_disable(void)
2769{
2770 static_branch_disable(&ftrace_exports_enabled);
2771}
2772
2773static void ftrace_exports(struct ring_buffer_event *event)
2774{
2775 struct trace_export *export;
2776
2777 preempt_disable_notrace();
2778
2779 export = rcu_dereference_raw_check(ftrace_exports_list);
2780 while (export) {
2781 trace_process_export(export, event);
2782 export = rcu_dereference_raw_check(export->next);
2783 }
2784
2785 preempt_enable_notrace();
2786}
2787
2788static inline void
2789add_trace_export(struct trace_export **list, struct trace_export *export)
2790{
2791 rcu_assign_pointer(export->next, *list);
2792 /*
2793 * We are entering export into the list but another
2794 * CPU might be walking that list. We need to make sure
2795 * the export->next pointer is valid before another CPU sees
2796 * the export pointer included into the list.
2797 */
2798 rcu_assign_pointer(*list, export);
2799}
2800
2801static inline int
2802rm_trace_export(struct trace_export **list, struct trace_export *export)
2803{
2804 struct trace_export **p;
2805
2806 for (p = list; *p != NULL; p = &(*p)->next)
2807 if (*p == export)
2808 break;
2809
2810 if (*p != export)
2811 return -1;
2812
2813 rcu_assign_pointer(*p, (*p)->next);
2814
2815 return 0;
2816}
2817
2818static inline void
2819add_ftrace_export(struct trace_export **list, struct trace_export *export)
2820{
2821 if (*list == NULL)
2822 ftrace_exports_enable();
2823
2824 add_trace_export(list, export);
2825}
2826
2827static inline int
2828rm_ftrace_export(struct trace_export **list, struct trace_export *export)
2829{
2830 int ret;
2831
2832 ret = rm_trace_export(list, export);
2833 if (*list == NULL)
2834 ftrace_exports_disable();
2835
2836 return ret;
2837}
2838
2839int register_ftrace_export(struct trace_export *export)
2840{
2841 if (WARN_ON_ONCE(!export->write))
2842 return -1;
2843
2844 mutex_lock(&ftrace_export_lock);
2845
2846 add_ftrace_export(&ftrace_exports_list, export);
2847
2848 mutex_unlock(&ftrace_export_lock);
2849
2850 return 0;
2851}
2852EXPORT_SYMBOL_GPL(register_ftrace_export);
2853
2854int unregister_ftrace_export(struct trace_export *export)
2855{
2856 int ret;
2857
2858 mutex_lock(&ftrace_export_lock);
2859
2860 ret = rm_ftrace_export(&ftrace_exports_list, export);
2861
2862 mutex_unlock(&ftrace_export_lock);
2863
2864 return ret;
2865}
2866EXPORT_SYMBOL_GPL(unregister_ftrace_export);
2867
2868void
2869trace_function(struct trace_array *tr,
2870 unsigned long ip, unsigned long parent_ip, unsigned long flags,
2871 int pc)
2872{
2873 struct trace_event_call *call = &event_function;
2874 struct trace_buffer *buffer = tr->array_buffer.buffer;
2875 struct ring_buffer_event *event;
2876 struct ftrace_entry *entry;
2877
2878 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2879 flags, pc);
2880 if (!event)
2881 return;
2882 entry = ring_buffer_event_data(event);
2883 entry->ip = ip;
2884 entry->parent_ip = parent_ip;
2885
2886 if (!call_filter_check_discard(call, entry, buffer, event)) {
2887 if (static_branch_unlikely(&ftrace_exports_enabled))
2888 ftrace_exports(event);
2889 __buffer_unlock_commit(buffer, event);
2890 }
2891}
2892
2893#ifdef CONFIG_STACKTRACE
2894
2895/* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2896#define FTRACE_KSTACK_NESTING 4
2897
2898#define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
2899
2900struct ftrace_stack {
2901 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2902};
2903
2904
2905struct ftrace_stacks {
2906 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
2907};
2908
2909static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
2910static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2911
2912static void __ftrace_trace_stack(struct trace_buffer *buffer,
2913 unsigned long flags,
2914 int skip, int pc, struct pt_regs *regs)
2915{
2916 struct trace_event_call *call = &event_kernel_stack;
2917 struct ring_buffer_event *event;
2918 unsigned int size, nr_entries;
2919 struct ftrace_stack *fstack;
2920 struct stack_entry *entry;
2921 int stackidx;
2922
2923 /*
2924 * Add one, for this function and the call to save_stack_trace()
2925 * If regs is set, then these functions will not be in the way.
2926 */
2927#ifndef CONFIG_UNWINDER_ORC
2928 if (!regs)
2929 skip++;
2930#endif
2931
2932 preempt_disable_notrace();
2933
2934 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
2935
2936 /* This should never happen. If it does, yell once and skip */
2937 if (WARN_ON_ONCE(stackidx > FTRACE_KSTACK_NESTING))
2938 goto out;
2939
2940 /*
2941 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
2942 * interrupt will either see the value pre increment or post
2943 * increment. If the interrupt happens pre increment it will have
2944 * restored the counter when it returns. We just need a barrier to
2945 * keep gcc from moving things around.
2946 */
2947 barrier();
2948
2949 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
2950 size = ARRAY_SIZE(fstack->calls);
2951
2952 if (regs) {
2953 nr_entries = stack_trace_save_regs(regs, fstack->calls,
2954 size, skip);
2955 } else {
2956 nr_entries = stack_trace_save(fstack->calls, size, skip);
2957 }
2958
2959 size = nr_entries * sizeof(unsigned long);
2960 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2961 sizeof(*entry) + size, flags, pc);
2962 if (!event)
2963 goto out;
2964 entry = ring_buffer_event_data(event);
2965
2966 memcpy(&entry->caller, fstack->calls, size);
2967 entry->size = nr_entries;
2968
2969 if (!call_filter_check_discard(call, entry, buffer, event))
2970 __buffer_unlock_commit(buffer, event);
2971
2972 out:
2973 /* Again, don't let gcc optimize things here */
2974 barrier();
2975 __this_cpu_dec(ftrace_stack_reserve);
2976 preempt_enable_notrace();
2977
2978}
2979
2980static inline void ftrace_trace_stack(struct trace_array *tr,
2981 struct trace_buffer *buffer,
2982 unsigned long flags,
2983 int skip, int pc, struct pt_regs *regs)
2984{
2985 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
2986 return;
2987
2988 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
2989}
2990
2991void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
2992 int pc)
2993{
2994 struct trace_buffer *buffer = tr->array_buffer.buffer;
2995
2996 if (rcu_is_watching()) {
2997 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
2998 return;
2999 }
3000
3001 /*
3002 * When an NMI triggers, RCU is enabled via rcu_nmi_enter(),
3003 * but if the above rcu_is_watching() failed, then the NMI
3004 * triggered someplace critical, and rcu_irq_enter() should
3005 * not be called from NMI.
3006 */
3007 if (unlikely(in_nmi()))
3008 return;
3009
3010 rcu_irq_enter_irqson();
3011 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
3012 rcu_irq_exit_irqson();
3013}
3014
3015/**
3016 * trace_dump_stack - record a stack back trace in the trace buffer
3017 * @skip: Number of functions to skip (helper handlers)
3018 */
3019void trace_dump_stack(int skip)
3020{
3021 unsigned long flags;
3022
3023 if (tracing_disabled || tracing_selftest_running)
3024 return;
3025
3026 local_save_flags(flags);
3027
3028#ifndef CONFIG_UNWINDER_ORC
3029 /* Skip 1 to skip this function. */
3030 skip++;
3031#endif
3032 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3033 flags, skip, preempt_count(), NULL);
3034}
3035EXPORT_SYMBOL_GPL(trace_dump_stack);
3036
3037#ifdef CONFIG_USER_STACKTRACE_SUPPORT
3038static DEFINE_PER_CPU(int, user_stack_count);
3039
3040static void
3041ftrace_trace_userstack(struct trace_buffer *buffer, unsigned long flags, int pc)
3042{
3043 struct trace_event_call *call = &event_user_stack;
3044 struct ring_buffer_event *event;
3045 struct userstack_entry *entry;
3046
3047 if (!(global_trace.trace_flags & TRACE_ITER_USERSTACKTRACE))
3048 return;
3049
3050 /*
3051 * NMIs can not handle page faults, even with fix ups.
3052 * The save user stack can (and often does) fault.
3053 */
3054 if (unlikely(in_nmi()))
3055 return;
3056
3057 /*
3058 * prevent recursion, since the user stack tracing may
3059 * trigger other kernel events.
3060 */
3061 preempt_disable();
3062 if (__this_cpu_read(user_stack_count))
3063 goto out;
3064
3065 __this_cpu_inc(user_stack_count);
3066
3067 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3068 sizeof(*entry), flags, pc);
3069 if (!event)
3070 goto out_drop_count;
3071 entry = ring_buffer_event_data(event);
3072
3073 entry->tgid = current->tgid;
3074 memset(&entry->caller, 0, sizeof(entry->caller));
3075
3076 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3077 if (!call_filter_check_discard(call, entry, buffer, event))
3078 __buffer_unlock_commit(buffer, event);
3079
3080 out_drop_count:
3081 __this_cpu_dec(user_stack_count);
3082 out:
3083 preempt_enable();
3084}
3085#else /* CONFIG_USER_STACKTRACE_SUPPORT */
3086static void ftrace_trace_userstack(struct trace_buffer *buffer,
3087 unsigned long flags, int pc)
3088{
3089}
3090#endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3091
3092#endif /* CONFIG_STACKTRACE */
3093
3094/* created for use with alloc_percpu */
3095struct trace_buffer_struct {
3096 int nesting;
3097 char buffer[4][TRACE_BUF_SIZE];
3098};
3099
3100static struct trace_buffer_struct *trace_percpu_buffer;
3101
3102/*
3103 * Thise allows for lockless recording. If we're nested too deeply, then
3104 * this returns NULL.
3105 */
3106static char *get_trace_buf(void)
3107{
3108 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3109
3110 if (!buffer || buffer->nesting >= 4)
3111 return NULL;
3112
3113 buffer->nesting++;
3114
3115 /* Interrupts must see nesting incremented before we use the buffer */
3116 barrier();
3117 return &buffer->buffer[buffer->nesting][0];
3118}
3119
3120static void put_trace_buf(void)
3121{
3122 /* Don't let the decrement of nesting leak before this */
3123 barrier();
3124 this_cpu_dec(trace_percpu_buffer->nesting);
3125}
3126
3127static int alloc_percpu_trace_buffer(void)
3128{
3129 struct trace_buffer_struct *buffers;
3130
3131 if (trace_percpu_buffer)
3132 return 0;
3133
3134 buffers = alloc_percpu(struct trace_buffer_struct);
3135 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3136 return -ENOMEM;
3137
3138 trace_percpu_buffer = buffers;
3139 return 0;
3140}
3141
3142static int buffers_allocated;
3143
3144void trace_printk_init_buffers(void)
3145{
3146 if (buffers_allocated)
3147 return;
3148
3149 if (alloc_percpu_trace_buffer())
3150 return;
3151
3152 /* trace_printk() is for debug use only. Don't use it in production. */
3153
3154 pr_warn("\n");
3155 pr_warn("**********************************************************\n");
3156 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3157 pr_warn("** **\n");
3158 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3159 pr_warn("** **\n");
3160 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3161 pr_warn("** unsafe for production use. **\n");
3162 pr_warn("** **\n");
3163 pr_warn("** If you see this message and you are not debugging **\n");
3164 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3165 pr_warn("** **\n");
3166 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3167 pr_warn("**********************************************************\n");
3168
3169 /* Expand the buffers to set size */
3170 tracing_update_buffers();
3171
3172 buffers_allocated = 1;
3173
3174 /*
3175 * trace_printk_init_buffers() can be called by modules.
3176 * If that happens, then we need to start cmdline recording
3177 * directly here. If the global_trace.buffer is already
3178 * allocated here, then this was called by module code.
3179 */
3180 if (global_trace.array_buffer.buffer)
3181 tracing_start_cmdline_record();
3182}
3183EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3184
3185void trace_printk_start_comm(void)
3186{
3187 /* Start tracing comms if trace printk is set */
3188 if (!buffers_allocated)
3189 return;
3190 tracing_start_cmdline_record();
3191}
3192
3193static void trace_printk_start_stop_comm(int enabled)
3194{
3195 if (!buffers_allocated)
3196 return;
3197
3198 if (enabled)
3199 tracing_start_cmdline_record();
3200 else
3201 tracing_stop_cmdline_record();
3202}
3203
3204/**
3205 * trace_vbprintk - write binary msg to tracing buffer
3206 * @ip: The address of the caller
3207 * @fmt: The string format to write to the buffer
3208 * @args: Arguments for @fmt
3209 */
3210int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3211{
3212 struct trace_event_call *call = &event_bprint;
3213 struct ring_buffer_event *event;
3214 struct trace_buffer *buffer;
3215 struct trace_array *tr = &global_trace;
3216 struct bprint_entry *entry;
3217 unsigned long flags;
3218 char *tbuffer;
3219 int len = 0, size, pc;
3220
3221 if (unlikely(tracing_selftest_running || tracing_disabled))
3222 return 0;
3223
3224 /* Don't pollute graph traces with trace_vprintk internals */
3225 pause_graph_tracing();
3226
3227 pc = preempt_count();
3228 preempt_disable_notrace();
3229
3230 tbuffer = get_trace_buf();
3231 if (!tbuffer) {
3232 len = 0;
3233 goto out_nobuffer;
3234 }
3235
3236 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3237
3238 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3239 goto out_put;
3240
3241 local_save_flags(flags);
3242 size = sizeof(*entry) + sizeof(u32) * len;
3243 buffer = tr->array_buffer.buffer;
3244 ring_buffer_nest_start(buffer);
3245 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3246 flags, pc);
3247 if (!event)
3248 goto out;
3249 entry = ring_buffer_event_data(event);
3250 entry->ip = ip;
3251 entry->fmt = fmt;
3252
3253 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3254 if (!call_filter_check_discard(call, entry, buffer, event)) {
3255 __buffer_unlock_commit(buffer, event);
3256 ftrace_trace_stack(tr, buffer, flags, 6, pc, NULL);
3257 }
3258
3259out:
3260 ring_buffer_nest_end(buffer);
3261out_put:
3262 put_trace_buf();
3263
3264out_nobuffer:
3265 preempt_enable_notrace();
3266 unpause_graph_tracing();
3267
3268 return len;
3269}
3270EXPORT_SYMBOL_GPL(trace_vbprintk);
3271
3272__printf(3, 0)
3273static int
3274__trace_array_vprintk(struct trace_buffer *buffer,
3275 unsigned long ip, const char *fmt, va_list args)
3276{
3277 struct trace_event_call *call = &event_print;
3278 struct ring_buffer_event *event;
3279 int len = 0, size, pc;
3280 struct print_entry *entry;
3281 unsigned long flags;
3282 char *tbuffer;
3283
3284 if (tracing_disabled || tracing_selftest_running)
3285 return 0;
3286
3287 /* Don't pollute graph traces with trace_vprintk internals */
3288 pause_graph_tracing();
3289
3290 pc = preempt_count();
3291 preempt_disable_notrace();
3292
3293
3294 tbuffer = get_trace_buf();
3295 if (!tbuffer) {
3296 len = 0;
3297 goto out_nobuffer;
3298 }
3299
3300 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3301
3302 local_save_flags(flags);
3303 size = sizeof(*entry) + len + 1;
3304 ring_buffer_nest_start(buffer);
3305 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3306 flags, pc);
3307 if (!event)
3308 goto out;
3309 entry = ring_buffer_event_data(event);
3310 entry->ip = ip;
3311
3312 memcpy(&entry->buf, tbuffer, len + 1);
3313 if (!call_filter_check_discard(call, entry, buffer, event)) {
3314 __buffer_unlock_commit(buffer, event);
3315 ftrace_trace_stack(&global_trace, buffer, flags, 6, pc, NULL);
3316 }
3317
3318out:
3319 ring_buffer_nest_end(buffer);
3320 put_trace_buf();
3321
3322out_nobuffer:
3323 preempt_enable_notrace();
3324 unpause_graph_tracing();
3325
3326 return len;
3327}
3328
3329__printf(3, 0)
3330int trace_array_vprintk(struct trace_array *tr,
3331 unsigned long ip, const char *fmt, va_list args)
3332{
3333 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3334}
3335
3336/**
3337 * trace_array_printk - Print a message to a specific instance
3338 * @tr: The instance trace_array descriptor
3339 * @ip: The instruction pointer that this is called from.
3340 * @fmt: The format to print (printf format)
3341 *
3342 * If a subsystem sets up its own instance, they have the right to
3343 * printk strings into their tracing instance buffer using this
3344 * function. Note, this function will not write into the top level
3345 * buffer (use trace_printk() for that), as writing into the top level
3346 * buffer should only have events that can be individually disabled.
3347 * trace_printk() is only used for debugging a kernel, and should not
3348 * be ever encorporated in normal use.
3349 *
3350 * trace_array_printk() can be used, as it will not add noise to the
3351 * top level tracing buffer.
3352 *
3353 * Note, trace_array_init_printk() must be called on @tr before this
3354 * can be used.
3355 */
3356__printf(3, 0)
3357int trace_array_printk(struct trace_array *tr,
3358 unsigned long ip, const char *fmt, ...)
3359{
3360 int ret;
3361 va_list ap;
3362
3363 if (!tr)
3364 return -ENOENT;
3365
3366 /* This is only allowed for created instances */
3367 if (tr == &global_trace)
3368 return 0;
3369
3370 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3371 return 0;
3372
3373 va_start(ap, fmt);
3374 ret = trace_array_vprintk(tr, ip, fmt, ap);
3375 va_end(ap);
3376 return ret;
3377}
3378EXPORT_SYMBOL_GPL(trace_array_printk);
3379
3380/**
3381 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3382 * @tr: The trace array to initialize the buffers for
3383 *
3384 * As trace_array_printk() only writes into instances, they are OK to
3385 * have in the kernel (unlike trace_printk()). This needs to be called
3386 * before trace_array_printk() can be used on a trace_array.
3387 */
3388int trace_array_init_printk(struct trace_array *tr)
3389{
3390 if (!tr)
3391 return -ENOENT;
3392
3393 /* This is only allowed for created instances */
3394 if (tr == &global_trace)
3395 return -EINVAL;
3396
3397 return alloc_percpu_trace_buffer();
3398}
3399EXPORT_SYMBOL_GPL(trace_array_init_printk);
3400
3401__printf(3, 4)
3402int trace_array_printk_buf(struct trace_buffer *buffer,
3403 unsigned long ip, const char *fmt, ...)
3404{
3405 int ret;
3406 va_list ap;
3407
3408 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3409 return 0;
3410
3411 va_start(ap, fmt);
3412 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3413 va_end(ap);
3414 return ret;
3415}
3416
3417__printf(2, 0)
3418int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3419{
3420 return trace_array_vprintk(&global_trace, ip, fmt, args);
3421}
3422EXPORT_SYMBOL_GPL(trace_vprintk);
3423
3424static void trace_iterator_increment(struct trace_iterator *iter)
3425{
3426 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3427
3428 iter->idx++;
3429 if (buf_iter)
3430 ring_buffer_iter_advance(buf_iter);
3431}
3432
3433static struct trace_entry *
3434peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3435 unsigned long *lost_events)
3436{
3437 struct ring_buffer_event *event;
3438 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3439
3440 if (buf_iter) {
3441 event = ring_buffer_iter_peek(buf_iter, ts);
3442 if (lost_events)
3443 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3444 (unsigned long)-1 : 0;
3445 } else {
3446 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3447 lost_events);
3448 }
3449
3450 if (event) {
3451 iter->ent_size = ring_buffer_event_length(event);
3452 return ring_buffer_event_data(event);
3453 }
3454 iter->ent_size = 0;
3455 return NULL;
3456}
3457
3458static struct trace_entry *
3459__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3460 unsigned long *missing_events, u64 *ent_ts)
3461{
3462 struct trace_buffer *buffer = iter->array_buffer->buffer;
3463 struct trace_entry *ent, *next = NULL;
3464 unsigned long lost_events = 0, next_lost = 0;
3465 int cpu_file = iter->cpu_file;
3466 u64 next_ts = 0, ts;
3467 int next_cpu = -1;
3468 int next_size = 0;
3469 int cpu;
3470
3471 /*
3472 * If we are in a per_cpu trace file, don't bother by iterating over
3473 * all cpu and peek directly.
3474 */
3475 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3476 if (ring_buffer_empty_cpu(buffer, cpu_file))
3477 return NULL;
3478 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3479 if (ent_cpu)
3480 *ent_cpu = cpu_file;
3481
3482 return ent;
3483 }
3484
3485 for_each_tracing_cpu(cpu) {
3486
3487 if (ring_buffer_empty_cpu(buffer, cpu))
3488 continue;
3489
3490 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3491
3492 /*
3493 * Pick the entry with the smallest timestamp:
3494 */
3495 if (ent && (!next || ts < next_ts)) {
3496 next = ent;
3497 next_cpu = cpu;
3498 next_ts = ts;
3499 next_lost = lost_events;
3500 next_size = iter->ent_size;
3501 }
3502 }
3503
3504 iter->ent_size = next_size;
3505
3506 if (ent_cpu)
3507 *ent_cpu = next_cpu;
3508
3509 if (ent_ts)
3510 *ent_ts = next_ts;
3511
3512 if (missing_events)
3513 *missing_events = next_lost;
3514
3515 return next;
3516}
3517
3518#define STATIC_TEMP_BUF_SIZE 128
3519static char static_temp_buf[STATIC_TEMP_BUF_SIZE];
3520
3521/* Find the next real entry, without updating the iterator itself */
3522struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3523 int *ent_cpu, u64 *ent_ts)
3524{
3525 /* __find_next_entry will reset ent_size */
3526 int ent_size = iter->ent_size;
3527 struct trace_entry *entry;
3528
3529 /*
3530 * If called from ftrace_dump(), then the iter->temp buffer
3531 * will be the static_temp_buf and not created from kmalloc.
3532 * If the entry size is greater than the buffer, we can
3533 * not save it. Just return NULL in that case. This is only
3534 * used to add markers when two consecutive events' time
3535 * stamps have a large delta. See trace_print_lat_context()
3536 */
3537 if (iter->temp == static_temp_buf &&
3538 STATIC_TEMP_BUF_SIZE < ent_size)
3539 return NULL;
3540
3541 /*
3542 * The __find_next_entry() may call peek_next_entry(), which may
3543 * call ring_buffer_peek() that may make the contents of iter->ent
3544 * undefined. Need to copy iter->ent now.
3545 */
3546 if (iter->ent && iter->ent != iter->temp) {
3547 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3548 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3549 void *temp;
3550 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3551 if (!temp)
3552 return NULL;
3553 kfree(iter->temp);
3554 iter->temp = temp;
3555 iter->temp_size = iter->ent_size;
3556 }
3557 memcpy(iter->temp, iter->ent, iter->ent_size);
3558 iter->ent = iter->temp;
3559 }
3560 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3561 /* Put back the original ent_size */
3562 iter->ent_size = ent_size;
3563
3564 return entry;
3565}
3566
3567/* Find the next real entry, and increment the iterator to the next entry */
3568void *trace_find_next_entry_inc(struct trace_iterator *iter)
3569{
3570 iter->ent = __find_next_entry(iter, &iter->cpu,
3571 &iter->lost_events, &iter->ts);
3572
3573 if (iter->ent)
3574 trace_iterator_increment(iter);
3575
3576 return iter->ent ? iter : NULL;
3577}
3578
3579static void trace_consume(struct trace_iterator *iter)
3580{
3581 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3582 &iter->lost_events);
3583}
3584
3585static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3586{
3587 struct trace_iterator *iter = m->private;
3588 int i = (int)*pos;
3589 void *ent;
3590
3591 WARN_ON_ONCE(iter->leftover);
3592
3593 (*pos)++;
3594
3595 /* can't go backwards */
3596 if (iter->idx > i)
3597 return NULL;
3598
3599 if (iter->idx < 0)
3600 ent = trace_find_next_entry_inc(iter);
3601 else
3602 ent = iter;
3603
3604 while (ent && iter->idx < i)
3605 ent = trace_find_next_entry_inc(iter);
3606
3607 iter->pos = *pos;
3608
3609 return ent;
3610}
3611
3612void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3613{
3614 struct ring_buffer_iter *buf_iter;
3615 unsigned long entries = 0;
3616 u64 ts;
3617
3618 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
3619
3620 buf_iter = trace_buffer_iter(iter, cpu);
3621 if (!buf_iter)
3622 return;
3623
3624 ring_buffer_iter_reset(buf_iter);
3625
3626 /*
3627 * We could have the case with the max latency tracers
3628 * that a reset never took place on a cpu. This is evident
3629 * by the timestamp being before the start of the buffer.
3630 */
3631 while (ring_buffer_iter_peek(buf_iter, &ts)) {
3632 if (ts >= iter->array_buffer->time_start)
3633 break;
3634 entries++;
3635 ring_buffer_iter_advance(buf_iter);
3636 }
3637
3638 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
3639}
3640
3641/*
3642 * The current tracer is copied to avoid a global locking
3643 * all around.
3644 */
3645static void *s_start(struct seq_file *m, loff_t *pos)
3646{
3647 struct trace_iterator *iter = m->private;
3648 struct trace_array *tr = iter->tr;
3649 int cpu_file = iter->cpu_file;
3650 void *p = NULL;
3651 loff_t l = 0;
3652 int cpu;
3653
3654 /*
3655 * copy the tracer to avoid using a global lock all around.
3656 * iter->trace is a copy of current_trace, the pointer to the
3657 * name may be used instead of a strcmp(), as iter->trace->name
3658 * will point to the same string as current_trace->name.
3659 */
3660 mutex_lock(&trace_types_lock);
3661 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
3662 *iter->trace = *tr->current_trace;
3663 mutex_unlock(&trace_types_lock);
3664
3665#ifdef CONFIG_TRACER_MAX_TRACE
3666 if (iter->snapshot && iter->trace->use_max_tr)
3667 return ERR_PTR(-EBUSY);
3668#endif
3669
3670 if (!iter->snapshot)
3671 atomic_inc(&trace_record_taskinfo_disabled);
3672
3673 if (*pos != iter->pos) {
3674 iter->ent = NULL;
3675 iter->cpu = 0;
3676 iter->idx = -1;
3677
3678 if (cpu_file == RING_BUFFER_ALL_CPUS) {
3679 for_each_tracing_cpu(cpu)
3680 tracing_iter_reset(iter, cpu);
3681 } else
3682 tracing_iter_reset(iter, cpu_file);
3683
3684 iter->leftover = 0;
3685 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
3686 ;
3687
3688 } else {
3689 /*
3690 * If we overflowed the seq_file before, then we want
3691 * to just reuse the trace_seq buffer again.
3692 */
3693 if (iter->leftover)
3694 p = iter;
3695 else {
3696 l = *pos - 1;
3697 p = s_next(m, p, &l);
3698 }
3699 }
3700
3701 trace_event_read_lock();
3702 trace_access_lock(cpu_file);
3703 return p;
3704}
3705
3706static void s_stop(struct seq_file *m, void *p)
3707{
3708 struct trace_iterator *iter = m->private;
3709
3710#ifdef CONFIG_TRACER_MAX_TRACE
3711 if (iter->snapshot && iter->trace->use_max_tr)
3712 return;
3713#endif
3714
3715 if (!iter->snapshot)
3716 atomic_dec(&trace_record_taskinfo_disabled);
3717
3718 trace_access_unlock(iter->cpu_file);
3719 trace_event_read_unlock();
3720}
3721
3722static void
3723get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
3724 unsigned long *entries, int cpu)
3725{
3726 unsigned long count;
3727
3728 count = ring_buffer_entries_cpu(buf->buffer, cpu);
3729 /*
3730 * If this buffer has skipped entries, then we hold all
3731 * entries for the trace and we need to ignore the
3732 * ones before the time stamp.
3733 */
3734 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
3735 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
3736 /* total is the same as the entries */
3737 *total = count;
3738 } else
3739 *total = count +
3740 ring_buffer_overrun_cpu(buf->buffer, cpu);
3741 *entries = count;
3742}
3743
3744static void
3745get_total_entries(struct array_buffer *buf,
3746 unsigned long *total, unsigned long *entries)
3747{
3748 unsigned long t, e;
3749 int cpu;
3750
3751 *total = 0;
3752 *entries = 0;
3753
3754 for_each_tracing_cpu(cpu) {
3755 get_total_entries_cpu(buf, &t, &e, cpu);
3756 *total += t;
3757 *entries += e;
3758 }
3759}
3760
3761unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
3762{
3763 unsigned long total, entries;
3764
3765 if (!tr)
3766 tr = &global_trace;
3767
3768 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
3769
3770 return entries;
3771}
3772
3773unsigned long trace_total_entries(struct trace_array *tr)
3774{
3775 unsigned long total, entries;
3776
3777 if (!tr)
3778 tr = &global_trace;
3779
3780 get_total_entries(&tr->array_buffer, &total, &entries);
3781
3782 return entries;
3783}
3784
3785static void print_lat_help_header(struct seq_file *m)
3786{
3787 seq_puts(m, "# _------=> CPU# \n"
3788 "# / _-----=> irqs-off \n"
3789 "# | / _----=> need-resched \n"
3790 "# || / _---=> hardirq/softirq \n"
3791 "# ||| / _--=> preempt-depth \n"
3792 "# |||| / delay \n"
3793 "# cmd pid ||||| time | caller \n"
3794 "# \\ / ||||| \\ | / \n");
3795}
3796
3797static void print_event_info(struct array_buffer *buf, struct seq_file *m)
3798{
3799 unsigned long total;
3800 unsigned long entries;
3801
3802 get_total_entries(buf, &total, &entries);
3803 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
3804 entries, total, num_online_cpus());
3805 seq_puts(m, "#\n");
3806}
3807
3808static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
3809 unsigned int flags)
3810{
3811 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3812
3813 print_event_info(buf, m);
3814
3815 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
3816 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
3817}
3818
3819static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
3820 unsigned int flags)
3821{
3822 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3823 const char *space = " ";
3824 int prec = tgid ? 12 : 2;
3825
3826 print_event_info(buf, m);
3827
3828 seq_printf(m, "# %.*s _-----=> irqs-off\n", prec, space);
3829 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
3830 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
3831 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
3832 seq_printf(m, "# %.*s||| / delay\n", prec, space);
3833 seq_printf(m, "# TASK-PID %.*s CPU# |||| TIMESTAMP FUNCTION\n", prec, " TGID ");
3834 seq_printf(m, "# | | %.*s | |||| | |\n", prec, " | ");
3835}
3836
3837void
3838print_trace_header(struct seq_file *m, struct trace_iterator *iter)
3839{
3840 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
3841 struct array_buffer *buf = iter->array_buffer;
3842 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
3843 struct tracer *type = iter->trace;
3844 unsigned long entries;
3845 unsigned long total;
3846 const char *name = "preemption";
3847
3848 name = type->name;
3849
3850 get_total_entries(buf, &total, &entries);
3851
3852 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
3853 name, UTS_RELEASE);
3854 seq_puts(m, "# -----------------------------------"
3855 "---------------------------------\n");
3856 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
3857 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
3858 nsecs_to_usecs(data->saved_latency),
3859 entries,
3860 total,
3861 buf->cpu,
3862#if defined(CONFIG_PREEMPT_NONE)
3863 "server",
3864#elif defined(CONFIG_PREEMPT_VOLUNTARY)
3865 "desktop",
3866#elif defined(CONFIG_PREEMPT)
3867 "preempt",
3868#elif defined(CONFIG_PREEMPT_RT)
3869 "preempt_rt",
3870#else
3871 "unknown",
3872#endif
3873 /* These are reserved for later use */
3874 0, 0, 0, 0);
3875#ifdef CONFIG_SMP
3876 seq_printf(m, " #P:%d)\n", num_online_cpus());
3877#else
3878 seq_puts(m, ")\n");
3879#endif
3880 seq_puts(m, "# -----------------\n");
3881 seq_printf(m, "# | task: %.16s-%d "
3882 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
3883 data->comm, data->pid,
3884 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
3885 data->policy, data->rt_priority);
3886 seq_puts(m, "# -----------------\n");
3887
3888 if (data->critical_start) {
3889 seq_puts(m, "# => started at: ");
3890 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
3891 trace_print_seq(m, &iter->seq);
3892 seq_puts(m, "\n# => ended at: ");
3893 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
3894 trace_print_seq(m, &iter->seq);
3895 seq_puts(m, "\n#\n");
3896 }
3897
3898 seq_puts(m, "#\n");
3899}
3900
3901static void test_cpu_buff_start(struct trace_iterator *iter)
3902{
3903 struct trace_seq *s = &iter->seq;
3904 struct trace_array *tr = iter->tr;
3905
3906 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
3907 return;
3908
3909 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
3910 return;
3911
3912 if (cpumask_available(iter->started) &&
3913 cpumask_test_cpu(iter->cpu, iter->started))
3914 return;
3915
3916 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
3917 return;
3918
3919 if (cpumask_available(iter->started))
3920 cpumask_set_cpu(iter->cpu, iter->started);
3921
3922 /* Don't print started cpu buffer for the first entry of the trace */
3923 if (iter->idx > 1)
3924 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
3925 iter->cpu);
3926}
3927
3928static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
3929{
3930 struct trace_array *tr = iter->tr;
3931 struct trace_seq *s = &iter->seq;
3932 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
3933 struct trace_entry *entry;
3934 struct trace_event *event;
3935
3936 entry = iter->ent;
3937
3938 test_cpu_buff_start(iter);
3939
3940 event = ftrace_find_event(entry->type);
3941
3942 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3943 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
3944 trace_print_lat_context(iter);
3945 else
3946 trace_print_context(iter);
3947 }
3948
3949 if (trace_seq_has_overflowed(s))
3950 return TRACE_TYPE_PARTIAL_LINE;
3951
3952 if (event)
3953 return event->funcs->trace(iter, sym_flags, event);
3954
3955 trace_seq_printf(s, "Unknown type %d\n", entry->type);
3956
3957 return trace_handle_return(s);
3958}
3959
3960static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
3961{
3962 struct trace_array *tr = iter->tr;
3963 struct trace_seq *s = &iter->seq;
3964 struct trace_entry *entry;
3965 struct trace_event *event;
3966
3967 entry = iter->ent;
3968
3969 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
3970 trace_seq_printf(s, "%d %d %llu ",
3971 entry->pid, iter->cpu, iter->ts);
3972
3973 if (trace_seq_has_overflowed(s))
3974 return TRACE_TYPE_PARTIAL_LINE;
3975
3976 event = ftrace_find_event(entry->type);
3977 if (event)
3978 return event->funcs->raw(iter, 0, event);
3979
3980 trace_seq_printf(s, "%d ?\n", entry->type);
3981
3982 return trace_handle_return(s);
3983}
3984
3985static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
3986{
3987 struct trace_array *tr = iter->tr;
3988 struct trace_seq *s = &iter->seq;
3989 unsigned char newline = '\n';
3990 struct trace_entry *entry;
3991 struct trace_event *event;
3992
3993 entry = iter->ent;
3994
3995 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3996 SEQ_PUT_HEX_FIELD(s, entry->pid);
3997 SEQ_PUT_HEX_FIELD(s, iter->cpu);
3998 SEQ_PUT_HEX_FIELD(s, iter->ts);
3999 if (trace_seq_has_overflowed(s))
4000 return TRACE_TYPE_PARTIAL_LINE;
4001 }
4002
4003 event = ftrace_find_event(entry->type);
4004 if (event) {
4005 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4006 if (ret != TRACE_TYPE_HANDLED)
4007 return ret;
4008 }
4009
4010 SEQ_PUT_FIELD(s, newline);
4011
4012 return trace_handle_return(s);
4013}
4014
4015static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4016{
4017 struct trace_array *tr = iter->tr;
4018 struct trace_seq *s = &iter->seq;
4019 struct trace_entry *entry;
4020 struct trace_event *event;
4021
4022 entry = iter->ent;
4023
4024 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4025 SEQ_PUT_FIELD(s, entry->pid);
4026 SEQ_PUT_FIELD(s, iter->cpu);
4027 SEQ_PUT_FIELD(s, iter->ts);
4028 if (trace_seq_has_overflowed(s))
4029 return TRACE_TYPE_PARTIAL_LINE;
4030 }
4031
4032 event = ftrace_find_event(entry->type);
4033 return event ? event->funcs->binary(iter, 0, event) :
4034 TRACE_TYPE_HANDLED;
4035}
4036
4037int trace_empty(struct trace_iterator *iter)
4038{
4039 struct ring_buffer_iter *buf_iter;
4040 int cpu;
4041
4042 /* If we are looking at one CPU buffer, only check that one */
4043 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4044 cpu = iter->cpu_file;
4045 buf_iter = trace_buffer_iter(iter, cpu);
4046 if (buf_iter) {
4047 if (!ring_buffer_iter_empty(buf_iter))
4048 return 0;
4049 } else {
4050 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4051 return 0;
4052 }
4053 return 1;
4054 }
4055
4056 for_each_tracing_cpu(cpu) {
4057 buf_iter = trace_buffer_iter(iter, cpu);
4058 if (buf_iter) {
4059 if (!ring_buffer_iter_empty(buf_iter))
4060 return 0;
4061 } else {
4062 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4063 return 0;
4064 }
4065 }
4066
4067 return 1;
4068}
4069
4070/* Called with trace_event_read_lock() held. */
4071enum print_line_t print_trace_line(struct trace_iterator *iter)
4072{
4073 struct trace_array *tr = iter->tr;
4074 unsigned long trace_flags = tr->trace_flags;
4075 enum print_line_t ret;
4076
4077 if (iter->lost_events) {
4078 if (iter->lost_events == (unsigned long)-1)
4079 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4080 iter->cpu);
4081 else
4082 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4083 iter->cpu, iter->lost_events);
4084 if (trace_seq_has_overflowed(&iter->seq))
4085 return TRACE_TYPE_PARTIAL_LINE;
4086 }
4087
4088 if (iter->trace && iter->trace->print_line) {
4089 ret = iter->trace->print_line(iter);
4090 if (ret != TRACE_TYPE_UNHANDLED)
4091 return ret;
4092 }
4093
4094 if (iter->ent->type == TRACE_BPUTS &&
4095 trace_flags & TRACE_ITER_PRINTK &&
4096 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4097 return trace_print_bputs_msg_only(iter);
4098
4099 if (iter->ent->type == TRACE_BPRINT &&
4100 trace_flags & TRACE_ITER_PRINTK &&
4101 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4102 return trace_print_bprintk_msg_only(iter);
4103
4104 if (iter->ent->type == TRACE_PRINT &&
4105 trace_flags & TRACE_ITER_PRINTK &&
4106 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4107 return trace_print_printk_msg_only(iter);
4108
4109 if (trace_flags & TRACE_ITER_BIN)
4110 return print_bin_fmt(iter);
4111
4112 if (trace_flags & TRACE_ITER_HEX)
4113 return print_hex_fmt(iter);
4114
4115 if (trace_flags & TRACE_ITER_RAW)
4116 return print_raw_fmt(iter);
4117
4118 return print_trace_fmt(iter);
4119}
4120
4121void trace_latency_header(struct seq_file *m)
4122{
4123 struct trace_iterator *iter = m->private;
4124 struct trace_array *tr = iter->tr;
4125
4126 /* print nothing if the buffers are empty */
4127 if (trace_empty(iter))
4128 return;
4129
4130 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4131 print_trace_header(m, iter);
4132
4133 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4134 print_lat_help_header(m);
4135}
4136
4137void trace_default_header(struct seq_file *m)
4138{
4139 struct trace_iterator *iter = m->private;
4140 struct trace_array *tr = iter->tr;
4141 unsigned long trace_flags = tr->trace_flags;
4142
4143 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4144 return;
4145
4146 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4147 /* print nothing if the buffers are empty */
4148 if (trace_empty(iter))
4149 return;
4150 print_trace_header(m, iter);
4151 if (!(trace_flags & TRACE_ITER_VERBOSE))
4152 print_lat_help_header(m);
4153 } else {
4154 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4155 if (trace_flags & TRACE_ITER_IRQ_INFO)
4156 print_func_help_header_irq(iter->array_buffer,
4157 m, trace_flags);
4158 else
4159 print_func_help_header(iter->array_buffer, m,
4160 trace_flags);
4161 }
4162 }
4163}
4164
4165static void test_ftrace_alive(struct seq_file *m)
4166{
4167 if (!ftrace_is_dead())
4168 return;
4169 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4170 "# MAY BE MISSING FUNCTION EVENTS\n");
4171}
4172
4173#ifdef CONFIG_TRACER_MAX_TRACE
4174static void show_snapshot_main_help(struct seq_file *m)
4175{
4176 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4177 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4178 "# Takes a snapshot of the main buffer.\n"
4179 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4180 "# (Doesn't have to be '2' works with any number that\n"
4181 "# is not a '0' or '1')\n");
4182}
4183
4184static void show_snapshot_percpu_help(struct seq_file *m)
4185{
4186 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4187#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4188 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4189 "# Takes a snapshot of the main buffer for this cpu.\n");
4190#else
4191 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4192 "# Must use main snapshot file to allocate.\n");
4193#endif
4194 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4195 "# (Doesn't have to be '2' works with any number that\n"
4196 "# is not a '0' or '1')\n");
4197}
4198
4199static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4200{
4201 if (iter->tr->allocated_snapshot)
4202 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4203 else
4204 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4205
4206 seq_puts(m, "# Snapshot commands:\n");
4207 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4208 show_snapshot_main_help(m);
4209 else
4210 show_snapshot_percpu_help(m);
4211}
4212#else
4213/* Should never be called */
4214static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4215#endif
4216
4217static int s_show(struct seq_file *m, void *v)
4218{
4219 struct trace_iterator *iter = v;
4220 int ret;
4221
4222 if (iter->ent == NULL) {
4223 if (iter->tr) {
4224 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4225 seq_puts(m, "#\n");
4226 test_ftrace_alive(m);
4227 }
4228 if (iter->snapshot && trace_empty(iter))
4229 print_snapshot_help(m, iter);
4230 else if (iter->trace && iter->trace->print_header)
4231 iter->trace->print_header(m);
4232 else
4233 trace_default_header(m);
4234
4235 } else if (iter->leftover) {
4236 /*
4237 * If we filled the seq_file buffer earlier, we
4238 * want to just show it now.
4239 */
4240 ret = trace_print_seq(m, &iter->seq);
4241
4242 /* ret should this time be zero, but you never know */
4243 iter->leftover = ret;
4244
4245 } else {
4246 print_trace_line(iter);
4247 ret = trace_print_seq(m, &iter->seq);
4248 /*
4249 * If we overflow the seq_file buffer, then it will
4250 * ask us for this data again at start up.
4251 * Use that instead.
4252 * ret is 0 if seq_file write succeeded.
4253 * -1 otherwise.
4254 */
4255 iter->leftover = ret;
4256 }
4257
4258 return 0;
4259}
4260
4261/*
4262 * Should be used after trace_array_get(), trace_types_lock
4263 * ensures that i_cdev was already initialized.
4264 */
4265static inline int tracing_get_cpu(struct inode *inode)
4266{
4267 if (inode->i_cdev) /* See trace_create_cpu_file() */
4268 return (long)inode->i_cdev - 1;
4269 return RING_BUFFER_ALL_CPUS;
4270}
4271
4272static const struct seq_operations tracer_seq_ops = {
4273 .start = s_start,
4274 .next = s_next,
4275 .stop = s_stop,
4276 .show = s_show,
4277};
4278
4279static struct trace_iterator *
4280__tracing_open(struct inode *inode, struct file *file, bool snapshot)
4281{
4282 struct trace_array *tr = inode->i_private;
4283 struct trace_iterator *iter;
4284 int cpu;
4285
4286 if (tracing_disabled)
4287 return ERR_PTR(-ENODEV);
4288
4289 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4290 if (!iter)
4291 return ERR_PTR(-ENOMEM);
4292
4293 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4294 GFP_KERNEL);
4295 if (!iter->buffer_iter)
4296 goto release;
4297
4298 /*
4299 * trace_find_next_entry() may need to save off iter->ent.
4300 * It will place it into the iter->temp buffer. As most
4301 * events are less than 128, allocate a buffer of that size.
4302 * If one is greater, then trace_find_next_entry() will
4303 * allocate a new buffer to adjust for the bigger iter->ent.
4304 * It's not critical if it fails to get allocated here.
4305 */
4306 iter->temp = kmalloc(128, GFP_KERNEL);
4307 if (iter->temp)
4308 iter->temp_size = 128;
4309
4310 /*
4311 * We make a copy of the current tracer to avoid concurrent
4312 * changes on it while we are reading.
4313 */
4314 mutex_lock(&trace_types_lock);
4315 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4316 if (!iter->trace)
4317 goto fail;
4318
4319 *iter->trace = *tr->current_trace;
4320
4321 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4322 goto fail;
4323
4324 iter->tr = tr;
4325
4326#ifdef CONFIG_TRACER_MAX_TRACE
4327 /* Currently only the top directory has a snapshot */
4328 if (tr->current_trace->print_max || snapshot)
4329 iter->array_buffer = &tr->max_buffer;
4330 else
4331#endif
4332 iter->array_buffer = &tr->array_buffer;
4333 iter->snapshot = snapshot;
4334 iter->pos = -1;
4335 iter->cpu_file = tracing_get_cpu(inode);
4336 mutex_init(&iter->mutex);
4337
4338 /* Notify the tracer early; before we stop tracing. */
4339 if (iter->trace->open)
4340 iter->trace->open(iter);
4341
4342 /* Annotate start of buffers if we had overruns */
4343 if (ring_buffer_overruns(iter->array_buffer->buffer))
4344 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4345
4346 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4347 if (trace_clocks[tr->clock_id].in_ns)
4348 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4349
4350 /*
4351 * If pause-on-trace is enabled, then stop the trace while
4352 * dumping, unless this is the "snapshot" file
4353 */
4354 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4355 tracing_stop_tr(tr);
4356
4357 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4358 for_each_tracing_cpu(cpu) {
4359 iter->buffer_iter[cpu] =
4360 ring_buffer_read_prepare(iter->array_buffer->buffer,
4361 cpu, GFP_KERNEL);
4362 }
4363 ring_buffer_read_prepare_sync();
4364 for_each_tracing_cpu(cpu) {
4365 ring_buffer_read_start(iter->buffer_iter[cpu]);
4366 tracing_iter_reset(iter, cpu);
4367 }
4368 } else {
4369 cpu = iter->cpu_file;
4370 iter->buffer_iter[cpu] =
4371 ring_buffer_read_prepare(iter->array_buffer->buffer,
4372 cpu, GFP_KERNEL);
4373 ring_buffer_read_prepare_sync();
4374 ring_buffer_read_start(iter->buffer_iter[cpu]);
4375 tracing_iter_reset(iter, cpu);
4376 }
4377
4378 mutex_unlock(&trace_types_lock);
4379
4380 return iter;
4381
4382 fail:
4383 mutex_unlock(&trace_types_lock);
4384 kfree(iter->trace);
4385 kfree(iter->temp);
4386 kfree(iter->buffer_iter);
4387release:
4388 seq_release_private(inode, file);
4389 return ERR_PTR(-ENOMEM);
4390}
4391
4392int tracing_open_generic(struct inode *inode, struct file *filp)
4393{
4394 int ret;
4395
4396 ret = tracing_check_open_get_tr(NULL);
4397 if (ret)
4398 return ret;
4399
4400 filp->private_data = inode->i_private;
4401 return 0;
4402}
4403
4404bool tracing_is_disabled(void)
4405{
4406 return (tracing_disabled) ? true: false;
4407}
4408
4409/*
4410 * Open and update trace_array ref count.
4411 * Must have the current trace_array passed to it.
4412 */
4413int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4414{
4415 struct trace_array *tr = inode->i_private;
4416 int ret;
4417
4418 ret = tracing_check_open_get_tr(tr);
4419 if (ret)
4420 return ret;
4421
4422 filp->private_data = inode->i_private;
4423
4424 return 0;
4425}
4426
4427static int tracing_release(struct inode *inode, struct file *file)
4428{
4429 struct trace_array *tr = inode->i_private;
4430 struct seq_file *m = file->private_data;
4431 struct trace_iterator *iter;
4432 int cpu;
4433
4434 if (!(file->f_mode & FMODE_READ)) {
4435 trace_array_put(tr);
4436 return 0;
4437 }
4438
4439 /* Writes do not use seq_file */
4440 iter = m->private;
4441 mutex_lock(&trace_types_lock);
4442
4443 for_each_tracing_cpu(cpu) {
4444 if (iter->buffer_iter[cpu])
4445 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4446 }
4447
4448 if (iter->trace && iter->trace->close)
4449 iter->trace->close(iter);
4450
4451 if (!iter->snapshot && tr->stop_count)
4452 /* reenable tracing if it was previously enabled */
4453 tracing_start_tr(tr);
4454
4455 __trace_array_put(tr);
4456
4457 mutex_unlock(&trace_types_lock);
4458
4459 mutex_destroy(&iter->mutex);
4460 free_cpumask_var(iter->started);
4461 kfree(iter->temp);
4462 kfree(iter->trace);
4463 kfree(iter->buffer_iter);
4464 seq_release_private(inode, file);
4465
4466 return 0;
4467}
4468
4469static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4470{
4471 struct trace_array *tr = inode->i_private;
4472
4473 trace_array_put(tr);
4474 return 0;
4475}
4476
4477static int tracing_single_release_tr(struct inode *inode, struct file *file)
4478{
4479 struct trace_array *tr = inode->i_private;
4480
4481 trace_array_put(tr);
4482
4483 return single_release(inode, file);
4484}
4485
4486static int tracing_open(struct inode *inode, struct file *file)
4487{
4488 struct trace_array *tr = inode->i_private;
4489 struct trace_iterator *iter;
4490 int ret;
4491
4492 ret = tracing_check_open_get_tr(tr);
4493 if (ret)
4494 return ret;
4495
4496 /* If this file was open for write, then erase contents */
4497 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4498 int cpu = tracing_get_cpu(inode);
4499 struct array_buffer *trace_buf = &tr->array_buffer;
4500
4501#ifdef CONFIG_TRACER_MAX_TRACE
4502 if (tr->current_trace->print_max)
4503 trace_buf = &tr->max_buffer;
4504#endif
4505
4506 if (cpu == RING_BUFFER_ALL_CPUS)
4507 tracing_reset_online_cpus(trace_buf);
4508 else
4509 tracing_reset_cpu(trace_buf, cpu);
4510 }
4511
4512 if (file->f_mode & FMODE_READ) {
4513 iter = __tracing_open(inode, file, false);
4514 if (IS_ERR(iter))
4515 ret = PTR_ERR(iter);
4516 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4517 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4518 }
4519
4520 if (ret < 0)
4521 trace_array_put(tr);
4522
4523 return ret;
4524}
4525
4526/*
4527 * Some tracers are not suitable for instance buffers.
4528 * A tracer is always available for the global array (toplevel)
4529 * or if it explicitly states that it is.
4530 */
4531static bool
4532trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4533{
4534 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4535}
4536
4537/* Find the next tracer that this trace array may use */
4538static struct tracer *
4539get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4540{
4541 while (t && !trace_ok_for_array(t, tr))
4542 t = t->next;
4543
4544 return t;
4545}
4546
4547static void *
4548t_next(struct seq_file *m, void *v, loff_t *pos)
4549{
4550 struct trace_array *tr = m->private;
4551 struct tracer *t = v;
4552
4553 (*pos)++;
4554
4555 if (t)
4556 t = get_tracer_for_array(tr, t->next);
4557
4558 return t;
4559}
4560
4561static void *t_start(struct seq_file *m, loff_t *pos)
4562{
4563 struct trace_array *tr = m->private;
4564 struct tracer *t;
4565 loff_t l = 0;
4566
4567 mutex_lock(&trace_types_lock);
4568
4569 t = get_tracer_for_array(tr, trace_types);
4570 for (; t && l < *pos; t = t_next(m, t, &l))
4571 ;
4572
4573 return t;
4574}
4575
4576static void t_stop(struct seq_file *m, void *p)
4577{
4578 mutex_unlock(&trace_types_lock);
4579}
4580
4581static int t_show(struct seq_file *m, void *v)
4582{
4583 struct tracer *t = v;
4584
4585 if (!t)
4586 return 0;
4587
4588 seq_puts(m, t->name);
4589 if (t->next)
4590 seq_putc(m, ' ');
4591 else
4592 seq_putc(m, '\n');
4593
4594 return 0;
4595}
4596
4597static const struct seq_operations show_traces_seq_ops = {
4598 .start = t_start,
4599 .next = t_next,
4600 .stop = t_stop,
4601 .show = t_show,
4602};
4603
4604static int show_traces_open(struct inode *inode, struct file *file)
4605{
4606 struct trace_array *tr = inode->i_private;
4607 struct seq_file *m;
4608 int ret;
4609
4610 ret = tracing_check_open_get_tr(tr);
4611 if (ret)
4612 return ret;
4613
4614 ret = seq_open(file, &show_traces_seq_ops);
4615 if (ret) {
4616 trace_array_put(tr);
4617 return ret;
4618 }
4619
4620 m = file->private_data;
4621 m->private = tr;
4622
4623 return 0;
4624}
4625
4626static int show_traces_release(struct inode *inode, struct file *file)
4627{
4628 struct trace_array *tr = inode->i_private;
4629
4630 trace_array_put(tr);
4631 return seq_release(inode, file);
4632}
4633
4634static ssize_t
4635tracing_write_stub(struct file *filp, const char __user *ubuf,
4636 size_t count, loff_t *ppos)
4637{
4638 return count;
4639}
4640
4641loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
4642{
4643 int ret;
4644
4645 if (file->f_mode & FMODE_READ)
4646 ret = seq_lseek(file, offset, whence);
4647 else
4648 file->f_pos = ret = 0;
4649
4650 return ret;
4651}
4652
4653static const struct file_operations tracing_fops = {
4654 .open = tracing_open,
4655 .read = seq_read,
4656 .write = tracing_write_stub,
4657 .llseek = tracing_lseek,
4658 .release = tracing_release,
4659};
4660
4661static const struct file_operations show_traces_fops = {
4662 .open = show_traces_open,
4663 .read = seq_read,
4664 .llseek = seq_lseek,
4665 .release = show_traces_release,
4666};
4667
4668static ssize_t
4669tracing_cpumask_read(struct file *filp, char __user *ubuf,
4670 size_t count, loff_t *ppos)
4671{
4672 struct trace_array *tr = file_inode(filp)->i_private;
4673 char *mask_str;
4674 int len;
4675
4676 len = snprintf(NULL, 0, "%*pb\n",
4677 cpumask_pr_args(tr->tracing_cpumask)) + 1;
4678 mask_str = kmalloc(len, GFP_KERNEL);
4679 if (!mask_str)
4680 return -ENOMEM;
4681
4682 len = snprintf(mask_str, len, "%*pb\n",
4683 cpumask_pr_args(tr->tracing_cpumask));
4684 if (len >= count) {
4685 count = -EINVAL;
4686 goto out_err;
4687 }
4688 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
4689
4690out_err:
4691 kfree(mask_str);
4692
4693 return count;
4694}
4695
4696int tracing_set_cpumask(struct trace_array *tr,
4697 cpumask_var_t tracing_cpumask_new)
4698{
4699 int cpu;
4700
4701 if (!tr)
4702 return -EINVAL;
4703
4704 local_irq_disable();
4705 arch_spin_lock(&tr->max_lock);
4706 for_each_tracing_cpu(cpu) {
4707 /*
4708 * Increase/decrease the disabled counter if we are
4709 * about to flip a bit in the cpumask:
4710 */
4711 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4712 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4713 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
4714 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
4715 }
4716 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4717 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4718 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
4719 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
4720 }
4721 }
4722 arch_spin_unlock(&tr->max_lock);
4723 local_irq_enable();
4724
4725 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
4726
4727 return 0;
4728}
4729
4730static ssize_t
4731tracing_cpumask_write(struct file *filp, const char __user *ubuf,
4732 size_t count, loff_t *ppos)
4733{
4734 struct trace_array *tr = file_inode(filp)->i_private;
4735 cpumask_var_t tracing_cpumask_new;
4736 int err;
4737
4738 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
4739 return -ENOMEM;
4740
4741 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
4742 if (err)
4743 goto err_free;
4744
4745 err = tracing_set_cpumask(tr, tracing_cpumask_new);
4746 if (err)
4747 goto err_free;
4748
4749 free_cpumask_var(tracing_cpumask_new);
4750
4751 return count;
4752
4753err_free:
4754 free_cpumask_var(tracing_cpumask_new);
4755
4756 return err;
4757}
4758
4759static const struct file_operations tracing_cpumask_fops = {
4760 .open = tracing_open_generic_tr,
4761 .read = tracing_cpumask_read,
4762 .write = tracing_cpumask_write,
4763 .release = tracing_release_generic_tr,
4764 .llseek = generic_file_llseek,
4765};
4766
4767static int tracing_trace_options_show(struct seq_file *m, void *v)
4768{
4769 struct tracer_opt *trace_opts;
4770 struct trace_array *tr = m->private;
4771 u32 tracer_flags;
4772 int i;
4773
4774 mutex_lock(&trace_types_lock);
4775 tracer_flags = tr->current_trace->flags->val;
4776 trace_opts = tr->current_trace->flags->opts;
4777
4778 for (i = 0; trace_options[i]; i++) {
4779 if (tr->trace_flags & (1 << i))
4780 seq_printf(m, "%s\n", trace_options[i]);
4781 else
4782 seq_printf(m, "no%s\n", trace_options[i]);
4783 }
4784
4785 for (i = 0; trace_opts[i].name; i++) {
4786 if (tracer_flags & trace_opts[i].bit)
4787 seq_printf(m, "%s\n", trace_opts[i].name);
4788 else
4789 seq_printf(m, "no%s\n", trace_opts[i].name);
4790 }
4791 mutex_unlock(&trace_types_lock);
4792
4793 return 0;
4794}
4795
4796static int __set_tracer_option(struct trace_array *tr,
4797 struct tracer_flags *tracer_flags,
4798 struct tracer_opt *opts, int neg)
4799{
4800 struct tracer *trace = tracer_flags->trace;
4801 int ret;
4802
4803 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
4804 if (ret)
4805 return ret;
4806
4807 if (neg)
4808 tracer_flags->val &= ~opts->bit;
4809 else
4810 tracer_flags->val |= opts->bit;
4811 return 0;
4812}
4813
4814/* Try to assign a tracer specific option */
4815static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
4816{
4817 struct tracer *trace = tr->current_trace;
4818 struct tracer_flags *tracer_flags = trace->flags;
4819 struct tracer_opt *opts = NULL;
4820 int i;
4821
4822 for (i = 0; tracer_flags->opts[i].name; i++) {
4823 opts = &tracer_flags->opts[i];
4824
4825 if (strcmp(cmp, opts->name) == 0)
4826 return __set_tracer_option(tr, trace->flags, opts, neg);
4827 }
4828
4829 return -EINVAL;
4830}
4831
4832/* Some tracers require overwrite to stay enabled */
4833int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
4834{
4835 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
4836 return -1;
4837
4838 return 0;
4839}
4840
4841int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
4842{
4843 if ((mask == TRACE_ITER_RECORD_TGID) ||
4844 (mask == TRACE_ITER_RECORD_CMD))
4845 lockdep_assert_held(&event_mutex);
4846
4847 /* do nothing if flag is already set */
4848 if (!!(tr->trace_flags & mask) == !!enabled)
4849 return 0;
4850
4851 /* Give the tracer a chance to approve the change */
4852 if (tr->current_trace->flag_changed)
4853 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
4854 return -EINVAL;
4855
4856 if (enabled)
4857 tr->trace_flags |= mask;
4858 else
4859 tr->trace_flags &= ~mask;
4860
4861 if (mask == TRACE_ITER_RECORD_CMD)
4862 trace_event_enable_cmd_record(enabled);
4863
4864 if (mask == TRACE_ITER_RECORD_TGID) {
4865 if (!tgid_map)
4866 tgid_map = kvcalloc(PID_MAX_DEFAULT + 1,
4867 sizeof(*tgid_map),
4868 GFP_KERNEL);
4869 if (!tgid_map) {
4870 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
4871 return -ENOMEM;
4872 }
4873
4874 trace_event_enable_tgid_record(enabled);
4875 }
4876
4877 if (mask == TRACE_ITER_EVENT_FORK)
4878 trace_event_follow_fork(tr, enabled);
4879
4880 if (mask == TRACE_ITER_FUNC_FORK)
4881 ftrace_pid_follow_fork(tr, enabled);
4882
4883 if (mask == TRACE_ITER_OVERWRITE) {
4884 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
4885#ifdef CONFIG_TRACER_MAX_TRACE
4886 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
4887#endif
4888 }
4889
4890 if (mask == TRACE_ITER_PRINTK) {
4891 trace_printk_start_stop_comm(enabled);
4892 trace_printk_control(enabled);
4893 }
4894
4895 return 0;
4896}
4897
4898int trace_set_options(struct trace_array *tr, char *option)
4899{
4900 char *cmp;
4901 int neg = 0;
4902 int ret;
4903 size_t orig_len = strlen(option);
4904 int len;
4905
4906 cmp = strstrip(option);
4907
4908 len = str_has_prefix(cmp, "no");
4909 if (len)
4910 neg = 1;
4911
4912 cmp += len;
4913
4914 mutex_lock(&event_mutex);
4915 mutex_lock(&trace_types_lock);
4916
4917 ret = match_string(trace_options, -1, cmp);
4918 /* If no option could be set, test the specific tracer options */
4919 if (ret < 0)
4920 ret = set_tracer_option(tr, cmp, neg);
4921 else
4922 ret = set_tracer_flag(tr, 1 << ret, !neg);
4923
4924 mutex_unlock(&trace_types_lock);
4925 mutex_unlock(&event_mutex);
4926
4927 /*
4928 * If the first trailing whitespace is replaced with '\0' by strstrip,
4929 * turn it back into a space.
4930 */
4931 if (orig_len > strlen(option))
4932 option[strlen(option)] = ' ';
4933
4934 return ret;
4935}
4936
4937static void __init apply_trace_boot_options(void)
4938{
4939 char *buf = trace_boot_options_buf;
4940 char *option;
4941
4942 while (true) {
4943 option = strsep(&buf, ",");
4944
4945 if (!option)
4946 break;
4947
4948 if (*option)
4949 trace_set_options(&global_trace, option);
4950
4951 /* Put back the comma to allow this to be called again */
4952 if (buf)
4953 *(buf - 1) = ',';
4954 }
4955}
4956
4957static ssize_t
4958tracing_trace_options_write(struct file *filp, const char __user *ubuf,
4959 size_t cnt, loff_t *ppos)
4960{
4961 struct seq_file *m = filp->private_data;
4962 struct trace_array *tr = m->private;
4963 char buf[64];
4964 int ret;
4965
4966 if (cnt >= sizeof(buf))
4967 return -EINVAL;
4968
4969 if (copy_from_user(buf, ubuf, cnt))
4970 return -EFAULT;
4971
4972 buf[cnt] = 0;
4973
4974 ret = trace_set_options(tr, buf);
4975 if (ret < 0)
4976 return ret;
4977
4978 *ppos += cnt;
4979
4980 return cnt;
4981}
4982
4983static int tracing_trace_options_open(struct inode *inode, struct file *file)
4984{
4985 struct trace_array *tr = inode->i_private;
4986 int ret;
4987
4988 ret = tracing_check_open_get_tr(tr);
4989 if (ret)
4990 return ret;
4991
4992 ret = single_open(file, tracing_trace_options_show, inode->i_private);
4993 if (ret < 0)
4994 trace_array_put(tr);
4995
4996 return ret;
4997}
4998
4999static const struct file_operations tracing_iter_fops = {
5000 .open = tracing_trace_options_open,
5001 .read = seq_read,
5002 .llseek = seq_lseek,
5003 .release = tracing_single_release_tr,
5004 .write = tracing_trace_options_write,
5005};
5006
5007static const char readme_msg[] =
5008 "tracing mini-HOWTO:\n\n"
5009 "# echo 0 > tracing_on : quick way to disable tracing\n"
5010 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5011 " Important files:\n"
5012 " trace\t\t\t- The static contents of the buffer\n"
5013 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5014 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5015 " current_tracer\t- function and latency tracers\n"
5016 " available_tracers\t- list of configured tracers for current_tracer\n"
5017 " error_log\t- error log for failed commands (that support it)\n"
5018 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5019 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5020 " trace_clock\t\t-change the clock used to order events\n"
5021 " local: Per cpu clock but may not be synced across CPUs\n"
5022 " global: Synced across CPUs but slows tracing down.\n"
5023 " counter: Not a clock, but just an increment\n"
5024 " uptime: Jiffy counter from time of boot\n"
5025 " perf: Same clock that perf events use\n"
5026#ifdef CONFIG_X86_64
5027 " x86-tsc: TSC cycle counter\n"
5028#endif
5029 "\n timestamp_mode\t-view the mode used to timestamp events\n"
5030 " delta: Delta difference against a buffer-wide timestamp\n"
5031 " absolute: Absolute (standalone) timestamp\n"
5032 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5033 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5034 " tracing_cpumask\t- Limit which CPUs to trace\n"
5035 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5036 "\t\t\t Remove sub-buffer with rmdir\n"
5037 " trace_options\t\t- Set format or modify how tracing happens\n"
5038 "\t\t\t Disable an option by prefixing 'no' to the\n"
5039 "\t\t\t option name\n"
5040 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5041#ifdef CONFIG_DYNAMIC_FTRACE
5042 "\n available_filter_functions - list of functions that can be filtered on\n"
5043 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5044 "\t\t\t functions\n"
5045 "\t accepts: func_full_name or glob-matching-pattern\n"
5046 "\t modules: Can select a group via module\n"
5047 "\t Format: :mod:<module-name>\n"
5048 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5049 "\t triggers: a command to perform when function is hit\n"
5050 "\t Format: <function>:<trigger>[:count]\n"
5051 "\t trigger: traceon, traceoff\n"
5052 "\t\t enable_event:<system>:<event>\n"
5053 "\t\t disable_event:<system>:<event>\n"
5054#ifdef CONFIG_STACKTRACE
5055 "\t\t stacktrace\n"
5056#endif
5057#ifdef CONFIG_TRACER_SNAPSHOT
5058 "\t\t snapshot\n"
5059#endif
5060 "\t\t dump\n"
5061 "\t\t cpudump\n"
5062 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5063 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5064 "\t The first one will disable tracing every time do_fault is hit\n"
5065 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5066 "\t The first time do trap is hit and it disables tracing, the\n"
5067 "\t counter will decrement to 2. If tracing is already disabled,\n"
5068 "\t the counter will not decrement. It only decrements when the\n"
5069 "\t trigger did work\n"
5070 "\t To remove trigger without count:\n"
5071 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5072 "\t To remove trigger with a count:\n"
5073 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5074 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5075 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5076 "\t modules: Can select a group via module command :mod:\n"
5077 "\t Does not accept triggers\n"
5078#endif /* CONFIG_DYNAMIC_FTRACE */
5079#ifdef CONFIG_FUNCTION_TRACER
5080 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5081 "\t\t (function)\n"
5082 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5083 "\t\t (function)\n"
5084#endif
5085#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5086 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5087 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5088 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5089#endif
5090#ifdef CONFIG_TRACER_SNAPSHOT
5091 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5092 "\t\t\t snapshot buffer. Read the contents for more\n"
5093 "\t\t\t information\n"
5094#endif
5095#ifdef CONFIG_STACK_TRACER
5096 " stack_trace\t\t- Shows the max stack trace when active\n"
5097 " stack_max_size\t- Shows current max stack size that was traced\n"
5098 "\t\t\t Write into this file to reset the max size (trigger a\n"
5099 "\t\t\t new trace)\n"
5100#ifdef CONFIG_DYNAMIC_FTRACE
5101 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5102 "\t\t\t traces\n"
5103#endif
5104#endif /* CONFIG_STACK_TRACER */
5105#ifdef CONFIG_DYNAMIC_EVENTS
5106 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5107 "\t\t\t Write into this file to define/undefine new trace events.\n"
5108#endif
5109#ifdef CONFIG_KPROBE_EVENTS
5110 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5111 "\t\t\t Write into this file to define/undefine new trace events.\n"
5112#endif
5113#ifdef CONFIG_UPROBE_EVENTS
5114 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5115 "\t\t\t Write into this file to define/undefine new trace events.\n"
5116#endif
5117#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5118 "\t accepts: event-definitions (one definition per line)\n"
5119 "\t Format: p[:[<group>/]<event>] <place> [<args>]\n"
5120 "\t r[maxactive][:[<group>/]<event>] <place> [<args>]\n"
5121#ifdef CONFIG_HIST_TRIGGERS
5122 "\t s:[synthetic/]<event> <field> [<field>]\n"
5123#endif
5124 "\t -:[<group>/]<event>\n"
5125#ifdef CONFIG_KPROBE_EVENTS
5126 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5127 "place (kretprobe): [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5128#endif
5129#ifdef CONFIG_UPROBE_EVENTS
5130 " place (uprobe): <path>:<offset>[(ref_ctr_offset)]\n"
5131#endif
5132 "\t args: <name>=fetcharg[:type]\n"
5133 "\t fetcharg: %<register>, @<address>, @<symbol>[+|-<offset>],\n"
5134#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5135 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5136#else
5137 "\t $stack<index>, $stack, $retval, $comm,\n"
5138#endif
5139 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5140 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string, symbol,\n"
5141 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5142 "\t <type>\\[<array-size>\\]\n"
5143#ifdef CONFIG_HIST_TRIGGERS
5144 "\t field: <stype> <name>;\n"
5145 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5146 "\t [unsigned] char/int/long\n"
5147#endif
5148#endif
5149 " events/\t\t- Directory containing all trace event subsystems:\n"
5150 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5151 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5152 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5153 "\t\t\t events\n"
5154 " filter\t\t- If set, only events passing filter are traced\n"
5155 " events/<system>/<event>/\t- Directory containing control files for\n"
5156 "\t\t\t <event>:\n"
5157 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5158 " filter\t\t- If set, only events passing filter are traced\n"
5159 " trigger\t\t- If set, a command to perform when event is hit\n"
5160 "\t Format: <trigger>[:count][if <filter>]\n"
5161 "\t trigger: traceon, traceoff\n"
5162 "\t enable_event:<system>:<event>\n"
5163 "\t disable_event:<system>:<event>\n"
5164#ifdef CONFIG_HIST_TRIGGERS
5165 "\t enable_hist:<system>:<event>\n"
5166 "\t disable_hist:<system>:<event>\n"
5167#endif
5168#ifdef CONFIG_STACKTRACE
5169 "\t\t stacktrace\n"
5170#endif
5171#ifdef CONFIG_TRACER_SNAPSHOT
5172 "\t\t snapshot\n"
5173#endif
5174#ifdef CONFIG_HIST_TRIGGERS
5175 "\t\t hist (see below)\n"
5176#endif
5177 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5178 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5179 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5180 "\t events/block/block_unplug/trigger\n"
5181 "\t The first disables tracing every time block_unplug is hit.\n"
5182 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5183 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5184 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5185 "\t Like function triggers, the counter is only decremented if it\n"
5186 "\t enabled or disabled tracing.\n"
5187 "\t To remove a trigger without a count:\n"
5188 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5189 "\t To remove a trigger with a count:\n"
5190 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5191 "\t Filters can be ignored when removing a trigger.\n"
5192#ifdef CONFIG_HIST_TRIGGERS
5193 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5194 "\t Format: hist:keys=<field1[,field2,...]>\n"
5195 "\t [:values=<field1[,field2,...]>]\n"
5196 "\t [:sort=<field1[,field2,...]>]\n"
5197 "\t [:size=#entries]\n"
5198 "\t [:pause][:continue][:clear]\n"
5199 "\t [:name=histname1]\n"
5200 "\t [:<handler>.<action>]\n"
5201 "\t [if <filter>]\n\n"
5202 "\t When a matching event is hit, an entry is added to a hash\n"
5203 "\t table using the key(s) and value(s) named, and the value of a\n"
5204 "\t sum called 'hitcount' is incremented. Keys and values\n"
5205 "\t correspond to fields in the event's format description. Keys\n"
5206 "\t can be any field, or the special string 'stacktrace'.\n"
5207 "\t Compound keys consisting of up to two fields can be specified\n"
5208 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5209 "\t fields. Sort keys consisting of up to two fields can be\n"
5210 "\t specified using the 'sort' keyword. The sort direction can\n"
5211 "\t be modified by appending '.descending' or '.ascending' to a\n"
5212 "\t sort field. The 'size' parameter can be used to specify more\n"
5213 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5214 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5215 "\t its histogram data will be shared with other triggers of the\n"
5216 "\t same name, and trigger hits will update this common data.\n\n"
5217 "\t Reading the 'hist' file for the event will dump the hash\n"
5218 "\t table in its entirety to stdout. If there are multiple hist\n"
5219 "\t triggers attached to an event, there will be a table for each\n"
5220 "\t trigger in the output. The table displayed for a named\n"
5221 "\t trigger will be the same as any other instance having the\n"
5222 "\t same name. The default format used to display a given field\n"
5223 "\t can be modified by appending any of the following modifiers\n"
5224 "\t to the field name, as applicable:\n\n"
5225 "\t .hex display a number as a hex value\n"
5226 "\t .sym display an address as a symbol\n"
5227 "\t .sym-offset display an address as a symbol and offset\n"
5228 "\t .execname display a common_pid as a program name\n"
5229 "\t .syscall display a syscall id as a syscall name\n"
5230 "\t .log2 display log2 value rather than raw number\n"
5231 "\t .usecs display a common_timestamp in microseconds\n\n"
5232 "\t The 'pause' parameter can be used to pause an existing hist\n"
5233 "\t trigger or to start a hist trigger but not log any events\n"
5234 "\t until told to do so. 'continue' can be used to start or\n"
5235 "\t restart a paused hist trigger.\n\n"
5236 "\t The 'clear' parameter will clear the contents of a running\n"
5237 "\t hist trigger and leave its current paused/active state\n"
5238 "\t unchanged.\n\n"
5239 "\t The enable_hist and disable_hist triggers can be used to\n"
5240 "\t have one event conditionally start and stop another event's\n"
5241 "\t already-attached hist trigger. The syntax is analogous to\n"
5242 "\t the enable_event and disable_event triggers.\n\n"
5243 "\t Hist trigger handlers and actions are executed whenever a\n"
5244 "\t a histogram entry is added or updated. They take the form:\n\n"
5245 "\t <handler>.<action>\n\n"
5246 "\t The available handlers are:\n\n"
5247 "\t onmatch(matching.event) - invoke on addition or update\n"
5248 "\t onmax(var) - invoke if var exceeds current max\n"
5249 "\t onchange(var) - invoke action if var changes\n\n"
5250 "\t The available actions are:\n\n"
5251 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5252 "\t save(field,...) - save current event fields\n"
5253#ifdef CONFIG_TRACER_SNAPSHOT
5254 "\t snapshot() - snapshot the trace buffer\n"
5255#endif
5256#endif
5257;
5258
5259static ssize_t
5260tracing_readme_read(struct file *filp, char __user *ubuf,
5261 size_t cnt, loff_t *ppos)
5262{
5263 return simple_read_from_buffer(ubuf, cnt, ppos,
5264 readme_msg, strlen(readme_msg));
5265}
5266
5267static const struct file_operations tracing_readme_fops = {
5268 .open = tracing_open_generic,
5269 .read = tracing_readme_read,
5270 .llseek = generic_file_llseek,
5271};
5272
5273static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5274{
5275 int *ptr = v;
5276
5277 if (*pos || m->count)
5278 ptr++;
5279
5280 (*pos)++;
5281
5282 for (; ptr <= &tgid_map[PID_MAX_DEFAULT]; ptr++) {
5283 if (trace_find_tgid(*ptr))
5284 return ptr;
5285 }
5286
5287 return NULL;
5288}
5289
5290static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5291{
5292 void *v;
5293 loff_t l = 0;
5294
5295 if (!tgid_map)
5296 return NULL;
5297
5298 v = &tgid_map[0];
5299 while (l <= *pos) {
5300 v = saved_tgids_next(m, v, &l);
5301 if (!v)
5302 return NULL;
5303 }
5304
5305 return v;
5306}
5307
5308static void saved_tgids_stop(struct seq_file *m, void *v)
5309{
5310}
5311
5312static int saved_tgids_show(struct seq_file *m, void *v)
5313{
5314 int pid = (int *)v - tgid_map;
5315
5316 seq_printf(m, "%d %d\n", pid, trace_find_tgid(pid));
5317 return 0;
5318}
5319
5320static const struct seq_operations tracing_saved_tgids_seq_ops = {
5321 .start = saved_tgids_start,
5322 .stop = saved_tgids_stop,
5323 .next = saved_tgids_next,
5324 .show = saved_tgids_show,
5325};
5326
5327static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5328{
5329 int ret;
5330
5331 ret = tracing_check_open_get_tr(NULL);
5332 if (ret)
5333 return ret;
5334
5335 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5336}
5337
5338
5339static const struct file_operations tracing_saved_tgids_fops = {
5340 .open = tracing_saved_tgids_open,
5341 .read = seq_read,
5342 .llseek = seq_lseek,
5343 .release = seq_release,
5344};
5345
5346static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5347{
5348 unsigned int *ptr = v;
5349
5350 if (*pos || m->count)
5351 ptr++;
5352
5353 (*pos)++;
5354
5355 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5356 ptr++) {
5357 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5358 continue;
5359
5360 return ptr;
5361 }
5362
5363 return NULL;
5364}
5365
5366static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5367{
5368 void *v;
5369 loff_t l = 0;
5370
5371 preempt_disable();
5372 arch_spin_lock(&trace_cmdline_lock);
5373
5374 v = &savedcmd->map_cmdline_to_pid[0];
5375 while (l <= *pos) {
5376 v = saved_cmdlines_next(m, v, &l);
5377 if (!v)
5378 return NULL;
5379 }
5380
5381 return v;
5382}
5383
5384static void saved_cmdlines_stop(struct seq_file *m, void *v)
5385{
5386 arch_spin_unlock(&trace_cmdline_lock);
5387 preempt_enable();
5388}
5389
5390static int saved_cmdlines_show(struct seq_file *m, void *v)
5391{
5392 char buf[TASK_COMM_LEN];
5393 unsigned int *pid = v;
5394
5395 __trace_find_cmdline(*pid, buf);
5396 seq_printf(m, "%d %s\n", *pid, buf);
5397 return 0;
5398}
5399
5400static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5401 .start = saved_cmdlines_start,
5402 .next = saved_cmdlines_next,
5403 .stop = saved_cmdlines_stop,
5404 .show = saved_cmdlines_show,
5405};
5406
5407static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5408{
5409 int ret;
5410
5411 ret = tracing_check_open_get_tr(NULL);
5412 if (ret)
5413 return ret;
5414
5415 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5416}
5417
5418static const struct file_operations tracing_saved_cmdlines_fops = {
5419 .open = tracing_saved_cmdlines_open,
5420 .read = seq_read,
5421 .llseek = seq_lseek,
5422 .release = seq_release,
5423};
5424
5425static ssize_t
5426tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5427 size_t cnt, loff_t *ppos)
5428{
5429 char buf[64];
5430 int r;
5431
5432 arch_spin_lock(&trace_cmdline_lock);
5433 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5434 arch_spin_unlock(&trace_cmdline_lock);
5435
5436 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5437}
5438
5439static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5440{
5441 kfree(s->saved_cmdlines);
5442 kfree(s->map_cmdline_to_pid);
5443 kfree(s);
5444}
5445
5446static int tracing_resize_saved_cmdlines(unsigned int val)
5447{
5448 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5449
5450 s = kmalloc(sizeof(*s), GFP_KERNEL);
5451 if (!s)
5452 return -ENOMEM;
5453
5454 if (allocate_cmdlines_buffer(val, s) < 0) {
5455 kfree(s);
5456 return -ENOMEM;
5457 }
5458
5459 arch_spin_lock(&trace_cmdline_lock);
5460 savedcmd_temp = savedcmd;
5461 savedcmd = s;
5462 arch_spin_unlock(&trace_cmdline_lock);
5463 free_saved_cmdlines_buffer(savedcmd_temp);
5464
5465 return 0;
5466}
5467
5468static ssize_t
5469tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
5470 size_t cnt, loff_t *ppos)
5471{
5472 unsigned long val;
5473 int ret;
5474
5475 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5476 if (ret)
5477 return ret;
5478
5479 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
5480 if (!val || val > PID_MAX_DEFAULT)
5481 return -EINVAL;
5482
5483 ret = tracing_resize_saved_cmdlines((unsigned int)val);
5484 if (ret < 0)
5485 return ret;
5486
5487 *ppos += cnt;
5488
5489 return cnt;
5490}
5491
5492static const struct file_operations tracing_saved_cmdlines_size_fops = {
5493 .open = tracing_open_generic,
5494 .read = tracing_saved_cmdlines_size_read,
5495 .write = tracing_saved_cmdlines_size_write,
5496};
5497
5498#ifdef CONFIG_TRACE_EVAL_MAP_FILE
5499static union trace_eval_map_item *
5500update_eval_map(union trace_eval_map_item *ptr)
5501{
5502 if (!ptr->map.eval_string) {
5503 if (ptr->tail.next) {
5504 ptr = ptr->tail.next;
5505 /* Set ptr to the next real item (skip head) */
5506 ptr++;
5507 } else
5508 return NULL;
5509 }
5510 return ptr;
5511}
5512
5513static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5514{
5515 union trace_eval_map_item *ptr = v;
5516
5517 /*
5518 * Paranoid! If ptr points to end, we don't want to increment past it.
5519 * This really should never happen.
5520 */
5521 (*pos)++;
5522 ptr = update_eval_map(ptr);
5523 if (WARN_ON_ONCE(!ptr))
5524 return NULL;
5525
5526 ptr++;
5527 ptr = update_eval_map(ptr);
5528
5529 return ptr;
5530}
5531
5532static void *eval_map_start(struct seq_file *m, loff_t *pos)
5533{
5534 union trace_eval_map_item *v;
5535 loff_t l = 0;
5536
5537 mutex_lock(&trace_eval_mutex);
5538
5539 v = trace_eval_maps;
5540 if (v)
5541 v++;
5542
5543 while (v && l < *pos) {
5544 v = eval_map_next(m, v, &l);
5545 }
5546
5547 return v;
5548}
5549
5550static void eval_map_stop(struct seq_file *m, void *v)
5551{
5552 mutex_unlock(&trace_eval_mutex);
5553}
5554
5555static int eval_map_show(struct seq_file *m, void *v)
5556{
5557 union trace_eval_map_item *ptr = v;
5558
5559 seq_printf(m, "%s %ld (%s)\n",
5560 ptr->map.eval_string, ptr->map.eval_value,
5561 ptr->map.system);
5562
5563 return 0;
5564}
5565
5566static const struct seq_operations tracing_eval_map_seq_ops = {
5567 .start = eval_map_start,
5568 .next = eval_map_next,
5569 .stop = eval_map_stop,
5570 .show = eval_map_show,
5571};
5572
5573static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5574{
5575 int ret;
5576
5577 ret = tracing_check_open_get_tr(NULL);
5578 if (ret)
5579 return ret;
5580
5581 return seq_open(filp, &tracing_eval_map_seq_ops);
5582}
5583
5584static const struct file_operations tracing_eval_map_fops = {
5585 .open = tracing_eval_map_open,
5586 .read = seq_read,
5587 .llseek = seq_lseek,
5588 .release = seq_release,
5589};
5590
5591static inline union trace_eval_map_item *
5592trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
5593{
5594 /* Return tail of array given the head */
5595 return ptr + ptr->head.length + 1;
5596}
5597
5598static void
5599trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
5600 int len)
5601{
5602 struct trace_eval_map **stop;
5603 struct trace_eval_map **map;
5604 union trace_eval_map_item *map_array;
5605 union trace_eval_map_item *ptr;
5606
5607 stop = start + len;
5608
5609 /*
5610 * The trace_eval_maps contains the map plus a head and tail item,
5611 * where the head holds the module and length of array, and the
5612 * tail holds a pointer to the next list.
5613 */
5614 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
5615 if (!map_array) {
5616 pr_warn("Unable to allocate trace eval mapping\n");
5617 return;
5618 }
5619
5620 mutex_lock(&trace_eval_mutex);
5621
5622 if (!trace_eval_maps)
5623 trace_eval_maps = map_array;
5624 else {
5625 ptr = trace_eval_maps;
5626 for (;;) {
5627 ptr = trace_eval_jmp_to_tail(ptr);
5628 if (!ptr->tail.next)
5629 break;
5630 ptr = ptr->tail.next;
5631
5632 }
5633 ptr->tail.next = map_array;
5634 }
5635 map_array->head.mod = mod;
5636 map_array->head.length = len;
5637 map_array++;
5638
5639 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
5640 map_array->map = **map;
5641 map_array++;
5642 }
5643 memset(map_array, 0, sizeof(*map_array));
5644
5645 mutex_unlock(&trace_eval_mutex);
5646}
5647
5648static void trace_create_eval_file(struct dentry *d_tracer)
5649{
5650 trace_create_file("eval_map", 0444, d_tracer,
5651 NULL, &tracing_eval_map_fops);
5652}
5653
5654#else /* CONFIG_TRACE_EVAL_MAP_FILE */
5655static inline void trace_create_eval_file(struct dentry *d_tracer) { }
5656static inline void trace_insert_eval_map_file(struct module *mod,
5657 struct trace_eval_map **start, int len) { }
5658#endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
5659
5660static void trace_insert_eval_map(struct module *mod,
5661 struct trace_eval_map **start, int len)
5662{
5663 struct trace_eval_map **map;
5664
5665 if (len <= 0)
5666 return;
5667
5668 map = start;
5669
5670 trace_event_eval_update(map, len);
5671
5672 trace_insert_eval_map_file(mod, start, len);
5673}
5674
5675static ssize_t
5676tracing_set_trace_read(struct file *filp, char __user *ubuf,
5677 size_t cnt, loff_t *ppos)
5678{
5679 struct trace_array *tr = filp->private_data;
5680 char buf[MAX_TRACER_SIZE+2];
5681 int r;
5682
5683 mutex_lock(&trace_types_lock);
5684 r = sprintf(buf, "%s\n", tr->current_trace->name);
5685 mutex_unlock(&trace_types_lock);
5686
5687 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5688}
5689
5690int tracer_init(struct tracer *t, struct trace_array *tr)
5691{
5692 tracing_reset_online_cpus(&tr->array_buffer);
5693 return t->init(tr);
5694}
5695
5696static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
5697{
5698 int cpu;
5699
5700 for_each_tracing_cpu(cpu)
5701 per_cpu_ptr(buf->data, cpu)->entries = val;
5702}
5703
5704#ifdef CONFIG_TRACER_MAX_TRACE
5705/* resize @tr's buffer to the size of @size_tr's entries */
5706static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
5707 struct array_buffer *size_buf, int cpu_id)
5708{
5709 int cpu, ret = 0;
5710
5711 if (cpu_id == RING_BUFFER_ALL_CPUS) {
5712 for_each_tracing_cpu(cpu) {
5713 ret = ring_buffer_resize(trace_buf->buffer,
5714 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
5715 if (ret < 0)
5716 break;
5717 per_cpu_ptr(trace_buf->data, cpu)->entries =
5718 per_cpu_ptr(size_buf->data, cpu)->entries;
5719 }
5720 } else {
5721 ret = ring_buffer_resize(trace_buf->buffer,
5722 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
5723 if (ret == 0)
5724 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
5725 per_cpu_ptr(size_buf->data, cpu_id)->entries;
5726 }
5727
5728 return ret;
5729}
5730#endif /* CONFIG_TRACER_MAX_TRACE */
5731
5732static int __tracing_resize_ring_buffer(struct trace_array *tr,
5733 unsigned long size, int cpu)
5734{
5735 int ret;
5736
5737 /*
5738 * If kernel or user changes the size of the ring buffer
5739 * we use the size that was given, and we can forget about
5740 * expanding it later.
5741 */
5742 ring_buffer_expanded = true;
5743
5744 /* May be called before buffers are initialized */
5745 if (!tr->array_buffer.buffer)
5746 return 0;
5747
5748 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
5749 if (ret < 0)
5750 return ret;
5751
5752#ifdef CONFIG_TRACER_MAX_TRACE
5753 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
5754 !tr->current_trace->use_max_tr)
5755 goto out;
5756
5757 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
5758 if (ret < 0) {
5759 int r = resize_buffer_duplicate_size(&tr->array_buffer,
5760 &tr->array_buffer, cpu);
5761 if (r < 0) {
5762 /*
5763 * AARGH! We are left with different
5764 * size max buffer!!!!
5765 * The max buffer is our "snapshot" buffer.
5766 * When a tracer needs a snapshot (one of the
5767 * latency tracers), it swaps the max buffer
5768 * with the saved snap shot. We succeeded to
5769 * update the size of the main buffer, but failed to
5770 * update the size of the max buffer. But when we tried
5771 * to reset the main buffer to the original size, we
5772 * failed there too. This is very unlikely to
5773 * happen, but if it does, warn and kill all
5774 * tracing.
5775 */
5776 WARN_ON(1);
5777 tracing_disabled = 1;
5778 }
5779 return ret;
5780 }
5781
5782 if (cpu == RING_BUFFER_ALL_CPUS)
5783 set_buffer_entries(&tr->max_buffer, size);
5784 else
5785 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
5786
5787 out:
5788#endif /* CONFIG_TRACER_MAX_TRACE */
5789
5790 if (cpu == RING_BUFFER_ALL_CPUS)
5791 set_buffer_entries(&tr->array_buffer, size);
5792 else
5793 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
5794
5795 return ret;
5796}
5797
5798ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
5799 unsigned long size, int cpu_id)
5800{
5801 int ret = size;
5802
5803 mutex_lock(&trace_types_lock);
5804
5805 if (cpu_id != RING_BUFFER_ALL_CPUS) {
5806 /* make sure, this cpu is enabled in the mask */
5807 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
5808 ret = -EINVAL;
5809 goto out;
5810 }
5811 }
5812
5813 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
5814 if (ret < 0)
5815 ret = -ENOMEM;
5816
5817out:
5818 mutex_unlock(&trace_types_lock);
5819
5820 return ret;
5821}
5822
5823
5824/**
5825 * tracing_update_buffers - used by tracing facility to expand ring buffers
5826 *
5827 * To save on memory when the tracing is never used on a system with it
5828 * configured in. The ring buffers are set to a minimum size. But once
5829 * a user starts to use the tracing facility, then they need to grow
5830 * to their default size.
5831 *
5832 * This function is to be called when a tracer is about to be used.
5833 */
5834int tracing_update_buffers(void)
5835{
5836 int ret = 0;
5837
5838 mutex_lock(&trace_types_lock);
5839 if (!ring_buffer_expanded)
5840 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
5841 RING_BUFFER_ALL_CPUS);
5842 mutex_unlock(&trace_types_lock);
5843
5844 return ret;
5845}
5846
5847struct trace_option_dentry;
5848
5849static void
5850create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
5851
5852/*
5853 * Used to clear out the tracer before deletion of an instance.
5854 * Must have trace_types_lock held.
5855 */
5856static void tracing_set_nop(struct trace_array *tr)
5857{
5858 if (tr->current_trace == &nop_trace)
5859 return;
5860
5861 tr->current_trace->enabled--;
5862
5863 if (tr->current_trace->reset)
5864 tr->current_trace->reset(tr);
5865
5866 tr->current_trace = &nop_trace;
5867}
5868
5869static void add_tracer_options(struct trace_array *tr, struct tracer *t)
5870{
5871 /* Only enable if the directory has been created already. */
5872 if (!tr->dir)
5873 return;
5874
5875 create_trace_option_files(tr, t);
5876}
5877
5878int tracing_set_tracer(struct trace_array *tr, const char *buf)
5879{
5880 struct tracer *t;
5881#ifdef CONFIG_TRACER_MAX_TRACE
5882 bool had_max_tr;
5883#endif
5884 int ret = 0;
5885
5886 mutex_lock(&trace_types_lock);
5887
5888 if (!ring_buffer_expanded) {
5889 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
5890 RING_BUFFER_ALL_CPUS);
5891 if (ret < 0)
5892 goto out;
5893 ret = 0;
5894 }
5895
5896 for (t = trace_types; t; t = t->next) {
5897 if (strcmp(t->name, buf) == 0)
5898 break;
5899 }
5900 if (!t) {
5901 ret = -EINVAL;
5902 goto out;
5903 }
5904 if (t == tr->current_trace)
5905 goto out;
5906
5907#ifdef CONFIG_TRACER_SNAPSHOT
5908 if (t->use_max_tr) {
5909 arch_spin_lock(&tr->max_lock);
5910 if (tr->cond_snapshot)
5911 ret = -EBUSY;
5912 arch_spin_unlock(&tr->max_lock);
5913 if (ret)
5914 goto out;
5915 }
5916#endif
5917 /* Some tracers won't work on kernel command line */
5918 if (system_state < SYSTEM_RUNNING && t->noboot) {
5919 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
5920 t->name);
5921 goto out;
5922 }
5923
5924 /* Some tracers are only allowed for the top level buffer */
5925 if (!trace_ok_for_array(t, tr)) {
5926 ret = -EINVAL;
5927 goto out;
5928 }
5929
5930 /* If trace pipe files are being read, we can't change the tracer */
5931 if (tr->trace_ref) {
5932 ret = -EBUSY;
5933 goto out;
5934 }
5935
5936 trace_branch_disable();
5937
5938 tr->current_trace->enabled--;
5939
5940 if (tr->current_trace->reset)
5941 tr->current_trace->reset(tr);
5942
5943 /* Current trace needs to be nop_trace before synchronize_rcu */
5944 tr->current_trace = &nop_trace;
5945
5946#ifdef CONFIG_TRACER_MAX_TRACE
5947 had_max_tr = tr->allocated_snapshot;
5948
5949 if (had_max_tr && !t->use_max_tr) {
5950 /*
5951 * We need to make sure that the update_max_tr sees that
5952 * current_trace changed to nop_trace to keep it from
5953 * swapping the buffers after we resize it.
5954 * The update_max_tr is called from interrupts disabled
5955 * so a synchronized_sched() is sufficient.
5956 */
5957 synchronize_rcu();
5958 free_snapshot(tr);
5959 }
5960#endif
5961
5962#ifdef CONFIG_TRACER_MAX_TRACE
5963 if (t->use_max_tr && !had_max_tr) {
5964 ret = tracing_alloc_snapshot_instance(tr);
5965 if (ret < 0)
5966 goto out;
5967 }
5968#endif
5969
5970 if (t->init) {
5971 ret = tracer_init(t, tr);
5972 if (ret)
5973 goto out;
5974 }
5975
5976 tr->current_trace = t;
5977 tr->current_trace->enabled++;
5978 trace_branch_enable(tr);
5979 out:
5980 mutex_unlock(&trace_types_lock);
5981
5982 return ret;
5983}
5984
5985static ssize_t
5986tracing_set_trace_write(struct file *filp, const char __user *ubuf,
5987 size_t cnt, loff_t *ppos)
5988{
5989 struct trace_array *tr = filp->private_data;
5990 char buf[MAX_TRACER_SIZE+1];
5991 int i;
5992 size_t ret;
5993 int err;
5994
5995 ret = cnt;
5996
5997 if (cnt > MAX_TRACER_SIZE)
5998 cnt = MAX_TRACER_SIZE;
5999
6000 if (copy_from_user(buf, ubuf, cnt))
6001 return -EFAULT;
6002
6003 buf[cnt] = 0;
6004
6005 /* strip ending whitespace. */
6006 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
6007 buf[i] = 0;
6008
6009 err = tracing_set_tracer(tr, buf);
6010 if (err)
6011 return err;
6012
6013 *ppos += ret;
6014
6015 return ret;
6016}
6017
6018static ssize_t
6019tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6020 size_t cnt, loff_t *ppos)
6021{
6022 char buf[64];
6023 int r;
6024
6025 r = snprintf(buf, sizeof(buf), "%ld\n",
6026 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6027 if (r > sizeof(buf))
6028 r = sizeof(buf);
6029 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6030}
6031
6032static ssize_t
6033tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6034 size_t cnt, loff_t *ppos)
6035{
6036 unsigned long val;
6037 int ret;
6038
6039 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6040 if (ret)
6041 return ret;
6042
6043 *ptr = val * 1000;
6044
6045 return cnt;
6046}
6047
6048static ssize_t
6049tracing_thresh_read(struct file *filp, char __user *ubuf,
6050 size_t cnt, loff_t *ppos)
6051{
6052 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6053}
6054
6055static ssize_t
6056tracing_thresh_write(struct file *filp, const char __user *ubuf,
6057 size_t cnt, loff_t *ppos)
6058{
6059 struct trace_array *tr = filp->private_data;
6060 int ret;
6061
6062 mutex_lock(&trace_types_lock);
6063 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6064 if (ret < 0)
6065 goto out;
6066
6067 if (tr->current_trace->update_thresh) {
6068 ret = tr->current_trace->update_thresh(tr);
6069 if (ret < 0)
6070 goto out;
6071 }
6072
6073 ret = cnt;
6074out:
6075 mutex_unlock(&trace_types_lock);
6076
6077 return ret;
6078}
6079
6080#if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
6081
6082static ssize_t
6083tracing_max_lat_read(struct file *filp, char __user *ubuf,
6084 size_t cnt, loff_t *ppos)
6085{
6086 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6087}
6088
6089static ssize_t
6090tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6091 size_t cnt, loff_t *ppos)
6092{
6093 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6094}
6095
6096#endif
6097
6098static int tracing_open_pipe(struct inode *inode, struct file *filp)
6099{
6100 struct trace_array *tr = inode->i_private;
6101 struct trace_iterator *iter;
6102 int ret;
6103
6104 ret = tracing_check_open_get_tr(tr);
6105 if (ret)
6106 return ret;
6107
6108 mutex_lock(&trace_types_lock);
6109
6110 /* create a buffer to store the information to pass to userspace */
6111 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6112 if (!iter) {
6113 ret = -ENOMEM;
6114 __trace_array_put(tr);
6115 goto out;
6116 }
6117
6118 trace_seq_init(&iter->seq);
6119 iter->trace = tr->current_trace;
6120
6121 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6122 ret = -ENOMEM;
6123 goto fail;
6124 }
6125
6126 /* trace pipe does not show start of buffer */
6127 cpumask_setall(iter->started);
6128
6129 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6130 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6131
6132 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6133 if (trace_clocks[tr->clock_id].in_ns)
6134 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6135
6136 iter->tr = tr;
6137 iter->array_buffer = &tr->array_buffer;
6138 iter->cpu_file = tracing_get_cpu(inode);
6139 mutex_init(&iter->mutex);
6140 filp->private_data = iter;
6141
6142 if (iter->trace->pipe_open)
6143 iter->trace->pipe_open(iter);
6144
6145 nonseekable_open(inode, filp);
6146
6147 tr->trace_ref++;
6148out:
6149 mutex_unlock(&trace_types_lock);
6150 return ret;
6151
6152fail:
6153 kfree(iter);
6154 __trace_array_put(tr);
6155 mutex_unlock(&trace_types_lock);
6156 return ret;
6157}
6158
6159static int tracing_release_pipe(struct inode *inode, struct file *file)
6160{
6161 struct trace_iterator *iter = file->private_data;
6162 struct trace_array *tr = inode->i_private;
6163
6164 mutex_lock(&trace_types_lock);
6165
6166 tr->trace_ref--;
6167
6168 if (iter->trace->pipe_close)
6169 iter->trace->pipe_close(iter);
6170
6171 mutex_unlock(&trace_types_lock);
6172
6173 free_cpumask_var(iter->started);
6174 mutex_destroy(&iter->mutex);
6175 kfree(iter);
6176
6177 trace_array_put(tr);
6178
6179 return 0;
6180}
6181
6182static __poll_t
6183trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6184{
6185 struct trace_array *tr = iter->tr;
6186
6187 /* Iterators are static, they should be filled or empty */
6188 if (trace_buffer_iter(iter, iter->cpu_file))
6189 return EPOLLIN | EPOLLRDNORM;
6190
6191 if (tr->trace_flags & TRACE_ITER_BLOCK)
6192 /*
6193 * Always select as readable when in blocking mode
6194 */
6195 return EPOLLIN | EPOLLRDNORM;
6196 else
6197 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6198 filp, poll_table);
6199}
6200
6201static __poll_t
6202tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6203{
6204 struct trace_iterator *iter = filp->private_data;
6205
6206 return trace_poll(iter, filp, poll_table);
6207}
6208
6209/* Must be called with iter->mutex held. */
6210static int tracing_wait_pipe(struct file *filp)
6211{
6212 struct trace_iterator *iter = filp->private_data;
6213 int ret;
6214
6215 while (trace_empty(iter)) {
6216
6217 if ((filp->f_flags & O_NONBLOCK)) {
6218 return -EAGAIN;
6219 }
6220
6221 /*
6222 * We block until we read something and tracing is disabled.
6223 * We still block if tracing is disabled, but we have never
6224 * read anything. This allows a user to cat this file, and
6225 * then enable tracing. But after we have read something,
6226 * we give an EOF when tracing is again disabled.
6227 *
6228 * iter->pos will be 0 if we haven't read anything.
6229 */
6230 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6231 break;
6232
6233 mutex_unlock(&iter->mutex);
6234
6235 ret = wait_on_pipe(iter, 0);
6236
6237 mutex_lock(&iter->mutex);
6238
6239 if (ret)
6240 return ret;
6241 }
6242
6243 return 1;
6244}
6245
6246/*
6247 * Consumer reader.
6248 */
6249static ssize_t
6250tracing_read_pipe(struct file *filp, char __user *ubuf,
6251 size_t cnt, loff_t *ppos)
6252{
6253 struct trace_iterator *iter = filp->private_data;
6254 ssize_t sret;
6255
6256 /*
6257 * Avoid more than one consumer on a single file descriptor
6258 * This is just a matter of traces coherency, the ring buffer itself
6259 * is protected.
6260 */
6261 mutex_lock(&iter->mutex);
6262
6263 /* return any leftover data */
6264 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6265 if (sret != -EBUSY)
6266 goto out;
6267
6268 trace_seq_init(&iter->seq);
6269
6270 if (iter->trace->read) {
6271 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6272 if (sret)
6273 goto out;
6274 }
6275
6276waitagain:
6277 sret = tracing_wait_pipe(filp);
6278 if (sret <= 0)
6279 goto out;
6280
6281 /* stop when tracing is finished */
6282 if (trace_empty(iter)) {
6283 sret = 0;
6284 goto out;
6285 }
6286
6287 if (cnt >= PAGE_SIZE)
6288 cnt = PAGE_SIZE - 1;
6289
6290 /* reset all but tr, trace, and overruns */
6291 memset(&iter->seq, 0,
6292 sizeof(struct trace_iterator) -
6293 offsetof(struct trace_iterator, seq));
6294 cpumask_clear(iter->started);
6295 trace_seq_init(&iter->seq);
6296 iter->pos = -1;
6297
6298 trace_event_read_lock();
6299 trace_access_lock(iter->cpu_file);
6300 while (trace_find_next_entry_inc(iter) != NULL) {
6301 enum print_line_t ret;
6302 int save_len = iter->seq.seq.len;
6303
6304 ret = print_trace_line(iter);
6305 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6306 /* don't print partial lines */
6307 iter->seq.seq.len = save_len;
6308 break;
6309 }
6310 if (ret != TRACE_TYPE_NO_CONSUME)
6311 trace_consume(iter);
6312
6313 if (trace_seq_used(&iter->seq) >= cnt)
6314 break;
6315
6316 /*
6317 * Setting the full flag means we reached the trace_seq buffer
6318 * size and we should leave by partial output condition above.
6319 * One of the trace_seq_* functions is not used properly.
6320 */
6321 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6322 iter->ent->type);
6323 }
6324 trace_access_unlock(iter->cpu_file);
6325 trace_event_read_unlock();
6326
6327 /* Now copy what we have to the user */
6328 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6329 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6330 trace_seq_init(&iter->seq);
6331
6332 /*
6333 * If there was nothing to send to user, in spite of consuming trace
6334 * entries, go back to wait for more entries.
6335 */
6336 if (sret == -EBUSY)
6337 goto waitagain;
6338
6339out:
6340 mutex_unlock(&iter->mutex);
6341
6342 return sret;
6343}
6344
6345static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6346 unsigned int idx)
6347{
6348 __free_page(spd->pages[idx]);
6349}
6350
6351static size_t
6352tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6353{
6354 size_t count;
6355 int save_len;
6356 int ret;
6357
6358 /* Seq buffer is page-sized, exactly what we need. */
6359 for (;;) {
6360 save_len = iter->seq.seq.len;
6361 ret = print_trace_line(iter);
6362
6363 if (trace_seq_has_overflowed(&iter->seq)) {
6364 iter->seq.seq.len = save_len;
6365 break;
6366 }
6367
6368 /*
6369 * This should not be hit, because it should only
6370 * be set if the iter->seq overflowed. But check it
6371 * anyway to be safe.
6372 */
6373 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6374 iter->seq.seq.len = save_len;
6375 break;
6376 }
6377
6378 count = trace_seq_used(&iter->seq) - save_len;
6379 if (rem < count) {
6380 rem = 0;
6381 iter->seq.seq.len = save_len;
6382 break;
6383 }
6384
6385 if (ret != TRACE_TYPE_NO_CONSUME)
6386 trace_consume(iter);
6387 rem -= count;
6388 if (!trace_find_next_entry_inc(iter)) {
6389 rem = 0;
6390 iter->ent = NULL;
6391 break;
6392 }
6393 }
6394
6395 return rem;
6396}
6397
6398static ssize_t tracing_splice_read_pipe(struct file *filp,
6399 loff_t *ppos,
6400 struct pipe_inode_info *pipe,
6401 size_t len,
6402 unsigned int flags)
6403{
6404 struct page *pages_def[PIPE_DEF_BUFFERS];
6405 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6406 struct trace_iterator *iter = filp->private_data;
6407 struct splice_pipe_desc spd = {
6408 .pages = pages_def,
6409 .partial = partial_def,
6410 .nr_pages = 0, /* This gets updated below. */
6411 .nr_pages_max = PIPE_DEF_BUFFERS,
6412 .ops = &default_pipe_buf_ops,
6413 .spd_release = tracing_spd_release_pipe,
6414 };
6415 ssize_t ret;
6416 size_t rem;
6417 unsigned int i;
6418
6419 if (splice_grow_spd(pipe, &spd))
6420 return -ENOMEM;
6421
6422 mutex_lock(&iter->mutex);
6423
6424 if (iter->trace->splice_read) {
6425 ret = iter->trace->splice_read(iter, filp,
6426 ppos, pipe, len, flags);
6427 if (ret)
6428 goto out_err;
6429 }
6430
6431 ret = tracing_wait_pipe(filp);
6432 if (ret <= 0)
6433 goto out_err;
6434
6435 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6436 ret = -EFAULT;
6437 goto out_err;
6438 }
6439
6440 trace_event_read_lock();
6441 trace_access_lock(iter->cpu_file);
6442
6443 /* Fill as many pages as possible. */
6444 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6445 spd.pages[i] = alloc_page(GFP_KERNEL);
6446 if (!spd.pages[i])
6447 break;
6448
6449 rem = tracing_fill_pipe_page(rem, iter);
6450
6451 /* Copy the data into the page, so we can start over. */
6452 ret = trace_seq_to_buffer(&iter->seq,
6453 page_address(spd.pages[i]),
6454 trace_seq_used(&iter->seq));
6455 if (ret < 0) {
6456 __free_page(spd.pages[i]);
6457 break;
6458 }
6459 spd.partial[i].offset = 0;
6460 spd.partial[i].len = trace_seq_used(&iter->seq);
6461
6462 trace_seq_init(&iter->seq);
6463 }
6464
6465 trace_access_unlock(iter->cpu_file);
6466 trace_event_read_unlock();
6467 mutex_unlock(&iter->mutex);
6468
6469 spd.nr_pages = i;
6470
6471 if (i)
6472 ret = splice_to_pipe(pipe, &spd);
6473 else
6474 ret = 0;
6475out:
6476 splice_shrink_spd(&spd);
6477 return ret;
6478
6479out_err:
6480 mutex_unlock(&iter->mutex);
6481 goto out;
6482}
6483
6484static ssize_t
6485tracing_entries_read(struct file *filp, char __user *ubuf,
6486 size_t cnt, loff_t *ppos)
6487{
6488 struct inode *inode = file_inode(filp);
6489 struct trace_array *tr = inode->i_private;
6490 int cpu = tracing_get_cpu(inode);
6491 char buf[64];
6492 int r = 0;
6493 ssize_t ret;
6494
6495 mutex_lock(&trace_types_lock);
6496
6497 if (cpu == RING_BUFFER_ALL_CPUS) {
6498 int cpu, buf_size_same;
6499 unsigned long size;
6500
6501 size = 0;
6502 buf_size_same = 1;
6503 /* check if all cpu sizes are same */
6504 for_each_tracing_cpu(cpu) {
6505 /* fill in the size from first enabled cpu */
6506 if (size == 0)
6507 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6508 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6509 buf_size_same = 0;
6510 break;
6511 }
6512 }
6513
6514 if (buf_size_same) {
6515 if (!ring_buffer_expanded)
6516 r = sprintf(buf, "%lu (expanded: %lu)\n",
6517 size >> 10,
6518 trace_buf_size >> 10);
6519 else
6520 r = sprintf(buf, "%lu\n", size >> 10);
6521 } else
6522 r = sprintf(buf, "X\n");
6523 } else
6524 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6525
6526 mutex_unlock(&trace_types_lock);
6527
6528 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6529 return ret;
6530}
6531
6532static ssize_t
6533tracing_entries_write(struct file *filp, const char __user *ubuf,
6534 size_t cnt, loff_t *ppos)
6535{
6536 struct inode *inode = file_inode(filp);
6537 struct trace_array *tr = inode->i_private;
6538 unsigned long val;
6539 int ret;
6540
6541 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6542 if (ret)
6543 return ret;
6544
6545 /* must have at least 1 entry */
6546 if (!val)
6547 return -EINVAL;
6548
6549 /* value is in KB */
6550 val <<= 10;
6551 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6552 if (ret < 0)
6553 return ret;
6554
6555 *ppos += cnt;
6556
6557 return cnt;
6558}
6559
6560static ssize_t
6561tracing_total_entries_read(struct file *filp, char __user *ubuf,
6562 size_t cnt, loff_t *ppos)
6563{
6564 struct trace_array *tr = filp->private_data;
6565 char buf[64];
6566 int r, cpu;
6567 unsigned long size = 0, expanded_size = 0;
6568
6569 mutex_lock(&trace_types_lock);
6570 for_each_tracing_cpu(cpu) {
6571 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6572 if (!ring_buffer_expanded)
6573 expanded_size += trace_buf_size >> 10;
6574 }
6575 if (ring_buffer_expanded)
6576 r = sprintf(buf, "%lu\n", size);
6577 else
6578 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6579 mutex_unlock(&trace_types_lock);
6580
6581 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6582}
6583
6584static ssize_t
6585tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
6586 size_t cnt, loff_t *ppos)
6587{
6588 /*
6589 * There is no need to read what the user has written, this function
6590 * is just to make sure that there is no error when "echo" is used
6591 */
6592
6593 *ppos += cnt;
6594
6595 return cnt;
6596}
6597
6598static int
6599tracing_free_buffer_release(struct inode *inode, struct file *filp)
6600{
6601 struct trace_array *tr = inode->i_private;
6602
6603 /* disable tracing ? */
6604 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
6605 tracer_tracing_off(tr);
6606 /* resize the ring buffer to 0 */
6607 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
6608
6609 trace_array_put(tr);
6610
6611 return 0;
6612}
6613
6614static ssize_t
6615tracing_mark_write(struct file *filp, const char __user *ubuf,
6616 size_t cnt, loff_t *fpos)
6617{
6618 struct trace_array *tr = filp->private_data;
6619 struct ring_buffer_event *event;
6620 enum event_trigger_type tt = ETT_NONE;
6621 struct trace_buffer *buffer;
6622 struct print_entry *entry;
6623 unsigned long irq_flags;
6624 ssize_t written;
6625 int size;
6626 int len;
6627
6628/* Used in tracing_mark_raw_write() as well */
6629#define FAULTED_STR "<faulted>"
6630#define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
6631
6632 if (tracing_disabled)
6633 return -EINVAL;
6634
6635 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6636 return -EINVAL;
6637
6638 if (cnt > TRACE_BUF_SIZE)
6639 cnt = TRACE_BUF_SIZE;
6640
6641 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6642
6643 local_save_flags(irq_flags);
6644 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
6645
6646 /* If less than "<faulted>", then make sure we can still add that */
6647 if (cnt < FAULTED_SIZE)
6648 size += FAULTED_SIZE - cnt;
6649
6650 buffer = tr->array_buffer.buffer;
6651 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
6652 irq_flags, preempt_count());
6653 if (unlikely(!event))
6654 /* Ring buffer disabled, return as if not open for write */
6655 return -EBADF;
6656
6657 entry = ring_buffer_event_data(event);
6658 entry->ip = _THIS_IP_;
6659
6660 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
6661 if (len) {
6662 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6663 cnt = FAULTED_SIZE;
6664 written = -EFAULT;
6665 } else
6666 written = cnt;
6667 len = cnt;
6668
6669 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
6670 /* do not add \n before testing triggers, but add \0 */
6671 entry->buf[cnt] = '\0';
6672 tt = event_triggers_call(tr->trace_marker_file, entry, event);
6673 }
6674
6675 if (entry->buf[cnt - 1] != '\n') {
6676 entry->buf[cnt] = '\n';
6677 entry->buf[cnt + 1] = '\0';
6678 } else
6679 entry->buf[cnt] = '\0';
6680
6681 __buffer_unlock_commit(buffer, event);
6682
6683 if (tt)
6684 event_triggers_post_call(tr->trace_marker_file, tt);
6685
6686 if (written > 0)
6687 *fpos += written;
6688
6689 return written;
6690}
6691
6692/* Limit it for now to 3K (including tag) */
6693#define RAW_DATA_MAX_SIZE (1024*3)
6694
6695static ssize_t
6696tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
6697 size_t cnt, loff_t *fpos)
6698{
6699 struct trace_array *tr = filp->private_data;
6700 struct ring_buffer_event *event;
6701 struct trace_buffer *buffer;
6702 struct raw_data_entry *entry;
6703 unsigned long irq_flags;
6704 ssize_t written;
6705 int size;
6706 int len;
6707
6708#define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
6709
6710 if (tracing_disabled)
6711 return -EINVAL;
6712
6713 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6714 return -EINVAL;
6715
6716 /* The marker must at least have a tag id */
6717 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
6718 return -EINVAL;
6719
6720 if (cnt > TRACE_BUF_SIZE)
6721 cnt = TRACE_BUF_SIZE;
6722
6723 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6724
6725 local_save_flags(irq_flags);
6726 size = sizeof(*entry) + cnt;
6727 if (cnt < FAULT_SIZE_ID)
6728 size += FAULT_SIZE_ID - cnt;
6729
6730 buffer = tr->array_buffer.buffer;
6731 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
6732 irq_flags, preempt_count());
6733 if (!event)
6734 /* Ring buffer disabled, return as if not open for write */
6735 return -EBADF;
6736
6737 entry = ring_buffer_event_data(event);
6738
6739 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
6740 if (len) {
6741 entry->id = -1;
6742 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6743 written = -EFAULT;
6744 } else
6745 written = cnt;
6746
6747 __buffer_unlock_commit(buffer, event);
6748
6749 if (written > 0)
6750 *fpos += written;
6751
6752 return written;
6753}
6754
6755static int tracing_clock_show(struct seq_file *m, void *v)
6756{
6757 struct trace_array *tr = m->private;
6758 int i;
6759
6760 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
6761 seq_printf(m,
6762 "%s%s%s%s", i ? " " : "",
6763 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
6764 i == tr->clock_id ? "]" : "");
6765 seq_putc(m, '\n');
6766
6767 return 0;
6768}
6769
6770int tracing_set_clock(struct trace_array *tr, const char *clockstr)
6771{
6772 int i;
6773
6774 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
6775 if (strcmp(trace_clocks[i].name, clockstr) == 0)
6776 break;
6777 }
6778 if (i == ARRAY_SIZE(trace_clocks))
6779 return -EINVAL;
6780
6781 mutex_lock(&trace_types_lock);
6782
6783 tr->clock_id = i;
6784
6785 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
6786
6787 /*
6788 * New clock may not be consistent with the previous clock.
6789 * Reset the buffer so that it doesn't have incomparable timestamps.
6790 */
6791 tracing_reset_online_cpus(&tr->array_buffer);
6792
6793#ifdef CONFIG_TRACER_MAX_TRACE
6794 if (tr->max_buffer.buffer)
6795 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
6796 tracing_reset_online_cpus(&tr->max_buffer);
6797#endif
6798
6799 mutex_unlock(&trace_types_lock);
6800
6801 return 0;
6802}
6803
6804static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
6805 size_t cnt, loff_t *fpos)
6806{
6807 struct seq_file *m = filp->private_data;
6808 struct trace_array *tr = m->private;
6809 char buf[64];
6810 const char *clockstr;
6811 int ret;
6812
6813 if (cnt >= sizeof(buf))
6814 return -EINVAL;
6815
6816 if (copy_from_user(buf, ubuf, cnt))
6817 return -EFAULT;
6818
6819 buf[cnt] = 0;
6820
6821 clockstr = strstrip(buf);
6822
6823 ret = tracing_set_clock(tr, clockstr);
6824 if (ret)
6825 return ret;
6826
6827 *fpos += cnt;
6828
6829 return cnt;
6830}
6831
6832static int tracing_clock_open(struct inode *inode, struct file *file)
6833{
6834 struct trace_array *tr = inode->i_private;
6835 int ret;
6836
6837 ret = tracing_check_open_get_tr(tr);
6838 if (ret)
6839 return ret;
6840
6841 ret = single_open(file, tracing_clock_show, inode->i_private);
6842 if (ret < 0)
6843 trace_array_put(tr);
6844
6845 return ret;
6846}
6847
6848static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
6849{
6850 struct trace_array *tr = m->private;
6851
6852 mutex_lock(&trace_types_lock);
6853
6854 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
6855 seq_puts(m, "delta [absolute]\n");
6856 else
6857 seq_puts(m, "[delta] absolute\n");
6858
6859 mutex_unlock(&trace_types_lock);
6860
6861 return 0;
6862}
6863
6864static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
6865{
6866 struct trace_array *tr = inode->i_private;
6867 int ret;
6868
6869 ret = tracing_check_open_get_tr(tr);
6870 if (ret)
6871 return ret;
6872
6873 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
6874 if (ret < 0)
6875 trace_array_put(tr);
6876
6877 return ret;
6878}
6879
6880int tracing_set_time_stamp_abs(struct trace_array *tr, bool abs)
6881{
6882 int ret = 0;
6883
6884 mutex_lock(&trace_types_lock);
6885
6886 if (abs && tr->time_stamp_abs_ref++)
6887 goto out;
6888
6889 if (!abs) {
6890 if (WARN_ON_ONCE(!tr->time_stamp_abs_ref)) {
6891 ret = -EINVAL;
6892 goto out;
6893 }
6894
6895 if (--tr->time_stamp_abs_ref)
6896 goto out;
6897 }
6898
6899 ring_buffer_set_time_stamp_abs(tr->array_buffer.buffer, abs);
6900
6901#ifdef CONFIG_TRACER_MAX_TRACE
6902 if (tr->max_buffer.buffer)
6903 ring_buffer_set_time_stamp_abs(tr->max_buffer.buffer, abs);
6904#endif
6905 out:
6906 mutex_unlock(&trace_types_lock);
6907
6908 return ret;
6909}
6910
6911struct ftrace_buffer_info {
6912 struct trace_iterator iter;
6913 void *spare;
6914 unsigned int spare_cpu;
6915 unsigned int read;
6916};
6917
6918#ifdef CONFIG_TRACER_SNAPSHOT
6919static int tracing_snapshot_open(struct inode *inode, struct file *file)
6920{
6921 struct trace_array *tr = inode->i_private;
6922 struct trace_iterator *iter;
6923 struct seq_file *m;
6924 int ret;
6925
6926 ret = tracing_check_open_get_tr(tr);
6927 if (ret)
6928 return ret;
6929
6930 if (file->f_mode & FMODE_READ) {
6931 iter = __tracing_open(inode, file, true);
6932 if (IS_ERR(iter))
6933 ret = PTR_ERR(iter);
6934 } else {
6935 /* Writes still need the seq_file to hold the private data */
6936 ret = -ENOMEM;
6937 m = kzalloc(sizeof(*m), GFP_KERNEL);
6938 if (!m)
6939 goto out;
6940 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6941 if (!iter) {
6942 kfree(m);
6943 goto out;
6944 }
6945 ret = 0;
6946
6947 iter->tr = tr;
6948 iter->array_buffer = &tr->max_buffer;
6949 iter->cpu_file = tracing_get_cpu(inode);
6950 m->private = iter;
6951 file->private_data = m;
6952 }
6953out:
6954 if (ret < 0)
6955 trace_array_put(tr);
6956
6957 return ret;
6958}
6959
6960static ssize_t
6961tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
6962 loff_t *ppos)
6963{
6964 struct seq_file *m = filp->private_data;
6965 struct trace_iterator *iter = m->private;
6966 struct trace_array *tr = iter->tr;
6967 unsigned long val;
6968 int ret;
6969
6970 ret = tracing_update_buffers();
6971 if (ret < 0)
6972 return ret;
6973
6974 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6975 if (ret)
6976 return ret;
6977
6978 mutex_lock(&trace_types_lock);
6979
6980 if (tr->current_trace->use_max_tr) {
6981 ret = -EBUSY;
6982 goto out;
6983 }
6984
6985 arch_spin_lock(&tr->max_lock);
6986 if (tr->cond_snapshot)
6987 ret = -EBUSY;
6988 arch_spin_unlock(&tr->max_lock);
6989 if (ret)
6990 goto out;
6991
6992 switch (val) {
6993 case 0:
6994 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
6995 ret = -EINVAL;
6996 break;
6997 }
6998 if (tr->allocated_snapshot)
6999 free_snapshot(tr);
7000 break;
7001 case 1:
7002/* Only allow per-cpu swap if the ring buffer supports it */
7003#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7004 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7005 ret = -EINVAL;
7006 break;
7007 }
7008#endif
7009 if (tr->allocated_snapshot)
7010 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7011 &tr->array_buffer, iter->cpu_file);
7012 else
7013 ret = tracing_alloc_snapshot_instance(tr);
7014 if (ret < 0)
7015 break;
7016 local_irq_disable();
7017 /* Now, we're going to swap */
7018 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7019 update_max_tr(tr, current, smp_processor_id(), NULL);
7020 else
7021 update_max_tr_single(tr, current, iter->cpu_file);
7022 local_irq_enable();
7023 break;
7024 default:
7025 if (tr->allocated_snapshot) {
7026 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7027 tracing_reset_online_cpus(&tr->max_buffer);
7028 else
7029 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7030 }
7031 break;
7032 }
7033
7034 if (ret >= 0) {
7035 *ppos += cnt;
7036 ret = cnt;
7037 }
7038out:
7039 mutex_unlock(&trace_types_lock);
7040 return ret;
7041}
7042
7043static int tracing_snapshot_release(struct inode *inode, struct file *file)
7044{
7045 struct seq_file *m = file->private_data;
7046 int ret;
7047
7048 ret = tracing_release(inode, file);
7049
7050 if (file->f_mode & FMODE_READ)
7051 return ret;
7052
7053 /* If write only, the seq_file is just a stub */
7054 if (m)
7055 kfree(m->private);
7056 kfree(m);
7057
7058 return 0;
7059}
7060
7061static int tracing_buffers_open(struct inode *inode, struct file *filp);
7062static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7063 size_t count, loff_t *ppos);
7064static int tracing_buffers_release(struct inode *inode, struct file *file);
7065static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7066 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7067
7068static int snapshot_raw_open(struct inode *inode, struct file *filp)
7069{
7070 struct ftrace_buffer_info *info;
7071 int ret;
7072
7073 /* The following checks for tracefs lockdown */
7074 ret = tracing_buffers_open(inode, filp);
7075 if (ret < 0)
7076 return ret;
7077
7078 info = filp->private_data;
7079
7080 if (info->iter.trace->use_max_tr) {
7081 tracing_buffers_release(inode, filp);
7082 return -EBUSY;
7083 }
7084
7085 info->iter.snapshot = true;
7086 info->iter.array_buffer = &info->iter.tr->max_buffer;
7087
7088 return ret;
7089}
7090
7091#endif /* CONFIG_TRACER_SNAPSHOT */
7092
7093
7094static const struct file_operations tracing_thresh_fops = {
7095 .open = tracing_open_generic,
7096 .read = tracing_thresh_read,
7097 .write = tracing_thresh_write,
7098 .llseek = generic_file_llseek,
7099};
7100
7101#if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
7102static const struct file_operations tracing_max_lat_fops = {
7103 .open = tracing_open_generic,
7104 .read = tracing_max_lat_read,
7105 .write = tracing_max_lat_write,
7106 .llseek = generic_file_llseek,
7107};
7108#endif
7109
7110static const struct file_operations set_tracer_fops = {
7111 .open = tracing_open_generic,
7112 .read = tracing_set_trace_read,
7113 .write = tracing_set_trace_write,
7114 .llseek = generic_file_llseek,
7115};
7116
7117static const struct file_operations tracing_pipe_fops = {
7118 .open = tracing_open_pipe,
7119 .poll = tracing_poll_pipe,
7120 .read = tracing_read_pipe,
7121 .splice_read = tracing_splice_read_pipe,
7122 .release = tracing_release_pipe,
7123 .llseek = no_llseek,
7124};
7125
7126static const struct file_operations tracing_entries_fops = {
7127 .open = tracing_open_generic_tr,
7128 .read = tracing_entries_read,
7129 .write = tracing_entries_write,
7130 .llseek = generic_file_llseek,
7131 .release = tracing_release_generic_tr,
7132};
7133
7134static const struct file_operations tracing_total_entries_fops = {
7135 .open = tracing_open_generic_tr,
7136 .read = tracing_total_entries_read,
7137 .llseek = generic_file_llseek,
7138 .release = tracing_release_generic_tr,
7139};
7140
7141static const struct file_operations tracing_free_buffer_fops = {
7142 .open = tracing_open_generic_tr,
7143 .write = tracing_free_buffer_write,
7144 .release = tracing_free_buffer_release,
7145};
7146
7147static const struct file_operations tracing_mark_fops = {
7148 .open = tracing_open_generic_tr,
7149 .write = tracing_mark_write,
7150 .llseek = generic_file_llseek,
7151 .release = tracing_release_generic_tr,
7152};
7153
7154static const struct file_operations tracing_mark_raw_fops = {
7155 .open = tracing_open_generic_tr,
7156 .write = tracing_mark_raw_write,
7157 .llseek = generic_file_llseek,
7158 .release = tracing_release_generic_tr,
7159};
7160
7161static const struct file_operations trace_clock_fops = {
7162 .open = tracing_clock_open,
7163 .read = seq_read,
7164 .llseek = seq_lseek,
7165 .release = tracing_single_release_tr,
7166 .write = tracing_clock_write,
7167};
7168
7169static const struct file_operations trace_time_stamp_mode_fops = {
7170 .open = tracing_time_stamp_mode_open,
7171 .read = seq_read,
7172 .llseek = seq_lseek,
7173 .release = tracing_single_release_tr,
7174};
7175
7176#ifdef CONFIG_TRACER_SNAPSHOT
7177static const struct file_operations snapshot_fops = {
7178 .open = tracing_snapshot_open,
7179 .read = seq_read,
7180 .write = tracing_snapshot_write,
7181 .llseek = tracing_lseek,
7182 .release = tracing_snapshot_release,
7183};
7184
7185static const struct file_operations snapshot_raw_fops = {
7186 .open = snapshot_raw_open,
7187 .read = tracing_buffers_read,
7188 .release = tracing_buffers_release,
7189 .splice_read = tracing_buffers_splice_read,
7190 .llseek = no_llseek,
7191};
7192
7193#endif /* CONFIG_TRACER_SNAPSHOT */
7194
7195#define TRACING_LOG_ERRS_MAX 8
7196#define TRACING_LOG_LOC_MAX 128
7197
7198#define CMD_PREFIX " Command: "
7199
7200struct err_info {
7201 const char **errs; /* ptr to loc-specific array of err strings */
7202 u8 type; /* index into errs -> specific err string */
7203 u8 pos; /* MAX_FILTER_STR_VAL = 256 */
7204 u64 ts;
7205};
7206
7207struct tracing_log_err {
7208 struct list_head list;
7209 struct err_info info;
7210 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7211 char cmd[MAX_FILTER_STR_VAL]; /* what caused err */
7212};
7213
7214static DEFINE_MUTEX(tracing_err_log_lock);
7215
7216static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr)
7217{
7218 struct tracing_log_err *err;
7219
7220 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7221 err = kzalloc(sizeof(*err), GFP_KERNEL);
7222 if (!err)
7223 err = ERR_PTR(-ENOMEM);
7224 tr->n_err_log_entries++;
7225
7226 return err;
7227 }
7228
7229 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7230 list_del(&err->list);
7231
7232 return err;
7233}
7234
7235/**
7236 * err_pos - find the position of a string within a command for error careting
7237 * @cmd: The tracing command that caused the error
7238 * @str: The string to position the caret at within @cmd
7239 *
7240 * Finds the position of the first occurence of @str within @cmd. The
7241 * return value can be passed to tracing_log_err() for caret placement
7242 * within @cmd.
7243 *
7244 * Returns the index within @cmd of the first occurence of @str or 0
7245 * if @str was not found.
7246 */
7247unsigned int err_pos(char *cmd, const char *str)
7248{
7249 char *found;
7250
7251 if (WARN_ON(!strlen(cmd)))
7252 return 0;
7253
7254 found = strstr(cmd, str);
7255 if (found)
7256 return found - cmd;
7257
7258 return 0;
7259}
7260
7261/**
7262 * tracing_log_err - write an error to the tracing error log
7263 * @tr: The associated trace array for the error (NULL for top level array)
7264 * @loc: A string describing where the error occurred
7265 * @cmd: The tracing command that caused the error
7266 * @errs: The array of loc-specific static error strings
7267 * @type: The index into errs[], which produces the specific static err string
7268 * @pos: The position the caret should be placed in the cmd
7269 *
7270 * Writes an error into tracing/error_log of the form:
7271 *
7272 * <loc>: error: <text>
7273 * Command: <cmd>
7274 * ^
7275 *
7276 * tracing/error_log is a small log file containing the last
7277 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7278 * unless there has been a tracing error, and the error log can be
7279 * cleared and have its memory freed by writing the empty string in
7280 * truncation mode to it i.e. echo > tracing/error_log.
7281 *
7282 * NOTE: the @errs array along with the @type param are used to
7283 * produce a static error string - this string is not copied and saved
7284 * when the error is logged - only a pointer to it is saved. See
7285 * existing callers for examples of how static strings are typically
7286 * defined for use with tracing_log_err().
7287 */
7288void tracing_log_err(struct trace_array *tr,
7289 const char *loc, const char *cmd,
7290 const char **errs, u8 type, u8 pos)
7291{
7292 struct tracing_log_err *err;
7293
7294 if (!tr)
7295 tr = &global_trace;
7296
7297 mutex_lock(&tracing_err_log_lock);
7298 err = get_tracing_log_err(tr);
7299 if (PTR_ERR(err) == -ENOMEM) {
7300 mutex_unlock(&tracing_err_log_lock);
7301 return;
7302 }
7303
7304 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7305 snprintf(err->cmd, MAX_FILTER_STR_VAL,"\n" CMD_PREFIX "%s\n", cmd);
7306
7307 err->info.errs = errs;
7308 err->info.type = type;
7309 err->info.pos = pos;
7310 err->info.ts = local_clock();
7311
7312 list_add_tail(&err->list, &tr->err_log);
7313 mutex_unlock(&tracing_err_log_lock);
7314}
7315
7316static void clear_tracing_err_log(struct trace_array *tr)
7317{
7318 struct tracing_log_err *err, *next;
7319
7320 mutex_lock(&tracing_err_log_lock);
7321 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7322 list_del(&err->list);
7323 kfree(err);
7324 }
7325
7326 tr->n_err_log_entries = 0;
7327 mutex_unlock(&tracing_err_log_lock);
7328}
7329
7330static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7331{
7332 struct trace_array *tr = m->private;
7333
7334 mutex_lock(&tracing_err_log_lock);
7335
7336 return seq_list_start(&tr->err_log, *pos);
7337}
7338
7339static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7340{
7341 struct trace_array *tr = m->private;
7342
7343 return seq_list_next(v, &tr->err_log, pos);
7344}
7345
7346static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7347{
7348 mutex_unlock(&tracing_err_log_lock);
7349}
7350
7351static void tracing_err_log_show_pos(struct seq_file *m, u8 pos)
7352{
7353 u8 i;
7354
7355 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7356 seq_putc(m, ' ');
7357 for (i = 0; i < pos; i++)
7358 seq_putc(m, ' ');
7359 seq_puts(m, "^\n");
7360}
7361
7362static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7363{
7364 struct tracing_log_err *err = v;
7365
7366 if (err) {
7367 const char *err_text = err->info.errs[err->info.type];
7368 u64 sec = err->info.ts;
7369 u32 nsec;
7370
7371 nsec = do_div(sec, NSEC_PER_SEC);
7372 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7373 err->loc, err_text);
7374 seq_printf(m, "%s", err->cmd);
7375 tracing_err_log_show_pos(m, err->info.pos);
7376 }
7377
7378 return 0;
7379}
7380
7381static const struct seq_operations tracing_err_log_seq_ops = {
7382 .start = tracing_err_log_seq_start,
7383 .next = tracing_err_log_seq_next,
7384 .stop = tracing_err_log_seq_stop,
7385 .show = tracing_err_log_seq_show
7386};
7387
7388static int tracing_err_log_open(struct inode *inode, struct file *file)
7389{
7390 struct trace_array *tr = inode->i_private;
7391 int ret = 0;
7392
7393 ret = tracing_check_open_get_tr(tr);
7394 if (ret)
7395 return ret;
7396
7397 /* If this file was opened for write, then erase contents */
7398 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7399 clear_tracing_err_log(tr);
7400
7401 if (file->f_mode & FMODE_READ) {
7402 ret = seq_open(file, &tracing_err_log_seq_ops);
7403 if (!ret) {
7404 struct seq_file *m = file->private_data;
7405 m->private = tr;
7406 } else {
7407 trace_array_put(tr);
7408 }
7409 }
7410 return ret;
7411}
7412
7413static ssize_t tracing_err_log_write(struct file *file,
7414 const char __user *buffer,
7415 size_t count, loff_t *ppos)
7416{
7417 return count;
7418}
7419
7420static int tracing_err_log_release(struct inode *inode, struct file *file)
7421{
7422 struct trace_array *tr = inode->i_private;
7423
7424 trace_array_put(tr);
7425
7426 if (file->f_mode & FMODE_READ)
7427 seq_release(inode, file);
7428
7429 return 0;
7430}
7431
7432static const struct file_operations tracing_err_log_fops = {
7433 .open = tracing_err_log_open,
7434 .write = tracing_err_log_write,
7435 .read = seq_read,
7436 .llseek = seq_lseek,
7437 .release = tracing_err_log_release,
7438};
7439
7440static int tracing_buffers_open(struct inode *inode, struct file *filp)
7441{
7442 struct trace_array *tr = inode->i_private;
7443 struct ftrace_buffer_info *info;
7444 int ret;
7445
7446 ret = tracing_check_open_get_tr(tr);
7447 if (ret)
7448 return ret;
7449
7450 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7451 if (!info) {
7452 trace_array_put(tr);
7453 return -ENOMEM;
7454 }
7455
7456 mutex_lock(&trace_types_lock);
7457
7458 info->iter.tr = tr;
7459 info->iter.cpu_file = tracing_get_cpu(inode);
7460 info->iter.trace = tr->current_trace;
7461 info->iter.array_buffer = &tr->array_buffer;
7462 info->spare = NULL;
7463 /* Force reading ring buffer for first read */
7464 info->read = (unsigned int)-1;
7465
7466 filp->private_data = info;
7467
7468 tr->trace_ref++;
7469
7470 mutex_unlock(&trace_types_lock);
7471
7472 ret = nonseekable_open(inode, filp);
7473 if (ret < 0)
7474 trace_array_put(tr);
7475
7476 return ret;
7477}
7478
7479static __poll_t
7480tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7481{
7482 struct ftrace_buffer_info *info = filp->private_data;
7483 struct trace_iterator *iter = &info->iter;
7484
7485 return trace_poll(iter, filp, poll_table);
7486}
7487
7488static ssize_t
7489tracing_buffers_read(struct file *filp, char __user *ubuf,
7490 size_t count, loff_t *ppos)
7491{
7492 struct ftrace_buffer_info *info = filp->private_data;
7493 struct trace_iterator *iter = &info->iter;
7494 ssize_t ret = 0;
7495 ssize_t size;
7496
7497 if (!count)
7498 return 0;
7499
7500#ifdef CONFIG_TRACER_MAX_TRACE
7501 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7502 return -EBUSY;
7503#endif
7504
7505 if (!info->spare) {
7506 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
7507 iter->cpu_file);
7508 if (IS_ERR(info->spare)) {
7509 ret = PTR_ERR(info->spare);
7510 info->spare = NULL;
7511 } else {
7512 info->spare_cpu = iter->cpu_file;
7513 }
7514 }
7515 if (!info->spare)
7516 return ret;
7517
7518 /* Do we have previous read data to read? */
7519 if (info->read < PAGE_SIZE)
7520 goto read;
7521
7522 again:
7523 trace_access_lock(iter->cpu_file);
7524 ret = ring_buffer_read_page(iter->array_buffer->buffer,
7525 &info->spare,
7526 count,
7527 iter->cpu_file, 0);
7528 trace_access_unlock(iter->cpu_file);
7529
7530 if (ret < 0) {
7531 if (trace_empty(iter)) {
7532 if ((filp->f_flags & O_NONBLOCK))
7533 return -EAGAIN;
7534
7535 ret = wait_on_pipe(iter, 0);
7536 if (ret)
7537 return ret;
7538
7539 goto again;
7540 }
7541 return 0;
7542 }
7543
7544 info->read = 0;
7545 read:
7546 size = PAGE_SIZE - info->read;
7547 if (size > count)
7548 size = count;
7549
7550 ret = copy_to_user(ubuf, info->spare + info->read, size);
7551 if (ret == size)
7552 return -EFAULT;
7553
7554 size -= ret;
7555
7556 *ppos += size;
7557 info->read += size;
7558
7559 return size;
7560}
7561
7562static int tracing_buffers_release(struct inode *inode, struct file *file)
7563{
7564 struct ftrace_buffer_info *info = file->private_data;
7565 struct trace_iterator *iter = &info->iter;
7566
7567 mutex_lock(&trace_types_lock);
7568
7569 iter->tr->trace_ref--;
7570
7571 __trace_array_put(iter->tr);
7572
7573 if (info->spare)
7574 ring_buffer_free_read_page(iter->array_buffer->buffer,
7575 info->spare_cpu, info->spare);
7576 kvfree(info);
7577
7578 mutex_unlock(&trace_types_lock);
7579
7580 return 0;
7581}
7582
7583struct buffer_ref {
7584 struct trace_buffer *buffer;
7585 void *page;
7586 int cpu;
7587 refcount_t refcount;
7588};
7589
7590static void buffer_ref_release(struct buffer_ref *ref)
7591{
7592 if (!refcount_dec_and_test(&ref->refcount))
7593 return;
7594 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
7595 kfree(ref);
7596}
7597
7598static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
7599 struct pipe_buffer *buf)
7600{
7601 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
7602
7603 buffer_ref_release(ref);
7604 buf->private = 0;
7605}
7606
7607static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
7608 struct pipe_buffer *buf)
7609{
7610 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
7611
7612 if (refcount_read(&ref->refcount) > INT_MAX/2)
7613 return false;
7614
7615 refcount_inc(&ref->refcount);
7616 return true;
7617}
7618
7619/* Pipe buffer operations for a buffer. */
7620static const struct pipe_buf_operations buffer_pipe_buf_ops = {
7621 .release = buffer_pipe_buf_release,
7622 .get = buffer_pipe_buf_get,
7623};
7624
7625/*
7626 * Callback from splice_to_pipe(), if we need to release some pages
7627 * at the end of the spd in case we error'ed out in filling the pipe.
7628 */
7629static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
7630{
7631 struct buffer_ref *ref =
7632 (struct buffer_ref *)spd->partial[i].private;
7633
7634 buffer_ref_release(ref);
7635 spd->partial[i].private = 0;
7636}
7637
7638static ssize_t
7639tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7640 struct pipe_inode_info *pipe, size_t len,
7641 unsigned int flags)
7642{
7643 struct ftrace_buffer_info *info = file->private_data;
7644 struct trace_iterator *iter = &info->iter;
7645 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7646 struct page *pages_def[PIPE_DEF_BUFFERS];
7647 struct splice_pipe_desc spd = {
7648 .pages = pages_def,
7649 .partial = partial_def,
7650 .nr_pages_max = PIPE_DEF_BUFFERS,
7651 .ops = &buffer_pipe_buf_ops,
7652 .spd_release = buffer_spd_release,
7653 };
7654 struct buffer_ref *ref;
7655 int entries, i;
7656 ssize_t ret = 0;
7657
7658#ifdef CONFIG_TRACER_MAX_TRACE
7659 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7660 return -EBUSY;
7661#endif
7662
7663 if (*ppos & (PAGE_SIZE - 1))
7664 return -EINVAL;
7665
7666 if (len & (PAGE_SIZE - 1)) {
7667 if (len < PAGE_SIZE)
7668 return -EINVAL;
7669 len &= PAGE_MASK;
7670 }
7671
7672 if (splice_grow_spd(pipe, &spd))
7673 return -ENOMEM;
7674
7675 again:
7676 trace_access_lock(iter->cpu_file);
7677 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
7678
7679 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
7680 struct page *page;
7681 int r;
7682
7683 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
7684 if (!ref) {
7685 ret = -ENOMEM;
7686 break;
7687 }
7688
7689 refcount_set(&ref->refcount, 1);
7690 ref->buffer = iter->array_buffer->buffer;
7691 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
7692 if (IS_ERR(ref->page)) {
7693 ret = PTR_ERR(ref->page);
7694 ref->page = NULL;
7695 kfree(ref);
7696 break;
7697 }
7698 ref->cpu = iter->cpu_file;
7699
7700 r = ring_buffer_read_page(ref->buffer, &ref->page,
7701 len, iter->cpu_file, 1);
7702 if (r < 0) {
7703 ring_buffer_free_read_page(ref->buffer, ref->cpu,
7704 ref->page);
7705 kfree(ref);
7706 break;
7707 }
7708
7709 page = virt_to_page(ref->page);
7710
7711 spd.pages[i] = page;
7712 spd.partial[i].len = PAGE_SIZE;
7713 spd.partial[i].offset = 0;
7714 spd.partial[i].private = (unsigned long)ref;
7715 spd.nr_pages++;
7716 *ppos += PAGE_SIZE;
7717
7718 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
7719 }
7720
7721 trace_access_unlock(iter->cpu_file);
7722 spd.nr_pages = i;
7723
7724 /* did we read anything? */
7725 if (!spd.nr_pages) {
7726 if (ret)
7727 goto out;
7728
7729 ret = -EAGAIN;
7730 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
7731 goto out;
7732
7733 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
7734 if (ret)
7735 goto out;
7736
7737 goto again;
7738 }
7739
7740 ret = splice_to_pipe(pipe, &spd);
7741out:
7742 splice_shrink_spd(&spd);
7743
7744 return ret;
7745}
7746
7747static const struct file_operations tracing_buffers_fops = {
7748 .open = tracing_buffers_open,
7749 .read = tracing_buffers_read,
7750 .poll = tracing_buffers_poll,
7751 .release = tracing_buffers_release,
7752 .splice_read = tracing_buffers_splice_read,
7753 .llseek = no_llseek,
7754};
7755
7756static ssize_t
7757tracing_stats_read(struct file *filp, char __user *ubuf,
7758 size_t count, loff_t *ppos)
7759{
7760 struct inode *inode = file_inode(filp);
7761 struct trace_array *tr = inode->i_private;
7762 struct array_buffer *trace_buf = &tr->array_buffer;
7763 int cpu = tracing_get_cpu(inode);
7764 struct trace_seq *s;
7765 unsigned long cnt;
7766 unsigned long long t;
7767 unsigned long usec_rem;
7768
7769 s = kmalloc(sizeof(*s), GFP_KERNEL);
7770 if (!s)
7771 return -ENOMEM;
7772
7773 trace_seq_init(s);
7774
7775 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
7776 trace_seq_printf(s, "entries: %ld\n", cnt);
7777
7778 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
7779 trace_seq_printf(s, "overrun: %ld\n", cnt);
7780
7781 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
7782 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
7783
7784 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
7785 trace_seq_printf(s, "bytes: %ld\n", cnt);
7786
7787 if (trace_clocks[tr->clock_id].in_ns) {
7788 /* local or global for trace_clock */
7789 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
7790 usec_rem = do_div(t, USEC_PER_SEC);
7791 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
7792 t, usec_rem);
7793
7794 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
7795 usec_rem = do_div(t, USEC_PER_SEC);
7796 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
7797 } else {
7798 /* counter or tsc mode for trace_clock */
7799 trace_seq_printf(s, "oldest event ts: %llu\n",
7800 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
7801
7802 trace_seq_printf(s, "now ts: %llu\n",
7803 ring_buffer_time_stamp(trace_buf->buffer, cpu));
7804 }
7805
7806 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
7807 trace_seq_printf(s, "dropped events: %ld\n", cnt);
7808
7809 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
7810 trace_seq_printf(s, "read events: %ld\n", cnt);
7811
7812 count = simple_read_from_buffer(ubuf, count, ppos,
7813 s->buffer, trace_seq_used(s));
7814
7815 kfree(s);
7816
7817 return count;
7818}
7819
7820static const struct file_operations tracing_stats_fops = {
7821 .open = tracing_open_generic_tr,
7822 .read = tracing_stats_read,
7823 .llseek = generic_file_llseek,
7824 .release = tracing_release_generic_tr,
7825};
7826
7827#ifdef CONFIG_DYNAMIC_FTRACE
7828
7829static ssize_t
7830tracing_read_dyn_info(struct file *filp, char __user *ubuf,
7831 size_t cnt, loff_t *ppos)
7832{
7833 ssize_t ret;
7834 char *buf;
7835 int r;
7836
7837 /* 256 should be plenty to hold the amount needed */
7838 buf = kmalloc(256, GFP_KERNEL);
7839 if (!buf)
7840 return -ENOMEM;
7841
7842 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
7843 ftrace_update_tot_cnt,
7844 ftrace_number_of_pages,
7845 ftrace_number_of_groups);
7846
7847 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7848 kfree(buf);
7849 return ret;
7850}
7851
7852static const struct file_operations tracing_dyn_info_fops = {
7853 .open = tracing_open_generic,
7854 .read = tracing_read_dyn_info,
7855 .llseek = generic_file_llseek,
7856};
7857#endif /* CONFIG_DYNAMIC_FTRACE */
7858
7859#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
7860static void
7861ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
7862 struct trace_array *tr, struct ftrace_probe_ops *ops,
7863 void *data)
7864{
7865 tracing_snapshot_instance(tr);
7866}
7867
7868static void
7869ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
7870 struct trace_array *tr, struct ftrace_probe_ops *ops,
7871 void *data)
7872{
7873 struct ftrace_func_mapper *mapper = data;
7874 long *count = NULL;
7875
7876 if (mapper)
7877 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7878
7879 if (count) {
7880
7881 if (*count <= 0)
7882 return;
7883
7884 (*count)--;
7885 }
7886
7887 tracing_snapshot_instance(tr);
7888}
7889
7890static int
7891ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
7892 struct ftrace_probe_ops *ops, void *data)
7893{
7894 struct ftrace_func_mapper *mapper = data;
7895 long *count = NULL;
7896
7897 seq_printf(m, "%ps:", (void *)ip);
7898
7899 seq_puts(m, "snapshot");
7900
7901 if (mapper)
7902 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7903
7904 if (count)
7905 seq_printf(m, ":count=%ld\n", *count);
7906 else
7907 seq_puts(m, ":unlimited\n");
7908
7909 return 0;
7910}
7911
7912static int
7913ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
7914 unsigned long ip, void *init_data, void **data)
7915{
7916 struct ftrace_func_mapper *mapper = *data;
7917
7918 if (!mapper) {
7919 mapper = allocate_ftrace_func_mapper();
7920 if (!mapper)
7921 return -ENOMEM;
7922 *data = mapper;
7923 }
7924
7925 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
7926}
7927
7928static void
7929ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
7930 unsigned long ip, void *data)
7931{
7932 struct ftrace_func_mapper *mapper = data;
7933
7934 if (!ip) {
7935 if (!mapper)
7936 return;
7937 free_ftrace_func_mapper(mapper, NULL);
7938 return;
7939 }
7940
7941 ftrace_func_mapper_remove_ip(mapper, ip);
7942}
7943
7944static struct ftrace_probe_ops snapshot_probe_ops = {
7945 .func = ftrace_snapshot,
7946 .print = ftrace_snapshot_print,
7947};
7948
7949static struct ftrace_probe_ops snapshot_count_probe_ops = {
7950 .func = ftrace_count_snapshot,
7951 .print = ftrace_snapshot_print,
7952 .init = ftrace_snapshot_init,
7953 .free = ftrace_snapshot_free,
7954};
7955
7956static int
7957ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
7958 char *glob, char *cmd, char *param, int enable)
7959{
7960 struct ftrace_probe_ops *ops;
7961 void *count = (void *)-1;
7962 char *number;
7963 int ret;
7964
7965 if (!tr)
7966 return -ENODEV;
7967
7968 /* hash funcs only work with set_ftrace_filter */
7969 if (!enable)
7970 return -EINVAL;
7971
7972 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
7973
7974 if (glob[0] == '!')
7975 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
7976
7977 if (!param)
7978 goto out_reg;
7979
7980 number = strsep(¶m, ":");
7981
7982 if (!strlen(number))
7983 goto out_reg;
7984
7985 /*
7986 * We use the callback data field (which is a pointer)
7987 * as our counter.
7988 */
7989 ret = kstrtoul(number, 0, (unsigned long *)&count);
7990 if (ret)
7991 return ret;
7992
7993 out_reg:
7994 ret = tracing_alloc_snapshot_instance(tr);
7995 if (ret < 0)
7996 goto out;
7997
7998 ret = register_ftrace_function_probe(glob, tr, ops, count);
7999
8000 out:
8001 return ret < 0 ? ret : 0;
8002}
8003
8004static struct ftrace_func_command ftrace_snapshot_cmd = {
8005 .name = "snapshot",
8006 .func = ftrace_trace_snapshot_callback,
8007};
8008
8009static __init int register_snapshot_cmd(void)
8010{
8011 return register_ftrace_command(&ftrace_snapshot_cmd);
8012}
8013#else
8014static inline __init int register_snapshot_cmd(void) { return 0; }
8015#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8016
8017static struct dentry *tracing_get_dentry(struct trace_array *tr)
8018{
8019 if (WARN_ON(!tr->dir))
8020 return ERR_PTR(-ENODEV);
8021
8022 /* Top directory uses NULL as the parent */
8023 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8024 return NULL;
8025
8026 /* All sub buffers have a descriptor */
8027 return tr->dir;
8028}
8029
8030static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8031{
8032 struct dentry *d_tracer;
8033
8034 if (tr->percpu_dir)
8035 return tr->percpu_dir;
8036
8037 d_tracer = tracing_get_dentry(tr);
8038 if (IS_ERR(d_tracer))
8039 return NULL;
8040
8041 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8042
8043 MEM_FAIL(!tr->percpu_dir,
8044 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8045
8046 return tr->percpu_dir;
8047}
8048
8049static struct dentry *
8050trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8051 void *data, long cpu, const struct file_operations *fops)
8052{
8053 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8054
8055 if (ret) /* See tracing_get_cpu() */
8056 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8057 return ret;
8058}
8059
8060static void
8061tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8062{
8063 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8064 struct dentry *d_cpu;
8065 char cpu_dir[30]; /* 30 characters should be more than enough */
8066
8067 if (!d_percpu)
8068 return;
8069
8070 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8071 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8072 if (!d_cpu) {
8073 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8074 return;
8075 }
8076
8077 /* per cpu trace_pipe */
8078 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
8079 tr, cpu, &tracing_pipe_fops);
8080
8081 /* per cpu trace */
8082 trace_create_cpu_file("trace", 0644, d_cpu,
8083 tr, cpu, &tracing_fops);
8084
8085 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
8086 tr, cpu, &tracing_buffers_fops);
8087
8088 trace_create_cpu_file("stats", 0444, d_cpu,
8089 tr, cpu, &tracing_stats_fops);
8090
8091 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
8092 tr, cpu, &tracing_entries_fops);
8093
8094#ifdef CONFIG_TRACER_SNAPSHOT
8095 trace_create_cpu_file("snapshot", 0644, d_cpu,
8096 tr, cpu, &snapshot_fops);
8097
8098 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
8099 tr, cpu, &snapshot_raw_fops);
8100#endif
8101}
8102
8103#ifdef CONFIG_FTRACE_SELFTEST
8104/* Let selftest have access to static functions in this file */
8105#include "trace_selftest.c"
8106#endif
8107
8108static ssize_t
8109trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8110 loff_t *ppos)
8111{
8112 struct trace_option_dentry *topt = filp->private_data;
8113 char *buf;
8114
8115 if (topt->flags->val & topt->opt->bit)
8116 buf = "1\n";
8117 else
8118 buf = "0\n";
8119
8120 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8121}
8122
8123static ssize_t
8124trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8125 loff_t *ppos)
8126{
8127 struct trace_option_dentry *topt = filp->private_data;
8128 unsigned long val;
8129 int ret;
8130
8131 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8132 if (ret)
8133 return ret;
8134
8135 if (val != 0 && val != 1)
8136 return -EINVAL;
8137
8138 if (!!(topt->flags->val & topt->opt->bit) != val) {
8139 mutex_lock(&trace_types_lock);
8140 ret = __set_tracer_option(topt->tr, topt->flags,
8141 topt->opt, !val);
8142 mutex_unlock(&trace_types_lock);
8143 if (ret)
8144 return ret;
8145 }
8146
8147 *ppos += cnt;
8148
8149 return cnt;
8150}
8151
8152
8153static const struct file_operations trace_options_fops = {
8154 .open = tracing_open_generic,
8155 .read = trace_options_read,
8156 .write = trace_options_write,
8157 .llseek = generic_file_llseek,
8158};
8159
8160/*
8161 * In order to pass in both the trace_array descriptor as well as the index
8162 * to the flag that the trace option file represents, the trace_array
8163 * has a character array of trace_flags_index[], which holds the index
8164 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8165 * The address of this character array is passed to the flag option file
8166 * read/write callbacks.
8167 *
8168 * In order to extract both the index and the trace_array descriptor,
8169 * get_tr_index() uses the following algorithm.
8170 *
8171 * idx = *ptr;
8172 *
8173 * As the pointer itself contains the address of the index (remember
8174 * index[1] == 1).
8175 *
8176 * Then to get the trace_array descriptor, by subtracting that index
8177 * from the ptr, we get to the start of the index itself.
8178 *
8179 * ptr - idx == &index[0]
8180 *
8181 * Then a simple container_of() from that pointer gets us to the
8182 * trace_array descriptor.
8183 */
8184static void get_tr_index(void *data, struct trace_array **ptr,
8185 unsigned int *pindex)
8186{
8187 *pindex = *(unsigned char *)data;
8188
8189 *ptr = container_of(data - *pindex, struct trace_array,
8190 trace_flags_index);
8191}
8192
8193static ssize_t
8194trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8195 loff_t *ppos)
8196{
8197 void *tr_index = filp->private_data;
8198 struct trace_array *tr;
8199 unsigned int index;
8200 char *buf;
8201
8202 get_tr_index(tr_index, &tr, &index);
8203
8204 if (tr->trace_flags & (1 << index))
8205 buf = "1\n";
8206 else
8207 buf = "0\n";
8208
8209 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8210}
8211
8212static ssize_t
8213trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8214 loff_t *ppos)
8215{
8216 void *tr_index = filp->private_data;
8217 struct trace_array *tr;
8218 unsigned int index;
8219 unsigned long val;
8220 int ret;
8221
8222 get_tr_index(tr_index, &tr, &index);
8223
8224 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8225 if (ret)
8226 return ret;
8227
8228 if (val != 0 && val != 1)
8229 return -EINVAL;
8230
8231 mutex_lock(&event_mutex);
8232 mutex_lock(&trace_types_lock);
8233 ret = set_tracer_flag(tr, 1 << index, val);
8234 mutex_unlock(&trace_types_lock);
8235 mutex_unlock(&event_mutex);
8236
8237 if (ret < 0)
8238 return ret;
8239
8240 *ppos += cnt;
8241
8242 return cnt;
8243}
8244
8245static const struct file_operations trace_options_core_fops = {
8246 .open = tracing_open_generic,
8247 .read = trace_options_core_read,
8248 .write = trace_options_core_write,
8249 .llseek = generic_file_llseek,
8250};
8251
8252struct dentry *trace_create_file(const char *name,
8253 umode_t mode,
8254 struct dentry *parent,
8255 void *data,
8256 const struct file_operations *fops)
8257{
8258 struct dentry *ret;
8259
8260 ret = tracefs_create_file(name, mode, parent, data, fops);
8261 if (!ret)
8262 pr_warn("Could not create tracefs '%s' entry\n", name);
8263
8264 return ret;
8265}
8266
8267
8268static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8269{
8270 struct dentry *d_tracer;
8271
8272 if (tr->options)
8273 return tr->options;
8274
8275 d_tracer = tracing_get_dentry(tr);
8276 if (IS_ERR(d_tracer))
8277 return NULL;
8278
8279 tr->options = tracefs_create_dir("options", d_tracer);
8280 if (!tr->options) {
8281 pr_warn("Could not create tracefs directory 'options'\n");
8282 return NULL;
8283 }
8284
8285 return tr->options;
8286}
8287
8288static void
8289create_trace_option_file(struct trace_array *tr,
8290 struct trace_option_dentry *topt,
8291 struct tracer_flags *flags,
8292 struct tracer_opt *opt)
8293{
8294 struct dentry *t_options;
8295
8296 t_options = trace_options_init_dentry(tr);
8297 if (!t_options)
8298 return;
8299
8300 topt->flags = flags;
8301 topt->opt = opt;
8302 topt->tr = tr;
8303
8304 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
8305 &trace_options_fops);
8306
8307}
8308
8309static void
8310create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8311{
8312 struct trace_option_dentry *topts;
8313 struct trace_options *tr_topts;
8314 struct tracer_flags *flags;
8315 struct tracer_opt *opts;
8316 int cnt;
8317 int i;
8318
8319 if (!tracer)
8320 return;
8321
8322 flags = tracer->flags;
8323
8324 if (!flags || !flags->opts)
8325 return;
8326
8327 /*
8328 * If this is an instance, only create flags for tracers
8329 * the instance may have.
8330 */
8331 if (!trace_ok_for_array(tracer, tr))
8332 return;
8333
8334 for (i = 0; i < tr->nr_topts; i++) {
8335 /* Make sure there's no duplicate flags. */
8336 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8337 return;
8338 }
8339
8340 opts = flags->opts;
8341
8342 for (cnt = 0; opts[cnt].name; cnt++)
8343 ;
8344
8345 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8346 if (!topts)
8347 return;
8348
8349 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8350 GFP_KERNEL);
8351 if (!tr_topts) {
8352 kfree(topts);
8353 return;
8354 }
8355
8356 tr->topts = tr_topts;
8357 tr->topts[tr->nr_topts].tracer = tracer;
8358 tr->topts[tr->nr_topts].topts = topts;
8359 tr->nr_topts++;
8360
8361 for (cnt = 0; opts[cnt].name; cnt++) {
8362 create_trace_option_file(tr, &topts[cnt], flags,
8363 &opts[cnt]);
8364 MEM_FAIL(topts[cnt].entry == NULL,
8365 "Failed to create trace option: %s",
8366 opts[cnt].name);
8367 }
8368}
8369
8370static struct dentry *
8371create_trace_option_core_file(struct trace_array *tr,
8372 const char *option, long index)
8373{
8374 struct dentry *t_options;
8375
8376 t_options = trace_options_init_dentry(tr);
8377 if (!t_options)
8378 return NULL;
8379
8380 return trace_create_file(option, 0644, t_options,
8381 (void *)&tr->trace_flags_index[index],
8382 &trace_options_core_fops);
8383}
8384
8385static void create_trace_options_dir(struct trace_array *tr)
8386{
8387 struct dentry *t_options;
8388 bool top_level = tr == &global_trace;
8389 int i;
8390
8391 t_options = trace_options_init_dentry(tr);
8392 if (!t_options)
8393 return;
8394
8395 for (i = 0; trace_options[i]; i++) {
8396 if (top_level ||
8397 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8398 create_trace_option_core_file(tr, trace_options[i], i);
8399 }
8400}
8401
8402static ssize_t
8403rb_simple_read(struct file *filp, char __user *ubuf,
8404 size_t cnt, loff_t *ppos)
8405{
8406 struct trace_array *tr = filp->private_data;
8407 char buf[64];
8408 int r;
8409
8410 r = tracer_tracing_is_on(tr);
8411 r = sprintf(buf, "%d\n", r);
8412
8413 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8414}
8415
8416static ssize_t
8417rb_simple_write(struct file *filp, const char __user *ubuf,
8418 size_t cnt, loff_t *ppos)
8419{
8420 struct trace_array *tr = filp->private_data;
8421 struct trace_buffer *buffer = tr->array_buffer.buffer;
8422 unsigned long val;
8423 int ret;
8424
8425 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8426 if (ret)
8427 return ret;
8428
8429 if (buffer) {
8430 mutex_lock(&trace_types_lock);
8431 if (!!val == tracer_tracing_is_on(tr)) {
8432 val = 0; /* do nothing */
8433 } else if (val) {
8434 tracer_tracing_on(tr);
8435 if (tr->current_trace->start)
8436 tr->current_trace->start(tr);
8437 } else {
8438 tracer_tracing_off(tr);
8439 if (tr->current_trace->stop)
8440 tr->current_trace->stop(tr);
8441 }
8442 mutex_unlock(&trace_types_lock);
8443 }
8444
8445 (*ppos)++;
8446
8447 return cnt;
8448}
8449
8450static const struct file_operations rb_simple_fops = {
8451 .open = tracing_open_generic_tr,
8452 .read = rb_simple_read,
8453 .write = rb_simple_write,
8454 .release = tracing_release_generic_tr,
8455 .llseek = default_llseek,
8456};
8457
8458static ssize_t
8459buffer_percent_read(struct file *filp, char __user *ubuf,
8460 size_t cnt, loff_t *ppos)
8461{
8462 struct trace_array *tr = filp->private_data;
8463 char buf[64];
8464 int r;
8465
8466 r = tr->buffer_percent;
8467 r = sprintf(buf, "%d\n", r);
8468
8469 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8470}
8471
8472static ssize_t
8473buffer_percent_write(struct file *filp, const char __user *ubuf,
8474 size_t cnt, loff_t *ppos)
8475{
8476 struct trace_array *tr = filp->private_data;
8477 unsigned long val;
8478 int ret;
8479
8480 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8481 if (ret)
8482 return ret;
8483
8484 if (val > 100)
8485 return -EINVAL;
8486
8487 if (!val)
8488 val = 1;
8489
8490 tr->buffer_percent = val;
8491
8492 (*ppos)++;
8493
8494 return cnt;
8495}
8496
8497static const struct file_operations buffer_percent_fops = {
8498 .open = tracing_open_generic_tr,
8499 .read = buffer_percent_read,
8500 .write = buffer_percent_write,
8501 .release = tracing_release_generic_tr,
8502 .llseek = default_llseek,
8503};
8504
8505static struct dentry *trace_instance_dir;
8506
8507static void
8508init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
8509
8510static int
8511allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
8512{
8513 enum ring_buffer_flags rb_flags;
8514
8515 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
8516
8517 buf->tr = tr;
8518
8519 buf->buffer = ring_buffer_alloc(size, rb_flags);
8520 if (!buf->buffer)
8521 return -ENOMEM;
8522
8523 buf->data = alloc_percpu(struct trace_array_cpu);
8524 if (!buf->data) {
8525 ring_buffer_free(buf->buffer);
8526 buf->buffer = NULL;
8527 return -ENOMEM;
8528 }
8529
8530 /* Allocate the first page for all buffers */
8531 set_buffer_entries(&tr->array_buffer,
8532 ring_buffer_size(tr->array_buffer.buffer, 0));
8533
8534 return 0;
8535}
8536
8537static int allocate_trace_buffers(struct trace_array *tr, int size)
8538{
8539 int ret;
8540
8541 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
8542 if (ret)
8543 return ret;
8544
8545#ifdef CONFIG_TRACER_MAX_TRACE
8546 ret = allocate_trace_buffer(tr, &tr->max_buffer,
8547 allocate_snapshot ? size : 1);
8548 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
8549 ring_buffer_free(tr->array_buffer.buffer);
8550 tr->array_buffer.buffer = NULL;
8551 free_percpu(tr->array_buffer.data);
8552 tr->array_buffer.data = NULL;
8553 return -ENOMEM;
8554 }
8555 tr->allocated_snapshot = allocate_snapshot;
8556
8557 /*
8558 * Only the top level trace array gets its snapshot allocated
8559 * from the kernel command line.
8560 */
8561 allocate_snapshot = false;
8562#endif
8563
8564 return 0;
8565}
8566
8567static void free_trace_buffer(struct array_buffer *buf)
8568{
8569 if (buf->buffer) {
8570 ring_buffer_free(buf->buffer);
8571 buf->buffer = NULL;
8572 free_percpu(buf->data);
8573 buf->data = NULL;
8574 }
8575}
8576
8577static void free_trace_buffers(struct trace_array *tr)
8578{
8579 if (!tr)
8580 return;
8581
8582 free_trace_buffer(&tr->array_buffer);
8583
8584#ifdef CONFIG_TRACER_MAX_TRACE
8585 free_trace_buffer(&tr->max_buffer);
8586#endif
8587}
8588
8589static void init_trace_flags_index(struct trace_array *tr)
8590{
8591 int i;
8592
8593 /* Used by the trace options files */
8594 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
8595 tr->trace_flags_index[i] = i;
8596}
8597
8598static void __update_tracer_options(struct trace_array *tr)
8599{
8600 struct tracer *t;
8601
8602 for (t = trace_types; t; t = t->next)
8603 add_tracer_options(tr, t);
8604}
8605
8606static void update_tracer_options(struct trace_array *tr)
8607{
8608 mutex_lock(&trace_types_lock);
8609 __update_tracer_options(tr);
8610 mutex_unlock(&trace_types_lock);
8611}
8612
8613/* Must have trace_types_lock held */
8614struct trace_array *trace_array_find(const char *instance)
8615{
8616 struct trace_array *tr, *found = NULL;
8617
8618 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8619 if (tr->name && strcmp(tr->name, instance) == 0) {
8620 found = tr;
8621 break;
8622 }
8623 }
8624
8625 return found;
8626}
8627
8628struct trace_array *trace_array_find_get(const char *instance)
8629{
8630 struct trace_array *tr;
8631
8632 mutex_lock(&trace_types_lock);
8633 tr = trace_array_find(instance);
8634 if (tr)
8635 tr->ref++;
8636 mutex_unlock(&trace_types_lock);
8637
8638 return tr;
8639}
8640
8641static struct trace_array *trace_array_create(const char *name)
8642{
8643 struct trace_array *tr;
8644 int ret;
8645
8646 ret = -ENOMEM;
8647 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
8648 if (!tr)
8649 return ERR_PTR(ret);
8650
8651 tr->name = kstrdup(name, GFP_KERNEL);
8652 if (!tr->name)
8653 goto out_free_tr;
8654
8655 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
8656 goto out_free_tr;
8657
8658 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
8659
8660 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
8661
8662 raw_spin_lock_init(&tr->start_lock);
8663
8664 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
8665
8666 tr->current_trace = &nop_trace;
8667
8668 INIT_LIST_HEAD(&tr->systems);
8669 INIT_LIST_HEAD(&tr->events);
8670 INIT_LIST_HEAD(&tr->hist_vars);
8671 INIT_LIST_HEAD(&tr->err_log);
8672
8673 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
8674 goto out_free_tr;
8675
8676 tr->dir = tracefs_create_dir(name, trace_instance_dir);
8677 if (!tr->dir)
8678 goto out_free_tr;
8679
8680 ret = event_trace_add_tracer(tr->dir, tr);
8681 if (ret) {
8682 tracefs_remove(tr->dir);
8683 goto out_free_tr;
8684 }
8685
8686 ftrace_init_trace_array(tr);
8687
8688 init_tracer_tracefs(tr, tr->dir);
8689 init_trace_flags_index(tr);
8690 __update_tracer_options(tr);
8691
8692 list_add(&tr->list, &ftrace_trace_arrays);
8693
8694 tr->ref++;
8695
8696
8697 return tr;
8698
8699 out_free_tr:
8700 free_trace_buffers(tr);
8701 free_cpumask_var(tr->tracing_cpumask);
8702 kfree(tr->name);
8703 kfree(tr);
8704
8705 return ERR_PTR(ret);
8706}
8707
8708static int instance_mkdir(const char *name)
8709{
8710 struct trace_array *tr;
8711 int ret;
8712
8713 mutex_lock(&event_mutex);
8714 mutex_lock(&trace_types_lock);
8715
8716 ret = -EEXIST;
8717 if (trace_array_find(name))
8718 goto out_unlock;
8719
8720 tr = trace_array_create(name);
8721
8722 ret = PTR_ERR_OR_ZERO(tr);
8723
8724out_unlock:
8725 mutex_unlock(&trace_types_lock);
8726 mutex_unlock(&event_mutex);
8727 return ret;
8728}
8729
8730/**
8731 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
8732 * @name: The name of the trace array to be looked up/created.
8733 *
8734 * Returns pointer to trace array with given name.
8735 * NULL, if it cannot be created.
8736 *
8737 * NOTE: This function increments the reference counter associated with the
8738 * trace array returned. This makes sure it cannot be freed while in use.
8739 * Use trace_array_put() once the trace array is no longer needed.
8740 * If the trace_array is to be freed, trace_array_destroy() needs to
8741 * be called after the trace_array_put(), or simply let user space delete
8742 * it from the tracefs instances directory. But until the
8743 * trace_array_put() is called, user space can not delete it.
8744 *
8745 */
8746struct trace_array *trace_array_get_by_name(const char *name)
8747{
8748 struct trace_array *tr;
8749
8750 mutex_lock(&event_mutex);
8751 mutex_lock(&trace_types_lock);
8752
8753 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8754 if (tr->name && strcmp(tr->name, name) == 0)
8755 goto out_unlock;
8756 }
8757
8758 tr = trace_array_create(name);
8759
8760 if (IS_ERR(tr))
8761 tr = NULL;
8762out_unlock:
8763 if (tr)
8764 tr->ref++;
8765
8766 mutex_unlock(&trace_types_lock);
8767 mutex_unlock(&event_mutex);
8768 return tr;
8769}
8770EXPORT_SYMBOL_GPL(trace_array_get_by_name);
8771
8772static int __remove_instance(struct trace_array *tr)
8773{
8774 int i;
8775
8776 /* Reference counter for a newly created trace array = 1. */
8777 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
8778 return -EBUSY;
8779
8780 list_del(&tr->list);
8781
8782 /* Disable all the flags that were enabled coming in */
8783 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
8784 if ((1 << i) & ZEROED_TRACE_FLAGS)
8785 set_tracer_flag(tr, 1 << i, 0);
8786 }
8787
8788 tracing_set_nop(tr);
8789 clear_ftrace_function_probes(tr);
8790 event_trace_del_tracer(tr);
8791 ftrace_clear_pids(tr);
8792 ftrace_destroy_function_files(tr);
8793 tracefs_remove(tr->dir);
8794 free_trace_buffers(tr);
8795
8796 for (i = 0; i < tr->nr_topts; i++) {
8797 kfree(tr->topts[i].topts);
8798 }
8799 kfree(tr->topts);
8800
8801 free_cpumask_var(tr->tracing_cpumask);
8802 kfree(tr->name);
8803 kfree(tr);
8804 tr = NULL;
8805
8806 return 0;
8807}
8808
8809int trace_array_destroy(struct trace_array *this_tr)
8810{
8811 struct trace_array *tr;
8812 int ret;
8813
8814 if (!this_tr)
8815 return -EINVAL;
8816
8817 mutex_lock(&event_mutex);
8818 mutex_lock(&trace_types_lock);
8819
8820 ret = -ENODEV;
8821
8822 /* Making sure trace array exists before destroying it. */
8823 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8824 if (tr == this_tr) {
8825 ret = __remove_instance(tr);
8826 break;
8827 }
8828 }
8829
8830 mutex_unlock(&trace_types_lock);
8831 mutex_unlock(&event_mutex);
8832
8833 return ret;
8834}
8835EXPORT_SYMBOL_GPL(trace_array_destroy);
8836
8837static int instance_rmdir(const char *name)
8838{
8839 struct trace_array *tr;
8840 int ret;
8841
8842 mutex_lock(&event_mutex);
8843 mutex_lock(&trace_types_lock);
8844
8845 ret = -ENODEV;
8846 tr = trace_array_find(name);
8847 if (tr)
8848 ret = __remove_instance(tr);
8849
8850 mutex_unlock(&trace_types_lock);
8851 mutex_unlock(&event_mutex);
8852
8853 return ret;
8854}
8855
8856static __init void create_trace_instances(struct dentry *d_tracer)
8857{
8858 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
8859 instance_mkdir,
8860 instance_rmdir);
8861 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
8862 return;
8863}
8864
8865static void
8866init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
8867{
8868 struct trace_event_file *file;
8869 int cpu;
8870
8871 trace_create_file("available_tracers", 0444, d_tracer,
8872 tr, &show_traces_fops);
8873
8874 trace_create_file("current_tracer", 0644, d_tracer,
8875 tr, &set_tracer_fops);
8876
8877 trace_create_file("tracing_cpumask", 0644, d_tracer,
8878 tr, &tracing_cpumask_fops);
8879
8880 trace_create_file("trace_options", 0644, d_tracer,
8881 tr, &tracing_iter_fops);
8882
8883 trace_create_file("trace", 0644, d_tracer,
8884 tr, &tracing_fops);
8885
8886 trace_create_file("trace_pipe", 0444, d_tracer,
8887 tr, &tracing_pipe_fops);
8888
8889 trace_create_file("buffer_size_kb", 0644, d_tracer,
8890 tr, &tracing_entries_fops);
8891
8892 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
8893 tr, &tracing_total_entries_fops);
8894
8895 trace_create_file("free_buffer", 0200, d_tracer,
8896 tr, &tracing_free_buffer_fops);
8897
8898 trace_create_file("trace_marker", 0220, d_tracer,
8899 tr, &tracing_mark_fops);
8900
8901 file = __find_event_file(tr, "ftrace", "print");
8902 if (file && file->dir)
8903 trace_create_file("trigger", 0644, file->dir, file,
8904 &event_trigger_fops);
8905 tr->trace_marker_file = file;
8906
8907 trace_create_file("trace_marker_raw", 0220, d_tracer,
8908 tr, &tracing_mark_raw_fops);
8909
8910 trace_create_file("trace_clock", 0644, d_tracer, tr,
8911 &trace_clock_fops);
8912
8913 trace_create_file("tracing_on", 0644, d_tracer,
8914 tr, &rb_simple_fops);
8915
8916 trace_create_file("timestamp_mode", 0444, d_tracer, tr,
8917 &trace_time_stamp_mode_fops);
8918
8919 tr->buffer_percent = 50;
8920
8921 trace_create_file("buffer_percent", 0444, d_tracer,
8922 tr, &buffer_percent_fops);
8923
8924 create_trace_options_dir(tr);
8925
8926#if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
8927 trace_create_maxlat_file(tr, d_tracer);
8928#endif
8929
8930 if (ftrace_create_function_files(tr, d_tracer))
8931 MEM_FAIL(1, "Could not allocate function filter files");
8932
8933#ifdef CONFIG_TRACER_SNAPSHOT
8934 trace_create_file("snapshot", 0644, d_tracer,
8935 tr, &snapshot_fops);
8936#endif
8937
8938 trace_create_file("error_log", 0644, d_tracer,
8939 tr, &tracing_err_log_fops);
8940
8941 for_each_tracing_cpu(cpu)
8942 tracing_init_tracefs_percpu(tr, cpu);
8943
8944 ftrace_init_tracefs(tr, d_tracer);
8945}
8946
8947static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
8948{
8949 struct vfsmount *mnt;
8950 struct file_system_type *type;
8951
8952 /*
8953 * To maintain backward compatibility for tools that mount
8954 * debugfs to get to the tracing facility, tracefs is automatically
8955 * mounted to the debugfs/tracing directory.
8956 */
8957 type = get_fs_type("tracefs");
8958 if (!type)
8959 return NULL;
8960 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
8961 put_filesystem(type);
8962 if (IS_ERR(mnt))
8963 return NULL;
8964 mntget(mnt);
8965
8966 return mnt;
8967}
8968
8969/**
8970 * tracing_init_dentry - initialize top level trace array
8971 *
8972 * This is called when creating files or directories in the tracing
8973 * directory. It is called via fs_initcall() by any of the boot up code
8974 * and expects to return the dentry of the top level tracing directory.
8975 */
8976struct dentry *tracing_init_dentry(void)
8977{
8978 struct trace_array *tr = &global_trace;
8979
8980 if (security_locked_down(LOCKDOWN_TRACEFS)) {
8981 pr_warn("Tracing disabled due to lockdown\n");
8982 return ERR_PTR(-EPERM);
8983 }
8984
8985 /* The top level trace array uses NULL as parent */
8986 if (tr->dir)
8987 return NULL;
8988
8989 if (WARN_ON(!tracefs_initialized()))
8990 return ERR_PTR(-ENODEV);
8991
8992 /*
8993 * As there may still be users that expect the tracing
8994 * files to exist in debugfs/tracing, we must automount
8995 * the tracefs file system there, so older tools still
8996 * work with the newer kerenl.
8997 */
8998 tr->dir = debugfs_create_automount("tracing", NULL,
8999 trace_automount, NULL);
9000
9001 return NULL;
9002}
9003
9004extern struct trace_eval_map *__start_ftrace_eval_maps[];
9005extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9006
9007static void __init trace_eval_init(void)
9008{
9009 int len;
9010
9011 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9012 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9013}
9014
9015#ifdef CONFIG_MODULES
9016static void trace_module_add_evals(struct module *mod)
9017{
9018 if (!mod->num_trace_evals)
9019 return;
9020
9021 /*
9022 * Modules with bad taint do not have events created, do
9023 * not bother with enums either.
9024 */
9025 if (trace_module_has_bad_taint(mod))
9026 return;
9027
9028 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9029}
9030
9031#ifdef CONFIG_TRACE_EVAL_MAP_FILE
9032static void trace_module_remove_evals(struct module *mod)
9033{
9034 union trace_eval_map_item *map;
9035 union trace_eval_map_item **last = &trace_eval_maps;
9036
9037 if (!mod->num_trace_evals)
9038 return;
9039
9040 mutex_lock(&trace_eval_mutex);
9041
9042 map = trace_eval_maps;
9043
9044 while (map) {
9045 if (map->head.mod == mod)
9046 break;
9047 map = trace_eval_jmp_to_tail(map);
9048 last = &map->tail.next;
9049 map = map->tail.next;
9050 }
9051 if (!map)
9052 goto out;
9053
9054 *last = trace_eval_jmp_to_tail(map)->tail.next;
9055 kfree(map);
9056 out:
9057 mutex_unlock(&trace_eval_mutex);
9058}
9059#else
9060static inline void trace_module_remove_evals(struct module *mod) { }
9061#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9062
9063static int trace_module_notify(struct notifier_block *self,
9064 unsigned long val, void *data)
9065{
9066 struct module *mod = data;
9067
9068 switch (val) {
9069 case MODULE_STATE_COMING:
9070 trace_module_add_evals(mod);
9071 break;
9072 case MODULE_STATE_GOING:
9073 trace_module_remove_evals(mod);
9074 break;
9075 }
9076
9077 return 0;
9078}
9079
9080static struct notifier_block trace_module_nb = {
9081 .notifier_call = trace_module_notify,
9082 .priority = 0,
9083};
9084#endif /* CONFIG_MODULES */
9085
9086static __init int tracer_init_tracefs(void)
9087{
9088 struct dentry *d_tracer;
9089
9090 trace_access_lock_init();
9091
9092 d_tracer = tracing_init_dentry();
9093 if (IS_ERR(d_tracer))
9094 return 0;
9095
9096 event_trace_init();
9097
9098 init_tracer_tracefs(&global_trace, d_tracer);
9099 ftrace_init_tracefs_toplevel(&global_trace, d_tracer);
9100
9101 trace_create_file("tracing_thresh", 0644, d_tracer,
9102 &global_trace, &tracing_thresh_fops);
9103
9104 trace_create_file("README", 0444, d_tracer,
9105 NULL, &tracing_readme_fops);
9106
9107 trace_create_file("saved_cmdlines", 0444, d_tracer,
9108 NULL, &tracing_saved_cmdlines_fops);
9109
9110 trace_create_file("saved_cmdlines_size", 0644, d_tracer,
9111 NULL, &tracing_saved_cmdlines_size_fops);
9112
9113 trace_create_file("saved_tgids", 0444, d_tracer,
9114 NULL, &tracing_saved_tgids_fops);
9115
9116 trace_eval_init();
9117
9118 trace_create_eval_file(d_tracer);
9119
9120#ifdef CONFIG_MODULES
9121 register_module_notifier(&trace_module_nb);
9122#endif
9123
9124#ifdef CONFIG_DYNAMIC_FTRACE
9125 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
9126 NULL, &tracing_dyn_info_fops);
9127#endif
9128
9129 create_trace_instances(d_tracer);
9130
9131 update_tracer_options(&global_trace);
9132
9133 return 0;
9134}
9135
9136static int trace_panic_handler(struct notifier_block *this,
9137 unsigned long event, void *unused)
9138{
9139 if (ftrace_dump_on_oops)
9140 ftrace_dump(ftrace_dump_on_oops);
9141 return NOTIFY_OK;
9142}
9143
9144static struct notifier_block trace_panic_notifier = {
9145 .notifier_call = trace_panic_handler,
9146 .next = NULL,
9147 .priority = 150 /* priority: INT_MAX >= x >= 0 */
9148};
9149
9150static int trace_die_handler(struct notifier_block *self,
9151 unsigned long val,
9152 void *data)
9153{
9154 switch (val) {
9155 case DIE_OOPS:
9156 if (ftrace_dump_on_oops)
9157 ftrace_dump(ftrace_dump_on_oops);
9158 break;
9159 default:
9160 break;
9161 }
9162 return NOTIFY_OK;
9163}
9164
9165static struct notifier_block trace_die_notifier = {
9166 .notifier_call = trace_die_handler,
9167 .priority = 200
9168};
9169
9170/*
9171 * printk is set to max of 1024, we really don't need it that big.
9172 * Nothing should be printing 1000 characters anyway.
9173 */
9174#define TRACE_MAX_PRINT 1000
9175
9176/*
9177 * Define here KERN_TRACE so that we have one place to modify
9178 * it if we decide to change what log level the ftrace dump
9179 * should be at.
9180 */
9181#define KERN_TRACE KERN_EMERG
9182
9183void
9184trace_printk_seq(struct trace_seq *s)
9185{
9186 /* Probably should print a warning here. */
9187 if (s->seq.len >= TRACE_MAX_PRINT)
9188 s->seq.len = TRACE_MAX_PRINT;
9189
9190 /*
9191 * More paranoid code. Although the buffer size is set to
9192 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9193 * an extra layer of protection.
9194 */
9195 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9196 s->seq.len = s->seq.size - 1;
9197
9198 /* should be zero ended, but we are paranoid. */
9199 s->buffer[s->seq.len] = 0;
9200
9201 printk(KERN_TRACE "%s", s->buffer);
9202
9203 trace_seq_init(s);
9204}
9205
9206void trace_init_global_iter(struct trace_iterator *iter)
9207{
9208 iter->tr = &global_trace;
9209 iter->trace = iter->tr->current_trace;
9210 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9211 iter->array_buffer = &global_trace.array_buffer;
9212
9213 if (iter->trace && iter->trace->open)
9214 iter->trace->open(iter);
9215
9216 /* Annotate start of buffers if we had overruns */
9217 if (ring_buffer_overruns(iter->array_buffer->buffer))
9218 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9219
9220 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9221 if (trace_clocks[iter->tr->clock_id].in_ns)
9222 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9223}
9224
9225void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
9226{
9227 /* use static because iter can be a bit big for the stack */
9228 static struct trace_iterator iter;
9229 static atomic_t dump_running;
9230 struct trace_array *tr = &global_trace;
9231 unsigned int old_userobj;
9232 unsigned long flags;
9233 int cnt = 0, cpu;
9234
9235 /* Only allow one dump user at a time. */
9236 if (atomic_inc_return(&dump_running) != 1) {
9237 atomic_dec(&dump_running);
9238 return;
9239 }
9240
9241 /*
9242 * Always turn off tracing when we dump.
9243 * We don't need to show trace output of what happens
9244 * between multiple crashes.
9245 *
9246 * If the user does a sysrq-z, then they can re-enable
9247 * tracing with echo 1 > tracing_on.
9248 */
9249 tracing_off();
9250
9251 local_irq_save(flags);
9252 printk_nmi_direct_enter();
9253
9254 /* Simulate the iterator */
9255 trace_init_global_iter(&iter);
9256 /* Can not use kmalloc for iter.temp */
9257 iter.temp = static_temp_buf;
9258 iter.temp_size = STATIC_TEMP_BUF_SIZE;
9259
9260 for_each_tracing_cpu(cpu) {
9261 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9262 }
9263
9264 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
9265
9266 /* don't look at user memory in panic mode */
9267 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
9268
9269 switch (oops_dump_mode) {
9270 case DUMP_ALL:
9271 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9272 break;
9273 case DUMP_ORIG:
9274 iter.cpu_file = raw_smp_processor_id();
9275 break;
9276 case DUMP_NONE:
9277 goto out_enable;
9278 default:
9279 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
9280 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9281 }
9282
9283 printk(KERN_TRACE "Dumping ftrace buffer:\n");
9284
9285 /* Did function tracer already get disabled? */
9286 if (ftrace_is_dead()) {
9287 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
9288 printk("# MAY BE MISSING FUNCTION EVENTS\n");
9289 }
9290
9291 /*
9292 * We need to stop all tracing on all CPUS to read the
9293 * the next buffer. This is a bit expensive, but is
9294 * not done often. We fill all what we can read,
9295 * and then release the locks again.
9296 */
9297
9298 while (!trace_empty(&iter)) {
9299
9300 if (!cnt)
9301 printk(KERN_TRACE "---------------------------------\n");
9302
9303 cnt++;
9304
9305 trace_iterator_reset(&iter);
9306 iter.iter_flags |= TRACE_FILE_LAT_FMT;
9307
9308 if (trace_find_next_entry_inc(&iter) != NULL) {
9309 int ret;
9310
9311 ret = print_trace_line(&iter);
9312 if (ret != TRACE_TYPE_NO_CONSUME)
9313 trace_consume(&iter);
9314 }
9315 touch_nmi_watchdog();
9316
9317 trace_printk_seq(&iter.seq);
9318 }
9319
9320 if (!cnt)
9321 printk(KERN_TRACE " (ftrace buffer empty)\n");
9322 else
9323 printk(KERN_TRACE "---------------------------------\n");
9324
9325 out_enable:
9326 tr->trace_flags |= old_userobj;
9327
9328 for_each_tracing_cpu(cpu) {
9329 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9330 }
9331 atomic_dec(&dump_running);
9332 printk_nmi_direct_exit();
9333 local_irq_restore(flags);
9334}
9335EXPORT_SYMBOL_GPL(ftrace_dump);
9336
9337int trace_run_command(const char *buf, int (*createfn)(int, char **))
9338{
9339 char **argv;
9340 int argc, ret;
9341
9342 argc = 0;
9343 ret = 0;
9344 argv = argv_split(GFP_KERNEL, buf, &argc);
9345 if (!argv)
9346 return -ENOMEM;
9347
9348 if (argc)
9349 ret = createfn(argc, argv);
9350
9351 argv_free(argv);
9352
9353 return ret;
9354}
9355
9356#define WRITE_BUFSIZE 4096
9357
9358ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
9359 size_t count, loff_t *ppos,
9360 int (*createfn)(int, char **))
9361{
9362 char *kbuf, *buf, *tmp;
9363 int ret = 0;
9364 size_t done = 0;
9365 size_t size;
9366
9367 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
9368 if (!kbuf)
9369 return -ENOMEM;
9370
9371 while (done < count) {
9372 size = count - done;
9373
9374 if (size >= WRITE_BUFSIZE)
9375 size = WRITE_BUFSIZE - 1;
9376
9377 if (copy_from_user(kbuf, buffer + done, size)) {
9378 ret = -EFAULT;
9379 goto out;
9380 }
9381 kbuf[size] = '\0';
9382 buf = kbuf;
9383 do {
9384 tmp = strchr(buf, '\n');
9385 if (tmp) {
9386 *tmp = '\0';
9387 size = tmp - buf + 1;
9388 } else {
9389 size = strlen(buf);
9390 if (done + size < count) {
9391 if (buf != kbuf)
9392 break;
9393 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
9394 pr_warn("Line length is too long: Should be less than %d\n",
9395 WRITE_BUFSIZE - 2);
9396 ret = -EINVAL;
9397 goto out;
9398 }
9399 }
9400 done += size;
9401
9402 /* Remove comments */
9403 tmp = strchr(buf, '#');
9404
9405 if (tmp)
9406 *tmp = '\0';
9407
9408 ret = trace_run_command(buf, createfn);
9409 if (ret)
9410 goto out;
9411 buf += size;
9412
9413 } while (done < count);
9414 }
9415 ret = done;
9416
9417out:
9418 kfree(kbuf);
9419
9420 return ret;
9421}
9422
9423__init static int tracer_alloc_buffers(void)
9424{
9425 int ring_buf_size;
9426 int ret = -ENOMEM;
9427
9428
9429 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9430 pr_warn("Tracing disabled due to lockdown\n");
9431 return -EPERM;
9432 }
9433
9434 /*
9435 * Make sure we don't accidently add more trace options
9436 * than we have bits for.
9437 */
9438 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
9439
9440 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
9441 goto out;
9442
9443 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
9444 goto out_free_buffer_mask;
9445
9446 /* Only allocate trace_printk buffers if a trace_printk exists */
9447 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
9448 /* Must be called before global_trace.buffer is allocated */
9449 trace_printk_init_buffers();
9450
9451 /* To save memory, keep the ring buffer size to its minimum */
9452 if (ring_buffer_expanded)
9453 ring_buf_size = trace_buf_size;
9454 else
9455 ring_buf_size = 1;
9456
9457 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
9458 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
9459
9460 raw_spin_lock_init(&global_trace.start_lock);
9461
9462 /*
9463 * The prepare callbacks allocates some memory for the ring buffer. We
9464 * don't free the buffer if the if the CPU goes down. If we were to free
9465 * the buffer, then the user would lose any trace that was in the
9466 * buffer. The memory will be removed once the "instance" is removed.
9467 */
9468 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
9469 "trace/RB:preapre", trace_rb_cpu_prepare,
9470 NULL);
9471 if (ret < 0)
9472 goto out_free_cpumask;
9473 /* Used for event triggers */
9474 ret = -ENOMEM;
9475 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
9476 if (!temp_buffer)
9477 goto out_rm_hp_state;
9478
9479 if (trace_create_savedcmd() < 0)
9480 goto out_free_temp_buffer;
9481
9482 /* TODO: make the number of buffers hot pluggable with CPUS */
9483 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
9484 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
9485 goto out_free_savedcmd;
9486 }
9487
9488 if (global_trace.buffer_disabled)
9489 tracing_off();
9490
9491 if (trace_boot_clock) {
9492 ret = tracing_set_clock(&global_trace, trace_boot_clock);
9493 if (ret < 0)
9494 pr_warn("Trace clock %s not defined, going back to default\n",
9495 trace_boot_clock);
9496 }
9497
9498 /*
9499 * register_tracer() might reference current_trace, so it
9500 * needs to be set before we register anything. This is
9501 * just a bootstrap of current_trace anyway.
9502 */
9503 global_trace.current_trace = &nop_trace;
9504
9505 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9506
9507 ftrace_init_global_array_ops(&global_trace);
9508
9509 init_trace_flags_index(&global_trace);
9510
9511 register_tracer(&nop_trace);
9512
9513 /* Function tracing may start here (via kernel command line) */
9514 init_function_trace();
9515
9516 /* All seems OK, enable tracing */
9517 tracing_disabled = 0;
9518
9519 atomic_notifier_chain_register(&panic_notifier_list,
9520 &trace_panic_notifier);
9521
9522 register_die_notifier(&trace_die_notifier);
9523
9524 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
9525
9526 INIT_LIST_HEAD(&global_trace.systems);
9527 INIT_LIST_HEAD(&global_trace.events);
9528 INIT_LIST_HEAD(&global_trace.hist_vars);
9529 INIT_LIST_HEAD(&global_trace.err_log);
9530 list_add(&global_trace.list, &ftrace_trace_arrays);
9531
9532 apply_trace_boot_options();
9533
9534 register_snapshot_cmd();
9535
9536 return 0;
9537
9538out_free_savedcmd:
9539 free_saved_cmdlines_buffer(savedcmd);
9540out_free_temp_buffer:
9541 ring_buffer_free(temp_buffer);
9542out_rm_hp_state:
9543 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
9544out_free_cpumask:
9545 free_cpumask_var(global_trace.tracing_cpumask);
9546out_free_buffer_mask:
9547 free_cpumask_var(tracing_buffer_mask);
9548out:
9549 return ret;
9550}
9551
9552void __init early_trace_init(void)
9553{
9554 if (tracepoint_printk) {
9555 tracepoint_print_iter =
9556 kmalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
9557 if (MEM_FAIL(!tracepoint_print_iter,
9558 "Failed to allocate trace iterator\n"))
9559 tracepoint_printk = 0;
9560 else
9561 static_key_enable(&tracepoint_printk_key.key);
9562 }
9563 tracer_alloc_buffers();
9564}
9565
9566void __init trace_init(void)
9567{
9568 trace_event_init();
9569}
9570
9571__init static int clear_boot_tracer(void)
9572{
9573 /*
9574 * The default tracer at boot buffer is an init section.
9575 * This function is called in lateinit. If we did not
9576 * find the boot tracer, then clear it out, to prevent
9577 * later registration from accessing the buffer that is
9578 * about to be freed.
9579 */
9580 if (!default_bootup_tracer)
9581 return 0;
9582
9583 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
9584 default_bootup_tracer);
9585 default_bootup_tracer = NULL;
9586
9587 return 0;
9588}
9589
9590fs_initcall(tracer_init_tracefs);
9591late_initcall_sync(clear_boot_tracer);
9592
9593#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
9594__init static int tracing_set_default_clock(void)
9595{
9596 /* sched_clock_stable() is determined in late_initcall */
9597 if (!trace_boot_clock && !sched_clock_stable()) {
9598 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9599 pr_warn("Can not set tracing clock due to lockdown\n");
9600 return -EPERM;
9601 }
9602
9603 printk(KERN_WARNING
9604 "Unstable clock detected, switching default tracing clock to \"global\"\n"
9605 "If you want to keep using the local clock, then add:\n"
9606 " \"trace_clock=local\"\n"
9607 "on the kernel command line\n");
9608 tracing_set_clock(&global_trace, "global");
9609 }
9610
9611 return 0;
9612}
9613late_initcall_sync(tracing_set_default_clock);
9614#endif