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