projects/linux·blob v5.18 · refs

10247 lines · 249.0 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * ring buffer based function tracer
4 *
5 * Copyright (C) 2007-2012 Steven Rostedt <[email protected]>
6 * Copyright (C) 2008 Ingo Molnar <[email protected]>
7 *
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <[email protected]>
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;
188static bool snapshot_at_boot;
189
190static int __init set_cmdline_ftrace(char *str)
191{
192 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
193 default_bootup_tracer = bootup_tracer_buf;
194 /* We are using ftrace early, expand it */
195 ring_buffer_expanded = true;
196 return 1;
197}
198__setup("ftrace=", set_cmdline_ftrace);
199
200static int __init set_ftrace_dump_on_oops(char *str)
201{
202 if (*str++ != '=' || !*str || !strcmp("1", str)) {
203 ftrace_dump_on_oops = DUMP_ALL;
204 return 1;
205 }
206
207 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
208 ftrace_dump_on_oops = DUMP_ORIG;
209 return 1;
210 }
211
212 return 0;
213}
214__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
215
216static int __init stop_trace_on_warning(char *str)
217{
218 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
219 __disable_trace_on_warning = 1;
220 return 1;
221}
222__setup("traceoff_on_warning", stop_trace_on_warning);
223
224static int __init boot_alloc_snapshot(char *str)
225{
226 allocate_snapshot = true;
227 /* We also need the main ring buffer expanded */
228 ring_buffer_expanded = true;
229 return 1;
230}
231__setup("alloc_snapshot", boot_alloc_snapshot);
232
233
234static int __init boot_snapshot(char *str)
235{
236 snapshot_at_boot = true;
237 boot_alloc_snapshot(str);
238 return 1;
239}
240__setup("ftrace_boot_snapshot", boot_snapshot);
241
242
243static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
244
245static int __init set_trace_boot_options(char *str)
246{
247 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
248 return 1;
249}
250__setup("trace_options=", set_trace_boot_options);
251
252static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
253static char *trace_boot_clock __initdata;
254
255static int __init set_trace_boot_clock(char *str)
256{
257 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
258 trace_boot_clock = trace_boot_clock_buf;
259 return 1;
260}
261__setup("trace_clock=", set_trace_boot_clock);
262
263static int __init set_tracepoint_printk(char *str)
264{
265 /* Ignore the "tp_printk_stop_on_boot" param */
266 if (*str == '_')
267 return 0;
268
269 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
270 tracepoint_printk = 1;
271 return 1;
272}
273__setup("tp_printk", set_tracepoint_printk);
274
275static int __init set_tracepoint_printk_stop(char *str)
276{
277 tracepoint_printk_stop_on_boot = true;
278 return 1;
279}
280__setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
281
282unsigned long long ns2usecs(u64 nsec)
283{
284 nsec += 500;
285 do_div(nsec, 1000);
286 return nsec;
287}
288
289static void
290trace_process_export(struct trace_export *export,
291 struct ring_buffer_event *event, int flag)
292{
293 struct trace_entry *entry;
294 unsigned int size = 0;
295
296 if (export->flags & flag) {
297 entry = ring_buffer_event_data(event);
298 size = ring_buffer_event_length(event);
299 export->write(export, entry, size);
300 }
301}
302
303static DEFINE_MUTEX(ftrace_export_lock);
304
305static struct trace_export __rcu *ftrace_exports_list __read_mostly;
306
307static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
308static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
309static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
310
311static inline void ftrace_exports_enable(struct trace_export *export)
312{
313 if (export->flags & TRACE_EXPORT_FUNCTION)
314 static_branch_inc(&trace_function_exports_enabled);
315
316 if (export->flags & TRACE_EXPORT_EVENT)
317 static_branch_inc(&trace_event_exports_enabled);
318
319 if (export->flags & TRACE_EXPORT_MARKER)
320 static_branch_inc(&trace_marker_exports_enabled);
321}
322
323static inline void ftrace_exports_disable(struct trace_export *export)
324{
325 if (export->flags & TRACE_EXPORT_FUNCTION)
326 static_branch_dec(&trace_function_exports_enabled);
327
328 if (export->flags & TRACE_EXPORT_EVENT)
329 static_branch_dec(&trace_event_exports_enabled);
330
331 if (export->flags & TRACE_EXPORT_MARKER)
332 static_branch_dec(&trace_marker_exports_enabled);
333}
334
335static void ftrace_exports(struct ring_buffer_event *event, int flag)
336{
337 struct trace_export *export;
338
339 preempt_disable_notrace();
340
341 export = rcu_dereference_raw_check(ftrace_exports_list);
342 while (export) {
343 trace_process_export(export, event, flag);
344 export = rcu_dereference_raw_check(export->next);
345 }
346
347 preempt_enable_notrace();
348}
349
350static inline void
351add_trace_export(struct trace_export **list, struct trace_export *export)
352{
353 rcu_assign_pointer(export->next, *list);
354 /*
355 * We are entering export into the list but another
356 * CPU might be walking that list. We need to make sure
357 * the export->next pointer is valid before another CPU sees
358 * the export pointer included into the list.
359 */
360 rcu_assign_pointer(*list, export);
361}
362
363static inline int
364rm_trace_export(struct trace_export **list, struct trace_export *export)
365{
366 struct trace_export **p;
367
368 for (p = list; *p != NULL; p = &(*p)->next)
369 if (*p == export)
370 break;
371
372 if (*p != export)
373 return -1;
374
375 rcu_assign_pointer(*p, (*p)->next);
376
377 return 0;
378}
379
380static inline void
381add_ftrace_export(struct trace_export **list, struct trace_export *export)
382{
383 ftrace_exports_enable(export);
384
385 add_trace_export(list, export);
386}
387
388static inline int
389rm_ftrace_export(struct trace_export **list, struct trace_export *export)
390{
391 int ret;
392
393 ret = rm_trace_export(list, export);
394 ftrace_exports_disable(export);
395
396 return ret;
397}
398
399int register_ftrace_export(struct trace_export *export)
400{
401 if (WARN_ON_ONCE(!export->write))
402 return -1;
403
404 mutex_lock(&ftrace_export_lock);
405
406 add_ftrace_export(&ftrace_exports_list, export);
407
408 mutex_unlock(&ftrace_export_lock);
409
410 return 0;
411}
412EXPORT_SYMBOL_GPL(register_ftrace_export);
413
414int unregister_ftrace_export(struct trace_export *export)
415{
416 int ret;
417
418 mutex_lock(&ftrace_export_lock);
419
420 ret = rm_ftrace_export(&ftrace_exports_list, export);
421
422 mutex_unlock(&ftrace_export_lock);
423
424 return ret;
425}
426EXPORT_SYMBOL_GPL(unregister_ftrace_export);
427
428/* trace_flags holds trace_options default values */
429#define TRACE_DEFAULT_FLAGS \
430 (FUNCTION_DEFAULT_FLAGS | \
431 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
432 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
433 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
434 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
435 TRACE_ITER_HASH_PTR)
436
437/* trace_options that are only supported by global_trace */
438#define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
439 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
440
441/* trace_flags that are default zero for instances */
442#define ZEROED_TRACE_FLAGS \
443 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
444
445/*
446 * The global_trace is the descriptor that holds the top-level tracing
447 * buffers for the live tracing.
448 */
449static struct trace_array global_trace = {
450 .trace_flags = TRACE_DEFAULT_FLAGS,
451};
452
453LIST_HEAD(ftrace_trace_arrays);
454
455int trace_array_get(struct trace_array *this_tr)
456{
457 struct trace_array *tr;
458 int ret = -ENODEV;
459
460 mutex_lock(&trace_types_lock);
461 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
462 if (tr == this_tr) {
463 tr->ref++;
464 ret = 0;
465 break;
466 }
467 }
468 mutex_unlock(&trace_types_lock);
469
470 return ret;
471}
472
473static void __trace_array_put(struct trace_array *this_tr)
474{
475 WARN_ON(!this_tr->ref);
476 this_tr->ref--;
477}
478
479/**
480 * trace_array_put - Decrement the reference counter for this trace array.
481 * @this_tr : pointer to the trace array
482 *
483 * NOTE: Use this when we no longer need the trace array returned by
484 * trace_array_get_by_name(). This ensures the trace array can be later
485 * destroyed.
486 *
487 */
488void trace_array_put(struct trace_array *this_tr)
489{
490 if (!this_tr)
491 return;
492
493 mutex_lock(&trace_types_lock);
494 __trace_array_put(this_tr);
495 mutex_unlock(&trace_types_lock);
496}
497EXPORT_SYMBOL_GPL(trace_array_put);
498
499int tracing_check_open_get_tr(struct trace_array *tr)
500{
501 int ret;
502
503 ret = security_locked_down(LOCKDOWN_TRACEFS);
504 if (ret)
505 return ret;
506
507 if (tracing_disabled)
508 return -ENODEV;
509
510 if (tr && trace_array_get(tr) < 0)
511 return -ENODEV;
512
513 return 0;
514}
515
516int call_filter_check_discard(struct trace_event_call *call, void *rec,
517 struct trace_buffer *buffer,
518 struct ring_buffer_event *event)
519{
520 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
521 !filter_match_preds(call->filter, rec)) {
522 __trace_event_discard_commit(buffer, event);
523 return 1;
524 }
525
526 return 0;
527}
528
529/**
530 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
531 * @filtered_pids: The list of pids to check
532 * @search_pid: The PID to find in @filtered_pids
533 *
534 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
535 */
536bool
537trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
538{
539 return trace_pid_list_is_set(filtered_pids, search_pid);
540}
541
542/**
543 * trace_ignore_this_task - should a task be ignored for tracing
544 * @filtered_pids: The list of pids to check
545 * @filtered_no_pids: The list of pids not to be traced
546 * @task: The task that should be ignored if not filtered
547 *
548 * Checks if @task should be traced or not from @filtered_pids.
549 * Returns true if @task should *NOT* be traced.
550 * Returns false if @task should be traced.
551 */
552bool
553trace_ignore_this_task(struct trace_pid_list *filtered_pids,
554 struct trace_pid_list *filtered_no_pids,
555 struct task_struct *task)
556{
557 /*
558 * If filtered_no_pids is not empty, and the task's pid is listed
559 * in filtered_no_pids, then return true.
560 * Otherwise, if filtered_pids is empty, that means we can
561 * trace all tasks. If it has content, then only trace pids
562 * within filtered_pids.
563 */
564
565 return (filtered_pids &&
566 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
567 (filtered_no_pids &&
568 trace_find_filtered_pid(filtered_no_pids, task->pid));
569}
570
571/**
572 * trace_filter_add_remove_task - Add or remove a task from a pid_list
573 * @pid_list: The list to modify
574 * @self: The current task for fork or NULL for exit
575 * @task: The task to add or remove
576 *
577 * If adding a task, if @self is defined, the task is only added if @self
578 * is also included in @pid_list. This happens on fork and tasks should
579 * only be added when the parent is listed. If @self is NULL, then the
580 * @task pid will be removed from the list, which would happen on exit
581 * of a task.
582 */
583void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
584 struct task_struct *self,
585 struct task_struct *task)
586{
587 if (!pid_list)
588 return;
589
590 /* For forks, we only add if the forking task is listed */
591 if (self) {
592 if (!trace_find_filtered_pid(pid_list, self->pid))
593 return;
594 }
595
596 /* "self" is set for forks, and NULL for exits */
597 if (self)
598 trace_pid_list_set(pid_list, task->pid);
599 else
600 trace_pid_list_clear(pid_list, task->pid);
601}
602
603/**
604 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
605 * @pid_list: The pid list to show
606 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
607 * @pos: The position of the file
608 *
609 * This is used by the seq_file "next" operation to iterate the pids
610 * listed in a trace_pid_list structure.
611 *
612 * Returns the pid+1 as we want to display pid of zero, but NULL would
613 * stop the iteration.
614 */
615void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
616{
617 long pid = (unsigned long)v;
618 unsigned int next;
619
620 (*pos)++;
621
622 /* pid already is +1 of the actual previous bit */
623 if (trace_pid_list_next(pid_list, pid, &next) < 0)
624 return NULL;
625
626 pid = next;
627
628 /* Return pid + 1 to allow zero to be represented */
629 return (void *)(pid + 1);
630}
631
632/**
633 * trace_pid_start - Used for seq_file to start reading pid lists
634 * @pid_list: The pid list to show
635 * @pos: The position of the file
636 *
637 * This is used by seq_file "start" operation to start the iteration
638 * of listing pids.
639 *
640 * Returns the pid+1 as we want to display pid of zero, but NULL would
641 * stop the iteration.
642 */
643void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
644{
645 unsigned long pid;
646 unsigned int first;
647 loff_t l = 0;
648
649 if (trace_pid_list_first(pid_list, &first) < 0)
650 return NULL;
651
652 pid = first;
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;
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 = trace_pid_list_alloc();
703 if (!pid_list) {
704 trace_parser_put(&parser);
705 return -ENOMEM;
706 }
707
708 if (filtered_pids) {
709 /* copy the current bits to the new max */
710 ret = trace_pid_list_first(filtered_pids, &pid);
711 while (!ret) {
712 trace_pid_list_set(pid_list, pid);
713 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
714 nr_pids++;
715 }
716 }
717
718 ret = 0;
719 while (cnt > 0) {
720
721 pos = 0;
722
723 ret = trace_get_user(&parser, ubuf, cnt, &pos);
724 if (ret < 0 || !trace_parser_loaded(&parser))
725 break;
726
727 read += ret;
728 ubuf += ret;
729 cnt -= ret;
730
731 ret = -EINVAL;
732 if (kstrtoul(parser.buffer, 0, &val))
733 break;
734
735 pid = (pid_t)val;
736
737 if (trace_pid_list_set(pid_list, pid) < 0) {
738 ret = -1;
739 break;
740 }
741 nr_pids++;
742
743 trace_parser_clear(&parser);
744 ret = 0;
745 }
746 trace_parser_put(&parser);
747
748 if (ret < 0) {
749 trace_pid_list_free(pid_list);
750 return ret;
751 }
752
753 if (!nr_pids) {
754 /* Cleared the list of pids */
755 trace_pid_list_free(pid_list);
756 read = ret;
757 pid_list = NULL;
758 }
759
760 *new_pid_list = pid_list;
761
762 return read;
763}
764
765static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
766{
767 u64 ts;
768
769 /* Early boot up does not have a buffer yet */
770 if (!buf->buffer)
771 return trace_clock_local();
772
773 ts = ring_buffer_time_stamp(buf->buffer);
774 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
775
776 return ts;
777}
778
779u64 ftrace_now(int cpu)
780{
781 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
782}
783
784/**
785 * tracing_is_enabled - Show if global_trace has been enabled
786 *
787 * Shows if the global trace has been enabled or not. It uses the
788 * mirror flag "buffer_disabled" to be used in fast paths such as for
789 * the irqsoff tracer. But it may be inaccurate due to races. If you
790 * need to know the accurate state, use tracing_is_on() which is a little
791 * slower, but accurate.
792 */
793int tracing_is_enabled(void)
794{
795 /*
796 * For quick access (irqsoff uses this in fast path), just
797 * return the mirror variable of the state of the ring buffer.
798 * It's a little racy, but we don't really care.
799 */
800 smp_rmb();
801 return !global_trace.buffer_disabled;
802}
803
804/*
805 * trace_buf_size is the size in bytes that is allocated
806 * for a buffer. Note, the number of bytes is always rounded
807 * to page size.
808 *
809 * This number is purposely set to a low number of 16384.
810 * If the dump on oops happens, it will be much appreciated
811 * to not have to wait for all that output. Anyway this can be
812 * boot time and run time configurable.
813 */
814#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
815
816static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
817
818/* trace_types holds a link list of available tracers. */
819static struct tracer *trace_types __read_mostly;
820
821/*
822 * trace_types_lock is used to protect the trace_types list.
823 */
824DEFINE_MUTEX(trace_types_lock);
825
826/*
827 * serialize the access of the ring buffer
828 *
829 * ring buffer serializes readers, but it is low level protection.
830 * The validity of the events (which returns by ring_buffer_peek() ..etc)
831 * are not protected by ring buffer.
832 *
833 * The content of events may become garbage if we allow other process consumes
834 * these events concurrently:
835 * A) the page of the consumed events may become a normal page
836 * (not reader page) in ring buffer, and this page will be rewritten
837 * by events producer.
838 * B) The page of the consumed events may become a page for splice_read,
839 * and this page will be returned to system.
840 *
841 * These primitives allow multi process access to different cpu ring buffer
842 * concurrently.
843 *
844 * These primitives don't distinguish read-only and read-consume access.
845 * Multi read-only access are also serialized.
846 */
847
848#ifdef CONFIG_SMP
849static DECLARE_RWSEM(all_cpu_access_lock);
850static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
851
852static inline void trace_access_lock(int cpu)
853{
854 if (cpu == RING_BUFFER_ALL_CPUS) {
855 /* gain it for accessing the whole ring buffer. */
856 down_write(&all_cpu_access_lock);
857 } else {
858 /* gain it for accessing a cpu ring buffer. */
859
860 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
861 down_read(&all_cpu_access_lock);
862
863 /* Secondly block other access to this @cpu ring buffer. */
864 mutex_lock(&per_cpu(cpu_access_lock, cpu));
865 }
866}
867
868static inline void trace_access_unlock(int cpu)
869{
870 if (cpu == RING_BUFFER_ALL_CPUS) {
871 up_write(&all_cpu_access_lock);
872 } else {
873 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
874 up_read(&all_cpu_access_lock);
875 }
876}
877
878static inline void trace_access_lock_init(void)
879{
880 int cpu;
881
882 for_each_possible_cpu(cpu)
883 mutex_init(&per_cpu(cpu_access_lock, cpu));
884}
885
886#else
887
888static DEFINE_MUTEX(access_lock);
889
890static inline void trace_access_lock(int cpu)
891{
892 (void)cpu;
893 mutex_lock(&access_lock);
894}
895
896static inline void trace_access_unlock(int cpu)
897{
898 (void)cpu;
899 mutex_unlock(&access_lock);
900}
901
902static inline void trace_access_lock_init(void)
903{
904}
905
906#endif
907
908#ifdef CONFIG_STACKTRACE
909static void __ftrace_trace_stack(struct trace_buffer *buffer,
910 unsigned int trace_ctx,
911 int skip, struct pt_regs *regs);
912static inline void ftrace_trace_stack(struct trace_array *tr,
913 struct trace_buffer *buffer,
914 unsigned int trace_ctx,
915 int skip, struct pt_regs *regs);
916
917#else
918static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
919 unsigned int trace_ctx,
920 int skip, struct pt_regs *regs)
921{
922}
923static inline void ftrace_trace_stack(struct trace_array *tr,
924 struct trace_buffer *buffer,
925 unsigned long trace_ctx,
926 int skip, struct pt_regs *regs)
927{
928}
929
930#endif
931
932static __always_inline void
933trace_event_setup(struct ring_buffer_event *event,
934 int type, unsigned int trace_ctx)
935{
936 struct trace_entry *ent = ring_buffer_event_data(event);
937
938 tracing_generic_entry_update(ent, type, trace_ctx);
939}
940
941static __always_inline struct ring_buffer_event *
942__trace_buffer_lock_reserve(struct trace_buffer *buffer,
943 int type,
944 unsigned long len,
945 unsigned int trace_ctx)
946{
947 struct ring_buffer_event *event;
948
949 event = ring_buffer_lock_reserve(buffer, len);
950 if (event != NULL)
951 trace_event_setup(event, type, trace_ctx);
952
953 return event;
954}
955
956void tracer_tracing_on(struct trace_array *tr)
957{
958 if (tr->array_buffer.buffer)
959 ring_buffer_record_on(tr->array_buffer.buffer);
960 /*
961 * This flag is looked at when buffers haven't been allocated
962 * yet, or by some tracers (like irqsoff), that just want to
963 * know if the ring buffer has been disabled, but it can handle
964 * races of where it gets disabled but we still do a record.
965 * As the check is in the fast path of the tracers, it is more
966 * important to be fast than accurate.
967 */
968 tr->buffer_disabled = 0;
969 /* Make the flag seen by readers */
970 smp_wmb();
971}
972
973/**
974 * tracing_on - enable tracing buffers
975 *
976 * This function enables tracing buffers that may have been
977 * disabled with tracing_off.
978 */
979void tracing_on(void)
980{
981 tracer_tracing_on(&global_trace);
982}
983EXPORT_SYMBOL_GPL(tracing_on);
984
985
986static __always_inline void
987__buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
988{
989 __this_cpu_write(trace_taskinfo_save, true);
990
991 /* If this is the temp buffer, we need to commit fully */
992 if (this_cpu_read(trace_buffered_event) == event) {
993 /* Length is in event->array[0] */
994 ring_buffer_write(buffer, event->array[0], &event->array[1]);
995 /* Release the temp buffer */
996 this_cpu_dec(trace_buffered_event_cnt);
997 /* ring_buffer_unlock_commit() enables preemption */
998 preempt_enable_notrace();
999 } else
1000 ring_buffer_unlock_commit(buffer, event);
1001}
1002
1003/**
1004 * __trace_puts - write a constant string into the trace buffer.
1005 * @ip: The address of the caller
1006 * @str: The constant string to write
1007 * @size: The size of the string.
1008 */
1009int __trace_puts(unsigned long ip, const char *str, int size)
1010{
1011 struct ring_buffer_event *event;
1012 struct trace_buffer *buffer;
1013 struct print_entry *entry;
1014 unsigned int trace_ctx;
1015 int alloc;
1016
1017 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1018 return 0;
1019
1020 if (unlikely(tracing_selftest_running || tracing_disabled))
1021 return 0;
1022
1023 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1024
1025 trace_ctx = tracing_gen_ctx();
1026 buffer = global_trace.array_buffer.buffer;
1027 ring_buffer_nest_start(buffer);
1028 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1029 trace_ctx);
1030 if (!event) {
1031 size = 0;
1032 goto out;
1033 }
1034
1035 entry = ring_buffer_event_data(event);
1036 entry->ip = ip;
1037
1038 memcpy(&entry->buf, str, size);
1039
1040 /* Add a newline if necessary */
1041 if (entry->buf[size - 1] != '\n') {
1042 entry->buf[size] = '\n';
1043 entry->buf[size + 1] = '\0';
1044 } else
1045 entry->buf[size] = '\0';
1046
1047 __buffer_unlock_commit(buffer, event);
1048 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1049 out:
1050 ring_buffer_nest_end(buffer);
1051 return size;
1052}
1053EXPORT_SYMBOL_GPL(__trace_puts);
1054
1055/**
1056 * __trace_bputs - write the pointer to a constant string into trace buffer
1057 * @ip: The address of the caller
1058 * @str: The constant string to write to the buffer to
1059 */
1060int __trace_bputs(unsigned long ip, const char *str)
1061{
1062 struct ring_buffer_event *event;
1063 struct trace_buffer *buffer;
1064 struct bputs_entry *entry;
1065 unsigned int trace_ctx;
1066 int size = sizeof(struct bputs_entry);
1067 int ret = 0;
1068
1069 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1070 return 0;
1071
1072 if (unlikely(tracing_selftest_running || tracing_disabled))
1073 return 0;
1074
1075 trace_ctx = tracing_gen_ctx();
1076 buffer = global_trace.array_buffer.buffer;
1077
1078 ring_buffer_nest_start(buffer);
1079 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1080 trace_ctx);
1081 if (!event)
1082 goto out;
1083
1084 entry = ring_buffer_event_data(event);
1085 entry->ip = ip;
1086 entry->str = str;
1087
1088 __buffer_unlock_commit(buffer, event);
1089 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1090
1091 ret = 1;
1092 out:
1093 ring_buffer_nest_end(buffer);
1094 return ret;
1095}
1096EXPORT_SYMBOL_GPL(__trace_bputs);
1097
1098#ifdef CONFIG_TRACER_SNAPSHOT
1099static void tracing_snapshot_instance_cond(struct trace_array *tr,
1100 void *cond_data)
1101{
1102 struct tracer *tracer = tr->current_trace;
1103 unsigned long flags;
1104
1105 if (in_nmi()) {
1106 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1107 internal_trace_puts("*** snapshot is being ignored ***\n");
1108 return;
1109 }
1110
1111 if (!tr->allocated_snapshot) {
1112 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
1113 internal_trace_puts("*** stopping trace here! ***\n");
1114 tracing_off();
1115 return;
1116 }
1117
1118 /* Note, snapshot can not be used when the tracer uses it */
1119 if (tracer->use_max_tr) {
1120 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
1121 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
1122 return;
1123 }
1124
1125 local_irq_save(flags);
1126 update_max_tr(tr, current, smp_processor_id(), cond_data);
1127 local_irq_restore(flags);
1128}
1129
1130void tracing_snapshot_instance(struct trace_array *tr)
1131{
1132 tracing_snapshot_instance_cond(tr, NULL);
1133}
1134
1135/**
1136 * tracing_snapshot - take a snapshot of the current buffer.
1137 *
1138 * This causes a swap between the snapshot buffer and the current live
1139 * tracing buffer. You can use this to take snapshots of the live
1140 * trace when some condition is triggered, but continue to trace.
1141 *
1142 * Note, make sure to allocate the snapshot with either
1143 * a tracing_snapshot_alloc(), or by doing it manually
1144 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
1145 *
1146 * If the snapshot buffer is not allocated, it will stop tracing.
1147 * Basically making a permanent snapshot.
1148 */
1149void tracing_snapshot(void)
1150{
1151 struct trace_array *tr = &global_trace;
1152
1153 tracing_snapshot_instance(tr);
1154}
1155EXPORT_SYMBOL_GPL(tracing_snapshot);
1156
1157/**
1158 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1159 * @tr: The tracing instance to snapshot
1160 * @cond_data: The data to be tested conditionally, and possibly saved
1161 *
1162 * This is the same as tracing_snapshot() except that the snapshot is
1163 * conditional - the snapshot will only happen if the
1164 * cond_snapshot.update() implementation receiving the cond_data
1165 * returns true, which means that the trace array's cond_snapshot
1166 * update() operation used the cond_data to determine whether the
1167 * snapshot should be taken, and if it was, presumably saved it along
1168 * with the snapshot.
1169 */
1170void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1171{
1172 tracing_snapshot_instance_cond(tr, cond_data);
1173}
1174EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1175
1176/**
1177 * tracing_snapshot_cond_data - get the user data associated with a snapshot
1178 * @tr: The tracing instance
1179 *
1180 * When the user enables a conditional snapshot using
1181 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1182 * with the snapshot. This accessor is used to retrieve it.
1183 *
1184 * Should not be called from cond_snapshot.update(), since it takes
1185 * the tr->max_lock lock, which the code calling
1186 * cond_snapshot.update() has already done.
1187 *
1188 * Returns the cond_data associated with the trace array's snapshot.
1189 */
1190void *tracing_cond_snapshot_data(struct trace_array *tr)
1191{
1192 void *cond_data = NULL;
1193
1194 arch_spin_lock(&tr->max_lock);
1195
1196 if (tr->cond_snapshot)
1197 cond_data = tr->cond_snapshot->cond_data;
1198
1199 arch_spin_unlock(&tr->max_lock);
1200
1201 return cond_data;
1202}
1203EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1204
1205static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1206 struct array_buffer *size_buf, int cpu_id);
1207static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1208
1209int tracing_alloc_snapshot_instance(struct trace_array *tr)
1210{
1211 int ret;
1212
1213 if (!tr->allocated_snapshot) {
1214
1215 /* allocate spare buffer */
1216 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1217 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1218 if (ret < 0)
1219 return ret;
1220
1221 tr->allocated_snapshot = true;
1222 }
1223
1224 return 0;
1225}
1226
1227static void free_snapshot(struct trace_array *tr)
1228{
1229 /*
1230 * We don't free the ring buffer. instead, resize it because
1231 * The max_tr ring buffer has some state (e.g. ring->clock) and
1232 * we want preserve it.
1233 */
1234 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1235 set_buffer_entries(&tr->max_buffer, 1);
1236 tracing_reset_online_cpus(&tr->max_buffer);
1237 tr->allocated_snapshot = false;
1238}
1239
1240/**
1241 * tracing_alloc_snapshot - allocate snapshot buffer.
1242 *
1243 * This only allocates the snapshot buffer if it isn't already
1244 * allocated - it doesn't also take a snapshot.
1245 *
1246 * This is meant to be used in cases where the snapshot buffer needs
1247 * to be set up for events that can't sleep but need to be able to
1248 * trigger a snapshot.
1249 */
1250int tracing_alloc_snapshot(void)
1251{
1252 struct trace_array *tr = &global_trace;
1253 int ret;
1254
1255 ret = tracing_alloc_snapshot_instance(tr);
1256 WARN_ON(ret < 0);
1257
1258 return ret;
1259}
1260EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1261
1262/**
1263 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1264 *
1265 * This is similar to tracing_snapshot(), but it will allocate the
1266 * snapshot buffer if it isn't already allocated. Use this only
1267 * where it is safe to sleep, as the allocation may sleep.
1268 *
1269 * This causes a swap between the snapshot buffer and the current live
1270 * tracing buffer. You can use this to take snapshots of the live
1271 * trace when some condition is triggered, but continue to trace.
1272 */
1273void tracing_snapshot_alloc(void)
1274{
1275 int ret;
1276
1277 ret = tracing_alloc_snapshot();
1278 if (ret < 0)
1279 return;
1280
1281 tracing_snapshot();
1282}
1283EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1284
1285/**
1286 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1287 * @tr: The tracing instance
1288 * @cond_data: User data to associate with the snapshot
1289 * @update: Implementation of the cond_snapshot update function
1290 *
1291 * Check whether the conditional snapshot for the given instance has
1292 * already been enabled, or if the current tracer is already using a
1293 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1294 * save the cond_data and update function inside.
1295 *
1296 * Returns 0 if successful, error otherwise.
1297 */
1298int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1299 cond_update_fn_t update)
1300{
1301 struct cond_snapshot *cond_snapshot;
1302 int ret = 0;
1303
1304 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1305 if (!cond_snapshot)
1306 return -ENOMEM;
1307
1308 cond_snapshot->cond_data = cond_data;
1309 cond_snapshot->update = update;
1310
1311 mutex_lock(&trace_types_lock);
1312
1313 ret = tracing_alloc_snapshot_instance(tr);
1314 if (ret)
1315 goto fail_unlock;
1316
1317 if (tr->current_trace->use_max_tr) {
1318 ret = -EBUSY;
1319 goto fail_unlock;
1320 }
1321
1322 /*
1323 * The cond_snapshot can only change to NULL without the
1324 * trace_types_lock. We don't care if we race with it going
1325 * to NULL, but we want to make sure that it's not set to
1326 * something other than NULL when we get here, which we can
1327 * do safely with only holding the trace_types_lock and not
1328 * having to take the max_lock.
1329 */
1330 if (tr->cond_snapshot) {
1331 ret = -EBUSY;
1332 goto fail_unlock;
1333 }
1334
1335 arch_spin_lock(&tr->max_lock);
1336 tr->cond_snapshot = cond_snapshot;
1337 arch_spin_unlock(&tr->max_lock);
1338
1339 mutex_unlock(&trace_types_lock);
1340
1341 return ret;
1342
1343 fail_unlock:
1344 mutex_unlock(&trace_types_lock);
1345 kfree(cond_snapshot);
1346 return ret;
1347}
1348EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1349
1350/**
1351 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1352 * @tr: The tracing instance
1353 *
1354 * Check whether the conditional snapshot for the given instance is
1355 * enabled; if so, free the cond_snapshot associated with it,
1356 * otherwise return -EINVAL.
1357 *
1358 * Returns 0 if successful, error otherwise.
1359 */
1360int tracing_snapshot_cond_disable(struct trace_array *tr)
1361{
1362 int ret = 0;
1363
1364 arch_spin_lock(&tr->max_lock);
1365
1366 if (!tr->cond_snapshot)
1367 ret = -EINVAL;
1368 else {
1369 kfree(tr->cond_snapshot);
1370 tr->cond_snapshot = NULL;
1371 }
1372
1373 arch_spin_unlock(&tr->max_lock);
1374
1375 return ret;
1376}
1377EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1378#else
1379void tracing_snapshot(void)
1380{
1381 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1382}
1383EXPORT_SYMBOL_GPL(tracing_snapshot);
1384void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1385{
1386 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1387}
1388EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1389int tracing_alloc_snapshot(void)
1390{
1391 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1392 return -ENODEV;
1393}
1394EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1395void tracing_snapshot_alloc(void)
1396{
1397 /* Give warning */
1398 tracing_snapshot();
1399}
1400EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1401void *tracing_cond_snapshot_data(struct trace_array *tr)
1402{
1403 return NULL;
1404}
1405EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1406int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1407{
1408 return -ENODEV;
1409}
1410EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1411int tracing_snapshot_cond_disable(struct trace_array *tr)
1412{
1413 return false;
1414}
1415EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1416#endif /* CONFIG_TRACER_SNAPSHOT */
1417
1418void tracer_tracing_off(struct trace_array *tr)
1419{
1420 if (tr->array_buffer.buffer)
1421 ring_buffer_record_off(tr->array_buffer.buffer);
1422 /*
1423 * This flag is looked at when buffers haven't been allocated
1424 * yet, or by some tracers (like irqsoff), that just want to
1425 * know if the ring buffer has been disabled, but it can handle
1426 * races of where it gets disabled but we still do a record.
1427 * As the check is in the fast path of the tracers, it is more
1428 * important to be fast than accurate.
1429 */
1430 tr->buffer_disabled = 1;
1431 /* Make the flag seen by readers */
1432 smp_wmb();
1433}
1434
1435/**
1436 * tracing_off - turn off tracing buffers
1437 *
1438 * This function stops the tracing buffers from recording data.
1439 * It does not disable any overhead the tracers themselves may
1440 * be causing. This function simply causes all recording to
1441 * the ring buffers to fail.
1442 */
1443void tracing_off(void)
1444{
1445 tracer_tracing_off(&global_trace);
1446}
1447EXPORT_SYMBOL_GPL(tracing_off);
1448
1449void disable_trace_on_warning(void)
1450{
1451 if (__disable_trace_on_warning) {
1452 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1453 "Disabling tracing due to warning\n");
1454 tracing_off();
1455 }
1456}
1457
1458/**
1459 * tracer_tracing_is_on - show real state of ring buffer enabled
1460 * @tr : the trace array to know if ring buffer is enabled
1461 *
1462 * Shows real state of the ring buffer if it is enabled or not.
1463 */
1464bool tracer_tracing_is_on(struct trace_array *tr)
1465{
1466 if (tr->array_buffer.buffer)
1467 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1468 return !tr->buffer_disabled;
1469}
1470
1471/**
1472 * tracing_is_on - show state of ring buffers enabled
1473 */
1474int tracing_is_on(void)
1475{
1476 return tracer_tracing_is_on(&global_trace);
1477}
1478EXPORT_SYMBOL_GPL(tracing_is_on);
1479
1480static int __init set_buf_size(char *str)
1481{
1482 unsigned long buf_size;
1483
1484 if (!str)
1485 return 0;
1486 buf_size = memparse(str, &str);
1487 /*
1488 * nr_entries can not be zero and the startup
1489 * tests require some buffer space. Therefore
1490 * ensure we have at least 4096 bytes of buffer.
1491 */
1492 trace_buf_size = max(4096UL, buf_size);
1493 return 1;
1494}
1495__setup("trace_buf_size=", set_buf_size);
1496
1497static int __init set_tracing_thresh(char *str)
1498{
1499 unsigned long threshold;
1500 int ret;
1501
1502 if (!str)
1503 return 0;
1504 ret = kstrtoul(str, 0, &threshold);
1505 if (ret < 0)
1506 return 0;
1507 tracing_thresh = threshold * 1000;
1508 return 1;
1509}
1510__setup("tracing_thresh=", set_tracing_thresh);
1511
1512unsigned long nsecs_to_usecs(unsigned long nsecs)
1513{
1514 return nsecs / 1000;
1515}
1516
1517/*
1518 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1519 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1520 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1521 * of strings in the order that the evals (enum) were defined.
1522 */
1523#undef C
1524#define C(a, b) b
1525
1526/* These must match the bit positions in trace_iterator_flags */
1527static const char *trace_options[] = {
1528 TRACE_FLAGS
1529 NULL
1530};
1531
1532static struct {
1533 u64 (*func)(void);
1534 const char *name;
1535 int in_ns; /* is this clock in nanoseconds? */
1536} trace_clocks[] = {
1537 { trace_clock_local, "local", 1 },
1538 { trace_clock_global, "global", 1 },
1539 { trace_clock_counter, "counter", 0 },
1540 { trace_clock_jiffies, "uptime", 0 },
1541 { trace_clock, "perf", 1 },
1542 { ktime_get_mono_fast_ns, "mono", 1 },
1543 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1544 { ktime_get_boot_fast_ns, "boot", 1 },
1545 ARCH_TRACE_CLOCKS
1546};
1547
1548bool trace_clock_in_ns(struct trace_array *tr)
1549{
1550 if (trace_clocks[tr->clock_id].in_ns)
1551 return true;
1552
1553 return false;
1554}
1555
1556/*
1557 * trace_parser_get_init - gets the buffer for trace parser
1558 */
1559int trace_parser_get_init(struct trace_parser *parser, int size)
1560{
1561 memset(parser, 0, sizeof(*parser));
1562
1563 parser->buffer = kmalloc(size, GFP_KERNEL);
1564 if (!parser->buffer)
1565 return 1;
1566
1567 parser->size = size;
1568 return 0;
1569}
1570
1571/*
1572 * trace_parser_put - frees the buffer for trace parser
1573 */
1574void trace_parser_put(struct trace_parser *parser)
1575{
1576 kfree(parser->buffer);
1577 parser->buffer = NULL;
1578}
1579
1580/*
1581 * trace_get_user - reads the user input string separated by space
1582 * (matched by isspace(ch))
1583 *
1584 * For each string found the 'struct trace_parser' is updated,
1585 * and the function returns.
1586 *
1587 * Returns number of bytes read.
1588 *
1589 * See kernel/trace/trace.h for 'struct trace_parser' details.
1590 */
1591int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1592 size_t cnt, loff_t *ppos)
1593{
1594 char ch;
1595 size_t read = 0;
1596 ssize_t ret;
1597
1598 if (!*ppos)
1599 trace_parser_clear(parser);
1600
1601 ret = get_user(ch, ubuf++);
1602 if (ret)
1603 goto out;
1604
1605 read++;
1606 cnt--;
1607
1608 /*
1609 * The parser is not finished with the last write,
1610 * continue reading the user input without skipping spaces.
1611 */
1612 if (!parser->cont) {
1613 /* skip white space */
1614 while (cnt && isspace(ch)) {
1615 ret = get_user(ch, ubuf++);
1616 if (ret)
1617 goto out;
1618 read++;
1619 cnt--;
1620 }
1621
1622 parser->idx = 0;
1623
1624 /* only spaces were written */
1625 if (isspace(ch) || !ch) {
1626 *ppos += read;
1627 ret = read;
1628 goto out;
1629 }
1630 }
1631
1632 /* read the non-space input */
1633 while (cnt && !isspace(ch) && ch) {
1634 if (parser->idx < parser->size - 1)
1635 parser->buffer[parser->idx++] = ch;
1636 else {
1637 ret = -EINVAL;
1638 goto out;
1639 }
1640 ret = get_user(ch, ubuf++);
1641 if (ret)
1642 goto out;
1643 read++;
1644 cnt--;
1645 }
1646
1647 /* We either got finished input or we have to wait for another call. */
1648 if (isspace(ch) || !ch) {
1649 parser->buffer[parser->idx] = 0;
1650 parser->cont = false;
1651 } else if (parser->idx < parser->size - 1) {
1652 parser->cont = true;
1653 parser->buffer[parser->idx++] = ch;
1654 /* Make sure the parsed string always terminates with '\0'. */
1655 parser->buffer[parser->idx] = 0;
1656 } else {
1657 ret = -EINVAL;
1658 goto out;
1659 }
1660
1661 *ppos += read;
1662 ret = read;
1663
1664out:
1665 return ret;
1666}
1667
1668/* TODO add a seq_buf_to_buffer() */
1669static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1670{
1671 int len;
1672
1673 if (trace_seq_used(s) <= s->seq.readpos)
1674 return -EBUSY;
1675
1676 len = trace_seq_used(s) - s->seq.readpos;
1677 if (cnt > len)
1678 cnt = len;
1679 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1680
1681 s->seq.readpos += cnt;
1682 return cnt;
1683}
1684
1685unsigned long __read_mostly tracing_thresh;
1686static const struct file_operations tracing_max_lat_fops;
1687
1688#ifdef LATENCY_FS_NOTIFY
1689
1690static struct workqueue_struct *fsnotify_wq;
1691
1692static void latency_fsnotify_workfn(struct work_struct *work)
1693{
1694 struct trace_array *tr = container_of(work, struct trace_array,
1695 fsnotify_work);
1696 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1697}
1698
1699static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1700{
1701 struct trace_array *tr = container_of(iwork, struct trace_array,
1702 fsnotify_irqwork);
1703 queue_work(fsnotify_wq, &tr->fsnotify_work);
1704}
1705
1706static void trace_create_maxlat_file(struct trace_array *tr,
1707 struct dentry *d_tracer)
1708{
1709 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1710 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1711 tr->d_max_latency = trace_create_file("tracing_max_latency",
1712 TRACE_MODE_WRITE,
1713 d_tracer, &tr->max_latency,
1714 &tracing_max_lat_fops);
1715}
1716
1717__init static int latency_fsnotify_init(void)
1718{
1719 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1720 WQ_UNBOUND | WQ_HIGHPRI, 0);
1721 if (!fsnotify_wq) {
1722 pr_err("Unable to allocate tr_max_lat_wq\n");
1723 return -ENOMEM;
1724 }
1725 return 0;
1726}
1727
1728late_initcall_sync(latency_fsnotify_init);
1729
1730void latency_fsnotify(struct trace_array *tr)
1731{
1732 if (!fsnotify_wq)
1733 return;
1734 /*
1735 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1736 * possible that we are called from __schedule() or do_idle(), which
1737 * could cause a deadlock.
1738 */
1739 irq_work_queue(&tr->fsnotify_irqwork);
1740}
1741
1742#elif defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER) \
1743 || defined(CONFIG_OSNOISE_TRACER)
1744
1745#define trace_create_maxlat_file(tr, d_tracer) \
1746 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1747 d_tracer, &tr->max_latency, &tracing_max_lat_fops)
1748
1749#else
1750#define trace_create_maxlat_file(tr, d_tracer) do { } while (0)
1751#endif
1752
1753#ifdef CONFIG_TRACER_MAX_TRACE
1754/*
1755 * Copy the new maximum trace into the separate maximum-trace
1756 * structure. (this way the maximum trace is permanently saved,
1757 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1758 */
1759static void
1760__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1761{
1762 struct array_buffer *trace_buf = &tr->array_buffer;
1763 struct array_buffer *max_buf = &tr->max_buffer;
1764 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1765 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1766
1767 max_buf->cpu = cpu;
1768 max_buf->time_start = data->preempt_timestamp;
1769
1770 max_data->saved_latency = tr->max_latency;
1771 max_data->critical_start = data->critical_start;
1772 max_data->critical_end = data->critical_end;
1773
1774 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1775 max_data->pid = tsk->pid;
1776 /*
1777 * If tsk == current, then use current_uid(), as that does not use
1778 * RCU. The irq tracer can be called out of RCU scope.
1779 */
1780 if (tsk == current)
1781 max_data->uid = current_uid();
1782 else
1783 max_data->uid = task_uid(tsk);
1784
1785 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1786 max_data->policy = tsk->policy;
1787 max_data->rt_priority = tsk->rt_priority;
1788
1789 /* record this tasks comm */
1790 tracing_record_cmdline(tsk);
1791 latency_fsnotify(tr);
1792}
1793
1794/**
1795 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1796 * @tr: tracer
1797 * @tsk: the task with the latency
1798 * @cpu: The cpu that initiated the trace.
1799 * @cond_data: User data associated with a conditional snapshot
1800 *
1801 * Flip the buffers between the @tr and the max_tr and record information
1802 * about which task was the cause of this latency.
1803 */
1804void
1805update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1806 void *cond_data)
1807{
1808 if (tr->stop_count)
1809 return;
1810
1811 WARN_ON_ONCE(!irqs_disabled());
1812
1813 if (!tr->allocated_snapshot) {
1814 /* Only the nop tracer should hit this when disabling */
1815 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1816 return;
1817 }
1818
1819 arch_spin_lock(&tr->max_lock);
1820
1821 /* Inherit the recordable setting from array_buffer */
1822 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1823 ring_buffer_record_on(tr->max_buffer.buffer);
1824 else
1825 ring_buffer_record_off(tr->max_buffer.buffer);
1826
1827#ifdef CONFIG_TRACER_SNAPSHOT
1828 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data))
1829 goto out_unlock;
1830#endif
1831 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1832
1833 __update_max_tr(tr, tsk, cpu);
1834
1835 out_unlock:
1836 arch_spin_unlock(&tr->max_lock);
1837}
1838
1839/**
1840 * update_max_tr_single - only copy one trace over, and reset the rest
1841 * @tr: tracer
1842 * @tsk: task with the latency
1843 * @cpu: the cpu of the buffer to copy.
1844 *
1845 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1846 */
1847void
1848update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1849{
1850 int ret;
1851
1852 if (tr->stop_count)
1853 return;
1854
1855 WARN_ON_ONCE(!irqs_disabled());
1856 if (!tr->allocated_snapshot) {
1857 /* Only the nop tracer should hit this when disabling */
1858 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1859 return;
1860 }
1861
1862 arch_spin_lock(&tr->max_lock);
1863
1864 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1865
1866 if (ret == -EBUSY) {
1867 /*
1868 * We failed to swap the buffer due to a commit taking
1869 * place on this CPU. We fail to record, but we reset
1870 * the max trace buffer (no one writes directly to it)
1871 * and flag that it failed.
1872 */
1873 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1874 "Failed to swap buffers due to commit in progress\n");
1875 }
1876
1877 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1878
1879 __update_max_tr(tr, tsk, cpu);
1880 arch_spin_unlock(&tr->max_lock);
1881}
1882#endif /* CONFIG_TRACER_MAX_TRACE */
1883
1884static int wait_on_pipe(struct trace_iterator *iter, int full)
1885{
1886 /* Iterators are static, they should be filled or empty */
1887 if (trace_buffer_iter(iter, iter->cpu_file))
1888 return 0;
1889
1890 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1891 full);
1892}
1893
1894#ifdef CONFIG_FTRACE_STARTUP_TEST
1895static bool selftests_can_run;
1896
1897struct trace_selftests {
1898 struct list_head list;
1899 struct tracer *type;
1900};
1901
1902static LIST_HEAD(postponed_selftests);
1903
1904static int save_selftest(struct tracer *type)
1905{
1906 struct trace_selftests *selftest;
1907
1908 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1909 if (!selftest)
1910 return -ENOMEM;
1911
1912 selftest->type = type;
1913 list_add(&selftest->list, &postponed_selftests);
1914 return 0;
1915}
1916
1917static int run_tracer_selftest(struct tracer *type)
1918{
1919 struct trace_array *tr = &global_trace;
1920 struct tracer *saved_tracer = tr->current_trace;
1921 int ret;
1922
1923 if (!type->selftest || tracing_selftest_disabled)
1924 return 0;
1925
1926 /*
1927 * If a tracer registers early in boot up (before scheduling is
1928 * initialized and such), then do not run its selftests yet.
1929 * Instead, run it a little later in the boot process.
1930 */
1931 if (!selftests_can_run)
1932 return save_selftest(type);
1933
1934 if (!tracing_is_on()) {
1935 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
1936 type->name);
1937 return 0;
1938 }
1939
1940 /*
1941 * Run a selftest on this tracer.
1942 * Here we reset the trace buffer, and set the current
1943 * tracer to be this tracer. The tracer can then run some
1944 * internal tracing to verify that everything is in order.
1945 * If we fail, we do not register this tracer.
1946 */
1947 tracing_reset_online_cpus(&tr->array_buffer);
1948
1949 tr->current_trace = type;
1950
1951#ifdef CONFIG_TRACER_MAX_TRACE
1952 if (type->use_max_tr) {
1953 /* If we expanded the buffers, make sure the max is expanded too */
1954 if (ring_buffer_expanded)
1955 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1956 RING_BUFFER_ALL_CPUS);
1957 tr->allocated_snapshot = true;
1958 }
1959#endif
1960
1961 /* the test is responsible for initializing and enabling */
1962 pr_info("Testing tracer %s: ", type->name);
1963 ret = type->selftest(type, tr);
1964 /* the test is responsible for resetting too */
1965 tr->current_trace = saved_tracer;
1966 if (ret) {
1967 printk(KERN_CONT "FAILED!\n");
1968 /* Add the warning after printing 'FAILED' */
1969 WARN_ON(1);
1970 return -1;
1971 }
1972 /* Only reset on passing, to avoid touching corrupted buffers */
1973 tracing_reset_online_cpus(&tr->array_buffer);
1974
1975#ifdef CONFIG_TRACER_MAX_TRACE
1976 if (type->use_max_tr) {
1977 tr->allocated_snapshot = false;
1978
1979 /* Shrink the max buffer again */
1980 if (ring_buffer_expanded)
1981 ring_buffer_resize(tr->max_buffer.buffer, 1,
1982 RING_BUFFER_ALL_CPUS);
1983 }
1984#endif
1985
1986 printk(KERN_CONT "PASSED\n");
1987 return 0;
1988}
1989
1990static __init int init_trace_selftests(void)
1991{
1992 struct trace_selftests *p, *n;
1993 struct tracer *t, **last;
1994 int ret;
1995
1996 selftests_can_run = true;
1997
1998 mutex_lock(&trace_types_lock);
1999
2000 if (list_empty(&postponed_selftests))
2001 goto out;
2002
2003 pr_info("Running postponed tracer tests:\n");
2004
2005 tracing_selftest_running = true;
2006 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2007 /* This loop can take minutes when sanitizers are enabled, so
2008 * lets make sure we allow RCU processing.
2009 */
2010 cond_resched();
2011 ret = run_tracer_selftest(p->type);
2012 /* If the test fails, then warn and remove from available_tracers */
2013 if (ret < 0) {
2014 WARN(1, "tracer: %s failed selftest, disabling\n",
2015 p->type->name);
2016 last = &trace_types;
2017 for (t = trace_types; t; t = t->next) {
2018 if (t == p->type) {
2019 *last = t->next;
2020 break;
2021 }
2022 last = &t->next;
2023 }
2024 }
2025 list_del(&p->list);
2026 kfree(p);
2027 }
2028 tracing_selftest_running = false;
2029
2030 out:
2031 mutex_unlock(&trace_types_lock);
2032
2033 return 0;
2034}
2035core_initcall(init_trace_selftests);
2036#else
2037static inline int run_tracer_selftest(struct tracer *type)
2038{
2039 return 0;
2040}
2041#endif /* CONFIG_FTRACE_STARTUP_TEST */
2042
2043static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2044
2045static void __init apply_trace_boot_options(void);
2046
2047/**
2048 * register_tracer - register a tracer with the ftrace system.
2049 * @type: the plugin for the tracer
2050 *
2051 * Register a new plugin tracer.
2052 */
2053int __init register_tracer(struct tracer *type)
2054{
2055 struct tracer *t;
2056 int ret = 0;
2057
2058 if (!type->name) {
2059 pr_info("Tracer must have a name\n");
2060 return -1;
2061 }
2062
2063 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2064 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2065 return -1;
2066 }
2067
2068 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2069 pr_warn("Can not register tracer %s due to lockdown\n",
2070 type->name);
2071 return -EPERM;
2072 }
2073
2074 mutex_lock(&trace_types_lock);
2075
2076 tracing_selftest_running = true;
2077
2078 for (t = trace_types; t; t = t->next) {
2079 if (strcmp(type->name, t->name) == 0) {
2080 /* already found */
2081 pr_info("Tracer %s already registered\n",
2082 type->name);
2083 ret = -1;
2084 goto out;
2085 }
2086 }
2087
2088 if (!type->set_flag)
2089 type->set_flag = &dummy_set_flag;
2090 if (!type->flags) {
2091 /*allocate a dummy tracer_flags*/
2092 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2093 if (!type->flags) {
2094 ret = -ENOMEM;
2095 goto out;
2096 }
2097 type->flags->val = 0;
2098 type->flags->opts = dummy_tracer_opt;
2099 } else
2100 if (!type->flags->opts)
2101 type->flags->opts = dummy_tracer_opt;
2102
2103 /* store the tracer for __set_tracer_option */
2104 type->flags->trace = type;
2105
2106 ret = run_tracer_selftest(type);
2107 if (ret < 0)
2108 goto out;
2109
2110 type->next = trace_types;
2111 trace_types = type;
2112 add_tracer_options(&global_trace, type);
2113
2114 out:
2115 tracing_selftest_running = false;
2116 mutex_unlock(&trace_types_lock);
2117
2118 if (ret || !default_bootup_tracer)
2119 goto out_unlock;
2120
2121 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2122 goto out_unlock;
2123
2124 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2125 /* Do we want this tracer to start on bootup? */
2126 tracing_set_tracer(&global_trace, type->name);
2127 default_bootup_tracer = NULL;
2128
2129 apply_trace_boot_options();
2130
2131 /* disable other selftests, since this will break it. */
2132 disable_tracing_selftest("running a tracer");
2133
2134 out_unlock:
2135 return ret;
2136}
2137
2138static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2139{
2140 struct trace_buffer *buffer = buf->buffer;
2141
2142 if (!buffer)
2143 return;
2144
2145 ring_buffer_record_disable(buffer);
2146
2147 /* Make sure all commits have finished */
2148 synchronize_rcu();
2149 ring_buffer_reset_cpu(buffer, cpu);
2150
2151 ring_buffer_record_enable(buffer);
2152}
2153
2154void tracing_reset_online_cpus(struct array_buffer *buf)
2155{
2156 struct trace_buffer *buffer = buf->buffer;
2157
2158 if (!buffer)
2159 return;
2160
2161 ring_buffer_record_disable(buffer);
2162
2163 /* Make sure all commits have finished */
2164 synchronize_rcu();
2165
2166 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2167
2168 ring_buffer_reset_online_cpus(buffer);
2169
2170 ring_buffer_record_enable(buffer);
2171}
2172
2173/* Must have trace_types_lock held */
2174void tracing_reset_all_online_cpus(void)
2175{
2176 struct trace_array *tr;
2177
2178 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2179 if (!tr->clear_trace)
2180 continue;
2181 tr->clear_trace = false;
2182 tracing_reset_online_cpus(&tr->array_buffer);
2183#ifdef CONFIG_TRACER_MAX_TRACE
2184 tracing_reset_online_cpus(&tr->max_buffer);
2185#endif
2186 }
2187}
2188
2189/*
2190 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2191 * is the tgid last observed corresponding to pid=i.
2192 */
2193static int *tgid_map;
2194
2195/* The maximum valid index into tgid_map. */
2196static size_t tgid_map_max;
2197
2198#define SAVED_CMDLINES_DEFAULT 128
2199#define NO_CMDLINE_MAP UINT_MAX
2200static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2201struct saved_cmdlines_buffer {
2202 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2203 unsigned *map_cmdline_to_pid;
2204 unsigned cmdline_num;
2205 int cmdline_idx;
2206 char *saved_cmdlines;
2207};
2208static struct saved_cmdlines_buffer *savedcmd;
2209
2210static inline char *get_saved_cmdlines(int idx)
2211{
2212 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2213}
2214
2215static inline void set_cmdline(int idx, const char *cmdline)
2216{
2217 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2218}
2219
2220static int allocate_cmdlines_buffer(unsigned int val,
2221 struct saved_cmdlines_buffer *s)
2222{
2223 s->map_cmdline_to_pid = kmalloc_array(val,
2224 sizeof(*s->map_cmdline_to_pid),
2225 GFP_KERNEL);
2226 if (!s->map_cmdline_to_pid)
2227 return -ENOMEM;
2228
2229 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2230 if (!s->saved_cmdlines) {
2231 kfree(s->map_cmdline_to_pid);
2232 return -ENOMEM;
2233 }
2234
2235 s->cmdline_idx = 0;
2236 s->cmdline_num = val;
2237 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2238 sizeof(s->map_pid_to_cmdline));
2239 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2240 val * sizeof(*s->map_cmdline_to_pid));
2241
2242 return 0;
2243}
2244
2245static int trace_create_savedcmd(void)
2246{
2247 int ret;
2248
2249 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2250 if (!savedcmd)
2251 return -ENOMEM;
2252
2253 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2254 if (ret < 0) {
2255 kfree(savedcmd);
2256 savedcmd = NULL;
2257 return -ENOMEM;
2258 }
2259
2260 return 0;
2261}
2262
2263int is_tracing_stopped(void)
2264{
2265 return global_trace.stop_count;
2266}
2267
2268/**
2269 * tracing_start - quick start of the tracer
2270 *
2271 * If tracing is enabled but was stopped by tracing_stop,
2272 * this will start the tracer back up.
2273 */
2274void tracing_start(void)
2275{
2276 struct trace_buffer *buffer;
2277 unsigned long flags;
2278
2279 if (tracing_disabled)
2280 return;
2281
2282 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2283 if (--global_trace.stop_count) {
2284 if (global_trace.stop_count < 0) {
2285 /* Someone screwed up their debugging */
2286 WARN_ON_ONCE(1);
2287 global_trace.stop_count = 0;
2288 }
2289 goto out;
2290 }
2291
2292 /* Prevent the buffers from switching */
2293 arch_spin_lock(&global_trace.max_lock);
2294
2295 buffer = global_trace.array_buffer.buffer;
2296 if (buffer)
2297 ring_buffer_record_enable(buffer);
2298
2299#ifdef CONFIG_TRACER_MAX_TRACE
2300 buffer = global_trace.max_buffer.buffer;
2301 if (buffer)
2302 ring_buffer_record_enable(buffer);
2303#endif
2304
2305 arch_spin_unlock(&global_trace.max_lock);
2306
2307 out:
2308 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2309}
2310
2311static void tracing_start_tr(struct trace_array *tr)
2312{
2313 struct trace_buffer *buffer;
2314 unsigned long flags;
2315
2316 if (tracing_disabled)
2317 return;
2318
2319 /* If global, we need to also start the max tracer */
2320 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2321 return tracing_start();
2322
2323 raw_spin_lock_irqsave(&tr->start_lock, flags);
2324
2325 if (--tr->stop_count) {
2326 if (tr->stop_count < 0) {
2327 /* Someone screwed up their debugging */
2328 WARN_ON_ONCE(1);
2329 tr->stop_count = 0;
2330 }
2331 goto out;
2332 }
2333
2334 buffer = tr->array_buffer.buffer;
2335 if (buffer)
2336 ring_buffer_record_enable(buffer);
2337
2338 out:
2339 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2340}
2341
2342/**
2343 * tracing_stop - quick stop of the tracer
2344 *
2345 * Light weight way to stop tracing. Use in conjunction with
2346 * tracing_start.
2347 */
2348void tracing_stop(void)
2349{
2350 struct trace_buffer *buffer;
2351 unsigned long flags;
2352
2353 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2354 if (global_trace.stop_count++)
2355 goto out;
2356
2357 /* Prevent the buffers from switching */
2358 arch_spin_lock(&global_trace.max_lock);
2359
2360 buffer = global_trace.array_buffer.buffer;
2361 if (buffer)
2362 ring_buffer_record_disable(buffer);
2363
2364#ifdef CONFIG_TRACER_MAX_TRACE
2365 buffer = global_trace.max_buffer.buffer;
2366 if (buffer)
2367 ring_buffer_record_disable(buffer);
2368#endif
2369
2370 arch_spin_unlock(&global_trace.max_lock);
2371
2372 out:
2373 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2374}
2375
2376static void tracing_stop_tr(struct trace_array *tr)
2377{
2378 struct trace_buffer *buffer;
2379 unsigned long flags;
2380
2381 /* If global, we need to also stop the max tracer */
2382 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2383 return tracing_stop();
2384
2385 raw_spin_lock_irqsave(&tr->start_lock, flags);
2386 if (tr->stop_count++)
2387 goto out;
2388
2389 buffer = tr->array_buffer.buffer;
2390 if (buffer)
2391 ring_buffer_record_disable(buffer);
2392
2393 out:
2394 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2395}
2396
2397static int trace_save_cmdline(struct task_struct *tsk)
2398{
2399 unsigned tpid, idx;
2400
2401 /* treat recording of idle task as a success */
2402 if (!tsk->pid)
2403 return 1;
2404
2405 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2406
2407 /*
2408 * It's not the end of the world if we don't get
2409 * the lock, but we also don't want to spin
2410 * nor do we want to disable interrupts,
2411 * so if we miss here, then better luck next time.
2412 */
2413 if (!arch_spin_trylock(&trace_cmdline_lock))
2414 return 0;
2415
2416 idx = savedcmd->map_pid_to_cmdline[tpid];
2417 if (idx == NO_CMDLINE_MAP) {
2418 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2419
2420 savedcmd->map_pid_to_cmdline[tpid] = idx;
2421 savedcmd->cmdline_idx = idx;
2422 }
2423
2424 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2425 set_cmdline(idx, tsk->comm);
2426
2427 arch_spin_unlock(&trace_cmdline_lock);
2428
2429 return 1;
2430}
2431
2432static void __trace_find_cmdline(int pid, char comm[])
2433{
2434 unsigned map;
2435 int tpid;
2436
2437 if (!pid) {
2438 strcpy(comm, "<idle>");
2439 return;
2440 }
2441
2442 if (WARN_ON_ONCE(pid < 0)) {
2443 strcpy(comm, "<XXX>");
2444 return;
2445 }
2446
2447 tpid = pid & (PID_MAX_DEFAULT - 1);
2448 map = savedcmd->map_pid_to_cmdline[tpid];
2449 if (map != NO_CMDLINE_MAP) {
2450 tpid = savedcmd->map_cmdline_to_pid[map];
2451 if (tpid == pid) {
2452 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2453 return;
2454 }
2455 }
2456 strcpy(comm, "<...>");
2457}
2458
2459void trace_find_cmdline(int pid, char comm[])
2460{
2461 preempt_disable();
2462 arch_spin_lock(&trace_cmdline_lock);
2463
2464 __trace_find_cmdline(pid, comm);
2465
2466 arch_spin_unlock(&trace_cmdline_lock);
2467 preempt_enable();
2468}
2469
2470static int *trace_find_tgid_ptr(int pid)
2471{
2472 /*
2473 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2474 * if we observe a non-NULL tgid_map then we also observe the correct
2475 * tgid_map_max.
2476 */
2477 int *map = smp_load_acquire(&tgid_map);
2478
2479 if (unlikely(!map || pid > tgid_map_max))
2480 return NULL;
2481
2482 return &map[pid];
2483}
2484
2485int trace_find_tgid(int pid)
2486{
2487 int *ptr = trace_find_tgid_ptr(pid);
2488
2489 return ptr ? *ptr : 0;
2490}
2491
2492static int trace_save_tgid(struct task_struct *tsk)
2493{
2494 int *ptr;
2495
2496 /* treat recording of idle task as a success */
2497 if (!tsk->pid)
2498 return 1;
2499
2500 ptr = trace_find_tgid_ptr(tsk->pid);
2501 if (!ptr)
2502 return 0;
2503
2504 *ptr = tsk->tgid;
2505 return 1;
2506}
2507
2508static bool tracing_record_taskinfo_skip(int flags)
2509{
2510 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2511 return true;
2512 if (!__this_cpu_read(trace_taskinfo_save))
2513 return true;
2514 return false;
2515}
2516
2517/**
2518 * tracing_record_taskinfo - record the task info of a task
2519 *
2520 * @task: task to record
2521 * @flags: TRACE_RECORD_CMDLINE for recording comm
2522 * TRACE_RECORD_TGID for recording tgid
2523 */
2524void tracing_record_taskinfo(struct task_struct *task, int flags)
2525{
2526 bool done;
2527
2528 if (tracing_record_taskinfo_skip(flags))
2529 return;
2530
2531 /*
2532 * Record as much task information as possible. If some fail, continue
2533 * to try to record the others.
2534 */
2535 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2536 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2537
2538 /* If recording any information failed, retry again soon. */
2539 if (!done)
2540 return;
2541
2542 __this_cpu_write(trace_taskinfo_save, false);
2543}
2544
2545/**
2546 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2547 *
2548 * @prev: previous task during sched_switch
2549 * @next: next task during sched_switch
2550 * @flags: TRACE_RECORD_CMDLINE for recording comm
2551 * TRACE_RECORD_TGID for recording tgid
2552 */
2553void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2554 struct task_struct *next, int flags)
2555{
2556 bool done;
2557
2558 if (tracing_record_taskinfo_skip(flags))
2559 return;
2560
2561 /*
2562 * Record as much task information as possible. If some fail, continue
2563 * to try to record the others.
2564 */
2565 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2566 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2567 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2568 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2569
2570 /* If recording any information failed, retry again soon. */
2571 if (!done)
2572 return;
2573
2574 __this_cpu_write(trace_taskinfo_save, false);
2575}
2576
2577/* Helpers to record a specific task information */
2578void tracing_record_cmdline(struct task_struct *task)
2579{
2580 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2581}
2582
2583void tracing_record_tgid(struct task_struct *task)
2584{
2585 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2586}
2587
2588/*
2589 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2590 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2591 * simplifies those functions and keeps them in sync.
2592 */
2593enum print_line_t trace_handle_return(struct trace_seq *s)
2594{
2595 return trace_seq_has_overflowed(s) ?
2596 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2597}
2598EXPORT_SYMBOL_GPL(trace_handle_return);
2599
2600static unsigned short migration_disable_value(void)
2601{
2602#if defined(CONFIG_SMP)
2603 return current->migration_disabled;
2604#else
2605 return 0;
2606#endif
2607}
2608
2609unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2610{
2611 unsigned int trace_flags = irqs_status;
2612 unsigned int pc;
2613
2614 pc = preempt_count();
2615
2616 if (pc & NMI_MASK)
2617 trace_flags |= TRACE_FLAG_NMI;
2618 if (pc & HARDIRQ_MASK)
2619 trace_flags |= TRACE_FLAG_HARDIRQ;
2620 if (in_serving_softirq())
2621 trace_flags |= TRACE_FLAG_SOFTIRQ;
2622 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2623 trace_flags |= TRACE_FLAG_BH_OFF;
2624
2625 if (tif_need_resched())
2626 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2627 if (test_preempt_need_resched())
2628 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2629 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2630 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2631}
2632
2633struct ring_buffer_event *
2634trace_buffer_lock_reserve(struct trace_buffer *buffer,
2635 int type,
2636 unsigned long len,
2637 unsigned int trace_ctx)
2638{
2639 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2640}
2641
2642DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2643DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2644static int trace_buffered_event_ref;
2645
2646/**
2647 * trace_buffered_event_enable - enable buffering events
2648 *
2649 * When events are being filtered, it is quicker to use a temporary
2650 * buffer to write the event data into if there's a likely chance
2651 * that it will not be committed. The discard of the ring buffer
2652 * is not as fast as committing, and is much slower than copying
2653 * a commit.
2654 *
2655 * When an event is to be filtered, allocate per cpu buffers to
2656 * write the event data into, and if the event is filtered and discarded
2657 * it is simply dropped, otherwise, the entire data is to be committed
2658 * in one shot.
2659 */
2660void trace_buffered_event_enable(void)
2661{
2662 struct ring_buffer_event *event;
2663 struct page *page;
2664 int cpu;
2665
2666 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2667
2668 if (trace_buffered_event_ref++)
2669 return;
2670
2671 for_each_tracing_cpu(cpu) {
2672 page = alloc_pages_node(cpu_to_node(cpu),
2673 GFP_KERNEL | __GFP_NORETRY, 0);
2674 if (!page)
2675 goto failed;
2676
2677 event = page_address(page);
2678 memset(event, 0, sizeof(*event));
2679
2680 per_cpu(trace_buffered_event, cpu) = event;
2681
2682 preempt_disable();
2683 if (cpu == smp_processor_id() &&
2684 __this_cpu_read(trace_buffered_event) !=
2685 per_cpu(trace_buffered_event, cpu))
2686 WARN_ON_ONCE(1);
2687 preempt_enable();
2688 }
2689
2690 return;
2691 failed:
2692 trace_buffered_event_disable();
2693}
2694
2695static void enable_trace_buffered_event(void *data)
2696{
2697 /* Probably not needed, but do it anyway */
2698 smp_rmb();
2699 this_cpu_dec(trace_buffered_event_cnt);
2700}
2701
2702static void disable_trace_buffered_event(void *data)
2703{
2704 this_cpu_inc(trace_buffered_event_cnt);
2705}
2706
2707/**
2708 * trace_buffered_event_disable - disable buffering events
2709 *
2710 * When a filter is removed, it is faster to not use the buffered
2711 * events, and to commit directly into the ring buffer. Free up
2712 * the temp buffers when there are no more users. This requires
2713 * special synchronization with current events.
2714 */
2715void trace_buffered_event_disable(void)
2716{
2717 int cpu;
2718
2719 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2720
2721 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2722 return;
2723
2724 if (--trace_buffered_event_ref)
2725 return;
2726
2727 preempt_disable();
2728 /* For each CPU, set the buffer as used. */
2729 smp_call_function_many(tracing_buffer_mask,
2730 disable_trace_buffered_event, NULL, 1);
2731 preempt_enable();
2732
2733 /* Wait for all current users to finish */
2734 synchronize_rcu();
2735
2736 for_each_tracing_cpu(cpu) {
2737 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2738 per_cpu(trace_buffered_event, cpu) = NULL;
2739 }
2740 /*
2741 * Make sure trace_buffered_event is NULL before clearing
2742 * trace_buffered_event_cnt.
2743 */
2744 smp_wmb();
2745
2746 preempt_disable();
2747 /* Do the work on each cpu */
2748 smp_call_function_many(tracing_buffer_mask,
2749 enable_trace_buffered_event, NULL, 1);
2750 preempt_enable();
2751}
2752
2753static struct trace_buffer *temp_buffer;
2754
2755struct ring_buffer_event *
2756trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2757 struct trace_event_file *trace_file,
2758 int type, unsigned long len,
2759 unsigned int trace_ctx)
2760{
2761 struct ring_buffer_event *entry;
2762 struct trace_array *tr = trace_file->tr;
2763 int val;
2764
2765 *current_rb = tr->array_buffer.buffer;
2766
2767 if (!tr->no_filter_buffering_ref &&
2768 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2769 preempt_disable_notrace();
2770 /*
2771 * Filtering is on, so try to use the per cpu buffer first.
2772 * This buffer will simulate a ring_buffer_event,
2773 * where the type_len is zero and the array[0] will
2774 * hold the full length.
2775 * (see include/linux/ring-buffer.h for details on
2776 * how the ring_buffer_event is structured).
2777 *
2778 * Using a temp buffer during filtering and copying it
2779 * on a matched filter is quicker than writing directly
2780 * into the ring buffer and then discarding it when
2781 * it doesn't match. That is because the discard
2782 * requires several atomic operations to get right.
2783 * Copying on match and doing nothing on a failed match
2784 * is still quicker than no copy on match, but having
2785 * to discard out of the ring buffer on a failed match.
2786 */
2787 if ((entry = __this_cpu_read(trace_buffered_event))) {
2788 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2789
2790 val = this_cpu_inc_return(trace_buffered_event_cnt);
2791
2792 /*
2793 * Preemption is disabled, but interrupts and NMIs
2794 * can still come in now. If that happens after
2795 * the above increment, then it will have to go
2796 * back to the old method of allocating the event
2797 * on the ring buffer, and if the filter fails, it
2798 * will have to call ring_buffer_discard_commit()
2799 * to remove it.
2800 *
2801 * Need to also check the unlikely case that the
2802 * length is bigger than the temp buffer size.
2803 * If that happens, then the reserve is pretty much
2804 * guaranteed to fail, as the ring buffer currently
2805 * only allows events less than a page. But that may
2806 * change in the future, so let the ring buffer reserve
2807 * handle the failure in that case.
2808 */
2809 if (val == 1 && likely(len <= max_len)) {
2810 trace_event_setup(entry, type, trace_ctx);
2811 entry->array[0] = len;
2812 /* Return with preemption disabled */
2813 return entry;
2814 }
2815 this_cpu_dec(trace_buffered_event_cnt);
2816 }
2817 /* __trace_buffer_lock_reserve() disables preemption */
2818 preempt_enable_notrace();
2819 }
2820
2821 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2822 trace_ctx);
2823 /*
2824 * If tracing is off, but we have triggers enabled
2825 * we still need to look at the event data. Use the temp_buffer
2826 * to store the trace event for the trigger to use. It's recursive
2827 * safe and will not be recorded anywhere.
2828 */
2829 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2830 *current_rb = temp_buffer;
2831 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2832 trace_ctx);
2833 }
2834 return entry;
2835}
2836EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2837
2838static DEFINE_SPINLOCK(tracepoint_iter_lock);
2839static DEFINE_MUTEX(tracepoint_printk_mutex);
2840
2841static void output_printk(struct trace_event_buffer *fbuffer)
2842{
2843 struct trace_event_call *event_call;
2844 struct trace_event_file *file;
2845 struct trace_event *event;
2846 unsigned long flags;
2847 struct trace_iterator *iter = tracepoint_print_iter;
2848
2849 /* We should never get here if iter is NULL */
2850 if (WARN_ON_ONCE(!iter))
2851 return;
2852
2853 event_call = fbuffer->trace_file->event_call;
2854 if (!event_call || !event_call->event.funcs ||
2855 !event_call->event.funcs->trace)
2856 return;
2857
2858 file = fbuffer->trace_file;
2859 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2860 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2861 !filter_match_preds(file->filter, fbuffer->entry)))
2862 return;
2863
2864 event = &fbuffer->trace_file->event_call->event;
2865
2866 spin_lock_irqsave(&tracepoint_iter_lock, flags);
2867 trace_seq_init(&iter->seq);
2868 iter->ent = fbuffer->entry;
2869 event_call->event.funcs->trace(iter, 0, event);
2870 trace_seq_putc(&iter->seq, 0);
2871 printk("%s", iter->seq.buffer);
2872
2873 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2874}
2875
2876int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2877 void *buffer, size_t *lenp,
2878 loff_t *ppos)
2879{
2880 int save_tracepoint_printk;
2881 int ret;
2882
2883 mutex_lock(&tracepoint_printk_mutex);
2884 save_tracepoint_printk = tracepoint_printk;
2885
2886 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2887
2888 /*
2889 * This will force exiting early, as tracepoint_printk
2890 * is always zero when tracepoint_printk_iter is not allocated
2891 */
2892 if (!tracepoint_print_iter)
2893 tracepoint_printk = 0;
2894
2895 if (save_tracepoint_printk == tracepoint_printk)
2896 goto out;
2897
2898 if (tracepoint_printk)
2899 static_key_enable(&tracepoint_printk_key.key);
2900 else
2901 static_key_disable(&tracepoint_printk_key.key);
2902
2903 out:
2904 mutex_unlock(&tracepoint_printk_mutex);
2905
2906 return ret;
2907}
2908
2909void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2910{
2911 enum event_trigger_type tt = ETT_NONE;
2912 struct trace_event_file *file = fbuffer->trace_file;
2913
2914 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2915 fbuffer->entry, &tt))
2916 goto discard;
2917
2918 if (static_key_false(&tracepoint_printk_key.key))
2919 output_printk(fbuffer);
2920
2921 if (static_branch_unlikely(&trace_event_exports_enabled))
2922 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2923
2924 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2925 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
2926
2927discard:
2928 if (tt)
2929 event_triggers_post_call(file, tt);
2930
2931}
2932EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2933
2934/*
2935 * Skip 3:
2936 *
2937 * trace_buffer_unlock_commit_regs()
2938 * trace_event_buffer_commit()
2939 * trace_event_raw_event_xxx()
2940 */
2941# define STACK_SKIP 3
2942
2943void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2944 struct trace_buffer *buffer,
2945 struct ring_buffer_event *event,
2946 unsigned int trace_ctx,
2947 struct pt_regs *regs)
2948{
2949 __buffer_unlock_commit(buffer, event);
2950
2951 /*
2952 * If regs is not set, then skip the necessary functions.
2953 * Note, we can still get here via blktrace, wakeup tracer
2954 * and mmiotrace, but that's ok if they lose a function or
2955 * two. They are not that meaningful.
2956 */
2957 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
2958 ftrace_trace_userstack(tr, buffer, trace_ctx);
2959}
2960
2961/*
2962 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2963 */
2964void
2965trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2966 struct ring_buffer_event *event)
2967{
2968 __buffer_unlock_commit(buffer, event);
2969}
2970
2971void
2972trace_function(struct trace_array *tr, unsigned long ip, unsigned long
2973 parent_ip, unsigned int trace_ctx)
2974{
2975 struct trace_event_call *call = &event_function;
2976 struct trace_buffer *buffer = tr->array_buffer.buffer;
2977 struct ring_buffer_event *event;
2978 struct ftrace_entry *entry;
2979
2980 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2981 trace_ctx);
2982 if (!event)
2983 return;
2984 entry = ring_buffer_event_data(event);
2985 entry->ip = ip;
2986 entry->parent_ip = parent_ip;
2987
2988 if (!call_filter_check_discard(call, entry, buffer, event)) {
2989 if (static_branch_unlikely(&trace_function_exports_enabled))
2990 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
2991 __buffer_unlock_commit(buffer, event);
2992 }
2993}
2994
2995#ifdef CONFIG_STACKTRACE
2996
2997/* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2998#define FTRACE_KSTACK_NESTING 4
2999
3000#define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3001
3002struct ftrace_stack {
3003 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3004};
3005
3006
3007struct ftrace_stacks {
3008 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3009};
3010
3011static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3012static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3013
3014static void __ftrace_trace_stack(struct trace_buffer *buffer,
3015 unsigned int trace_ctx,
3016 int skip, struct pt_regs *regs)
3017{
3018 struct trace_event_call *call = &event_kernel_stack;
3019 struct ring_buffer_event *event;
3020 unsigned int size, nr_entries;
3021 struct ftrace_stack *fstack;
3022 struct stack_entry *entry;
3023 int stackidx;
3024
3025 /*
3026 * Add one, for this function and the call to save_stack_trace()
3027 * If regs is set, then these functions will not be in the way.
3028 */
3029#ifndef CONFIG_UNWINDER_ORC
3030 if (!regs)
3031 skip++;
3032#endif
3033
3034 preempt_disable_notrace();
3035
3036 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3037
3038 /* This should never happen. If it does, yell once and skip */
3039 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3040 goto out;
3041
3042 /*
3043 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3044 * interrupt will either see the value pre increment or post
3045 * increment. If the interrupt happens pre increment it will have
3046 * restored the counter when it returns. We just need a barrier to
3047 * keep gcc from moving things around.
3048 */
3049 barrier();
3050
3051 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3052 size = ARRAY_SIZE(fstack->calls);
3053
3054 if (regs) {
3055 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3056 size, skip);
3057 } else {
3058 nr_entries = stack_trace_save(fstack->calls, size, skip);
3059 }
3060
3061 size = nr_entries * sizeof(unsigned long);
3062 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3063 (sizeof(*entry) - sizeof(entry->caller)) + size,
3064 trace_ctx);
3065 if (!event)
3066 goto out;
3067 entry = ring_buffer_event_data(event);
3068
3069 memcpy(&entry->caller, fstack->calls, size);
3070 entry->size = nr_entries;
3071
3072 if (!call_filter_check_discard(call, entry, buffer, event))
3073 __buffer_unlock_commit(buffer, event);
3074
3075 out:
3076 /* Again, don't let gcc optimize things here */
3077 barrier();
3078 __this_cpu_dec(ftrace_stack_reserve);
3079 preempt_enable_notrace();
3080
3081}
3082
3083static inline void ftrace_trace_stack(struct trace_array *tr,
3084 struct trace_buffer *buffer,
3085 unsigned int trace_ctx,
3086 int skip, struct pt_regs *regs)
3087{
3088 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3089 return;
3090
3091 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3092}
3093
3094void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3095 int skip)
3096{
3097 struct trace_buffer *buffer = tr->array_buffer.buffer;
3098
3099 if (rcu_is_watching()) {
3100 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3101 return;
3102 }
3103
3104 /*
3105 * When an NMI triggers, RCU is enabled via rcu_nmi_enter(),
3106 * but if the above rcu_is_watching() failed, then the NMI
3107 * triggered someplace critical, and rcu_irq_enter() should
3108 * not be called from NMI.
3109 */
3110 if (unlikely(in_nmi()))
3111 return;
3112
3113 rcu_irq_enter_irqson();
3114 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3115 rcu_irq_exit_irqson();
3116}
3117
3118/**
3119 * trace_dump_stack - record a stack back trace in the trace buffer
3120 * @skip: Number of functions to skip (helper handlers)
3121 */
3122void trace_dump_stack(int skip)
3123{
3124 if (tracing_disabled || tracing_selftest_running)
3125 return;
3126
3127#ifndef CONFIG_UNWINDER_ORC
3128 /* Skip 1 to skip this function. */
3129 skip++;
3130#endif
3131 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3132 tracing_gen_ctx(), skip, NULL);
3133}
3134EXPORT_SYMBOL_GPL(trace_dump_stack);
3135
3136#ifdef CONFIG_USER_STACKTRACE_SUPPORT
3137static DEFINE_PER_CPU(int, user_stack_count);
3138
3139static void
3140ftrace_trace_userstack(struct trace_array *tr,
3141 struct trace_buffer *buffer, unsigned int trace_ctx)
3142{
3143 struct trace_event_call *call = &event_user_stack;
3144 struct ring_buffer_event *event;
3145 struct userstack_entry *entry;
3146
3147 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3148 return;
3149
3150 /*
3151 * NMIs can not handle page faults, even with fix ups.
3152 * The save user stack can (and often does) fault.
3153 */
3154 if (unlikely(in_nmi()))
3155 return;
3156
3157 /*
3158 * prevent recursion, since the user stack tracing may
3159 * trigger other kernel events.
3160 */
3161 preempt_disable();
3162 if (__this_cpu_read(user_stack_count))
3163 goto out;
3164
3165 __this_cpu_inc(user_stack_count);
3166
3167 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3168 sizeof(*entry), trace_ctx);
3169 if (!event)
3170 goto out_drop_count;
3171 entry = ring_buffer_event_data(event);
3172
3173 entry->tgid = current->tgid;
3174 memset(&entry->caller, 0, sizeof(entry->caller));
3175
3176 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3177 if (!call_filter_check_discard(call, entry, buffer, event))
3178 __buffer_unlock_commit(buffer, event);
3179
3180 out_drop_count:
3181 __this_cpu_dec(user_stack_count);
3182 out:
3183 preempt_enable();
3184}
3185#else /* CONFIG_USER_STACKTRACE_SUPPORT */
3186static void ftrace_trace_userstack(struct trace_array *tr,
3187 struct trace_buffer *buffer,
3188 unsigned int trace_ctx)
3189{
3190}
3191#endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3192
3193#endif /* CONFIG_STACKTRACE */
3194
3195static inline void
3196func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3197 unsigned long long delta)
3198{
3199 entry->bottom_delta_ts = delta & U32_MAX;
3200 entry->top_delta_ts = (delta >> 32);
3201}
3202
3203void trace_last_func_repeats(struct trace_array *tr,
3204 struct trace_func_repeats *last_info,
3205 unsigned int trace_ctx)
3206{
3207 struct trace_buffer *buffer = tr->array_buffer.buffer;
3208 struct func_repeats_entry *entry;
3209 struct ring_buffer_event *event;
3210 u64 delta;
3211
3212 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3213 sizeof(*entry), trace_ctx);
3214 if (!event)
3215 return;
3216
3217 delta = ring_buffer_event_time_stamp(buffer, event) -
3218 last_info->ts_last_call;
3219
3220 entry = ring_buffer_event_data(event);
3221 entry->ip = last_info->ip;
3222 entry->parent_ip = last_info->parent_ip;
3223 entry->count = last_info->count;
3224 func_repeats_set_delta_ts(entry, delta);
3225
3226 __buffer_unlock_commit(buffer, event);
3227}
3228
3229/* created for use with alloc_percpu */
3230struct trace_buffer_struct {
3231 int nesting;
3232 char buffer[4][TRACE_BUF_SIZE];
3233};
3234
3235static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3236
3237/*
3238 * This allows for lockless recording. If we're nested too deeply, then
3239 * this returns NULL.
3240 */
3241static char *get_trace_buf(void)
3242{
3243 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3244
3245 if (!trace_percpu_buffer || buffer->nesting >= 4)
3246 return NULL;
3247
3248 buffer->nesting++;
3249
3250 /* Interrupts must see nesting incremented before we use the buffer */
3251 barrier();
3252 return &buffer->buffer[buffer->nesting - 1][0];
3253}
3254
3255static void put_trace_buf(void)
3256{
3257 /* Don't let the decrement of nesting leak before this */
3258 barrier();
3259 this_cpu_dec(trace_percpu_buffer->nesting);
3260}
3261
3262static int alloc_percpu_trace_buffer(void)
3263{
3264 struct trace_buffer_struct __percpu *buffers;
3265
3266 if (trace_percpu_buffer)
3267 return 0;
3268
3269 buffers = alloc_percpu(struct trace_buffer_struct);
3270 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3271 return -ENOMEM;
3272
3273 trace_percpu_buffer = buffers;
3274 return 0;
3275}
3276
3277static int buffers_allocated;
3278
3279void trace_printk_init_buffers(void)
3280{
3281 if (buffers_allocated)
3282 return;
3283
3284 if (alloc_percpu_trace_buffer())
3285 return;
3286
3287 /* trace_printk() is for debug use only. Don't use it in production. */
3288
3289 pr_warn("\n");
3290 pr_warn("**********************************************************\n");
3291 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3292 pr_warn("** **\n");
3293 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3294 pr_warn("** **\n");
3295 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3296 pr_warn("** unsafe for production use. **\n");
3297 pr_warn("** **\n");
3298 pr_warn("** If you see this message and you are not debugging **\n");
3299 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3300 pr_warn("** **\n");
3301 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3302 pr_warn("**********************************************************\n");
3303
3304 /* Expand the buffers to set size */
3305 tracing_update_buffers();
3306
3307 buffers_allocated = 1;
3308
3309 /*
3310 * trace_printk_init_buffers() can be called by modules.
3311 * If that happens, then we need to start cmdline recording
3312 * directly here. If the global_trace.buffer is already
3313 * allocated here, then this was called by module code.
3314 */
3315 if (global_trace.array_buffer.buffer)
3316 tracing_start_cmdline_record();
3317}
3318EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3319
3320void trace_printk_start_comm(void)
3321{
3322 /* Start tracing comms if trace printk is set */
3323 if (!buffers_allocated)
3324 return;
3325 tracing_start_cmdline_record();
3326}
3327
3328static void trace_printk_start_stop_comm(int enabled)
3329{
3330 if (!buffers_allocated)
3331 return;
3332
3333 if (enabled)
3334 tracing_start_cmdline_record();
3335 else
3336 tracing_stop_cmdline_record();
3337}
3338
3339/**
3340 * trace_vbprintk - write binary msg to tracing buffer
3341 * @ip: The address of the caller
3342 * @fmt: The string format to write to the buffer
3343 * @args: Arguments for @fmt
3344 */
3345int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3346{
3347 struct trace_event_call *call = &event_bprint;
3348 struct ring_buffer_event *event;
3349 struct trace_buffer *buffer;
3350 struct trace_array *tr = &global_trace;
3351 struct bprint_entry *entry;
3352 unsigned int trace_ctx;
3353 char *tbuffer;
3354 int len = 0, size;
3355
3356 if (unlikely(tracing_selftest_running || tracing_disabled))
3357 return 0;
3358
3359 /* Don't pollute graph traces with trace_vprintk internals */
3360 pause_graph_tracing();
3361
3362 trace_ctx = tracing_gen_ctx();
3363 preempt_disable_notrace();
3364
3365 tbuffer = get_trace_buf();
3366 if (!tbuffer) {
3367 len = 0;
3368 goto out_nobuffer;
3369 }
3370
3371 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3372
3373 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3374 goto out_put;
3375
3376 size = sizeof(*entry) + sizeof(u32) * len;
3377 buffer = tr->array_buffer.buffer;
3378 ring_buffer_nest_start(buffer);
3379 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3380 trace_ctx);
3381 if (!event)
3382 goto out;
3383 entry = ring_buffer_event_data(event);
3384 entry->ip = ip;
3385 entry->fmt = fmt;
3386
3387 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3388 if (!call_filter_check_discard(call, entry, buffer, event)) {
3389 __buffer_unlock_commit(buffer, event);
3390 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3391 }
3392
3393out:
3394 ring_buffer_nest_end(buffer);
3395out_put:
3396 put_trace_buf();
3397
3398out_nobuffer:
3399 preempt_enable_notrace();
3400 unpause_graph_tracing();
3401
3402 return len;
3403}
3404EXPORT_SYMBOL_GPL(trace_vbprintk);
3405
3406__printf(3, 0)
3407static int
3408__trace_array_vprintk(struct trace_buffer *buffer,
3409 unsigned long ip, const char *fmt, va_list args)
3410{
3411 struct trace_event_call *call = &event_print;
3412 struct ring_buffer_event *event;
3413 int len = 0, size;
3414 struct print_entry *entry;
3415 unsigned int trace_ctx;
3416 char *tbuffer;
3417
3418 if (tracing_disabled || tracing_selftest_running)
3419 return 0;
3420
3421 /* Don't pollute graph traces with trace_vprintk internals */
3422 pause_graph_tracing();
3423
3424 trace_ctx = tracing_gen_ctx();
3425 preempt_disable_notrace();
3426
3427
3428 tbuffer = get_trace_buf();
3429 if (!tbuffer) {
3430 len = 0;
3431 goto out_nobuffer;
3432 }
3433
3434 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3435
3436 size = sizeof(*entry) + len + 1;
3437 ring_buffer_nest_start(buffer);
3438 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3439 trace_ctx);
3440 if (!event)
3441 goto out;
3442 entry = ring_buffer_event_data(event);
3443 entry->ip = ip;
3444
3445 memcpy(&entry->buf, tbuffer, len + 1);
3446 if (!call_filter_check_discard(call, entry, buffer, event)) {
3447 __buffer_unlock_commit(buffer, event);
3448 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3449 }
3450
3451out:
3452 ring_buffer_nest_end(buffer);
3453 put_trace_buf();
3454
3455out_nobuffer:
3456 preempt_enable_notrace();
3457 unpause_graph_tracing();
3458
3459 return len;
3460}
3461
3462__printf(3, 0)
3463int trace_array_vprintk(struct trace_array *tr,
3464 unsigned long ip, const char *fmt, va_list args)
3465{
3466 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3467}
3468
3469/**
3470 * trace_array_printk - Print a message to a specific instance
3471 * @tr: The instance trace_array descriptor
3472 * @ip: The instruction pointer that this is called from.
3473 * @fmt: The format to print (printf format)
3474 *
3475 * If a subsystem sets up its own instance, they have the right to
3476 * printk strings into their tracing instance buffer using this
3477 * function. Note, this function will not write into the top level
3478 * buffer (use trace_printk() for that), as writing into the top level
3479 * buffer should only have events that can be individually disabled.
3480 * trace_printk() is only used for debugging a kernel, and should not
3481 * be ever incorporated in normal use.
3482 *
3483 * trace_array_printk() can be used, as it will not add noise to the
3484 * top level tracing buffer.
3485 *
3486 * Note, trace_array_init_printk() must be called on @tr before this
3487 * can be used.
3488 */
3489__printf(3, 0)
3490int trace_array_printk(struct trace_array *tr,
3491 unsigned long ip, const char *fmt, ...)
3492{
3493 int ret;
3494 va_list ap;
3495
3496 if (!tr)
3497 return -ENOENT;
3498
3499 /* This is only allowed for created instances */
3500 if (tr == &global_trace)
3501 return 0;
3502
3503 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3504 return 0;
3505
3506 va_start(ap, fmt);
3507 ret = trace_array_vprintk(tr, ip, fmt, ap);
3508 va_end(ap);
3509 return ret;
3510}
3511EXPORT_SYMBOL_GPL(trace_array_printk);
3512
3513/**
3514 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3515 * @tr: The trace array to initialize the buffers for
3516 *
3517 * As trace_array_printk() only writes into instances, they are OK to
3518 * have in the kernel (unlike trace_printk()). This needs to be called
3519 * before trace_array_printk() can be used on a trace_array.
3520 */
3521int trace_array_init_printk(struct trace_array *tr)
3522{
3523 if (!tr)
3524 return -ENOENT;
3525
3526 /* This is only allowed for created instances */
3527 if (tr == &global_trace)
3528 return -EINVAL;
3529
3530 return alloc_percpu_trace_buffer();
3531}
3532EXPORT_SYMBOL_GPL(trace_array_init_printk);
3533
3534__printf(3, 4)
3535int trace_array_printk_buf(struct trace_buffer *buffer,
3536 unsigned long ip, const char *fmt, ...)
3537{
3538 int ret;
3539 va_list ap;
3540
3541 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3542 return 0;
3543
3544 va_start(ap, fmt);
3545 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3546 va_end(ap);
3547 return ret;
3548}
3549
3550__printf(2, 0)
3551int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3552{
3553 return trace_array_vprintk(&global_trace, ip, fmt, args);
3554}
3555EXPORT_SYMBOL_GPL(trace_vprintk);
3556
3557static void trace_iterator_increment(struct trace_iterator *iter)
3558{
3559 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3560
3561 iter->idx++;
3562 if (buf_iter)
3563 ring_buffer_iter_advance(buf_iter);
3564}
3565
3566static struct trace_entry *
3567peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3568 unsigned long *lost_events)
3569{
3570 struct ring_buffer_event *event;
3571 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3572
3573 if (buf_iter) {
3574 event = ring_buffer_iter_peek(buf_iter, ts);
3575 if (lost_events)
3576 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3577 (unsigned long)-1 : 0;
3578 } else {
3579 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3580 lost_events);
3581 }
3582
3583 if (event) {
3584 iter->ent_size = ring_buffer_event_length(event);
3585 return ring_buffer_event_data(event);
3586 }
3587 iter->ent_size = 0;
3588 return NULL;
3589}
3590
3591static struct trace_entry *
3592__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3593 unsigned long *missing_events, u64 *ent_ts)
3594{
3595 struct trace_buffer *buffer = iter->array_buffer->buffer;
3596 struct trace_entry *ent, *next = NULL;
3597 unsigned long lost_events = 0, next_lost = 0;
3598 int cpu_file = iter->cpu_file;
3599 u64 next_ts = 0, ts;
3600 int next_cpu = -1;
3601 int next_size = 0;
3602 int cpu;
3603
3604 /*
3605 * If we are in a per_cpu trace file, don't bother by iterating over
3606 * all cpu and peek directly.
3607 */
3608 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3609 if (ring_buffer_empty_cpu(buffer, cpu_file))
3610 return NULL;
3611 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3612 if (ent_cpu)
3613 *ent_cpu = cpu_file;
3614
3615 return ent;
3616 }
3617
3618 for_each_tracing_cpu(cpu) {
3619
3620 if (ring_buffer_empty_cpu(buffer, cpu))
3621 continue;
3622
3623 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3624
3625 /*
3626 * Pick the entry with the smallest timestamp:
3627 */
3628 if (ent && (!next || ts < next_ts)) {
3629 next = ent;
3630 next_cpu = cpu;
3631 next_ts = ts;
3632 next_lost = lost_events;
3633 next_size = iter->ent_size;
3634 }
3635 }
3636
3637 iter->ent_size = next_size;
3638
3639 if (ent_cpu)
3640 *ent_cpu = next_cpu;
3641
3642 if (ent_ts)
3643 *ent_ts = next_ts;
3644
3645 if (missing_events)
3646 *missing_events = next_lost;
3647
3648 return next;
3649}
3650
3651#define STATIC_FMT_BUF_SIZE 128
3652static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3653
3654static char *trace_iter_expand_format(struct trace_iterator *iter)
3655{
3656 char *tmp;
3657
3658 /*
3659 * iter->tr is NULL when used with tp_printk, which makes
3660 * this get called where it is not safe to call krealloc().
3661 */
3662 if (!iter->tr || iter->fmt == static_fmt_buf)
3663 return NULL;
3664
3665 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3666 GFP_KERNEL);
3667 if (tmp) {
3668 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3669 iter->fmt = tmp;
3670 }
3671
3672 return tmp;
3673}
3674
3675/* Returns true if the string is safe to dereference from an event */
3676static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3677 bool star, int len)
3678{
3679 unsigned long addr = (unsigned long)str;
3680 struct trace_event *trace_event;
3681 struct trace_event_call *event;
3682
3683 /* Ignore strings with no length */
3684 if (star && !len)
3685 return true;
3686
3687 /* OK if part of the event data */
3688 if ((addr >= (unsigned long)iter->ent) &&
3689 (addr < (unsigned long)iter->ent + iter->ent_size))
3690 return true;
3691
3692 /* OK if part of the temp seq buffer */
3693 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3694 (addr < (unsigned long)iter->tmp_seq.buffer + PAGE_SIZE))
3695 return true;
3696
3697 /* Core rodata can not be freed */
3698 if (is_kernel_rodata(addr))
3699 return true;
3700
3701 if (trace_is_tracepoint_string(str))
3702 return true;
3703
3704 /*
3705 * Now this could be a module event, referencing core module
3706 * data, which is OK.
3707 */
3708 if (!iter->ent)
3709 return false;
3710
3711 trace_event = ftrace_find_event(iter->ent->type);
3712 if (!trace_event)
3713 return false;
3714
3715 event = container_of(trace_event, struct trace_event_call, event);
3716 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3717 return false;
3718
3719 /* Would rather have rodata, but this will suffice */
3720 if (within_module_core(addr, event->module))
3721 return true;
3722
3723 return false;
3724}
3725
3726static const char *show_buffer(struct trace_seq *s)
3727{
3728 struct seq_buf *seq = &s->seq;
3729
3730 seq_buf_terminate(seq);
3731
3732 return seq->buffer;
3733}
3734
3735static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3736
3737static int test_can_verify_check(const char *fmt, ...)
3738{
3739 char buf[16];
3740 va_list ap;
3741 int ret;
3742
3743 /*
3744 * The verifier is dependent on vsnprintf() modifies the va_list
3745 * passed to it, where it is sent as a reference. Some architectures
3746 * (like x86_32) passes it by value, which means that vsnprintf()
3747 * does not modify the va_list passed to it, and the verifier
3748 * would then need to be able to understand all the values that
3749 * vsnprintf can use. If it is passed by value, then the verifier
3750 * is disabled.
3751 */
3752 va_start(ap, fmt);
3753 vsnprintf(buf, 16, "%d", ap);
3754 ret = va_arg(ap, int);
3755 va_end(ap);
3756
3757 return ret;
3758}
3759
3760static void test_can_verify(void)
3761{
3762 if (!test_can_verify_check("%d %d", 0, 1)) {
3763 pr_info("trace event string verifier disabled\n");
3764 static_branch_inc(&trace_no_verify);
3765 }
3766}
3767
3768/**
3769 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3770 * @iter: The iterator that holds the seq buffer and the event being printed
3771 * @fmt: The format used to print the event
3772 * @ap: The va_list holding the data to print from @fmt.
3773 *
3774 * This writes the data into the @iter->seq buffer using the data from
3775 * @fmt and @ap. If the format has a %s, then the source of the string
3776 * is examined to make sure it is safe to print, otherwise it will
3777 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3778 * pointer.
3779 */
3780void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3781 va_list ap)
3782{
3783 const char *p = fmt;
3784 const char *str;
3785 int i, j;
3786
3787 if (WARN_ON_ONCE(!fmt))
3788 return;
3789
3790 if (static_branch_unlikely(&trace_no_verify))
3791 goto print;
3792
3793 /* Don't bother checking when doing a ftrace_dump() */
3794 if (iter->fmt == static_fmt_buf)
3795 goto print;
3796
3797 while (*p) {
3798 bool star = false;
3799 int len = 0;
3800
3801 j = 0;
3802
3803 /* We only care about %s and variants */
3804 for (i = 0; p[i]; i++) {
3805 if (i + 1 >= iter->fmt_size) {
3806 /*
3807 * If we can't expand the copy buffer,
3808 * just print it.
3809 */
3810 if (!trace_iter_expand_format(iter))
3811 goto print;
3812 }
3813
3814 if (p[i] == '\\' && p[i+1]) {
3815 i++;
3816 continue;
3817 }
3818 if (p[i] == '%') {
3819 /* Need to test cases like %08.*s */
3820 for (j = 1; p[i+j]; j++) {
3821 if (isdigit(p[i+j]) ||
3822 p[i+j] == '.')
3823 continue;
3824 if (p[i+j] == '*') {
3825 star = true;
3826 continue;
3827 }
3828 break;
3829 }
3830 if (p[i+j] == 's')
3831 break;
3832 star = false;
3833 }
3834 j = 0;
3835 }
3836 /* If no %s found then just print normally */
3837 if (!p[i])
3838 break;
3839
3840 /* Copy up to the %s, and print that */
3841 strncpy(iter->fmt, p, i);
3842 iter->fmt[i] = '\0';
3843 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3844
3845 /*
3846 * If iter->seq is full, the above call no longer guarantees
3847 * that ap is in sync with fmt processing, and further calls
3848 * to va_arg() can return wrong positional arguments.
3849 *
3850 * Ensure that ap is no longer used in this case.
3851 */
3852 if (iter->seq.full) {
3853 p = "";
3854 break;
3855 }
3856
3857 if (star)
3858 len = va_arg(ap, int);
3859
3860 /* The ap now points to the string data of the %s */
3861 str = va_arg(ap, const char *);
3862
3863 /*
3864 * If you hit this warning, it is likely that the
3865 * trace event in question used %s on a string that
3866 * was saved at the time of the event, but may not be
3867 * around when the trace is read. Use __string(),
3868 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3869 * instead. See samples/trace_events/trace-events-sample.h
3870 * for reference.
3871 */
3872 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3873 "fmt: '%s' current_buffer: '%s'",
3874 fmt, show_buffer(&iter->seq))) {
3875 int ret;
3876
3877 /* Try to safely read the string */
3878 if (star) {
3879 if (len + 1 > iter->fmt_size)
3880 len = iter->fmt_size - 1;
3881 if (len < 0)
3882 len = 0;
3883 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3884 iter->fmt[len] = 0;
3885 star = false;
3886 } else {
3887 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3888 iter->fmt_size);
3889 }
3890 if (ret < 0)
3891 trace_seq_printf(&iter->seq, "(0x%px)", str);
3892 else
3893 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3894 str, iter->fmt);
3895 str = "[UNSAFE-MEMORY]";
3896 strcpy(iter->fmt, "%s");
3897 } else {
3898 strncpy(iter->fmt, p + i, j + 1);
3899 iter->fmt[j+1] = '\0';
3900 }
3901 if (star)
3902 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3903 else
3904 trace_seq_printf(&iter->seq, iter->fmt, str);
3905
3906 p += i + j + 1;
3907 }
3908 print:
3909 if (*p)
3910 trace_seq_vprintf(&iter->seq, p, ap);
3911}
3912
3913const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3914{
3915 const char *p, *new_fmt;
3916 char *q;
3917
3918 if (WARN_ON_ONCE(!fmt))
3919 return fmt;
3920
3921 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
3922 return fmt;
3923
3924 p = fmt;
3925 new_fmt = q = iter->fmt;
3926 while (*p) {
3927 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
3928 if (!trace_iter_expand_format(iter))
3929 return fmt;
3930
3931 q += iter->fmt - new_fmt;
3932 new_fmt = iter->fmt;
3933 }
3934
3935 *q++ = *p++;
3936
3937 /* Replace %p with %px */
3938 if (p[-1] == '%') {
3939 if (p[0] == '%') {
3940 *q++ = *p++;
3941 } else if (p[0] == 'p' && !isalnum(p[1])) {
3942 *q++ = *p++;
3943 *q++ = 'x';
3944 }
3945 }
3946 }
3947 *q = '\0';
3948
3949 return new_fmt;
3950}
3951
3952#define STATIC_TEMP_BUF_SIZE 128
3953static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3954
3955/* Find the next real entry, without updating the iterator itself */
3956struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3957 int *ent_cpu, u64 *ent_ts)
3958{
3959 /* __find_next_entry will reset ent_size */
3960 int ent_size = iter->ent_size;
3961 struct trace_entry *entry;
3962
3963 /*
3964 * If called from ftrace_dump(), then the iter->temp buffer
3965 * will be the static_temp_buf and not created from kmalloc.
3966 * If the entry size is greater than the buffer, we can
3967 * not save it. Just return NULL in that case. This is only
3968 * used to add markers when two consecutive events' time
3969 * stamps have a large delta. See trace_print_lat_context()
3970 */
3971 if (iter->temp == static_temp_buf &&
3972 STATIC_TEMP_BUF_SIZE < ent_size)
3973 return NULL;
3974
3975 /*
3976 * The __find_next_entry() may call peek_next_entry(), which may
3977 * call ring_buffer_peek() that may make the contents of iter->ent
3978 * undefined. Need to copy iter->ent now.
3979 */
3980 if (iter->ent && iter->ent != iter->temp) {
3981 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3982 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3983 void *temp;
3984 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3985 if (!temp)
3986 return NULL;
3987 kfree(iter->temp);
3988 iter->temp = temp;
3989 iter->temp_size = iter->ent_size;
3990 }
3991 memcpy(iter->temp, iter->ent, iter->ent_size);
3992 iter->ent = iter->temp;
3993 }
3994 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3995 /* Put back the original ent_size */
3996 iter->ent_size = ent_size;
3997
3998 return entry;
3999}
4000
4001/* Find the next real entry, and increment the iterator to the next entry */
4002void *trace_find_next_entry_inc(struct trace_iterator *iter)
4003{
4004 iter->ent = __find_next_entry(iter, &iter->cpu,
4005 &iter->lost_events, &iter->ts);
4006
4007 if (iter->ent)
4008 trace_iterator_increment(iter);
4009
4010 return iter->ent ? iter : NULL;
4011}
4012
4013static void trace_consume(struct trace_iterator *iter)
4014{
4015 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4016 &iter->lost_events);
4017}
4018
4019static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4020{
4021 struct trace_iterator *iter = m->private;
4022 int i = (int)*pos;
4023 void *ent;
4024
4025 WARN_ON_ONCE(iter->leftover);
4026
4027 (*pos)++;
4028
4029 /* can't go backwards */
4030 if (iter->idx > i)
4031 return NULL;
4032
4033 if (iter->idx < 0)
4034 ent = trace_find_next_entry_inc(iter);
4035 else
4036 ent = iter;
4037
4038 while (ent && iter->idx < i)
4039 ent = trace_find_next_entry_inc(iter);
4040
4041 iter->pos = *pos;
4042
4043 return ent;
4044}
4045
4046void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4047{
4048 struct ring_buffer_iter *buf_iter;
4049 unsigned long entries = 0;
4050 u64 ts;
4051
4052 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4053
4054 buf_iter = trace_buffer_iter(iter, cpu);
4055 if (!buf_iter)
4056 return;
4057
4058 ring_buffer_iter_reset(buf_iter);
4059
4060 /*
4061 * We could have the case with the max latency tracers
4062 * that a reset never took place on a cpu. This is evident
4063 * by the timestamp being before the start of the buffer.
4064 */
4065 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4066 if (ts >= iter->array_buffer->time_start)
4067 break;
4068 entries++;
4069 ring_buffer_iter_advance(buf_iter);
4070 }
4071
4072 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4073}
4074
4075/*
4076 * The current tracer is copied to avoid a global locking
4077 * all around.
4078 */
4079static void *s_start(struct seq_file *m, loff_t *pos)
4080{
4081 struct trace_iterator *iter = m->private;
4082 struct trace_array *tr = iter->tr;
4083 int cpu_file = iter->cpu_file;
4084 void *p = NULL;
4085 loff_t l = 0;
4086 int cpu;
4087
4088 /*
4089 * copy the tracer to avoid using a global lock all around.
4090 * iter->trace is a copy of current_trace, the pointer to the
4091 * name may be used instead of a strcmp(), as iter->trace->name
4092 * will point to the same string as current_trace->name.
4093 */
4094 mutex_lock(&trace_types_lock);
4095 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
4096 *iter->trace = *tr->current_trace;
4097 mutex_unlock(&trace_types_lock);
4098
4099#ifdef CONFIG_TRACER_MAX_TRACE
4100 if (iter->snapshot && iter->trace->use_max_tr)
4101 return ERR_PTR(-EBUSY);
4102#endif
4103
4104 if (*pos != iter->pos) {
4105 iter->ent = NULL;
4106 iter->cpu = 0;
4107 iter->idx = -1;
4108
4109 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4110 for_each_tracing_cpu(cpu)
4111 tracing_iter_reset(iter, cpu);
4112 } else
4113 tracing_iter_reset(iter, cpu_file);
4114
4115 iter->leftover = 0;
4116 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4117 ;
4118
4119 } else {
4120 /*
4121 * If we overflowed the seq_file before, then we want
4122 * to just reuse the trace_seq buffer again.
4123 */
4124 if (iter->leftover)
4125 p = iter;
4126 else {
4127 l = *pos - 1;
4128 p = s_next(m, p, &l);
4129 }
4130 }
4131
4132 trace_event_read_lock();
4133 trace_access_lock(cpu_file);
4134 return p;
4135}
4136
4137static void s_stop(struct seq_file *m, void *p)
4138{
4139 struct trace_iterator *iter = m->private;
4140
4141#ifdef CONFIG_TRACER_MAX_TRACE
4142 if (iter->snapshot && iter->trace->use_max_tr)
4143 return;
4144#endif
4145
4146 trace_access_unlock(iter->cpu_file);
4147 trace_event_read_unlock();
4148}
4149
4150static void
4151get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4152 unsigned long *entries, int cpu)
4153{
4154 unsigned long count;
4155
4156 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4157 /*
4158 * If this buffer has skipped entries, then we hold all
4159 * entries for the trace and we need to ignore the
4160 * ones before the time stamp.
4161 */
4162 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4163 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4164 /* total is the same as the entries */
4165 *total = count;
4166 } else
4167 *total = count +
4168 ring_buffer_overrun_cpu(buf->buffer, cpu);
4169 *entries = count;
4170}
4171
4172static void
4173get_total_entries(struct array_buffer *buf,
4174 unsigned long *total, unsigned long *entries)
4175{
4176 unsigned long t, e;
4177 int cpu;
4178
4179 *total = 0;
4180 *entries = 0;
4181
4182 for_each_tracing_cpu(cpu) {
4183 get_total_entries_cpu(buf, &t, &e, cpu);
4184 *total += t;
4185 *entries += e;
4186 }
4187}
4188
4189unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4190{
4191 unsigned long total, entries;
4192
4193 if (!tr)
4194 tr = &global_trace;
4195
4196 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4197
4198 return entries;
4199}
4200
4201unsigned long trace_total_entries(struct trace_array *tr)
4202{
4203 unsigned long total, entries;
4204
4205 if (!tr)
4206 tr = &global_trace;
4207
4208 get_total_entries(&tr->array_buffer, &total, &entries);
4209
4210 return entries;
4211}
4212
4213static void print_lat_help_header(struct seq_file *m)
4214{
4215 seq_puts(m, "# _------=> CPU# \n"
4216 "# / _-----=> irqs-off/BH-disabled\n"
4217 "# | / _----=> need-resched \n"
4218 "# || / _---=> hardirq/softirq \n"
4219 "# ||| / _--=> preempt-depth \n"
4220 "# |||| / _-=> migrate-disable \n"
4221 "# ||||| / delay \n"
4222 "# cmd pid |||||| time | caller \n"
4223 "# \\ / |||||| \\ | / \n");
4224}
4225
4226static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4227{
4228 unsigned long total;
4229 unsigned long entries;
4230
4231 get_total_entries(buf, &total, &entries);
4232 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4233 entries, total, num_online_cpus());
4234 seq_puts(m, "#\n");
4235}
4236
4237static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4238 unsigned int flags)
4239{
4240 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4241
4242 print_event_info(buf, m);
4243
4244 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4245 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4246}
4247
4248static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4249 unsigned int flags)
4250{
4251 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4252 const char *space = " ";
4253 int prec = tgid ? 12 : 2;
4254
4255 print_event_info(buf, m);
4256
4257 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4258 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4259 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4260 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4261 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4262 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4263 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4264 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4265}
4266
4267void
4268print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4269{
4270 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4271 struct array_buffer *buf = iter->array_buffer;
4272 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4273 struct tracer *type = iter->trace;
4274 unsigned long entries;
4275 unsigned long total;
4276 const char *name = "preemption";
4277
4278 name = type->name;
4279
4280 get_total_entries(buf, &total, &entries);
4281
4282 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4283 name, UTS_RELEASE);
4284 seq_puts(m, "# -----------------------------------"
4285 "---------------------------------\n");
4286 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4287 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4288 nsecs_to_usecs(data->saved_latency),
4289 entries,
4290 total,
4291 buf->cpu,
4292#if defined(CONFIG_PREEMPT_NONE)
4293 "server",
4294#elif defined(CONFIG_PREEMPT_VOLUNTARY)
4295 "desktop",
4296#elif defined(CONFIG_PREEMPT)
4297 "preempt",
4298#elif defined(CONFIG_PREEMPT_RT)
4299 "preempt_rt",
4300#else
4301 "unknown",
4302#endif
4303 /* These are reserved for later use */
4304 0, 0, 0, 0);
4305#ifdef CONFIG_SMP
4306 seq_printf(m, " #P:%d)\n", num_online_cpus());
4307#else
4308 seq_puts(m, ")\n");
4309#endif
4310 seq_puts(m, "# -----------------\n");
4311 seq_printf(m, "# | task: %.16s-%d "
4312 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4313 data->comm, data->pid,
4314 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4315 data->policy, data->rt_priority);
4316 seq_puts(m, "# -----------------\n");
4317
4318 if (data->critical_start) {
4319 seq_puts(m, "# => started at: ");
4320 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4321 trace_print_seq(m, &iter->seq);
4322 seq_puts(m, "\n# => ended at: ");
4323 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4324 trace_print_seq(m, &iter->seq);
4325 seq_puts(m, "\n#\n");
4326 }
4327
4328 seq_puts(m, "#\n");
4329}
4330
4331static void test_cpu_buff_start(struct trace_iterator *iter)
4332{
4333 struct trace_seq *s = &iter->seq;
4334 struct trace_array *tr = iter->tr;
4335
4336 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4337 return;
4338
4339 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4340 return;
4341
4342 if (cpumask_available(iter->started) &&
4343 cpumask_test_cpu(iter->cpu, iter->started))
4344 return;
4345
4346 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4347 return;
4348
4349 if (cpumask_available(iter->started))
4350 cpumask_set_cpu(iter->cpu, iter->started);
4351
4352 /* Don't print started cpu buffer for the first entry of the trace */
4353 if (iter->idx > 1)
4354 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4355 iter->cpu);
4356}
4357
4358static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4359{
4360 struct trace_array *tr = iter->tr;
4361 struct trace_seq *s = &iter->seq;
4362 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4363 struct trace_entry *entry;
4364 struct trace_event *event;
4365
4366 entry = iter->ent;
4367
4368 test_cpu_buff_start(iter);
4369
4370 event = ftrace_find_event(entry->type);
4371
4372 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4373 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4374 trace_print_lat_context(iter);
4375 else
4376 trace_print_context(iter);
4377 }
4378
4379 if (trace_seq_has_overflowed(s))
4380 return TRACE_TYPE_PARTIAL_LINE;
4381
4382 if (event)
4383 return event->funcs->trace(iter, sym_flags, event);
4384
4385 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4386
4387 return trace_handle_return(s);
4388}
4389
4390static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4391{
4392 struct trace_array *tr = iter->tr;
4393 struct trace_seq *s = &iter->seq;
4394 struct trace_entry *entry;
4395 struct trace_event *event;
4396
4397 entry = iter->ent;
4398
4399 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4400 trace_seq_printf(s, "%d %d %llu ",
4401 entry->pid, iter->cpu, iter->ts);
4402
4403 if (trace_seq_has_overflowed(s))
4404 return TRACE_TYPE_PARTIAL_LINE;
4405
4406 event = ftrace_find_event(entry->type);
4407 if (event)
4408 return event->funcs->raw(iter, 0, event);
4409
4410 trace_seq_printf(s, "%d ?\n", entry->type);
4411
4412 return trace_handle_return(s);
4413}
4414
4415static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4416{
4417 struct trace_array *tr = iter->tr;
4418 struct trace_seq *s = &iter->seq;
4419 unsigned char newline = '\n';
4420 struct trace_entry *entry;
4421 struct trace_event *event;
4422
4423 entry = iter->ent;
4424
4425 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4426 SEQ_PUT_HEX_FIELD(s, entry->pid);
4427 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4428 SEQ_PUT_HEX_FIELD(s, iter->ts);
4429 if (trace_seq_has_overflowed(s))
4430 return TRACE_TYPE_PARTIAL_LINE;
4431 }
4432
4433 event = ftrace_find_event(entry->type);
4434 if (event) {
4435 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4436 if (ret != TRACE_TYPE_HANDLED)
4437 return ret;
4438 }
4439
4440 SEQ_PUT_FIELD(s, newline);
4441
4442 return trace_handle_return(s);
4443}
4444
4445static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4446{
4447 struct trace_array *tr = iter->tr;
4448 struct trace_seq *s = &iter->seq;
4449 struct trace_entry *entry;
4450 struct trace_event *event;
4451
4452 entry = iter->ent;
4453
4454 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4455 SEQ_PUT_FIELD(s, entry->pid);
4456 SEQ_PUT_FIELD(s, iter->cpu);
4457 SEQ_PUT_FIELD(s, iter->ts);
4458 if (trace_seq_has_overflowed(s))
4459 return TRACE_TYPE_PARTIAL_LINE;
4460 }
4461
4462 event = ftrace_find_event(entry->type);
4463 return event ? event->funcs->binary(iter, 0, event) :
4464 TRACE_TYPE_HANDLED;
4465}
4466
4467int trace_empty(struct trace_iterator *iter)
4468{
4469 struct ring_buffer_iter *buf_iter;
4470 int cpu;
4471
4472 /* If we are looking at one CPU buffer, only check that one */
4473 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4474 cpu = iter->cpu_file;
4475 buf_iter = trace_buffer_iter(iter, cpu);
4476 if (buf_iter) {
4477 if (!ring_buffer_iter_empty(buf_iter))
4478 return 0;
4479 } else {
4480 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4481 return 0;
4482 }
4483 return 1;
4484 }
4485
4486 for_each_tracing_cpu(cpu) {
4487 buf_iter = trace_buffer_iter(iter, cpu);
4488 if (buf_iter) {
4489 if (!ring_buffer_iter_empty(buf_iter))
4490 return 0;
4491 } else {
4492 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4493 return 0;
4494 }
4495 }
4496
4497 return 1;
4498}
4499
4500/* Called with trace_event_read_lock() held. */
4501enum print_line_t print_trace_line(struct trace_iterator *iter)
4502{
4503 struct trace_array *tr = iter->tr;
4504 unsigned long trace_flags = tr->trace_flags;
4505 enum print_line_t ret;
4506
4507 if (iter->lost_events) {
4508 if (iter->lost_events == (unsigned long)-1)
4509 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4510 iter->cpu);
4511 else
4512 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4513 iter->cpu, iter->lost_events);
4514 if (trace_seq_has_overflowed(&iter->seq))
4515 return TRACE_TYPE_PARTIAL_LINE;
4516 }
4517
4518 if (iter->trace && iter->trace->print_line) {
4519 ret = iter->trace->print_line(iter);
4520 if (ret != TRACE_TYPE_UNHANDLED)
4521 return ret;
4522 }
4523
4524 if (iter->ent->type == TRACE_BPUTS &&
4525 trace_flags & TRACE_ITER_PRINTK &&
4526 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4527 return trace_print_bputs_msg_only(iter);
4528
4529 if (iter->ent->type == TRACE_BPRINT &&
4530 trace_flags & TRACE_ITER_PRINTK &&
4531 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4532 return trace_print_bprintk_msg_only(iter);
4533
4534 if (iter->ent->type == TRACE_PRINT &&
4535 trace_flags & TRACE_ITER_PRINTK &&
4536 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4537 return trace_print_printk_msg_only(iter);
4538
4539 if (trace_flags & TRACE_ITER_BIN)
4540 return print_bin_fmt(iter);
4541
4542 if (trace_flags & TRACE_ITER_HEX)
4543 return print_hex_fmt(iter);
4544
4545 if (trace_flags & TRACE_ITER_RAW)
4546 return print_raw_fmt(iter);
4547
4548 return print_trace_fmt(iter);
4549}
4550
4551void trace_latency_header(struct seq_file *m)
4552{
4553 struct trace_iterator *iter = m->private;
4554 struct trace_array *tr = iter->tr;
4555
4556 /* print nothing if the buffers are empty */
4557 if (trace_empty(iter))
4558 return;
4559
4560 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4561 print_trace_header(m, iter);
4562
4563 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4564 print_lat_help_header(m);
4565}
4566
4567void trace_default_header(struct seq_file *m)
4568{
4569 struct trace_iterator *iter = m->private;
4570 struct trace_array *tr = iter->tr;
4571 unsigned long trace_flags = tr->trace_flags;
4572
4573 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4574 return;
4575
4576 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4577 /* print nothing if the buffers are empty */
4578 if (trace_empty(iter))
4579 return;
4580 print_trace_header(m, iter);
4581 if (!(trace_flags & TRACE_ITER_VERBOSE))
4582 print_lat_help_header(m);
4583 } else {
4584 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4585 if (trace_flags & TRACE_ITER_IRQ_INFO)
4586 print_func_help_header_irq(iter->array_buffer,
4587 m, trace_flags);
4588 else
4589 print_func_help_header(iter->array_buffer, m,
4590 trace_flags);
4591 }
4592 }
4593}
4594
4595static void test_ftrace_alive(struct seq_file *m)
4596{
4597 if (!ftrace_is_dead())
4598 return;
4599 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4600 "# MAY BE MISSING FUNCTION EVENTS\n");
4601}
4602
4603#ifdef CONFIG_TRACER_MAX_TRACE
4604static void show_snapshot_main_help(struct seq_file *m)
4605{
4606 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4607 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4608 "# Takes a snapshot of the main buffer.\n"
4609 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4610 "# (Doesn't have to be '2' works with any number that\n"
4611 "# is not a '0' or '1')\n");
4612}
4613
4614static void show_snapshot_percpu_help(struct seq_file *m)
4615{
4616 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4617#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4618 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4619 "# Takes a snapshot of the main buffer for this cpu.\n");
4620#else
4621 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4622 "# Must use main snapshot file to allocate.\n");
4623#endif
4624 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4625 "# (Doesn't have to be '2' works with any number that\n"
4626 "# is not a '0' or '1')\n");
4627}
4628
4629static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4630{
4631 if (iter->tr->allocated_snapshot)
4632 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4633 else
4634 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4635
4636 seq_puts(m, "# Snapshot commands:\n");
4637 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4638 show_snapshot_main_help(m);
4639 else
4640 show_snapshot_percpu_help(m);
4641}
4642#else
4643/* Should never be called */
4644static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4645#endif
4646
4647static int s_show(struct seq_file *m, void *v)
4648{
4649 struct trace_iterator *iter = v;
4650 int ret;
4651
4652 if (iter->ent == NULL) {
4653 if (iter->tr) {
4654 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4655 seq_puts(m, "#\n");
4656 test_ftrace_alive(m);
4657 }
4658 if (iter->snapshot && trace_empty(iter))
4659 print_snapshot_help(m, iter);
4660 else if (iter->trace && iter->trace->print_header)
4661 iter->trace->print_header(m);
4662 else
4663 trace_default_header(m);
4664
4665 } else if (iter->leftover) {
4666 /*
4667 * If we filled the seq_file buffer earlier, we
4668 * want to just show it now.
4669 */
4670 ret = trace_print_seq(m, &iter->seq);
4671
4672 /* ret should this time be zero, but you never know */
4673 iter->leftover = ret;
4674
4675 } else {
4676 print_trace_line(iter);
4677 ret = trace_print_seq(m, &iter->seq);
4678 /*
4679 * If we overflow the seq_file buffer, then it will
4680 * ask us for this data again at start up.
4681 * Use that instead.
4682 * ret is 0 if seq_file write succeeded.
4683 * -1 otherwise.
4684 */
4685 iter->leftover = ret;
4686 }
4687
4688 return 0;
4689}
4690
4691/*
4692 * Should be used after trace_array_get(), trace_types_lock
4693 * ensures that i_cdev was already initialized.
4694 */
4695static inline int tracing_get_cpu(struct inode *inode)
4696{
4697 if (inode->i_cdev) /* See trace_create_cpu_file() */
4698 return (long)inode->i_cdev - 1;
4699 return RING_BUFFER_ALL_CPUS;
4700}
4701
4702static const struct seq_operations tracer_seq_ops = {
4703 .start = s_start,
4704 .next = s_next,
4705 .stop = s_stop,
4706 .show = s_show,
4707};
4708
4709static struct trace_iterator *
4710__tracing_open(struct inode *inode, struct file *file, bool snapshot)
4711{
4712 struct trace_array *tr = inode->i_private;
4713 struct trace_iterator *iter;
4714 int cpu;
4715
4716 if (tracing_disabled)
4717 return ERR_PTR(-ENODEV);
4718
4719 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4720 if (!iter)
4721 return ERR_PTR(-ENOMEM);
4722
4723 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4724 GFP_KERNEL);
4725 if (!iter->buffer_iter)
4726 goto release;
4727
4728 /*
4729 * trace_find_next_entry() may need to save off iter->ent.
4730 * It will place it into the iter->temp buffer. As most
4731 * events are less than 128, allocate a buffer of that size.
4732 * If one is greater, then trace_find_next_entry() will
4733 * allocate a new buffer to adjust for the bigger iter->ent.
4734 * It's not critical if it fails to get allocated here.
4735 */
4736 iter->temp = kmalloc(128, GFP_KERNEL);
4737 if (iter->temp)
4738 iter->temp_size = 128;
4739
4740 /*
4741 * trace_event_printf() may need to modify given format
4742 * string to replace %p with %px so that it shows real address
4743 * instead of hash value. However, that is only for the event
4744 * tracing, other tracer may not need. Defer the allocation
4745 * until it is needed.
4746 */
4747 iter->fmt = NULL;
4748 iter->fmt_size = 0;
4749
4750 /*
4751 * We make a copy of the current tracer to avoid concurrent
4752 * changes on it while we are reading.
4753 */
4754 mutex_lock(&trace_types_lock);
4755 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4756 if (!iter->trace)
4757 goto fail;
4758
4759 *iter->trace = *tr->current_trace;
4760
4761 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4762 goto fail;
4763
4764 iter->tr = tr;
4765
4766#ifdef CONFIG_TRACER_MAX_TRACE
4767 /* Currently only the top directory has a snapshot */
4768 if (tr->current_trace->print_max || snapshot)
4769 iter->array_buffer = &tr->max_buffer;
4770 else
4771#endif
4772 iter->array_buffer = &tr->array_buffer;
4773 iter->snapshot = snapshot;
4774 iter->pos = -1;
4775 iter->cpu_file = tracing_get_cpu(inode);
4776 mutex_init(&iter->mutex);
4777
4778 /* Notify the tracer early; before we stop tracing. */
4779 if (iter->trace->open)
4780 iter->trace->open(iter);
4781
4782 /* Annotate start of buffers if we had overruns */
4783 if (ring_buffer_overruns(iter->array_buffer->buffer))
4784 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4785
4786 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4787 if (trace_clocks[tr->clock_id].in_ns)
4788 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4789
4790 /*
4791 * If pause-on-trace is enabled, then stop the trace while
4792 * dumping, unless this is the "snapshot" file
4793 */
4794 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4795 tracing_stop_tr(tr);
4796
4797 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4798 for_each_tracing_cpu(cpu) {
4799 iter->buffer_iter[cpu] =
4800 ring_buffer_read_prepare(iter->array_buffer->buffer,
4801 cpu, GFP_KERNEL);
4802 }
4803 ring_buffer_read_prepare_sync();
4804 for_each_tracing_cpu(cpu) {
4805 ring_buffer_read_start(iter->buffer_iter[cpu]);
4806 tracing_iter_reset(iter, cpu);
4807 }
4808 } else {
4809 cpu = iter->cpu_file;
4810 iter->buffer_iter[cpu] =
4811 ring_buffer_read_prepare(iter->array_buffer->buffer,
4812 cpu, GFP_KERNEL);
4813 ring_buffer_read_prepare_sync();
4814 ring_buffer_read_start(iter->buffer_iter[cpu]);
4815 tracing_iter_reset(iter, cpu);
4816 }
4817
4818 mutex_unlock(&trace_types_lock);
4819
4820 return iter;
4821
4822 fail:
4823 mutex_unlock(&trace_types_lock);
4824 kfree(iter->trace);
4825 kfree(iter->temp);
4826 kfree(iter->buffer_iter);
4827release:
4828 seq_release_private(inode, file);
4829 return ERR_PTR(-ENOMEM);
4830}
4831
4832int tracing_open_generic(struct inode *inode, struct file *filp)
4833{
4834 int ret;
4835
4836 ret = tracing_check_open_get_tr(NULL);
4837 if (ret)
4838 return ret;
4839
4840 filp->private_data = inode->i_private;
4841 return 0;
4842}
4843
4844bool tracing_is_disabled(void)
4845{
4846 return (tracing_disabled) ? true: false;
4847}
4848
4849/*
4850 * Open and update trace_array ref count.
4851 * Must have the current trace_array passed to it.
4852 */
4853int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4854{
4855 struct trace_array *tr = inode->i_private;
4856 int ret;
4857
4858 ret = tracing_check_open_get_tr(tr);
4859 if (ret)
4860 return ret;
4861
4862 filp->private_data = inode->i_private;
4863
4864 return 0;
4865}
4866
4867static int tracing_mark_open(struct inode *inode, struct file *filp)
4868{
4869 stream_open(inode, filp);
4870 return tracing_open_generic_tr(inode, filp);
4871}
4872
4873static int tracing_release(struct inode *inode, struct file *file)
4874{
4875 struct trace_array *tr = inode->i_private;
4876 struct seq_file *m = file->private_data;
4877 struct trace_iterator *iter;
4878 int cpu;
4879
4880 if (!(file->f_mode & FMODE_READ)) {
4881 trace_array_put(tr);
4882 return 0;
4883 }
4884
4885 /* Writes do not use seq_file */
4886 iter = m->private;
4887 mutex_lock(&trace_types_lock);
4888
4889 for_each_tracing_cpu(cpu) {
4890 if (iter->buffer_iter[cpu])
4891 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4892 }
4893
4894 if (iter->trace && iter->trace->close)
4895 iter->trace->close(iter);
4896
4897 if (!iter->snapshot && tr->stop_count)
4898 /* reenable tracing if it was previously enabled */
4899 tracing_start_tr(tr);
4900
4901 __trace_array_put(tr);
4902
4903 mutex_unlock(&trace_types_lock);
4904
4905 mutex_destroy(&iter->mutex);
4906 free_cpumask_var(iter->started);
4907 kfree(iter->fmt);
4908 kfree(iter->temp);
4909 kfree(iter->trace);
4910 kfree(iter->buffer_iter);
4911 seq_release_private(inode, file);
4912
4913 return 0;
4914}
4915
4916static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4917{
4918 struct trace_array *tr = inode->i_private;
4919
4920 trace_array_put(tr);
4921 return 0;
4922}
4923
4924static int tracing_single_release_tr(struct inode *inode, struct file *file)
4925{
4926 struct trace_array *tr = inode->i_private;
4927
4928 trace_array_put(tr);
4929
4930 return single_release(inode, file);
4931}
4932
4933static int tracing_open(struct inode *inode, struct file *file)
4934{
4935 struct trace_array *tr = inode->i_private;
4936 struct trace_iterator *iter;
4937 int ret;
4938
4939 ret = tracing_check_open_get_tr(tr);
4940 if (ret)
4941 return ret;
4942
4943 /* If this file was open for write, then erase contents */
4944 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4945 int cpu = tracing_get_cpu(inode);
4946 struct array_buffer *trace_buf = &tr->array_buffer;
4947
4948#ifdef CONFIG_TRACER_MAX_TRACE
4949 if (tr->current_trace->print_max)
4950 trace_buf = &tr->max_buffer;
4951#endif
4952
4953 if (cpu == RING_BUFFER_ALL_CPUS)
4954 tracing_reset_online_cpus(trace_buf);
4955 else
4956 tracing_reset_cpu(trace_buf, cpu);
4957 }
4958
4959 if (file->f_mode & FMODE_READ) {
4960 iter = __tracing_open(inode, file, false);
4961 if (IS_ERR(iter))
4962 ret = PTR_ERR(iter);
4963 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4964 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4965 }
4966
4967 if (ret < 0)
4968 trace_array_put(tr);
4969
4970 return ret;
4971}
4972
4973/*
4974 * Some tracers are not suitable for instance buffers.
4975 * A tracer is always available for the global array (toplevel)
4976 * or if it explicitly states that it is.
4977 */
4978static bool
4979trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4980{
4981 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4982}
4983
4984/* Find the next tracer that this trace array may use */
4985static struct tracer *
4986get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4987{
4988 while (t && !trace_ok_for_array(t, tr))
4989 t = t->next;
4990
4991 return t;
4992}
4993
4994static void *
4995t_next(struct seq_file *m, void *v, loff_t *pos)
4996{
4997 struct trace_array *tr = m->private;
4998 struct tracer *t = v;
4999
5000 (*pos)++;
5001
5002 if (t)
5003 t = get_tracer_for_array(tr, t->next);
5004
5005 return t;
5006}
5007
5008static void *t_start(struct seq_file *m, loff_t *pos)
5009{
5010 struct trace_array *tr = m->private;
5011 struct tracer *t;
5012 loff_t l = 0;
5013
5014 mutex_lock(&trace_types_lock);
5015
5016 t = get_tracer_for_array(tr, trace_types);
5017 for (; t && l < *pos; t = t_next(m, t, &l))
5018 ;
5019
5020 return t;
5021}
5022
5023static void t_stop(struct seq_file *m, void *p)
5024{
5025 mutex_unlock(&trace_types_lock);
5026}
5027
5028static int t_show(struct seq_file *m, void *v)
5029{
5030 struct tracer *t = v;
5031
5032 if (!t)
5033 return 0;
5034
5035 seq_puts(m, t->name);
5036 if (t->next)
5037 seq_putc(m, ' ');
5038 else
5039 seq_putc(m, '\n');
5040
5041 return 0;
5042}
5043
5044static const struct seq_operations show_traces_seq_ops = {
5045 .start = t_start,
5046 .next = t_next,
5047 .stop = t_stop,
5048 .show = t_show,
5049};
5050
5051static int show_traces_open(struct inode *inode, struct file *file)
5052{
5053 struct trace_array *tr = inode->i_private;
5054 struct seq_file *m;
5055 int ret;
5056
5057 ret = tracing_check_open_get_tr(tr);
5058 if (ret)
5059 return ret;
5060
5061 ret = seq_open(file, &show_traces_seq_ops);
5062 if (ret) {
5063 trace_array_put(tr);
5064 return ret;
5065 }
5066
5067 m = file->private_data;
5068 m->private = tr;
5069
5070 return 0;
5071}
5072
5073static int show_traces_release(struct inode *inode, struct file *file)
5074{
5075 struct trace_array *tr = inode->i_private;
5076
5077 trace_array_put(tr);
5078 return seq_release(inode, file);
5079}
5080
5081static ssize_t
5082tracing_write_stub(struct file *filp, const char __user *ubuf,
5083 size_t count, loff_t *ppos)
5084{
5085 return count;
5086}
5087
5088loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5089{
5090 int ret;
5091
5092 if (file->f_mode & FMODE_READ)
5093 ret = seq_lseek(file, offset, whence);
5094 else
5095 file->f_pos = ret = 0;
5096
5097 return ret;
5098}
5099
5100static const struct file_operations tracing_fops = {
5101 .open = tracing_open,
5102 .read = seq_read,
5103 .write = tracing_write_stub,
5104 .llseek = tracing_lseek,
5105 .release = tracing_release,
5106};
5107
5108static const struct file_operations show_traces_fops = {
5109 .open = show_traces_open,
5110 .read = seq_read,
5111 .llseek = seq_lseek,
5112 .release = show_traces_release,
5113};
5114
5115static ssize_t
5116tracing_cpumask_read(struct file *filp, char __user *ubuf,
5117 size_t count, loff_t *ppos)
5118{
5119 struct trace_array *tr = file_inode(filp)->i_private;
5120 char *mask_str;
5121 int len;
5122
5123 len = snprintf(NULL, 0, "%*pb\n",
5124 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5125 mask_str = kmalloc(len, GFP_KERNEL);
5126 if (!mask_str)
5127 return -ENOMEM;
5128
5129 len = snprintf(mask_str, len, "%*pb\n",
5130 cpumask_pr_args(tr->tracing_cpumask));
5131 if (len >= count) {
5132 count = -EINVAL;
5133 goto out_err;
5134 }
5135 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5136
5137out_err:
5138 kfree(mask_str);
5139
5140 return count;
5141}
5142
5143int tracing_set_cpumask(struct trace_array *tr,
5144 cpumask_var_t tracing_cpumask_new)
5145{
5146 int cpu;
5147
5148 if (!tr)
5149 return -EINVAL;
5150
5151 local_irq_disable();
5152 arch_spin_lock(&tr->max_lock);
5153 for_each_tracing_cpu(cpu) {
5154 /*
5155 * Increase/decrease the disabled counter if we are
5156 * about to flip a bit in the cpumask:
5157 */
5158 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5159 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5160 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5161 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5162 }
5163 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5164 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5165 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5166 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5167 }
5168 }
5169 arch_spin_unlock(&tr->max_lock);
5170 local_irq_enable();
5171
5172 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5173
5174 return 0;
5175}
5176
5177static ssize_t
5178tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5179 size_t count, loff_t *ppos)
5180{
5181 struct trace_array *tr = file_inode(filp)->i_private;
5182 cpumask_var_t tracing_cpumask_new;
5183 int err;
5184
5185 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5186 return -ENOMEM;
5187
5188 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5189 if (err)
5190 goto err_free;
5191
5192 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5193 if (err)
5194 goto err_free;
5195
5196 free_cpumask_var(tracing_cpumask_new);
5197
5198 return count;
5199
5200err_free:
5201 free_cpumask_var(tracing_cpumask_new);
5202
5203 return err;
5204}
5205
5206static const struct file_operations tracing_cpumask_fops = {
5207 .open = tracing_open_generic_tr,
5208 .read = tracing_cpumask_read,
5209 .write = tracing_cpumask_write,
5210 .release = tracing_release_generic_tr,
5211 .llseek = generic_file_llseek,
5212};
5213
5214static int tracing_trace_options_show(struct seq_file *m, void *v)
5215{
5216 struct tracer_opt *trace_opts;
5217 struct trace_array *tr = m->private;
5218 u32 tracer_flags;
5219 int i;
5220
5221 mutex_lock(&trace_types_lock);
5222 tracer_flags = tr->current_trace->flags->val;
5223 trace_opts = tr->current_trace->flags->opts;
5224
5225 for (i = 0; trace_options[i]; i++) {
5226 if (tr->trace_flags & (1 << i))
5227 seq_printf(m, "%s\n", trace_options[i]);
5228 else
5229 seq_printf(m, "no%s\n", trace_options[i]);
5230 }
5231
5232 for (i = 0; trace_opts[i].name; i++) {
5233 if (tracer_flags & trace_opts[i].bit)
5234 seq_printf(m, "%s\n", trace_opts[i].name);
5235 else
5236 seq_printf(m, "no%s\n", trace_opts[i].name);
5237 }
5238 mutex_unlock(&trace_types_lock);
5239
5240 return 0;
5241}
5242
5243static int __set_tracer_option(struct trace_array *tr,
5244 struct tracer_flags *tracer_flags,
5245 struct tracer_opt *opts, int neg)
5246{
5247 struct tracer *trace = tracer_flags->trace;
5248 int ret;
5249
5250 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5251 if (ret)
5252 return ret;
5253
5254 if (neg)
5255 tracer_flags->val &= ~opts->bit;
5256 else
5257 tracer_flags->val |= opts->bit;
5258 return 0;
5259}
5260
5261/* Try to assign a tracer specific option */
5262static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5263{
5264 struct tracer *trace = tr->current_trace;
5265 struct tracer_flags *tracer_flags = trace->flags;
5266 struct tracer_opt *opts = NULL;
5267 int i;
5268
5269 for (i = 0; tracer_flags->opts[i].name; i++) {
5270 opts = &tracer_flags->opts[i];
5271
5272 if (strcmp(cmp, opts->name) == 0)
5273 return __set_tracer_option(tr, trace->flags, opts, neg);
5274 }
5275
5276 return -EINVAL;
5277}
5278
5279/* Some tracers require overwrite to stay enabled */
5280int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5281{
5282 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5283 return -1;
5284
5285 return 0;
5286}
5287
5288int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5289{
5290 int *map;
5291
5292 if ((mask == TRACE_ITER_RECORD_TGID) ||
5293 (mask == TRACE_ITER_RECORD_CMD))
5294 lockdep_assert_held(&event_mutex);
5295
5296 /* do nothing if flag is already set */
5297 if (!!(tr->trace_flags & mask) == !!enabled)
5298 return 0;
5299
5300 /* Give the tracer a chance to approve the change */
5301 if (tr->current_trace->flag_changed)
5302 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5303 return -EINVAL;
5304
5305 if (enabled)
5306 tr->trace_flags |= mask;
5307 else
5308 tr->trace_flags &= ~mask;
5309
5310 if (mask == TRACE_ITER_RECORD_CMD)
5311 trace_event_enable_cmd_record(enabled);
5312
5313 if (mask == TRACE_ITER_RECORD_TGID) {
5314 if (!tgid_map) {
5315 tgid_map_max = pid_max;
5316 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5317 GFP_KERNEL);
5318
5319 /*
5320 * Pairs with smp_load_acquire() in
5321 * trace_find_tgid_ptr() to ensure that if it observes
5322 * the tgid_map we just allocated then it also observes
5323 * the corresponding tgid_map_max value.
5324 */
5325 smp_store_release(&tgid_map, map);
5326 }
5327 if (!tgid_map) {
5328 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5329 return -ENOMEM;
5330 }
5331
5332 trace_event_enable_tgid_record(enabled);
5333 }
5334
5335 if (mask == TRACE_ITER_EVENT_FORK)
5336 trace_event_follow_fork(tr, enabled);
5337
5338 if (mask == TRACE_ITER_FUNC_FORK)
5339 ftrace_pid_follow_fork(tr, enabled);
5340
5341 if (mask == TRACE_ITER_OVERWRITE) {
5342 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5343#ifdef CONFIG_TRACER_MAX_TRACE
5344 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5345#endif
5346 }
5347
5348 if (mask == TRACE_ITER_PRINTK) {
5349 trace_printk_start_stop_comm(enabled);
5350 trace_printk_control(enabled);
5351 }
5352
5353 return 0;
5354}
5355
5356int trace_set_options(struct trace_array *tr, char *option)
5357{
5358 char *cmp;
5359 int neg = 0;
5360 int ret;
5361 size_t orig_len = strlen(option);
5362 int len;
5363
5364 cmp = strstrip(option);
5365
5366 len = str_has_prefix(cmp, "no");
5367 if (len)
5368 neg = 1;
5369
5370 cmp += len;
5371
5372 mutex_lock(&event_mutex);
5373 mutex_lock(&trace_types_lock);
5374
5375 ret = match_string(trace_options, -1, cmp);
5376 /* If no option could be set, test the specific tracer options */
5377 if (ret < 0)
5378 ret = set_tracer_option(tr, cmp, neg);
5379 else
5380 ret = set_tracer_flag(tr, 1 << ret, !neg);
5381
5382 mutex_unlock(&trace_types_lock);
5383 mutex_unlock(&event_mutex);
5384
5385 /*
5386 * If the first trailing whitespace is replaced with '\0' by strstrip,
5387 * turn it back into a space.
5388 */
5389 if (orig_len > strlen(option))
5390 option[strlen(option)] = ' ';
5391
5392 return ret;
5393}
5394
5395static void __init apply_trace_boot_options(void)
5396{
5397 char *buf = trace_boot_options_buf;
5398 char *option;
5399
5400 while (true) {
5401 option = strsep(&buf, ",");
5402
5403 if (!option)
5404 break;
5405
5406 if (*option)
5407 trace_set_options(&global_trace, option);
5408
5409 /* Put back the comma to allow this to be called again */
5410 if (buf)
5411 *(buf - 1) = ',';
5412 }
5413}
5414
5415static ssize_t
5416tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5417 size_t cnt, loff_t *ppos)
5418{
5419 struct seq_file *m = filp->private_data;
5420 struct trace_array *tr = m->private;
5421 char buf[64];
5422 int ret;
5423
5424 if (cnt >= sizeof(buf))
5425 return -EINVAL;
5426
5427 if (copy_from_user(buf, ubuf, cnt))
5428 return -EFAULT;
5429
5430 buf[cnt] = 0;
5431
5432 ret = trace_set_options(tr, buf);
5433 if (ret < 0)
5434 return ret;
5435
5436 *ppos += cnt;
5437
5438 return cnt;
5439}
5440
5441static int tracing_trace_options_open(struct inode *inode, struct file *file)
5442{
5443 struct trace_array *tr = inode->i_private;
5444 int ret;
5445
5446 ret = tracing_check_open_get_tr(tr);
5447 if (ret)
5448 return ret;
5449
5450 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5451 if (ret < 0)
5452 trace_array_put(tr);
5453
5454 return ret;
5455}
5456
5457static const struct file_operations tracing_iter_fops = {
5458 .open = tracing_trace_options_open,
5459 .read = seq_read,
5460 .llseek = seq_lseek,
5461 .release = tracing_single_release_tr,
5462 .write = tracing_trace_options_write,
5463};
5464
5465static const char readme_msg[] =
5466 "tracing mini-HOWTO:\n\n"
5467 "# echo 0 > tracing_on : quick way to disable tracing\n"
5468 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5469 " Important files:\n"
5470 " trace\t\t\t- The static contents of the buffer\n"
5471 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5472 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5473 " current_tracer\t- function and latency tracers\n"
5474 " available_tracers\t- list of configured tracers for current_tracer\n"
5475 " error_log\t- error log for failed commands (that support it)\n"
5476 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5477 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5478 " trace_clock\t\t-change the clock used to order events\n"
5479 " local: Per cpu clock but may not be synced across CPUs\n"
5480 " global: Synced across CPUs but slows tracing down.\n"
5481 " counter: Not a clock, but just an increment\n"
5482 " uptime: Jiffy counter from time of boot\n"
5483 " perf: Same clock that perf events use\n"
5484#ifdef CONFIG_X86_64
5485 " x86-tsc: TSC cycle counter\n"
5486#endif
5487 "\n timestamp_mode\t-view the mode used to timestamp events\n"
5488 " delta: Delta difference against a buffer-wide timestamp\n"
5489 " absolute: Absolute (standalone) timestamp\n"
5490 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5491 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5492 " tracing_cpumask\t- Limit which CPUs to trace\n"
5493 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5494 "\t\t\t Remove sub-buffer with rmdir\n"
5495 " trace_options\t\t- Set format or modify how tracing happens\n"
5496 "\t\t\t Disable an option by prefixing 'no' to the\n"
5497 "\t\t\t option name\n"
5498 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5499#ifdef CONFIG_DYNAMIC_FTRACE
5500 "\n available_filter_functions - list of functions that can be filtered on\n"
5501 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5502 "\t\t\t functions\n"
5503 "\t accepts: func_full_name or glob-matching-pattern\n"
5504 "\t modules: Can select a group via module\n"
5505 "\t Format: :mod:<module-name>\n"
5506 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5507 "\t triggers: a command to perform when function is hit\n"
5508 "\t Format: <function>:<trigger>[:count]\n"
5509 "\t trigger: traceon, traceoff\n"
5510 "\t\t enable_event:<system>:<event>\n"
5511 "\t\t disable_event:<system>:<event>\n"
5512#ifdef CONFIG_STACKTRACE
5513 "\t\t stacktrace\n"
5514#endif
5515#ifdef CONFIG_TRACER_SNAPSHOT
5516 "\t\t snapshot\n"
5517#endif
5518 "\t\t dump\n"
5519 "\t\t cpudump\n"
5520 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5521 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5522 "\t The first one will disable tracing every time do_fault is hit\n"
5523 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5524 "\t The first time do trap is hit and it disables tracing, the\n"
5525 "\t counter will decrement to 2. If tracing is already disabled,\n"
5526 "\t the counter will not decrement. It only decrements when the\n"
5527 "\t trigger did work\n"
5528 "\t To remove trigger without count:\n"
5529 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5530 "\t To remove trigger with a count:\n"
5531 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5532 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5533 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5534 "\t modules: Can select a group via module command :mod:\n"
5535 "\t Does not accept triggers\n"
5536#endif /* CONFIG_DYNAMIC_FTRACE */
5537#ifdef CONFIG_FUNCTION_TRACER
5538 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5539 "\t\t (function)\n"
5540 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5541 "\t\t (function)\n"
5542#endif
5543#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5544 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5545 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5546 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5547#endif
5548#ifdef CONFIG_TRACER_SNAPSHOT
5549 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5550 "\t\t\t snapshot buffer. Read the contents for more\n"
5551 "\t\t\t information\n"
5552#endif
5553#ifdef CONFIG_STACK_TRACER
5554 " stack_trace\t\t- Shows the max stack trace when active\n"
5555 " stack_max_size\t- Shows current max stack size that was traced\n"
5556 "\t\t\t Write into this file to reset the max size (trigger a\n"
5557 "\t\t\t new trace)\n"
5558#ifdef CONFIG_DYNAMIC_FTRACE
5559 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5560 "\t\t\t traces\n"
5561#endif
5562#endif /* CONFIG_STACK_TRACER */
5563#ifdef CONFIG_DYNAMIC_EVENTS
5564 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5565 "\t\t\t Write into this file to define/undefine new trace events.\n"
5566#endif
5567#ifdef CONFIG_KPROBE_EVENTS
5568 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5569 "\t\t\t Write into this file to define/undefine new trace events.\n"
5570#endif
5571#ifdef CONFIG_UPROBE_EVENTS
5572 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5573 "\t\t\t Write into this file to define/undefine new trace events.\n"
5574#endif
5575#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5576 "\t accepts: event-definitions (one definition per line)\n"
5577 "\t Format: p[:[<group>/]<event>] <place> [<args>]\n"
5578 "\t r[maxactive][:[<group>/]<event>] <place> [<args>]\n"
5579#ifdef CONFIG_HIST_TRIGGERS
5580 "\t s:[synthetic/]<event> <field> [<field>]\n"
5581#endif
5582 "\t e[:[<group>/]<event>] <attached-group>.<attached-event> [<args>]\n"
5583 "\t -:[<group>/]<event>\n"
5584#ifdef CONFIG_KPROBE_EVENTS
5585 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5586 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5587#endif
5588#ifdef CONFIG_UPROBE_EVENTS
5589 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5590#endif
5591 "\t args: <name>=fetcharg[:type]\n"
5592 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5593#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5594 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5595#else
5596 "\t $stack<index>, $stack, $retval, $comm,\n"
5597#endif
5598 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5599 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string, symbol,\n"
5600 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5601 "\t <type>\\[<array-size>\\]\n"
5602#ifdef CONFIG_HIST_TRIGGERS
5603 "\t field: <stype> <name>;\n"
5604 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5605 "\t [unsigned] char/int/long\n"
5606#endif
5607 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5608 "\t of the <attached-group>/<attached-event>.\n"
5609#endif
5610 " events/\t\t- Directory containing all trace event subsystems:\n"
5611 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5612 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5613 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5614 "\t\t\t events\n"
5615 " filter\t\t- If set, only events passing filter are traced\n"
5616 " events/<system>/<event>/\t- Directory containing control files for\n"
5617 "\t\t\t <event>:\n"
5618 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5619 " filter\t\t- If set, only events passing filter are traced\n"
5620 " trigger\t\t- If set, a command to perform when event is hit\n"
5621 "\t Format: <trigger>[:count][if <filter>]\n"
5622 "\t trigger: traceon, traceoff\n"
5623 "\t enable_event:<system>:<event>\n"
5624 "\t disable_event:<system>:<event>\n"
5625#ifdef CONFIG_HIST_TRIGGERS
5626 "\t enable_hist:<system>:<event>\n"
5627 "\t disable_hist:<system>:<event>\n"
5628#endif
5629#ifdef CONFIG_STACKTRACE
5630 "\t\t stacktrace\n"
5631#endif
5632#ifdef CONFIG_TRACER_SNAPSHOT
5633 "\t\t snapshot\n"
5634#endif
5635#ifdef CONFIG_HIST_TRIGGERS
5636 "\t\t hist (see below)\n"
5637#endif
5638 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5639 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5640 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5641 "\t events/block/block_unplug/trigger\n"
5642 "\t The first disables tracing every time block_unplug is hit.\n"
5643 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5644 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5645 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5646 "\t Like function triggers, the counter is only decremented if it\n"
5647 "\t enabled or disabled tracing.\n"
5648 "\t To remove a trigger without a count:\n"
5649 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5650 "\t To remove a trigger with a count:\n"
5651 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5652 "\t Filters can be ignored when removing a trigger.\n"
5653#ifdef CONFIG_HIST_TRIGGERS
5654 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5655 "\t Format: hist:keys=<field1[,field2,...]>\n"
5656 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5657 "\t [:values=<field1[,field2,...]>]\n"
5658 "\t [:sort=<field1[,field2,...]>]\n"
5659 "\t [:size=#entries]\n"
5660 "\t [:pause][:continue][:clear]\n"
5661 "\t [:name=histname1]\n"
5662 "\t [:<handler>.<action>]\n"
5663 "\t [if <filter>]\n\n"
5664 "\t Note, special fields can be used as well:\n"
5665 "\t common_timestamp - to record current timestamp\n"
5666 "\t common_cpu - to record the CPU the event happened on\n"
5667 "\n"
5668 "\t A hist trigger variable can be:\n"
5669 "\t - a reference to a field e.g. x=current_timestamp,\n"
5670 "\t - a reference to another variable e.g. y=$x,\n"
5671 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5672 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5673 "\n"
5674 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5675 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5676 "\t variable reference, field or numeric literal.\n"
5677 "\n"
5678 "\t When a matching event is hit, an entry is added to a hash\n"
5679 "\t table using the key(s) and value(s) named, and the value of a\n"
5680 "\t sum called 'hitcount' is incremented. Keys and values\n"
5681 "\t correspond to fields in the event's format description. Keys\n"
5682 "\t can be any field, or the special string 'stacktrace'.\n"
5683 "\t Compound keys consisting of up to two fields can be specified\n"
5684 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5685 "\t fields. Sort keys consisting of up to two fields can be\n"
5686 "\t specified using the 'sort' keyword. The sort direction can\n"
5687 "\t be modified by appending '.descending' or '.ascending' to a\n"
5688 "\t sort field. The 'size' parameter can be used to specify more\n"
5689 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5690 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5691 "\t its histogram data will be shared with other triggers of the\n"
5692 "\t same name, and trigger hits will update this common data.\n\n"
5693 "\t Reading the 'hist' file for the event will dump the hash\n"
5694 "\t table in its entirety to stdout. If there are multiple hist\n"
5695 "\t triggers attached to an event, there will be a table for each\n"
5696 "\t trigger in the output. The table displayed for a named\n"
5697 "\t trigger will be the same as any other instance having the\n"
5698 "\t same name. The default format used to display a given field\n"
5699 "\t can be modified by appending any of the following modifiers\n"
5700 "\t to the field name, as applicable:\n\n"
5701 "\t .hex display a number as a hex value\n"
5702 "\t .sym display an address as a symbol\n"
5703 "\t .sym-offset display an address as a symbol and offset\n"
5704 "\t .execname display a common_pid as a program name\n"
5705 "\t .syscall display a syscall id as a syscall name\n"
5706 "\t .log2 display log2 value rather than raw number\n"
5707 "\t .buckets=size display values in groups of size rather than raw number\n"
5708 "\t .usecs display a common_timestamp in microseconds\n\n"
5709 "\t The 'pause' parameter can be used to pause an existing hist\n"
5710 "\t trigger or to start a hist trigger but not log any events\n"
5711 "\t until told to do so. 'continue' can be used to start or\n"
5712 "\t restart a paused hist trigger.\n\n"
5713 "\t The 'clear' parameter will clear the contents of a running\n"
5714 "\t hist trigger and leave its current paused/active state\n"
5715 "\t unchanged.\n\n"
5716 "\t The enable_hist and disable_hist triggers can be used to\n"
5717 "\t have one event conditionally start and stop another event's\n"
5718 "\t already-attached hist trigger. The syntax is analogous to\n"
5719 "\t the enable_event and disable_event triggers.\n\n"
5720 "\t Hist trigger handlers and actions are executed whenever a\n"
5721 "\t a histogram entry is added or updated. They take the form:\n\n"
5722 "\t <handler>.<action>\n\n"
5723 "\t The available handlers are:\n\n"
5724 "\t onmatch(matching.event) - invoke on addition or update\n"
5725 "\t onmax(var) - invoke if var exceeds current max\n"
5726 "\t onchange(var) - invoke action if var changes\n\n"
5727 "\t The available actions are:\n\n"
5728 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5729 "\t save(field,...) - save current event fields\n"
5730#ifdef CONFIG_TRACER_SNAPSHOT
5731 "\t snapshot() - snapshot the trace buffer\n\n"
5732#endif
5733#ifdef CONFIG_SYNTH_EVENTS
5734 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5735 "\t Write into this file to define/undefine new synthetic events.\n"
5736 "\t example: echo 'myevent u64 lat; char name[]' >> synthetic_events\n"
5737#endif
5738#endif
5739;
5740
5741static ssize_t
5742tracing_readme_read(struct file *filp, char __user *ubuf,
5743 size_t cnt, loff_t *ppos)
5744{
5745 return simple_read_from_buffer(ubuf, cnt, ppos,
5746 readme_msg, strlen(readme_msg));
5747}
5748
5749static const struct file_operations tracing_readme_fops = {
5750 .open = tracing_open_generic,
5751 .read = tracing_readme_read,
5752 .llseek = generic_file_llseek,
5753};
5754
5755static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5756{
5757 int pid = ++(*pos);
5758
5759 return trace_find_tgid_ptr(pid);
5760}
5761
5762static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5763{
5764 int pid = *pos;
5765
5766 return trace_find_tgid_ptr(pid);
5767}
5768
5769static void saved_tgids_stop(struct seq_file *m, void *v)
5770{
5771}
5772
5773static int saved_tgids_show(struct seq_file *m, void *v)
5774{
5775 int *entry = (int *)v;
5776 int pid = entry - tgid_map;
5777 int tgid = *entry;
5778
5779 if (tgid == 0)
5780 return SEQ_SKIP;
5781
5782 seq_printf(m, "%d %d\n", pid, tgid);
5783 return 0;
5784}
5785
5786static const struct seq_operations tracing_saved_tgids_seq_ops = {
5787 .start = saved_tgids_start,
5788 .stop = saved_tgids_stop,
5789 .next = saved_tgids_next,
5790 .show = saved_tgids_show,
5791};
5792
5793static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5794{
5795 int ret;
5796
5797 ret = tracing_check_open_get_tr(NULL);
5798 if (ret)
5799 return ret;
5800
5801 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5802}
5803
5804
5805static const struct file_operations tracing_saved_tgids_fops = {
5806 .open = tracing_saved_tgids_open,
5807 .read = seq_read,
5808 .llseek = seq_lseek,
5809 .release = seq_release,
5810};
5811
5812static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5813{
5814 unsigned int *ptr = v;
5815
5816 if (*pos || m->count)
5817 ptr++;
5818
5819 (*pos)++;
5820
5821 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5822 ptr++) {
5823 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5824 continue;
5825
5826 return ptr;
5827 }
5828
5829 return NULL;
5830}
5831
5832static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5833{
5834 void *v;
5835 loff_t l = 0;
5836
5837 preempt_disable();
5838 arch_spin_lock(&trace_cmdline_lock);
5839
5840 v = &savedcmd->map_cmdline_to_pid[0];
5841 while (l <= *pos) {
5842 v = saved_cmdlines_next(m, v, &l);
5843 if (!v)
5844 return NULL;
5845 }
5846
5847 return v;
5848}
5849
5850static void saved_cmdlines_stop(struct seq_file *m, void *v)
5851{
5852 arch_spin_unlock(&trace_cmdline_lock);
5853 preempt_enable();
5854}
5855
5856static int saved_cmdlines_show(struct seq_file *m, void *v)
5857{
5858 char buf[TASK_COMM_LEN];
5859 unsigned int *pid = v;
5860
5861 __trace_find_cmdline(*pid, buf);
5862 seq_printf(m, "%d %s\n", *pid, buf);
5863 return 0;
5864}
5865
5866static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5867 .start = saved_cmdlines_start,
5868 .next = saved_cmdlines_next,
5869 .stop = saved_cmdlines_stop,
5870 .show = saved_cmdlines_show,
5871};
5872
5873static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5874{
5875 int ret;
5876
5877 ret = tracing_check_open_get_tr(NULL);
5878 if (ret)
5879 return ret;
5880
5881 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5882}
5883
5884static const struct file_operations tracing_saved_cmdlines_fops = {
5885 .open = tracing_saved_cmdlines_open,
5886 .read = seq_read,
5887 .llseek = seq_lseek,
5888 .release = seq_release,
5889};
5890
5891static ssize_t
5892tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5893 size_t cnt, loff_t *ppos)
5894{
5895 char buf[64];
5896 int r;
5897
5898 arch_spin_lock(&trace_cmdline_lock);
5899 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5900 arch_spin_unlock(&trace_cmdline_lock);
5901
5902 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5903}
5904
5905static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5906{
5907 kfree(s->saved_cmdlines);
5908 kfree(s->map_cmdline_to_pid);
5909 kfree(s);
5910}
5911
5912static int tracing_resize_saved_cmdlines(unsigned int val)
5913{
5914 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5915
5916 s = kmalloc(sizeof(*s), GFP_KERNEL);
5917 if (!s)
5918 return -ENOMEM;
5919
5920 if (allocate_cmdlines_buffer(val, s) < 0) {
5921 kfree(s);
5922 return -ENOMEM;
5923 }
5924
5925 arch_spin_lock(&trace_cmdline_lock);
5926 savedcmd_temp = savedcmd;
5927 savedcmd = s;
5928 arch_spin_unlock(&trace_cmdline_lock);
5929 free_saved_cmdlines_buffer(savedcmd_temp);
5930
5931 return 0;
5932}
5933
5934static ssize_t
5935tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
5936 size_t cnt, loff_t *ppos)
5937{
5938 unsigned long val;
5939 int ret;
5940
5941 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5942 if (ret)
5943 return ret;
5944
5945 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
5946 if (!val || val > PID_MAX_DEFAULT)
5947 return -EINVAL;
5948
5949 ret = tracing_resize_saved_cmdlines((unsigned int)val);
5950 if (ret < 0)
5951 return ret;
5952
5953 *ppos += cnt;
5954
5955 return cnt;
5956}
5957
5958static const struct file_operations tracing_saved_cmdlines_size_fops = {
5959 .open = tracing_open_generic,
5960 .read = tracing_saved_cmdlines_size_read,
5961 .write = tracing_saved_cmdlines_size_write,
5962};
5963
5964#ifdef CONFIG_TRACE_EVAL_MAP_FILE
5965static union trace_eval_map_item *
5966update_eval_map(union trace_eval_map_item *ptr)
5967{
5968 if (!ptr->map.eval_string) {
5969 if (ptr->tail.next) {
5970 ptr = ptr->tail.next;
5971 /* Set ptr to the next real item (skip head) */
5972 ptr++;
5973 } else
5974 return NULL;
5975 }
5976 return ptr;
5977}
5978
5979static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5980{
5981 union trace_eval_map_item *ptr = v;
5982
5983 /*
5984 * Paranoid! If ptr points to end, we don't want to increment past it.
5985 * This really should never happen.
5986 */
5987 (*pos)++;
5988 ptr = update_eval_map(ptr);
5989 if (WARN_ON_ONCE(!ptr))
5990 return NULL;
5991
5992 ptr++;
5993 ptr = update_eval_map(ptr);
5994
5995 return ptr;
5996}
5997
5998static void *eval_map_start(struct seq_file *m, loff_t *pos)
5999{
6000 union trace_eval_map_item *v;
6001 loff_t l = 0;
6002
6003 mutex_lock(&trace_eval_mutex);
6004
6005 v = trace_eval_maps;
6006 if (v)
6007 v++;
6008
6009 while (v && l < *pos) {
6010 v = eval_map_next(m, v, &l);
6011 }
6012
6013 return v;
6014}
6015
6016static void eval_map_stop(struct seq_file *m, void *v)
6017{
6018 mutex_unlock(&trace_eval_mutex);
6019}
6020
6021static int eval_map_show(struct seq_file *m, void *v)
6022{
6023 union trace_eval_map_item *ptr = v;
6024
6025 seq_printf(m, "%s %ld (%s)\n",
6026 ptr->map.eval_string, ptr->map.eval_value,
6027 ptr->map.system);
6028
6029 return 0;
6030}
6031
6032static const struct seq_operations tracing_eval_map_seq_ops = {
6033 .start = eval_map_start,
6034 .next = eval_map_next,
6035 .stop = eval_map_stop,
6036 .show = eval_map_show,
6037};
6038
6039static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6040{
6041 int ret;
6042
6043 ret = tracing_check_open_get_tr(NULL);
6044 if (ret)
6045 return ret;
6046
6047 return seq_open(filp, &tracing_eval_map_seq_ops);
6048}
6049
6050static const struct file_operations tracing_eval_map_fops = {
6051 .open = tracing_eval_map_open,
6052 .read = seq_read,
6053 .llseek = seq_lseek,
6054 .release = seq_release,
6055};
6056
6057static inline union trace_eval_map_item *
6058trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6059{
6060 /* Return tail of array given the head */
6061 return ptr + ptr->head.length + 1;
6062}
6063
6064static void
6065trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6066 int len)
6067{
6068 struct trace_eval_map **stop;
6069 struct trace_eval_map **map;
6070 union trace_eval_map_item *map_array;
6071 union trace_eval_map_item *ptr;
6072
6073 stop = start + len;
6074
6075 /*
6076 * The trace_eval_maps contains the map plus a head and tail item,
6077 * where the head holds the module and length of array, and the
6078 * tail holds a pointer to the next list.
6079 */
6080 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6081 if (!map_array) {
6082 pr_warn("Unable to allocate trace eval mapping\n");
6083 return;
6084 }
6085
6086 mutex_lock(&trace_eval_mutex);
6087
6088 if (!trace_eval_maps)
6089 trace_eval_maps = map_array;
6090 else {
6091 ptr = trace_eval_maps;
6092 for (;;) {
6093 ptr = trace_eval_jmp_to_tail(ptr);
6094 if (!ptr->tail.next)
6095 break;
6096 ptr = ptr->tail.next;
6097
6098 }
6099 ptr->tail.next = map_array;
6100 }
6101 map_array->head.mod = mod;
6102 map_array->head.length = len;
6103 map_array++;
6104
6105 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6106 map_array->map = **map;
6107 map_array++;
6108 }
6109 memset(map_array, 0, sizeof(*map_array));
6110
6111 mutex_unlock(&trace_eval_mutex);
6112}
6113
6114static void trace_create_eval_file(struct dentry *d_tracer)
6115{
6116 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6117 NULL, &tracing_eval_map_fops);
6118}
6119
6120#else /* CONFIG_TRACE_EVAL_MAP_FILE */
6121static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6122static inline void trace_insert_eval_map_file(struct module *mod,
6123 struct trace_eval_map **start, int len) { }
6124#endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6125
6126static void trace_insert_eval_map(struct module *mod,
6127 struct trace_eval_map **start, int len)
6128{
6129 struct trace_eval_map **map;
6130
6131 if (len <= 0)
6132 return;
6133
6134 map = start;
6135
6136 trace_event_eval_update(map, len);
6137
6138 trace_insert_eval_map_file(mod, start, len);
6139}
6140
6141static ssize_t
6142tracing_set_trace_read(struct file *filp, char __user *ubuf,
6143 size_t cnt, loff_t *ppos)
6144{
6145 struct trace_array *tr = filp->private_data;
6146 char buf[MAX_TRACER_SIZE+2];
6147 int r;
6148
6149 mutex_lock(&trace_types_lock);
6150 r = sprintf(buf, "%s\n", tr->current_trace->name);
6151 mutex_unlock(&trace_types_lock);
6152
6153 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6154}
6155
6156int tracer_init(struct tracer *t, struct trace_array *tr)
6157{
6158 tracing_reset_online_cpus(&tr->array_buffer);
6159 return t->init(tr);
6160}
6161
6162static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6163{
6164 int cpu;
6165
6166 for_each_tracing_cpu(cpu)
6167 per_cpu_ptr(buf->data, cpu)->entries = val;
6168}
6169
6170#ifdef CONFIG_TRACER_MAX_TRACE
6171/* resize @tr's buffer to the size of @size_tr's entries */
6172static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6173 struct array_buffer *size_buf, int cpu_id)
6174{
6175 int cpu, ret = 0;
6176
6177 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6178 for_each_tracing_cpu(cpu) {
6179 ret = ring_buffer_resize(trace_buf->buffer,
6180 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6181 if (ret < 0)
6182 break;
6183 per_cpu_ptr(trace_buf->data, cpu)->entries =
6184 per_cpu_ptr(size_buf->data, cpu)->entries;
6185 }
6186 } else {
6187 ret = ring_buffer_resize(trace_buf->buffer,
6188 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6189 if (ret == 0)
6190 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6191 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6192 }
6193
6194 return ret;
6195}
6196#endif /* CONFIG_TRACER_MAX_TRACE */
6197
6198static int __tracing_resize_ring_buffer(struct trace_array *tr,
6199 unsigned long size, int cpu)
6200{
6201 int ret;
6202
6203 /*
6204 * If kernel or user changes the size of the ring buffer
6205 * we use the size that was given, and we can forget about
6206 * expanding it later.
6207 */
6208 ring_buffer_expanded = true;
6209
6210 /* May be called before buffers are initialized */
6211 if (!tr->array_buffer.buffer)
6212 return 0;
6213
6214 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6215 if (ret < 0)
6216 return ret;
6217
6218#ifdef CONFIG_TRACER_MAX_TRACE
6219 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
6220 !tr->current_trace->use_max_tr)
6221 goto out;
6222
6223 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6224 if (ret < 0) {
6225 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6226 &tr->array_buffer, cpu);
6227 if (r < 0) {
6228 /*
6229 * AARGH! We are left with different
6230 * size max buffer!!!!
6231 * The max buffer is our "snapshot" buffer.
6232 * When a tracer needs a snapshot (one of the
6233 * latency tracers), it swaps the max buffer
6234 * with the saved snap shot. We succeeded to
6235 * update the size of the main buffer, but failed to
6236 * update the size of the max buffer. But when we tried
6237 * to reset the main buffer to the original size, we
6238 * failed there too. This is very unlikely to
6239 * happen, but if it does, warn and kill all
6240 * tracing.
6241 */
6242 WARN_ON(1);
6243 tracing_disabled = 1;
6244 }
6245 return ret;
6246 }
6247
6248 if (cpu == RING_BUFFER_ALL_CPUS)
6249 set_buffer_entries(&tr->max_buffer, size);
6250 else
6251 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
6252
6253 out:
6254#endif /* CONFIG_TRACER_MAX_TRACE */
6255
6256 if (cpu == RING_BUFFER_ALL_CPUS)
6257 set_buffer_entries(&tr->array_buffer, size);
6258 else
6259 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
6260
6261 return ret;
6262}
6263
6264ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6265 unsigned long size, int cpu_id)
6266{
6267 int ret;
6268
6269 mutex_lock(&trace_types_lock);
6270
6271 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6272 /* make sure, this cpu is enabled in the mask */
6273 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6274 ret = -EINVAL;
6275 goto out;
6276 }
6277 }
6278
6279 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6280 if (ret < 0)
6281 ret = -ENOMEM;
6282
6283out:
6284 mutex_unlock(&trace_types_lock);
6285
6286 return ret;
6287}
6288
6289
6290/**
6291 * tracing_update_buffers - used by tracing facility to expand ring buffers
6292 *
6293 * To save on memory when the tracing is never used on a system with it
6294 * configured in. The ring buffers are set to a minimum size. But once
6295 * a user starts to use the tracing facility, then they need to grow
6296 * to their default size.
6297 *
6298 * This function is to be called when a tracer is about to be used.
6299 */
6300int tracing_update_buffers(void)
6301{
6302 int ret = 0;
6303
6304 mutex_lock(&trace_types_lock);
6305 if (!ring_buffer_expanded)
6306 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
6307 RING_BUFFER_ALL_CPUS);
6308 mutex_unlock(&trace_types_lock);
6309
6310 return ret;
6311}
6312
6313struct trace_option_dentry;
6314
6315static void
6316create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6317
6318/*
6319 * Used to clear out the tracer before deletion of an instance.
6320 * Must have trace_types_lock held.
6321 */
6322static void tracing_set_nop(struct trace_array *tr)
6323{
6324 if (tr->current_trace == &nop_trace)
6325 return;
6326
6327 tr->current_trace->enabled--;
6328
6329 if (tr->current_trace->reset)
6330 tr->current_trace->reset(tr);
6331
6332 tr->current_trace = &nop_trace;
6333}
6334
6335static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6336{
6337 /* Only enable if the directory has been created already. */
6338 if (!tr->dir)
6339 return;
6340
6341 create_trace_option_files(tr, t);
6342}
6343
6344int tracing_set_tracer(struct trace_array *tr, const char *buf)
6345{
6346 struct tracer *t;
6347#ifdef CONFIG_TRACER_MAX_TRACE
6348 bool had_max_tr;
6349#endif
6350 int ret = 0;
6351
6352 mutex_lock(&trace_types_lock);
6353
6354 if (!ring_buffer_expanded) {
6355 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6356 RING_BUFFER_ALL_CPUS);
6357 if (ret < 0)
6358 goto out;
6359 ret = 0;
6360 }
6361
6362 for (t = trace_types; t; t = t->next) {
6363 if (strcmp(t->name, buf) == 0)
6364 break;
6365 }
6366 if (!t) {
6367 ret = -EINVAL;
6368 goto out;
6369 }
6370 if (t == tr->current_trace)
6371 goto out;
6372
6373#ifdef CONFIG_TRACER_SNAPSHOT
6374 if (t->use_max_tr) {
6375 arch_spin_lock(&tr->max_lock);
6376 if (tr->cond_snapshot)
6377 ret = -EBUSY;
6378 arch_spin_unlock(&tr->max_lock);
6379 if (ret)
6380 goto out;
6381 }
6382#endif
6383 /* Some tracers won't work on kernel command line */
6384 if (system_state < SYSTEM_RUNNING && t->noboot) {
6385 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6386 t->name);
6387 goto out;
6388 }
6389
6390 /* Some tracers are only allowed for the top level buffer */
6391 if (!trace_ok_for_array(t, tr)) {
6392 ret = -EINVAL;
6393 goto out;
6394 }
6395
6396 /* If trace pipe files are being read, we can't change the tracer */
6397 if (tr->trace_ref) {
6398 ret = -EBUSY;
6399 goto out;
6400 }
6401
6402 trace_branch_disable();
6403
6404 tr->current_trace->enabled--;
6405
6406 if (tr->current_trace->reset)
6407 tr->current_trace->reset(tr);
6408
6409 /* Current trace needs to be nop_trace before synchronize_rcu */
6410 tr->current_trace = &nop_trace;
6411
6412#ifdef CONFIG_TRACER_MAX_TRACE
6413 had_max_tr = tr->allocated_snapshot;
6414
6415 if (had_max_tr && !t->use_max_tr) {
6416 /*
6417 * We need to make sure that the update_max_tr sees that
6418 * current_trace changed to nop_trace to keep it from
6419 * swapping the buffers after we resize it.
6420 * The update_max_tr is called from interrupts disabled
6421 * so a synchronized_sched() is sufficient.
6422 */
6423 synchronize_rcu();
6424 free_snapshot(tr);
6425 }
6426#endif
6427
6428#ifdef CONFIG_TRACER_MAX_TRACE
6429 if (t->use_max_tr && !had_max_tr) {
6430 ret = tracing_alloc_snapshot_instance(tr);
6431 if (ret < 0)
6432 goto out;
6433 }
6434#endif
6435
6436 if (t->init) {
6437 ret = tracer_init(t, tr);
6438 if (ret)
6439 goto out;
6440 }
6441
6442 tr->current_trace = t;
6443 tr->current_trace->enabled++;
6444 trace_branch_enable(tr);
6445 out:
6446 mutex_unlock(&trace_types_lock);
6447
6448 return ret;
6449}
6450
6451static ssize_t
6452tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6453 size_t cnt, loff_t *ppos)
6454{
6455 struct trace_array *tr = filp->private_data;
6456 char buf[MAX_TRACER_SIZE+1];
6457 int i;
6458 size_t ret;
6459 int err;
6460
6461 ret = cnt;
6462
6463 if (cnt > MAX_TRACER_SIZE)
6464 cnt = MAX_TRACER_SIZE;
6465
6466 if (copy_from_user(buf, ubuf, cnt))
6467 return -EFAULT;
6468
6469 buf[cnt] = 0;
6470
6471 /* strip ending whitespace. */
6472 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
6473 buf[i] = 0;
6474
6475 err = tracing_set_tracer(tr, buf);
6476 if (err)
6477 return err;
6478
6479 *ppos += ret;
6480
6481 return ret;
6482}
6483
6484static ssize_t
6485tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6486 size_t cnt, loff_t *ppos)
6487{
6488 char buf[64];
6489 int r;
6490
6491 r = snprintf(buf, sizeof(buf), "%ld\n",
6492 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6493 if (r > sizeof(buf))
6494 r = sizeof(buf);
6495 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6496}
6497
6498static ssize_t
6499tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6500 size_t cnt, loff_t *ppos)
6501{
6502 unsigned long val;
6503 int ret;
6504
6505 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6506 if (ret)
6507 return ret;
6508
6509 *ptr = val * 1000;
6510
6511 return cnt;
6512}
6513
6514static ssize_t
6515tracing_thresh_read(struct file *filp, char __user *ubuf,
6516 size_t cnt, loff_t *ppos)
6517{
6518 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6519}
6520
6521static ssize_t
6522tracing_thresh_write(struct file *filp, const char __user *ubuf,
6523 size_t cnt, loff_t *ppos)
6524{
6525 struct trace_array *tr = filp->private_data;
6526 int ret;
6527
6528 mutex_lock(&trace_types_lock);
6529 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6530 if (ret < 0)
6531 goto out;
6532
6533 if (tr->current_trace->update_thresh) {
6534 ret = tr->current_trace->update_thresh(tr);
6535 if (ret < 0)
6536 goto out;
6537 }
6538
6539 ret = cnt;
6540out:
6541 mutex_unlock(&trace_types_lock);
6542
6543 return ret;
6544}
6545
6546#if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
6547
6548static ssize_t
6549tracing_max_lat_read(struct file *filp, char __user *ubuf,
6550 size_t cnt, loff_t *ppos)
6551{
6552 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6553}
6554
6555static ssize_t
6556tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6557 size_t cnt, loff_t *ppos)
6558{
6559 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6560}
6561
6562#endif
6563
6564static int tracing_open_pipe(struct inode *inode, struct file *filp)
6565{
6566 struct trace_array *tr = inode->i_private;
6567 struct trace_iterator *iter;
6568 int ret;
6569
6570 ret = tracing_check_open_get_tr(tr);
6571 if (ret)
6572 return ret;
6573
6574 mutex_lock(&trace_types_lock);
6575
6576 /* create a buffer to store the information to pass to userspace */
6577 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6578 if (!iter) {
6579 ret = -ENOMEM;
6580 __trace_array_put(tr);
6581 goto out;
6582 }
6583
6584 trace_seq_init(&iter->seq);
6585 iter->trace = tr->current_trace;
6586
6587 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6588 ret = -ENOMEM;
6589 goto fail;
6590 }
6591
6592 /* trace pipe does not show start of buffer */
6593 cpumask_setall(iter->started);
6594
6595 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6596 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6597
6598 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6599 if (trace_clocks[tr->clock_id].in_ns)
6600 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6601
6602 iter->tr = tr;
6603 iter->array_buffer = &tr->array_buffer;
6604 iter->cpu_file = tracing_get_cpu(inode);
6605 mutex_init(&iter->mutex);
6606 filp->private_data = iter;
6607
6608 if (iter->trace->pipe_open)
6609 iter->trace->pipe_open(iter);
6610
6611 nonseekable_open(inode, filp);
6612
6613 tr->trace_ref++;
6614out:
6615 mutex_unlock(&trace_types_lock);
6616 return ret;
6617
6618fail:
6619 kfree(iter);
6620 __trace_array_put(tr);
6621 mutex_unlock(&trace_types_lock);
6622 return ret;
6623}
6624
6625static int tracing_release_pipe(struct inode *inode, struct file *file)
6626{
6627 struct trace_iterator *iter = file->private_data;
6628 struct trace_array *tr = inode->i_private;
6629
6630 mutex_lock(&trace_types_lock);
6631
6632 tr->trace_ref--;
6633
6634 if (iter->trace->pipe_close)
6635 iter->trace->pipe_close(iter);
6636
6637 mutex_unlock(&trace_types_lock);
6638
6639 free_cpumask_var(iter->started);
6640 mutex_destroy(&iter->mutex);
6641 kfree(iter);
6642
6643 trace_array_put(tr);
6644
6645 return 0;
6646}
6647
6648static __poll_t
6649trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6650{
6651 struct trace_array *tr = iter->tr;
6652
6653 /* Iterators are static, they should be filled or empty */
6654 if (trace_buffer_iter(iter, iter->cpu_file))
6655 return EPOLLIN | EPOLLRDNORM;
6656
6657 if (tr->trace_flags & TRACE_ITER_BLOCK)
6658 /*
6659 * Always select as readable when in blocking mode
6660 */
6661 return EPOLLIN | EPOLLRDNORM;
6662 else
6663 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6664 filp, poll_table);
6665}
6666
6667static __poll_t
6668tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6669{
6670 struct trace_iterator *iter = filp->private_data;
6671
6672 return trace_poll(iter, filp, poll_table);
6673}
6674
6675/* Must be called with iter->mutex held. */
6676static int tracing_wait_pipe(struct file *filp)
6677{
6678 struct trace_iterator *iter = filp->private_data;
6679 int ret;
6680
6681 while (trace_empty(iter)) {
6682
6683 if ((filp->f_flags & O_NONBLOCK)) {
6684 return -EAGAIN;
6685 }
6686
6687 /*
6688 * We block until we read something and tracing is disabled.
6689 * We still block if tracing is disabled, but we have never
6690 * read anything. This allows a user to cat this file, and
6691 * then enable tracing. But after we have read something,
6692 * we give an EOF when tracing is again disabled.
6693 *
6694 * iter->pos will be 0 if we haven't read anything.
6695 */
6696 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6697 break;
6698
6699 mutex_unlock(&iter->mutex);
6700
6701 ret = wait_on_pipe(iter, 0);
6702
6703 mutex_lock(&iter->mutex);
6704
6705 if (ret)
6706 return ret;
6707 }
6708
6709 return 1;
6710}
6711
6712/*
6713 * Consumer reader.
6714 */
6715static ssize_t
6716tracing_read_pipe(struct file *filp, char __user *ubuf,
6717 size_t cnt, loff_t *ppos)
6718{
6719 struct trace_iterator *iter = filp->private_data;
6720 ssize_t sret;
6721
6722 /*
6723 * Avoid more than one consumer on a single file descriptor
6724 * This is just a matter of traces coherency, the ring buffer itself
6725 * is protected.
6726 */
6727 mutex_lock(&iter->mutex);
6728
6729 /* return any leftover data */
6730 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6731 if (sret != -EBUSY)
6732 goto out;
6733
6734 trace_seq_init(&iter->seq);
6735
6736 if (iter->trace->read) {
6737 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6738 if (sret)
6739 goto out;
6740 }
6741
6742waitagain:
6743 sret = tracing_wait_pipe(filp);
6744 if (sret <= 0)
6745 goto out;
6746
6747 /* stop when tracing is finished */
6748 if (trace_empty(iter)) {
6749 sret = 0;
6750 goto out;
6751 }
6752
6753 if (cnt >= PAGE_SIZE)
6754 cnt = PAGE_SIZE - 1;
6755
6756 /* reset all but tr, trace, and overruns */
6757 trace_iterator_reset(iter);
6758 cpumask_clear(iter->started);
6759 trace_seq_init(&iter->seq);
6760
6761 trace_event_read_lock();
6762 trace_access_lock(iter->cpu_file);
6763 while (trace_find_next_entry_inc(iter) != NULL) {
6764 enum print_line_t ret;
6765 int save_len = iter->seq.seq.len;
6766
6767 ret = print_trace_line(iter);
6768 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6769 /* don't print partial lines */
6770 iter->seq.seq.len = save_len;
6771 break;
6772 }
6773 if (ret != TRACE_TYPE_NO_CONSUME)
6774 trace_consume(iter);
6775
6776 if (trace_seq_used(&iter->seq) >= cnt)
6777 break;
6778
6779 /*
6780 * Setting the full flag means we reached the trace_seq buffer
6781 * size and we should leave by partial output condition above.
6782 * One of the trace_seq_* functions is not used properly.
6783 */
6784 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6785 iter->ent->type);
6786 }
6787 trace_access_unlock(iter->cpu_file);
6788 trace_event_read_unlock();
6789
6790 /* Now copy what we have to the user */
6791 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6792 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6793 trace_seq_init(&iter->seq);
6794
6795 /*
6796 * If there was nothing to send to user, in spite of consuming trace
6797 * entries, go back to wait for more entries.
6798 */
6799 if (sret == -EBUSY)
6800 goto waitagain;
6801
6802out:
6803 mutex_unlock(&iter->mutex);
6804
6805 return sret;
6806}
6807
6808static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6809 unsigned int idx)
6810{
6811 __free_page(spd->pages[idx]);
6812}
6813
6814static size_t
6815tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6816{
6817 size_t count;
6818 int save_len;
6819 int ret;
6820
6821 /* Seq buffer is page-sized, exactly what we need. */
6822 for (;;) {
6823 save_len = iter->seq.seq.len;
6824 ret = print_trace_line(iter);
6825
6826 if (trace_seq_has_overflowed(&iter->seq)) {
6827 iter->seq.seq.len = save_len;
6828 break;
6829 }
6830
6831 /*
6832 * This should not be hit, because it should only
6833 * be set if the iter->seq overflowed. But check it
6834 * anyway to be safe.
6835 */
6836 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6837 iter->seq.seq.len = save_len;
6838 break;
6839 }
6840
6841 count = trace_seq_used(&iter->seq) - save_len;
6842 if (rem < count) {
6843 rem = 0;
6844 iter->seq.seq.len = save_len;
6845 break;
6846 }
6847
6848 if (ret != TRACE_TYPE_NO_CONSUME)
6849 trace_consume(iter);
6850 rem -= count;
6851 if (!trace_find_next_entry_inc(iter)) {
6852 rem = 0;
6853 iter->ent = NULL;
6854 break;
6855 }
6856 }
6857
6858 return rem;
6859}
6860
6861static ssize_t tracing_splice_read_pipe(struct file *filp,
6862 loff_t *ppos,
6863 struct pipe_inode_info *pipe,
6864 size_t len,
6865 unsigned int flags)
6866{
6867 struct page *pages_def[PIPE_DEF_BUFFERS];
6868 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6869 struct trace_iterator *iter = filp->private_data;
6870 struct splice_pipe_desc spd = {
6871 .pages = pages_def,
6872 .partial = partial_def,
6873 .nr_pages = 0, /* This gets updated below. */
6874 .nr_pages_max = PIPE_DEF_BUFFERS,
6875 .ops = &default_pipe_buf_ops,
6876 .spd_release = tracing_spd_release_pipe,
6877 };
6878 ssize_t ret;
6879 size_t rem;
6880 unsigned int i;
6881
6882 if (splice_grow_spd(pipe, &spd))
6883 return -ENOMEM;
6884
6885 mutex_lock(&iter->mutex);
6886
6887 if (iter->trace->splice_read) {
6888 ret = iter->trace->splice_read(iter, filp,
6889 ppos, pipe, len, flags);
6890 if (ret)
6891 goto out_err;
6892 }
6893
6894 ret = tracing_wait_pipe(filp);
6895 if (ret <= 0)
6896 goto out_err;
6897
6898 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6899 ret = -EFAULT;
6900 goto out_err;
6901 }
6902
6903 trace_event_read_lock();
6904 trace_access_lock(iter->cpu_file);
6905
6906 /* Fill as many pages as possible. */
6907 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6908 spd.pages[i] = alloc_page(GFP_KERNEL);
6909 if (!spd.pages[i])
6910 break;
6911
6912 rem = tracing_fill_pipe_page(rem, iter);
6913
6914 /* Copy the data into the page, so we can start over. */
6915 ret = trace_seq_to_buffer(&iter->seq,
6916 page_address(spd.pages[i]),
6917 trace_seq_used(&iter->seq));
6918 if (ret < 0) {
6919 __free_page(spd.pages[i]);
6920 break;
6921 }
6922 spd.partial[i].offset = 0;
6923 spd.partial[i].len = trace_seq_used(&iter->seq);
6924
6925 trace_seq_init(&iter->seq);
6926 }
6927
6928 trace_access_unlock(iter->cpu_file);
6929 trace_event_read_unlock();
6930 mutex_unlock(&iter->mutex);
6931
6932 spd.nr_pages = i;
6933
6934 if (i)
6935 ret = splice_to_pipe(pipe, &spd);
6936 else
6937 ret = 0;
6938out:
6939 splice_shrink_spd(&spd);
6940 return ret;
6941
6942out_err:
6943 mutex_unlock(&iter->mutex);
6944 goto out;
6945}
6946
6947static ssize_t
6948tracing_entries_read(struct file *filp, char __user *ubuf,
6949 size_t cnt, loff_t *ppos)
6950{
6951 struct inode *inode = file_inode(filp);
6952 struct trace_array *tr = inode->i_private;
6953 int cpu = tracing_get_cpu(inode);
6954 char buf[64];
6955 int r = 0;
6956 ssize_t ret;
6957
6958 mutex_lock(&trace_types_lock);
6959
6960 if (cpu == RING_BUFFER_ALL_CPUS) {
6961 int cpu, buf_size_same;
6962 unsigned long size;
6963
6964 size = 0;
6965 buf_size_same = 1;
6966 /* check if all cpu sizes are same */
6967 for_each_tracing_cpu(cpu) {
6968 /* fill in the size from first enabled cpu */
6969 if (size == 0)
6970 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6971 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6972 buf_size_same = 0;
6973 break;
6974 }
6975 }
6976
6977 if (buf_size_same) {
6978 if (!ring_buffer_expanded)
6979 r = sprintf(buf, "%lu (expanded: %lu)\n",
6980 size >> 10,
6981 trace_buf_size >> 10);
6982 else
6983 r = sprintf(buf, "%lu\n", size >> 10);
6984 } else
6985 r = sprintf(buf, "X\n");
6986 } else
6987 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6988
6989 mutex_unlock(&trace_types_lock);
6990
6991 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6992 return ret;
6993}
6994
6995static ssize_t
6996tracing_entries_write(struct file *filp, const char __user *ubuf,
6997 size_t cnt, loff_t *ppos)
6998{
6999 struct inode *inode = file_inode(filp);
7000 struct trace_array *tr = inode->i_private;
7001 unsigned long val;
7002 int ret;
7003
7004 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7005 if (ret)
7006 return ret;
7007
7008 /* must have at least 1 entry */
7009 if (!val)
7010 return -EINVAL;
7011
7012 /* value is in KB */
7013 val <<= 10;
7014 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7015 if (ret < 0)
7016 return ret;
7017
7018 *ppos += cnt;
7019
7020 return cnt;
7021}
7022
7023static ssize_t
7024tracing_total_entries_read(struct file *filp, char __user *ubuf,
7025 size_t cnt, loff_t *ppos)
7026{
7027 struct trace_array *tr = filp->private_data;
7028 char buf[64];
7029 int r, cpu;
7030 unsigned long size = 0, expanded_size = 0;
7031
7032 mutex_lock(&trace_types_lock);
7033 for_each_tracing_cpu(cpu) {
7034 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7035 if (!ring_buffer_expanded)
7036 expanded_size += trace_buf_size >> 10;
7037 }
7038 if (ring_buffer_expanded)
7039 r = sprintf(buf, "%lu\n", size);
7040 else
7041 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7042 mutex_unlock(&trace_types_lock);
7043
7044 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7045}
7046
7047static ssize_t
7048tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7049 size_t cnt, loff_t *ppos)
7050{
7051 /*
7052 * There is no need to read what the user has written, this function
7053 * is just to make sure that there is no error when "echo" is used
7054 */
7055
7056 *ppos += cnt;
7057
7058 return cnt;
7059}
7060
7061static int
7062tracing_free_buffer_release(struct inode *inode, struct file *filp)
7063{
7064 struct trace_array *tr = inode->i_private;
7065
7066 /* disable tracing ? */
7067 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7068 tracer_tracing_off(tr);
7069 /* resize the ring buffer to 0 */
7070 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7071
7072 trace_array_put(tr);
7073
7074 return 0;
7075}
7076
7077static ssize_t
7078tracing_mark_write(struct file *filp, const char __user *ubuf,
7079 size_t cnt, loff_t *fpos)
7080{
7081 struct trace_array *tr = filp->private_data;
7082 struct ring_buffer_event *event;
7083 enum event_trigger_type tt = ETT_NONE;
7084 struct trace_buffer *buffer;
7085 struct print_entry *entry;
7086 ssize_t written;
7087 int size;
7088 int len;
7089
7090/* Used in tracing_mark_raw_write() as well */
7091#define FAULTED_STR "<faulted>"
7092#define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7093
7094 if (tracing_disabled)
7095 return -EINVAL;
7096
7097 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7098 return -EINVAL;
7099
7100 if (cnt > TRACE_BUF_SIZE)
7101 cnt = TRACE_BUF_SIZE;
7102
7103 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7104
7105 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
7106
7107 /* If less than "<faulted>", then make sure we can still add that */
7108 if (cnt < FAULTED_SIZE)
7109 size += FAULTED_SIZE - cnt;
7110
7111 buffer = tr->array_buffer.buffer;
7112 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7113 tracing_gen_ctx());
7114 if (unlikely(!event))
7115 /* Ring buffer disabled, return as if not open for write */
7116 return -EBADF;
7117
7118 entry = ring_buffer_event_data(event);
7119 entry->ip = _THIS_IP_;
7120
7121 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7122 if (len) {
7123 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7124 cnt = FAULTED_SIZE;
7125 written = -EFAULT;
7126 } else
7127 written = cnt;
7128
7129 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7130 /* do not add \n before testing triggers, but add \0 */
7131 entry->buf[cnt] = '\0';
7132 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7133 }
7134
7135 if (entry->buf[cnt - 1] != '\n') {
7136 entry->buf[cnt] = '\n';
7137 entry->buf[cnt + 1] = '\0';
7138 } else
7139 entry->buf[cnt] = '\0';
7140
7141 if (static_branch_unlikely(&trace_marker_exports_enabled))
7142 ftrace_exports(event, TRACE_EXPORT_MARKER);
7143 __buffer_unlock_commit(buffer, event);
7144
7145 if (tt)
7146 event_triggers_post_call(tr->trace_marker_file, tt);
7147
7148 return written;
7149}
7150
7151/* Limit it for now to 3K (including tag) */
7152#define RAW_DATA_MAX_SIZE (1024*3)
7153
7154static ssize_t
7155tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7156 size_t cnt, loff_t *fpos)
7157{
7158 struct trace_array *tr = filp->private_data;
7159 struct ring_buffer_event *event;
7160 struct trace_buffer *buffer;
7161 struct raw_data_entry *entry;
7162 ssize_t written;
7163 int size;
7164 int len;
7165
7166#define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7167
7168 if (tracing_disabled)
7169 return -EINVAL;
7170
7171 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7172 return -EINVAL;
7173
7174 /* The marker must at least have a tag id */
7175 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
7176 return -EINVAL;
7177
7178 if (cnt > TRACE_BUF_SIZE)
7179 cnt = TRACE_BUF_SIZE;
7180
7181 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7182
7183 size = sizeof(*entry) + cnt;
7184 if (cnt < FAULT_SIZE_ID)
7185 size += FAULT_SIZE_ID - cnt;
7186
7187 buffer = tr->array_buffer.buffer;
7188 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7189 tracing_gen_ctx());
7190 if (!event)
7191 /* Ring buffer disabled, return as if not open for write */
7192 return -EBADF;
7193
7194 entry = ring_buffer_event_data(event);
7195
7196 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7197 if (len) {
7198 entry->id = -1;
7199 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7200 written = -EFAULT;
7201 } else
7202 written = cnt;
7203
7204 __buffer_unlock_commit(buffer, event);
7205
7206 return written;
7207}
7208
7209static int tracing_clock_show(struct seq_file *m, void *v)
7210{
7211 struct trace_array *tr = m->private;
7212 int i;
7213
7214 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7215 seq_printf(m,
7216 "%s%s%s%s", i ? " " : "",
7217 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7218 i == tr->clock_id ? "]" : "");
7219 seq_putc(m, '\n');
7220
7221 return 0;
7222}
7223
7224int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7225{
7226 int i;
7227
7228 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7229 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7230 break;
7231 }
7232 if (i == ARRAY_SIZE(trace_clocks))
7233 return -EINVAL;
7234
7235 mutex_lock(&trace_types_lock);
7236
7237 tr->clock_id = i;
7238
7239 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7240
7241 /*
7242 * New clock may not be consistent with the previous clock.
7243 * Reset the buffer so that it doesn't have incomparable timestamps.
7244 */
7245 tracing_reset_online_cpus(&tr->array_buffer);
7246
7247#ifdef CONFIG_TRACER_MAX_TRACE
7248 if (tr->max_buffer.buffer)
7249 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7250 tracing_reset_online_cpus(&tr->max_buffer);
7251#endif
7252
7253 mutex_unlock(&trace_types_lock);
7254
7255 return 0;
7256}
7257
7258static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7259 size_t cnt, loff_t *fpos)
7260{
7261 struct seq_file *m = filp->private_data;
7262 struct trace_array *tr = m->private;
7263 char buf[64];
7264 const char *clockstr;
7265 int ret;
7266
7267 if (cnt >= sizeof(buf))
7268 return -EINVAL;
7269
7270 if (copy_from_user(buf, ubuf, cnt))
7271 return -EFAULT;
7272
7273 buf[cnt] = 0;
7274
7275 clockstr = strstrip(buf);
7276
7277 ret = tracing_set_clock(tr, clockstr);
7278 if (ret)
7279 return ret;
7280
7281 *fpos += cnt;
7282
7283 return cnt;
7284}
7285
7286static int tracing_clock_open(struct inode *inode, struct file *file)
7287{
7288 struct trace_array *tr = inode->i_private;
7289 int ret;
7290
7291 ret = tracing_check_open_get_tr(tr);
7292 if (ret)
7293 return ret;
7294
7295 ret = single_open(file, tracing_clock_show, inode->i_private);
7296 if (ret < 0)
7297 trace_array_put(tr);
7298
7299 return ret;
7300}
7301
7302static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7303{
7304 struct trace_array *tr = m->private;
7305
7306 mutex_lock(&trace_types_lock);
7307
7308 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7309 seq_puts(m, "delta [absolute]\n");
7310 else
7311 seq_puts(m, "[delta] absolute\n");
7312
7313 mutex_unlock(&trace_types_lock);
7314
7315 return 0;
7316}
7317
7318static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7319{
7320 struct trace_array *tr = inode->i_private;
7321 int ret;
7322
7323 ret = tracing_check_open_get_tr(tr);
7324 if (ret)
7325 return ret;
7326
7327 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7328 if (ret < 0)
7329 trace_array_put(tr);
7330
7331 return ret;
7332}
7333
7334u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7335{
7336 if (rbe == this_cpu_read(trace_buffered_event))
7337 return ring_buffer_time_stamp(buffer);
7338
7339 return ring_buffer_event_time_stamp(buffer, rbe);
7340}
7341
7342/*
7343 * Set or disable using the per CPU trace_buffer_event when possible.
7344 */
7345int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7346{
7347 int ret = 0;
7348
7349 mutex_lock(&trace_types_lock);
7350
7351 if (set && tr->no_filter_buffering_ref++)
7352 goto out;
7353
7354 if (!set) {
7355 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7356 ret = -EINVAL;
7357 goto out;
7358 }
7359
7360 --tr->no_filter_buffering_ref;
7361 }
7362 out:
7363 mutex_unlock(&trace_types_lock);
7364
7365 return ret;
7366}
7367
7368struct ftrace_buffer_info {
7369 struct trace_iterator iter;
7370 void *spare;
7371 unsigned int spare_cpu;
7372 unsigned int read;
7373};
7374
7375#ifdef CONFIG_TRACER_SNAPSHOT
7376static int tracing_snapshot_open(struct inode *inode, struct file *file)
7377{
7378 struct trace_array *tr = inode->i_private;
7379 struct trace_iterator *iter;
7380 struct seq_file *m;
7381 int ret;
7382
7383 ret = tracing_check_open_get_tr(tr);
7384 if (ret)
7385 return ret;
7386
7387 if (file->f_mode & FMODE_READ) {
7388 iter = __tracing_open(inode, file, true);
7389 if (IS_ERR(iter))
7390 ret = PTR_ERR(iter);
7391 } else {
7392 /* Writes still need the seq_file to hold the private data */
7393 ret = -ENOMEM;
7394 m = kzalloc(sizeof(*m), GFP_KERNEL);
7395 if (!m)
7396 goto out;
7397 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7398 if (!iter) {
7399 kfree(m);
7400 goto out;
7401 }
7402 ret = 0;
7403
7404 iter->tr = tr;
7405 iter->array_buffer = &tr->max_buffer;
7406 iter->cpu_file = tracing_get_cpu(inode);
7407 m->private = iter;
7408 file->private_data = m;
7409 }
7410out:
7411 if (ret < 0)
7412 trace_array_put(tr);
7413
7414 return ret;
7415}
7416
7417static ssize_t
7418tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7419 loff_t *ppos)
7420{
7421 struct seq_file *m = filp->private_data;
7422 struct trace_iterator *iter = m->private;
7423 struct trace_array *tr = iter->tr;
7424 unsigned long val;
7425 int ret;
7426
7427 ret = tracing_update_buffers();
7428 if (ret < 0)
7429 return ret;
7430
7431 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7432 if (ret)
7433 return ret;
7434
7435 mutex_lock(&trace_types_lock);
7436
7437 if (tr->current_trace->use_max_tr) {
7438 ret = -EBUSY;
7439 goto out;
7440 }
7441
7442 arch_spin_lock(&tr->max_lock);
7443 if (tr->cond_snapshot)
7444 ret = -EBUSY;
7445 arch_spin_unlock(&tr->max_lock);
7446 if (ret)
7447 goto out;
7448
7449 switch (val) {
7450 case 0:
7451 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7452 ret = -EINVAL;
7453 break;
7454 }
7455 if (tr->allocated_snapshot)
7456 free_snapshot(tr);
7457 break;
7458 case 1:
7459/* Only allow per-cpu swap if the ring buffer supports it */
7460#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7461 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7462 ret = -EINVAL;
7463 break;
7464 }
7465#endif
7466 if (tr->allocated_snapshot)
7467 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7468 &tr->array_buffer, iter->cpu_file);
7469 else
7470 ret = tracing_alloc_snapshot_instance(tr);
7471 if (ret < 0)
7472 break;
7473 local_irq_disable();
7474 /* Now, we're going to swap */
7475 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7476 update_max_tr(tr, current, smp_processor_id(), NULL);
7477 else
7478 update_max_tr_single(tr, current, iter->cpu_file);
7479 local_irq_enable();
7480 break;
7481 default:
7482 if (tr->allocated_snapshot) {
7483 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7484 tracing_reset_online_cpus(&tr->max_buffer);
7485 else
7486 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7487 }
7488 break;
7489 }
7490
7491 if (ret >= 0) {
7492 *ppos += cnt;
7493 ret = cnt;
7494 }
7495out:
7496 mutex_unlock(&trace_types_lock);
7497 return ret;
7498}
7499
7500static int tracing_snapshot_release(struct inode *inode, struct file *file)
7501{
7502 struct seq_file *m = file->private_data;
7503 int ret;
7504
7505 ret = tracing_release(inode, file);
7506
7507 if (file->f_mode & FMODE_READ)
7508 return ret;
7509
7510 /* If write only, the seq_file is just a stub */
7511 if (m)
7512 kfree(m->private);
7513 kfree(m);
7514
7515 return 0;
7516}
7517
7518static int tracing_buffers_open(struct inode *inode, struct file *filp);
7519static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7520 size_t count, loff_t *ppos);
7521static int tracing_buffers_release(struct inode *inode, struct file *file);
7522static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7523 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7524
7525static int snapshot_raw_open(struct inode *inode, struct file *filp)
7526{
7527 struct ftrace_buffer_info *info;
7528 int ret;
7529
7530 /* The following checks for tracefs lockdown */
7531 ret = tracing_buffers_open(inode, filp);
7532 if (ret < 0)
7533 return ret;
7534
7535 info = filp->private_data;
7536
7537 if (info->iter.trace->use_max_tr) {
7538 tracing_buffers_release(inode, filp);
7539 return -EBUSY;
7540 }
7541
7542 info->iter.snapshot = true;
7543 info->iter.array_buffer = &info->iter.tr->max_buffer;
7544
7545 return ret;
7546}
7547
7548#endif /* CONFIG_TRACER_SNAPSHOT */
7549
7550
7551static const struct file_operations tracing_thresh_fops = {
7552 .open = tracing_open_generic,
7553 .read = tracing_thresh_read,
7554 .write = tracing_thresh_write,
7555 .llseek = generic_file_llseek,
7556};
7557
7558#if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
7559static const struct file_operations tracing_max_lat_fops = {
7560 .open = tracing_open_generic,
7561 .read = tracing_max_lat_read,
7562 .write = tracing_max_lat_write,
7563 .llseek = generic_file_llseek,
7564};
7565#endif
7566
7567static const struct file_operations set_tracer_fops = {
7568 .open = tracing_open_generic,
7569 .read = tracing_set_trace_read,
7570 .write = tracing_set_trace_write,
7571 .llseek = generic_file_llseek,
7572};
7573
7574static const struct file_operations tracing_pipe_fops = {
7575 .open = tracing_open_pipe,
7576 .poll = tracing_poll_pipe,
7577 .read = tracing_read_pipe,
7578 .splice_read = tracing_splice_read_pipe,
7579 .release = tracing_release_pipe,
7580 .llseek = no_llseek,
7581};
7582
7583static const struct file_operations tracing_entries_fops = {
7584 .open = tracing_open_generic_tr,
7585 .read = tracing_entries_read,
7586 .write = tracing_entries_write,
7587 .llseek = generic_file_llseek,
7588 .release = tracing_release_generic_tr,
7589};
7590
7591static const struct file_operations tracing_total_entries_fops = {
7592 .open = tracing_open_generic_tr,
7593 .read = tracing_total_entries_read,
7594 .llseek = generic_file_llseek,
7595 .release = tracing_release_generic_tr,
7596};
7597
7598static const struct file_operations tracing_free_buffer_fops = {
7599 .open = tracing_open_generic_tr,
7600 .write = tracing_free_buffer_write,
7601 .release = tracing_free_buffer_release,
7602};
7603
7604static const struct file_operations tracing_mark_fops = {
7605 .open = tracing_mark_open,
7606 .write = tracing_mark_write,
7607 .release = tracing_release_generic_tr,
7608};
7609
7610static const struct file_operations tracing_mark_raw_fops = {
7611 .open = tracing_mark_open,
7612 .write = tracing_mark_raw_write,
7613 .release = tracing_release_generic_tr,
7614};
7615
7616static const struct file_operations trace_clock_fops = {
7617 .open = tracing_clock_open,
7618 .read = seq_read,
7619 .llseek = seq_lseek,
7620 .release = tracing_single_release_tr,
7621 .write = tracing_clock_write,
7622};
7623
7624static const struct file_operations trace_time_stamp_mode_fops = {
7625 .open = tracing_time_stamp_mode_open,
7626 .read = seq_read,
7627 .llseek = seq_lseek,
7628 .release = tracing_single_release_tr,
7629};
7630
7631#ifdef CONFIG_TRACER_SNAPSHOT
7632static const struct file_operations snapshot_fops = {
7633 .open = tracing_snapshot_open,
7634 .read = seq_read,
7635 .write = tracing_snapshot_write,
7636 .llseek = tracing_lseek,
7637 .release = tracing_snapshot_release,
7638};
7639
7640static const struct file_operations snapshot_raw_fops = {
7641 .open = snapshot_raw_open,
7642 .read = tracing_buffers_read,
7643 .release = tracing_buffers_release,
7644 .splice_read = tracing_buffers_splice_read,
7645 .llseek = no_llseek,
7646};
7647
7648#endif /* CONFIG_TRACER_SNAPSHOT */
7649
7650/*
7651 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7652 * @filp: The active open file structure
7653 * @ubuf: The userspace provided buffer to read value into
7654 * @cnt: The maximum number of bytes to read
7655 * @ppos: The current "file" position
7656 *
7657 * This function implements the write interface for a struct trace_min_max_param.
7658 * The filp->private_data must point to a trace_min_max_param structure that
7659 * defines where to write the value, the min and the max acceptable values,
7660 * and a lock to protect the write.
7661 */
7662static ssize_t
7663trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7664{
7665 struct trace_min_max_param *param = filp->private_data;
7666 u64 val;
7667 int err;
7668
7669 if (!param)
7670 return -EFAULT;
7671
7672 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7673 if (err)
7674 return err;
7675
7676 if (param->lock)
7677 mutex_lock(param->lock);
7678
7679 if (param->min && val < *param->min)
7680 err = -EINVAL;
7681
7682 if (param->max && val > *param->max)
7683 err = -EINVAL;
7684
7685 if (!err)
7686 *param->val = val;
7687
7688 if (param->lock)
7689 mutex_unlock(param->lock);
7690
7691 if (err)
7692 return err;
7693
7694 return cnt;
7695}
7696
7697/*
7698 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7699 * @filp: The active open file structure
7700 * @ubuf: The userspace provided buffer to read value into
7701 * @cnt: The maximum number of bytes to read
7702 * @ppos: The current "file" position
7703 *
7704 * This function implements the read interface for a struct trace_min_max_param.
7705 * The filp->private_data must point to a trace_min_max_param struct with valid
7706 * data.
7707 */
7708static ssize_t
7709trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7710{
7711 struct trace_min_max_param *param = filp->private_data;
7712 char buf[U64_STR_SIZE];
7713 int len;
7714 u64 val;
7715
7716 if (!param)
7717 return -EFAULT;
7718
7719 val = *param->val;
7720
7721 if (cnt > sizeof(buf))
7722 cnt = sizeof(buf);
7723
7724 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7725
7726 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7727}
7728
7729const struct file_operations trace_min_max_fops = {
7730 .open = tracing_open_generic,
7731 .read = trace_min_max_read,
7732 .write = trace_min_max_write,
7733};
7734
7735#define TRACING_LOG_ERRS_MAX 8
7736#define TRACING_LOG_LOC_MAX 128
7737
7738#define CMD_PREFIX " Command: "
7739
7740struct err_info {
7741 const char **errs; /* ptr to loc-specific array of err strings */
7742 u8 type; /* index into errs -> specific err string */
7743 u16 pos; /* caret position */
7744 u64 ts;
7745};
7746
7747struct tracing_log_err {
7748 struct list_head list;
7749 struct err_info info;
7750 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7751 char *cmd; /* what caused err */
7752};
7753
7754static DEFINE_MUTEX(tracing_err_log_lock);
7755
7756static struct tracing_log_err *alloc_tracing_log_err(int len)
7757{
7758 struct tracing_log_err *err;
7759
7760 err = kzalloc(sizeof(*err), GFP_KERNEL);
7761 if (!err)
7762 return ERR_PTR(-ENOMEM);
7763
7764 err->cmd = kzalloc(len, GFP_KERNEL);
7765 if (!err->cmd) {
7766 kfree(err);
7767 return ERR_PTR(-ENOMEM);
7768 }
7769
7770 return err;
7771}
7772
7773static void free_tracing_log_err(struct tracing_log_err *err)
7774{
7775 kfree(err->cmd);
7776 kfree(err);
7777}
7778
7779static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7780 int len)
7781{
7782 struct tracing_log_err *err;
7783
7784 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7785 err = alloc_tracing_log_err(len);
7786 if (PTR_ERR(err) != -ENOMEM)
7787 tr->n_err_log_entries++;
7788
7789 return err;
7790 }
7791
7792 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7793 kfree(err->cmd);
7794 err->cmd = kzalloc(len, GFP_KERNEL);
7795 if (!err->cmd)
7796 return ERR_PTR(-ENOMEM);
7797 list_del(&err->list);
7798
7799 return err;
7800}
7801
7802/**
7803 * err_pos - find the position of a string within a command for error careting
7804 * @cmd: The tracing command that caused the error
7805 * @str: The string to position the caret at within @cmd
7806 *
7807 * Finds the position of the first occurrence of @str within @cmd. The
7808 * return value can be passed to tracing_log_err() for caret placement
7809 * within @cmd.
7810 *
7811 * Returns the index within @cmd of the first occurrence of @str or 0
7812 * if @str was not found.
7813 */
7814unsigned int err_pos(char *cmd, const char *str)
7815{
7816 char *found;
7817
7818 if (WARN_ON(!strlen(cmd)))
7819 return 0;
7820
7821 found = strstr(cmd, str);
7822 if (found)
7823 return found - cmd;
7824
7825 return 0;
7826}
7827
7828/**
7829 * tracing_log_err - write an error to the tracing error log
7830 * @tr: The associated trace array for the error (NULL for top level array)
7831 * @loc: A string describing where the error occurred
7832 * @cmd: The tracing command that caused the error
7833 * @errs: The array of loc-specific static error strings
7834 * @type: The index into errs[], which produces the specific static err string
7835 * @pos: The position the caret should be placed in the cmd
7836 *
7837 * Writes an error into tracing/error_log of the form:
7838 *
7839 * <loc>: error: <text>
7840 * Command: <cmd>
7841 * ^
7842 *
7843 * tracing/error_log is a small log file containing the last
7844 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7845 * unless there has been a tracing error, and the error log can be
7846 * cleared and have its memory freed by writing the empty string in
7847 * truncation mode to it i.e. echo > tracing/error_log.
7848 *
7849 * NOTE: the @errs array along with the @type param are used to
7850 * produce a static error string - this string is not copied and saved
7851 * when the error is logged - only a pointer to it is saved. See
7852 * existing callers for examples of how static strings are typically
7853 * defined for use with tracing_log_err().
7854 */
7855void tracing_log_err(struct trace_array *tr,
7856 const char *loc, const char *cmd,
7857 const char **errs, u8 type, u16 pos)
7858{
7859 struct tracing_log_err *err;
7860 int len = 0;
7861
7862 if (!tr)
7863 tr = &global_trace;
7864
7865 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
7866
7867 mutex_lock(&tracing_err_log_lock);
7868 err = get_tracing_log_err(tr, len);
7869 if (PTR_ERR(err) == -ENOMEM) {
7870 mutex_unlock(&tracing_err_log_lock);
7871 return;
7872 }
7873
7874 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7875 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
7876
7877 err->info.errs = errs;
7878 err->info.type = type;
7879 err->info.pos = pos;
7880 err->info.ts = local_clock();
7881
7882 list_add_tail(&err->list, &tr->err_log);
7883 mutex_unlock(&tracing_err_log_lock);
7884}
7885
7886static void clear_tracing_err_log(struct trace_array *tr)
7887{
7888 struct tracing_log_err *err, *next;
7889
7890 mutex_lock(&tracing_err_log_lock);
7891 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7892 list_del(&err->list);
7893 free_tracing_log_err(err);
7894 }
7895
7896 tr->n_err_log_entries = 0;
7897 mutex_unlock(&tracing_err_log_lock);
7898}
7899
7900static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7901{
7902 struct trace_array *tr = m->private;
7903
7904 mutex_lock(&tracing_err_log_lock);
7905
7906 return seq_list_start(&tr->err_log, *pos);
7907}
7908
7909static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7910{
7911 struct trace_array *tr = m->private;
7912
7913 return seq_list_next(v, &tr->err_log, pos);
7914}
7915
7916static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7917{
7918 mutex_unlock(&tracing_err_log_lock);
7919}
7920
7921static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
7922{
7923 u16 i;
7924
7925 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7926 seq_putc(m, ' ');
7927 for (i = 0; i < pos; i++)
7928 seq_putc(m, ' ');
7929 seq_puts(m, "^\n");
7930}
7931
7932static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7933{
7934 struct tracing_log_err *err = v;
7935
7936 if (err) {
7937 const char *err_text = err->info.errs[err->info.type];
7938 u64 sec = err->info.ts;
7939 u32 nsec;
7940
7941 nsec = do_div(sec, NSEC_PER_SEC);
7942 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7943 err->loc, err_text);
7944 seq_printf(m, "%s", err->cmd);
7945 tracing_err_log_show_pos(m, err->info.pos);
7946 }
7947
7948 return 0;
7949}
7950
7951static const struct seq_operations tracing_err_log_seq_ops = {
7952 .start = tracing_err_log_seq_start,
7953 .next = tracing_err_log_seq_next,
7954 .stop = tracing_err_log_seq_stop,
7955 .show = tracing_err_log_seq_show
7956};
7957
7958static int tracing_err_log_open(struct inode *inode, struct file *file)
7959{
7960 struct trace_array *tr = inode->i_private;
7961 int ret = 0;
7962
7963 ret = tracing_check_open_get_tr(tr);
7964 if (ret)
7965 return ret;
7966
7967 /* If this file was opened for write, then erase contents */
7968 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7969 clear_tracing_err_log(tr);
7970
7971 if (file->f_mode & FMODE_READ) {
7972 ret = seq_open(file, &tracing_err_log_seq_ops);
7973 if (!ret) {
7974 struct seq_file *m = file->private_data;
7975 m->private = tr;
7976 } else {
7977 trace_array_put(tr);
7978 }
7979 }
7980 return ret;
7981}
7982
7983static ssize_t tracing_err_log_write(struct file *file,
7984 const char __user *buffer,
7985 size_t count, loff_t *ppos)
7986{
7987 return count;
7988}
7989
7990static int tracing_err_log_release(struct inode *inode, struct file *file)
7991{
7992 struct trace_array *tr = inode->i_private;
7993
7994 trace_array_put(tr);
7995
7996 if (file->f_mode & FMODE_READ)
7997 seq_release(inode, file);
7998
7999 return 0;
8000}
8001
8002static const struct file_operations tracing_err_log_fops = {
8003 .open = tracing_err_log_open,
8004 .write = tracing_err_log_write,
8005 .read = seq_read,
8006 .llseek = seq_lseek,
8007 .release = tracing_err_log_release,
8008};
8009
8010static int tracing_buffers_open(struct inode *inode, struct file *filp)
8011{
8012 struct trace_array *tr = inode->i_private;
8013 struct ftrace_buffer_info *info;
8014 int ret;
8015
8016 ret = tracing_check_open_get_tr(tr);
8017 if (ret)
8018 return ret;
8019
8020 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8021 if (!info) {
8022 trace_array_put(tr);
8023 return -ENOMEM;
8024 }
8025
8026 mutex_lock(&trace_types_lock);
8027
8028 info->iter.tr = tr;
8029 info->iter.cpu_file = tracing_get_cpu(inode);
8030 info->iter.trace = tr->current_trace;
8031 info->iter.array_buffer = &tr->array_buffer;
8032 info->spare = NULL;
8033 /* Force reading ring buffer for first read */
8034 info->read = (unsigned int)-1;
8035
8036 filp->private_data = info;
8037
8038 tr->trace_ref++;
8039
8040 mutex_unlock(&trace_types_lock);
8041
8042 ret = nonseekable_open(inode, filp);
8043 if (ret < 0)
8044 trace_array_put(tr);
8045
8046 return ret;
8047}
8048
8049static __poll_t
8050tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8051{
8052 struct ftrace_buffer_info *info = filp->private_data;
8053 struct trace_iterator *iter = &info->iter;
8054
8055 return trace_poll(iter, filp, poll_table);
8056}
8057
8058static ssize_t
8059tracing_buffers_read(struct file *filp, char __user *ubuf,
8060 size_t count, loff_t *ppos)
8061{
8062 struct ftrace_buffer_info *info = filp->private_data;
8063 struct trace_iterator *iter = &info->iter;
8064 ssize_t ret = 0;
8065 ssize_t size;
8066
8067 if (!count)
8068 return 0;
8069
8070#ifdef CONFIG_TRACER_MAX_TRACE
8071 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8072 return -EBUSY;
8073#endif
8074
8075 if (!info->spare) {
8076 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8077 iter->cpu_file);
8078 if (IS_ERR(info->spare)) {
8079 ret = PTR_ERR(info->spare);
8080 info->spare = NULL;
8081 } else {
8082 info->spare_cpu = iter->cpu_file;
8083 }
8084 }
8085 if (!info->spare)
8086 return ret;
8087
8088 /* Do we have previous read data to read? */
8089 if (info->read < PAGE_SIZE)
8090 goto read;
8091
8092 again:
8093 trace_access_lock(iter->cpu_file);
8094 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8095 &info->spare,
8096 count,
8097 iter->cpu_file, 0);
8098 trace_access_unlock(iter->cpu_file);
8099
8100 if (ret < 0) {
8101 if (trace_empty(iter)) {
8102 if ((filp->f_flags & O_NONBLOCK))
8103 return -EAGAIN;
8104
8105 ret = wait_on_pipe(iter, 0);
8106 if (ret)
8107 return ret;
8108
8109 goto again;
8110 }
8111 return 0;
8112 }
8113
8114 info->read = 0;
8115 read:
8116 size = PAGE_SIZE - info->read;
8117 if (size > count)
8118 size = count;
8119
8120 ret = copy_to_user(ubuf, info->spare + info->read, size);
8121 if (ret == size)
8122 return -EFAULT;
8123
8124 size -= ret;
8125
8126 *ppos += size;
8127 info->read += size;
8128
8129 return size;
8130}
8131
8132static int tracing_buffers_release(struct inode *inode, struct file *file)
8133{
8134 struct ftrace_buffer_info *info = file->private_data;
8135 struct trace_iterator *iter = &info->iter;
8136
8137 mutex_lock(&trace_types_lock);
8138
8139 iter->tr->trace_ref--;
8140
8141 __trace_array_put(iter->tr);
8142
8143 if (info->spare)
8144 ring_buffer_free_read_page(iter->array_buffer->buffer,
8145 info->spare_cpu, info->spare);
8146 kvfree(info);
8147
8148 mutex_unlock(&trace_types_lock);
8149
8150 return 0;
8151}
8152
8153struct buffer_ref {
8154 struct trace_buffer *buffer;
8155 void *page;
8156 int cpu;
8157 refcount_t refcount;
8158};
8159
8160static void buffer_ref_release(struct buffer_ref *ref)
8161{
8162 if (!refcount_dec_and_test(&ref->refcount))
8163 return;
8164 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8165 kfree(ref);
8166}
8167
8168static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8169 struct pipe_buffer *buf)
8170{
8171 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8172
8173 buffer_ref_release(ref);
8174 buf->private = 0;
8175}
8176
8177static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8178 struct pipe_buffer *buf)
8179{
8180 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8181
8182 if (refcount_read(&ref->refcount) > INT_MAX/2)
8183 return false;
8184
8185 refcount_inc(&ref->refcount);
8186 return true;
8187}
8188
8189/* Pipe buffer operations for a buffer. */
8190static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8191 .release = buffer_pipe_buf_release,
8192 .get = buffer_pipe_buf_get,
8193};
8194
8195/*
8196 * Callback from splice_to_pipe(), if we need to release some pages
8197 * at the end of the spd in case we error'ed out in filling the pipe.
8198 */
8199static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8200{
8201 struct buffer_ref *ref =
8202 (struct buffer_ref *)spd->partial[i].private;
8203
8204 buffer_ref_release(ref);
8205 spd->partial[i].private = 0;
8206}
8207
8208static ssize_t
8209tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8210 struct pipe_inode_info *pipe, size_t len,
8211 unsigned int flags)
8212{
8213 struct ftrace_buffer_info *info = file->private_data;
8214 struct trace_iterator *iter = &info->iter;
8215 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8216 struct page *pages_def[PIPE_DEF_BUFFERS];
8217 struct splice_pipe_desc spd = {
8218 .pages = pages_def,
8219 .partial = partial_def,
8220 .nr_pages_max = PIPE_DEF_BUFFERS,
8221 .ops = &buffer_pipe_buf_ops,
8222 .spd_release = buffer_spd_release,
8223 };
8224 struct buffer_ref *ref;
8225 int entries, i;
8226 ssize_t ret = 0;
8227
8228#ifdef CONFIG_TRACER_MAX_TRACE
8229 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8230 return -EBUSY;
8231#endif
8232
8233 if (*ppos & (PAGE_SIZE - 1))
8234 return -EINVAL;
8235
8236 if (len & (PAGE_SIZE - 1)) {
8237 if (len < PAGE_SIZE)
8238 return -EINVAL;
8239 len &= PAGE_MASK;
8240 }
8241
8242 if (splice_grow_spd(pipe, &spd))
8243 return -ENOMEM;
8244
8245 again:
8246 trace_access_lock(iter->cpu_file);
8247 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8248
8249 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
8250 struct page *page;
8251 int r;
8252
8253 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8254 if (!ref) {
8255 ret = -ENOMEM;
8256 break;
8257 }
8258
8259 refcount_set(&ref->refcount, 1);
8260 ref->buffer = iter->array_buffer->buffer;
8261 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8262 if (IS_ERR(ref->page)) {
8263 ret = PTR_ERR(ref->page);
8264 ref->page = NULL;
8265 kfree(ref);
8266 break;
8267 }
8268 ref->cpu = iter->cpu_file;
8269
8270 r = ring_buffer_read_page(ref->buffer, &ref->page,
8271 len, iter->cpu_file, 1);
8272 if (r < 0) {
8273 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8274 ref->page);
8275 kfree(ref);
8276 break;
8277 }
8278
8279 page = virt_to_page(ref->page);
8280
8281 spd.pages[i] = page;
8282 spd.partial[i].len = PAGE_SIZE;
8283 spd.partial[i].offset = 0;
8284 spd.partial[i].private = (unsigned long)ref;
8285 spd.nr_pages++;
8286 *ppos += PAGE_SIZE;
8287
8288 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8289 }
8290
8291 trace_access_unlock(iter->cpu_file);
8292 spd.nr_pages = i;
8293
8294 /* did we read anything? */
8295 if (!spd.nr_pages) {
8296 if (ret)
8297 goto out;
8298
8299 ret = -EAGAIN;
8300 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8301 goto out;
8302
8303 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
8304 if (ret)
8305 goto out;
8306
8307 goto again;
8308 }
8309
8310 ret = splice_to_pipe(pipe, &spd);
8311out:
8312 splice_shrink_spd(&spd);
8313
8314 return ret;
8315}
8316
8317static const struct file_operations tracing_buffers_fops = {
8318 .open = tracing_buffers_open,
8319 .read = tracing_buffers_read,
8320 .poll = tracing_buffers_poll,
8321 .release = tracing_buffers_release,
8322 .splice_read = tracing_buffers_splice_read,
8323 .llseek = no_llseek,
8324};
8325
8326static ssize_t
8327tracing_stats_read(struct file *filp, char __user *ubuf,
8328 size_t count, loff_t *ppos)
8329{
8330 struct inode *inode = file_inode(filp);
8331 struct trace_array *tr = inode->i_private;
8332 struct array_buffer *trace_buf = &tr->array_buffer;
8333 int cpu = tracing_get_cpu(inode);
8334 struct trace_seq *s;
8335 unsigned long cnt;
8336 unsigned long long t;
8337 unsigned long usec_rem;
8338
8339 s = kmalloc(sizeof(*s), GFP_KERNEL);
8340 if (!s)
8341 return -ENOMEM;
8342
8343 trace_seq_init(s);
8344
8345 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8346 trace_seq_printf(s, "entries: %ld\n", cnt);
8347
8348 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8349 trace_seq_printf(s, "overrun: %ld\n", cnt);
8350
8351 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8352 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8353
8354 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8355 trace_seq_printf(s, "bytes: %ld\n", cnt);
8356
8357 if (trace_clocks[tr->clock_id].in_ns) {
8358 /* local or global for trace_clock */
8359 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8360 usec_rem = do_div(t, USEC_PER_SEC);
8361 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8362 t, usec_rem);
8363
8364 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8365 usec_rem = do_div(t, USEC_PER_SEC);
8366 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8367 } else {
8368 /* counter or tsc mode for trace_clock */
8369 trace_seq_printf(s, "oldest event ts: %llu\n",
8370 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8371
8372 trace_seq_printf(s, "now ts: %llu\n",
8373 ring_buffer_time_stamp(trace_buf->buffer));
8374 }
8375
8376 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8377 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8378
8379 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8380 trace_seq_printf(s, "read events: %ld\n", cnt);
8381
8382 count = simple_read_from_buffer(ubuf, count, ppos,
8383 s->buffer, trace_seq_used(s));
8384
8385 kfree(s);
8386
8387 return count;
8388}
8389
8390static const struct file_operations tracing_stats_fops = {
8391 .open = tracing_open_generic_tr,
8392 .read = tracing_stats_read,
8393 .llseek = generic_file_llseek,
8394 .release = tracing_release_generic_tr,
8395};
8396
8397#ifdef CONFIG_DYNAMIC_FTRACE
8398
8399static ssize_t
8400tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8401 size_t cnt, loff_t *ppos)
8402{
8403 ssize_t ret;
8404 char *buf;
8405 int r;
8406
8407 /* 256 should be plenty to hold the amount needed */
8408 buf = kmalloc(256, GFP_KERNEL);
8409 if (!buf)
8410 return -ENOMEM;
8411
8412 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8413 ftrace_update_tot_cnt,
8414 ftrace_number_of_pages,
8415 ftrace_number_of_groups);
8416
8417 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8418 kfree(buf);
8419 return ret;
8420}
8421
8422static const struct file_operations tracing_dyn_info_fops = {
8423 .open = tracing_open_generic,
8424 .read = tracing_read_dyn_info,
8425 .llseek = generic_file_llseek,
8426};
8427#endif /* CONFIG_DYNAMIC_FTRACE */
8428
8429#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8430static void
8431ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8432 struct trace_array *tr, struct ftrace_probe_ops *ops,
8433 void *data)
8434{
8435 tracing_snapshot_instance(tr);
8436}
8437
8438static void
8439ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8440 struct trace_array *tr, struct ftrace_probe_ops *ops,
8441 void *data)
8442{
8443 struct ftrace_func_mapper *mapper = data;
8444 long *count = NULL;
8445
8446 if (mapper)
8447 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8448
8449 if (count) {
8450
8451 if (*count <= 0)
8452 return;
8453
8454 (*count)--;
8455 }
8456
8457 tracing_snapshot_instance(tr);
8458}
8459
8460static int
8461ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8462 struct ftrace_probe_ops *ops, void *data)
8463{
8464 struct ftrace_func_mapper *mapper = data;
8465 long *count = NULL;
8466
8467 seq_printf(m, "%ps:", (void *)ip);
8468
8469 seq_puts(m, "snapshot");
8470
8471 if (mapper)
8472 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8473
8474 if (count)
8475 seq_printf(m, ":count=%ld\n", *count);
8476 else
8477 seq_puts(m, ":unlimited\n");
8478
8479 return 0;
8480}
8481
8482static int
8483ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8484 unsigned long ip, void *init_data, void **data)
8485{
8486 struct ftrace_func_mapper *mapper = *data;
8487
8488 if (!mapper) {
8489 mapper = allocate_ftrace_func_mapper();
8490 if (!mapper)
8491 return -ENOMEM;
8492 *data = mapper;
8493 }
8494
8495 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8496}
8497
8498static void
8499ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8500 unsigned long ip, void *data)
8501{
8502 struct ftrace_func_mapper *mapper = data;
8503
8504 if (!ip) {
8505 if (!mapper)
8506 return;
8507 free_ftrace_func_mapper(mapper, NULL);
8508 return;
8509 }
8510
8511 ftrace_func_mapper_remove_ip(mapper, ip);
8512}
8513
8514static struct ftrace_probe_ops snapshot_probe_ops = {
8515 .func = ftrace_snapshot,
8516 .print = ftrace_snapshot_print,
8517};
8518
8519static struct ftrace_probe_ops snapshot_count_probe_ops = {
8520 .func = ftrace_count_snapshot,
8521 .print = ftrace_snapshot_print,
8522 .init = ftrace_snapshot_init,
8523 .free = ftrace_snapshot_free,
8524};
8525
8526static int
8527ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8528 char *glob, char *cmd, char *param, int enable)
8529{
8530 struct ftrace_probe_ops *ops;
8531 void *count = (void *)-1;
8532 char *number;
8533 int ret;
8534
8535 if (!tr)
8536 return -ENODEV;
8537
8538 /* hash funcs only work with set_ftrace_filter */
8539 if (!enable)
8540 return -EINVAL;
8541
8542 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8543
8544 if (glob[0] == '!')
8545 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8546
8547 if (!param)
8548 goto out_reg;
8549
8550 number = strsep(&param, ":");
8551
8552 if (!strlen(number))
8553 goto out_reg;
8554
8555 /*
8556 * We use the callback data field (which is a pointer)
8557 * as our counter.
8558 */
8559 ret = kstrtoul(number, 0, (unsigned long *)&count);
8560 if (ret)
8561 return ret;
8562
8563 out_reg:
8564 ret = tracing_alloc_snapshot_instance(tr);
8565 if (ret < 0)
8566 goto out;
8567
8568 ret = register_ftrace_function_probe(glob, tr, ops, count);
8569
8570 out:
8571 return ret < 0 ? ret : 0;
8572}
8573
8574static struct ftrace_func_command ftrace_snapshot_cmd = {
8575 .name = "snapshot",
8576 .func = ftrace_trace_snapshot_callback,
8577};
8578
8579static __init int register_snapshot_cmd(void)
8580{
8581 return register_ftrace_command(&ftrace_snapshot_cmd);
8582}
8583#else
8584static inline __init int register_snapshot_cmd(void) { return 0; }
8585#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8586
8587static struct dentry *tracing_get_dentry(struct trace_array *tr)
8588{
8589 if (WARN_ON(!tr->dir))
8590 return ERR_PTR(-ENODEV);
8591
8592 /* Top directory uses NULL as the parent */
8593 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8594 return NULL;
8595
8596 /* All sub buffers have a descriptor */
8597 return tr->dir;
8598}
8599
8600static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8601{
8602 struct dentry *d_tracer;
8603
8604 if (tr->percpu_dir)
8605 return tr->percpu_dir;
8606
8607 d_tracer = tracing_get_dentry(tr);
8608 if (IS_ERR(d_tracer))
8609 return NULL;
8610
8611 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8612
8613 MEM_FAIL(!tr->percpu_dir,
8614 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8615
8616 return tr->percpu_dir;
8617}
8618
8619static struct dentry *
8620trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8621 void *data, long cpu, const struct file_operations *fops)
8622{
8623 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8624
8625 if (ret) /* See tracing_get_cpu() */
8626 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8627 return ret;
8628}
8629
8630static void
8631tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8632{
8633 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8634 struct dentry *d_cpu;
8635 char cpu_dir[30]; /* 30 characters should be more than enough */
8636
8637 if (!d_percpu)
8638 return;
8639
8640 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8641 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8642 if (!d_cpu) {
8643 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8644 return;
8645 }
8646
8647 /* per cpu trace_pipe */
8648 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8649 tr, cpu, &tracing_pipe_fops);
8650
8651 /* per cpu trace */
8652 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8653 tr, cpu, &tracing_fops);
8654
8655 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8656 tr, cpu, &tracing_buffers_fops);
8657
8658 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8659 tr, cpu, &tracing_stats_fops);
8660
8661 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8662 tr, cpu, &tracing_entries_fops);
8663
8664#ifdef CONFIG_TRACER_SNAPSHOT
8665 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8666 tr, cpu, &snapshot_fops);
8667
8668 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8669 tr, cpu, &snapshot_raw_fops);
8670#endif
8671}
8672
8673#ifdef CONFIG_FTRACE_SELFTEST
8674/* Let selftest have access to static functions in this file */
8675#include "trace_selftest.c"
8676#endif
8677
8678static ssize_t
8679trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8680 loff_t *ppos)
8681{
8682 struct trace_option_dentry *topt = filp->private_data;
8683 char *buf;
8684
8685 if (topt->flags->val & topt->opt->bit)
8686 buf = "1\n";
8687 else
8688 buf = "0\n";
8689
8690 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8691}
8692
8693static ssize_t
8694trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8695 loff_t *ppos)
8696{
8697 struct trace_option_dentry *topt = filp->private_data;
8698 unsigned long val;
8699 int ret;
8700
8701 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8702 if (ret)
8703 return ret;
8704
8705 if (val != 0 && val != 1)
8706 return -EINVAL;
8707
8708 if (!!(topt->flags->val & topt->opt->bit) != val) {
8709 mutex_lock(&trace_types_lock);
8710 ret = __set_tracer_option(topt->tr, topt->flags,
8711 topt->opt, !val);
8712 mutex_unlock(&trace_types_lock);
8713 if (ret)
8714 return ret;
8715 }
8716
8717 *ppos += cnt;
8718
8719 return cnt;
8720}
8721
8722
8723static const struct file_operations trace_options_fops = {
8724 .open = tracing_open_generic,
8725 .read = trace_options_read,
8726 .write = trace_options_write,
8727 .llseek = generic_file_llseek,
8728};
8729
8730/*
8731 * In order to pass in both the trace_array descriptor as well as the index
8732 * to the flag that the trace option file represents, the trace_array
8733 * has a character array of trace_flags_index[], which holds the index
8734 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8735 * The address of this character array is passed to the flag option file
8736 * read/write callbacks.
8737 *
8738 * In order to extract both the index and the trace_array descriptor,
8739 * get_tr_index() uses the following algorithm.
8740 *
8741 * idx = *ptr;
8742 *
8743 * As the pointer itself contains the address of the index (remember
8744 * index[1] == 1).
8745 *
8746 * Then to get the trace_array descriptor, by subtracting that index
8747 * from the ptr, we get to the start of the index itself.
8748 *
8749 * ptr - idx == &index[0]
8750 *
8751 * Then a simple container_of() from that pointer gets us to the
8752 * trace_array descriptor.
8753 */
8754static void get_tr_index(void *data, struct trace_array **ptr,
8755 unsigned int *pindex)
8756{
8757 *pindex = *(unsigned char *)data;
8758
8759 *ptr = container_of(data - *pindex, struct trace_array,
8760 trace_flags_index);
8761}
8762
8763static ssize_t
8764trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8765 loff_t *ppos)
8766{
8767 void *tr_index = filp->private_data;
8768 struct trace_array *tr;
8769 unsigned int index;
8770 char *buf;
8771
8772 get_tr_index(tr_index, &tr, &index);
8773
8774 if (tr->trace_flags & (1 << index))
8775 buf = "1\n";
8776 else
8777 buf = "0\n";
8778
8779 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8780}
8781
8782static ssize_t
8783trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8784 loff_t *ppos)
8785{
8786 void *tr_index = filp->private_data;
8787 struct trace_array *tr;
8788 unsigned int index;
8789 unsigned long val;
8790 int ret;
8791
8792 get_tr_index(tr_index, &tr, &index);
8793
8794 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8795 if (ret)
8796 return ret;
8797
8798 if (val != 0 && val != 1)
8799 return -EINVAL;
8800
8801 mutex_lock(&event_mutex);
8802 mutex_lock(&trace_types_lock);
8803 ret = set_tracer_flag(tr, 1 << index, val);
8804 mutex_unlock(&trace_types_lock);
8805 mutex_unlock(&event_mutex);
8806
8807 if (ret < 0)
8808 return ret;
8809
8810 *ppos += cnt;
8811
8812 return cnt;
8813}
8814
8815static const struct file_operations trace_options_core_fops = {
8816 .open = tracing_open_generic,
8817 .read = trace_options_core_read,
8818 .write = trace_options_core_write,
8819 .llseek = generic_file_llseek,
8820};
8821
8822struct dentry *trace_create_file(const char *name,
8823 umode_t mode,
8824 struct dentry *parent,
8825 void *data,
8826 const struct file_operations *fops)
8827{
8828 struct dentry *ret;
8829
8830 ret = tracefs_create_file(name, mode, parent, data, fops);
8831 if (!ret)
8832 pr_warn("Could not create tracefs '%s' entry\n", name);
8833
8834 return ret;
8835}
8836
8837
8838static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8839{
8840 struct dentry *d_tracer;
8841
8842 if (tr->options)
8843 return tr->options;
8844
8845 d_tracer = tracing_get_dentry(tr);
8846 if (IS_ERR(d_tracer))
8847 return NULL;
8848
8849 tr->options = tracefs_create_dir("options", d_tracer);
8850 if (!tr->options) {
8851 pr_warn("Could not create tracefs directory 'options'\n");
8852 return NULL;
8853 }
8854
8855 return tr->options;
8856}
8857
8858static void
8859create_trace_option_file(struct trace_array *tr,
8860 struct trace_option_dentry *topt,
8861 struct tracer_flags *flags,
8862 struct tracer_opt *opt)
8863{
8864 struct dentry *t_options;
8865
8866 t_options = trace_options_init_dentry(tr);
8867 if (!t_options)
8868 return;
8869
8870 topt->flags = flags;
8871 topt->opt = opt;
8872 topt->tr = tr;
8873
8874 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
8875 t_options, topt, &trace_options_fops);
8876
8877}
8878
8879static void
8880create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8881{
8882 struct trace_option_dentry *topts;
8883 struct trace_options *tr_topts;
8884 struct tracer_flags *flags;
8885 struct tracer_opt *opts;
8886 int cnt;
8887 int i;
8888
8889 if (!tracer)
8890 return;
8891
8892 flags = tracer->flags;
8893
8894 if (!flags || !flags->opts)
8895 return;
8896
8897 /*
8898 * If this is an instance, only create flags for tracers
8899 * the instance may have.
8900 */
8901 if (!trace_ok_for_array(tracer, tr))
8902 return;
8903
8904 for (i = 0; i < tr->nr_topts; i++) {
8905 /* Make sure there's no duplicate flags. */
8906 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8907 return;
8908 }
8909
8910 opts = flags->opts;
8911
8912 for (cnt = 0; opts[cnt].name; cnt++)
8913 ;
8914
8915 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8916 if (!topts)
8917 return;
8918
8919 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8920 GFP_KERNEL);
8921 if (!tr_topts) {
8922 kfree(topts);
8923 return;
8924 }
8925
8926 tr->topts = tr_topts;
8927 tr->topts[tr->nr_topts].tracer = tracer;
8928 tr->topts[tr->nr_topts].topts = topts;
8929 tr->nr_topts++;
8930
8931 for (cnt = 0; opts[cnt].name; cnt++) {
8932 create_trace_option_file(tr, &topts[cnt], flags,
8933 &opts[cnt]);
8934 MEM_FAIL(topts[cnt].entry == NULL,
8935 "Failed to create trace option: %s",
8936 opts[cnt].name);
8937 }
8938}
8939
8940static struct dentry *
8941create_trace_option_core_file(struct trace_array *tr,
8942 const char *option, long index)
8943{
8944 struct dentry *t_options;
8945
8946 t_options = trace_options_init_dentry(tr);
8947 if (!t_options)
8948 return NULL;
8949
8950 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
8951 (void *)&tr->trace_flags_index[index],
8952 &trace_options_core_fops);
8953}
8954
8955static void create_trace_options_dir(struct trace_array *tr)
8956{
8957 struct dentry *t_options;
8958 bool top_level = tr == &global_trace;
8959 int i;
8960
8961 t_options = trace_options_init_dentry(tr);
8962 if (!t_options)
8963 return;
8964
8965 for (i = 0; trace_options[i]; i++) {
8966 if (top_level ||
8967 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8968 create_trace_option_core_file(tr, trace_options[i], i);
8969 }
8970}
8971
8972static ssize_t
8973rb_simple_read(struct file *filp, char __user *ubuf,
8974 size_t cnt, loff_t *ppos)
8975{
8976 struct trace_array *tr = filp->private_data;
8977 char buf[64];
8978 int r;
8979
8980 r = tracer_tracing_is_on(tr);
8981 r = sprintf(buf, "%d\n", r);
8982
8983 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8984}
8985
8986static ssize_t
8987rb_simple_write(struct file *filp, const char __user *ubuf,
8988 size_t cnt, loff_t *ppos)
8989{
8990 struct trace_array *tr = filp->private_data;
8991 struct trace_buffer *buffer = tr->array_buffer.buffer;
8992 unsigned long val;
8993 int ret;
8994
8995 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8996 if (ret)
8997 return ret;
8998
8999 if (buffer) {
9000 mutex_lock(&trace_types_lock);
9001 if (!!val == tracer_tracing_is_on(tr)) {
9002 val = 0; /* do nothing */
9003 } else if (val) {
9004 tracer_tracing_on(tr);
9005 if (tr->current_trace->start)
9006 tr->current_trace->start(tr);
9007 } else {
9008 tracer_tracing_off(tr);
9009 if (tr->current_trace->stop)
9010 tr->current_trace->stop(tr);
9011 }
9012 mutex_unlock(&trace_types_lock);
9013 }
9014
9015 (*ppos)++;
9016
9017 return cnt;
9018}
9019
9020static const struct file_operations rb_simple_fops = {
9021 .open = tracing_open_generic_tr,
9022 .read = rb_simple_read,
9023 .write = rb_simple_write,
9024 .release = tracing_release_generic_tr,
9025 .llseek = default_llseek,
9026};
9027
9028static ssize_t
9029buffer_percent_read(struct file *filp, char __user *ubuf,
9030 size_t cnt, loff_t *ppos)
9031{
9032 struct trace_array *tr = filp->private_data;
9033 char buf[64];
9034 int r;
9035
9036 r = tr->buffer_percent;
9037 r = sprintf(buf, "%d\n", r);
9038
9039 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9040}
9041
9042static ssize_t
9043buffer_percent_write(struct file *filp, const char __user *ubuf,
9044 size_t cnt, loff_t *ppos)
9045{
9046 struct trace_array *tr = filp->private_data;
9047 unsigned long val;
9048 int ret;
9049
9050 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9051 if (ret)
9052 return ret;
9053
9054 if (val > 100)
9055 return -EINVAL;
9056
9057 if (!val)
9058 val = 1;
9059
9060 tr->buffer_percent = val;
9061
9062 (*ppos)++;
9063
9064 return cnt;
9065}
9066
9067static const struct file_operations buffer_percent_fops = {
9068 .open = tracing_open_generic_tr,
9069 .read = buffer_percent_read,
9070 .write = buffer_percent_write,
9071 .release = tracing_release_generic_tr,
9072 .llseek = default_llseek,
9073};
9074
9075static struct dentry *trace_instance_dir;
9076
9077static void
9078init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9079
9080static int
9081allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9082{
9083 enum ring_buffer_flags rb_flags;
9084
9085 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9086
9087 buf->tr = tr;
9088
9089 buf->buffer = ring_buffer_alloc(size, rb_flags);
9090 if (!buf->buffer)
9091 return -ENOMEM;
9092
9093 buf->data = alloc_percpu(struct trace_array_cpu);
9094 if (!buf->data) {
9095 ring_buffer_free(buf->buffer);
9096 buf->buffer = NULL;
9097 return -ENOMEM;
9098 }
9099
9100 /* Allocate the first page for all buffers */
9101 set_buffer_entries(&tr->array_buffer,
9102 ring_buffer_size(tr->array_buffer.buffer, 0));
9103
9104 return 0;
9105}
9106
9107static int allocate_trace_buffers(struct trace_array *tr, int size)
9108{
9109 int ret;
9110
9111 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9112 if (ret)
9113 return ret;
9114
9115#ifdef CONFIG_TRACER_MAX_TRACE
9116 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9117 allocate_snapshot ? size : 1);
9118 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9119 ring_buffer_free(tr->array_buffer.buffer);
9120 tr->array_buffer.buffer = NULL;
9121 free_percpu(tr->array_buffer.data);
9122 tr->array_buffer.data = NULL;
9123 return -ENOMEM;
9124 }
9125 tr->allocated_snapshot = allocate_snapshot;
9126
9127 /*
9128 * Only the top level trace array gets its snapshot allocated
9129 * from the kernel command line.
9130 */
9131 allocate_snapshot = false;
9132#endif
9133
9134 return 0;
9135}
9136
9137static void free_trace_buffer(struct array_buffer *buf)
9138{
9139 if (buf->buffer) {
9140 ring_buffer_free(buf->buffer);
9141 buf->buffer = NULL;
9142 free_percpu(buf->data);
9143 buf->data = NULL;
9144 }
9145}
9146
9147static void free_trace_buffers(struct trace_array *tr)
9148{
9149 if (!tr)
9150 return;
9151
9152 free_trace_buffer(&tr->array_buffer);
9153
9154#ifdef CONFIG_TRACER_MAX_TRACE
9155 free_trace_buffer(&tr->max_buffer);
9156#endif
9157}
9158
9159static void init_trace_flags_index(struct trace_array *tr)
9160{
9161 int i;
9162
9163 /* Used by the trace options files */
9164 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9165 tr->trace_flags_index[i] = i;
9166}
9167
9168static void __update_tracer_options(struct trace_array *tr)
9169{
9170 struct tracer *t;
9171
9172 for (t = trace_types; t; t = t->next)
9173 add_tracer_options(tr, t);
9174}
9175
9176static void update_tracer_options(struct trace_array *tr)
9177{
9178 mutex_lock(&trace_types_lock);
9179 __update_tracer_options(tr);
9180 mutex_unlock(&trace_types_lock);
9181}
9182
9183/* Must have trace_types_lock held */
9184struct trace_array *trace_array_find(const char *instance)
9185{
9186 struct trace_array *tr, *found = NULL;
9187
9188 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9189 if (tr->name && strcmp(tr->name, instance) == 0) {
9190 found = tr;
9191 break;
9192 }
9193 }
9194
9195 return found;
9196}
9197
9198struct trace_array *trace_array_find_get(const char *instance)
9199{
9200 struct trace_array *tr;
9201
9202 mutex_lock(&trace_types_lock);
9203 tr = trace_array_find(instance);
9204 if (tr)
9205 tr->ref++;
9206 mutex_unlock(&trace_types_lock);
9207
9208 return tr;
9209}
9210
9211static int trace_array_create_dir(struct trace_array *tr)
9212{
9213 int ret;
9214
9215 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9216 if (!tr->dir)
9217 return -EINVAL;
9218
9219 ret = event_trace_add_tracer(tr->dir, tr);
9220 if (ret) {
9221 tracefs_remove(tr->dir);
9222 return ret;
9223 }
9224
9225 init_tracer_tracefs(tr, tr->dir);
9226 __update_tracer_options(tr);
9227
9228 return ret;
9229}
9230
9231static struct trace_array *trace_array_create(const char *name)
9232{
9233 struct trace_array *tr;
9234 int ret;
9235
9236 ret = -ENOMEM;
9237 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9238 if (!tr)
9239 return ERR_PTR(ret);
9240
9241 tr->name = kstrdup(name, GFP_KERNEL);
9242 if (!tr->name)
9243 goto out_free_tr;
9244
9245 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9246 goto out_free_tr;
9247
9248 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9249
9250 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9251
9252 raw_spin_lock_init(&tr->start_lock);
9253
9254 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9255
9256 tr->current_trace = &nop_trace;
9257
9258 INIT_LIST_HEAD(&tr->systems);
9259 INIT_LIST_HEAD(&tr->events);
9260 INIT_LIST_HEAD(&tr->hist_vars);
9261 INIT_LIST_HEAD(&tr->err_log);
9262
9263 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9264 goto out_free_tr;
9265
9266 if (ftrace_allocate_ftrace_ops(tr) < 0)
9267 goto out_free_tr;
9268
9269 ftrace_init_trace_array(tr);
9270
9271 init_trace_flags_index(tr);
9272
9273 if (trace_instance_dir) {
9274 ret = trace_array_create_dir(tr);
9275 if (ret)
9276 goto out_free_tr;
9277 } else
9278 __trace_early_add_events(tr);
9279
9280 list_add(&tr->list, &ftrace_trace_arrays);
9281
9282 tr->ref++;
9283
9284 return tr;
9285
9286 out_free_tr:
9287 ftrace_free_ftrace_ops(tr);
9288 free_trace_buffers(tr);
9289 free_cpumask_var(tr->tracing_cpumask);
9290 kfree(tr->name);
9291 kfree(tr);
9292
9293 return ERR_PTR(ret);
9294}
9295
9296static int instance_mkdir(const char *name)
9297{
9298 struct trace_array *tr;
9299 int ret;
9300
9301 mutex_lock(&event_mutex);
9302 mutex_lock(&trace_types_lock);
9303
9304 ret = -EEXIST;
9305 if (trace_array_find(name))
9306 goto out_unlock;
9307
9308 tr = trace_array_create(name);
9309
9310 ret = PTR_ERR_OR_ZERO(tr);
9311
9312out_unlock:
9313 mutex_unlock(&trace_types_lock);
9314 mutex_unlock(&event_mutex);
9315 return ret;
9316}
9317
9318/**
9319 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9320 * @name: The name of the trace array to be looked up/created.
9321 *
9322 * Returns pointer to trace array with given name.
9323 * NULL, if it cannot be created.
9324 *
9325 * NOTE: This function increments the reference counter associated with the
9326 * trace array returned. This makes sure it cannot be freed while in use.
9327 * Use trace_array_put() once the trace array is no longer needed.
9328 * If the trace_array is to be freed, trace_array_destroy() needs to
9329 * be called after the trace_array_put(), or simply let user space delete
9330 * it from the tracefs instances directory. But until the
9331 * trace_array_put() is called, user space can not delete it.
9332 *
9333 */
9334struct trace_array *trace_array_get_by_name(const char *name)
9335{
9336 struct trace_array *tr;
9337
9338 mutex_lock(&event_mutex);
9339 mutex_lock(&trace_types_lock);
9340
9341 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9342 if (tr->name && strcmp(tr->name, name) == 0)
9343 goto out_unlock;
9344 }
9345
9346 tr = trace_array_create(name);
9347
9348 if (IS_ERR(tr))
9349 tr = NULL;
9350out_unlock:
9351 if (tr)
9352 tr->ref++;
9353
9354 mutex_unlock(&trace_types_lock);
9355 mutex_unlock(&event_mutex);
9356 return tr;
9357}
9358EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9359
9360static int __remove_instance(struct trace_array *tr)
9361{
9362 int i;
9363
9364 /* Reference counter for a newly created trace array = 1. */
9365 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9366 return -EBUSY;
9367
9368 list_del(&tr->list);
9369
9370 /* Disable all the flags that were enabled coming in */
9371 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9372 if ((1 << i) & ZEROED_TRACE_FLAGS)
9373 set_tracer_flag(tr, 1 << i, 0);
9374 }
9375
9376 tracing_set_nop(tr);
9377 clear_ftrace_function_probes(tr);
9378 event_trace_del_tracer(tr);
9379 ftrace_clear_pids(tr);
9380 ftrace_destroy_function_files(tr);
9381 tracefs_remove(tr->dir);
9382 free_percpu(tr->last_func_repeats);
9383 free_trace_buffers(tr);
9384
9385 for (i = 0; i < tr->nr_topts; i++) {
9386 kfree(tr->topts[i].topts);
9387 }
9388 kfree(tr->topts);
9389
9390 free_cpumask_var(tr->tracing_cpumask);
9391 kfree(tr->name);
9392 kfree(tr);
9393
9394 return 0;
9395}
9396
9397int trace_array_destroy(struct trace_array *this_tr)
9398{
9399 struct trace_array *tr;
9400 int ret;
9401
9402 if (!this_tr)
9403 return -EINVAL;
9404
9405 mutex_lock(&event_mutex);
9406 mutex_lock(&trace_types_lock);
9407
9408 ret = -ENODEV;
9409
9410 /* Making sure trace array exists before destroying it. */
9411 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9412 if (tr == this_tr) {
9413 ret = __remove_instance(tr);
9414 break;
9415 }
9416 }
9417
9418 mutex_unlock(&trace_types_lock);
9419 mutex_unlock(&event_mutex);
9420
9421 return ret;
9422}
9423EXPORT_SYMBOL_GPL(trace_array_destroy);
9424
9425static int instance_rmdir(const char *name)
9426{
9427 struct trace_array *tr;
9428 int ret;
9429
9430 mutex_lock(&event_mutex);
9431 mutex_lock(&trace_types_lock);
9432
9433 ret = -ENODEV;
9434 tr = trace_array_find(name);
9435 if (tr)
9436 ret = __remove_instance(tr);
9437
9438 mutex_unlock(&trace_types_lock);
9439 mutex_unlock(&event_mutex);
9440
9441 return ret;
9442}
9443
9444static __init void create_trace_instances(struct dentry *d_tracer)
9445{
9446 struct trace_array *tr;
9447
9448 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9449 instance_mkdir,
9450 instance_rmdir);
9451 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9452 return;
9453
9454 mutex_lock(&event_mutex);
9455 mutex_lock(&trace_types_lock);
9456
9457 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9458 if (!tr->name)
9459 continue;
9460 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9461 "Failed to create instance directory\n"))
9462 break;
9463 }
9464
9465 mutex_unlock(&trace_types_lock);
9466 mutex_unlock(&event_mutex);
9467}
9468
9469static void
9470init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9471{
9472 struct trace_event_file *file;
9473 int cpu;
9474
9475 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9476 tr, &show_traces_fops);
9477
9478 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9479 tr, &set_tracer_fops);
9480
9481 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9482 tr, &tracing_cpumask_fops);
9483
9484 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9485 tr, &tracing_iter_fops);
9486
9487 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9488 tr, &tracing_fops);
9489
9490 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9491 tr, &tracing_pipe_fops);
9492
9493 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9494 tr, &tracing_entries_fops);
9495
9496 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9497 tr, &tracing_total_entries_fops);
9498
9499 trace_create_file("free_buffer", 0200, d_tracer,
9500 tr, &tracing_free_buffer_fops);
9501
9502 trace_create_file("trace_marker", 0220, d_tracer,
9503 tr, &tracing_mark_fops);
9504
9505 file = __find_event_file(tr, "ftrace", "print");
9506 if (file && file->dir)
9507 trace_create_file("trigger", TRACE_MODE_WRITE, file->dir,
9508 file, &event_trigger_fops);
9509 tr->trace_marker_file = file;
9510
9511 trace_create_file("trace_marker_raw", 0220, d_tracer,
9512 tr, &tracing_mark_raw_fops);
9513
9514 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9515 &trace_clock_fops);
9516
9517 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9518 tr, &rb_simple_fops);
9519
9520 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9521 &trace_time_stamp_mode_fops);
9522
9523 tr->buffer_percent = 50;
9524
9525 trace_create_file("buffer_percent", TRACE_MODE_READ, d_tracer,
9526 tr, &buffer_percent_fops);
9527
9528 create_trace_options_dir(tr);
9529
9530 trace_create_maxlat_file(tr, d_tracer);
9531
9532 if (ftrace_create_function_files(tr, d_tracer))
9533 MEM_FAIL(1, "Could not allocate function filter files");
9534
9535#ifdef CONFIG_TRACER_SNAPSHOT
9536 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9537 tr, &snapshot_fops);
9538#endif
9539
9540 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9541 tr, &tracing_err_log_fops);
9542
9543 for_each_tracing_cpu(cpu)
9544 tracing_init_tracefs_percpu(tr, cpu);
9545
9546 ftrace_init_tracefs(tr, d_tracer);
9547}
9548
9549static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9550{
9551 struct vfsmount *mnt;
9552 struct file_system_type *type;
9553
9554 /*
9555 * To maintain backward compatibility for tools that mount
9556 * debugfs to get to the tracing facility, tracefs is automatically
9557 * mounted to the debugfs/tracing directory.
9558 */
9559 type = get_fs_type("tracefs");
9560 if (!type)
9561 return NULL;
9562 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9563 put_filesystem(type);
9564 if (IS_ERR(mnt))
9565 return NULL;
9566 mntget(mnt);
9567
9568 return mnt;
9569}
9570
9571/**
9572 * tracing_init_dentry - initialize top level trace array
9573 *
9574 * This is called when creating files or directories in the tracing
9575 * directory. It is called via fs_initcall() by any of the boot up code
9576 * and expects to return the dentry of the top level tracing directory.
9577 */
9578int tracing_init_dentry(void)
9579{
9580 struct trace_array *tr = &global_trace;
9581
9582 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9583 pr_warn("Tracing disabled due to lockdown\n");
9584 return -EPERM;
9585 }
9586
9587 /* The top level trace array uses NULL as parent */
9588 if (tr->dir)
9589 return 0;
9590
9591 if (WARN_ON(!tracefs_initialized()))
9592 return -ENODEV;
9593
9594 /*
9595 * As there may still be users that expect the tracing
9596 * files to exist in debugfs/tracing, we must automount
9597 * the tracefs file system there, so older tools still
9598 * work with the newer kernel.
9599 */
9600 tr->dir = debugfs_create_automount("tracing", NULL,
9601 trace_automount, NULL);
9602
9603 return 0;
9604}
9605
9606extern struct trace_eval_map *__start_ftrace_eval_maps[];
9607extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9608
9609static struct workqueue_struct *eval_map_wq __initdata;
9610static struct work_struct eval_map_work __initdata;
9611
9612static void __init eval_map_work_func(struct work_struct *work)
9613{
9614 int len;
9615
9616 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9617 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9618}
9619
9620static int __init trace_eval_init(void)
9621{
9622 INIT_WORK(&eval_map_work, eval_map_work_func);
9623
9624 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9625 if (!eval_map_wq) {
9626 pr_err("Unable to allocate eval_map_wq\n");
9627 /* Do work here */
9628 eval_map_work_func(&eval_map_work);
9629 return -ENOMEM;
9630 }
9631
9632 queue_work(eval_map_wq, &eval_map_work);
9633 return 0;
9634}
9635
9636static int __init trace_eval_sync(void)
9637{
9638 /* Make sure the eval map updates are finished */
9639 if (eval_map_wq)
9640 destroy_workqueue(eval_map_wq);
9641 return 0;
9642}
9643
9644late_initcall_sync(trace_eval_sync);
9645
9646
9647#ifdef CONFIG_MODULES
9648static void trace_module_add_evals(struct module *mod)
9649{
9650 if (!mod->num_trace_evals)
9651 return;
9652
9653 /*
9654 * Modules with bad taint do not have events created, do
9655 * not bother with enums either.
9656 */
9657 if (trace_module_has_bad_taint(mod))
9658 return;
9659
9660 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9661}
9662
9663#ifdef CONFIG_TRACE_EVAL_MAP_FILE
9664static void trace_module_remove_evals(struct module *mod)
9665{
9666 union trace_eval_map_item *map;
9667 union trace_eval_map_item **last = &trace_eval_maps;
9668
9669 if (!mod->num_trace_evals)
9670 return;
9671
9672 mutex_lock(&trace_eval_mutex);
9673
9674 map = trace_eval_maps;
9675
9676 while (map) {
9677 if (map->head.mod == mod)
9678 break;
9679 map = trace_eval_jmp_to_tail(map);
9680 last = &map->tail.next;
9681 map = map->tail.next;
9682 }
9683 if (!map)
9684 goto out;
9685
9686 *last = trace_eval_jmp_to_tail(map)->tail.next;
9687 kfree(map);
9688 out:
9689 mutex_unlock(&trace_eval_mutex);
9690}
9691#else
9692static inline void trace_module_remove_evals(struct module *mod) { }
9693#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9694
9695static int trace_module_notify(struct notifier_block *self,
9696 unsigned long val, void *data)
9697{
9698 struct module *mod = data;
9699
9700 switch (val) {
9701 case MODULE_STATE_COMING:
9702 trace_module_add_evals(mod);
9703 break;
9704 case MODULE_STATE_GOING:
9705 trace_module_remove_evals(mod);
9706 break;
9707 }
9708
9709 return NOTIFY_OK;
9710}
9711
9712static struct notifier_block trace_module_nb = {
9713 .notifier_call = trace_module_notify,
9714 .priority = 0,
9715};
9716#endif /* CONFIG_MODULES */
9717
9718static __init int tracer_init_tracefs(void)
9719{
9720 int ret;
9721
9722 trace_access_lock_init();
9723
9724 ret = tracing_init_dentry();
9725 if (ret)
9726 return 0;
9727
9728 event_trace_init();
9729
9730 init_tracer_tracefs(&global_trace, NULL);
9731 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9732
9733 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9734 &global_trace, &tracing_thresh_fops);
9735
9736 trace_create_file("README", TRACE_MODE_READ, NULL,
9737 NULL, &tracing_readme_fops);
9738
9739 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9740 NULL, &tracing_saved_cmdlines_fops);
9741
9742 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9743 NULL, &tracing_saved_cmdlines_size_fops);
9744
9745 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9746 NULL, &tracing_saved_tgids_fops);
9747
9748 trace_eval_init();
9749
9750 trace_create_eval_file(NULL);
9751
9752#ifdef CONFIG_MODULES
9753 register_module_notifier(&trace_module_nb);
9754#endif
9755
9756#ifdef CONFIG_DYNAMIC_FTRACE
9757 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
9758 NULL, &tracing_dyn_info_fops);
9759#endif
9760
9761 create_trace_instances(NULL);
9762
9763 update_tracer_options(&global_trace);
9764
9765 return 0;
9766}
9767
9768fs_initcall(tracer_init_tracefs);
9769
9770static int trace_panic_handler(struct notifier_block *this,
9771 unsigned long event, void *unused)
9772{
9773 if (ftrace_dump_on_oops)
9774 ftrace_dump(ftrace_dump_on_oops);
9775 return NOTIFY_OK;
9776}
9777
9778static struct notifier_block trace_panic_notifier = {
9779 .notifier_call = trace_panic_handler,
9780 .next = NULL,
9781 .priority = 150 /* priority: INT_MAX >= x >= 0 */
9782};
9783
9784static int trace_die_handler(struct notifier_block *self,
9785 unsigned long val,
9786 void *data)
9787{
9788 switch (val) {
9789 case DIE_OOPS:
9790 if (ftrace_dump_on_oops)
9791 ftrace_dump(ftrace_dump_on_oops);
9792 break;
9793 default:
9794 break;
9795 }
9796 return NOTIFY_OK;
9797}
9798
9799static struct notifier_block trace_die_notifier = {
9800 .notifier_call = trace_die_handler,
9801 .priority = 200
9802};
9803
9804/*
9805 * printk is set to max of 1024, we really don't need it that big.
9806 * Nothing should be printing 1000 characters anyway.
9807 */
9808#define TRACE_MAX_PRINT 1000
9809
9810/*
9811 * Define here KERN_TRACE so that we have one place to modify
9812 * it if we decide to change what log level the ftrace dump
9813 * should be at.
9814 */
9815#define KERN_TRACE KERN_EMERG
9816
9817void
9818trace_printk_seq(struct trace_seq *s)
9819{
9820 /* Probably should print a warning here. */
9821 if (s->seq.len >= TRACE_MAX_PRINT)
9822 s->seq.len = TRACE_MAX_PRINT;
9823
9824 /*
9825 * More paranoid code. Although the buffer size is set to
9826 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9827 * an extra layer of protection.
9828 */
9829 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9830 s->seq.len = s->seq.size - 1;
9831
9832 /* should be zero ended, but we are paranoid. */
9833 s->buffer[s->seq.len] = 0;
9834
9835 printk(KERN_TRACE "%s", s->buffer);
9836
9837 trace_seq_init(s);
9838}
9839
9840void trace_init_global_iter(struct trace_iterator *iter)
9841{
9842 iter->tr = &global_trace;
9843 iter->trace = iter->tr->current_trace;
9844 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9845 iter->array_buffer = &global_trace.array_buffer;
9846
9847 if (iter->trace && iter->trace->open)
9848 iter->trace->open(iter);
9849
9850 /* Annotate start of buffers if we had overruns */
9851 if (ring_buffer_overruns(iter->array_buffer->buffer))
9852 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9853
9854 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9855 if (trace_clocks[iter->tr->clock_id].in_ns)
9856 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9857}
9858
9859void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
9860{
9861 /* use static because iter can be a bit big for the stack */
9862 static struct trace_iterator iter;
9863 static atomic_t dump_running;
9864 struct trace_array *tr = &global_trace;
9865 unsigned int old_userobj;
9866 unsigned long flags;
9867 int cnt = 0, cpu;
9868
9869 /* Only allow one dump user at a time. */
9870 if (atomic_inc_return(&dump_running) != 1) {
9871 atomic_dec(&dump_running);
9872 return;
9873 }
9874
9875 /*
9876 * Always turn off tracing when we dump.
9877 * We don't need to show trace output of what happens
9878 * between multiple crashes.
9879 *
9880 * If the user does a sysrq-z, then they can re-enable
9881 * tracing with echo 1 > tracing_on.
9882 */
9883 tracing_off();
9884
9885 local_irq_save(flags);
9886
9887 /* Simulate the iterator */
9888 trace_init_global_iter(&iter);
9889 /* Can not use kmalloc for iter.temp and iter.fmt */
9890 iter.temp = static_temp_buf;
9891 iter.temp_size = STATIC_TEMP_BUF_SIZE;
9892 iter.fmt = static_fmt_buf;
9893 iter.fmt_size = STATIC_FMT_BUF_SIZE;
9894
9895 for_each_tracing_cpu(cpu) {
9896 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9897 }
9898
9899 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
9900
9901 /* don't look at user memory in panic mode */
9902 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
9903
9904 switch (oops_dump_mode) {
9905 case DUMP_ALL:
9906 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9907 break;
9908 case DUMP_ORIG:
9909 iter.cpu_file = raw_smp_processor_id();
9910 break;
9911 case DUMP_NONE:
9912 goto out_enable;
9913 default:
9914 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
9915 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9916 }
9917
9918 printk(KERN_TRACE "Dumping ftrace buffer:\n");
9919
9920 /* Did function tracer already get disabled? */
9921 if (ftrace_is_dead()) {
9922 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
9923 printk("# MAY BE MISSING FUNCTION EVENTS\n");
9924 }
9925
9926 /*
9927 * We need to stop all tracing on all CPUS to read
9928 * the next buffer. This is a bit expensive, but is
9929 * not done often. We fill all what we can read,
9930 * and then release the locks again.
9931 */
9932
9933 while (!trace_empty(&iter)) {
9934
9935 if (!cnt)
9936 printk(KERN_TRACE "---------------------------------\n");
9937
9938 cnt++;
9939
9940 trace_iterator_reset(&iter);
9941 iter.iter_flags |= TRACE_FILE_LAT_FMT;
9942
9943 if (trace_find_next_entry_inc(&iter) != NULL) {
9944 int ret;
9945
9946 ret = print_trace_line(&iter);
9947 if (ret != TRACE_TYPE_NO_CONSUME)
9948 trace_consume(&iter);
9949 }
9950 touch_nmi_watchdog();
9951
9952 trace_printk_seq(&iter.seq);
9953 }
9954
9955 if (!cnt)
9956 printk(KERN_TRACE " (ftrace buffer empty)\n");
9957 else
9958 printk(KERN_TRACE "---------------------------------\n");
9959
9960 out_enable:
9961 tr->trace_flags |= old_userobj;
9962
9963 for_each_tracing_cpu(cpu) {
9964 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9965 }
9966 atomic_dec(&dump_running);
9967 local_irq_restore(flags);
9968}
9969EXPORT_SYMBOL_GPL(ftrace_dump);
9970
9971#define WRITE_BUFSIZE 4096
9972
9973ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
9974 size_t count, loff_t *ppos,
9975 int (*createfn)(const char *))
9976{
9977 char *kbuf, *buf, *tmp;
9978 int ret = 0;
9979 size_t done = 0;
9980 size_t size;
9981
9982 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
9983 if (!kbuf)
9984 return -ENOMEM;
9985
9986 while (done < count) {
9987 size = count - done;
9988
9989 if (size >= WRITE_BUFSIZE)
9990 size = WRITE_BUFSIZE - 1;
9991
9992 if (copy_from_user(kbuf, buffer + done, size)) {
9993 ret = -EFAULT;
9994 goto out;
9995 }
9996 kbuf[size] = '\0';
9997 buf = kbuf;
9998 do {
9999 tmp = strchr(buf, '\n');
10000 if (tmp) {
10001 *tmp = '\0';
10002 size = tmp - buf + 1;
10003 } else {
10004 size = strlen(buf);
10005 if (done + size < count) {
10006 if (buf != kbuf)
10007 break;
10008 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10009 pr_warn("Line length is too long: Should be less than %d\n",
10010 WRITE_BUFSIZE - 2);
10011 ret = -EINVAL;
10012 goto out;
10013 }
10014 }
10015 done += size;
10016
10017 /* Remove comments */
10018 tmp = strchr(buf, '#');
10019
10020 if (tmp)
10021 *tmp = '\0';
10022
10023 ret = createfn(buf);
10024 if (ret)
10025 goto out;
10026 buf += size;
10027
10028 } while (done < count);
10029 }
10030 ret = done;
10031
10032out:
10033 kfree(kbuf);
10034
10035 return ret;
10036}
10037
10038__init static int tracer_alloc_buffers(void)
10039{
10040 int ring_buf_size;
10041 int ret = -ENOMEM;
10042
10043
10044 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10045 pr_warn("Tracing disabled due to lockdown\n");
10046 return -EPERM;
10047 }
10048
10049 /*
10050 * Make sure we don't accidentally add more trace options
10051 * than we have bits for.
10052 */
10053 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10054
10055 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10056 goto out;
10057
10058 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10059 goto out_free_buffer_mask;
10060
10061 /* Only allocate trace_printk buffers if a trace_printk exists */
10062 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10063 /* Must be called before global_trace.buffer is allocated */
10064 trace_printk_init_buffers();
10065
10066 /* To save memory, keep the ring buffer size to its minimum */
10067 if (ring_buffer_expanded)
10068 ring_buf_size = trace_buf_size;
10069 else
10070 ring_buf_size = 1;
10071
10072 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10073 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10074
10075 raw_spin_lock_init(&global_trace.start_lock);
10076
10077 /*
10078 * The prepare callbacks allocates some memory for the ring buffer. We
10079 * don't free the buffer if the CPU goes down. If we were to free
10080 * the buffer, then the user would lose any trace that was in the
10081 * buffer. The memory will be removed once the "instance" is removed.
10082 */
10083 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10084 "trace/RB:preapre", trace_rb_cpu_prepare,
10085 NULL);
10086 if (ret < 0)
10087 goto out_free_cpumask;
10088 /* Used for event triggers */
10089 ret = -ENOMEM;
10090 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10091 if (!temp_buffer)
10092 goto out_rm_hp_state;
10093
10094 if (trace_create_savedcmd() < 0)
10095 goto out_free_temp_buffer;
10096
10097 /* TODO: make the number of buffers hot pluggable with CPUS */
10098 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10099 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10100 goto out_free_savedcmd;
10101 }
10102
10103 if (global_trace.buffer_disabled)
10104 tracing_off();
10105
10106 if (trace_boot_clock) {
10107 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10108 if (ret < 0)
10109 pr_warn("Trace clock %s not defined, going back to default\n",
10110 trace_boot_clock);
10111 }
10112
10113 /*
10114 * register_tracer() might reference current_trace, so it
10115 * needs to be set before we register anything. This is
10116 * just a bootstrap of current_trace anyway.
10117 */
10118 global_trace.current_trace = &nop_trace;
10119
10120 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10121
10122 ftrace_init_global_array_ops(&global_trace);
10123
10124 init_trace_flags_index(&global_trace);
10125
10126 register_tracer(&nop_trace);
10127
10128 /* Function tracing may start here (via kernel command line) */
10129 init_function_trace();
10130
10131 /* All seems OK, enable tracing */
10132 tracing_disabled = 0;
10133
10134 atomic_notifier_chain_register(&panic_notifier_list,
10135 &trace_panic_notifier);
10136
10137 register_die_notifier(&trace_die_notifier);
10138
10139 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10140
10141 INIT_LIST_HEAD(&global_trace.systems);
10142 INIT_LIST_HEAD(&global_trace.events);
10143 INIT_LIST_HEAD(&global_trace.hist_vars);
10144 INIT_LIST_HEAD(&global_trace.err_log);
10145 list_add(&global_trace.list, &ftrace_trace_arrays);
10146
10147 apply_trace_boot_options();
10148
10149 register_snapshot_cmd();
10150
10151 test_can_verify();
10152
10153 return 0;
10154
10155out_free_savedcmd:
10156 free_saved_cmdlines_buffer(savedcmd);
10157out_free_temp_buffer:
10158 ring_buffer_free(temp_buffer);
10159out_rm_hp_state:
10160 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10161out_free_cpumask:
10162 free_cpumask_var(global_trace.tracing_cpumask);
10163out_free_buffer_mask:
10164 free_cpumask_var(tracing_buffer_mask);
10165out:
10166 return ret;
10167}
10168
10169void __init ftrace_boot_snapshot(void)
10170{
10171 if (snapshot_at_boot) {
10172 tracing_snapshot();
10173 internal_trace_puts("** Boot snapshot taken **\n");
10174 }
10175}
10176
10177void __init early_trace_init(void)
10178{
10179 if (tracepoint_printk) {
10180 tracepoint_print_iter =
10181 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10182 if (MEM_FAIL(!tracepoint_print_iter,
10183 "Failed to allocate trace iterator\n"))
10184 tracepoint_printk = 0;
10185 else
10186 static_key_enable(&tracepoint_printk_key.key);
10187 }
10188 tracer_alloc_buffers();
10189}
10190
10191void __init trace_init(void)
10192{
10193 trace_event_init();
10194}
10195
10196__init static void clear_boot_tracer(void)
10197{
10198 /*
10199 * The default tracer at boot buffer is an init section.
10200 * This function is called in lateinit. If we did not
10201 * find the boot tracer, then clear it out, to prevent
10202 * later registration from accessing the buffer that is
10203 * about to be freed.
10204 */
10205 if (!default_bootup_tracer)
10206 return;
10207
10208 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10209 default_bootup_tracer);
10210 default_bootup_tracer = NULL;
10211}
10212
10213#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10214__init static void tracing_set_default_clock(void)
10215{
10216 /* sched_clock_stable() is determined in late_initcall */
10217 if (!trace_boot_clock && !sched_clock_stable()) {
10218 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10219 pr_warn("Can not set tracing clock due to lockdown\n");
10220 return;
10221 }
10222
10223 printk(KERN_WARNING
10224 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10225 "If you want to keep using the local clock, then add:\n"
10226 " \"trace_clock=local\"\n"
10227 "on the kernel command line\n");
10228 tracing_set_clock(&global_trace, "global");
10229 }
10230}
10231#else
10232static inline void tracing_set_default_clock(void) { }
10233#endif
10234
10235__init static int late_trace_init(void)
10236{
10237 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10238 static_key_disable(&tracepoint_printk_key.key);
10239 tracepoint_printk = 0;
10240 }
10241
10242 tracing_set_default_clock();
10243 clear_boot_tracer();
10244 return 0;
10245}
10246
10247late_initcall_sync(late_trace_init);