projects/linux·blob v6.2 · refs

10355 lines · 252.1 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/irqflags.h>
23#include <linux/debugfs.h>
24#include <linux/tracefs.h>
25#include <linux/pagemap.h>
26#include <linux/hardirq.h>
27#include <linux/linkage.h>
28#include <linux/uaccess.h>
29#include <linux/vmalloc.h>
30#include <linux/ftrace.h>
31#include <linux/module.h>
32#include <linux/percpu.h>
33#include <linux/splice.h>
34#include <linux/kdebug.h>
35#include <linux/string.h>
36#include <linux/mount.h>
37#include <linux/rwsem.h>
38#include <linux/slab.h>
39#include <linux/ctype.h>
40#include <linux/init.h>
41#include <linux/panic_notifier.h>
42#include <linux/poll.h>
43#include <linux/nmi.h>
44#include <linux/fs.h>
45#include <linux/trace.h>
46#include <linux/sched/clock.h>
47#include <linux/sched/rt.h>
48#include <linux/fsnotify.h>
49#include <linux/irq_work.h>
50#include <linux/workqueue.h>
51
52#include "trace.h"
53#include "trace_output.h"
54
55/*
56 * On boot up, the ring buffer is set to the minimum size, so that
57 * we do not waste memory on systems that are not using tracing.
58 */
59bool ring_buffer_expanded;
60
61/*
62 * We need to change this state when a selftest is running.
63 * A selftest will lurk into the ring-buffer to count the
64 * entries inserted during the selftest although some concurrent
65 * insertions into the ring-buffer such as trace_printk could occurred
66 * at the same time, giving false positive or negative results.
67 */
68static bool __read_mostly tracing_selftest_running;
69
70/*
71 * If boot-time tracing including tracers/events via kernel cmdline
72 * is running, we do not want to run SELFTEST.
73 */
74bool __read_mostly tracing_selftest_disabled;
75
76#ifdef CONFIG_FTRACE_STARTUP_TEST
77void __init disable_tracing_selftest(const char *reason)
78{
79 if (!tracing_selftest_disabled) {
80 tracing_selftest_disabled = true;
81 pr_info("Ftrace startup test is disabled due to %s\n", reason);
82 }
83}
84#endif
85
86/* Pipe tracepoints to printk */
87static struct trace_iterator *tracepoint_print_iter;
88int tracepoint_printk;
89static bool tracepoint_printk_stop_on_boot __initdata;
90static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
91
92/* For tracers that don't implement custom flags */
93static struct tracer_opt dummy_tracer_opt[] = {
94 { }
95};
96
97static int
98dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
99{
100 return 0;
101}
102
103/*
104 * To prevent the comm cache from being overwritten when no
105 * tracing is active, only save the comm when a trace event
106 * occurred.
107 */
108static DEFINE_PER_CPU(bool, trace_taskinfo_save);
109
110/*
111 * Kill all tracing for good (never come back).
112 * It is initialized to 1 but will turn to zero if the initialization
113 * of the tracer is successful. But that is the only place that sets
114 * this back to zero.
115 */
116static int tracing_disabled = 1;
117
118cpumask_var_t __read_mostly tracing_buffer_mask;
119
120/*
121 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
122 *
123 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
124 * is set, then ftrace_dump is called. This will output the contents
125 * of the ftrace buffers to the console. This is very useful for
126 * capturing traces that lead to crashes and outputing it to a
127 * serial console.
128 *
129 * It is default off, but you can enable it with either specifying
130 * "ftrace_dump_on_oops" in the kernel command line, or setting
131 * /proc/sys/kernel/ftrace_dump_on_oops
132 * Set 1 if you want to dump buffers of all CPUs
133 * Set 2 if you want to dump the buffer of the CPU that triggered oops
134 */
135
136enum ftrace_dump_mode ftrace_dump_on_oops;
137
138/* When set, tracing will stop when a WARN*() is hit */
139int __disable_trace_on_warning;
140
141#ifdef CONFIG_TRACE_EVAL_MAP_FILE
142/* Map of enums to their values, for "eval_map" file */
143struct trace_eval_map_head {
144 struct module *mod;
145 unsigned long length;
146};
147
148union trace_eval_map_item;
149
150struct trace_eval_map_tail {
151 /*
152 * "end" is first and points to NULL as it must be different
153 * than "mod" or "eval_string"
154 */
155 union trace_eval_map_item *next;
156 const char *end; /* points to NULL */
157};
158
159static DEFINE_MUTEX(trace_eval_mutex);
160
161/*
162 * The trace_eval_maps are saved in an array with two extra elements,
163 * one at the beginning, and one at the end. The beginning item contains
164 * the count of the saved maps (head.length), and the module they
165 * belong to if not built in (head.mod). The ending item contains a
166 * pointer to the next array of saved eval_map items.
167 */
168union trace_eval_map_item {
169 struct trace_eval_map map;
170 struct trace_eval_map_head head;
171 struct trace_eval_map_tail tail;
172};
173
174static union trace_eval_map_item *trace_eval_maps;
175#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
176
177int tracing_set_tracer(struct trace_array *tr, const char *buf);
178static void ftrace_trace_userstack(struct trace_array *tr,
179 struct trace_buffer *buffer,
180 unsigned int trace_ctx);
181
182#define MAX_TRACER_SIZE 100
183static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
184static char *default_bootup_tracer;
185
186static bool allocate_snapshot;
187static bool snapshot_at_boot;
188
189static int __init set_cmdline_ftrace(char *str)
190{
191 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
192 default_bootup_tracer = bootup_tracer_buf;
193 /* We are using ftrace early, expand it */
194 ring_buffer_expanded = true;
195 return 1;
196}
197__setup("ftrace=", set_cmdline_ftrace);
198
199static int __init set_ftrace_dump_on_oops(char *str)
200{
201 if (*str++ != '=' || !*str || !strcmp("1", str)) {
202 ftrace_dump_on_oops = DUMP_ALL;
203 return 1;
204 }
205
206 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
207 ftrace_dump_on_oops = DUMP_ORIG;
208 return 1;
209 }
210
211 return 0;
212}
213__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
214
215static int __init stop_trace_on_warning(char *str)
216{
217 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
218 __disable_trace_on_warning = 1;
219 return 1;
220}
221__setup("traceoff_on_warning", stop_trace_on_warning);
222
223static int __init boot_alloc_snapshot(char *str)
224{
225 allocate_snapshot = true;
226 /* We also need the main ring buffer expanded */
227 ring_buffer_expanded = true;
228 return 1;
229}
230__setup("alloc_snapshot", boot_alloc_snapshot);
231
232
233static int __init boot_snapshot(char *str)
234{
235 snapshot_at_boot = true;
236 boot_alloc_snapshot(str);
237 return 1;
238}
239__setup("ftrace_boot_snapshot", boot_snapshot);
240
241
242static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
243
244static int __init set_trace_boot_options(char *str)
245{
246 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
247 return 1;
248}
249__setup("trace_options=", set_trace_boot_options);
250
251static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
252static char *trace_boot_clock __initdata;
253
254static int __init set_trace_boot_clock(char *str)
255{
256 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
257 trace_boot_clock = trace_boot_clock_buf;
258 return 1;
259}
260__setup("trace_clock=", set_trace_boot_clock);
261
262static int __init set_tracepoint_printk(char *str)
263{
264 /* Ignore the "tp_printk_stop_on_boot" param */
265 if (*str == '_')
266 return 0;
267
268 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
269 tracepoint_printk = 1;
270 return 1;
271}
272__setup("tp_printk", set_tracepoint_printk);
273
274static int __init set_tracepoint_printk_stop(char *str)
275{
276 tracepoint_printk_stop_on_boot = true;
277 return 1;
278}
279__setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
280
281unsigned long long ns2usecs(u64 nsec)
282{
283 nsec += 500;
284 do_div(nsec, 1000);
285 return nsec;
286}
287
288static void
289trace_process_export(struct trace_export *export,
290 struct ring_buffer_event *event, int flag)
291{
292 struct trace_entry *entry;
293 unsigned int size = 0;
294
295 if (export->flags & flag) {
296 entry = ring_buffer_event_data(event);
297 size = ring_buffer_event_length(event);
298 export->write(export, entry, size);
299 }
300}
301
302static DEFINE_MUTEX(ftrace_export_lock);
303
304static struct trace_export __rcu *ftrace_exports_list __read_mostly;
305
306static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
307static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
308static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
309
310static inline void ftrace_exports_enable(struct trace_export *export)
311{
312 if (export->flags & TRACE_EXPORT_FUNCTION)
313 static_branch_inc(&trace_function_exports_enabled);
314
315 if (export->flags & TRACE_EXPORT_EVENT)
316 static_branch_inc(&trace_event_exports_enabled);
317
318 if (export->flags & TRACE_EXPORT_MARKER)
319 static_branch_inc(&trace_marker_exports_enabled);
320}
321
322static inline void ftrace_exports_disable(struct trace_export *export)
323{
324 if (export->flags & TRACE_EXPORT_FUNCTION)
325 static_branch_dec(&trace_function_exports_enabled);
326
327 if (export->flags & TRACE_EXPORT_EVENT)
328 static_branch_dec(&trace_event_exports_enabled);
329
330 if (export->flags & TRACE_EXPORT_MARKER)
331 static_branch_dec(&trace_marker_exports_enabled);
332}
333
334static void ftrace_exports(struct ring_buffer_event *event, int flag)
335{
336 struct trace_export *export;
337
338 preempt_disable_notrace();
339
340 export = rcu_dereference_raw_check(ftrace_exports_list);
341 while (export) {
342 trace_process_export(export, event, flag);
343 export = rcu_dereference_raw_check(export->next);
344 }
345
346 preempt_enable_notrace();
347}
348
349static inline void
350add_trace_export(struct trace_export **list, struct trace_export *export)
351{
352 rcu_assign_pointer(export->next, *list);
353 /*
354 * We are entering export into the list but another
355 * CPU might be walking that list. We need to make sure
356 * the export->next pointer is valid before another CPU sees
357 * the export pointer included into the list.
358 */
359 rcu_assign_pointer(*list, export);
360}
361
362static inline int
363rm_trace_export(struct trace_export **list, struct trace_export *export)
364{
365 struct trace_export **p;
366
367 for (p = list; *p != NULL; p = &(*p)->next)
368 if (*p == export)
369 break;
370
371 if (*p != export)
372 return -1;
373
374 rcu_assign_pointer(*p, (*p)->next);
375
376 return 0;
377}
378
379static inline void
380add_ftrace_export(struct trace_export **list, struct trace_export *export)
381{
382 ftrace_exports_enable(export);
383
384 add_trace_export(list, export);
385}
386
387static inline int
388rm_ftrace_export(struct trace_export **list, struct trace_export *export)
389{
390 int ret;
391
392 ret = rm_trace_export(list, export);
393 ftrace_exports_disable(export);
394
395 return ret;
396}
397
398int register_ftrace_export(struct trace_export *export)
399{
400 if (WARN_ON_ONCE(!export->write))
401 return -1;
402
403 mutex_lock(&ftrace_export_lock);
404
405 add_ftrace_export(&ftrace_exports_list, export);
406
407 mutex_unlock(&ftrace_export_lock);
408
409 return 0;
410}
411EXPORT_SYMBOL_GPL(register_ftrace_export);
412
413int unregister_ftrace_export(struct trace_export *export)
414{
415 int ret;
416
417 mutex_lock(&ftrace_export_lock);
418
419 ret = rm_ftrace_export(&ftrace_exports_list, export);
420
421 mutex_unlock(&ftrace_export_lock);
422
423 return ret;
424}
425EXPORT_SYMBOL_GPL(unregister_ftrace_export);
426
427/* trace_flags holds trace_options default values */
428#define TRACE_DEFAULT_FLAGS \
429 (FUNCTION_DEFAULT_FLAGS | \
430 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
431 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
432 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
433 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
434 TRACE_ITER_HASH_PTR)
435
436/* trace_options that are only supported by global_trace */
437#define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
438 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
439
440/* trace_flags that are default zero for instances */
441#define ZEROED_TRACE_FLAGS \
442 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
443
444/*
445 * The global_trace is the descriptor that holds the top-level tracing
446 * buffers for the live tracing.
447 */
448static struct trace_array global_trace = {
449 .trace_flags = TRACE_DEFAULT_FLAGS,
450};
451
452LIST_HEAD(ftrace_trace_arrays);
453
454int trace_array_get(struct trace_array *this_tr)
455{
456 struct trace_array *tr;
457 int ret = -ENODEV;
458
459 mutex_lock(&trace_types_lock);
460 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
461 if (tr == this_tr) {
462 tr->ref++;
463 ret = 0;
464 break;
465 }
466 }
467 mutex_unlock(&trace_types_lock);
468
469 return ret;
470}
471
472static void __trace_array_put(struct trace_array *this_tr)
473{
474 WARN_ON(!this_tr->ref);
475 this_tr->ref--;
476}
477
478/**
479 * trace_array_put - Decrement the reference counter for this trace array.
480 * @this_tr : pointer to the trace array
481 *
482 * NOTE: Use this when we no longer need the trace array returned by
483 * trace_array_get_by_name(). This ensures the trace array can be later
484 * destroyed.
485 *
486 */
487void trace_array_put(struct trace_array *this_tr)
488{
489 if (!this_tr)
490 return;
491
492 mutex_lock(&trace_types_lock);
493 __trace_array_put(this_tr);
494 mutex_unlock(&trace_types_lock);
495}
496EXPORT_SYMBOL_GPL(trace_array_put);
497
498int tracing_check_open_get_tr(struct trace_array *tr)
499{
500 int ret;
501
502 ret = security_locked_down(LOCKDOWN_TRACEFS);
503 if (ret)
504 return ret;
505
506 if (tracing_disabled)
507 return -ENODEV;
508
509 if (tr && trace_array_get(tr) < 0)
510 return -ENODEV;
511
512 return 0;
513}
514
515int call_filter_check_discard(struct trace_event_call *call, void *rec,
516 struct trace_buffer *buffer,
517 struct ring_buffer_event *event)
518{
519 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
520 !filter_match_preds(call->filter, rec)) {
521 __trace_event_discard_commit(buffer, event);
522 return 1;
523 }
524
525 return 0;
526}
527
528/**
529 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
530 * @filtered_pids: The list of pids to check
531 * @search_pid: The PID to find in @filtered_pids
532 *
533 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
534 */
535bool
536trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
537{
538 return trace_pid_list_is_set(filtered_pids, search_pid);
539}
540
541/**
542 * trace_ignore_this_task - should a task be ignored for tracing
543 * @filtered_pids: The list of pids to check
544 * @filtered_no_pids: The list of pids not to be traced
545 * @task: The task that should be ignored if not filtered
546 *
547 * Checks if @task should be traced or not from @filtered_pids.
548 * Returns true if @task should *NOT* be traced.
549 * Returns false if @task should be traced.
550 */
551bool
552trace_ignore_this_task(struct trace_pid_list *filtered_pids,
553 struct trace_pid_list *filtered_no_pids,
554 struct task_struct *task)
555{
556 /*
557 * If filtered_no_pids is not empty, and the task's pid is listed
558 * in filtered_no_pids, then return true.
559 * Otherwise, if filtered_pids is empty, that means we can
560 * trace all tasks. If it has content, then only trace pids
561 * within filtered_pids.
562 */
563
564 return (filtered_pids &&
565 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
566 (filtered_no_pids &&
567 trace_find_filtered_pid(filtered_no_pids, task->pid));
568}
569
570/**
571 * trace_filter_add_remove_task - Add or remove a task from a pid_list
572 * @pid_list: The list to modify
573 * @self: The current task for fork or NULL for exit
574 * @task: The task to add or remove
575 *
576 * If adding a task, if @self is defined, the task is only added if @self
577 * is also included in @pid_list. This happens on fork and tasks should
578 * only be added when the parent is listed. If @self is NULL, then the
579 * @task pid will be removed from the list, which would happen on exit
580 * of a task.
581 */
582void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
583 struct task_struct *self,
584 struct task_struct *task)
585{
586 if (!pid_list)
587 return;
588
589 /* For forks, we only add if the forking task is listed */
590 if (self) {
591 if (!trace_find_filtered_pid(pid_list, self->pid))
592 return;
593 }
594
595 /* "self" is set for forks, and NULL for exits */
596 if (self)
597 trace_pid_list_set(pid_list, task->pid);
598 else
599 trace_pid_list_clear(pid_list, task->pid);
600}
601
602/**
603 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
604 * @pid_list: The pid list to show
605 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
606 * @pos: The position of the file
607 *
608 * This is used by the seq_file "next" operation to iterate the pids
609 * listed in a trace_pid_list structure.
610 *
611 * Returns the pid+1 as we want to display pid of zero, but NULL would
612 * stop the iteration.
613 */
614void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
615{
616 long pid = (unsigned long)v;
617 unsigned int next;
618
619 (*pos)++;
620
621 /* pid already is +1 of the actual previous bit */
622 if (trace_pid_list_next(pid_list, pid, &next) < 0)
623 return NULL;
624
625 pid = next;
626
627 /* Return pid + 1 to allow zero to be represented */
628 return (void *)(pid + 1);
629}
630
631/**
632 * trace_pid_start - Used for seq_file to start reading pid lists
633 * @pid_list: The pid list to show
634 * @pos: The position of the file
635 *
636 * This is used by seq_file "start" operation to start the iteration
637 * of listing pids.
638 *
639 * Returns the pid+1 as we want to display pid of zero, but NULL would
640 * stop the iteration.
641 */
642void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
643{
644 unsigned long pid;
645 unsigned int first;
646 loff_t l = 0;
647
648 if (trace_pid_list_first(pid_list, &first) < 0)
649 return NULL;
650
651 pid = first;
652
653 /* Return pid + 1 so that zero can be the exit value */
654 for (pid++; pid && l < *pos;
655 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
656 ;
657 return (void *)pid;
658}
659
660/**
661 * trace_pid_show - show the current pid in seq_file processing
662 * @m: The seq_file structure to write into
663 * @v: A void pointer of the pid (+1) value to display
664 *
665 * Can be directly used by seq_file operations to display the current
666 * pid value.
667 */
668int trace_pid_show(struct seq_file *m, void *v)
669{
670 unsigned long pid = (unsigned long)v - 1;
671
672 seq_printf(m, "%lu\n", pid);
673 return 0;
674}
675
676/* 128 should be much more than enough */
677#define PID_BUF_SIZE 127
678
679int trace_pid_write(struct trace_pid_list *filtered_pids,
680 struct trace_pid_list **new_pid_list,
681 const char __user *ubuf, size_t cnt)
682{
683 struct trace_pid_list *pid_list;
684 struct trace_parser parser;
685 unsigned long val;
686 int nr_pids = 0;
687 ssize_t read = 0;
688 ssize_t ret;
689 loff_t pos;
690 pid_t pid;
691
692 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
693 return -ENOMEM;
694
695 /*
696 * Always recreate a new array. The write is an all or nothing
697 * operation. Always create a new array when adding new pids by
698 * the user. If the operation fails, then the current list is
699 * not modified.
700 */
701 pid_list = trace_pid_list_alloc();
702 if (!pid_list) {
703 trace_parser_put(&parser);
704 return -ENOMEM;
705 }
706
707 if (filtered_pids) {
708 /* copy the current bits to the new max */
709 ret = trace_pid_list_first(filtered_pids, &pid);
710 while (!ret) {
711 trace_pid_list_set(pid_list, pid);
712 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
713 nr_pids++;
714 }
715 }
716
717 ret = 0;
718 while (cnt > 0) {
719
720 pos = 0;
721
722 ret = trace_get_user(&parser, ubuf, cnt, &pos);
723 if (ret < 0)
724 break;
725
726 read += ret;
727 ubuf += ret;
728 cnt -= ret;
729
730 if (!trace_parser_loaded(&parser))
731 break;
732
733 ret = -EINVAL;
734 if (kstrtoul(parser.buffer, 0, &val))
735 break;
736
737 pid = (pid_t)val;
738
739 if (trace_pid_list_set(pid_list, pid) < 0) {
740 ret = -1;
741 break;
742 }
743 nr_pids++;
744
745 trace_parser_clear(&parser);
746 ret = 0;
747 }
748 trace_parser_put(&parser);
749
750 if (ret < 0) {
751 trace_pid_list_free(pid_list);
752 return ret;
753 }
754
755 if (!nr_pids) {
756 /* Cleared the list of pids */
757 trace_pid_list_free(pid_list);
758 pid_list = NULL;
759 }
760
761 *new_pid_list = pid_list;
762
763 return read;
764}
765
766static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
767{
768 u64 ts;
769
770 /* Early boot up does not have a buffer yet */
771 if (!buf->buffer)
772 return trace_clock_local();
773
774 ts = ring_buffer_time_stamp(buf->buffer);
775 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
776
777 return ts;
778}
779
780u64 ftrace_now(int cpu)
781{
782 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
783}
784
785/**
786 * tracing_is_enabled - Show if global_trace has been enabled
787 *
788 * Shows if the global trace has been enabled or not. It uses the
789 * mirror flag "buffer_disabled" to be used in fast paths such as for
790 * the irqsoff tracer. But it may be inaccurate due to races. If you
791 * need to know the accurate state, use tracing_is_on() which is a little
792 * slower, but accurate.
793 */
794int tracing_is_enabled(void)
795{
796 /*
797 * For quick access (irqsoff uses this in fast path), just
798 * return the mirror variable of the state of the ring buffer.
799 * It's a little racy, but we don't really care.
800 */
801 smp_rmb();
802 return !global_trace.buffer_disabled;
803}
804
805/*
806 * trace_buf_size is the size in bytes that is allocated
807 * for a buffer. Note, the number of bytes is always rounded
808 * to page size.
809 *
810 * This number is purposely set to a low number of 16384.
811 * If the dump on oops happens, it will be much appreciated
812 * to not have to wait for all that output. Anyway this can be
813 * boot time and run time configurable.
814 */
815#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
816
817static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
818
819/* trace_types holds a link list of available tracers. */
820static struct tracer *trace_types __read_mostly;
821
822/*
823 * trace_types_lock is used to protect the trace_types list.
824 */
825DEFINE_MUTEX(trace_types_lock);
826
827/*
828 * serialize the access of the ring buffer
829 *
830 * ring buffer serializes readers, but it is low level protection.
831 * The validity of the events (which returns by ring_buffer_peek() ..etc)
832 * are not protected by ring buffer.
833 *
834 * The content of events may become garbage if we allow other process consumes
835 * these events concurrently:
836 * A) the page of the consumed events may become a normal page
837 * (not reader page) in ring buffer, and this page will be rewritten
838 * by events producer.
839 * B) The page of the consumed events may become a page for splice_read,
840 * and this page will be returned to system.
841 *
842 * These primitives allow multi process access to different cpu ring buffer
843 * concurrently.
844 *
845 * These primitives don't distinguish read-only and read-consume access.
846 * Multi read-only access are also serialized.
847 */
848
849#ifdef CONFIG_SMP
850static DECLARE_RWSEM(all_cpu_access_lock);
851static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
852
853static inline void trace_access_lock(int cpu)
854{
855 if (cpu == RING_BUFFER_ALL_CPUS) {
856 /* gain it for accessing the whole ring buffer. */
857 down_write(&all_cpu_access_lock);
858 } else {
859 /* gain it for accessing a cpu ring buffer. */
860
861 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
862 down_read(&all_cpu_access_lock);
863
864 /* Secondly block other access to this @cpu ring buffer. */
865 mutex_lock(&per_cpu(cpu_access_lock, cpu));
866 }
867}
868
869static inline void trace_access_unlock(int cpu)
870{
871 if (cpu == RING_BUFFER_ALL_CPUS) {
872 up_write(&all_cpu_access_lock);
873 } else {
874 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
875 up_read(&all_cpu_access_lock);
876 }
877}
878
879static inline void trace_access_lock_init(void)
880{
881 int cpu;
882
883 for_each_possible_cpu(cpu)
884 mutex_init(&per_cpu(cpu_access_lock, cpu));
885}
886
887#else
888
889static DEFINE_MUTEX(access_lock);
890
891static inline void trace_access_lock(int cpu)
892{
893 (void)cpu;
894 mutex_lock(&access_lock);
895}
896
897static inline void trace_access_unlock(int cpu)
898{
899 (void)cpu;
900 mutex_unlock(&access_lock);
901}
902
903static inline void trace_access_lock_init(void)
904{
905}
906
907#endif
908
909#ifdef CONFIG_STACKTRACE
910static void __ftrace_trace_stack(struct trace_buffer *buffer,
911 unsigned int trace_ctx,
912 int skip, struct pt_regs *regs);
913static inline void ftrace_trace_stack(struct trace_array *tr,
914 struct trace_buffer *buffer,
915 unsigned int trace_ctx,
916 int skip, struct pt_regs *regs);
917
918#else
919static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
920 unsigned int trace_ctx,
921 int skip, struct pt_regs *regs)
922{
923}
924static inline void ftrace_trace_stack(struct trace_array *tr,
925 struct trace_buffer *buffer,
926 unsigned long trace_ctx,
927 int skip, struct pt_regs *regs)
928{
929}
930
931#endif
932
933static __always_inline void
934trace_event_setup(struct ring_buffer_event *event,
935 int type, unsigned int trace_ctx)
936{
937 struct trace_entry *ent = ring_buffer_event_data(event);
938
939 tracing_generic_entry_update(ent, type, trace_ctx);
940}
941
942static __always_inline struct ring_buffer_event *
943__trace_buffer_lock_reserve(struct trace_buffer *buffer,
944 int type,
945 unsigned long len,
946 unsigned int trace_ctx)
947{
948 struct ring_buffer_event *event;
949
950 event = ring_buffer_lock_reserve(buffer, len);
951 if (event != NULL)
952 trace_event_setup(event, type, trace_ctx);
953
954 return event;
955}
956
957void tracer_tracing_on(struct trace_array *tr)
958{
959 if (tr->array_buffer.buffer)
960 ring_buffer_record_on(tr->array_buffer.buffer);
961 /*
962 * This flag is looked at when buffers haven't been allocated
963 * yet, or by some tracers (like irqsoff), that just want to
964 * know if the ring buffer has been disabled, but it can handle
965 * races of where it gets disabled but we still do a record.
966 * As the check is in the fast path of the tracers, it is more
967 * important to be fast than accurate.
968 */
969 tr->buffer_disabled = 0;
970 /* Make the flag seen by readers */
971 smp_wmb();
972}
973
974/**
975 * tracing_on - enable tracing buffers
976 *
977 * This function enables tracing buffers that may have been
978 * disabled with tracing_off.
979 */
980void tracing_on(void)
981{
982 tracer_tracing_on(&global_trace);
983}
984EXPORT_SYMBOL_GPL(tracing_on);
985
986
987static __always_inline void
988__buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
989{
990 __this_cpu_write(trace_taskinfo_save, true);
991
992 /* If this is the temp buffer, we need to commit fully */
993 if (this_cpu_read(trace_buffered_event) == event) {
994 /* Length is in event->array[0] */
995 ring_buffer_write(buffer, event->array[0], &event->array[1]);
996 /* Release the temp buffer */
997 this_cpu_dec(trace_buffered_event_cnt);
998 /* ring_buffer_unlock_commit() enables preemption */
999 preempt_enable_notrace();
1000 } else
1001 ring_buffer_unlock_commit(buffer);
1002}
1003
1004/**
1005 * __trace_puts - write a constant string into the trace buffer.
1006 * @ip: The address of the caller
1007 * @str: The constant string to write
1008 * @size: The size of the string.
1009 */
1010int __trace_puts(unsigned long ip, const char *str, int size)
1011{
1012 struct ring_buffer_event *event;
1013 struct trace_buffer *buffer;
1014 struct print_entry *entry;
1015 unsigned int trace_ctx;
1016 int alloc;
1017
1018 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1019 return 0;
1020
1021 if (unlikely(tracing_selftest_running || tracing_disabled))
1022 return 0;
1023
1024 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1025
1026 trace_ctx = tracing_gen_ctx();
1027 buffer = global_trace.array_buffer.buffer;
1028 ring_buffer_nest_start(buffer);
1029 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1030 trace_ctx);
1031 if (!event) {
1032 size = 0;
1033 goto out;
1034 }
1035
1036 entry = ring_buffer_event_data(event);
1037 entry->ip = ip;
1038
1039 memcpy(&entry->buf, str, size);
1040
1041 /* Add a newline if necessary */
1042 if (entry->buf[size - 1] != '\n') {
1043 entry->buf[size] = '\n';
1044 entry->buf[size + 1] = '\0';
1045 } else
1046 entry->buf[size] = '\0';
1047
1048 __buffer_unlock_commit(buffer, event);
1049 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1050 out:
1051 ring_buffer_nest_end(buffer);
1052 return size;
1053}
1054EXPORT_SYMBOL_GPL(__trace_puts);
1055
1056/**
1057 * __trace_bputs - write the pointer to a constant string into trace buffer
1058 * @ip: The address of the caller
1059 * @str: The constant string to write to the buffer to
1060 */
1061int __trace_bputs(unsigned long ip, const char *str)
1062{
1063 struct ring_buffer_event *event;
1064 struct trace_buffer *buffer;
1065 struct bputs_entry *entry;
1066 unsigned int trace_ctx;
1067 int size = sizeof(struct bputs_entry);
1068 int ret = 0;
1069
1070 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1071 return 0;
1072
1073 if (unlikely(tracing_selftest_running || tracing_disabled))
1074 return 0;
1075
1076 trace_ctx = tracing_gen_ctx();
1077 buffer = global_trace.array_buffer.buffer;
1078
1079 ring_buffer_nest_start(buffer);
1080 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1081 trace_ctx);
1082 if (!event)
1083 goto out;
1084
1085 entry = ring_buffer_event_data(event);
1086 entry->ip = ip;
1087 entry->str = str;
1088
1089 __buffer_unlock_commit(buffer, event);
1090 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1091
1092 ret = 1;
1093 out:
1094 ring_buffer_nest_end(buffer);
1095 return ret;
1096}
1097EXPORT_SYMBOL_GPL(__trace_bputs);
1098
1099#ifdef CONFIG_TRACER_SNAPSHOT
1100static void tracing_snapshot_instance_cond(struct trace_array *tr,
1101 void *cond_data)
1102{
1103 struct tracer *tracer = tr->current_trace;
1104 unsigned long flags;
1105
1106 if (in_nmi()) {
1107 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1108 internal_trace_puts("*** snapshot is being ignored ***\n");
1109 return;
1110 }
1111
1112 if (!tr->allocated_snapshot) {
1113 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
1114 internal_trace_puts("*** stopping trace here! ***\n");
1115 tracing_off();
1116 return;
1117 }
1118
1119 /* Note, snapshot can not be used when the tracer uses it */
1120 if (tracer->use_max_tr) {
1121 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
1122 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
1123 return;
1124 }
1125
1126 local_irq_save(flags);
1127 update_max_tr(tr, current, smp_processor_id(), cond_data);
1128 local_irq_restore(flags);
1129}
1130
1131void tracing_snapshot_instance(struct trace_array *tr)
1132{
1133 tracing_snapshot_instance_cond(tr, NULL);
1134}
1135
1136/**
1137 * tracing_snapshot - take a snapshot of the current buffer.
1138 *
1139 * This causes a swap between the snapshot buffer and the current live
1140 * tracing buffer. You can use this to take snapshots of the live
1141 * trace when some condition is triggered, but continue to trace.
1142 *
1143 * Note, make sure to allocate the snapshot with either
1144 * a tracing_snapshot_alloc(), or by doing it manually
1145 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
1146 *
1147 * If the snapshot buffer is not allocated, it will stop tracing.
1148 * Basically making a permanent snapshot.
1149 */
1150void tracing_snapshot(void)
1151{
1152 struct trace_array *tr = &global_trace;
1153
1154 tracing_snapshot_instance(tr);
1155}
1156EXPORT_SYMBOL_GPL(tracing_snapshot);
1157
1158/**
1159 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1160 * @tr: The tracing instance to snapshot
1161 * @cond_data: The data to be tested conditionally, and possibly saved
1162 *
1163 * This is the same as tracing_snapshot() except that the snapshot is
1164 * conditional - the snapshot will only happen if the
1165 * cond_snapshot.update() implementation receiving the cond_data
1166 * returns true, which means that the trace array's cond_snapshot
1167 * update() operation used the cond_data to determine whether the
1168 * snapshot should be taken, and if it was, presumably saved it along
1169 * with the snapshot.
1170 */
1171void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1172{
1173 tracing_snapshot_instance_cond(tr, cond_data);
1174}
1175EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1176
1177/**
1178 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1179 * @tr: The tracing instance
1180 *
1181 * When the user enables a conditional snapshot using
1182 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1183 * with the snapshot. This accessor is used to retrieve it.
1184 *
1185 * Should not be called from cond_snapshot.update(), since it takes
1186 * the tr->max_lock lock, which the code calling
1187 * cond_snapshot.update() has already done.
1188 *
1189 * Returns the cond_data associated with the trace array's snapshot.
1190 */
1191void *tracing_cond_snapshot_data(struct trace_array *tr)
1192{
1193 void *cond_data = NULL;
1194
1195 local_irq_disable();
1196 arch_spin_lock(&tr->max_lock);
1197
1198 if (tr->cond_snapshot)
1199 cond_data = tr->cond_snapshot->cond_data;
1200
1201 arch_spin_unlock(&tr->max_lock);
1202 local_irq_enable();
1203
1204 return cond_data;
1205}
1206EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1207
1208static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1209 struct array_buffer *size_buf, int cpu_id);
1210static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1211
1212int tracing_alloc_snapshot_instance(struct trace_array *tr)
1213{
1214 int ret;
1215
1216 if (!tr->allocated_snapshot) {
1217
1218 /* allocate spare buffer */
1219 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1220 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1221 if (ret < 0)
1222 return ret;
1223
1224 tr->allocated_snapshot = true;
1225 }
1226
1227 return 0;
1228}
1229
1230static void free_snapshot(struct trace_array *tr)
1231{
1232 /*
1233 * We don't free the ring buffer. instead, resize it because
1234 * The max_tr ring buffer has some state (e.g. ring->clock) and
1235 * we want preserve it.
1236 */
1237 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1238 set_buffer_entries(&tr->max_buffer, 1);
1239 tracing_reset_online_cpus(&tr->max_buffer);
1240 tr->allocated_snapshot = false;
1241}
1242
1243/**
1244 * tracing_alloc_snapshot - allocate snapshot buffer.
1245 *
1246 * This only allocates the snapshot buffer if it isn't already
1247 * allocated - it doesn't also take a snapshot.
1248 *
1249 * This is meant to be used in cases where the snapshot buffer needs
1250 * to be set up for events that can't sleep but need to be able to
1251 * trigger a snapshot.
1252 */
1253int tracing_alloc_snapshot(void)
1254{
1255 struct trace_array *tr = &global_trace;
1256 int ret;
1257
1258 ret = tracing_alloc_snapshot_instance(tr);
1259 WARN_ON(ret < 0);
1260
1261 return ret;
1262}
1263EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1264
1265/**
1266 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1267 *
1268 * This is similar to tracing_snapshot(), but it will allocate the
1269 * snapshot buffer if it isn't already allocated. Use this only
1270 * where it is safe to sleep, as the allocation may sleep.
1271 *
1272 * This causes a swap between the snapshot buffer and the current live
1273 * tracing buffer. You can use this to take snapshots of the live
1274 * trace when some condition is triggered, but continue to trace.
1275 */
1276void tracing_snapshot_alloc(void)
1277{
1278 int ret;
1279
1280 ret = tracing_alloc_snapshot();
1281 if (ret < 0)
1282 return;
1283
1284 tracing_snapshot();
1285}
1286EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1287
1288/**
1289 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1290 * @tr: The tracing instance
1291 * @cond_data: User data to associate with the snapshot
1292 * @update: Implementation of the cond_snapshot update function
1293 *
1294 * Check whether the conditional snapshot for the given instance has
1295 * already been enabled, or if the current tracer is already using a
1296 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1297 * save the cond_data and update function inside.
1298 *
1299 * Returns 0 if successful, error otherwise.
1300 */
1301int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1302 cond_update_fn_t update)
1303{
1304 struct cond_snapshot *cond_snapshot;
1305 int ret = 0;
1306
1307 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1308 if (!cond_snapshot)
1309 return -ENOMEM;
1310
1311 cond_snapshot->cond_data = cond_data;
1312 cond_snapshot->update = update;
1313
1314 mutex_lock(&trace_types_lock);
1315
1316 ret = tracing_alloc_snapshot_instance(tr);
1317 if (ret)
1318 goto fail_unlock;
1319
1320 if (tr->current_trace->use_max_tr) {
1321 ret = -EBUSY;
1322 goto fail_unlock;
1323 }
1324
1325 /*
1326 * The cond_snapshot can only change to NULL without the
1327 * trace_types_lock. We don't care if we race with it going
1328 * to NULL, but we want to make sure that it's not set to
1329 * something other than NULL when we get here, which we can
1330 * do safely with only holding the trace_types_lock and not
1331 * having to take the max_lock.
1332 */
1333 if (tr->cond_snapshot) {
1334 ret = -EBUSY;
1335 goto fail_unlock;
1336 }
1337
1338 local_irq_disable();
1339 arch_spin_lock(&tr->max_lock);
1340 tr->cond_snapshot = cond_snapshot;
1341 arch_spin_unlock(&tr->max_lock);
1342 local_irq_enable();
1343
1344 mutex_unlock(&trace_types_lock);
1345
1346 return ret;
1347
1348 fail_unlock:
1349 mutex_unlock(&trace_types_lock);
1350 kfree(cond_snapshot);
1351 return ret;
1352}
1353EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1354
1355/**
1356 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1357 * @tr: The tracing instance
1358 *
1359 * Check whether the conditional snapshot for the given instance is
1360 * enabled; if so, free the cond_snapshot associated with it,
1361 * otherwise return -EINVAL.
1362 *
1363 * Returns 0 if successful, error otherwise.
1364 */
1365int tracing_snapshot_cond_disable(struct trace_array *tr)
1366{
1367 int ret = 0;
1368
1369 local_irq_disable();
1370 arch_spin_lock(&tr->max_lock);
1371
1372 if (!tr->cond_snapshot)
1373 ret = -EINVAL;
1374 else {
1375 kfree(tr->cond_snapshot);
1376 tr->cond_snapshot = NULL;
1377 }
1378
1379 arch_spin_unlock(&tr->max_lock);
1380 local_irq_enable();
1381
1382 return ret;
1383}
1384EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1385#else
1386void tracing_snapshot(void)
1387{
1388 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1389}
1390EXPORT_SYMBOL_GPL(tracing_snapshot);
1391void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1392{
1393 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1394}
1395EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1396int tracing_alloc_snapshot(void)
1397{
1398 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1399 return -ENODEV;
1400}
1401EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1402void tracing_snapshot_alloc(void)
1403{
1404 /* Give warning */
1405 tracing_snapshot();
1406}
1407EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1408void *tracing_cond_snapshot_data(struct trace_array *tr)
1409{
1410 return NULL;
1411}
1412EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1413int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1414{
1415 return -ENODEV;
1416}
1417EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1418int tracing_snapshot_cond_disable(struct trace_array *tr)
1419{
1420 return false;
1421}
1422EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1423#define free_snapshot(tr) do { } while (0)
1424#endif /* CONFIG_TRACER_SNAPSHOT */
1425
1426void tracer_tracing_off(struct trace_array *tr)
1427{
1428 if (tr->array_buffer.buffer)
1429 ring_buffer_record_off(tr->array_buffer.buffer);
1430 /*
1431 * This flag is looked at when buffers haven't been allocated
1432 * yet, or by some tracers (like irqsoff), that just want to
1433 * know if the ring buffer has been disabled, but it can handle
1434 * races of where it gets disabled but we still do a record.
1435 * As the check is in the fast path of the tracers, it is more
1436 * important to be fast than accurate.
1437 */
1438 tr->buffer_disabled = 1;
1439 /* Make the flag seen by readers */
1440 smp_wmb();
1441}
1442
1443/**
1444 * tracing_off - turn off tracing buffers
1445 *
1446 * This function stops the tracing buffers from recording data.
1447 * It does not disable any overhead the tracers themselves may
1448 * be causing. This function simply causes all recording to
1449 * the ring buffers to fail.
1450 */
1451void tracing_off(void)
1452{
1453 tracer_tracing_off(&global_trace);
1454}
1455EXPORT_SYMBOL_GPL(tracing_off);
1456
1457void disable_trace_on_warning(void)
1458{
1459 if (__disable_trace_on_warning) {
1460 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1461 "Disabling tracing due to warning\n");
1462 tracing_off();
1463 }
1464}
1465
1466/**
1467 * tracer_tracing_is_on - show real state of ring buffer enabled
1468 * @tr : the trace array to know if ring buffer is enabled
1469 *
1470 * Shows real state of the ring buffer if it is enabled or not.
1471 */
1472bool tracer_tracing_is_on(struct trace_array *tr)
1473{
1474 if (tr->array_buffer.buffer)
1475 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1476 return !tr->buffer_disabled;
1477}
1478
1479/**
1480 * tracing_is_on - show state of ring buffers enabled
1481 */
1482int tracing_is_on(void)
1483{
1484 return tracer_tracing_is_on(&global_trace);
1485}
1486EXPORT_SYMBOL_GPL(tracing_is_on);
1487
1488static int __init set_buf_size(char *str)
1489{
1490 unsigned long buf_size;
1491
1492 if (!str)
1493 return 0;
1494 buf_size = memparse(str, &str);
1495 /*
1496 * nr_entries can not be zero and the startup
1497 * tests require some buffer space. Therefore
1498 * ensure we have at least 4096 bytes of buffer.
1499 */
1500 trace_buf_size = max(4096UL, buf_size);
1501 return 1;
1502}
1503__setup("trace_buf_size=", set_buf_size);
1504
1505static int __init set_tracing_thresh(char *str)
1506{
1507 unsigned long threshold;
1508 int ret;
1509
1510 if (!str)
1511 return 0;
1512 ret = kstrtoul(str, 0, &threshold);
1513 if (ret < 0)
1514 return 0;
1515 tracing_thresh = threshold * 1000;
1516 return 1;
1517}
1518__setup("tracing_thresh=", set_tracing_thresh);
1519
1520unsigned long nsecs_to_usecs(unsigned long nsecs)
1521{
1522 return nsecs / 1000;
1523}
1524
1525/*
1526 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1527 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1528 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1529 * of strings in the order that the evals (enum) were defined.
1530 */
1531#undef C
1532#define C(a, b) b
1533
1534/* These must match the bit positions in trace_iterator_flags */
1535static const char *trace_options[] = {
1536 TRACE_FLAGS
1537 NULL
1538};
1539
1540static struct {
1541 u64 (*func)(void);
1542 const char *name;
1543 int in_ns; /* is this clock in nanoseconds? */
1544} trace_clocks[] = {
1545 { trace_clock_local, "local", 1 },
1546 { trace_clock_global, "global", 1 },
1547 { trace_clock_counter, "counter", 0 },
1548 { trace_clock_jiffies, "uptime", 0 },
1549 { trace_clock, "perf", 1 },
1550 { ktime_get_mono_fast_ns, "mono", 1 },
1551 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1552 { ktime_get_boot_fast_ns, "boot", 1 },
1553 { ktime_get_tai_fast_ns, "tai", 1 },
1554 ARCH_TRACE_CLOCKS
1555};
1556
1557bool trace_clock_in_ns(struct trace_array *tr)
1558{
1559 if (trace_clocks[tr->clock_id].in_ns)
1560 return true;
1561
1562 return false;
1563}
1564
1565/*
1566 * trace_parser_get_init - gets the buffer for trace parser
1567 */
1568int trace_parser_get_init(struct trace_parser *parser, int size)
1569{
1570 memset(parser, 0, sizeof(*parser));
1571
1572 parser->buffer = kmalloc(size, GFP_KERNEL);
1573 if (!parser->buffer)
1574 return 1;
1575
1576 parser->size = size;
1577 return 0;
1578}
1579
1580/*
1581 * trace_parser_put - frees the buffer for trace parser
1582 */
1583void trace_parser_put(struct trace_parser *parser)
1584{
1585 kfree(parser->buffer);
1586 parser->buffer = NULL;
1587}
1588
1589/*
1590 * trace_get_user - reads the user input string separated by space
1591 * (matched by isspace(ch))
1592 *
1593 * For each string found the 'struct trace_parser' is updated,
1594 * and the function returns.
1595 *
1596 * Returns number of bytes read.
1597 *
1598 * See kernel/trace/trace.h for 'struct trace_parser' details.
1599 */
1600int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1601 size_t cnt, loff_t *ppos)
1602{
1603 char ch;
1604 size_t read = 0;
1605 ssize_t ret;
1606
1607 if (!*ppos)
1608 trace_parser_clear(parser);
1609
1610 ret = get_user(ch, ubuf++);
1611 if (ret)
1612 goto out;
1613
1614 read++;
1615 cnt--;
1616
1617 /*
1618 * The parser is not finished with the last write,
1619 * continue reading the user input without skipping spaces.
1620 */
1621 if (!parser->cont) {
1622 /* skip white space */
1623 while (cnt && isspace(ch)) {
1624 ret = get_user(ch, ubuf++);
1625 if (ret)
1626 goto out;
1627 read++;
1628 cnt--;
1629 }
1630
1631 parser->idx = 0;
1632
1633 /* only spaces were written */
1634 if (isspace(ch) || !ch) {
1635 *ppos += read;
1636 ret = read;
1637 goto out;
1638 }
1639 }
1640
1641 /* read the non-space input */
1642 while (cnt && !isspace(ch) && ch) {
1643 if (parser->idx < parser->size - 1)
1644 parser->buffer[parser->idx++] = ch;
1645 else {
1646 ret = -EINVAL;
1647 goto out;
1648 }
1649 ret = get_user(ch, ubuf++);
1650 if (ret)
1651 goto out;
1652 read++;
1653 cnt--;
1654 }
1655
1656 /* We either got finished input or we have to wait for another call. */
1657 if (isspace(ch) || !ch) {
1658 parser->buffer[parser->idx] = 0;
1659 parser->cont = false;
1660 } else if (parser->idx < parser->size - 1) {
1661 parser->cont = true;
1662 parser->buffer[parser->idx++] = ch;
1663 /* Make sure the parsed string always terminates with '\0'. */
1664 parser->buffer[parser->idx] = 0;
1665 } else {
1666 ret = -EINVAL;
1667 goto out;
1668 }
1669
1670 *ppos += read;
1671 ret = read;
1672
1673out:
1674 return ret;
1675}
1676
1677/* TODO add a seq_buf_to_buffer() */
1678static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1679{
1680 int len;
1681
1682 if (trace_seq_used(s) <= s->seq.readpos)
1683 return -EBUSY;
1684
1685 len = trace_seq_used(s) - s->seq.readpos;
1686 if (cnt > len)
1687 cnt = len;
1688 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1689
1690 s->seq.readpos += cnt;
1691 return cnt;
1692}
1693
1694unsigned long __read_mostly tracing_thresh;
1695
1696#ifdef CONFIG_TRACER_MAX_TRACE
1697static const struct file_operations tracing_max_lat_fops;
1698
1699#ifdef LATENCY_FS_NOTIFY
1700
1701static struct workqueue_struct *fsnotify_wq;
1702
1703static void latency_fsnotify_workfn(struct work_struct *work)
1704{
1705 struct trace_array *tr = container_of(work, struct trace_array,
1706 fsnotify_work);
1707 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1708}
1709
1710static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1711{
1712 struct trace_array *tr = container_of(iwork, struct trace_array,
1713 fsnotify_irqwork);
1714 queue_work(fsnotify_wq, &tr->fsnotify_work);
1715}
1716
1717static void trace_create_maxlat_file(struct trace_array *tr,
1718 struct dentry *d_tracer)
1719{
1720 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1721 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1722 tr->d_max_latency = trace_create_file("tracing_max_latency",
1723 TRACE_MODE_WRITE,
1724 d_tracer, &tr->max_latency,
1725 &tracing_max_lat_fops);
1726}
1727
1728__init static int latency_fsnotify_init(void)
1729{
1730 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1731 WQ_UNBOUND | WQ_HIGHPRI, 0);
1732 if (!fsnotify_wq) {
1733 pr_err("Unable to allocate tr_max_lat_wq\n");
1734 return -ENOMEM;
1735 }
1736 return 0;
1737}
1738
1739late_initcall_sync(latency_fsnotify_init);
1740
1741void latency_fsnotify(struct trace_array *tr)
1742{
1743 if (!fsnotify_wq)
1744 return;
1745 /*
1746 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1747 * possible that we are called from __schedule() or do_idle(), which
1748 * could cause a deadlock.
1749 */
1750 irq_work_queue(&tr->fsnotify_irqwork);
1751}
1752
1753#else /* !LATENCY_FS_NOTIFY */
1754
1755#define trace_create_maxlat_file(tr, d_tracer) \
1756 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1757 d_tracer, &tr->max_latency, &tracing_max_lat_fops)
1758
1759#endif
1760
1761/*
1762 * Copy the new maximum trace into the separate maximum-trace
1763 * structure. (this way the maximum trace is permanently saved,
1764 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1765 */
1766static void
1767__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1768{
1769 struct array_buffer *trace_buf = &tr->array_buffer;
1770 struct array_buffer *max_buf = &tr->max_buffer;
1771 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1772 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1773
1774 max_buf->cpu = cpu;
1775 max_buf->time_start = data->preempt_timestamp;
1776
1777 max_data->saved_latency = tr->max_latency;
1778 max_data->critical_start = data->critical_start;
1779 max_data->critical_end = data->critical_end;
1780
1781 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1782 max_data->pid = tsk->pid;
1783 /*
1784 * If tsk == current, then use current_uid(), as that does not use
1785 * RCU. The irq tracer can be called out of RCU scope.
1786 */
1787 if (tsk == current)
1788 max_data->uid = current_uid();
1789 else
1790 max_data->uid = task_uid(tsk);
1791
1792 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1793 max_data->policy = tsk->policy;
1794 max_data->rt_priority = tsk->rt_priority;
1795
1796 /* record this tasks comm */
1797 tracing_record_cmdline(tsk);
1798 latency_fsnotify(tr);
1799}
1800
1801/**
1802 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1803 * @tr: tracer
1804 * @tsk: the task with the latency
1805 * @cpu: The cpu that initiated the trace.
1806 * @cond_data: User data associated with a conditional snapshot
1807 *
1808 * Flip the buffers between the @tr and the max_tr and record information
1809 * about which task was the cause of this latency.
1810 */
1811void
1812update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1813 void *cond_data)
1814{
1815 if (tr->stop_count)
1816 return;
1817
1818 WARN_ON_ONCE(!irqs_disabled());
1819
1820 if (!tr->allocated_snapshot) {
1821 /* Only the nop tracer should hit this when disabling */
1822 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1823 return;
1824 }
1825
1826 arch_spin_lock(&tr->max_lock);
1827
1828 /* Inherit the recordable setting from array_buffer */
1829 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1830 ring_buffer_record_on(tr->max_buffer.buffer);
1831 else
1832 ring_buffer_record_off(tr->max_buffer.buffer);
1833
1834#ifdef CONFIG_TRACER_SNAPSHOT
1835 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1836 arch_spin_unlock(&tr->max_lock);
1837 return;
1838 }
1839#endif
1840 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1841
1842 __update_max_tr(tr, tsk, cpu);
1843
1844 arch_spin_unlock(&tr->max_lock);
1845}
1846
1847/**
1848 * update_max_tr_single - only copy one trace over, and reset the rest
1849 * @tr: tracer
1850 * @tsk: task with the latency
1851 * @cpu: the cpu of the buffer to copy.
1852 *
1853 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1854 */
1855void
1856update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1857{
1858 int ret;
1859
1860 if (tr->stop_count)
1861 return;
1862
1863 WARN_ON_ONCE(!irqs_disabled());
1864 if (!tr->allocated_snapshot) {
1865 /* Only the nop tracer should hit this when disabling */
1866 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1867 return;
1868 }
1869
1870 arch_spin_lock(&tr->max_lock);
1871
1872 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1873
1874 if (ret == -EBUSY) {
1875 /*
1876 * We failed to swap the buffer due to a commit taking
1877 * place on this CPU. We fail to record, but we reset
1878 * the max trace buffer (no one writes directly to it)
1879 * and flag that it failed.
1880 */
1881 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1882 "Failed to swap buffers due to commit in progress\n");
1883 }
1884
1885 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1886
1887 __update_max_tr(tr, tsk, cpu);
1888 arch_spin_unlock(&tr->max_lock);
1889}
1890
1891#endif /* CONFIG_TRACER_MAX_TRACE */
1892
1893static int wait_on_pipe(struct trace_iterator *iter, int full)
1894{
1895 /* Iterators are static, they should be filled or empty */
1896 if (trace_buffer_iter(iter, iter->cpu_file))
1897 return 0;
1898
1899 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1900 full);
1901}
1902
1903#ifdef CONFIG_FTRACE_STARTUP_TEST
1904static bool selftests_can_run;
1905
1906struct trace_selftests {
1907 struct list_head list;
1908 struct tracer *type;
1909};
1910
1911static LIST_HEAD(postponed_selftests);
1912
1913static int save_selftest(struct tracer *type)
1914{
1915 struct trace_selftests *selftest;
1916
1917 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1918 if (!selftest)
1919 return -ENOMEM;
1920
1921 selftest->type = type;
1922 list_add(&selftest->list, &postponed_selftests);
1923 return 0;
1924}
1925
1926static int run_tracer_selftest(struct tracer *type)
1927{
1928 struct trace_array *tr = &global_trace;
1929 struct tracer *saved_tracer = tr->current_trace;
1930 int ret;
1931
1932 if (!type->selftest || tracing_selftest_disabled)
1933 return 0;
1934
1935 /*
1936 * If a tracer registers early in boot up (before scheduling is
1937 * initialized and such), then do not run its selftests yet.
1938 * Instead, run it a little later in the boot process.
1939 */
1940 if (!selftests_can_run)
1941 return save_selftest(type);
1942
1943 if (!tracing_is_on()) {
1944 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
1945 type->name);
1946 return 0;
1947 }
1948
1949 /*
1950 * Run a selftest on this tracer.
1951 * Here we reset the trace buffer, and set the current
1952 * tracer to be this tracer. The tracer can then run some
1953 * internal tracing to verify that everything is in order.
1954 * If we fail, we do not register this tracer.
1955 */
1956 tracing_reset_online_cpus(&tr->array_buffer);
1957
1958 tr->current_trace = type;
1959
1960#ifdef CONFIG_TRACER_MAX_TRACE
1961 if (type->use_max_tr) {
1962 /* If we expanded the buffers, make sure the max is expanded too */
1963 if (ring_buffer_expanded)
1964 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1965 RING_BUFFER_ALL_CPUS);
1966 tr->allocated_snapshot = true;
1967 }
1968#endif
1969
1970 /* the test is responsible for initializing and enabling */
1971 pr_info("Testing tracer %s: ", type->name);
1972 ret = type->selftest(type, tr);
1973 /* the test is responsible for resetting too */
1974 tr->current_trace = saved_tracer;
1975 if (ret) {
1976 printk(KERN_CONT "FAILED!\n");
1977 /* Add the warning after printing 'FAILED' */
1978 WARN_ON(1);
1979 return -1;
1980 }
1981 /* Only reset on passing, to avoid touching corrupted buffers */
1982 tracing_reset_online_cpus(&tr->array_buffer);
1983
1984#ifdef CONFIG_TRACER_MAX_TRACE
1985 if (type->use_max_tr) {
1986 tr->allocated_snapshot = false;
1987
1988 /* Shrink the max buffer again */
1989 if (ring_buffer_expanded)
1990 ring_buffer_resize(tr->max_buffer.buffer, 1,
1991 RING_BUFFER_ALL_CPUS);
1992 }
1993#endif
1994
1995 printk(KERN_CONT "PASSED\n");
1996 return 0;
1997}
1998
1999static __init int init_trace_selftests(void)
2000{
2001 struct trace_selftests *p, *n;
2002 struct tracer *t, **last;
2003 int ret;
2004
2005 selftests_can_run = true;
2006
2007 mutex_lock(&trace_types_lock);
2008
2009 if (list_empty(&postponed_selftests))
2010 goto out;
2011
2012 pr_info("Running postponed tracer tests:\n");
2013
2014 tracing_selftest_running = true;
2015 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2016 /* This loop can take minutes when sanitizers are enabled, so
2017 * lets make sure we allow RCU processing.
2018 */
2019 cond_resched();
2020 ret = run_tracer_selftest(p->type);
2021 /* If the test fails, then warn and remove from available_tracers */
2022 if (ret < 0) {
2023 WARN(1, "tracer: %s failed selftest, disabling\n",
2024 p->type->name);
2025 last = &trace_types;
2026 for (t = trace_types; t; t = t->next) {
2027 if (t == p->type) {
2028 *last = t->next;
2029 break;
2030 }
2031 last = &t->next;
2032 }
2033 }
2034 list_del(&p->list);
2035 kfree(p);
2036 }
2037 tracing_selftest_running = false;
2038
2039 out:
2040 mutex_unlock(&trace_types_lock);
2041
2042 return 0;
2043}
2044core_initcall(init_trace_selftests);
2045#else
2046static inline int run_tracer_selftest(struct tracer *type)
2047{
2048 return 0;
2049}
2050#endif /* CONFIG_FTRACE_STARTUP_TEST */
2051
2052static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2053
2054static void __init apply_trace_boot_options(void);
2055
2056/**
2057 * register_tracer - register a tracer with the ftrace system.
2058 * @type: the plugin for the tracer
2059 *
2060 * Register a new plugin tracer.
2061 */
2062int __init register_tracer(struct tracer *type)
2063{
2064 struct tracer *t;
2065 int ret = 0;
2066
2067 if (!type->name) {
2068 pr_info("Tracer must have a name\n");
2069 return -1;
2070 }
2071
2072 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2073 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2074 return -1;
2075 }
2076
2077 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2078 pr_warn("Can not register tracer %s due to lockdown\n",
2079 type->name);
2080 return -EPERM;
2081 }
2082
2083 mutex_lock(&trace_types_lock);
2084
2085 tracing_selftest_running = true;
2086
2087 for (t = trace_types; t; t = t->next) {
2088 if (strcmp(type->name, t->name) == 0) {
2089 /* already found */
2090 pr_info("Tracer %s already registered\n",
2091 type->name);
2092 ret = -1;
2093 goto out;
2094 }
2095 }
2096
2097 if (!type->set_flag)
2098 type->set_flag = &dummy_set_flag;
2099 if (!type->flags) {
2100 /*allocate a dummy tracer_flags*/
2101 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2102 if (!type->flags) {
2103 ret = -ENOMEM;
2104 goto out;
2105 }
2106 type->flags->val = 0;
2107 type->flags->opts = dummy_tracer_opt;
2108 } else
2109 if (!type->flags->opts)
2110 type->flags->opts = dummy_tracer_opt;
2111
2112 /* store the tracer for __set_tracer_option */
2113 type->flags->trace = type;
2114
2115 ret = run_tracer_selftest(type);
2116 if (ret < 0)
2117 goto out;
2118
2119 type->next = trace_types;
2120 trace_types = type;
2121 add_tracer_options(&global_trace, type);
2122
2123 out:
2124 tracing_selftest_running = false;
2125 mutex_unlock(&trace_types_lock);
2126
2127 if (ret || !default_bootup_tracer)
2128 goto out_unlock;
2129
2130 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2131 goto out_unlock;
2132
2133 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2134 /* Do we want this tracer to start on bootup? */
2135 tracing_set_tracer(&global_trace, type->name);
2136 default_bootup_tracer = NULL;
2137
2138 apply_trace_boot_options();
2139
2140 /* disable other selftests, since this will break it. */
2141 disable_tracing_selftest("running a tracer");
2142
2143 out_unlock:
2144 return ret;
2145}
2146
2147static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2148{
2149 struct trace_buffer *buffer = buf->buffer;
2150
2151 if (!buffer)
2152 return;
2153
2154 ring_buffer_record_disable(buffer);
2155
2156 /* Make sure all commits have finished */
2157 synchronize_rcu();
2158 ring_buffer_reset_cpu(buffer, cpu);
2159
2160 ring_buffer_record_enable(buffer);
2161}
2162
2163void tracing_reset_online_cpus(struct array_buffer *buf)
2164{
2165 struct trace_buffer *buffer = buf->buffer;
2166
2167 if (!buffer)
2168 return;
2169
2170 ring_buffer_record_disable(buffer);
2171
2172 /* Make sure all commits have finished */
2173 synchronize_rcu();
2174
2175 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2176
2177 ring_buffer_reset_online_cpus(buffer);
2178
2179 ring_buffer_record_enable(buffer);
2180}
2181
2182/* Must have trace_types_lock held */
2183void tracing_reset_all_online_cpus_unlocked(void)
2184{
2185 struct trace_array *tr;
2186
2187 lockdep_assert_held(&trace_types_lock);
2188
2189 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2190 if (!tr->clear_trace)
2191 continue;
2192 tr->clear_trace = false;
2193 tracing_reset_online_cpus(&tr->array_buffer);
2194#ifdef CONFIG_TRACER_MAX_TRACE
2195 tracing_reset_online_cpus(&tr->max_buffer);
2196#endif
2197 }
2198}
2199
2200void tracing_reset_all_online_cpus(void)
2201{
2202 mutex_lock(&trace_types_lock);
2203 tracing_reset_all_online_cpus_unlocked();
2204 mutex_unlock(&trace_types_lock);
2205}
2206
2207/*
2208 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2209 * is the tgid last observed corresponding to pid=i.
2210 */
2211static int *tgid_map;
2212
2213/* The maximum valid index into tgid_map. */
2214static size_t tgid_map_max;
2215
2216#define SAVED_CMDLINES_DEFAULT 128
2217#define NO_CMDLINE_MAP UINT_MAX
2218/*
2219 * Preemption must be disabled before acquiring trace_cmdline_lock.
2220 * The various trace_arrays' max_lock must be acquired in a context
2221 * where interrupt is disabled.
2222 */
2223static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2224struct saved_cmdlines_buffer {
2225 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2226 unsigned *map_cmdline_to_pid;
2227 unsigned cmdline_num;
2228 int cmdline_idx;
2229 char *saved_cmdlines;
2230};
2231static struct saved_cmdlines_buffer *savedcmd;
2232
2233static inline char *get_saved_cmdlines(int idx)
2234{
2235 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2236}
2237
2238static inline void set_cmdline(int idx, const char *cmdline)
2239{
2240 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2241}
2242
2243static int allocate_cmdlines_buffer(unsigned int val,
2244 struct saved_cmdlines_buffer *s)
2245{
2246 s->map_cmdline_to_pid = kmalloc_array(val,
2247 sizeof(*s->map_cmdline_to_pid),
2248 GFP_KERNEL);
2249 if (!s->map_cmdline_to_pid)
2250 return -ENOMEM;
2251
2252 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2253 if (!s->saved_cmdlines) {
2254 kfree(s->map_cmdline_to_pid);
2255 return -ENOMEM;
2256 }
2257
2258 s->cmdline_idx = 0;
2259 s->cmdline_num = val;
2260 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2261 sizeof(s->map_pid_to_cmdline));
2262 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2263 val * sizeof(*s->map_cmdline_to_pid));
2264
2265 return 0;
2266}
2267
2268static int trace_create_savedcmd(void)
2269{
2270 int ret;
2271
2272 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2273 if (!savedcmd)
2274 return -ENOMEM;
2275
2276 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2277 if (ret < 0) {
2278 kfree(savedcmd);
2279 savedcmd = NULL;
2280 return -ENOMEM;
2281 }
2282
2283 return 0;
2284}
2285
2286int is_tracing_stopped(void)
2287{
2288 return global_trace.stop_count;
2289}
2290
2291/**
2292 * tracing_start - quick start of the tracer
2293 *
2294 * If tracing is enabled but was stopped by tracing_stop,
2295 * this will start the tracer back up.
2296 */
2297void tracing_start(void)
2298{
2299 struct trace_buffer *buffer;
2300 unsigned long flags;
2301
2302 if (tracing_disabled)
2303 return;
2304
2305 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2306 if (--global_trace.stop_count) {
2307 if (global_trace.stop_count < 0) {
2308 /* Someone screwed up their debugging */
2309 WARN_ON_ONCE(1);
2310 global_trace.stop_count = 0;
2311 }
2312 goto out;
2313 }
2314
2315 /* Prevent the buffers from switching */
2316 arch_spin_lock(&global_trace.max_lock);
2317
2318 buffer = global_trace.array_buffer.buffer;
2319 if (buffer)
2320 ring_buffer_record_enable(buffer);
2321
2322#ifdef CONFIG_TRACER_MAX_TRACE
2323 buffer = global_trace.max_buffer.buffer;
2324 if (buffer)
2325 ring_buffer_record_enable(buffer);
2326#endif
2327
2328 arch_spin_unlock(&global_trace.max_lock);
2329
2330 out:
2331 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2332}
2333
2334static void tracing_start_tr(struct trace_array *tr)
2335{
2336 struct trace_buffer *buffer;
2337 unsigned long flags;
2338
2339 if (tracing_disabled)
2340 return;
2341
2342 /* If global, we need to also start the max tracer */
2343 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2344 return tracing_start();
2345
2346 raw_spin_lock_irqsave(&tr->start_lock, flags);
2347
2348 if (--tr->stop_count) {
2349 if (tr->stop_count < 0) {
2350 /* Someone screwed up their debugging */
2351 WARN_ON_ONCE(1);
2352 tr->stop_count = 0;
2353 }
2354 goto out;
2355 }
2356
2357 buffer = tr->array_buffer.buffer;
2358 if (buffer)
2359 ring_buffer_record_enable(buffer);
2360
2361 out:
2362 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2363}
2364
2365/**
2366 * tracing_stop - quick stop of the tracer
2367 *
2368 * Light weight way to stop tracing. Use in conjunction with
2369 * tracing_start.
2370 */
2371void tracing_stop(void)
2372{
2373 struct trace_buffer *buffer;
2374 unsigned long flags;
2375
2376 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2377 if (global_trace.stop_count++)
2378 goto out;
2379
2380 /* Prevent the buffers from switching */
2381 arch_spin_lock(&global_trace.max_lock);
2382
2383 buffer = global_trace.array_buffer.buffer;
2384 if (buffer)
2385 ring_buffer_record_disable(buffer);
2386
2387#ifdef CONFIG_TRACER_MAX_TRACE
2388 buffer = global_trace.max_buffer.buffer;
2389 if (buffer)
2390 ring_buffer_record_disable(buffer);
2391#endif
2392
2393 arch_spin_unlock(&global_trace.max_lock);
2394
2395 out:
2396 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2397}
2398
2399static void tracing_stop_tr(struct trace_array *tr)
2400{
2401 struct trace_buffer *buffer;
2402 unsigned long flags;
2403
2404 /* If global, we need to also stop the max tracer */
2405 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2406 return tracing_stop();
2407
2408 raw_spin_lock_irqsave(&tr->start_lock, flags);
2409 if (tr->stop_count++)
2410 goto out;
2411
2412 buffer = tr->array_buffer.buffer;
2413 if (buffer)
2414 ring_buffer_record_disable(buffer);
2415
2416 out:
2417 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2418}
2419
2420static int trace_save_cmdline(struct task_struct *tsk)
2421{
2422 unsigned tpid, idx;
2423
2424 /* treat recording of idle task as a success */
2425 if (!tsk->pid)
2426 return 1;
2427
2428 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2429
2430 /*
2431 * It's not the end of the world if we don't get
2432 * the lock, but we also don't want to spin
2433 * nor do we want to disable interrupts,
2434 * so if we miss here, then better luck next time.
2435 *
2436 * This is called within the scheduler and wake up, so interrupts
2437 * had better been disabled and run queue lock been held.
2438 */
2439 lockdep_assert_preemption_disabled();
2440 if (!arch_spin_trylock(&trace_cmdline_lock))
2441 return 0;
2442
2443 idx = savedcmd->map_pid_to_cmdline[tpid];
2444 if (idx == NO_CMDLINE_MAP) {
2445 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2446
2447 savedcmd->map_pid_to_cmdline[tpid] = idx;
2448 savedcmd->cmdline_idx = idx;
2449 }
2450
2451 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2452 set_cmdline(idx, tsk->comm);
2453
2454 arch_spin_unlock(&trace_cmdline_lock);
2455
2456 return 1;
2457}
2458
2459static void __trace_find_cmdline(int pid, char comm[])
2460{
2461 unsigned map;
2462 int tpid;
2463
2464 if (!pid) {
2465 strcpy(comm, "<idle>");
2466 return;
2467 }
2468
2469 if (WARN_ON_ONCE(pid < 0)) {
2470 strcpy(comm, "<XXX>");
2471 return;
2472 }
2473
2474 tpid = pid & (PID_MAX_DEFAULT - 1);
2475 map = savedcmd->map_pid_to_cmdline[tpid];
2476 if (map != NO_CMDLINE_MAP) {
2477 tpid = savedcmd->map_cmdline_to_pid[map];
2478 if (tpid == pid) {
2479 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2480 return;
2481 }
2482 }
2483 strcpy(comm, "<...>");
2484}
2485
2486void trace_find_cmdline(int pid, char comm[])
2487{
2488 preempt_disable();
2489 arch_spin_lock(&trace_cmdline_lock);
2490
2491 __trace_find_cmdline(pid, comm);
2492
2493 arch_spin_unlock(&trace_cmdline_lock);
2494 preempt_enable();
2495}
2496
2497static int *trace_find_tgid_ptr(int pid)
2498{
2499 /*
2500 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2501 * if we observe a non-NULL tgid_map then we also observe the correct
2502 * tgid_map_max.
2503 */
2504 int *map = smp_load_acquire(&tgid_map);
2505
2506 if (unlikely(!map || pid > tgid_map_max))
2507 return NULL;
2508
2509 return &map[pid];
2510}
2511
2512int trace_find_tgid(int pid)
2513{
2514 int *ptr = trace_find_tgid_ptr(pid);
2515
2516 return ptr ? *ptr : 0;
2517}
2518
2519static int trace_save_tgid(struct task_struct *tsk)
2520{
2521 int *ptr;
2522
2523 /* treat recording of idle task as a success */
2524 if (!tsk->pid)
2525 return 1;
2526
2527 ptr = trace_find_tgid_ptr(tsk->pid);
2528 if (!ptr)
2529 return 0;
2530
2531 *ptr = tsk->tgid;
2532 return 1;
2533}
2534
2535static bool tracing_record_taskinfo_skip(int flags)
2536{
2537 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2538 return true;
2539 if (!__this_cpu_read(trace_taskinfo_save))
2540 return true;
2541 return false;
2542}
2543
2544/**
2545 * tracing_record_taskinfo - record the task info of a task
2546 *
2547 * @task: task to record
2548 * @flags: TRACE_RECORD_CMDLINE for recording comm
2549 * TRACE_RECORD_TGID for recording tgid
2550 */
2551void tracing_record_taskinfo(struct task_struct *task, int flags)
2552{
2553 bool done;
2554
2555 if (tracing_record_taskinfo_skip(flags))
2556 return;
2557
2558 /*
2559 * Record as much task information as possible. If some fail, continue
2560 * to try to record the others.
2561 */
2562 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2563 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2564
2565 /* If recording any information failed, retry again soon. */
2566 if (!done)
2567 return;
2568
2569 __this_cpu_write(trace_taskinfo_save, false);
2570}
2571
2572/**
2573 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2574 *
2575 * @prev: previous task during sched_switch
2576 * @next: next task during sched_switch
2577 * @flags: TRACE_RECORD_CMDLINE for recording comm
2578 * TRACE_RECORD_TGID for recording tgid
2579 */
2580void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2581 struct task_struct *next, int flags)
2582{
2583 bool done;
2584
2585 if (tracing_record_taskinfo_skip(flags))
2586 return;
2587
2588 /*
2589 * Record as much task information as possible. If some fail, continue
2590 * to try to record the others.
2591 */
2592 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2593 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2594 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2595 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2596
2597 /* If recording any information failed, retry again soon. */
2598 if (!done)
2599 return;
2600
2601 __this_cpu_write(trace_taskinfo_save, false);
2602}
2603
2604/* Helpers to record a specific task information */
2605void tracing_record_cmdline(struct task_struct *task)
2606{
2607 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2608}
2609
2610void tracing_record_tgid(struct task_struct *task)
2611{
2612 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2613}
2614
2615/*
2616 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2617 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2618 * simplifies those functions and keeps them in sync.
2619 */
2620enum print_line_t trace_handle_return(struct trace_seq *s)
2621{
2622 return trace_seq_has_overflowed(s) ?
2623 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2624}
2625EXPORT_SYMBOL_GPL(trace_handle_return);
2626
2627static unsigned short migration_disable_value(void)
2628{
2629#if defined(CONFIG_SMP)
2630 return current->migration_disabled;
2631#else
2632 return 0;
2633#endif
2634}
2635
2636unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2637{
2638 unsigned int trace_flags = irqs_status;
2639 unsigned int pc;
2640
2641 pc = preempt_count();
2642
2643 if (pc & NMI_MASK)
2644 trace_flags |= TRACE_FLAG_NMI;
2645 if (pc & HARDIRQ_MASK)
2646 trace_flags |= TRACE_FLAG_HARDIRQ;
2647 if (in_serving_softirq())
2648 trace_flags |= TRACE_FLAG_SOFTIRQ;
2649 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2650 trace_flags |= TRACE_FLAG_BH_OFF;
2651
2652 if (tif_need_resched())
2653 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2654 if (test_preempt_need_resched())
2655 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2656 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2657 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2658}
2659
2660struct ring_buffer_event *
2661trace_buffer_lock_reserve(struct trace_buffer *buffer,
2662 int type,
2663 unsigned long len,
2664 unsigned int trace_ctx)
2665{
2666 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2667}
2668
2669DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2670DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2671static int trace_buffered_event_ref;
2672
2673/**
2674 * trace_buffered_event_enable - enable buffering events
2675 *
2676 * When events are being filtered, it is quicker to use a temporary
2677 * buffer to write the event data into if there's a likely chance
2678 * that it will not be committed. The discard of the ring buffer
2679 * is not as fast as committing, and is much slower than copying
2680 * a commit.
2681 *
2682 * When an event is to be filtered, allocate per cpu buffers to
2683 * write the event data into, and if the event is filtered and discarded
2684 * it is simply dropped, otherwise, the entire data is to be committed
2685 * in one shot.
2686 */
2687void trace_buffered_event_enable(void)
2688{
2689 struct ring_buffer_event *event;
2690 struct page *page;
2691 int cpu;
2692
2693 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2694
2695 if (trace_buffered_event_ref++)
2696 return;
2697
2698 for_each_tracing_cpu(cpu) {
2699 page = alloc_pages_node(cpu_to_node(cpu),
2700 GFP_KERNEL | __GFP_NORETRY, 0);
2701 if (!page)
2702 goto failed;
2703
2704 event = page_address(page);
2705 memset(event, 0, sizeof(*event));
2706
2707 per_cpu(trace_buffered_event, cpu) = event;
2708
2709 preempt_disable();
2710 if (cpu == smp_processor_id() &&
2711 __this_cpu_read(trace_buffered_event) !=
2712 per_cpu(trace_buffered_event, cpu))
2713 WARN_ON_ONCE(1);
2714 preempt_enable();
2715 }
2716
2717 return;
2718 failed:
2719 trace_buffered_event_disable();
2720}
2721
2722static void enable_trace_buffered_event(void *data)
2723{
2724 /* Probably not needed, but do it anyway */
2725 smp_rmb();
2726 this_cpu_dec(trace_buffered_event_cnt);
2727}
2728
2729static void disable_trace_buffered_event(void *data)
2730{
2731 this_cpu_inc(trace_buffered_event_cnt);
2732}
2733
2734/**
2735 * trace_buffered_event_disable - disable buffering events
2736 *
2737 * When a filter is removed, it is faster to not use the buffered
2738 * events, and to commit directly into the ring buffer. Free up
2739 * the temp buffers when there are no more users. This requires
2740 * special synchronization with current events.
2741 */
2742void trace_buffered_event_disable(void)
2743{
2744 int cpu;
2745
2746 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2747
2748 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2749 return;
2750
2751 if (--trace_buffered_event_ref)
2752 return;
2753
2754 preempt_disable();
2755 /* For each CPU, set the buffer as used. */
2756 smp_call_function_many(tracing_buffer_mask,
2757 disable_trace_buffered_event, NULL, 1);
2758 preempt_enable();
2759
2760 /* Wait for all current users to finish */
2761 synchronize_rcu();
2762
2763 for_each_tracing_cpu(cpu) {
2764 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2765 per_cpu(trace_buffered_event, cpu) = NULL;
2766 }
2767 /*
2768 * Make sure trace_buffered_event is NULL before clearing
2769 * trace_buffered_event_cnt.
2770 */
2771 smp_wmb();
2772
2773 preempt_disable();
2774 /* Do the work on each cpu */
2775 smp_call_function_many(tracing_buffer_mask,
2776 enable_trace_buffered_event, NULL, 1);
2777 preempt_enable();
2778}
2779
2780static struct trace_buffer *temp_buffer;
2781
2782struct ring_buffer_event *
2783trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2784 struct trace_event_file *trace_file,
2785 int type, unsigned long len,
2786 unsigned int trace_ctx)
2787{
2788 struct ring_buffer_event *entry;
2789 struct trace_array *tr = trace_file->tr;
2790 int val;
2791
2792 *current_rb = tr->array_buffer.buffer;
2793
2794 if (!tr->no_filter_buffering_ref &&
2795 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2796 preempt_disable_notrace();
2797 /*
2798 * Filtering is on, so try to use the per cpu buffer first.
2799 * This buffer will simulate a ring_buffer_event,
2800 * where the type_len is zero and the array[0] will
2801 * hold the full length.
2802 * (see include/linux/ring-buffer.h for details on
2803 * how the ring_buffer_event is structured).
2804 *
2805 * Using a temp buffer during filtering and copying it
2806 * on a matched filter is quicker than writing directly
2807 * into the ring buffer and then discarding it when
2808 * it doesn't match. That is because the discard
2809 * requires several atomic operations to get right.
2810 * Copying on match and doing nothing on a failed match
2811 * is still quicker than no copy on match, but having
2812 * to discard out of the ring buffer on a failed match.
2813 */
2814 if ((entry = __this_cpu_read(trace_buffered_event))) {
2815 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2816
2817 val = this_cpu_inc_return(trace_buffered_event_cnt);
2818
2819 /*
2820 * Preemption is disabled, but interrupts and NMIs
2821 * can still come in now. If that happens after
2822 * the above increment, then it will have to go
2823 * back to the old method of allocating the event
2824 * on the ring buffer, and if the filter fails, it
2825 * will have to call ring_buffer_discard_commit()
2826 * to remove it.
2827 *
2828 * Need to also check the unlikely case that the
2829 * length is bigger than the temp buffer size.
2830 * If that happens, then the reserve is pretty much
2831 * guaranteed to fail, as the ring buffer currently
2832 * only allows events less than a page. But that may
2833 * change in the future, so let the ring buffer reserve
2834 * handle the failure in that case.
2835 */
2836 if (val == 1 && likely(len <= max_len)) {
2837 trace_event_setup(entry, type, trace_ctx);
2838 entry->array[0] = len;
2839 /* Return with preemption disabled */
2840 return entry;
2841 }
2842 this_cpu_dec(trace_buffered_event_cnt);
2843 }
2844 /* __trace_buffer_lock_reserve() disables preemption */
2845 preempt_enable_notrace();
2846 }
2847
2848 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2849 trace_ctx);
2850 /*
2851 * If tracing is off, but we have triggers enabled
2852 * we still need to look at the event data. Use the temp_buffer
2853 * to store the trace event for the trigger to use. It's recursive
2854 * safe and will not be recorded anywhere.
2855 */
2856 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2857 *current_rb = temp_buffer;
2858 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2859 trace_ctx);
2860 }
2861 return entry;
2862}
2863EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2864
2865static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2866static DEFINE_MUTEX(tracepoint_printk_mutex);
2867
2868static void output_printk(struct trace_event_buffer *fbuffer)
2869{
2870 struct trace_event_call *event_call;
2871 struct trace_event_file *file;
2872 struct trace_event *event;
2873 unsigned long flags;
2874 struct trace_iterator *iter = tracepoint_print_iter;
2875
2876 /* We should never get here if iter is NULL */
2877 if (WARN_ON_ONCE(!iter))
2878 return;
2879
2880 event_call = fbuffer->trace_file->event_call;
2881 if (!event_call || !event_call->event.funcs ||
2882 !event_call->event.funcs->trace)
2883 return;
2884
2885 file = fbuffer->trace_file;
2886 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2887 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2888 !filter_match_preds(file->filter, fbuffer->entry)))
2889 return;
2890
2891 event = &fbuffer->trace_file->event_call->event;
2892
2893 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2894 trace_seq_init(&iter->seq);
2895 iter->ent = fbuffer->entry;
2896 event_call->event.funcs->trace(iter, 0, event);
2897 trace_seq_putc(&iter->seq, 0);
2898 printk("%s", iter->seq.buffer);
2899
2900 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2901}
2902
2903int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2904 void *buffer, size_t *lenp,
2905 loff_t *ppos)
2906{
2907 int save_tracepoint_printk;
2908 int ret;
2909
2910 mutex_lock(&tracepoint_printk_mutex);
2911 save_tracepoint_printk = tracepoint_printk;
2912
2913 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2914
2915 /*
2916 * This will force exiting early, as tracepoint_printk
2917 * is always zero when tracepoint_printk_iter is not allocated
2918 */
2919 if (!tracepoint_print_iter)
2920 tracepoint_printk = 0;
2921
2922 if (save_tracepoint_printk == tracepoint_printk)
2923 goto out;
2924
2925 if (tracepoint_printk)
2926 static_key_enable(&tracepoint_printk_key.key);
2927 else
2928 static_key_disable(&tracepoint_printk_key.key);
2929
2930 out:
2931 mutex_unlock(&tracepoint_printk_mutex);
2932
2933 return ret;
2934}
2935
2936void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2937{
2938 enum event_trigger_type tt = ETT_NONE;
2939 struct trace_event_file *file = fbuffer->trace_file;
2940
2941 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2942 fbuffer->entry, &tt))
2943 goto discard;
2944
2945 if (static_key_false(&tracepoint_printk_key.key))
2946 output_printk(fbuffer);
2947
2948 if (static_branch_unlikely(&trace_event_exports_enabled))
2949 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2950
2951 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2952 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
2953
2954discard:
2955 if (tt)
2956 event_triggers_post_call(file, tt);
2957
2958}
2959EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2960
2961/*
2962 * Skip 3:
2963 *
2964 * trace_buffer_unlock_commit_regs()
2965 * trace_event_buffer_commit()
2966 * trace_event_raw_event_xxx()
2967 */
2968# define STACK_SKIP 3
2969
2970void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2971 struct trace_buffer *buffer,
2972 struct ring_buffer_event *event,
2973 unsigned int trace_ctx,
2974 struct pt_regs *regs)
2975{
2976 __buffer_unlock_commit(buffer, event);
2977
2978 /*
2979 * If regs is not set, then skip the necessary functions.
2980 * Note, we can still get here via blktrace, wakeup tracer
2981 * and mmiotrace, but that's ok if they lose a function or
2982 * two. They are not that meaningful.
2983 */
2984 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
2985 ftrace_trace_userstack(tr, buffer, trace_ctx);
2986}
2987
2988/*
2989 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2990 */
2991void
2992trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2993 struct ring_buffer_event *event)
2994{
2995 __buffer_unlock_commit(buffer, event);
2996}
2997
2998void
2999trace_function(struct trace_array *tr, unsigned long ip, unsigned long
3000 parent_ip, unsigned int trace_ctx)
3001{
3002 struct trace_event_call *call = &event_function;
3003 struct trace_buffer *buffer = tr->array_buffer.buffer;
3004 struct ring_buffer_event *event;
3005 struct ftrace_entry *entry;
3006
3007 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
3008 trace_ctx);
3009 if (!event)
3010 return;
3011 entry = ring_buffer_event_data(event);
3012 entry->ip = ip;
3013 entry->parent_ip = parent_ip;
3014
3015 if (!call_filter_check_discard(call, entry, buffer, event)) {
3016 if (static_branch_unlikely(&trace_function_exports_enabled))
3017 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
3018 __buffer_unlock_commit(buffer, event);
3019 }
3020}
3021
3022#ifdef CONFIG_STACKTRACE
3023
3024/* Allow 4 levels of nesting: normal, softirq, irq, NMI */
3025#define FTRACE_KSTACK_NESTING 4
3026
3027#define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3028
3029struct ftrace_stack {
3030 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3031};
3032
3033
3034struct ftrace_stacks {
3035 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3036};
3037
3038static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3039static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3040
3041static void __ftrace_trace_stack(struct trace_buffer *buffer,
3042 unsigned int trace_ctx,
3043 int skip, struct pt_regs *regs)
3044{
3045 struct trace_event_call *call = &event_kernel_stack;
3046 struct ring_buffer_event *event;
3047 unsigned int size, nr_entries;
3048 struct ftrace_stack *fstack;
3049 struct stack_entry *entry;
3050 int stackidx;
3051
3052 /*
3053 * Add one, for this function and the call to save_stack_trace()
3054 * If regs is set, then these functions will not be in the way.
3055 */
3056#ifndef CONFIG_UNWINDER_ORC
3057 if (!regs)
3058 skip++;
3059#endif
3060
3061 preempt_disable_notrace();
3062
3063 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3064
3065 /* This should never happen. If it does, yell once and skip */
3066 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3067 goto out;
3068
3069 /*
3070 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3071 * interrupt will either see the value pre increment or post
3072 * increment. If the interrupt happens pre increment it will have
3073 * restored the counter when it returns. We just need a barrier to
3074 * keep gcc from moving things around.
3075 */
3076 barrier();
3077
3078 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3079 size = ARRAY_SIZE(fstack->calls);
3080
3081 if (regs) {
3082 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3083 size, skip);
3084 } else {
3085 nr_entries = stack_trace_save(fstack->calls, size, skip);
3086 }
3087
3088 size = nr_entries * sizeof(unsigned long);
3089 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3090 (sizeof(*entry) - sizeof(entry->caller)) + size,
3091 trace_ctx);
3092 if (!event)
3093 goto out;
3094 entry = ring_buffer_event_data(event);
3095
3096 memcpy(&entry->caller, fstack->calls, size);
3097 entry->size = nr_entries;
3098
3099 if (!call_filter_check_discard(call, entry, buffer, event))
3100 __buffer_unlock_commit(buffer, event);
3101
3102 out:
3103 /* Again, don't let gcc optimize things here */
3104 barrier();
3105 __this_cpu_dec(ftrace_stack_reserve);
3106 preempt_enable_notrace();
3107
3108}
3109
3110static inline void ftrace_trace_stack(struct trace_array *tr,
3111 struct trace_buffer *buffer,
3112 unsigned int trace_ctx,
3113 int skip, struct pt_regs *regs)
3114{
3115 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3116 return;
3117
3118 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3119}
3120
3121void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3122 int skip)
3123{
3124 struct trace_buffer *buffer = tr->array_buffer.buffer;
3125
3126 if (rcu_is_watching()) {
3127 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3128 return;
3129 }
3130
3131 /*
3132 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3133 * but if the above rcu_is_watching() failed, then the NMI
3134 * triggered someplace critical, and ct_irq_enter() should
3135 * not be called from NMI.
3136 */
3137 if (unlikely(in_nmi()))
3138 return;
3139
3140 ct_irq_enter_irqson();
3141 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3142 ct_irq_exit_irqson();
3143}
3144
3145/**
3146 * trace_dump_stack - record a stack back trace in the trace buffer
3147 * @skip: Number of functions to skip (helper handlers)
3148 */
3149void trace_dump_stack(int skip)
3150{
3151 if (tracing_disabled || tracing_selftest_running)
3152 return;
3153
3154#ifndef CONFIG_UNWINDER_ORC
3155 /* Skip 1 to skip this function. */
3156 skip++;
3157#endif
3158 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3159 tracing_gen_ctx(), skip, NULL);
3160}
3161EXPORT_SYMBOL_GPL(trace_dump_stack);
3162
3163#ifdef CONFIG_USER_STACKTRACE_SUPPORT
3164static DEFINE_PER_CPU(int, user_stack_count);
3165
3166static void
3167ftrace_trace_userstack(struct trace_array *tr,
3168 struct trace_buffer *buffer, unsigned int trace_ctx)
3169{
3170 struct trace_event_call *call = &event_user_stack;
3171 struct ring_buffer_event *event;
3172 struct userstack_entry *entry;
3173
3174 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3175 return;
3176
3177 /*
3178 * NMIs can not handle page faults, even with fix ups.
3179 * The save user stack can (and often does) fault.
3180 */
3181 if (unlikely(in_nmi()))
3182 return;
3183
3184 /*
3185 * prevent recursion, since the user stack tracing may
3186 * trigger other kernel events.
3187 */
3188 preempt_disable();
3189 if (__this_cpu_read(user_stack_count))
3190 goto out;
3191
3192 __this_cpu_inc(user_stack_count);
3193
3194 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3195 sizeof(*entry), trace_ctx);
3196 if (!event)
3197 goto out_drop_count;
3198 entry = ring_buffer_event_data(event);
3199
3200 entry->tgid = current->tgid;
3201 memset(&entry->caller, 0, sizeof(entry->caller));
3202
3203 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3204 if (!call_filter_check_discard(call, entry, buffer, event))
3205 __buffer_unlock_commit(buffer, event);
3206
3207 out_drop_count:
3208 __this_cpu_dec(user_stack_count);
3209 out:
3210 preempt_enable();
3211}
3212#else /* CONFIG_USER_STACKTRACE_SUPPORT */
3213static void ftrace_trace_userstack(struct trace_array *tr,
3214 struct trace_buffer *buffer,
3215 unsigned int trace_ctx)
3216{
3217}
3218#endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3219
3220#endif /* CONFIG_STACKTRACE */
3221
3222static inline void
3223func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3224 unsigned long long delta)
3225{
3226 entry->bottom_delta_ts = delta & U32_MAX;
3227 entry->top_delta_ts = (delta >> 32);
3228}
3229
3230void trace_last_func_repeats(struct trace_array *tr,
3231 struct trace_func_repeats *last_info,
3232 unsigned int trace_ctx)
3233{
3234 struct trace_buffer *buffer = tr->array_buffer.buffer;
3235 struct func_repeats_entry *entry;
3236 struct ring_buffer_event *event;
3237 u64 delta;
3238
3239 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3240 sizeof(*entry), trace_ctx);
3241 if (!event)
3242 return;
3243
3244 delta = ring_buffer_event_time_stamp(buffer, event) -
3245 last_info->ts_last_call;
3246
3247 entry = ring_buffer_event_data(event);
3248 entry->ip = last_info->ip;
3249 entry->parent_ip = last_info->parent_ip;
3250 entry->count = last_info->count;
3251 func_repeats_set_delta_ts(entry, delta);
3252
3253 __buffer_unlock_commit(buffer, event);
3254}
3255
3256/* created for use with alloc_percpu */
3257struct trace_buffer_struct {
3258 int nesting;
3259 char buffer[4][TRACE_BUF_SIZE];
3260};
3261
3262static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3263
3264/*
3265 * This allows for lockless recording. If we're nested too deeply, then
3266 * this returns NULL.
3267 */
3268static char *get_trace_buf(void)
3269{
3270 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3271
3272 if (!trace_percpu_buffer || buffer->nesting >= 4)
3273 return NULL;
3274
3275 buffer->nesting++;
3276
3277 /* Interrupts must see nesting incremented before we use the buffer */
3278 barrier();
3279 return &buffer->buffer[buffer->nesting - 1][0];
3280}
3281
3282static void put_trace_buf(void)
3283{
3284 /* Don't let the decrement of nesting leak before this */
3285 barrier();
3286 this_cpu_dec(trace_percpu_buffer->nesting);
3287}
3288
3289static int alloc_percpu_trace_buffer(void)
3290{
3291 struct trace_buffer_struct __percpu *buffers;
3292
3293 if (trace_percpu_buffer)
3294 return 0;
3295
3296 buffers = alloc_percpu(struct trace_buffer_struct);
3297 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3298 return -ENOMEM;
3299
3300 trace_percpu_buffer = buffers;
3301 return 0;
3302}
3303
3304static int buffers_allocated;
3305
3306void trace_printk_init_buffers(void)
3307{
3308 if (buffers_allocated)
3309 return;
3310
3311 if (alloc_percpu_trace_buffer())
3312 return;
3313
3314 /* trace_printk() is for debug use only. Don't use it in production. */
3315
3316 pr_warn("\n");
3317 pr_warn("**********************************************************\n");
3318 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3319 pr_warn("** **\n");
3320 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3321 pr_warn("** **\n");
3322 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3323 pr_warn("** unsafe for production use. **\n");
3324 pr_warn("** **\n");
3325 pr_warn("** If you see this message and you are not debugging **\n");
3326 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3327 pr_warn("** **\n");
3328 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3329 pr_warn("**********************************************************\n");
3330
3331 /* Expand the buffers to set size */
3332 tracing_update_buffers();
3333
3334 buffers_allocated = 1;
3335
3336 /*
3337 * trace_printk_init_buffers() can be called by modules.
3338 * If that happens, then we need to start cmdline recording
3339 * directly here. If the global_trace.buffer is already
3340 * allocated here, then this was called by module code.
3341 */
3342 if (global_trace.array_buffer.buffer)
3343 tracing_start_cmdline_record();
3344}
3345EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3346
3347void trace_printk_start_comm(void)
3348{
3349 /* Start tracing comms if trace printk is set */
3350 if (!buffers_allocated)
3351 return;
3352 tracing_start_cmdline_record();
3353}
3354
3355static void trace_printk_start_stop_comm(int enabled)
3356{
3357 if (!buffers_allocated)
3358 return;
3359
3360 if (enabled)
3361 tracing_start_cmdline_record();
3362 else
3363 tracing_stop_cmdline_record();
3364}
3365
3366/**
3367 * trace_vbprintk - write binary msg to tracing buffer
3368 * @ip: The address of the caller
3369 * @fmt: The string format to write to the buffer
3370 * @args: Arguments for @fmt
3371 */
3372int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3373{
3374 struct trace_event_call *call = &event_bprint;
3375 struct ring_buffer_event *event;
3376 struct trace_buffer *buffer;
3377 struct trace_array *tr = &global_trace;
3378 struct bprint_entry *entry;
3379 unsigned int trace_ctx;
3380 char *tbuffer;
3381 int len = 0, size;
3382
3383 if (unlikely(tracing_selftest_running || tracing_disabled))
3384 return 0;
3385
3386 /* Don't pollute graph traces with trace_vprintk internals */
3387 pause_graph_tracing();
3388
3389 trace_ctx = tracing_gen_ctx();
3390 preempt_disable_notrace();
3391
3392 tbuffer = get_trace_buf();
3393 if (!tbuffer) {
3394 len = 0;
3395 goto out_nobuffer;
3396 }
3397
3398 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3399
3400 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3401 goto out_put;
3402
3403 size = sizeof(*entry) + sizeof(u32) * len;
3404 buffer = tr->array_buffer.buffer;
3405 ring_buffer_nest_start(buffer);
3406 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3407 trace_ctx);
3408 if (!event)
3409 goto out;
3410 entry = ring_buffer_event_data(event);
3411 entry->ip = ip;
3412 entry->fmt = fmt;
3413
3414 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3415 if (!call_filter_check_discard(call, entry, buffer, event)) {
3416 __buffer_unlock_commit(buffer, event);
3417 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3418 }
3419
3420out:
3421 ring_buffer_nest_end(buffer);
3422out_put:
3423 put_trace_buf();
3424
3425out_nobuffer:
3426 preempt_enable_notrace();
3427 unpause_graph_tracing();
3428
3429 return len;
3430}
3431EXPORT_SYMBOL_GPL(trace_vbprintk);
3432
3433__printf(3, 0)
3434static int
3435__trace_array_vprintk(struct trace_buffer *buffer,
3436 unsigned long ip, const char *fmt, va_list args)
3437{
3438 struct trace_event_call *call = &event_print;
3439 struct ring_buffer_event *event;
3440 int len = 0, size;
3441 struct print_entry *entry;
3442 unsigned int trace_ctx;
3443 char *tbuffer;
3444
3445 if (tracing_disabled || tracing_selftest_running)
3446 return 0;
3447
3448 /* Don't pollute graph traces with trace_vprintk internals */
3449 pause_graph_tracing();
3450
3451 trace_ctx = tracing_gen_ctx();
3452 preempt_disable_notrace();
3453
3454
3455 tbuffer = get_trace_buf();
3456 if (!tbuffer) {
3457 len = 0;
3458 goto out_nobuffer;
3459 }
3460
3461 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3462
3463 size = sizeof(*entry) + len + 1;
3464 ring_buffer_nest_start(buffer);
3465 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3466 trace_ctx);
3467 if (!event)
3468 goto out;
3469 entry = ring_buffer_event_data(event);
3470 entry->ip = ip;
3471
3472 memcpy(&entry->buf, tbuffer, len + 1);
3473 if (!call_filter_check_discard(call, entry, buffer, event)) {
3474 __buffer_unlock_commit(buffer, event);
3475 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3476 }
3477
3478out:
3479 ring_buffer_nest_end(buffer);
3480 put_trace_buf();
3481
3482out_nobuffer:
3483 preempt_enable_notrace();
3484 unpause_graph_tracing();
3485
3486 return len;
3487}
3488
3489__printf(3, 0)
3490int trace_array_vprintk(struct trace_array *tr,
3491 unsigned long ip, const char *fmt, va_list args)
3492{
3493 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3494}
3495
3496/**
3497 * trace_array_printk - Print a message to a specific instance
3498 * @tr: The instance trace_array descriptor
3499 * @ip: The instruction pointer that this is called from.
3500 * @fmt: The format to print (printf format)
3501 *
3502 * If a subsystem sets up its own instance, they have the right to
3503 * printk strings into their tracing instance buffer using this
3504 * function. Note, this function will not write into the top level
3505 * buffer (use trace_printk() for that), as writing into the top level
3506 * buffer should only have events that can be individually disabled.
3507 * trace_printk() is only used for debugging a kernel, and should not
3508 * be ever incorporated in normal use.
3509 *
3510 * trace_array_printk() can be used, as it will not add noise to the
3511 * top level tracing buffer.
3512 *
3513 * Note, trace_array_init_printk() must be called on @tr before this
3514 * can be used.
3515 */
3516__printf(3, 0)
3517int trace_array_printk(struct trace_array *tr,
3518 unsigned long ip, const char *fmt, ...)
3519{
3520 int ret;
3521 va_list ap;
3522
3523 if (!tr)
3524 return -ENOENT;
3525
3526 /* This is only allowed for created instances */
3527 if (tr == &global_trace)
3528 return 0;
3529
3530 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3531 return 0;
3532
3533 va_start(ap, fmt);
3534 ret = trace_array_vprintk(tr, ip, fmt, ap);
3535 va_end(ap);
3536 return ret;
3537}
3538EXPORT_SYMBOL_GPL(trace_array_printk);
3539
3540/**
3541 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3542 * @tr: The trace array to initialize the buffers for
3543 *
3544 * As trace_array_printk() only writes into instances, they are OK to
3545 * have in the kernel (unlike trace_printk()). This needs to be called
3546 * before trace_array_printk() can be used on a trace_array.
3547 */
3548int trace_array_init_printk(struct trace_array *tr)
3549{
3550 if (!tr)
3551 return -ENOENT;
3552
3553 /* This is only allowed for created instances */
3554 if (tr == &global_trace)
3555 return -EINVAL;
3556
3557 return alloc_percpu_trace_buffer();
3558}
3559EXPORT_SYMBOL_GPL(trace_array_init_printk);
3560
3561__printf(3, 4)
3562int trace_array_printk_buf(struct trace_buffer *buffer,
3563 unsigned long ip, const char *fmt, ...)
3564{
3565 int ret;
3566 va_list ap;
3567
3568 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3569 return 0;
3570
3571 va_start(ap, fmt);
3572 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3573 va_end(ap);
3574 return ret;
3575}
3576
3577__printf(2, 0)
3578int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3579{
3580 return trace_array_vprintk(&global_trace, ip, fmt, args);
3581}
3582EXPORT_SYMBOL_GPL(trace_vprintk);
3583
3584static void trace_iterator_increment(struct trace_iterator *iter)
3585{
3586 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3587
3588 iter->idx++;
3589 if (buf_iter)
3590 ring_buffer_iter_advance(buf_iter);
3591}
3592
3593static struct trace_entry *
3594peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3595 unsigned long *lost_events)
3596{
3597 struct ring_buffer_event *event;
3598 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3599
3600 if (buf_iter) {
3601 event = ring_buffer_iter_peek(buf_iter, ts);
3602 if (lost_events)
3603 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3604 (unsigned long)-1 : 0;
3605 } else {
3606 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3607 lost_events);
3608 }
3609
3610 if (event) {
3611 iter->ent_size = ring_buffer_event_length(event);
3612 return ring_buffer_event_data(event);
3613 }
3614 iter->ent_size = 0;
3615 return NULL;
3616}
3617
3618static struct trace_entry *
3619__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3620 unsigned long *missing_events, u64 *ent_ts)
3621{
3622 struct trace_buffer *buffer = iter->array_buffer->buffer;
3623 struct trace_entry *ent, *next = NULL;
3624 unsigned long lost_events = 0, next_lost = 0;
3625 int cpu_file = iter->cpu_file;
3626 u64 next_ts = 0, ts;
3627 int next_cpu = -1;
3628 int next_size = 0;
3629 int cpu;
3630
3631 /*
3632 * If we are in a per_cpu trace file, don't bother by iterating over
3633 * all cpu and peek directly.
3634 */
3635 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3636 if (ring_buffer_empty_cpu(buffer, cpu_file))
3637 return NULL;
3638 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3639 if (ent_cpu)
3640 *ent_cpu = cpu_file;
3641
3642 return ent;
3643 }
3644
3645 for_each_tracing_cpu(cpu) {
3646
3647 if (ring_buffer_empty_cpu(buffer, cpu))
3648 continue;
3649
3650 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3651
3652 /*
3653 * Pick the entry with the smallest timestamp:
3654 */
3655 if (ent && (!next || ts < next_ts)) {
3656 next = ent;
3657 next_cpu = cpu;
3658 next_ts = ts;
3659 next_lost = lost_events;
3660 next_size = iter->ent_size;
3661 }
3662 }
3663
3664 iter->ent_size = next_size;
3665
3666 if (ent_cpu)
3667 *ent_cpu = next_cpu;
3668
3669 if (ent_ts)
3670 *ent_ts = next_ts;
3671
3672 if (missing_events)
3673 *missing_events = next_lost;
3674
3675 return next;
3676}
3677
3678#define STATIC_FMT_BUF_SIZE 128
3679static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3680
3681static char *trace_iter_expand_format(struct trace_iterator *iter)
3682{
3683 char *tmp;
3684
3685 /*
3686 * iter->tr is NULL when used with tp_printk, which makes
3687 * this get called where it is not safe to call krealloc().
3688 */
3689 if (!iter->tr || iter->fmt == static_fmt_buf)
3690 return NULL;
3691
3692 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3693 GFP_KERNEL);
3694 if (tmp) {
3695 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3696 iter->fmt = tmp;
3697 }
3698
3699 return tmp;
3700}
3701
3702/* Returns true if the string is safe to dereference from an event */
3703static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3704 bool star, int len)
3705{
3706 unsigned long addr = (unsigned long)str;
3707 struct trace_event *trace_event;
3708 struct trace_event_call *event;
3709
3710 /* Ignore strings with no length */
3711 if (star && !len)
3712 return true;
3713
3714 /* OK if part of the event data */
3715 if ((addr >= (unsigned long)iter->ent) &&
3716 (addr < (unsigned long)iter->ent + iter->ent_size))
3717 return true;
3718
3719 /* OK if part of the temp seq buffer */
3720 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3721 (addr < (unsigned long)iter->tmp_seq.buffer + PAGE_SIZE))
3722 return true;
3723
3724 /* Core rodata can not be freed */
3725 if (is_kernel_rodata(addr))
3726 return true;
3727
3728 if (trace_is_tracepoint_string(str))
3729 return true;
3730
3731 /*
3732 * Now this could be a module event, referencing core module
3733 * data, which is OK.
3734 */
3735 if (!iter->ent)
3736 return false;
3737
3738 trace_event = ftrace_find_event(iter->ent->type);
3739 if (!trace_event)
3740 return false;
3741
3742 event = container_of(trace_event, struct trace_event_call, event);
3743 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3744 return false;
3745
3746 /* Would rather have rodata, but this will suffice */
3747 if (within_module_core(addr, event->module))
3748 return true;
3749
3750 return false;
3751}
3752
3753static const char *show_buffer(struct trace_seq *s)
3754{
3755 struct seq_buf *seq = &s->seq;
3756
3757 seq_buf_terminate(seq);
3758
3759 return seq->buffer;
3760}
3761
3762static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3763
3764static int test_can_verify_check(const char *fmt, ...)
3765{
3766 char buf[16];
3767 va_list ap;
3768 int ret;
3769
3770 /*
3771 * The verifier is dependent on vsnprintf() modifies the va_list
3772 * passed to it, where it is sent as a reference. Some architectures
3773 * (like x86_32) passes it by value, which means that vsnprintf()
3774 * does not modify the va_list passed to it, and the verifier
3775 * would then need to be able to understand all the values that
3776 * vsnprintf can use. If it is passed by value, then the verifier
3777 * is disabled.
3778 */
3779 va_start(ap, fmt);
3780 vsnprintf(buf, 16, "%d", ap);
3781 ret = va_arg(ap, int);
3782 va_end(ap);
3783
3784 return ret;
3785}
3786
3787static void test_can_verify(void)
3788{
3789 if (!test_can_verify_check("%d %d", 0, 1)) {
3790 pr_info("trace event string verifier disabled\n");
3791 static_branch_inc(&trace_no_verify);
3792 }
3793}
3794
3795/**
3796 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3797 * @iter: The iterator that holds the seq buffer and the event being printed
3798 * @fmt: The format used to print the event
3799 * @ap: The va_list holding the data to print from @fmt.
3800 *
3801 * This writes the data into the @iter->seq buffer using the data from
3802 * @fmt and @ap. If the format has a %s, then the source of the string
3803 * is examined to make sure it is safe to print, otherwise it will
3804 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3805 * pointer.
3806 */
3807void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3808 va_list ap)
3809{
3810 const char *p = fmt;
3811 const char *str;
3812 int i, j;
3813
3814 if (WARN_ON_ONCE(!fmt))
3815 return;
3816
3817 if (static_branch_unlikely(&trace_no_verify))
3818 goto print;
3819
3820 /* Don't bother checking when doing a ftrace_dump() */
3821 if (iter->fmt == static_fmt_buf)
3822 goto print;
3823
3824 while (*p) {
3825 bool star = false;
3826 int len = 0;
3827
3828 j = 0;
3829
3830 /* We only care about %s and variants */
3831 for (i = 0; p[i]; i++) {
3832 if (i + 1 >= iter->fmt_size) {
3833 /*
3834 * If we can't expand the copy buffer,
3835 * just print it.
3836 */
3837 if (!trace_iter_expand_format(iter))
3838 goto print;
3839 }
3840
3841 if (p[i] == '\\' && p[i+1]) {
3842 i++;
3843 continue;
3844 }
3845 if (p[i] == '%') {
3846 /* Need to test cases like %08.*s */
3847 for (j = 1; p[i+j]; j++) {
3848 if (isdigit(p[i+j]) ||
3849 p[i+j] == '.')
3850 continue;
3851 if (p[i+j] == '*') {
3852 star = true;
3853 continue;
3854 }
3855 break;
3856 }
3857 if (p[i+j] == 's')
3858 break;
3859 star = false;
3860 }
3861 j = 0;
3862 }
3863 /* If no %s found then just print normally */
3864 if (!p[i])
3865 break;
3866
3867 /* Copy up to the %s, and print that */
3868 strncpy(iter->fmt, p, i);
3869 iter->fmt[i] = '\0';
3870 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3871
3872 /*
3873 * If iter->seq is full, the above call no longer guarantees
3874 * that ap is in sync with fmt processing, and further calls
3875 * to va_arg() can return wrong positional arguments.
3876 *
3877 * Ensure that ap is no longer used in this case.
3878 */
3879 if (iter->seq.full) {
3880 p = "";
3881 break;
3882 }
3883
3884 if (star)
3885 len = va_arg(ap, int);
3886
3887 /* The ap now points to the string data of the %s */
3888 str = va_arg(ap, const char *);
3889
3890 /*
3891 * If you hit this warning, it is likely that the
3892 * trace event in question used %s on a string that
3893 * was saved at the time of the event, but may not be
3894 * around when the trace is read. Use __string(),
3895 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3896 * instead. See samples/trace_events/trace-events-sample.h
3897 * for reference.
3898 */
3899 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3900 "fmt: '%s' current_buffer: '%s'",
3901 fmt, show_buffer(&iter->seq))) {
3902 int ret;
3903
3904 /* Try to safely read the string */
3905 if (star) {
3906 if (len + 1 > iter->fmt_size)
3907 len = iter->fmt_size - 1;
3908 if (len < 0)
3909 len = 0;
3910 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3911 iter->fmt[len] = 0;
3912 star = false;
3913 } else {
3914 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3915 iter->fmt_size);
3916 }
3917 if (ret < 0)
3918 trace_seq_printf(&iter->seq, "(0x%px)", str);
3919 else
3920 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3921 str, iter->fmt);
3922 str = "[UNSAFE-MEMORY]";
3923 strcpy(iter->fmt, "%s");
3924 } else {
3925 strncpy(iter->fmt, p + i, j + 1);
3926 iter->fmt[j+1] = '\0';
3927 }
3928 if (star)
3929 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3930 else
3931 trace_seq_printf(&iter->seq, iter->fmt, str);
3932
3933 p += i + j + 1;
3934 }
3935 print:
3936 if (*p)
3937 trace_seq_vprintf(&iter->seq, p, ap);
3938}
3939
3940const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3941{
3942 const char *p, *new_fmt;
3943 char *q;
3944
3945 if (WARN_ON_ONCE(!fmt))
3946 return fmt;
3947
3948 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
3949 return fmt;
3950
3951 p = fmt;
3952 new_fmt = q = iter->fmt;
3953 while (*p) {
3954 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
3955 if (!trace_iter_expand_format(iter))
3956 return fmt;
3957
3958 q += iter->fmt - new_fmt;
3959 new_fmt = iter->fmt;
3960 }
3961
3962 *q++ = *p++;
3963
3964 /* Replace %p with %px */
3965 if (p[-1] == '%') {
3966 if (p[0] == '%') {
3967 *q++ = *p++;
3968 } else if (p[0] == 'p' && !isalnum(p[1])) {
3969 *q++ = *p++;
3970 *q++ = 'x';
3971 }
3972 }
3973 }
3974 *q = '\0';
3975
3976 return new_fmt;
3977}
3978
3979#define STATIC_TEMP_BUF_SIZE 128
3980static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3981
3982/* Find the next real entry, without updating the iterator itself */
3983struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3984 int *ent_cpu, u64 *ent_ts)
3985{
3986 /* __find_next_entry will reset ent_size */
3987 int ent_size = iter->ent_size;
3988 struct trace_entry *entry;
3989
3990 /*
3991 * If called from ftrace_dump(), then the iter->temp buffer
3992 * will be the static_temp_buf and not created from kmalloc.
3993 * If the entry size is greater than the buffer, we can
3994 * not save it. Just return NULL in that case. This is only
3995 * used to add markers when two consecutive events' time
3996 * stamps have a large delta. See trace_print_lat_context()
3997 */
3998 if (iter->temp == static_temp_buf &&
3999 STATIC_TEMP_BUF_SIZE < ent_size)
4000 return NULL;
4001
4002 /*
4003 * The __find_next_entry() may call peek_next_entry(), which may
4004 * call ring_buffer_peek() that may make the contents of iter->ent
4005 * undefined. Need to copy iter->ent now.
4006 */
4007 if (iter->ent && iter->ent != iter->temp) {
4008 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4009 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4010 void *temp;
4011 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4012 if (!temp)
4013 return NULL;
4014 kfree(iter->temp);
4015 iter->temp = temp;
4016 iter->temp_size = iter->ent_size;
4017 }
4018 memcpy(iter->temp, iter->ent, iter->ent_size);
4019 iter->ent = iter->temp;
4020 }
4021 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4022 /* Put back the original ent_size */
4023 iter->ent_size = ent_size;
4024
4025 return entry;
4026}
4027
4028/* Find the next real entry, and increment the iterator to the next entry */
4029void *trace_find_next_entry_inc(struct trace_iterator *iter)
4030{
4031 iter->ent = __find_next_entry(iter, &iter->cpu,
4032 &iter->lost_events, &iter->ts);
4033
4034 if (iter->ent)
4035 trace_iterator_increment(iter);
4036
4037 return iter->ent ? iter : NULL;
4038}
4039
4040static void trace_consume(struct trace_iterator *iter)
4041{
4042 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4043 &iter->lost_events);
4044}
4045
4046static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4047{
4048 struct trace_iterator *iter = m->private;
4049 int i = (int)*pos;
4050 void *ent;
4051
4052 WARN_ON_ONCE(iter->leftover);
4053
4054 (*pos)++;
4055
4056 /* can't go backwards */
4057 if (iter->idx > i)
4058 return NULL;
4059
4060 if (iter->idx < 0)
4061 ent = trace_find_next_entry_inc(iter);
4062 else
4063 ent = iter;
4064
4065 while (ent && iter->idx < i)
4066 ent = trace_find_next_entry_inc(iter);
4067
4068 iter->pos = *pos;
4069
4070 return ent;
4071}
4072
4073void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4074{
4075 struct ring_buffer_iter *buf_iter;
4076 unsigned long entries = 0;
4077 u64 ts;
4078
4079 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4080
4081 buf_iter = trace_buffer_iter(iter, cpu);
4082 if (!buf_iter)
4083 return;
4084
4085 ring_buffer_iter_reset(buf_iter);
4086
4087 /*
4088 * We could have the case with the max latency tracers
4089 * that a reset never took place on a cpu. This is evident
4090 * by the timestamp being before the start of the buffer.
4091 */
4092 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4093 if (ts >= iter->array_buffer->time_start)
4094 break;
4095 entries++;
4096 ring_buffer_iter_advance(buf_iter);
4097 }
4098
4099 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4100}
4101
4102/*
4103 * The current tracer is copied to avoid a global locking
4104 * all around.
4105 */
4106static void *s_start(struct seq_file *m, loff_t *pos)
4107{
4108 struct trace_iterator *iter = m->private;
4109 struct trace_array *tr = iter->tr;
4110 int cpu_file = iter->cpu_file;
4111 void *p = NULL;
4112 loff_t l = 0;
4113 int cpu;
4114
4115 /*
4116 * copy the tracer to avoid using a global lock all around.
4117 * iter->trace is a copy of current_trace, the pointer to the
4118 * name may be used instead of a strcmp(), as iter->trace->name
4119 * will point to the same string as current_trace->name.
4120 */
4121 mutex_lock(&trace_types_lock);
4122 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
4123 *iter->trace = *tr->current_trace;
4124 mutex_unlock(&trace_types_lock);
4125
4126#ifdef CONFIG_TRACER_MAX_TRACE
4127 if (iter->snapshot && iter->trace->use_max_tr)
4128 return ERR_PTR(-EBUSY);
4129#endif
4130
4131 if (*pos != iter->pos) {
4132 iter->ent = NULL;
4133 iter->cpu = 0;
4134 iter->idx = -1;
4135
4136 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4137 for_each_tracing_cpu(cpu)
4138 tracing_iter_reset(iter, cpu);
4139 } else
4140 tracing_iter_reset(iter, cpu_file);
4141
4142 iter->leftover = 0;
4143 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4144 ;
4145
4146 } else {
4147 /*
4148 * If we overflowed the seq_file before, then we want
4149 * to just reuse the trace_seq buffer again.
4150 */
4151 if (iter->leftover)
4152 p = iter;
4153 else {
4154 l = *pos - 1;
4155 p = s_next(m, p, &l);
4156 }
4157 }
4158
4159 trace_event_read_lock();
4160 trace_access_lock(cpu_file);
4161 return p;
4162}
4163
4164static void s_stop(struct seq_file *m, void *p)
4165{
4166 struct trace_iterator *iter = m->private;
4167
4168#ifdef CONFIG_TRACER_MAX_TRACE
4169 if (iter->snapshot && iter->trace->use_max_tr)
4170 return;
4171#endif
4172
4173 trace_access_unlock(iter->cpu_file);
4174 trace_event_read_unlock();
4175}
4176
4177static void
4178get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4179 unsigned long *entries, int cpu)
4180{
4181 unsigned long count;
4182
4183 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4184 /*
4185 * If this buffer has skipped entries, then we hold all
4186 * entries for the trace and we need to ignore the
4187 * ones before the time stamp.
4188 */
4189 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4190 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4191 /* total is the same as the entries */
4192 *total = count;
4193 } else
4194 *total = count +
4195 ring_buffer_overrun_cpu(buf->buffer, cpu);
4196 *entries = count;
4197}
4198
4199static void
4200get_total_entries(struct array_buffer *buf,
4201 unsigned long *total, unsigned long *entries)
4202{
4203 unsigned long t, e;
4204 int cpu;
4205
4206 *total = 0;
4207 *entries = 0;
4208
4209 for_each_tracing_cpu(cpu) {
4210 get_total_entries_cpu(buf, &t, &e, cpu);
4211 *total += t;
4212 *entries += e;
4213 }
4214}
4215
4216unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4217{
4218 unsigned long total, entries;
4219
4220 if (!tr)
4221 tr = &global_trace;
4222
4223 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4224
4225 return entries;
4226}
4227
4228unsigned long trace_total_entries(struct trace_array *tr)
4229{
4230 unsigned long total, entries;
4231
4232 if (!tr)
4233 tr = &global_trace;
4234
4235 get_total_entries(&tr->array_buffer, &total, &entries);
4236
4237 return entries;
4238}
4239
4240static void print_lat_help_header(struct seq_file *m)
4241{
4242 seq_puts(m, "# _------=> CPU# \n"
4243 "# / _-----=> irqs-off/BH-disabled\n"
4244 "# | / _----=> need-resched \n"
4245 "# || / _---=> hardirq/softirq \n"
4246 "# ||| / _--=> preempt-depth \n"
4247 "# |||| / _-=> migrate-disable \n"
4248 "# ||||| / delay \n"
4249 "# cmd pid |||||| time | caller \n"
4250 "# \\ / |||||| \\ | / \n");
4251}
4252
4253static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4254{
4255 unsigned long total;
4256 unsigned long entries;
4257
4258 get_total_entries(buf, &total, &entries);
4259 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4260 entries, total, num_online_cpus());
4261 seq_puts(m, "#\n");
4262}
4263
4264static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4265 unsigned int flags)
4266{
4267 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4268
4269 print_event_info(buf, m);
4270
4271 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4272 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4273}
4274
4275static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4276 unsigned int flags)
4277{
4278 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4279 static const char space[] = " ";
4280 int prec = tgid ? 12 : 2;
4281
4282 print_event_info(buf, m);
4283
4284 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4285 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4286 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4287 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4288 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4289 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4290 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4291 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4292}
4293
4294void
4295print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4296{
4297 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4298 struct array_buffer *buf = iter->array_buffer;
4299 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4300 struct tracer *type = iter->trace;
4301 unsigned long entries;
4302 unsigned long total;
4303 const char *name = type->name;
4304
4305 get_total_entries(buf, &total, &entries);
4306
4307 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4308 name, UTS_RELEASE);
4309 seq_puts(m, "# -----------------------------------"
4310 "---------------------------------\n");
4311 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4312 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4313 nsecs_to_usecs(data->saved_latency),
4314 entries,
4315 total,
4316 buf->cpu,
4317 preempt_model_none() ? "server" :
4318 preempt_model_voluntary() ? "desktop" :
4319 preempt_model_full() ? "preempt" :
4320 preempt_model_rt() ? "preempt_rt" :
4321 "unknown",
4322 /* These are reserved for later use */
4323 0, 0, 0, 0);
4324#ifdef CONFIG_SMP
4325 seq_printf(m, " #P:%d)\n", num_online_cpus());
4326#else
4327 seq_puts(m, ")\n");
4328#endif
4329 seq_puts(m, "# -----------------\n");
4330 seq_printf(m, "# | task: %.16s-%d "
4331 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4332 data->comm, data->pid,
4333 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4334 data->policy, data->rt_priority);
4335 seq_puts(m, "# -----------------\n");
4336
4337 if (data->critical_start) {
4338 seq_puts(m, "# => started at: ");
4339 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4340 trace_print_seq(m, &iter->seq);
4341 seq_puts(m, "\n# => ended at: ");
4342 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4343 trace_print_seq(m, &iter->seq);
4344 seq_puts(m, "\n#\n");
4345 }
4346
4347 seq_puts(m, "#\n");
4348}
4349
4350static void test_cpu_buff_start(struct trace_iterator *iter)
4351{
4352 struct trace_seq *s = &iter->seq;
4353 struct trace_array *tr = iter->tr;
4354
4355 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4356 return;
4357
4358 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4359 return;
4360
4361 if (cpumask_available(iter->started) &&
4362 cpumask_test_cpu(iter->cpu, iter->started))
4363 return;
4364
4365 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4366 return;
4367
4368 if (cpumask_available(iter->started))
4369 cpumask_set_cpu(iter->cpu, iter->started);
4370
4371 /* Don't print started cpu buffer for the first entry of the trace */
4372 if (iter->idx > 1)
4373 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4374 iter->cpu);
4375}
4376
4377static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4378{
4379 struct trace_array *tr = iter->tr;
4380 struct trace_seq *s = &iter->seq;
4381 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4382 struct trace_entry *entry;
4383 struct trace_event *event;
4384
4385 entry = iter->ent;
4386
4387 test_cpu_buff_start(iter);
4388
4389 event = ftrace_find_event(entry->type);
4390
4391 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4392 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4393 trace_print_lat_context(iter);
4394 else
4395 trace_print_context(iter);
4396 }
4397
4398 if (trace_seq_has_overflowed(s))
4399 return TRACE_TYPE_PARTIAL_LINE;
4400
4401 if (event)
4402 return event->funcs->trace(iter, sym_flags, event);
4403
4404 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4405
4406 return trace_handle_return(s);
4407}
4408
4409static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4410{
4411 struct trace_array *tr = iter->tr;
4412 struct trace_seq *s = &iter->seq;
4413 struct trace_entry *entry;
4414 struct trace_event *event;
4415
4416 entry = iter->ent;
4417
4418 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4419 trace_seq_printf(s, "%d %d %llu ",
4420 entry->pid, iter->cpu, iter->ts);
4421
4422 if (trace_seq_has_overflowed(s))
4423 return TRACE_TYPE_PARTIAL_LINE;
4424
4425 event = ftrace_find_event(entry->type);
4426 if (event)
4427 return event->funcs->raw(iter, 0, event);
4428
4429 trace_seq_printf(s, "%d ?\n", entry->type);
4430
4431 return trace_handle_return(s);
4432}
4433
4434static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4435{
4436 struct trace_array *tr = iter->tr;
4437 struct trace_seq *s = &iter->seq;
4438 unsigned char newline = '\n';
4439 struct trace_entry *entry;
4440 struct trace_event *event;
4441
4442 entry = iter->ent;
4443
4444 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4445 SEQ_PUT_HEX_FIELD(s, entry->pid);
4446 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4447 SEQ_PUT_HEX_FIELD(s, iter->ts);
4448 if (trace_seq_has_overflowed(s))
4449 return TRACE_TYPE_PARTIAL_LINE;
4450 }
4451
4452 event = ftrace_find_event(entry->type);
4453 if (event) {
4454 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4455 if (ret != TRACE_TYPE_HANDLED)
4456 return ret;
4457 }
4458
4459 SEQ_PUT_FIELD(s, newline);
4460
4461 return trace_handle_return(s);
4462}
4463
4464static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4465{
4466 struct trace_array *tr = iter->tr;
4467 struct trace_seq *s = &iter->seq;
4468 struct trace_entry *entry;
4469 struct trace_event *event;
4470
4471 entry = iter->ent;
4472
4473 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4474 SEQ_PUT_FIELD(s, entry->pid);
4475 SEQ_PUT_FIELD(s, iter->cpu);
4476 SEQ_PUT_FIELD(s, iter->ts);
4477 if (trace_seq_has_overflowed(s))
4478 return TRACE_TYPE_PARTIAL_LINE;
4479 }
4480
4481 event = ftrace_find_event(entry->type);
4482 return event ? event->funcs->binary(iter, 0, event) :
4483 TRACE_TYPE_HANDLED;
4484}
4485
4486int trace_empty(struct trace_iterator *iter)
4487{
4488 struct ring_buffer_iter *buf_iter;
4489 int cpu;
4490
4491 /* If we are looking at one CPU buffer, only check that one */
4492 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4493 cpu = iter->cpu_file;
4494 buf_iter = trace_buffer_iter(iter, cpu);
4495 if (buf_iter) {
4496 if (!ring_buffer_iter_empty(buf_iter))
4497 return 0;
4498 } else {
4499 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4500 return 0;
4501 }
4502 return 1;
4503 }
4504
4505 for_each_tracing_cpu(cpu) {
4506 buf_iter = trace_buffer_iter(iter, cpu);
4507 if (buf_iter) {
4508 if (!ring_buffer_iter_empty(buf_iter))
4509 return 0;
4510 } else {
4511 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4512 return 0;
4513 }
4514 }
4515
4516 return 1;
4517}
4518
4519/* Called with trace_event_read_lock() held. */
4520enum print_line_t print_trace_line(struct trace_iterator *iter)
4521{
4522 struct trace_array *tr = iter->tr;
4523 unsigned long trace_flags = tr->trace_flags;
4524 enum print_line_t ret;
4525
4526 if (iter->lost_events) {
4527 if (iter->lost_events == (unsigned long)-1)
4528 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4529 iter->cpu);
4530 else
4531 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4532 iter->cpu, iter->lost_events);
4533 if (trace_seq_has_overflowed(&iter->seq))
4534 return TRACE_TYPE_PARTIAL_LINE;
4535 }
4536
4537 if (iter->trace && iter->trace->print_line) {
4538 ret = iter->trace->print_line(iter);
4539 if (ret != TRACE_TYPE_UNHANDLED)
4540 return ret;
4541 }
4542
4543 if (iter->ent->type == TRACE_BPUTS &&
4544 trace_flags & TRACE_ITER_PRINTK &&
4545 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4546 return trace_print_bputs_msg_only(iter);
4547
4548 if (iter->ent->type == TRACE_BPRINT &&
4549 trace_flags & TRACE_ITER_PRINTK &&
4550 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4551 return trace_print_bprintk_msg_only(iter);
4552
4553 if (iter->ent->type == TRACE_PRINT &&
4554 trace_flags & TRACE_ITER_PRINTK &&
4555 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4556 return trace_print_printk_msg_only(iter);
4557
4558 if (trace_flags & TRACE_ITER_BIN)
4559 return print_bin_fmt(iter);
4560
4561 if (trace_flags & TRACE_ITER_HEX)
4562 return print_hex_fmt(iter);
4563
4564 if (trace_flags & TRACE_ITER_RAW)
4565 return print_raw_fmt(iter);
4566
4567 return print_trace_fmt(iter);
4568}
4569
4570void trace_latency_header(struct seq_file *m)
4571{
4572 struct trace_iterator *iter = m->private;
4573 struct trace_array *tr = iter->tr;
4574
4575 /* print nothing if the buffers are empty */
4576 if (trace_empty(iter))
4577 return;
4578
4579 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4580 print_trace_header(m, iter);
4581
4582 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4583 print_lat_help_header(m);
4584}
4585
4586void trace_default_header(struct seq_file *m)
4587{
4588 struct trace_iterator *iter = m->private;
4589 struct trace_array *tr = iter->tr;
4590 unsigned long trace_flags = tr->trace_flags;
4591
4592 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4593 return;
4594
4595 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4596 /* print nothing if the buffers are empty */
4597 if (trace_empty(iter))
4598 return;
4599 print_trace_header(m, iter);
4600 if (!(trace_flags & TRACE_ITER_VERBOSE))
4601 print_lat_help_header(m);
4602 } else {
4603 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4604 if (trace_flags & TRACE_ITER_IRQ_INFO)
4605 print_func_help_header_irq(iter->array_buffer,
4606 m, trace_flags);
4607 else
4608 print_func_help_header(iter->array_buffer, m,
4609 trace_flags);
4610 }
4611 }
4612}
4613
4614static void test_ftrace_alive(struct seq_file *m)
4615{
4616 if (!ftrace_is_dead())
4617 return;
4618 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4619 "# MAY BE MISSING FUNCTION EVENTS\n");
4620}
4621
4622#ifdef CONFIG_TRACER_MAX_TRACE
4623static void show_snapshot_main_help(struct seq_file *m)
4624{
4625 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4626 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4627 "# Takes a snapshot of the main buffer.\n"
4628 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4629 "# (Doesn't have to be '2' works with any number that\n"
4630 "# is not a '0' or '1')\n");
4631}
4632
4633static void show_snapshot_percpu_help(struct seq_file *m)
4634{
4635 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4636#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4637 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4638 "# Takes a snapshot of the main buffer for this cpu.\n");
4639#else
4640 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4641 "# Must use main snapshot file to allocate.\n");
4642#endif
4643 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4644 "# (Doesn't have to be '2' works with any number that\n"
4645 "# is not a '0' or '1')\n");
4646}
4647
4648static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4649{
4650 if (iter->tr->allocated_snapshot)
4651 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4652 else
4653 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4654
4655 seq_puts(m, "# Snapshot commands:\n");
4656 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4657 show_snapshot_main_help(m);
4658 else
4659 show_snapshot_percpu_help(m);
4660}
4661#else
4662/* Should never be called */
4663static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4664#endif
4665
4666static int s_show(struct seq_file *m, void *v)
4667{
4668 struct trace_iterator *iter = v;
4669 int ret;
4670
4671 if (iter->ent == NULL) {
4672 if (iter->tr) {
4673 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4674 seq_puts(m, "#\n");
4675 test_ftrace_alive(m);
4676 }
4677 if (iter->snapshot && trace_empty(iter))
4678 print_snapshot_help(m, iter);
4679 else if (iter->trace && iter->trace->print_header)
4680 iter->trace->print_header(m);
4681 else
4682 trace_default_header(m);
4683
4684 } else if (iter->leftover) {
4685 /*
4686 * If we filled the seq_file buffer earlier, we
4687 * want to just show it now.
4688 */
4689 ret = trace_print_seq(m, &iter->seq);
4690
4691 /* ret should this time be zero, but you never know */
4692 iter->leftover = ret;
4693
4694 } else {
4695 print_trace_line(iter);
4696 ret = trace_print_seq(m, &iter->seq);
4697 /*
4698 * If we overflow the seq_file buffer, then it will
4699 * ask us for this data again at start up.
4700 * Use that instead.
4701 * ret is 0 if seq_file write succeeded.
4702 * -1 otherwise.
4703 */
4704 iter->leftover = ret;
4705 }
4706
4707 return 0;
4708}
4709
4710/*
4711 * Should be used after trace_array_get(), trace_types_lock
4712 * ensures that i_cdev was already initialized.
4713 */
4714static inline int tracing_get_cpu(struct inode *inode)
4715{
4716 if (inode->i_cdev) /* See trace_create_cpu_file() */
4717 return (long)inode->i_cdev - 1;
4718 return RING_BUFFER_ALL_CPUS;
4719}
4720
4721static const struct seq_operations tracer_seq_ops = {
4722 .start = s_start,
4723 .next = s_next,
4724 .stop = s_stop,
4725 .show = s_show,
4726};
4727
4728static struct trace_iterator *
4729__tracing_open(struct inode *inode, struct file *file, bool snapshot)
4730{
4731 struct trace_array *tr = inode->i_private;
4732 struct trace_iterator *iter;
4733 int cpu;
4734
4735 if (tracing_disabled)
4736 return ERR_PTR(-ENODEV);
4737
4738 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4739 if (!iter)
4740 return ERR_PTR(-ENOMEM);
4741
4742 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4743 GFP_KERNEL);
4744 if (!iter->buffer_iter)
4745 goto release;
4746
4747 /*
4748 * trace_find_next_entry() may need to save off iter->ent.
4749 * It will place it into the iter->temp buffer. As most
4750 * events are less than 128, allocate a buffer of that size.
4751 * If one is greater, then trace_find_next_entry() will
4752 * allocate a new buffer to adjust for the bigger iter->ent.
4753 * It's not critical if it fails to get allocated here.
4754 */
4755 iter->temp = kmalloc(128, GFP_KERNEL);
4756 if (iter->temp)
4757 iter->temp_size = 128;
4758
4759 /*
4760 * trace_event_printf() may need to modify given format
4761 * string to replace %p with %px so that it shows real address
4762 * instead of hash value. However, that is only for the event
4763 * tracing, other tracer may not need. Defer the allocation
4764 * until it is needed.
4765 */
4766 iter->fmt = NULL;
4767 iter->fmt_size = 0;
4768
4769 /*
4770 * We make a copy of the current tracer to avoid concurrent
4771 * changes on it while we are reading.
4772 */
4773 mutex_lock(&trace_types_lock);
4774 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4775 if (!iter->trace)
4776 goto fail;
4777
4778 *iter->trace = *tr->current_trace;
4779
4780 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4781 goto fail;
4782
4783 iter->tr = tr;
4784
4785#ifdef CONFIG_TRACER_MAX_TRACE
4786 /* Currently only the top directory has a snapshot */
4787 if (tr->current_trace->print_max || snapshot)
4788 iter->array_buffer = &tr->max_buffer;
4789 else
4790#endif
4791 iter->array_buffer = &tr->array_buffer;
4792 iter->snapshot = snapshot;
4793 iter->pos = -1;
4794 iter->cpu_file = tracing_get_cpu(inode);
4795 mutex_init(&iter->mutex);
4796
4797 /* Notify the tracer early; before we stop tracing. */
4798 if (iter->trace->open)
4799 iter->trace->open(iter);
4800
4801 /* Annotate start of buffers if we had overruns */
4802 if (ring_buffer_overruns(iter->array_buffer->buffer))
4803 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4804
4805 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4806 if (trace_clocks[tr->clock_id].in_ns)
4807 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4808
4809 /*
4810 * If pause-on-trace is enabled, then stop the trace while
4811 * dumping, unless this is the "snapshot" file
4812 */
4813 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4814 tracing_stop_tr(tr);
4815
4816 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4817 for_each_tracing_cpu(cpu) {
4818 iter->buffer_iter[cpu] =
4819 ring_buffer_read_prepare(iter->array_buffer->buffer,
4820 cpu, GFP_KERNEL);
4821 }
4822 ring_buffer_read_prepare_sync();
4823 for_each_tracing_cpu(cpu) {
4824 ring_buffer_read_start(iter->buffer_iter[cpu]);
4825 tracing_iter_reset(iter, cpu);
4826 }
4827 } else {
4828 cpu = iter->cpu_file;
4829 iter->buffer_iter[cpu] =
4830 ring_buffer_read_prepare(iter->array_buffer->buffer,
4831 cpu, GFP_KERNEL);
4832 ring_buffer_read_prepare_sync();
4833 ring_buffer_read_start(iter->buffer_iter[cpu]);
4834 tracing_iter_reset(iter, cpu);
4835 }
4836
4837 mutex_unlock(&trace_types_lock);
4838
4839 return iter;
4840
4841 fail:
4842 mutex_unlock(&trace_types_lock);
4843 kfree(iter->trace);
4844 kfree(iter->temp);
4845 kfree(iter->buffer_iter);
4846release:
4847 seq_release_private(inode, file);
4848 return ERR_PTR(-ENOMEM);
4849}
4850
4851int tracing_open_generic(struct inode *inode, struct file *filp)
4852{
4853 int ret;
4854
4855 ret = tracing_check_open_get_tr(NULL);
4856 if (ret)
4857 return ret;
4858
4859 filp->private_data = inode->i_private;
4860 return 0;
4861}
4862
4863bool tracing_is_disabled(void)
4864{
4865 return (tracing_disabled) ? true: false;
4866}
4867
4868/*
4869 * Open and update trace_array ref count.
4870 * Must have the current trace_array passed to it.
4871 */
4872int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4873{
4874 struct trace_array *tr = inode->i_private;
4875 int ret;
4876
4877 ret = tracing_check_open_get_tr(tr);
4878 if (ret)
4879 return ret;
4880
4881 filp->private_data = inode->i_private;
4882
4883 return 0;
4884}
4885
4886static int tracing_mark_open(struct inode *inode, struct file *filp)
4887{
4888 stream_open(inode, filp);
4889 return tracing_open_generic_tr(inode, filp);
4890}
4891
4892static int tracing_release(struct inode *inode, struct file *file)
4893{
4894 struct trace_array *tr = inode->i_private;
4895 struct seq_file *m = file->private_data;
4896 struct trace_iterator *iter;
4897 int cpu;
4898
4899 if (!(file->f_mode & FMODE_READ)) {
4900 trace_array_put(tr);
4901 return 0;
4902 }
4903
4904 /* Writes do not use seq_file */
4905 iter = m->private;
4906 mutex_lock(&trace_types_lock);
4907
4908 for_each_tracing_cpu(cpu) {
4909 if (iter->buffer_iter[cpu])
4910 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4911 }
4912
4913 if (iter->trace && iter->trace->close)
4914 iter->trace->close(iter);
4915
4916 if (!iter->snapshot && tr->stop_count)
4917 /* reenable tracing if it was previously enabled */
4918 tracing_start_tr(tr);
4919
4920 __trace_array_put(tr);
4921
4922 mutex_unlock(&trace_types_lock);
4923
4924 mutex_destroy(&iter->mutex);
4925 free_cpumask_var(iter->started);
4926 kfree(iter->fmt);
4927 kfree(iter->temp);
4928 kfree(iter->trace);
4929 kfree(iter->buffer_iter);
4930 seq_release_private(inode, file);
4931
4932 return 0;
4933}
4934
4935static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4936{
4937 struct trace_array *tr = inode->i_private;
4938
4939 trace_array_put(tr);
4940 return 0;
4941}
4942
4943static int tracing_single_release_tr(struct inode *inode, struct file *file)
4944{
4945 struct trace_array *tr = inode->i_private;
4946
4947 trace_array_put(tr);
4948
4949 return single_release(inode, file);
4950}
4951
4952static int tracing_open(struct inode *inode, struct file *file)
4953{
4954 struct trace_array *tr = inode->i_private;
4955 struct trace_iterator *iter;
4956 int ret;
4957
4958 ret = tracing_check_open_get_tr(tr);
4959 if (ret)
4960 return ret;
4961
4962 /* If this file was open for write, then erase contents */
4963 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4964 int cpu = tracing_get_cpu(inode);
4965 struct array_buffer *trace_buf = &tr->array_buffer;
4966
4967#ifdef CONFIG_TRACER_MAX_TRACE
4968 if (tr->current_trace->print_max)
4969 trace_buf = &tr->max_buffer;
4970#endif
4971
4972 if (cpu == RING_BUFFER_ALL_CPUS)
4973 tracing_reset_online_cpus(trace_buf);
4974 else
4975 tracing_reset_cpu(trace_buf, cpu);
4976 }
4977
4978 if (file->f_mode & FMODE_READ) {
4979 iter = __tracing_open(inode, file, false);
4980 if (IS_ERR(iter))
4981 ret = PTR_ERR(iter);
4982 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4983 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4984 }
4985
4986 if (ret < 0)
4987 trace_array_put(tr);
4988
4989 return ret;
4990}
4991
4992/*
4993 * Some tracers are not suitable for instance buffers.
4994 * A tracer is always available for the global array (toplevel)
4995 * or if it explicitly states that it is.
4996 */
4997static bool
4998trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4999{
5000 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5001}
5002
5003/* Find the next tracer that this trace array may use */
5004static struct tracer *
5005get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5006{
5007 while (t && !trace_ok_for_array(t, tr))
5008 t = t->next;
5009
5010 return t;
5011}
5012
5013static void *
5014t_next(struct seq_file *m, void *v, loff_t *pos)
5015{
5016 struct trace_array *tr = m->private;
5017 struct tracer *t = v;
5018
5019 (*pos)++;
5020
5021 if (t)
5022 t = get_tracer_for_array(tr, t->next);
5023
5024 return t;
5025}
5026
5027static void *t_start(struct seq_file *m, loff_t *pos)
5028{
5029 struct trace_array *tr = m->private;
5030 struct tracer *t;
5031 loff_t l = 0;
5032
5033 mutex_lock(&trace_types_lock);
5034
5035 t = get_tracer_for_array(tr, trace_types);
5036 for (; t && l < *pos; t = t_next(m, t, &l))
5037 ;
5038
5039 return t;
5040}
5041
5042static void t_stop(struct seq_file *m, void *p)
5043{
5044 mutex_unlock(&trace_types_lock);
5045}
5046
5047static int t_show(struct seq_file *m, void *v)
5048{
5049 struct tracer *t = v;
5050
5051 if (!t)
5052 return 0;
5053
5054 seq_puts(m, t->name);
5055 if (t->next)
5056 seq_putc(m, ' ');
5057 else
5058 seq_putc(m, '\n');
5059
5060 return 0;
5061}
5062
5063static const struct seq_operations show_traces_seq_ops = {
5064 .start = t_start,
5065 .next = t_next,
5066 .stop = t_stop,
5067 .show = t_show,
5068};
5069
5070static int show_traces_open(struct inode *inode, struct file *file)
5071{
5072 struct trace_array *tr = inode->i_private;
5073 struct seq_file *m;
5074 int ret;
5075
5076 ret = tracing_check_open_get_tr(tr);
5077 if (ret)
5078 return ret;
5079
5080 ret = seq_open(file, &show_traces_seq_ops);
5081 if (ret) {
5082 trace_array_put(tr);
5083 return ret;
5084 }
5085
5086 m = file->private_data;
5087 m->private = tr;
5088
5089 return 0;
5090}
5091
5092static int show_traces_release(struct inode *inode, struct file *file)
5093{
5094 struct trace_array *tr = inode->i_private;
5095
5096 trace_array_put(tr);
5097 return seq_release(inode, file);
5098}
5099
5100static ssize_t
5101tracing_write_stub(struct file *filp, const char __user *ubuf,
5102 size_t count, loff_t *ppos)
5103{
5104 return count;
5105}
5106
5107loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5108{
5109 int ret;
5110
5111 if (file->f_mode & FMODE_READ)
5112 ret = seq_lseek(file, offset, whence);
5113 else
5114 file->f_pos = ret = 0;
5115
5116 return ret;
5117}
5118
5119static const struct file_operations tracing_fops = {
5120 .open = tracing_open,
5121 .read = seq_read,
5122 .write = tracing_write_stub,
5123 .llseek = tracing_lseek,
5124 .release = tracing_release,
5125};
5126
5127static const struct file_operations show_traces_fops = {
5128 .open = show_traces_open,
5129 .read = seq_read,
5130 .llseek = seq_lseek,
5131 .release = show_traces_release,
5132};
5133
5134static ssize_t
5135tracing_cpumask_read(struct file *filp, char __user *ubuf,
5136 size_t count, loff_t *ppos)
5137{
5138 struct trace_array *tr = file_inode(filp)->i_private;
5139 char *mask_str;
5140 int len;
5141
5142 len = snprintf(NULL, 0, "%*pb\n",
5143 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5144 mask_str = kmalloc(len, GFP_KERNEL);
5145 if (!mask_str)
5146 return -ENOMEM;
5147
5148 len = snprintf(mask_str, len, "%*pb\n",
5149 cpumask_pr_args(tr->tracing_cpumask));
5150 if (len >= count) {
5151 count = -EINVAL;
5152 goto out_err;
5153 }
5154 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5155
5156out_err:
5157 kfree(mask_str);
5158
5159 return count;
5160}
5161
5162int tracing_set_cpumask(struct trace_array *tr,
5163 cpumask_var_t tracing_cpumask_new)
5164{
5165 int cpu;
5166
5167 if (!tr)
5168 return -EINVAL;
5169
5170 local_irq_disable();
5171 arch_spin_lock(&tr->max_lock);
5172 for_each_tracing_cpu(cpu) {
5173 /*
5174 * Increase/decrease the disabled counter if we are
5175 * about to flip a bit in the cpumask:
5176 */
5177 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5178 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5179 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5180 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5181 }
5182 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5183 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5184 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5185 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5186 }
5187 }
5188 arch_spin_unlock(&tr->max_lock);
5189 local_irq_enable();
5190
5191 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5192
5193 return 0;
5194}
5195
5196static ssize_t
5197tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5198 size_t count, loff_t *ppos)
5199{
5200 struct trace_array *tr = file_inode(filp)->i_private;
5201 cpumask_var_t tracing_cpumask_new;
5202 int err;
5203
5204 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5205 return -ENOMEM;
5206
5207 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5208 if (err)
5209 goto err_free;
5210
5211 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5212 if (err)
5213 goto err_free;
5214
5215 free_cpumask_var(tracing_cpumask_new);
5216
5217 return count;
5218
5219err_free:
5220 free_cpumask_var(tracing_cpumask_new);
5221
5222 return err;
5223}
5224
5225static const struct file_operations tracing_cpumask_fops = {
5226 .open = tracing_open_generic_tr,
5227 .read = tracing_cpumask_read,
5228 .write = tracing_cpumask_write,
5229 .release = tracing_release_generic_tr,
5230 .llseek = generic_file_llseek,
5231};
5232
5233static int tracing_trace_options_show(struct seq_file *m, void *v)
5234{
5235 struct tracer_opt *trace_opts;
5236 struct trace_array *tr = m->private;
5237 u32 tracer_flags;
5238 int i;
5239
5240 mutex_lock(&trace_types_lock);
5241 tracer_flags = tr->current_trace->flags->val;
5242 trace_opts = tr->current_trace->flags->opts;
5243
5244 for (i = 0; trace_options[i]; i++) {
5245 if (tr->trace_flags & (1 << i))
5246 seq_printf(m, "%s\n", trace_options[i]);
5247 else
5248 seq_printf(m, "no%s\n", trace_options[i]);
5249 }
5250
5251 for (i = 0; trace_opts[i].name; i++) {
5252 if (tracer_flags & trace_opts[i].bit)
5253 seq_printf(m, "%s\n", trace_opts[i].name);
5254 else
5255 seq_printf(m, "no%s\n", trace_opts[i].name);
5256 }
5257 mutex_unlock(&trace_types_lock);
5258
5259 return 0;
5260}
5261
5262static int __set_tracer_option(struct trace_array *tr,
5263 struct tracer_flags *tracer_flags,
5264 struct tracer_opt *opts, int neg)
5265{
5266 struct tracer *trace = tracer_flags->trace;
5267 int ret;
5268
5269 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5270 if (ret)
5271 return ret;
5272
5273 if (neg)
5274 tracer_flags->val &= ~opts->bit;
5275 else
5276 tracer_flags->val |= opts->bit;
5277 return 0;
5278}
5279
5280/* Try to assign a tracer specific option */
5281static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5282{
5283 struct tracer *trace = tr->current_trace;
5284 struct tracer_flags *tracer_flags = trace->flags;
5285 struct tracer_opt *opts = NULL;
5286 int i;
5287
5288 for (i = 0; tracer_flags->opts[i].name; i++) {
5289 opts = &tracer_flags->opts[i];
5290
5291 if (strcmp(cmp, opts->name) == 0)
5292 return __set_tracer_option(tr, trace->flags, opts, neg);
5293 }
5294
5295 return -EINVAL;
5296}
5297
5298/* Some tracers require overwrite to stay enabled */
5299int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5300{
5301 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5302 return -1;
5303
5304 return 0;
5305}
5306
5307int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5308{
5309 int *map;
5310
5311 if ((mask == TRACE_ITER_RECORD_TGID) ||
5312 (mask == TRACE_ITER_RECORD_CMD))
5313 lockdep_assert_held(&event_mutex);
5314
5315 /* do nothing if flag is already set */
5316 if (!!(tr->trace_flags & mask) == !!enabled)
5317 return 0;
5318
5319 /* Give the tracer a chance to approve the change */
5320 if (tr->current_trace->flag_changed)
5321 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5322 return -EINVAL;
5323
5324 if (enabled)
5325 tr->trace_flags |= mask;
5326 else
5327 tr->trace_flags &= ~mask;
5328
5329 if (mask == TRACE_ITER_RECORD_CMD)
5330 trace_event_enable_cmd_record(enabled);
5331
5332 if (mask == TRACE_ITER_RECORD_TGID) {
5333 if (!tgid_map) {
5334 tgid_map_max = pid_max;
5335 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5336 GFP_KERNEL);
5337
5338 /*
5339 * Pairs with smp_load_acquire() in
5340 * trace_find_tgid_ptr() to ensure that if it observes
5341 * the tgid_map we just allocated then it also observes
5342 * the corresponding tgid_map_max value.
5343 */
5344 smp_store_release(&tgid_map, map);
5345 }
5346 if (!tgid_map) {
5347 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5348 return -ENOMEM;
5349 }
5350
5351 trace_event_enable_tgid_record(enabled);
5352 }
5353
5354 if (mask == TRACE_ITER_EVENT_FORK)
5355 trace_event_follow_fork(tr, enabled);
5356
5357 if (mask == TRACE_ITER_FUNC_FORK)
5358 ftrace_pid_follow_fork(tr, enabled);
5359
5360 if (mask == TRACE_ITER_OVERWRITE) {
5361 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5362#ifdef CONFIG_TRACER_MAX_TRACE
5363 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5364#endif
5365 }
5366
5367 if (mask == TRACE_ITER_PRINTK) {
5368 trace_printk_start_stop_comm(enabled);
5369 trace_printk_control(enabled);
5370 }
5371
5372 return 0;
5373}
5374
5375int trace_set_options(struct trace_array *tr, char *option)
5376{
5377 char *cmp;
5378 int neg = 0;
5379 int ret;
5380 size_t orig_len = strlen(option);
5381 int len;
5382
5383 cmp = strstrip(option);
5384
5385 len = str_has_prefix(cmp, "no");
5386 if (len)
5387 neg = 1;
5388
5389 cmp += len;
5390
5391 mutex_lock(&event_mutex);
5392 mutex_lock(&trace_types_lock);
5393
5394 ret = match_string(trace_options, -1, cmp);
5395 /* If no option could be set, test the specific tracer options */
5396 if (ret < 0)
5397 ret = set_tracer_option(tr, cmp, neg);
5398 else
5399 ret = set_tracer_flag(tr, 1 << ret, !neg);
5400
5401 mutex_unlock(&trace_types_lock);
5402 mutex_unlock(&event_mutex);
5403
5404 /*
5405 * If the first trailing whitespace is replaced with '\0' by strstrip,
5406 * turn it back into a space.
5407 */
5408 if (orig_len > strlen(option))
5409 option[strlen(option)] = ' ';
5410
5411 return ret;
5412}
5413
5414static void __init apply_trace_boot_options(void)
5415{
5416 char *buf = trace_boot_options_buf;
5417 char *option;
5418
5419 while (true) {
5420 option = strsep(&buf, ",");
5421
5422 if (!option)
5423 break;
5424
5425 if (*option)
5426 trace_set_options(&global_trace, option);
5427
5428 /* Put back the comma to allow this to be called again */
5429 if (buf)
5430 *(buf - 1) = ',';
5431 }
5432}
5433
5434static ssize_t
5435tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5436 size_t cnt, loff_t *ppos)
5437{
5438 struct seq_file *m = filp->private_data;
5439 struct trace_array *tr = m->private;
5440 char buf[64];
5441 int ret;
5442
5443 if (cnt >= sizeof(buf))
5444 return -EINVAL;
5445
5446 if (copy_from_user(buf, ubuf, cnt))
5447 return -EFAULT;
5448
5449 buf[cnt] = 0;
5450
5451 ret = trace_set_options(tr, buf);
5452 if (ret < 0)
5453 return ret;
5454
5455 *ppos += cnt;
5456
5457 return cnt;
5458}
5459
5460static int tracing_trace_options_open(struct inode *inode, struct file *file)
5461{
5462 struct trace_array *tr = inode->i_private;
5463 int ret;
5464
5465 ret = tracing_check_open_get_tr(tr);
5466 if (ret)
5467 return ret;
5468
5469 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5470 if (ret < 0)
5471 trace_array_put(tr);
5472
5473 return ret;
5474}
5475
5476static const struct file_operations tracing_iter_fops = {
5477 .open = tracing_trace_options_open,
5478 .read = seq_read,
5479 .llseek = seq_lseek,
5480 .release = tracing_single_release_tr,
5481 .write = tracing_trace_options_write,
5482};
5483
5484static const char readme_msg[] =
5485 "tracing mini-HOWTO:\n\n"
5486 "# echo 0 > tracing_on : quick way to disable tracing\n"
5487 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5488 " Important files:\n"
5489 " trace\t\t\t- The static contents of the buffer\n"
5490 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5491 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5492 " current_tracer\t- function and latency tracers\n"
5493 " available_tracers\t- list of configured tracers for current_tracer\n"
5494 " error_log\t- error log for failed commands (that support it)\n"
5495 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5496 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5497 " trace_clock\t\t- change the clock used to order events\n"
5498 " local: Per cpu clock but may not be synced across CPUs\n"
5499 " global: Synced across CPUs but slows tracing down.\n"
5500 " counter: Not a clock, but just an increment\n"
5501 " uptime: Jiffy counter from time of boot\n"
5502 " perf: Same clock that perf events use\n"
5503#ifdef CONFIG_X86_64
5504 " x86-tsc: TSC cycle counter\n"
5505#endif
5506 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5507 " delta: Delta difference against a buffer-wide timestamp\n"
5508 " absolute: Absolute (standalone) timestamp\n"
5509 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5510 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5511 " tracing_cpumask\t- Limit which CPUs to trace\n"
5512 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5513 "\t\t\t Remove sub-buffer with rmdir\n"
5514 " trace_options\t\t- Set format or modify how tracing happens\n"
5515 "\t\t\t Disable an option by prefixing 'no' to the\n"
5516 "\t\t\t option name\n"
5517 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5518#ifdef CONFIG_DYNAMIC_FTRACE
5519 "\n available_filter_functions - list of functions that can be filtered on\n"
5520 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5521 "\t\t\t functions\n"
5522 "\t accepts: func_full_name or glob-matching-pattern\n"
5523 "\t modules: Can select a group via module\n"
5524 "\t Format: :mod:<module-name>\n"
5525 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5526 "\t triggers: a command to perform when function is hit\n"
5527 "\t Format: <function>:<trigger>[:count]\n"
5528 "\t trigger: traceon, traceoff\n"
5529 "\t\t enable_event:<system>:<event>\n"
5530 "\t\t disable_event:<system>:<event>\n"
5531#ifdef CONFIG_STACKTRACE
5532 "\t\t stacktrace\n"
5533#endif
5534#ifdef CONFIG_TRACER_SNAPSHOT
5535 "\t\t snapshot\n"
5536#endif
5537 "\t\t dump\n"
5538 "\t\t cpudump\n"
5539 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5540 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5541 "\t The first one will disable tracing every time do_fault is hit\n"
5542 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5543 "\t The first time do trap is hit and it disables tracing, the\n"
5544 "\t counter will decrement to 2. If tracing is already disabled,\n"
5545 "\t the counter will not decrement. It only decrements when the\n"
5546 "\t trigger did work\n"
5547 "\t To remove trigger without count:\n"
5548 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5549 "\t To remove trigger with a count:\n"
5550 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5551 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5552 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5553 "\t modules: Can select a group via module command :mod:\n"
5554 "\t Does not accept triggers\n"
5555#endif /* CONFIG_DYNAMIC_FTRACE */
5556#ifdef CONFIG_FUNCTION_TRACER
5557 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5558 "\t\t (function)\n"
5559 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5560 "\t\t (function)\n"
5561#endif
5562#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5563 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5564 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5565 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5566#endif
5567#ifdef CONFIG_TRACER_SNAPSHOT
5568 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5569 "\t\t\t snapshot buffer. Read the contents for more\n"
5570 "\t\t\t information\n"
5571#endif
5572#ifdef CONFIG_STACK_TRACER
5573 " stack_trace\t\t- Shows the max stack trace when active\n"
5574 " stack_max_size\t- Shows current max stack size that was traced\n"
5575 "\t\t\t Write into this file to reset the max size (trigger a\n"
5576 "\t\t\t new trace)\n"
5577#ifdef CONFIG_DYNAMIC_FTRACE
5578 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5579 "\t\t\t traces\n"
5580#endif
5581#endif /* CONFIG_STACK_TRACER */
5582#ifdef CONFIG_DYNAMIC_EVENTS
5583 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5584 "\t\t\t Write into this file to define/undefine new trace events.\n"
5585#endif
5586#ifdef CONFIG_KPROBE_EVENTS
5587 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5588 "\t\t\t Write into this file to define/undefine new trace events.\n"
5589#endif
5590#ifdef CONFIG_UPROBE_EVENTS
5591 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5592 "\t\t\t Write into this file to define/undefine new trace events.\n"
5593#endif
5594#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5595 "\t accepts: event-definitions (one definition per line)\n"
5596 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5597 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5598#ifdef CONFIG_HIST_TRIGGERS
5599 "\t s:[synthetic/]<event> <field> [<field>]\n"
5600#endif
5601 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>]\n"
5602 "\t -:[<group>/][<event>]\n"
5603#ifdef CONFIG_KPROBE_EVENTS
5604 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5605 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5606#endif
5607#ifdef CONFIG_UPROBE_EVENTS
5608 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5609#endif
5610 "\t args: <name>=fetcharg[:type]\n"
5611 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5612#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5613 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5614#else
5615 "\t $stack<index>, $stack, $retval, $comm,\n"
5616#endif
5617 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5618 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string, symbol,\n"
5619 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5620 "\t symstr, <type>\\[<array-size>\\]\n"
5621#ifdef CONFIG_HIST_TRIGGERS
5622 "\t field: <stype> <name>;\n"
5623 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5624 "\t [unsigned] char/int/long\n"
5625#endif
5626 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5627 "\t of the <attached-group>/<attached-event>.\n"
5628#endif
5629 " events/\t\t- Directory containing all trace event subsystems:\n"
5630 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5631 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5632 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5633 "\t\t\t events\n"
5634 " filter\t\t- If set, only events passing filter are traced\n"
5635 " events/<system>/<event>/\t- Directory containing control files for\n"
5636 "\t\t\t <event>:\n"
5637 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5638 " filter\t\t- If set, only events passing filter are traced\n"
5639 " trigger\t\t- If set, a command to perform when event is hit\n"
5640 "\t Format: <trigger>[:count][if <filter>]\n"
5641 "\t trigger: traceon, traceoff\n"
5642 "\t enable_event:<system>:<event>\n"
5643 "\t disable_event:<system>:<event>\n"
5644#ifdef CONFIG_HIST_TRIGGERS
5645 "\t enable_hist:<system>:<event>\n"
5646 "\t disable_hist:<system>:<event>\n"
5647#endif
5648#ifdef CONFIG_STACKTRACE
5649 "\t\t stacktrace\n"
5650#endif
5651#ifdef CONFIG_TRACER_SNAPSHOT
5652 "\t\t snapshot\n"
5653#endif
5654#ifdef CONFIG_HIST_TRIGGERS
5655 "\t\t hist (see below)\n"
5656#endif
5657 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5658 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5659 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5660 "\t events/block/block_unplug/trigger\n"
5661 "\t The first disables tracing every time block_unplug is hit.\n"
5662 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5663 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5664 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5665 "\t Like function triggers, the counter is only decremented if it\n"
5666 "\t enabled or disabled tracing.\n"
5667 "\t To remove a trigger without a count:\n"
5668 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5669 "\t To remove a trigger with a count:\n"
5670 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5671 "\t Filters can be ignored when removing a trigger.\n"
5672#ifdef CONFIG_HIST_TRIGGERS
5673 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5674 "\t Format: hist:keys=<field1[,field2,...]>\n"
5675 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5676 "\t [:values=<field1[,field2,...]>]\n"
5677 "\t [:sort=<field1[,field2,...]>]\n"
5678 "\t [:size=#entries]\n"
5679 "\t [:pause][:continue][:clear]\n"
5680 "\t [:name=histname1]\n"
5681 "\t [:nohitcount]\n"
5682 "\t [:<handler>.<action>]\n"
5683 "\t [if <filter>]\n\n"
5684 "\t Note, special fields can be used as well:\n"
5685 "\t common_timestamp - to record current timestamp\n"
5686 "\t common_cpu - to record the CPU the event happened on\n"
5687 "\n"
5688 "\t A hist trigger variable can be:\n"
5689 "\t - a reference to a field e.g. x=current_timestamp,\n"
5690 "\t - a reference to another variable e.g. y=$x,\n"
5691 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5692 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5693 "\n"
5694 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5695 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5696 "\t variable reference, field or numeric literal.\n"
5697 "\n"
5698 "\t When a matching event is hit, an entry is added to a hash\n"
5699 "\t table using the key(s) and value(s) named, and the value of a\n"
5700 "\t sum called 'hitcount' is incremented. Keys and values\n"
5701 "\t correspond to fields in the event's format description. Keys\n"
5702 "\t can be any field, or the special string 'stacktrace'.\n"
5703 "\t Compound keys consisting of up to two fields can be specified\n"
5704 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5705 "\t fields. Sort keys consisting of up to two fields can be\n"
5706 "\t specified using the 'sort' keyword. The sort direction can\n"
5707 "\t be modified by appending '.descending' or '.ascending' to a\n"
5708 "\t sort field. The 'size' parameter can be used to specify more\n"
5709 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5710 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5711 "\t its histogram data will be shared with other triggers of the\n"
5712 "\t same name, and trigger hits will update this common data.\n\n"
5713 "\t Reading the 'hist' file for the event will dump the hash\n"
5714 "\t table in its entirety to stdout. If there are multiple hist\n"
5715 "\t triggers attached to an event, there will be a table for each\n"
5716 "\t trigger in the output. The table displayed for a named\n"
5717 "\t trigger will be the same as any other instance having the\n"
5718 "\t same name. The default format used to display a given field\n"
5719 "\t can be modified by appending any of the following modifiers\n"
5720 "\t to the field name, as applicable:\n\n"
5721 "\t .hex display a number as a hex value\n"
5722 "\t .sym display an address as a symbol\n"
5723 "\t .sym-offset display an address as a symbol and offset\n"
5724 "\t .execname display a common_pid as a program name\n"
5725 "\t .syscall display a syscall id as a syscall name\n"
5726 "\t .log2 display log2 value rather than raw number\n"
5727 "\t .buckets=size display values in groups of size rather than raw number\n"
5728 "\t .usecs display a common_timestamp in microseconds\n"
5729 "\t .percent display a number of percentage value\n"
5730 "\t .graph display a bar-graph of a value\n\n"
5731 "\t The 'pause' parameter can be used to pause an existing hist\n"
5732 "\t trigger or to start a hist trigger but not log any events\n"
5733 "\t until told to do so. 'continue' can be used to start or\n"
5734 "\t restart a paused hist trigger.\n\n"
5735 "\t The 'clear' parameter will clear the contents of a running\n"
5736 "\t hist trigger and leave its current paused/active state\n"
5737 "\t unchanged.\n\n"
5738 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5739 "\t raw hitcount in the histogram.\n\n"
5740 "\t The enable_hist and disable_hist triggers can be used to\n"
5741 "\t have one event conditionally start and stop another event's\n"
5742 "\t already-attached hist trigger. The syntax is analogous to\n"
5743 "\t the enable_event and disable_event triggers.\n\n"
5744 "\t Hist trigger handlers and actions are executed whenever a\n"
5745 "\t a histogram entry is added or updated. They take the form:\n\n"
5746 "\t <handler>.<action>\n\n"
5747 "\t The available handlers are:\n\n"
5748 "\t onmatch(matching.event) - invoke on addition or update\n"
5749 "\t onmax(var) - invoke if var exceeds current max\n"
5750 "\t onchange(var) - invoke action if var changes\n\n"
5751 "\t The available actions are:\n\n"
5752 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5753 "\t save(field,...) - save current event fields\n"
5754#ifdef CONFIG_TRACER_SNAPSHOT
5755 "\t snapshot() - snapshot the trace buffer\n\n"
5756#endif
5757#ifdef CONFIG_SYNTH_EVENTS
5758 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5759 "\t Write into this file to define/undefine new synthetic events.\n"
5760 "\t example: echo 'myevent u64 lat; char name[]' >> synthetic_events\n"
5761#endif
5762#endif
5763;
5764
5765static ssize_t
5766tracing_readme_read(struct file *filp, char __user *ubuf,
5767 size_t cnt, loff_t *ppos)
5768{
5769 return simple_read_from_buffer(ubuf, cnt, ppos,
5770 readme_msg, strlen(readme_msg));
5771}
5772
5773static const struct file_operations tracing_readme_fops = {
5774 .open = tracing_open_generic,
5775 .read = tracing_readme_read,
5776 .llseek = generic_file_llseek,
5777};
5778
5779static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5780{
5781 int pid = ++(*pos);
5782
5783 return trace_find_tgid_ptr(pid);
5784}
5785
5786static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5787{
5788 int pid = *pos;
5789
5790 return trace_find_tgid_ptr(pid);
5791}
5792
5793static void saved_tgids_stop(struct seq_file *m, void *v)
5794{
5795}
5796
5797static int saved_tgids_show(struct seq_file *m, void *v)
5798{
5799 int *entry = (int *)v;
5800 int pid = entry - tgid_map;
5801 int tgid = *entry;
5802
5803 if (tgid == 0)
5804 return SEQ_SKIP;
5805
5806 seq_printf(m, "%d %d\n", pid, tgid);
5807 return 0;
5808}
5809
5810static const struct seq_operations tracing_saved_tgids_seq_ops = {
5811 .start = saved_tgids_start,
5812 .stop = saved_tgids_stop,
5813 .next = saved_tgids_next,
5814 .show = saved_tgids_show,
5815};
5816
5817static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5818{
5819 int ret;
5820
5821 ret = tracing_check_open_get_tr(NULL);
5822 if (ret)
5823 return ret;
5824
5825 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5826}
5827
5828
5829static const struct file_operations tracing_saved_tgids_fops = {
5830 .open = tracing_saved_tgids_open,
5831 .read = seq_read,
5832 .llseek = seq_lseek,
5833 .release = seq_release,
5834};
5835
5836static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5837{
5838 unsigned int *ptr = v;
5839
5840 if (*pos || m->count)
5841 ptr++;
5842
5843 (*pos)++;
5844
5845 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5846 ptr++) {
5847 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5848 continue;
5849
5850 return ptr;
5851 }
5852
5853 return NULL;
5854}
5855
5856static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5857{
5858 void *v;
5859 loff_t l = 0;
5860
5861 preempt_disable();
5862 arch_spin_lock(&trace_cmdline_lock);
5863
5864 v = &savedcmd->map_cmdline_to_pid[0];
5865 while (l <= *pos) {
5866 v = saved_cmdlines_next(m, v, &l);
5867 if (!v)
5868 return NULL;
5869 }
5870
5871 return v;
5872}
5873
5874static void saved_cmdlines_stop(struct seq_file *m, void *v)
5875{
5876 arch_spin_unlock(&trace_cmdline_lock);
5877 preempt_enable();
5878}
5879
5880static int saved_cmdlines_show(struct seq_file *m, void *v)
5881{
5882 char buf[TASK_COMM_LEN];
5883 unsigned int *pid = v;
5884
5885 __trace_find_cmdline(*pid, buf);
5886 seq_printf(m, "%d %s\n", *pid, buf);
5887 return 0;
5888}
5889
5890static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5891 .start = saved_cmdlines_start,
5892 .next = saved_cmdlines_next,
5893 .stop = saved_cmdlines_stop,
5894 .show = saved_cmdlines_show,
5895};
5896
5897static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5898{
5899 int ret;
5900
5901 ret = tracing_check_open_get_tr(NULL);
5902 if (ret)
5903 return ret;
5904
5905 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5906}
5907
5908static const struct file_operations tracing_saved_cmdlines_fops = {
5909 .open = tracing_saved_cmdlines_open,
5910 .read = seq_read,
5911 .llseek = seq_lseek,
5912 .release = seq_release,
5913};
5914
5915static ssize_t
5916tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5917 size_t cnt, loff_t *ppos)
5918{
5919 char buf[64];
5920 int r;
5921
5922 preempt_disable();
5923 arch_spin_lock(&trace_cmdline_lock);
5924 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5925 arch_spin_unlock(&trace_cmdline_lock);
5926 preempt_enable();
5927
5928 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5929}
5930
5931static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5932{
5933 kfree(s->saved_cmdlines);
5934 kfree(s->map_cmdline_to_pid);
5935 kfree(s);
5936}
5937
5938static int tracing_resize_saved_cmdlines(unsigned int val)
5939{
5940 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5941
5942 s = kmalloc(sizeof(*s), GFP_KERNEL);
5943 if (!s)
5944 return -ENOMEM;
5945
5946 if (allocate_cmdlines_buffer(val, s) < 0) {
5947 kfree(s);
5948 return -ENOMEM;
5949 }
5950
5951 preempt_disable();
5952 arch_spin_lock(&trace_cmdline_lock);
5953 savedcmd_temp = savedcmd;
5954 savedcmd = s;
5955 arch_spin_unlock(&trace_cmdline_lock);
5956 preempt_enable();
5957 free_saved_cmdlines_buffer(savedcmd_temp);
5958
5959 return 0;
5960}
5961
5962static ssize_t
5963tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
5964 size_t cnt, loff_t *ppos)
5965{
5966 unsigned long val;
5967 int ret;
5968
5969 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5970 if (ret)
5971 return ret;
5972
5973 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
5974 if (!val || val > PID_MAX_DEFAULT)
5975 return -EINVAL;
5976
5977 ret = tracing_resize_saved_cmdlines((unsigned int)val);
5978 if (ret < 0)
5979 return ret;
5980
5981 *ppos += cnt;
5982
5983 return cnt;
5984}
5985
5986static const struct file_operations tracing_saved_cmdlines_size_fops = {
5987 .open = tracing_open_generic,
5988 .read = tracing_saved_cmdlines_size_read,
5989 .write = tracing_saved_cmdlines_size_write,
5990};
5991
5992#ifdef CONFIG_TRACE_EVAL_MAP_FILE
5993static union trace_eval_map_item *
5994update_eval_map(union trace_eval_map_item *ptr)
5995{
5996 if (!ptr->map.eval_string) {
5997 if (ptr->tail.next) {
5998 ptr = ptr->tail.next;
5999 /* Set ptr to the next real item (skip head) */
6000 ptr++;
6001 } else
6002 return NULL;
6003 }
6004 return ptr;
6005}
6006
6007static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6008{
6009 union trace_eval_map_item *ptr = v;
6010
6011 /*
6012 * Paranoid! If ptr points to end, we don't want to increment past it.
6013 * This really should never happen.
6014 */
6015 (*pos)++;
6016 ptr = update_eval_map(ptr);
6017 if (WARN_ON_ONCE(!ptr))
6018 return NULL;
6019
6020 ptr++;
6021 ptr = update_eval_map(ptr);
6022
6023 return ptr;
6024}
6025
6026static void *eval_map_start(struct seq_file *m, loff_t *pos)
6027{
6028 union trace_eval_map_item *v;
6029 loff_t l = 0;
6030
6031 mutex_lock(&trace_eval_mutex);
6032
6033 v = trace_eval_maps;
6034 if (v)
6035 v++;
6036
6037 while (v && l < *pos) {
6038 v = eval_map_next(m, v, &l);
6039 }
6040
6041 return v;
6042}
6043
6044static void eval_map_stop(struct seq_file *m, void *v)
6045{
6046 mutex_unlock(&trace_eval_mutex);
6047}
6048
6049static int eval_map_show(struct seq_file *m, void *v)
6050{
6051 union trace_eval_map_item *ptr = v;
6052
6053 seq_printf(m, "%s %ld (%s)\n",
6054 ptr->map.eval_string, ptr->map.eval_value,
6055 ptr->map.system);
6056
6057 return 0;
6058}
6059
6060static const struct seq_operations tracing_eval_map_seq_ops = {
6061 .start = eval_map_start,
6062 .next = eval_map_next,
6063 .stop = eval_map_stop,
6064 .show = eval_map_show,
6065};
6066
6067static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6068{
6069 int ret;
6070
6071 ret = tracing_check_open_get_tr(NULL);
6072 if (ret)
6073 return ret;
6074
6075 return seq_open(filp, &tracing_eval_map_seq_ops);
6076}
6077
6078static const struct file_operations tracing_eval_map_fops = {
6079 .open = tracing_eval_map_open,
6080 .read = seq_read,
6081 .llseek = seq_lseek,
6082 .release = seq_release,
6083};
6084
6085static inline union trace_eval_map_item *
6086trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6087{
6088 /* Return tail of array given the head */
6089 return ptr + ptr->head.length + 1;
6090}
6091
6092static void
6093trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6094 int len)
6095{
6096 struct trace_eval_map **stop;
6097 struct trace_eval_map **map;
6098 union trace_eval_map_item *map_array;
6099 union trace_eval_map_item *ptr;
6100
6101 stop = start + len;
6102
6103 /*
6104 * The trace_eval_maps contains the map plus a head and tail item,
6105 * where the head holds the module and length of array, and the
6106 * tail holds a pointer to the next list.
6107 */
6108 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6109 if (!map_array) {
6110 pr_warn("Unable to allocate trace eval mapping\n");
6111 return;
6112 }
6113
6114 mutex_lock(&trace_eval_mutex);
6115
6116 if (!trace_eval_maps)
6117 trace_eval_maps = map_array;
6118 else {
6119 ptr = trace_eval_maps;
6120 for (;;) {
6121 ptr = trace_eval_jmp_to_tail(ptr);
6122 if (!ptr->tail.next)
6123 break;
6124 ptr = ptr->tail.next;
6125
6126 }
6127 ptr->tail.next = map_array;
6128 }
6129 map_array->head.mod = mod;
6130 map_array->head.length = len;
6131 map_array++;
6132
6133 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6134 map_array->map = **map;
6135 map_array++;
6136 }
6137 memset(map_array, 0, sizeof(*map_array));
6138
6139 mutex_unlock(&trace_eval_mutex);
6140}
6141
6142static void trace_create_eval_file(struct dentry *d_tracer)
6143{
6144 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6145 NULL, &tracing_eval_map_fops);
6146}
6147
6148#else /* CONFIG_TRACE_EVAL_MAP_FILE */
6149static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6150static inline void trace_insert_eval_map_file(struct module *mod,
6151 struct trace_eval_map **start, int len) { }
6152#endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6153
6154static void trace_insert_eval_map(struct module *mod,
6155 struct trace_eval_map **start, int len)
6156{
6157 struct trace_eval_map **map;
6158
6159 if (len <= 0)
6160 return;
6161
6162 map = start;
6163
6164 trace_event_eval_update(map, len);
6165
6166 trace_insert_eval_map_file(mod, start, len);
6167}
6168
6169static ssize_t
6170tracing_set_trace_read(struct file *filp, char __user *ubuf,
6171 size_t cnt, loff_t *ppos)
6172{
6173 struct trace_array *tr = filp->private_data;
6174 char buf[MAX_TRACER_SIZE+2];
6175 int r;
6176
6177 mutex_lock(&trace_types_lock);
6178 r = sprintf(buf, "%s\n", tr->current_trace->name);
6179 mutex_unlock(&trace_types_lock);
6180
6181 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6182}
6183
6184int tracer_init(struct tracer *t, struct trace_array *tr)
6185{
6186 tracing_reset_online_cpus(&tr->array_buffer);
6187 return t->init(tr);
6188}
6189
6190static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6191{
6192 int cpu;
6193
6194 for_each_tracing_cpu(cpu)
6195 per_cpu_ptr(buf->data, cpu)->entries = val;
6196}
6197
6198#ifdef CONFIG_TRACER_MAX_TRACE
6199/* resize @tr's buffer to the size of @size_tr's entries */
6200static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6201 struct array_buffer *size_buf, int cpu_id)
6202{
6203 int cpu, ret = 0;
6204
6205 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6206 for_each_tracing_cpu(cpu) {
6207 ret = ring_buffer_resize(trace_buf->buffer,
6208 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6209 if (ret < 0)
6210 break;
6211 per_cpu_ptr(trace_buf->data, cpu)->entries =
6212 per_cpu_ptr(size_buf->data, cpu)->entries;
6213 }
6214 } else {
6215 ret = ring_buffer_resize(trace_buf->buffer,
6216 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6217 if (ret == 0)
6218 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6219 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6220 }
6221
6222 return ret;
6223}
6224#endif /* CONFIG_TRACER_MAX_TRACE */
6225
6226static int __tracing_resize_ring_buffer(struct trace_array *tr,
6227 unsigned long size, int cpu)
6228{
6229 int ret;
6230
6231 /*
6232 * If kernel or user changes the size of the ring buffer
6233 * we use the size that was given, and we can forget about
6234 * expanding it later.
6235 */
6236 ring_buffer_expanded = true;
6237
6238 /* May be called before buffers are initialized */
6239 if (!tr->array_buffer.buffer)
6240 return 0;
6241
6242 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6243 if (ret < 0)
6244 return ret;
6245
6246#ifdef CONFIG_TRACER_MAX_TRACE
6247 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
6248 !tr->current_trace->use_max_tr)
6249 goto out;
6250
6251 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6252 if (ret < 0) {
6253 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6254 &tr->array_buffer, cpu);
6255 if (r < 0) {
6256 /*
6257 * AARGH! We are left with different
6258 * size max buffer!!!!
6259 * The max buffer is our "snapshot" buffer.
6260 * When a tracer needs a snapshot (one of the
6261 * latency tracers), it swaps the max buffer
6262 * with the saved snap shot. We succeeded to
6263 * update the size of the main buffer, but failed to
6264 * update the size of the max buffer. But when we tried
6265 * to reset the main buffer to the original size, we
6266 * failed there too. This is very unlikely to
6267 * happen, but if it does, warn and kill all
6268 * tracing.
6269 */
6270 WARN_ON(1);
6271 tracing_disabled = 1;
6272 }
6273 return ret;
6274 }
6275
6276 if (cpu == RING_BUFFER_ALL_CPUS)
6277 set_buffer_entries(&tr->max_buffer, size);
6278 else
6279 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
6280
6281 out:
6282#endif /* CONFIG_TRACER_MAX_TRACE */
6283
6284 if (cpu == RING_BUFFER_ALL_CPUS)
6285 set_buffer_entries(&tr->array_buffer, size);
6286 else
6287 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
6288
6289 return ret;
6290}
6291
6292ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6293 unsigned long size, int cpu_id)
6294{
6295 int ret;
6296
6297 mutex_lock(&trace_types_lock);
6298
6299 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6300 /* make sure, this cpu is enabled in the mask */
6301 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6302 ret = -EINVAL;
6303 goto out;
6304 }
6305 }
6306
6307 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6308 if (ret < 0)
6309 ret = -ENOMEM;
6310
6311out:
6312 mutex_unlock(&trace_types_lock);
6313
6314 return ret;
6315}
6316
6317
6318/**
6319 * tracing_update_buffers - used by tracing facility to expand ring buffers
6320 *
6321 * To save on memory when the tracing is never used on a system with it
6322 * configured in. The ring buffers are set to a minimum size. But once
6323 * a user starts to use the tracing facility, then they need to grow
6324 * to their default size.
6325 *
6326 * This function is to be called when a tracer is about to be used.
6327 */
6328int tracing_update_buffers(void)
6329{
6330 int ret = 0;
6331
6332 mutex_lock(&trace_types_lock);
6333 if (!ring_buffer_expanded)
6334 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
6335 RING_BUFFER_ALL_CPUS);
6336 mutex_unlock(&trace_types_lock);
6337
6338 return ret;
6339}
6340
6341struct trace_option_dentry;
6342
6343static void
6344create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6345
6346/*
6347 * Used to clear out the tracer before deletion of an instance.
6348 * Must have trace_types_lock held.
6349 */
6350static void tracing_set_nop(struct trace_array *tr)
6351{
6352 if (tr->current_trace == &nop_trace)
6353 return;
6354
6355 tr->current_trace->enabled--;
6356
6357 if (tr->current_trace->reset)
6358 tr->current_trace->reset(tr);
6359
6360 tr->current_trace = &nop_trace;
6361}
6362
6363static bool tracer_options_updated;
6364
6365static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6366{
6367 /* Only enable if the directory has been created already. */
6368 if (!tr->dir)
6369 return;
6370
6371 /* Only create trace option files after update_tracer_options finish */
6372 if (!tracer_options_updated)
6373 return;
6374
6375 create_trace_option_files(tr, t);
6376}
6377
6378int tracing_set_tracer(struct trace_array *tr, const char *buf)
6379{
6380 struct tracer *t;
6381#ifdef CONFIG_TRACER_MAX_TRACE
6382 bool had_max_tr;
6383#endif
6384 int ret = 0;
6385
6386 mutex_lock(&trace_types_lock);
6387
6388 if (!ring_buffer_expanded) {
6389 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6390 RING_BUFFER_ALL_CPUS);
6391 if (ret < 0)
6392 goto out;
6393 ret = 0;
6394 }
6395
6396 for (t = trace_types; t; t = t->next) {
6397 if (strcmp(t->name, buf) == 0)
6398 break;
6399 }
6400 if (!t) {
6401 ret = -EINVAL;
6402 goto out;
6403 }
6404 if (t == tr->current_trace)
6405 goto out;
6406
6407#ifdef CONFIG_TRACER_SNAPSHOT
6408 if (t->use_max_tr) {
6409 local_irq_disable();
6410 arch_spin_lock(&tr->max_lock);
6411 if (tr->cond_snapshot)
6412 ret = -EBUSY;
6413 arch_spin_unlock(&tr->max_lock);
6414 local_irq_enable();
6415 if (ret)
6416 goto out;
6417 }
6418#endif
6419 /* Some tracers won't work on kernel command line */
6420 if (system_state < SYSTEM_RUNNING && t->noboot) {
6421 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6422 t->name);
6423 goto out;
6424 }
6425
6426 /* Some tracers are only allowed for the top level buffer */
6427 if (!trace_ok_for_array(t, tr)) {
6428 ret = -EINVAL;
6429 goto out;
6430 }
6431
6432 /* If trace pipe files are being read, we can't change the tracer */
6433 if (tr->trace_ref) {
6434 ret = -EBUSY;
6435 goto out;
6436 }
6437
6438 trace_branch_disable();
6439
6440 tr->current_trace->enabled--;
6441
6442 if (tr->current_trace->reset)
6443 tr->current_trace->reset(tr);
6444
6445#ifdef CONFIG_TRACER_MAX_TRACE
6446 had_max_tr = tr->current_trace->use_max_tr;
6447
6448 /* Current trace needs to be nop_trace before synchronize_rcu */
6449 tr->current_trace = &nop_trace;
6450
6451 if (had_max_tr && !t->use_max_tr) {
6452 /*
6453 * We need to make sure that the update_max_tr sees that
6454 * current_trace changed to nop_trace to keep it from
6455 * swapping the buffers after we resize it.
6456 * The update_max_tr is called from interrupts disabled
6457 * so a synchronized_sched() is sufficient.
6458 */
6459 synchronize_rcu();
6460 free_snapshot(tr);
6461 }
6462
6463 if (t->use_max_tr && !tr->allocated_snapshot) {
6464 ret = tracing_alloc_snapshot_instance(tr);
6465 if (ret < 0)
6466 goto out;
6467 }
6468#else
6469 tr->current_trace = &nop_trace;
6470#endif
6471
6472 if (t->init) {
6473 ret = tracer_init(t, tr);
6474 if (ret)
6475 goto out;
6476 }
6477
6478 tr->current_trace = t;
6479 tr->current_trace->enabled++;
6480 trace_branch_enable(tr);
6481 out:
6482 mutex_unlock(&trace_types_lock);
6483
6484 return ret;
6485}
6486
6487static ssize_t
6488tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6489 size_t cnt, loff_t *ppos)
6490{
6491 struct trace_array *tr = filp->private_data;
6492 char buf[MAX_TRACER_SIZE+1];
6493 char *name;
6494 size_t ret;
6495 int err;
6496
6497 ret = cnt;
6498
6499 if (cnt > MAX_TRACER_SIZE)
6500 cnt = MAX_TRACER_SIZE;
6501
6502 if (copy_from_user(buf, ubuf, cnt))
6503 return -EFAULT;
6504
6505 buf[cnt] = 0;
6506
6507 name = strim(buf);
6508
6509 err = tracing_set_tracer(tr, name);
6510 if (err)
6511 return err;
6512
6513 *ppos += ret;
6514
6515 return ret;
6516}
6517
6518static ssize_t
6519tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6520 size_t cnt, loff_t *ppos)
6521{
6522 char buf[64];
6523 int r;
6524
6525 r = snprintf(buf, sizeof(buf), "%ld\n",
6526 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6527 if (r > sizeof(buf))
6528 r = sizeof(buf);
6529 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6530}
6531
6532static ssize_t
6533tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6534 size_t cnt, loff_t *ppos)
6535{
6536 unsigned long val;
6537 int ret;
6538
6539 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6540 if (ret)
6541 return ret;
6542
6543 *ptr = val * 1000;
6544
6545 return cnt;
6546}
6547
6548static ssize_t
6549tracing_thresh_read(struct file *filp, char __user *ubuf,
6550 size_t cnt, loff_t *ppos)
6551{
6552 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6553}
6554
6555static ssize_t
6556tracing_thresh_write(struct file *filp, const char __user *ubuf,
6557 size_t cnt, loff_t *ppos)
6558{
6559 struct trace_array *tr = filp->private_data;
6560 int ret;
6561
6562 mutex_lock(&trace_types_lock);
6563 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6564 if (ret < 0)
6565 goto out;
6566
6567 if (tr->current_trace->update_thresh) {
6568 ret = tr->current_trace->update_thresh(tr);
6569 if (ret < 0)
6570 goto out;
6571 }
6572
6573 ret = cnt;
6574out:
6575 mutex_unlock(&trace_types_lock);
6576
6577 return ret;
6578}
6579
6580#ifdef CONFIG_TRACER_MAX_TRACE
6581
6582static ssize_t
6583tracing_max_lat_read(struct file *filp, char __user *ubuf,
6584 size_t cnt, loff_t *ppos)
6585{
6586 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6587}
6588
6589static ssize_t
6590tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6591 size_t cnt, loff_t *ppos)
6592{
6593 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6594}
6595
6596#endif
6597
6598static int tracing_open_pipe(struct inode *inode, struct file *filp)
6599{
6600 struct trace_array *tr = inode->i_private;
6601 struct trace_iterator *iter;
6602 int ret;
6603
6604 ret = tracing_check_open_get_tr(tr);
6605 if (ret)
6606 return ret;
6607
6608 mutex_lock(&trace_types_lock);
6609
6610 /* create a buffer to store the information to pass to userspace */
6611 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6612 if (!iter) {
6613 ret = -ENOMEM;
6614 __trace_array_put(tr);
6615 goto out;
6616 }
6617
6618 trace_seq_init(&iter->seq);
6619 iter->trace = tr->current_trace;
6620
6621 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6622 ret = -ENOMEM;
6623 goto fail;
6624 }
6625
6626 /* trace pipe does not show start of buffer */
6627 cpumask_setall(iter->started);
6628
6629 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6630 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6631
6632 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6633 if (trace_clocks[tr->clock_id].in_ns)
6634 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6635
6636 iter->tr = tr;
6637 iter->array_buffer = &tr->array_buffer;
6638 iter->cpu_file = tracing_get_cpu(inode);
6639 mutex_init(&iter->mutex);
6640 filp->private_data = iter;
6641
6642 if (iter->trace->pipe_open)
6643 iter->trace->pipe_open(iter);
6644
6645 nonseekable_open(inode, filp);
6646
6647 tr->trace_ref++;
6648out:
6649 mutex_unlock(&trace_types_lock);
6650 return ret;
6651
6652fail:
6653 kfree(iter);
6654 __trace_array_put(tr);
6655 mutex_unlock(&trace_types_lock);
6656 return ret;
6657}
6658
6659static int tracing_release_pipe(struct inode *inode, struct file *file)
6660{
6661 struct trace_iterator *iter = file->private_data;
6662 struct trace_array *tr = inode->i_private;
6663
6664 mutex_lock(&trace_types_lock);
6665
6666 tr->trace_ref--;
6667
6668 if (iter->trace->pipe_close)
6669 iter->trace->pipe_close(iter);
6670
6671 mutex_unlock(&trace_types_lock);
6672
6673 free_cpumask_var(iter->started);
6674 kfree(iter->fmt);
6675 mutex_destroy(&iter->mutex);
6676 kfree(iter);
6677
6678 trace_array_put(tr);
6679
6680 return 0;
6681}
6682
6683static __poll_t
6684trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6685{
6686 struct trace_array *tr = iter->tr;
6687
6688 /* Iterators are static, they should be filled or empty */
6689 if (trace_buffer_iter(iter, iter->cpu_file))
6690 return EPOLLIN | EPOLLRDNORM;
6691
6692 if (tr->trace_flags & TRACE_ITER_BLOCK)
6693 /*
6694 * Always select as readable when in blocking mode
6695 */
6696 return EPOLLIN | EPOLLRDNORM;
6697 else
6698 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6699 filp, poll_table, iter->tr->buffer_percent);
6700}
6701
6702static __poll_t
6703tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6704{
6705 struct trace_iterator *iter = filp->private_data;
6706
6707 return trace_poll(iter, filp, poll_table);
6708}
6709
6710/* Must be called with iter->mutex held. */
6711static int tracing_wait_pipe(struct file *filp)
6712{
6713 struct trace_iterator *iter = filp->private_data;
6714 int ret;
6715
6716 while (trace_empty(iter)) {
6717
6718 if ((filp->f_flags & O_NONBLOCK)) {
6719 return -EAGAIN;
6720 }
6721
6722 /*
6723 * We block until we read something and tracing is disabled.
6724 * We still block if tracing is disabled, but we have never
6725 * read anything. This allows a user to cat this file, and
6726 * then enable tracing. But after we have read something,
6727 * we give an EOF when tracing is again disabled.
6728 *
6729 * iter->pos will be 0 if we haven't read anything.
6730 */
6731 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6732 break;
6733
6734 mutex_unlock(&iter->mutex);
6735
6736 ret = wait_on_pipe(iter, 0);
6737
6738 mutex_lock(&iter->mutex);
6739
6740 if (ret)
6741 return ret;
6742 }
6743
6744 return 1;
6745}
6746
6747/*
6748 * Consumer reader.
6749 */
6750static ssize_t
6751tracing_read_pipe(struct file *filp, char __user *ubuf,
6752 size_t cnt, loff_t *ppos)
6753{
6754 struct trace_iterator *iter = filp->private_data;
6755 ssize_t sret;
6756
6757 /*
6758 * Avoid more than one consumer on a single file descriptor
6759 * This is just a matter of traces coherency, the ring buffer itself
6760 * is protected.
6761 */
6762 mutex_lock(&iter->mutex);
6763
6764 /* return any leftover data */
6765 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6766 if (sret != -EBUSY)
6767 goto out;
6768
6769 trace_seq_init(&iter->seq);
6770
6771 if (iter->trace->read) {
6772 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6773 if (sret)
6774 goto out;
6775 }
6776
6777waitagain:
6778 sret = tracing_wait_pipe(filp);
6779 if (sret <= 0)
6780 goto out;
6781
6782 /* stop when tracing is finished */
6783 if (trace_empty(iter)) {
6784 sret = 0;
6785 goto out;
6786 }
6787
6788 if (cnt >= PAGE_SIZE)
6789 cnt = PAGE_SIZE - 1;
6790
6791 /* reset all but tr, trace, and overruns */
6792 trace_iterator_reset(iter);
6793 cpumask_clear(iter->started);
6794 trace_seq_init(&iter->seq);
6795
6796 trace_event_read_lock();
6797 trace_access_lock(iter->cpu_file);
6798 while (trace_find_next_entry_inc(iter) != NULL) {
6799 enum print_line_t ret;
6800 int save_len = iter->seq.seq.len;
6801
6802 ret = print_trace_line(iter);
6803 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6804 /*
6805 * If one print_trace_line() fills entire trace_seq in one shot,
6806 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6807 * In this case, we need to consume it, otherwise, loop will peek
6808 * this event next time, resulting in an infinite loop.
6809 */
6810 if (save_len == 0) {
6811 iter->seq.full = 0;
6812 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6813 trace_consume(iter);
6814 break;
6815 }
6816
6817 /* In other cases, don't print partial lines */
6818 iter->seq.seq.len = save_len;
6819 break;
6820 }
6821 if (ret != TRACE_TYPE_NO_CONSUME)
6822 trace_consume(iter);
6823
6824 if (trace_seq_used(&iter->seq) >= cnt)
6825 break;
6826
6827 /*
6828 * Setting the full flag means we reached the trace_seq buffer
6829 * size and we should leave by partial output condition above.
6830 * One of the trace_seq_* functions is not used properly.
6831 */
6832 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6833 iter->ent->type);
6834 }
6835 trace_access_unlock(iter->cpu_file);
6836 trace_event_read_unlock();
6837
6838 /* Now copy what we have to the user */
6839 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6840 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6841 trace_seq_init(&iter->seq);
6842
6843 /*
6844 * If there was nothing to send to user, in spite of consuming trace
6845 * entries, go back to wait for more entries.
6846 */
6847 if (sret == -EBUSY)
6848 goto waitagain;
6849
6850out:
6851 mutex_unlock(&iter->mutex);
6852
6853 return sret;
6854}
6855
6856static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6857 unsigned int idx)
6858{
6859 __free_page(spd->pages[idx]);
6860}
6861
6862static size_t
6863tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6864{
6865 size_t count;
6866 int save_len;
6867 int ret;
6868
6869 /* Seq buffer is page-sized, exactly what we need. */
6870 for (;;) {
6871 save_len = iter->seq.seq.len;
6872 ret = print_trace_line(iter);
6873
6874 if (trace_seq_has_overflowed(&iter->seq)) {
6875 iter->seq.seq.len = save_len;
6876 break;
6877 }
6878
6879 /*
6880 * This should not be hit, because it should only
6881 * be set if the iter->seq overflowed. But check it
6882 * anyway to be safe.
6883 */
6884 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6885 iter->seq.seq.len = save_len;
6886 break;
6887 }
6888
6889 count = trace_seq_used(&iter->seq) - save_len;
6890 if (rem < count) {
6891 rem = 0;
6892 iter->seq.seq.len = save_len;
6893 break;
6894 }
6895
6896 if (ret != TRACE_TYPE_NO_CONSUME)
6897 trace_consume(iter);
6898 rem -= count;
6899 if (!trace_find_next_entry_inc(iter)) {
6900 rem = 0;
6901 iter->ent = NULL;
6902 break;
6903 }
6904 }
6905
6906 return rem;
6907}
6908
6909static ssize_t tracing_splice_read_pipe(struct file *filp,
6910 loff_t *ppos,
6911 struct pipe_inode_info *pipe,
6912 size_t len,
6913 unsigned int flags)
6914{
6915 struct page *pages_def[PIPE_DEF_BUFFERS];
6916 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6917 struct trace_iterator *iter = filp->private_data;
6918 struct splice_pipe_desc spd = {
6919 .pages = pages_def,
6920 .partial = partial_def,
6921 .nr_pages = 0, /* This gets updated below. */
6922 .nr_pages_max = PIPE_DEF_BUFFERS,
6923 .ops = &default_pipe_buf_ops,
6924 .spd_release = tracing_spd_release_pipe,
6925 };
6926 ssize_t ret;
6927 size_t rem;
6928 unsigned int i;
6929
6930 if (splice_grow_spd(pipe, &spd))
6931 return -ENOMEM;
6932
6933 mutex_lock(&iter->mutex);
6934
6935 if (iter->trace->splice_read) {
6936 ret = iter->trace->splice_read(iter, filp,
6937 ppos, pipe, len, flags);
6938 if (ret)
6939 goto out_err;
6940 }
6941
6942 ret = tracing_wait_pipe(filp);
6943 if (ret <= 0)
6944 goto out_err;
6945
6946 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6947 ret = -EFAULT;
6948 goto out_err;
6949 }
6950
6951 trace_event_read_lock();
6952 trace_access_lock(iter->cpu_file);
6953
6954 /* Fill as many pages as possible. */
6955 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6956 spd.pages[i] = alloc_page(GFP_KERNEL);
6957 if (!spd.pages[i])
6958 break;
6959
6960 rem = tracing_fill_pipe_page(rem, iter);
6961
6962 /* Copy the data into the page, so we can start over. */
6963 ret = trace_seq_to_buffer(&iter->seq,
6964 page_address(spd.pages[i]),
6965 trace_seq_used(&iter->seq));
6966 if (ret < 0) {
6967 __free_page(spd.pages[i]);
6968 break;
6969 }
6970 spd.partial[i].offset = 0;
6971 spd.partial[i].len = trace_seq_used(&iter->seq);
6972
6973 trace_seq_init(&iter->seq);
6974 }
6975
6976 trace_access_unlock(iter->cpu_file);
6977 trace_event_read_unlock();
6978 mutex_unlock(&iter->mutex);
6979
6980 spd.nr_pages = i;
6981
6982 if (i)
6983 ret = splice_to_pipe(pipe, &spd);
6984 else
6985 ret = 0;
6986out:
6987 splice_shrink_spd(&spd);
6988 return ret;
6989
6990out_err:
6991 mutex_unlock(&iter->mutex);
6992 goto out;
6993}
6994
6995static ssize_t
6996tracing_entries_read(struct file *filp, 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 int cpu = tracing_get_cpu(inode);
7002 char buf[64];
7003 int r = 0;
7004 ssize_t ret;
7005
7006 mutex_lock(&trace_types_lock);
7007
7008 if (cpu == RING_BUFFER_ALL_CPUS) {
7009 int cpu, buf_size_same;
7010 unsigned long size;
7011
7012 size = 0;
7013 buf_size_same = 1;
7014 /* check if all cpu sizes are same */
7015 for_each_tracing_cpu(cpu) {
7016 /* fill in the size from first enabled cpu */
7017 if (size == 0)
7018 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7019 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7020 buf_size_same = 0;
7021 break;
7022 }
7023 }
7024
7025 if (buf_size_same) {
7026 if (!ring_buffer_expanded)
7027 r = sprintf(buf, "%lu (expanded: %lu)\n",
7028 size >> 10,
7029 trace_buf_size >> 10);
7030 else
7031 r = sprintf(buf, "%lu\n", size >> 10);
7032 } else
7033 r = sprintf(buf, "X\n");
7034 } else
7035 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7036
7037 mutex_unlock(&trace_types_lock);
7038
7039 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7040 return ret;
7041}
7042
7043static ssize_t
7044tracing_entries_write(struct file *filp, const char __user *ubuf,
7045 size_t cnt, loff_t *ppos)
7046{
7047 struct inode *inode = file_inode(filp);
7048 struct trace_array *tr = inode->i_private;
7049 unsigned long val;
7050 int ret;
7051
7052 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7053 if (ret)
7054 return ret;
7055
7056 /* must have at least 1 entry */
7057 if (!val)
7058 return -EINVAL;
7059
7060 /* value is in KB */
7061 val <<= 10;
7062 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7063 if (ret < 0)
7064 return ret;
7065
7066 *ppos += cnt;
7067
7068 return cnt;
7069}
7070
7071static ssize_t
7072tracing_total_entries_read(struct file *filp, char __user *ubuf,
7073 size_t cnt, loff_t *ppos)
7074{
7075 struct trace_array *tr = filp->private_data;
7076 char buf[64];
7077 int r, cpu;
7078 unsigned long size = 0, expanded_size = 0;
7079
7080 mutex_lock(&trace_types_lock);
7081 for_each_tracing_cpu(cpu) {
7082 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7083 if (!ring_buffer_expanded)
7084 expanded_size += trace_buf_size >> 10;
7085 }
7086 if (ring_buffer_expanded)
7087 r = sprintf(buf, "%lu\n", size);
7088 else
7089 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7090 mutex_unlock(&trace_types_lock);
7091
7092 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7093}
7094
7095static ssize_t
7096tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7097 size_t cnt, loff_t *ppos)
7098{
7099 /*
7100 * There is no need to read what the user has written, this function
7101 * is just to make sure that there is no error when "echo" is used
7102 */
7103
7104 *ppos += cnt;
7105
7106 return cnt;
7107}
7108
7109static int
7110tracing_free_buffer_release(struct inode *inode, struct file *filp)
7111{
7112 struct trace_array *tr = inode->i_private;
7113
7114 /* disable tracing ? */
7115 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7116 tracer_tracing_off(tr);
7117 /* resize the ring buffer to 0 */
7118 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7119
7120 trace_array_put(tr);
7121
7122 return 0;
7123}
7124
7125static ssize_t
7126tracing_mark_write(struct file *filp, const char __user *ubuf,
7127 size_t cnt, loff_t *fpos)
7128{
7129 struct trace_array *tr = filp->private_data;
7130 struct ring_buffer_event *event;
7131 enum event_trigger_type tt = ETT_NONE;
7132 struct trace_buffer *buffer;
7133 struct print_entry *entry;
7134 ssize_t written;
7135 int size;
7136 int len;
7137
7138/* Used in tracing_mark_raw_write() as well */
7139#define FAULTED_STR "<faulted>"
7140#define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7141
7142 if (tracing_disabled)
7143 return -EINVAL;
7144
7145 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7146 return -EINVAL;
7147
7148 if (cnt > TRACE_BUF_SIZE)
7149 cnt = TRACE_BUF_SIZE;
7150
7151 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7152
7153 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
7154
7155 /* If less than "<faulted>", then make sure we can still add that */
7156 if (cnt < FAULTED_SIZE)
7157 size += FAULTED_SIZE - cnt;
7158
7159 buffer = tr->array_buffer.buffer;
7160 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7161 tracing_gen_ctx());
7162 if (unlikely(!event))
7163 /* Ring buffer disabled, return as if not open for write */
7164 return -EBADF;
7165
7166 entry = ring_buffer_event_data(event);
7167 entry->ip = _THIS_IP_;
7168
7169 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7170 if (len) {
7171 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7172 cnt = FAULTED_SIZE;
7173 written = -EFAULT;
7174 } else
7175 written = cnt;
7176
7177 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7178 /* do not add \n before testing triggers, but add \0 */
7179 entry->buf[cnt] = '\0';
7180 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7181 }
7182
7183 if (entry->buf[cnt - 1] != '\n') {
7184 entry->buf[cnt] = '\n';
7185 entry->buf[cnt + 1] = '\0';
7186 } else
7187 entry->buf[cnt] = '\0';
7188
7189 if (static_branch_unlikely(&trace_marker_exports_enabled))
7190 ftrace_exports(event, TRACE_EXPORT_MARKER);
7191 __buffer_unlock_commit(buffer, event);
7192
7193 if (tt)
7194 event_triggers_post_call(tr->trace_marker_file, tt);
7195
7196 return written;
7197}
7198
7199/* Limit it for now to 3K (including tag) */
7200#define RAW_DATA_MAX_SIZE (1024*3)
7201
7202static ssize_t
7203tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7204 size_t cnt, loff_t *fpos)
7205{
7206 struct trace_array *tr = filp->private_data;
7207 struct ring_buffer_event *event;
7208 struct trace_buffer *buffer;
7209 struct raw_data_entry *entry;
7210 ssize_t written;
7211 int size;
7212 int len;
7213
7214#define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7215
7216 if (tracing_disabled)
7217 return -EINVAL;
7218
7219 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7220 return -EINVAL;
7221
7222 /* The marker must at least have a tag id */
7223 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
7224 return -EINVAL;
7225
7226 if (cnt > TRACE_BUF_SIZE)
7227 cnt = TRACE_BUF_SIZE;
7228
7229 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7230
7231 size = sizeof(*entry) + cnt;
7232 if (cnt < FAULT_SIZE_ID)
7233 size += FAULT_SIZE_ID - cnt;
7234
7235 buffer = tr->array_buffer.buffer;
7236 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7237 tracing_gen_ctx());
7238 if (!event)
7239 /* Ring buffer disabled, return as if not open for write */
7240 return -EBADF;
7241
7242 entry = ring_buffer_event_data(event);
7243
7244 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7245 if (len) {
7246 entry->id = -1;
7247 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7248 written = -EFAULT;
7249 } else
7250 written = cnt;
7251
7252 __buffer_unlock_commit(buffer, event);
7253
7254 return written;
7255}
7256
7257static int tracing_clock_show(struct seq_file *m, void *v)
7258{
7259 struct trace_array *tr = m->private;
7260 int i;
7261
7262 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7263 seq_printf(m,
7264 "%s%s%s%s", i ? " " : "",
7265 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7266 i == tr->clock_id ? "]" : "");
7267 seq_putc(m, '\n');
7268
7269 return 0;
7270}
7271
7272int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7273{
7274 int i;
7275
7276 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7277 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7278 break;
7279 }
7280 if (i == ARRAY_SIZE(trace_clocks))
7281 return -EINVAL;
7282
7283 mutex_lock(&trace_types_lock);
7284
7285 tr->clock_id = i;
7286
7287 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7288
7289 /*
7290 * New clock may not be consistent with the previous clock.
7291 * Reset the buffer so that it doesn't have incomparable timestamps.
7292 */
7293 tracing_reset_online_cpus(&tr->array_buffer);
7294
7295#ifdef CONFIG_TRACER_MAX_TRACE
7296 if (tr->max_buffer.buffer)
7297 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7298 tracing_reset_online_cpus(&tr->max_buffer);
7299#endif
7300
7301 mutex_unlock(&trace_types_lock);
7302
7303 return 0;
7304}
7305
7306static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7307 size_t cnt, loff_t *fpos)
7308{
7309 struct seq_file *m = filp->private_data;
7310 struct trace_array *tr = m->private;
7311 char buf[64];
7312 const char *clockstr;
7313 int ret;
7314
7315 if (cnt >= sizeof(buf))
7316 return -EINVAL;
7317
7318 if (copy_from_user(buf, ubuf, cnt))
7319 return -EFAULT;
7320
7321 buf[cnt] = 0;
7322
7323 clockstr = strstrip(buf);
7324
7325 ret = tracing_set_clock(tr, clockstr);
7326 if (ret)
7327 return ret;
7328
7329 *fpos += cnt;
7330
7331 return cnt;
7332}
7333
7334static int tracing_clock_open(struct inode *inode, struct file *file)
7335{
7336 struct trace_array *tr = inode->i_private;
7337 int ret;
7338
7339 ret = tracing_check_open_get_tr(tr);
7340 if (ret)
7341 return ret;
7342
7343 ret = single_open(file, tracing_clock_show, inode->i_private);
7344 if (ret < 0)
7345 trace_array_put(tr);
7346
7347 return ret;
7348}
7349
7350static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7351{
7352 struct trace_array *tr = m->private;
7353
7354 mutex_lock(&trace_types_lock);
7355
7356 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7357 seq_puts(m, "delta [absolute]\n");
7358 else
7359 seq_puts(m, "[delta] absolute\n");
7360
7361 mutex_unlock(&trace_types_lock);
7362
7363 return 0;
7364}
7365
7366static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7367{
7368 struct trace_array *tr = inode->i_private;
7369 int ret;
7370
7371 ret = tracing_check_open_get_tr(tr);
7372 if (ret)
7373 return ret;
7374
7375 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7376 if (ret < 0)
7377 trace_array_put(tr);
7378
7379 return ret;
7380}
7381
7382u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7383{
7384 if (rbe == this_cpu_read(trace_buffered_event))
7385 return ring_buffer_time_stamp(buffer);
7386
7387 return ring_buffer_event_time_stamp(buffer, rbe);
7388}
7389
7390/*
7391 * Set or disable using the per CPU trace_buffer_event when possible.
7392 */
7393int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7394{
7395 int ret = 0;
7396
7397 mutex_lock(&trace_types_lock);
7398
7399 if (set && tr->no_filter_buffering_ref++)
7400 goto out;
7401
7402 if (!set) {
7403 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7404 ret = -EINVAL;
7405 goto out;
7406 }
7407
7408 --tr->no_filter_buffering_ref;
7409 }
7410 out:
7411 mutex_unlock(&trace_types_lock);
7412
7413 return ret;
7414}
7415
7416struct ftrace_buffer_info {
7417 struct trace_iterator iter;
7418 void *spare;
7419 unsigned int spare_cpu;
7420 unsigned int read;
7421};
7422
7423#ifdef CONFIG_TRACER_SNAPSHOT
7424static int tracing_snapshot_open(struct inode *inode, struct file *file)
7425{
7426 struct trace_array *tr = inode->i_private;
7427 struct trace_iterator *iter;
7428 struct seq_file *m;
7429 int ret;
7430
7431 ret = tracing_check_open_get_tr(tr);
7432 if (ret)
7433 return ret;
7434
7435 if (file->f_mode & FMODE_READ) {
7436 iter = __tracing_open(inode, file, true);
7437 if (IS_ERR(iter))
7438 ret = PTR_ERR(iter);
7439 } else {
7440 /* Writes still need the seq_file to hold the private data */
7441 ret = -ENOMEM;
7442 m = kzalloc(sizeof(*m), GFP_KERNEL);
7443 if (!m)
7444 goto out;
7445 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7446 if (!iter) {
7447 kfree(m);
7448 goto out;
7449 }
7450 ret = 0;
7451
7452 iter->tr = tr;
7453 iter->array_buffer = &tr->max_buffer;
7454 iter->cpu_file = tracing_get_cpu(inode);
7455 m->private = iter;
7456 file->private_data = m;
7457 }
7458out:
7459 if (ret < 0)
7460 trace_array_put(tr);
7461
7462 return ret;
7463}
7464
7465static ssize_t
7466tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7467 loff_t *ppos)
7468{
7469 struct seq_file *m = filp->private_data;
7470 struct trace_iterator *iter = m->private;
7471 struct trace_array *tr = iter->tr;
7472 unsigned long val;
7473 int ret;
7474
7475 ret = tracing_update_buffers();
7476 if (ret < 0)
7477 return ret;
7478
7479 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7480 if (ret)
7481 return ret;
7482
7483 mutex_lock(&trace_types_lock);
7484
7485 if (tr->current_trace->use_max_tr) {
7486 ret = -EBUSY;
7487 goto out;
7488 }
7489
7490 local_irq_disable();
7491 arch_spin_lock(&tr->max_lock);
7492 if (tr->cond_snapshot)
7493 ret = -EBUSY;
7494 arch_spin_unlock(&tr->max_lock);
7495 local_irq_enable();
7496 if (ret)
7497 goto out;
7498
7499 switch (val) {
7500 case 0:
7501 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7502 ret = -EINVAL;
7503 break;
7504 }
7505 if (tr->allocated_snapshot)
7506 free_snapshot(tr);
7507 break;
7508 case 1:
7509/* Only allow per-cpu swap if the ring buffer supports it */
7510#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7511 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7512 ret = -EINVAL;
7513 break;
7514 }
7515#endif
7516 if (tr->allocated_snapshot)
7517 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7518 &tr->array_buffer, iter->cpu_file);
7519 else
7520 ret = tracing_alloc_snapshot_instance(tr);
7521 if (ret < 0)
7522 break;
7523 local_irq_disable();
7524 /* Now, we're going to swap */
7525 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7526 update_max_tr(tr, current, smp_processor_id(), NULL);
7527 else
7528 update_max_tr_single(tr, current, iter->cpu_file);
7529 local_irq_enable();
7530 break;
7531 default:
7532 if (tr->allocated_snapshot) {
7533 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7534 tracing_reset_online_cpus(&tr->max_buffer);
7535 else
7536 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7537 }
7538 break;
7539 }
7540
7541 if (ret >= 0) {
7542 *ppos += cnt;
7543 ret = cnt;
7544 }
7545out:
7546 mutex_unlock(&trace_types_lock);
7547 return ret;
7548}
7549
7550static int tracing_snapshot_release(struct inode *inode, struct file *file)
7551{
7552 struct seq_file *m = file->private_data;
7553 int ret;
7554
7555 ret = tracing_release(inode, file);
7556
7557 if (file->f_mode & FMODE_READ)
7558 return ret;
7559
7560 /* If write only, the seq_file is just a stub */
7561 if (m)
7562 kfree(m->private);
7563 kfree(m);
7564
7565 return 0;
7566}
7567
7568static int tracing_buffers_open(struct inode *inode, struct file *filp);
7569static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7570 size_t count, loff_t *ppos);
7571static int tracing_buffers_release(struct inode *inode, struct file *file);
7572static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7573 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7574
7575static int snapshot_raw_open(struct inode *inode, struct file *filp)
7576{
7577 struct ftrace_buffer_info *info;
7578 int ret;
7579
7580 /* The following checks for tracefs lockdown */
7581 ret = tracing_buffers_open(inode, filp);
7582 if (ret < 0)
7583 return ret;
7584
7585 info = filp->private_data;
7586
7587 if (info->iter.trace->use_max_tr) {
7588 tracing_buffers_release(inode, filp);
7589 return -EBUSY;
7590 }
7591
7592 info->iter.snapshot = true;
7593 info->iter.array_buffer = &info->iter.tr->max_buffer;
7594
7595 return ret;
7596}
7597
7598#endif /* CONFIG_TRACER_SNAPSHOT */
7599
7600
7601static const struct file_operations tracing_thresh_fops = {
7602 .open = tracing_open_generic,
7603 .read = tracing_thresh_read,
7604 .write = tracing_thresh_write,
7605 .llseek = generic_file_llseek,
7606};
7607
7608#ifdef CONFIG_TRACER_MAX_TRACE
7609static const struct file_operations tracing_max_lat_fops = {
7610 .open = tracing_open_generic,
7611 .read = tracing_max_lat_read,
7612 .write = tracing_max_lat_write,
7613 .llseek = generic_file_llseek,
7614};
7615#endif
7616
7617static const struct file_operations set_tracer_fops = {
7618 .open = tracing_open_generic,
7619 .read = tracing_set_trace_read,
7620 .write = tracing_set_trace_write,
7621 .llseek = generic_file_llseek,
7622};
7623
7624static const struct file_operations tracing_pipe_fops = {
7625 .open = tracing_open_pipe,
7626 .poll = tracing_poll_pipe,
7627 .read = tracing_read_pipe,
7628 .splice_read = tracing_splice_read_pipe,
7629 .release = tracing_release_pipe,
7630 .llseek = no_llseek,
7631};
7632
7633static const struct file_operations tracing_entries_fops = {
7634 .open = tracing_open_generic_tr,
7635 .read = tracing_entries_read,
7636 .write = tracing_entries_write,
7637 .llseek = generic_file_llseek,
7638 .release = tracing_release_generic_tr,
7639};
7640
7641static const struct file_operations tracing_total_entries_fops = {
7642 .open = tracing_open_generic_tr,
7643 .read = tracing_total_entries_read,
7644 .llseek = generic_file_llseek,
7645 .release = tracing_release_generic_tr,
7646};
7647
7648static const struct file_operations tracing_free_buffer_fops = {
7649 .open = tracing_open_generic_tr,
7650 .write = tracing_free_buffer_write,
7651 .release = tracing_free_buffer_release,
7652};
7653
7654static const struct file_operations tracing_mark_fops = {
7655 .open = tracing_mark_open,
7656 .write = tracing_mark_write,
7657 .release = tracing_release_generic_tr,
7658};
7659
7660static const struct file_operations tracing_mark_raw_fops = {
7661 .open = tracing_mark_open,
7662 .write = tracing_mark_raw_write,
7663 .release = tracing_release_generic_tr,
7664};
7665
7666static const struct file_operations trace_clock_fops = {
7667 .open = tracing_clock_open,
7668 .read = seq_read,
7669 .llseek = seq_lseek,
7670 .release = tracing_single_release_tr,
7671 .write = tracing_clock_write,
7672};
7673
7674static const struct file_operations trace_time_stamp_mode_fops = {
7675 .open = tracing_time_stamp_mode_open,
7676 .read = seq_read,
7677 .llseek = seq_lseek,
7678 .release = tracing_single_release_tr,
7679};
7680
7681#ifdef CONFIG_TRACER_SNAPSHOT
7682static const struct file_operations snapshot_fops = {
7683 .open = tracing_snapshot_open,
7684 .read = seq_read,
7685 .write = tracing_snapshot_write,
7686 .llseek = tracing_lseek,
7687 .release = tracing_snapshot_release,
7688};
7689
7690static const struct file_operations snapshot_raw_fops = {
7691 .open = snapshot_raw_open,
7692 .read = tracing_buffers_read,
7693 .release = tracing_buffers_release,
7694 .splice_read = tracing_buffers_splice_read,
7695 .llseek = no_llseek,
7696};
7697
7698#endif /* CONFIG_TRACER_SNAPSHOT */
7699
7700/*
7701 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7702 * @filp: The active open file structure
7703 * @ubuf: The userspace provided buffer to read value into
7704 * @cnt: The maximum number of bytes to read
7705 * @ppos: The current "file" position
7706 *
7707 * This function implements the write interface for a struct trace_min_max_param.
7708 * The filp->private_data must point to a trace_min_max_param structure that
7709 * defines where to write the value, the min and the max acceptable values,
7710 * and a lock to protect the write.
7711 */
7712static ssize_t
7713trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7714{
7715 struct trace_min_max_param *param = filp->private_data;
7716 u64 val;
7717 int err;
7718
7719 if (!param)
7720 return -EFAULT;
7721
7722 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7723 if (err)
7724 return err;
7725
7726 if (param->lock)
7727 mutex_lock(param->lock);
7728
7729 if (param->min && val < *param->min)
7730 err = -EINVAL;
7731
7732 if (param->max && val > *param->max)
7733 err = -EINVAL;
7734
7735 if (!err)
7736 *param->val = val;
7737
7738 if (param->lock)
7739 mutex_unlock(param->lock);
7740
7741 if (err)
7742 return err;
7743
7744 return cnt;
7745}
7746
7747/*
7748 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7749 * @filp: The active open file structure
7750 * @ubuf: The userspace provided buffer to read value into
7751 * @cnt: The maximum number of bytes to read
7752 * @ppos: The current "file" position
7753 *
7754 * This function implements the read interface for a struct trace_min_max_param.
7755 * The filp->private_data must point to a trace_min_max_param struct with valid
7756 * data.
7757 */
7758static ssize_t
7759trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7760{
7761 struct trace_min_max_param *param = filp->private_data;
7762 char buf[U64_STR_SIZE];
7763 int len;
7764 u64 val;
7765
7766 if (!param)
7767 return -EFAULT;
7768
7769 val = *param->val;
7770
7771 if (cnt > sizeof(buf))
7772 cnt = sizeof(buf);
7773
7774 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7775
7776 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7777}
7778
7779const struct file_operations trace_min_max_fops = {
7780 .open = tracing_open_generic,
7781 .read = trace_min_max_read,
7782 .write = trace_min_max_write,
7783};
7784
7785#define TRACING_LOG_ERRS_MAX 8
7786#define TRACING_LOG_LOC_MAX 128
7787
7788#define CMD_PREFIX " Command: "
7789
7790struct err_info {
7791 const char **errs; /* ptr to loc-specific array of err strings */
7792 u8 type; /* index into errs -> specific err string */
7793 u16 pos; /* caret position */
7794 u64 ts;
7795};
7796
7797struct tracing_log_err {
7798 struct list_head list;
7799 struct err_info info;
7800 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7801 char *cmd; /* what caused err */
7802};
7803
7804static DEFINE_MUTEX(tracing_err_log_lock);
7805
7806static struct tracing_log_err *alloc_tracing_log_err(int len)
7807{
7808 struct tracing_log_err *err;
7809
7810 err = kzalloc(sizeof(*err), GFP_KERNEL);
7811 if (!err)
7812 return ERR_PTR(-ENOMEM);
7813
7814 err->cmd = kzalloc(len, GFP_KERNEL);
7815 if (!err->cmd) {
7816 kfree(err);
7817 return ERR_PTR(-ENOMEM);
7818 }
7819
7820 return err;
7821}
7822
7823static void free_tracing_log_err(struct tracing_log_err *err)
7824{
7825 kfree(err->cmd);
7826 kfree(err);
7827}
7828
7829static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7830 int len)
7831{
7832 struct tracing_log_err *err;
7833 char *cmd;
7834
7835 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7836 err = alloc_tracing_log_err(len);
7837 if (PTR_ERR(err) != -ENOMEM)
7838 tr->n_err_log_entries++;
7839
7840 return err;
7841 }
7842 cmd = kzalloc(len, GFP_KERNEL);
7843 if (!cmd)
7844 return ERR_PTR(-ENOMEM);
7845 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7846 kfree(err->cmd);
7847 err->cmd = cmd;
7848 list_del(&err->list);
7849
7850 return err;
7851}
7852
7853/**
7854 * err_pos - find the position of a string within a command for error careting
7855 * @cmd: The tracing command that caused the error
7856 * @str: The string to position the caret at within @cmd
7857 *
7858 * Finds the position of the first occurrence of @str within @cmd. The
7859 * return value can be passed to tracing_log_err() for caret placement
7860 * within @cmd.
7861 *
7862 * Returns the index within @cmd of the first occurrence of @str or 0
7863 * if @str was not found.
7864 */
7865unsigned int err_pos(char *cmd, const char *str)
7866{
7867 char *found;
7868
7869 if (WARN_ON(!strlen(cmd)))
7870 return 0;
7871
7872 found = strstr(cmd, str);
7873 if (found)
7874 return found - cmd;
7875
7876 return 0;
7877}
7878
7879/**
7880 * tracing_log_err - write an error to the tracing error log
7881 * @tr: The associated trace array for the error (NULL for top level array)
7882 * @loc: A string describing where the error occurred
7883 * @cmd: The tracing command that caused the error
7884 * @errs: The array of loc-specific static error strings
7885 * @type: The index into errs[], which produces the specific static err string
7886 * @pos: The position the caret should be placed in the cmd
7887 *
7888 * Writes an error into tracing/error_log of the form:
7889 *
7890 * <loc>: error: <text>
7891 * Command: <cmd>
7892 * ^
7893 *
7894 * tracing/error_log is a small log file containing the last
7895 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7896 * unless there has been a tracing error, and the error log can be
7897 * cleared and have its memory freed by writing the empty string in
7898 * truncation mode to it i.e. echo > tracing/error_log.
7899 *
7900 * NOTE: the @errs array along with the @type param are used to
7901 * produce a static error string - this string is not copied and saved
7902 * when the error is logged - only a pointer to it is saved. See
7903 * existing callers for examples of how static strings are typically
7904 * defined for use with tracing_log_err().
7905 */
7906void tracing_log_err(struct trace_array *tr,
7907 const char *loc, const char *cmd,
7908 const char **errs, u8 type, u16 pos)
7909{
7910 struct tracing_log_err *err;
7911 int len = 0;
7912
7913 if (!tr)
7914 tr = &global_trace;
7915
7916 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
7917
7918 mutex_lock(&tracing_err_log_lock);
7919 err = get_tracing_log_err(tr, len);
7920 if (PTR_ERR(err) == -ENOMEM) {
7921 mutex_unlock(&tracing_err_log_lock);
7922 return;
7923 }
7924
7925 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7926 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
7927
7928 err->info.errs = errs;
7929 err->info.type = type;
7930 err->info.pos = pos;
7931 err->info.ts = local_clock();
7932
7933 list_add_tail(&err->list, &tr->err_log);
7934 mutex_unlock(&tracing_err_log_lock);
7935}
7936
7937static void clear_tracing_err_log(struct trace_array *tr)
7938{
7939 struct tracing_log_err *err, *next;
7940
7941 mutex_lock(&tracing_err_log_lock);
7942 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7943 list_del(&err->list);
7944 free_tracing_log_err(err);
7945 }
7946
7947 tr->n_err_log_entries = 0;
7948 mutex_unlock(&tracing_err_log_lock);
7949}
7950
7951static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7952{
7953 struct trace_array *tr = m->private;
7954
7955 mutex_lock(&tracing_err_log_lock);
7956
7957 return seq_list_start(&tr->err_log, *pos);
7958}
7959
7960static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7961{
7962 struct trace_array *tr = m->private;
7963
7964 return seq_list_next(v, &tr->err_log, pos);
7965}
7966
7967static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7968{
7969 mutex_unlock(&tracing_err_log_lock);
7970}
7971
7972static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
7973{
7974 u16 i;
7975
7976 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7977 seq_putc(m, ' ');
7978 for (i = 0; i < pos; i++)
7979 seq_putc(m, ' ');
7980 seq_puts(m, "^\n");
7981}
7982
7983static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7984{
7985 struct tracing_log_err *err = v;
7986
7987 if (err) {
7988 const char *err_text = err->info.errs[err->info.type];
7989 u64 sec = err->info.ts;
7990 u32 nsec;
7991
7992 nsec = do_div(sec, NSEC_PER_SEC);
7993 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7994 err->loc, err_text);
7995 seq_printf(m, "%s", err->cmd);
7996 tracing_err_log_show_pos(m, err->info.pos);
7997 }
7998
7999 return 0;
8000}
8001
8002static const struct seq_operations tracing_err_log_seq_ops = {
8003 .start = tracing_err_log_seq_start,
8004 .next = tracing_err_log_seq_next,
8005 .stop = tracing_err_log_seq_stop,
8006 .show = tracing_err_log_seq_show
8007};
8008
8009static int tracing_err_log_open(struct inode *inode, struct file *file)
8010{
8011 struct trace_array *tr = inode->i_private;
8012 int ret = 0;
8013
8014 ret = tracing_check_open_get_tr(tr);
8015 if (ret)
8016 return ret;
8017
8018 /* If this file was opened for write, then erase contents */
8019 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8020 clear_tracing_err_log(tr);
8021
8022 if (file->f_mode & FMODE_READ) {
8023 ret = seq_open(file, &tracing_err_log_seq_ops);
8024 if (!ret) {
8025 struct seq_file *m = file->private_data;
8026 m->private = tr;
8027 } else {
8028 trace_array_put(tr);
8029 }
8030 }
8031 return ret;
8032}
8033
8034static ssize_t tracing_err_log_write(struct file *file,
8035 const char __user *buffer,
8036 size_t count, loff_t *ppos)
8037{
8038 return count;
8039}
8040
8041static int tracing_err_log_release(struct inode *inode, struct file *file)
8042{
8043 struct trace_array *tr = inode->i_private;
8044
8045 trace_array_put(tr);
8046
8047 if (file->f_mode & FMODE_READ)
8048 seq_release(inode, file);
8049
8050 return 0;
8051}
8052
8053static const struct file_operations tracing_err_log_fops = {
8054 .open = tracing_err_log_open,
8055 .write = tracing_err_log_write,
8056 .read = seq_read,
8057 .llseek = seq_lseek,
8058 .release = tracing_err_log_release,
8059};
8060
8061static int tracing_buffers_open(struct inode *inode, struct file *filp)
8062{
8063 struct trace_array *tr = inode->i_private;
8064 struct ftrace_buffer_info *info;
8065 int ret;
8066
8067 ret = tracing_check_open_get_tr(tr);
8068 if (ret)
8069 return ret;
8070
8071 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8072 if (!info) {
8073 trace_array_put(tr);
8074 return -ENOMEM;
8075 }
8076
8077 mutex_lock(&trace_types_lock);
8078
8079 info->iter.tr = tr;
8080 info->iter.cpu_file = tracing_get_cpu(inode);
8081 info->iter.trace = tr->current_trace;
8082 info->iter.array_buffer = &tr->array_buffer;
8083 info->spare = NULL;
8084 /* Force reading ring buffer for first read */
8085 info->read = (unsigned int)-1;
8086
8087 filp->private_data = info;
8088
8089 tr->trace_ref++;
8090
8091 mutex_unlock(&trace_types_lock);
8092
8093 ret = nonseekable_open(inode, filp);
8094 if (ret < 0)
8095 trace_array_put(tr);
8096
8097 return ret;
8098}
8099
8100static __poll_t
8101tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8102{
8103 struct ftrace_buffer_info *info = filp->private_data;
8104 struct trace_iterator *iter = &info->iter;
8105
8106 return trace_poll(iter, filp, poll_table);
8107}
8108
8109static ssize_t
8110tracing_buffers_read(struct file *filp, char __user *ubuf,
8111 size_t count, loff_t *ppos)
8112{
8113 struct ftrace_buffer_info *info = filp->private_data;
8114 struct trace_iterator *iter = &info->iter;
8115 ssize_t ret = 0;
8116 ssize_t size;
8117
8118 if (!count)
8119 return 0;
8120
8121#ifdef CONFIG_TRACER_MAX_TRACE
8122 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8123 return -EBUSY;
8124#endif
8125
8126 if (!info->spare) {
8127 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8128 iter->cpu_file);
8129 if (IS_ERR(info->spare)) {
8130 ret = PTR_ERR(info->spare);
8131 info->spare = NULL;
8132 } else {
8133 info->spare_cpu = iter->cpu_file;
8134 }
8135 }
8136 if (!info->spare)
8137 return ret;
8138
8139 /* Do we have previous read data to read? */
8140 if (info->read < PAGE_SIZE)
8141 goto read;
8142
8143 again:
8144 trace_access_lock(iter->cpu_file);
8145 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8146 &info->spare,
8147 count,
8148 iter->cpu_file, 0);
8149 trace_access_unlock(iter->cpu_file);
8150
8151 if (ret < 0) {
8152 if (trace_empty(iter)) {
8153 if ((filp->f_flags & O_NONBLOCK))
8154 return -EAGAIN;
8155
8156 ret = wait_on_pipe(iter, 0);
8157 if (ret)
8158 return ret;
8159
8160 goto again;
8161 }
8162 return 0;
8163 }
8164
8165 info->read = 0;
8166 read:
8167 size = PAGE_SIZE - info->read;
8168 if (size > count)
8169 size = count;
8170
8171 ret = copy_to_user(ubuf, info->spare + info->read, size);
8172 if (ret == size)
8173 return -EFAULT;
8174
8175 size -= ret;
8176
8177 *ppos += size;
8178 info->read += size;
8179
8180 return size;
8181}
8182
8183static int tracing_buffers_release(struct inode *inode, struct file *file)
8184{
8185 struct ftrace_buffer_info *info = file->private_data;
8186 struct trace_iterator *iter = &info->iter;
8187
8188 mutex_lock(&trace_types_lock);
8189
8190 iter->tr->trace_ref--;
8191
8192 __trace_array_put(iter->tr);
8193
8194 iter->wait_index++;
8195 /* Make sure the waiters see the new wait_index */
8196 smp_wmb();
8197
8198 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8199
8200 if (info->spare)
8201 ring_buffer_free_read_page(iter->array_buffer->buffer,
8202 info->spare_cpu, info->spare);
8203 kvfree(info);
8204
8205 mutex_unlock(&trace_types_lock);
8206
8207 return 0;
8208}
8209
8210struct buffer_ref {
8211 struct trace_buffer *buffer;
8212 void *page;
8213 int cpu;
8214 refcount_t refcount;
8215};
8216
8217static void buffer_ref_release(struct buffer_ref *ref)
8218{
8219 if (!refcount_dec_and_test(&ref->refcount))
8220 return;
8221 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8222 kfree(ref);
8223}
8224
8225static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8226 struct pipe_buffer *buf)
8227{
8228 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8229
8230 buffer_ref_release(ref);
8231 buf->private = 0;
8232}
8233
8234static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8235 struct pipe_buffer *buf)
8236{
8237 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8238
8239 if (refcount_read(&ref->refcount) > INT_MAX/2)
8240 return false;
8241
8242 refcount_inc(&ref->refcount);
8243 return true;
8244}
8245
8246/* Pipe buffer operations for a buffer. */
8247static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8248 .release = buffer_pipe_buf_release,
8249 .get = buffer_pipe_buf_get,
8250};
8251
8252/*
8253 * Callback from splice_to_pipe(), if we need to release some pages
8254 * at the end of the spd in case we error'ed out in filling the pipe.
8255 */
8256static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8257{
8258 struct buffer_ref *ref =
8259 (struct buffer_ref *)spd->partial[i].private;
8260
8261 buffer_ref_release(ref);
8262 spd->partial[i].private = 0;
8263}
8264
8265static ssize_t
8266tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8267 struct pipe_inode_info *pipe, size_t len,
8268 unsigned int flags)
8269{
8270 struct ftrace_buffer_info *info = file->private_data;
8271 struct trace_iterator *iter = &info->iter;
8272 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8273 struct page *pages_def[PIPE_DEF_BUFFERS];
8274 struct splice_pipe_desc spd = {
8275 .pages = pages_def,
8276 .partial = partial_def,
8277 .nr_pages_max = PIPE_DEF_BUFFERS,
8278 .ops = &buffer_pipe_buf_ops,
8279 .spd_release = buffer_spd_release,
8280 };
8281 struct buffer_ref *ref;
8282 int entries, i;
8283 ssize_t ret = 0;
8284
8285#ifdef CONFIG_TRACER_MAX_TRACE
8286 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8287 return -EBUSY;
8288#endif
8289
8290 if (*ppos & (PAGE_SIZE - 1))
8291 return -EINVAL;
8292
8293 if (len & (PAGE_SIZE - 1)) {
8294 if (len < PAGE_SIZE)
8295 return -EINVAL;
8296 len &= PAGE_MASK;
8297 }
8298
8299 if (splice_grow_spd(pipe, &spd))
8300 return -ENOMEM;
8301
8302 again:
8303 trace_access_lock(iter->cpu_file);
8304 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8305
8306 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
8307 struct page *page;
8308 int r;
8309
8310 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8311 if (!ref) {
8312 ret = -ENOMEM;
8313 break;
8314 }
8315
8316 refcount_set(&ref->refcount, 1);
8317 ref->buffer = iter->array_buffer->buffer;
8318 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8319 if (IS_ERR(ref->page)) {
8320 ret = PTR_ERR(ref->page);
8321 ref->page = NULL;
8322 kfree(ref);
8323 break;
8324 }
8325 ref->cpu = iter->cpu_file;
8326
8327 r = ring_buffer_read_page(ref->buffer, &ref->page,
8328 len, iter->cpu_file, 1);
8329 if (r < 0) {
8330 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8331 ref->page);
8332 kfree(ref);
8333 break;
8334 }
8335
8336 page = virt_to_page(ref->page);
8337
8338 spd.pages[i] = page;
8339 spd.partial[i].len = PAGE_SIZE;
8340 spd.partial[i].offset = 0;
8341 spd.partial[i].private = (unsigned long)ref;
8342 spd.nr_pages++;
8343 *ppos += PAGE_SIZE;
8344
8345 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8346 }
8347
8348 trace_access_unlock(iter->cpu_file);
8349 spd.nr_pages = i;
8350
8351 /* did we read anything? */
8352 if (!spd.nr_pages) {
8353 long wait_index;
8354
8355 if (ret)
8356 goto out;
8357
8358 ret = -EAGAIN;
8359 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8360 goto out;
8361
8362 wait_index = READ_ONCE(iter->wait_index);
8363
8364 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
8365 if (ret)
8366 goto out;
8367
8368 /* No need to wait after waking up when tracing is off */
8369 if (!tracer_tracing_is_on(iter->tr))
8370 goto out;
8371
8372 /* Make sure we see the new wait_index */
8373 smp_rmb();
8374 if (wait_index != iter->wait_index)
8375 goto out;
8376
8377 goto again;
8378 }
8379
8380 ret = splice_to_pipe(pipe, &spd);
8381out:
8382 splice_shrink_spd(&spd);
8383
8384 return ret;
8385}
8386
8387/* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8388static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8389{
8390 struct ftrace_buffer_info *info = file->private_data;
8391 struct trace_iterator *iter = &info->iter;
8392
8393 if (cmd)
8394 return -ENOIOCTLCMD;
8395
8396 mutex_lock(&trace_types_lock);
8397
8398 iter->wait_index++;
8399 /* Make sure the waiters see the new wait_index */
8400 smp_wmb();
8401
8402 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8403
8404 mutex_unlock(&trace_types_lock);
8405 return 0;
8406}
8407
8408static const struct file_operations tracing_buffers_fops = {
8409 .open = tracing_buffers_open,
8410 .read = tracing_buffers_read,
8411 .poll = tracing_buffers_poll,
8412 .release = tracing_buffers_release,
8413 .splice_read = tracing_buffers_splice_read,
8414 .unlocked_ioctl = tracing_buffers_ioctl,
8415 .llseek = no_llseek,
8416};
8417
8418static ssize_t
8419tracing_stats_read(struct file *filp, char __user *ubuf,
8420 size_t count, loff_t *ppos)
8421{
8422 struct inode *inode = file_inode(filp);
8423 struct trace_array *tr = inode->i_private;
8424 struct array_buffer *trace_buf = &tr->array_buffer;
8425 int cpu = tracing_get_cpu(inode);
8426 struct trace_seq *s;
8427 unsigned long cnt;
8428 unsigned long long t;
8429 unsigned long usec_rem;
8430
8431 s = kmalloc(sizeof(*s), GFP_KERNEL);
8432 if (!s)
8433 return -ENOMEM;
8434
8435 trace_seq_init(s);
8436
8437 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8438 trace_seq_printf(s, "entries: %ld\n", cnt);
8439
8440 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8441 trace_seq_printf(s, "overrun: %ld\n", cnt);
8442
8443 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8444 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8445
8446 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8447 trace_seq_printf(s, "bytes: %ld\n", cnt);
8448
8449 if (trace_clocks[tr->clock_id].in_ns) {
8450 /* local or global for trace_clock */
8451 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8452 usec_rem = do_div(t, USEC_PER_SEC);
8453 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8454 t, usec_rem);
8455
8456 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8457 usec_rem = do_div(t, USEC_PER_SEC);
8458 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8459 } else {
8460 /* counter or tsc mode for trace_clock */
8461 trace_seq_printf(s, "oldest event ts: %llu\n",
8462 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8463
8464 trace_seq_printf(s, "now ts: %llu\n",
8465 ring_buffer_time_stamp(trace_buf->buffer));
8466 }
8467
8468 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8469 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8470
8471 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8472 trace_seq_printf(s, "read events: %ld\n", cnt);
8473
8474 count = simple_read_from_buffer(ubuf, count, ppos,
8475 s->buffer, trace_seq_used(s));
8476
8477 kfree(s);
8478
8479 return count;
8480}
8481
8482static const struct file_operations tracing_stats_fops = {
8483 .open = tracing_open_generic_tr,
8484 .read = tracing_stats_read,
8485 .llseek = generic_file_llseek,
8486 .release = tracing_release_generic_tr,
8487};
8488
8489#ifdef CONFIG_DYNAMIC_FTRACE
8490
8491static ssize_t
8492tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8493 size_t cnt, loff_t *ppos)
8494{
8495 ssize_t ret;
8496 char *buf;
8497 int r;
8498
8499 /* 256 should be plenty to hold the amount needed */
8500 buf = kmalloc(256, GFP_KERNEL);
8501 if (!buf)
8502 return -ENOMEM;
8503
8504 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8505 ftrace_update_tot_cnt,
8506 ftrace_number_of_pages,
8507 ftrace_number_of_groups);
8508
8509 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8510 kfree(buf);
8511 return ret;
8512}
8513
8514static const struct file_operations tracing_dyn_info_fops = {
8515 .open = tracing_open_generic,
8516 .read = tracing_read_dyn_info,
8517 .llseek = generic_file_llseek,
8518};
8519#endif /* CONFIG_DYNAMIC_FTRACE */
8520
8521#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8522static void
8523ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8524 struct trace_array *tr, struct ftrace_probe_ops *ops,
8525 void *data)
8526{
8527 tracing_snapshot_instance(tr);
8528}
8529
8530static void
8531ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8532 struct trace_array *tr, struct ftrace_probe_ops *ops,
8533 void *data)
8534{
8535 struct ftrace_func_mapper *mapper = data;
8536 long *count = NULL;
8537
8538 if (mapper)
8539 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8540
8541 if (count) {
8542
8543 if (*count <= 0)
8544 return;
8545
8546 (*count)--;
8547 }
8548
8549 tracing_snapshot_instance(tr);
8550}
8551
8552static int
8553ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8554 struct ftrace_probe_ops *ops, void *data)
8555{
8556 struct ftrace_func_mapper *mapper = data;
8557 long *count = NULL;
8558
8559 seq_printf(m, "%ps:", (void *)ip);
8560
8561 seq_puts(m, "snapshot");
8562
8563 if (mapper)
8564 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8565
8566 if (count)
8567 seq_printf(m, ":count=%ld\n", *count);
8568 else
8569 seq_puts(m, ":unlimited\n");
8570
8571 return 0;
8572}
8573
8574static int
8575ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8576 unsigned long ip, void *init_data, void **data)
8577{
8578 struct ftrace_func_mapper *mapper = *data;
8579
8580 if (!mapper) {
8581 mapper = allocate_ftrace_func_mapper();
8582 if (!mapper)
8583 return -ENOMEM;
8584 *data = mapper;
8585 }
8586
8587 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8588}
8589
8590static void
8591ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8592 unsigned long ip, void *data)
8593{
8594 struct ftrace_func_mapper *mapper = data;
8595
8596 if (!ip) {
8597 if (!mapper)
8598 return;
8599 free_ftrace_func_mapper(mapper, NULL);
8600 return;
8601 }
8602
8603 ftrace_func_mapper_remove_ip(mapper, ip);
8604}
8605
8606static struct ftrace_probe_ops snapshot_probe_ops = {
8607 .func = ftrace_snapshot,
8608 .print = ftrace_snapshot_print,
8609};
8610
8611static struct ftrace_probe_ops snapshot_count_probe_ops = {
8612 .func = ftrace_count_snapshot,
8613 .print = ftrace_snapshot_print,
8614 .init = ftrace_snapshot_init,
8615 .free = ftrace_snapshot_free,
8616};
8617
8618static int
8619ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8620 char *glob, char *cmd, char *param, int enable)
8621{
8622 struct ftrace_probe_ops *ops;
8623 void *count = (void *)-1;
8624 char *number;
8625 int ret;
8626
8627 if (!tr)
8628 return -ENODEV;
8629
8630 /* hash funcs only work with set_ftrace_filter */
8631 if (!enable)
8632 return -EINVAL;
8633
8634 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8635
8636 if (glob[0] == '!')
8637 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8638
8639 if (!param)
8640 goto out_reg;
8641
8642 number = strsep(&param, ":");
8643
8644 if (!strlen(number))
8645 goto out_reg;
8646
8647 /*
8648 * We use the callback data field (which is a pointer)
8649 * as our counter.
8650 */
8651 ret = kstrtoul(number, 0, (unsigned long *)&count);
8652 if (ret)
8653 return ret;
8654
8655 out_reg:
8656 ret = tracing_alloc_snapshot_instance(tr);
8657 if (ret < 0)
8658 goto out;
8659
8660 ret = register_ftrace_function_probe(glob, tr, ops, count);
8661
8662 out:
8663 return ret < 0 ? ret : 0;
8664}
8665
8666static struct ftrace_func_command ftrace_snapshot_cmd = {
8667 .name = "snapshot",
8668 .func = ftrace_trace_snapshot_callback,
8669};
8670
8671static __init int register_snapshot_cmd(void)
8672{
8673 return register_ftrace_command(&ftrace_snapshot_cmd);
8674}
8675#else
8676static inline __init int register_snapshot_cmd(void) { return 0; }
8677#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8678
8679static struct dentry *tracing_get_dentry(struct trace_array *tr)
8680{
8681 if (WARN_ON(!tr->dir))
8682 return ERR_PTR(-ENODEV);
8683
8684 /* Top directory uses NULL as the parent */
8685 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8686 return NULL;
8687
8688 /* All sub buffers have a descriptor */
8689 return tr->dir;
8690}
8691
8692static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8693{
8694 struct dentry *d_tracer;
8695
8696 if (tr->percpu_dir)
8697 return tr->percpu_dir;
8698
8699 d_tracer = tracing_get_dentry(tr);
8700 if (IS_ERR(d_tracer))
8701 return NULL;
8702
8703 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8704
8705 MEM_FAIL(!tr->percpu_dir,
8706 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8707
8708 return tr->percpu_dir;
8709}
8710
8711static struct dentry *
8712trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8713 void *data, long cpu, const struct file_operations *fops)
8714{
8715 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8716
8717 if (ret) /* See tracing_get_cpu() */
8718 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8719 return ret;
8720}
8721
8722static void
8723tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8724{
8725 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8726 struct dentry *d_cpu;
8727 char cpu_dir[30]; /* 30 characters should be more than enough */
8728
8729 if (!d_percpu)
8730 return;
8731
8732 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8733 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8734 if (!d_cpu) {
8735 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8736 return;
8737 }
8738
8739 /* per cpu trace_pipe */
8740 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8741 tr, cpu, &tracing_pipe_fops);
8742
8743 /* per cpu trace */
8744 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8745 tr, cpu, &tracing_fops);
8746
8747 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8748 tr, cpu, &tracing_buffers_fops);
8749
8750 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8751 tr, cpu, &tracing_stats_fops);
8752
8753 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8754 tr, cpu, &tracing_entries_fops);
8755
8756#ifdef CONFIG_TRACER_SNAPSHOT
8757 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8758 tr, cpu, &snapshot_fops);
8759
8760 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8761 tr, cpu, &snapshot_raw_fops);
8762#endif
8763}
8764
8765#ifdef CONFIG_FTRACE_SELFTEST
8766/* Let selftest have access to static functions in this file */
8767#include "trace_selftest.c"
8768#endif
8769
8770static ssize_t
8771trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8772 loff_t *ppos)
8773{
8774 struct trace_option_dentry *topt = filp->private_data;
8775 char *buf;
8776
8777 if (topt->flags->val & topt->opt->bit)
8778 buf = "1\n";
8779 else
8780 buf = "0\n";
8781
8782 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8783}
8784
8785static ssize_t
8786trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8787 loff_t *ppos)
8788{
8789 struct trace_option_dentry *topt = filp->private_data;
8790 unsigned long val;
8791 int ret;
8792
8793 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8794 if (ret)
8795 return ret;
8796
8797 if (val != 0 && val != 1)
8798 return -EINVAL;
8799
8800 if (!!(topt->flags->val & topt->opt->bit) != val) {
8801 mutex_lock(&trace_types_lock);
8802 ret = __set_tracer_option(topt->tr, topt->flags,
8803 topt->opt, !val);
8804 mutex_unlock(&trace_types_lock);
8805 if (ret)
8806 return ret;
8807 }
8808
8809 *ppos += cnt;
8810
8811 return cnt;
8812}
8813
8814
8815static const struct file_operations trace_options_fops = {
8816 .open = tracing_open_generic,
8817 .read = trace_options_read,
8818 .write = trace_options_write,
8819 .llseek = generic_file_llseek,
8820};
8821
8822/*
8823 * In order to pass in both the trace_array descriptor as well as the index
8824 * to the flag that the trace option file represents, the trace_array
8825 * has a character array of trace_flags_index[], which holds the index
8826 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8827 * The address of this character array is passed to the flag option file
8828 * read/write callbacks.
8829 *
8830 * In order to extract both the index and the trace_array descriptor,
8831 * get_tr_index() uses the following algorithm.
8832 *
8833 * idx = *ptr;
8834 *
8835 * As the pointer itself contains the address of the index (remember
8836 * index[1] == 1).
8837 *
8838 * Then to get the trace_array descriptor, by subtracting that index
8839 * from the ptr, we get to the start of the index itself.
8840 *
8841 * ptr - idx == &index[0]
8842 *
8843 * Then a simple container_of() from that pointer gets us to the
8844 * trace_array descriptor.
8845 */
8846static void get_tr_index(void *data, struct trace_array **ptr,
8847 unsigned int *pindex)
8848{
8849 *pindex = *(unsigned char *)data;
8850
8851 *ptr = container_of(data - *pindex, struct trace_array,
8852 trace_flags_index);
8853}
8854
8855static ssize_t
8856trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8857 loff_t *ppos)
8858{
8859 void *tr_index = filp->private_data;
8860 struct trace_array *tr;
8861 unsigned int index;
8862 char *buf;
8863
8864 get_tr_index(tr_index, &tr, &index);
8865
8866 if (tr->trace_flags & (1 << index))
8867 buf = "1\n";
8868 else
8869 buf = "0\n";
8870
8871 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8872}
8873
8874static ssize_t
8875trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8876 loff_t *ppos)
8877{
8878 void *tr_index = filp->private_data;
8879 struct trace_array *tr;
8880 unsigned int index;
8881 unsigned long val;
8882 int ret;
8883
8884 get_tr_index(tr_index, &tr, &index);
8885
8886 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8887 if (ret)
8888 return ret;
8889
8890 if (val != 0 && val != 1)
8891 return -EINVAL;
8892
8893 mutex_lock(&event_mutex);
8894 mutex_lock(&trace_types_lock);
8895 ret = set_tracer_flag(tr, 1 << index, val);
8896 mutex_unlock(&trace_types_lock);
8897 mutex_unlock(&event_mutex);
8898
8899 if (ret < 0)
8900 return ret;
8901
8902 *ppos += cnt;
8903
8904 return cnt;
8905}
8906
8907static const struct file_operations trace_options_core_fops = {
8908 .open = tracing_open_generic,
8909 .read = trace_options_core_read,
8910 .write = trace_options_core_write,
8911 .llseek = generic_file_llseek,
8912};
8913
8914struct dentry *trace_create_file(const char *name,
8915 umode_t mode,
8916 struct dentry *parent,
8917 void *data,
8918 const struct file_operations *fops)
8919{
8920 struct dentry *ret;
8921
8922 ret = tracefs_create_file(name, mode, parent, data, fops);
8923 if (!ret)
8924 pr_warn("Could not create tracefs '%s' entry\n", name);
8925
8926 return ret;
8927}
8928
8929
8930static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8931{
8932 struct dentry *d_tracer;
8933
8934 if (tr->options)
8935 return tr->options;
8936
8937 d_tracer = tracing_get_dentry(tr);
8938 if (IS_ERR(d_tracer))
8939 return NULL;
8940
8941 tr->options = tracefs_create_dir("options", d_tracer);
8942 if (!tr->options) {
8943 pr_warn("Could not create tracefs directory 'options'\n");
8944 return NULL;
8945 }
8946
8947 return tr->options;
8948}
8949
8950static void
8951create_trace_option_file(struct trace_array *tr,
8952 struct trace_option_dentry *topt,
8953 struct tracer_flags *flags,
8954 struct tracer_opt *opt)
8955{
8956 struct dentry *t_options;
8957
8958 t_options = trace_options_init_dentry(tr);
8959 if (!t_options)
8960 return;
8961
8962 topt->flags = flags;
8963 topt->opt = opt;
8964 topt->tr = tr;
8965
8966 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
8967 t_options, topt, &trace_options_fops);
8968
8969}
8970
8971static void
8972create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8973{
8974 struct trace_option_dentry *topts;
8975 struct trace_options *tr_topts;
8976 struct tracer_flags *flags;
8977 struct tracer_opt *opts;
8978 int cnt;
8979 int i;
8980
8981 if (!tracer)
8982 return;
8983
8984 flags = tracer->flags;
8985
8986 if (!flags || !flags->opts)
8987 return;
8988
8989 /*
8990 * If this is an instance, only create flags for tracers
8991 * the instance may have.
8992 */
8993 if (!trace_ok_for_array(tracer, tr))
8994 return;
8995
8996 for (i = 0; i < tr->nr_topts; i++) {
8997 /* Make sure there's no duplicate flags. */
8998 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8999 return;
9000 }
9001
9002 opts = flags->opts;
9003
9004 for (cnt = 0; opts[cnt].name; cnt++)
9005 ;
9006
9007 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
9008 if (!topts)
9009 return;
9010
9011 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9012 GFP_KERNEL);
9013 if (!tr_topts) {
9014 kfree(topts);
9015 return;
9016 }
9017
9018 tr->topts = tr_topts;
9019 tr->topts[tr->nr_topts].tracer = tracer;
9020 tr->topts[tr->nr_topts].topts = topts;
9021 tr->nr_topts++;
9022
9023 for (cnt = 0; opts[cnt].name; cnt++) {
9024 create_trace_option_file(tr, &topts[cnt], flags,
9025 &opts[cnt]);
9026 MEM_FAIL(topts[cnt].entry == NULL,
9027 "Failed to create trace option: %s",
9028 opts[cnt].name);
9029 }
9030}
9031
9032static struct dentry *
9033create_trace_option_core_file(struct trace_array *tr,
9034 const char *option, long index)
9035{
9036 struct dentry *t_options;
9037
9038 t_options = trace_options_init_dentry(tr);
9039 if (!t_options)
9040 return NULL;
9041
9042 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9043 (void *)&tr->trace_flags_index[index],
9044 &trace_options_core_fops);
9045}
9046
9047static void create_trace_options_dir(struct trace_array *tr)
9048{
9049 struct dentry *t_options;
9050 bool top_level = tr == &global_trace;
9051 int i;
9052
9053 t_options = trace_options_init_dentry(tr);
9054 if (!t_options)
9055 return;
9056
9057 for (i = 0; trace_options[i]; i++) {
9058 if (top_level ||
9059 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9060 create_trace_option_core_file(tr, trace_options[i], i);
9061 }
9062}
9063
9064static ssize_t
9065rb_simple_read(struct file *filp, char __user *ubuf,
9066 size_t cnt, loff_t *ppos)
9067{
9068 struct trace_array *tr = filp->private_data;
9069 char buf[64];
9070 int r;
9071
9072 r = tracer_tracing_is_on(tr);
9073 r = sprintf(buf, "%d\n", r);
9074
9075 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9076}
9077
9078static ssize_t
9079rb_simple_write(struct file *filp, const char __user *ubuf,
9080 size_t cnt, loff_t *ppos)
9081{
9082 struct trace_array *tr = filp->private_data;
9083 struct trace_buffer *buffer = tr->array_buffer.buffer;
9084 unsigned long val;
9085 int ret;
9086
9087 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9088 if (ret)
9089 return ret;
9090
9091 if (buffer) {
9092 mutex_lock(&trace_types_lock);
9093 if (!!val == tracer_tracing_is_on(tr)) {
9094 val = 0; /* do nothing */
9095 } else if (val) {
9096 tracer_tracing_on(tr);
9097 if (tr->current_trace->start)
9098 tr->current_trace->start(tr);
9099 } else {
9100 tracer_tracing_off(tr);
9101 if (tr->current_trace->stop)
9102 tr->current_trace->stop(tr);
9103 /* Wake up any waiters */
9104 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9105 }
9106 mutex_unlock(&trace_types_lock);
9107 }
9108
9109 (*ppos)++;
9110
9111 return cnt;
9112}
9113
9114static const struct file_operations rb_simple_fops = {
9115 .open = tracing_open_generic_tr,
9116 .read = rb_simple_read,
9117 .write = rb_simple_write,
9118 .release = tracing_release_generic_tr,
9119 .llseek = default_llseek,
9120};
9121
9122static ssize_t
9123buffer_percent_read(struct file *filp, char __user *ubuf,
9124 size_t cnt, loff_t *ppos)
9125{
9126 struct trace_array *tr = filp->private_data;
9127 char buf[64];
9128 int r;
9129
9130 r = tr->buffer_percent;
9131 r = sprintf(buf, "%d\n", r);
9132
9133 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9134}
9135
9136static ssize_t
9137buffer_percent_write(struct file *filp, const char __user *ubuf,
9138 size_t cnt, loff_t *ppos)
9139{
9140 struct trace_array *tr = filp->private_data;
9141 unsigned long val;
9142 int ret;
9143
9144 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9145 if (ret)
9146 return ret;
9147
9148 if (val > 100)
9149 return -EINVAL;
9150
9151 tr->buffer_percent = val;
9152
9153 (*ppos)++;
9154
9155 return cnt;
9156}
9157
9158static const struct file_operations buffer_percent_fops = {
9159 .open = tracing_open_generic_tr,
9160 .read = buffer_percent_read,
9161 .write = buffer_percent_write,
9162 .release = tracing_release_generic_tr,
9163 .llseek = default_llseek,
9164};
9165
9166static struct dentry *trace_instance_dir;
9167
9168static void
9169init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9170
9171static int
9172allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9173{
9174 enum ring_buffer_flags rb_flags;
9175
9176 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9177
9178 buf->tr = tr;
9179
9180 buf->buffer = ring_buffer_alloc(size, rb_flags);
9181 if (!buf->buffer)
9182 return -ENOMEM;
9183
9184 buf->data = alloc_percpu(struct trace_array_cpu);
9185 if (!buf->data) {
9186 ring_buffer_free(buf->buffer);
9187 buf->buffer = NULL;
9188 return -ENOMEM;
9189 }
9190
9191 /* Allocate the first page for all buffers */
9192 set_buffer_entries(&tr->array_buffer,
9193 ring_buffer_size(tr->array_buffer.buffer, 0));
9194
9195 return 0;
9196}
9197
9198static void free_trace_buffer(struct array_buffer *buf)
9199{
9200 if (buf->buffer) {
9201 ring_buffer_free(buf->buffer);
9202 buf->buffer = NULL;
9203 free_percpu(buf->data);
9204 buf->data = NULL;
9205 }
9206}
9207
9208static int allocate_trace_buffers(struct trace_array *tr, int size)
9209{
9210 int ret;
9211
9212 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9213 if (ret)
9214 return ret;
9215
9216#ifdef CONFIG_TRACER_MAX_TRACE
9217 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9218 allocate_snapshot ? size : 1);
9219 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9220 free_trace_buffer(&tr->array_buffer);
9221 return -ENOMEM;
9222 }
9223 tr->allocated_snapshot = allocate_snapshot;
9224
9225 /*
9226 * Only the top level trace array gets its snapshot allocated
9227 * from the kernel command line.
9228 */
9229 allocate_snapshot = false;
9230#endif
9231
9232 return 0;
9233}
9234
9235static void free_trace_buffers(struct trace_array *tr)
9236{
9237 if (!tr)
9238 return;
9239
9240 free_trace_buffer(&tr->array_buffer);
9241
9242#ifdef CONFIG_TRACER_MAX_TRACE
9243 free_trace_buffer(&tr->max_buffer);
9244#endif
9245}
9246
9247static void init_trace_flags_index(struct trace_array *tr)
9248{
9249 int i;
9250
9251 /* Used by the trace options files */
9252 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9253 tr->trace_flags_index[i] = i;
9254}
9255
9256static void __update_tracer_options(struct trace_array *tr)
9257{
9258 struct tracer *t;
9259
9260 for (t = trace_types; t; t = t->next)
9261 add_tracer_options(tr, t);
9262}
9263
9264static void update_tracer_options(struct trace_array *tr)
9265{
9266 mutex_lock(&trace_types_lock);
9267 tracer_options_updated = true;
9268 __update_tracer_options(tr);
9269 mutex_unlock(&trace_types_lock);
9270}
9271
9272/* Must have trace_types_lock held */
9273struct trace_array *trace_array_find(const char *instance)
9274{
9275 struct trace_array *tr, *found = NULL;
9276
9277 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9278 if (tr->name && strcmp(tr->name, instance) == 0) {
9279 found = tr;
9280 break;
9281 }
9282 }
9283
9284 return found;
9285}
9286
9287struct trace_array *trace_array_find_get(const char *instance)
9288{
9289 struct trace_array *tr;
9290
9291 mutex_lock(&trace_types_lock);
9292 tr = trace_array_find(instance);
9293 if (tr)
9294 tr->ref++;
9295 mutex_unlock(&trace_types_lock);
9296
9297 return tr;
9298}
9299
9300static int trace_array_create_dir(struct trace_array *tr)
9301{
9302 int ret;
9303
9304 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9305 if (!tr->dir)
9306 return -EINVAL;
9307
9308 ret = event_trace_add_tracer(tr->dir, tr);
9309 if (ret) {
9310 tracefs_remove(tr->dir);
9311 return ret;
9312 }
9313
9314 init_tracer_tracefs(tr, tr->dir);
9315 __update_tracer_options(tr);
9316
9317 return ret;
9318}
9319
9320static struct trace_array *trace_array_create(const char *name)
9321{
9322 struct trace_array *tr;
9323 int ret;
9324
9325 ret = -ENOMEM;
9326 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9327 if (!tr)
9328 return ERR_PTR(ret);
9329
9330 tr->name = kstrdup(name, GFP_KERNEL);
9331 if (!tr->name)
9332 goto out_free_tr;
9333
9334 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9335 goto out_free_tr;
9336
9337 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9338
9339 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9340
9341 raw_spin_lock_init(&tr->start_lock);
9342
9343 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9344
9345 tr->current_trace = &nop_trace;
9346
9347 INIT_LIST_HEAD(&tr->systems);
9348 INIT_LIST_HEAD(&tr->events);
9349 INIT_LIST_HEAD(&tr->hist_vars);
9350 INIT_LIST_HEAD(&tr->err_log);
9351
9352 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9353 goto out_free_tr;
9354
9355 if (ftrace_allocate_ftrace_ops(tr) < 0)
9356 goto out_free_tr;
9357
9358 ftrace_init_trace_array(tr);
9359
9360 init_trace_flags_index(tr);
9361
9362 if (trace_instance_dir) {
9363 ret = trace_array_create_dir(tr);
9364 if (ret)
9365 goto out_free_tr;
9366 } else
9367 __trace_early_add_events(tr);
9368
9369 list_add(&tr->list, &ftrace_trace_arrays);
9370
9371 tr->ref++;
9372
9373 return tr;
9374
9375 out_free_tr:
9376 ftrace_free_ftrace_ops(tr);
9377 free_trace_buffers(tr);
9378 free_cpumask_var(tr->tracing_cpumask);
9379 kfree(tr->name);
9380 kfree(tr);
9381
9382 return ERR_PTR(ret);
9383}
9384
9385static int instance_mkdir(const char *name)
9386{
9387 struct trace_array *tr;
9388 int ret;
9389
9390 mutex_lock(&event_mutex);
9391 mutex_lock(&trace_types_lock);
9392
9393 ret = -EEXIST;
9394 if (trace_array_find(name))
9395 goto out_unlock;
9396
9397 tr = trace_array_create(name);
9398
9399 ret = PTR_ERR_OR_ZERO(tr);
9400
9401out_unlock:
9402 mutex_unlock(&trace_types_lock);
9403 mutex_unlock(&event_mutex);
9404 return ret;
9405}
9406
9407/**
9408 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9409 * @name: The name of the trace array to be looked up/created.
9410 *
9411 * Returns pointer to trace array with given name.
9412 * NULL, if it cannot be created.
9413 *
9414 * NOTE: This function increments the reference counter associated with the
9415 * trace array returned. This makes sure it cannot be freed while in use.
9416 * Use trace_array_put() once the trace array is no longer needed.
9417 * If the trace_array is to be freed, trace_array_destroy() needs to
9418 * be called after the trace_array_put(), or simply let user space delete
9419 * it from the tracefs instances directory. But until the
9420 * trace_array_put() is called, user space can not delete it.
9421 *
9422 */
9423struct trace_array *trace_array_get_by_name(const char *name)
9424{
9425 struct trace_array *tr;
9426
9427 mutex_lock(&event_mutex);
9428 mutex_lock(&trace_types_lock);
9429
9430 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9431 if (tr->name && strcmp(tr->name, name) == 0)
9432 goto out_unlock;
9433 }
9434
9435 tr = trace_array_create(name);
9436
9437 if (IS_ERR(tr))
9438 tr = NULL;
9439out_unlock:
9440 if (tr)
9441 tr->ref++;
9442
9443 mutex_unlock(&trace_types_lock);
9444 mutex_unlock(&event_mutex);
9445 return tr;
9446}
9447EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9448
9449static int __remove_instance(struct trace_array *tr)
9450{
9451 int i;
9452
9453 /* Reference counter for a newly created trace array = 1. */
9454 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9455 return -EBUSY;
9456
9457 list_del(&tr->list);
9458
9459 /* Disable all the flags that were enabled coming in */
9460 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9461 if ((1 << i) & ZEROED_TRACE_FLAGS)
9462 set_tracer_flag(tr, 1 << i, 0);
9463 }
9464
9465 tracing_set_nop(tr);
9466 clear_ftrace_function_probes(tr);
9467 event_trace_del_tracer(tr);
9468 ftrace_clear_pids(tr);
9469 ftrace_destroy_function_files(tr);
9470 tracefs_remove(tr->dir);
9471 free_percpu(tr->last_func_repeats);
9472 free_trace_buffers(tr);
9473
9474 for (i = 0; i < tr->nr_topts; i++) {
9475 kfree(tr->topts[i].topts);
9476 }
9477 kfree(tr->topts);
9478
9479 free_cpumask_var(tr->tracing_cpumask);
9480 kfree(tr->name);
9481 kfree(tr);
9482
9483 return 0;
9484}
9485
9486int trace_array_destroy(struct trace_array *this_tr)
9487{
9488 struct trace_array *tr;
9489 int ret;
9490
9491 if (!this_tr)
9492 return -EINVAL;
9493
9494 mutex_lock(&event_mutex);
9495 mutex_lock(&trace_types_lock);
9496
9497 ret = -ENODEV;
9498
9499 /* Making sure trace array exists before destroying it. */
9500 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9501 if (tr == this_tr) {
9502 ret = __remove_instance(tr);
9503 break;
9504 }
9505 }
9506
9507 mutex_unlock(&trace_types_lock);
9508 mutex_unlock(&event_mutex);
9509
9510 return ret;
9511}
9512EXPORT_SYMBOL_GPL(trace_array_destroy);
9513
9514static int instance_rmdir(const char *name)
9515{
9516 struct trace_array *tr;
9517 int ret;
9518
9519 mutex_lock(&event_mutex);
9520 mutex_lock(&trace_types_lock);
9521
9522 ret = -ENODEV;
9523 tr = trace_array_find(name);
9524 if (tr)
9525 ret = __remove_instance(tr);
9526
9527 mutex_unlock(&trace_types_lock);
9528 mutex_unlock(&event_mutex);
9529
9530 return ret;
9531}
9532
9533static __init void create_trace_instances(struct dentry *d_tracer)
9534{
9535 struct trace_array *tr;
9536
9537 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9538 instance_mkdir,
9539 instance_rmdir);
9540 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9541 return;
9542
9543 mutex_lock(&event_mutex);
9544 mutex_lock(&trace_types_lock);
9545
9546 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9547 if (!tr->name)
9548 continue;
9549 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9550 "Failed to create instance directory\n"))
9551 break;
9552 }
9553
9554 mutex_unlock(&trace_types_lock);
9555 mutex_unlock(&event_mutex);
9556}
9557
9558static void
9559init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9560{
9561 struct trace_event_file *file;
9562 int cpu;
9563
9564 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9565 tr, &show_traces_fops);
9566
9567 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9568 tr, &set_tracer_fops);
9569
9570 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9571 tr, &tracing_cpumask_fops);
9572
9573 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9574 tr, &tracing_iter_fops);
9575
9576 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9577 tr, &tracing_fops);
9578
9579 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9580 tr, &tracing_pipe_fops);
9581
9582 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9583 tr, &tracing_entries_fops);
9584
9585 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9586 tr, &tracing_total_entries_fops);
9587
9588 trace_create_file("free_buffer", 0200, d_tracer,
9589 tr, &tracing_free_buffer_fops);
9590
9591 trace_create_file("trace_marker", 0220, d_tracer,
9592 tr, &tracing_mark_fops);
9593
9594 file = __find_event_file(tr, "ftrace", "print");
9595 if (file && file->dir)
9596 trace_create_file("trigger", TRACE_MODE_WRITE, file->dir,
9597 file, &event_trigger_fops);
9598 tr->trace_marker_file = file;
9599
9600 trace_create_file("trace_marker_raw", 0220, d_tracer,
9601 tr, &tracing_mark_raw_fops);
9602
9603 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9604 &trace_clock_fops);
9605
9606 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9607 tr, &rb_simple_fops);
9608
9609 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9610 &trace_time_stamp_mode_fops);
9611
9612 tr->buffer_percent = 50;
9613
9614 trace_create_file("buffer_percent", TRACE_MODE_READ, d_tracer,
9615 tr, &buffer_percent_fops);
9616
9617 create_trace_options_dir(tr);
9618
9619#ifdef CONFIG_TRACER_MAX_TRACE
9620 trace_create_maxlat_file(tr, d_tracer);
9621#endif
9622
9623 if (ftrace_create_function_files(tr, d_tracer))
9624 MEM_FAIL(1, "Could not allocate function filter files");
9625
9626#ifdef CONFIG_TRACER_SNAPSHOT
9627 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9628 tr, &snapshot_fops);
9629#endif
9630
9631 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9632 tr, &tracing_err_log_fops);
9633
9634 for_each_tracing_cpu(cpu)
9635 tracing_init_tracefs_percpu(tr, cpu);
9636
9637 ftrace_init_tracefs(tr, d_tracer);
9638}
9639
9640static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9641{
9642 struct vfsmount *mnt;
9643 struct file_system_type *type;
9644
9645 /*
9646 * To maintain backward compatibility for tools that mount
9647 * debugfs to get to the tracing facility, tracefs is automatically
9648 * mounted to the debugfs/tracing directory.
9649 */
9650 type = get_fs_type("tracefs");
9651 if (!type)
9652 return NULL;
9653 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9654 put_filesystem(type);
9655 if (IS_ERR(mnt))
9656 return NULL;
9657 mntget(mnt);
9658
9659 return mnt;
9660}
9661
9662/**
9663 * tracing_init_dentry - initialize top level trace array
9664 *
9665 * This is called when creating files or directories in the tracing
9666 * directory. It is called via fs_initcall() by any of the boot up code
9667 * and expects to return the dentry of the top level tracing directory.
9668 */
9669int tracing_init_dentry(void)
9670{
9671 struct trace_array *tr = &global_trace;
9672
9673 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9674 pr_warn("Tracing disabled due to lockdown\n");
9675 return -EPERM;
9676 }
9677
9678 /* The top level trace array uses NULL as parent */
9679 if (tr->dir)
9680 return 0;
9681
9682 if (WARN_ON(!tracefs_initialized()))
9683 return -ENODEV;
9684
9685 /*
9686 * As there may still be users that expect the tracing
9687 * files to exist in debugfs/tracing, we must automount
9688 * the tracefs file system there, so older tools still
9689 * work with the newer kernel.
9690 */
9691 tr->dir = debugfs_create_automount("tracing", NULL,
9692 trace_automount, NULL);
9693
9694 return 0;
9695}
9696
9697extern struct trace_eval_map *__start_ftrace_eval_maps[];
9698extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9699
9700static struct workqueue_struct *eval_map_wq __initdata;
9701static struct work_struct eval_map_work __initdata;
9702static struct work_struct tracerfs_init_work __initdata;
9703
9704static void __init eval_map_work_func(struct work_struct *work)
9705{
9706 int len;
9707
9708 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9709 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9710}
9711
9712static int __init trace_eval_init(void)
9713{
9714 INIT_WORK(&eval_map_work, eval_map_work_func);
9715
9716 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9717 if (!eval_map_wq) {
9718 pr_err("Unable to allocate eval_map_wq\n");
9719 /* Do work here */
9720 eval_map_work_func(&eval_map_work);
9721 return -ENOMEM;
9722 }
9723
9724 queue_work(eval_map_wq, &eval_map_work);
9725 return 0;
9726}
9727
9728subsys_initcall(trace_eval_init);
9729
9730static int __init trace_eval_sync(void)
9731{
9732 /* Make sure the eval map updates are finished */
9733 if (eval_map_wq)
9734 destroy_workqueue(eval_map_wq);
9735 return 0;
9736}
9737
9738late_initcall_sync(trace_eval_sync);
9739
9740
9741#ifdef CONFIG_MODULES
9742static void trace_module_add_evals(struct module *mod)
9743{
9744 if (!mod->num_trace_evals)
9745 return;
9746
9747 /*
9748 * Modules with bad taint do not have events created, do
9749 * not bother with enums either.
9750 */
9751 if (trace_module_has_bad_taint(mod))
9752 return;
9753
9754 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9755}
9756
9757#ifdef CONFIG_TRACE_EVAL_MAP_FILE
9758static void trace_module_remove_evals(struct module *mod)
9759{
9760 union trace_eval_map_item *map;
9761 union trace_eval_map_item **last = &trace_eval_maps;
9762
9763 if (!mod->num_trace_evals)
9764 return;
9765
9766 mutex_lock(&trace_eval_mutex);
9767
9768 map = trace_eval_maps;
9769
9770 while (map) {
9771 if (map->head.mod == mod)
9772 break;
9773 map = trace_eval_jmp_to_tail(map);
9774 last = &map->tail.next;
9775 map = map->tail.next;
9776 }
9777 if (!map)
9778 goto out;
9779
9780 *last = trace_eval_jmp_to_tail(map)->tail.next;
9781 kfree(map);
9782 out:
9783 mutex_unlock(&trace_eval_mutex);
9784}
9785#else
9786static inline void trace_module_remove_evals(struct module *mod) { }
9787#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9788
9789static int trace_module_notify(struct notifier_block *self,
9790 unsigned long val, void *data)
9791{
9792 struct module *mod = data;
9793
9794 switch (val) {
9795 case MODULE_STATE_COMING:
9796 trace_module_add_evals(mod);
9797 break;
9798 case MODULE_STATE_GOING:
9799 trace_module_remove_evals(mod);
9800 break;
9801 }
9802
9803 return NOTIFY_OK;
9804}
9805
9806static struct notifier_block trace_module_nb = {
9807 .notifier_call = trace_module_notify,
9808 .priority = 0,
9809};
9810#endif /* CONFIG_MODULES */
9811
9812static __init void tracer_init_tracefs_work_func(struct work_struct *work)
9813{
9814
9815 event_trace_init();
9816
9817 init_tracer_tracefs(&global_trace, NULL);
9818 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9819
9820 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9821 &global_trace, &tracing_thresh_fops);
9822
9823 trace_create_file("README", TRACE_MODE_READ, NULL,
9824 NULL, &tracing_readme_fops);
9825
9826 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9827 NULL, &tracing_saved_cmdlines_fops);
9828
9829 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9830 NULL, &tracing_saved_cmdlines_size_fops);
9831
9832 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9833 NULL, &tracing_saved_tgids_fops);
9834
9835 trace_create_eval_file(NULL);
9836
9837#ifdef CONFIG_MODULES
9838 register_module_notifier(&trace_module_nb);
9839#endif
9840
9841#ifdef CONFIG_DYNAMIC_FTRACE
9842 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
9843 NULL, &tracing_dyn_info_fops);
9844#endif
9845
9846 create_trace_instances(NULL);
9847
9848 update_tracer_options(&global_trace);
9849}
9850
9851static __init int tracer_init_tracefs(void)
9852{
9853 int ret;
9854
9855 trace_access_lock_init();
9856
9857 ret = tracing_init_dentry();
9858 if (ret)
9859 return 0;
9860
9861 if (eval_map_wq) {
9862 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
9863 queue_work(eval_map_wq, &tracerfs_init_work);
9864 } else {
9865 tracer_init_tracefs_work_func(NULL);
9866 }
9867
9868 rv_init_interface();
9869
9870 return 0;
9871}
9872
9873fs_initcall(tracer_init_tracefs);
9874
9875static int trace_die_panic_handler(struct notifier_block *self,
9876 unsigned long ev, void *unused);
9877
9878static struct notifier_block trace_panic_notifier = {
9879 .notifier_call = trace_die_panic_handler,
9880 .priority = INT_MAX - 1,
9881};
9882
9883static struct notifier_block trace_die_notifier = {
9884 .notifier_call = trace_die_panic_handler,
9885 .priority = INT_MAX - 1,
9886};
9887
9888/*
9889 * The idea is to execute the following die/panic callback early, in order
9890 * to avoid showing irrelevant information in the trace (like other panic
9891 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
9892 * warnings get disabled (to prevent potential log flooding).
9893 */
9894static int trace_die_panic_handler(struct notifier_block *self,
9895 unsigned long ev, void *unused)
9896{
9897 if (!ftrace_dump_on_oops)
9898 return NOTIFY_DONE;
9899
9900 /* The die notifier requires DIE_OOPS to trigger */
9901 if (self == &trace_die_notifier && ev != DIE_OOPS)
9902 return NOTIFY_DONE;
9903
9904 ftrace_dump(ftrace_dump_on_oops);
9905
9906 return NOTIFY_DONE;
9907}
9908
9909/*
9910 * printk is set to max of 1024, we really don't need it that big.
9911 * Nothing should be printing 1000 characters anyway.
9912 */
9913#define TRACE_MAX_PRINT 1000
9914
9915/*
9916 * Define here KERN_TRACE so that we have one place to modify
9917 * it if we decide to change what log level the ftrace dump
9918 * should be at.
9919 */
9920#define KERN_TRACE KERN_EMERG
9921
9922void
9923trace_printk_seq(struct trace_seq *s)
9924{
9925 /* Probably should print a warning here. */
9926 if (s->seq.len >= TRACE_MAX_PRINT)
9927 s->seq.len = TRACE_MAX_PRINT;
9928
9929 /*
9930 * More paranoid code. Although the buffer size is set to
9931 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9932 * an extra layer of protection.
9933 */
9934 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9935 s->seq.len = s->seq.size - 1;
9936
9937 /* should be zero ended, but we are paranoid. */
9938 s->buffer[s->seq.len] = 0;
9939
9940 printk(KERN_TRACE "%s", s->buffer);
9941
9942 trace_seq_init(s);
9943}
9944
9945void trace_init_global_iter(struct trace_iterator *iter)
9946{
9947 iter->tr = &global_trace;
9948 iter->trace = iter->tr->current_trace;
9949 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9950 iter->array_buffer = &global_trace.array_buffer;
9951
9952 if (iter->trace && iter->trace->open)
9953 iter->trace->open(iter);
9954
9955 /* Annotate start of buffers if we had overruns */
9956 if (ring_buffer_overruns(iter->array_buffer->buffer))
9957 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9958
9959 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9960 if (trace_clocks[iter->tr->clock_id].in_ns)
9961 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9962
9963 /* Can not use kmalloc for iter.temp and iter.fmt */
9964 iter->temp = static_temp_buf;
9965 iter->temp_size = STATIC_TEMP_BUF_SIZE;
9966 iter->fmt = static_fmt_buf;
9967 iter->fmt_size = STATIC_FMT_BUF_SIZE;
9968}
9969
9970void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
9971{
9972 /* use static because iter can be a bit big for the stack */
9973 static struct trace_iterator iter;
9974 static atomic_t dump_running;
9975 struct trace_array *tr = &global_trace;
9976 unsigned int old_userobj;
9977 unsigned long flags;
9978 int cnt = 0, cpu;
9979
9980 /* Only allow one dump user at a time. */
9981 if (atomic_inc_return(&dump_running) != 1) {
9982 atomic_dec(&dump_running);
9983 return;
9984 }
9985
9986 /*
9987 * Always turn off tracing when we dump.
9988 * We don't need to show trace output of what happens
9989 * between multiple crashes.
9990 *
9991 * If the user does a sysrq-z, then they can re-enable
9992 * tracing with echo 1 > tracing_on.
9993 */
9994 tracing_off();
9995
9996 local_irq_save(flags);
9997
9998 /* Simulate the iterator */
9999 trace_init_global_iter(&iter);
10000
10001 for_each_tracing_cpu(cpu) {
10002 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10003 }
10004
10005 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10006
10007 /* don't look at user memory in panic mode */
10008 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10009
10010 switch (oops_dump_mode) {
10011 case DUMP_ALL:
10012 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10013 break;
10014 case DUMP_ORIG:
10015 iter.cpu_file = raw_smp_processor_id();
10016 break;
10017 case DUMP_NONE:
10018 goto out_enable;
10019 default:
10020 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10021 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10022 }
10023
10024 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10025
10026 /* Did function tracer already get disabled? */
10027 if (ftrace_is_dead()) {
10028 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10029 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10030 }
10031
10032 /*
10033 * We need to stop all tracing on all CPUS to read
10034 * the next buffer. This is a bit expensive, but is
10035 * not done often. We fill all what we can read,
10036 * and then release the locks again.
10037 */
10038
10039 while (!trace_empty(&iter)) {
10040
10041 if (!cnt)
10042 printk(KERN_TRACE "---------------------------------\n");
10043
10044 cnt++;
10045
10046 trace_iterator_reset(&iter);
10047 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10048
10049 if (trace_find_next_entry_inc(&iter) != NULL) {
10050 int ret;
10051
10052 ret = print_trace_line(&iter);
10053 if (ret != TRACE_TYPE_NO_CONSUME)
10054 trace_consume(&iter);
10055 }
10056 touch_nmi_watchdog();
10057
10058 trace_printk_seq(&iter.seq);
10059 }
10060
10061 if (!cnt)
10062 printk(KERN_TRACE " (ftrace buffer empty)\n");
10063 else
10064 printk(KERN_TRACE "---------------------------------\n");
10065
10066 out_enable:
10067 tr->trace_flags |= old_userobj;
10068
10069 for_each_tracing_cpu(cpu) {
10070 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10071 }
10072 atomic_dec(&dump_running);
10073 local_irq_restore(flags);
10074}
10075EXPORT_SYMBOL_GPL(ftrace_dump);
10076
10077#define WRITE_BUFSIZE 4096
10078
10079ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10080 size_t count, loff_t *ppos,
10081 int (*createfn)(const char *))
10082{
10083 char *kbuf, *buf, *tmp;
10084 int ret = 0;
10085 size_t done = 0;
10086 size_t size;
10087
10088 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10089 if (!kbuf)
10090 return -ENOMEM;
10091
10092 while (done < count) {
10093 size = count - done;
10094
10095 if (size >= WRITE_BUFSIZE)
10096 size = WRITE_BUFSIZE - 1;
10097
10098 if (copy_from_user(kbuf, buffer + done, size)) {
10099 ret = -EFAULT;
10100 goto out;
10101 }
10102 kbuf[size] = '\0';
10103 buf = kbuf;
10104 do {
10105 tmp = strchr(buf, '\n');
10106 if (tmp) {
10107 *tmp = '\0';
10108 size = tmp - buf + 1;
10109 } else {
10110 size = strlen(buf);
10111 if (done + size < count) {
10112 if (buf != kbuf)
10113 break;
10114 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10115 pr_warn("Line length is too long: Should be less than %d\n",
10116 WRITE_BUFSIZE - 2);
10117 ret = -EINVAL;
10118 goto out;
10119 }
10120 }
10121 done += size;
10122
10123 /* Remove comments */
10124 tmp = strchr(buf, '#');
10125
10126 if (tmp)
10127 *tmp = '\0';
10128
10129 ret = createfn(buf);
10130 if (ret)
10131 goto out;
10132 buf += size;
10133
10134 } while (done < count);
10135 }
10136 ret = done;
10137
10138out:
10139 kfree(kbuf);
10140
10141 return ret;
10142}
10143
10144__init static int tracer_alloc_buffers(void)
10145{
10146 int ring_buf_size;
10147 int ret = -ENOMEM;
10148
10149
10150 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10151 pr_warn("Tracing disabled due to lockdown\n");
10152 return -EPERM;
10153 }
10154
10155 /*
10156 * Make sure we don't accidentally add more trace options
10157 * than we have bits for.
10158 */
10159 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10160
10161 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10162 goto out;
10163
10164 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10165 goto out_free_buffer_mask;
10166
10167 /* Only allocate trace_printk buffers if a trace_printk exists */
10168 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10169 /* Must be called before global_trace.buffer is allocated */
10170 trace_printk_init_buffers();
10171
10172 /* To save memory, keep the ring buffer size to its minimum */
10173 if (ring_buffer_expanded)
10174 ring_buf_size = trace_buf_size;
10175 else
10176 ring_buf_size = 1;
10177
10178 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10179 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10180
10181 raw_spin_lock_init(&global_trace.start_lock);
10182
10183 /*
10184 * The prepare callbacks allocates some memory for the ring buffer. We
10185 * don't free the buffer if the CPU goes down. If we were to free
10186 * the buffer, then the user would lose any trace that was in the
10187 * buffer. The memory will be removed once the "instance" is removed.
10188 */
10189 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10190 "trace/RB:prepare", trace_rb_cpu_prepare,
10191 NULL);
10192 if (ret < 0)
10193 goto out_free_cpumask;
10194 /* Used for event triggers */
10195 ret = -ENOMEM;
10196 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10197 if (!temp_buffer)
10198 goto out_rm_hp_state;
10199
10200 if (trace_create_savedcmd() < 0)
10201 goto out_free_temp_buffer;
10202
10203 /* TODO: make the number of buffers hot pluggable with CPUS */
10204 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10205 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10206 goto out_free_savedcmd;
10207 }
10208
10209 if (global_trace.buffer_disabled)
10210 tracing_off();
10211
10212 if (trace_boot_clock) {
10213 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10214 if (ret < 0)
10215 pr_warn("Trace clock %s not defined, going back to default\n",
10216 trace_boot_clock);
10217 }
10218
10219 /*
10220 * register_tracer() might reference current_trace, so it
10221 * needs to be set before we register anything. This is
10222 * just a bootstrap of current_trace anyway.
10223 */
10224 global_trace.current_trace = &nop_trace;
10225
10226 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10227
10228 ftrace_init_global_array_ops(&global_trace);
10229
10230 init_trace_flags_index(&global_trace);
10231
10232 register_tracer(&nop_trace);
10233
10234 /* Function tracing may start here (via kernel command line) */
10235 init_function_trace();
10236
10237 /* All seems OK, enable tracing */
10238 tracing_disabled = 0;
10239
10240 atomic_notifier_chain_register(&panic_notifier_list,
10241 &trace_panic_notifier);
10242
10243 register_die_notifier(&trace_die_notifier);
10244
10245 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10246
10247 INIT_LIST_HEAD(&global_trace.systems);
10248 INIT_LIST_HEAD(&global_trace.events);
10249 INIT_LIST_HEAD(&global_trace.hist_vars);
10250 INIT_LIST_HEAD(&global_trace.err_log);
10251 list_add(&global_trace.list, &ftrace_trace_arrays);
10252
10253 apply_trace_boot_options();
10254
10255 register_snapshot_cmd();
10256
10257 test_can_verify();
10258
10259 return 0;
10260
10261out_free_savedcmd:
10262 free_saved_cmdlines_buffer(savedcmd);
10263out_free_temp_buffer:
10264 ring_buffer_free(temp_buffer);
10265out_rm_hp_state:
10266 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10267out_free_cpumask:
10268 free_cpumask_var(global_trace.tracing_cpumask);
10269out_free_buffer_mask:
10270 free_cpumask_var(tracing_buffer_mask);
10271out:
10272 return ret;
10273}
10274
10275void __init ftrace_boot_snapshot(void)
10276{
10277 if (snapshot_at_boot) {
10278 tracing_snapshot();
10279 internal_trace_puts("** Boot snapshot taken **\n");
10280 }
10281}
10282
10283void __init early_trace_init(void)
10284{
10285 if (tracepoint_printk) {
10286 tracepoint_print_iter =
10287 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10288 if (MEM_FAIL(!tracepoint_print_iter,
10289 "Failed to allocate trace iterator\n"))
10290 tracepoint_printk = 0;
10291 else
10292 static_key_enable(&tracepoint_printk_key.key);
10293 }
10294 tracer_alloc_buffers();
10295
10296 init_events();
10297}
10298
10299void __init trace_init(void)
10300{
10301 trace_event_init();
10302}
10303
10304__init static void clear_boot_tracer(void)
10305{
10306 /*
10307 * The default tracer at boot buffer is an init section.
10308 * This function is called in lateinit. If we did not
10309 * find the boot tracer, then clear it out, to prevent
10310 * later registration from accessing the buffer that is
10311 * about to be freed.
10312 */
10313 if (!default_bootup_tracer)
10314 return;
10315
10316 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10317 default_bootup_tracer);
10318 default_bootup_tracer = NULL;
10319}
10320
10321#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10322__init static void tracing_set_default_clock(void)
10323{
10324 /* sched_clock_stable() is determined in late_initcall */
10325 if (!trace_boot_clock && !sched_clock_stable()) {
10326 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10327 pr_warn("Can not set tracing clock due to lockdown\n");
10328 return;
10329 }
10330
10331 printk(KERN_WARNING
10332 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10333 "If you want to keep using the local clock, then add:\n"
10334 " \"trace_clock=local\"\n"
10335 "on the kernel command line\n");
10336 tracing_set_clock(&global_trace, "global");
10337 }
10338}
10339#else
10340static inline void tracing_set_default_clock(void) { }
10341#endif
10342
10343__init static int late_trace_init(void)
10344{
10345 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10346 static_key_disable(&tracepoint_printk_key.key);
10347 tracepoint_printk = 0;
10348 }
10349
10350 tracing_set_default_clock();
10351 clear_boot_tracer();
10352 return 0;
10353}
10354
10355late_initcall_sync(late_trace_init);