Loading...
1/*
2 * ring buffer based function tracer
3 *
4 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6 *
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
9 *
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 Nadia Yvette Chambers
13 */
14#include <linux/ring_buffer.h>
15#include <generated/utsrelease.h>
16#include <linux/stacktrace.h>
17#include <linux/writeback.h>
18#include <linux/kallsyms.h>
19#include <linux/seq_file.h>
20#include <linux/notifier.h>
21#include <linux/irqflags.h>
22#include <linux/debugfs.h>
23#include <linux/tracefs.h>
24#include <linux/pagemap.h>
25#include <linux/hardirq.h>
26#include <linux/linkage.h>
27#include <linux/uaccess.h>
28#include <linux/kprobes.h>
29#include <linux/ftrace.h>
30#include <linux/module.h>
31#include <linux/percpu.h>
32#include <linux/splice.h>
33#include <linux/kdebug.h>
34#include <linux/string.h>
35#include <linux/mount.h>
36#include <linux/rwsem.h>
37#include <linux/slab.h>
38#include <linux/ctype.h>
39#include <linux/init.h>
40#include <linux/poll.h>
41#include <linux/nmi.h>
42#include <linux/fs.h>
43#include <linux/sched/rt.h>
44
45#include "trace.h"
46#include "trace_output.h"
47
48/*
49 * On boot up, the ring buffer is set to the minimum size, so that
50 * we do not waste memory on systems that are not using tracing.
51 */
52bool ring_buffer_expanded;
53
54/*
55 * We need to change this state when a selftest is running.
56 * A selftest will lurk into the ring-buffer to count the
57 * entries inserted during the selftest although some concurrent
58 * insertions into the ring-buffer such as trace_printk could occurred
59 * at the same time, giving false positive or negative results.
60 */
61static bool __read_mostly tracing_selftest_running;
62
63/*
64 * If a tracer is running, we do not want to run SELFTEST.
65 */
66bool __read_mostly tracing_selftest_disabled;
67
68/* Pipe tracepoints to printk */
69struct trace_iterator *tracepoint_print_iter;
70int tracepoint_printk;
71
72/* For tracers that don't implement custom flags */
73static struct tracer_opt dummy_tracer_opt[] = {
74 { }
75};
76
77static int
78dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
79{
80 return 0;
81}
82
83/*
84 * To prevent the comm cache from being overwritten when no
85 * tracing is active, only save the comm when a trace event
86 * occurred.
87 */
88static DEFINE_PER_CPU(bool, trace_cmdline_save);
89
90/*
91 * Kill all tracing for good (never come back).
92 * It is initialized to 1 but will turn to zero if the initialization
93 * of the tracer is successful. But that is the only place that sets
94 * this back to zero.
95 */
96static int tracing_disabled = 1;
97
98cpumask_var_t __read_mostly tracing_buffer_mask;
99
100/*
101 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
102 *
103 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
104 * is set, then ftrace_dump is called. This will output the contents
105 * of the ftrace buffers to the console. This is very useful for
106 * capturing traces that lead to crashes and outputing it to a
107 * serial console.
108 *
109 * It is default off, but you can enable it with either specifying
110 * "ftrace_dump_on_oops" in the kernel command line, or setting
111 * /proc/sys/kernel/ftrace_dump_on_oops
112 * Set 1 if you want to dump buffers of all CPUs
113 * Set 2 if you want to dump the buffer of the CPU that triggered oops
114 */
115
116enum ftrace_dump_mode ftrace_dump_on_oops;
117
118/* When set, tracing will stop when a WARN*() is hit */
119int __disable_trace_on_warning;
120
121#ifdef CONFIG_TRACE_ENUM_MAP_FILE
122/* Map of enums to their values, for "enum_map" file */
123struct trace_enum_map_head {
124 struct module *mod;
125 unsigned long length;
126};
127
128union trace_enum_map_item;
129
130struct trace_enum_map_tail {
131 /*
132 * "end" is first and points to NULL as it must be different
133 * than "mod" or "enum_string"
134 */
135 union trace_enum_map_item *next;
136 const char *end; /* points to NULL */
137};
138
139static DEFINE_MUTEX(trace_enum_mutex);
140
141/*
142 * The trace_enum_maps are saved in an array with two extra elements,
143 * one at the beginning, and one at the end. The beginning item contains
144 * the count of the saved maps (head.length), and the module they
145 * belong to if not built in (head.mod). The ending item contains a
146 * pointer to the next array of saved enum_map items.
147 */
148union trace_enum_map_item {
149 struct trace_enum_map map;
150 struct trace_enum_map_head head;
151 struct trace_enum_map_tail tail;
152};
153
154static union trace_enum_map_item *trace_enum_maps;
155#endif /* CONFIG_TRACE_ENUM_MAP_FILE */
156
157static int tracing_set_tracer(struct trace_array *tr, const char *buf);
158
159#define MAX_TRACER_SIZE 100
160static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
161static char *default_bootup_tracer;
162
163static bool allocate_snapshot;
164
165static int __init set_cmdline_ftrace(char *str)
166{
167 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
168 default_bootup_tracer = bootup_tracer_buf;
169 /* We are using ftrace early, expand it */
170 ring_buffer_expanded = true;
171 return 1;
172}
173__setup("ftrace=", set_cmdline_ftrace);
174
175static int __init set_ftrace_dump_on_oops(char *str)
176{
177 if (*str++ != '=' || !*str) {
178 ftrace_dump_on_oops = DUMP_ALL;
179 return 1;
180 }
181
182 if (!strcmp("orig_cpu", str)) {
183 ftrace_dump_on_oops = DUMP_ORIG;
184 return 1;
185 }
186
187 return 0;
188}
189__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
190
191static int __init stop_trace_on_warning(char *str)
192{
193 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
194 __disable_trace_on_warning = 1;
195 return 1;
196}
197__setup("traceoff_on_warning", stop_trace_on_warning);
198
199static int __init boot_alloc_snapshot(char *str)
200{
201 allocate_snapshot = true;
202 /* We also need the main ring buffer expanded */
203 ring_buffer_expanded = true;
204 return 1;
205}
206__setup("alloc_snapshot", boot_alloc_snapshot);
207
208
209static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
210
211static int __init set_trace_boot_options(char *str)
212{
213 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
214 return 0;
215}
216__setup("trace_options=", set_trace_boot_options);
217
218static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
219static char *trace_boot_clock __initdata;
220
221static int __init set_trace_boot_clock(char *str)
222{
223 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
224 trace_boot_clock = trace_boot_clock_buf;
225 return 0;
226}
227__setup("trace_clock=", set_trace_boot_clock);
228
229static int __init set_tracepoint_printk(char *str)
230{
231 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
232 tracepoint_printk = 1;
233 return 1;
234}
235__setup("tp_printk", set_tracepoint_printk);
236
237unsigned long long ns2usecs(cycle_t nsec)
238{
239 nsec += 500;
240 do_div(nsec, 1000);
241 return nsec;
242}
243
244/* trace_flags holds trace_options default values */
245#define TRACE_DEFAULT_FLAGS \
246 (FUNCTION_DEFAULT_FLAGS | \
247 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
248 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
249 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
250 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS)
251
252/* trace_options that are only supported by global_trace */
253#define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
254 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
255
256
257/*
258 * The global_trace is the descriptor that holds the tracing
259 * buffers for the live tracing. For each CPU, it contains
260 * a link list of pages that will store trace entries. The
261 * page descriptor of the pages in the memory is used to hold
262 * the link list by linking the lru item in the page descriptor
263 * to each of the pages in the buffer per CPU.
264 *
265 * For each active CPU there is a data field that holds the
266 * pages for the buffer for that CPU. Each CPU has the same number
267 * of pages allocated for its buffer.
268 */
269static struct trace_array global_trace = {
270 .trace_flags = TRACE_DEFAULT_FLAGS,
271};
272
273LIST_HEAD(ftrace_trace_arrays);
274
275int trace_array_get(struct trace_array *this_tr)
276{
277 struct trace_array *tr;
278 int ret = -ENODEV;
279
280 mutex_lock(&trace_types_lock);
281 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
282 if (tr == this_tr) {
283 tr->ref++;
284 ret = 0;
285 break;
286 }
287 }
288 mutex_unlock(&trace_types_lock);
289
290 return ret;
291}
292
293static void __trace_array_put(struct trace_array *this_tr)
294{
295 WARN_ON(!this_tr->ref);
296 this_tr->ref--;
297}
298
299void trace_array_put(struct trace_array *this_tr)
300{
301 mutex_lock(&trace_types_lock);
302 __trace_array_put(this_tr);
303 mutex_unlock(&trace_types_lock);
304}
305
306int filter_check_discard(struct trace_event_file *file, void *rec,
307 struct ring_buffer *buffer,
308 struct ring_buffer_event *event)
309{
310 if (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
311 !filter_match_preds(file->filter, rec)) {
312 ring_buffer_discard_commit(buffer, event);
313 return 1;
314 }
315
316 return 0;
317}
318EXPORT_SYMBOL_GPL(filter_check_discard);
319
320int call_filter_check_discard(struct trace_event_call *call, void *rec,
321 struct ring_buffer *buffer,
322 struct ring_buffer_event *event)
323{
324 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
325 !filter_match_preds(call->filter, rec)) {
326 ring_buffer_discard_commit(buffer, event);
327 return 1;
328 }
329
330 return 0;
331}
332EXPORT_SYMBOL_GPL(call_filter_check_discard);
333
334static cycle_t buffer_ftrace_now(struct trace_buffer *buf, int cpu)
335{
336 u64 ts;
337
338 /* Early boot up does not have a buffer yet */
339 if (!buf->buffer)
340 return trace_clock_local();
341
342 ts = ring_buffer_time_stamp(buf->buffer, cpu);
343 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
344
345 return ts;
346}
347
348cycle_t ftrace_now(int cpu)
349{
350 return buffer_ftrace_now(&global_trace.trace_buffer, cpu);
351}
352
353/**
354 * tracing_is_enabled - Show if global_trace has been disabled
355 *
356 * Shows if the global trace has been enabled or not. It uses the
357 * mirror flag "buffer_disabled" to be used in fast paths such as for
358 * the irqsoff tracer. But it may be inaccurate due to races. If you
359 * need to know the accurate state, use tracing_is_on() which is a little
360 * slower, but accurate.
361 */
362int tracing_is_enabled(void)
363{
364 /*
365 * For quick access (irqsoff uses this in fast path), just
366 * return the mirror variable of the state of the ring buffer.
367 * It's a little racy, but we don't really care.
368 */
369 smp_rmb();
370 return !global_trace.buffer_disabled;
371}
372
373/*
374 * trace_buf_size is the size in bytes that is allocated
375 * for a buffer. Note, the number of bytes is always rounded
376 * to page size.
377 *
378 * This number is purposely set to a low number of 16384.
379 * If the dump on oops happens, it will be much appreciated
380 * to not have to wait for all that output. Anyway this can be
381 * boot time and run time configurable.
382 */
383#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
384
385static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
386
387/* trace_types holds a link list of available tracers. */
388static struct tracer *trace_types __read_mostly;
389
390/*
391 * trace_types_lock is used to protect the trace_types list.
392 */
393DEFINE_MUTEX(trace_types_lock);
394
395/*
396 * serialize the access of the ring buffer
397 *
398 * ring buffer serializes readers, but it is low level protection.
399 * The validity of the events (which returns by ring_buffer_peek() ..etc)
400 * are not protected by ring buffer.
401 *
402 * The content of events may become garbage if we allow other process consumes
403 * these events concurrently:
404 * A) the page of the consumed events may become a normal page
405 * (not reader page) in ring buffer, and this page will be rewrited
406 * by events producer.
407 * B) The page of the consumed events may become a page for splice_read,
408 * and this page will be returned to system.
409 *
410 * These primitives allow multi process access to different cpu ring buffer
411 * concurrently.
412 *
413 * These primitives don't distinguish read-only and read-consume access.
414 * Multi read-only access are also serialized.
415 */
416
417#ifdef CONFIG_SMP
418static DECLARE_RWSEM(all_cpu_access_lock);
419static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
420
421static inline void trace_access_lock(int cpu)
422{
423 if (cpu == RING_BUFFER_ALL_CPUS) {
424 /* gain it for accessing the whole ring buffer. */
425 down_write(&all_cpu_access_lock);
426 } else {
427 /* gain it for accessing a cpu ring buffer. */
428
429 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
430 down_read(&all_cpu_access_lock);
431
432 /* Secondly block other access to this @cpu ring buffer. */
433 mutex_lock(&per_cpu(cpu_access_lock, cpu));
434 }
435}
436
437static inline void trace_access_unlock(int cpu)
438{
439 if (cpu == RING_BUFFER_ALL_CPUS) {
440 up_write(&all_cpu_access_lock);
441 } else {
442 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
443 up_read(&all_cpu_access_lock);
444 }
445}
446
447static inline void trace_access_lock_init(void)
448{
449 int cpu;
450
451 for_each_possible_cpu(cpu)
452 mutex_init(&per_cpu(cpu_access_lock, cpu));
453}
454
455#else
456
457static DEFINE_MUTEX(access_lock);
458
459static inline void trace_access_lock(int cpu)
460{
461 (void)cpu;
462 mutex_lock(&access_lock);
463}
464
465static inline void trace_access_unlock(int cpu)
466{
467 (void)cpu;
468 mutex_unlock(&access_lock);
469}
470
471static inline void trace_access_lock_init(void)
472{
473}
474
475#endif
476
477#ifdef CONFIG_STACKTRACE
478static void __ftrace_trace_stack(struct ring_buffer *buffer,
479 unsigned long flags,
480 int skip, int pc, struct pt_regs *regs);
481static inline void ftrace_trace_stack(struct trace_array *tr,
482 struct ring_buffer *buffer,
483 unsigned long flags,
484 int skip, int pc, struct pt_regs *regs);
485
486#else
487static inline void __ftrace_trace_stack(struct ring_buffer *buffer,
488 unsigned long flags,
489 int skip, int pc, struct pt_regs *regs)
490{
491}
492static inline void ftrace_trace_stack(struct trace_array *tr,
493 struct ring_buffer *buffer,
494 unsigned long flags,
495 int skip, int pc, struct pt_regs *regs)
496{
497}
498
499#endif
500
501static void tracer_tracing_on(struct trace_array *tr)
502{
503 if (tr->trace_buffer.buffer)
504 ring_buffer_record_on(tr->trace_buffer.buffer);
505 /*
506 * This flag is looked at when buffers haven't been allocated
507 * yet, or by some tracers (like irqsoff), that just want to
508 * know if the ring buffer has been disabled, but it can handle
509 * races of where it gets disabled but we still do a record.
510 * As the check is in the fast path of the tracers, it is more
511 * important to be fast than accurate.
512 */
513 tr->buffer_disabled = 0;
514 /* Make the flag seen by readers */
515 smp_wmb();
516}
517
518/**
519 * tracing_on - enable tracing buffers
520 *
521 * This function enables tracing buffers that may have been
522 * disabled with tracing_off.
523 */
524void tracing_on(void)
525{
526 tracer_tracing_on(&global_trace);
527}
528EXPORT_SYMBOL_GPL(tracing_on);
529
530/**
531 * __trace_puts - write a constant string into the trace buffer.
532 * @ip: The address of the caller
533 * @str: The constant string to write
534 * @size: The size of the string.
535 */
536int __trace_puts(unsigned long ip, const char *str, int size)
537{
538 struct ring_buffer_event *event;
539 struct ring_buffer *buffer;
540 struct print_entry *entry;
541 unsigned long irq_flags;
542 int alloc;
543 int pc;
544
545 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
546 return 0;
547
548 pc = preempt_count();
549
550 if (unlikely(tracing_selftest_running || tracing_disabled))
551 return 0;
552
553 alloc = sizeof(*entry) + size + 2; /* possible \n added */
554
555 local_save_flags(irq_flags);
556 buffer = global_trace.trace_buffer.buffer;
557 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
558 irq_flags, pc);
559 if (!event)
560 return 0;
561
562 entry = ring_buffer_event_data(event);
563 entry->ip = ip;
564
565 memcpy(&entry->buf, str, size);
566
567 /* Add a newline if necessary */
568 if (entry->buf[size - 1] != '\n') {
569 entry->buf[size] = '\n';
570 entry->buf[size + 1] = '\0';
571 } else
572 entry->buf[size] = '\0';
573
574 __buffer_unlock_commit(buffer, event);
575 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
576
577 return size;
578}
579EXPORT_SYMBOL_GPL(__trace_puts);
580
581/**
582 * __trace_bputs - write the pointer to a constant string into trace buffer
583 * @ip: The address of the caller
584 * @str: The constant string to write to the buffer to
585 */
586int __trace_bputs(unsigned long ip, const char *str)
587{
588 struct ring_buffer_event *event;
589 struct ring_buffer *buffer;
590 struct bputs_entry *entry;
591 unsigned long irq_flags;
592 int size = sizeof(struct bputs_entry);
593 int pc;
594
595 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
596 return 0;
597
598 pc = preempt_count();
599
600 if (unlikely(tracing_selftest_running || tracing_disabled))
601 return 0;
602
603 local_save_flags(irq_flags);
604 buffer = global_trace.trace_buffer.buffer;
605 event = trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
606 irq_flags, pc);
607 if (!event)
608 return 0;
609
610 entry = ring_buffer_event_data(event);
611 entry->ip = ip;
612 entry->str = str;
613
614 __buffer_unlock_commit(buffer, event);
615 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
616
617 return 1;
618}
619EXPORT_SYMBOL_GPL(__trace_bputs);
620
621#ifdef CONFIG_TRACER_SNAPSHOT
622/**
623 * trace_snapshot - take a snapshot of the current buffer.
624 *
625 * This causes a swap between the snapshot buffer and the current live
626 * tracing buffer. You can use this to take snapshots of the live
627 * trace when some condition is triggered, but continue to trace.
628 *
629 * Note, make sure to allocate the snapshot with either
630 * a tracing_snapshot_alloc(), or by doing it manually
631 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
632 *
633 * If the snapshot buffer is not allocated, it will stop tracing.
634 * Basically making a permanent snapshot.
635 */
636void tracing_snapshot(void)
637{
638 struct trace_array *tr = &global_trace;
639 struct tracer *tracer = tr->current_trace;
640 unsigned long flags;
641
642 if (in_nmi()) {
643 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
644 internal_trace_puts("*** snapshot is being ignored ***\n");
645 return;
646 }
647
648 if (!tr->allocated_snapshot) {
649 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
650 internal_trace_puts("*** stopping trace here! ***\n");
651 tracing_off();
652 return;
653 }
654
655 /* Note, snapshot can not be used when the tracer uses it */
656 if (tracer->use_max_tr) {
657 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
658 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
659 return;
660 }
661
662 local_irq_save(flags);
663 update_max_tr(tr, current, smp_processor_id());
664 local_irq_restore(flags);
665}
666EXPORT_SYMBOL_GPL(tracing_snapshot);
667
668static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
669 struct trace_buffer *size_buf, int cpu_id);
670static void set_buffer_entries(struct trace_buffer *buf, unsigned long val);
671
672static int alloc_snapshot(struct trace_array *tr)
673{
674 int ret;
675
676 if (!tr->allocated_snapshot) {
677
678 /* allocate spare buffer */
679 ret = resize_buffer_duplicate_size(&tr->max_buffer,
680 &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
681 if (ret < 0)
682 return ret;
683
684 tr->allocated_snapshot = true;
685 }
686
687 return 0;
688}
689
690static void free_snapshot(struct trace_array *tr)
691{
692 /*
693 * We don't free the ring buffer. instead, resize it because
694 * The max_tr ring buffer has some state (e.g. ring->clock) and
695 * we want preserve it.
696 */
697 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
698 set_buffer_entries(&tr->max_buffer, 1);
699 tracing_reset_online_cpus(&tr->max_buffer);
700 tr->allocated_snapshot = false;
701}
702
703/**
704 * tracing_alloc_snapshot - allocate snapshot buffer.
705 *
706 * This only allocates the snapshot buffer if it isn't already
707 * allocated - it doesn't also take a snapshot.
708 *
709 * This is meant to be used in cases where the snapshot buffer needs
710 * to be set up for events that can't sleep but need to be able to
711 * trigger a snapshot.
712 */
713int tracing_alloc_snapshot(void)
714{
715 struct trace_array *tr = &global_trace;
716 int ret;
717
718 ret = alloc_snapshot(tr);
719 WARN_ON(ret < 0);
720
721 return ret;
722}
723EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
724
725/**
726 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
727 *
728 * This is similar to trace_snapshot(), but it will allocate the
729 * snapshot buffer if it isn't already allocated. Use this only
730 * where it is safe to sleep, as the allocation may sleep.
731 *
732 * This causes a swap between the snapshot buffer and the current live
733 * tracing buffer. You can use this to take snapshots of the live
734 * trace when some condition is triggered, but continue to trace.
735 */
736void tracing_snapshot_alloc(void)
737{
738 int ret;
739
740 ret = tracing_alloc_snapshot();
741 if (ret < 0)
742 return;
743
744 tracing_snapshot();
745}
746EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
747#else
748void tracing_snapshot(void)
749{
750 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
751}
752EXPORT_SYMBOL_GPL(tracing_snapshot);
753int tracing_alloc_snapshot(void)
754{
755 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
756 return -ENODEV;
757}
758EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
759void tracing_snapshot_alloc(void)
760{
761 /* Give warning */
762 tracing_snapshot();
763}
764EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
765#endif /* CONFIG_TRACER_SNAPSHOT */
766
767static void tracer_tracing_off(struct trace_array *tr)
768{
769 if (tr->trace_buffer.buffer)
770 ring_buffer_record_off(tr->trace_buffer.buffer);
771 /*
772 * This flag is looked at when buffers haven't been allocated
773 * yet, or by some tracers (like irqsoff), that just want to
774 * know if the ring buffer has been disabled, but it can handle
775 * races of where it gets disabled but we still do a record.
776 * As the check is in the fast path of the tracers, it is more
777 * important to be fast than accurate.
778 */
779 tr->buffer_disabled = 1;
780 /* Make the flag seen by readers */
781 smp_wmb();
782}
783
784/**
785 * tracing_off - turn off tracing buffers
786 *
787 * This function stops the tracing buffers from recording data.
788 * It does not disable any overhead the tracers themselves may
789 * be causing. This function simply causes all recording to
790 * the ring buffers to fail.
791 */
792void tracing_off(void)
793{
794 tracer_tracing_off(&global_trace);
795}
796EXPORT_SYMBOL_GPL(tracing_off);
797
798void disable_trace_on_warning(void)
799{
800 if (__disable_trace_on_warning)
801 tracing_off();
802}
803
804/**
805 * tracer_tracing_is_on - show real state of ring buffer enabled
806 * @tr : the trace array to know if ring buffer is enabled
807 *
808 * Shows real state of the ring buffer if it is enabled or not.
809 */
810static int tracer_tracing_is_on(struct trace_array *tr)
811{
812 if (tr->trace_buffer.buffer)
813 return ring_buffer_record_is_on(tr->trace_buffer.buffer);
814 return !tr->buffer_disabled;
815}
816
817/**
818 * tracing_is_on - show state of ring buffers enabled
819 */
820int tracing_is_on(void)
821{
822 return tracer_tracing_is_on(&global_trace);
823}
824EXPORT_SYMBOL_GPL(tracing_is_on);
825
826static int __init set_buf_size(char *str)
827{
828 unsigned long buf_size;
829
830 if (!str)
831 return 0;
832 buf_size = memparse(str, &str);
833 /* nr_entries can not be zero */
834 if (buf_size == 0)
835 return 0;
836 trace_buf_size = buf_size;
837 return 1;
838}
839__setup("trace_buf_size=", set_buf_size);
840
841static int __init set_tracing_thresh(char *str)
842{
843 unsigned long threshold;
844 int ret;
845
846 if (!str)
847 return 0;
848 ret = kstrtoul(str, 0, &threshold);
849 if (ret < 0)
850 return 0;
851 tracing_thresh = threshold * 1000;
852 return 1;
853}
854__setup("tracing_thresh=", set_tracing_thresh);
855
856unsigned long nsecs_to_usecs(unsigned long nsecs)
857{
858 return nsecs / 1000;
859}
860
861/*
862 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
863 * It uses C(a, b) where 'a' is the enum name and 'b' is the string that
864 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
865 * of strings in the order that the enums were defined.
866 */
867#undef C
868#define C(a, b) b
869
870/* These must match the bit postions in trace_iterator_flags */
871static const char *trace_options[] = {
872 TRACE_FLAGS
873 NULL
874};
875
876static struct {
877 u64 (*func)(void);
878 const char *name;
879 int in_ns; /* is this clock in nanoseconds? */
880} trace_clocks[] = {
881 { trace_clock_local, "local", 1 },
882 { trace_clock_global, "global", 1 },
883 { trace_clock_counter, "counter", 0 },
884 { trace_clock_jiffies, "uptime", 0 },
885 { trace_clock, "perf", 1 },
886 { ktime_get_mono_fast_ns, "mono", 1 },
887 { ktime_get_raw_fast_ns, "mono_raw", 1 },
888 ARCH_TRACE_CLOCKS
889};
890
891/*
892 * trace_parser_get_init - gets the buffer for trace parser
893 */
894int trace_parser_get_init(struct trace_parser *parser, int size)
895{
896 memset(parser, 0, sizeof(*parser));
897
898 parser->buffer = kmalloc(size, GFP_KERNEL);
899 if (!parser->buffer)
900 return 1;
901
902 parser->size = size;
903 return 0;
904}
905
906/*
907 * trace_parser_put - frees the buffer for trace parser
908 */
909void trace_parser_put(struct trace_parser *parser)
910{
911 kfree(parser->buffer);
912}
913
914/*
915 * trace_get_user - reads the user input string separated by space
916 * (matched by isspace(ch))
917 *
918 * For each string found the 'struct trace_parser' is updated,
919 * and the function returns.
920 *
921 * Returns number of bytes read.
922 *
923 * See kernel/trace/trace.h for 'struct trace_parser' details.
924 */
925int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
926 size_t cnt, loff_t *ppos)
927{
928 char ch;
929 size_t read = 0;
930 ssize_t ret;
931
932 if (!*ppos)
933 trace_parser_clear(parser);
934
935 ret = get_user(ch, ubuf++);
936 if (ret)
937 goto out;
938
939 read++;
940 cnt--;
941
942 /*
943 * The parser is not finished with the last write,
944 * continue reading the user input without skipping spaces.
945 */
946 if (!parser->cont) {
947 /* skip white space */
948 while (cnt && isspace(ch)) {
949 ret = get_user(ch, ubuf++);
950 if (ret)
951 goto out;
952 read++;
953 cnt--;
954 }
955
956 /* only spaces were written */
957 if (isspace(ch)) {
958 *ppos += read;
959 ret = read;
960 goto out;
961 }
962
963 parser->idx = 0;
964 }
965
966 /* read the non-space input */
967 while (cnt && !isspace(ch)) {
968 if (parser->idx < parser->size - 1)
969 parser->buffer[parser->idx++] = ch;
970 else {
971 ret = -EINVAL;
972 goto out;
973 }
974 ret = get_user(ch, ubuf++);
975 if (ret)
976 goto out;
977 read++;
978 cnt--;
979 }
980
981 /* We either got finished input or we have to wait for another call. */
982 if (isspace(ch)) {
983 parser->buffer[parser->idx] = 0;
984 parser->cont = false;
985 } else if (parser->idx < parser->size - 1) {
986 parser->cont = true;
987 parser->buffer[parser->idx++] = ch;
988 } else {
989 ret = -EINVAL;
990 goto out;
991 }
992
993 *ppos += read;
994 ret = read;
995
996out:
997 return ret;
998}
999
1000/* TODO add a seq_buf_to_buffer() */
1001static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1002{
1003 int len;
1004
1005 if (trace_seq_used(s) <= s->seq.readpos)
1006 return -EBUSY;
1007
1008 len = trace_seq_used(s) - s->seq.readpos;
1009 if (cnt > len)
1010 cnt = len;
1011 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1012
1013 s->seq.readpos += cnt;
1014 return cnt;
1015}
1016
1017unsigned long __read_mostly tracing_thresh;
1018
1019#ifdef CONFIG_TRACER_MAX_TRACE
1020/*
1021 * Copy the new maximum trace into the separate maximum-trace
1022 * structure. (this way the maximum trace is permanently saved,
1023 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
1024 */
1025static void
1026__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1027{
1028 struct trace_buffer *trace_buf = &tr->trace_buffer;
1029 struct trace_buffer *max_buf = &tr->max_buffer;
1030 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1031 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1032
1033 max_buf->cpu = cpu;
1034 max_buf->time_start = data->preempt_timestamp;
1035
1036 max_data->saved_latency = tr->max_latency;
1037 max_data->critical_start = data->critical_start;
1038 max_data->critical_end = data->critical_end;
1039
1040 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1041 max_data->pid = tsk->pid;
1042 /*
1043 * If tsk == current, then use current_uid(), as that does not use
1044 * RCU. The irq tracer can be called out of RCU scope.
1045 */
1046 if (tsk == current)
1047 max_data->uid = current_uid();
1048 else
1049 max_data->uid = task_uid(tsk);
1050
1051 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1052 max_data->policy = tsk->policy;
1053 max_data->rt_priority = tsk->rt_priority;
1054
1055 /* record this tasks comm */
1056 tracing_record_cmdline(tsk);
1057}
1058
1059/**
1060 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1061 * @tr: tracer
1062 * @tsk: the task with the latency
1063 * @cpu: The cpu that initiated the trace.
1064 *
1065 * Flip the buffers between the @tr and the max_tr and record information
1066 * about which task was the cause of this latency.
1067 */
1068void
1069update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1070{
1071 struct ring_buffer *buf;
1072
1073 if (tr->stop_count)
1074 return;
1075
1076 WARN_ON_ONCE(!irqs_disabled());
1077
1078 if (!tr->allocated_snapshot) {
1079 /* Only the nop tracer should hit this when disabling */
1080 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1081 return;
1082 }
1083
1084 arch_spin_lock(&tr->max_lock);
1085
1086 buf = tr->trace_buffer.buffer;
1087 tr->trace_buffer.buffer = tr->max_buffer.buffer;
1088 tr->max_buffer.buffer = buf;
1089
1090 __update_max_tr(tr, tsk, cpu);
1091 arch_spin_unlock(&tr->max_lock);
1092}
1093
1094/**
1095 * update_max_tr_single - only copy one trace over, and reset the rest
1096 * @tr - tracer
1097 * @tsk - task with the latency
1098 * @cpu - the cpu of the buffer to copy.
1099 *
1100 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1101 */
1102void
1103update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1104{
1105 int ret;
1106
1107 if (tr->stop_count)
1108 return;
1109
1110 WARN_ON_ONCE(!irqs_disabled());
1111 if (!tr->allocated_snapshot) {
1112 /* Only the nop tracer should hit this when disabling */
1113 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1114 return;
1115 }
1116
1117 arch_spin_lock(&tr->max_lock);
1118
1119 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1120
1121 if (ret == -EBUSY) {
1122 /*
1123 * We failed to swap the buffer due to a commit taking
1124 * place on this CPU. We fail to record, but we reset
1125 * the max trace buffer (no one writes directly to it)
1126 * and flag that it failed.
1127 */
1128 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1129 "Failed to swap buffers due to commit in progress\n");
1130 }
1131
1132 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1133
1134 __update_max_tr(tr, tsk, cpu);
1135 arch_spin_unlock(&tr->max_lock);
1136}
1137#endif /* CONFIG_TRACER_MAX_TRACE */
1138
1139static int wait_on_pipe(struct trace_iterator *iter, bool full)
1140{
1141 /* Iterators are static, they should be filled or empty */
1142 if (trace_buffer_iter(iter, iter->cpu_file))
1143 return 0;
1144
1145 return ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file,
1146 full);
1147}
1148
1149#ifdef CONFIG_FTRACE_STARTUP_TEST
1150static int run_tracer_selftest(struct tracer *type)
1151{
1152 struct trace_array *tr = &global_trace;
1153 struct tracer *saved_tracer = tr->current_trace;
1154 int ret;
1155
1156 if (!type->selftest || tracing_selftest_disabled)
1157 return 0;
1158
1159 /*
1160 * Run a selftest on this tracer.
1161 * Here we reset the trace buffer, and set the current
1162 * tracer to be this tracer. The tracer can then run some
1163 * internal tracing to verify that everything is in order.
1164 * If we fail, we do not register this tracer.
1165 */
1166 tracing_reset_online_cpus(&tr->trace_buffer);
1167
1168 tr->current_trace = type;
1169
1170#ifdef CONFIG_TRACER_MAX_TRACE
1171 if (type->use_max_tr) {
1172 /* If we expanded the buffers, make sure the max is expanded too */
1173 if (ring_buffer_expanded)
1174 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1175 RING_BUFFER_ALL_CPUS);
1176 tr->allocated_snapshot = true;
1177 }
1178#endif
1179
1180 /* the test is responsible for initializing and enabling */
1181 pr_info("Testing tracer %s: ", type->name);
1182 ret = type->selftest(type, tr);
1183 /* the test is responsible for resetting too */
1184 tr->current_trace = saved_tracer;
1185 if (ret) {
1186 printk(KERN_CONT "FAILED!\n");
1187 /* Add the warning after printing 'FAILED' */
1188 WARN_ON(1);
1189 return -1;
1190 }
1191 /* Only reset on passing, to avoid touching corrupted buffers */
1192 tracing_reset_online_cpus(&tr->trace_buffer);
1193
1194#ifdef CONFIG_TRACER_MAX_TRACE
1195 if (type->use_max_tr) {
1196 tr->allocated_snapshot = false;
1197
1198 /* Shrink the max buffer again */
1199 if (ring_buffer_expanded)
1200 ring_buffer_resize(tr->max_buffer.buffer, 1,
1201 RING_BUFFER_ALL_CPUS);
1202 }
1203#endif
1204
1205 printk(KERN_CONT "PASSED\n");
1206 return 0;
1207}
1208#else
1209static inline int run_tracer_selftest(struct tracer *type)
1210{
1211 return 0;
1212}
1213#endif /* CONFIG_FTRACE_STARTUP_TEST */
1214
1215static void add_tracer_options(struct trace_array *tr, struct tracer *t);
1216
1217static void __init apply_trace_boot_options(void);
1218
1219/**
1220 * register_tracer - register a tracer with the ftrace system.
1221 * @type - the plugin for the tracer
1222 *
1223 * Register a new plugin tracer.
1224 */
1225int __init register_tracer(struct tracer *type)
1226{
1227 struct tracer *t;
1228 int ret = 0;
1229
1230 if (!type->name) {
1231 pr_info("Tracer must have a name\n");
1232 return -1;
1233 }
1234
1235 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1236 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1237 return -1;
1238 }
1239
1240 mutex_lock(&trace_types_lock);
1241
1242 tracing_selftest_running = true;
1243
1244 for (t = trace_types; t; t = t->next) {
1245 if (strcmp(type->name, t->name) == 0) {
1246 /* already found */
1247 pr_info("Tracer %s already registered\n",
1248 type->name);
1249 ret = -1;
1250 goto out;
1251 }
1252 }
1253
1254 if (!type->set_flag)
1255 type->set_flag = &dummy_set_flag;
1256 if (!type->flags) {
1257 /*allocate a dummy tracer_flags*/
1258 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
1259 if (!type->flags) {
1260 ret = -ENOMEM;
1261 goto out;
1262 }
1263 type->flags->val = 0;
1264 type->flags->opts = dummy_tracer_opt;
1265 } else
1266 if (!type->flags->opts)
1267 type->flags->opts = dummy_tracer_opt;
1268
1269 /* store the tracer for __set_tracer_option */
1270 type->flags->trace = type;
1271
1272 ret = run_tracer_selftest(type);
1273 if (ret < 0)
1274 goto out;
1275
1276 type->next = trace_types;
1277 trace_types = type;
1278 add_tracer_options(&global_trace, type);
1279
1280 out:
1281 tracing_selftest_running = false;
1282 mutex_unlock(&trace_types_lock);
1283
1284 if (ret || !default_bootup_tracer)
1285 goto out_unlock;
1286
1287 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1288 goto out_unlock;
1289
1290 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1291 /* Do we want this tracer to start on bootup? */
1292 tracing_set_tracer(&global_trace, type->name);
1293 default_bootup_tracer = NULL;
1294
1295 apply_trace_boot_options();
1296
1297 /* disable other selftests, since this will break it. */
1298 tracing_selftest_disabled = true;
1299#ifdef CONFIG_FTRACE_STARTUP_TEST
1300 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1301 type->name);
1302#endif
1303
1304 out_unlock:
1305 return ret;
1306}
1307
1308void tracing_reset(struct trace_buffer *buf, int cpu)
1309{
1310 struct ring_buffer *buffer = buf->buffer;
1311
1312 if (!buffer)
1313 return;
1314
1315 ring_buffer_record_disable(buffer);
1316
1317 /* Make sure all commits have finished */
1318 synchronize_sched();
1319 ring_buffer_reset_cpu(buffer, cpu);
1320
1321 ring_buffer_record_enable(buffer);
1322}
1323
1324void tracing_reset_online_cpus(struct trace_buffer *buf)
1325{
1326 struct ring_buffer *buffer = buf->buffer;
1327 int cpu;
1328
1329 if (!buffer)
1330 return;
1331
1332 ring_buffer_record_disable(buffer);
1333
1334 /* Make sure all commits have finished */
1335 synchronize_sched();
1336
1337 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
1338
1339 for_each_online_cpu(cpu)
1340 ring_buffer_reset_cpu(buffer, cpu);
1341
1342 ring_buffer_record_enable(buffer);
1343}
1344
1345/* Must have trace_types_lock held */
1346void tracing_reset_all_online_cpus(void)
1347{
1348 struct trace_array *tr;
1349
1350 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1351 tracing_reset_online_cpus(&tr->trace_buffer);
1352#ifdef CONFIG_TRACER_MAX_TRACE
1353 tracing_reset_online_cpus(&tr->max_buffer);
1354#endif
1355 }
1356}
1357
1358#define SAVED_CMDLINES_DEFAULT 128
1359#define NO_CMDLINE_MAP UINT_MAX
1360static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1361struct saved_cmdlines_buffer {
1362 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1363 unsigned *map_cmdline_to_pid;
1364 unsigned cmdline_num;
1365 int cmdline_idx;
1366 char *saved_cmdlines;
1367};
1368static struct saved_cmdlines_buffer *savedcmd;
1369
1370/* temporary disable recording */
1371static atomic_t trace_record_cmdline_disabled __read_mostly;
1372
1373static inline char *get_saved_cmdlines(int idx)
1374{
1375 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
1376}
1377
1378static inline void set_cmdline(int idx, const char *cmdline)
1379{
1380 memcpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
1381}
1382
1383static int allocate_cmdlines_buffer(unsigned int val,
1384 struct saved_cmdlines_buffer *s)
1385{
1386 s->map_cmdline_to_pid = kmalloc(val * sizeof(*s->map_cmdline_to_pid),
1387 GFP_KERNEL);
1388 if (!s->map_cmdline_to_pid)
1389 return -ENOMEM;
1390
1391 s->saved_cmdlines = kmalloc(val * TASK_COMM_LEN, GFP_KERNEL);
1392 if (!s->saved_cmdlines) {
1393 kfree(s->map_cmdline_to_pid);
1394 return -ENOMEM;
1395 }
1396
1397 s->cmdline_idx = 0;
1398 s->cmdline_num = val;
1399 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
1400 sizeof(s->map_pid_to_cmdline));
1401 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
1402 val * sizeof(*s->map_cmdline_to_pid));
1403
1404 return 0;
1405}
1406
1407static int trace_create_savedcmd(void)
1408{
1409 int ret;
1410
1411 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
1412 if (!savedcmd)
1413 return -ENOMEM;
1414
1415 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
1416 if (ret < 0) {
1417 kfree(savedcmd);
1418 savedcmd = NULL;
1419 return -ENOMEM;
1420 }
1421
1422 return 0;
1423}
1424
1425int is_tracing_stopped(void)
1426{
1427 return global_trace.stop_count;
1428}
1429
1430/**
1431 * tracing_start - quick start of the tracer
1432 *
1433 * If tracing is enabled but was stopped by tracing_stop,
1434 * this will start the tracer back up.
1435 */
1436void tracing_start(void)
1437{
1438 struct ring_buffer *buffer;
1439 unsigned long flags;
1440
1441 if (tracing_disabled)
1442 return;
1443
1444 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1445 if (--global_trace.stop_count) {
1446 if (global_trace.stop_count < 0) {
1447 /* Someone screwed up their debugging */
1448 WARN_ON_ONCE(1);
1449 global_trace.stop_count = 0;
1450 }
1451 goto out;
1452 }
1453
1454 /* Prevent the buffers from switching */
1455 arch_spin_lock(&global_trace.max_lock);
1456
1457 buffer = global_trace.trace_buffer.buffer;
1458 if (buffer)
1459 ring_buffer_record_enable(buffer);
1460
1461#ifdef CONFIG_TRACER_MAX_TRACE
1462 buffer = global_trace.max_buffer.buffer;
1463 if (buffer)
1464 ring_buffer_record_enable(buffer);
1465#endif
1466
1467 arch_spin_unlock(&global_trace.max_lock);
1468
1469 out:
1470 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1471}
1472
1473static void tracing_start_tr(struct trace_array *tr)
1474{
1475 struct ring_buffer *buffer;
1476 unsigned long flags;
1477
1478 if (tracing_disabled)
1479 return;
1480
1481 /* If global, we need to also start the max tracer */
1482 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1483 return tracing_start();
1484
1485 raw_spin_lock_irqsave(&tr->start_lock, flags);
1486
1487 if (--tr->stop_count) {
1488 if (tr->stop_count < 0) {
1489 /* Someone screwed up their debugging */
1490 WARN_ON_ONCE(1);
1491 tr->stop_count = 0;
1492 }
1493 goto out;
1494 }
1495
1496 buffer = tr->trace_buffer.buffer;
1497 if (buffer)
1498 ring_buffer_record_enable(buffer);
1499
1500 out:
1501 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1502}
1503
1504/**
1505 * tracing_stop - quick stop of the tracer
1506 *
1507 * Light weight way to stop tracing. Use in conjunction with
1508 * tracing_start.
1509 */
1510void tracing_stop(void)
1511{
1512 struct ring_buffer *buffer;
1513 unsigned long flags;
1514
1515 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1516 if (global_trace.stop_count++)
1517 goto out;
1518
1519 /* Prevent the buffers from switching */
1520 arch_spin_lock(&global_trace.max_lock);
1521
1522 buffer = global_trace.trace_buffer.buffer;
1523 if (buffer)
1524 ring_buffer_record_disable(buffer);
1525
1526#ifdef CONFIG_TRACER_MAX_TRACE
1527 buffer = global_trace.max_buffer.buffer;
1528 if (buffer)
1529 ring_buffer_record_disable(buffer);
1530#endif
1531
1532 arch_spin_unlock(&global_trace.max_lock);
1533
1534 out:
1535 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1536}
1537
1538static void tracing_stop_tr(struct trace_array *tr)
1539{
1540 struct ring_buffer *buffer;
1541 unsigned long flags;
1542
1543 /* If global, we need to also stop the max tracer */
1544 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1545 return tracing_stop();
1546
1547 raw_spin_lock_irqsave(&tr->start_lock, flags);
1548 if (tr->stop_count++)
1549 goto out;
1550
1551 buffer = tr->trace_buffer.buffer;
1552 if (buffer)
1553 ring_buffer_record_disable(buffer);
1554
1555 out:
1556 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1557}
1558
1559void trace_stop_cmdline_recording(void);
1560
1561static int trace_save_cmdline(struct task_struct *tsk)
1562{
1563 unsigned pid, idx;
1564
1565 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1566 return 0;
1567
1568 /*
1569 * It's not the end of the world if we don't get
1570 * the lock, but we also don't want to spin
1571 * nor do we want to disable interrupts,
1572 * so if we miss here, then better luck next time.
1573 */
1574 if (!arch_spin_trylock(&trace_cmdline_lock))
1575 return 0;
1576
1577 idx = savedcmd->map_pid_to_cmdline[tsk->pid];
1578 if (idx == NO_CMDLINE_MAP) {
1579 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
1580
1581 /*
1582 * Check whether the cmdline buffer at idx has a pid
1583 * mapped. We are going to overwrite that entry so we
1584 * need to clear the map_pid_to_cmdline. Otherwise we
1585 * would read the new comm for the old pid.
1586 */
1587 pid = savedcmd->map_cmdline_to_pid[idx];
1588 if (pid != NO_CMDLINE_MAP)
1589 savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1590
1591 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
1592 savedcmd->map_pid_to_cmdline[tsk->pid] = idx;
1593
1594 savedcmd->cmdline_idx = idx;
1595 }
1596
1597 set_cmdline(idx, tsk->comm);
1598
1599 arch_spin_unlock(&trace_cmdline_lock);
1600
1601 return 1;
1602}
1603
1604static void __trace_find_cmdline(int pid, char comm[])
1605{
1606 unsigned map;
1607
1608 if (!pid) {
1609 strcpy(comm, "<idle>");
1610 return;
1611 }
1612
1613 if (WARN_ON_ONCE(pid < 0)) {
1614 strcpy(comm, "<XXX>");
1615 return;
1616 }
1617
1618 if (pid > PID_MAX_DEFAULT) {
1619 strcpy(comm, "<...>");
1620 return;
1621 }
1622
1623 map = savedcmd->map_pid_to_cmdline[pid];
1624 if (map != NO_CMDLINE_MAP)
1625 strcpy(comm, get_saved_cmdlines(map));
1626 else
1627 strcpy(comm, "<...>");
1628}
1629
1630void trace_find_cmdline(int pid, char comm[])
1631{
1632 preempt_disable();
1633 arch_spin_lock(&trace_cmdline_lock);
1634
1635 __trace_find_cmdline(pid, comm);
1636
1637 arch_spin_unlock(&trace_cmdline_lock);
1638 preempt_enable();
1639}
1640
1641void tracing_record_cmdline(struct task_struct *tsk)
1642{
1643 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1644 return;
1645
1646 if (!__this_cpu_read(trace_cmdline_save))
1647 return;
1648
1649 if (trace_save_cmdline(tsk))
1650 __this_cpu_write(trace_cmdline_save, false);
1651}
1652
1653void
1654tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1655 int pc)
1656{
1657 struct task_struct *tsk = current;
1658
1659 entry->preempt_count = pc & 0xff;
1660 entry->pid = (tsk) ? tsk->pid : 0;
1661 entry->flags =
1662#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1663 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1664#else
1665 TRACE_FLAG_IRQS_NOSUPPORT |
1666#endif
1667 ((pc & NMI_MASK ) ? TRACE_FLAG_NMI : 0) |
1668 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1669 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1670 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
1671 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
1672}
1673EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1674
1675struct ring_buffer_event *
1676trace_buffer_lock_reserve(struct ring_buffer *buffer,
1677 int type,
1678 unsigned long len,
1679 unsigned long flags, int pc)
1680{
1681 struct ring_buffer_event *event;
1682
1683 event = ring_buffer_lock_reserve(buffer, len);
1684 if (event != NULL) {
1685 struct trace_entry *ent = ring_buffer_event_data(event);
1686
1687 tracing_generic_entry_update(ent, flags, pc);
1688 ent->type = type;
1689 }
1690
1691 return event;
1692}
1693
1694void
1695__buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1696{
1697 __this_cpu_write(trace_cmdline_save, true);
1698 ring_buffer_unlock_commit(buffer, event);
1699}
1700
1701void trace_buffer_unlock_commit(struct trace_array *tr,
1702 struct ring_buffer *buffer,
1703 struct ring_buffer_event *event,
1704 unsigned long flags, int pc)
1705{
1706 __buffer_unlock_commit(buffer, event);
1707
1708 ftrace_trace_stack(tr, buffer, flags, 6, pc, NULL);
1709 ftrace_trace_userstack(buffer, flags, pc);
1710}
1711EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1712
1713static struct ring_buffer *temp_buffer;
1714
1715struct ring_buffer_event *
1716trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1717 struct trace_event_file *trace_file,
1718 int type, unsigned long len,
1719 unsigned long flags, int pc)
1720{
1721 struct ring_buffer_event *entry;
1722
1723 *current_rb = trace_file->tr->trace_buffer.buffer;
1724 entry = trace_buffer_lock_reserve(*current_rb,
1725 type, len, flags, pc);
1726 /*
1727 * If tracing is off, but we have triggers enabled
1728 * we still need to look at the event data. Use the temp_buffer
1729 * to store the trace event for the tigger to use. It's recusive
1730 * safe and will not be recorded anywhere.
1731 */
1732 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
1733 *current_rb = temp_buffer;
1734 entry = trace_buffer_lock_reserve(*current_rb,
1735 type, len, flags, pc);
1736 }
1737 return entry;
1738}
1739EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1740
1741struct ring_buffer_event *
1742trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1743 int type, unsigned long len,
1744 unsigned long flags, int pc)
1745{
1746 *current_rb = global_trace.trace_buffer.buffer;
1747 return trace_buffer_lock_reserve(*current_rb,
1748 type, len, flags, pc);
1749}
1750EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1751
1752void trace_buffer_unlock_commit_regs(struct trace_array *tr,
1753 struct ring_buffer *buffer,
1754 struct ring_buffer_event *event,
1755 unsigned long flags, int pc,
1756 struct pt_regs *regs)
1757{
1758 __buffer_unlock_commit(buffer, event);
1759
1760 ftrace_trace_stack(tr, buffer, flags, 0, pc, regs);
1761 ftrace_trace_userstack(buffer, flags, pc);
1762}
1763EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1764
1765void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1766 struct ring_buffer_event *event)
1767{
1768 ring_buffer_discard_commit(buffer, event);
1769}
1770EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1771
1772void
1773trace_function(struct trace_array *tr,
1774 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1775 int pc)
1776{
1777 struct trace_event_call *call = &event_function;
1778 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1779 struct ring_buffer_event *event;
1780 struct ftrace_entry *entry;
1781
1782 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1783 flags, pc);
1784 if (!event)
1785 return;
1786 entry = ring_buffer_event_data(event);
1787 entry->ip = ip;
1788 entry->parent_ip = parent_ip;
1789
1790 if (!call_filter_check_discard(call, entry, buffer, event))
1791 __buffer_unlock_commit(buffer, event);
1792}
1793
1794#ifdef CONFIG_STACKTRACE
1795
1796#define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1797struct ftrace_stack {
1798 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1799};
1800
1801static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1802static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1803
1804static void __ftrace_trace_stack(struct ring_buffer *buffer,
1805 unsigned long flags,
1806 int skip, int pc, struct pt_regs *regs)
1807{
1808 struct trace_event_call *call = &event_kernel_stack;
1809 struct ring_buffer_event *event;
1810 struct stack_entry *entry;
1811 struct stack_trace trace;
1812 int use_stack;
1813 int size = FTRACE_STACK_ENTRIES;
1814
1815 trace.nr_entries = 0;
1816 trace.skip = skip;
1817
1818 /*
1819 * Since events can happen in NMIs there's no safe way to
1820 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1821 * or NMI comes in, it will just have to use the default
1822 * FTRACE_STACK_SIZE.
1823 */
1824 preempt_disable_notrace();
1825
1826 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1827 /*
1828 * We don't need any atomic variables, just a barrier.
1829 * If an interrupt comes in, we don't care, because it would
1830 * have exited and put the counter back to what we want.
1831 * We just need a barrier to keep gcc from moving things
1832 * around.
1833 */
1834 barrier();
1835 if (use_stack == 1) {
1836 trace.entries = this_cpu_ptr(ftrace_stack.calls);
1837 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1838
1839 if (regs)
1840 save_stack_trace_regs(regs, &trace);
1841 else
1842 save_stack_trace(&trace);
1843
1844 if (trace.nr_entries > size)
1845 size = trace.nr_entries;
1846 } else
1847 /* From now on, use_stack is a boolean */
1848 use_stack = 0;
1849
1850 size *= sizeof(unsigned long);
1851
1852 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1853 sizeof(*entry) + size, flags, pc);
1854 if (!event)
1855 goto out;
1856 entry = ring_buffer_event_data(event);
1857
1858 memset(&entry->caller, 0, size);
1859
1860 if (use_stack)
1861 memcpy(&entry->caller, trace.entries,
1862 trace.nr_entries * sizeof(unsigned long));
1863 else {
1864 trace.max_entries = FTRACE_STACK_ENTRIES;
1865 trace.entries = entry->caller;
1866 if (regs)
1867 save_stack_trace_regs(regs, &trace);
1868 else
1869 save_stack_trace(&trace);
1870 }
1871
1872 entry->size = trace.nr_entries;
1873
1874 if (!call_filter_check_discard(call, entry, buffer, event))
1875 __buffer_unlock_commit(buffer, event);
1876
1877 out:
1878 /* Again, don't let gcc optimize things here */
1879 barrier();
1880 __this_cpu_dec(ftrace_stack_reserve);
1881 preempt_enable_notrace();
1882
1883}
1884
1885static inline void ftrace_trace_stack(struct trace_array *tr,
1886 struct ring_buffer *buffer,
1887 unsigned long flags,
1888 int skip, int pc, struct pt_regs *regs)
1889{
1890 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
1891 return;
1892
1893 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1894}
1895
1896void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1897 int pc)
1898{
1899 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1900}
1901
1902/**
1903 * trace_dump_stack - record a stack back trace in the trace buffer
1904 * @skip: Number of functions to skip (helper handlers)
1905 */
1906void trace_dump_stack(int skip)
1907{
1908 unsigned long flags;
1909
1910 if (tracing_disabled || tracing_selftest_running)
1911 return;
1912
1913 local_save_flags(flags);
1914
1915 /*
1916 * Skip 3 more, seems to get us at the caller of
1917 * this function.
1918 */
1919 skip += 3;
1920 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1921 flags, skip, preempt_count(), NULL);
1922}
1923
1924static DEFINE_PER_CPU(int, user_stack_count);
1925
1926void
1927ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1928{
1929 struct trace_event_call *call = &event_user_stack;
1930 struct ring_buffer_event *event;
1931 struct userstack_entry *entry;
1932 struct stack_trace trace;
1933
1934 if (!(global_trace.trace_flags & TRACE_ITER_USERSTACKTRACE))
1935 return;
1936
1937 /*
1938 * NMIs can not handle page faults, even with fix ups.
1939 * The save user stack can (and often does) fault.
1940 */
1941 if (unlikely(in_nmi()))
1942 return;
1943
1944 /*
1945 * prevent recursion, since the user stack tracing may
1946 * trigger other kernel events.
1947 */
1948 preempt_disable();
1949 if (__this_cpu_read(user_stack_count))
1950 goto out;
1951
1952 __this_cpu_inc(user_stack_count);
1953
1954 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1955 sizeof(*entry), flags, pc);
1956 if (!event)
1957 goto out_drop_count;
1958 entry = ring_buffer_event_data(event);
1959
1960 entry->tgid = current->tgid;
1961 memset(&entry->caller, 0, sizeof(entry->caller));
1962
1963 trace.nr_entries = 0;
1964 trace.max_entries = FTRACE_STACK_ENTRIES;
1965 trace.skip = 0;
1966 trace.entries = entry->caller;
1967
1968 save_stack_trace_user(&trace);
1969 if (!call_filter_check_discard(call, entry, buffer, event))
1970 __buffer_unlock_commit(buffer, event);
1971
1972 out_drop_count:
1973 __this_cpu_dec(user_stack_count);
1974 out:
1975 preempt_enable();
1976}
1977
1978#ifdef UNUSED
1979static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1980{
1981 ftrace_trace_userstack(tr, flags, preempt_count());
1982}
1983#endif /* UNUSED */
1984
1985#endif /* CONFIG_STACKTRACE */
1986
1987/* created for use with alloc_percpu */
1988struct trace_buffer_struct {
1989 char buffer[TRACE_BUF_SIZE];
1990};
1991
1992static struct trace_buffer_struct *trace_percpu_buffer;
1993static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1994static struct trace_buffer_struct *trace_percpu_irq_buffer;
1995static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1996
1997/*
1998 * The buffer used is dependent on the context. There is a per cpu
1999 * buffer for normal context, softirq contex, hard irq context and
2000 * for NMI context. Thise allows for lockless recording.
2001 *
2002 * Note, if the buffers failed to be allocated, then this returns NULL
2003 */
2004static char *get_trace_buf(void)
2005{
2006 struct trace_buffer_struct *percpu_buffer;
2007
2008 /*
2009 * If we have allocated per cpu buffers, then we do not
2010 * need to do any locking.
2011 */
2012 if (in_nmi())
2013 percpu_buffer = trace_percpu_nmi_buffer;
2014 else if (in_irq())
2015 percpu_buffer = trace_percpu_irq_buffer;
2016 else if (in_softirq())
2017 percpu_buffer = trace_percpu_sirq_buffer;
2018 else
2019 percpu_buffer = trace_percpu_buffer;
2020
2021 if (!percpu_buffer)
2022 return NULL;
2023
2024 return this_cpu_ptr(&percpu_buffer->buffer[0]);
2025}
2026
2027static int alloc_percpu_trace_buffer(void)
2028{
2029 struct trace_buffer_struct *buffers;
2030 struct trace_buffer_struct *sirq_buffers;
2031 struct trace_buffer_struct *irq_buffers;
2032 struct trace_buffer_struct *nmi_buffers;
2033
2034 buffers = alloc_percpu(struct trace_buffer_struct);
2035 if (!buffers)
2036 goto err_warn;
2037
2038 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
2039 if (!sirq_buffers)
2040 goto err_sirq;
2041
2042 irq_buffers = alloc_percpu(struct trace_buffer_struct);
2043 if (!irq_buffers)
2044 goto err_irq;
2045
2046 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
2047 if (!nmi_buffers)
2048 goto err_nmi;
2049
2050 trace_percpu_buffer = buffers;
2051 trace_percpu_sirq_buffer = sirq_buffers;
2052 trace_percpu_irq_buffer = irq_buffers;
2053 trace_percpu_nmi_buffer = nmi_buffers;
2054
2055 return 0;
2056
2057 err_nmi:
2058 free_percpu(irq_buffers);
2059 err_irq:
2060 free_percpu(sirq_buffers);
2061 err_sirq:
2062 free_percpu(buffers);
2063 err_warn:
2064 WARN(1, "Could not allocate percpu trace_printk buffer");
2065 return -ENOMEM;
2066}
2067
2068static int buffers_allocated;
2069
2070void trace_printk_init_buffers(void)
2071{
2072 if (buffers_allocated)
2073 return;
2074
2075 if (alloc_percpu_trace_buffer())
2076 return;
2077
2078 /* trace_printk() is for debug use only. Don't use it in production. */
2079
2080 pr_warn("\n");
2081 pr_warn("**********************************************************\n");
2082 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
2083 pr_warn("** **\n");
2084 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
2085 pr_warn("** **\n");
2086 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
2087 pr_warn("** unsafe for production use. **\n");
2088 pr_warn("** **\n");
2089 pr_warn("** If you see this message and you are not debugging **\n");
2090 pr_warn("** the kernel, report this immediately to your vendor! **\n");
2091 pr_warn("** **\n");
2092 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
2093 pr_warn("**********************************************************\n");
2094
2095 /* Expand the buffers to set size */
2096 tracing_update_buffers();
2097
2098 buffers_allocated = 1;
2099
2100 /*
2101 * trace_printk_init_buffers() can be called by modules.
2102 * If that happens, then we need to start cmdline recording
2103 * directly here. If the global_trace.buffer is already
2104 * allocated here, then this was called by module code.
2105 */
2106 if (global_trace.trace_buffer.buffer)
2107 tracing_start_cmdline_record();
2108}
2109
2110void trace_printk_start_comm(void)
2111{
2112 /* Start tracing comms if trace printk is set */
2113 if (!buffers_allocated)
2114 return;
2115 tracing_start_cmdline_record();
2116}
2117
2118static void trace_printk_start_stop_comm(int enabled)
2119{
2120 if (!buffers_allocated)
2121 return;
2122
2123 if (enabled)
2124 tracing_start_cmdline_record();
2125 else
2126 tracing_stop_cmdline_record();
2127}
2128
2129/**
2130 * trace_vbprintk - write binary msg to tracing buffer
2131 *
2132 */
2133int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
2134{
2135 struct trace_event_call *call = &event_bprint;
2136 struct ring_buffer_event *event;
2137 struct ring_buffer *buffer;
2138 struct trace_array *tr = &global_trace;
2139 struct bprint_entry *entry;
2140 unsigned long flags;
2141 char *tbuffer;
2142 int len = 0, size, pc;
2143
2144 if (unlikely(tracing_selftest_running || tracing_disabled))
2145 return 0;
2146
2147 /* Don't pollute graph traces with trace_vprintk internals */
2148 pause_graph_tracing();
2149
2150 pc = preempt_count();
2151 preempt_disable_notrace();
2152
2153 tbuffer = get_trace_buf();
2154 if (!tbuffer) {
2155 len = 0;
2156 goto out;
2157 }
2158
2159 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
2160
2161 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
2162 goto out;
2163
2164 local_save_flags(flags);
2165 size = sizeof(*entry) + sizeof(u32) * len;
2166 buffer = tr->trace_buffer.buffer;
2167 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2168 flags, pc);
2169 if (!event)
2170 goto out;
2171 entry = ring_buffer_event_data(event);
2172 entry->ip = ip;
2173 entry->fmt = fmt;
2174
2175 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2176 if (!call_filter_check_discard(call, entry, buffer, event)) {
2177 __buffer_unlock_commit(buffer, event);
2178 ftrace_trace_stack(tr, buffer, flags, 6, pc, NULL);
2179 }
2180
2181out:
2182 preempt_enable_notrace();
2183 unpause_graph_tracing();
2184
2185 return len;
2186}
2187EXPORT_SYMBOL_GPL(trace_vbprintk);
2188
2189static int
2190__trace_array_vprintk(struct ring_buffer *buffer,
2191 unsigned long ip, const char *fmt, va_list args)
2192{
2193 struct trace_event_call *call = &event_print;
2194 struct ring_buffer_event *event;
2195 int len = 0, size, pc;
2196 struct print_entry *entry;
2197 unsigned long flags;
2198 char *tbuffer;
2199
2200 if (tracing_disabled || tracing_selftest_running)
2201 return 0;
2202
2203 /* Don't pollute graph traces with trace_vprintk internals */
2204 pause_graph_tracing();
2205
2206 pc = preempt_count();
2207 preempt_disable_notrace();
2208
2209
2210 tbuffer = get_trace_buf();
2211 if (!tbuffer) {
2212 len = 0;
2213 goto out;
2214 }
2215
2216 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2217
2218 local_save_flags(flags);
2219 size = sizeof(*entry) + len + 1;
2220 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2221 flags, pc);
2222 if (!event)
2223 goto out;
2224 entry = ring_buffer_event_data(event);
2225 entry->ip = ip;
2226
2227 memcpy(&entry->buf, tbuffer, len + 1);
2228 if (!call_filter_check_discard(call, entry, buffer, event)) {
2229 __buffer_unlock_commit(buffer, event);
2230 ftrace_trace_stack(&global_trace, buffer, flags, 6, pc, NULL);
2231 }
2232 out:
2233 preempt_enable_notrace();
2234 unpause_graph_tracing();
2235
2236 return len;
2237}
2238
2239int trace_array_vprintk(struct trace_array *tr,
2240 unsigned long ip, const char *fmt, va_list args)
2241{
2242 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2243}
2244
2245int trace_array_printk(struct trace_array *tr,
2246 unsigned long ip, const char *fmt, ...)
2247{
2248 int ret;
2249 va_list ap;
2250
2251 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
2252 return 0;
2253
2254 va_start(ap, fmt);
2255 ret = trace_array_vprintk(tr, ip, fmt, ap);
2256 va_end(ap);
2257 return ret;
2258}
2259
2260int trace_array_printk_buf(struct ring_buffer *buffer,
2261 unsigned long ip, const char *fmt, ...)
2262{
2263 int ret;
2264 va_list ap;
2265
2266 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
2267 return 0;
2268
2269 va_start(ap, fmt);
2270 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2271 va_end(ap);
2272 return ret;
2273}
2274
2275int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2276{
2277 return trace_array_vprintk(&global_trace, ip, fmt, args);
2278}
2279EXPORT_SYMBOL_GPL(trace_vprintk);
2280
2281static void trace_iterator_increment(struct trace_iterator *iter)
2282{
2283 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2284
2285 iter->idx++;
2286 if (buf_iter)
2287 ring_buffer_read(buf_iter, NULL);
2288}
2289
2290static struct trace_entry *
2291peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2292 unsigned long *lost_events)
2293{
2294 struct ring_buffer_event *event;
2295 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2296
2297 if (buf_iter)
2298 event = ring_buffer_iter_peek(buf_iter, ts);
2299 else
2300 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2301 lost_events);
2302
2303 if (event) {
2304 iter->ent_size = ring_buffer_event_length(event);
2305 return ring_buffer_event_data(event);
2306 }
2307 iter->ent_size = 0;
2308 return NULL;
2309}
2310
2311static struct trace_entry *
2312__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2313 unsigned long *missing_events, u64 *ent_ts)
2314{
2315 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2316 struct trace_entry *ent, *next = NULL;
2317 unsigned long lost_events = 0, next_lost = 0;
2318 int cpu_file = iter->cpu_file;
2319 u64 next_ts = 0, ts;
2320 int next_cpu = -1;
2321 int next_size = 0;
2322 int cpu;
2323
2324 /*
2325 * If we are in a per_cpu trace file, don't bother by iterating over
2326 * all cpu and peek directly.
2327 */
2328 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2329 if (ring_buffer_empty_cpu(buffer, cpu_file))
2330 return NULL;
2331 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2332 if (ent_cpu)
2333 *ent_cpu = cpu_file;
2334
2335 return ent;
2336 }
2337
2338 for_each_tracing_cpu(cpu) {
2339
2340 if (ring_buffer_empty_cpu(buffer, cpu))
2341 continue;
2342
2343 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2344
2345 /*
2346 * Pick the entry with the smallest timestamp:
2347 */
2348 if (ent && (!next || ts < next_ts)) {
2349 next = ent;
2350 next_cpu = cpu;
2351 next_ts = ts;
2352 next_lost = lost_events;
2353 next_size = iter->ent_size;
2354 }
2355 }
2356
2357 iter->ent_size = next_size;
2358
2359 if (ent_cpu)
2360 *ent_cpu = next_cpu;
2361
2362 if (ent_ts)
2363 *ent_ts = next_ts;
2364
2365 if (missing_events)
2366 *missing_events = next_lost;
2367
2368 return next;
2369}
2370
2371/* Find the next real entry, without updating the iterator itself */
2372struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2373 int *ent_cpu, u64 *ent_ts)
2374{
2375 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2376}
2377
2378/* Find the next real entry, and increment the iterator to the next entry */
2379void *trace_find_next_entry_inc(struct trace_iterator *iter)
2380{
2381 iter->ent = __find_next_entry(iter, &iter->cpu,
2382 &iter->lost_events, &iter->ts);
2383
2384 if (iter->ent)
2385 trace_iterator_increment(iter);
2386
2387 return iter->ent ? iter : NULL;
2388}
2389
2390static void trace_consume(struct trace_iterator *iter)
2391{
2392 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2393 &iter->lost_events);
2394}
2395
2396static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2397{
2398 struct trace_iterator *iter = m->private;
2399 int i = (int)*pos;
2400 void *ent;
2401
2402 WARN_ON_ONCE(iter->leftover);
2403
2404 (*pos)++;
2405
2406 /* can't go backwards */
2407 if (iter->idx > i)
2408 return NULL;
2409
2410 if (iter->idx < 0)
2411 ent = trace_find_next_entry_inc(iter);
2412 else
2413 ent = iter;
2414
2415 while (ent && iter->idx < i)
2416 ent = trace_find_next_entry_inc(iter);
2417
2418 iter->pos = *pos;
2419
2420 return ent;
2421}
2422
2423void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2424{
2425 struct ring_buffer_event *event;
2426 struct ring_buffer_iter *buf_iter;
2427 unsigned long entries = 0;
2428 u64 ts;
2429
2430 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2431
2432 buf_iter = trace_buffer_iter(iter, cpu);
2433 if (!buf_iter)
2434 return;
2435
2436 ring_buffer_iter_reset(buf_iter);
2437
2438 /*
2439 * We could have the case with the max latency tracers
2440 * that a reset never took place on a cpu. This is evident
2441 * by the timestamp being before the start of the buffer.
2442 */
2443 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2444 if (ts >= iter->trace_buffer->time_start)
2445 break;
2446 entries++;
2447 ring_buffer_read(buf_iter, NULL);
2448 }
2449
2450 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2451}
2452
2453/*
2454 * The current tracer is copied to avoid a global locking
2455 * all around.
2456 */
2457static void *s_start(struct seq_file *m, loff_t *pos)
2458{
2459 struct trace_iterator *iter = m->private;
2460 struct trace_array *tr = iter->tr;
2461 int cpu_file = iter->cpu_file;
2462 void *p = NULL;
2463 loff_t l = 0;
2464 int cpu;
2465
2466 /*
2467 * copy the tracer to avoid using a global lock all around.
2468 * iter->trace is a copy of current_trace, the pointer to the
2469 * name may be used instead of a strcmp(), as iter->trace->name
2470 * will point to the same string as current_trace->name.
2471 */
2472 mutex_lock(&trace_types_lock);
2473 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2474 *iter->trace = *tr->current_trace;
2475 mutex_unlock(&trace_types_lock);
2476
2477#ifdef CONFIG_TRACER_MAX_TRACE
2478 if (iter->snapshot && iter->trace->use_max_tr)
2479 return ERR_PTR(-EBUSY);
2480#endif
2481
2482 if (!iter->snapshot)
2483 atomic_inc(&trace_record_cmdline_disabled);
2484
2485 if (*pos != iter->pos) {
2486 iter->ent = NULL;
2487 iter->cpu = 0;
2488 iter->idx = -1;
2489
2490 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2491 for_each_tracing_cpu(cpu)
2492 tracing_iter_reset(iter, cpu);
2493 } else
2494 tracing_iter_reset(iter, cpu_file);
2495
2496 iter->leftover = 0;
2497 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2498 ;
2499
2500 } else {
2501 /*
2502 * If we overflowed the seq_file before, then we want
2503 * to just reuse the trace_seq buffer again.
2504 */
2505 if (iter->leftover)
2506 p = iter;
2507 else {
2508 l = *pos - 1;
2509 p = s_next(m, p, &l);
2510 }
2511 }
2512
2513 trace_event_read_lock();
2514 trace_access_lock(cpu_file);
2515 return p;
2516}
2517
2518static void s_stop(struct seq_file *m, void *p)
2519{
2520 struct trace_iterator *iter = m->private;
2521
2522#ifdef CONFIG_TRACER_MAX_TRACE
2523 if (iter->snapshot && iter->trace->use_max_tr)
2524 return;
2525#endif
2526
2527 if (!iter->snapshot)
2528 atomic_dec(&trace_record_cmdline_disabled);
2529
2530 trace_access_unlock(iter->cpu_file);
2531 trace_event_read_unlock();
2532}
2533
2534static void
2535get_total_entries(struct trace_buffer *buf,
2536 unsigned long *total, unsigned long *entries)
2537{
2538 unsigned long count;
2539 int cpu;
2540
2541 *total = 0;
2542 *entries = 0;
2543
2544 for_each_tracing_cpu(cpu) {
2545 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2546 /*
2547 * If this buffer has skipped entries, then we hold all
2548 * entries for the trace and we need to ignore the
2549 * ones before the time stamp.
2550 */
2551 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2552 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2553 /* total is the same as the entries */
2554 *total += count;
2555 } else
2556 *total += count +
2557 ring_buffer_overrun_cpu(buf->buffer, cpu);
2558 *entries += count;
2559 }
2560}
2561
2562static void print_lat_help_header(struct seq_file *m)
2563{
2564 seq_puts(m, "# _------=> CPU# \n"
2565 "# / _-----=> irqs-off \n"
2566 "# | / _----=> need-resched \n"
2567 "# || / _---=> hardirq/softirq \n"
2568 "# ||| / _--=> preempt-depth \n"
2569 "# |||| / delay \n"
2570 "# cmd pid ||||| time | caller \n"
2571 "# \\ / ||||| \\ | / \n");
2572}
2573
2574static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2575{
2576 unsigned long total;
2577 unsigned long entries;
2578
2579 get_total_entries(buf, &total, &entries);
2580 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2581 entries, total, num_online_cpus());
2582 seq_puts(m, "#\n");
2583}
2584
2585static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2586{
2587 print_event_info(buf, m);
2588 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n"
2589 "# | | | | |\n");
2590}
2591
2592static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2593{
2594 print_event_info(buf, m);
2595 seq_puts(m, "# _-----=> irqs-off\n"
2596 "# / _----=> need-resched\n"
2597 "# | / _---=> hardirq/softirq\n"
2598 "# || / _--=> preempt-depth\n"
2599 "# ||| / delay\n"
2600 "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n"
2601 "# | | | |||| | |\n");
2602}
2603
2604void
2605print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2606{
2607 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
2608 struct trace_buffer *buf = iter->trace_buffer;
2609 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2610 struct tracer *type = iter->trace;
2611 unsigned long entries;
2612 unsigned long total;
2613 const char *name = "preemption";
2614
2615 name = type->name;
2616
2617 get_total_entries(buf, &total, &entries);
2618
2619 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2620 name, UTS_RELEASE);
2621 seq_puts(m, "# -----------------------------------"
2622 "---------------------------------\n");
2623 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2624 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2625 nsecs_to_usecs(data->saved_latency),
2626 entries,
2627 total,
2628 buf->cpu,
2629#if defined(CONFIG_PREEMPT_NONE)
2630 "server",
2631#elif defined(CONFIG_PREEMPT_VOLUNTARY)
2632 "desktop",
2633#elif defined(CONFIG_PREEMPT)
2634 "preempt",
2635#else
2636 "unknown",
2637#endif
2638 /* These are reserved for later use */
2639 0, 0, 0, 0);
2640#ifdef CONFIG_SMP
2641 seq_printf(m, " #P:%d)\n", num_online_cpus());
2642#else
2643 seq_puts(m, ")\n");
2644#endif
2645 seq_puts(m, "# -----------------\n");
2646 seq_printf(m, "# | task: %.16s-%d "
2647 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2648 data->comm, data->pid,
2649 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2650 data->policy, data->rt_priority);
2651 seq_puts(m, "# -----------------\n");
2652
2653 if (data->critical_start) {
2654 seq_puts(m, "# => started at: ");
2655 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2656 trace_print_seq(m, &iter->seq);
2657 seq_puts(m, "\n# => ended at: ");
2658 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2659 trace_print_seq(m, &iter->seq);
2660 seq_puts(m, "\n#\n");
2661 }
2662
2663 seq_puts(m, "#\n");
2664}
2665
2666static void test_cpu_buff_start(struct trace_iterator *iter)
2667{
2668 struct trace_seq *s = &iter->seq;
2669 struct trace_array *tr = iter->tr;
2670
2671 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
2672 return;
2673
2674 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2675 return;
2676
2677 if (iter->started && cpumask_test_cpu(iter->cpu, iter->started))
2678 return;
2679
2680 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2681 return;
2682
2683 if (iter->started)
2684 cpumask_set_cpu(iter->cpu, iter->started);
2685
2686 /* Don't print started cpu buffer for the first entry of the trace */
2687 if (iter->idx > 1)
2688 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2689 iter->cpu);
2690}
2691
2692static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2693{
2694 struct trace_array *tr = iter->tr;
2695 struct trace_seq *s = &iter->seq;
2696 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
2697 struct trace_entry *entry;
2698 struct trace_event *event;
2699
2700 entry = iter->ent;
2701
2702 test_cpu_buff_start(iter);
2703
2704 event = ftrace_find_event(entry->type);
2705
2706 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
2707 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2708 trace_print_lat_context(iter);
2709 else
2710 trace_print_context(iter);
2711 }
2712
2713 if (trace_seq_has_overflowed(s))
2714 return TRACE_TYPE_PARTIAL_LINE;
2715
2716 if (event)
2717 return event->funcs->trace(iter, sym_flags, event);
2718
2719 trace_seq_printf(s, "Unknown type %d\n", entry->type);
2720
2721 return trace_handle_return(s);
2722}
2723
2724static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2725{
2726 struct trace_array *tr = iter->tr;
2727 struct trace_seq *s = &iter->seq;
2728 struct trace_entry *entry;
2729 struct trace_event *event;
2730
2731 entry = iter->ent;
2732
2733 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
2734 trace_seq_printf(s, "%d %d %llu ",
2735 entry->pid, iter->cpu, iter->ts);
2736
2737 if (trace_seq_has_overflowed(s))
2738 return TRACE_TYPE_PARTIAL_LINE;
2739
2740 event = ftrace_find_event(entry->type);
2741 if (event)
2742 return event->funcs->raw(iter, 0, event);
2743
2744 trace_seq_printf(s, "%d ?\n", entry->type);
2745
2746 return trace_handle_return(s);
2747}
2748
2749static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2750{
2751 struct trace_array *tr = iter->tr;
2752 struct trace_seq *s = &iter->seq;
2753 unsigned char newline = '\n';
2754 struct trace_entry *entry;
2755 struct trace_event *event;
2756
2757 entry = iter->ent;
2758
2759 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
2760 SEQ_PUT_HEX_FIELD(s, entry->pid);
2761 SEQ_PUT_HEX_FIELD(s, iter->cpu);
2762 SEQ_PUT_HEX_FIELD(s, iter->ts);
2763 if (trace_seq_has_overflowed(s))
2764 return TRACE_TYPE_PARTIAL_LINE;
2765 }
2766
2767 event = ftrace_find_event(entry->type);
2768 if (event) {
2769 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2770 if (ret != TRACE_TYPE_HANDLED)
2771 return ret;
2772 }
2773
2774 SEQ_PUT_FIELD(s, newline);
2775
2776 return trace_handle_return(s);
2777}
2778
2779static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2780{
2781 struct trace_array *tr = iter->tr;
2782 struct trace_seq *s = &iter->seq;
2783 struct trace_entry *entry;
2784 struct trace_event *event;
2785
2786 entry = iter->ent;
2787
2788 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
2789 SEQ_PUT_FIELD(s, entry->pid);
2790 SEQ_PUT_FIELD(s, iter->cpu);
2791 SEQ_PUT_FIELD(s, iter->ts);
2792 if (trace_seq_has_overflowed(s))
2793 return TRACE_TYPE_PARTIAL_LINE;
2794 }
2795
2796 event = ftrace_find_event(entry->type);
2797 return event ? event->funcs->binary(iter, 0, event) :
2798 TRACE_TYPE_HANDLED;
2799}
2800
2801int trace_empty(struct trace_iterator *iter)
2802{
2803 struct ring_buffer_iter *buf_iter;
2804 int cpu;
2805
2806 /* If we are looking at one CPU buffer, only check that one */
2807 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2808 cpu = iter->cpu_file;
2809 buf_iter = trace_buffer_iter(iter, cpu);
2810 if (buf_iter) {
2811 if (!ring_buffer_iter_empty(buf_iter))
2812 return 0;
2813 } else {
2814 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2815 return 0;
2816 }
2817 return 1;
2818 }
2819
2820 for_each_tracing_cpu(cpu) {
2821 buf_iter = trace_buffer_iter(iter, cpu);
2822 if (buf_iter) {
2823 if (!ring_buffer_iter_empty(buf_iter))
2824 return 0;
2825 } else {
2826 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2827 return 0;
2828 }
2829 }
2830
2831 return 1;
2832}
2833
2834/* Called with trace_event_read_lock() held. */
2835enum print_line_t print_trace_line(struct trace_iterator *iter)
2836{
2837 struct trace_array *tr = iter->tr;
2838 unsigned long trace_flags = tr->trace_flags;
2839 enum print_line_t ret;
2840
2841 if (iter->lost_events) {
2842 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2843 iter->cpu, iter->lost_events);
2844 if (trace_seq_has_overflowed(&iter->seq))
2845 return TRACE_TYPE_PARTIAL_LINE;
2846 }
2847
2848 if (iter->trace && iter->trace->print_line) {
2849 ret = iter->trace->print_line(iter);
2850 if (ret != TRACE_TYPE_UNHANDLED)
2851 return ret;
2852 }
2853
2854 if (iter->ent->type == TRACE_BPUTS &&
2855 trace_flags & TRACE_ITER_PRINTK &&
2856 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2857 return trace_print_bputs_msg_only(iter);
2858
2859 if (iter->ent->type == TRACE_BPRINT &&
2860 trace_flags & TRACE_ITER_PRINTK &&
2861 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2862 return trace_print_bprintk_msg_only(iter);
2863
2864 if (iter->ent->type == TRACE_PRINT &&
2865 trace_flags & TRACE_ITER_PRINTK &&
2866 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2867 return trace_print_printk_msg_only(iter);
2868
2869 if (trace_flags & TRACE_ITER_BIN)
2870 return print_bin_fmt(iter);
2871
2872 if (trace_flags & TRACE_ITER_HEX)
2873 return print_hex_fmt(iter);
2874
2875 if (trace_flags & TRACE_ITER_RAW)
2876 return print_raw_fmt(iter);
2877
2878 return print_trace_fmt(iter);
2879}
2880
2881void trace_latency_header(struct seq_file *m)
2882{
2883 struct trace_iterator *iter = m->private;
2884 struct trace_array *tr = iter->tr;
2885
2886 /* print nothing if the buffers are empty */
2887 if (trace_empty(iter))
2888 return;
2889
2890 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2891 print_trace_header(m, iter);
2892
2893 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
2894 print_lat_help_header(m);
2895}
2896
2897void trace_default_header(struct seq_file *m)
2898{
2899 struct trace_iterator *iter = m->private;
2900 struct trace_array *tr = iter->tr;
2901 unsigned long trace_flags = tr->trace_flags;
2902
2903 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2904 return;
2905
2906 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2907 /* print nothing if the buffers are empty */
2908 if (trace_empty(iter))
2909 return;
2910 print_trace_header(m, iter);
2911 if (!(trace_flags & TRACE_ITER_VERBOSE))
2912 print_lat_help_header(m);
2913 } else {
2914 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2915 if (trace_flags & TRACE_ITER_IRQ_INFO)
2916 print_func_help_header_irq(iter->trace_buffer, m);
2917 else
2918 print_func_help_header(iter->trace_buffer, m);
2919 }
2920 }
2921}
2922
2923static void test_ftrace_alive(struct seq_file *m)
2924{
2925 if (!ftrace_is_dead())
2926 return;
2927 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
2928 "# MAY BE MISSING FUNCTION EVENTS\n");
2929}
2930
2931#ifdef CONFIG_TRACER_MAX_TRACE
2932static void show_snapshot_main_help(struct seq_file *m)
2933{
2934 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
2935 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
2936 "# Takes a snapshot of the main buffer.\n"
2937 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
2938 "# (Doesn't have to be '2' works with any number that\n"
2939 "# is not a '0' or '1')\n");
2940}
2941
2942static void show_snapshot_percpu_help(struct seq_file *m)
2943{
2944 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2945#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2946 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
2947 "# Takes a snapshot of the main buffer for this cpu.\n");
2948#else
2949 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
2950 "# Must use main snapshot file to allocate.\n");
2951#endif
2952 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
2953 "# (Doesn't have to be '2' works with any number that\n"
2954 "# is not a '0' or '1')\n");
2955}
2956
2957static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2958{
2959 if (iter->tr->allocated_snapshot)
2960 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
2961 else
2962 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
2963
2964 seq_puts(m, "# Snapshot commands:\n");
2965 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2966 show_snapshot_main_help(m);
2967 else
2968 show_snapshot_percpu_help(m);
2969}
2970#else
2971/* Should never be called */
2972static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2973#endif
2974
2975static int s_show(struct seq_file *m, void *v)
2976{
2977 struct trace_iterator *iter = v;
2978 int ret;
2979
2980 if (iter->ent == NULL) {
2981 if (iter->tr) {
2982 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2983 seq_puts(m, "#\n");
2984 test_ftrace_alive(m);
2985 }
2986 if (iter->snapshot && trace_empty(iter))
2987 print_snapshot_help(m, iter);
2988 else if (iter->trace && iter->trace->print_header)
2989 iter->trace->print_header(m);
2990 else
2991 trace_default_header(m);
2992
2993 } else if (iter->leftover) {
2994 /*
2995 * If we filled the seq_file buffer earlier, we
2996 * want to just show it now.
2997 */
2998 ret = trace_print_seq(m, &iter->seq);
2999
3000 /* ret should this time be zero, but you never know */
3001 iter->leftover = ret;
3002
3003 } else {
3004 print_trace_line(iter);
3005 ret = trace_print_seq(m, &iter->seq);
3006 /*
3007 * If we overflow the seq_file buffer, then it will
3008 * ask us for this data again at start up.
3009 * Use that instead.
3010 * ret is 0 if seq_file write succeeded.
3011 * -1 otherwise.
3012 */
3013 iter->leftover = ret;
3014 }
3015
3016 return 0;
3017}
3018
3019/*
3020 * Should be used after trace_array_get(), trace_types_lock
3021 * ensures that i_cdev was already initialized.
3022 */
3023static inline int tracing_get_cpu(struct inode *inode)
3024{
3025 if (inode->i_cdev) /* See trace_create_cpu_file() */
3026 return (long)inode->i_cdev - 1;
3027 return RING_BUFFER_ALL_CPUS;
3028}
3029
3030static const struct seq_operations tracer_seq_ops = {
3031 .start = s_start,
3032 .next = s_next,
3033 .stop = s_stop,
3034 .show = s_show,
3035};
3036
3037static struct trace_iterator *
3038__tracing_open(struct inode *inode, struct file *file, bool snapshot)
3039{
3040 struct trace_array *tr = inode->i_private;
3041 struct trace_iterator *iter;
3042 int cpu;
3043
3044 if (tracing_disabled)
3045 return ERR_PTR(-ENODEV);
3046
3047 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
3048 if (!iter)
3049 return ERR_PTR(-ENOMEM);
3050
3051 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
3052 GFP_KERNEL);
3053 if (!iter->buffer_iter)
3054 goto release;
3055
3056 /*
3057 * We make a copy of the current tracer to avoid concurrent
3058 * changes on it while we are reading.
3059 */
3060 mutex_lock(&trace_types_lock);
3061 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
3062 if (!iter->trace)
3063 goto fail;
3064
3065 *iter->trace = *tr->current_trace;
3066
3067 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
3068 goto fail;
3069
3070 iter->tr = tr;
3071
3072#ifdef CONFIG_TRACER_MAX_TRACE
3073 /* Currently only the top directory has a snapshot */
3074 if (tr->current_trace->print_max || snapshot)
3075 iter->trace_buffer = &tr->max_buffer;
3076 else
3077#endif
3078 iter->trace_buffer = &tr->trace_buffer;
3079 iter->snapshot = snapshot;
3080 iter->pos = -1;
3081 iter->cpu_file = tracing_get_cpu(inode);
3082 mutex_init(&iter->mutex);
3083
3084 /* Notify the tracer early; before we stop tracing. */
3085 if (iter->trace && iter->trace->open)
3086 iter->trace->open(iter);
3087
3088 /* Annotate start of buffers if we had overruns */
3089 if (ring_buffer_overruns(iter->trace_buffer->buffer))
3090 iter->iter_flags |= TRACE_FILE_ANNOTATE;
3091
3092 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3093 if (trace_clocks[tr->clock_id].in_ns)
3094 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3095
3096 /* stop the trace while dumping if we are not opening "snapshot" */
3097 if (!iter->snapshot)
3098 tracing_stop_tr(tr);
3099
3100 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
3101 for_each_tracing_cpu(cpu) {
3102 iter->buffer_iter[cpu] =
3103 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
3104 }
3105 ring_buffer_read_prepare_sync();
3106 for_each_tracing_cpu(cpu) {
3107 ring_buffer_read_start(iter->buffer_iter[cpu]);
3108 tracing_iter_reset(iter, cpu);
3109 }
3110 } else {
3111 cpu = iter->cpu_file;
3112 iter->buffer_iter[cpu] =
3113 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
3114 ring_buffer_read_prepare_sync();
3115 ring_buffer_read_start(iter->buffer_iter[cpu]);
3116 tracing_iter_reset(iter, cpu);
3117 }
3118
3119 mutex_unlock(&trace_types_lock);
3120
3121 return iter;
3122
3123 fail:
3124 mutex_unlock(&trace_types_lock);
3125 kfree(iter->trace);
3126 kfree(iter->buffer_iter);
3127release:
3128 seq_release_private(inode, file);
3129 return ERR_PTR(-ENOMEM);
3130}
3131
3132int tracing_open_generic(struct inode *inode, struct file *filp)
3133{
3134 if (tracing_disabled)
3135 return -ENODEV;
3136
3137 filp->private_data = inode->i_private;
3138 return 0;
3139}
3140
3141bool tracing_is_disabled(void)
3142{
3143 return (tracing_disabled) ? true: false;
3144}
3145
3146/*
3147 * Open and update trace_array ref count.
3148 * Must have the current trace_array passed to it.
3149 */
3150static int tracing_open_generic_tr(struct inode *inode, struct file *filp)
3151{
3152 struct trace_array *tr = inode->i_private;
3153
3154 if (tracing_disabled)
3155 return -ENODEV;
3156
3157 if (trace_array_get(tr) < 0)
3158 return -ENODEV;
3159
3160 filp->private_data = inode->i_private;
3161
3162 return 0;
3163}
3164
3165static int tracing_release(struct inode *inode, struct file *file)
3166{
3167 struct trace_array *tr = inode->i_private;
3168 struct seq_file *m = file->private_data;
3169 struct trace_iterator *iter;
3170 int cpu;
3171
3172 if (!(file->f_mode & FMODE_READ)) {
3173 trace_array_put(tr);
3174 return 0;
3175 }
3176
3177 /* Writes do not use seq_file */
3178 iter = m->private;
3179 mutex_lock(&trace_types_lock);
3180
3181 for_each_tracing_cpu(cpu) {
3182 if (iter->buffer_iter[cpu])
3183 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3184 }
3185
3186 if (iter->trace && iter->trace->close)
3187 iter->trace->close(iter);
3188
3189 if (!iter->snapshot)
3190 /* reenable tracing if it was previously enabled */
3191 tracing_start_tr(tr);
3192
3193 __trace_array_put(tr);
3194
3195 mutex_unlock(&trace_types_lock);
3196
3197 mutex_destroy(&iter->mutex);
3198 free_cpumask_var(iter->started);
3199 kfree(iter->trace);
3200 kfree(iter->buffer_iter);
3201 seq_release_private(inode, file);
3202
3203 return 0;
3204}
3205
3206static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3207{
3208 struct trace_array *tr = inode->i_private;
3209
3210 trace_array_put(tr);
3211 return 0;
3212}
3213
3214static int tracing_single_release_tr(struct inode *inode, struct file *file)
3215{
3216 struct trace_array *tr = inode->i_private;
3217
3218 trace_array_put(tr);
3219
3220 return single_release(inode, file);
3221}
3222
3223static int tracing_open(struct inode *inode, struct file *file)
3224{
3225 struct trace_array *tr = inode->i_private;
3226 struct trace_iterator *iter;
3227 int ret = 0;
3228
3229 if (trace_array_get(tr) < 0)
3230 return -ENODEV;
3231
3232 /* If this file was open for write, then erase contents */
3233 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
3234 int cpu = tracing_get_cpu(inode);
3235
3236 if (cpu == RING_BUFFER_ALL_CPUS)
3237 tracing_reset_online_cpus(&tr->trace_buffer);
3238 else
3239 tracing_reset(&tr->trace_buffer, cpu);
3240 }
3241
3242 if (file->f_mode & FMODE_READ) {
3243 iter = __tracing_open(inode, file, false);
3244 if (IS_ERR(iter))
3245 ret = PTR_ERR(iter);
3246 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
3247 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3248 }
3249
3250 if (ret < 0)
3251 trace_array_put(tr);
3252
3253 return ret;
3254}
3255
3256/*
3257 * Some tracers are not suitable for instance buffers.
3258 * A tracer is always available for the global array (toplevel)
3259 * or if it explicitly states that it is.
3260 */
3261static bool
3262trace_ok_for_array(struct tracer *t, struct trace_array *tr)
3263{
3264 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
3265}
3266
3267/* Find the next tracer that this trace array may use */
3268static struct tracer *
3269get_tracer_for_array(struct trace_array *tr, struct tracer *t)
3270{
3271 while (t && !trace_ok_for_array(t, tr))
3272 t = t->next;
3273
3274 return t;
3275}
3276
3277static void *
3278t_next(struct seq_file *m, void *v, loff_t *pos)
3279{
3280 struct trace_array *tr = m->private;
3281 struct tracer *t = v;
3282
3283 (*pos)++;
3284
3285 if (t)
3286 t = get_tracer_for_array(tr, t->next);
3287
3288 return t;
3289}
3290
3291static void *t_start(struct seq_file *m, loff_t *pos)
3292{
3293 struct trace_array *tr = m->private;
3294 struct tracer *t;
3295 loff_t l = 0;
3296
3297 mutex_lock(&trace_types_lock);
3298
3299 t = get_tracer_for_array(tr, trace_types);
3300 for (; t && l < *pos; t = t_next(m, t, &l))
3301 ;
3302
3303 return t;
3304}
3305
3306static void t_stop(struct seq_file *m, void *p)
3307{
3308 mutex_unlock(&trace_types_lock);
3309}
3310
3311static int t_show(struct seq_file *m, void *v)
3312{
3313 struct tracer *t = v;
3314
3315 if (!t)
3316 return 0;
3317
3318 seq_puts(m, t->name);
3319 if (t->next)
3320 seq_putc(m, ' ');
3321 else
3322 seq_putc(m, '\n');
3323
3324 return 0;
3325}
3326
3327static const struct seq_operations show_traces_seq_ops = {
3328 .start = t_start,
3329 .next = t_next,
3330 .stop = t_stop,
3331 .show = t_show,
3332};
3333
3334static int show_traces_open(struct inode *inode, struct file *file)
3335{
3336 struct trace_array *tr = inode->i_private;
3337 struct seq_file *m;
3338 int ret;
3339
3340 if (tracing_disabled)
3341 return -ENODEV;
3342
3343 ret = seq_open(file, &show_traces_seq_ops);
3344 if (ret)
3345 return ret;
3346
3347 m = file->private_data;
3348 m->private = tr;
3349
3350 return 0;
3351}
3352
3353static ssize_t
3354tracing_write_stub(struct file *filp, const char __user *ubuf,
3355 size_t count, loff_t *ppos)
3356{
3357 return count;
3358}
3359
3360loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
3361{
3362 int ret;
3363
3364 if (file->f_mode & FMODE_READ)
3365 ret = seq_lseek(file, offset, whence);
3366 else
3367 file->f_pos = ret = 0;
3368
3369 return ret;
3370}
3371
3372static const struct file_operations tracing_fops = {
3373 .open = tracing_open,
3374 .read = seq_read,
3375 .write = tracing_write_stub,
3376 .llseek = tracing_lseek,
3377 .release = tracing_release,
3378};
3379
3380static const struct file_operations show_traces_fops = {
3381 .open = show_traces_open,
3382 .read = seq_read,
3383 .release = seq_release,
3384 .llseek = seq_lseek,
3385};
3386
3387/*
3388 * The tracer itself will not take this lock, but still we want
3389 * to provide a consistent cpumask to user-space:
3390 */
3391static DEFINE_MUTEX(tracing_cpumask_update_lock);
3392
3393/*
3394 * Temporary storage for the character representation of the
3395 * CPU bitmask (and one more byte for the newline):
3396 */
3397static char mask_str[NR_CPUS + 1];
3398
3399static ssize_t
3400tracing_cpumask_read(struct file *filp, char __user *ubuf,
3401 size_t count, loff_t *ppos)
3402{
3403 struct trace_array *tr = file_inode(filp)->i_private;
3404 int len;
3405
3406 mutex_lock(&tracing_cpumask_update_lock);
3407
3408 len = snprintf(mask_str, count, "%*pb\n",
3409 cpumask_pr_args(tr->tracing_cpumask));
3410 if (len >= count) {
3411 count = -EINVAL;
3412 goto out_err;
3413 }
3414 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3415
3416out_err:
3417 mutex_unlock(&tracing_cpumask_update_lock);
3418
3419 return count;
3420}
3421
3422static ssize_t
3423tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3424 size_t count, loff_t *ppos)
3425{
3426 struct trace_array *tr = file_inode(filp)->i_private;
3427 cpumask_var_t tracing_cpumask_new;
3428 int err, cpu;
3429
3430 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3431 return -ENOMEM;
3432
3433 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3434 if (err)
3435 goto err_unlock;
3436
3437 mutex_lock(&tracing_cpumask_update_lock);
3438
3439 local_irq_disable();
3440 arch_spin_lock(&tr->max_lock);
3441 for_each_tracing_cpu(cpu) {
3442 /*
3443 * Increase/decrease the disabled counter if we are
3444 * about to flip a bit in the cpumask:
3445 */
3446 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3447 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3448 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3449 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3450 }
3451 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3452 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3453 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3454 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3455 }
3456 }
3457 arch_spin_unlock(&tr->max_lock);
3458 local_irq_enable();
3459
3460 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
3461
3462 mutex_unlock(&tracing_cpumask_update_lock);
3463 free_cpumask_var(tracing_cpumask_new);
3464
3465 return count;
3466
3467err_unlock:
3468 free_cpumask_var(tracing_cpumask_new);
3469
3470 return err;
3471}
3472
3473static const struct file_operations tracing_cpumask_fops = {
3474 .open = tracing_open_generic_tr,
3475 .read = tracing_cpumask_read,
3476 .write = tracing_cpumask_write,
3477 .release = tracing_release_generic_tr,
3478 .llseek = generic_file_llseek,
3479};
3480
3481static int tracing_trace_options_show(struct seq_file *m, void *v)
3482{
3483 struct tracer_opt *trace_opts;
3484 struct trace_array *tr = m->private;
3485 u32 tracer_flags;
3486 int i;
3487
3488 mutex_lock(&trace_types_lock);
3489 tracer_flags = tr->current_trace->flags->val;
3490 trace_opts = tr->current_trace->flags->opts;
3491
3492 for (i = 0; trace_options[i]; i++) {
3493 if (tr->trace_flags & (1 << i))
3494 seq_printf(m, "%s\n", trace_options[i]);
3495 else
3496 seq_printf(m, "no%s\n", trace_options[i]);
3497 }
3498
3499 for (i = 0; trace_opts[i].name; i++) {
3500 if (tracer_flags & trace_opts[i].bit)
3501 seq_printf(m, "%s\n", trace_opts[i].name);
3502 else
3503 seq_printf(m, "no%s\n", trace_opts[i].name);
3504 }
3505 mutex_unlock(&trace_types_lock);
3506
3507 return 0;
3508}
3509
3510static int __set_tracer_option(struct trace_array *tr,
3511 struct tracer_flags *tracer_flags,
3512 struct tracer_opt *opts, int neg)
3513{
3514 struct tracer *trace = tracer_flags->trace;
3515 int ret;
3516
3517 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
3518 if (ret)
3519 return ret;
3520
3521 if (neg)
3522 tracer_flags->val &= ~opts->bit;
3523 else
3524 tracer_flags->val |= opts->bit;
3525 return 0;
3526}
3527
3528/* Try to assign a tracer specific option */
3529static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
3530{
3531 struct tracer *trace = tr->current_trace;
3532 struct tracer_flags *tracer_flags = trace->flags;
3533 struct tracer_opt *opts = NULL;
3534 int i;
3535
3536 for (i = 0; tracer_flags->opts[i].name; i++) {
3537 opts = &tracer_flags->opts[i];
3538
3539 if (strcmp(cmp, opts->name) == 0)
3540 return __set_tracer_option(tr, trace->flags, opts, neg);
3541 }
3542
3543 return -EINVAL;
3544}
3545
3546/* Some tracers require overwrite to stay enabled */
3547int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3548{
3549 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3550 return -1;
3551
3552 return 0;
3553}
3554
3555int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3556{
3557 /* do nothing if flag is already set */
3558 if (!!(tr->trace_flags & mask) == !!enabled)
3559 return 0;
3560
3561 /* Give the tracer a chance to approve the change */
3562 if (tr->current_trace->flag_changed)
3563 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
3564 return -EINVAL;
3565
3566 if (enabled)
3567 tr->trace_flags |= mask;
3568 else
3569 tr->trace_flags &= ~mask;
3570
3571 if (mask == TRACE_ITER_RECORD_CMD)
3572 trace_event_enable_cmd_record(enabled);
3573
3574 if (mask == TRACE_ITER_OVERWRITE) {
3575 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3576#ifdef CONFIG_TRACER_MAX_TRACE
3577 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3578#endif
3579 }
3580
3581 if (mask == TRACE_ITER_PRINTK) {
3582 trace_printk_start_stop_comm(enabled);
3583 trace_printk_control(enabled);
3584 }
3585
3586 return 0;
3587}
3588
3589static int trace_set_options(struct trace_array *tr, char *option)
3590{
3591 char *cmp;
3592 int neg = 0;
3593 int ret = -ENODEV;
3594 int i;
3595 size_t orig_len = strlen(option);
3596
3597 cmp = strstrip(option);
3598
3599 if (strncmp(cmp, "no", 2) == 0) {
3600 neg = 1;
3601 cmp += 2;
3602 }
3603
3604 mutex_lock(&trace_types_lock);
3605
3606 for (i = 0; trace_options[i]; i++) {
3607 if (strcmp(cmp, trace_options[i]) == 0) {
3608 ret = set_tracer_flag(tr, 1 << i, !neg);
3609 break;
3610 }
3611 }
3612
3613 /* If no option could be set, test the specific tracer options */
3614 if (!trace_options[i])
3615 ret = set_tracer_option(tr, cmp, neg);
3616
3617 mutex_unlock(&trace_types_lock);
3618
3619 /*
3620 * If the first trailing whitespace is replaced with '\0' by strstrip,
3621 * turn it back into a space.
3622 */
3623 if (orig_len > strlen(option))
3624 option[strlen(option)] = ' ';
3625
3626 return ret;
3627}
3628
3629static void __init apply_trace_boot_options(void)
3630{
3631 char *buf = trace_boot_options_buf;
3632 char *option;
3633
3634 while (true) {
3635 option = strsep(&buf, ",");
3636
3637 if (!option)
3638 break;
3639
3640 if (*option)
3641 trace_set_options(&global_trace, option);
3642
3643 /* Put back the comma to allow this to be called again */
3644 if (buf)
3645 *(buf - 1) = ',';
3646 }
3647}
3648
3649static ssize_t
3650tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3651 size_t cnt, loff_t *ppos)
3652{
3653 struct seq_file *m = filp->private_data;
3654 struct trace_array *tr = m->private;
3655 char buf[64];
3656 int ret;
3657
3658 if (cnt >= sizeof(buf))
3659 return -EINVAL;
3660
3661 if (copy_from_user(&buf, ubuf, cnt))
3662 return -EFAULT;
3663
3664 buf[cnt] = 0;
3665
3666 ret = trace_set_options(tr, buf);
3667 if (ret < 0)
3668 return ret;
3669
3670 *ppos += cnt;
3671
3672 return cnt;
3673}
3674
3675static int tracing_trace_options_open(struct inode *inode, struct file *file)
3676{
3677 struct trace_array *tr = inode->i_private;
3678 int ret;
3679
3680 if (tracing_disabled)
3681 return -ENODEV;
3682
3683 if (trace_array_get(tr) < 0)
3684 return -ENODEV;
3685
3686 ret = single_open(file, tracing_trace_options_show, inode->i_private);
3687 if (ret < 0)
3688 trace_array_put(tr);
3689
3690 return ret;
3691}
3692
3693static const struct file_operations tracing_iter_fops = {
3694 .open = tracing_trace_options_open,
3695 .read = seq_read,
3696 .llseek = seq_lseek,
3697 .release = tracing_single_release_tr,
3698 .write = tracing_trace_options_write,
3699};
3700
3701static const char readme_msg[] =
3702 "tracing mini-HOWTO:\n\n"
3703 "# echo 0 > tracing_on : quick way to disable tracing\n"
3704 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3705 " Important files:\n"
3706 " trace\t\t\t- The static contents of the buffer\n"
3707 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3708 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3709 " current_tracer\t- function and latency tracers\n"
3710 " available_tracers\t- list of configured tracers for current_tracer\n"
3711 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3712 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3713 " trace_clock\t\t-change the clock used to order events\n"
3714 " local: Per cpu clock but may not be synced across CPUs\n"
3715 " global: Synced across CPUs but slows tracing down.\n"
3716 " counter: Not a clock, but just an increment\n"
3717 " uptime: Jiffy counter from time of boot\n"
3718 " perf: Same clock that perf events use\n"
3719#ifdef CONFIG_X86_64
3720 " x86-tsc: TSC cycle counter\n"
3721#endif
3722 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3723 " tracing_cpumask\t- Limit which CPUs to trace\n"
3724 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3725 "\t\t\t Remove sub-buffer with rmdir\n"
3726 " trace_options\t\t- Set format or modify how tracing happens\n"
3727 "\t\t\t Disable an option by adding a suffix 'no' to the\n"
3728 "\t\t\t option name\n"
3729 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
3730#ifdef CONFIG_DYNAMIC_FTRACE
3731 "\n available_filter_functions - list of functions that can be filtered on\n"
3732 " set_ftrace_filter\t- echo function name in here to only trace these\n"
3733 "\t\t\t functions\n"
3734 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3735 "\t modules: Can select a group via module\n"
3736 "\t Format: :mod:<module-name>\n"
3737 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
3738 "\t triggers: a command to perform when function is hit\n"
3739 "\t Format: <function>:<trigger>[:count]\n"
3740 "\t trigger: traceon, traceoff\n"
3741 "\t\t enable_event:<system>:<event>\n"
3742 "\t\t disable_event:<system>:<event>\n"
3743#ifdef CONFIG_STACKTRACE
3744 "\t\t stacktrace\n"
3745#endif
3746#ifdef CONFIG_TRACER_SNAPSHOT
3747 "\t\t snapshot\n"
3748#endif
3749 "\t\t dump\n"
3750 "\t\t cpudump\n"
3751 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
3752 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
3753 "\t The first one will disable tracing every time do_fault is hit\n"
3754 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
3755 "\t The first time do trap is hit and it disables tracing, the\n"
3756 "\t counter will decrement to 2. If tracing is already disabled,\n"
3757 "\t the counter will not decrement. It only decrements when the\n"
3758 "\t trigger did work\n"
3759 "\t To remove trigger without count:\n"
3760 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
3761 "\t To remove trigger with a count:\n"
3762 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3763 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3764 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3765 "\t modules: Can select a group via module command :mod:\n"
3766 "\t Does not accept triggers\n"
3767#endif /* CONFIG_DYNAMIC_FTRACE */
3768#ifdef CONFIG_FUNCTION_TRACER
3769 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
3770 "\t\t (function)\n"
3771#endif
3772#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3773 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3774 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
3775 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3776#endif
3777#ifdef CONFIG_TRACER_SNAPSHOT
3778 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
3779 "\t\t\t snapshot buffer. Read the contents for more\n"
3780 "\t\t\t information\n"
3781#endif
3782#ifdef CONFIG_STACK_TRACER
3783 " stack_trace\t\t- Shows the max stack trace when active\n"
3784 " stack_max_size\t- Shows current max stack size that was traced\n"
3785 "\t\t\t Write into this file to reset the max size (trigger a\n"
3786 "\t\t\t new trace)\n"
3787#ifdef CONFIG_DYNAMIC_FTRACE
3788 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
3789 "\t\t\t traces\n"
3790#endif
3791#endif /* CONFIG_STACK_TRACER */
3792 " events/\t\t- Directory containing all trace event subsystems:\n"
3793 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
3794 " events/<system>/\t- Directory containing all trace events for <system>:\n"
3795 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
3796 "\t\t\t events\n"
3797 " filter\t\t- If set, only events passing filter are traced\n"
3798 " events/<system>/<event>/\t- Directory containing control files for\n"
3799 "\t\t\t <event>:\n"
3800 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
3801 " filter\t\t- If set, only events passing filter are traced\n"
3802 " trigger\t\t- If set, a command to perform when event is hit\n"
3803 "\t Format: <trigger>[:count][if <filter>]\n"
3804 "\t trigger: traceon, traceoff\n"
3805 "\t enable_event:<system>:<event>\n"
3806 "\t disable_event:<system>:<event>\n"
3807#ifdef CONFIG_STACKTRACE
3808 "\t\t stacktrace\n"
3809#endif
3810#ifdef CONFIG_TRACER_SNAPSHOT
3811 "\t\t snapshot\n"
3812#endif
3813 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
3814 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
3815 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
3816 "\t events/block/block_unplug/trigger\n"
3817 "\t The first disables tracing every time block_unplug is hit.\n"
3818 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
3819 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
3820 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
3821 "\t Like function triggers, the counter is only decremented if it\n"
3822 "\t enabled or disabled tracing.\n"
3823 "\t To remove a trigger without a count:\n"
3824 "\t echo '!<trigger> > <system>/<event>/trigger\n"
3825 "\t To remove a trigger with a count:\n"
3826 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
3827 "\t Filters can be ignored when removing a trigger.\n"
3828;
3829
3830static ssize_t
3831tracing_readme_read(struct file *filp, char __user *ubuf,
3832 size_t cnt, loff_t *ppos)
3833{
3834 return simple_read_from_buffer(ubuf, cnt, ppos,
3835 readme_msg, strlen(readme_msg));
3836}
3837
3838static const struct file_operations tracing_readme_fops = {
3839 .open = tracing_open_generic,
3840 .read = tracing_readme_read,
3841 .llseek = generic_file_llseek,
3842};
3843
3844static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
3845{
3846 unsigned int *ptr = v;
3847
3848 if (*pos || m->count)
3849 ptr++;
3850
3851 (*pos)++;
3852
3853 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
3854 ptr++) {
3855 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
3856 continue;
3857
3858 return ptr;
3859 }
3860
3861 return NULL;
3862}
3863
3864static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
3865{
3866 void *v;
3867 loff_t l = 0;
3868
3869 preempt_disable();
3870 arch_spin_lock(&trace_cmdline_lock);
3871
3872 v = &savedcmd->map_cmdline_to_pid[0];
3873 while (l <= *pos) {
3874 v = saved_cmdlines_next(m, v, &l);
3875 if (!v)
3876 return NULL;
3877 }
3878
3879 return v;
3880}
3881
3882static void saved_cmdlines_stop(struct seq_file *m, void *v)
3883{
3884 arch_spin_unlock(&trace_cmdline_lock);
3885 preempt_enable();
3886}
3887
3888static int saved_cmdlines_show(struct seq_file *m, void *v)
3889{
3890 char buf[TASK_COMM_LEN];
3891 unsigned int *pid = v;
3892
3893 __trace_find_cmdline(*pid, buf);
3894 seq_printf(m, "%d %s\n", *pid, buf);
3895 return 0;
3896}
3897
3898static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
3899 .start = saved_cmdlines_start,
3900 .next = saved_cmdlines_next,
3901 .stop = saved_cmdlines_stop,
3902 .show = saved_cmdlines_show,
3903};
3904
3905static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
3906{
3907 if (tracing_disabled)
3908 return -ENODEV;
3909
3910 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
3911}
3912
3913static const struct file_operations tracing_saved_cmdlines_fops = {
3914 .open = tracing_saved_cmdlines_open,
3915 .read = seq_read,
3916 .llseek = seq_lseek,
3917 .release = seq_release,
3918};
3919
3920static ssize_t
3921tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
3922 size_t cnt, loff_t *ppos)
3923{
3924 char buf[64];
3925 int r;
3926
3927 arch_spin_lock(&trace_cmdline_lock);
3928 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
3929 arch_spin_unlock(&trace_cmdline_lock);
3930
3931 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3932}
3933
3934static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
3935{
3936 kfree(s->saved_cmdlines);
3937 kfree(s->map_cmdline_to_pid);
3938 kfree(s);
3939}
3940
3941static int tracing_resize_saved_cmdlines(unsigned int val)
3942{
3943 struct saved_cmdlines_buffer *s, *savedcmd_temp;
3944
3945 s = kmalloc(sizeof(*s), GFP_KERNEL);
3946 if (!s)
3947 return -ENOMEM;
3948
3949 if (allocate_cmdlines_buffer(val, s) < 0) {
3950 kfree(s);
3951 return -ENOMEM;
3952 }
3953
3954 arch_spin_lock(&trace_cmdline_lock);
3955 savedcmd_temp = savedcmd;
3956 savedcmd = s;
3957 arch_spin_unlock(&trace_cmdline_lock);
3958 free_saved_cmdlines_buffer(savedcmd_temp);
3959
3960 return 0;
3961}
3962
3963static ssize_t
3964tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
3965 size_t cnt, loff_t *ppos)
3966{
3967 unsigned long val;
3968 int ret;
3969
3970 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3971 if (ret)
3972 return ret;
3973
3974 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
3975 if (!val || val > PID_MAX_DEFAULT)
3976 return -EINVAL;
3977
3978 ret = tracing_resize_saved_cmdlines((unsigned int)val);
3979 if (ret < 0)
3980 return ret;
3981
3982 *ppos += cnt;
3983
3984 return cnt;
3985}
3986
3987static const struct file_operations tracing_saved_cmdlines_size_fops = {
3988 .open = tracing_open_generic,
3989 .read = tracing_saved_cmdlines_size_read,
3990 .write = tracing_saved_cmdlines_size_write,
3991};
3992
3993#ifdef CONFIG_TRACE_ENUM_MAP_FILE
3994static union trace_enum_map_item *
3995update_enum_map(union trace_enum_map_item *ptr)
3996{
3997 if (!ptr->map.enum_string) {
3998 if (ptr->tail.next) {
3999 ptr = ptr->tail.next;
4000 /* Set ptr to the next real item (skip head) */
4001 ptr++;
4002 } else
4003 return NULL;
4004 }
4005 return ptr;
4006}
4007
4008static void *enum_map_next(struct seq_file *m, void *v, loff_t *pos)
4009{
4010 union trace_enum_map_item *ptr = v;
4011
4012 /*
4013 * Paranoid! If ptr points to end, we don't want to increment past it.
4014 * This really should never happen.
4015 */
4016 ptr = update_enum_map(ptr);
4017 if (WARN_ON_ONCE(!ptr))
4018 return NULL;
4019
4020 ptr++;
4021
4022 (*pos)++;
4023
4024 ptr = update_enum_map(ptr);
4025
4026 return ptr;
4027}
4028
4029static void *enum_map_start(struct seq_file *m, loff_t *pos)
4030{
4031 union trace_enum_map_item *v;
4032 loff_t l = 0;
4033
4034 mutex_lock(&trace_enum_mutex);
4035
4036 v = trace_enum_maps;
4037 if (v)
4038 v++;
4039
4040 while (v && l < *pos) {
4041 v = enum_map_next(m, v, &l);
4042 }
4043
4044 return v;
4045}
4046
4047static void enum_map_stop(struct seq_file *m, void *v)
4048{
4049 mutex_unlock(&trace_enum_mutex);
4050}
4051
4052static int enum_map_show(struct seq_file *m, void *v)
4053{
4054 union trace_enum_map_item *ptr = v;
4055
4056 seq_printf(m, "%s %ld (%s)\n",
4057 ptr->map.enum_string, ptr->map.enum_value,
4058 ptr->map.system);
4059
4060 return 0;
4061}
4062
4063static const struct seq_operations tracing_enum_map_seq_ops = {
4064 .start = enum_map_start,
4065 .next = enum_map_next,
4066 .stop = enum_map_stop,
4067 .show = enum_map_show,
4068};
4069
4070static int tracing_enum_map_open(struct inode *inode, struct file *filp)
4071{
4072 if (tracing_disabled)
4073 return -ENODEV;
4074
4075 return seq_open(filp, &tracing_enum_map_seq_ops);
4076}
4077
4078static const struct file_operations tracing_enum_map_fops = {
4079 .open = tracing_enum_map_open,
4080 .read = seq_read,
4081 .llseek = seq_lseek,
4082 .release = seq_release,
4083};
4084
4085static inline union trace_enum_map_item *
4086trace_enum_jmp_to_tail(union trace_enum_map_item *ptr)
4087{
4088 /* Return tail of array given the head */
4089 return ptr + ptr->head.length + 1;
4090}
4091
4092static void
4093trace_insert_enum_map_file(struct module *mod, struct trace_enum_map **start,
4094 int len)
4095{
4096 struct trace_enum_map **stop;
4097 struct trace_enum_map **map;
4098 union trace_enum_map_item *map_array;
4099 union trace_enum_map_item *ptr;
4100
4101 stop = start + len;
4102
4103 /*
4104 * The trace_enum_maps contains the map plus a head and tail item,
4105 * where the head holds the module and length of array, and the
4106 * tail holds a pointer to the next list.
4107 */
4108 map_array = kmalloc(sizeof(*map_array) * (len + 2), GFP_KERNEL);
4109 if (!map_array) {
4110 pr_warn("Unable to allocate trace enum mapping\n");
4111 return;
4112 }
4113
4114 mutex_lock(&trace_enum_mutex);
4115
4116 if (!trace_enum_maps)
4117 trace_enum_maps = map_array;
4118 else {
4119 ptr = trace_enum_maps;
4120 for (;;) {
4121 ptr = trace_enum_jmp_to_tail(ptr);
4122 if (!ptr->tail.next)
4123 break;
4124 ptr = ptr->tail.next;
4125
4126 }
4127 ptr->tail.next = map_array;
4128 }
4129 map_array->head.mod = mod;
4130 map_array->head.length = len;
4131 map_array++;
4132
4133 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
4134 map_array->map = **map;
4135 map_array++;
4136 }
4137 memset(map_array, 0, sizeof(*map_array));
4138
4139 mutex_unlock(&trace_enum_mutex);
4140}
4141
4142static void trace_create_enum_file(struct dentry *d_tracer)
4143{
4144 trace_create_file("enum_map", 0444, d_tracer,
4145 NULL, &tracing_enum_map_fops);
4146}
4147
4148#else /* CONFIG_TRACE_ENUM_MAP_FILE */
4149static inline void trace_create_enum_file(struct dentry *d_tracer) { }
4150static inline void trace_insert_enum_map_file(struct module *mod,
4151 struct trace_enum_map **start, int len) { }
4152#endif /* !CONFIG_TRACE_ENUM_MAP_FILE */
4153
4154static void trace_insert_enum_map(struct module *mod,
4155 struct trace_enum_map **start, int len)
4156{
4157 struct trace_enum_map **map;
4158
4159 if (len <= 0)
4160 return;
4161
4162 map = start;
4163
4164 trace_event_enum_update(map, len);
4165
4166 trace_insert_enum_map_file(mod, start, len);
4167}
4168
4169static ssize_t
4170tracing_set_trace_read(struct file *filp, char __user *ubuf,
4171 size_t cnt, loff_t *ppos)
4172{
4173 struct trace_array *tr = filp->private_data;
4174 char buf[MAX_TRACER_SIZE+2];
4175 int r;
4176
4177 mutex_lock(&trace_types_lock);
4178 r = sprintf(buf, "%s\n", tr->current_trace->name);
4179 mutex_unlock(&trace_types_lock);
4180
4181 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4182}
4183
4184int tracer_init(struct tracer *t, struct trace_array *tr)
4185{
4186 tracing_reset_online_cpus(&tr->trace_buffer);
4187 return t->init(tr);
4188}
4189
4190static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
4191{
4192 int cpu;
4193
4194 for_each_tracing_cpu(cpu)
4195 per_cpu_ptr(buf->data, cpu)->entries = val;
4196}
4197
4198#ifdef CONFIG_TRACER_MAX_TRACE
4199/* resize @tr's buffer to the size of @size_tr's entries */
4200static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
4201 struct trace_buffer *size_buf, int cpu_id)
4202{
4203 int cpu, ret = 0;
4204
4205 if (cpu_id == RING_BUFFER_ALL_CPUS) {
4206 for_each_tracing_cpu(cpu) {
4207 ret = ring_buffer_resize(trace_buf->buffer,
4208 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
4209 if (ret < 0)
4210 break;
4211 per_cpu_ptr(trace_buf->data, cpu)->entries =
4212 per_cpu_ptr(size_buf->data, cpu)->entries;
4213 }
4214 } else {
4215 ret = ring_buffer_resize(trace_buf->buffer,
4216 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
4217 if (ret == 0)
4218 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
4219 per_cpu_ptr(size_buf->data, cpu_id)->entries;
4220 }
4221
4222 return ret;
4223}
4224#endif /* CONFIG_TRACER_MAX_TRACE */
4225
4226static int __tracing_resize_ring_buffer(struct trace_array *tr,
4227 unsigned long size, int cpu)
4228{
4229 int ret;
4230
4231 /*
4232 * If kernel or user changes the size of the ring buffer
4233 * we use the size that was given, and we can forget about
4234 * expanding it later.
4235 */
4236 ring_buffer_expanded = true;
4237
4238 /* May be called before buffers are initialized */
4239 if (!tr->trace_buffer.buffer)
4240 return 0;
4241
4242 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
4243 if (ret < 0)
4244 return ret;
4245
4246#ifdef CONFIG_TRACER_MAX_TRACE
4247 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
4248 !tr->current_trace->use_max_tr)
4249 goto out;
4250
4251 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
4252 if (ret < 0) {
4253 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
4254 &tr->trace_buffer, cpu);
4255 if (r < 0) {
4256 /*
4257 * AARGH! We are left with different
4258 * size max buffer!!!!
4259 * The max buffer is our "snapshot" buffer.
4260 * When a tracer needs a snapshot (one of the
4261 * latency tracers), it swaps the max buffer
4262 * with the saved snap shot. We succeeded to
4263 * update the size of the main buffer, but failed to
4264 * update the size of the max buffer. But when we tried
4265 * to reset the main buffer to the original size, we
4266 * failed there too. This is very unlikely to
4267 * happen, but if it does, warn and kill all
4268 * tracing.
4269 */
4270 WARN_ON(1);
4271 tracing_disabled = 1;
4272 }
4273 return ret;
4274 }
4275
4276 if (cpu == RING_BUFFER_ALL_CPUS)
4277 set_buffer_entries(&tr->max_buffer, size);
4278 else
4279 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
4280
4281 out:
4282#endif /* CONFIG_TRACER_MAX_TRACE */
4283
4284 if (cpu == RING_BUFFER_ALL_CPUS)
4285 set_buffer_entries(&tr->trace_buffer, size);
4286 else
4287 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
4288
4289 return ret;
4290}
4291
4292static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
4293 unsigned long size, int cpu_id)
4294{
4295 int ret = size;
4296
4297 mutex_lock(&trace_types_lock);
4298
4299 if (cpu_id != RING_BUFFER_ALL_CPUS) {
4300 /* make sure, this cpu is enabled in the mask */
4301 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
4302 ret = -EINVAL;
4303 goto out;
4304 }
4305 }
4306
4307 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
4308 if (ret < 0)
4309 ret = -ENOMEM;
4310
4311out:
4312 mutex_unlock(&trace_types_lock);
4313
4314 return ret;
4315}
4316
4317
4318/**
4319 * tracing_update_buffers - used by tracing facility to expand ring buffers
4320 *
4321 * To save on memory when the tracing is never used on a system with it
4322 * configured in. The ring buffers are set to a minimum size. But once
4323 * a user starts to use the tracing facility, then they need to grow
4324 * to their default size.
4325 *
4326 * This function is to be called when a tracer is about to be used.
4327 */
4328int tracing_update_buffers(void)
4329{
4330 int ret = 0;
4331
4332 mutex_lock(&trace_types_lock);
4333 if (!ring_buffer_expanded)
4334 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
4335 RING_BUFFER_ALL_CPUS);
4336 mutex_unlock(&trace_types_lock);
4337
4338 return ret;
4339}
4340
4341struct trace_option_dentry;
4342
4343static void
4344create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
4345
4346/*
4347 * Used to clear out the tracer before deletion of an instance.
4348 * Must have trace_types_lock held.
4349 */
4350static void tracing_set_nop(struct trace_array *tr)
4351{
4352 if (tr->current_trace == &nop_trace)
4353 return;
4354
4355 tr->current_trace->enabled--;
4356
4357 if (tr->current_trace->reset)
4358 tr->current_trace->reset(tr);
4359
4360 tr->current_trace = &nop_trace;
4361}
4362
4363static void add_tracer_options(struct trace_array *tr, struct tracer *t)
4364{
4365 /* Only enable if the directory has been created already. */
4366 if (!tr->dir)
4367 return;
4368
4369 create_trace_option_files(tr, t);
4370}
4371
4372static int tracing_set_tracer(struct trace_array *tr, const char *buf)
4373{
4374 struct tracer *t;
4375#ifdef CONFIG_TRACER_MAX_TRACE
4376 bool had_max_tr;
4377#endif
4378 int ret = 0;
4379
4380 mutex_lock(&trace_types_lock);
4381
4382 if (!ring_buffer_expanded) {
4383 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
4384 RING_BUFFER_ALL_CPUS);
4385 if (ret < 0)
4386 goto out;
4387 ret = 0;
4388 }
4389
4390 for (t = trace_types; t; t = t->next) {
4391 if (strcmp(t->name, buf) == 0)
4392 break;
4393 }
4394 if (!t) {
4395 ret = -EINVAL;
4396 goto out;
4397 }
4398 if (t == tr->current_trace)
4399 goto out;
4400
4401 /* Some tracers are only allowed for the top level buffer */
4402 if (!trace_ok_for_array(t, tr)) {
4403 ret = -EINVAL;
4404 goto out;
4405 }
4406
4407 /* If trace pipe files are being read, we can't change the tracer */
4408 if (tr->current_trace->ref) {
4409 ret = -EBUSY;
4410 goto out;
4411 }
4412
4413 trace_branch_disable();
4414
4415 tr->current_trace->enabled--;
4416
4417 if (tr->current_trace->reset)
4418 tr->current_trace->reset(tr);
4419
4420 /* Current trace needs to be nop_trace before synchronize_sched */
4421 tr->current_trace = &nop_trace;
4422
4423#ifdef CONFIG_TRACER_MAX_TRACE
4424 had_max_tr = tr->allocated_snapshot;
4425
4426 if (had_max_tr && !t->use_max_tr) {
4427 /*
4428 * We need to make sure that the update_max_tr sees that
4429 * current_trace changed to nop_trace to keep it from
4430 * swapping the buffers after we resize it.
4431 * The update_max_tr is called from interrupts disabled
4432 * so a synchronized_sched() is sufficient.
4433 */
4434 synchronize_sched();
4435 free_snapshot(tr);
4436 }
4437#endif
4438
4439#ifdef CONFIG_TRACER_MAX_TRACE
4440 if (t->use_max_tr && !had_max_tr) {
4441 ret = alloc_snapshot(tr);
4442 if (ret < 0)
4443 goto out;
4444 }
4445#endif
4446
4447 if (t->init) {
4448 ret = tracer_init(t, tr);
4449 if (ret)
4450 goto out;
4451 }
4452
4453 tr->current_trace = t;
4454 tr->current_trace->enabled++;
4455 trace_branch_enable(tr);
4456 out:
4457 mutex_unlock(&trace_types_lock);
4458
4459 return ret;
4460}
4461
4462static ssize_t
4463tracing_set_trace_write(struct file *filp, const char __user *ubuf,
4464 size_t cnt, loff_t *ppos)
4465{
4466 struct trace_array *tr = filp->private_data;
4467 char buf[MAX_TRACER_SIZE+1];
4468 int i;
4469 size_t ret;
4470 int err;
4471
4472 ret = cnt;
4473
4474 if (cnt > MAX_TRACER_SIZE)
4475 cnt = MAX_TRACER_SIZE;
4476
4477 if (copy_from_user(&buf, ubuf, cnt))
4478 return -EFAULT;
4479
4480 buf[cnt] = 0;
4481
4482 /* strip ending whitespace. */
4483 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
4484 buf[i] = 0;
4485
4486 err = tracing_set_tracer(tr, buf);
4487 if (err)
4488 return err;
4489
4490 *ppos += ret;
4491
4492 return ret;
4493}
4494
4495static ssize_t
4496tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
4497 size_t cnt, loff_t *ppos)
4498{
4499 char buf[64];
4500 int r;
4501
4502 r = snprintf(buf, sizeof(buf), "%ld\n",
4503 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
4504 if (r > sizeof(buf))
4505 r = sizeof(buf);
4506 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4507}
4508
4509static ssize_t
4510tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
4511 size_t cnt, loff_t *ppos)
4512{
4513 unsigned long val;
4514 int ret;
4515
4516 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4517 if (ret)
4518 return ret;
4519
4520 *ptr = val * 1000;
4521
4522 return cnt;
4523}
4524
4525static ssize_t
4526tracing_thresh_read(struct file *filp, char __user *ubuf,
4527 size_t cnt, loff_t *ppos)
4528{
4529 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
4530}
4531
4532static ssize_t
4533tracing_thresh_write(struct file *filp, const char __user *ubuf,
4534 size_t cnt, loff_t *ppos)
4535{
4536 struct trace_array *tr = filp->private_data;
4537 int ret;
4538
4539 mutex_lock(&trace_types_lock);
4540 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
4541 if (ret < 0)
4542 goto out;
4543
4544 if (tr->current_trace->update_thresh) {
4545 ret = tr->current_trace->update_thresh(tr);
4546 if (ret < 0)
4547 goto out;
4548 }
4549
4550 ret = cnt;
4551out:
4552 mutex_unlock(&trace_types_lock);
4553
4554 return ret;
4555}
4556
4557#ifdef CONFIG_TRACER_MAX_TRACE
4558
4559static ssize_t
4560tracing_max_lat_read(struct file *filp, char __user *ubuf,
4561 size_t cnt, loff_t *ppos)
4562{
4563 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
4564}
4565
4566static ssize_t
4567tracing_max_lat_write(struct file *filp, const char __user *ubuf,
4568 size_t cnt, loff_t *ppos)
4569{
4570 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
4571}
4572
4573#endif
4574
4575static int tracing_open_pipe(struct inode *inode, struct file *filp)
4576{
4577 struct trace_array *tr = inode->i_private;
4578 struct trace_iterator *iter;
4579 int ret = 0;
4580
4581 if (tracing_disabled)
4582 return -ENODEV;
4583
4584 if (trace_array_get(tr) < 0)
4585 return -ENODEV;
4586
4587 mutex_lock(&trace_types_lock);
4588
4589 /* create a buffer to store the information to pass to userspace */
4590 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4591 if (!iter) {
4592 ret = -ENOMEM;
4593 __trace_array_put(tr);
4594 goto out;
4595 }
4596
4597 trace_seq_init(&iter->seq);
4598 iter->trace = tr->current_trace;
4599
4600 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
4601 ret = -ENOMEM;
4602 goto fail;
4603 }
4604
4605 /* trace pipe does not show start of buffer */
4606 cpumask_setall(iter->started);
4607
4608 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4609 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4610
4611 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4612 if (trace_clocks[tr->clock_id].in_ns)
4613 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4614
4615 iter->tr = tr;
4616 iter->trace_buffer = &tr->trace_buffer;
4617 iter->cpu_file = tracing_get_cpu(inode);
4618 mutex_init(&iter->mutex);
4619 filp->private_data = iter;
4620
4621 if (iter->trace->pipe_open)
4622 iter->trace->pipe_open(iter);
4623
4624 nonseekable_open(inode, filp);
4625
4626 tr->current_trace->ref++;
4627out:
4628 mutex_unlock(&trace_types_lock);
4629 return ret;
4630
4631fail:
4632 kfree(iter->trace);
4633 kfree(iter);
4634 __trace_array_put(tr);
4635 mutex_unlock(&trace_types_lock);
4636 return ret;
4637}
4638
4639static int tracing_release_pipe(struct inode *inode, struct file *file)
4640{
4641 struct trace_iterator *iter = file->private_data;
4642 struct trace_array *tr = inode->i_private;
4643
4644 mutex_lock(&trace_types_lock);
4645
4646 tr->current_trace->ref--;
4647
4648 if (iter->trace->pipe_close)
4649 iter->trace->pipe_close(iter);
4650
4651 mutex_unlock(&trace_types_lock);
4652
4653 free_cpumask_var(iter->started);
4654 mutex_destroy(&iter->mutex);
4655 kfree(iter);
4656
4657 trace_array_put(tr);
4658
4659 return 0;
4660}
4661
4662static unsigned int
4663trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4664{
4665 struct trace_array *tr = iter->tr;
4666
4667 /* Iterators are static, they should be filled or empty */
4668 if (trace_buffer_iter(iter, iter->cpu_file))
4669 return POLLIN | POLLRDNORM;
4670
4671 if (tr->trace_flags & TRACE_ITER_BLOCK)
4672 /*
4673 * Always select as readable when in blocking mode
4674 */
4675 return POLLIN | POLLRDNORM;
4676 else
4677 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4678 filp, poll_table);
4679}
4680
4681static unsigned int
4682tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4683{
4684 struct trace_iterator *iter = filp->private_data;
4685
4686 return trace_poll(iter, filp, poll_table);
4687}
4688
4689/* Must be called with iter->mutex held. */
4690static int tracing_wait_pipe(struct file *filp)
4691{
4692 struct trace_iterator *iter = filp->private_data;
4693 int ret;
4694
4695 while (trace_empty(iter)) {
4696
4697 if ((filp->f_flags & O_NONBLOCK)) {
4698 return -EAGAIN;
4699 }
4700
4701 /*
4702 * We block until we read something and tracing is disabled.
4703 * We still block if tracing is disabled, but we have never
4704 * read anything. This allows a user to cat this file, and
4705 * then enable tracing. But after we have read something,
4706 * we give an EOF when tracing is again disabled.
4707 *
4708 * iter->pos will be 0 if we haven't read anything.
4709 */
4710 if (!tracing_is_on() && iter->pos)
4711 break;
4712
4713 mutex_unlock(&iter->mutex);
4714
4715 ret = wait_on_pipe(iter, false);
4716
4717 mutex_lock(&iter->mutex);
4718
4719 if (ret)
4720 return ret;
4721 }
4722
4723 return 1;
4724}
4725
4726/*
4727 * Consumer reader.
4728 */
4729static ssize_t
4730tracing_read_pipe(struct file *filp, char __user *ubuf,
4731 size_t cnt, loff_t *ppos)
4732{
4733 struct trace_iterator *iter = filp->private_data;
4734 ssize_t sret;
4735
4736 /* return any leftover data */
4737 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4738 if (sret != -EBUSY)
4739 return sret;
4740
4741 trace_seq_init(&iter->seq);
4742
4743 /*
4744 * Avoid more than one consumer on a single file descriptor
4745 * This is just a matter of traces coherency, the ring buffer itself
4746 * is protected.
4747 */
4748 mutex_lock(&iter->mutex);
4749 if (iter->trace->read) {
4750 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4751 if (sret)
4752 goto out;
4753 }
4754
4755waitagain:
4756 sret = tracing_wait_pipe(filp);
4757 if (sret <= 0)
4758 goto out;
4759
4760 /* stop when tracing is finished */
4761 if (trace_empty(iter)) {
4762 sret = 0;
4763 goto out;
4764 }
4765
4766 if (cnt >= PAGE_SIZE)
4767 cnt = PAGE_SIZE - 1;
4768
4769 /* reset all but tr, trace, and overruns */
4770 memset(&iter->seq, 0,
4771 sizeof(struct trace_iterator) -
4772 offsetof(struct trace_iterator, seq));
4773 cpumask_clear(iter->started);
4774 iter->pos = -1;
4775
4776 trace_event_read_lock();
4777 trace_access_lock(iter->cpu_file);
4778 while (trace_find_next_entry_inc(iter) != NULL) {
4779 enum print_line_t ret;
4780 int save_len = iter->seq.seq.len;
4781
4782 ret = print_trace_line(iter);
4783 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4784 /* don't print partial lines */
4785 iter->seq.seq.len = save_len;
4786 break;
4787 }
4788 if (ret != TRACE_TYPE_NO_CONSUME)
4789 trace_consume(iter);
4790
4791 if (trace_seq_used(&iter->seq) >= cnt)
4792 break;
4793
4794 /*
4795 * Setting the full flag means we reached the trace_seq buffer
4796 * size and we should leave by partial output condition above.
4797 * One of the trace_seq_* functions is not used properly.
4798 */
4799 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4800 iter->ent->type);
4801 }
4802 trace_access_unlock(iter->cpu_file);
4803 trace_event_read_unlock();
4804
4805 /* Now copy what we have to the user */
4806 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4807 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
4808 trace_seq_init(&iter->seq);
4809
4810 /*
4811 * If there was nothing to send to user, in spite of consuming trace
4812 * entries, go back to wait for more entries.
4813 */
4814 if (sret == -EBUSY)
4815 goto waitagain;
4816
4817out:
4818 mutex_unlock(&iter->mutex);
4819
4820 return sret;
4821}
4822
4823static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4824 unsigned int idx)
4825{
4826 __free_page(spd->pages[idx]);
4827}
4828
4829static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4830 .can_merge = 0,
4831 .confirm = generic_pipe_buf_confirm,
4832 .release = generic_pipe_buf_release,
4833 .steal = generic_pipe_buf_steal,
4834 .get = generic_pipe_buf_get,
4835};
4836
4837static size_t
4838tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4839{
4840 size_t count;
4841 int save_len;
4842 int ret;
4843
4844 /* Seq buffer is page-sized, exactly what we need. */
4845 for (;;) {
4846 save_len = iter->seq.seq.len;
4847 ret = print_trace_line(iter);
4848
4849 if (trace_seq_has_overflowed(&iter->seq)) {
4850 iter->seq.seq.len = save_len;
4851 break;
4852 }
4853
4854 /*
4855 * This should not be hit, because it should only
4856 * be set if the iter->seq overflowed. But check it
4857 * anyway to be safe.
4858 */
4859 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4860 iter->seq.seq.len = save_len;
4861 break;
4862 }
4863
4864 count = trace_seq_used(&iter->seq) - save_len;
4865 if (rem < count) {
4866 rem = 0;
4867 iter->seq.seq.len = save_len;
4868 break;
4869 }
4870
4871 if (ret != TRACE_TYPE_NO_CONSUME)
4872 trace_consume(iter);
4873 rem -= count;
4874 if (!trace_find_next_entry_inc(iter)) {
4875 rem = 0;
4876 iter->ent = NULL;
4877 break;
4878 }
4879 }
4880
4881 return rem;
4882}
4883
4884static ssize_t tracing_splice_read_pipe(struct file *filp,
4885 loff_t *ppos,
4886 struct pipe_inode_info *pipe,
4887 size_t len,
4888 unsigned int flags)
4889{
4890 struct page *pages_def[PIPE_DEF_BUFFERS];
4891 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4892 struct trace_iterator *iter = filp->private_data;
4893 struct splice_pipe_desc spd = {
4894 .pages = pages_def,
4895 .partial = partial_def,
4896 .nr_pages = 0, /* This gets updated below. */
4897 .nr_pages_max = PIPE_DEF_BUFFERS,
4898 .flags = flags,
4899 .ops = &tracing_pipe_buf_ops,
4900 .spd_release = tracing_spd_release_pipe,
4901 };
4902 ssize_t ret;
4903 size_t rem;
4904 unsigned int i;
4905
4906 if (splice_grow_spd(pipe, &spd))
4907 return -ENOMEM;
4908
4909 mutex_lock(&iter->mutex);
4910
4911 if (iter->trace->splice_read) {
4912 ret = iter->trace->splice_read(iter, filp,
4913 ppos, pipe, len, flags);
4914 if (ret)
4915 goto out_err;
4916 }
4917
4918 ret = tracing_wait_pipe(filp);
4919 if (ret <= 0)
4920 goto out_err;
4921
4922 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4923 ret = -EFAULT;
4924 goto out_err;
4925 }
4926
4927 trace_event_read_lock();
4928 trace_access_lock(iter->cpu_file);
4929
4930 /* Fill as many pages as possible. */
4931 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
4932 spd.pages[i] = alloc_page(GFP_KERNEL);
4933 if (!spd.pages[i])
4934 break;
4935
4936 rem = tracing_fill_pipe_page(rem, iter);
4937
4938 /* Copy the data into the page, so we can start over. */
4939 ret = trace_seq_to_buffer(&iter->seq,
4940 page_address(spd.pages[i]),
4941 trace_seq_used(&iter->seq));
4942 if (ret < 0) {
4943 __free_page(spd.pages[i]);
4944 break;
4945 }
4946 spd.partial[i].offset = 0;
4947 spd.partial[i].len = trace_seq_used(&iter->seq);
4948
4949 trace_seq_init(&iter->seq);
4950 }
4951
4952 trace_access_unlock(iter->cpu_file);
4953 trace_event_read_unlock();
4954 mutex_unlock(&iter->mutex);
4955
4956 spd.nr_pages = i;
4957
4958 if (i)
4959 ret = splice_to_pipe(pipe, &spd);
4960 else
4961 ret = 0;
4962out:
4963 splice_shrink_spd(&spd);
4964 return ret;
4965
4966out_err:
4967 mutex_unlock(&iter->mutex);
4968 goto out;
4969}
4970
4971static ssize_t
4972tracing_entries_read(struct file *filp, char __user *ubuf,
4973 size_t cnt, loff_t *ppos)
4974{
4975 struct inode *inode = file_inode(filp);
4976 struct trace_array *tr = inode->i_private;
4977 int cpu = tracing_get_cpu(inode);
4978 char buf[64];
4979 int r = 0;
4980 ssize_t ret;
4981
4982 mutex_lock(&trace_types_lock);
4983
4984 if (cpu == RING_BUFFER_ALL_CPUS) {
4985 int cpu, buf_size_same;
4986 unsigned long size;
4987
4988 size = 0;
4989 buf_size_same = 1;
4990 /* check if all cpu sizes are same */
4991 for_each_tracing_cpu(cpu) {
4992 /* fill in the size from first enabled cpu */
4993 if (size == 0)
4994 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4995 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4996 buf_size_same = 0;
4997 break;
4998 }
4999 }
5000
5001 if (buf_size_same) {
5002 if (!ring_buffer_expanded)
5003 r = sprintf(buf, "%lu (expanded: %lu)\n",
5004 size >> 10,
5005 trace_buf_size >> 10);
5006 else
5007 r = sprintf(buf, "%lu\n", size >> 10);
5008 } else
5009 r = sprintf(buf, "X\n");
5010 } else
5011 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
5012
5013 mutex_unlock(&trace_types_lock);
5014
5015 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5016 return ret;
5017}
5018
5019static ssize_t
5020tracing_entries_write(struct file *filp, const char __user *ubuf,
5021 size_t cnt, loff_t *ppos)
5022{
5023 struct inode *inode = file_inode(filp);
5024 struct trace_array *tr = inode->i_private;
5025 unsigned long val;
5026 int ret;
5027
5028 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5029 if (ret)
5030 return ret;
5031
5032 /* must have at least 1 entry */
5033 if (!val)
5034 return -EINVAL;
5035
5036 /* value is in KB */
5037 val <<= 10;
5038 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
5039 if (ret < 0)
5040 return ret;
5041
5042 *ppos += cnt;
5043
5044 return cnt;
5045}
5046
5047static ssize_t
5048tracing_total_entries_read(struct file *filp, char __user *ubuf,
5049 size_t cnt, loff_t *ppos)
5050{
5051 struct trace_array *tr = filp->private_data;
5052 char buf[64];
5053 int r, cpu;
5054 unsigned long size = 0, expanded_size = 0;
5055
5056 mutex_lock(&trace_types_lock);
5057 for_each_tracing_cpu(cpu) {
5058 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
5059 if (!ring_buffer_expanded)
5060 expanded_size += trace_buf_size >> 10;
5061 }
5062 if (ring_buffer_expanded)
5063 r = sprintf(buf, "%lu\n", size);
5064 else
5065 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
5066 mutex_unlock(&trace_types_lock);
5067
5068 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5069}
5070
5071static ssize_t
5072tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
5073 size_t cnt, loff_t *ppos)
5074{
5075 /*
5076 * There is no need to read what the user has written, this function
5077 * is just to make sure that there is no error when "echo" is used
5078 */
5079
5080 *ppos += cnt;
5081
5082 return cnt;
5083}
5084
5085static int
5086tracing_free_buffer_release(struct inode *inode, struct file *filp)
5087{
5088 struct trace_array *tr = inode->i_private;
5089
5090 /* disable tracing ? */
5091 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
5092 tracer_tracing_off(tr);
5093 /* resize the ring buffer to 0 */
5094 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
5095
5096 trace_array_put(tr);
5097
5098 return 0;
5099}
5100
5101static ssize_t
5102tracing_mark_write(struct file *filp, const char __user *ubuf,
5103 size_t cnt, loff_t *fpos)
5104{
5105 unsigned long addr = (unsigned long)ubuf;
5106 struct trace_array *tr = filp->private_data;
5107 struct ring_buffer_event *event;
5108 struct ring_buffer *buffer;
5109 struct print_entry *entry;
5110 unsigned long irq_flags;
5111 struct page *pages[2];
5112 void *map_page[2];
5113 int nr_pages = 1;
5114 ssize_t written;
5115 int offset;
5116 int size;
5117 int len;
5118 int ret;
5119 int i;
5120
5121 if (tracing_disabled)
5122 return -EINVAL;
5123
5124 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
5125 return -EINVAL;
5126
5127 if (cnt > TRACE_BUF_SIZE)
5128 cnt = TRACE_BUF_SIZE;
5129
5130 /*
5131 * Userspace is injecting traces into the kernel trace buffer.
5132 * We want to be as non intrusive as possible.
5133 * To do so, we do not want to allocate any special buffers
5134 * or take any locks, but instead write the userspace data
5135 * straight into the ring buffer.
5136 *
5137 * First we need to pin the userspace buffer into memory,
5138 * which, most likely it is, because it just referenced it.
5139 * But there's no guarantee that it is. By using get_user_pages_fast()
5140 * and kmap_atomic/kunmap_atomic() we can get access to the
5141 * pages directly. We then write the data directly into the
5142 * ring buffer.
5143 */
5144 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
5145
5146 /* check if we cross pages */
5147 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
5148 nr_pages = 2;
5149
5150 offset = addr & (PAGE_SIZE - 1);
5151 addr &= PAGE_MASK;
5152
5153 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
5154 if (ret < nr_pages) {
5155 while (--ret >= 0)
5156 put_page(pages[ret]);
5157 written = -EFAULT;
5158 goto out;
5159 }
5160
5161 for (i = 0; i < nr_pages; i++)
5162 map_page[i] = kmap_atomic(pages[i]);
5163
5164 local_save_flags(irq_flags);
5165 size = sizeof(*entry) + cnt + 2; /* possible \n added */
5166 buffer = tr->trace_buffer.buffer;
5167 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
5168 irq_flags, preempt_count());
5169 if (!event) {
5170 /* Ring buffer disabled, return as if not open for write */
5171 written = -EBADF;
5172 goto out_unlock;
5173 }
5174
5175 entry = ring_buffer_event_data(event);
5176 entry->ip = _THIS_IP_;
5177
5178 if (nr_pages == 2) {
5179 len = PAGE_SIZE - offset;
5180 memcpy(&entry->buf, map_page[0] + offset, len);
5181 memcpy(&entry->buf[len], map_page[1], cnt - len);
5182 } else
5183 memcpy(&entry->buf, map_page[0] + offset, cnt);
5184
5185 if (entry->buf[cnt - 1] != '\n') {
5186 entry->buf[cnt] = '\n';
5187 entry->buf[cnt + 1] = '\0';
5188 } else
5189 entry->buf[cnt] = '\0';
5190
5191 __buffer_unlock_commit(buffer, event);
5192
5193 written = cnt;
5194
5195 *fpos += written;
5196
5197 out_unlock:
5198 for (i = nr_pages - 1; i >= 0; i--) {
5199 kunmap_atomic(map_page[i]);
5200 put_page(pages[i]);
5201 }
5202 out:
5203 return written;
5204}
5205
5206static int tracing_clock_show(struct seq_file *m, void *v)
5207{
5208 struct trace_array *tr = m->private;
5209 int i;
5210
5211 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
5212 seq_printf(m,
5213 "%s%s%s%s", i ? " " : "",
5214 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
5215 i == tr->clock_id ? "]" : "");
5216 seq_putc(m, '\n');
5217
5218 return 0;
5219}
5220
5221static int tracing_set_clock(struct trace_array *tr, const char *clockstr)
5222{
5223 int i;
5224
5225 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
5226 if (strcmp(trace_clocks[i].name, clockstr) == 0)
5227 break;
5228 }
5229 if (i == ARRAY_SIZE(trace_clocks))
5230 return -EINVAL;
5231
5232 mutex_lock(&trace_types_lock);
5233
5234 tr->clock_id = i;
5235
5236 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
5237
5238 /*
5239 * New clock may not be consistent with the previous clock.
5240 * Reset the buffer so that it doesn't have incomparable timestamps.
5241 */
5242 tracing_reset_online_cpus(&tr->trace_buffer);
5243
5244#ifdef CONFIG_TRACER_MAX_TRACE
5245 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
5246 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
5247 tracing_reset_online_cpus(&tr->max_buffer);
5248#endif
5249
5250 mutex_unlock(&trace_types_lock);
5251
5252 return 0;
5253}
5254
5255static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
5256 size_t cnt, loff_t *fpos)
5257{
5258 struct seq_file *m = filp->private_data;
5259 struct trace_array *tr = m->private;
5260 char buf[64];
5261 const char *clockstr;
5262 int ret;
5263
5264 if (cnt >= sizeof(buf))
5265 return -EINVAL;
5266
5267 if (copy_from_user(&buf, ubuf, cnt))
5268 return -EFAULT;
5269
5270 buf[cnt] = 0;
5271
5272 clockstr = strstrip(buf);
5273
5274 ret = tracing_set_clock(tr, clockstr);
5275 if (ret)
5276 return ret;
5277
5278 *fpos += cnt;
5279
5280 return cnt;
5281}
5282
5283static int tracing_clock_open(struct inode *inode, struct file *file)
5284{
5285 struct trace_array *tr = inode->i_private;
5286 int ret;
5287
5288 if (tracing_disabled)
5289 return -ENODEV;
5290
5291 if (trace_array_get(tr))
5292 return -ENODEV;
5293
5294 ret = single_open(file, tracing_clock_show, inode->i_private);
5295 if (ret < 0)
5296 trace_array_put(tr);
5297
5298 return ret;
5299}
5300
5301struct ftrace_buffer_info {
5302 struct trace_iterator iter;
5303 void *spare;
5304 unsigned int read;
5305};
5306
5307#ifdef CONFIG_TRACER_SNAPSHOT
5308static int tracing_snapshot_open(struct inode *inode, struct file *file)
5309{
5310 struct trace_array *tr = inode->i_private;
5311 struct trace_iterator *iter;
5312 struct seq_file *m;
5313 int ret = 0;
5314
5315 if (trace_array_get(tr) < 0)
5316 return -ENODEV;
5317
5318 if (file->f_mode & FMODE_READ) {
5319 iter = __tracing_open(inode, file, true);
5320 if (IS_ERR(iter))
5321 ret = PTR_ERR(iter);
5322 } else {
5323 /* Writes still need the seq_file to hold the private data */
5324 ret = -ENOMEM;
5325 m = kzalloc(sizeof(*m), GFP_KERNEL);
5326 if (!m)
5327 goto out;
5328 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
5329 if (!iter) {
5330 kfree(m);
5331 goto out;
5332 }
5333 ret = 0;
5334
5335 iter->tr = tr;
5336 iter->trace_buffer = &tr->max_buffer;
5337 iter->cpu_file = tracing_get_cpu(inode);
5338 m->private = iter;
5339 file->private_data = m;
5340 }
5341out:
5342 if (ret < 0)
5343 trace_array_put(tr);
5344
5345 return ret;
5346}
5347
5348static ssize_t
5349tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
5350 loff_t *ppos)
5351{
5352 struct seq_file *m = filp->private_data;
5353 struct trace_iterator *iter = m->private;
5354 struct trace_array *tr = iter->tr;
5355 unsigned long val;
5356 int ret;
5357
5358 ret = tracing_update_buffers();
5359 if (ret < 0)
5360 return ret;
5361
5362 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5363 if (ret)
5364 return ret;
5365
5366 mutex_lock(&trace_types_lock);
5367
5368 if (tr->current_trace->use_max_tr) {
5369 ret = -EBUSY;
5370 goto out;
5371 }
5372
5373 switch (val) {
5374 case 0:
5375 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
5376 ret = -EINVAL;
5377 break;
5378 }
5379 if (tr->allocated_snapshot)
5380 free_snapshot(tr);
5381 break;
5382 case 1:
5383/* Only allow per-cpu swap if the ring buffer supports it */
5384#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
5385 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
5386 ret = -EINVAL;
5387 break;
5388 }
5389#endif
5390 if (!tr->allocated_snapshot) {
5391 ret = alloc_snapshot(tr);
5392 if (ret < 0)
5393 break;
5394 }
5395 local_irq_disable();
5396 /* Now, we're going to swap */
5397 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
5398 update_max_tr(tr, current, smp_processor_id());
5399 else
5400 update_max_tr_single(tr, current, iter->cpu_file);
5401 local_irq_enable();
5402 break;
5403 default:
5404 if (tr->allocated_snapshot) {
5405 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
5406 tracing_reset_online_cpus(&tr->max_buffer);
5407 else
5408 tracing_reset(&tr->max_buffer, iter->cpu_file);
5409 }
5410 break;
5411 }
5412
5413 if (ret >= 0) {
5414 *ppos += cnt;
5415 ret = cnt;
5416 }
5417out:
5418 mutex_unlock(&trace_types_lock);
5419 return ret;
5420}
5421
5422static int tracing_snapshot_release(struct inode *inode, struct file *file)
5423{
5424 struct seq_file *m = file->private_data;
5425 int ret;
5426
5427 ret = tracing_release(inode, file);
5428
5429 if (file->f_mode & FMODE_READ)
5430 return ret;
5431
5432 /* If write only, the seq_file is just a stub */
5433 if (m)
5434 kfree(m->private);
5435 kfree(m);
5436
5437 return 0;
5438}
5439
5440static int tracing_buffers_open(struct inode *inode, struct file *filp);
5441static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
5442 size_t count, loff_t *ppos);
5443static int tracing_buffers_release(struct inode *inode, struct file *file);
5444static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5445 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
5446
5447static int snapshot_raw_open(struct inode *inode, struct file *filp)
5448{
5449 struct ftrace_buffer_info *info;
5450 int ret;
5451
5452 ret = tracing_buffers_open(inode, filp);
5453 if (ret < 0)
5454 return ret;
5455
5456 info = filp->private_data;
5457
5458 if (info->iter.trace->use_max_tr) {
5459 tracing_buffers_release(inode, filp);
5460 return -EBUSY;
5461 }
5462
5463 info->iter.snapshot = true;
5464 info->iter.trace_buffer = &info->iter.tr->max_buffer;
5465
5466 return ret;
5467}
5468
5469#endif /* CONFIG_TRACER_SNAPSHOT */
5470
5471
5472static const struct file_operations tracing_thresh_fops = {
5473 .open = tracing_open_generic,
5474 .read = tracing_thresh_read,
5475 .write = tracing_thresh_write,
5476 .llseek = generic_file_llseek,
5477};
5478
5479#ifdef CONFIG_TRACER_MAX_TRACE
5480static const struct file_operations tracing_max_lat_fops = {
5481 .open = tracing_open_generic,
5482 .read = tracing_max_lat_read,
5483 .write = tracing_max_lat_write,
5484 .llseek = generic_file_llseek,
5485};
5486#endif
5487
5488static const struct file_operations set_tracer_fops = {
5489 .open = tracing_open_generic,
5490 .read = tracing_set_trace_read,
5491 .write = tracing_set_trace_write,
5492 .llseek = generic_file_llseek,
5493};
5494
5495static const struct file_operations tracing_pipe_fops = {
5496 .open = tracing_open_pipe,
5497 .poll = tracing_poll_pipe,
5498 .read = tracing_read_pipe,
5499 .splice_read = tracing_splice_read_pipe,
5500 .release = tracing_release_pipe,
5501 .llseek = no_llseek,
5502};
5503
5504static const struct file_operations tracing_entries_fops = {
5505 .open = tracing_open_generic_tr,
5506 .read = tracing_entries_read,
5507 .write = tracing_entries_write,
5508 .llseek = generic_file_llseek,
5509 .release = tracing_release_generic_tr,
5510};
5511
5512static const struct file_operations tracing_total_entries_fops = {
5513 .open = tracing_open_generic_tr,
5514 .read = tracing_total_entries_read,
5515 .llseek = generic_file_llseek,
5516 .release = tracing_release_generic_tr,
5517};
5518
5519static const struct file_operations tracing_free_buffer_fops = {
5520 .open = tracing_open_generic_tr,
5521 .write = tracing_free_buffer_write,
5522 .release = tracing_free_buffer_release,
5523};
5524
5525static const struct file_operations tracing_mark_fops = {
5526 .open = tracing_open_generic_tr,
5527 .write = tracing_mark_write,
5528 .llseek = generic_file_llseek,
5529 .release = tracing_release_generic_tr,
5530};
5531
5532static const struct file_operations trace_clock_fops = {
5533 .open = tracing_clock_open,
5534 .read = seq_read,
5535 .llseek = seq_lseek,
5536 .release = tracing_single_release_tr,
5537 .write = tracing_clock_write,
5538};
5539
5540#ifdef CONFIG_TRACER_SNAPSHOT
5541static const struct file_operations snapshot_fops = {
5542 .open = tracing_snapshot_open,
5543 .read = seq_read,
5544 .write = tracing_snapshot_write,
5545 .llseek = tracing_lseek,
5546 .release = tracing_snapshot_release,
5547};
5548
5549static const struct file_operations snapshot_raw_fops = {
5550 .open = snapshot_raw_open,
5551 .read = tracing_buffers_read,
5552 .release = tracing_buffers_release,
5553 .splice_read = tracing_buffers_splice_read,
5554 .llseek = no_llseek,
5555};
5556
5557#endif /* CONFIG_TRACER_SNAPSHOT */
5558
5559static int tracing_buffers_open(struct inode *inode, struct file *filp)
5560{
5561 struct trace_array *tr = inode->i_private;
5562 struct ftrace_buffer_info *info;
5563 int ret;
5564
5565 if (tracing_disabled)
5566 return -ENODEV;
5567
5568 if (trace_array_get(tr) < 0)
5569 return -ENODEV;
5570
5571 info = kzalloc(sizeof(*info), GFP_KERNEL);
5572 if (!info) {
5573 trace_array_put(tr);
5574 return -ENOMEM;
5575 }
5576
5577 mutex_lock(&trace_types_lock);
5578
5579 info->iter.tr = tr;
5580 info->iter.cpu_file = tracing_get_cpu(inode);
5581 info->iter.trace = tr->current_trace;
5582 info->iter.trace_buffer = &tr->trace_buffer;
5583 info->spare = NULL;
5584 /* Force reading ring buffer for first read */
5585 info->read = (unsigned int)-1;
5586
5587 filp->private_data = info;
5588
5589 tr->current_trace->ref++;
5590
5591 mutex_unlock(&trace_types_lock);
5592
5593 ret = nonseekable_open(inode, filp);
5594 if (ret < 0)
5595 trace_array_put(tr);
5596
5597 return ret;
5598}
5599
5600static unsigned int
5601tracing_buffers_poll(struct file *filp, poll_table *poll_table)
5602{
5603 struct ftrace_buffer_info *info = filp->private_data;
5604 struct trace_iterator *iter = &info->iter;
5605
5606 return trace_poll(iter, filp, poll_table);
5607}
5608
5609static ssize_t
5610tracing_buffers_read(struct file *filp, char __user *ubuf,
5611 size_t count, loff_t *ppos)
5612{
5613 struct ftrace_buffer_info *info = filp->private_data;
5614 struct trace_iterator *iter = &info->iter;
5615 ssize_t ret;
5616 ssize_t size;
5617
5618 if (!count)
5619 return 0;
5620
5621#ifdef CONFIG_TRACER_MAX_TRACE
5622 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
5623 return -EBUSY;
5624#endif
5625
5626 if (!info->spare)
5627 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
5628 iter->cpu_file);
5629 if (!info->spare)
5630 return -ENOMEM;
5631
5632 /* Do we have previous read data to read? */
5633 if (info->read < PAGE_SIZE)
5634 goto read;
5635
5636 again:
5637 trace_access_lock(iter->cpu_file);
5638 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5639 &info->spare,
5640 count,
5641 iter->cpu_file, 0);
5642 trace_access_unlock(iter->cpu_file);
5643
5644 if (ret < 0) {
5645 if (trace_empty(iter)) {
5646 if ((filp->f_flags & O_NONBLOCK))
5647 return -EAGAIN;
5648
5649 ret = wait_on_pipe(iter, false);
5650 if (ret)
5651 return ret;
5652
5653 goto again;
5654 }
5655 return 0;
5656 }
5657
5658 info->read = 0;
5659 read:
5660 size = PAGE_SIZE - info->read;
5661 if (size > count)
5662 size = count;
5663
5664 ret = copy_to_user(ubuf, info->spare + info->read, size);
5665 if (ret == size)
5666 return -EFAULT;
5667
5668 size -= ret;
5669
5670 *ppos += size;
5671 info->read += size;
5672
5673 return size;
5674}
5675
5676static int tracing_buffers_release(struct inode *inode, struct file *file)
5677{
5678 struct ftrace_buffer_info *info = file->private_data;
5679 struct trace_iterator *iter = &info->iter;
5680
5681 mutex_lock(&trace_types_lock);
5682
5683 iter->tr->current_trace->ref--;
5684
5685 __trace_array_put(iter->tr);
5686
5687 if (info->spare)
5688 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5689 kfree(info);
5690
5691 mutex_unlock(&trace_types_lock);
5692
5693 return 0;
5694}
5695
5696struct buffer_ref {
5697 struct ring_buffer *buffer;
5698 void *page;
5699 int ref;
5700};
5701
5702static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5703 struct pipe_buffer *buf)
5704{
5705 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5706
5707 if (--ref->ref)
5708 return;
5709
5710 ring_buffer_free_read_page(ref->buffer, ref->page);
5711 kfree(ref);
5712 buf->private = 0;
5713}
5714
5715static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5716 struct pipe_buffer *buf)
5717{
5718 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5719
5720 ref->ref++;
5721}
5722
5723/* Pipe buffer operations for a buffer. */
5724static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5725 .can_merge = 0,
5726 .confirm = generic_pipe_buf_confirm,
5727 .release = buffer_pipe_buf_release,
5728 .steal = generic_pipe_buf_steal,
5729 .get = buffer_pipe_buf_get,
5730};
5731
5732/*
5733 * Callback from splice_to_pipe(), if we need to release some pages
5734 * at the end of the spd in case we error'ed out in filling the pipe.
5735 */
5736static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5737{
5738 struct buffer_ref *ref =
5739 (struct buffer_ref *)spd->partial[i].private;
5740
5741 if (--ref->ref)
5742 return;
5743
5744 ring_buffer_free_read_page(ref->buffer, ref->page);
5745 kfree(ref);
5746 spd->partial[i].private = 0;
5747}
5748
5749static ssize_t
5750tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5751 struct pipe_inode_info *pipe, size_t len,
5752 unsigned int flags)
5753{
5754 struct ftrace_buffer_info *info = file->private_data;
5755 struct trace_iterator *iter = &info->iter;
5756 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5757 struct page *pages_def[PIPE_DEF_BUFFERS];
5758 struct splice_pipe_desc spd = {
5759 .pages = pages_def,
5760 .partial = partial_def,
5761 .nr_pages_max = PIPE_DEF_BUFFERS,
5762 .flags = flags,
5763 .ops = &buffer_pipe_buf_ops,
5764 .spd_release = buffer_spd_release,
5765 };
5766 struct buffer_ref *ref;
5767 int entries, size, i;
5768 ssize_t ret = 0;
5769
5770#ifdef CONFIG_TRACER_MAX_TRACE
5771 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
5772 return -EBUSY;
5773#endif
5774
5775 if (splice_grow_spd(pipe, &spd))
5776 return -ENOMEM;
5777
5778 if (*ppos & (PAGE_SIZE - 1))
5779 return -EINVAL;
5780
5781 if (len & (PAGE_SIZE - 1)) {
5782 if (len < PAGE_SIZE)
5783 return -EINVAL;
5784 len &= PAGE_MASK;
5785 }
5786
5787 again:
5788 trace_access_lock(iter->cpu_file);
5789 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5790
5791 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
5792 struct page *page;
5793 int r;
5794
5795 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5796 if (!ref) {
5797 ret = -ENOMEM;
5798 break;
5799 }
5800
5801 ref->ref = 1;
5802 ref->buffer = iter->trace_buffer->buffer;
5803 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5804 if (!ref->page) {
5805 ret = -ENOMEM;
5806 kfree(ref);
5807 break;
5808 }
5809
5810 r = ring_buffer_read_page(ref->buffer, &ref->page,
5811 len, iter->cpu_file, 1);
5812 if (r < 0) {
5813 ring_buffer_free_read_page(ref->buffer, ref->page);
5814 kfree(ref);
5815 break;
5816 }
5817
5818 /*
5819 * zero out any left over data, this is going to
5820 * user land.
5821 */
5822 size = ring_buffer_page_len(ref->page);
5823 if (size < PAGE_SIZE)
5824 memset(ref->page + size, 0, PAGE_SIZE - size);
5825
5826 page = virt_to_page(ref->page);
5827
5828 spd.pages[i] = page;
5829 spd.partial[i].len = PAGE_SIZE;
5830 spd.partial[i].offset = 0;
5831 spd.partial[i].private = (unsigned long)ref;
5832 spd.nr_pages++;
5833 *ppos += PAGE_SIZE;
5834
5835 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5836 }
5837
5838 trace_access_unlock(iter->cpu_file);
5839 spd.nr_pages = i;
5840
5841 /* did we read anything? */
5842 if (!spd.nr_pages) {
5843 if (ret)
5844 return ret;
5845
5846 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
5847 return -EAGAIN;
5848
5849 ret = wait_on_pipe(iter, true);
5850 if (ret)
5851 return ret;
5852
5853 goto again;
5854 }
5855
5856 ret = splice_to_pipe(pipe, &spd);
5857 splice_shrink_spd(&spd);
5858
5859 return ret;
5860}
5861
5862static const struct file_operations tracing_buffers_fops = {
5863 .open = tracing_buffers_open,
5864 .read = tracing_buffers_read,
5865 .poll = tracing_buffers_poll,
5866 .release = tracing_buffers_release,
5867 .splice_read = tracing_buffers_splice_read,
5868 .llseek = no_llseek,
5869};
5870
5871static ssize_t
5872tracing_stats_read(struct file *filp, char __user *ubuf,
5873 size_t count, loff_t *ppos)
5874{
5875 struct inode *inode = file_inode(filp);
5876 struct trace_array *tr = inode->i_private;
5877 struct trace_buffer *trace_buf = &tr->trace_buffer;
5878 int cpu = tracing_get_cpu(inode);
5879 struct trace_seq *s;
5880 unsigned long cnt;
5881 unsigned long long t;
5882 unsigned long usec_rem;
5883
5884 s = kmalloc(sizeof(*s), GFP_KERNEL);
5885 if (!s)
5886 return -ENOMEM;
5887
5888 trace_seq_init(s);
5889
5890 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5891 trace_seq_printf(s, "entries: %ld\n", cnt);
5892
5893 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5894 trace_seq_printf(s, "overrun: %ld\n", cnt);
5895
5896 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5897 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5898
5899 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5900 trace_seq_printf(s, "bytes: %ld\n", cnt);
5901
5902 if (trace_clocks[tr->clock_id].in_ns) {
5903 /* local or global for trace_clock */
5904 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5905 usec_rem = do_div(t, USEC_PER_SEC);
5906 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5907 t, usec_rem);
5908
5909 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5910 usec_rem = do_div(t, USEC_PER_SEC);
5911 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5912 } else {
5913 /* counter or tsc mode for trace_clock */
5914 trace_seq_printf(s, "oldest event ts: %llu\n",
5915 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5916
5917 trace_seq_printf(s, "now ts: %llu\n",
5918 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5919 }
5920
5921 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5922 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5923
5924 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5925 trace_seq_printf(s, "read events: %ld\n", cnt);
5926
5927 count = simple_read_from_buffer(ubuf, count, ppos,
5928 s->buffer, trace_seq_used(s));
5929
5930 kfree(s);
5931
5932 return count;
5933}
5934
5935static const struct file_operations tracing_stats_fops = {
5936 .open = tracing_open_generic_tr,
5937 .read = tracing_stats_read,
5938 .llseek = generic_file_llseek,
5939 .release = tracing_release_generic_tr,
5940};
5941
5942#ifdef CONFIG_DYNAMIC_FTRACE
5943
5944int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5945{
5946 return 0;
5947}
5948
5949static ssize_t
5950tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5951 size_t cnt, loff_t *ppos)
5952{
5953 static char ftrace_dyn_info_buffer[1024];
5954 static DEFINE_MUTEX(dyn_info_mutex);
5955 unsigned long *p = filp->private_data;
5956 char *buf = ftrace_dyn_info_buffer;
5957 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5958 int r;
5959
5960 mutex_lock(&dyn_info_mutex);
5961 r = sprintf(buf, "%ld ", *p);
5962
5963 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5964 buf[r++] = '\n';
5965
5966 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5967
5968 mutex_unlock(&dyn_info_mutex);
5969
5970 return r;
5971}
5972
5973static const struct file_operations tracing_dyn_info_fops = {
5974 .open = tracing_open_generic,
5975 .read = tracing_read_dyn_info,
5976 .llseek = generic_file_llseek,
5977};
5978#endif /* CONFIG_DYNAMIC_FTRACE */
5979
5980#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5981static void
5982ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5983{
5984 tracing_snapshot();
5985}
5986
5987static void
5988ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5989{
5990 unsigned long *count = (long *)data;
5991
5992 if (!*count)
5993 return;
5994
5995 if (*count != -1)
5996 (*count)--;
5997
5998 tracing_snapshot();
5999}
6000
6001static int
6002ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
6003 struct ftrace_probe_ops *ops, void *data)
6004{
6005 long count = (long)data;
6006
6007 seq_printf(m, "%ps:", (void *)ip);
6008
6009 seq_puts(m, "snapshot");
6010
6011 if (count == -1)
6012 seq_puts(m, ":unlimited\n");
6013 else
6014 seq_printf(m, ":count=%ld\n", count);
6015
6016 return 0;
6017}
6018
6019static struct ftrace_probe_ops snapshot_probe_ops = {
6020 .func = ftrace_snapshot,
6021 .print = ftrace_snapshot_print,
6022};
6023
6024static struct ftrace_probe_ops snapshot_count_probe_ops = {
6025 .func = ftrace_count_snapshot,
6026 .print = ftrace_snapshot_print,
6027};
6028
6029static int
6030ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
6031 char *glob, char *cmd, char *param, int enable)
6032{
6033 struct ftrace_probe_ops *ops;
6034 void *count = (void *)-1;
6035 char *number;
6036 int ret;
6037
6038 /* hash funcs only work with set_ftrace_filter */
6039 if (!enable)
6040 return -EINVAL;
6041
6042 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
6043
6044 if (glob[0] == '!') {
6045 unregister_ftrace_function_probe_func(glob+1, ops);
6046 return 0;
6047 }
6048
6049 if (!param)
6050 goto out_reg;
6051
6052 number = strsep(¶m, ":");
6053
6054 if (!strlen(number))
6055 goto out_reg;
6056
6057 /*
6058 * We use the callback data field (which is a pointer)
6059 * as our counter.
6060 */
6061 ret = kstrtoul(number, 0, (unsigned long *)&count);
6062 if (ret)
6063 return ret;
6064
6065 out_reg:
6066 ret = register_ftrace_function_probe(glob, ops, count);
6067
6068 if (ret >= 0)
6069 alloc_snapshot(&global_trace);
6070
6071 return ret < 0 ? ret : 0;
6072}
6073
6074static struct ftrace_func_command ftrace_snapshot_cmd = {
6075 .name = "snapshot",
6076 .func = ftrace_trace_snapshot_callback,
6077};
6078
6079static __init int register_snapshot_cmd(void)
6080{
6081 return register_ftrace_command(&ftrace_snapshot_cmd);
6082}
6083#else
6084static inline __init int register_snapshot_cmd(void) { return 0; }
6085#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
6086
6087static struct dentry *tracing_get_dentry(struct trace_array *tr)
6088{
6089 if (WARN_ON(!tr->dir))
6090 return ERR_PTR(-ENODEV);
6091
6092 /* Top directory uses NULL as the parent */
6093 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
6094 return NULL;
6095
6096 /* All sub buffers have a descriptor */
6097 return tr->dir;
6098}
6099
6100static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
6101{
6102 struct dentry *d_tracer;
6103
6104 if (tr->percpu_dir)
6105 return tr->percpu_dir;
6106
6107 d_tracer = tracing_get_dentry(tr);
6108 if (IS_ERR(d_tracer))
6109 return NULL;
6110
6111 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
6112
6113 WARN_ONCE(!tr->percpu_dir,
6114 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
6115
6116 return tr->percpu_dir;
6117}
6118
6119static struct dentry *
6120trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
6121 void *data, long cpu, const struct file_operations *fops)
6122{
6123 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
6124
6125 if (ret) /* See tracing_get_cpu() */
6126 d_inode(ret)->i_cdev = (void *)(cpu + 1);
6127 return ret;
6128}
6129
6130static void
6131tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
6132{
6133 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
6134 struct dentry *d_cpu;
6135 char cpu_dir[30]; /* 30 characters should be more than enough */
6136
6137 if (!d_percpu)
6138 return;
6139
6140 snprintf(cpu_dir, 30, "cpu%ld", cpu);
6141 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
6142 if (!d_cpu) {
6143 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
6144 return;
6145 }
6146
6147 /* per cpu trace_pipe */
6148 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
6149 tr, cpu, &tracing_pipe_fops);
6150
6151 /* per cpu trace */
6152 trace_create_cpu_file("trace", 0644, d_cpu,
6153 tr, cpu, &tracing_fops);
6154
6155 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
6156 tr, cpu, &tracing_buffers_fops);
6157
6158 trace_create_cpu_file("stats", 0444, d_cpu,
6159 tr, cpu, &tracing_stats_fops);
6160
6161 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
6162 tr, cpu, &tracing_entries_fops);
6163
6164#ifdef CONFIG_TRACER_SNAPSHOT
6165 trace_create_cpu_file("snapshot", 0644, d_cpu,
6166 tr, cpu, &snapshot_fops);
6167
6168 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
6169 tr, cpu, &snapshot_raw_fops);
6170#endif
6171}
6172
6173#ifdef CONFIG_FTRACE_SELFTEST
6174/* Let selftest have access to static functions in this file */
6175#include "trace_selftest.c"
6176#endif
6177
6178static ssize_t
6179trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
6180 loff_t *ppos)
6181{
6182 struct trace_option_dentry *topt = filp->private_data;
6183 char *buf;
6184
6185 if (topt->flags->val & topt->opt->bit)
6186 buf = "1\n";
6187 else
6188 buf = "0\n";
6189
6190 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
6191}
6192
6193static ssize_t
6194trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
6195 loff_t *ppos)
6196{
6197 struct trace_option_dentry *topt = filp->private_data;
6198 unsigned long val;
6199 int ret;
6200
6201 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6202 if (ret)
6203 return ret;
6204
6205 if (val != 0 && val != 1)
6206 return -EINVAL;
6207
6208 if (!!(topt->flags->val & topt->opt->bit) != val) {
6209 mutex_lock(&trace_types_lock);
6210 ret = __set_tracer_option(topt->tr, topt->flags,
6211 topt->opt, !val);
6212 mutex_unlock(&trace_types_lock);
6213 if (ret)
6214 return ret;
6215 }
6216
6217 *ppos += cnt;
6218
6219 return cnt;
6220}
6221
6222
6223static const struct file_operations trace_options_fops = {
6224 .open = tracing_open_generic,
6225 .read = trace_options_read,
6226 .write = trace_options_write,
6227 .llseek = generic_file_llseek,
6228};
6229
6230/*
6231 * In order to pass in both the trace_array descriptor as well as the index
6232 * to the flag that the trace option file represents, the trace_array
6233 * has a character array of trace_flags_index[], which holds the index
6234 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
6235 * The address of this character array is passed to the flag option file
6236 * read/write callbacks.
6237 *
6238 * In order to extract both the index and the trace_array descriptor,
6239 * get_tr_index() uses the following algorithm.
6240 *
6241 * idx = *ptr;
6242 *
6243 * As the pointer itself contains the address of the index (remember
6244 * index[1] == 1).
6245 *
6246 * Then to get the trace_array descriptor, by subtracting that index
6247 * from the ptr, we get to the start of the index itself.
6248 *
6249 * ptr - idx == &index[0]
6250 *
6251 * Then a simple container_of() from that pointer gets us to the
6252 * trace_array descriptor.
6253 */
6254static void get_tr_index(void *data, struct trace_array **ptr,
6255 unsigned int *pindex)
6256{
6257 *pindex = *(unsigned char *)data;
6258
6259 *ptr = container_of(data - *pindex, struct trace_array,
6260 trace_flags_index);
6261}
6262
6263static ssize_t
6264trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
6265 loff_t *ppos)
6266{
6267 void *tr_index = filp->private_data;
6268 struct trace_array *tr;
6269 unsigned int index;
6270 char *buf;
6271
6272 get_tr_index(tr_index, &tr, &index);
6273
6274 if (tr->trace_flags & (1 << index))
6275 buf = "1\n";
6276 else
6277 buf = "0\n";
6278
6279 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
6280}
6281
6282static ssize_t
6283trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
6284 loff_t *ppos)
6285{
6286 void *tr_index = filp->private_data;
6287 struct trace_array *tr;
6288 unsigned int index;
6289 unsigned long val;
6290 int ret;
6291
6292 get_tr_index(tr_index, &tr, &index);
6293
6294 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6295 if (ret)
6296 return ret;
6297
6298 if (val != 0 && val != 1)
6299 return -EINVAL;
6300
6301 mutex_lock(&trace_types_lock);
6302 ret = set_tracer_flag(tr, 1 << index, val);
6303 mutex_unlock(&trace_types_lock);
6304
6305 if (ret < 0)
6306 return ret;
6307
6308 *ppos += cnt;
6309
6310 return cnt;
6311}
6312
6313static const struct file_operations trace_options_core_fops = {
6314 .open = tracing_open_generic,
6315 .read = trace_options_core_read,
6316 .write = trace_options_core_write,
6317 .llseek = generic_file_llseek,
6318};
6319
6320struct dentry *trace_create_file(const char *name,
6321 umode_t mode,
6322 struct dentry *parent,
6323 void *data,
6324 const struct file_operations *fops)
6325{
6326 struct dentry *ret;
6327
6328 ret = tracefs_create_file(name, mode, parent, data, fops);
6329 if (!ret)
6330 pr_warn("Could not create tracefs '%s' entry\n", name);
6331
6332 return ret;
6333}
6334
6335
6336static struct dentry *trace_options_init_dentry(struct trace_array *tr)
6337{
6338 struct dentry *d_tracer;
6339
6340 if (tr->options)
6341 return tr->options;
6342
6343 d_tracer = tracing_get_dentry(tr);
6344 if (IS_ERR(d_tracer))
6345 return NULL;
6346
6347 tr->options = tracefs_create_dir("options", d_tracer);
6348 if (!tr->options) {
6349 pr_warn("Could not create tracefs directory 'options'\n");
6350 return NULL;
6351 }
6352
6353 return tr->options;
6354}
6355
6356static void
6357create_trace_option_file(struct trace_array *tr,
6358 struct trace_option_dentry *topt,
6359 struct tracer_flags *flags,
6360 struct tracer_opt *opt)
6361{
6362 struct dentry *t_options;
6363
6364 t_options = trace_options_init_dentry(tr);
6365 if (!t_options)
6366 return;
6367
6368 topt->flags = flags;
6369 topt->opt = opt;
6370 topt->tr = tr;
6371
6372 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
6373 &trace_options_fops);
6374
6375}
6376
6377static void
6378create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
6379{
6380 struct trace_option_dentry *topts;
6381 struct trace_options *tr_topts;
6382 struct tracer_flags *flags;
6383 struct tracer_opt *opts;
6384 int cnt;
6385 int i;
6386
6387 if (!tracer)
6388 return;
6389
6390 flags = tracer->flags;
6391
6392 if (!flags || !flags->opts)
6393 return;
6394
6395 /*
6396 * If this is an instance, only create flags for tracers
6397 * the instance may have.
6398 */
6399 if (!trace_ok_for_array(tracer, tr))
6400 return;
6401
6402 for (i = 0; i < tr->nr_topts; i++) {
6403 /* Make sure there's no duplicate flags. */
6404 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
6405 return;
6406 }
6407
6408 opts = flags->opts;
6409
6410 for (cnt = 0; opts[cnt].name; cnt++)
6411 ;
6412
6413 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
6414 if (!topts)
6415 return;
6416
6417 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
6418 GFP_KERNEL);
6419 if (!tr_topts) {
6420 kfree(topts);
6421 return;
6422 }
6423
6424 tr->topts = tr_topts;
6425 tr->topts[tr->nr_topts].tracer = tracer;
6426 tr->topts[tr->nr_topts].topts = topts;
6427 tr->nr_topts++;
6428
6429 for (cnt = 0; opts[cnt].name; cnt++) {
6430 create_trace_option_file(tr, &topts[cnt], flags,
6431 &opts[cnt]);
6432 WARN_ONCE(topts[cnt].entry == NULL,
6433 "Failed to create trace option: %s",
6434 opts[cnt].name);
6435 }
6436}
6437
6438static struct dentry *
6439create_trace_option_core_file(struct trace_array *tr,
6440 const char *option, long index)
6441{
6442 struct dentry *t_options;
6443
6444 t_options = trace_options_init_dentry(tr);
6445 if (!t_options)
6446 return NULL;
6447
6448 return trace_create_file(option, 0644, t_options,
6449 (void *)&tr->trace_flags_index[index],
6450 &trace_options_core_fops);
6451}
6452
6453static void create_trace_options_dir(struct trace_array *tr)
6454{
6455 struct dentry *t_options;
6456 bool top_level = tr == &global_trace;
6457 int i;
6458
6459 t_options = trace_options_init_dentry(tr);
6460 if (!t_options)
6461 return;
6462
6463 for (i = 0; trace_options[i]; i++) {
6464 if (top_level ||
6465 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
6466 create_trace_option_core_file(tr, trace_options[i], i);
6467 }
6468}
6469
6470static ssize_t
6471rb_simple_read(struct file *filp, char __user *ubuf,
6472 size_t cnt, loff_t *ppos)
6473{
6474 struct trace_array *tr = filp->private_data;
6475 char buf[64];
6476 int r;
6477
6478 r = tracer_tracing_is_on(tr);
6479 r = sprintf(buf, "%d\n", r);
6480
6481 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6482}
6483
6484static ssize_t
6485rb_simple_write(struct file *filp, const char __user *ubuf,
6486 size_t cnt, loff_t *ppos)
6487{
6488 struct trace_array *tr = filp->private_data;
6489 struct ring_buffer *buffer = tr->trace_buffer.buffer;
6490 unsigned long val;
6491 int ret;
6492
6493 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6494 if (ret)
6495 return ret;
6496
6497 if (buffer) {
6498 mutex_lock(&trace_types_lock);
6499 if (val) {
6500 tracer_tracing_on(tr);
6501 if (tr->current_trace->start)
6502 tr->current_trace->start(tr);
6503 } else {
6504 tracer_tracing_off(tr);
6505 if (tr->current_trace->stop)
6506 tr->current_trace->stop(tr);
6507 }
6508 mutex_unlock(&trace_types_lock);
6509 }
6510
6511 (*ppos)++;
6512
6513 return cnt;
6514}
6515
6516static const struct file_operations rb_simple_fops = {
6517 .open = tracing_open_generic_tr,
6518 .read = rb_simple_read,
6519 .write = rb_simple_write,
6520 .release = tracing_release_generic_tr,
6521 .llseek = default_llseek,
6522};
6523
6524struct dentry *trace_instance_dir;
6525
6526static void
6527init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
6528
6529static int
6530allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
6531{
6532 enum ring_buffer_flags rb_flags;
6533
6534 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
6535
6536 buf->tr = tr;
6537
6538 buf->buffer = ring_buffer_alloc(size, rb_flags);
6539 if (!buf->buffer)
6540 return -ENOMEM;
6541
6542 buf->data = alloc_percpu(struct trace_array_cpu);
6543 if (!buf->data) {
6544 ring_buffer_free(buf->buffer);
6545 return -ENOMEM;
6546 }
6547
6548 /* Allocate the first page for all buffers */
6549 set_buffer_entries(&tr->trace_buffer,
6550 ring_buffer_size(tr->trace_buffer.buffer, 0));
6551
6552 return 0;
6553}
6554
6555static int allocate_trace_buffers(struct trace_array *tr, int size)
6556{
6557 int ret;
6558
6559 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
6560 if (ret)
6561 return ret;
6562
6563#ifdef CONFIG_TRACER_MAX_TRACE
6564 ret = allocate_trace_buffer(tr, &tr->max_buffer,
6565 allocate_snapshot ? size : 1);
6566 if (WARN_ON(ret)) {
6567 ring_buffer_free(tr->trace_buffer.buffer);
6568 free_percpu(tr->trace_buffer.data);
6569 return -ENOMEM;
6570 }
6571 tr->allocated_snapshot = allocate_snapshot;
6572
6573 /*
6574 * Only the top level trace array gets its snapshot allocated
6575 * from the kernel command line.
6576 */
6577 allocate_snapshot = false;
6578#endif
6579 return 0;
6580}
6581
6582static void free_trace_buffer(struct trace_buffer *buf)
6583{
6584 if (buf->buffer) {
6585 ring_buffer_free(buf->buffer);
6586 buf->buffer = NULL;
6587 free_percpu(buf->data);
6588 buf->data = NULL;
6589 }
6590}
6591
6592static void free_trace_buffers(struct trace_array *tr)
6593{
6594 if (!tr)
6595 return;
6596
6597 free_trace_buffer(&tr->trace_buffer);
6598
6599#ifdef CONFIG_TRACER_MAX_TRACE
6600 free_trace_buffer(&tr->max_buffer);
6601#endif
6602}
6603
6604static void init_trace_flags_index(struct trace_array *tr)
6605{
6606 int i;
6607
6608 /* Used by the trace options files */
6609 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
6610 tr->trace_flags_index[i] = i;
6611}
6612
6613static void __update_tracer_options(struct trace_array *tr)
6614{
6615 struct tracer *t;
6616
6617 for (t = trace_types; t; t = t->next)
6618 add_tracer_options(tr, t);
6619}
6620
6621static void update_tracer_options(struct trace_array *tr)
6622{
6623 mutex_lock(&trace_types_lock);
6624 __update_tracer_options(tr);
6625 mutex_unlock(&trace_types_lock);
6626}
6627
6628static int instance_mkdir(const char *name)
6629{
6630 struct trace_array *tr;
6631 int ret;
6632
6633 mutex_lock(&trace_types_lock);
6634
6635 ret = -EEXIST;
6636 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6637 if (tr->name && strcmp(tr->name, name) == 0)
6638 goto out_unlock;
6639 }
6640
6641 ret = -ENOMEM;
6642 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
6643 if (!tr)
6644 goto out_unlock;
6645
6646 tr->name = kstrdup(name, GFP_KERNEL);
6647 if (!tr->name)
6648 goto out_free_tr;
6649
6650 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
6651 goto out_free_tr;
6652
6653 tr->trace_flags = global_trace.trace_flags;
6654
6655 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
6656
6657 raw_spin_lock_init(&tr->start_lock);
6658
6659 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
6660
6661 tr->current_trace = &nop_trace;
6662
6663 INIT_LIST_HEAD(&tr->systems);
6664 INIT_LIST_HEAD(&tr->events);
6665
6666 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
6667 goto out_free_tr;
6668
6669 tr->dir = tracefs_create_dir(name, trace_instance_dir);
6670 if (!tr->dir)
6671 goto out_free_tr;
6672
6673 ret = event_trace_add_tracer(tr->dir, tr);
6674 if (ret) {
6675 tracefs_remove_recursive(tr->dir);
6676 goto out_free_tr;
6677 }
6678
6679 init_tracer_tracefs(tr, tr->dir);
6680 init_trace_flags_index(tr);
6681 __update_tracer_options(tr);
6682
6683 list_add(&tr->list, &ftrace_trace_arrays);
6684
6685 mutex_unlock(&trace_types_lock);
6686
6687 return 0;
6688
6689 out_free_tr:
6690 free_trace_buffers(tr);
6691 free_cpumask_var(tr->tracing_cpumask);
6692 kfree(tr->name);
6693 kfree(tr);
6694
6695 out_unlock:
6696 mutex_unlock(&trace_types_lock);
6697
6698 return ret;
6699
6700}
6701
6702static int instance_rmdir(const char *name)
6703{
6704 struct trace_array *tr;
6705 int found = 0;
6706 int ret;
6707 int i;
6708
6709 mutex_lock(&trace_types_lock);
6710
6711 ret = -ENODEV;
6712 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6713 if (tr->name && strcmp(tr->name, name) == 0) {
6714 found = 1;
6715 break;
6716 }
6717 }
6718 if (!found)
6719 goto out_unlock;
6720
6721 ret = -EBUSY;
6722 if (tr->ref || (tr->current_trace && tr->current_trace->ref))
6723 goto out_unlock;
6724
6725 list_del(&tr->list);
6726
6727 tracing_set_nop(tr);
6728 event_trace_del_tracer(tr);
6729 ftrace_destroy_function_files(tr);
6730 tracefs_remove_recursive(tr->dir);
6731 free_trace_buffers(tr);
6732
6733 for (i = 0; i < tr->nr_topts; i++) {
6734 kfree(tr->topts[i].topts);
6735 }
6736 kfree(tr->topts);
6737
6738 kfree(tr->name);
6739 kfree(tr);
6740
6741 ret = 0;
6742
6743 out_unlock:
6744 mutex_unlock(&trace_types_lock);
6745
6746 return ret;
6747}
6748
6749static __init void create_trace_instances(struct dentry *d_tracer)
6750{
6751 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
6752 instance_mkdir,
6753 instance_rmdir);
6754 if (WARN_ON(!trace_instance_dir))
6755 return;
6756}
6757
6758static void
6759init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
6760{
6761 int cpu;
6762
6763 trace_create_file("available_tracers", 0444, d_tracer,
6764 tr, &show_traces_fops);
6765
6766 trace_create_file("current_tracer", 0644, d_tracer,
6767 tr, &set_tracer_fops);
6768
6769 trace_create_file("tracing_cpumask", 0644, d_tracer,
6770 tr, &tracing_cpumask_fops);
6771
6772 trace_create_file("trace_options", 0644, d_tracer,
6773 tr, &tracing_iter_fops);
6774
6775 trace_create_file("trace", 0644, d_tracer,
6776 tr, &tracing_fops);
6777
6778 trace_create_file("trace_pipe", 0444, d_tracer,
6779 tr, &tracing_pipe_fops);
6780
6781 trace_create_file("buffer_size_kb", 0644, d_tracer,
6782 tr, &tracing_entries_fops);
6783
6784 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6785 tr, &tracing_total_entries_fops);
6786
6787 trace_create_file("free_buffer", 0200, d_tracer,
6788 tr, &tracing_free_buffer_fops);
6789
6790 trace_create_file("trace_marker", 0220, d_tracer,
6791 tr, &tracing_mark_fops);
6792
6793 trace_create_file("trace_clock", 0644, d_tracer, tr,
6794 &trace_clock_fops);
6795
6796 trace_create_file("tracing_on", 0644, d_tracer,
6797 tr, &rb_simple_fops);
6798
6799 create_trace_options_dir(tr);
6800
6801#ifdef CONFIG_TRACER_MAX_TRACE
6802 trace_create_file("tracing_max_latency", 0644, d_tracer,
6803 &tr->max_latency, &tracing_max_lat_fops);
6804#endif
6805
6806 if (ftrace_create_function_files(tr, d_tracer))
6807 WARN(1, "Could not allocate function filter files");
6808
6809#ifdef CONFIG_TRACER_SNAPSHOT
6810 trace_create_file("snapshot", 0644, d_tracer,
6811 tr, &snapshot_fops);
6812#endif
6813
6814 for_each_tracing_cpu(cpu)
6815 tracing_init_tracefs_percpu(tr, cpu);
6816
6817}
6818
6819static struct vfsmount *trace_automount(void *ingore)
6820{
6821 struct vfsmount *mnt;
6822 struct file_system_type *type;
6823
6824 /*
6825 * To maintain backward compatibility for tools that mount
6826 * debugfs to get to the tracing facility, tracefs is automatically
6827 * mounted to the debugfs/tracing directory.
6828 */
6829 type = get_fs_type("tracefs");
6830 if (!type)
6831 return NULL;
6832 mnt = vfs_kern_mount(type, 0, "tracefs", NULL);
6833 put_filesystem(type);
6834 if (IS_ERR(mnt))
6835 return NULL;
6836 mntget(mnt);
6837
6838 return mnt;
6839}
6840
6841/**
6842 * tracing_init_dentry - initialize top level trace array
6843 *
6844 * This is called when creating files or directories in the tracing
6845 * directory. It is called via fs_initcall() by any of the boot up code
6846 * and expects to return the dentry of the top level tracing directory.
6847 */
6848struct dentry *tracing_init_dentry(void)
6849{
6850 struct trace_array *tr = &global_trace;
6851
6852 /* The top level trace array uses NULL as parent */
6853 if (tr->dir)
6854 return NULL;
6855
6856 if (WARN_ON(!tracefs_initialized()) ||
6857 (IS_ENABLED(CONFIG_DEBUG_FS) &&
6858 WARN_ON(!debugfs_initialized())))
6859 return ERR_PTR(-ENODEV);
6860
6861 /*
6862 * As there may still be users that expect the tracing
6863 * files to exist in debugfs/tracing, we must automount
6864 * the tracefs file system there, so older tools still
6865 * work with the newer kerenl.
6866 */
6867 tr->dir = debugfs_create_automount("tracing", NULL,
6868 trace_automount, NULL);
6869 if (!tr->dir) {
6870 pr_warn_once("Could not create debugfs directory 'tracing'\n");
6871 return ERR_PTR(-ENOMEM);
6872 }
6873
6874 return NULL;
6875}
6876
6877extern struct trace_enum_map *__start_ftrace_enum_maps[];
6878extern struct trace_enum_map *__stop_ftrace_enum_maps[];
6879
6880static void __init trace_enum_init(void)
6881{
6882 int len;
6883
6884 len = __stop_ftrace_enum_maps - __start_ftrace_enum_maps;
6885 trace_insert_enum_map(NULL, __start_ftrace_enum_maps, len);
6886}
6887
6888#ifdef CONFIG_MODULES
6889static void trace_module_add_enums(struct module *mod)
6890{
6891 if (!mod->num_trace_enums)
6892 return;
6893
6894 /*
6895 * Modules with bad taint do not have events created, do
6896 * not bother with enums either.
6897 */
6898 if (trace_module_has_bad_taint(mod))
6899 return;
6900
6901 trace_insert_enum_map(mod, mod->trace_enums, mod->num_trace_enums);
6902}
6903
6904#ifdef CONFIG_TRACE_ENUM_MAP_FILE
6905static void trace_module_remove_enums(struct module *mod)
6906{
6907 union trace_enum_map_item *map;
6908 union trace_enum_map_item **last = &trace_enum_maps;
6909
6910 if (!mod->num_trace_enums)
6911 return;
6912
6913 mutex_lock(&trace_enum_mutex);
6914
6915 map = trace_enum_maps;
6916
6917 while (map) {
6918 if (map->head.mod == mod)
6919 break;
6920 map = trace_enum_jmp_to_tail(map);
6921 last = &map->tail.next;
6922 map = map->tail.next;
6923 }
6924 if (!map)
6925 goto out;
6926
6927 *last = trace_enum_jmp_to_tail(map)->tail.next;
6928 kfree(map);
6929 out:
6930 mutex_unlock(&trace_enum_mutex);
6931}
6932#else
6933static inline void trace_module_remove_enums(struct module *mod) { }
6934#endif /* CONFIG_TRACE_ENUM_MAP_FILE */
6935
6936static int trace_module_notify(struct notifier_block *self,
6937 unsigned long val, void *data)
6938{
6939 struct module *mod = data;
6940
6941 switch (val) {
6942 case MODULE_STATE_COMING:
6943 trace_module_add_enums(mod);
6944 break;
6945 case MODULE_STATE_GOING:
6946 trace_module_remove_enums(mod);
6947 break;
6948 }
6949
6950 return 0;
6951}
6952
6953static struct notifier_block trace_module_nb = {
6954 .notifier_call = trace_module_notify,
6955 .priority = 0,
6956};
6957#endif /* CONFIG_MODULES */
6958
6959static __init int tracer_init_tracefs(void)
6960{
6961 struct dentry *d_tracer;
6962
6963 trace_access_lock_init();
6964
6965 d_tracer = tracing_init_dentry();
6966 if (IS_ERR(d_tracer))
6967 return 0;
6968
6969 init_tracer_tracefs(&global_trace, d_tracer);
6970
6971 trace_create_file("tracing_thresh", 0644, d_tracer,
6972 &global_trace, &tracing_thresh_fops);
6973
6974 trace_create_file("README", 0444, d_tracer,
6975 NULL, &tracing_readme_fops);
6976
6977 trace_create_file("saved_cmdlines", 0444, d_tracer,
6978 NULL, &tracing_saved_cmdlines_fops);
6979
6980 trace_create_file("saved_cmdlines_size", 0644, d_tracer,
6981 NULL, &tracing_saved_cmdlines_size_fops);
6982
6983 trace_enum_init();
6984
6985 trace_create_enum_file(d_tracer);
6986
6987#ifdef CONFIG_MODULES
6988 register_module_notifier(&trace_module_nb);
6989#endif
6990
6991#ifdef CONFIG_DYNAMIC_FTRACE
6992 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6993 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6994#endif
6995
6996 create_trace_instances(d_tracer);
6997
6998 update_tracer_options(&global_trace);
6999
7000 return 0;
7001}
7002
7003static int trace_panic_handler(struct notifier_block *this,
7004 unsigned long event, void *unused)
7005{
7006 if (ftrace_dump_on_oops)
7007 ftrace_dump(ftrace_dump_on_oops);
7008 return NOTIFY_OK;
7009}
7010
7011static struct notifier_block trace_panic_notifier = {
7012 .notifier_call = trace_panic_handler,
7013 .next = NULL,
7014 .priority = 150 /* priority: INT_MAX >= x >= 0 */
7015};
7016
7017static int trace_die_handler(struct notifier_block *self,
7018 unsigned long val,
7019 void *data)
7020{
7021 switch (val) {
7022 case DIE_OOPS:
7023 if (ftrace_dump_on_oops)
7024 ftrace_dump(ftrace_dump_on_oops);
7025 break;
7026 default:
7027 break;
7028 }
7029 return NOTIFY_OK;
7030}
7031
7032static struct notifier_block trace_die_notifier = {
7033 .notifier_call = trace_die_handler,
7034 .priority = 200
7035};
7036
7037/*
7038 * printk is set to max of 1024, we really don't need it that big.
7039 * Nothing should be printing 1000 characters anyway.
7040 */
7041#define TRACE_MAX_PRINT 1000
7042
7043/*
7044 * Define here KERN_TRACE so that we have one place to modify
7045 * it if we decide to change what log level the ftrace dump
7046 * should be at.
7047 */
7048#define KERN_TRACE KERN_EMERG
7049
7050void
7051trace_printk_seq(struct trace_seq *s)
7052{
7053 /* Probably should print a warning here. */
7054 if (s->seq.len >= TRACE_MAX_PRINT)
7055 s->seq.len = TRACE_MAX_PRINT;
7056
7057 /*
7058 * More paranoid code. Although the buffer size is set to
7059 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
7060 * an extra layer of protection.
7061 */
7062 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
7063 s->seq.len = s->seq.size - 1;
7064
7065 /* should be zero ended, but we are paranoid. */
7066 s->buffer[s->seq.len] = 0;
7067
7068 printk(KERN_TRACE "%s", s->buffer);
7069
7070 trace_seq_init(s);
7071}
7072
7073void trace_init_global_iter(struct trace_iterator *iter)
7074{
7075 iter->tr = &global_trace;
7076 iter->trace = iter->tr->current_trace;
7077 iter->cpu_file = RING_BUFFER_ALL_CPUS;
7078 iter->trace_buffer = &global_trace.trace_buffer;
7079
7080 if (iter->trace && iter->trace->open)
7081 iter->trace->open(iter);
7082
7083 /* Annotate start of buffers if we had overruns */
7084 if (ring_buffer_overruns(iter->trace_buffer->buffer))
7085 iter->iter_flags |= TRACE_FILE_ANNOTATE;
7086
7087 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
7088 if (trace_clocks[iter->tr->clock_id].in_ns)
7089 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
7090}
7091
7092void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
7093{
7094 /* use static because iter can be a bit big for the stack */
7095 static struct trace_iterator iter;
7096 static atomic_t dump_running;
7097 struct trace_array *tr = &global_trace;
7098 unsigned int old_userobj;
7099 unsigned long flags;
7100 int cnt = 0, cpu;
7101
7102 /* Only allow one dump user at a time. */
7103 if (atomic_inc_return(&dump_running) != 1) {
7104 atomic_dec(&dump_running);
7105 return;
7106 }
7107
7108 /*
7109 * Always turn off tracing when we dump.
7110 * We don't need to show trace output of what happens
7111 * between multiple crashes.
7112 *
7113 * If the user does a sysrq-z, then they can re-enable
7114 * tracing with echo 1 > tracing_on.
7115 */
7116 tracing_off();
7117
7118 local_irq_save(flags);
7119
7120 /* Simulate the iterator */
7121 trace_init_global_iter(&iter);
7122
7123 for_each_tracing_cpu(cpu) {
7124 atomic_inc(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
7125 }
7126
7127 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
7128
7129 /* don't look at user memory in panic mode */
7130 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
7131
7132 switch (oops_dump_mode) {
7133 case DUMP_ALL:
7134 iter.cpu_file = RING_BUFFER_ALL_CPUS;
7135 break;
7136 case DUMP_ORIG:
7137 iter.cpu_file = raw_smp_processor_id();
7138 break;
7139 case DUMP_NONE:
7140 goto out_enable;
7141 default:
7142 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
7143 iter.cpu_file = RING_BUFFER_ALL_CPUS;
7144 }
7145
7146 printk(KERN_TRACE "Dumping ftrace buffer:\n");
7147
7148 /* Did function tracer already get disabled? */
7149 if (ftrace_is_dead()) {
7150 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
7151 printk("# MAY BE MISSING FUNCTION EVENTS\n");
7152 }
7153
7154 /*
7155 * We need to stop all tracing on all CPUS to read the
7156 * the next buffer. This is a bit expensive, but is
7157 * not done often. We fill all what we can read,
7158 * and then release the locks again.
7159 */
7160
7161 while (!trace_empty(&iter)) {
7162
7163 if (!cnt)
7164 printk(KERN_TRACE "---------------------------------\n");
7165
7166 cnt++;
7167
7168 /* reset all but tr, trace, and overruns */
7169 memset(&iter.seq, 0,
7170 sizeof(struct trace_iterator) -
7171 offsetof(struct trace_iterator, seq));
7172 iter.iter_flags |= TRACE_FILE_LAT_FMT;
7173 iter.pos = -1;
7174
7175 if (trace_find_next_entry_inc(&iter) != NULL) {
7176 int ret;
7177
7178 ret = print_trace_line(&iter);
7179 if (ret != TRACE_TYPE_NO_CONSUME)
7180 trace_consume(&iter);
7181 }
7182 touch_nmi_watchdog();
7183
7184 trace_printk_seq(&iter.seq);
7185 }
7186
7187 if (!cnt)
7188 printk(KERN_TRACE " (ftrace buffer empty)\n");
7189 else
7190 printk(KERN_TRACE "---------------------------------\n");
7191
7192 out_enable:
7193 tr->trace_flags |= old_userobj;
7194
7195 for_each_tracing_cpu(cpu) {
7196 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
7197 }
7198 atomic_dec(&dump_running);
7199 local_irq_restore(flags);
7200}
7201EXPORT_SYMBOL_GPL(ftrace_dump);
7202
7203__init static int tracer_alloc_buffers(void)
7204{
7205 int ring_buf_size;
7206 int ret = -ENOMEM;
7207
7208 /*
7209 * Make sure we don't accidently add more trace options
7210 * than we have bits for.
7211 */
7212 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
7213
7214 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
7215 goto out;
7216
7217 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
7218 goto out_free_buffer_mask;
7219
7220 /* Only allocate trace_printk buffers if a trace_printk exists */
7221 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
7222 /* Must be called before global_trace.buffer is allocated */
7223 trace_printk_init_buffers();
7224
7225 /* To save memory, keep the ring buffer size to its minimum */
7226 if (ring_buffer_expanded)
7227 ring_buf_size = trace_buf_size;
7228 else
7229 ring_buf_size = 1;
7230
7231 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
7232 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
7233
7234 raw_spin_lock_init(&global_trace.start_lock);
7235
7236 /* Used for event triggers */
7237 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
7238 if (!temp_buffer)
7239 goto out_free_cpumask;
7240
7241 if (trace_create_savedcmd() < 0)
7242 goto out_free_temp_buffer;
7243
7244 /* TODO: make the number of buffers hot pluggable with CPUS */
7245 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
7246 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
7247 WARN_ON(1);
7248 goto out_free_savedcmd;
7249 }
7250
7251 if (global_trace.buffer_disabled)
7252 tracing_off();
7253
7254 if (trace_boot_clock) {
7255 ret = tracing_set_clock(&global_trace, trace_boot_clock);
7256 if (ret < 0)
7257 pr_warn("Trace clock %s not defined, going back to default\n",
7258 trace_boot_clock);
7259 }
7260
7261 /*
7262 * register_tracer() might reference current_trace, so it
7263 * needs to be set before we register anything. This is
7264 * just a bootstrap of current_trace anyway.
7265 */
7266 global_trace.current_trace = &nop_trace;
7267
7268 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
7269
7270 ftrace_init_global_array_ops(&global_trace);
7271
7272 init_trace_flags_index(&global_trace);
7273
7274 register_tracer(&nop_trace);
7275
7276 /* All seems OK, enable tracing */
7277 tracing_disabled = 0;
7278
7279 atomic_notifier_chain_register(&panic_notifier_list,
7280 &trace_panic_notifier);
7281
7282 register_die_notifier(&trace_die_notifier);
7283
7284 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
7285
7286 INIT_LIST_HEAD(&global_trace.systems);
7287 INIT_LIST_HEAD(&global_trace.events);
7288 list_add(&global_trace.list, &ftrace_trace_arrays);
7289
7290 apply_trace_boot_options();
7291
7292 register_snapshot_cmd();
7293
7294 return 0;
7295
7296out_free_savedcmd:
7297 free_saved_cmdlines_buffer(savedcmd);
7298out_free_temp_buffer:
7299 ring_buffer_free(temp_buffer);
7300out_free_cpumask:
7301 free_cpumask_var(global_trace.tracing_cpumask);
7302out_free_buffer_mask:
7303 free_cpumask_var(tracing_buffer_mask);
7304out:
7305 return ret;
7306}
7307
7308void __init trace_init(void)
7309{
7310 if (tracepoint_printk) {
7311 tracepoint_print_iter =
7312 kmalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
7313 if (WARN_ON(!tracepoint_print_iter))
7314 tracepoint_printk = 0;
7315 }
7316 tracer_alloc_buffers();
7317 trace_event_init();
7318}
7319
7320__init static int clear_boot_tracer(void)
7321{
7322 /*
7323 * The default tracer at boot buffer is an init section.
7324 * This function is called in lateinit. If we did not
7325 * find the boot tracer, then clear it out, to prevent
7326 * later registration from accessing the buffer that is
7327 * about to be freed.
7328 */
7329 if (!default_bootup_tracer)
7330 return 0;
7331
7332 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
7333 default_bootup_tracer);
7334 default_bootup_tracer = NULL;
7335
7336 return 0;
7337}
7338
7339fs_initcall(tracer_init_tracefs);
7340late_initcall(clear_boot_tracer);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * ring buffer based function tracer
4 *
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 *
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 *
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
14 */
15#include <linux/ring_buffer.h>
16#include <linux/utsname.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 <asm/setup.h> /* COMMAND_LINE_SIZE */
53
54#include "trace.h"
55#include "trace_output.h"
56
57#ifdef CONFIG_FTRACE_STARTUP_TEST
58/*
59 * We need to change this state when a selftest is running.
60 * A selftest will lurk into the ring-buffer to count the
61 * entries inserted during the selftest although some concurrent
62 * insertions into the ring-buffer such as trace_printk could occurred
63 * at the same time, giving false positive or negative results.
64 */
65static bool __read_mostly tracing_selftest_running;
66
67/*
68 * If boot-time tracing including tracers/events via kernel cmdline
69 * is running, we do not want to run SELFTEST.
70 */
71bool __read_mostly tracing_selftest_disabled;
72
73void __init disable_tracing_selftest(const char *reason)
74{
75 if (!tracing_selftest_disabled) {
76 tracing_selftest_disabled = true;
77 pr_info("Ftrace startup test is disabled due to %s\n", reason);
78 }
79}
80#else
81#define tracing_selftest_running 0
82#define tracing_selftest_disabled 0
83#endif
84
85/* Pipe tracepoints to printk */
86static struct trace_iterator *tracepoint_print_iter;
87int tracepoint_printk;
88static bool tracepoint_printk_stop_on_boot __initdata;
89static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
90
91/* For tracers that don't implement custom flags */
92static struct tracer_opt dummy_tracer_opt[] = {
93 { }
94};
95
96static int
97dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
98{
99 return 0;
100}
101
102/*
103 * To prevent the comm cache from being overwritten when no
104 * tracing is active, only save the comm when a trace event
105 * occurred.
106 */
107DEFINE_PER_CPU(bool, trace_taskinfo_save);
108
109/*
110 * Kill all tracing for good (never come back).
111 * It is initialized to 1 but will turn to zero if the initialization
112 * of the tracer is successful. But that is the only place that sets
113 * this back to zero.
114 */
115static int tracing_disabled = 1;
116
117cpumask_var_t __read_mostly tracing_buffer_mask;
118
119/*
120 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
121 *
122 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
123 * is set, then ftrace_dump is called. This will output the contents
124 * of the ftrace buffers to the console. This is very useful for
125 * capturing traces that lead to crashes and outputing it to a
126 * serial console.
127 *
128 * It is default off, but you can enable it with either specifying
129 * "ftrace_dump_on_oops" in the kernel command line, or setting
130 * /proc/sys/kernel/ftrace_dump_on_oops
131 * Set 1 if you want to dump buffers of all CPUs
132 * Set 2 if you want to dump the buffer of the CPU that triggered oops
133 * Set instance name if you want to dump the specific trace instance
134 * Multiple instance dump is also supported, and instances are seperated
135 * by commas.
136 */
137/* Set to string format zero to disable by default */
138char ftrace_dump_on_oops[MAX_TRACER_SIZE] = "0";
139
140/* When set, tracing will stop when a WARN*() is hit */
141int __disable_trace_on_warning;
142
143#ifdef CONFIG_TRACE_EVAL_MAP_FILE
144/* Map of enums to their values, for "eval_map" file */
145struct trace_eval_map_head {
146 struct module *mod;
147 unsigned long length;
148};
149
150union trace_eval_map_item;
151
152struct trace_eval_map_tail {
153 /*
154 * "end" is first and points to NULL as it must be different
155 * than "mod" or "eval_string"
156 */
157 union trace_eval_map_item *next;
158 const char *end; /* points to NULL */
159};
160
161static DEFINE_MUTEX(trace_eval_mutex);
162
163/*
164 * The trace_eval_maps are saved in an array with two extra elements,
165 * one at the beginning, and one at the end. The beginning item contains
166 * the count of the saved maps (head.length), and the module they
167 * belong to if not built in (head.mod). The ending item contains a
168 * pointer to the next array of saved eval_map items.
169 */
170union trace_eval_map_item {
171 struct trace_eval_map map;
172 struct trace_eval_map_head head;
173 struct trace_eval_map_tail tail;
174};
175
176static union trace_eval_map_item *trace_eval_maps;
177#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
178
179int tracing_set_tracer(struct trace_array *tr, const char *buf);
180static void ftrace_trace_userstack(struct trace_array *tr,
181 struct trace_buffer *buffer,
182 unsigned int trace_ctx);
183
184static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
185static char *default_bootup_tracer;
186
187static bool allocate_snapshot;
188static bool snapshot_at_boot;
189
190static char boot_instance_info[COMMAND_LINE_SIZE] __initdata;
191static int boot_instance_index;
192
193static char boot_snapshot_info[COMMAND_LINE_SIZE] __initdata;
194static int boot_snapshot_index;
195
196static int __init set_cmdline_ftrace(char *str)
197{
198 strscpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
199 default_bootup_tracer = bootup_tracer_buf;
200 /* We are using ftrace early, expand it */
201 trace_set_ring_buffer_expanded(NULL);
202 return 1;
203}
204__setup("ftrace=", set_cmdline_ftrace);
205
206int ftrace_dump_on_oops_enabled(void)
207{
208 if (!strcmp("0", ftrace_dump_on_oops))
209 return 0;
210 else
211 return 1;
212}
213
214static int __init set_ftrace_dump_on_oops(char *str)
215{
216 if (!*str) {
217 strscpy(ftrace_dump_on_oops, "1", MAX_TRACER_SIZE);
218 return 1;
219 }
220
221 if (*str == ',') {
222 strscpy(ftrace_dump_on_oops, "1", MAX_TRACER_SIZE);
223 strscpy(ftrace_dump_on_oops + 1, str, MAX_TRACER_SIZE - 1);
224 return 1;
225 }
226
227 if (*str++ == '=') {
228 strscpy(ftrace_dump_on_oops, str, MAX_TRACER_SIZE);
229 return 1;
230 }
231
232 return 0;
233}
234__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
235
236static int __init stop_trace_on_warning(char *str)
237{
238 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
239 __disable_trace_on_warning = 1;
240 return 1;
241}
242__setup("traceoff_on_warning", stop_trace_on_warning);
243
244static int __init boot_alloc_snapshot(char *str)
245{
246 char *slot = boot_snapshot_info + boot_snapshot_index;
247 int left = sizeof(boot_snapshot_info) - boot_snapshot_index;
248 int ret;
249
250 if (str[0] == '=') {
251 str++;
252 if (strlen(str) >= left)
253 return -1;
254
255 ret = snprintf(slot, left, "%s\t", str);
256 boot_snapshot_index += ret;
257 } else {
258 allocate_snapshot = true;
259 /* We also need the main ring buffer expanded */
260 trace_set_ring_buffer_expanded(NULL);
261 }
262 return 1;
263}
264__setup("alloc_snapshot", boot_alloc_snapshot);
265
266
267static int __init boot_snapshot(char *str)
268{
269 snapshot_at_boot = true;
270 boot_alloc_snapshot(str);
271 return 1;
272}
273__setup("ftrace_boot_snapshot", boot_snapshot);
274
275
276static int __init boot_instance(char *str)
277{
278 char *slot = boot_instance_info + boot_instance_index;
279 int left = sizeof(boot_instance_info) - boot_instance_index;
280 int ret;
281
282 if (strlen(str) >= left)
283 return -1;
284
285 ret = snprintf(slot, left, "%s\t", str);
286 boot_instance_index += ret;
287
288 return 1;
289}
290__setup("trace_instance=", boot_instance);
291
292
293static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
294
295static int __init set_trace_boot_options(char *str)
296{
297 strscpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
298 return 1;
299}
300__setup("trace_options=", set_trace_boot_options);
301
302static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
303static char *trace_boot_clock __initdata;
304
305static int __init set_trace_boot_clock(char *str)
306{
307 strscpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
308 trace_boot_clock = trace_boot_clock_buf;
309 return 1;
310}
311__setup("trace_clock=", set_trace_boot_clock);
312
313static int __init set_tracepoint_printk(char *str)
314{
315 /* Ignore the "tp_printk_stop_on_boot" param */
316 if (*str == '_')
317 return 0;
318
319 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
320 tracepoint_printk = 1;
321 return 1;
322}
323__setup("tp_printk", set_tracepoint_printk);
324
325static int __init set_tracepoint_printk_stop(char *str)
326{
327 tracepoint_printk_stop_on_boot = true;
328 return 1;
329}
330__setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
331
332unsigned long long ns2usecs(u64 nsec)
333{
334 nsec += 500;
335 do_div(nsec, 1000);
336 return nsec;
337}
338
339static void
340trace_process_export(struct trace_export *export,
341 struct ring_buffer_event *event, int flag)
342{
343 struct trace_entry *entry;
344 unsigned int size = 0;
345
346 if (export->flags & flag) {
347 entry = ring_buffer_event_data(event);
348 size = ring_buffer_event_length(event);
349 export->write(export, entry, size);
350 }
351}
352
353static DEFINE_MUTEX(ftrace_export_lock);
354
355static struct trace_export __rcu *ftrace_exports_list __read_mostly;
356
357static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
358static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
359static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
360
361static inline void ftrace_exports_enable(struct trace_export *export)
362{
363 if (export->flags & TRACE_EXPORT_FUNCTION)
364 static_branch_inc(&trace_function_exports_enabled);
365
366 if (export->flags & TRACE_EXPORT_EVENT)
367 static_branch_inc(&trace_event_exports_enabled);
368
369 if (export->flags & TRACE_EXPORT_MARKER)
370 static_branch_inc(&trace_marker_exports_enabled);
371}
372
373static inline void ftrace_exports_disable(struct trace_export *export)
374{
375 if (export->flags & TRACE_EXPORT_FUNCTION)
376 static_branch_dec(&trace_function_exports_enabled);
377
378 if (export->flags & TRACE_EXPORT_EVENT)
379 static_branch_dec(&trace_event_exports_enabled);
380
381 if (export->flags & TRACE_EXPORT_MARKER)
382 static_branch_dec(&trace_marker_exports_enabled);
383}
384
385static void ftrace_exports(struct ring_buffer_event *event, int flag)
386{
387 struct trace_export *export;
388
389 preempt_disable_notrace();
390
391 export = rcu_dereference_raw_check(ftrace_exports_list);
392 while (export) {
393 trace_process_export(export, event, flag);
394 export = rcu_dereference_raw_check(export->next);
395 }
396
397 preempt_enable_notrace();
398}
399
400static inline void
401add_trace_export(struct trace_export **list, struct trace_export *export)
402{
403 rcu_assign_pointer(export->next, *list);
404 /*
405 * We are entering export into the list but another
406 * CPU might be walking that list. We need to make sure
407 * the export->next pointer is valid before another CPU sees
408 * the export pointer included into the list.
409 */
410 rcu_assign_pointer(*list, export);
411}
412
413static inline int
414rm_trace_export(struct trace_export **list, struct trace_export *export)
415{
416 struct trace_export **p;
417
418 for (p = list; *p != NULL; p = &(*p)->next)
419 if (*p == export)
420 break;
421
422 if (*p != export)
423 return -1;
424
425 rcu_assign_pointer(*p, (*p)->next);
426
427 return 0;
428}
429
430static inline void
431add_ftrace_export(struct trace_export **list, struct trace_export *export)
432{
433 ftrace_exports_enable(export);
434
435 add_trace_export(list, export);
436}
437
438static inline int
439rm_ftrace_export(struct trace_export **list, struct trace_export *export)
440{
441 int ret;
442
443 ret = rm_trace_export(list, export);
444 ftrace_exports_disable(export);
445
446 return ret;
447}
448
449int register_ftrace_export(struct trace_export *export)
450{
451 if (WARN_ON_ONCE(!export->write))
452 return -1;
453
454 mutex_lock(&ftrace_export_lock);
455
456 add_ftrace_export(&ftrace_exports_list, export);
457
458 mutex_unlock(&ftrace_export_lock);
459
460 return 0;
461}
462EXPORT_SYMBOL_GPL(register_ftrace_export);
463
464int unregister_ftrace_export(struct trace_export *export)
465{
466 int ret;
467
468 mutex_lock(&ftrace_export_lock);
469
470 ret = rm_ftrace_export(&ftrace_exports_list, export);
471
472 mutex_unlock(&ftrace_export_lock);
473
474 return ret;
475}
476EXPORT_SYMBOL_GPL(unregister_ftrace_export);
477
478/* trace_flags holds trace_options default values */
479#define TRACE_DEFAULT_FLAGS \
480 (FUNCTION_DEFAULT_FLAGS | \
481 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
482 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
483 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
484 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
485 TRACE_ITER_HASH_PTR)
486
487/* trace_options that are only supported by global_trace */
488#define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
489 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
490
491/* trace_flags that are default zero for instances */
492#define ZEROED_TRACE_FLAGS \
493 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
494
495/*
496 * The global_trace is the descriptor that holds the top-level tracing
497 * buffers for the live tracing.
498 */
499static struct trace_array global_trace = {
500 .trace_flags = TRACE_DEFAULT_FLAGS,
501};
502
503void trace_set_ring_buffer_expanded(struct trace_array *tr)
504{
505 if (!tr)
506 tr = &global_trace;
507 tr->ring_buffer_expanded = true;
508}
509
510LIST_HEAD(ftrace_trace_arrays);
511
512int trace_array_get(struct trace_array *this_tr)
513{
514 struct trace_array *tr;
515 int ret = -ENODEV;
516
517 mutex_lock(&trace_types_lock);
518 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
519 if (tr == this_tr) {
520 tr->ref++;
521 ret = 0;
522 break;
523 }
524 }
525 mutex_unlock(&trace_types_lock);
526
527 return ret;
528}
529
530static void __trace_array_put(struct trace_array *this_tr)
531{
532 WARN_ON(!this_tr->ref);
533 this_tr->ref--;
534}
535
536/**
537 * trace_array_put - Decrement the reference counter for this trace array.
538 * @this_tr : pointer to the trace array
539 *
540 * NOTE: Use this when we no longer need the trace array returned by
541 * trace_array_get_by_name(). This ensures the trace array can be later
542 * destroyed.
543 *
544 */
545void trace_array_put(struct trace_array *this_tr)
546{
547 if (!this_tr)
548 return;
549
550 mutex_lock(&trace_types_lock);
551 __trace_array_put(this_tr);
552 mutex_unlock(&trace_types_lock);
553}
554EXPORT_SYMBOL_GPL(trace_array_put);
555
556int tracing_check_open_get_tr(struct trace_array *tr)
557{
558 int ret;
559
560 ret = security_locked_down(LOCKDOWN_TRACEFS);
561 if (ret)
562 return ret;
563
564 if (tracing_disabled)
565 return -ENODEV;
566
567 if (tr && trace_array_get(tr) < 0)
568 return -ENODEV;
569
570 return 0;
571}
572
573int call_filter_check_discard(struct trace_event_call *call, void *rec,
574 struct trace_buffer *buffer,
575 struct ring_buffer_event *event)
576{
577 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
578 !filter_match_preds(call->filter, rec)) {
579 __trace_event_discard_commit(buffer, event);
580 return 1;
581 }
582
583 return 0;
584}
585
586/**
587 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
588 * @filtered_pids: The list of pids to check
589 * @search_pid: The PID to find in @filtered_pids
590 *
591 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
592 */
593bool
594trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
595{
596 return trace_pid_list_is_set(filtered_pids, search_pid);
597}
598
599/**
600 * trace_ignore_this_task - should a task be ignored for tracing
601 * @filtered_pids: The list of pids to check
602 * @filtered_no_pids: The list of pids not to be traced
603 * @task: The task that should be ignored if not filtered
604 *
605 * Checks if @task should be traced or not from @filtered_pids.
606 * Returns true if @task should *NOT* be traced.
607 * Returns false if @task should be traced.
608 */
609bool
610trace_ignore_this_task(struct trace_pid_list *filtered_pids,
611 struct trace_pid_list *filtered_no_pids,
612 struct task_struct *task)
613{
614 /*
615 * If filtered_no_pids is not empty, and the task's pid is listed
616 * in filtered_no_pids, then return true.
617 * Otherwise, if filtered_pids is empty, that means we can
618 * trace all tasks. If it has content, then only trace pids
619 * within filtered_pids.
620 */
621
622 return (filtered_pids &&
623 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
624 (filtered_no_pids &&
625 trace_find_filtered_pid(filtered_no_pids, task->pid));
626}
627
628/**
629 * trace_filter_add_remove_task - Add or remove a task from a pid_list
630 * @pid_list: The list to modify
631 * @self: The current task for fork or NULL for exit
632 * @task: The task to add or remove
633 *
634 * If adding a task, if @self is defined, the task is only added if @self
635 * is also included in @pid_list. This happens on fork and tasks should
636 * only be added when the parent is listed. If @self is NULL, then the
637 * @task pid will be removed from the list, which would happen on exit
638 * of a task.
639 */
640void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
641 struct task_struct *self,
642 struct task_struct *task)
643{
644 if (!pid_list)
645 return;
646
647 /* For forks, we only add if the forking task is listed */
648 if (self) {
649 if (!trace_find_filtered_pid(pid_list, self->pid))
650 return;
651 }
652
653 /* "self" is set for forks, and NULL for exits */
654 if (self)
655 trace_pid_list_set(pid_list, task->pid);
656 else
657 trace_pid_list_clear(pid_list, task->pid);
658}
659
660/**
661 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
662 * @pid_list: The pid list to show
663 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
664 * @pos: The position of the file
665 *
666 * This is used by the seq_file "next" operation to iterate the pids
667 * listed in a trace_pid_list structure.
668 *
669 * Returns the pid+1 as we want to display pid of zero, but NULL would
670 * stop the iteration.
671 */
672void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
673{
674 long pid = (unsigned long)v;
675 unsigned int next;
676
677 (*pos)++;
678
679 /* pid already is +1 of the actual previous bit */
680 if (trace_pid_list_next(pid_list, pid, &next) < 0)
681 return NULL;
682
683 pid = next;
684
685 /* Return pid + 1 to allow zero to be represented */
686 return (void *)(pid + 1);
687}
688
689/**
690 * trace_pid_start - Used for seq_file to start reading pid lists
691 * @pid_list: The pid list to show
692 * @pos: The position of the file
693 *
694 * This is used by seq_file "start" operation to start the iteration
695 * of listing pids.
696 *
697 * Returns the pid+1 as we want to display pid of zero, but NULL would
698 * stop the iteration.
699 */
700void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
701{
702 unsigned long pid;
703 unsigned int first;
704 loff_t l = 0;
705
706 if (trace_pid_list_first(pid_list, &first) < 0)
707 return NULL;
708
709 pid = first;
710
711 /* Return pid + 1 so that zero can be the exit value */
712 for (pid++; pid && l < *pos;
713 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
714 ;
715 return (void *)pid;
716}
717
718/**
719 * trace_pid_show - show the current pid in seq_file processing
720 * @m: The seq_file structure to write into
721 * @v: A void pointer of the pid (+1) value to display
722 *
723 * Can be directly used by seq_file operations to display the current
724 * pid value.
725 */
726int trace_pid_show(struct seq_file *m, void *v)
727{
728 unsigned long pid = (unsigned long)v - 1;
729
730 seq_printf(m, "%lu\n", pid);
731 return 0;
732}
733
734/* 128 should be much more than enough */
735#define PID_BUF_SIZE 127
736
737int trace_pid_write(struct trace_pid_list *filtered_pids,
738 struct trace_pid_list **new_pid_list,
739 const char __user *ubuf, size_t cnt)
740{
741 struct trace_pid_list *pid_list;
742 struct trace_parser parser;
743 unsigned long val;
744 int nr_pids = 0;
745 ssize_t read = 0;
746 ssize_t ret;
747 loff_t pos;
748 pid_t pid;
749
750 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
751 return -ENOMEM;
752
753 /*
754 * Always recreate a new array. The write is an all or nothing
755 * operation. Always create a new array when adding new pids by
756 * the user. If the operation fails, then the current list is
757 * not modified.
758 */
759 pid_list = trace_pid_list_alloc();
760 if (!pid_list) {
761 trace_parser_put(&parser);
762 return -ENOMEM;
763 }
764
765 if (filtered_pids) {
766 /* copy the current bits to the new max */
767 ret = trace_pid_list_first(filtered_pids, &pid);
768 while (!ret) {
769 trace_pid_list_set(pid_list, pid);
770 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
771 nr_pids++;
772 }
773 }
774
775 ret = 0;
776 while (cnt > 0) {
777
778 pos = 0;
779
780 ret = trace_get_user(&parser, ubuf, cnt, &pos);
781 if (ret < 0)
782 break;
783
784 read += ret;
785 ubuf += ret;
786 cnt -= ret;
787
788 if (!trace_parser_loaded(&parser))
789 break;
790
791 ret = -EINVAL;
792 if (kstrtoul(parser.buffer, 0, &val))
793 break;
794
795 pid = (pid_t)val;
796
797 if (trace_pid_list_set(pid_list, pid) < 0) {
798 ret = -1;
799 break;
800 }
801 nr_pids++;
802
803 trace_parser_clear(&parser);
804 ret = 0;
805 }
806 trace_parser_put(&parser);
807
808 if (ret < 0) {
809 trace_pid_list_free(pid_list);
810 return ret;
811 }
812
813 if (!nr_pids) {
814 /* Cleared the list of pids */
815 trace_pid_list_free(pid_list);
816 pid_list = NULL;
817 }
818
819 *new_pid_list = pid_list;
820
821 return read;
822}
823
824static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
825{
826 u64 ts;
827
828 /* Early boot up does not have a buffer yet */
829 if (!buf->buffer)
830 return trace_clock_local();
831
832 ts = ring_buffer_time_stamp(buf->buffer);
833 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
834
835 return ts;
836}
837
838u64 ftrace_now(int cpu)
839{
840 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
841}
842
843/**
844 * tracing_is_enabled - Show if global_trace has been enabled
845 *
846 * Shows if the global trace has been enabled or not. It uses the
847 * mirror flag "buffer_disabled" to be used in fast paths such as for
848 * the irqsoff tracer. But it may be inaccurate due to races. If you
849 * need to know the accurate state, use tracing_is_on() which is a little
850 * slower, but accurate.
851 */
852int tracing_is_enabled(void)
853{
854 /*
855 * For quick access (irqsoff uses this in fast path), just
856 * return the mirror variable of the state of the ring buffer.
857 * It's a little racy, but we don't really care.
858 */
859 smp_rmb();
860 return !global_trace.buffer_disabled;
861}
862
863/*
864 * trace_buf_size is the size in bytes that is allocated
865 * for a buffer. Note, the number of bytes is always rounded
866 * to page size.
867 *
868 * This number is purposely set to a low number of 16384.
869 * If the dump on oops happens, it will be much appreciated
870 * to not have to wait for all that output. Anyway this can be
871 * boot time and run time configurable.
872 */
873#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
874
875static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
876
877/* trace_types holds a link list of available tracers. */
878static struct tracer *trace_types __read_mostly;
879
880/*
881 * trace_types_lock is used to protect the trace_types list.
882 */
883DEFINE_MUTEX(trace_types_lock);
884
885/*
886 * serialize the access of the ring buffer
887 *
888 * ring buffer serializes readers, but it is low level protection.
889 * The validity of the events (which returns by ring_buffer_peek() ..etc)
890 * are not protected by ring buffer.
891 *
892 * The content of events may become garbage if we allow other process consumes
893 * these events concurrently:
894 * A) the page of the consumed events may become a normal page
895 * (not reader page) in ring buffer, and this page will be rewritten
896 * by events producer.
897 * B) The page of the consumed events may become a page for splice_read,
898 * and this page will be returned to system.
899 *
900 * These primitives allow multi process access to different cpu ring buffer
901 * concurrently.
902 *
903 * These primitives don't distinguish read-only and read-consume access.
904 * Multi read-only access are also serialized.
905 */
906
907#ifdef CONFIG_SMP
908static DECLARE_RWSEM(all_cpu_access_lock);
909static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
910
911static inline void trace_access_lock(int cpu)
912{
913 if (cpu == RING_BUFFER_ALL_CPUS) {
914 /* gain it for accessing the whole ring buffer. */
915 down_write(&all_cpu_access_lock);
916 } else {
917 /* gain it for accessing a cpu ring buffer. */
918
919 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
920 down_read(&all_cpu_access_lock);
921
922 /* Secondly block other access to this @cpu ring buffer. */
923 mutex_lock(&per_cpu(cpu_access_lock, cpu));
924 }
925}
926
927static inline void trace_access_unlock(int cpu)
928{
929 if (cpu == RING_BUFFER_ALL_CPUS) {
930 up_write(&all_cpu_access_lock);
931 } else {
932 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
933 up_read(&all_cpu_access_lock);
934 }
935}
936
937static inline void trace_access_lock_init(void)
938{
939 int cpu;
940
941 for_each_possible_cpu(cpu)
942 mutex_init(&per_cpu(cpu_access_lock, cpu));
943}
944
945#else
946
947static DEFINE_MUTEX(access_lock);
948
949static inline void trace_access_lock(int cpu)
950{
951 (void)cpu;
952 mutex_lock(&access_lock);
953}
954
955static inline void trace_access_unlock(int cpu)
956{
957 (void)cpu;
958 mutex_unlock(&access_lock);
959}
960
961static inline void trace_access_lock_init(void)
962{
963}
964
965#endif
966
967#ifdef CONFIG_STACKTRACE
968static void __ftrace_trace_stack(struct trace_buffer *buffer,
969 unsigned int trace_ctx,
970 int skip, struct pt_regs *regs);
971static inline void ftrace_trace_stack(struct trace_array *tr,
972 struct trace_buffer *buffer,
973 unsigned int trace_ctx,
974 int skip, struct pt_regs *regs);
975
976#else
977static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
978 unsigned int trace_ctx,
979 int skip, struct pt_regs *regs)
980{
981}
982static inline void ftrace_trace_stack(struct trace_array *tr,
983 struct trace_buffer *buffer,
984 unsigned long trace_ctx,
985 int skip, struct pt_regs *regs)
986{
987}
988
989#endif
990
991static __always_inline void
992trace_event_setup(struct ring_buffer_event *event,
993 int type, unsigned int trace_ctx)
994{
995 struct trace_entry *ent = ring_buffer_event_data(event);
996
997 tracing_generic_entry_update(ent, type, trace_ctx);
998}
999
1000static __always_inline struct ring_buffer_event *
1001__trace_buffer_lock_reserve(struct trace_buffer *buffer,
1002 int type,
1003 unsigned long len,
1004 unsigned int trace_ctx)
1005{
1006 struct ring_buffer_event *event;
1007
1008 event = ring_buffer_lock_reserve(buffer, len);
1009 if (event != NULL)
1010 trace_event_setup(event, type, trace_ctx);
1011
1012 return event;
1013}
1014
1015void tracer_tracing_on(struct trace_array *tr)
1016{
1017 if (tr->array_buffer.buffer)
1018 ring_buffer_record_on(tr->array_buffer.buffer);
1019 /*
1020 * This flag is looked at when buffers haven't been allocated
1021 * yet, or by some tracers (like irqsoff), that just want to
1022 * know if the ring buffer has been disabled, but it can handle
1023 * races of where it gets disabled but we still do a record.
1024 * As the check is in the fast path of the tracers, it is more
1025 * important to be fast than accurate.
1026 */
1027 tr->buffer_disabled = 0;
1028 /* Make the flag seen by readers */
1029 smp_wmb();
1030}
1031
1032/**
1033 * tracing_on - enable tracing buffers
1034 *
1035 * This function enables tracing buffers that may have been
1036 * disabled with tracing_off.
1037 */
1038void tracing_on(void)
1039{
1040 tracer_tracing_on(&global_trace);
1041}
1042EXPORT_SYMBOL_GPL(tracing_on);
1043
1044
1045static __always_inline void
1046__buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
1047{
1048 __this_cpu_write(trace_taskinfo_save, true);
1049
1050 /* If this is the temp buffer, we need to commit fully */
1051 if (this_cpu_read(trace_buffered_event) == event) {
1052 /* Length is in event->array[0] */
1053 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1054 /* Release the temp buffer */
1055 this_cpu_dec(trace_buffered_event_cnt);
1056 /* ring_buffer_unlock_commit() enables preemption */
1057 preempt_enable_notrace();
1058 } else
1059 ring_buffer_unlock_commit(buffer);
1060}
1061
1062int __trace_array_puts(struct trace_array *tr, unsigned long ip,
1063 const char *str, int size)
1064{
1065 struct ring_buffer_event *event;
1066 struct trace_buffer *buffer;
1067 struct print_entry *entry;
1068 unsigned int trace_ctx;
1069 int alloc;
1070
1071 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1072 return 0;
1073
1074 if (unlikely(tracing_selftest_running && tr == &global_trace))
1075 return 0;
1076
1077 if (unlikely(tracing_disabled))
1078 return 0;
1079
1080 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1081
1082 trace_ctx = tracing_gen_ctx();
1083 buffer = tr->array_buffer.buffer;
1084 ring_buffer_nest_start(buffer);
1085 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1086 trace_ctx);
1087 if (!event) {
1088 size = 0;
1089 goto out;
1090 }
1091
1092 entry = ring_buffer_event_data(event);
1093 entry->ip = ip;
1094
1095 memcpy(&entry->buf, str, size);
1096
1097 /* Add a newline if necessary */
1098 if (entry->buf[size - 1] != '\n') {
1099 entry->buf[size] = '\n';
1100 entry->buf[size + 1] = '\0';
1101 } else
1102 entry->buf[size] = '\0';
1103
1104 __buffer_unlock_commit(buffer, event);
1105 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1106 out:
1107 ring_buffer_nest_end(buffer);
1108 return size;
1109}
1110EXPORT_SYMBOL_GPL(__trace_array_puts);
1111
1112/**
1113 * __trace_puts - write a constant string into the trace buffer.
1114 * @ip: The address of the caller
1115 * @str: The constant string to write
1116 * @size: The size of the string.
1117 */
1118int __trace_puts(unsigned long ip, const char *str, int size)
1119{
1120 return __trace_array_puts(&global_trace, ip, str, size);
1121}
1122EXPORT_SYMBOL_GPL(__trace_puts);
1123
1124/**
1125 * __trace_bputs - write the pointer to a constant string into trace buffer
1126 * @ip: The address of the caller
1127 * @str: The constant string to write to the buffer to
1128 */
1129int __trace_bputs(unsigned long ip, const char *str)
1130{
1131 struct ring_buffer_event *event;
1132 struct trace_buffer *buffer;
1133 struct bputs_entry *entry;
1134 unsigned int trace_ctx;
1135 int size = sizeof(struct bputs_entry);
1136 int ret = 0;
1137
1138 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1139 return 0;
1140
1141 if (unlikely(tracing_selftest_running || tracing_disabled))
1142 return 0;
1143
1144 trace_ctx = tracing_gen_ctx();
1145 buffer = global_trace.array_buffer.buffer;
1146
1147 ring_buffer_nest_start(buffer);
1148 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1149 trace_ctx);
1150 if (!event)
1151 goto out;
1152
1153 entry = ring_buffer_event_data(event);
1154 entry->ip = ip;
1155 entry->str = str;
1156
1157 __buffer_unlock_commit(buffer, event);
1158 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1159
1160 ret = 1;
1161 out:
1162 ring_buffer_nest_end(buffer);
1163 return ret;
1164}
1165EXPORT_SYMBOL_GPL(__trace_bputs);
1166
1167#ifdef CONFIG_TRACER_SNAPSHOT
1168static void tracing_snapshot_instance_cond(struct trace_array *tr,
1169 void *cond_data)
1170{
1171 struct tracer *tracer = tr->current_trace;
1172 unsigned long flags;
1173
1174 if (in_nmi()) {
1175 trace_array_puts(tr, "*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1176 trace_array_puts(tr, "*** snapshot is being ignored ***\n");
1177 return;
1178 }
1179
1180 if (!tr->allocated_snapshot) {
1181 trace_array_puts(tr, "*** SNAPSHOT NOT ALLOCATED ***\n");
1182 trace_array_puts(tr, "*** stopping trace here! ***\n");
1183 tracer_tracing_off(tr);
1184 return;
1185 }
1186
1187 /* Note, snapshot can not be used when the tracer uses it */
1188 if (tracer->use_max_tr) {
1189 trace_array_puts(tr, "*** LATENCY TRACER ACTIVE ***\n");
1190 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1191 return;
1192 }
1193
1194 local_irq_save(flags);
1195 update_max_tr(tr, current, smp_processor_id(), cond_data);
1196 local_irq_restore(flags);
1197}
1198
1199void tracing_snapshot_instance(struct trace_array *tr)
1200{
1201 tracing_snapshot_instance_cond(tr, NULL);
1202}
1203
1204/**
1205 * tracing_snapshot - take a snapshot of the current buffer.
1206 *
1207 * This causes a swap between the snapshot buffer and the current live
1208 * tracing buffer. You can use this to take snapshots of the live
1209 * trace when some condition is triggered, but continue to trace.
1210 *
1211 * Note, make sure to allocate the snapshot with either
1212 * a tracing_snapshot_alloc(), or by doing it manually
1213 * with: echo 1 > /sys/kernel/tracing/snapshot
1214 *
1215 * If the snapshot buffer is not allocated, it will stop tracing.
1216 * Basically making a permanent snapshot.
1217 */
1218void tracing_snapshot(void)
1219{
1220 struct trace_array *tr = &global_trace;
1221
1222 tracing_snapshot_instance(tr);
1223}
1224EXPORT_SYMBOL_GPL(tracing_snapshot);
1225
1226/**
1227 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1228 * @tr: The tracing instance to snapshot
1229 * @cond_data: The data to be tested conditionally, and possibly saved
1230 *
1231 * This is the same as tracing_snapshot() except that the snapshot is
1232 * conditional - the snapshot will only happen if the
1233 * cond_snapshot.update() implementation receiving the cond_data
1234 * returns true, which means that the trace array's cond_snapshot
1235 * update() operation used the cond_data to determine whether the
1236 * snapshot should be taken, and if it was, presumably saved it along
1237 * with the snapshot.
1238 */
1239void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1240{
1241 tracing_snapshot_instance_cond(tr, cond_data);
1242}
1243EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1244
1245/**
1246 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1247 * @tr: The tracing instance
1248 *
1249 * When the user enables a conditional snapshot using
1250 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1251 * with the snapshot. This accessor is used to retrieve it.
1252 *
1253 * Should not be called from cond_snapshot.update(), since it takes
1254 * the tr->max_lock lock, which the code calling
1255 * cond_snapshot.update() has already done.
1256 *
1257 * Returns the cond_data associated with the trace array's snapshot.
1258 */
1259void *tracing_cond_snapshot_data(struct trace_array *tr)
1260{
1261 void *cond_data = NULL;
1262
1263 local_irq_disable();
1264 arch_spin_lock(&tr->max_lock);
1265
1266 if (tr->cond_snapshot)
1267 cond_data = tr->cond_snapshot->cond_data;
1268
1269 arch_spin_unlock(&tr->max_lock);
1270 local_irq_enable();
1271
1272 return cond_data;
1273}
1274EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1275
1276static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1277 struct array_buffer *size_buf, int cpu_id);
1278static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1279
1280int tracing_alloc_snapshot_instance(struct trace_array *tr)
1281{
1282 int order;
1283 int ret;
1284
1285 if (!tr->allocated_snapshot) {
1286
1287 /* Make the snapshot buffer have the same order as main buffer */
1288 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
1289 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
1290 if (ret < 0)
1291 return ret;
1292
1293 /* allocate spare buffer */
1294 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1295 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1296 if (ret < 0)
1297 return ret;
1298
1299 tr->allocated_snapshot = true;
1300 }
1301
1302 return 0;
1303}
1304
1305static void free_snapshot(struct trace_array *tr)
1306{
1307 /*
1308 * We don't free the ring buffer. instead, resize it because
1309 * The max_tr ring buffer has some state (e.g. ring->clock) and
1310 * we want preserve it.
1311 */
1312 ring_buffer_subbuf_order_set(tr->max_buffer.buffer, 0);
1313 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1314 set_buffer_entries(&tr->max_buffer, 1);
1315 tracing_reset_online_cpus(&tr->max_buffer);
1316 tr->allocated_snapshot = false;
1317}
1318
1319static int tracing_arm_snapshot_locked(struct trace_array *tr)
1320{
1321 int ret;
1322
1323 lockdep_assert_held(&trace_types_lock);
1324
1325 spin_lock(&tr->snapshot_trigger_lock);
1326 if (tr->snapshot == UINT_MAX) {
1327 spin_unlock(&tr->snapshot_trigger_lock);
1328 return -EBUSY;
1329 }
1330
1331 tr->snapshot++;
1332 spin_unlock(&tr->snapshot_trigger_lock);
1333
1334 ret = tracing_alloc_snapshot_instance(tr);
1335 if (ret) {
1336 spin_lock(&tr->snapshot_trigger_lock);
1337 tr->snapshot--;
1338 spin_unlock(&tr->snapshot_trigger_lock);
1339 }
1340
1341 return ret;
1342}
1343
1344int tracing_arm_snapshot(struct trace_array *tr)
1345{
1346 int ret;
1347
1348 mutex_lock(&trace_types_lock);
1349 ret = tracing_arm_snapshot_locked(tr);
1350 mutex_unlock(&trace_types_lock);
1351
1352 return ret;
1353}
1354
1355void tracing_disarm_snapshot(struct trace_array *tr)
1356{
1357 spin_lock(&tr->snapshot_trigger_lock);
1358 if (!WARN_ON(!tr->snapshot))
1359 tr->snapshot--;
1360 spin_unlock(&tr->snapshot_trigger_lock);
1361}
1362
1363/**
1364 * tracing_alloc_snapshot - allocate snapshot buffer.
1365 *
1366 * This only allocates the snapshot buffer if it isn't already
1367 * allocated - it doesn't also take a snapshot.
1368 *
1369 * This is meant to be used in cases where the snapshot buffer needs
1370 * to be set up for events that can't sleep but need to be able to
1371 * trigger a snapshot.
1372 */
1373int tracing_alloc_snapshot(void)
1374{
1375 struct trace_array *tr = &global_trace;
1376 int ret;
1377
1378 ret = tracing_alloc_snapshot_instance(tr);
1379 WARN_ON(ret < 0);
1380
1381 return ret;
1382}
1383EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1384
1385/**
1386 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1387 *
1388 * This is similar to tracing_snapshot(), but it will allocate the
1389 * snapshot buffer if it isn't already allocated. Use this only
1390 * where it is safe to sleep, as the allocation may sleep.
1391 *
1392 * This causes a swap between the snapshot buffer and the current live
1393 * tracing buffer. You can use this to take snapshots of the live
1394 * trace when some condition is triggered, but continue to trace.
1395 */
1396void tracing_snapshot_alloc(void)
1397{
1398 int ret;
1399
1400 ret = tracing_alloc_snapshot();
1401 if (ret < 0)
1402 return;
1403
1404 tracing_snapshot();
1405}
1406EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1407
1408/**
1409 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1410 * @tr: The tracing instance
1411 * @cond_data: User data to associate with the snapshot
1412 * @update: Implementation of the cond_snapshot update function
1413 *
1414 * Check whether the conditional snapshot for the given instance has
1415 * already been enabled, or if the current tracer is already using a
1416 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1417 * save the cond_data and update function inside.
1418 *
1419 * Returns 0 if successful, error otherwise.
1420 */
1421int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1422 cond_update_fn_t update)
1423{
1424 struct cond_snapshot *cond_snapshot;
1425 int ret = 0;
1426
1427 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1428 if (!cond_snapshot)
1429 return -ENOMEM;
1430
1431 cond_snapshot->cond_data = cond_data;
1432 cond_snapshot->update = update;
1433
1434 mutex_lock(&trace_types_lock);
1435
1436 if (tr->current_trace->use_max_tr) {
1437 ret = -EBUSY;
1438 goto fail_unlock;
1439 }
1440
1441 /*
1442 * The cond_snapshot can only change to NULL without the
1443 * trace_types_lock. We don't care if we race with it going
1444 * to NULL, but we want to make sure that it's not set to
1445 * something other than NULL when we get here, which we can
1446 * do safely with only holding the trace_types_lock and not
1447 * having to take the max_lock.
1448 */
1449 if (tr->cond_snapshot) {
1450 ret = -EBUSY;
1451 goto fail_unlock;
1452 }
1453
1454 ret = tracing_arm_snapshot_locked(tr);
1455 if (ret)
1456 goto fail_unlock;
1457
1458 local_irq_disable();
1459 arch_spin_lock(&tr->max_lock);
1460 tr->cond_snapshot = cond_snapshot;
1461 arch_spin_unlock(&tr->max_lock);
1462 local_irq_enable();
1463
1464 mutex_unlock(&trace_types_lock);
1465
1466 return ret;
1467
1468 fail_unlock:
1469 mutex_unlock(&trace_types_lock);
1470 kfree(cond_snapshot);
1471 return ret;
1472}
1473EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1474
1475/**
1476 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1477 * @tr: The tracing instance
1478 *
1479 * Check whether the conditional snapshot for the given instance is
1480 * enabled; if so, free the cond_snapshot associated with it,
1481 * otherwise return -EINVAL.
1482 *
1483 * Returns 0 if successful, error otherwise.
1484 */
1485int tracing_snapshot_cond_disable(struct trace_array *tr)
1486{
1487 int ret = 0;
1488
1489 local_irq_disable();
1490 arch_spin_lock(&tr->max_lock);
1491
1492 if (!tr->cond_snapshot)
1493 ret = -EINVAL;
1494 else {
1495 kfree(tr->cond_snapshot);
1496 tr->cond_snapshot = NULL;
1497 }
1498
1499 arch_spin_unlock(&tr->max_lock);
1500 local_irq_enable();
1501
1502 tracing_disarm_snapshot(tr);
1503
1504 return ret;
1505}
1506EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1507#else
1508void tracing_snapshot(void)
1509{
1510 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1511}
1512EXPORT_SYMBOL_GPL(tracing_snapshot);
1513void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1514{
1515 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1516}
1517EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1518int tracing_alloc_snapshot(void)
1519{
1520 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1521 return -ENODEV;
1522}
1523EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1524void tracing_snapshot_alloc(void)
1525{
1526 /* Give warning */
1527 tracing_snapshot();
1528}
1529EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1530void *tracing_cond_snapshot_data(struct trace_array *tr)
1531{
1532 return NULL;
1533}
1534EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1535int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1536{
1537 return -ENODEV;
1538}
1539EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1540int tracing_snapshot_cond_disable(struct trace_array *tr)
1541{
1542 return false;
1543}
1544EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1545#define free_snapshot(tr) do { } while (0)
1546#define tracing_arm_snapshot_locked(tr) ({ -EBUSY; })
1547#endif /* CONFIG_TRACER_SNAPSHOT */
1548
1549void tracer_tracing_off(struct trace_array *tr)
1550{
1551 if (tr->array_buffer.buffer)
1552 ring_buffer_record_off(tr->array_buffer.buffer);
1553 /*
1554 * This flag is looked at when buffers haven't been allocated
1555 * yet, or by some tracers (like irqsoff), that just want to
1556 * know if the ring buffer has been disabled, but it can handle
1557 * races of where it gets disabled but we still do a record.
1558 * As the check is in the fast path of the tracers, it is more
1559 * important to be fast than accurate.
1560 */
1561 tr->buffer_disabled = 1;
1562 /* Make the flag seen by readers */
1563 smp_wmb();
1564}
1565
1566/**
1567 * tracing_off - turn off tracing buffers
1568 *
1569 * This function stops the tracing buffers from recording data.
1570 * It does not disable any overhead the tracers themselves may
1571 * be causing. This function simply causes all recording to
1572 * the ring buffers to fail.
1573 */
1574void tracing_off(void)
1575{
1576 tracer_tracing_off(&global_trace);
1577}
1578EXPORT_SYMBOL_GPL(tracing_off);
1579
1580void disable_trace_on_warning(void)
1581{
1582 if (__disable_trace_on_warning) {
1583 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1584 "Disabling tracing due to warning\n");
1585 tracing_off();
1586 }
1587}
1588
1589/**
1590 * tracer_tracing_is_on - show real state of ring buffer enabled
1591 * @tr : the trace array to know if ring buffer is enabled
1592 *
1593 * Shows real state of the ring buffer if it is enabled or not.
1594 */
1595bool tracer_tracing_is_on(struct trace_array *tr)
1596{
1597 if (tr->array_buffer.buffer)
1598 return ring_buffer_record_is_set_on(tr->array_buffer.buffer);
1599 return !tr->buffer_disabled;
1600}
1601
1602/**
1603 * tracing_is_on - show state of ring buffers enabled
1604 */
1605int tracing_is_on(void)
1606{
1607 return tracer_tracing_is_on(&global_trace);
1608}
1609EXPORT_SYMBOL_GPL(tracing_is_on);
1610
1611static int __init set_buf_size(char *str)
1612{
1613 unsigned long buf_size;
1614
1615 if (!str)
1616 return 0;
1617 buf_size = memparse(str, &str);
1618 /*
1619 * nr_entries can not be zero and the startup
1620 * tests require some buffer space. Therefore
1621 * ensure we have at least 4096 bytes of buffer.
1622 */
1623 trace_buf_size = max(4096UL, buf_size);
1624 return 1;
1625}
1626__setup("trace_buf_size=", set_buf_size);
1627
1628static int __init set_tracing_thresh(char *str)
1629{
1630 unsigned long threshold;
1631 int ret;
1632
1633 if (!str)
1634 return 0;
1635 ret = kstrtoul(str, 0, &threshold);
1636 if (ret < 0)
1637 return 0;
1638 tracing_thresh = threshold * 1000;
1639 return 1;
1640}
1641__setup("tracing_thresh=", set_tracing_thresh);
1642
1643unsigned long nsecs_to_usecs(unsigned long nsecs)
1644{
1645 return nsecs / 1000;
1646}
1647
1648/*
1649 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1650 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1651 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1652 * of strings in the order that the evals (enum) were defined.
1653 */
1654#undef C
1655#define C(a, b) b
1656
1657/* These must match the bit positions in trace_iterator_flags */
1658static const char *trace_options[] = {
1659 TRACE_FLAGS
1660 NULL
1661};
1662
1663static struct {
1664 u64 (*func)(void);
1665 const char *name;
1666 int in_ns; /* is this clock in nanoseconds? */
1667} trace_clocks[] = {
1668 { trace_clock_local, "local", 1 },
1669 { trace_clock_global, "global", 1 },
1670 { trace_clock_counter, "counter", 0 },
1671 { trace_clock_jiffies, "uptime", 0 },
1672 { trace_clock, "perf", 1 },
1673 { ktime_get_mono_fast_ns, "mono", 1 },
1674 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1675 { ktime_get_boot_fast_ns, "boot", 1 },
1676 { ktime_get_tai_fast_ns, "tai", 1 },
1677 ARCH_TRACE_CLOCKS
1678};
1679
1680bool trace_clock_in_ns(struct trace_array *tr)
1681{
1682 if (trace_clocks[tr->clock_id].in_ns)
1683 return true;
1684
1685 return false;
1686}
1687
1688/*
1689 * trace_parser_get_init - gets the buffer for trace parser
1690 */
1691int trace_parser_get_init(struct trace_parser *parser, int size)
1692{
1693 memset(parser, 0, sizeof(*parser));
1694
1695 parser->buffer = kmalloc(size, GFP_KERNEL);
1696 if (!parser->buffer)
1697 return 1;
1698
1699 parser->size = size;
1700 return 0;
1701}
1702
1703/*
1704 * trace_parser_put - frees the buffer for trace parser
1705 */
1706void trace_parser_put(struct trace_parser *parser)
1707{
1708 kfree(parser->buffer);
1709 parser->buffer = NULL;
1710}
1711
1712/*
1713 * trace_get_user - reads the user input string separated by space
1714 * (matched by isspace(ch))
1715 *
1716 * For each string found the 'struct trace_parser' is updated,
1717 * and the function returns.
1718 *
1719 * Returns number of bytes read.
1720 *
1721 * See kernel/trace/trace.h for 'struct trace_parser' details.
1722 */
1723int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1724 size_t cnt, loff_t *ppos)
1725{
1726 char ch;
1727 size_t read = 0;
1728 ssize_t ret;
1729
1730 if (!*ppos)
1731 trace_parser_clear(parser);
1732
1733 ret = get_user(ch, ubuf++);
1734 if (ret)
1735 goto out;
1736
1737 read++;
1738 cnt--;
1739
1740 /*
1741 * The parser is not finished with the last write,
1742 * continue reading the user input without skipping spaces.
1743 */
1744 if (!parser->cont) {
1745 /* skip white space */
1746 while (cnt && isspace(ch)) {
1747 ret = get_user(ch, ubuf++);
1748 if (ret)
1749 goto out;
1750 read++;
1751 cnt--;
1752 }
1753
1754 parser->idx = 0;
1755
1756 /* only spaces were written */
1757 if (isspace(ch) || !ch) {
1758 *ppos += read;
1759 ret = read;
1760 goto out;
1761 }
1762 }
1763
1764 /* read the non-space input */
1765 while (cnt && !isspace(ch) && ch) {
1766 if (parser->idx < parser->size - 1)
1767 parser->buffer[parser->idx++] = ch;
1768 else {
1769 ret = -EINVAL;
1770 goto out;
1771 }
1772 ret = get_user(ch, ubuf++);
1773 if (ret)
1774 goto out;
1775 read++;
1776 cnt--;
1777 }
1778
1779 /* We either got finished input or we have to wait for another call. */
1780 if (isspace(ch) || !ch) {
1781 parser->buffer[parser->idx] = 0;
1782 parser->cont = false;
1783 } else if (parser->idx < parser->size - 1) {
1784 parser->cont = true;
1785 parser->buffer[parser->idx++] = ch;
1786 /* Make sure the parsed string always terminates with '\0'. */
1787 parser->buffer[parser->idx] = 0;
1788 } else {
1789 ret = -EINVAL;
1790 goto out;
1791 }
1792
1793 *ppos += read;
1794 ret = read;
1795
1796out:
1797 return ret;
1798}
1799
1800/* TODO add a seq_buf_to_buffer() */
1801static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1802{
1803 int len;
1804
1805 if (trace_seq_used(s) <= s->readpos)
1806 return -EBUSY;
1807
1808 len = trace_seq_used(s) - s->readpos;
1809 if (cnt > len)
1810 cnt = len;
1811 memcpy(buf, s->buffer + s->readpos, cnt);
1812
1813 s->readpos += cnt;
1814 return cnt;
1815}
1816
1817unsigned long __read_mostly tracing_thresh;
1818
1819#ifdef CONFIG_TRACER_MAX_TRACE
1820static const struct file_operations tracing_max_lat_fops;
1821
1822#ifdef LATENCY_FS_NOTIFY
1823
1824static struct workqueue_struct *fsnotify_wq;
1825
1826static void latency_fsnotify_workfn(struct work_struct *work)
1827{
1828 struct trace_array *tr = container_of(work, struct trace_array,
1829 fsnotify_work);
1830 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1831}
1832
1833static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1834{
1835 struct trace_array *tr = container_of(iwork, struct trace_array,
1836 fsnotify_irqwork);
1837 queue_work(fsnotify_wq, &tr->fsnotify_work);
1838}
1839
1840static void trace_create_maxlat_file(struct trace_array *tr,
1841 struct dentry *d_tracer)
1842{
1843 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1844 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1845 tr->d_max_latency = trace_create_file("tracing_max_latency",
1846 TRACE_MODE_WRITE,
1847 d_tracer, tr,
1848 &tracing_max_lat_fops);
1849}
1850
1851__init static int latency_fsnotify_init(void)
1852{
1853 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1854 WQ_UNBOUND | WQ_HIGHPRI, 0);
1855 if (!fsnotify_wq) {
1856 pr_err("Unable to allocate tr_max_lat_wq\n");
1857 return -ENOMEM;
1858 }
1859 return 0;
1860}
1861
1862late_initcall_sync(latency_fsnotify_init);
1863
1864void latency_fsnotify(struct trace_array *tr)
1865{
1866 if (!fsnotify_wq)
1867 return;
1868 /*
1869 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1870 * possible that we are called from __schedule() or do_idle(), which
1871 * could cause a deadlock.
1872 */
1873 irq_work_queue(&tr->fsnotify_irqwork);
1874}
1875
1876#else /* !LATENCY_FS_NOTIFY */
1877
1878#define trace_create_maxlat_file(tr, d_tracer) \
1879 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1880 d_tracer, tr, &tracing_max_lat_fops)
1881
1882#endif
1883
1884/*
1885 * Copy the new maximum trace into the separate maximum-trace
1886 * structure. (this way the maximum trace is permanently saved,
1887 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1888 */
1889static void
1890__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1891{
1892 struct array_buffer *trace_buf = &tr->array_buffer;
1893 struct array_buffer *max_buf = &tr->max_buffer;
1894 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1895 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1896
1897 max_buf->cpu = cpu;
1898 max_buf->time_start = data->preempt_timestamp;
1899
1900 max_data->saved_latency = tr->max_latency;
1901 max_data->critical_start = data->critical_start;
1902 max_data->critical_end = data->critical_end;
1903
1904 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1905 max_data->pid = tsk->pid;
1906 /*
1907 * If tsk == current, then use current_uid(), as that does not use
1908 * RCU. The irq tracer can be called out of RCU scope.
1909 */
1910 if (tsk == current)
1911 max_data->uid = current_uid();
1912 else
1913 max_data->uid = task_uid(tsk);
1914
1915 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1916 max_data->policy = tsk->policy;
1917 max_data->rt_priority = tsk->rt_priority;
1918
1919 /* record this tasks comm */
1920 tracing_record_cmdline(tsk);
1921 latency_fsnotify(tr);
1922}
1923
1924/**
1925 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1926 * @tr: tracer
1927 * @tsk: the task with the latency
1928 * @cpu: The cpu that initiated the trace.
1929 * @cond_data: User data associated with a conditional snapshot
1930 *
1931 * Flip the buffers between the @tr and the max_tr and record information
1932 * about which task was the cause of this latency.
1933 */
1934void
1935update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1936 void *cond_data)
1937{
1938 if (tr->stop_count)
1939 return;
1940
1941 WARN_ON_ONCE(!irqs_disabled());
1942
1943 if (!tr->allocated_snapshot) {
1944 /* Only the nop tracer should hit this when disabling */
1945 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1946 return;
1947 }
1948
1949 arch_spin_lock(&tr->max_lock);
1950
1951 /* Inherit the recordable setting from array_buffer */
1952 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1953 ring_buffer_record_on(tr->max_buffer.buffer);
1954 else
1955 ring_buffer_record_off(tr->max_buffer.buffer);
1956
1957#ifdef CONFIG_TRACER_SNAPSHOT
1958 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1959 arch_spin_unlock(&tr->max_lock);
1960 return;
1961 }
1962#endif
1963 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1964
1965 __update_max_tr(tr, tsk, cpu);
1966
1967 arch_spin_unlock(&tr->max_lock);
1968
1969 /* Any waiters on the old snapshot buffer need to wake up */
1970 ring_buffer_wake_waiters(tr->array_buffer.buffer, RING_BUFFER_ALL_CPUS);
1971}
1972
1973/**
1974 * update_max_tr_single - only copy one trace over, and reset the rest
1975 * @tr: tracer
1976 * @tsk: task with the latency
1977 * @cpu: the cpu of the buffer to copy.
1978 *
1979 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1980 */
1981void
1982update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1983{
1984 int ret;
1985
1986 if (tr->stop_count)
1987 return;
1988
1989 WARN_ON_ONCE(!irqs_disabled());
1990 if (!tr->allocated_snapshot) {
1991 /* Only the nop tracer should hit this when disabling */
1992 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1993 return;
1994 }
1995
1996 arch_spin_lock(&tr->max_lock);
1997
1998 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1999
2000 if (ret == -EBUSY) {
2001 /*
2002 * We failed to swap the buffer due to a commit taking
2003 * place on this CPU. We fail to record, but we reset
2004 * the max trace buffer (no one writes directly to it)
2005 * and flag that it failed.
2006 * Another reason is resize is in progress.
2007 */
2008 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
2009 "Failed to swap buffers due to commit or resize in progress\n");
2010 }
2011
2012 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
2013
2014 __update_max_tr(tr, tsk, cpu);
2015 arch_spin_unlock(&tr->max_lock);
2016}
2017
2018#endif /* CONFIG_TRACER_MAX_TRACE */
2019
2020struct pipe_wait {
2021 struct trace_iterator *iter;
2022 int wait_index;
2023};
2024
2025static bool wait_pipe_cond(void *data)
2026{
2027 struct pipe_wait *pwait = data;
2028 struct trace_iterator *iter = pwait->iter;
2029
2030 if (atomic_read_acquire(&iter->wait_index) != pwait->wait_index)
2031 return true;
2032
2033 return iter->closed;
2034}
2035
2036static int wait_on_pipe(struct trace_iterator *iter, int full)
2037{
2038 struct pipe_wait pwait;
2039 int ret;
2040
2041 /* Iterators are static, they should be filled or empty */
2042 if (trace_buffer_iter(iter, iter->cpu_file))
2043 return 0;
2044
2045 pwait.wait_index = atomic_read_acquire(&iter->wait_index);
2046 pwait.iter = iter;
2047
2048 ret = ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file, full,
2049 wait_pipe_cond, &pwait);
2050
2051#ifdef CONFIG_TRACER_MAX_TRACE
2052 /*
2053 * Make sure this is still the snapshot buffer, as if a snapshot were
2054 * to happen, this would now be the main buffer.
2055 */
2056 if (iter->snapshot)
2057 iter->array_buffer = &iter->tr->max_buffer;
2058#endif
2059 return ret;
2060}
2061
2062#ifdef CONFIG_FTRACE_STARTUP_TEST
2063static bool selftests_can_run;
2064
2065struct trace_selftests {
2066 struct list_head list;
2067 struct tracer *type;
2068};
2069
2070static LIST_HEAD(postponed_selftests);
2071
2072static int save_selftest(struct tracer *type)
2073{
2074 struct trace_selftests *selftest;
2075
2076 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
2077 if (!selftest)
2078 return -ENOMEM;
2079
2080 selftest->type = type;
2081 list_add(&selftest->list, &postponed_selftests);
2082 return 0;
2083}
2084
2085static int run_tracer_selftest(struct tracer *type)
2086{
2087 struct trace_array *tr = &global_trace;
2088 struct tracer *saved_tracer = tr->current_trace;
2089 int ret;
2090
2091 if (!type->selftest || tracing_selftest_disabled)
2092 return 0;
2093
2094 /*
2095 * If a tracer registers early in boot up (before scheduling is
2096 * initialized and such), then do not run its selftests yet.
2097 * Instead, run it a little later in the boot process.
2098 */
2099 if (!selftests_can_run)
2100 return save_selftest(type);
2101
2102 if (!tracing_is_on()) {
2103 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2104 type->name);
2105 return 0;
2106 }
2107
2108 /*
2109 * Run a selftest on this tracer.
2110 * Here we reset the trace buffer, and set the current
2111 * tracer to be this tracer. The tracer can then run some
2112 * internal tracing to verify that everything is in order.
2113 * If we fail, we do not register this tracer.
2114 */
2115 tracing_reset_online_cpus(&tr->array_buffer);
2116
2117 tr->current_trace = type;
2118
2119#ifdef CONFIG_TRACER_MAX_TRACE
2120 if (type->use_max_tr) {
2121 /* If we expanded the buffers, make sure the max is expanded too */
2122 if (tr->ring_buffer_expanded)
2123 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2124 RING_BUFFER_ALL_CPUS);
2125 tr->allocated_snapshot = true;
2126 }
2127#endif
2128
2129 /* the test is responsible for initializing and enabling */
2130 pr_info("Testing tracer %s: ", type->name);
2131 ret = type->selftest(type, tr);
2132 /* the test is responsible for resetting too */
2133 tr->current_trace = saved_tracer;
2134 if (ret) {
2135 printk(KERN_CONT "FAILED!\n");
2136 /* Add the warning after printing 'FAILED' */
2137 WARN_ON(1);
2138 return -1;
2139 }
2140 /* Only reset on passing, to avoid touching corrupted buffers */
2141 tracing_reset_online_cpus(&tr->array_buffer);
2142
2143#ifdef CONFIG_TRACER_MAX_TRACE
2144 if (type->use_max_tr) {
2145 tr->allocated_snapshot = false;
2146
2147 /* Shrink the max buffer again */
2148 if (tr->ring_buffer_expanded)
2149 ring_buffer_resize(tr->max_buffer.buffer, 1,
2150 RING_BUFFER_ALL_CPUS);
2151 }
2152#endif
2153
2154 printk(KERN_CONT "PASSED\n");
2155 return 0;
2156}
2157
2158static int do_run_tracer_selftest(struct tracer *type)
2159{
2160 int ret;
2161
2162 /*
2163 * Tests can take a long time, especially if they are run one after the
2164 * other, as does happen during bootup when all the tracers are
2165 * registered. This could cause the soft lockup watchdog to trigger.
2166 */
2167 cond_resched();
2168
2169 tracing_selftest_running = true;
2170 ret = run_tracer_selftest(type);
2171 tracing_selftest_running = false;
2172
2173 return ret;
2174}
2175
2176static __init int init_trace_selftests(void)
2177{
2178 struct trace_selftests *p, *n;
2179 struct tracer *t, **last;
2180 int ret;
2181
2182 selftests_can_run = true;
2183
2184 mutex_lock(&trace_types_lock);
2185
2186 if (list_empty(&postponed_selftests))
2187 goto out;
2188
2189 pr_info("Running postponed tracer tests:\n");
2190
2191 tracing_selftest_running = true;
2192 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2193 /* This loop can take minutes when sanitizers are enabled, so
2194 * lets make sure we allow RCU processing.
2195 */
2196 cond_resched();
2197 ret = run_tracer_selftest(p->type);
2198 /* If the test fails, then warn and remove from available_tracers */
2199 if (ret < 0) {
2200 WARN(1, "tracer: %s failed selftest, disabling\n",
2201 p->type->name);
2202 last = &trace_types;
2203 for (t = trace_types; t; t = t->next) {
2204 if (t == p->type) {
2205 *last = t->next;
2206 break;
2207 }
2208 last = &t->next;
2209 }
2210 }
2211 list_del(&p->list);
2212 kfree(p);
2213 }
2214 tracing_selftest_running = false;
2215
2216 out:
2217 mutex_unlock(&trace_types_lock);
2218
2219 return 0;
2220}
2221core_initcall(init_trace_selftests);
2222#else
2223static inline int run_tracer_selftest(struct tracer *type)
2224{
2225 return 0;
2226}
2227static inline int do_run_tracer_selftest(struct tracer *type)
2228{
2229 return 0;
2230}
2231#endif /* CONFIG_FTRACE_STARTUP_TEST */
2232
2233static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2234
2235static void __init apply_trace_boot_options(void);
2236
2237/**
2238 * register_tracer - register a tracer with the ftrace system.
2239 * @type: the plugin for the tracer
2240 *
2241 * Register a new plugin tracer.
2242 */
2243int __init register_tracer(struct tracer *type)
2244{
2245 struct tracer *t;
2246 int ret = 0;
2247
2248 if (!type->name) {
2249 pr_info("Tracer must have a name\n");
2250 return -1;
2251 }
2252
2253 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2254 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2255 return -1;
2256 }
2257
2258 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2259 pr_warn("Can not register tracer %s due to lockdown\n",
2260 type->name);
2261 return -EPERM;
2262 }
2263
2264 mutex_lock(&trace_types_lock);
2265
2266 for (t = trace_types; t; t = t->next) {
2267 if (strcmp(type->name, t->name) == 0) {
2268 /* already found */
2269 pr_info("Tracer %s already registered\n",
2270 type->name);
2271 ret = -1;
2272 goto out;
2273 }
2274 }
2275
2276 if (!type->set_flag)
2277 type->set_flag = &dummy_set_flag;
2278 if (!type->flags) {
2279 /*allocate a dummy tracer_flags*/
2280 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2281 if (!type->flags) {
2282 ret = -ENOMEM;
2283 goto out;
2284 }
2285 type->flags->val = 0;
2286 type->flags->opts = dummy_tracer_opt;
2287 } else
2288 if (!type->flags->opts)
2289 type->flags->opts = dummy_tracer_opt;
2290
2291 /* store the tracer for __set_tracer_option */
2292 type->flags->trace = type;
2293
2294 ret = do_run_tracer_selftest(type);
2295 if (ret < 0)
2296 goto out;
2297
2298 type->next = trace_types;
2299 trace_types = type;
2300 add_tracer_options(&global_trace, type);
2301
2302 out:
2303 mutex_unlock(&trace_types_lock);
2304
2305 if (ret || !default_bootup_tracer)
2306 goto out_unlock;
2307
2308 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2309 goto out_unlock;
2310
2311 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2312 /* Do we want this tracer to start on bootup? */
2313 tracing_set_tracer(&global_trace, type->name);
2314 default_bootup_tracer = NULL;
2315
2316 apply_trace_boot_options();
2317
2318 /* disable other selftests, since this will break it. */
2319 disable_tracing_selftest("running a tracer");
2320
2321 out_unlock:
2322 return ret;
2323}
2324
2325static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2326{
2327 struct trace_buffer *buffer = buf->buffer;
2328
2329 if (!buffer)
2330 return;
2331
2332 ring_buffer_record_disable(buffer);
2333
2334 /* Make sure all commits have finished */
2335 synchronize_rcu();
2336 ring_buffer_reset_cpu(buffer, cpu);
2337
2338 ring_buffer_record_enable(buffer);
2339}
2340
2341void tracing_reset_online_cpus(struct array_buffer *buf)
2342{
2343 struct trace_buffer *buffer = buf->buffer;
2344
2345 if (!buffer)
2346 return;
2347
2348 ring_buffer_record_disable(buffer);
2349
2350 /* Make sure all commits have finished */
2351 synchronize_rcu();
2352
2353 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2354
2355 ring_buffer_reset_online_cpus(buffer);
2356
2357 ring_buffer_record_enable(buffer);
2358}
2359
2360/* Must have trace_types_lock held */
2361void tracing_reset_all_online_cpus_unlocked(void)
2362{
2363 struct trace_array *tr;
2364
2365 lockdep_assert_held(&trace_types_lock);
2366
2367 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2368 if (!tr->clear_trace)
2369 continue;
2370 tr->clear_trace = false;
2371 tracing_reset_online_cpus(&tr->array_buffer);
2372#ifdef CONFIG_TRACER_MAX_TRACE
2373 tracing_reset_online_cpus(&tr->max_buffer);
2374#endif
2375 }
2376}
2377
2378void tracing_reset_all_online_cpus(void)
2379{
2380 mutex_lock(&trace_types_lock);
2381 tracing_reset_all_online_cpus_unlocked();
2382 mutex_unlock(&trace_types_lock);
2383}
2384
2385int is_tracing_stopped(void)
2386{
2387 return global_trace.stop_count;
2388}
2389
2390static void tracing_start_tr(struct trace_array *tr)
2391{
2392 struct trace_buffer *buffer;
2393 unsigned long flags;
2394
2395 if (tracing_disabled)
2396 return;
2397
2398 raw_spin_lock_irqsave(&tr->start_lock, flags);
2399 if (--tr->stop_count) {
2400 if (WARN_ON_ONCE(tr->stop_count < 0)) {
2401 /* Someone screwed up their debugging */
2402 tr->stop_count = 0;
2403 }
2404 goto out;
2405 }
2406
2407 /* Prevent the buffers from switching */
2408 arch_spin_lock(&tr->max_lock);
2409
2410 buffer = tr->array_buffer.buffer;
2411 if (buffer)
2412 ring_buffer_record_enable(buffer);
2413
2414#ifdef CONFIG_TRACER_MAX_TRACE
2415 buffer = tr->max_buffer.buffer;
2416 if (buffer)
2417 ring_buffer_record_enable(buffer);
2418#endif
2419
2420 arch_spin_unlock(&tr->max_lock);
2421
2422 out:
2423 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2424}
2425
2426/**
2427 * tracing_start - quick start of the tracer
2428 *
2429 * If tracing is enabled but was stopped by tracing_stop,
2430 * this will start the tracer back up.
2431 */
2432void tracing_start(void)
2433
2434{
2435 return tracing_start_tr(&global_trace);
2436}
2437
2438static void tracing_stop_tr(struct trace_array *tr)
2439{
2440 struct trace_buffer *buffer;
2441 unsigned long flags;
2442
2443 raw_spin_lock_irqsave(&tr->start_lock, flags);
2444 if (tr->stop_count++)
2445 goto out;
2446
2447 /* Prevent the buffers from switching */
2448 arch_spin_lock(&tr->max_lock);
2449
2450 buffer = tr->array_buffer.buffer;
2451 if (buffer)
2452 ring_buffer_record_disable(buffer);
2453
2454#ifdef CONFIG_TRACER_MAX_TRACE
2455 buffer = tr->max_buffer.buffer;
2456 if (buffer)
2457 ring_buffer_record_disable(buffer);
2458#endif
2459
2460 arch_spin_unlock(&tr->max_lock);
2461
2462 out:
2463 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2464}
2465
2466/**
2467 * tracing_stop - quick stop of the tracer
2468 *
2469 * Light weight way to stop tracing. Use in conjunction with
2470 * tracing_start.
2471 */
2472void tracing_stop(void)
2473{
2474 return tracing_stop_tr(&global_trace);
2475}
2476
2477/*
2478 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2479 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2480 * simplifies those functions and keeps them in sync.
2481 */
2482enum print_line_t trace_handle_return(struct trace_seq *s)
2483{
2484 return trace_seq_has_overflowed(s) ?
2485 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2486}
2487EXPORT_SYMBOL_GPL(trace_handle_return);
2488
2489static unsigned short migration_disable_value(void)
2490{
2491#if defined(CONFIG_SMP)
2492 return current->migration_disabled;
2493#else
2494 return 0;
2495#endif
2496}
2497
2498unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2499{
2500 unsigned int trace_flags = irqs_status;
2501 unsigned int pc;
2502
2503 pc = preempt_count();
2504
2505 if (pc & NMI_MASK)
2506 trace_flags |= TRACE_FLAG_NMI;
2507 if (pc & HARDIRQ_MASK)
2508 trace_flags |= TRACE_FLAG_HARDIRQ;
2509 if (in_serving_softirq())
2510 trace_flags |= TRACE_FLAG_SOFTIRQ;
2511 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2512 trace_flags |= TRACE_FLAG_BH_OFF;
2513
2514 if (tif_need_resched())
2515 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2516 if (test_preempt_need_resched())
2517 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2518 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2519 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2520}
2521
2522struct ring_buffer_event *
2523trace_buffer_lock_reserve(struct trace_buffer *buffer,
2524 int type,
2525 unsigned long len,
2526 unsigned int trace_ctx)
2527{
2528 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2529}
2530
2531DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2532DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2533static int trace_buffered_event_ref;
2534
2535/**
2536 * trace_buffered_event_enable - enable buffering events
2537 *
2538 * When events are being filtered, it is quicker to use a temporary
2539 * buffer to write the event data into if there's a likely chance
2540 * that it will not be committed. The discard of the ring buffer
2541 * is not as fast as committing, and is much slower than copying
2542 * a commit.
2543 *
2544 * When an event is to be filtered, allocate per cpu buffers to
2545 * write the event data into, and if the event is filtered and discarded
2546 * it is simply dropped, otherwise, the entire data is to be committed
2547 * in one shot.
2548 */
2549void trace_buffered_event_enable(void)
2550{
2551 struct ring_buffer_event *event;
2552 struct page *page;
2553 int cpu;
2554
2555 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2556
2557 if (trace_buffered_event_ref++)
2558 return;
2559
2560 for_each_tracing_cpu(cpu) {
2561 page = alloc_pages_node(cpu_to_node(cpu),
2562 GFP_KERNEL | __GFP_NORETRY, 0);
2563 /* This is just an optimization and can handle failures */
2564 if (!page) {
2565 pr_err("Failed to allocate event buffer\n");
2566 break;
2567 }
2568
2569 event = page_address(page);
2570 memset(event, 0, sizeof(*event));
2571
2572 per_cpu(trace_buffered_event, cpu) = event;
2573
2574 preempt_disable();
2575 if (cpu == smp_processor_id() &&
2576 __this_cpu_read(trace_buffered_event) !=
2577 per_cpu(trace_buffered_event, cpu))
2578 WARN_ON_ONCE(1);
2579 preempt_enable();
2580 }
2581}
2582
2583static void enable_trace_buffered_event(void *data)
2584{
2585 /* Probably not needed, but do it anyway */
2586 smp_rmb();
2587 this_cpu_dec(trace_buffered_event_cnt);
2588}
2589
2590static void disable_trace_buffered_event(void *data)
2591{
2592 this_cpu_inc(trace_buffered_event_cnt);
2593}
2594
2595/**
2596 * trace_buffered_event_disable - disable buffering events
2597 *
2598 * When a filter is removed, it is faster to not use the buffered
2599 * events, and to commit directly into the ring buffer. Free up
2600 * the temp buffers when there are no more users. This requires
2601 * special synchronization with current events.
2602 */
2603void trace_buffered_event_disable(void)
2604{
2605 int cpu;
2606
2607 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2608
2609 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2610 return;
2611
2612 if (--trace_buffered_event_ref)
2613 return;
2614
2615 /* For each CPU, set the buffer as used. */
2616 on_each_cpu_mask(tracing_buffer_mask, disable_trace_buffered_event,
2617 NULL, true);
2618
2619 /* Wait for all current users to finish */
2620 synchronize_rcu();
2621
2622 for_each_tracing_cpu(cpu) {
2623 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2624 per_cpu(trace_buffered_event, cpu) = NULL;
2625 }
2626
2627 /*
2628 * Wait for all CPUs that potentially started checking if they can use
2629 * their event buffer only after the previous synchronize_rcu() call and
2630 * they still read a valid pointer from trace_buffered_event. It must be
2631 * ensured they don't see cleared trace_buffered_event_cnt else they
2632 * could wrongly decide to use the pointed-to buffer which is now freed.
2633 */
2634 synchronize_rcu();
2635
2636 /* For each CPU, relinquish the buffer */
2637 on_each_cpu_mask(tracing_buffer_mask, enable_trace_buffered_event, NULL,
2638 true);
2639}
2640
2641static struct trace_buffer *temp_buffer;
2642
2643struct ring_buffer_event *
2644trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2645 struct trace_event_file *trace_file,
2646 int type, unsigned long len,
2647 unsigned int trace_ctx)
2648{
2649 struct ring_buffer_event *entry;
2650 struct trace_array *tr = trace_file->tr;
2651 int val;
2652
2653 *current_rb = tr->array_buffer.buffer;
2654
2655 if (!tr->no_filter_buffering_ref &&
2656 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2657 preempt_disable_notrace();
2658 /*
2659 * Filtering is on, so try to use the per cpu buffer first.
2660 * This buffer will simulate a ring_buffer_event,
2661 * where the type_len is zero and the array[0] will
2662 * hold the full length.
2663 * (see include/linux/ring-buffer.h for details on
2664 * how the ring_buffer_event is structured).
2665 *
2666 * Using a temp buffer during filtering and copying it
2667 * on a matched filter is quicker than writing directly
2668 * into the ring buffer and then discarding it when
2669 * it doesn't match. That is because the discard
2670 * requires several atomic operations to get right.
2671 * Copying on match and doing nothing on a failed match
2672 * is still quicker than no copy on match, but having
2673 * to discard out of the ring buffer on a failed match.
2674 */
2675 if ((entry = __this_cpu_read(trace_buffered_event))) {
2676 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2677
2678 val = this_cpu_inc_return(trace_buffered_event_cnt);
2679
2680 /*
2681 * Preemption is disabled, but interrupts and NMIs
2682 * can still come in now. If that happens after
2683 * the above increment, then it will have to go
2684 * back to the old method of allocating the event
2685 * on the ring buffer, and if the filter fails, it
2686 * will have to call ring_buffer_discard_commit()
2687 * to remove it.
2688 *
2689 * Need to also check the unlikely case that the
2690 * length is bigger than the temp buffer size.
2691 * If that happens, then the reserve is pretty much
2692 * guaranteed to fail, as the ring buffer currently
2693 * only allows events less than a page. But that may
2694 * change in the future, so let the ring buffer reserve
2695 * handle the failure in that case.
2696 */
2697 if (val == 1 && likely(len <= max_len)) {
2698 trace_event_setup(entry, type, trace_ctx);
2699 entry->array[0] = len;
2700 /* Return with preemption disabled */
2701 return entry;
2702 }
2703 this_cpu_dec(trace_buffered_event_cnt);
2704 }
2705 /* __trace_buffer_lock_reserve() disables preemption */
2706 preempt_enable_notrace();
2707 }
2708
2709 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2710 trace_ctx);
2711 /*
2712 * If tracing is off, but we have triggers enabled
2713 * we still need to look at the event data. Use the temp_buffer
2714 * to store the trace event for the trigger to use. It's recursive
2715 * safe and will not be recorded anywhere.
2716 */
2717 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2718 *current_rb = temp_buffer;
2719 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2720 trace_ctx);
2721 }
2722 return entry;
2723}
2724EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2725
2726static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2727static DEFINE_MUTEX(tracepoint_printk_mutex);
2728
2729static void output_printk(struct trace_event_buffer *fbuffer)
2730{
2731 struct trace_event_call *event_call;
2732 struct trace_event_file *file;
2733 struct trace_event *event;
2734 unsigned long flags;
2735 struct trace_iterator *iter = tracepoint_print_iter;
2736
2737 /* We should never get here if iter is NULL */
2738 if (WARN_ON_ONCE(!iter))
2739 return;
2740
2741 event_call = fbuffer->trace_file->event_call;
2742 if (!event_call || !event_call->event.funcs ||
2743 !event_call->event.funcs->trace)
2744 return;
2745
2746 file = fbuffer->trace_file;
2747 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2748 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2749 !filter_match_preds(file->filter, fbuffer->entry)))
2750 return;
2751
2752 event = &fbuffer->trace_file->event_call->event;
2753
2754 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2755 trace_seq_init(&iter->seq);
2756 iter->ent = fbuffer->entry;
2757 event_call->event.funcs->trace(iter, 0, event);
2758 trace_seq_putc(&iter->seq, 0);
2759 printk("%s", iter->seq.buffer);
2760
2761 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2762}
2763
2764int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2765 void *buffer, size_t *lenp,
2766 loff_t *ppos)
2767{
2768 int save_tracepoint_printk;
2769 int ret;
2770
2771 mutex_lock(&tracepoint_printk_mutex);
2772 save_tracepoint_printk = tracepoint_printk;
2773
2774 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2775
2776 /*
2777 * This will force exiting early, as tracepoint_printk
2778 * is always zero when tracepoint_printk_iter is not allocated
2779 */
2780 if (!tracepoint_print_iter)
2781 tracepoint_printk = 0;
2782
2783 if (save_tracepoint_printk == tracepoint_printk)
2784 goto out;
2785
2786 if (tracepoint_printk)
2787 static_key_enable(&tracepoint_printk_key.key);
2788 else
2789 static_key_disable(&tracepoint_printk_key.key);
2790
2791 out:
2792 mutex_unlock(&tracepoint_printk_mutex);
2793
2794 return ret;
2795}
2796
2797void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2798{
2799 enum event_trigger_type tt = ETT_NONE;
2800 struct trace_event_file *file = fbuffer->trace_file;
2801
2802 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2803 fbuffer->entry, &tt))
2804 goto discard;
2805
2806 if (static_key_false(&tracepoint_printk_key.key))
2807 output_printk(fbuffer);
2808
2809 if (static_branch_unlikely(&trace_event_exports_enabled))
2810 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2811
2812 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2813 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
2814
2815discard:
2816 if (tt)
2817 event_triggers_post_call(file, tt);
2818
2819}
2820EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2821
2822/*
2823 * Skip 3:
2824 *
2825 * trace_buffer_unlock_commit_regs()
2826 * trace_event_buffer_commit()
2827 * trace_event_raw_event_xxx()
2828 */
2829# define STACK_SKIP 3
2830
2831void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2832 struct trace_buffer *buffer,
2833 struct ring_buffer_event *event,
2834 unsigned int trace_ctx,
2835 struct pt_regs *regs)
2836{
2837 __buffer_unlock_commit(buffer, event);
2838
2839 /*
2840 * If regs is not set, then skip the necessary functions.
2841 * Note, we can still get here via blktrace, wakeup tracer
2842 * and mmiotrace, but that's ok if they lose a function or
2843 * two. They are not that meaningful.
2844 */
2845 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
2846 ftrace_trace_userstack(tr, buffer, trace_ctx);
2847}
2848
2849/*
2850 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2851 */
2852void
2853trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2854 struct ring_buffer_event *event)
2855{
2856 __buffer_unlock_commit(buffer, event);
2857}
2858
2859void
2860trace_function(struct trace_array *tr, unsigned long ip, unsigned long
2861 parent_ip, unsigned int trace_ctx)
2862{
2863 struct trace_event_call *call = &event_function;
2864 struct trace_buffer *buffer = tr->array_buffer.buffer;
2865 struct ring_buffer_event *event;
2866 struct ftrace_entry *entry;
2867
2868 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2869 trace_ctx);
2870 if (!event)
2871 return;
2872 entry = ring_buffer_event_data(event);
2873 entry->ip = ip;
2874 entry->parent_ip = parent_ip;
2875
2876 if (!call_filter_check_discard(call, entry, buffer, event)) {
2877 if (static_branch_unlikely(&trace_function_exports_enabled))
2878 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
2879 __buffer_unlock_commit(buffer, event);
2880 }
2881}
2882
2883#ifdef CONFIG_STACKTRACE
2884
2885/* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2886#define FTRACE_KSTACK_NESTING 4
2887
2888#define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
2889
2890struct ftrace_stack {
2891 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2892};
2893
2894
2895struct ftrace_stacks {
2896 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
2897};
2898
2899static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
2900static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2901
2902static void __ftrace_trace_stack(struct trace_buffer *buffer,
2903 unsigned int trace_ctx,
2904 int skip, struct pt_regs *regs)
2905{
2906 struct trace_event_call *call = &event_kernel_stack;
2907 struct ring_buffer_event *event;
2908 unsigned int size, nr_entries;
2909 struct ftrace_stack *fstack;
2910 struct stack_entry *entry;
2911 int stackidx;
2912
2913 /*
2914 * Add one, for this function and the call to save_stack_trace()
2915 * If regs is set, then these functions will not be in the way.
2916 */
2917#ifndef CONFIG_UNWINDER_ORC
2918 if (!regs)
2919 skip++;
2920#endif
2921
2922 preempt_disable_notrace();
2923
2924 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
2925
2926 /* This should never happen. If it does, yell once and skip */
2927 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
2928 goto out;
2929
2930 /*
2931 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
2932 * interrupt will either see the value pre increment or post
2933 * increment. If the interrupt happens pre increment it will have
2934 * restored the counter when it returns. We just need a barrier to
2935 * keep gcc from moving things around.
2936 */
2937 barrier();
2938
2939 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
2940 size = ARRAY_SIZE(fstack->calls);
2941
2942 if (regs) {
2943 nr_entries = stack_trace_save_regs(regs, fstack->calls,
2944 size, skip);
2945 } else {
2946 nr_entries = stack_trace_save(fstack->calls, size, skip);
2947 }
2948
2949 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2950 struct_size(entry, caller, nr_entries),
2951 trace_ctx);
2952 if (!event)
2953 goto out;
2954 entry = ring_buffer_event_data(event);
2955
2956 entry->size = nr_entries;
2957 memcpy(&entry->caller, fstack->calls,
2958 flex_array_size(entry, caller, nr_entries));
2959
2960 if (!call_filter_check_discard(call, entry, buffer, event))
2961 __buffer_unlock_commit(buffer, event);
2962
2963 out:
2964 /* Again, don't let gcc optimize things here */
2965 barrier();
2966 __this_cpu_dec(ftrace_stack_reserve);
2967 preempt_enable_notrace();
2968
2969}
2970
2971static inline void ftrace_trace_stack(struct trace_array *tr,
2972 struct trace_buffer *buffer,
2973 unsigned int trace_ctx,
2974 int skip, struct pt_regs *regs)
2975{
2976 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
2977 return;
2978
2979 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
2980}
2981
2982void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
2983 int skip)
2984{
2985 struct trace_buffer *buffer = tr->array_buffer.buffer;
2986
2987 if (rcu_is_watching()) {
2988 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
2989 return;
2990 }
2991
2992 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
2993 return;
2994
2995 /*
2996 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
2997 * but if the above rcu_is_watching() failed, then the NMI
2998 * triggered someplace critical, and ct_irq_enter() should
2999 * not be called from NMI.
3000 */
3001 if (unlikely(in_nmi()))
3002 return;
3003
3004 ct_irq_enter_irqson();
3005 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3006 ct_irq_exit_irqson();
3007}
3008
3009/**
3010 * trace_dump_stack - record a stack back trace in the trace buffer
3011 * @skip: Number of functions to skip (helper handlers)
3012 */
3013void trace_dump_stack(int skip)
3014{
3015 if (tracing_disabled || tracing_selftest_running)
3016 return;
3017
3018#ifndef CONFIG_UNWINDER_ORC
3019 /* Skip 1 to skip this function. */
3020 skip++;
3021#endif
3022 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3023 tracing_gen_ctx(), skip, NULL);
3024}
3025EXPORT_SYMBOL_GPL(trace_dump_stack);
3026
3027#ifdef CONFIG_USER_STACKTRACE_SUPPORT
3028static DEFINE_PER_CPU(int, user_stack_count);
3029
3030static void
3031ftrace_trace_userstack(struct trace_array *tr,
3032 struct trace_buffer *buffer, unsigned int trace_ctx)
3033{
3034 struct trace_event_call *call = &event_user_stack;
3035 struct ring_buffer_event *event;
3036 struct userstack_entry *entry;
3037
3038 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3039 return;
3040
3041 /*
3042 * NMIs can not handle page faults, even with fix ups.
3043 * The save user stack can (and often does) fault.
3044 */
3045 if (unlikely(in_nmi()))
3046 return;
3047
3048 /*
3049 * prevent recursion, since the user stack tracing may
3050 * trigger other kernel events.
3051 */
3052 preempt_disable();
3053 if (__this_cpu_read(user_stack_count))
3054 goto out;
3055
3056 __this_cpu_inc(user_stack_count);
3057
3058 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3059 sizeof(*entry), trace_ctx);
3060 if (!event)
3061 goto out_drop_count;
3062 entry = ring_buffer_event_data(event);
3063
3064 entry->tgid = current->tgid;
3065 memset(&entry->caller, 0, sizeof(entry->caller));
3066
3067 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3068 if (!call_filter_check_discard(call, entry, buffer, event))
3069 __buffer_unlock_commit(buffer, event);
3070
3071 out_drop_count:
3072 __this_cpu_dec(user_stack_count);
3073 out:
3074 preempt_enable();
3075}
3076#else /* CONFIG_USER_STACKTRACE_SUPPORT */
3077static void ftrace_trace_userstack(struct trace_array *tr,
3078 struct trace_buffer *buffer,
3079 unsigned int trace_ctx)
3080{
3081}
3082#endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3083
3084#endif /* CONFIG_STACKTRACE */
3085
3086static inline void
3087func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3088 unsigned long long delta)
3089{
3090 entry->bottom_delta_ts = delta & U32_MAX;
3091 entry->top_delta_ts = (delta >> 32);
3092}
3093
3094void trace_last_func_repeats(struct trace_array *tr,
3095 struct trace_func_repeats *last_info,
3096 unsigned int trace_ctx)
3097{
3098 struct trace_buffer *buffer = tr->array_buffer.buffer;
3099 struct func_repeats_entry *entry;
3100 struct ring_buffer_event *event;
3101 u64 delta;
3102
3103 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3104 sizeof(*entry), trace_ctx);
3105 if (!event)
3106 return;
3107
3108 delta = ring_buffer_event_time_stamp(buffer, event) -
3109 last_info->ts_last_call;
3110
3111 entry = ring_buffer_event_data(event);
3112 entry->ip = last_info->ip;
3113 entry->parent_ip = last_info->parent_ip;
3114 entry->count = last_info->count;
3115 func_repeats_set_delta_ts(entry, delta);
3116
3117 __buffer_unlock_commit(buffer, event);
3118}
3119
3120/* created for use with alloc_percpu */
3121struct trace_buffer_struct {
3122 int nesting;
3123 char buffer[4][TRACE_BUF_SIZE];
3124};
3125
3126static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3127
3128/*
3129 * This allows for lockless recording. If we're nested too deeply, then
3130 * this returns NULL.
3131 */
3132static char *get_trace_buf(void)
3133{
3134 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3135
3136 if (!trace_percpu_buffer || buffer->nesting >= 4)
3137 return NULL;
3138
3139 buffer->nesting++;
3140
3141 /* Interrupts must see nesting incremented before we use the buffer */
3142 barrier();
3143 return &buffer->buffer[buffer->nesting - 1][0];
3144}
3145
3146static void put_trace_buf(void)
3147{
3148 /* Don't let the decrement of nesting leak before this */
3149 barrier();
3150 this_cpu_dec(trace_percpu_buffer->nesting);
3151}
3152
3153static int alloc_percpu_trace_buffer(void)
3154{
3155 struct trace_buffer_struct __percpu *buffers;
3156
3157 if (trace_percpu_buffer)
3158 return 0;
3159
3160 buffers = alloc_percpu(struct trace_buffer_struct);
3161 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3162 return -ENOMEM;
3163
3164 trace_percpu_buffer = buffers;
3165 return 0;
3166}
3167
3168static int buffers_allocated;
3169
3170void trace_printk_init_buffers(void)
3171{
3172 if (buffers_allocated)
3173 return;
3174
3175 if (alloc_percpu_trace_buffer())
3176 return;
3177
3178 /* trace_printk() is for debug use only. Don't use it in production. */
3179
3180 pr_warn("\n");
3181 pr_warn("**********************************************************\n");
3182 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3183 pr_warn("** **\n");
3184 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3185 pr_warn("** **\n");
3186 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3187 pr_warn("** unsafe for production use. **\n");
3188 pr_warn("** **\n");
3189 pr_warn("** If you see this message and you are not debugging **\n");
3190 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3191 pr_warn("** **\n");
3192 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3193 pr_warn("**********************************************************\n");
3194
3195 /* Expand the buffers to set size */
3196 tracing_update_buffers(&global_trace);
3197
3198 buffers_allocated = 1;
3199
3200 /*
3201 * trace_printk_init_buffers() can be called by modules.
3202 * If that happens, then we need to start cmdline recording
3203 * directly here. If the global_trace.buffer is already
3204 * allocated here, then this was called by module code.
3205 */
3206 if (global_trace.array_buffer.buffer)
3207 tracing_start_cmdline_record();
3208}
3209EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3210
3211void trace_printk_start_comm(void)
3212{
3213 /* Start tracing comms if trace printk is set */
3214 if (!buffers_allocated)
3215 return;
3216 tracing_start_cmdline_record();
3217}
3218
3219static void trace_printk_start_stop_comm(int enabled)
3220{
3221 if (!buffers_allocated)
3222 return;
3223
3224 if (enabled)
3225 tracing_start_cmdline_record();
3226 else
3227 tracing_stop_cmdline_record();
3228}
3229
3230/**
3231 * trace_vbprintk - write binary msg to tracing buffer
3232 * @ip: The address of the caller
3233 * @fmt: The string format to write to the buffer
3234 * @args: Arguments for @fmt
3235 */
3236int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3237{
3238 struct trace_event_call *call = &event_bprint;
3239 struct ring_buffer_event *event;
3240 struct trace_buffer *buffer;
3241 struct trace_array *tr = &global_trace;
3242 struct bprint_entry *entry;
3243 unsigned int trace_ctx;
3244 char *tbuffer;
3245 int len = 0, size;
3246
3247 if (unlikely(tracing_selftest_running || tracing_disabled))
3248 return 0;
3249
3250 /* Don't pollute graph traces with trace_vprintk internals */
3251 pause_graph_tracing();
3252
3253 trace_ctx = tracing_gen_ctx();
3254 preempt_disable_notrace();
3255
3256 tbuffer = get_trace_buf();
3257 if (!tbuffer) {
3258 len = 0;
3259 goto out_nobuffer;
3260 }
3261
3262 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3263
3264 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3265 goto out_put;
3266
3267 size = sizeof(*entry) + sizeof(u32) * len;
3268 buffer = tr->array_buffer.buffer;
3269 ring_buffer_nest_start(buffer);
3270 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3271 trace_ctx);
3272 if (!event)
3273 goto out;
3274 entry = ring_buffer_event_data(event);
3275 entry->ip = ip;
3276 entry->fmt = fmt;
3277
3278 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3279 if (!call_filter_check_discard(call, entry, buffer, event)) {
3280 __buffer_unlock_commit(buffer, event);
3281 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3282 }
3283
3284out:
3285 ring_buffer_nest_end(buffer);
3286out_put:
3287 put_trace_buf();
3288
3289out_nobuffer:
3290 preempt_enable_notrace();
3291 unpause_graph_tracing();
3292
3293 return len;
3294}
3295EXPORT_SYMBOL_GPL(trace_vbprintk);
3296
3297__printf(3, 0)
3298static int
3299__trace_array_vprintk(struct trace_buffer *buffer,
3300 unsigned long ip, const char *fmt, va_list args)
3301{
3302 struct trace_event_call *call = &event_print;
3303 struct ring_buffer_event *event;
3304 int len = 0, size;
3305 struct print_entry *entry;
3306 unsigned int trace_ctx;
3307 char *tbuffer;
3308
3309 if (tracing_disabled)
3310 return 0;
3311
3312 /* Don't pollute graph traces with trace_vprintk internals */
3313 pause_graph_tracing();
3314
3315 trace_ctx = tracing_gen_ctx();
3316 preempt_disable_notrace();
3317
3318
3319 tbuffer = get_trace_buf();
3320 if (!tbuffer) {
3321 len = 0;
3322 goto out_nobuffer;
3323 }
3324
3325 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3326
3327 size = sizeof(*entry) + len + 1;
3328 ring_buffer_nest_start(buffer);
3329 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3330 trace_ctx);
3331 if (!event)
3332 goto out;
3333 entry = ring_buffer_event_data(event);
3334 entry->ip = ip;
3335
3336 memcpy(&entry->buf, tbuffer, len + 1);
3337 if (!call_filter_check_discard(call, entry, buffer, event)) {
3338 __buffer_unlock_commit(buffer, event);
3339 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3340 }
3341
3342out:
3343 ring_buffer_nest_end(buffer);
3344 put_trace_buf();
3345
3346out_nobuffer:
3347 preempt_enable_notrace();
3348 unpause_graph_tracing();
3349
3350 return len;
3351}
3352
3353__printf(3, 0)
3354int trace_array_vprintk(struct trace_array *tr,
3355 unsigned long ip, const char *fmt, va_list args)
3356{
3357 if (tracing_selftest_running && tr == &global_trace)
3358 return 0;
3359
3360 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3361}
3362
3363/**
3364 * trace_array_printk - Print a message to a specific instance
3365 * @tr: The instance trace_array descriptor
3366 * @ip: The instruction pointer that this is called from.
3367 * @fmt: The format to print (printf format)
3368 *
3369 * If a subsystem sets up its own instance, they have the right to
3370 * printk strings into their tracing instance buffer using this
3371 * function. Note, this function will not write into the top level
3372 * buffer (use trace_printk() for that), as writing into the top level
3373 * buffer should only have events that can be individually disabled.
3374 * trace_printk() is only used for debugging a kernel, and should not
3375 * be ever incorporated in normal use.
3376 *
3377 * trace_array_printk() can be used, as it will not add noise to the
3378 * top level tracing buffer.
3379 *
3380 * Note, trace_array_init_printk() must be called on @tr before this
3381 * can be used.
3382 */
3383__printf(3, 0)
3384int trace_array_printk(struct trace_array *tr,
3385 unsigned long ip, const char *fmt, ...)
3386{
3387 int ret;
3388 va_list ap;
3389
3390 if (!tr)
3391 return -ENOENT;
3392
3393 /* This is only allowed for created instances */
3394 if (tr == &global_trace)
3395 return 0;
3396
3397 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3398 return 0;
3399
3400 va_start(ap, fmt);
3401 ret = trace_array_vprintk(tr, ip, fmt, ap);
3402 va_end(ap);
3403 return ret;
3404}
3405EXPORT_SYMBOL_GPL(trace_array_printk);
3406
3407/**
3408 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3409 * @tr: The trace array to initialize the buffers for
3410 *
3411 * As trace_array_printk() only writes into instances, they are OK to
3412 * have in the kernel (unlike trace_printk()). This needs to be called
3413 * before trace_array_printk() can be used on a trace_array.
3414 */
3415int trace_array_init_printk(struct trace_array *tr)
3416{
3417 if (!tr)
3418 return -ENOENT;
3419
3420 /* This is only allowed for created instances */
3421 if (tr == &global_trace)
3422 return -EINVAL;
3423
3424 return alloc_percpu_trace_buffer();
3425}
3426EXPORT_SYMBOL_GPL(trace_array_init_printk);
3427
3428__printf(3, 4)
3429int trace_array_printk_buf(struct trace_buffer *buffer,
3430 unsigned long ip, const char *fmt, ...)
3431{
3432 int ret;
3433 va_list ap;
3434
3435 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3436 return 0;
3437
3438 va_start(ap, fmt);
3439 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3440 va_end(ap);
3441 return ret;
3442}
3443
3444__printf(2, 0)
3445int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3446{
3447 return trace_array_vprintk(&global_trace, ip, fmt, args);
3448}
3449EXPORT_SYMBOL_GPL(trace_vprintk);
3450
3451static void trace_iterator_increment(struct trace_iterator *iter)
3452{
3453 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3454
3455 iter->idx++;
3456 if (buf_iter)
3457 ring_buffer_iter_advance(buf_iter);
3458}
3459
3460static struct trace_entry *
3461peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3462 unsigned long *lost_events)
3463{
3464 struct ring_buffer_event *event;
3465 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3466
3467 if (buf_iter) {
3468 event = ring_buffer_iter_peek(buf_iter, ts);
3469 if (lost_events)
3470 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3471 (unsigned long)-1 : 0;
3472 } else {
3473 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3474 lost_events);
3475 }
3476
3477 if (event) {
3478 iter->ent_size = ring_buffer_event_length(event);
3479 return ring_buffer_event_data(event);
3480 }
3481 iter->ent_size = 0;
3482 return NULL;
3483}
3484
3485static struct trace_entry *
3486__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3487 unsigned long *missing_events, u64 *ent_ts)
3488{
3489 struct trace_buffer *buffer = iter->array_buffer->buffer;
3490 struct trace_entry *ent, *next = NULL;
3491 unsigned long lost_events = 0, next_lost = 0;
3492 int cpu_file = iter->cpu_file;
3493 u64 next_ts = 0, ts;
3494 int next_cpu = -1;
3495 int next_size = 0;
3496 int cpu;
3497
3498 /*
3499 * If we are in a per_cpu trace file, don't bother by iterating over
3500 * all cpu and peek directly.
3501 */
3502 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3503 if (ring_buffer_empty_cpu(buffer, cpu_file))
3504 return NULL;
3505 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3506 if (ent_cpu)
3507 *ent_cpu = cpu_file;
3508
3509 return ent;
3510 }
3511
3512 for_each_tracing_cpu(cpu) {
3513
3514 if (ring_buffer_empty_cpu(buffer, cpu))
3515 continue;
3516
3517 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3518
3519 /*
3520 * Pick the entry with the smallest timestamp:
3521 */
3522 if (ent && (!next || ts < next_ts)) {
3523 next = ent;
3524 next_cpu = cpu;
3525 next_ts = ts;
3526 next_lost = lost_events;
3527 next_size = iter->ent_size;
3528 }
3529 }
3530
3531 iter->ent_size = next_size;
3532
3533 if (ent_cpu)
3534 *ent_cpu = next_cpu;
3535
3536 if (ent_ts)
3537 *ent_ts = next_ts;
3538
3539 if (missing_events)
3540 *missing_events = next_lost;
3541
3542 return next;
3543}
3544
3545#define STATIC_FMT_BUF_SIZE 128
3546static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3547
3548char *trace_iter_expand_format(struct trace_iterator *iter)
3549{
3550 char *tmp;
3551
3552 /*
3553 * iter->tr is NULL when used with tp_printk, which makes
3554 * this get called where it is not safe to call krealloc().
3555 */
3556 if (!iter->tr || iter->fmt == static_fmt_buf)
3557 return NULL;
3558
3559 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3560 GFP_KERNEL);
3561 if (tmp) {
3562 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3563 iter->fmt = tmp;
3564 }
3565
3566 return tmp;
3567}
3568
3569/* Returns true if the string is safe to dereference from an event */
3570static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3571 bool star, int len)
3572{
3573 unsigned long addr = (unsigned long)str;
3574 struct trace_event *trace_event;
3575 struct trace_event_call *event;
3576
3577 /* Ignore strings with no length */
3578 if (star && !len)
3579 return true;
3580
3581 /* OK if part of the event data */
3582 if ((addr >= (unsigned long)iter->ent) &&
3583 (addr < (unsigned long)iter->ent + iter->ent_size))
3584 return true;
3585
3586 /* OK if part of the temp seq buffer */
3587 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3588 (addr < (unsigned long)iter->tmp_seq.buffer + TRACE_SEQ_BUFFER_SIZE))
3589 return true;
3590
3591 /* Core rodata can not be freed */
3592 if (is_kernel_rodata(addr))
3593 return true;
3594
3595 if (trace_is_tracepoint_string(str))
3596 return true;
3597
3598 /*
3599 * Now this could be a module event, referencing core module
3600 * data, which is OK.
3601 */
3602 if (!iter->ent)
3603 return false;
3604
3605 trace_event = ftrace_find_event(iter->ent->type);
3606 if (!trace_event)
3607 return false;
3608
3609 event = container_of(trace_event, struct trace_event_call, event);
3610 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3611 return false;
3612
3613 /* Would rather have rodata, but this will suffice */
3614 if (within_module_core(addr, event->module))
3615 return true;
3616
3617 return false;
3618}
3619
3620static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3621
3622static int test_can_verify_check(const char *fmt, ...)
3623{
3624 char buf[16];
3625 va_list ap;
3626 int ret;
3627
3628 /*
3629 * The verifier is dependent on vsnprintf() modifies the va_list
3630 * passed to it, where it is sent as a reference. Some architectures
3631 * (like x86_32) passes it by value, which means that vsnprintf()
3632 * does not modify the va_list passed to it, and the verifier
3633 * would then need to be able to understand all the values that
3634 * vsnprintf can use. If it is passed by value, then the verifier
3635 * is disabled.
3636 */
3637 va_start(ap, fmt);
3638 vsnprintf(buf, 16, "%d", ap);
3639 ret = va_arg(ap, int);
3640 va_end(ap);
3641
3642 return ret;
3643}
3644
3645static void test_can_verify(void)
3646{
3647 if (!test_can_verify_check("%d %d", 0, 1)) {
3648 pr_info("trace event string verifier disabled\n");
3649 static_branch_inc(&trace_no_verify);
3650 }
3651}
3652
3653/**
3654 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3655 * @iter: The iterator that holds the seq buffer and the event being printed
3656 * @fmt: The format used to print the event
3657 * @ap: The va_list holding the data to print from @fmt.
3658 *
3659 * This writes the data into the @iter->seq buffer using the data from
3660 * @fmt and @ap. If the format has a %s, then the source of the string
3661 * is examined to make sure it is safe to print, otherwise it will
3662 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3663 * pointer.
3664 */
3665void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3666 va_list ap)
3667{
3668 const char *p = fmt;
3669 const char *str;
3670 int i, j;
3671
3672 if (WARN_ON_ONCE(!fmt))
3673 return;
3674
3675 if (static_branch_unlikely(&trace_no_verify))
3676 goto print;
3677
3678 /* Don't bother checking when doing a ftrace_dump() */
3679 if (iter->fmt == static_fmt_buf)
3680 goto print;
3681
3682 while (*p) {
3683 bool star = false;
3684 int len = 0;
3685
3686 j = 0;
3687
3688 /* We only care about %s and variants */
3689 for (i = 0; p[i]; i++) {
3690 if (i + 1 >= iter->fmt_size) {
3691 /*
3692 * If we can't expand the copy buffer,
3693 * just print it.
3694 */
3695 if (!trace_iter_expand_format(iter))
3696 goto print;
3697 }
3698
3699 if (p[i] == '\\' && p[i+1]) {
3700 i++;
3701 continue;
3702 }
3703 if (p[i] == '%') {
3704 /* Need to test cases like %08.*s */
3705 for (j = 1; p[i+j]; j++) {
3706 if (isdigit(p[i+j]) ||
3707 p[i+j] == '.')
3708 continue;
3709 if (p[i+j] == '*') {
3710 star = true;
3711 continue;
3712 }
3713 break;
3714 }
3715 if (p[i+j] == 's')
3716 break;
3717 star = false;
3718 }
3719 j = 0;
3720 }
3721 /* If no %s found then just print normally */
3722 if (!p[i])
3723 break;
3724
3725 /* Copy up to the %s, and print that */
3726 strncpy(iter->fmt, p, i);
3727 iter->fmt[i] = '\0';
3728 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3729
3730 /*
3731 * If iter->seq is full, the above call no longer guarantees
3732 * that ap is in sync with fmt processing, and further calls
3733 * to va_arg() can return wrong positional arguments.
3734 *
3735 * Ensure that ap is no longer used in this case.
3736 */
3737 if (iter->seq.full) {
3738 p = "";
3739 break;
3740 }
3741
3742 if (star)
3743 len = va_arg(ap, int);
3744
3745 /* The ap now points to the string data of the %s */
3746 str = va_arg(ap, const char *);
3747
3748 /*
3749 * If you hit this warning, it is likely that the
3750 * trace event in question used %s on a string that
3751 * was saved at the time of the event, but may not be
3752 * around when the trace is read. Use __string(),
3753 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3754 * instead. See samples/trace_events/trace-events-sample.h
3755 * for reference.
3756 */
3757 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3758 "fmt: '%s' current_buffer: '%s'",
3759 fmt, seq_buf_str(&iter->seq.seq))) {
3760 int ret;
3761
3762 /* Try to safely read the string */
3763 if (star) {
3764 if (len + 1 > iter->fmt_size)
3765 len = iter->fmt_size - 1;
3766 if (len < 0)
3767 len = 0;
3768 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3769 iter->fmt[len] = 0;
3770 star = false;
3771 } else {
3772 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3773 iter->fmt_size);
3774 }
3775 if (ret < 0)
3776 trace_seq_printf(&iter->seq, "(0x%px)", str);
3777 else
3778 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3779 str, iter->fmt);
3780 str = "[UNSAFE-MEMORY]";
3781 strcpy(iter->fmt, "%s");
3782 } else {
3783 strncpy(iter->fmt, p + i, j + 1);
3784 iter->fmt[j+1] = '\0';
3785 }
3786 if (star)
3787 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3788 else
3789 trace_seq_printf(&iter->seq, iter->fmt, str);
3790
3791 p += i + j + 1;
3792 }
3793 print:
3794 if (*p)
3795 trace_seq_vprintf(&iter->seq, p, ap);
3796}
3797
3798const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3799{
3800 const char *p, *new_fmt;
3801 char *q;
3802
3803 if (WARN_ON_ONCE(!fmt))
3804 return fmt;
3805
3806 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
3807 return fmt;
3808
3809 p = fmt;
3810 new_fmt = q = iter->fmt;
3811 while (*p) {
3812 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
3813 if (!trace_iter_expand_format(iter))
3814 return fmt;
3815
3816 q += iter->fmt - new_fmt;
3817 new_fmt = iter->fmt;
3818 }
3819
3820 *q++ = *p++;
3821
3822 /* Replace %p with %px */
3823 if (p[-1] == '%') {
3824 if (p[0] == '%') {
3825 *q++ = *p++;
3826 } else if (p[0] == 'p' && !isalnum(p[1])) {
3827 *q++ = *p++;
3828 *q++ = 'x';
3829 }
3830 }
3831 }
3832 *q = '\0';
3833
3834 return new_fmt;
3835}
3836
3837#define STATIC_TEMP_BUF_SIZE 128
3838static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3839
3840/* Find the next real entry, without updating the iterator itself */
3841struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3842 int *ent_cpu, u64 *ent_ts)
3843{
3844 /* __find_next_entry will reset ent_size */
3845 int ent_size = iter->ent_size;
3846 struct trace_entry *entry;
3847
3848 /*
3849 * If called from ftrace_dump(), then the iter->temp buffer
3850 * will be the static_temp_buf and not created from kmalloc.
3851 * If the entry size is greater than the buffer, we can
3852 * not save it. Just return NULL in that case. This is only
3853 * used to add markers when two consecutive events' time
3854 * stamps have a large delta. See trace_print_lat_context()
3855 */
3856 if (iter->temp == static_temp_buf &&
3857 STATIC_TEMP_BUF_SIZE < ent_size)
3858 return NULL;
3859
3860 /*
3861 * The __find_next_entry() may call peek_next_entry(), which may
3862 * call ring_buffer_peek() that may make the contents of iter->ent
3863 * undefined. Need to copy iter->ent now.
3864 */
3865 if (iter->ent && iter->ent != iter->temp) {
3866 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3867 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3868 void *temp;
3869 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3870 if (!temp)
3871 return NULL;
3872 kfree(iter->temp);
3873 iter->temp = temp;
3874 iter->temp_size = iter->ent_size;
3875 }
3876 memcpy(iter->temp, iter->ent, iter->ent_size);
3877 iter->ent = iter->temp;
3878 }
3879 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3880 /* Put back the original ent_size */
3881 iter->ent_size = ent_size;
3882
3883 return entry;
3884}
3885
3886/* Find the next real entry, and increment the iterator to the next entry */
3887void *trace_find_next_entry_inc(struct trace_iterator *iter)
3888{
3889 iter->ent = __find_next_entry(iter, &iter->cpu,
3890 &iter->lost_events, &iter->ts);
3891
3892 if (iter->ent)
3893 trace_iterator_increment(iter);
3894
3895 return iter->ent ? iter : NULL;
3896}
3897
3898static void trace_consume(struct trace_iterator *iter)
3899{
3900 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3901 &iter->lost_events);
3902}
3903
3904static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3905{
3906 struct trace_iterator *iter = m->private;
3907 int i = (int)*pos;
3908 void *ent;
3909
3910 WARN_ON_ONCE(iter->leftover);
3911
3912 (*pos)++;
3913
3914 /* can't go backwards */
3915 if (iter->idx > i)
3916 return NULL;
3917
3918 if (iter->idx < 0)
3919 ent = trace_find_next_entry_inc(iter);
3920 else
3921 ent = iter;
3922
3923 while (ent && iter->idx < i)
3924 ent = trace_find_next_entry_inc(iter);
3925
3926 iter->pos = *pos;
3927
3928 return ent;
3929}
3930
3931void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3932{
3933 struct ring_buffer_iter *buf_iter;
3934 unsigned long entries = 0;
3935 u64 ts;
3936
3937 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
3938
3939 buf_iter = trace_buffer_iter(iter, cpu);
3940 if (!buf_iter)
3941 return;
3942
3943 ring_buffer_iter_reset(buf_iter);
3944
3945 /*
3946 * We could have the case with the max latency tracers
3947 * that a reset never took place on a cpu. This is evident
3948 * by the timestamp being before the start of the buffer.
3949 */
3950 while (ring_buffer_iter_peek(buf_iter, &ts)) {
3951 if (ts >= iter->array_buffer->time_start)
3952 break;
3953 entries++;
3954 ring_buffer_iter_advance(buf_iter);
3955 }
3956
3957 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
3958}
3959
3960/*
3961 * The current tracer is copied to avoid a global locking
3962 * all around.
3963 */
3964static void *s_start(struct seq_file *m, loff_t *pos)
3965{
3966 struct trace_iterator *iter = m->private;
3967 struct trace_array *tr = iter->tr;
3968 int cpu_file = iter->cpu_file;
3969 void *p = NULL;
3970 loff_t l = 0;
3971 int cpu;
3972
3973 mutex_lock(&trace_types_lock);
3974 if (unlikely(tr->current_trace != iter->trace)) {
3975 /* Close iter->trace before switching to the new current tracer */
3976 if (iter->trace->close)
3977 iter->trace->close(iter);
3978 iter->trace = tr->current_trace;
3979 /* Reopen the new current tracer */
3980 if (iter->trace->open)
3981 iter->trace->open(iter);
3982 }
3983 mutex_unlock(&trace_types_lock);
3984
3985#ifdef CONFIG_TRACER_MAX_TRACE
3986 if (iter->snapshot && iter->trace->use_max_tr)
3987 return ERR_PTR(-EBUSY);
3988#endif
3989
3990 if (*pos != iter->pos) {
3991 iter->ent = NULL;
3992 iter->cpu = 0;
3993 iter->idx = -1;
3994
3995 if (cpu_file == RING_BUFFER_ALL_CPUS) {
3996 for_each_tracing_cpu(cpu)
3997 tracing_iter_reset(iter, cpu);
3998 } else
3999 tracing_iter_reset(iter, cpu_file);
4000
4001 iter->leftover = 0;
4002 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4003 ;
4004
4005 } else {
4006 /*
4007 * If we overflowed the seq_file before, then we want
4008 * to just reuse the trace_seq buffer again.
4009 */
4010 if (iter->leftover)
4011 p = iter;
4012 else {
4013 l = *pos - 1;
4014 p = s_next(m, p, &l);
4015 }
4016 }
4017
4018 trace_event_read_lock();
4019 trace_access_lock(cpu_file);
4020 return p;
4021}
4022
4023static void s_stop(struct seq_file *m, void *p)
4024{
4025 struct trace_iterator *iter = m->private;
4026
4027#ifdef CONFIG_TRACER_MAX_TRACE
4028 if (iter->snapshot && iter->trace->use_max_tr)
4029 return;
4030#endif
4031
4032 trace_access_unlock(iter->cpu_file);
4033 trace_event_read_unlock();
4034}
4035
4036static void
4037get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4038 unsigned long *entries, int cpu)
4039{
4040 unsigned long count;
4041
4042 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4043 /*
4044 * If this buffer has skipped entries, then we hold all
4045 * entries for the trace and we need to ignore the
4046 * ones before the time stamp.
4047 */
4048 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4049 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4050 /* total is the same as the entries */
4051 *total = count;
4052 } else
4053 *total = count +
4054 ring_buffer_overrun_cpu(buf->buffer, cpu);
4055 *entries = count;
4056}
4057
4058static void
4059get_total_entries(struct array_buffer *buf,
4060 unsigned long *total, unsigned long *entries)
4061{
4062 unsigned long t, e;
4063 int cpu;
4064
4065 *total = 0;
4066 *entries = 0;
4067
4068 for_each_tracing_cpu(cpu) {
4069 get_total_entries_cpu(buf, &t, &e, cpu);
4070 *total += t;
4071 *entries += e;
4072 }
4073}
4074
4075unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4076{
4077 unsigned long total, entries;
4078
4079 if (!tr)
4080 tr = &global_trace;
4081
4082 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4083
4084 return entries;
4085}
4086
4087unsigned long trace_total_entries(struct trace_array *tr)
4088{
4089 unsigned long total, entries;
4090
4091 if (!tr)
4092 tr = &global_trace;
4093
4094 get_total_entries(&tr->array_buffer, &total, &entries);
4095
4096 return entries;
4097}
4098
4099static void print_lat_help_header(struct seq_file *m)
4100{
4101 seq_puts(m, "# _------=> CPU# \n"
4102 "# / _-----=> irqs-off/BH-disabled\n"
4103 "# | / _----=> need-resched \n"
4104 "# || / _---=> hardirq/softirq \n"
4105 "# ||| / _--=> preempt-depth \n"
4106 "# |||| / _-=> migrate-disable \n"
4107 "# ||||| / delay \n"
4108 "# cmd pid |||||| time | caller \n"
4109 "# \\ / |||||| \\ | / \n");
4110}
4111
4112static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4113{
4114 unsigned long total;
4115 unsigned long entries;
4116
4117 get_total_entries(buf, &total, &entries);
4118 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4119 entries, total, num_online_cpus());
4120 seq_puts(m, "#\n");
4121}
4122
4123static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4124 unsigned int flags)
4125{
4126 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4127
4128 print_event_info(buf, m);
4129
4130 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4131 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4132}
4133
4134static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4135 unsigned int flags)
4136{
4137 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4138 static const char space[] = " ";
4139 int prec = tgid ? 12 : 2;
4140
4141 print_event_info(buf, m);
4142
4143 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4144 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4145 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4146 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4147 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4148 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4149 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4150 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4151}
4152
4153void
4154print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4155{
4156 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4157 struct array_buffer *buf = iter->array_buffer;
4158 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4159 struct tracer *type = iter->trace;
4160 unsigned long entries;
4161 unsigned long total;
4162 const char *name = type->name;
4163
4164 get_total_entries(buf, &total, &entries);
4165
4166 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4167 name, init_utsname()->release);
4168 seq_puts(m, "# -----------------------------------"
4169 "---------------------------------\n");
4170 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4171 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4172 nsecs_to_usecs(data->saved_latency),
4173 entries,
4174 total,
4175 buf->cpu,
4176 preempt_model_none() ? "server" :
4177 preempt_model_voluntary() ? "desktop" :
4178 preempt_model_full() ? "preempt" :
4179 preempt_model_rt() ? "preempt_rt" :
4180 "unknown",
4181 /* These are reserved for later use */
4182 0, 0, 0, 0);
4183#ifdef CONFIG_SMP
4184 seq_printf(m, " #P:%d)\n", num_online_cpus());
4185#else
4186 seq_puts(m, ")\n");
4187#endif
4188 seq_puts(m, "# -----------------\n");
4189 seq_printf(m, "# | task: %.16s-%d "
4190 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4191 data->comm, data->pid,
4192 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4193 data->policy, data->rt_priority);
4194 seq_puts(m, "# -----------------\n");
4195
4196 if (data->critical_start) {
4197 seq_puts(m, "# => started at: ");
4198 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4199 trace_print_seq(m, &iter->seq);
4200 seq_puts(m, "\n# => ended at: ");
4201 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4202 trace_print_seq(m, &iter->seq);
4203 seq_puts(m, "\n#\n");
4204 }
4205
4206 seq_puts(m, "#\n");
4207}
4208
4209static void test_cpu_buff_start(struct trace_iterator *iter)
4210{
4211 struct trace_seq *s = &iter->seq;
4212 struct trace_array *tr = iter->tr;
4213
4214 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4215 return;
4216
4217 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4218 return;
4219
4220 if (cpumask_available(iter->started) &&
4221 cpumask_test_cpu(iter->cpu, iter->started))
4222 return;
4223
4224 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4225 return;
4226
4227 if (cpumask_available(iter->started))
4228 cpumask_set_cpu(iter->cpu, iter->started);
4229
4230 /* Don't print started cpu buffer for the first entry of the trace */
4231 if (iter->idx > 1)
4232 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4233 iter->cpu);
4234}
4235
4236static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4237{
4238 struct trace_array *tr = iter->tr;
4239 struct trace_seq *s = &iter->seq;
4240 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4241 struct trace_entry *entry;
4242 struct trace_event *event;
4243
4244 entry = iter->ent;
4245
4246 test_cpu_buff_start(iter);
4247
4248 event = ftrace_find_event(entry->type);
4249
4250 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4251 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4252 trace_print_lat_context(iter);
4253 else
4254 trace_print_context(iter);
4255 }
4256
4257 if (trace_seq_has_overflowed(s))
4258 return TRACE_TYPE_PARTIAL_LINE;
4259
4260 if (event) {
4261 if (tr->trace_flags & TRACE_ITER_FIELDS)
4262 return print_event_fields(iter, event);
4263 return event->funcs->trace(iter, sym_flags, event);
4264 }
4265
4266 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4267
4268 return trace_handle_return(s);
4269}
4270
4271static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4272{
4273 struct trace_array *tr = iter->tr;
4274 struct trace_seq *s = &iter->seq;
4275 struct trace_entry *entry;
4276 struct trace_event *event;
4277
4278 entry = iter->ent;
4279
4280 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4281 trace_seq_printf(s, "%d %d %llu ",
4282 entry->pid, iter->cpu, iter->ts);
4283
4284 if (trace_seq_has_overflowed(s))
4285 return TRACE_TYPE_PARTIAL_LINE;
4286
4287 event = ftrace_find_event(entry->type);
4288 if (event)
4289 return event->funcs->raw(iter, 0, event);
4290
4291 trace_seq_printf(s, "%d ?\n", entry->type);
4292
4293 return trace_handle_return(s);
4294}
4295
4296static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4297{
4298 struct trace_array *tr = iter->tr;
4299 struct trace_seq *s = &iter->seq;
4300 unsigned char newline = '\n';
4301 struct trace_entry *entry;
4302 struct trace_event *event;
4303
4304 entry = iter->ent;
4305
4306 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4307 SEQ_PUT_HEX_FIELD(s, entry->pid);
4308 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4309 SEQ_PUT_HEX_FIELD(s, iter->ts);
4310 if (trace_seq_has_overflowed(s))
4311 return TRACE_TYPE_PARTIAL_LINE;
4312 }
4313
4314 event = ftrace_find_event(entry->type);
4315 if (event) {
4316 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4317 if (ret != TRACE_TYPE_HANDLED)
4318 return ret;
4319 }
4320
4321 SEQ_PUT_FIELD(s, newline);
4322
4323 return trace_handle_return(s);
4324}
4325
4326static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4327{
4328 struct trace_array *tr = iter->tr;
4329 struct trace_seq *s = &iter->seq;
4330 struct trace_entry *entry;
4331 struct trace_event *event;
4332
4333 entry = iter->ent;
4334
4335 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4336 SEQ_PUT_FIELD(s, entry->pid);
4337 SEQ_PUT_FIELD(s, iter->cpu);
4338 SEQ_PUT_FIELD(s, iter->ts);
4339 if (trace_seq_has_overflowed(s))
4340 return TRACE_TYPE_PARTIAL_LINE;
4341 }
4342
4343 event = ftrace_find_event(entry->type);
4344 return event ? event->funcs->binary(iter, 0, event) :
4345 TRACE_TYPE_HANDLED;
4346}
4347
4348int trace_empty(struct trace_iterator *iter)
4349{
4350 struct ring_buffer_iter *buf_iter;
4351 int cpu;
4352
4353 /* If we are looking at one CPU buffer, only check that one */
4354 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4355 cpu = iter->cpu_file;
4356 buf_iter = trace_buffer_iter(iter, cpu);
4357 if (buf_iter) {
4358 if (!ring_buffer_iter_empty(buf_iter))
4359 return 0;
4360 } else {
4361 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4362 return 0;
4363 }
4364 return 1;
4365 }
4366
4367 for_each_tracing_cpu(cpu) {
4368 buf_iter = trace_buffer_iter(iter, cpu);
4369 if (buf_iter) {
4370 if (!ring_buffer_iter_empty(buf_iter))
4371 return 0;
4372 } else {
4373 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4374 return 0;
4375 }
4376 }
4377
4378 return 1;
4379}
4380
4381/* Called with trace_event_read_lock() held. */
4382enum print_line_t print_trace_line(struct trace_iterator *iter)
4383{
4384 struct trace_array *tr = iter->tr;
4385 unsigned long trace_flags = tr->trace_flags;
4386 enum print_line_t ret;
4387
4388 if (iter->lost_events) {
4389 if (iter->lost_events == (unsigned long)-1)
4390 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4391 iter->cpu);
4392 else
4393 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4394 iter->cpu, iter->lost_events);
4395 if (trace_seq_has_overflowed(&iter->seq))
4396 return TRACE_TYPE_PARTIAL_LINE;
4397 }
4398
4399 if (iter->trace && iter->trace->print_line) {
4400 ret = iter->trace->print_line(iter);
4401 if (ret != TRACE_TYPE_UNHANDLED)
4402 return ret;
4403 }
4404
4405 if (iter->ent->type == TRACE_BPUTS &&
4406 trace_flags & TRACE_ITER_PRINTK &&
4407 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4408 return trace_print_bputs_msg_only(iter);
4409
4410 if (iter->ent->type == TRACE_BPRINT &&
4411 trace_flags & TRACE_ITER_PRINTK &&
4412 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4413 return trace_print_bprintk_msg_only(iter);
4414
4415 if (iter->ent->type == TRACE_PRINT &&
4416 trace_flags & TRACE_ITER_PRINTK &&
4417 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4418 return trace_print_printk_msg_only(iter);
4419
4420 if (trace_flags & TRACE_ITER_BIN)
4421 return print_bin_fmt(iter);
4422
4423 if (trace_flags & TRACE_ITER_HEX)
4424 return print_hex_fmt(iter);
4425
4426 if (trace_flags & TRACE_ITER_RAW)
4427 return print_raw_fmt(iter);
4428
4429 return print_trace_fmt(iter);
4430}
4431
4432void trace_latency_header(struct seq_file *m)
4433{
4434 struct trace_iterator *iter = m->private;
4435 struct trace_array *tr = iter->tr;
4436
4437 /* print nothing if the buffers are empty */
4438 if (trace_empty(iter))
4439 return;
4440
4441 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4442 print_trace_header(m, iter);
4443
4444 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4445 print_lat_help_header(m);
4446}
4447
4448void trace_default_header(struct seq_file *m)
4449{
4450 struct trace_iterator *iter = m->private;
4451 struct trace_array *tr = iter->tr;
4452 unsigned long trace_flags = tr->trace_flags;
4453
4454 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4455 return;
4456
4457 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4458 /* print nothing if the buffers are empty */
4459 if (trace_empty(iter))
4460 return;
4461 print_trace_header(m, iter);
4462 if (!(trace_flags & TRACE_ITER_VERBOSE))
4463 print_lat_help_header(m);
4464 } else {
4465 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4466 if (trace_flags & TRACE_ITER_IRQ_INFO)
4467 print_func_help_header_irq(iter->array_buffer,
4468 m, trace_flags);
4469 else
4470 print_func_help_header(iter->array_buffer, m,
4471 trace_flags);
4472 }
4473 }
4474}
4475
4476static void test_ftrace_alive(struct seq_file *m)
4477{
4478 if (!ftrace_is_dead())
4479 return;
4480 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4481 "# MAY BE MISSING FUNCTION EVENTS\n");
4482}
4483
4484#ifdef CONFIG_TRACER_MAX_TRACE
4485static void show_snapshot_main_help(struct seq_file *m)
4486{
4487 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4488 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4489 "# Takes a snapshot of the main buffer.\n"
4490 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4491 "# (Doesn't have to be '2' works with any number that\n"
4492 "# is not a '0' or '1')\n");
4493}
4494
4495static void show_snapshot_percpu_help(struct seq_file *m)
4496{
4497 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4498#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4499 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4500 "# Takes a snapshot of the main buffer for this cpu.\n");
4501#else
4502 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4503 "# Must use main snapshot file to allocate.\n");
4504#endif
4505 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4506 "# (Doesn't have to be '2' works with any number that\n"
4507 "# is not a '0' or '1')\n");
4508}
4509
4510static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4511{
4512 if (iter->tr->allocated_snapshot)
4513 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4514 else
4515 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4516
4517 seq_puts(m, "# Snapshot commands:\n");
4518 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4519 show_snapshot_main_help(m);
4520 else
4521 show_snapshot_percpu_help(m);
4522}
4523#else
4524/* Should never be called */
4525static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4526#endif
4527
4528static int s_show(struct seq_file *m, void *v)
4529{
4530 struct trace_iterator *iter = v;
4531 int ret;
4532
4533 if (iter->ent == NULL) {
4534 if (iter->tr) {
4535 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4536 seq_puts(m, "#\n");
4537 test_ftrace_alive(m);
4538 }
4539 if (iter->snapshot && trace_empty(iter))
4540 print_snapshot_help(m, iter);
4541 else if (iter->trace && iter->trace->print_header)
4542 iter->trace->print_header(m);
4543 else
4544 trace_default_header(m);
4545
4546 } else if (iter->leftover) {
4547 /*
4548 * If we filled the seq_file buffer earlier, we
4549 * want to just show it now.
4550 */
4551 ret = trace_print_seq(m, &iter->seq);
4552
4553 /* ret should this time be zero, but you never know */
4554 iter->leftover = ret;
4555
4556 } else {
4557 ret = print_trace_line(iter);
4558 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4559 iter->seq.full = 0;
4560 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
4561 }
4562 ret = trace_print_seq(m, &iter->seq);
4563 /*
4564 * If we overflow the seq_file buffer, then it will
4565 * ask us for this data again at start up.
4566 * Use that instead.
4567 * ret is 0 if seq_file write succeeded.
4568 * -1 otherwise.
4569 */
4570 iter->leftover = ret;
4571 }
4572
4573 return 0;
4574}
4575
4576/*
4577 * Should be used after trace_array_get(), trace_types_lock
4578 * ensures that i_cdev was already initialized.
4579 */
4580static inline int tracing_get_cpu(struct inode *inode)
4581{
4582 if (inode->i_cdev) /* See trace_create_cpu_file() */
4583 return (long)inode->i_cdev - 1;
4584 return RING_BUFFER_ALL_CPUS;
4585}
4586
4587static const struct seq_operations tracer_seq_ops = {
4588 .start = s_start,
4589 .next = s_next,
4590 .stop = s_stop,
4591 .show = s_show,
4592};
4593
4594/*
4595 * Note, as iter itself can be allocated and freed in different
4596 * ways, this function is only used to free its content, and not
4597 * the iterator itself. The only requirement to all the allocations
4598 * is that it must zero all fields (kzalloc), as freeing works with
4599 * ethier allocated content or NULL.
4600 */
4601static void free_trace_iter_content(struct trace_iterator *iter)
4602{
4603 /* The fmt is either NULL, allocated or points to static_fmt_buf */
4604 if (iter->fmt != static_fmt_buf)
4605 kfree(iter->fmt);
4606
4607 kfree(iter->temp);
4608 kfree(iter->buffer_iter);
4609 mutex_destroy(&iter->mutex);
4610 free_cpumask_var(iter->started);
4611}
4612
4613static struct trace_iterator *
4614__tracing_open(struct inode *inode, struct file *file, bool snapshot)
4615{
4616 struct trace_array *tr = inode->i_private;
4617 struct trace_iterator *iter;
4618 int cpu;
4619
4620 if (tracing_disabled)
4621 return ERR_PTR(-ENODEV);
4622
4623 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4624 if (!iter)
4625 return ERR_PTR(-ENOMEM);
4626
4627 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4628 GFP_KERNEL);
4629 if (!iter->buffer_iter)
4630 goto release;
4631
4632 /*
4633 * trace_find_next_entry() may need to save off iter->ent.
4634 * It will place it into the iter->temp buffer. As most
4635 * events are less than 128, allocate a buffer of that size.
4636 * If one is greater, then trace_find_next_entry() will
4637 * allocate a new buffer to adjust for the bigger iter->ent.
4638 * It's not critical if it fails to get allocated here.
4639 */
4640 iter->temp = kmalloc(128, GFP_KERNEL);
4641 if (iter->temp)
4642 iter->temp_size = 128;
4643
4644 /*
4645 * trace_event_printf() may need to modify given format
4646 * string to replace %p with %px so that it shows real address
4647 * instead of hash value. However, that is only for the event
4648 * tracing, other tracer may not need. Defer the allocation
4649 * until it is needed.
4650 */
4651 iter->fmt = NULL;
4652 iter->fmt_size = 0;
4653
4654 mutex_lock(&trace_types_lock);
4655 iter->trace = tr->current_trace;
4656
4657 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4658 goto fail;
4659
4660 iter->tr = tr;
4661
4662#ifdef CONFIG_TRACER_MAX_TRACE
4663 /* Currently only the top directory has a snapshot */
4664 if (tr->current_trace->print_max || snapshot)
4665 iter->array_buffer = &tr->max_buffer;
4666 else
4667#endif
4668 iter->array_buffer = &tr->array_buffer;
4669 iter->snapshot = snapshot;
4670 iter->pos = -1;
4671 iter->cpu_file = tracing_get_cpu(inode);
4672 mutex_init(&iter->mutex);
4673
4674 /* Notify the tracer early; before we stop tracing. */
4675 if (iter->trace->open)
4676 iter->trace->open(iter);
4677
4678 /* Annotate start of buffers if we had overruns */
4679 if (ring_buffer_overruns(iter->array_buffer->buffer))
4680 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4681
4682 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4683 if (trace_clocks[tr->clock_id].in_ns)
4684 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4685
4686 /*
4687 * If pause-on-trace is enabled, then stop the trace while
4688 * dumping, unless this is the "snapshot" file
4689 */
4690 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4691 tracing_stop_tr(tr);
4692
4693 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4694 for_each_tracing_cpu(cpu) {
4695 iter->buffer_iter[cpu] =
4696 ring_buffer_read_prepare(iter->array_buffer->buffer,
4697 cpu, GFP_KERNEL);
4698 }
4699 ring_buffer_read_prepare_sync();
4700 for_each_tracing_cpu(cpu) {
4701 ring_buffer_read_start(iter->buffer_iter[cpu]);
4702 tracing_iter_reset(iter, cpu);
4703 }
4704 } else {
4705 cpu = iter->cpu_file;
4706 iter->buffer_iter[cpu] =
4707 ring_buffer_read_prepare(iter->array_buffer->buffer,
4708 cpu, GFP_KERNEL);
4709 ring_buffer_read_prepare_sync();
4710 ring_buffer_read_start(iter->buffer_iter[cpu]);
4711 tracing_iter_reset(iter, cpu);
4712 }
4713
4714 mutex_unlock(&trace_types_lock);
4715
4716 return iter;
4717
4718 fail:
4719 mutex_unlock(&trace_types_lock);
4720 free_trace_iter_content(iter);
4721release:
4722 seq_release_private(inode, file);
4723 return ERR_PTR(-ENOMEM);
4724}
4725
4726int tracing_open_generic(struct inode *inode, struct file *filp)
4727{
4728 int ret;
4729
4730 ret = tracing_check_open_get_tr(NULL);
4731 if (ret)
4732 return ret;
4733
4734 filp->private_data = inode->i_private;
4735 return 0;
4736}
4737
4738bool tracing_is_disabled(void)
4739{
4740 return (tracing_disabled) ? true: false;
4741}
4742
4743/*
4744 * Open and update trace_array ref count.
4745 * Must have the current trace_array passed to it.
4746 */
4747int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4748{
4749 struct trace_array *tr = inode->i_private;
4750 int ret;
4751
4752 ret = tracing_check_open_get_tr(tr);
4753 if (ret)
4754 return ret;
4755
4756 filp->private_data = inode->i_private;
4757
4758 return 0;
4759}
4760
4761/*
4762 * The private pointer of the inode is the trace_event_file.
4763 * Update the tr ref count associated to it.
4764 */
4765int tracing_open_file_tr(struct inode *inode, struct file *filp)
4766{
4767 struct trace_event_file *file = inode->i_private;
4768 int ret;
4769
4770 ret = tracing_check_open_get_tr(file->tr);
4771 if (ret)
4772 return ret;
4773
4774 mutex_lock(&event_mutex);
4775
4776 /* Fail if the file is marked for removal */
4777 if (file->flags & EVENT_FILE_FL_FREED) {
4778 trace_array_put(file->tr);
4779 ret = -ENODEV;
4780 } else {
4781 event_file_get(file);
4782 }
4783
4784 mutex_unlock(&event_mutex);
4785 if (ret)
4786 return ret;
4787
4788 filp->private_data = inode->i_private;
4789
4790 return 0;
4791}
4792
4793int tracing_release_file_tr(struct inode *inode, struct file *filp)
4794{
4795 struct trace_event_file *file = inode->i_private;
4796
4797 trace_array_put(file->tr);
4798 event_file_put(file);
4799
4800 return 0;
4801}
4802
4803int tracing_single_release_file_tr(struct inode *inode, struct file *filp)
4804{
4805 tracing_release_file_tr(inode, filp);
4806 return single_release(inode, filp);
4807}
4808
4809static int tracing_mark_open(struct inode *inode, struct file *filp)
4810{
4811 stream_open(inode, filp);
4812 return tracing_open_generic_tr(inode, filp);
4813}
4814
4815static int tracing_release(struct inode *inode, struct file *file)
4816{
4817 struct trace_array *tr = inode->i_private;
4818 struct seq_file *m = file->private_data;
4819 struct trace_iterator *iter;
4820 int cpu;
4821
4822 if (!(file->f_mode & FMODE_READ)) {
4823 trace_array_put(tr);
4824 return 0;
4825 }
4826
4827 /* Writes do not use seq_file */
4828 iter = m->private;
4829 mutex_lock(&trace_types_lock);
4830
4831 for_each_tracing_cpu(cpu) {
4832 if (iter->buffer_iter[cpu])
4833 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4834 }
4835
4836 if (iter->trace && iter->trace->close)
4837 iter->trace->close(iter);
4838
4839 if (!iter->snapshot && tr->stop_count)
4840 /* reenable tracing if it was previously enabled */
4841 tracing_start_tr(tr);
4842
4843 __trace_array_put(tr);
4844
4845 mutex_unlock(&trace_types_lock);
4846
4847 free_trace_iter_content(iter);
4848 seq_release_private(inode, file);
4849
4850 return 0;
4851}
4852
4853int tracing_release_generic_tr(struct inode *inode, struct file *file)
4854{
4855 struct trace_array *tr = inode->i_private;
4856
4857 trace_array_put(tr);
4858 return 0;
4859}
4860
4861static int tracing_single_release_tr(struct inode *inode, struct file *file)
4862{
4863 struct trace_array *tr = inode->i_private;
4864
4865 trace_array_put(tr);
4866
4867 return single_release(inode, file);
4868}
4869
4870static int tracing_open(struct inode *inode, struct file *file)
4871{
4872 struct trace_array *tr = inode->i_private;
4873 struct trace_iterator *iter;
4874 int ret;
4875
4876 ret = tracing_check_open_get_tr(tr);
4877 if (ret)
4878 return ret;
4879
4880 /* If this file was open for write, then erase contents */
4881 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4882 int cpu = tracing_get_cpu(inode);
4883 struct array_buffer *trace_buf = &tr->array_buffer;
4884
4885#ifdef CONFIG_TRACER_MAX_TRACE
4886 if (tr->current_trace->print_max)
4887 trace_buf = &tr->max_buffer;
4888#endif
4889
4890 if (cpu == RING_BUFFER_ALL_CPUS)
4891 tracing_reset_online_cpus(trace_buf);
4892 else
4893 tracing_reset_cpu(trace_buf, cpu);
4894 }
4895
4896 if (file->f_mode & FMODE_READ) {
4897 iter = __tracing_open(inode, file, false);
4898 if (IS_ERR(iter))
4899 ret = PTR_ERR(iter);
4900 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4901 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4902 }
4903
4904 if (ret < 0)
4905 trace_array_put(tr);
4906
4907 return ret;
4908}
4909
4910/*
4911 * Some tracers are not suitable for instance buffers.
4912 * A tracer is always available for the global array (toplevel)
4913 * or if it explicitly states that it is.
4914 */
4915static bool
4916trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4917{
4918 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4919}
4920
4921/* Find the next tracer that this trace array may use */
4922static struct tracer *
4923get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4924{
4925 while (t && !trace_ok_for_array(t, tr))
4926 t = t->next;
4927
4928 return t;
4929}
4930
4931static void *
4932t_next(struct seq_file *m, void *v, loff_t *pos)
4933{
4934 struct trace_array *tr = m->private;
4935 struct tracer *t = v;
4936
4937 (*pos)++;
4938
4939 if (t)
4940 t = get_tracer_for_array(tr, t->next);
4941
4942 return t;
4943}
4944
4945static void *t_start(struct seq_file *m, loff_t *pos)
4946{
4947 struct trace_array *tr = m->private;
4948 struct tracer *t;
4949 loff_t l = 0;
4950
4951 mutex_lock(&trace_types_lock);
4952
4953 t = get_tracer_for_array(tr, trace_types);
4954 for (; t && l < *pos; t = t_next(m, t, &l))
4955 ;
4956
4957 return t;
4958}
4959
4960static void t_stop(struct seq_file *m, void *p)
4961{
4962 mutex_unlock(&trace_types_lock);
4963}
4964
4965static int t_show(struct seq_file *m, void *v)
4966{
4967 struct tracer *t = v;
4968
4969 if (!t)
4970 return 0;
4971
4972 seq_puts(m, t->name);
4973 if (t->next)
4974 seq_putc(m, ' ');
4975 else
4976 seq_putc(m, '\n');
4977
4978 return 0;
4979}
4980
4981static const struct seq_operations show_traces_seq_ops = {
4982 .start = t_start,
4983 .next = t_next,
4984 .stop = t_stop,
4985 .show = t_show,
4986};
4987
4988static int show_traces_open(struct inode *inode, struct file *file)
4989{
4990 struct trace_array *tr = inode->i_private;
4991 struct seq_file *m;
4992 int ret;
4993
4994 ret = tracing_check_open_get_tr(tr);
4995 if (ret)
4996 return ret;
4997
4998 ret = seq_open(file, &show_traces_seq_ops);
4999 if (ret) {
5000 trace_array_put(tr);
5001 return ret;
5002 }
5003
5004 m = file->private_data;
5005 m->private = tr;
5006
5007 return 0;
5008}
5009
5010static int show_traces_release(struct inode *inode, struct file *file)
5011{
5012 struct trace_array *tr = inode->i_private;
5013
5014 trace_array_put(tr);
5015 return seq_release(inode, file);
5016}
5017
5018static ssize_t
5019tracing_write_stub(struct file *filp, const char __user *ubuf,
5020 size_t count, loff_t *ppos)
5021{
5022 return count;
5023}
5024
5025loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5026{
5027 int ret;
5028
5029 if (file->f_mode & FMODE_READ)
5030 ret = seq_lseek(file, offset, whence);
5031 else
5032 file->f_pos = ret = 0;
5033
5034 return ret;
5035}
5036
5037static const struct file_operations tracing_fops = {
5038 .open = tracing_open,
5039 .read = seq_read,
5040 .read_iter = seq_read_iter,
5041 .splice_read = copy_splice_read,
5042 .write = tracing_write_stub,
5043 .llseek = tracing_lseek,
5044 .release = tracing_release,
5045};
5046
5047static const struct file_operations show_traces_fops = {
5048 .open = show_traces_open,
5049 .read = seq_read,
5050 .llseek = seq_lseek,
5051 .release = show_traces_release,
5052};
5053
5054static ssize_t
5055tracing_cpumask_read(struct file *filp, char __user *ubuf,
5056 size_t count, loff_t *ppos)
5057{
5058 struct trace_array *tr = file_inode(filp)->i_private;
5059 char *mask_str;
5060 int len;
5061
5062 len = snprintf(NULL, 0, "%*pb\n",
5063 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5064 mask_str = kmalloc(len, GFP_KERNEL);
5065 if (!mask_str)
5066 return -ENOMEM;
5067
5068 len = snprintf(mask_str, len, "%*pb\n",
5069 cpumask_pr_args(tr->tracing_cpumask));
5070 if (len >= count) {
5071 count = -EINVAL;
5072 goto out_err;
5073 }
5074 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5075
5076out_err:
5077 kfree(mask_str);
5078
5079 return count;
5080}
5081
5082int tracing_set_cpumask(struct trace_array *tr,
5083 cpumask_var_t tracing_cpumask_new)
5084{
5085 int cpu;
5086
5087 if (!tr)
5088 return -EINVAL;
5089
5090 local_irq_disable();
5091 arch_spin_lock(&tr->max_lock);
5092 for_each_tracing_cpu(cpu) {
5093 /*
5094 * Increase/decrease the disabled counter if we are
5095 * about to flip a bit in the cpumask:
5096 */
5097 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5098 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5099 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5100 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5101#ifdef CONFIG_TRACER_MAX_TRACE
5102 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5103#endif
5104 }
5105 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5106 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5107 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5108 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5109#ifdef CONFIG_TRACER_MAX_TRACE
5110 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5111#endif
5112 }
5113 }
5114 arch_spin_unlock(&tr->max_lock);
5115 local_irq_enable();
5116
5117 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5118
5119 return 0;
5120}
5121
5122static ssize_t
5123tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5124 size_t count, loff_t *ppos)
5125{
5126 struct trace_array *tr = file_inode(filp)->i_private;
5127 cpumask_var_t tracing_cpumask_new;
5128 int err;
5129
5130 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5131 return -ENOMEM;
5132
5133 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5134 if (err)
5135 goto err_free;
5136
5137 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5138 if (err)
5139 goto err_free;
5140
5141 free_cpumask_var(tracing_cpumask_new);
5142
5143 return count;
5144
5145err_free:
5146 free_cpumask_var(tracing_cpumask_new);
5147
5148 return err;
5149}
5150
5151static const struct file_operations tracing_cpumask_fops = {
5152 .open = tracing_open_generic_tr,
5153 .read = tracing_cpumask_read,
5154 .write = tracing_cpumask_write,
5155 .release = tracing_release_generic_tr,
5156 .llseek = generic_file_llseek,
5157};
5158
5159static int tracing_trace_options_show(struct seq_file *m, void *v)
5160{
5161 struct tracer_opt *trace_opts;
5162 struct trace_array *tr = m->private;
5163 u32 tracer_flags;
5164 int i;
5165
5166 mutex_lock(&trace_types_lock);
5167 tracer_flags = tr->current_trace->flags->val;
5168 trace_opts = tr->current_trace->flags->opts;
5169
5170 for (i = 0; trace_options[i]; i++) {
5171 if (tr->trace_flags & (1 << i))
5172 seq_printf(m, "%s\n", trace_options[i]);
5173 else
5174 seq_printf(m, "no%s\n", trace_options[i]);
5175 }
5176
5177 for (i = 0; trace_opts[i].name; i++) {
5178 if (tracer_flags & trace_opts[i].bit)
5179 seq_printf(m, "%s\n", trace_opts[i].name);
5180 else
5181 seq_printf(m, "no%s\n", trace_opts[i].name);
5182 }
5183 mutex_unlock(&trace_types_lock);
5184
5185 return 0;
5186}
5187
5188static int __set_tracer_option(struct trace_array *tr,
5189 struct tracer_flags *tracer_flags,
5190 struct tracer_opt *opts, int neg)
5191{
5192 struct tracer *trace = tracer_flags->trace;
5193 int ret;
5194
5195 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5196 if (ret)
5197 return ret;
5198
5199 if (neg)
5200 tracer_flags->val &= ~opts->bit;
5201 else
5202 tracer_flags->val |= opts->bit;
5203 return 0;
5204}
5205
5206/* Try to assign a tracer specific option */
5207static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5208{
5209 struct tracer *trace = tr->current_trace;
5210 struct tracer_flags *tracer_flags = trace->flags;
5211 struct tracer_opt *opts = NULL;
5212 int i;
5213
5214 for (i = 0; tracer_flags->opts[i].name; i++) {
5215 opts = &tracer_flags->opts[i];
5216
5217 if (strcmp(cmp, opts->name) == 0)
5218 return __set_tracer_option(tr, trace->flags, opts, neg);
5219 }
5220
5221 return -EINVAL;
5222}
5223
5224/* Some tracers require overwrite to stay enabled */
5225int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5226{
5227 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5228 return -1;
5229
5230 return 0;
5231}
5232
5233int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5234{
5235 if ((mask == TRACE_ITER_RECORD_TGID) ||
5236 (mask == TRACE_ITER_RECORD_CMD))
5237 lockdep_assert_held(&event_mutex);
5238
5239 /* do nothing if flag is already set */
5240 if (!!(tr->trace_flags & mask) == !!enabled)
5241 return 0;
5242
5243 /* Give the tracer a chance to approve the change */
5244 if (tr->current_trace->flag_changed)
5245 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5246 return -EINVAL;
5247
5248 if (enabled)
5249 tr->trace_flags |= mask;
5250 else
5251 tr->trace_flags &= ~mask;
5252
5253 if (mask == TRACE_ITER_RECORD_CMD)
5254 trace_event_enable_cmd_record(enabled);
5255
5256 if (mask == TRACE_ITER_RECORD_TGID) {
5257
5258 if (trace_alloc_tgid_map() < 0) {
5259 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5260 return -ENOMEM;
5261 }
5262
5263 trace_event_enable_tgid_record(enabled);
5264 }
5265
5266 if (mask == TRACE_ITER_EVENT_FORK)
5267 trace_event_follow_fork(tr, enabled);
5268
5269 if (mask == TRACE_ITER_FUNC_FORK)
5270 ftrace_pid_follow_fork(tr, enabled);
5271
5272 if (mask == TRACE_ITER_OVERWRITE) {
5273 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5274#ifdef CONFIG_TRACER_MAX_TRACE
5275 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5276#endif
5277 }
5278
5279 if (mask == TRACE_ITER_PRINTK) {
5280 trace_printk_start_stop_comm(enabled);
5281 trace_printk_control(enabled);
5282 }
5283
5284 return 0;
5285}
5286
5287int trace_set_options(struct trace_array *tr, char *option)
5288{
5289 char *cmp;
5290 int neg = 0;
5291 int ret;
5292 size_t orig_len = strlen(option);
5293 int len;
5294
5295 cmp = strstrip(option);
5296
5297 len = str_has_prefix(cmp, "no");
5298 if (len)
5299 neg = 1;
5300
5301 cmp += len;
5302
5303 mutex_lock(&event_mutex);
5304 mutex_lock(&trace_types_lock);
5305
5306 ret = match_string(trace_options, -1, cmp);
5307 /* If no option could be set, test the specific tracer options */
5308 if (ret < 0)
5309 ret = set_tracer_option(tr, cmp, neg);
5310 else
5311 ret = set_tracer_flag(tr, 1 << ret, !neg);
5312
5313 mutex_unlock(&trace_types_lock);
5314 mutex_unlock(&event_mutex);
5315
5316 /*
5317 * If the first trailing whitespace is replaced with '\0' by strstrip,
5318 * turn it back into a space.
5319 */
5320 if (orig_len > strlen(option))
5321 option[strlen(option)] = ' ';
5322
5323 return ret;
5324}
5325
5326static void __init apply_trace_boot_options(void)
5327{
5328 char *buf = trace_boot_options_buf;
5329 char *option;
5330
5331 while (true) {
5332 option = strsep(&buf, ",");
5333
5334 if (!option)
5335 break;
5336
5337 if (*option)
5338 trace_set_options(&global_trace, option);
5339
5340 /* Put back the comma to allow this to be called again */
5341 if (buf)
5342 *(buf - 1) = ',';
5343 }
5344}
5345
5346static ssize_t
5347tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5348 size_t cnt, loff_t *ppos)
5349{
5350 struct seq_file *m = filp->private_data;
5351 struct trace_array *tr = m->private;
5352 char buf[64];
5353 int ret;
5354
5355 if (cnt >= sizeof(buf))
5356 return -EINVAL;
5357
5358 if (copy_from_user(buf, ubuf, cnt))
5359 return -EFAULT;
5360
5361 buf[cnt] = 0;
5362
5363 ret = trace_set_options(tr, buf);
5364 if (ret < 0)
5365 return ret;
5366
5367 *ppos += cnt;
5368
5369 return cnt;
5370}
5371
5372static int tracing_trace_options_open(struct inode *inode, struct file *file)
5373{
5374 struct trace_array *tr = inode->i_private;
5375 int ret;
5376
5377 ret = tracing_check_open_get_tr(tr);
5378 if (ret)
5379 return ret;
5380
5381 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5382 if (ret < 0)
5383 trace_array_put(tr);
5384
5385 return ret;
5386}
5387
5388static const struct file_operations tracing_iter_fops = {
5389 .open = tracing_trace_options_open,
5390 .read = seq_read,
5391 .llseek = seq_lseek,
5392 .release = tracing_single_release_tr,
5393 .write = tracing_trace_options_write,
5394};
5395
5396static const char readme_msg[] =
5397 "tracing mini-HOWTO:\n\n"
5398 "# echo 0 > tracing_on : quick way to disable tracing\n"
5399 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5400 " Important files:\n"
5401 " trace\t\t\t- The static contents of the buffer\n"
5402 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5403 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5404 " current_tracer\t- function and latency tracers\n"
5405 " available_tracers\t- list of configured tracers for current_tracer\n"
5406 " error_log\t- error log for failed commands (that support it)\n"
5407 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5408 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5409 " trace_clock\t\t- change the clock used to order events\n"
5410 " local: Per cpu clock but may not be synced across CPUs\n"
5411 " global: Synced across CPUs but slows tracing down.\n"
5412 " counter: Not a clock, but just an increment\n"
5413 " uptime: Jiffy counter from time of boot\n"
5414 " perf: Same clock that perf events use\n"
5415#ifdef CONFIG_X86_64
5416 " x86-tsc: TSC cycle counter\n"
5417#endif
5418 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5419 " delta: Delta difference against a buffer-wide timestamp\n"
5420 " absolute: Absolute (standalone) timestamp\n"
5421 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5422 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5423 " tracing_cpumask\t- Limit which CPUs to trace\n"
5424 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5425 "\t\t\t Remove sub-buffer with rmdir\n"
5426 " trace_options\t\t- Set format or modify how tracing happens\n"
5427 "\t\t\t Disable an option by prefixing 'no' to the\n"
5428 "\t\t\t option name\n"
5429 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5430#ifdef CONFIG_DYNAMIC_FTRACE
5431 "\n available_filter_functions - list of functions that can be filtered on\n"
5432 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5433 "\t\t\t functions\n"
5434 "\t accepts: func_full_name or glob-matching-pattern\n"
5435 "\t modules: Can select a group via module\n"
5436 "\t Format: :mod:<module-name>\n"
5437 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5438 "\t triggers: a command to perform when function is hit\n"
5439 "\t Format: <function>:<trigger>[:count]\n"
5440 "\t trigger: traceon, traceoff\n"
5441 "\t\t enable_event:<system>:<event>\n"
5442 "\t\t disable_event:<system>:<event>\n"
5443#ifdef CONFIG_STACKTRACE
5444 "\t\t stacktrace\n"
5445#endif
5446#ifdef CONFIG_TRACER_SNAPSHOT
5447 "\t\t snapshot\n"
5448#endif
5449 "\t\t dump\n"
5450 "\t\t cpudump\n"
5451 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5452 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5453 "\t The first one will disable tracing every time do_fault is hit\n"
5454 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5455 "\t The first time do trap is hit and it disables tracing, the\n"
5456 "\t counter will decrement to 2. If tracing is already disabled,\n"
5457 "\t the counter will not decrement. It only decrements when the\n"
5458 "\t trigger did work\n"
5459 "\t To remove trigger without count:\n"
5460 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5461 "\t To remove trigger with a count:\n"
5462 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5463 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5464 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5465 "\t modules: Can select a group via module command :mod:\n"
5466 "\t Does not accept triggers\n"
5467#endif /* CONFIG_DYNAMIC_FTRACE */
5468#ifdef CONFIG_FUNCTION_TRACER
5469 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5470 "\t\t (function)\n"
5471 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5472 "\t\t (function)\n"
5473#endif
5474#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5475 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5476 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5477 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5478#endif
5479#ifdef CONFIG_TRACER_SNAPSHOT
5480 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5481 "\t\t\t snapshot buffer. Read the contents for more\n"
5482 "\t\t\t information\n"
5483#endif
5484#ifdef CONFIG_STACK_TRACER
5485 " stack_trace\t\t- Shows the max stack trace when active\n"
5486 " stack_max_size\t- Shows current max stack size that was traced\n"
5487 "\t\t\t Write into this file to reset the max size (trigger a\n"
5488 "\t\t\t new trace)\n"
5489#ifdef CONFIG_DYNAMIC_FTRACE
5490 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5491 "\t\t\t traces\n"
5492#endif
5493#endif /* CONFIG_STACK_TRACER */
5494#ifdef CONFIG_DYNAMIC_EVENTS
5495 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5496 "\t\t\t Write into this file to define/undefine new trace events.\n"
5497#endif
5498#ifdef CONFIG_KPROBE_EVENTS
5499 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5500 "\t\t\t Write into this file to define/undefine new trace events.\n"
5501#endif
5502#ifdef CONFIG_UPROBE_EVENTS
5503 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5504 "\t\t\t Write into this file to define/undefine new trace events.\n"
5505#endif
5506#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5507 defined(CONFIG_FPROBE_EVENTS)
5508 "\t accepts: event-definitions (one definition per line)\n"
5509#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5510 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5511 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5512#endif
5513#ifdef CONFIG_FPROBE_EVENTS
5514 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5515 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5516#endif
5517#ifdef CONFIG_HIST_TRIGGERS
5518 "\t s:[synthetic/]<event> <field> [<field>]\n"
5519#endif
5520 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5521 "\t -:[<group>/][<event>]\n"
5522#ifdef CONFIG_KPROBE_EVENTS
5523 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5524 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5525#endif
5526#ifdef CONFIG_UPROBE_EVENTS
5527 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5528#endif
5529 "\t args: <name>=fetcharg[:type]\n"
5530 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5531#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5532 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5533#ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5534 "\t <argname>[->field[->field|.field...]],\n"
5535#endif
5536#else
5537 "\t $stack<index>, $stack, $retval, $comm,\n"
5538#endif
5539 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5540 "\t kernel return probes support: $retval, $arg<N>, $comm\n"
5541 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5542 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5543 "\t symstr, <type>\\[<array-size>\\]\n"
5544#ifdef CONFIG_HIST_TRIGGERS
5545 "\t field: <stype> <name>;\n"
5546 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5547 "\t [unsigned] char/int/long\n"
5548#endif
5549 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5550 "\t of the <attached-group>/<attached-event>.\n"
5551#endif
5552 " events/\t\t- Directory containing all trace event subsystems:\n"
5553 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5554 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5555 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5556 "\t\t\t events\n"
5557 " filter\t\t- If set, only events passing filter are traced\n"
5558 " events/<system>/<event>/\t- Directory containing control files for\n"
5559 "\t\t\t <event>:\n"
5560 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5561 " filter\t\t- If set, only events passing filter are traced\n"
5562 " trigger\t\t- If set, a command to perform when event is hit\n"
5563 "\t Format: <trigger>[:count][if <filter>]\n"
5564 "\t trigger: traceon, traceoff\n"
5565 "\t enable_event:<system>:<event>\n"
5566 "\t disable_event:<system>:<event>\n"
5567#ifdef CONFIG_HIST_TRIGGERS
5568 "\t enable_hist:<system>:<event>\n"
5569 "\t disable_hist:<system>:<event>\n"
5570#endif
5571#ifdef CONFIG_STACKTRACE
5572 "\t\t stacktrace\n"
5573#endif
5574#ifdef CONFIG_TRACER_SNAPSHOT
5575 "\t\t snapshot\n"
5576#endif
5577#ifdef CONFIG_HIST_TRIGGERS
5578 "\t\t hist (see below)\n"
5579#endif
5580 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5581 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5582 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5583 "\t events/block/block_unplug/trigger\n"
5584 "\t The first disables tracing every time block_unplug is hit.\n"
5585 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5586 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5587 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5588 "\t Like function triggers, the counter is only decremented if it\n"
5589 "\t enabled or disabled tracing.\n"
5590 "\t To remove a trigger without a count:\n"
5591 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5592 "\t To remove a trigger with a count:\n"
5593 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5594 "\t Filters can be ignored when removing a trigger.\n"
5595#ifdef CONFIG_HIST_TRIGGERS
5596 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5597 "\t Format: hist:keys=<field1[,field2,...]>\n"
5598 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5599 "\t [:values=<field1[,field2,...]>]\n"
5600 "\t [:sort=<field1[,field2,...]>]\n"
5601 "\t [:size=#entries]\n"
5602 "\t [:pause][:continue][:clear]\n"
5603 "\t [:name=histname1]\n"
5604 "\t [:nohitcount]\n"
5605 "\t [:<handler>.<action>]\n"
5606 "\t [if <filter>]\n\n"
5607 "\t Note, special fields can be used as well:\n"
5608 "\t common_timestamp - to record current timestamp\n"
5609 "\t common_cpu - to record the CPU the event happened on\n"
5610 "\n"
5611 "\t A hist trigger variable can be:\n"
5612 "\t - a reference to a field e.g. x=current_timestamp,\n"
5613 "\t - a reference to another variable e.g. y=$x,\n"
5614 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5615 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5616 "\n"
5617 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5618 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5619 "\t variable reference, field or numeric literal.\n"
5620 "\n"
5621 "\t When a matching event is hit, an entry is added to a hash\n"
5622 "\t table using the key(s) and value(s) named, and the value of a\n"
5623 "\t sum called 'hitcount' is incremented. Keys and values\n"
5624 "\t correspond to fields in the event's format description. Keys\n"
5625 "\t can be any field, or the special string 'common_stacktrace'.\n"
5626 "\t Compound keys consisting of up to two fields can be specified\n"
5627 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5628 "\t fields. Sort keys consisting of up to two fields can be\n"
5629 "\t specified using the 'sort' keyword. The sort direction can\n"
5630 "\t be modified by appending '.descending' or '.ascending' to a\n"
5631 "\t sort field. The 'size' parameter can be used to specify more\n"
5632 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5633 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5634 "\t its histogram data will be shared with other triggers of the\n"
5635 "\t same name, and trigger hits will update this common data.\n\n"
5636 "\t Reading the 'hist' file for the event will dump the hash\n"
5637 "\t table in its entirety to stdout. If there are multiple hist\n"
5638 "\t triggers attached to an event, there will be a table for each\n"
5639 "\t trigger in the output. The table displayed for a named\n"
5640 "\t trigger will be the same as any other instance having the\n"
5641 "\t same name. The default format used to display a given field\n"
5642 "\t can be modified by appending any of the following modifiers\n"
5643 "\t to the field name, as applicable:\n\n"
5644 "\t .hex display a number as a hex value\n"
5645 "\t .sym display an address as a symbol\n"
5646 "\t .sym-offset display an address as a symbol and offset\n"
5647 "\t .execname display a common_pid as a program name\n"
5648 "\t .syscall display a syscall id as a syscall name\n"
5649 "\t .log2 display log2 value rather than raw number\n"
5650 "\t .buckets=size display values in groups of size rather than raw number\n"
5651 "\t .usecs display a common_timestamp in microseconds\n"
5652 "\t .percent display a number of percentage value\n"
5653 "\t .graph display a bar-graph of a value\n\n"
5654 "\t The 'pause' parameter can be used to pause an existing hist\n"
5655 "\t trigger or to start a hist trigger but not log any events\n"
5656 "\t until told to do so. 'continue' can be used to start or\n"
5657 "\t restart a paused hist trigger.\n\n"
5658 "\t The 'clear' parameter will clear the contents of a running\n"
5659 "\t hist trigger and leave its current paused/active state\n"
5660 "\t unchanged.\n\n"
5661 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5662 "\t raw hitcount in the histogram.\n\n"
5663 "\t The enable_hist and disable_hist triggers can be used to\n"
5664 "\t have one event conditionally start and stop another event's\n"
5665 "\t already-attached hist trigger. The syntax is analogous to\n"
5666 "\t the enable_event and disable_event triggers.\n\n"
5667 "\t Hist trigger handlers and actions are executed whenever a\n"
5668 "\t a histogram entry is added or updated. They take the form:\n\n"
5669 "\t <handler>.<action>\n\n"
5670 "\t The available handlers are:\n\n"
5671 "\t onmatch(matching.event) - invoke on addition or update\n"
5672 "\t onmax(var) - invoke if var exceeds current max\n"
5673 "\t onchange(var) - invoke action if var changes\n\n"
5674 "\t The available actions are:\n\n"
5675 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5676 "\t save(field,...) - save current event fields\n"
5677#ifdef CONFIG_TRACER_SNAPSHOT
5678 "\t snapshot() - snapshot the trace buffer\n\n"
5679#endif
5680#ifdef CONFIG_SYNTH_EVENTS
5681 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5682 "\t Write into this file to define/undefine new synthetic events.\n"
5683 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5684#endif
5685#endif
5686;
5687
5688static ssize_t
5689tracing_readme_read(struct file *filp, char __user *ubuf,
5690 size_t cnt, loff_t *ppos)
5691{
5692 return simple_read_from_buffer(ubuf, cnt, ppos,
5693 readme_msg, strlen(readme_msg));
5694}
5695
5696static const struct file_operations tracing_readme_fops = {
5697 .open = tracing_open_generic,
5698 .read = tracing_readme_read,
5699 .llseek = generic_file_llseek,
5700};
5701
5702#ifdef CONFIG_TRACE_EVAL_MAP_FILE
5703static union trace_eval_map_item *
5704update_eval_map(union trace_eval_map_item *ptr)
5705{
5706 if (!ptr->map.eval_string) {
5707 if (ptr->tail.next) {
5708 ptr = ptr->tail.next;
5709 /* Set ptr to the next real item (skip head) */
5710 ptr++;
5711 } else
5712 return NULL;
5713 }
5714 return ptr;
5715}
5716
5717static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5718{
5719 union trace_eval_map_item *ptr = v;
5720
5721 /*
5722 * Paranoid! If ptr points to end, we don't want to increment past it.
5723 * This really should never happen.
5724 */
5725 (*pos)++;
5726 ptr = update_eval_map(ptr);
5727 if (WARN_ON_ONCE(!ptr))
5728 return NULL;
5729
5730 ptr++;
5731 ptr = update_eval_map(ptr);
5732
5733 return ptr;
5734}
5735
5736static void *eval_map_start(struct seq_file *m, loff_t *pos)
5737{
5738 union trace_eval_map_item *v;
5739 loff_t l = 0;
5740
5741 mutex_lock(&trace_eval_mutex);
5742
5743 v = trace_eval_maps;
5744 if (v)
5745 v++;
5746
5747 while (v && l < *pos) {
5748 v = eval_map_next(m, v, &l);
5749 }
5750
5751 return v;
5752}
5753
5754static void eval_map_stop(struct seq_file *m, void *v)
5755{
5756 mutex_unlock(&trace_eval_mutex);
5757}
5758
5759static int eval_map_show(struct seq_file *m, void *v)
5760{
5761 union trace_eval_map_item *ptr = v;
5762
5763 seq_printf(m, "%s %ld (%s)\n",
5764 ptr->map.eval_string, ptr->map.eval_value,
5765 ptr->map.system);
5766
5767 return 0;
5768}
5769
5770static const struct seq_operations tracing_eval_map_seq_ops = {
5771 .start = eval_map_start,
5772 .next = eval_map_next,
5773 .stop = eval_map_stop,
5774 .show = eval_map_show,
5775};
5776
5777static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5778{
5779 int ret;
5780
5781 ret = tracing_check_open_get_tr(NULL);
5782 if (ret)
5783 return ret;
5784
5785 return seq_open(filp, &tracing_eval_map_seq_ops);
5786}
5787
5788static const struct file_operations tracing_eval_map_fops = {
5789 .open = tracing_eval_map_open,
5790 .read = seq_read,
5791 .llseek = seq_lseek,
5792 .release = seq_release,
5793};
5794
5795static inline union trace_eval_map_item *
5796trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
5797{
5798 /* Return tail of array given the head */
5799 return ptr + ptr->head.length + 1;
5800}
5801
5802static void
5803trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
5804 int len)
5805{
5806 struct trace_eval_map **stop;
5807 struct trace_eval_map **map;
5808 union trace_eval_map_item *map_array;
5809 union trace_eval_map_item *ptr;
5810
5811 stop = start + len;
5812
5813 /*
5814 * The trace_eval_maps contains the map plus a head and tail item,
5815 * where the head holds the module and length of array, and the
5816 * tail holds a pointer to the next list.
5817 */
5818 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
5819 if (!map_array) {
5820 pr_warn("Unable to allocate trace eval mapping\n");
5821 return;
5822 }
5823
5824 mutex_lock(&trace_eval_mutex);
5825
5826 if (!trace_eval_maps)
5827 trace_eval_maps = map_array;
5828 else {
5829 ptr = trace_eval_maps;
5830 for (;;) {
5831 ptr = trace_eval_jmp_to_tail(ptr);
5832 if (!ptr->tail.next)
5833 break;
5834 ptr = ptr->tail.next;
5835
5836 }
5837 ptr->tail.next = map_array;
5838 }
5839 map_array->head.mod = mod;
5840 map_array->head.length = len;
5841 map_array++;
5842
5843 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
5844 map_array->map = **map;
5845 map_array++;
5846 }
5847 memset(map_array, 0, sizeof(*map_array));
5848
5849 mutex_unlock(&trace_eval_mutex);
5850}
5851
5852static void trace_create_eval_file(struct dentry *d_tracer)
5853{
5854 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
5855 NULL, &tracing_eval_map_fops);
5856}
5857
5858#else /* CONFIG_TRACE_EVAL_MAP_FILE */
5859static inline void trace_create_eval_file(struct dentry *d_tracer) { }
5860static inline void trace_insert_eval_map_file(struct module *mod,
5861 struct trace_eval_map **start, int len) { }
5862#endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
5863
5864static void trace_insert_eval_map(struct module *mod,
5865 struct trace_eval_map **start, int len)
5866{
5867 struct trace_eval_map **map;
5868
5869 if (len <= 0)
5870 return;
5871
5872 map = start;
5873
5874 trace_event_eval_update(map, len);
5875
5876 trace_insert_eval_map_file(mod, start, len);
5877}
5878
5879static ssize_t
5880tracing_set_trace_read(struct file *filp, char __user *ubuf,
5881 size_t cnt, loff_t *ppos)
5882{
5883 struct trace_array *tr = filp->private_data;
5884 char buf[MAX_TRACER_SIZE+2];
5885 int r;
5886
5887 mutex_lock(&trace_types_lock);
5888 r = sprintf(buf, "%s\n", tr->current_trace->name);
5889 mutex_unlock(&trace_types_lock);
5890
5891 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5892}
5893
5894int tracer_init(struct tracer *t, struct trace_array *tr)
5895{
5896 tracing_reset_online_cpus(&tr->array_buffer);
5897 return t->init(tr);
5898}
5899
5900static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
5901{
5902 int cpu;
5903
5904 for_each_tracing_cpu(cpu)
5905 per_cpu_ptr(buf->data, cpu)->entries = val;
5906}
5907
5908static void update_buffer_entries(struct array_buffer *buf, int cpu)
5909{
5910 if (cpu == RING_BUFFER_ALL_CPUS) {
5911 set_buffer_entries(buf, ring_buffer_size(buf->buffer, 0));
5912 } else {
5913 per_cpu_ptr(buf->data, cpu)->entries = ring_buffer_size(buf->buffer, cpu);
5914 }
5915}
5916
5917#ifdef CONFIG_TRACER_MAX_TRACE
5918/* resize @tr's buffer to the size of @size_tr's entries */
5919static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
5920 struct array_buffer *size_buf, int cpu_id)
5921{
5922 int cpu, ret = 0;
5923
5924 if (cpu_id == RING_BUFFER_ALL_CPUS) {
5925 for_each_tracing_cpu(cpu) {
5926 ret = ring_buffer_resize(trace_buf->buffer,
5927 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
5928 if (ret < 0)
5929 break;
5930 per_cpu_ptr(trace_buf->data, cpu)->entries =
5931 per_cpu_ptr(size_buf->data, cpu)->entries;
5932 }
5933 } else {
5934 ret = ring_buffer_resize(trace_buf->buffer,
5935 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
5936 if (ret == 0)
5937 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
5938 per_cpu_ptr(size_buf->data, cpu_id)->entries;
5939 }
5940
5941 return ret;
5942}
5943#endif /* CONFIG_TRACER_MAX_TRACE */
5944
5945static int __tracing_resize_ring_buffer(struct trace_array *tr,
5946 unsigned long size, int cpu)
5947{
5948 int ret;
5949
5950 /*
5951 * If kernel or user changes the size of the ring buffer
5952 * we use the size that was given, and we can forget about
5953 * expanding it later.
5954 */
5955 trace_set_ring_buffer_expanded(tr);
5956
5957 /* May be called before buffers are initialized */
5958 if (!tr->array_buffer.buffer)
5959 return 0;
5960
5961 /* Do not allow tracing while resizing ring buffer */
5962 tracing_stop_tr(tr);
5963
5964 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
5965 if (ret < 0)
5966 goto out_start;
5967
5968#ifdef CONFIG_TRACER_MAX_TRACE
5969 if (!tr->allocated_snapshot)
5970 goto out;
5971
5972 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
5973 if (ret < 0) {
5974 int r = resize_buffer_duplicate_size(&tr->array_buffer,
5975 &tr->array_buffer, cpu);
5976 if (r < 0) {
5977 /*
5978 * AARGH! We are left with different
5979 * size max buffer!!!!
5980 * The max buffer is our "snapshot" buffer.
5981 * When a tracer needs a snapshot (one of the
5982 * latency tracers), it swaps the max buffer
5983 * with the saved snap shot. We succeeded to
5984 * update the size of the main buffer, but failed to
5985 * update the size of the max buffer. But when we tried
5986 * to reset the main buffer to the original size, we
5987 * failed there too. This is very unlikely to
5988 * happen, but if it does, warn and kill all
5989 * tracing.
5990 */
5991 WARN_ON(1);
5992 tracing_disabled = 1;
5993 }
5994 goto out_start;
5995 }
5996
5997 update_buffer_entries(&tr->max_buffer, cpu);
5998
5999 out:
6000#endif /* CONFIG_TRACER_MAX_TRACE */
6001
6002 update_buffer_entries(&tr->array_buffer, cpu);
6003 out_start:
6004 tracing_start_tr(tr);
6005 return ret;
6006}
6007
6008ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6009 unsigned long size, int cpu_id)
6010{
6011 int ret;
6012
6013 mutex_lock(&trace_types_lock);
6014
6015 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6016 /* make sure, this cpu is enabled in the mask */
6017 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6018 ret = -EINVAL;
6019 goto out;
6020 }
6021 }
6022
6023 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6024 if (ret < 0)
6025 ret = -ENOMEM;
6026
6027out:
6028 mutex_unlock(&trace_types_lock);
6029
6030 return ret;
6031}
6032
6033
6034/**
6035 * tracing_update_buffers - used by tracing facility to expand ring buffers
6036 * @tr: The tracing instance
6037 *
6038 * To save on memory when the tracing is never used on a system with it
6039 * configured in. The ring buffers are set to a minimum size. But once
6040 * a user starts to use the tracing facility, then they need to grow
6041 * to their default size.
6042 *
6043 * This function is to be called when a tracer is about to be used.
6044 */
6045int tracing_update_buffers(struct trace_array *tr)
6046{
6047 int ret = 0;
6048
6049 mutex_lock(&trace_types_lock);
6050 if (!tr->ring_buffer_expanded)
6051 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6052 RING_BUFFER_ALL_CPUS);
6053 mutex_unlock(&trace_types_lock);
6054
6055 return ret;
6056}
6057
6058struct trace_option_dentry;
6059
6060static void
6061create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6062
6063/*
6064 * Used to clear out the tracer before deletion of an instance.
6065 * Must have trace_types_lock held.
6066 */
6067static void tracing_set_nop(struct trace_array *tr)
6068{
6069 if (tr->current_trace == &nop_trace)
6070 return;
6071
6072 tr->current_trace->enabled--;
6073
6074 if (tr->current_trace->reset)
6075 tr->current_trace->reset(tr);
6076
6077 tr->current_trace = &nop_trace;
6078}
6079
6080static bool tracer_options_updated;
6081
6082static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6083{
6084 /* Only enable if the directory has been created already. */
6085 if (!tr->dir)
6086 return;
6087
6088 /* Only create trace option files after update_tracer_options finish */
6089 if (!tracer_options_updated)
6090 return;
6091
6092 create_trace_option_files(tr, t);
6093}
6094
6095int tracing_set_tracer(struct trace_array *tr, const char *buf)
6096{
6097 struct tracer *t;
6098#ifdef CONFIG_TRACER_MAX_TRACE
6099 bool had_max_tr;
6100#endif
6101 int ret = 0;
6102
6103 mutex_lock(&trace_types_lock);
6104
6105 if (!tr->ring_buffer_expanded) {
6106 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6107 RING_BUFFER_ALL_CPUS);
6108 if (ret < 0)
6109 goto out;
6110 ret = 0;
6111 }
6112
6113 for (t = trace_types; t; t = t->next) {
6114 if (strcmp(t->name, buf) == 0)
6115 break;
6116 }
6117 if (!t) {
6118 ret = -EINVAL;
6119 goto out;
6120 }
6121 if (t == tr->current_trace)
6122 goto out;
6123
6124#ifdef CONFIG_TRACER_SNAPSHOT
6125 if (t->use_max_tr) {
6126 local_irq_disable();
6127 arch_spin_lock(&tr->max_lock);
6128 if (tr->cond_snapshot)
6129 ret = -EBUSY;
6130 arch_spin_unlock(&tr->max_lock);
6131 local_irq_enable();
6132 if (ret)
6133 goto out;
6134 }
6135#endif
6136 /* Some tracers won't work on kernel command line */
6137 if (system_state < SYSTEM_RUNNING && t->noboot) {
6138 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6139 t->name);
6140 goto out;
6141 }
6142
6143 /* Some tracers are only allowed for the top level buffer */
6144 if (!trace_ok_for_array(t, tr)) {
6145 ret = -EINVAL;
6146 goto out;
6147 }
6148
6149 /* If trace pipe files are being read, we can't change the tracer */
6150 if (tr->trace_ref) {
6151 ret = -EBUSY;
6152 goto out;
6153 }
6154
6155 trace_branch_disable();
6156
6157 tr->current_trace->enabled--;
6158
6159 if (tr->current_trace->reset)
6160 tr->current_trace->reset(tr);
6161
6162#ifdef CONFIG_TRACER_MAX_TRACE
6163 had_max_tr = tr->current_trace->use_max_tr;
6164
6165 /* Current trace needs to be nop_trace before synchronize_rcu */
6166 tr->current_trace = &nop_trace;
6167
6168 if (had_max_tr && !t->use_max_tr) {
6169 /*
6170 * We need to make sure that the update_max_tr sees that
6171 * current_trace changed to nop_trace to keep it from
6172 * swapping the buffers after we resize it.
6173 * The update_max_tr is called from interrupts disabled
6174 * so a synchronized_sched() is sufficient.
6175 */
6176 synchronize_rcu();
6177 free_snapshot(tr);
6178 tracing_disarm_snapshot(tr);
6179 }
6180
6181 if (!had_max_tr && t->use_max_tr) {
6182 ret = tracing_arm_snapshot_locked(tr);
6183 if (ret)
6184 goto out;
6185 }
6186#else
6187 tr->current_trace = &nop_trace;
6188#endif
6189
6190 if (t->init) {
6191 ret = tracer_init(t, tr);
6192 if (ret) {
6193#ifdef CONFIG_TRACER_MAX_TRACE
6194 if (t->use_max_tr)
6195 tracing_disarm_snapshot(tr);
6196#endif
6197 goto out;
6198 }
6199 }
6200
6201 tr->current_trace = t;
6202 tr->current_trace->enabled++;
6203 trace_branch_enable(tr);
6204 out:
6205 mutex_unlock(&trace_types_lock);
6206
6207 return ret;
6208}
6209
6210static ssize_t
6211tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6212 size_t cnt, loff_t *ppos)
6213{
6214 struct trace_array *tr = filp->private_data;
6215 char buf[MAX_TRACER_SIZE+1];
6216 char *name;
6217 size_t ret;
6218 int err;
6219
6220 ret = cnt;
6221
6222 if (cnt > MAX_TRACER_SIZE)
6223 cnt = MAX_TRACER_SIZE;
6224
6225 if (copy_from_user(buf, ubuf, cnt))
6226 return -EFAULT;
6227
6228 buf[cnt] = 0;
6229
6230 name = strim(buf);
6231
6232 err = tracing_set_tracer(tr, name);
6233 if (err)
6234 return err;
6235
6236 *ppos += ret;
6237
6238 return ret;
6239}
6240
6241static ssize_t
6242tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6243 size_t cnt, loff_t *ppos)
6244{
6245 char buf[64];
6246 int r;
6247
6248 r = snprintf(buf, sizeof(buf), "%ld\n",
6249 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6250 if (r > sizeof(buf))
6251 r = sizeof(buf);
6252 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6253}
6254
6255static ssize_t
6256tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6257 size_t cnt, loff_t *ppos)
6258{
6259 unsigned long val;
6260 int ret;
6261
6262 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6263 if (ret)
6264 return ret;
6265
6266 *ptr = val * 1000;
6267
6268 return cnt;
6269}
6270
6271static ssize_t
6272tracing_thresh_read(struct file *filp, char __user *ubuf,
6273 size_t cnt, loff_t *ppos)
6274{
6275 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6276}
6277
6278static ssize_t
6279tracing_thresh_write(struct file *filp, const char __user *ubuf,
6280 size_t cnt, loff_t *ppos)
6281{
6282 struct trace_array *tr = filp->private_data;
6283 int ret;
6284
6285 mutex_lock(&trace_types_lock);
6286 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6287 if (ret < 0)
6288 goto out;
6289
6290 if (tr->current_trace->update_thresh) {
6291 ret = tr->current_trace->update_thresh(tr);
6292 if (ret < 0)
6293 goto out;
6294 }
6295
6296 ret = cnt;
6297out:
6298 mutex_unlock(&trace_types_lock);
6299
6300 return ret;
6301}
6302
6303#ifdef CONFIG_TRACER_MAX_TRACE
6304
6305static ssize_t
6306tracing_max_lat_read(struct file *filp, char __user *ubuf,
6307 size_t cnt, loff_t *ppos)
6308{
6309 struct trace_array *tr = filp->private_data;
6310
6311 return tracing_nsecs_read(&tr->max_latency, ubuf, cnt, ppos);
6312}
6313
6314static ssize_t
6315tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6316 size_t cnt, loff_t *ppos)
6317{
6318 struct trace_array *tr = filp->private_data;
6319
6320 return tracing_nsecs_write(&tr->max_latency, ubuf, cnt, ppos);
6321}
6322
6323#endif
6324
6325static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6326{
6327 if (cpu == RING_BUFFER_ALL_CPUS) {
6328 if (cpumask_empty(tr->pipe_cpumask)) {
6329 cpumask_setall(tr->pipe_cpumask);
6330 return 0;
6331 }
6332 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6333 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6334 return 0;
6335 }
6336 return -EBUSY;
6337}
6338
6339static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6340{
6341 if (cpu == RING_BUFFER_ALL_CPUS) {
6342 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6343 cpumask_clear(tr->pipe_cpumask);
6344 } else {
6345 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6346 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6347 }
6348}
6349
6350static int tracing_open_pipe(struct inode *inode, struct file *filp)
6351{
6352 struct trace_array *tr = inode->i_private;
6353 struct trace_iterator *iter;
6354 int cpu;
6355 int ret;
6356
6357 ret = tracing_check_open_get_tr(tr);
6358 if (ret)
6359 return ret;
6360
6361 mutex_lock(&trace_types_lock);
6362 cpu = tracing_get_cpu(inode);
6363 ret = open_pipe_on_cpu(tr, cpu);
6364 if (ret)
6365 goto fail_pipe_on_cpu;
6366
6367 /* create a buffer to store the information to pass to userspace */
6368 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6369 if (!iter) {
6370 ret = -ENOMEM;
6371 goto fail_alloc_iter;
6372 }
6373
6374 trace_seq_init(&iter->seq);
6375 iter->trace = tr->current_trace;
6376
6377 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6378 ret = -ENOMEM;
6379 goto fail;
6380 }
6381
6382 /* trace pipe does not show start of buffer */
6383 cpumask_setall(iter->started);
6384
6385 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6386 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6387
6388 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6389 if (trace_clocks[tr->clock_id].in_ns)
6390 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6391
6392 iter->tr = tr;
6393 iter->array_buffer = &tr->array_buffer;
6394 iter->cpu_file = cpu;
6395 mutex_init(&iter->mutex);
6396 filp->private_data = iter;
6397
6398 if (iter->trace->pipe_open)
6399 iter->trace->pipe_open(iter);
6400
6401 nonseekable_open(inode, filp);
6402
6403 tr->trace_ref++;
6404
6405 mutex_unlock(&trace_types_lock);
6406 return ret;
6407
6408fail:
6409 kfree(iter);
6410fail_alloc_iter:
6411 close_pipe_on_cpu(tr, cpu);
6412fail_pipe_on_cpu:
6413 __trace_array_put(tr);
6414 mutex_unlock(&trace_types_lock);
6415 return ret;
6416}
6417
6418static int tracing_release_pipe(struct inode *inode, struct file *file)
6419{
6420 struct trace_iterator *iter = file->private_data;
6421 struct trace_array *tr = inode->i_private;
6422
6423 mutex_lock(&trace_types_lock);
6424
6425 tr->trace_ref--;
6426
6427 if (iter->trace->pipe_close)
6428 iter->trace->pipe_close(iter);
6429 close_pipe_on_cpu(tr, iter->cpu_file);
6430 mutex_unlock(&trace_types_lock);
6431
6432 free_trace_iter_content(iter);
6433 kfree(iter);
6434
6435 trace_array_put(tr);
6436
6437 return 0;
6438}
6439
6440static __poll_t
6441trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6442{
6443 struct trace_array *tr = iter->tr;
6444
6445 /* Iterators are static, they should be filled or empty */
6446 if (trace_buffer_iter(iter, iter->cpu_file))
6447 return EPOLLIN | EPOLLRDNORM;
6448
6449 if (tr->trace_flags & TRACE_ITER_BLOCK)
6450 /*
6451 * Always select as readable when in blocking mode
6452 */
6453 return EPOLLIN | EPOLLRDNORM;
6454 else
6455 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6456 filp, poll_table, iter->tr->buffer_percent);
6457}
6458
6459static __poll_t
6460tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6461{
6462 struct trace_iterator *iter = filp->private_data;
6463
6464 return trace_poll(iter, filp, poll_table);
6465}
6466
6467/* Must be called with iter->mutex held. */
6468static int tracing_wait_pipe(struct file *filp)
6469{
6470 struct trace_iterator *iter = filp->private_data;
6471 int ret;
6472
6473 while (trace_empty(iter)) {
6474
6475 if ((filp->f_flags & O_NONBLOCK)) {
6476 return -EAGAIN;
6477 }
6478
6479 /*
6480 * We block until we read something and tracing is disabled.
6481 * We still block if tracing is disabled, but we have never
6482 * read anything. This allows a user to cat this file, and
6483 * then enable tracing. But after we have read something,
6484 * we give an EOF when tracing is again disabled.
6485 *
6486 * iter->pos will be 0 if we haven't read anything.
6487 */
6488 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6489 break;
6490
6491 mutex_unlock(&iter->mutex);
6492
6493 ret = wait_on_pipe(iter, 0);
6494
6495 mutex_lock(&iter->mutex);
6496
6497 if (ret)
6498 return ret;
6499 }
6500
6501 return 1;
6502}
6503
6504/*
6505 * Consumer reader.
6506 */
6507static ssize_t
6508tracing_read_pipe(struct file *filp, char __user *ubuf,
6509 size_t cnt, loff_t *ppos)
6510{
6511 struct trace_iterator *iter = filp->private_data;
6512 ssize_t sret;
6513
6514 /*
6515 * Avoid more than one consumer on a single file descriptor
6516 * This is just a matter of traces coherency, the ring buffer itself
6517 * is protected.
6518 */
6519 mutex_lock(&iter->mutex);
6520
6521 /* return any leftover data */
6522 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6523 if (sret != -EBUSY)
6524 goto out;
6525
6526 trace_seq_init(&iter->seq);
6527
6528 if (iter->trace->read) {
6529 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6530 if (sret)
6531 goto out;
6532 }
6533
6534waitagain:
6535 sret = tracing_wait_pipe(filp);
6536 if (sret <= 0)
6537 goto out;
6538
6539 /* stop when tracing is finished */
6540 if (trace_empty(iter)) {
6541 sret = 0;
6542 goto out;
6543 }
6544
6545 if (cnt >= TRACE_SEQ_BUFFER_SIZE)
6546 cnt = TRACE_SEQ_BUFFER_SIZE - 1;
6547
6548 /* reset all but tr, trace, and overruns */
6549 trace_iterator_reset(iter);
6550 cpumask_clear(iter->started);
6551 trace_seq_init(&iter->seq);
6552
6553 trace_event_read_lock();
6554 trace_access_lock(iter->cpu_file);
6555 while (trace_find_next_entry_inc(iter) != NULL) {
6556 enum print_line_t ret;
6557 int save_len = iter->seq.seq.len;
6558
6559 ret = print_trace_line(iter);
6560 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6561 /*
6562 * If one print_trace_line() fills entire trace_seq in one shot,
6563 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6564 * In this case, we need to consume it, otherwise, loop will peek
6565 * this event next time, resulting in an infinite loop.
6566 */
6567 if (save_len == 0) {
6568 iter->seq.full = 0;
6569 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6570 trace_consume(iter);
6571 break;
6572 }
6573
6574 /* In other cases, don't print partial lines */
6575 iter->seq.seq.len = save_len;
6576 break;
6577 }
6578 if (ret != TRACE_TYPE_NO_CONSUME)
6579 trace_consume(iter);
6580
6581 if (trace_seq_used(&iter->seq) >= cnt)
6582 break;
6583
6584 /*
6585 * Setting the full flag means we reached the trace_seq buffer
6586 * size and we should leave by partial output condition above.
6587 * One of the trace_seq_* functions is not used properly.
6588 */
6589 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6590 iter->ent->type);
6591 }
6592 trace_access_unlock(iter->cpu_file);
6593 trace_event_read_unlock();
6594
6595 /* Now copy what we have to the user */
6596 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6597 if (iter->seq.readpos >= trace_seq_used(&iter->seq))
6598 trace_seq_init(&iter->seq);
6599
6600 /*
6601 * If there was nothing to send to user, in spite of consuming trace
6602 * entries, go back to wait for more entries.
6603 */
6604 if (sret == -EBUSY)
6605 goto waitagain;
6606
6607out:
6608 mutex_unlock(&iter->mutex);
6609
6610 return sret;
6611}
6612
6613static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6614 unsigned int idx)
6615{
6616 __free_page(spd->pages[idx]);
6617}
6618
6619static size_t
6620tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6621{
6622 size_t count;
6623 int save_len;
6624 int ret;
6625
6626 /* Seq buffer is page-sized, exactly what we need. */
6627 for (;;) {
6628 save_len = iter->seq.seq.len;
6629 ret = print_trace_line(iter);
6630
6631 if (trace_seq_has_overflowed(&iter->seq)) {
6632 iter->seq.seq.len = save_len;
6633 break;
6634 }
6635
6636 /*
6637 * This should not be hit, because it should only
6638 * be set if the iter->seq overflowed. But check it
6639 * anyway to be safe.
6640 */
6641 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6642 iter->seq.seq.len = save_len;
6643 break;
6644 }
6645
6646 count = trace_seq_used(&iter->seq) - save_len;
6647 if (rem < count) {
6648 rem = 0;
6649 iter->seq.seq.len = save_len;
6650 break;
6651 }
6652
6653 if (ret != TRACE_TYPE_NO_CONSUME)
6654 trace_consume(iter);
6655 rem -= count;
6656 if (!trace_find_next_entry_inc(iter)) {
6657 rem = 0;
6658 iter->ent = NULL;
6659 break;
6660 }
6661 }
6662
6663 return rem;
6664}
6665
6666static ssize_t tracing_splice_read_pipe(struct file *filp,
6667 loff_t *ppos,
6668 struct pipe_inode_info *pipe,
6669 size_t len,
6670 unsigned int flags)
6671{
6672 struct page *pages_def[PIPE_DEF_BUFFERS];
6673 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6674 struct trace_iterator *iter = filp->private_data;
6675 struct splice_pipe_desc spd = {
6676 .pages = pages_def,
6677 .partial = partial_def,
6678 .nr_pages = 0, /* This gets updated below. */
6679 .nr_pages_max = PIPE_DEF_BUFFERS,
6680 .ops = &default_pipe_buf_ops,
6681 .spd_release = tracing_spd_release_pipe,
6682 };
6683 ssize_t ret;
6684 size_t rem;
6685 unsigned int i;
6686
6687 if (splice_grow_spd(pipe, &spd))
6688 return -ENOMEM;
6689
6690 mutex_lock(&iter->mutex);
6691
6692 if (iter->trace->splice_read) {
6693 ret = iter->trace->splice_read(iter, filp,
6694 ppos, pipe, len, flags);
6695 if (ret)
6696 goto out_err;
6697 }
6698
6699 ret = tracing_wait_pipe(filp);
6700 if (ret <= 0)
6701 goto out_err;
6702
6703 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6704 ret = -EFAULT;
6705 goto out_err;
6706 }
6707
6708 trace_event_read_lock();
6709 trace_access_lock(iter->cpu_file);
6710
6711 /* Fill as many pages as possible. */
6712 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6713 spd.pages[i] = alloc_page(GFP_KERNEL);
6714 if (!spd.pages[i])
6715 break;
6716
6717 rem = tracing_fill_pipe_page(rem, iter);
6718
6719 /* Copy the data into the page, so we can start over. */
6720 ret = trace_seq_to_buffer(&iter->seq,
6721 page_address(spd.pages[i]),
6722 trace_seq_used(&iter->seq));
6723 if (ret < 0) {
6724 __free_page(spd.pages[i]);
6725 break;
6726 }
6727 spd.partial[i].offset = 0;
6728 spd.partial[i].len = trace_seq_used(&iter->seq);
6729
6730 trace_seq_init(&iter->seq);
6731 }
6732
6733 trace_access_unlock(iter->cpu_file);
6734 trace_event_read_unlock();
6735 mutex_unlock(&iter->mutex);
6736
6737 spd.nr_pages = i;
6738
6739 if (i)
6740 ret = splice_to_pipe(pipe, &spd);
6741 else
6742 ret = 0;
6743out:
6744 splice_shrink_spd(&spd);
6745 return ret;
6746
6747out_err:
6748 mutex_unlock(&iter->mutex);
6749 goto out;
6750}
6751
6752static ssize_t
6753tracing_entries_read(struct file *filp, char __user *ubuf,
6754 size_t cnt, loff_t *ppos)
6755{
6756 struct inode *inode = file_inode(filp);
6757 struct trace_array *tr = inode->i_private;
6758 int cpu = tracing_get_cpu(inode);
6759 char buf[64];
6760 int r = 0;
6761 ssize_t ret;
6762
6763 mutex_lock(&trace_types_lock);
6764
6765 if (cpu == RING_BUFFER_ALL_CPUS) {
6766 int cpu, buf_size_same;
6767 unsigned long size;
6768
6769 size = 0;
6770 buf_size_same = 1;
6771 /* check if all cpu sizes are same */
6772 for_each_tracing_cpu(cpu) {
6773 /* fill in the size from first enabled cpu */
6774 if (size == 0)
6775 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6776 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6777 buf_size_same = 0;
6778 break;
6779 }
6780 }
6781
6782 if (buf_size_same) {
6783 if (!tr->ring_buffer_expanded)
6784 r = sprintf(buf, "%lu (expanded: %lu)\n",
6785 size >> 10,
6786 trace_buf_size >> 10);
6787 else
6788 r = sprintf(buf, "%lu\n", size >> 10);
6789 } else
6790 r = sprintf(buf, "X\n");
6791 } else
6792 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6793
6794 mutex_unlock(&trace_types_lock);
6795
6796 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6797 return ret;
6798}
6799
6800static ssize_t
6801tracing_entries_write(struct file *filp, const char __user *ubuf,
6802 size_t cnt, loff_t *ppos)
6803{
6804 struct inode *inode = file_inode(filp);
6805 struct trace_array *tr = inode->i_private;
6806 unsigned long val;
6807 int ret;
6808
6809 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6810 if (ret)
6811 return ret;
6812
6813 /* must have at least 1 entry */
6814 if (!val)
6815 return -EINVAL;
6816
6817 /* value is in KB */
6818 val <<= 10;
6819 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6820 if (ret < 0)
6821 return ret;
6822
6823 *ppos += cnt;
6824
6825 return cnt;
6826}
6827
6828static ssize_t
6829tracing_total_entries_read(struct file *filp, char __user *ubuf,
6830 size_t cnt, loff_t *ppos)
6831{
6832 struct trace_array *tr = filp->private_data;
6833 char buf[64];
6834 int r, cpu;
6835 unsigned long size = 0, expanded_size = 0;
6836
6837 mutex_lock(&trace_types_lock);
6838 for_each_tracing_cpu(cpu) {
6839 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6840 if (!tr->ring_buffer_expanded)
6841 expanded_size += trace_buf_size >> 10;
6842 }
6843 if (tr->ring_buffer_expanded)
6844 r = sprintf(buf, "%lu\n", size);
6845 else
6846 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6847 mutex_unlock(&trace_types_lock);
6848
6849 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6850}
6851
6852static ssize_t
6853tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
6854 size_t cnt, loff_t *ppos)
6855{
6856 /*
6857 * There is no need to read what the user has written, this function
6858 * is just to make sure that there is no error when "echo" is used
6859 */
6860
6861 *ppos += cnt;
6862
6863 return cnt;
6864}
6865
6866static int
6867tracing_free_buffer_release(struct inode *inode, struct file *filp)
6868{
6869 struct trace_array *tr = inode->i_private;
6870
6871 /* disable tracing ? */
6872 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
6873 tracer_tracing_off(tr);
6874 /* resize the ring buffer to 0 */
6875 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
6876
6877 trace_array_put(tr);
6878
6879 return 0;
6880}
6881
6882#define TRACE_MARKER_MAX_SIZE 4096
6883
6884static ssize_t
6885tracing_mark_write(struct file *filp, const char __user *ubuf,
6886 size_t cnt, loff_t *fpos)
6887{
6888 struct trace_array *tr = filp->private_data;
6889 struct ring_buffer_event *event;
6890 enum event_trigger_type tt = ETT_NONE;
6891 struct trace_buffer *buffer;
6892 struct print_entry *entry;
6893 int meta_size;
6894 ssize_t written;
6895 size_t size;
6896 int len;
6897
6898/* Used in tracing_mark_raw_write() as well */
6899#define FAULTED_STR "<faulted>"
6900#define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
6901
6902 if (tracing_disabled)
6903 return -EINVAL;
6904
6905 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6906 return -EINVAL;
6907
6908 if ((ssize_t)cnt < 0)
6909 return -EINVAL;
6910
6911 if (cnt > TRACE_MARKER_MAX_SIZE)
6912 cnt = TRACE_MARKER_MAX_SIZE;
6913
6914 meta_size = sizeof(*entry) + 2; /* add '\0' and possible '\n' */
6915 again:
6916 size = cnt + meta_size;
6917
6918 /* If less than "<faulted>", then make sure we can still add that */
6919 if (cnt < FAULTED_SIZE)
6920 size += FAULTED_SIZE - cnt;
6921
6922 buffer = tr->array_buffer.buffer;
6923 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
6924 tracing_gen_ctx());
6925 if (unlikely(!event)) {
6926 /*
6927 * If the size was greater than what was allowed, then
6928 * make it smaller and try again.
6929 */
6930 if (size > ring_buffer_max_event_size(buffer)) {
6931 /* cnt < FAULTED size should never be bigger than max */
6932 if (WARN_ON_ONCE(cnt < FAULTED_SIZE))
6933 return -EBADF;
6934 cnt = ring_buffer_max_event_size(buffer) - meta_size;
6935 /* The above should only happen once */
6936 if (WARN_ON_ONCE(cnt + meta_size == size))
6937 return -EBADF;
6938 goto again;
6939 }
6940
6941 /* Ring buffer disabled, return as if not open for write */
6942 return -EBADF;
6943 }
6944
6945 entry = ring_buffer_event_data(event);
6946 entry->ip = _THIS_IP_;
6947
6948 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
6949 if (len) {
6950 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6951 cnt = FAULTED_SIZE;
6952 written = -EFAULT;
6953 } else
6954 written = cnt;
6955
6956 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
6957 /* do not add \n before testing triggers, but add \0 */
6958 entry->buf[cnt] = '\0';
6959 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
6960 }
6961
6962 if (entry->buf[cnt - 1] != '\n') {
6963 entry->buf[cnt] = '\n';
6964 entry->buf[cnt + 1] = '\0';
6965 } else
6966 entry->buf[cnt] = '\0';
6967
6968 if (static_branch_unlikely(&trace_marker_exports_enabled))
6969 ftrace_exports(event, TRACE_EXPORT_MARKER);
6970 __buffer_unlock_commit(buffer, event);
6971
6972 if (tt)
6973 event_triggers_post_call(tr->trace_marker_file, tt);
6974
6975 return written;
6976}
6977
6978static ssize_t
6979tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
6980 size_t cnt, loff_t *fpos)
6981{
6982 struct trace_array *tr = filp->private_data;
6983 struct ring_buffer_event *event;
6984 struct trace_buffer *buffer;
6985 struct raw_data_entry *entry;
6986 ssize_t written;
6987 int size;
6988 int len;
6989
6990#define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
6991
6992 if (tracing_disabled)
6993 return -EINVAL;
6994
6995 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6996 return -EINVAL;
6997
6998 /* The marker must at least have a tag id */
6999 if (cnt < sizeof(unsigned int))
7000 return -EINVAL;
7001
7002 size = sizeof(*entry) + cnt;
7003 if (cnt < FAULT_SIZE_ID)
7004 size += FAULT_SIZE_ID - cnt;
7005
7006 buffer = tr->array_buffer.buffer;
7007
7008 if (size > ring_buffer_max_event_size(buffer))
7009 return -EINVAL;
7010
7011 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7012 tracing_gen_ctx());
7013 if (!event)
7014 /* Ring buffer disabled, return as if not open for write */
7015 return -EBADF;
7016
7017 entry = ring_buffer_event_data(event);
7018
7019 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7020 if (len) {
7021 entry->id = -1;
7022 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7023 written = -EFAULT;
7024 } else
7025 written = cnt;
7026
7027 __buffer_unlock_commit(buffer, event);
7028
7029 return written;
7030}
7031
7032static int tracing_clock_show(struct seq_file *m, void *v)
7033{
7034 struct trace_array *tr = m->private;
7035 int i;
7036
7037 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7038 seq_printf(m,
7039 "%s%s%s%s", i ? " " : "",
7040 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7041 i == tr->clock_id ? "]" : "");
7042 seq_putc(m, '\n');
7043
7044 return 0;
7045}
7046
7047int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7048{
7049 int i;
7050
7051 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7052 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7053 break;
7054 }
7055 if (i == ARRAY_SIZE(trace_clocks))
7056 return -EINVAL;
7057
7058 mutex_lock(&trace_types_lock);
7059
7060 tr->clock_id = i;
7061
7062 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7063
7064 /*
7065 * New clock may not be consistent with the previous clock.
7066 * Reset the buffer so that it doesn't have incomparable timestamps.
7067 */
7068 tracing_reset_online_cpus(&tr->array_buffer);
7069
7070#ifdef CONFIG_TRACER_MAX_TRACE
7071 if (tr->max_buffer.buffer)
7072 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7073 tracing_reset_online_cpus(&tr->max_buffer);
7074#endif
7075
7076 mutex_unlock(&trace_types_lock);
7077
7078 return 0;
7079}
7080
7081static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7082 size_t cnt, loff_t *fpos)
7083{
7084 struct seq_file *m = filp->private_data;
7085 struct trace_array *tr = m->private;
7086 char buf[64];
7087 const char *clockstr;
7088 int ret;
7089
7090 if (cnt >= sizeof(buf))
7091 return -EINVAL;
7092
7093 if (copy_from_user(buf, ubuf, cnt))
7094 return -EFAULT;
7095
7096 buf[cnt] = 0;
7097
7098 clockstr = strstrip(buf);
7099
7100 ret = tracing_set_clock(tr, clockstr);
7101 if (ret)
7102 return ret;
7103
7104 *fpos += cnt;
7105
7106 return cnt;
7107}
7108
7109static int tracing_clock_open(struct inode *inode, struct file *file)
7110{
7111 struct trace_array *tr = inode->i_private;
7112 int ret;
7113
7114 ret = tracing_check_open_get_tr(tr);
7115 if (ret)
7116 return ret;
7117
7118 ret = single_open(file, tracing_clock_show, inode->i_private);
7119 if (ret < 0)
7120 trace_array_put(tr);
7121
7122 return ret;
7123}
7124
7125static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7126{
7127 struct trace_array *tr = m->private;
7128
7129 mutex_lock(&trace_types_lock);
7130
7131 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7132 seq_puts(m, "delta [absolute]\n");
7133 else
7134 seq_puts(m, "[delta] absolute\n");
7135
7136 mutex_unlock(&trace_types_lock);
7137
7138 return 0;
7139}
7140
7141static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7142{
7143 struct trace_array *tr = inode->i_private;
7144 int ret;
7145
7146 ret = tracing_check_open_get_tr(tr);
7147 if (ret)
7148 return ret;
7149
7150 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7151 if (ret < 0)
7152 trace_array_put(tr);
7153
7154 return ret;
7155}
7156
7157u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7158{
7159 if (rbe == this_cpu_read(trace_buffered_event))
7160 return ring_buffer_time_stamp(buffer);
7161
7162 return ring_buffer_event_time_stamp(buffer, rbe);
7163}
7164
7165/*
7166 * Set or disable using the per CPU trace_buffer_event when possible.
7167 */
7168int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7169{
7170 int ret = 0;
7171
7172 mutex_lock(&trace_types_lock);
7173
7174 if (set && tr->no_filter_buffering_ref++)
7175 goto out;
7176
7177 if (!set) {
7178 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7179 ret = -EINVAL;
7180 goto out;
7181 }
7182
7183 --tr->no_filter_buffering_ref;
7184 }
7185 out:
7186 mutex_unlock(&trace_types_lock);
7187
7188 return ret;
7189}
7190
7191struct ftrace_buffer_info {
7192 struct trace_iterator iter;
7193 void *spare;
7194 unsigned int spare_cpu;
7195 unsigned int spare_size;
7196 unsigned int read;
7197};
7198
7199#ifdef CONFIG_TRACER_SNAPSHOT
7200static int tracing_snapshot_open(struct inode *inode, struct file *file)
7201{
7202 struct trace_array *tr = inode->i_private;
7203 struct trace_iterator *iter;
7204 struct seq_file *m;
7205 int ret;
7206
7207 ret = tracing_check_open_get_tr(tr);
7208 if (ret)
7209 return ret;
7210
7211 if (file->f_mode & FMODE_READ) {
7212 iter = __tracing_open(inode, file, true);
7213 if (IS_ERR(iter))
7214 ret = PTR_ERR(iter);
7215 } else {
7216 /* Writes still need the seq_file to hold the private data */
7217 ret = -ENOMEM;
7218 m = kzalloc(sizeof(*m), GFP_KERNEL);
7219 if (!m)
7220 goto out;
7221 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7222 if (!iter) {
7223 kfree(m);
7224 goto out;
7225 }
7226 ret = 0;
7227
7228 iter->tr = tr;
7229 iter->array_buffer = &tr->max_buffer;
7230 iter->cpu_file = tracing_get_cpu(inode);
7231 m->private = iter;
7232 file->private_data = m;
7233 }
7234out:
7235 if (ret < 0)
7236 trace_array_put(tr);
7237
7238 return ret;
7239}
7240
7241static void tracing_swap_cpu_buffer(void *tr)
7242{
7243 update_max_tr_single((struct trace_array *)tr, current, smp_processor_id());
7244}
7245
7246static ssize_t
7247tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7248 loff_t *ppos)
7249{
7250 struct seq_file *m = filp->private_data;
7251 struct trace_iterator *iter = m->private;
7252 struct trace_array *tr = iter->tr;
7253 unsigned long val;
7254 int ret;
7255
7256 ret = tracing_update_buffers(tr);
7257 if (ret < 0)
7258 return ret;
7259
7260 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7261 if (ret)
7262 return ret;
7263
7264 mutex_lock(&trace_types_lock);
7265
7266 if (tr->current_trace->use_max_tr) {
7267 ret = -EBUSY;
7268 goto out;
7269 }
7270
7271 local_irq_disable();
7272 arch_spin_lock(&tr->max_lock);
7273 if (tr->cond_snapshot)
7274 ret = -EBUSY;
7275 arch_spin_unlock(&tr->max_lock);
7276 local_irq_enable();
7277 if (ret)
7278 goto out;
7279
7280 switch (val) {
7281 case 0:
7282 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7283 ret = -EINVAL;
7284 break;
7285 }
7286 if (tr->allocated_snapshot)
7287 free_snapshot(tr);
7288 break;
7289 case 1:
7290/* Only allow per-cpu swap if the ring buffer supports it */
7291#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7292 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7293 ret = -EINVAL;
7294 break;
7295 }
7296#endif
7297 if (tr->allocated_snapshot)
7298 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7299 &tr->array_buffer, iter->cpu_file);
7300
7301 ret = tracing_arm_snapshot_locked(tr);
7302 if (ret)
7303 break;
7304
7305 /* Now, we're going to swap */
7306 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
7307 local_irq_disable();
7308 update_max_tr(tr, current, smp_processor_id(), NULL);
7309 local_irq_enable();
7310 } else {
7311 smp_call_function_single(iter->cpu_file, tracing_swap_cpu_buffer,
7312 (void *)tr, 1);
7313 }
7314 tracing_disarm_snapshot(tr);
7315 break;
7316 default:
7317 if (tr->allocated_snapshot) {
7318 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7319 tracing_reset_online_cpus(&tr->max_buffer);
7320 else
7321 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7322 }
7323 break;
7324 }
7325
7326 if (ret >= 0) {
7327 *ppos += cnt;
7328 ret = cnt;
7329 }
7330out:
7331 mutex_unlock(&trace_types_lock);
7332 return ret;
7333}
7334
7335static int tracing_snapshot_release(struct inode *inode, struct file *file)
7336{
7337 struct seq_file *m = file->private_data;
7338 int ret;
7339
7340 ret = tracing_release(inode, file);
7341
7342 if (file->f_mode & FMODE_READ)
7343 return ret;
7344
7345 /* If write only, the seq_file is just a stub */
7346 if (m)
7347 kfree(m->private);
7348 kfree(m);
7349
7350 return 0;
7351}
7352
7353static int tracing_buffers_open(struct inode *inode, struct file *filp);
7354static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7355 size_t count, loff_t *ppos);
7356static int tracing_buffers_release(struct inode *inode, struct file *file);
7357static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7358 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7359
7360static int snapshot_raw_open(struct inode *inode, struct file *filp)
7361{
7362 struct ftrace_buffer_info *info;
7363 int ret;
7364
7365 /* The following checks for tracefs lockdown */
7366 ret = tracing_buffers_open(inode, filp);
7367 if (ret < 0)
7368 return ret;
7369
7370 info = filp->private_data;
7371
7372 if (info->iter.trace->use_max_tr) {
7373 tracing_buffers_release(inode, filp);
7374 return -EBUSY;
7375 }
7376
7377 info->iter.snapshot = true;
7378 info->iter.array_buffer = &info->iter.tr->max_buffer;
7379
7380 return ret;
7381}
7382
7383#endif /* CONFIG_TRACER_SNAPSHOT */
7384
7385
7386static const struct file_operations tracing_thresh_fops = {
7387 .open = tracing_open_generic,
7388 .read = tracing_thresh_read,
7389 .write = tracing_thresh_write,
7390 .llseek = generic_file_llseek,
7391};
7392
7393#ifdef CONFIG_TRACER_MAX_TRACE
7394static const struct file_operations tracing_max_lat_fops = {
7395 .open = tracing_open_generic_tr,
7396 .read = tracing_max_lat_read,
7397 .write = tracing_max_lat_write,
7398 .llseek = generic_file_llseek,
7399 .release = tracing_release_generic_tr,
7400};
7401#endif
7402
7403static const struct file_operations set_tracer_fops = {
7404 .open = tracing_open_generic_tr,
7405 .read = tracing_set_trace_read,
7406 .write = tracing_set_trace_write,
7407 .llseek = generic_file_llseek,
7408 .release = tracing_release_generic_tr,
7409};
7410
7411static const struct file_operations tracing_pipe_fops = {
7412 .open = tracing_open_pipe,
7413 .poll = tracing_poll_pipe,
7414 .read = tracing_read_pipe,
7415 .splice_read = tracing_splice_read_pipe,
7416 .release = tracing_release_pipe,
7417 .llseek = no_llseek,
7418};
7419
7420static const struct file_operations tracing_entries_fops = {
7421 .open = tracing_open_generic_tr,
7422 .read = tracing_entries_read,
7423 .write = tracing_entries_write,
7424 .llseek = generic_file_llseek,
7425 .release = tracing_release_generic_tr,
7426};
7427
7428static const struct file_operations tracing_total_entries_fops = {
7429 .open = tracing_open_generic_tr,
7430 .read = tracing_total_entries_read,
7431 .llseek = generic_file_llseek,
7432 .release = tracing_release_generic_tr,
7433};
7434
7435static const struct file_operations tracing_free_buffer_fops = {
7436 .open = tracing_open_generic_tr,
7437 .write = tracing_free_buffer_write,
7438 .release = tracing_free_buffer_release,
7439};
7440
7441static const struct file_operations tracing_mark_fops = {
7442 .open = tracing_mark_open,
7443 .write = tracing_mark_write,
7444 .release = tracing_release_generic_tr,
7445};
7446
7447static const struct file_operations tracing_mark_raw_fops = {
7448 .open = tracing_mark_open,
7449 .write = tracing_mark_raw_write,
7450 .release = tracing_release_generic_tr,
7451};
7452
7453static const struct file_operations trace_clock_fops = {
7454 .open = tracing_clock_open,
7455 .read = seq_read,
7456 .llseek = seq_lseek,
7457 .release = tracing_single_release_tr,
7458 .write = tracing_clock_write,
7459};
7460
7461static const struct file_operations trace_time_stamp_mode_fops = {
7462 .open = tracing_time_stamp_mode_open,
7463 .read = seq_read,
7464 .llseek = seq_lseek,
7465 .release = tracing_single_release_tr,
7466};
7467
7468#ifdef CONFIG_TRACER_SNAPSHOT
7469static const struct file_operations snapshot_fops = {
7470 .open = tracing_snapshot_open,
7471 .read = seq_read,
7472 .write = tracing_snapshot_write,
7473 .llseek = tracing_lseek,
7474 .release = tracing_snapshot_release,
7475};
7476
7477static const struct file_operations snapshot_raw_fops = {
7478 .open = snapshot_raw_open,
7479 .read = tracing_buffers_read,
7480 .release = tracing_buffers_release,
7481 .splice_read = tracing_buffers_splice_read,
7482 .llseek = no_llseek,
7483};
7484
7485#endif /* CONFIG_TRACER_SNAPSHOT */
7486
7487/*
7488 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7489 * @filp: The active open file structure
7490 * @ubuf: The userspace provided buffer to read value into
7491 * @cnt: The maximum number of bytes to read
7492 * @ppos: The current "file" position
7493 *
7494 * This function implements the write interface for a struct trace_min_max_param.
7495 * The filp->private_data must point to a trace_min_max_param structure that
7496 * defines where to write the value, the min and the max acceptable values,
7497 * and a lock to protect the write.
7498 */
7499static ssize_t
7500trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7501{
7502 struct trace_min_max_param *param = filp->private_data;
7503 u64 val;
7504 int err;
7505
7506 if (!param)
7507 return -EFAULT;
7508
7509 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7510 if (err)
7511 return err;
7512
7513 if (param->lock)
7514 mutex_lock(param->lock);
7515
7516 if (param->min && val < *param->min)
7517 err = -EINVAL;
7518
7519 if (param->max && val > *param->max)
7520 err = -EINVAL;
7521
7522 if (!err)
7523 *param->val = val;
7524
7525 if (param->lock)
7526 mutex_unlock(param->lock);
7527
7528 if (err)
7529 return err;
7530
7531 return cnt;
7532}
7533
7534/*
7535 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7536 * @filp: The active open file structure
7537 * @ubuf: The userspace provided buffer to read value into
7538 * @cnt: The maximum number of bytes to read
7539 * @ppos: The current "file" position
7540 *
7541 * This function implements the read interface for a struct trace_min_max_param.
7542 * The filp->private_data must point to a trace_min_max_param struct with valid
7543 * data.
7544 */
7545static ssize_t
7546trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7547{
7548 struct trace_min_max_param *param = filp->private_data;
7549 char buf[U64_STR_SIZE];
7550 int len;
7551 u64 val;
7552
7553 if (!param)
7554 return -EFAULT;
7555
7556 val = *param->val;
7557
7558 if (cnt > sizeof(buf))
7559 cnt = sizeof(buf);
7560
7561 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7562
7563 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7564}
7565
7566const struct file_operations trace_min_max_fops = {
7567 .open = tracing_open_generic,
7568 .read = trace_min_max_read,
7569 .write = trace_min_max_write,
7570};
7571
7572#define TRACING_LOG_ERRS_MAX 8
7573#define TRACING_LOG_LOC_MAX 128
7574
7575#define CMD_PREFIX " Command: "
7576
7577struct err_info {
7578 const char **errs; /* ptr to loc-specific array of err strings */
7579 u8 type; /* index into errs -> specific err string */
7580 u16 pos; /* caret position */
7581 u64 ts;
7582};
7583
7584struct tracing_log_err {
7585 struct list_head list;
7586 struct err_info info;
7587 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7588 char *cmd; /* what caused err */
7589};
7590
7591static DEFINE_MUTEX(tracing_err_log_lock);
7592
7593static struct tracing_log_err *alloc_tracing_log_err(int len)
7594{
7595 struct tracing_log_err *err;
7596
7597 err = kzalloc(sizeof(*err), GFP_KERNEL);
7598 if (!err)
7599 return ERR_PTR(-ENOMEM);
7600
7601 err->cmd = kzalloc(len, GFP_KERNEL);
7602 if (!err->cmd) {
7603 kfree(err);
7604 return ERR_PTR(-ENOMEM);
7605 }
7606
7607 return err;
7608}
7609
7610static void free_tracing_log_err(struct tracing_log_err *err)
7611{
7612 kfree(err->cmd);
7613 kfree(err);
7614}
7615
7616static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7617 int len)
7618{
7619 struct tracing_log_err *err;
7620 char *cmd;
7621
7622 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7623 err = alloc_tracing_log_err(len);
7624 if (PTR_ERR(err) != -ENOMEM)
7625 tr->n_err_log_entries++;
7626
7627 return err;
7628 }
7629 cmd = kzalloc(len, GFP_KERNEL);
7630 if (!cmd)
7631 return ERR_PTR(-ENOMEM);
7632 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7633 kfree(err->cmd);
7634 err->cmd = cmd;
7635 list_del(&err->list);
7636
7637 return err;
7638}
7639
7640/**
7641 * err_pos - find the position of a string within a command for error careting
7642 * @cmd: The tracing command that caused the error
7643 * @str: The string to position the caret at within @cmd
7644 *
7645 * Finds the position of the first occurrence of @str within @cmd. The
7646 * return value can be passed to tracing_log_err() for caret placement
7647 * within @cmd.
7648 *
7649 * Returns the index within @cmd of the first occurrence of @str or 0
7650 * if @str was not found.
7651 */
7652unsigned int err_pos(char *cmd, const char *str)
7653{
7654 char *found;
7655
7656 if (WARN_ON(!strlen(cmd)))
7657 return 0;
7658
7659 found = strstr(cmd, str);
7660 if (found)
7661 return found - cmd;
7662
7663 return 0;
7664}
7665
7666/**
7667 * tracing_log_err - write an error to the tracing error log
7668 * @tr: The associated trace array for the error (NULL for top level array)
7669 * @loc: A string describing where the error occurred
7670 * @cmd: The tracing command that caused the error
7671 * @errs: The array of loc-specific static error strings
7672 * @type: The index into errs[], which produces the specific static err string
7673 * @pos: The position the caret should be placed in the cmd
7674 *
7675 * Writes an error into tracing/error_log of the form:
7676 *
7677 * <loc>: error: <text>
7678 * Command: <cmd>
7679 * ^
7680 *
7681 * tracing/error_log is a small log file containing the last
7682 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7683 * unless there has been a tracing error, and the error log can be
7684 * cleared and have its memory freed by writing the empty string in
7685 * truncation mode to it i.e. echo > tracing/error_log.
7686 *
7687 * NOTE: the @errs array along with the @type param are used to
7688 * produce a static error string - this string is not copied and saved
7689 * when the error is logged - only a pointer to it is saved. See
7690 * existing callers for examples of how static strings are typically
7691 * defined for use with tracing_log_err().
7692 */
7693void tracing_log_err(struct trace_array *tr,
7694 const char *loc, const char *cmd,
7695 const char **errs, u8 type, u16 pos)
7696{
7697 struct tracing_log_err *err;
7698 int len = 0;
7699
7700 if (!tr)
7701 tr = &global_trace;
7702
7703 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
7704
7705 mutex_lock(&tracing_err_log_lock);
7706 err = get_tracing_log_err(tr, len);
7707 if (PTR_ERR(err) == -ENOMEM) {
7708 mutex_unlock(&tracing_err_log_lock);
7709 return;
7710 }
7711
7712 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7713 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
7714
7715 err->info.errs = errs;
7716 err->info.type = type;
7717 err->info.pos = pos;
7718 err->info.ts = local_clock();
7719
7720 list_add_tail(&err->list, &tr->err_log);
7721 mutex_unlock(&tracing_err_log_lock);
7722}
7723
7724static void clear_tracing_err_log(struct trace_array *tr)
7725{
7726 struct tracing_log_err *err, *next;
7727
7728 mutex_lock(&tracing_err_log_lock);
7729 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7730 list_del(&err->list);
7731 free_tracing_log_err(err);
7732 }
7733
7734 tr->n_err_log_entries = 0;
7735 mutex_unlock(&tracing_err_log_lock);
7736}
7737
7738static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7739{
7740 struct trace_array *tr = m->private;
7741
7742 mutex_lock(&tracing_err_log_lock);
7743
7744 return seq_list_start(&tr->err_log, *pos);
7745}
7746
7747static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7748{
7749 struct trace_array *tr = m->private;
7750
7751 return seq_list_next(v, &tr->err_log, pos);
7752}
7753
7754static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7755{
7756 mutex_unlock(&tracing_err_log_lock);
7757}
7758
7759static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
7760{
7761 u16 i;
7762
7763 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7764 seq_putc(m, ' ');
7765 for (i = 0; i < pos; i++)
7766 seq_putc(m, ' ');
7767 seq_puts(m, "^\n");
7768}
7769
7770static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7771{
7772 struct tracing_log_err *err = v;
7773
7774 if (err) {
7775 const char *err_text = err->info.errs[err->info.type];
7776 u64 sec = err->info.ts;
7777 u32 nsec;
7778
7779 nsec = do_div(sec, NSEC_PER_SEC);
7780 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7781 err->loc, err_text);
7782 seq_printf(m, "%s", err->cmd);
7783 tracing_err_log_show_pos(m, err->info.pos);
7784 }
7785
7786 return 0;
7787}
7788
7789static const struct seq_operations tracing_err_log_seq_ops = {
7790 .start = tracing_err_log_seq_start,
7791 .next = tracing_err_log_seq_next,
7792 .stop = tracing_err_log_seq_stop,
7793 .show = tracing_err_log_seq_show
7794};
7795
7796static int tracing_err_log_open(struct inode *inode, struct file *file)
7797{
7798 struct trace_array *tr = inode->i_private;
7799 int ret = 0;
7800
7801 ret = tracing_check_open_get_tr(tr);
7802 if (ret)
7803 return ret;
7804
7805 /* If this file was opened for write, then erase contents */
7806 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7807 clear_tracing_err_log(tr);
7808
7809 if (file->f_mode & FMODE_READ) {
7810 ret = seq_open(file, &tracing_err_log_seq_ops);
7811 if (!ret) {
7812 struct seq_file *m = file->private_data;
7813 m->private = tr;
7814 } else {
7815 trace_array_put(tr);
7816 }
7817 }
7818 return ret;
7819}
7820
7821static ssize_t tracing_err_log_write(struct file *file,
7822 const char __user *buffer,
7823 size_t count, loff_t *ppos)
7824{
7825 return count;
7826}
7827
7828static int tracing_err_log_release(struct inode *inode, struct file *file)
7829{
7830 struct trace_array *tr = inode->i_private;
7831
7832 trace_array_put(tr);
7833
7834 if (file->f_mode & FMODE_READ)
7835 seq_release(inode, file);
7836
7837 return 0;
7838}
7839
7840static const struct file_operations tracing_err_log_fops = {
7841 .open = tracing_err_log_open,
7842 .write = tracing_err_log_write,
7843 .read = seq_read,
7844 .llseek = tracing_lseek,
7845 .release = tracing_err_log_release,
7846};
7847
7848static int tracing_buffers_open(struct inode *inode, struct file *filp)
7849{
7850 struct trace_array *tr = inode->i_private;
7851 struct ftrace_buffer_info *info;
7852 int ret;
7853
7854 ret = tracing_check_open_get_tr(tr);
7855 if (ret)
7856 return ret;
7857
7858 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7859 if (!info) {
7860 trace_array_put(tr);
7861 return -ENOMEM;
7862 }
7863
7864 mutex_lock(&trace_types_lock);
7865
7866 info->iter.tr = tr;
7867 info->iter.cpu_file = tracing_get_cpu(inode);
7868 info->iter.trace = tr->current_trace;
7869 info->iter.array_buffer = &tr->array_buffer;
7870 info->spare = NULL;
7871 /* Force reading ring buffer for first read */
7872 info->read = (unsigned int)-1;
7873
7874 filp->private_data = info;
7875
7876 tr->trace_ref++;
7877
7878 mutex_unlock(&trace_types_lock);
7879
7880 ret = nonseekable_open(inode, filp);
7881 if (ret < 0)
7882 trace_array_put(tr);
7883
7884 return ret;
7885}
7886
7887static __poll_t
7888tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7889{
7890 struct ftrace_buffer_info *info = filp->private_data;
7891 struct trace_iterator *iter = &info->iter;
7892
7893 return trace_poll(iter, filp, poll_table);
7894}
7895
7896static ssize_t
7897tracing_buffers_read(struct file *filp, char __user *ubuf,
7898 size_t count, loff_t *ppos)
7899{
7900 struct ftrace_buffer_info *info = filp->private_data;
7901 struct trace_iterator *iter = &info->iter;
7902 void *trace_data;
7903 int page_size;
7904 ssize_t ret = 0;
7905 ssize_t size;
7906
7907 if (!count)
7908 return 0;
7909
7910#ifdef CONFIG_TRACER_MAX_TRACE
7911 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7912 return -EBUSY;
7913#endif
7914
7915 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
7916
7917 /* Make sure the spare matches the current sub buffer size */
7918 if (info->spare) {
7919 if (page_size != info->spare_size) {
7920 ring_buffer_free_read_page(iter->array_buffer->buffer,
7921 info->spare_cpu, info->spare);
7922 info->spare = NULL;
7923 }
7924 }
7925
7926 if (!info->spare) {
7927 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
7928 iter->cpu_file);
7929 if (IS_ERR(info->spare)) {
7930 ret = PTR_ERR(info->spare);
7931 info->spare = NULL;
7932 } else {
7933 info->spare_cpu = iter->cpu_file;
7934 info->spare_size = page_size;
7935 }
7936 }
7937 if (!info->spare)
7938 return ret;
7939
7940 /* Do we have previous read data to read? */
7941 if (info->read < page_size)
7942 goto read;
7943
7944 again:
7945 trace_access_lock(iter->cpu_file);
7946 ret = ring_buffer_read_page(iter->array_buffer->buffer,
7947 info->spare,
7948 count,
7949 iter->cpu_file, 0);
7950 trace_access_unlock(iter->cpu_file);
7951
7952 if (ret < 0) {
7953 if (trace_empty(iter)) {
7954 if ((filp->f_flags & O_NONBLOCK))
7955 return -EAGAIN;
7956
7957 ret = wait_on_pipe(iter, 0);
7958 if (ret)
7959 return ret;
7960
7961 goto again;
7962 }
7963 return 0;
7964 }
7965
7966 info->read = 0;
7967 read:
7968 size = page_size - info->read;
7969 if (size > count)
7970 size = count;
7971 trace_data = ring_buffer_read_page_data(info->spare);
7972 ret = copy_to_user(ubuf, trace_data + info->read, size);
7973 if (ret == size)
7974 return -EFAULT;
7975
7976 size -= ret;
7977
7978 *ppos += size;
7979 info->read += size;
7980
7981 return size;
7982}
7983
7984static int tracing_buffers_flush(struct file *file, fl_owner_t id)
7985{
7986 struct ftrace_buffer_info *info = file->private_data;
7987 struct trace_iterator *iter = &info->iter;
7988
7989 iter->closed = true;
7990 /* Make sure the waiters see the new wait_index */
7991 (void)atomic_fetch_inc_release(&iter->wait_index);
7992
7993 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
7994
7995 return 0;
7996}
7997
7998static int tracing_buffers_release(struct inode *inode, struct file *file)
7999{
8000 struct ftrace_buffer_info *info = file->private_data;
8001 struct trace_iterator *iter = &info->iter;
8002
8003 mutex_lock(&trace_types_lock);
8004
8005 iter->tr->trace_ref--;
8006
8007 __trace_array_put(iter->tr);
8008
8009 if (info->spare)
8010 ring_buffer_free_read_page(iter->array_buffer->buffer,
8011 info->spare_cpu, info->spare);
8012 kvfree(info);
8013
8014 mutex_unlock(&trace_types_lock);
8015
8016 return 0;
8017}
8018
8019struct buffer_ref {
8020 struct trace_buffer *buffer;
8021 void *page;
8022 int cpu;
8023 refcount_t refcount;
8024};
8025
8026static void buffer_ref_release(struct buffer_ref *ref)
8027{
8028 if (!refcount_dec_and_test(&ref->refcount))
8029 return;
8030 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8031 kfree(ref);
8032}
8033
8034static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8035 struct pipe_buffer *buf)
8036{
8037 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8038
8039 buffer_ref_release(ref);
8040 buf->private = 0;
8041}
8042
8043static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8044 struct pipe_buffer *buf)
8045{
8046 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8047
8048 if (refcount_read(&ref->refcount) > INT_MAX/2)
8049 return false;
8050
8051 refcount_inc(&ref->refcount);
8052 return true;
8053}
8054
8055/* Pipe buffer operations for a buffer. */
8056static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8057 .release = buffer_pipe_buf_release,
8058 .get = buffer_pipe_buf_get,
8059};
8060
8061/*
8062 * Callback from splice_to_pipe(), if we need to release some pages
8063 * at the end of the spd in case we error'ed out in filling the pipe.
8064 */
8065static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8066{
8067 struct buffer_ref *ref =
8068 (struct buffer_ref *)spd->partial[i].private;
8069
8070 buffer_ref_release(ref);
8071 spd->partial[i].private = 0;
8072}
8073
8074static ssize_t
8075tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8076 struct pipe_inode_info *pipe, size_t len,
8077 unsigned int flags)
8078{
8079 struct ftrace_buffer_info *info = file->private_data;
8080 struct trace_iterator *iter = &info->iter;
8081 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8082 struct page *pages_def[PIPE_DEF_BUFFERS];
8083 struct splice_pipe_desc spd = {
8084 .pages = pages_def,
8085 .partial = partial_def,
8086 .nr_pages_max = PIPE_DEF_BUFFERS,
8087 .ops = &buffer_pipe_buf_ops,
8088 .spd_release = buffer_spd_release,
8089 };
8090 struct buffer_ref *ref;
8091 bool woken = false;
8092 int page_size;
8093 int entries, i;
8094 ssize_t ret = 0;
8095
8096#ifdef CONFIG_TRACER_MAX_TRACE
8097 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8098 return -EBUSY;
8099#endif
8100
8101 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8102 if (*ppos & (page_size - 1))
8103 return -EINVAL;
8104
8105 if (len & (page_size - 1)) {
8106 if (len < page_size)
8107 return -EINVAL;
8108 len &= (~(page_size - 1));
8109 }
8110
8111 if (splice_grow_spd(pipe, &spd))
8112 return -ENOMEM;
8113
8114 again:
8115 trace_access_lock(iter->cpu_file);
8116 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8117
8118 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= page_size) {
8119 struct page *page;
8120 int r;
8121
8122 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8123 if (!ref) {
8124 ret = -ENOMEM;
8125 break;
8126 }
8127
8128 refcount_set(&ref->refcount, 1);
8129 ref->buffer = iter->array_buffer->buffer;
8130 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8131 if (IS_ERR(ref->page)) {
8132 ret = PTR_ERR(ref->page);
8133 ref->page = NULL;
8134 kfree(ref);
8135 break;
8136 }
8137 ref->cpu = iter->cpu_file;
8138
8139 r = ring_buffer_read_page(ref->buffer, ref->page,
8140 len, iter->cpu_file, 1);
8141 if (r < 0) {
8142 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8143 ref->page);
8144 kfree(ref);
8145 break;
8146 }
8147
8148 page = virt_to_page(ring_buffer_read_page_data(ref->page));
8149
8150 spd.pages[i] = page;
8151 spd.partial[i].len = page_size;
8152 spd.partial[i].offset = 0;
8153 spd.partial[i].private = (unsigned long)ref;
8154 spd.nr_pages++;
8155 *ppos += page_size;
8156
8157 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8158 }
8159
8160 trace_access_unlock(iter->cpu_file);
8161 spd.nr_pages = i;
8162
8163 /* did we read anything? */
8164 if (!spd.nr_pages) {
8165
8166 if (ret)
8167 goto out;
8168
8169 if (woken)
8170 goto out;
8171
8172 ret = -EAGAIN;
8173 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8174 goto out;
8175
8176 ret = wait_on_pipe(iter, iter->snapshot ? 0 : iter->tr->buffer_percent);
8177 if (ret)
8178 goto out;
8179
8180 /* No need to wait after waking up when tracing is off */
8181 if (!tracer_tracing_is_on(iter->tr))
8182 goto out;
8183
8184 /* Iterate one more time to collect any new data then exit */
8185 woken = true;
8186
8187 goto again;
8188 }
8189
8190 ret = splice_to_pipe(pipe, &spd);
8191out:
8192 splice_shrink_spd(&spd);
8193
8194 return ret;
8195}
8196
8197/* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8198static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8199{
8200 struct ftrace_buffer_info *info = file->private_data;
8201 struct trace_iterator *iter = &info->iter;
8202
8203 if (cmd)
8204 return -ENOIOCTLCMD;
8205
8206 mutex_lock(&trace_types_lock);
8207
8208 /* Make sure the waiters see the new wait_index */
8209 (void)atomic_fetch_inc_release(&iter->wait_index);
8210
8211 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8212
8213 mutex_unlock(&trace_types_lock);
8214 return 0;
8215}
8216
8217static const struct file_operations tracing_buffers_fops = {
8218 .open = tracing_buffers_open,
8219 .read = tracing_buffers_read,
8220 .poll = tracing_buffers_poll,
8221 .release = tracing_buffers_release,
8222 .flush = tracing_buffers_flush,
8223 .splice_read = tracing_buffers_splice_read,
8224 .unlocked_ioctl = tracing_buffers_ioctl,
8225 .llseek = no_llseek,
8226};
8227
8228static ssize_t
8229tracing_stats_read(struct file *filp, char __user *ubuf,
8230 size_t count, loff_t *ppos)
8231{
8232 struct inode *inode = file_inode(filp);
8233 struct trace_array *tr = inode->i_private;
8234 struct array_buffer *trace_buf = &tr->array_buffer;
8235 int cpu = tracing_get_cpu(inode);
8236 struct trace_seq *s;
8237 unsigned long cnt;
8238 unsigned long long t;
8239 unsigned long usec_rem;
8240
8241 s = kmalloc(sizeof(*s), GFP_KERNEL);
8242 if (!s)
8243 return -ENOMEM;
8244
8245 trace_seq_init(s);
8246
8247 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8248 trace_seq_printf(s, "entries: %ld\n", cnt);
8249
8250 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8251 trace_seq_printf(s, "overrun: %ld\n", cnt);
8252
8253 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8254 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8255
8256 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8257 trace_seq_printf(s, "bytes: %ld\n", cnt);
8258
8259 if (trace_clocks[tr->clock_id].in_ns) {
8260 /* local or global for trace_clock */
8261 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8262 usec_rem = do_div(t, USEC_PER_SEC);
8263 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8264 t, usec_rem);
8265
8266 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8267 usec_rem = do_div(t, USEC_PER_SEC);
8268 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8269 } else {
8270 /* counter or tsc mode for trace_clock */
8271 trace_seq_printf(s, "oldest event ts: %llu\n",
8272 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8273
8274 trace_seq_printf(s, "now ts: %llu\n",
8275 ring_buffer_time_stamp(trace_buf->buffer));
8276 }
8277
8278 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8279 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8280
8281 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8282 trace_seq_printf(s, "read events: %ld\n", cnt);
8283
8284 count = simple_read_from_buffer(ubuf, count, ppos,
8285 s->buffer, trace_seq_used(s));
8286
8287 kfree(s);
8288
8289 return count;
8290}
8291
8292static const struct file_operations tracing_stats_fops = {
8293 .open = tracing_open_generic_tr,
8294 .read = tracing_stats_read,
8295 .llseek = generic_file_llseek,
8296 .release = tracing_release_generic_tr,
8297};
8298
8299#ifdef CONFIG_DYNAMIC_FTRACE
8300
8301static ssize_t
8302tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8303 size_t cnt, loff_t *ppos)
8304{
8305 ssize_t ret;
8306 char *buf;
8307 int r;
8308
8309 /* 256 should be plenty to hold the amount needed */
8310 buf = kmalloc(256, GFP_KERNEL);
8311 if (!buf)
8312 return -ENOMEM;
8313
8314 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8315 ftrace_update_tot_cnt,
8316 ftrace_number_of_pages,
8317 ftrace_number_of_groups);
8318
8319 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8320 kfree(buf);
8321 return ret;
8322}
8323
8324static const struct file_operations tracing_dyn_info_fops = {
8325 .open = tracing_open_generic,
8326 .read = tracing_read_dyn_info,
8327 .llseek = generic_file_llseek,
8328};
8329#endif /* CONFIG_DYNAMIC_FTRACE */
8330
8331#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8332static void
8333ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8334 struct trace_array *tr, struct ftrace_probe_ops *ops,
8335 void *data)
8336{
8337 tracing_snapshot_instance(tr);
8338}
8339
8340static void
8341ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8342 struct trace_array *tr, struct ftrace_probe_ops *ops,
8343 void *data)
8344{
8345 struct ftrace_func_mapper *mapper = data;
8346 long *count = NULL;
8347
8348 if (mapper)
8349 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8350
8351 if (count) {
8352
8353 if (*count <= 0)
8354 return;
8355
8356 (*count)--;
8357 }
8358
8359 tracing_snapshot_instance(tr);
8360}
8361
8362static int
8363ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8364 struct ftrace_probe_ops *ops, void *data)
8365{
8366 struct ftrace_func_mapper *mapper = data;
8367 long *count = NULL;
8368
8369 seq_printf(m, "%ps:", (void *)ip);
8370
8371 seq_puts(m, "snapshot");
8372
8373 if (mapper)
8374 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8375
8376 if (count)
8377 seq_printf(m, ":count=%ld\n", *count);
8378 else
8379 seq_puts(m, ":unlimited\n");
8380
8381 return 0;
8382}
8383
8384static int
8385ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8386 unsigned long ip, void *init_data, void **data)
8387{
8388 struct ftrace_func_mapper *mapper = *data;
8389
8390 if (!mapper) {
8391 mapper = allocate_ftrace_func_mapper();
8392 if (!mapper)
8393 return -ENOMEM;
8394 *data = mapper;
8395 }
8396
8397 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8398}
8399
8400static void
8401ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8402 unsigned long ip, void *data)
8403{
8404 struct ftrace_func_mapper *mapper = data;
8405
8406 if (!ip) {
8407 if (!mapper)
8408 return;
8409 free_ftrace_func_mapper(mapper, NULL);
8410 return;
8411 }
8412
8413 ftrace_func_mapper_remove_ip(mapper, ip);
8414}
8415
8416static struct ftrace_probe_ops snapshot_probe_ops = {
8417 .func = ftrace_snapshot,
8418 .print = ftrace_snapshot_print,
8419};
8420
8421static struct ftrace_probe_ops snapshot_count_probe_ops = {
8422 .func = ftrace_count_snapshot,
8423 .print = ftrace_snapshot_print,
8424 .init = ftrace_snapshot_init,
8425 .free = ftrace_snapshot_free,
8426};
8427
8428static int
8429ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8430 char *glob, char *cmd, char *param, int enable)
8431{
8432 struct ftrace_probe_ops *ops;
8433 void *count = (void *)-1;
8434 char *number;
8435 int ret;
8436
8437 if (!tr)
8438 return -ENODEV;
8439
8440 /* hash funcs only work with set_ftrace_filter */
8441 if (!enable)
8442 return -EINVAL;
8443
8444 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8445
8446 if (glob[0] == '!') {
8447 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
8448 if (!ret)
8449 tracing_disarm_snapshot(tr);
8450
8451 return ret;
8452 }
8453
8454 if (!param)
8455 goto out_reg;
8456
8457 number = strsep(¶m, ":");
8458
8459 if (!strlen(number))
8460 goto out_reg;
8461
8462 /*
8463 * We use the callback data field (which is a pointer)
8464 * as our counter.
8465 */
8466 ret = kstrtoul(number, 0, (unsigned long *)&count);
8467 if (ret)
8468 return ret;
8469
8470 out_reg:
8471 ret = tracing_arm_snapshot(tr);
8472 if (ret < 0)
8473 goto out;
8474
8475 ret = register_ftrace_function_probe(glob, tr, ops, count);
8476 if (ret < 0)
8477 tracing_disarm_snapshot(tr);
8478 out:
8479 return ret < 0 ? ret : 0;
8480}
8481
8482static struct ftrace_func_command ftrace_snapshot_cmd = {
8483 .name = "snapshot",
8484 .func = ftrace_trace_snapshot_callback,
8485};
8486
8487static __init int register_snapshot_cmd(void)
8488{
8489 return register_ftrace_command(&ftrace_snapshot_cmd);
8490}
8491#else
8492static inline __init int register_snapshot_cmd(void) { return 0; }
8493#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8494
8495static struct dentry *tracing_get_dentry(struct trace_array *tr)
8496{
8497 if (WARN_ON(!tr->dir))
8498 return ERR_PTR(-ENODEV);
8499
8500 /* Top directory uses NULL as the parent */
8501 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8502 return NULL;
8503
8504 /* All sub buffers have a descriptor */
8505 return tr->dir;
8506}
8507
8508static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8509{
8510 struct dentry *d_tracer;
8511
8512 if (tr->percpu_dir)
8513 return tr->percpu_dir;
8514
8515 d_tracer = tracing_get_dentry(tr);
8516 if (IS_ERR(d_tracer))
8517 return NULL;
8518
8519 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8520
8521 MEM_FAIL(!tr->percpu_dir,
8522 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8523
8524 return tr->percpu_dir;
8525}
8526
8527static struct dentry *
8528trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8529 void *data, long cpu, const struct file_operations *fops)
8530{
8531 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8532
8533 if (ret) /* See tracing_get_cpu() */
8534 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8535 return ret;
8536}
8537
8538static void
8539tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8540{
8541 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8542 struct dentry *d_cpu;
8543 char cpu_dir[30]; /* 30 characters should be more than enough */
8544
8545 if (!d_percpu)
8546 return;
8547
8548 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8549 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8550 if (!d_cpu) {
8551 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8552 return;
8553 }
8554
8555 /* per cpu trace_pipe */
8556 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8557 tr, cpu, &tracing_pipe_fops);
8558
8559 /* per cpu trace */
8560 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8561 tr, cpu, &tracing_fops);
8562
8563 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8564 tr, cpu, &tracing_buffers_fops);
8565
8566 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8567 tr, cpu, &tracing_stats_fops);
8568
8569 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8570 tr, cpu, &tracing_entries_fops);
8571
8572#ifdef CONFIG_TRACER_SNAPSHOT
8573 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8574 tr, cpu, &snapshot_fops);
8575
8576 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8577 tr, cpu, &snapshot_raw_fops);
8578#endif
8579}
8580
8581#ifdef CONFIG_FTRACE_SELFTEST
8582/* Let selftest have access to static functions in this file */
8583#include "trace_selftest.c"
8584#endif
8585
8586static ssize_t
8587trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8588 loff_t *ppos)
8589{
8590 struct trace_option_dentry *topt = filp->private_data;
8591 char *buf;
8592
8593 if (topt->flags->val & topt->opt->bit)
8594 buf = "1\n";
8595 else
8596 buf = "0\n";
8597
8598 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8599}
8600
8601static ssize_t
8602trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8603 loff_t *ppos)
8604{
8605 struct trace_option_dentry *topt = filp->private_data;
8606 unsigned long val;
8607 int ret;
8608
8609 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8610 if (ret)
8611 return ret;
8612
8613 if (val != 0 && val != 1)
8614 return -EINVAL;
8615
8616 if (!!(topt->flags->val & topt->opt->bit) != val) {
8617 mutex_lock(&trace_types_lock);
8618 ret = __set_tracer_option(topt->tr, topt->flags,
8619 topt->opt, !val);
8620 mutex_unlock(&trace_types_lock);
8621 if (ret)
8622 return ret;
8623 }
8624
8625 *ppos += cnt;
8626
8627 return cnt;
8628}
8629
8630static int tracing_open_options(struct inode *inode, struct file *filp)
8631{
8632 struct trace_option_dentry *topt = inode->i_private;
8633 int ret;
8634
8635 ret = tracing_check_open_get_tr(topt->tr);
8636 if (ret)
8637 return ret;
8638
8639 filp->private_data = inode->i_private;
8640 return 0;
8641}
8642
8643static int tracing_release_options(struct inode *inode, struct file *file)
8644{
8645 struct trace_option_dentry *topt = file->private_data;
8646
8647 trace_array_put(topt->tr);
8648 return 0;
8649}
8650
8651static const struct file_operations trace_options_fops = {
8652 .open = tracing_open_options,
8653 .read = trace_options_read,
8654 .write = trace_options_write,
8655 .llseek = generic_file_llseek,
8656 .release = tracing_release_options,
8657};
8658
8659/*
8660 * In order to pass in both the trace_array descriptor as well as the index
8661 * to the flag that the trace option file represents, the trace_array
8662 * has a character array of trace_flags_index[], which holds the index
8663 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8664 * The address of this character array is passed to the flag option file
8665 * read/write callbacks.
8666 *
8667 * In order to extract both the index and the trace_array descriptor,
8668 * get_tr_index() uses the following algorithm.
8669 *
8670 * idx = *ptr;
8671 *
8672 * As the pointer itself contains the address of the index (remember
8673 * index[1] == 1).
8674 *
8675 * Then to get the trace_array descriptor, by subtracting that index
8676 * from the ptr, we get to the start of the index itself.
8677 *
8678 * ptr - idx == &index[0]
8679 *
8680 * Then a simple container_of() from that pointer gets us to the
8681 * trace_array descriptor.
8682 */
8683static void get_tr_index(void *data, struct trace_array **ptr,
8684 unsigned int *pindex)
8685{
8686 *pindex = *(unsigned char *)data;
8687
8688 *ptr = container_of(data - *pindex, struct trace_array,
8689 trace_flags_index);
8690}
8691
8692static ssize_t
8693trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8694 loff_t *ppos)
8695{
8696 void *tr_index = filp->private_data;
8697 struct trace_array *tr;
8698 unsigned int index;
8699 char *buf;
8700
8701 get_tr_index(tr_index, &tr, &index);
8702
8703 if (tr->trace_flags & (1 << index))
8704 buf = "1\n";
8705 else
8706 buf = "0\n";
8707
8708 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8709}
8710
8711static ssize_t
8712trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8713 loff_t *ppos)
8714{
8715 void *tr_index = filp->private_data;
8716 struct trace_array *tr;
8717 unsigned int index;
8718 unsigned long val;
8719 int ret;
8720
8721 get_tr_index(tr_index, &tr, &index);
8722
8723 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8724 if (ret)
8725 return ret;
8726
8727 if (val != 0 && val != 1)
8728 return -EINVAL;
8729
8730 mutex_lock(&event_mutex);
8731 mutex_lock(&trace_types_lock);
8732 ret = set_tracer_flag(tr, 1 << index, val);
8733 mutex_unlock(&trace_types_lock);
8734 mutex_unlock(&event_mutex);
8735
8736 if (ret < 0)
8737 return ret;
8738
8739 *ppos += cnt;
8740
8741 return cnt;
8742}
8743
8744static const struct file_operations trace_options_core_fops = {
8745 .open = tracing_open_generic,
8746 .read = trace_options_core_read,
8747 .write = trace_options_core_write,
8748 .llseek = generic_file_llseek,
8749};
8750
8751struct dentry *trace_create_file(const char *name,
8752 umode_t mode,
8753 struct dentry *parent,
8754 void *data,
8755 const struct file_operations *fops)
8756{
8757 struct dentry *ret;
8758
8759 ret = tracefs_create_file(name, mode, parent, data, fops);
8760 if (!ret)
8761 pr_warn("Could not create tracefs '%s' entry\n", name);
8762
8763 return ret;
8764}
8765
8766
8767static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8768{
8769 struct dentry *d_tracer;
8770
8771 if (tr->options)
8772 return tr->options;
8773
8774 d_tracer = tracing_get_dentry(tr);
8775 if (IS_ERR(d_tracer))
8776 return NULL;
8777
8778 tr->options = tracefs_create_dir("options", d_tracer);
8779 if (!tr->options) {
8780 pr_warn("Could not create tracefs directory 'options'\n");
8781 return NULL;
8782 }
8783
8784 return tr->options;
8785}
8786
8787static void
8788create_trace_option_file(struct trace_array *tr,
8789 struct trace_option_dentry *topt,
8790 struct tracer_flags *flags,
8791 struct tracer_opt *opt)
8792{
8793 struct dentry *t_options;
8794
8795 t_options = trace_options_init_dentry(tr);
8796 if (!t_options)
8797 return;
8798
8799 topt->flags = flags;
8800 topt->opt = opt;
8801 topt->tr = tr;
8802
8803 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
8804 t_options, topt, &trace_options_fops);
8805
8806}
8807
8808static void
8809create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8810{
8811 struct trace_option_dentry *topts;
8812 struct trace_options *tr_topts;
8813 struct tracer_flags *flags;
8814 struct tracer_opt *opts;
8815 int cnt;
8816 int i;
8817
8818 if (!tracer)
8819 return;
8820
8821 flags = tracer->flags;
8822
8823 if (!flags || !flags->opts)
8824 return;
8825
8826 /*
8827 * If this is an instance, only create flags for tracers
8828 * the instance may have.
8829 */
8830 if (!trace_ok_for_array(tracer, tr))
8831 return;
8832
8833 for (i = 0; i < tr->nr_topts; i++) {
8834 /* Make sure there's no duplicate flags. */
8835 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8836 return;
8837 }
8838
8839 opts = flags->opts;
8840
8841 for (cnt = 0; opts[cnt].name; cnt++)
8842 ;
8843
8844 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8845 if (!topts)
8846 return;
8847
8848 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8849 GFP_KERNEL);
8850 if (!tr_topts) {
8851 kfree(topts);
8852 return;
8853 }
8854
8855 tr->topts = tr_topts;
8856 tr->topts[tr->nr_topts].tracer = tracer;
8857 tr->topts[tr->nr_topts].topts = topts;
8858 tr->nr_topts++;
8859
8860 for (cnt = 0; opts[cnt].name; cnt++) {
8861 create_trace_option_file(tr, &topts[cnt], flags,
8862 &opts[cnt]);
8863 MEM_FAIL(topts[cnt].entry == NULL,
8864 "Failed to create trace option: %s",
8865 opts[cnt].name);
8866 }
8867}
8868
8869static struct dentry *
8870create_trace_option_core_file(struct trace_array *tr,
8871 const char *option, long index)
8872{
8873 struct dentry *t_options;
8874
8875 t_options = trace_options_init_dentry(tr);
8876 if (!t_options)
8877 return NULL;
8878
8879 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
8880 (void *)&tr->trace_flags_index[index],
8881 &trace_options_core_fops);
8882}
8883
8884static void create_trace_options_dir(struct trace_array *tr)
8885{
8886 struct dentry *t_options;
8887 bool top_level = tr == &global_trace;
8888 int i;
8889
8890 t_options = trace_options_init_dentry(tr);
8891 if (!t_options)
8892 return;
8893
8894 for (i = 0; trace_options[i]; i++) {
8895 if (top_level ||
8896 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8897 create_trace_option_core_file(tr, trace_options[i], i);
8898 }
8899}
8900
8901static ssize_t
8902rb_simple_read(struct file *filp, char __user *ubuf,
8903 size_t cnt, loff_t *ppos)
8904{
8905 struct trace_array *tr = filp->private_data;
8906 char buf[64];
8907 int r;
8908
8909 r = tracer_tracing_is_on(tr);
8910 r = sprintf(buf, "%d\n", r);
8911
8912 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8913}
8914
8915static ssize_t
8916rb_simple_write(struct file *filp, const char __user *ubuf,
8917 size_t cnt, loff_t *ppos)
8918{
8919 struct trace_array *tr = filp->private_data;
8920 struct trace_buffer *buffer = tr->array_buffer.buffer;
8921 unsigned long val;
8922 int ret;
8923
8924 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8925 if (ret)
8926 return ret;
8927
8928 if (buffer) {
8929 mutex_lock(&trace_types_lock);
8930 if (!!val == tracer_tracing_is_on(tr)) {
8931 val = 0; /* do nothing */
8932 } else if (val) {
8933 tracer_tracing_on(tr);
8934 if (tr->current_trace->start)
8935 tr->current_trace->start(tr);
8936 } else {
8937 tracer_tracing_off(tr);
8938 if (tr->current_trace->stop)
8939 tr->current_trace->stop(tr);
8940 /* Wake up any waiters */
8941 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
8942 }
8943 mutex_unlock(&trace_types_lock);
8944 }
8945
8946 (*ppos)++;
8947
8948 return cnt;
8949}
8950
8951static const struct file_operations rb_simple_fops = {
8952 .open = tracing_open_generic_tr,
8953 .read = rb_simple_read,
8954 .write = rb_simple_write,
8955 .release = tracing_release_generic_tr,
8956 .llseek = default_llseek,
8957};
8958
8959static ssize_t
8960buffer_percent_read(struct file *filp, char __user *ubuf,
8961 size_t cnt, loff_t *ppos)
8962{
8963 struct trace_array *tr = filp->private_data;
8964 char buf[64];
8965 int r;
8966
8967 r = tr->buffer_percent;
8968 r = sprintf(buf, "%d\n", r);
8969
8970 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8971}
8972
8973static ssize_t
8974buffer_percent_write(struct file *filp, const char __user *ubuf,
8975 size_t cnt, loff_t *ppos)
8976{
8977 struct trace_array *tr = filp->private_data;
8978 unsigned long val;
8979 int ret;
8980
8981 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8982 if (ret)
8983 return ret;
8984
8985 if (val > 100)
8986 return -EINVAL;
8987
8988 tr->buffer_percent = val;
8989
8990 (*ppos)++;
8991
8992 return cnt;
8993}
8994
8995static const struct file_operations buffer_percent_fops = {
8996 .open = tracing_open_generic_tr,
8997 .read = buffer_percent_read,
8998 .write = buffer_percent_write,
8999 .release = tracing_release_generic_tr,
9000 .llseek = default_llseek,
9001};
9002
9003static ssize_t
9004buffer_subbuf_size_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
9005{
9006 struct trace_array *tr = filp->private_data;
9007 size_t size;
9008 char buf[64];
9009 int order;
9010 int r;
9011
9012 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9013 size = (PAGE_SIZE << order) / 1024;
9014
9015 r = sprintf(buf, "%zd\n", size);
9016
9017 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9018}
9019
9020static ssize_t
9021buffer_subbuf_size_write(struct file *filp, const char __user *ubuf,
9022 size_t cnt, loff_t *ppos)
9023{
9024 struct trace_array *tr = filp->private_data;
9025 unsigned long val;
9026 int old_order;
9027 int order;
9028 int pages;
9029 int ret;
9030
9031 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9032 if (ret)
9033 return ret;
9034
9035 val *= 1024; /* value passed in is in KB */
9036
9037 pages = DIV_ROUND_UP(val, PAGE_SIZE);
9038 order = fls(pages - 1);
9039
9040 /* limit between 1 and 128 system pages */
9041 if (order < 0 || order > 7)
9042 return -EINVAL;
9043
9044 /* Do not allow tracing while changing the order of the ring buffer */
9045 tracing_stop_tr(tr);
9046
9047 old_order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9048 if (old_order == order)
9049 goto out;
9050
9051 ret = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, order);
9052 if (ret)
9053 goto out;
9054
9055#ifdef CONFIG_TRACER_MAX_TRACE
9056
9057 if (!tr->allocated_snapshot)
9058 goto out_max;
9059
9060 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
9061 if (ret) {
9062 /* Put back the old order */
9063 cnt = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, old_order);
9064 if (WARN_ON_ONCE(cnt)) {
9065 /*
9066 * AARGH! We are left with different orders!
9067 * The max buffer is our "snapshot" buffer.
9068 * When a tracer needs a snapshot (one of the
9069 * latency tracers), it swaps the max buffer
9070 * with the saved snap shot. We succeeded to
9071 * update the order of the main buffer, but failed to
9072 * update the order of the max buffer. But when we tried
9073 * to reset the main buffer to the original size, we
9074 * failed there too. This is very unlikely to
9075 * happen, but if it does, warn and kill all
9076 * tracing.
9077 */
9078 tracing_disabled = 1;
9079 }
9080 goto out;
9081 }
9082 out_max:
9083#endif
9084 (*ppos)++;
9085 out:
9086 if (ret)
9087 cnt = ret;
9088 tracing_start_tr(tr);
9089 return cnt;
9090}
9091
9092static const struct file_operations buffer_subbuf_size_fops = {
9093 .open = tracing_open_generic_tr,
9094 .read = buffer_subbuf_size_read,
9095 .write = buffer_subbuf_size_write,
9096 .release = tracing_release_generic_tr,
9097 .llseek = default_llseek,
9098};
9099
9100static struct dentry *trace_instance_dir;
9101
9102static void
9103init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9104
9105static int
9106allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9107{
9108 enum ring_buffer_flags rb_flags;
9109
9110 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9111
9112 buf->tr = tr;
9113
9114 buf->buffer = ring_buffer_alloc(size, rb_flags);
9115 if (!buf->buffer)
9116 return -ENOMEM;
9117
9118 buf->data = alloc_percpu(struct trace_array_cpu);
9119 if (!buf->data) {
9120 ring_buffer_free(buf->buffer);
9121 buf->buffer = NULL;
9122 return -ENOMEM;
9123 }
9124
9125 /* Allocate the first page for all buffers */
9126 set_buffer_entries(&tr->array_buffer,
9127 ring_buffer_size(tr->array_buffer.buffer, 0));
9128
9129 return 0;
9130}
9131
9132static void free_trace_buffer(struct array_buffer *buf)
9133{
9134 if (buf->buffer) {
9135 ring_buffer_free(buf->buffer);
9136 buf->buffer = NULL;
9137 free_percpu(buf->data);
9138 buf->data = NULL;
9139 }
9140}
9141
9142static int allocate_trace_buffers(struct trace_array *tr, int size)
9143{
9144 int ret;
9145
9146 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9147 if (ret)
9148 return ret;
9149
9150#ifdef CONFIG_TRACER_MAX_TRACE
9151 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9152 allocate_snapshot ? size : 1);
9153 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9154 free_trace_buffer(&tr->array_buffer);
9155 return -ENOMEM;
9156 }
9157 tr->allocated_snapshot = allocate_snapshot;
9158
9159 allocate_snapshot = false;
9160#endif
9161
9162 return 0;
9163}
9164
9165static void free_trace_buffers(struct trace_array *tr)
9166{
9167 if (!tr)
9168 return;
9169
9170 free_trace_buffer(&tr->array_buffer);
9171
9172#ifdef CONFIG_TRACER_MAX_TRACE
9173 free_trace_buffer(&tr->max_buffer);
9174#endif
9175}
9176
9177static void init_trace_flags_index(struct trace_array *tr)
9178{
9179 int i;
9180
9181 /* Used by the trace options files */
9182 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9183 tr->trace_flags_index[i] = i;
9184}
9185
9186static void __update_tracer_options(struct trace_array *tr)
9187{
9188 struct tracer *t;
9189
9190 for (t = trace_types; t; t = t->next)
9191 add_tracer_options(tr, t);
9192}
9193
9194static void update_tracer_options(struct trace_array *tr)
9195{
9196 mutex_lock(&trace_types_lock);
9197 tracer_options_updated = true;
9198 __update_tracer_options(tr);
9199 mutex_unlock(&trace_types_lock);
9200}
9201
9202/* Must have trace_types_lock held */
9203struct trace_array *trace_array_find(const char *instance)
9204{
9205 struct trace_array *tr, *found = NULL;
9206
9207 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9208 if (tr->name && strcmp(tr->name, instance) == 0) {
9209 found = tr;
9210 break;
9211 }
9212 }
9213
9214 return found;
9215}
9216
9217struct trace_array *trace_array_find_get(const char *instance)
9218{
9219 struct trace_array *tr;
9220
9221 mutex_lock(&trace_types_lock);
9222 tr = trace_array_find(instance);
9223 if (tr)
9224 tr->ref++;
9225 mutex_unlock(&trace_types_lock);
9226
9227 return tr;
9228}
9229
9230static int trace_array_create_dir(struct trace_array *tr)
9231{
9232 int ret;
9233
9234 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9235 if (!tr->dir)
9236 return -EINVAL;
9237
9238 ret = event_trace_add_tracer(tr->dir, tr);
9239 if (ret) {
9240 tracefs_remove(tr->dir);
9241 return ret;
9242 }
9243
9244 init_tracer_tracefs(tr, tr->dir);
9245 __update_tracer_options(tr);
9246
9247 return ret;
9248}
9249
9250static struct trace_array *
9251trace_array_create_systems(const char *name, const char *systems)
9252{
9253 struct trace_array *tr;
9254 int ret;
9255
9256 ret = -ENOMEM;
9257 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9258 if (!tr)
9259 return ERR_PTR(ret);
9260
9261 tr->name = kstrdup(name, GFP_KERNEL);
9262 if (!tr->name)
9263 goto out_free_tr;
9264
9265 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9266 goto out_free_tr;
9267
9268 if (!zalloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9269 goto out_free_tr;
9270
9271 if (systems) {
9272 tr->system_names = kstrdup_const(systems, GFP_KERNEL);
9273 if (!tr->system_names)
9274 goto out_free_tr;
9275 }
9276
9277 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9278
9279 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9280
9281 raw_spin_lock_init(&tr->start_lock);
9282
9283 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9284#ifdef CONFIG_TRACER_MAX_TRACE
9285 spin_lock_init(&tr->snapshot_trigger_lock);
9286#endif
9287 tr->current_trace = &nop_trace;
9288
9289 INIT_LIST_HEAD(&tr->systems);
9290 INIT_LIST_HEAD(&tr->events);
9291 INIT_LIST_HEAD(&tr->hist_vars);
9292 INIT_LIST_HEAD(&tr->err_log);
9293
9294 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9295 goto out_free_tr;
9296
9297 /* The ring buffer is defaultly expanded */
9298 trace_set_ring_buffer_expanded(tr);
9299
9300 if (ftrace_allocate_ftrace_ops(tr) < 0)
9301 goto out_free_tr;
9302
9303 ftrace_init_trace_array(tr);
9304
9305 init_trace_flags_index(tr);
9306
9307 if (trace_instance_dir) {
9308 ret = trace_array_create_dir(tr);
9309 if (ret)
9310 goto out_free_tr;
9311 } else
9312 __trace_early_add_events(tr);
9313
9314 list_add(&tr->list, &ftrace_trace_arrays);
9315
9316 tr->ref++;
9317
9318 return tr;
9319
9320 out_free_tr:
9321 ftrace_free_ftrace_ops(tr);
9322 free_trace_buffers(tr);
9323 free_cpumask_var(tr->pipe_cpumask);
9324 free_cpumask_var(tr->tracing_cpumask);
9325 kfree_const(tr->system_names);
9326 kfree(tr->name);
9327 kfree(tr);
9328
9329 return ERR_PTR(ret);
9330}
9331
9332static struct trace_array *trace_array_create(const char *name)
9333{
9334 return trace_array_create_systems(name, NULL);
9335}
9336
9337static int instance_mkdir(const char *name)
9338{
9339 struct trace_array *tr;
9340 int ret;
9341
9342 mutex_lock(&event_mutex);
9343 mutex_lock(&trace_types_lock);
9344
9345 ret = -EEXIST;
9346 if (trace_array_find(name))
9347 goto out_unlock;
9348
9349 tr = trace_array_create(name);
9350
9351 ret = PTR_ERR_OR_ZERO(tr);
9352
9353out_unlock:
9354 mutex_unlock(&trace_types_lock);
9355 mutex_unlock(&event_mutex);
9356 return ret;
9357}
9358
9359/**
9360 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9361 * @name: The name of the trace array to be looked up/created.
9362 * @systems: A list of systems to create event directories for (NULL for all)
9363 *
9364 * Returns pointer to trace array with given name.
9365 * NULL, if it cannot be created.
9366 *
9367 * NOTE: This function increments the reference counter associated with the
9368 * trace array returned. This makes sure it cannot be freed while in use.
9369 * Use trace_array_put() once the trace array is no longer needed.
9370 * If the trace_array is to be freed, trace_array_destroy() needs to
9371 * be called after the trace_array_put(), or simply let user space delete
9372 * it from the tracefs instances directory. But until the
9373 * trace_array_put() is called, user space can not delete it.
9374 *
9375 */
9376struct trace_array *trace_array_get_by_name(const char *name, const char *systems)
9377{
9378 struct trace_array *tr;
9379
9380 mutex_lock(&event_mutex);
9381 mutex_lock(&trace_types_lock);
9382
9383 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9384 if (tr->name && strcmp(tr->name, name) == 0)
9385 goto out_unlock;
9386 }
9387
9388 tr = trace_array_create_systems(name, systems);
9389
9390 if (IS_ERR(tr))
9391 tr = NULL;
9392out_unlock:
9393 if (tr)
9394 tr->ref++;
9395
9396 mutex_unlock(&trace_types_lock);
9397 mutex_unlock(&event_mutex);
9398 return tr;
9399}
9400EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9401
9402static int __remove_instance(struct trace_array *tr)
9403{
9404 int i;
9405
9406 /* Reference counter for a newly created trace array = 1. */
9407 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9408 return -EBUSY;
9409
9410 list_del(&tr->list);
9411
9412 /* Disable all the flags that were enabled coming in */
9413 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9414 if ((1 << i) & ZEROED_TRACE_FLAGS)
9415 set_tracer_flag(tr, 1 << i, 0);
9416 }
9417
9418 tracing_set_nop(tr);
9419 clear_ftrace_function_probes(tr);
9420 event_trace_del_tracer(tr);
9421 ftrace_clear_pids(tr);
9422 ftrace_destroy_function_files(tr);
9423 tracefs_remove(tr->dir);
9424 free_percpu(tr->last_func_repeats);
9425 free_trace_buffers(tr);
9426 clear_tracing_err_log(tr);
9427
9428 for (i = 0; i < tr->nr_topts; i++) {
9429 kfree(tr->topts[i].topts);
9430 }
9431 kfree(tr->topts);
9432
9433 free_cpumask_var(tr->pipe_cpumask);
9434 free_cpumask_var(tr->tracing_cpumask);
9435 kfree_const(tr->system_names);
9436 kfree(tr->name);
9437 kfree(tr);
9438
9439 return 0;
9440}
9441
9442int trace_array_destroy(struct trace_array *this_tr)
9443{
9444 struct trace_array *tr;
9445 int ret;
9446
9447 if (!this_tr)
9448 return -EINVAL;
9449
9450 mutex_lock(&event_mutex);
9451 mutex_lock(&trace_types_lock);
9452
9453 ret = -ENODEV;
9454
9455 /* Making sure trace array exists before destroying it. */
9456 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9457 if (tr == this_tr) {
9458 ret = __remove_instance(tr);
9459 break;
9460 }
9461 }
9462
9463 mutex_unlock(&trace_types_lock);
9464 mutex_unlock(&event_mutex);
9465
9466 return ret;
9467}
9468EXPORT_SYMBOL_GPL(trace_array_destroy);
9469
9470static int instance_rmdir(const char *name)
9471{
9472 struct trace_array *tr;
9473 int ret;
9474
9475 mutex_lock(&event_mutex);
9476 mutex_lock(&trace_types_lock);
9477
9478 ret = -ENODEV;
9479 tr = trace_array_find(name);
9480 if (tr)
9481 ret = __remove_instance(tr);
9482
9483 mutex_unlock(&trace_types_lock);
9484 mutex_unlock(&event_mutex);
9485
9486 return ret;
9487}
9488
9489static __init void create_trace_instances(struct dentry *d_tracer)
9490{
9491 struct trace_array *tr;
9492
9493 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9494 instance_mkdir,
9495 instance_rmdir);
9496 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9497 return;
9498
9499 mutex_lock(&event_mutex);
9500 mutex_lock(&trace_types_lock);
9501
9502 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9503 if (!tr->name)
9504 continue;
9505 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9506 "Failed to create instance directory\n"))
9507 break;
9508 }
9509
9510 mutex_unlock(&trace_types_lock);
9511 mutex_unlock(&event_mutex);
9512}
9513
9514static void
9515init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9516{
9517 int cpu;
9518
9519 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9520 tr, &show_traces_fops);
9521
9522 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9523 tr, &set_tracer_fops);
9524
9525 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9526 tr, &tracing_cpumask_fops);
9527
9528 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9529 tr, &tracing_iter_fops);
9530
9531 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9532 tr, &tracing_fops);
9533
9534 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9535 tr, &tracing_pipe_fops);
9536
9537 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9538 tr, &tracing_entries_fops);
9539
9540 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9541 tr, &tracing_total_entries_fops);
9542
9543 trace_create_file("free_buffer", 0200, d_tracer,
9544 tr, &tracing_free_buffer_fops);
9545
9546 trace_create_file("trace_marker", 0220, d_tracer,
9547 tr, &tracing_mark_fops);
9548
9549 tr->trace_marker_file = __find_event_file(tr, "ftrace", "print");
9550
9551 trace_create_file("trace_marker_raw", 0220, d_tracer,
9552 tr, &tracing_mark_raw_fops);
9553
9554 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9555 &trace_clock_fops);
9556
9557 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9558 tr, &rb_simple_fops);
9559
9560 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9561 &trace_time_stamp_mode_fops);
9562
9563 tr->buffer_percent = 50;
9564
9565 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9566 tr, &buffer_percent_fops);
9567
9568 trace_create_file("buffer_subbuf_size_kb", TRACE_MODE_WRITE, d_tracer,
9569 tr, &buffer_subbuf_size_fops);
9570
9571 create_trace_options_dir(tr);
9572
9573#ifdef CONFIG_TRACER_MAX_TRACE
9574 trace_create_maxlat_file(tr, d_tracer);
9575#endif
9576
9577 if (ftrace_create_function_files(tr, d_tracer))
9578 MEM_FAIL(1, "Could not allocate function filter files");
9579
9580#ifdef CONFIG_TRACER_SNAPSHOT
9581 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9582 tr, &snapshot_fops);
9583#endif
9584
9585 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9586 tr, &tracing_err_log_fops);
9587
9588 for_each_tracing_cpu(cpu)
9589 tracing_init_tracefs_percpu(tr, cpu);
9590
9591 ftrace_init_tracefs(tr, d_tracer);
9592}
9593
9594static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9595{
9596 struct vfsmount *mnt;
9597 struct file_system_type *type;
9598
9599 /*
9600 * To maintain backward compatibility for tools that mount
9601 * debugfs to get to the tracing facility, tracefs is automatically
9602 * mounted to the debugfs/tracing directory.
9603 */
9604 type = get_fs_type("tracefs");
9605 if (!type)
9606 return NULL;
9607 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9608 put_filesystem(type);
9609 if (IS_ERR(mnt))
9610 return NULL;
9611 mntget(mnt);
9612
9613 return mnt;
9614}
9615
9616/**
9617 * tracing_init_dentry - initialize top level trace array
9618 *
9619 * This is called when creating files or directories in the tracing
9620 * directory. It is called via fs_initcall() by any of the boot up code
9621 * and expects to return the dentry of the top level tracing directory.
9622 */
9623int tracing_init_dentry(void)
9624{
9625 struct trace_array *tr = &global_trace;
9626
9627 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9628 pr_warn("Tracing disabled due to lockdown\n");
9629 return -EPERM;
9630 }
9631
9632 /* The top level trace array uses NULL as parent */
9633 if (tr->dir)
9634 return 0;
9635
9636 if (WARN_ON(!tracefs_initialized()))
9637 return -ENODEV;
9638
9639 /*
9640 * As there may still be users that expect the tracing
9641 * files to exist in debugfs/tracing, we must automount
9642 * the tracefs file system there, so older tools still
9643 * work with the newer kernel.
9644 */
9645 tr->dir = debugfs_create_automount("tracing", NULL,
9646 trace_automount, NULL);
9647
9648 return 0;
9649}
9650
9651extern struct trace_eval_map *__start_ftrace_eval_maps[];
9652extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9653
9654static struct workqueue_struct *eval_map_wq __initdata;
9655static struct work_struct eval_map_work __initdata;
9656static struct work_struct tracerfs_init_work __initdata;
9657
9658static void __init eval_map_work_func(struct work_struct *work)
9659{
9660 int len;
9661
9662 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9663 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9664}
9665
9666static int __init trace_eval_init(void)
9667{
9668 INIT_WORK(&eval_map_work, eval_map_work_func);
9669
9670 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9671 if (!eval_map_wq) {
9672 pr_err("Unable to allocate eval_map_wq\n");
9673 /* Do work here */
9674 eval_map_work_func(&eval_map_work);
9675 return -ENOMEM;
9676 }
9677
9678 queue_work(eval_map_wq, &eval_map_work);
9679 return 0;
9680}
9681
9682subsys_initcall(trace_eval_init);
9683
9684static int __init trace_eval_sync(void)
9685{
9686 /* Make sure the eval map updates are finished */
9687 if (eval_map_wq)
9688 destroy_workqueue(eval_map_wq);
9689 return 0;
9690}
9691
9692late_initcall_sync(trace_eval_sync);
9693
9694
9695#ifdef CONFIG_MODULES
9696static void trace_module_add_evals(struct module *mod)
9697{
9698 if (!mod->num_trace_evals)
9699 return;
9700
9701 /*
9702 * Modules with bad taint do not have events created, do
9703 * not bother with enums either.
9704 */
9705 if (trace_module_has_bad_taint(mod))
9706 return;
9707
9708 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9709}
9710
9711#ifdef CONFIG_TRACE_EVAL_MAP_FILE
9712static void trace_module_remove_evals(struct module *mod)
9713{
9714 union trace_eval_map_item *map;
9715 union trace_eval_map_item **last = &trace_eval_maps;
9716
9717 if (!mod->num_trace_evals)
9718 return;
9719
9720 mutex_lock(&trace_eval_mutex);
9721
9722 map = trace_eval_maps;
9723
9724 while (map) {
9725 if (map->head.mod == mod)
9726 break;
9727 map = trace_eval_jmp_to_tail(map);
9728 last = &map->tail.next;
9729 map = map->tail.next;
9730 }
9731 if (!map)
9732 goto out;
9733
9734 *last = trace_eval_jmp_to_tail(map)->tail.next;
9735 kfree(map);
9736 out:
9737 mutex_unlock(&trace_eval_mutex);
9738}
9739#else
9740static inline void trace_module_remove_evals(struct module *mod) { }
9741#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9742
9743static int trace_module_notify(struct notifier_block *self,
9744 unsigned long val, void *data)
9745{
9746 struct module *mod = data;
9747
9748 switch (val) {
9749 case MODULE_STATE_COMING:
9750 trace_module_add_evals(mod);
9751 break;
9752 case MODULE_STATE_GOING:
9753 trace_module_remove_evals(mod);
9754 break;
9755 }
9756
9757 return NOTIFY_OK;
9758}
9759
9760static struct notifier_block trace_module_nb = {
9761 .notifier_call = trace_module_notify,
9762 .priority = 0,
9763};
9764#endif /* CONFIG_MODULES */
9765
9766static __init void tracer_init_tracefs_work_func(struct work_struct *work)
9767{
9768
9769 event_trace_init();
9770
9771 init_tracer_tracefs(&global_trace, NULL);
9772 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9773
9774 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9775 &global_trace, &tracing_thresh_fops);
9776
9777 trace_create_file("README", TRACE_MODE_READ, NULL,
9778 NULL, &tracing_readme_fops);
9779
9780 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9781 NULL, &tracing_saved_cmdlines_fops);
9782
9783 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9784 NULL, &tracing_saved_cmdlines_size_fops);
9785
9786 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9787 NULL, &tracing_saved_tgids_fops);
9788
9789 trace_create_eval_file(NULL);
9790
9791#ifdef CONFIG_MODULES
9792 register_module_notifier(&trace_module_nb);
9793#endif
9794
9795#ifdef CONFIG_DYNAMIC_FTRACE
9796 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
9797 NULL, &tracing_dyn_info_fops);
9798#endif
9799
9800 create_trace_instances(NULL);
9801
9802 update_tracer_options(&global_trace);
9803}
9804
9805static __init int tracer_init_tracefs(void)
9806{
9807 int ret;
9808
9809 trace_access_lock_init();
9810
9811 ret = tracing_init_dentry();
9812 if (ret)
9813 return 0;
9814
9815 if (eval_map_wq) {
9816 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
9817 queue_work(eval_map_wq, &tracerfs_init_work);
9818 } else {
9819 tracer_init_tracefs_work_func(NULL);
9820 }
9821
9822 rv_init_interface();
9823
9824 return 0;
9825}
9826
9827fs_initcall(tracer_init_tracefs);
9828
9829static int trace_die_panic_handler(struct notifier_block *self,
9830 unsigned long ev, void *unused);
9831
9832static struct notifier_block trace_panic_notifier = {
9833 .notifier_call = trace_die_panic_handler,
9834 .priority = INT_MAX - 1,
9835};
9836
9837static struct notifier_block trace_die_notifier = {
9838 .notifier_call = trace_die_panic_handler,
9839 .priority = INT_MAX - 1,
9840};
9841
9842/*
9843 * The idea is to execute the following die/panic callback early, in order
9844 * to avoid showing irrelevant information in the trace (like other panic
9845 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
9846 * warnings get disabled (to prevent potential log flooding).
9847 */
9848static int trace_die_panic_handler(struct notifier_block *self,
9849 unsigned long ev, void *unused)
9850{
9851 if (!ftrace_dump_on_oops_enabled())
9852 return NOTIFY_DONE;
9853
9854 /* The die notifier requires DIE_OOPS to trigger */
9855 if (self == &trace_die_notifier && ev != DIE_OOPS)
9856 return NOTIFY_DONE;
9857
9858 ftrace_dump(DUMP_PARAM);
9859
9860 return NOTIFY_DONE;
9861}
9862
9863/*
9864 * printk is set to max of 1024, we really don't need it that big.
9865 * Nothing should be printing 1000 characters anyway.
9866 */
9867#define TRACE_MAX_PRINT 1000
9868
9869/*
9870 * Define here KERN_TRACE so that we have one place to modify
9871 * it if we decide to change what log level the ftrace dump
9872 * should be at.
9873 */
9874#define KERN_TRACE KERN_EMERG
9875
9876void
9877trace_printk_seq(struct trace_seq *s)
9878{
9879 /* Probably should print a warning here. */
9880 if (s->seq.len >= TRACE_MAX_PRINT)
9881 s->seq.len = TRACE_MAX_PRINT;
9882
9883 /*
9884 * More paranoid code. Although the buffer size is set to
9885 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9886 * an extra layer of protection.
9887 */
9888 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9889 s->seq.len = s->seq.size - 1;
9890
9891 /* should be zero ended, but we are paranoid. */
9892 s->buffer[s->seq.len] = 0;
9893
9894 printk(KERN_TRACE "%s", s->buffer);
9895
9896 trace_seq_init(s);
9897}
9898
9899static void trace_init_iter(struct trace_iterator *iter, struct trace_array *tr)
9900{
9901 iter->tr = tr;
9902 iter->trace = iter->tr->current_trace;
9903 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9904 iter->array_buffer = &tr->array_buffer;
9905
9906 if (iter->trace && iter->trace->open)
9907 iter->trace->open(iter);
9908
9909 /* Annotate start of buffers if we had overruns */
9910 if (ring_buffer_overruns(iter->array_buffer->buffer))
9911 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9912
9913 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9914 if (trace_clocks[iter->tr->clock_id].in_ns)
9915 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9916
9917 /* Can not use kmalloc for iter.temp and iter.fmt */
9918 iter->temp = static_temp_buf;
9919 iter->temp_size = STATIC_TEMP_BUF_SIZE;
9920 iter->fmt = static_fmt_buf;
9921 iter->fmt_size = STATIC_FMT_BUF_SIZE;
9922}
9923
9924void trace_init_global_iter(struct trace_iterator *iter)
9925{
9926 trace_init_iter(iter, &global_trace);
9927}
9928
9929static void ftrace_dump_one(struct trace_array *tr, enum ftrace_dump_mode dump_mode)
9930{
9931 /* use static because iter can be a bit big for the stack */
9932 static struct trace_iterator iter;
9933 unsigned int old_userobj;
9934 unsigned long flags;
9935 int cnt = 0, cpu;
9936
9937 /*
9938 * Always turn off tracing when we dump.
9939 * We don't need to show trace output of what happens
9940 * between multiple crashes.
9941 *
9942 * If the user does a sysrq-z, then they can re-enable
9943 * tracing with echo 1 > tracing_on.
9944 */
9945 tracer_tracing_off(tr);
9946
9947 local_irq_save(flags);
9948
9949 /* Simulate the iterator */
9950 trace_init_iter(&iter, tr);
9951
9952 for_each_tracing_cpu(cpu) {
9953 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9954 }
9955
9956 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
9957
9958 /* don't look at user memory in panic mode */
9959 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
9960
9961 if (dump_mode == DUMP_ORIG)
9962 iter.cpu_file = raw_smp_processor_id();
9963 else
9964 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9965
9966 if (tr == &global_trace)
9967 printk(KERN_TRACE "Dumping ftrace buffer:\n");
9968 else
9969 printk(KERN_TRACE "Dumping ftrace instance %s buffer:\n", tr->name);
9970
9971 /* Did function tracer already get disabled? */
9972 if (ftrace_is_dead()) {
9973 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
9974 printk("# MAY BE MISSING FUNCTION EVENTS\n");
9975 }
9976
9977 /*
9978 * We need to stop all tracing on all CPUS to read
9979 * the next buffer. This is a bit expensive, but is
9980 * not done often. We fill all what we can read,
9981 * and then release the locks again.
9982 */
9983
9984 while (!trace_empty(&iter)) {
9985
9986 if (!cnt)
9987 printk(KERN_TRACE "---------------------------------\n");
9988
9989 cnt++;
9990
9991 trace_iterator_reset(&iter);
9992 iter.iter_flags |= TRACE_FILE_LAT_FMT;
9993
9994 if (trace_find_next_entry_inc(&iter) != NULL) {
9995 int ret;
9996
9997 ret = print_trace_line(&iter);
9998 if (ret != TRACE_TYPE_NO_CONSUME)
9999 trace_consume(&iter);
10000 }
10001 touch_nmi_watchdog();
10002
10003 trace_printk_seq(&iter.seq);
10004 }
10005
10006 if (!cnt)
10007 printk(KERN_TRACE " (ftrace buffer empty)\n");
10008 else
10009 printk(KERN_TRACE "---------------------------------\n");
10010
10011 tr->trace_flags |= old_userobj;
10012
10013 for_each_tracing_cpu(cpu) {
10014 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10015 }
10016 local_irq_restore(flags);
10017}
10018
10019static void ftrace_dump_by_param(void)
10020{
10021 bool first_param = true;
10022 char dump_param[MAX_TRACER_SIZE];
10023 char *buf, *token, *inst_name;
10024 struct trace_array *tr;
10025
10026 strscpy(dump_param, ftrace_dump_on_oops, MAX_TRACER_SIZE);
10027 buf = dump_param;
10028
10029 while ((token = strsep(&buf, ",")) != NULL) {
10030 if (first_param) {
10031 first_param = false;
10032 if (!strcmp("0", token))
10033 continue;
10034 else if (!strcmp("1", token)) {
10035 ftrace_dump_one(&global_trace, DUMP_ALL);
10036 continue;
10037 }
10038 else if (!strcmp("2", token) ||
10039 !strcmp("orig_cpu", token)) {
10040 ftrace_dump_one(&global_trace, DUMP_ORIG);
10041 continue;
10042 }
10043 }
10044
10045 inst_name = strsep(&token, "=");
10046 tr = trace_array_find(inst_name);
10047 if (!tr) {
10048 printk(KERN_TRACE "Instance %s not found\n", inst_name);
10049 continue;
10050 }
10051
10052 if (token && (!strcmp("2", token) ||
10053 !strcmp("orig_cpu", token)))
10054 ftrace_dump_one(tr, DUMP_ORIG);
10055 else
10056 ftrace_dump_one(tr, DUMP_ALL);
10057 }
10058}
10059
10060void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10061{
10062 static atomic_t dump_running;
10063
10064 /* Only allow one dump user at a time. */
10065 if (atomic_inc_return(&dump_running) != 1) {
10066 atomic_dec(&dump_running);
10067 return;
10068 }
10069
10070 switch (oops_dump_mode) {
10071 case DUMP_ALL:
10072 ftrace_dump_one(&global_trace, DUMP_ALL);
10073 break;
10074 case DUMP_ORIG:
10075 ftrace_dump_one(&global_trace, DUMP_ORIG);
10076 break;
10077 case DUMP_PARAM:
10078 ftrace_dump_by_param();
10079 break;
10080 case DUMP_NONE:
10081 break;
10082 default:
10083 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10084 ftrace_dump_one(&global_trace, DUMP_ALL);
10085 }
10086
10087 atomic_dec(&dump_running);
10088}
10089EXPORT_SYMBOL_GPL(ftrace_dump);
10090
10091#define WRITE_BUFSIZE 4096
10092
10093ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10094 size_t count, loff_t *ppos,
10095 int (*createfn)(const char *))
10096{
10097 char *kbuf, *buf, *tmp;
10098 int ret = 0;
10099 size_t done = 0;
10100 size_t size;
10101
10102 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10103 if (!kbuf)
10104 return -ENOMEM;
10105
10106 while (done < count) {
10107 size = count - done;
10108
10109 if (size >= WRITE_BUFSIZE)
10110 size = WRITE_BUFSIZE - 1;
10111
10112 if (copy_from_user(kbuf, buffer + done, size)) {
10113 ret = -EFAULT;
10114 goto out;
10115 }
10116 kbuf[size] = '\0';
10117 buf = kbuf;
10118 do {
10119 tmp = strchr(buf, '\n');
10120 if (tmp) {
10121 *tmp = '\0';
10122 size = tmp - buf + 1;
10123 } else {
10124 size = strlen(buf);
10125 if (done + size < count) {
10126 if (buf != kbuf)
10127 break;
10128 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10129 pr_warn("Line length is too long: Should be less than %d\n",
10130 WRITE_BUFSIZE - 2);
10131 ret = -EINVAL;
10132 goto out;
10133 }
10134 }
10135 done += size;
10136
10137 /* Remove comments */
10138 tmp = strchr(buf, '#');
10139
10140 if (tmp)
10141 *tmp = '\0';
10142
10143 ret = createfn(buf);
10144 if (ret)
10145 goto out;
10146 buf += size;
10147
10148 } while (done < count);
10149 }
10150 ret = done;
10151
10152out:
10153 kfree(kbuf);
10154
10155 return ret;
10156}
10157
10158#ifdef CONFIG_TRACER_MAX_TRACE
10159__init static bool tr_needs_alloc_snapshot(const char *name)
10160{
10161 char *test;
10162 int len = strlen(name);
10163 bool ret;
10164
10165 if (!boot_snapshot_index)
10166 return false;
10167
10168 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10169 boot_snapshot_info[len] == '\t')
10170 return true;
10171
10172 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10173 if (!test)
10174 return false;
10175
10176 sprintf(test, "\t%s\t", name);
10177 ret = strstr(boot_snapshot_info, test) == NULL;
10178 kfree(test);
10179 return ret;
10180}
10181
10182__init static void do_allocate_snapshot(const char *name)
10183{
10184 if (!tr_needs_alloc_snapshot(name))
10185 return;
10186
10187 /*
10188 * When allocate_snapshot is set, the next call to
10189 * allocate_trace_buffers() (called by trace_array_get_by_name())
10190 * will allocate the snapshot buffer. That will alse clear
10191 * this flag.
10192 */
10193 allocate_snapshot = true;
10194}
10195#else
10196static inline void do_allocate_snapshot(const char *name) { }
10197#endif
10198
10199__init static void enable_instances(void)
10200{
10201 struct trace_array *tr;
10202 char *curr_str;
10203 char *str;
10204 char *tok;
10205
10206 /* A tab is always appended */
10207 boot_instance_info[boot_instance_index - 1] = '\0';
10208 str = boot_instance_info;
10209
10210 while ((curr_str = strsep(&str, "\t"))) {
10211
10212 tok = strsep(&curr_str, ",");
10213
10214 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10215 do_allocate_snapshot(tok);
10216
10217 tr = trace_array_get_by_name(tok, NULL);
10218 if (!tr) {
10219 pr_warn("Failed to create instance buffer %s\n", curr_str);
10220 continue;
10221 }
10222 /* Allow user space to delete it */
10223 trace_array_put(tr);
10224
10225 while ((tok = strsep(&curr_str, ","))) {
10226 early_enable_events(tr, tok, true);
10227 }
10228 }
10229}
10230
10231__init static int tracer_alloc_buffers(void)
10232{
10233 int ring_buf_size;
10234 int ret = -ENOMEM;
10235
10236
10237 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10238 pr_warn("Tracing disabled due to lockdown\n");
10239 return -EPERM;
10240 }
10241
10242 /*
10243 * Make sure we don't accidentally add more trace options
10244 * than we have bits for.
10245 */
10246 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10247
10248 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10249 goto out;
10250
10251 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10252 goto out_free_buffer_mask;
10253
10254 /* Only allocate trace_printk buffers if a trace_printk exists */
10255 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10256 /* Must be called before global_trace.buffer is allocated */
10257 trace_printk_init_buffers();
10258
10259 /* To save memory, keep the ring buffer size to its minimum */
10260 if (global_trace.ring_buffer_expanded)
10261 ring_buf_size = trace_buf_size;
10262 else
10263 ring_buf_size = 1;
10264
10265 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10266 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10267
10268 raw_spin_lock_init(&global_trace.start_lock);
10269
10270 /*
10271 * The prepare callbacks allocates some memory for the ring buffer. We
10272 * don't free the buffer if the CPU goes down. If we were to free
10273 * the buffer, then the user would lose any trace that was in the
10274 * buffer. The memory will be removed once the "instance" is removed.
10275 */
10276 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10277 "trace/RB:prepare", trace_rb_cpu_prepare,
10278 NULL);
10279 if (ret < 0)
10280 goto out_free_cpumask;
10281 /* Used for event triggers */
10282 ret = -ENOMEM;
10283 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10284 if (!temp_buffer)
10285 goto out_rm_hp_state;
10286
10287 if (trace_create_savedcmd() < 0)
10288 goto out_free_temp_buffer;
10289
10290 if (!zalloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10291 goto out_free_savedcmd;
10292
10293 /* TODO: make the number of buffers hot pluggable with CPUS */
10294 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10295 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10296 goto out_free_pipe_cpumask;
10297 }
10298 if (global_trace.buffer_disabled)
10299 tracing_off();
10300
10301 if (trace_boot_clock) {
10302 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10303 if (ret < 0)
10304 pr_warn("Trace clock %s not defined, going back to default\n",
10305 trace_boot_clock);
10306 }
10307
10308 /*
10309 * register_tracer() might reference current_trace, so it
10310 * needs to be set before we register anything. This is
10311 * just a bootstrap of current_trace anyway.
10312 */
10313 global_trace.current_trace = &nop_trace;
10314
10315 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10316#ifdef CONFIG_TRACER_MAX_TRACE
10317 spin_lock_init(&global_trace.snapshot_trigger_lock);
10318#endif
10319 ftrace_init_global_array_ops(&global_trace);
10320
10321 init_trace_flags_index(&global_trace);
10322
10323 register_tracer(&nop_trace);
10324
10325 /* Function tracing may start here (via kernel command line) */
10326 init_function_trace();
10327
10328 /* All seems OK, enable tracing */
10329 tracing_disabled = 0;
10330
10331 atomic_notifier_chain_register(&panic_notifier_list,
10332 &trace_panic_notifier);
10333
10334 register_die_notifier(&trace_die_notifier);
10335
10336 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10337
10338 INIT_LIST_HEAD(&global_trace.systems);
10339 INIT_LIST_HEAD(&global_trace.events);
10340 INIT_LIST_HEAD(&global_trace.hist_vars);
10341 INIT_LIST_HEAD(&global_trace.err_log);
10342 list_add(&global_trace.list, &ftrace_trace_arrays);
10343
10344 apply_trace_boot_options();
10345
10346 register_snapshot_cmd();
10347
10348 test_can_verify();
10349
10350 return 0;
10351
10352out_free_pipe_cpumask:
10353 free_cpumask_var(global_trace.pipe_cpumask);
10354out_free_savedcmd:
10355 trace_free_saved_cmdlines_buffer();
10356out_free_temp_buffer:
10357 ring_buffer_free(temp_buffer);
10358out_rm_hp_state:
10359 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10360out_free_cpumask:
10361 free_cpumask_var(global_trace.tracing_cpumask);
10362out_free_buffer_mask:
10363 free_cpumask_var(tracing_buffer_mask);
10364out:
10365 return ret;
10366}
10367
10368void __init ftrace_boot_snapshot(void)
10369{
10370#ifdef CONFIG_TRACER_MAX_TRACE
10371 struct trace_array *tr;
10372
10373 if (!snapshot_at_boot)
10374 return;
10375
10376 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10377 if (!tr->allocated_snapshot)
10378 continue;
10379
10380 tracing_snapshot_instance(tr);
10381 trace_array_puts(tr, "** Boot snapshot taken **\n");
10382 }
10383#endif
10384}
10385
10386void __init early_trace_init(void)
10387{
10388 if (tracepoint_printk) {
10389 tracepoint_print_iter =
10390 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10391 if (MEM_FAIL(!tracepoint_print_iter,
10392 "Failed to allocate trace iterator\n"))
10393 tracepoint_printk = 0;
10394 else
10395 static_key_enable(&tracepoint_printk_key.key);
10396 }
10397 tracer_alloc_buffers();
10398
10399 init_events();
10400}
10401
10402void __init trace_init(void)
10403{
10404 trace_event_init();
10405
10406 if (boot_instance_index)
10407 enable_instances();
10408}
10409
10410__init static void clear_boot_tracer(void)
10411{
10412 /*
10413 * The default tracer at boot buffer is an init section.
10414 * This function is called in lateinit. If we did not
10415 * find the boot tracer, then clear it out, to prevent
10416 * later registration from accessing the buffer that is
10417 * about to be freed.
10418 */
10419 if (!default_bootup_tracer)
10420 return;
10421
10422 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10423 default_bootup_tracer);
10424 default_bootup_tracer = NULL;
10425}
10426
10427#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10428__init static void tracing_set_default_clock(void)
10429{
10430 /* sched_clock_stable() is determined in late_initcall */
10431 if (!trace_boot_clock && !sched_clock_stable()) {
10432 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10433 pr_warn("Can not set tracing clock due to lockdown\n");
10434 return;
10435 }
10436
10437 printk(KERN_WARNING
10438 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10439 "If you want to keep using the local clock, then add:\n"
10440 " \"trace_clock=local\"\n"
10441 "on the kernel command line\n");
10442 tracing_set_clock(&global_trace, "global");
10443 }
10444}
10445#else
10446static inline void tracing_set_default_clock(void) { }
10447#endif
10448
10449__init static int late_trace_init(void)
10450{
10451 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10452 static_key_disable(&tracepoint_printk_key.key);
10453 tracepoint_printk = 0;
10454 }
10455
10456 tracing_set_default_clock();
10457 clear_boot_tracer();
10458 return 0;
10459}
10460
10461late_initcall_sync(late_trace_init);