Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * ring buffer based function tracer
4 *
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 *
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 *
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
14 */
15#include <linux/ring_buffer.h>
16#include <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/cleanup.h>
30#include <linux/vmalloc.h>
31#include <linux/ftrace.h>
32#include <linux/module.h>
33#include <linux/percpu.h>
34#include <linux/splice.h>
35#include <linux/kdebug.h>
36#include <linux/string.h>
37#include <linux/mount.h>
38#include <linux/rwsem.h>
39#include <linux/slab.h>
40#include <linux/ctype.h>
41#include <linux/init.h>
42#include <linux/panic_notifier.h>
43#include <linux/poll.h>
44#include <linux/nmi.h>
45#include <linux/fs.h>
46#include <linux/trace.h>
47#include <linux/sched/clock.h>
48#include <linux/sched/rt.h>
49#include <linux/fsnotify.h>
50#include <linux/irq_work.h>
51#include <linux/workqueue.h>
52
53#include <asm/setup.h> /* COMMAND_LINE_SIZE */
54
55#include "trace.h"
56#include "trace_output.h"
57
58#ifdef CONFIG_FTRACE_STARTUP_TEST
59/*
60 * We need to change this state when a selftest is running.
61 * A selftest will lurk into the ring-buffer to count the
62 * entries inserted during the selftest although some concurrent
63 * insertions into the ring-buffer such as trace_printk could occurred
64 * at the same time, giving false positive or negative results.
65 */
66static bool __read_mostly tracing_selftest_running;
67
68/*
69 * If boot-time tracing including tracers/events via kernel cmdline
70 * is running, we do not want to run SELFTEST.
71 */
72bool __read_mostly tracing_selftest_disabled;
73
74void __init disable_tracing_selftest(const char *reason)
75{
76 if (!tracing_selftest_disabled) {
77 tracing_selftest_disabled = true;
78 pr_info("Ftrace startup test is disabled due to %s\n", reason);
79 }
80}
81#else
82#define tracing_selftest_running 0
83#define tracing_selftest_disabled 0
84#endif
85
86/* Pipe tracepoints to printk */
87static struct trace_iterator *tracepoint_print_iter;
88int tracepoint_printk;
89static bool tracepoint_printk_stop_on_boot __initdata;
90static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
91
92/* For tracers that don't implement custom flags */
93static struct tracer_opt dummy_tracer_opt[] = {
94 { }
95};
96
97static int
98dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
99{
100 return 0;
101}
102
103/*
104 * To prevent the comm cache from being overwritten when no
105 * tracing is active, only save the comm when a trace event
106 * occurred.
107 */
108DEFINE_PER_CPU(bool, trace_taskinfo_save);
109
110/*
111 * Kill all tracing for good (never come back).
112 * It is initialized to 1 but will turn to zero if the initialization
113 * of the tracer is successful. But that is the only place that sets
114 * this back to zero.
115 */
116static int tracing_disabled = 1;
117
118cpumask_var_t __read_mostly tracing_buffer_mask;
119
120/*
121 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
122 *
123 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
124 * is set, then ftrace_dump is called. This will output the contents
125 * of the ftrace buffers to the console. This is very useful for
126 * capturing traces that lead to crashes and outputing it to a
127 * serial console.
128 *
129 * It is default off, but you can enable it with either specifying
130 * "ftrace_dump_on_oops" in the kernel command line, or setting
131 * /proc/sys/kernel/ftrace_dump_on_oops
132 * Set 1 if you want to dump buffers of all CPUs
133 * Set 2 if you want to dump the buffer of the CPU that triggered oops
134 * Set instance name if you want to dump the specific trace instance
135 * Multiple instance dump is also supported, and instances are seperated
136 * by commas.
137 */
138/* Set to string format zero to disable by default */
139char ftrace_dump_on_oops[MAX_TRACER_SIZE] = "0";
140
141/* When set, tracing will stop when a WARN*() is hit */
142int __disable_trace_on_warning;
143
144#ifdef CONFIG_TRACE_EVAL_MAP_FILE
145/* Map of enums to their values, for "eval_map" file */
146struct trace_eval_map_head {
147 struct module *mod;
148 unsigned long length;
149};
150
151union trace_eval_map_item;
152
153struct trace_eval_map_tail {
154 /*
155 * "end" is first and points to NULL as it must be different
156 * than "mod" or "eval_string"
157 */
158 union trace_eval_map_item *next;
159 const char *end; /* points to NULL */
160};
161
162static DEFINE_MUTEX(trace_eval_mutex);
163
164/*
165 * The trace_eval_maps are saved in an array with two extra elements,
166 * one at the beginning, and one at the end. The beginning item contains
167 * the count of the saved maps (head.length), and the module they
168 * belong to if not built in (head.mod). The ending item contains a
169 * pointer to the next array of saved eval_map items.
170 */
171union trace_eval_map_item {
172 struct trace_eval_map map;
173 struct trace_eval_map_head head;
174 struct trace_eval_map_tail tail;
175};
176
177static union trace_eval_map_item *trace_eval_maps;
178#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
179
180int tracing_set_tracer(struct trace_array *tr, const char *buf);
181static void ftrace_trace_userstack(struct trace_array *tr,
182 struct trace_buffer *buffer,
183 unsigned int trace_ctx);
184
185static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
186static char *default_bootup_tracer;
187
188static bool allocate_snapshot;
189static bool snapshot_at_boot;
190
191static char boot_instance_info[COMMAND_LINE_SIZE] __initdata;
192static int boot_instance_index;
193
194static char boot_snapshot_info[COMMAND_LINE_SIZE] __initdata;
195static int boot_snapshot_index;
196
197static int __init set_cmdline_ftrace(char *str)
198{
199 strscpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
200 default_bootup_tracer = bootup_tracer_buf;
201 /* We are using ftrace early, expand it */
202 trace_set_ring_buffer_expanded(NULL);
203 return 1;
204}
205__setup("ftrace=", set_cmdline_ftrace);
206
207int ftrace_dump_on_oops_enabled(void)
208{
209 if (!strcmp("0", ftrace_dump_on_oops))
210 return 0;
211 else
212 return 1;
213}
214
215static int __init set_ftrace_dump_on_oops(char *str)
216{
217 if (!*str) {
218 strscpy(ftrace_dump_on_oops, "1", MAX_TRACER_SIZE);
219 return 1;
220 }
221
222 if (*str == ',') {
223 strscpy(ftrace_dump_on_oops, "1", MAX_TRACER_SIZE);
224 strscpy(ftrace_dump_on_oops + 1, str, MAX_TRACER_SIZE - 1);
225 return 1;
226 }
227
228 if (*str++ == '=') {
229 strscpy(ftrace_dump_on_oops, str, MAX_TRACER_SIZE);
230 return 1;
231 }
232
233 return 0;
234}
235__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
236
237static int __init stop_trace_on_warning(char *str)
238{
239 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
240 __disable_trace_on_warning = 1;
241 return 1;
242}
243__setup("traceoff_on_warning", stop_trace_on_warning);
244
245static int __init boot_alloc_snapshot(char *str)
246{
247 char *slot = boot_snapshot_info + boot_snapshot_index;
248 int left = sizeof(boot_snapshot_info) - boot_snapshot_index;
249 int ret;
250
251 if (str[0] == '=') {
252 str++;
253 if (strlen(str) >= left)
254 return -1;
255
256 ret = snprintf(slot, left, "%s\t", str);
257 boot_snapshot_index += ret;
258 } else {
259 allocate_snapshot = true;
260 /* We also need the main ring buffer expanded */
261 trace_set_ring_buffer_expanded(NULL);
262 }
263 return 1;
264}
265__setup("alloc_snapshot", boot_alloc_snapshot);
266
267
268static int __init boot_snapshot(char *str)
269{
270 snapshot_at_boot = true;
271 boot_alloc_snapshot(str);
272 return 1;
273}
274__setup("ftrace_boot_snapshot", boot_snapshot);
275
276
277static int __init boot_instance(char *str)
278{
279 char *slot = boot_instance_info + boot_instance_index;
280 int left = sizeof(boot_instance_info) - boot_instance_index;
281 int ret;
282
283 if (strlen(str) >= left)
284 return -1;
285
286 ret = snprintf(slot, left, "%s\t", str);
287 boot_instance_index += ret;
288
289 return 1;
290}
291__setup("trace_instance=", boot_instance);
292
293
294static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
295
296static int __init set_trace_boot_options(char *str)
297{
298 strscpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
299 return 1;
300}
301__setup("trace_options=", set_trace_boot_options);
302
303static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
304static char *trace_boot_clock __initdata;
305
306static int __init set_trace_boot_clock(char *str)
307{
308 strscpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
309 trace_boot_clock = trace_boot_clock_buf;
310 return 1;
311}
312__setup("trace_clock=", set_trace_boot_clock);
313
314static int __init set_tracepoint_printk(char *str)
315{
316 /* Ignore the "tp_printk_stop_on_boot" param */
317 if (*str == '_')
318 return 0;
319
320 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
321 tracepoint_printk = 1;
322 return 1;
323}
324__setup("tp_printk", set_tracepoint_printk);
325
326static int __init set_tracepoint_printk_stop(char *str)
327{
328 tracepoint_printk_stop_on_boot = true;
329 return 1;
330}
331__setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
332
333unsigned long long ns2usecs(u64 nsec)
334{
335 nsec += 500;
336 do_div(nsec, 1000);
337 return nsec;
338}
339
340static void
341trace_process_export(struct trace_export *export,
342 struct ring_buffer_event *event, int flag)
343{
344 struct trace_entry *entry;
345 unsigned int size = 0;
346
347 if (export->flags & flag) {
348 entry = ring_buffer_event_data(event);
349 size = ring_buffer_event_length(event);
350 export->write(export, entry, size);
351 }
352}
353
354static DEFINE_MUTEX(ftrace_export_lock);
355
356static struct trace_export __rcu *ftrace_exports_list __read_mostly;
357
358static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
359static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
360static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
361
362static inline void ftrace_exports_enable(struct trace_export *export)
363{
364 if (export->flags & TRACE_EXPORT_FUNCTION)
365 static_branch_inc(&trace_function_exports_enabled);
366
367 if (export->flags & TRACE_EXPORT_EVENT)
368 static_branch_inc(&trace_event_exports_enabled);
369
370 if (export->flags & TRACE_EXPORT_MARKER)
371 static_branch_inc(&trace_marker_exports_enabled);
372}
373
374static inline void ftrace_exports_disable(struct trace_export *export)
375{
376 if (export->flags & TRACE_EXPORT_FUNCTION)
377 static_branch_dec(&trace_function_exports_enabled);
378
379 if (export->flags & TRACE_EXPORT_EVENT)
380 static_branch_dec(&trace_event_exports_enabled);
381
382 if (export->flags & TRACE_EXPORT_MARKER)
383 static_branch_dec(&trace_marker_exports_enabled);
384}
385
386static void ftrace_exports(struct ring_buffer_event *event, int flag)
387{
388 struct trace_export *export;
389
390 preempt_disable_notrace();
391
392 export = rcu_dereference_raw_check(ftrace_exports_list);
393 while (export) {
394 trace_process_export(export, event, flag);
395 export = rcu_dereference_raw_check(export->next);
396 }
397
398 preempt_enable_notrace();
399}
400
401static inline void
402add_trace_export(struct trace_export **list, struct trace_export *export)
403{
404 rcu_assign_pointer(export->next, *list);
405 /*
406 * We are entering export into the list but another
407 * CPU might be walking that list. We need to make sure
408 * the export->next pointer is valid before another CPU sees
409 * the export pointer included into the list.
410 */
411 rcu_assign_pointer(*list, export);
412}
413
414static inline int
415rm_trace_export(struct trace_export **list, struct trace_export *export)
416{
417 struct trace_export **p;
418
419 for (p = list; *p != NULL; p = &(*p)->next)
420 if (*p == export)
421 break;
422
423 if (*p != export)
424 return -1;
425
426 rcu_assign_pointer(*p, (*p)->next);
427
428 return 0;
429}
430
431static inline void
432add_ftrace_export(struct trace_export **list, struct trace_export *export)
433{
434 ftrace_exports_enable(export);
435
436 add_trace_export(list, export);
437}
438
439static inline int
440rm_ftrace_export(struct trace_export **list, struct trace_export *export)
441{
442 int ret;
443
444 ret = rm_trace_export(list, export);
445 ftrace_exports_disable(export);
446
447 return ret;
448}
449
450int register_ftrace_export(struct trace_export *export)
451{
452 if (WARN_ON_ONCE(!export->write))
453 return -1;
454
455 mutex_lock(&ftrace_export_lock);
456
457 add_ftrace_export(&ftrace_exports_list, export);
458
459 mutex_unlock(&ftrace_export_lock);
460
461 return 0;
462}
463EXPORT_SYMBOL_GPL(register_ftrace_export);
464
465int unregister_ftrace_export(struct trace_export *export)
466{
467 int ret;
468
469 mutex_lock(&ftrace_export_lock);
470
471 ret = rm_ftrace_export(&ftrace_exports_list, export);
472
473 mutex_unlock(&ftrace_export_lock);
474
475 return ret;
476}
477EXPORT_SYMBOL_GPL(unregister_ftrace_export);
478
479/* trace_flags holds trace_options default values */
480#define TRACE_DEFAULT_FLAGS \
481 (FUNCTION_DEFAULT_FLAGS | \
482 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
483 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
484 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
485 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
486 TRACE_ITER_HASH_PTR | TRACE_ITER_TRACE_PRINTK)
487
488/* trace_options that are only supported by global_trace */
489#define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
490 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
491
492/* trace_flags that are default zero for instances */
493#define ZEROED_TRACE_FLAGS \
494 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK | TRACE_ITER_TRACE_PRINTK)
495
496/*
497 * The global_trace is the descriptor that holds the top-level tracing
498 * buffers for the live tracing.
499 */
500static struct trace_array global_trace = {
501 .trace_flags = TRACE_DEFAULT_FLAGS,
502};
503
504static struct trace_array *printk_trace = &global_trace;
505
506static __always_inline bool printk_binsafe(struct trace_array *tr)
507{
508 /*
509 * The binary format of traceprintk can cause a crash if used
510 * by a buffer from another boot. Force the use of the
511 * non binary version of trace_printk if the trace_printk
512 * buffer is a boot mapped ring buffer.
513 */
514 return !(tr->flags & TRACE_ARRAY_FL_BOOT);
515}
516
517static void update_printk_trace(struct trace_array *tr)
518{
519 if (printk_trace == tr)
520 return;
521
522 printk_trace->trace_flags &= ~TRACE_ITER_TRACE_PRINTK;
523 printk_trace = tr;
524 tr->trace_flags |= TRACE_ITER_TRACE_PRINTK;
525}
526
527void trace_set_ring_buffer_expanded(struct trace_array *tr)
528{
529 if (!tr)
530 tr = &global_trace;
531 tr->ring_buffer_expanded = true;
532}
533
534LIST_HEAD(ftrace_trace_arrays);
535
536int trace_array_get(struct trace_array *this_tr)
537{
538 struct trace_array *tr;
539
540 guard(mutex)(&trace_types_lock);
541 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
542 if (tr == this_tr) {
543 tr->ref++;
544 return 0;
545 }
546 }
547
548 return -ENODEV;
549}
550
551static void __trace_array_put(struct trace_array *this_tr)
552{
553 WARN_ON(!this_tr->ref);
554 this_tr->ref--;
555}
556
557/**
558 * trace_array_put - Decrement the reference counter for this trace array.
559 * @this_tr : pointer to the trace array
560 *
561 * NOTE: Use this when we no longer need the trace array returned by
562 * trace_array_get_by_name(). This ensures the trace array can be later
563 * destroyed.
564 *
565 */
566void trace_array_put(struct trace_array *this_tr)
567{
568 if (!this_tr)
569 return;
570
571 mutex_lock(&trace_types_lock);
572 __trace_array_put(this_tr);
573 mutex_unlock(&trace_types_lock);
574}
575EXPORT_SYMBOL_GPL(trace_array_put);
576
577int tracing_check_open_get_tr(struct trace_array *tr)
578{
579 int ret;
580
581 ret = security_locked_down(LOCKDOWN_TRACEFS);
582 if (ret)
583 return ret;
584
585 if (tracing_disabled)
586 return -ENODEV;
587
588 if (tr && trace_array_get(tr) < 0)
589 return -ENODEV;
590
591 return 0;
592}
593
594/**
595 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
596 * @filtered_pids: The list of pids to check
597 * @search_pid: The PID to find in @filtered_pids
598 *
599 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
600 */
601bool
602trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
603{
604 return trace_pid_list_is_set(filtered_pids, search_pid);
605}
606
607/**
608 * trace_ignore_this_task - should a task be ignored for tracing
609 * @filtered_pids: The list of pids to check
610 * @filtered_no_pids: The list of pids not to be traced
611 * @task: The task that should be ignored if not filtered
612 *
613 * Checks if @task should be traced or not from @filtered_pids.
614 * Returns true if @task should *NOT* be traced.
615 * Returns false if @task should be traced.
616 */
617bool
618trace_ignore_this_task(struct trace_pid_list *filtered_pids,
619 struct trace_pid_list *filtered_no_pids,
620 struct task_struct *task)
621{
622 /*
623 * If filtered_no_pids is not empty, and the task's pid is listed
624 * in filtered_no_pids, then return true.
625 * Otherwise, if filtered_pids is empty, that means we can
626 * trace all tasks. If it has content, then only trace pids
627 * within filtered_pids.
628 */
629
630 return (filtered_pids &&
631 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
632 (filtered_no_pids &&
633 trace_find_filtered_pid(filtered_no_pids, task->pid));
634}
635
636/**
637 * trace_filter_add_remove_task - Add or remove a task from a pid_list
638 * @pid_list: The list to modify
639 * @self: The current task for fork or NULL for exit
640 * @task: The task to add or remove
641 *
642 * If adding a task, if @self is defined, the task is only added if @self
643 * is also included in @pid_list. This happens on fork and tasks should
644 * only be added when the parent is listed. If @self is NULL, then the
645 * @task pid will be removed from the list, which would happen on exit
646 * of a task.
647 */
648void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
649 struct task_struct *self,
650 struct task_struct *task)
651{
652 if (!pid_list)
653 return;
654
655 /* For forks, we only add if the forking task is listed */
656 if (self) {
657 if (!trace_find_filtered_pid(pid_list, self->pid))
658 return;
659 }
660
661 /* "self" is set for forks, and NULL for exits */
662 if (self)
663 trace_pid_list_set(pid_list, task->pid);
664 else
665 trace_pid_list_clear(pid_list, task->pid);
666}
667
668/**
669 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
670 * @pid_list: The pid list to show
671 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
672 * @pos: The position of the file
673 *
674 * This is used by the seq_file "next" operation to iterate the pids
675 * listed in a trace_pid_list structure.
676 *
677 * Returns the pid+1 as we want to display pid of zero, but NULL would
678 * stop the iteration.
679 */
680void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
681{
682 long pid = (unsigned long)v;
683 unsigned int next;
684
685 (*pos)++;
686
687 /* pid already is +1 of the actual previous bit */
688 if (trace_pid_list_next(pid_list, pid, &next) < 0)
689 return NULL;
690
691 pid = next;
692
693 /* Return pid + 1 to allow zero to be represented */
694 return (void *)(pid + 1);
695}
696
697/**
698 * trace_pid_start - Used for seq_file to start reading pid lists
699 * @pid_list: The pid list to show
700 * @pos: The position of the file
701 *
702 * This is used by seq_file "start" operation to start the iteration
703 * of listing pids.
704 *
705 * Returns the pid+1 as we want to display pid of zero, but NULL would
706 * stop the iteration.
707 */
708void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
709{
710 unsigned long pid;
711 unsigned int first;
712 loff_t l = 0;
713
714 if (trace_pid_list_first(pid_list, &first) < 0)
715 return NULL;
716
717 pid = first;
718
719 /* Return pid + 1 so that zero can be the exit value */
720 for (pid++; pid && l < *pos;
721 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
722 ;
723 return (void *)pid;
724}
725
726/**
727 * trace_pid_show - show the current pid in seq_file processing
728 * @m: The seq_file structure to write into
729 * @v: A void pointer of the pid (+1) value to display
730 *
731 * Can be directly used by seq_file operations to display the current
732 * pid value.
733 */
734int trace_pid_show(struct seq_file *m, void *v)
735{
736 unsigned long pid = (unsigned long)v - 1;
737
738 seq_printf(m, "%lu\n", pid);
739 return 0;
740}
741
742/* 128 should be much more than enough */
743#define PID_BUF_SIZE 127
744
745int trace_pid_write(struct trace_pid_list *filtered_pids,
746 struct trace_pid_list **new_pid_list,
747 const char __user *ubuf, size_t cnt)
748{
749 struct trace_pid_list *pid_list;
750 struct trace_parser parser;
751 unsigned long val;
752 int nr_pids = 0;
753 ssize_t read = 0;
754 ssize_t ret;
755 loff_t pos;
756 pid_t pid;
757
758 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
759 return -ENOMEM;
760
761 /*
762 * Always recreate a new array. The write is an all or nothing
763 * operation. Always create a new array when adding new pids by
764 * the user. If the operation fails, then the current list is
765 * not modified.
766 */
767 pid_list = trace_pid_list_alloc();
768 if (!pid_list) {
769 trace_parser_put(&parser);
770 return -ENOMEM;
771 }
772
773 if (filtered_pids) {
774 /* copy the current bits to the new max */
775 ret = trace_pid_list_first(filtered_pids, &pid);
776 while (!ret) {
777 trace_pid_list_set(pid_list, pid);
778 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
779 nr_pids++;
780 }
781 }
782
783 ret = 0;
784 while (cnt > 0) {
785
786 pos = 0;
787
788 ret = trace_get_user(&parser, ubuf, cnt, &pos);
789 if (ret < 0)
790 break;
791
792 read += ret;
793 ubuf += ret;
794 cnt -= ret;
795
796 if (!trace_parser_loaded(&parser))
797 break;
798
799 ret = -EINVAL;
800 if (kstrtoul(parser.buffer, 0, &val))
801 break;
802
803 pid = (pid_t)val;
804
805 if (trace_pid_list_set(pid_list, pid) < 0) {
806 ret = -1;
807 break;
808 }
809 nr_pids++;
810
811 trace_parser_clear(&parser);
812 ret = 0;
813 }
814 trace_parser_put(&parser);
815
816 if (ret < 0) {
817 trace_pid_list_free(pid_list);
818 return ret;
819 }
820
821 if (!nr_pids) {
822 /* Cleared the list of pids */
823 trace_pid_list_free(pid_list);
824 pid_list = NULL;
825 }
826
827 *new_pid_list = pid_list;
828
829 return read;
830}
831
832static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
833{
834 u64 ts;
835
836 /* Early boot up does not have a buffer yet */
837 if (!buf->buffer)
838 return trace_clock_local();
839
840 ts = ring_buffer_time_stamp(buf->buffer);
841 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
842
843 return ts;
844}
845
846u64 ftrace_now(int cpu)
847{
848 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
849}
850
851/**
852 * tracing_is_enabled - Show if global_trace has been enabled
853 *
854 * Shows if the global trace has been enabled or not. It uses the
855 * mirror flag "buffer_disabled" to be used in fast paths such as for
856 * the irqsoff tracer. But it may be inaccurate due to races. If you
857 * need to know the accurate state, use tracing_is_on() which is a little
858 * slower, but accurate.
859 */
860int tracing_is_enabled(void)
861{
862 /*
863 * For quick access (irqsoff uses this in fast path), just
864 * return the mirror variable of the state of the ring buffer.
865 * It's a little racy, but we don't really care.
866 */
867 smp_rmb();
868 return !global_trace.buffer_disabled;
869}
870
871/*
872 * trace_buf_size is the size in bytes that is allocated
873 * for a buffer. Note, the number of bytes is always rounded
874 * to page size.
875 *
876 * This number is purposely set to a low number of 16384.
877 * If the dump on oops happens, it will be much appreciated
878 * to not have to wait for all that output. Anyway this can be
879 * boot time and run time configurable.
880 */
881#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
882
883static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
884
885/* trace_types holds a link list of available tracers. */
886static struct tracer *trace_types __read_mostly;
887
888/*
889 * trace_types_lock is used to protect the trace_types list.
890 */
891DEFINE_MUTEX(trace_types_lock);
892
893/*
894 * serialize the access of the ring buffer
895 *
896 * ring buffer serializes readers, but it is low level protection.
897 * The validity of the events (which returns by ring_buffer_peek() ..etc)
898 * are not protected by ring buffer.
899 *
900 * The content of events may become garbage if we allow other process consumes
901 * these events concurrently:
902 * A) the page of the consumed events may become a normal page
903 * (not reader page) in ring buffer, and this page will be rewritten
904 * by events producer.
905 * B) The page of the consumed events may become a page for splice_read,
906 * and this page will be returned to system.
907 *
908 * These primitives allow multi process access to different cpu ring buffer
909 * concurrently.
910 *
911 * These primitives don't distinguish read-only and read-consume access.
912 * Multi read-only access are also serialized.
913 */
914
915#ifdef CONFIG_SMP
916static DECLARE_RWSEM(all_cpu_access_lock);
917static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
918
919static inline void trace_access_lock(int cpu)
920{
921 if (cpu == RING_BUFFER_ALL_CPUS) {
922 /* gain it for accessing the whole ring buffer. */
923 down_write(&all_cpu_access_lock);
924 } else {
925 /* gain it for accessing a cpu ring buffer. */
926
927 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
928 down_read(&all_cpu_access_lock);
929
930 /* Secondly block other access to this @cpu ring buffer. */
931 mutex_lock(&per_cpu(cpu_access_lock, cpu));
932 }
933}
934
935static inline void trace_access_unlock(int cpu)
936{
937 if (cpu == RING_BUFFER_ALL_CPUS) {
938 up_write(&all_cpu_access_lock);
939 } else {
940 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
941 up_read(&all_cpu_access_lock);
942 }
943}
944
945static inline void trace_access_lock_init(void)
946{
947 int cpu;
948
949 for_each_possible_cpu(cpu)
950 mutex_init(&per_cpu(cpu_access_lock, cpu));
951}
952
953#else
954
955static DEFINE_MUTEX(access_lock);
956
957static inline void trace_access_lock(int cpu)
958{
959 (void)cpu;
960 mutex_lock(&access_lock);
961}
962
963static inline void trace_access_unlock(int cpu)
964{
965 (void)cpu;
966 mutex_unlock(&access_lock);
967}
968
969static inline void trace_access_lock_init(void)
970{
971}
972
973#endif
974
975#ifdef CONFIG_STACKTRACE
976static void __ftrace_trace_stack(struct trace_array *tr,
977 struct trace_buffer *buffer,
978 unsigned int trace_ctx,
979 int skip, struct pt_regs *regs);
980static inline void ftrace_trace_stack(struct trace_array *tr,
981 struct trace_buffer *buffer,
982 unsigned int trace_ctx,
983 int skip, struct pt_regs *regs);
984
985#else
986static inline void __ftrace_trace_stack(struct trace_array *tr,
987 struct trace_buffer *buffer,
988 unsigned int trace_ctx,
989 int skip, struct pt_regs *regs)
990{
991}
992static inline void ftrace_trace_stack(struct trace_array *tr,
993 struct trace_buffer *buffer,
994 unsigned long trace_ctx,
995 int skip, struct pt_regs *regs)
996{
997}
998
999#endif
1000
1001static __always_inline void
1002trace_event_setup(struct ring_buffer_event *event,
1003 int type, unsigned int trace_ctx)
1004{
1005 struct trace_entry *ent = ring_buffer_event_data(event);
1006
1007 tracing_generic_entry_update(ent, type, trace_ctx);
1008}
1009
1010static __always_inline struct ring_buffer_event *
1011__trace_buffer_lock_reserve(struct trace_buffer *buffer,
1012 int type,
1013 unsigned long len,
1014 unsigned int trace_ctx)
1015{
1016 struct ring_buffer_event *event;
1017
1018 event = ring_buffer_lock_reserve(buffer, len);
1019 if (event != NULL)
1020 trace_event_setup(event, type, trace_ctx);
1021
1022 return event;
1023}
1024
1025void tracer_tracing_on(struct trace_array *tr)
1026{
1027 if (tr->array_buffer.buffer)
1028 ring_buffer_record_on(tr->array_buffer.buffer);
1029 /*
1030 * This flag is looked at when buffers haven't been allocated
1031 * yet, or by some tracers (like irqsoff), that just want to
1032 * know if the ring buffer has been disabled, but it can handle
1033 * races of where it gets disabled but we still do a record.
1034 * As the check is in the fast path of the tracers, it is more
1035 * important to be fast than accurate.
1036 */
1037 tr->buffer_disabled = 0;
1038 /* Make the flag seen by readers */
1039 smp_wmb();
1040}
1041
1042/**
1043 * tracing_on - enable tracing buffers
1044 *
1045 * This function enables tracing buffers that may have been
1046 * disabled with tracing_off.
1047 */
1048void tracing_on(void)
1049{
1050 tracer_tracing_on(&global_trace);
1051}
1052EXPORT_SYMBOL_GPL(tracing_on);
1053
1054
1055static __always_inline void
1056__buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
1057{
1058 __this_cpu_write(trace_taskinfo_save, true);
1059
1060 /* If this is the temp buffer, we need to commit fully */
1061 if (this_cpu_read(trace_buffered_event) == event) {
1062 /* Length is in event->array[0] */
1063 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1064 /* Release the temp buffer */
1065 this_cpu_dec(trace_buffered_event_cnt);
1066 /* ring_buffer_unlock_commit() enables preemption */
1067 preempt_enable_notrace();
1068 } else
1069 ring_buffer_unlock_commit(buffer);
1070}
1071
1072int __trace_array_puts(struct trace_array *tr, unsigned long ip,
1073 const char *str, int size)
1074{
1075 struct ring_buffer_event *event;
1076 struct trace_buffer *buffer;
1077 struct print_entry *entry;
1078 unsigned int trace_ctx;
1079 int alloc;
1080
1081 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1082 return 0;
1083
1084 if (unlikely(tracing_selftest_running && tr == &global_trace))
1085 return 0;
1086
1087 if (unlikely(tracing_disabled))
1088 return 0;
1089
1090 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1091
1092 trace_ctx = tracing_gen_ctx();
1093 buffer = tr->array_buffer.buffer;
1094 ring_buffer_nest_start(buffer);
1095 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1096 trace_ctx);
1097 if (!event) {
1098 size = 0;
1099 goto out;
1100 }
1101
1102 entry = ring_buffer_event_data(event);
1103 entry->ip = ip;
1104
1105 memcpy(&entry->buf, str, size);
1106
1107 /* Add a newline if necessary */
1108 if (entry->buf[size - 1] != '\n') {
1109 entry->buf[size] = '\n';
1110 entry->buf[size + 1] = '\0';
1111 } else
1112 entry->buf[size] = '\0';
1113
1114 __buffer_unlock_commit(buffer, event);
1115 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1116 out:
1117 ring_buffer_nest_end(buffer);
1118 return size;
1119}
1120EXPORT_SYMBOL_GPL(__trace_array_puts);
1121
1122/**
1123 * __trace_puts - write a constant string into the trace buffer.
1124 * @ip: The address of the caller
1125 * @str: The constant string to write
1126 * @size: The size of the string.
1127 */
1128int __trace_puts(unsigned long ip, const char *str, int size)
1129{
1130 return __trace_array_puts(printk_trace, ip, str, size);
1131}
1132EXPORT_SYMBOL_GPL(__trace_puts);
1133
1134/**
1135 * __trace_bputs - write the pointer to a constant string into trace buffer
1136 * @ip: The address of the caller
1137 * @str: The constant string to write to the buffer to
1138 */
1139int __trace_bputs(unsigned long ip, const char *str)
1140{
1141 struct trace_array *tr = READ_ONCE(printk_trace);
1142 struct ring_buffer_event *event;
1143 struct trace_buffer *buffer;
1144 struct bputs_entry *entry;
1145 unsigned int trace_ctx;
1146 int size = sizeof(struct bputs_entry);
1147 int ret = 0;
1148
1149 if (!printk_binsafe(tr))
1150 return __trace_puts(ip, str, strlen(str));
1151
1152 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1153 return 0;
1154
1155 if (unlikely(tracing_selftest_running || tracing_disabled))
1156 return 0;
1157
1158 trace_ctx = tracing_gen_ctx();
1159 buffer = tr->array_buffer.buffer;
1160
1161 ring_buffer_nest_start(buffer);
1162 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1163 trace_ctx);
1164 if (!event)
1165 goto out;
1166
1167 entry = ring_buffer_event_data(event);
1168 entry->ip = ip;
1169 entry->str = str;
1170
1171 __buffer_unlock_commit(buffer, event);
1172 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1173
1174 ret = 1;
1175 out:
1176 ring_buffer_nest_end(buffer);
1177 return ret;
1178}
1179EXPORT_SYMBOL_GPL(__trace_bputs);
1180
1181#ifdef CONFIG_TRACER_SNAPSHOT
1182static void tracing_snapshot_instance_cond(struct trace_array *tr,
1183 void *cond_data)
1184{
1185 struct tracer *tracer = tr->current_trace;
1186 unsigned long flags;
1187
1188 if (in_nmi()) {
1189 trace_array_puts(tr, "*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1190 trace_array_puts(tr, "*** snapshot is being ignored ***\n");
1191 return;
1192 }
1193
1194 if (!tr->allocated_snapshot) {
1195 trace_array_puts(tr, "*** SNAPSHOT NOT ALLOCATED ***\n");
1196 trace_array_puts(tr, "*** stopping trace here! ***\n");
1197 tracer_tracing_off(tr);
1198 return;
1199 }
1200
1201 /* Note, snapshot can not be used when the tracer uses it */
1202 if (tracer->use_max_tr) {
1203 trace_array_puts(tr, "*** LATENCY TRACER ACTIVE ***\n");
1204 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1205 return;
1206 }
1207
1208 if (tr->mapped) {
1209 trace_array_puts(tr, "*** BUFFER MEMORY MAPPED ***\n");
1210 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1211 return;
1212 }
1213
1214 local_irq_save(flags);
1215 update_max_tr(tr, current, smp_processor_id(), cond_data);
1216 local_irq_restore(flags);
1217}
1218
1219void tracing_snapshot_instance(struct trace_array *tr)
1220{
1221 tracing_snapshot_instance_cond(tr, NULL);
1222}
1223
1224/**
1225 * tracing_snapshot - take a snapshot of the current buffer.
1226 *
1227 * This causes a swap between the snapshot buffer and the current live
1228 * tracing buffer. You can use this to take snapshots of the live
1229 * trace when some condition is triggered, but continue to trace.
1230 *
1231 * Note, make sure to allocate the snapshot with either
1232 * a tracing_snapshot_alloc(), or by doing it manually
1233 * with: echo 1 > /sys/kernel/tracing/snapshot
1234 *
1235 * If the snapshot buffer is not allocated, it will stop tracing.
1236 * Basically making a permanent snapshot.
1237 */
1238void tracing_snapshot(void)
1239{
1240 struct trace_array *tr = &global_trace;
1241
1242 tracing_snapshot_instance(tr);
1243}
1244EXPORT_SYMBOL_GPL(tracing_snapshot);
1245
1246/**
1247 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1248 * @tr: The tracing instance to snapshot
1249 * @cond_data: The data to be tested conditionally, and possibly saved
1250 *
1251 * This is the same as tracing_snapshot() except that the snapshot is
1252 * conditional - the snapshot will only happen if the
1253 * cond_snapshot.update() implementation receiving the cond_data
1254 * returns true, which means that the trace array's cond_snapshot
1255 * update() operation used the cond_data to determine whether the
1256 * snapshot should be taken, and if it was, presumably saved it along
1257 * with the snapshot.
1258 */
1259void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1260{
1261 tracing_snapshot_instance_cond(tr, cond_data);
1262}
1263EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1264
1265/**
1266 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1267 * @tr: The tracing instance
1268 *
1269 * When the user enables a conditional snapshot using
1270 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1271 * with the snapshot. This accessor is used to retrieve it.
1272 *
1273 * Should not be called from cond_snapshot.update(), since it takes
1274 * the tr->max_lock lock, which the code calling
1275 * cond_snapshot.update() has already done.
1276 *
1277 * Returns the cond_data associated with the trace array's snapshot.
1278 */
1279void *tracing_cond_snapshot_data(struct trace_array *tr)
1280{
1281 void *cond_data = NULL;
1282
1283 local_irq_disable();
1284 arch_spin_lock(&tr->max_lock);
1285
1286 if (tr->cond_snapshot)
1287 cond_data = tr->cond_snapshot->cond_data;
1288
1289 arch_spin_unlock(&tr->max_lock);
1290 local_irq_enable();
1291
1292 return cond_data;
1293}
1294EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1295
1296static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1297 struct array_buffer *size_buf, int cpu_id);
1298static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1299
1300int tracing_alloc_snapshot_instance(struct trace_array *tr)
1301{
1302 int order;
1303 int ret;
1304
1305 if (!tr->allocated_snapshot) {
1306
1307 /* Make the snapshot buffer have the same order as main buffer */
1308 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
1309 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
1310 if (ret < 0)
1311 return ret;
1312
1313 /* allocate spare buffer */
1314 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1315 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1316 if (ret < 0)
1317 return ret;
1318
1319 tr->allocated_snapshot = true;
1320 }
1321
1322 return 0;
1323}
1324
1325static void free_snapshot(struct trace_array *tr)
1326{
1327 /*
1328 * We don't free the ring buffer. instead, resize it because
1329 * The max_tr ring buffer has some state (e.g. ring->clock) and
1330 * we want preserve it.
1331 */
1332 ring_buffer_subbuf_order_set(tr->max_buffer.buffer, 0);
1333 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1334 set_buffer_entries(&tr->max_buffer, 1);
1335 tracing_reset_online_cpus(&tr->max_buffer);
1336 tr->allocated_snapshot = false;
1337}
1338
1339static int tracing_arm_snapshot_locked(struct trace_array *tr)
1340{
1341 int ret;
1342
1343 lockdep_assert_held(&trace_types_lock);
1344
1345 spin_lock(&tr->snapshot_trigger_lock);
1346 if (tr->snapshot == UINT_MAX || tr->mapped) {
1347 spin_unlock(&tr->snapshot_trigger_lock);
1348 return -EBUSY;
1349 }
1350
1351 tr->snapshot++;
1352 spin_unlock(&tr->snapshot_trigger_lock);
1353
1354 ret = tracing_alloc_snapshot_instance(tr);
1355 if (ret) {
1356 spin_lock(&tr->snapshot_trigger_lock);
1357 tr->snapshot--;
1358 spin_unlock(&tr->snapshot_trigger_lock);
1359 }
1360
1361 return ret;
1362}
1363
1364int tracing_arm_snapshot(struct trace_array *tr)
1365{
1366 int ret;
1367
1368 mutex_lock(&trace_types_lock);
1369 ret = tracing_arm_snapshot_locked(tr);
1370 mutex_unlock(&trace_types_lock);
1371
1372 return ret;
1373}
1374
1375void tracing_disarm_snapshot(struct trace_array *tr)
1376{
1377 spin_lock(&tr->snapshot_trigger_lock);
1378 if (!WARN_ON(!tr->snapshot))
1379 tr->snapshot--;
1380 spin_unlock(&tr->snapshot_trigger_lock);
1381}
1382
1383/**
1384 * tracing_alloc_snapshot - allocate snapshot buffer.
1385 *
1386 * This only allocates the snapshot buffer if it isn't already
1387 * allocated - it doesn't also take a snapshot.
1388 *
1389 * This is meant to be used in cases where the snapshot buffer needs
1390 * to be set up for events that can't sleep but need to be able to
1391 * trigger a snapshot.
1392 */
1393int tracing_alloc_snapshot(void)
1394{
1395 struct trace_array *tr = &global_trace;
1396 int ret;
1397
1398 ret = tracing_alloc_snapshot_instance(tr);
1399 WARN_ON(ret < 0);
1400
1401 return ret;
1402}
1403EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1404
1405/**
1406 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1407 *
1408 * This is similar to tracing_snapshot(), but it will allocate the
1409 * snapshot buffer if it isn't already allocated. Use this only
1410 * where it is safe to sleep, as the allocation may sleep.
1411 *
1412 * This causes a swap between the snapshot buffer and the current live
1413 * tracing buffer. You can use this to take snapshots of the live
1414 * trace when some condition is triggered, but continue to trace.
1415 */
1416void tracing_snapshot_alloc(void)
1417{
1418 int ret;
1419
1420 ret = tracing_alloc_snapshot();
1421 if (ret < 0)
1422 return;
1423
1424 tracing_snapshot();
1425}
1426EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1427
1428/**
1429 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1430 * @tr: The tracing instance
1431 * @cond_data: User data to associate with the snapshot
1432 * @update: Implementation of the cond_snapshot update function
1433 *
1434 * Check whether the conditional snapshot for the given instance has
1435 * already been enabled, or if the current tracer is already using a
1436 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1437 * save the cond_data and update function inside.
1438 *
1439 * Returns 0 if successful, error otherwise.
1440 */
1441int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1442 cond_update_fn_t update)
1443{
1444 struct cond_snapshot *cond_snapshot __free(kfree) =
1445 kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1446 int ret;
1447
1448 if (!cond_snapshot)
1449 return -ENOMEM;
1450
1451 cond_snapshot->cond_data = cond_data;
1452 cond_snapshot->update = update;
1453
1454 guard(mutex)(&trace_types_lock);
1455
1456 if (tr->current_trace->use_max_tr)
1457 return -EBUSY;
1458
1459 /*
1460 * The cond_snapshot can only change to NULL without the
1461 * trace_types_lock. We don't care if we race with it going
1462 * to NULL, but we want to make sure that it's not set to
1463 * something other than NULL when we get here, which we can
1464 * do safely with only holding the trace_types_lock and not
1465 * having to take the max_lock.
1466 */
1467 if (tr->cond_snapshot)
1468 return -EBUSY;
1469
1470 ret = tracing_arm_snapshot_locked(tr);
1471 if (ret)
1472 return ret;
1473
1474 local_irq_disable();
1475 arch_spin_lock(&tr->max_lock);
1476 tr->cond_snapshot = no_free_ptr(cond_snapshot);
1477 arch_spin_unlock(&tr->max_lock);
1478 local_irq_enable();
1479
1480 return 0;
1481}
1482EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1483
1484/**
1485 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1486 * @tr: The tracing instance
1487 *
1488 * Check whether the conditional snapshot for the given instance is
1489 * enabled; if so, free the cond_snapshot associated with it,
1490 * otherwise return -EINVAL.
1491 *
1492 * Returns 0 if successful, error otherwise.
1493 */
1494int tracing_snapshot_cond_disable(struct trace_array *tr)
1495{
1496 int ret = 0;
1497
1498 local_irq_disable();
1499 arch_spin_lock(&tr->max_lock);
1500
1501 if (!tr->cond_snapshot)
1502 ret = -EINVAL;
1503 else {
1504 kfree(tr->cond_snapshot);
1505 tr->cond_snapshot = NULL;
1506 }
1507
1508 arch_spin_unlock(&tr->max_lock);
1509 local_irq_enable();
1510
1511 tracing_disarm_snapshot(tr);
1512
1513 return ret;
1514}
1515EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1516#else
1517void tracing_snapshot(void)
1518{
1519 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1520}
1521EXPORT_SYMBOL_GPL(tracing_snapshot);
1522void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1523{
1524 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1525}
1526EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1527int tracing_alloc_snapshot(void)
1528{
1529 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1530 return -ENODEV;
1531}
1532EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1533void tracing_snapshot_alloc(void)
1534{
1535 /* Give warning */
1536 tracing_snapshot();
1537}
1538EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1539void *tracing_cond_snapshot_data(struct trace_array *tr)
1540{
1541 return NULL;
1542}
1543EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1544int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1545{
1546 return -ENODEV;
1547}
1548EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1549int tracing_snapshot_cond_disable(struct trace_array *tr)
1550{
1551 return false;
1552}
1553EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1554#define free_snapshot(tr) do { } while (0)
1555#define tracing_arm_snapshot_locked(tr) ({ -EBUSY; })
1556#endif /* CONFIG_TRACER_SNAPSHOT */
1557
1558void tracer_tracing_off(struct trace_array *tr)
1559{
1560 if (tr->array_buffer.buffer)
1561 ring_buffer_record_off(tr->array_buffer.buffer);
1562 /*
1563 * This flag is looked at when buffers haven't been allocated
1564 * yet, or by some tracers (like irqsoff), that just want to
1565 * know if the ring buffer has been disabled, but it can handle
1566 * races of where it gets disabled but we still do a record.
1567 * As the check is in the fast path of the tracers, it is more
1568 * important to be fast than accurate.
1569 */
1570 tr->buffer_disabled = 1;
1571 /* Make the flag seen by readers */
1572 smp_wmb();
1573}
1574
1575/**
1576 * tracing_off - turn off tracing buffers
1577 *
1578 * This function stops the tracing buffers from recording data.
1579 * It does not disable any overhead the tracers themselves may
1580 * be causing. This function simply causes all recording to
1581 * the ring buffers to fail.
1582 */
1583void tracing_off(void)
1584{
1585 tracer_tracing_off(&global_trace);
1586}
1587EXPORT_SYMBOL_GPL(tracing_off);
1588
1589void disable_trace_on_warning(void)
1590{
1591 if (__disable_trace_on_warning) {
1592 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1593 "Disabling tracing due to warning\n");
1594 tracing_off();
1595 }
1596}
1597
1598/**
1599 * tracer_tracing_is_on - show real state of ring buffer enabled
1600 * @tr : the trace array to know if ring buffer is enabled
1601 *
1602 * Shows real state of the ring buffer if it is enabled or not.
1603 */
1604bool tracer_tracing_is_on(struct trace_array *tr)
1605{
1606 if (tr->array_buffer.buffer)
1607 return ring_buffer_record_is_set_on(tr->array_buffer.buffer);
1608 return !tr->buffer_disabled;
1609}
1610
1611/**
1612 * tracing_is_on - show state of ring buffers enabled
1613 */
1614int tracing_is_on(void)
1615{
1616 return tracer_tracing_is_on(&global_trace);
1617}
1618EXPORT_SYMBOL_GPL(tracing_is_on);
1619
1620static int __init set_buf_size(char *str)
1621{
1622 unsigned long buf_size;
1623
1624 if (!str)
1625 return 0;
1626 buf_size = memparse(str, &str);
1627 /*
1628 * nr_entries can not be zero and the startup
1629 * tests require some buffer space. Therefore
1630 * ensure we have at least 4096 bytes of buffer.
1631 */
1632 trace_buf_size = max(4096UL, buf_size);
1633 return 1;
1634}
1635__setup("trace_buf_size=", set_buf_size);
1636
1637static int __init set_tracing_thresh(char *str)
1638{
1639 unsigned long threshold;
1640 int ret;
1641
1642 if (!str)
1643 return 0;
1644 ret = kstrtoul(str, 0, &threshold);
1645 if (ret < 0)
1646 return 0;
1647 tracing_thresh = threshold * 1000;
1648 return 1;
1649}
1650__setup("tracing_thresh=", set_tracing_thresh);
1651
1652unsigned long nsecs_to_usecs(unsigned long nsecs)
1653{
1654 return nsecs / 1000;
1655}
1656
1657/*
1658 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1659 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1660 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1661 * of strings in the order that the evals (enum) were defined.
1662 */
1663#undef C
1664#define C(a, b) b
1665
1666/* These must match the bit positions in trace_iterator_flags */
1667static const char *trace_options[] = {
1668 TRACE_FLAGS
1669 NULL
1670};
1671
1672static struct {
1673 u64 (*func)(void);
1674 const char *name;
1675 int in_ns; /* is this clock in nanoseconds? */
1676} trace_clocks[] = {
1677 { trace_clock_local, "local", 1 },
1678 { trace_clock_global, "global", 1 },
1679 { trace_clock_counter, "counter", 0 },
1680 { trace_clock_jiffies, "uptime", 0 },
1681 { trace_clock, "perf", 1 },
1682 { ktime_get_mono_fast_ns, "mono", 1 },
1683 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1684 { ktime_get_boot_fast_ns, "boot", 1 },
1685 { ktime_get_tai_fast_ns, "tai", 1 },
1686 ARCH_TRACE_CLOCKS
1687};
1688
1689bool trace_clock_in_ns(struct trace_array *tr)
1690{
1691 if (trace_clocks[tr->clock_id].in_ns)
1692 return true;
1693
1694 return false;
1695}
1696
1697/*
1698 * trace_parser_get_init - gets the buffer for trace parser
1699 */
1700int trace_parser_get_init(struct trace_parser *parser, int size)
1701{
1702 memset(parser, 0, sizeof(*parser));
1703
1704 parser->buffer = kmalloc(size, GFP_KERNEL);
1705 if (!parser->buffer)
1706 return 1;
1707
1708 parser->size = size;
1709 return 0;
1710}
1711
1712/*
1713 * trace_parser_put - frees the buffer for trace parser
1714 */
1715void trace_parser_put(struct trace_parser *parser)
1716{
1717 kfree(parser->buffer);
1718 parser->buffer = NULL;
1719}
1720
1721/*
1722 * trace_get_user - reads the user input string separated by space
1723 * (matched by isspace(ch))
1724 *
1725 * For each string found the 'struct trace_parser' is updated,
1726 * and the function returns.
1727 *
1728 * Returns number of bytes read.
1729 *
1730 * See kernel/trace/trace.h for 'struct trace_parser' details.
1731 */
1732int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1733 size_t cnt, loff_t *ppos)
1734{
1735 char ch;
1736 size_t read = 0;
1737 ssize_t ret;
1738
1739 if (!*ppos)
1740 trace_parser_clear(parser);
1741
1742 ret = get_user(ch, ubuf++);
1743 if (ret)
1744 goto out;
1745
1746 read++;
1747 cnt--;
1748
1749 /*
1750 * The parser is not finished with the last write,
1751 * continue reading the user input without skipping spaces.
1752 */
1753 if (!parser->cont) {
1754 /* skip white space */
1755 while (cnt && isspace(ch)) {
1756 ret = get_user(ch, ubuf++);
1757 if (ret)
1758 goto out;
1759 read++;
1760 cnt--;
1761 }
1762
1763 parser->idx = 0;
1764
1765 /* only spaces were written */
1766 if (isspace(ch) || !ch) {
1767 *ppos += read;
1768 ret = read;
1769 goto out;
1770 }
1771 }
1772
1773 /* read the non-space input */
1774 while (cnt && !isspace(ch) && ch) {
1775 if (parser->idx < parser->size - 1)
1776 parser->buffer[parser->idx++] = ch;
1777 else {
1778 ret = -EINVAL;
1779 goto out;
1780 }
1781 ret = get_user(ch, ubuf++);
1782 if (ret)
1783 goto out;
1784 read++;
1785 cnt--;
1786 }
1787
1788 /* We either got finished input or we have to wait for another call. */
1789 if (isspace(ch) || !ch) {
1790 parser->buffer[parser->idx] = 0;
1791 parser->cont = false;
1792 } else if (parser->idx < parser->size - 1) {
1793 parser->cont = true;
1794 parser->buffer[parser->idx++] = ch;
1795 /* Make sure the parsed string always terminates with '\0'. */
1796 parser->buffer[parser->idx] = 0;
1797 } else {
1798 ret = -EINVAL;
1799 goto out;
1800 }
1801
1802 *ppos += read;
1803 ret = read;
1804
1805out:
1806 return ret;
1807}
1808
1809/* TODO add a seq_buf_to_buffer() */
1810static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1811{
1812 int len;
1813
1814 if (trace_seq_used(s) <= s->readpos)
1815 return -EBUSY;
1816
1817 len = trace_seq_used(s) - s->readpos;
1818 if (cnt > len)
1819 cnt = len;
1820 memcpy(buf, s->buffer + s->readpos, cnt);
1821
1822 s->readpos += cnt;
1823 return cnt;
1824}
1825
1826unsigned long __read_mostly tracing_thresh;
1827
1828#ifdef CONFIG_TRACER_MAX_TRACE
1829static const struct file_operations tracing_max_lat_fops;
1830
1831#ifdef LATENCY_FS_NOTIFY
1832
1833static struct workqueue_struct *fsnotify_wq;
1834
1835static void latency_fsnotify_workfn(struct work_struct *work)
1836{
1837 struct trace_array *tr = container_of(work, struct trace_array,
1838 fsnotify_work);
1839 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1840}
1841
1842static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1843{
1844 struct trace_array *tr = container_of(iwork, struct trace_array,
1845 fsnotify_irqwork);
1846 queue_work(fsnotify_wq, &tr->fsnotify_work);
1847}
1848
1849static void trace_create_maxlat_file(struct trace_array *tr,
1850 struct dentry *d_tracer)
1851{
1852 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1853 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1854 tr->d_max_latency = trace_create_file("tracing_max_latency",
1855 TRACE_MODE_WRITE,
1856 d_tracer, tr,
1857 &tracing_max_lat_fops);
1858}
1859
1860__init static int latency_fsnotify_init(void)
1861{
1862 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1863 WQ_UNBOUND | WQ_HIGHPRI, 0);
1864 if (!fsnotify_wq) {
1865 pr_err("Unable to allocate tr_max_lat_wq\n");
1866 return -ENOMEM;
1867 }
1868 return 0;
1869}
1870
1871late_initcall_sync(latency_fsnotify_init);
1872
1873void latency_fsnotify(struct trace_array *tr)
1874{
1875 if (!fsnotify_wq)
1876 return;
1877 /*
1878 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1879 * possible that we are called from __schedule() or do_idle(), which
1880 * could cause a deadlock.
1881 */
1882 irq_work_queue(&tr->fsnotify_irqwork);
1883}
1884
1885#else /* !LATENCY_FS_NOTIFY */
1886
1887#define trace_create_maxlat_file(tr, d_tracer) \
1888 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1889 d_tracer, tr, &tracing_max_lat_fops)
1890
1891#endif
1892
1893/*
1894 * Copy the new maximum trace into the separate maximum-trace
1895 * structure. (this way the maximum trace is permanently saved,
1896 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1897 */
1898static void
1899__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1900{
1901 struct array_buffer *trace_buf = &tr->array_buffer;
1902 struct array_buffer *max_buf = &tr->max_buffer;
1903 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1904 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1905
1906 max_buf->cpu = cpu;
1907 max_buf->time_start = data->preempt_timestamp;
1908
1909 max_data->saved_latency = tr->max_latency;
1910 max_data->critical_start = data->critical_start;
1911 max_data->critical_end = data->critical_end;
1912
1913 strscpy(max_data->comm, tsk->comm);
1914 max_data->pid = tsk->pid;
1915 /*
1916 * If tsk == current, then use current_uid(), as that does not use
1917 * RCU. The irq tracer can be called out of RCU scope.
1918 */
1919 if (tsk == current)
1920 max_data->uid = current_uid();
1921 else
1922 max_data->uid = task_uid(tsk);
1923
1924 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1925 max_data->policy = tsk->policy;
1926 max_data->rt_priority = tsk->rt_priority;
1927
1928 /* record this tasks comm */
1929 tracing_record_cmdline(tsk);
1930 latency_fsnotify(tr);
1931}
1932
1933/**
1934 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1935 * @tr: tracer
1936 * @tsk: the task with the latency
1937 * @cpu: The cpu that initiated the trace.
1938 * @cond_data: User data associated with a conditional snapshot
1939 *
1940 * Flip the buffers between the @tr and the max_tr and record information
1941 * about which task was the cause of this latency.
1942 */
1943void
1944update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1945 void *cond_data)
1946{
1947 if (tr->stop_count)
1948 return;
1949
1950 WARN_ON_ONCE(!irqs_disabled());
1951
1952 if (!tr->allocated_snapshot) {
1953 /* Only the nop tracer should hit this when disabling */
1954 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1955 return;
1956 }
1957
1958 arch_spin_lock(&tr->max_lock);
1959
1960 /* Inherit the recordable setting from array_buffer */
1961 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1962 ring_buffer_record_on(tr->max_buffer.buffer);
1963 else
1964 ring_buffer_record_off(tr->max_buffer.buffer);
1965
1966#ifdef CONFIG_TRACER_SNAPSHOT
1967 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1968 arch_spin_unlock(&tr->max_lock);
1969 return;
1970 }
1971#endif
1972 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1973
1974 __update_max_tr(tr, tsk, cpu);
1975
1976 arch_spin_unlock(&tr->max_lock);
1977
1978 /* Any waiters on the old snapshot buffer need to wake up */
1979 ring_buffer_wake_waiters(tr->array_buffer.buffer, RING_BUFFER_ALL_CPUS);
1980}
1981
1982/**
1983 * update_max_tr_single - only copy one trace over, and reset the rest
1984 * @tr: tracer
1985 * @tsk: task with the latency
1986 * @cpu: the cpu of the buffer to copy.
1987 *
1988 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1989 */
1990void
1991update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1992{
1993 int ret;
1994
1995 if (tr->stop_count)
1996 return;
1997
1998 WARN_ON_ONCE(!irqs_disabled());
1999 if (!tr->allocated_snapshot) {
2000 /* Only the nop tracer should hit this when disabling */
2001 WARN_ON_ONCE(tr->current_trace != &nop_trace);
2002 return;
2003 }
2004
2005 arch_spin_lock(&tr->max_lock);
2006
2007 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
2008
2009 if (ret == -EBUSY) {
2010 /*
2011 * We failed to swap the buffer due to a commit taking
2012 * place on this CPU. We fail to record, but we reset
2013 * the max trace buffer (no one writes directly to it)
2014 * and flag that it failed.
2015 * Another reason is resize is in progress.
2016 */
2017 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
2018 "Failed to swap buffers due to commit or resize in progress\n");
2019 }
2020
2021 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
2022
2023 __update_max_tr(tr, tsk, cpu);
2024 arch_spin_unlock(&tr->max_lock);
2025}
2026
2027#endif /* CONFIG_TRACER_MAX_TRACE */
2028
2029struct pipe_wait {
2030 struct trace_iterator *iter;
2031 int wait_index;
2032};
2033
2034static bool wait_pipe_cond(void *data)
2035{
2036 struct pipe_wait *pwait = data;
2037 struct trace_iterator *iter = pwait->iter;
2038
2039 if (atomic_read_acquire(&iter->wait_index) != pwait->wait_index)
2040 return true;
2041
2042 return iter->closed;
2043}
2044
2045static int wait_on_pipe(struct trace_iterator *iter, int full)
2046{
2047 struct pipe_wait pwait;
2048 int ret;
2049
2050 /* Iterators are static, they should be filled or empty */
2051 if (trace_buffer_iter(iter, iter->cpu_file))
2052 return 0;
2053
2054 pwait.wait_index = atomic_read_acquire(&iter->wait_index);
2055 pwait.iter = iter;
2056
2057 ret = ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file, full,
2058 wait_pipe_cond, &pwait);
2059
2060#ifdef CONFIG_TRACER_MAX_TRACE
2061 /*
2062 * Make sure this is still the snapshot buffer, as if a snapshot were
2063 * to happen, this would now be the main buffer.
2064 */
2065 if (iter->snapshot)
2066 iter->array_buffer = &iter->tr->max_buffer;
2067#endif
2068 return ret;
2069}
2070
2071#ifdef CONFIG_FTRACE_STARTUP_TEST
2072static bool selftests_can_run;
2073
2074struct trace_selftests {
2075 struct list_head list;
2076 struct tracer *type;
2077};
2078
2079static LIST_HEAD(postponed_selftests);
2080
2081static int save_selftest(struct tracer *type)
2082{
2083 struct trace_selftests *selftest;
2084
2085 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
2086 if (!selftest)
2087 return -ENOMEM;
2088
2089 selftest->type = type;
2090 list_add(&selftest->list, &postponed_selftests);
2091 return 0;
2092}
2093
2094static int run_tracer_selftest(struct tracer *type)
2095{
2096 struct trace_array *tr = &global_trace;
2097 struct tracer *saved_tracer = tr->current_trace;
2098 int ret;
2099
2100 if (!type->selftest || tracing_selftest_disabled)
2101 return 0;
2102
2103 /*
2104 * If a tracer registers early in boot up (before scheduling is
2105 * initialized and such), then do not run its selftests yet.
2106 * Instead, run it a little later in the boot process.
2107 */
2108 if (!selftests_can_run)
2109 return save_selftest(type);
2110
2111 if (!tracing_is_on()) {
2112 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2113 type->name);
2114 return 0;
2115 }
2116
2117 /*
2118 * Run a selftest on this tracer.
2119 * Here we reset the trace buffer, and set the current
2120 * tracer to be this tracer. The tracer can then run some
2121 * internal tracing to verify that everything is in order.
2122 * If we fail, we do not register this tracer.
2123 */
2124 tracing_reset_online_cpus(&tr->array_buffer);
2125
2126 tr->current_trace = type;
2127
2128#ifdef CONFIG_TRACER_MAX_TRACE
2129 if (type->use_max_tr) {
2130 /* If we expanded the buffers, make sure the max is expanded too */
2131 if (tr->ring_buffer_expanded)
2132 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2133 RING_BUFFER_ALL_CPUS);
2134 tr->allocated_snapshot = true;
2135 }
2136#endif
2137
2138 /* the test is responsible for initializing and enabling */
2139 pr_info("Testing tracer %s: ", type->name);
2140 ret = type->selftest(type, tr);
2141 /* the test is responsible for resetting too */
2142 tr->current_trace = saved_tracer;
2143 if (ret) {
2144 printk(KERN_CONT "FAILED!\n");
2145 /* Add the warning after printing 'FAILED' */
2146 WARN_ON(1);
2147 return -1;
2148 }
2149 /* Only reset on passing, to avoid touching corrupted buffers */
2150 tracing_reset_online_cpus(&tr->array_buffer);
2151
2152#ifdef CONFIG_TRACER_MAX_TRACE
2153 if (type->use_max_tr) {
2154 tr->allocated_snapshot = false;
2155
2156 /* Shrink the max buffer again */
2157 if (tr->ring_buffer_expanded)
2158 ring_buffer_resize(tr->max_buffer.buffer, 1,
2159 RING_BUFFER_ALL_CPUS);
2160 }
2161#endif
2162
2163 printk(KERN_CONT "PASSED\n");
2164 return 0;
2165}
2166
2167static int do_run_tracer_selftest(struct tracer *type)
2168{
2169 int ret;
2170
2171 /*
2172 * Tests can take a long time, especially if they are run one after the
2173 * other, as does happen during bootup when all the tracers are
2174 * registered. This could cause the soft lockup watchdog to trigger.
2175 */
2176 cond_resched();
2177
2178 tracing_selftest_running = true;
2179 ret = run_tracer_selftest(type);
2180 tracing_selftest_running = false;
2181
2182 return ret;
2183}
2184
2185static __init int init_trace_selftests(void)
2186{
2187 struct trace_selftests *p, *n;
2188 struct tracer *t, **last;
2189 int ret;
2190
2191 selftests_can_run = true;
2192
2193 guard(mutex)(&trace_types_lock);
2194
2195 if (list_empty(&postponed_selftests))
2196 return 0;
2197
2198 pr_info("Running postponed tracer tests:\n");
2199
2200 tracing_selftest_running = true;
2201 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2202 /* This loop can take minutes when sanitizers are enabled, so
2203 * lets make sure we allow RCU processing.
2204 */
2205 cond_resched();
2206 ret = run_tracer_selftest(p->type);
2207 /* If the test fails, then warn and remove from available_tracers */
2208 if (ret < 0) {
2209 WARN(1, "tracer: %s failed selftest, disabling\n",
2210 p->type->name);
2211 last = &trace_types;
2212 for (t = trace_types; t; t = t->next) {
2213 if (t == p->type) {
2214 *last = t->next;
2215 break;
2216 }
2217 last = &t->next;
2218 }
2219 }
2220 list_del(&p->list);
2221 kfree(p);
2222 }
2223 tracing_selftest_running = false;
2224
2225 return 0;
2226}
2227core_initcall(init_trace_selftests);
2228#else
2229static inline int do_run_tracer_selftest(struct tracer *type)
2230{
2231 return 0;
2232}
2233#endif /* CONFIG_FTRACE_STARTUP_TEST */
2234
2235static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2236
2237static void __init apply_trace_boot_options(void);
2238
2239/**
2240 * register_tracer - register a tracer with the ftrace system.
2241 * @type: the plugin for the tracer
2242 *
2243 * Register a new plugin tracer.
2244 */
2245int __init register_tracer(struct tracer *type)
2246{
2247 struct tracer *t;
2248 int ret = 0;
2249
2250 if (!type->name) {
2251 pr_info("Tracer must have a name\n");
2252 return -1;
2253 }
2254
2255 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2256 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2257 return -1;
2258 }
2259
2260 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2261 pr_warn("Can not register tracer %s due to lockdown\n",
2262 type->name);
2263 return -EPERM;
2264 }
2265
2266 mutex_lock(&trace_types_lock);
2267
2268 for (t = trace_types; t; t = t->next) {
2269 if (strcmp(type->name, t->name) == 0) {
2270 /* already found */
2271 pr_info("Tracer %s already registered\n",
2272 type->name);
2273 ret = -1;
2274 goto out;
2275 }
2276 }
2277
2278 if (!type->set_flag)
2279 type->set_flag = &dummy_set_flag;
2280 if (!type->flags) {
2281 /*allocate a dummy tracer_flags*/
2282 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2283 if (!type->flags) {
2284 ret = -ENOMEM;
2285 goto out;
2286 }
2287 type->flags->val = 0;
2288 type->flags->opts = dummy_tracer_opt;
2289 } else
2290 if (!type->flags->opts)
2291 type->flags->opts = dummy_tracer_opt;
2292
2293 /* store the tracer for __set_tracer_option */
2294 type->flags->trace = type;
2295
2296 ret = do_run_tracer_selftest(type);
2297 if (ret < 0)
2298 goto out;
2299
2300 type->next = trace_types;
2301 trace_types = type;
2302 add_tracer_options(&global_trace, type);
2303
2304 out:
2305 mutex_unlock(&trace_types_lock);
2306
2307 if (ret || !default_bootup_tracer)
2308 goto out_unlock;
2309
2310 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2311 goto out_unlock;
2312
2313 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2314 /* Do we want this tracer to start on bootup? */
2315 tracing_set_tracer(&global_trace, type->name);
2316 default_bootup_tracer = NULL;
2317
2318 apply_trace_boot_options();
2319
2320 /* disable other selftests, since this will break it. */
2321 disable_tracing_selftest("running a tracer");
2322
2323 out_unlock:
2324 return ret;
2325}
2326
2327static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2328{
2329 struct trace_buffer *buffer = buf->buffer;
2330
2331 if (!buffer)
2332 return;
2333
2334 ring_buffer_record_disable(buffer);
2335
2336 /* Make sure all commits have finished */
2337 synchronize_rcu();
2338 ring_buffer_reset_cpu(buffer, cpu);
2339
2340 ring_buffer_record_enable(buffer);
2341}
2342
2343void tracing_reset_online_cpus(struct array_buffer *buf)
2344{
2345 struct trace_buffer *buffer = buf->buffer;
2346
2347 if (!buffer)
2348 return;
2349
2350 ring_buffer_record_disable(buffer);
2351
2352 /* Make sure all commits have finished */
2353 synchronize_rcu();
2354
2355 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2356
2357 ring_buffer_reset_online_cpus(buffer);
2358
2359 ring_buffer_record_enable(buffer);
2360}
2361
2362static void tracing_reset_all_cpus(struct array_buffer *buf)
2363{
2364 struct trace_buffer *buffer = buf->buffer;
2365
2366 if (!buffer)
2367 return;
2368
2369 ring_buffer_record_disable(buffer);
2370
2371 /* Make sure all commits have finished */
2372 synchronize_rcu();
2373
2374 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2375
2376 ring_buffer_reset(buffer);
2377
2378 ring_buffer_record_enable(buffer);
2379}
2380
2381/* Must have trace_types_lock held */
2382void tracing_reset_all_online_cpus_unlocked(void)
2383{
2384 struct trace_array *tr;
2385
2386 lockdep_assert_held(&trace_types_lock);
2387
2388 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2389 if (!tr->clear_trace)
2390 continue;
2391 tr->clear_trace = false;
2392 tracing_reset_online_cpus(&tr->array_buffer);
2393#ifdef CONFIG_TRACER_MAX_TRACE
2394 tracing_reset_online_cpus(&tr->max_buffer);
2395#endif
2396 }
2397}
2398
2399void tracing_reset_all_online_cpus(void)
2400{
2401 mutex_lock(&trace_types_lock);
2402 tracing_reset_all_online_cpus_unlocked();
2403 mutex_unlock(&trace_types_lock);
2404}
2405
2406int is_tracing_stopped(void)
2407{
2408 return global_trace.stop_count;
2409}
2410
2411static void tracing_start_tr(struct trace_array *tr)
2412{
2413 struct trace_buffer *buffer;
2414 unsigned long flags;
2415
2416 if (tracing_disabled)
2417 return;
2418
2419 raw_spin_lock_irqsave(&tr->start_lock, flags);
2420 if (--tr->stop_count) {
2421 if (WARN_ON_ONCE(tr->stop_count < 0)) {
2422 /* Someone screwed up their debugging */
2423 tr->stop_count = 0;
2424 }
2425 goto out;
2426 }
2427
2428 /* Prevent the buffers from switching */
2429 arch_spin_lock(&tr->max_lock);
2430
2431 buffer = tr->array_buffer.buffer;
2432 if (buffer)
2433 ring_buffer_record_enable(buffer);
2434
2435#ifdef CONFIG_TRACER_MAX_TRACE
2436 buffer = tr->max_buffer.buffer;
2437 if (buffer)
2438 ring_buffer_record_enable(buffer);
2439#endif
2440
2441 arch_spin_unlock(&tr->max_lock);
2442
2443 out:
2444 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2445}
2446
2447/**
2448 * tracing_start - quick start of the tracer
2449 *
2450 * If tracing is enabled but was stopped by tracing_stop,
2451 * this will start the tracer back up.
2452 */
2453void tracing_start(void)
2454
2455{
2456 return tracing_start_tr(&global_trace);
2457}
2458
2459static void tracing_stop_tr(struct trace_array *tr)
2460{
2461 struct trace_buffer *buffer;
2462 unsigned long flags;
2463
2464 raw_spin_lock_irqsave(&tr->start_lock, flags);
2465 if (tr->stop_count++)
2466 goto out;
2467
2468 /* Prevent the buffers from switching */
2469 arch_spin_lock(&tr->max_lock);
2470
2471 buffer = tr->array_buffer.buffer;
2472 if (buffer)
2473 ring_buffer_record_disable(buffer);
2474
2475#ifdef CONFIG_TRACER_MAX_TRACE
2476 buffer = tr->max_buffer.buffer;
2477 if (buffer)
2478 ring_buffer_record_disable(buffer);
2479#endif
2480
2481 arch_spin_unlock(&tr->max_lock);
2482
2483 out:
2484 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2485}
2486
2487/**
2488 * tracing_stop - quick stop of the tracer
2489 *
2490 * Light weight way to stop tracing. Use in conjunction with
2491 * tracing_start.
2492 */
2493void tracing_stop(void)
2494{
2495 return tracing_stop_tr(&global_trace);
2496}
2497
2498/*
2499 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2500 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2501 * simplifies those functions and keeps them in sync.
2502 */
2503enum print_line_t trace_handle_return(struct trace_seq *s)
2504{
2505 return trace_seq_has_overflowed(s) ?
2506 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2507}
2508EXPORT_SYMBOL_GPL(trace_handle_return);
2509
2510static unsigned short migration_disable_value(void)
2511{
2512#if defined(CONFIG_SMP)
2513 return current->migration_disabled;
2514#else
2515 return 0;
2516#endif
2517}
2518
2519unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2520{
2521 unsigned int trace_flags = irqs_status;
2522 unsigned int pc;
2523
2524 pc = preempt_count();
2525
2526 if (pc & NMI_MASK)
2527 trace_flags |= TRACE_FLAG_NMI;
2528 if (pc & HARDIRQ_MASK)
2529 trace_flags |= TRACE_FLAG_HARDIRQ;
2530 if (in_serving_softirq())
2531 trace_flags |= TRACE_FLAG_SOFTIRQ;
2532 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2533 trace_flags |= TRACE_FLAG_BH_OFF;
2534
2535 if (tif_need_resched())
2536 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2537 if (test_preempt_need_resched())
2538 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2539 if (IS_ENABLED(CONFIG_ARCH_HAS_PREEMPT_LAZY) && tif_test_bit(TIF_NEED_RESCHED_LAZY))
2540 trace_flags |= TRACE_FLAG_NEED_RESCHED_LAZY;
2541 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2542 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2543}
2544
2545struct ring_buffer_event *
2546trace_buffer_lock_reserve(struct trace_buffer *buffer,
2547 int type,
2548 unsigned long len,
2549 unsigned int trace_ctx)
2550{
2551 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2552}
2553
2554DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2555DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2556static int trace_buffered_event_ref;
2557
2558/**
2559 * trace_buffered_event_enable - enable buffering events
2560 *
2561 * When events are being filtered, it is quicker to use a temporary
2562 * buffer to write the event data into if there's a likely chance
2563 * that it will not be committed. The discard of the ring buffer
2564 * is not as fast as committing, and is much slower than copying
2565 * a commit.
2566 *
2567 * When an event is to be filtered, allocate per cpu buffers to
2568 * write the event data into, and if the event is filtered and discarded
2569 * it is simply dropped, otherwise, the entire data is to be committed
2570 * in one shot.
2571 */
2572void trace_buffered_event_enable(void)
2573{
2574 struct ring_buffer_event *event;
2575 struct page *page;
2576 int cpu;
2577
2578 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2579
2580 if (trace_buffered_event_ref++)
2581 return;
2582
2583 for_each_tracing_cpu(cpu) {
2584 page = alloc_pages_node(cpu_to_node(cpu),
2585 GFP_KERNEL | __GFP_NORETRY, 0);
2586 /* This is just an optimization and can handle failures */
2587 if (!page) {
2588 pr_err("Failed to allocate event buffer\n");
2589 break;
2590 }
2591
2592 event = page_address(page);
2593 memset(event, 0, sizeof(*event));
2594
2595 per_cpu(trace_buffered_event, cpu) = event;
2596
2597 preempt_disable();
2598 if (cpu == smp_processor_id() &&
2599 __this_cpu_read(trace_buffered_event) !=
2600 per_cpu(trace_buffered_event, cpu))
2601 WARN_ON_ONCE(1);
2602 preempt_enable();
2603 }
2604}
2605
2606static void enable_trace_buffered_event(void *data)
2607{
2608 /* Probably not needed, but do it anyway */
2609 smp_rmb();
2610 this_cpu_dec(trace_buffered_event_cnt);
2611}
2612
2613static void disable_trace_buffered_event(void *data)
2614{
2615 this_cpu_inc(trace_buffered_event_cnt);
2616}
2617
2618/**
2619 * trace_buffered_event_disable - disable buffering events
2620 *
2621 * When a filter is removed, it is faster to not use the buffered
2622 * events, and to commit directly into the ring buffer. Free up
2623 * the temp buffers when there are no more users. This requires
2624 * special synchronization with current events.
2625 */
2626void trace_buffered_event_disable(void)
2627{
2628 int cpu;
2629
2630 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2631
2632 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2633 return;
2634
2635 if (--trace_buffered_event_ref)
2636 return;
2637
2638 /* For each CPU, set the buffer as used. */
2639 on_each_cpu_mask(tracing_buffer_mask, disable_trace_buffered_event,
2640 NULL, true);
2641
2642 /* Wait for all current users to finish */
2643 synchronize_rcu();
2644
2645 for_each_tracing_cpu(cpu) {
2646 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2647 per_cpu(trace_buffered_event, cpu) = NULL;
2648 }
2649
2650 /*
2651 * Wait for all CPUs that potentially started checking if they can use
2652 * their event buffer only after the previous synchronize_rcu() call and
2653 * they still read a valid pointer from trace_buffered_event. It must be
2654 * ensured they don't see cleared trace_buffered_event_cnt else they
2655 * could wrongly decide to use the pointed-to buffer which is now freed.
2656 */
2657 synchronize_rcu();
2658
2659 /* For each CPU, relinquish the buffer */
2660 on_each_cpu_mask(tracing_buffer_mask, enable_trace_buffered_event, NULL,
2661 true);
2662}
2663
2664static struct trace_buffer *temp_buffer;
2665
2666struct ring_buffer_event *
2667trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2668 struct trace_event_file *trace_file,
2669 int type, unsigned long len,
2670 unsigned int trace_ctx)
2671{
2672 struct ring_buffer_event *entry;
2673 struct trace_array *tr = trace_file->tr;
2674 int val;
2675
2676 *current_rb = tr->array_buffer.buffer;
2677
2678 if (!tr->no_filter_buffering_ref &&
2679 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2680 preempt_disable_notrace();
2681 /*
2682 * Filtering is on, so try to use the per cpu buffer first.
2683 * This buffer will simulate a ring_buffer_event,
2684 * where the type_len is zero and the array[0] will
2685 * hold the full length.
2686 * (see include/linux/ring-buffer.h for details on
2687 * how the ring_buffer_event is structured).
2688 *
2689 * Using a temp buffer during filtering and copying it
2690 * on a matched filter is quicker than writing directly
2691 * into the ring buffer and then discarding it when
2692 * it doesn't match. That is because the discard
2693 * requires several atomic operations to get right.
2694 * Copying on match and doing nothing on a failed match
2695 * is still quicker than no copy on match, but having
2696 * to discard out of the ring buffer on a failed match.
2697 */
2698 if ((entry = __this_cpu_read(trace_buffered_event))) {
2699 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2700
2701 val = this_cpu_inc_return(trace_buffered_event_cnt);
2702
2703 /*
2704 * Preemption is disabled, but interrupts and NMIs
2705 * can still come in now. If that happens after
2706 * the above increment, then it will have to go
2707 * back to the old method of allocating the event
2708 * on the ring buffer, and if the filter fails, it
2709 * will have to call ring_buffer_discard_commit()
2710 * to remove it.
2711 *
2712 * Need to also check the unlikely case that the
2713 * length is bigger than the temp buffer size.
2714 * If that happens, then the reserve is pretty much
2715 * guaranteed to fail, as the ring buffer currently
2716 * only allows events less than a page. But that may
2717 * change in the future, so let the ring buffer reserve
2718 * handle the failure in that case.
2719 */
2720 if (val == 1 && likely(len <= max_len)) {
2721 trace_event_setup(entry, type, trace_ctx);
2722 entry->array[0] = len;
2723 /* Return with preemption disabled */
2724 return entry;
2725 }
2726 this_cpu_dec(trace_buffered_event_cnt);
2727 }
2728 /* __trace_buffer_lock_reserve() disables preemption */
2729 preempt_enable_notrace();
2730 }
2731
2732 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2733 trace_ctx);
2734 /*
2735 * If tracing is off, but we have triggers enabled
2736 * we still need to look at the event data. Use the temp_buffer
2737 * to store the trace event for the trigger to use. It's recursive
2738 * safe and will not be recorded anywhere.
2739 */
2740 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2741 *current_rb = temp_buffer;
2742 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2743 trace_ctx);
2744 }
2745 return entry;
2746}
2747EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2748
2749static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2750static DEFINE_MUTEX(tracepoint_printk_mutex);
2751
2752static void output_printk(struct trace_event_buffer *fbuffer)
2753{
2754 struct trace_event_call *event_call;
2755 struct trace_event_file *file;
2756 struct trace_event *event;
2757 unsigned long flags;
2758 struct trace_iterator *iter = tracepoint_print_iter;
2759
2760 /* We should never get here if iter is NULL */
2761 if (WARN_ON_ONCE(!iter))
2762 return;
2763
2764 event_call = fbuffer->trace_file->event_call;
2765 if (!event_call || !event_call->event.funcs ||
2766 !event_call->event.funcs->trace)
2767 return;
2768
2769 file = fbuffer->trace_file;
2770 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2771 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2772 !filter_match_preds(file->filter, fbuffer->entry)))
2773 return;
2774
2775 event = &fbuffer->trace_file->event_call->event;
2776
2777 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2778 trace_seq_init(&iter->seq);
2779 iter->ent = fbuffer->entry;
2780 event_call->event.funcs->trace(iter, 0, event);
2781 trace_seq_putc(&iter->seq, 0);
2782 printk("%s", iter->seq.buffer);
2783
2784 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2785}
2786
2787int tracepoint_printk_sysctl(const struct ctl_table *table, int write,
2788 void *buffer, size_t *lenp,
2789 loff_t *ppos)
2790{
2791 int save_tracepoint_printk;
2792 int ret;
2793
2794 guard(mutex)(&tracepoint_printk_mutex);
2795 save_tracepoint_printk = tracepoint_printk;
2796
2797 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2798
2799 /*
2800 * This will force exiting early, as tracepoint_printk
2801 * is always zero when tracepoint_printk_iter is not allocated
2802 */
2803 if (!tracepoint_print_iter)
2804 tracepoint_printk = 0;
2805
2806 if (save_tracepoint_printk == tracepoint_printk)
2807 return ret;
2808
2809 if (tracepoint_printk)
2810 static_key_enable(&tracepoint_printk_key.key);
2811 else
2812 static_key_disable(&tracepoint_printk_key.key);
2813
2814 return ret;
2815}
2816
2817void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2818{
2819 enum event_trigger_type tt = ETT_NONE;
2820 struct trace_event_file *file = fbuffer->trace_file;
2821
2822 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2823 fbuffer->entry, &tt))
2824 goto discard;
2825
2826 if (static_key_false(&tracepoint_printk_key.key))
2827 output_printk(fbuffer);
2828
2829 if (static_branch_unlikely(&trace_event_exports_enabled))
2830 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2831
2832 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2833 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
2834
2835discard:
2836 if (tt)
2837 event_triggers_post_call(file, tt);
2838
2839}
2840EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2841
2842/*
2843 * Skip 3:
2844 *
2845 * trace_buffer_unlock_commit_regs()
2846 * trace_event_buffer_commit()
2847 * trace_event_raw_event_xxx()
2848 */
2849# define STACK_SKIP 3
2850
2851void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2852 struct trace_buffer *buffer,
2853 struct ring_buffer_event *event,
2854 unsigned int trace_ctx,
2855 struct pt_regs *regs)
2856{
2857 __buffer_unlock_commit(buffer, event);
2858
2859 /*
2860 * If regs is not set, then skip the necessary functions.
2861 * Note, we can still get here via blktrace, wakeup tracer
2862 * and mmiotrace, but that's ok if they lose a function or
2863 * two. They are not that meaningful.
2864 */
2865 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
2866 ftrace_trace_userstack(tr, buffer, trace_ctx);
2867}
2868
2869/*
2870 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2871 */
2872void
2873trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2874 struct ring_buffer_event *event)
2875{
2876 __buffer_unlock_commit(buffer, event);
2877}
2878
2879void
2880trace_function(struct trace_array *tr, unsigned long ip, unsigned long
2881 parent_ip, unsigned int trace_ctx)
2882{
2883 struct trace_buffer *buffer = tr->array_buffer.buffer;
2884 struct ring_buffer_event *event;
2885 struct ftrace_entry *entry;
2886
2887 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2888 trace_ctx);
2889 if (!event)
2890 return;
2891 entry = ring_buffer_event_data(event);
2892 entry->ip = ip;
2893 entry->parent_ip = parent_ip;
2894
2895 if (static_branch_unlikely(&trace_function_exports_enabled))
2896 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
2897 __buffer_unlock_commit(buffer, event);
2898}
2899
2900#ifdef CONFIG_STACKTRACE
2901
2902/* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2903#define FTRACE_KSTACK_NESTING 4
2904
2905#define FTRACE_KSTACK_ENTRIES (SZ_4K / FTRACE_KSTACK_NESTING)
2906
2907struct ftrace_stack {
2908 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2909};
2910
2911
2912struct ftrace_stacks {
2913 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
2914};
2915
2916static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
2917static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2918
2919static void __ftrace_trace_stack(struct trace_array *tr,
2920 struct trace_buffer *buffer,
2921 unsigned int trace_ctx,
2922 int skip, struct pt_regs *regs)
2923{
2924 struct ring_buffer_event *event;
2925 unsigned int size, nr_entries;
2926 struct ftrace_stack *fstack;
2927 struct stack_entry *entry;
2928 int stackidx;
2929
2930 /*
2931 * Add one, for this function and the call to save_stack_trace()
2932 * If regs is set, then these functions will not be in the way.
2933 */
2934#ifndef CONFIG_UNWINDER_ORC
2935 if (!regs)
2936 skip++;
2937#endif
2938
2939 preempt_disable_notrace();
2940
2941 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
2942
2943 /* This should never happen. If it does, yell once and skip */
2944 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
2945 goto out;
2946
2947 /*
2948 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
2949 * interrupt will either see the value pre increment or post
2950 * increment. If the interrupt happens pre increment it will have
2951 * restored the counter when it returns. We just need a barrier to
2952 * keep gcc from moving things around.
2953 */
2954 barrier();
2955
2956 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
2957 size = ARRAY_SIZE(fstack->calls);
2958
2959 if (regs) {
2960 nr_entries = stack_trace_save_regs(regs, fstack->calls,
2961 size, skip);
2962 } else {
2963 nr_entries = stack_trace_save(fstack->calls, size, skip);
2964 }
2965
2966#ifdef CONFIG_DYNAMIC_FTRACE
2967 /* Mark entry of stack trace as trampoline code */
2968 if (tr->ops && tr->ops->trampoline) {
2969 unsigned long tramp_start = tr->ops->trampoline;
2970 unsigned long tramp_end = tramp_start + tr->ops->trampoline_size;
2971 unsigned long *calls = fstack->calls;
2972
2973 for (int i = 0; i < nr_entries; i++) {
2974 if (calls[i] >= tramp_start && calls[i] < tramp_end)
2975 calls[i] = FTRACE_TRAMPOLINE_MARKER;
2976 }
2977 }
2978#endif
2979
2980 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2981 struct_size(entry, caller, nr_entries),
2982 trace_ctx);
2983 if (!event)
2984 goto out;
2985 entry = ring_buffer_event_data(event);
2986
2987 entry->size = nr_entries;
2988 memcpy(&entry->caller, fstack->calls,
2989 flex_array_size(entry, caller, nr_entries));
2990
2991 __buffer_unlock_commit(buffer, event);
2992
2993 out:
2994 /* Again, don't let gcc optimize things here */
2995 barrier();
2996 __this_cpu_dec(ftrace_stack_reserve);
2997 preempt_enable_notrace();
2998
2999}
3000
3001static inline void ftrace_trace_stack(struct trace_array *tr,
3002 struct trace_buffer *buffer,
3003 unsigned int trace_ctx,
3004 int skip, struct pt_regs *regs)
3005{
3006 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3007 return;
3008
3009 __ftrace_trace_stack(tr, buffer, trace_ctx, skip, regs);
3010}
3011
3012void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3013 int skip)
3014{
3015 struct trace_buffer *buffer = tr->array_buffer.buffer;
3016
3017 if (rcu_is_watching()) {
3018 __ftrace_trace_stack(tr, buffer, trace_ctx, skip, NULL);
3019 return;
3020 }
3021
3022 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3023 return;
3024
3025 /*
3026 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3027 * but if the above rcu_is_watching() failed, then the NMI
3028 * triggered someplace critical, and ct_irq_enter() should
3029 * not be called from NMI.
3030 */
3031 if (unlikely(in_nmi()))
3032 return;
3033
3034 ct_irq_enter_irqson();
3035 __ftrace_trace_stack(tr, buffer, trace_ctx, skip, NULL);
3036 ct_irq_exit_irqson();
3037}
3038
3039/**
3040 * trace_dump_stack - record a stack back trace in the trace buffer
3041 * @skip: Number of functions to skip (helper handlers)
3042 */
3043void trace_dump_stack(int skip)
3044{
3045 if (tracing_disabled || tracing_selftest_running)
3046 return;
3047
3048#ifndef CONFIG_UNWINDER_ORC
3049 /* Skip 1 to skip this function. */
3050 skip++;
3051#endif
3052 __ftrace_trace_stack(printk_trace, printk_trace->array_buffer.buffer,
3053 tracing_gen_ctx(), skip, NULL);
3054}
3055EXPORT_SYMBOL_GPL(trace_dump_stack);
3056
3057#ifdef CONFIG_USER_STACKTRACE_SUPPORT
3058static DEFINE_PER_CPU(int, user_stack_count);
3059
3060static void
3061ftrace_trace_userstack(struct trace_array *tr,
3062 struct trace_buffer *buffer, unsigned int trace_ctx)
3063{
3064 struct ring_buffer_event *event;
3065 struct userstack_entry *entry;
3066
3067 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3068 return;
3069
3070 /*
3071 * NMIs can not handle page faults, even with fix ups.
3072 * The save user stack can (and often does) fault.
3073 */
3074 if (unlikely(in_nmi()))
3075 return;
3076
3077 /*
3078 * prevent recursion, since the user stack tracing may
3079 * trigger other kernel events.
3080 */
3081 preempt_disable();
3082 if (__this_cpu_read(user_stack_count))
3083 goto out;
3084
3085 __this_cpu_inc(user_stack_count);
3086
3087 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3088 sizeof(*entry), trace_ctx);
3089 if (!event)
3090 goto out_drop_count;
3091 entry = ring_buffer_event_data(event);
3092
3093 entry->tgid = current->tgid;
3094 memset(&entry->caller, 0, sizeof(entry->caller));
3095
3096 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3097 __buffer_unlock_commit(buffer, event);
3098
3099 out_drop_count:
3100 __this_cpu_dec(user_stack_count);
3101 out:
3102 preempt_enable();
3103}
3104#else /* CONFIG_USER_STACKTRACE_SUPPORT */
3105static void ftrace_trace_userstack(struct trace_array *tr,
3106 struct trace_buffer *buffer,
3107 unsigned int trace_ctx)
3108{
3109}
3110#endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3111
3112#endif /* CONFIG_STACKTRACE */
3113
3114static inline void
3115func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3116 unsigned long long delta)
3117{
3118 entry->bottom_delta_ts = delta & U32_MAX;
3119 entry->top_delta_ts = (delta >> 32);
3120}
3121
3122void trace_last_func_repeats(struct trace_array *tr,
3123 struct trace_func_repeats *last_info,
3124 unsigned int trace_ctx)
3125{
3126 struct trace_buffer *buffer = tr->array_buffer.buffer;
3127 struct func_repeats_entry *entry;
3128 struct ring_buffer_event *event;
3129 u64 delta;
3130
3131 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3132 sizeof(*entry), trace_ctx);
3133 if (!event)
3134 return;
3135
3136 delta = ring_buffer_event_time_stamp(buffer, event) -
3137 last_info->ts_last_call;
3138
3139 entry = ring_buffer_event_data(event);
3140 entry->ip = last_info->ip;
3141 entry->parent_ip = last_info->parent_ip;
3142 entry->count = last_info->count;
3143 func_repeats_set_delta_ts(entry, delta);
3144
3145 __buffer_unlock_commit(buffer, event);
3146}
3147
3148/* created for use with alloc_percpu */
3149struct trace_buffer_struct {
3150 int nesting;
3151 char buffer[4][TRACE_BUF_SIZE];
3152};
3153
3154static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3155
3156/*
3157 * This allows for lockless recording. If we're nested too deeply, then
3158 * this returns NULL.
3159 */
3160static char *get_trace_buf(void)
3161{
3162 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3163
3164 if (!trace_percpu_buffer || buffer->nesting >= 4)
3165 return NULL;
3166
3167 buffer->nesting++;
3168
3169 /* Interrupts must see nesting incremented before we use the buffer */
3170 barrier();
3171 return &buffer->buffer[buffer->nesting - 1][0];
3172}
3173
3174static void put_trace_buf(void)
3175{
3176 /* Don't let the decrement of nesting leak before this */
3177 barrier();
3178 this_cpu_dec(trace_percpu_buffer->nesting);
3179}
3180
3181static int alloc_percpu_trace_buffer(void)
3182{
3183 struct trace_buffer_struct __percpu *buffers;
3184
3185 if (trace_percpu_buffer)
3186 return 0;
3187
3188 buffers = alloc_percpu(struct trace_buffer_struct);
3189 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3190 return -ENOMEM;
3191
3192 trace_percpu_buffer = buffers;
3193 return 0;
3194}
3195
3196static int buffers_allocated;
3197
3198void trace_printk_init_buffers(void)
3199{
3200 if (buffers_allocated)
3201 return;
3202
3203 if (alloc_percpu_trace_buffer())
3204 return;
3205
3206 /* trace_printk() is for debug use only. Don't use it in production. */
3207
3208 pr_warn("\n");
3209 pr_warn("**********************************************************\n");
3210 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3211 pr_warn("** **\n");
3212 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3213 pr_warn("** **\n");
3214 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3215 pr_warn("** unsafe for production use. **\n");
3216 pr_warn("** **\n");
3217 pr_warn("** If you see this message and you are not debugging **\n");
3218 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3219 pr_warn("** **\n");
3220 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3221 pr_warn("**********************************************************\n");
3222
3223 /* Expand the buffers to set size */
3224 tracing_update_buffers(&global_trace);
3225
3226 buffers_allocated = 1;
3227
3228 /*
3229 * trace_printk_init_buffers() can be called by modules.
3230 * If that happens, then we need to start cmdline recording
3231 * directly here. If the global_trace.buffer is already
3232 * allocated here, then this was called by module code.
3233 */
3234 if (global_trace.array_buffer.buffer)
3235 tracing_start_cmdline_record();
3236}
3237EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3238
3239void trace_printk_start_comm(void)
3240{
3241 /* Start tracing comms if trace printk is set */
3242 if (!buffers_allocated)
3243 return;
3244 tracing_start_cmdline_record();
3245}
3246
3247static void trace_printk_start_stop_comm(int enabled)
3248{
3249 if (!buffers_allocated)
3250 return;
3251
3252 if (enabled)
3253 tracing_start_cmdline_record();
3254 else
3255 tracing_stop_cmdline_record();
3256}
3257
3258/**
3259 * trace_vbprintk - write binary msg to tracing buffer
3260 * @ip: The address of the caller
3261 * @fmt: The string format to write to the buffer
3262 * @args: Arguments for @fmt
3263 */
3264int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3265{
3266 struct ring_buffer_event *event;
3267 struct trace_buffer *buffer;
3268 struct trace_array *tr = READ_ONCE(printk_trace);
3269 struct bprint_entry *entry;
3270 unsigned int trace_ctx;
3271 char *tbuffer;
3272 int len = 0, size;
3273
3274 if (!printk_binsafe(tr))
3275 return trace_vprintk(ip, fmt, args);
3276
3277 if (unlikely(tracing_selftest_running || tracing_disabled))
3278 return 0;
3279
3280 /* Don't pollute graph traces with trace_vprintk internals */
3281 pause_graph_tracing();
3282
3283 trace_ctx = tracing_gen_ctx();
3284 preempt_disable_notrace();
3285
3286 tbuffer = get_trace_buf();
3287 if (!tbuffer) {
3288 len = 0;
3289 goto out_nobuffer;
3290 }
3291
3292 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3293
3294 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3295 goto out_put;
3296
3297 size = sizeof(*entry) + sizeof(u32) * len;
3298 buffer = tr->array_buffer.buffer;
3299 ring_buffer_nest_start(buffer);
3300 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3301 trace_ctx);
3302 if (!event)
3303 goto out;
3304 entry = ring_buffer_event_data(event);
3305 entry->ip = ip;
3306 entry->fmt = fmt;
3307
3308 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3309 __buffer_unlock_commit(buffer, event);
3310 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3311
3312out:
3313 ring_buffer_nest_end(buffer);
3314out_put:
3315 put_trace_buf();
3316
3317out_nobuffer:
3318 preempt_enable_notrace();
3319 unpause_graph_tracing();
3320
3321 return len;
3322}
3323EXPORT_SYMBOL_GPL(trace_vbprintk);
3324
3325__printf(3, 0)
3326static int
3327__trace_array_vprintk(struct trace_buffer *buffer,
3328 unsigned long ip, const char *fmt, va_list args)
3329{
3330 struct ring_buffer_event *event;
3331 int len = 0, size;
3332 struct print_entry *entry;
3333 unsigned int trace_ctx;
3334 char *tbuffer;
3335
3336 if (tracing_disabled)
3337 return 0;
3338
3339 /* Don't pollute graph traces with trace_vprintk internals */
3340 pause_graph_tracing();
3341
3342 trace_ctx = tracing_gen_ctx();
3343 preempt_disable_notrace();
3344
3345
3346 tbuffer = get_trace_buf();
3347 if (!tbuffer) {
3348 len = 0;
3349 goto out_nobuffer;
3350 }
3351
3352 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3353
3354 size = sizeof(*entry) + len + 1;
3355 ring_buffer_nest_start(buffer);
3356 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3357 trace_ctx);
3358 if (!event)
3359 goto out;
3360 entry = ring_buffer_event_data(event);
3361 entry->ip = ip;
3362
3363 memcpy(&entry->buf, tbuffer, len + 1);
3364 __buffer_unlock_commit(buffer, event);
3365 ftrace_trace_stack(printk_trace, buffer, trace_ctx, 6, NULL);
3366
3367out:
3368 ring_buffer_nest_end(buffer);
3369 put_trace_buf();
3370
3371out_nobuffer:
3372 preempt_enable_notrace();
3373 unpause_graph_tracing();
3374
3375 return len;
3376}
3377
3378__printf(3, 0)
3379int trace_array_vprintk(struct trace_array *tr,
3380 unsigned long ip, const char *fmt, va_list args)
3381{
3382 if (tracing_selftest_running && tr == &global_trace)
3383 return 0;
3384
3385 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3386}
3387
3388/**
3389 * trace_array_printk - Print a message to a specific instance
3390 * @tr: The instance trace_array descriptor
3391 * @ip: The instruction pointer that this is called from.
3392 * @fmt: The format to print (printf format)
3393 *
3394 * If a subsystem sets up its own instance, they have the right to
3395 * printk strings into their tracing instance buffer using this
3396 * function. Note, this function will not write into the top level
3397 * buffer (use trace_printk() for that), as writing into the top level
3398 * buffer should only have events that can be individually disabled.
3399 * trace_printk() is only used for debugging a kernel, and should not
3400 * be ever incorporated in normal use.
3401 *
3402 * trace_array_printk() can be used, as it will not add noise to the
3403 * top level tracing buffer.
3404 *
3405 * Note, trace_array_init_printk() must be called on @tr before this
3406 * can be used.
3407 */
3408__printf(3, 0)
3409int trace_array_printk(struct trace_array *tr,
3410 unsigned long ip, const char *fmt, ...)
3411{
3412 int ret;
3413 va_list ap;
3414
3415 if (!tr)
3416 return -ENOENT;
3417
3418 /* This is only allowed for created instances */
3419 if (tr == &global_trace)
3420 return 0;
3421
3422 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3423 return 0;
3424
3425 va_start(ap, fmt);
3426 ret = trace_array_vprintk(tr, ip, fmt, ap);
3427 va_end(ap);
3428 return ret;
3429}
3430EXPORT_SYMBOL_GPL(trace_array_printk);
3431
3432/**
3433 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3434 * @tr: The trace array to initialize the buffers for
3435 *
3436 * As trace_array_printk() only writes into instances, they are OK to
3437 * have in the kernel (unlike trace_printk()). This needs to be called
3438 * before trace_array_printk() can be used on a trace_array.
3439 */
3440int trace_array_init_printk(struct trace_array *tr)
3441{
3442 if (!tr)
3443 return -ENOENT;
3444
3445 /* This is only allowed for created instances */
3446 if (tr == &global_trace)
3447 return -EINVAL;
3448
3449 return alloc_percpu_trace_buffer();
3450}
3451EXPORT_SYMBOL_GPL(trace_array_init_printk);
3452
3453__printf(3, 4)
3454int trace_array_printk_buf(struct trace_buffer *buffer,
3455 unsigned long ip, const char *fmt, ...)
3456{
3457 int ret;
3458 va_list ap;
3459
3460 if (!(printk_trace->trace_flags & TRACE_ITER_PRINTK))
3461 return 0;
3462
3463 va_start(ap, fmt);
3464 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3465 va_end(ap);
3466 return ret;
3467}
3468
3469__printf(2, 0)
3470int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3471{
3472 return trace_array_vprintk(printk_trace, ip, fmt, args);
3473}
3474EXPORT_SYMBOL_GPL(trace_vprintk);
3475
3476static void trace_iterator_increment(struct trace_iterator *iter)
3477{
3478 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3479
3480 iter->idx++;
3481 if (buf_iter)
3482 ring_buffer_iter_advance(buf_iter);
3483}
3484
3485static struct trace_entry *
3486peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3487 unsigned long *lost_events)
3488{
3489 struct ring_buffer_event *event;
3490 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3491
3492 if (buf_iter) {
3493 event = ring_buffer_iter_peek(buf_iter, ts);
3494 if (lost_events)
3495 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3496 (unsigned long)-1 : 0;
3497 } else {
3498 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3499 lost_events);
3500 }
3501
3502 if (event) {
3503 iter->ent_size = ring_buffer_event_length(event);
3504 return ring_buffer_event_data(event);
3505 }
3506 iter->ent_size = 0;
3507 return NULL;
3508}
3509
3510static struct trace_entry *
3511__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3512 unsigned long *missing_events, u64 *ent_ts)
3513{
3514 struct trace_buffer *buffer = iter->array_buffer->buffer;
3515 struct trace_entry *ent, *next = NULL;
3516 unsigned long lost_events = 0, next_lost = 0;
3517 int cpu_file = iter->cpu_file;
3518 u64 next_ts = 0, ts;
3519 int next_cpu = -1;
3520 int next_size = 0;
3521 int cpu;
3522
3523 /*
3524 * If we are in a per_cpu trace file, don't bother by iterating over
3525 * all cpu and peek directly.
3526 */
3527 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3528 if (ring_buffer_empty_cpu(buffer, cpu_file))
3529 return NULL;
3530 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3531 if (ent_cpu)
3532 *ent_cpu = cpu_file;
3533
3534 return ent;
3535 }
3536
3537 for_each_tracing_cpu(cpu) {
3538
3539 if (ring_buffer_empty_cpu(buffer, cpu))
3540 continue;
3541
3542 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3543
3544 /*
3545 * Pick the entry with the smallest timestamp:
3546 */
3547 if (ent && (!next || ts < next_ts)) {
3548 next = ent;
3549 next_cpu = cpu;
3550 next_ts = ts;
3551 next_lost = lost_events;
3552 next_size = iter->ent_size;
3553 }
3554 }
3555
3556 iter->ent_size = next_size;
3557
3558 if (ent_cpu)
3559 *ent_cpu = next_cpu;
3560
3561 if (ent_ts)
3562 *ent_ts = next_ts;
3563
3564 if (missing_events)
3565 *missing_events = next_lost;
3566
3567 return next;
3568}
3569
3570#define STATIC_FMT_BUF_SIZE 128
3571static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3572
3573char *trace_iter_expand_format(struct trace_iterator *iter)
3574{
3575 char *tmp;
3576
3577 /*
3578 * iter->tr is NULL when used with tp_printk, which makes
3579 * this get called where it is not safe to call krealloc().
3580 */
3581 if (!iter->tr || iter->fmt == static_fmt_buf)
3582 return NULL;
3583
3584 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3585 GFP_KERNEL);
3586 if (tmp) {
3587 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3588 iter->fmt = tmp;
3589 }
3590
3591 return tmp;
3592}
3593
3594/* Returns true if the string is safe to dereference from an event */
3595static bool trace_safe_str(struct trace_iterator *iter, const char *str)
3596{
3597 unsigned long addr = (unsigned long)str;
3598 struct trace_event *trace_event;
3599 struct trace_event_call *event;
3600
3601 /* OK if part of the event data */
3602 if ((addr >= (unsigned long)iter->ent) &&
3603 (addr < (unsigned long)iter->ent + iter->ent_size))
3604 return true;
3605
3606 /* OK if part of the temp seq buffer */
3607 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3608 (addr < (unsigned long)iter->tmp_seq.buffer + TRACE_SEQ_BUFFER_SIZE))
3609 return true;
3610
3611 /* Core rodata can not be freed */
3612 if (is_kernel_rodata(addr))
3613 return true;
3614
3615 if (trace_is_tracepoint_string(str))
3616 return true;
3617
3618 /*
3619 * Now this could be a module event, referencing core module
3620 * data, which is OK.
3621 */
3622 if (!iter->ent)
3623 return false;
3624
3625 trace_event = ftrace_find_event(iter->ent->type);
3626 if (!trace_event)
3627 return false;
3628
3629 event = container_of(trace_event, struct trace_event_call, event);
3630 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3631 return false;
3632
3633 /* Would rather have rodata, but this will suffice */
3634 if (within_module_core(addr, event->module))
3635 return true;
3636
3637 return false;
3638}
3639
3640/**
3641 * ignore_event - Check dereferenced fields while writing to the seq buffer
3642 * @iter: The iterator that holds the seq buffer and the event being printed
3643 *
3644 * At boot up, test_event_printk() will flag any event that dereferences
3645 * a string with "%s" that does exist in the ring buffer. It may still
3646 * be valid, as the string may point to a static string in the kernel
3647 * rodata that never gets freed. But if the string pointer is pointing
3648 * to something that was allocated, there's a chance that it can be freed
3649 * by the time the user reads the trace. This would cause a bad memory
3650 * access by the kernel and possibly crash the system.
3651 *
3652 * This function will check if the event has any fields flagged as needing
3653 * to be checked at runtime and perform those checks.
3654 *
3655 * If it is found that a field is unsafe, it will write into the @iter->seq
3656 * a message stating what was found to be unsafe.
3657 *
3658 * @return: true if the event is unsafe and should be ignored,
3659 * false otherwise.
3660 */
3661bool ignore_event(struct trace_iterator *iter)
3662{
3663 struct ftrace_event_field *field;
3664 struct trace_event *trace_event;
3665 struct trace_event_call *event;
3666 struct list_head *head;
3667 struct trace_seq *seq;
3668 const void *ptr;
3669
3670 trace_event = ftrace_find_event(iter->ent->type);
3671
3672 seq = &iter->seq;
3673
3674 if (!trace_event) {
3675 trace_seq_printf(seq, "EVENT ID %d NOT FOUND?\n", iter->ent->type);
3676 return true;
3677 }
3678
3679 event = container_of(trace_event, struct trace_event_call, event);
3680 if (!(event->flags & TRACE_EVENT_FL_TEST_STR))
3681 return false;
3682
3683 head = trace_get_fields(event);
3684 if (!head) {
3685 trace_seq_printf(seq, "FIELDS FOR EVENT '%s' NOT FOUND?\n",
3686 trace_event_name(event));
3687 return true;
3688 }
3689
3690 /* Offsets are from the iter->ent that points to the raw event */
3691 ptr = iter->ent;
3692
3693 list_for_each_entry(field, head, link) {
3694 const char *str;
3695 bool good;
3696
3697 if (!field->needs_test)
3698 continue;
3699
3700 str = *(const char **)(ptr + field->offset);
3701
3702 good = trace_safe_str(iter, str);
3703
3704 /*
3705 * If you hit this warning, it is likely that the
3706 * trace event in question used %s on a string that
3707 * was saved at the time of the event, but may not be
3708 * around when the trace is read. Use __string(),
3709 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3710 * instead. See samples/trace_events/trace-events-sample.h
3711 * for reference.
3712 */
3713 if (WARN_ONCE(!good, "event '%s' has unsafe pointer field '%s'",
3714 trace_event_name(event), field->name)) {
3715 trace_seq_printf(seq, "EVENT %s: HAS UNSAFE POINTER FIELD '%s'\n",
3716 trace_event_name(event), field->name);
3717 return true;
3718 }
3719 }
3720 return false;
3721}
3722
3723const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3724{
3725 const char *p, *new_fmt;
3726 char *q;
3727
3728 if (WARN_ON_ONCE(!fmt))
3729 return fmt;
3730
3731 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
3732 return fmt;
3733
3734 p = fmt;
3735 new_fmt = q = iter->fmt;
3736 while (*p) {
3737 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
3738 if (!trace_iter_expand_format(iter))
3739 return fmt;
3740
3741 q += iter->fmt - new_fmt;
3742 new_fmt = iter->fmt;
3743 }
3744
3745 *q++ = *p++;
3746
3747 /* Replace %p with %px */
3748 if (p[-1] == '%') {
3749 if (p[0] == '%') {
3750 *q++ = *p++;
3751 } else if (p[0] == 'p' && !isalnum(p[1])) {
3752 *q++ = *p++;
3753 *q++ = 'x';
3754 }
3755 }
3756 }
3757 *q = '\0';
3758
3759 return new_fmt;
3760}
3761
3762#define STATIC_TEMP_BUF_SIZE 128
3763static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3764
3765/* Find the next real entry, without updating the iterator itself */
3766struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3767 int *ent_cpu, u64 *ent_ts)
3768{
3769 /* __find_next_entry will reset ent_size */
3770 int ent_size = iter->ent_size;
3771 struct trace_entry *entry;
3772
3773 /*
3774 * If called from ftrace_dump(), then the iter->temp buffer
3775 * will be the static_temp_buf and not created from kmalloc.
3776 * If the entry size is greater than the buffer, we can
3777 * not save it. Just return NULL in that case. This is only
3778 * used to add markers when two consecutive events' time
3779 * stamps have a large delta. See trace_print_lat_context()
3780 */
3781 if (iter->temp == static_temp_buf &&
3782 STATIC_TEMP_BUF_SIZE < ent_size)
3783 return NULL;
3784
3785 /*
3786 * The __find_next_entry() may call peek_next_entry(), which may
3787 * call ring_buffer_peek() that may make the contents of iter->ent
3788 * undefined. Need to copy iter->ent now.
3789 */
3790 if (iter->ent && iter->ent != iter->temp) {
3791 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3792 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3793 void *temp;
3794 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3795 if (!temp)
3796 return NULL;
3797 kfree(iter->temp);
3798 iter->temp = temp;
3799 iter->temp_size = iter->ent_size;
3800 }
3801 memcpy(iter->temp, iter->ent, iter->ent_size);
3802 iter->ent = iter->temp;
3803 }
3804 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3805 /* Put back the original ent_size */
3806 iter->ent_size = ent_size;
3807
3808 return entry;
3809}
3810
3811/* Find the next real entry, and increment the iterator to the next entry */
3812void *trace_find_next_entry_inc(struct trace_iterator *iter)
3813{
3814 iter->ent = __find_next_entry(iter, &iter->cpu,
3815 &iter->lost_events, &iter->ts);
3816
3817 if (iter->ent)
3818 trace_iterator_increment(iter);
3819
3820 return iter->ent ? iter : NULL;
3821}
3822
3823static void trace_consume(struct trace_iterator *iter)
3824{
3825 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3826 &iter->lost_events);
3827}
3828
3829static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3830{
3831 struct trace_iterator *iter = m->private;
3832 int i = (int)*pos;
3833 void *ent;
3834
3835 WARN_ON_ONCE(iter->leftover);
3836
3837 (*pos)++;
3838
3839 /* can't go backwards */
3840 if (iter->idx > i)
3841 return NULL;
3842
3843 if (iter->idx < 0)
3844 ent = trace_find_next_entry_inc(iter);
3845 else
3846 ent = iter;
3847
3848 while (ent && iter->idx < i)
3849 ent = trace_find_next_entry_inc(iter);
3850
3851 iter->pos = *pos;
3852
3853 return ent;
3854}
3855
3856void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3857{
3858 struct ring_buffer_iter *buf_iter;
3859 unsigned long entries = 0;
3860 u64 ts;
3861
3862 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
3863
3864 buf_iter = trace_buffer_iter(iter, cpu);
3865 if (!buf_iter)
3866 return;
3867
3868 ring_buffer_iter_reset(buf_iter);
3869
3870 /*
3871 * We could have the case with the max latency tracers
3872 * that a reset never took place on a cpu. This is evident
3873 * by the timestamp being before the start of the buffer.
3874 */
3875 while (ring_buffer_iter_peek(buf_iter, &ts)) {
3876 if (ts >= iter->array_buffer->time_start)
3877 break;
3878 entries++;
3879 ring_buffer_iter_advance(buf_iter);
3880 /* This could be a big loop */
3881 cond_resched();
3882 }
3883
3884 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
3885}
3886
3887/*
3888 * The current tracer is copied to avoid a global locking
3889 * all around.
3890 */
3891static void *s_start(struct seq_file *m, loff_t *pos)
3892{
3893 struct trace_iterator *iter = m->private;
3894 struct trace_array *tr = iter->tr;
3895 int cpu_file = iter->cpu_file;
3896 void *p = NULL;
3897 loff_t l = 0;
3898 int cpu;
3899
3900 mutex_lock(&trace_types_lock);
3901 if (unlikely(tr->current_trace != iter->trace)) {
3902 /* Close iter->trace before switching to the new current tracer */
3903 if (iter->trace->close)
3904 iter->trace->close(iter);
3905 iter->trace = tr->current_trace;
3906 /* Reopen the new current tracer */
3907 if (iter->trace->open)
3908 iter->trace->open(iter);
3909 }
3910 mutex_unlock(&trace_types_lock);
3911
3912#ifdef CONFIG_TRACER_MAX_TRACE
3913 if (iter->snapshot && iter->trace->use_max_tr)
3914 return ERR_PTR(-EBUSY);
3915#endif
3916
3917 if (*pos != iter->pos) {
3918 iter->ent = NULL;
3919 iter->cpu = 0;
3920 iter->idx = -1;
3921
3922 if (cpu_file == RING_BUFFER_ALL_CPUS) {
3923 for_each_tracing_cpu(cpu)
3924 tracing_iter_reset(iter, cpu);
3925 } else
3926 tracing_iter_reset(iter, cpu_file);
3927
3928 iter->leftover = 0;
3929 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
3930 ;
3931
3932 } else {
3933 /*
3934 * If we overflowed the seq_file before, then we want
3935 * to just reuse the trace_seq buffer again.
3936 */
3937 if (iter->leftover)
3938 p = iter;
3939 else {
3940 l = *pos - 1;
3941 p = s_next(m, p, &l);
3942 }
3943 }
3944
3945 trace_event_read_lock();
3946 trace_access_lock(cpu_file);
3947 return p;
3948}
3949
3950static void s_stop(struct seq_file *m, void *p)
3951{
3952 struct trace_iterator *iter = m->private;
3953
3954#ifdef CONFIG_TRACER_MAX_TRACE
3955 if (iter->snapshot && iter->trace->use_max_tr)
3956 return;
3957#endif
3958
3959 trace_access_unlock(iter->cpu_file);
3960 trace_event_read_unlock();
3961}
3962
3963static void
3964get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
3965 unsigned long *entries, int cpu)
3966{
3967 unsigned long count;
3968
3969 count = ring_buffer_entries_cpu(buf->buffer, cpu);
3970 /*
3971 * If this buffer has skipped entries, then we hold all
3972 * entries for the trace and we need to ignore the
3973 * ones before the time stamp.
3974 */
3975 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
3976 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
3977 /* total is the same as the entries */
3978 *total = count;
3979 } else
3980 *total = count +
3981 ring_buffer_overrun_cpu(buf->buffer, cpu);
3982 *entries = count;
3983}
3984
3985static void
3986get_total_entries(struct array_buffer *buf,
3987 unsigned long *total, unsigned long *entries)
3988{
3989 unsigned long t, e;
3990 int cpu;
3991
3992 *total = 0;
3993 *entries = 0;
3994
3995 for_each_tracing_cpu(cpu) {
3996 get_total_entries_cpu(buf, &t, &e, cpu);
3997 *total += t;
3998 *entries += e;
3999 }
4000}
4001
4002unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4003{
4004 unsigned long total, entries;
4005
4006 if (!tr)
4007 tr = &global_trace;
4008
4009 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4010
4011 return entries;
4012}
4013
4014unsigned long trace_total_entries(struct trace_array *tr)
4015{
4016 unsigned long total, entries;
4017
4018 if (!tr)
4019 tr = &global_trace;
4020
4021 get_total_entries(&tr->array_buffer, &total, &entries);
4022
4023 return entries;
4024}
4025
4026static void print_lat_help_header(struct seq_file *m)
4027{
4028 seq_puts(m, "# _------=> CPU# \n"
4029 "# / _-----=> irqs-off/BH-disabled\n"
4030 "# | / _----=> need-resched \n"
4031 "# || / _---=> hardirq/softirq \n"
4032 "# ||| / _--=> preempt-depth \n"
4033 "# |||| / _-=> migrate-disable \n"
4034 "# ||||| / delay \n"
4035 "# cmd pid |||||| time | caller \n"
4036 "# \\ / |||||| \\ | / \n");
4037}
4038
4039static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4040{
4041 unsigned long total;
4042 unsigned long entries;
4043
4044 get_total_entries(buf, &total, &entries);
4045 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4046 entries, total, num_online_cpus());
4047 seq_puts(m, "#\n");
4048}
4049
4050static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4051 unsigned int flags)
4052{
4053 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4054
4055 print_event_info(buf, m);
4056
4057 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4058 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4059}
4060
4061static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4062 unsigned int flags)
4063{
4064 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4065 static const char space[] = " ";
4066 int prec = tgid ? 12 : 2;
4067
4068 print_event_info(buf, m);
4069
4070 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4071 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4072 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4073 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4074 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4075 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4076 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4077 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4078}
4079
4080void
4081print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4082{
4083 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4084 struct array_buffer *buf = iter->array_buffer;
4085 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4086 struct tracer *type = iter->trace;
4087 unsigned long entries;
4088 unsigned long total;
4089 const char *name = type->name;
4090
4091 get_total_entries(buf, &total, &entries);
4092
4093 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4094 name, init_utsname()->release);
4095 seq_puts(m, "# -----------------------------------"
4096 "---------------------------------\n");
4097 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4098 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4099 nsecs_to_usecs(data->saved_latency),
4100 entries,
4101 total,
4102 buf->cpu,
4103 preempt_model_none() ? "server" :
4104 preempt_model_voluntary() ? "desktop" :
4105 preempt_model_full() ? "preempt" :
4106 preempt_model_lazy() ? "lazy" :
4107 preempt_model_rt() ? "preempt_rt" :
4108 "unknown",
4109 /* These are reserved for later use */
4110 0, 0, 0, 0);
4111#ifdef CONFIG_SMP
4112 seq_printf(m, " #P:%d)\n", num_online_cpus());
4113#else
4114 seq_puts(m, ")\n");
4115#endif
4116 seq_puts(m, "# -----------------\n");
4117 seq_printf(m, "# | task: %.16s-%d "
4118 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4119 data->comm, data->pid,
4120 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4121 data->policy, data->rt_priority);
4122 seq_puts(m, "# -----------------\n");
4123
4124 if (data->critical_start) {
4125 seq_puts(m, "# => started at: ");
4126 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4127 trace_print_seq(m, &iter->seq);
4128 seq_puts(m, "\n# => ended at: ");
4129 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4130 trace_print_seq(m, &iter->seq);
4131 seq_puts(m, "\n#\n");
4132 }
4133
4134 seq_puts(m, "#\n");
4135}
4136
4137static void test_cpu_buff_start(struct trace_iterator *iter)
4138{
4139 struct trace_seq *s = &iter->seq;
4140 struct trace_array *tr = iter->tr;
4141
4142 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4143 return;
4144
4145 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4146 return;
4147
4148 if (cpumask_available(iter->started) &&
4149 cpumask_test_cpu(iter->cpu, iter->started))
4150 return;
4151
4152 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4153 return;
4154
4155 if (cpumask_available(iter->started))
4156 cpumask_set_cpu(iter->cpu, iter->started);
4157
4158 /* Don't print started cpu buffer for the first entry of the trace */
4159 if (iter->idx > 1)
4160 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4161 iter->cpu);
4162}
4163
4164static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4165{
4166 struct trace_array *tr = iter->tr;
4167 struct trace_seq *s = &iter->seq;
4168 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4169 struct trace_entry *entry;
4170 struct trace_event *event;
4171
4172 entry = iter->ent;
4173
4174 test_cpu_buff_start(iter);
4175
4176 event = ftrace_find_event(entry->type);
4177
4178 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4179 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4180 trace_print_lat_context(iter);
4181 else
4182 trace_print_context(iter);
4183 }
4184
4185 if (trace_seq_has_overflowed(s))
4186 return TRACE_TYPE_PARTIAL_LINE;
4187
4188 if (event) {
4189 if (tr->trace_flags & TRACE_ITER_FIELDS)
4190 return print_event_fields(iter, event);
4191 /*
4192 * For TRACE_EVENT() events, the print_fmt is not
4193 * safe to use if the array has delta offsets
4194 * Force printing via the fields.
4195 */
4196 if ((tr->text_delta || tr->data_delta) &&
4197 event->type > __TRACE_LAST_TYPE)
4198 return print_event_fields(iter, event);
4199
4200 return event->funcs->trace(iter, sym_flags, event);
4201 }
4202
4203 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4204
4205 return trace_handle_return(s);
4206}
4207
4208static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4209{
4210 struct trace_array *tr = iter->tr;
4211 struct trace_seq *s = &iter->seq;
4212 struct trace_entry *entry;
4213 struct trace_event *event;
4214
4215 entry = iter->ent;
4216
4217 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4218 trace_seq_printf(s, "%d %d %llu ",
4219 entry->pid, iter->cpu, iter->ts);
4220
4221 if (trace_seq_has_overflowed(s))
4222 return TRACE_TYPE_PARTIAL_LINE;
4223
4224 event = ftrace_find_event(entry->type);
4225 if (event)
4226 return event->funcs->raw(iter, 0, event);
4227
4228 trace_seq_printf(s, "%d ?\n", entry->type);
4229
4230 return trace_handle_return(s);
4231}
4232
4233static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4234{
4235 struct trace_array *tr = iter->tr;
4236 struct trace_seq *s = &iter->seq;
4237 unsigned char newline = '\n';
4238 struct trace_entry *entry;
4239 struct trace_event *event;
4240
4241 entry = iter->ent;
4242
4243 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4244 SEQ_PUT_HEX_FIELD(s, entry->pid);
4245 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4246 SEQ_PUT_HEX_FIELD(s, iter->ts);
4247 if (trace_seq_has_overflowed(s))
4248 return TRACE_TYPE_PARTIAL_LINE;
4249 }
4250
4251 event = ftrace_find_event(entry->type);
4252 if (event) {
4253 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4254 if (ret != TRACE_TYPE_HANDLED)
4255 return ret;
4256 }
4257
4258 SEQ_PUT_FIELD(s, newline);
4259
4260 return trace_handle_return(s);
4261}
4262
4263static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4264{
4265 struct trace_array *tr = iter->tr;
4266 struct trace_seq *s = &iter->seq;
4267 struct trace_entry *entry;
4268 struct trace_event *event;
4269
4270 entry = iter->ent;
4271
4272 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4273 SEQ_PUT_FIELD(s, entry->pid);
4274 SEQ_PUT_FIELD(s, iter->cpu);
4275 SEQ_PUT_FIELD(s, iter->ts);
4276 if (trace_seq_has_overflowed(s))
4277 return TRACE_TYPE_PARTIAL_LINE;
4278 }
4279
4280 event = ftrace_find_event(entry->type);
4281 return event ? event->funcs->binary(iter, 0, event) :
4282 TRACE_TYPE_HANDLED;
4283}
4284
4285int trace_empty(struct trace_iterator *iter)
4286{
4287 struct ring_buffer_iter *buf_iter;
4288 int cpu;
4289
4290 /* If we are looking at one CPU buffer, only check that one */
4291 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4292 cpu = iter->cpu_file;
4293 buf_iter = trace_buffer_iter(iter, cpu);
4294 if (buf_iter) {
4295 if (!ring_buffer_iter_empty(buf_iter))
4296 return 0;
4297 } else {
4298 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4299 return 0;
4300 }
4301 return 1;
4302 }
4303
4304 for_each_tracing_cpu(cpu) {
4305 buf_iter = trace_buffer_iter(iter, cpu);
4306 if (buf_iter) {
4307 if (!ring_buffer_iter_empty(buf_iter))
4308 return 0;
4309 } else {
4310 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4311 return 0;
4312 }
4313 }
4314
4315 return 1;
4316}
4317
4318/* Called with trace_event_read_lock() held. */
4319enum print_line_t print_trace_line(struct trace_iterator *iter)
4320{
4321 struct trace_array *tr = iter->tr;
4322 unsigned long trace_flags = tr->trace_flags;
4323 enum print_line_t ret;
4324
4325 if (iter->lost_events) {
4326 if (iter->lost_events == (unsigned long)-1)
4327 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4328 iter->cpu);
4329 else
4330 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4331 iter->cpu, iter->lost_events);
4332 if (trace_seq_has_overflowed(&iter->seq))
4333 return TRACE_TYPE_PARTIAL_LINE;
4334 }
4335
4336 if (iter->trace && iter->trace->print_line) {
4337 ret = iter->trace->print_line(iter);
4338 if (ret != TRACE_TYPE_UNHANDLED)
4339 return ret;
4340 }
4341
4342 if (iter->ent->type == TRACE_BPUTS &&
4343 trace_flags & TRACE_ITER_PRINTK &&
4344 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4345 return trace_print_bputs_msg_only(iter);
4346
4347 if (iter->ent->type == TRACE_BPRINT &&
4348 trace_flags & TRACE_ITER_PRINTK &&
4349 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4350 return trace_print_bprintk_msg_only(iter);
4351
4352 if (iter->ent->type == TRACE_PRINT &&
4353 trace_flags & TRACE_ITER_PRINTK &&
4354 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4355 return trace_print_printk_msg_only(iter);
4356
4357 if (trace_flags & TRACE_ITER_BIN)
4358 return print_bin_fmt(iter);
4359
4360 if (trace_flags & TRACE_ITER_HEX)
4361 return print_hex_fmt(iter);
4362
4363 if (trace_flags & TRACE_ITER_RAW)
4364 return print_raw_fmt(iter);
4365
4366 return print_trace_fmt(iter);
4367}
4368
4369void trace_latency_header(struct seq_file *m)
4370{
4371 struct trace_iterator *iter = m->private;
4372 struct trace_array *tr = iter->tr;
4373
4374 /* print nothing if the buffers are empty */
4375 if (trace_empty(iter))
4376 return;
4377
4378 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4379 print_trace_header(m, iter);
4380
4381 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4382 print_lat_help_header(m);
4383}
4384
4385void trace_default_header(struct seq_file *m)
4386{
4387 struct trace_iterator *iter = m->private;
4388 struct trace_array *tr = iter->tr;
4389 unsigned long trace_flags = tr->trace_flags;
4390
4391 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4392 return;
4393
4394 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4395 /* print nothing if the buffers are empty */
4396 if (trace_empty(iter))
4397 return;
4398 print_trace_header(m, iter);
4399 if (!(trace_flags & TRACE_ITER_VERBOSE))
4400 print_lat_help_header(m);
4401 } else {
4402 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4403 if (trace_flags & TRACE_ITER_IRQ_INFO)
4404 print_func_help_header_irq(iter->array_buffer,
4405 m, trace_flags);
4406 else
4407 print_func_help_header(iter->array_buffer, m,
4408 trace_flags);
4409 }
4410 }
4411}
4412
4413static void test_ftrace_alive(struct seq_file *m)
4414{
4415 if (!ftrace_is_dead())
4416 return;
4417 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4418 "# MAY BE MISSING FUNCTION EVENTS\n");
4419}
4420
4421#ifdef CONFIG_TRACER_MAX_TRACE
4422static void show_snapshot_main_help(struct seq_file *m)
4423{
4424 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4425 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4426 "# Takes a snapshot of the main buffer.\n"
4427 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4428 "# (Doesn't have to be '2' works with any number that\n"
4429 "# is not a '0' or '1')\n");
4430}
4431
4432static void show_snapshot_percpu_help(struct seq_file *m)
4433{
4434 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4435#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4436 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4437 "# Takes a snapshot of the main buffer for this cpu.\n");
4438#else
4439 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4440 "# Must use main snapshot file to allocate.\n");
4441#endif
4442 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4443 "# (Doesn't have to be '2' works with any number that\n"
4444 "# is not a '0' or '1')\n");
4445}
4446
4447static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4448{
4449 if (iter->tr->allocated_snapshot)
4450 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4451 else
4452 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4453
4454 seq_puts(m, "# Snapshot commands:\n");
4455 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4456 show_snapshot_main_help(m);
4457 else
4458 show_snapshot_percpu_help(m);
4459}
4460#else
4461/* Should never be called */
4462static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4463#endif
4464
4465static int s_show(struct seq_file *m, void *v)
4466{
4467 struct trace_iterator *iter = v;
4468 int ret;
4469
4470 if (iter->ent == NULL) {
4471 if (iter->tr) {
4472 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4473 seq_puts(m, "#\n");
4474 test_ftrace_alive(m);
4475 }
4476 if (iter->snapshot && trace_empty(iter))
4477 print_snapshot_help(m, iter);
4478 else if (iter->trace && iter->trace->print_header)
4479 iter->trace->print_header(m);
4480 else
4481 trace_default_header(m);
4482
4483 } else if (iter->leftover) {
4484 /*
4485 * If we filled the seq_file buffer earlier, we
4486 * want to just show it now.
4487 */
4488 ret = trace_print_seq(m, &iter->seq);
4489
4490 /* ret should this time be zero, but you never know */
4491 iter->leftover = ret;
4492
4493 } else {
4494 ret = print_trace_line(iter);
4495 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4496 iter->seq.full = 0;
4497 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
4498 }
4499 ret = trace_print_seq(m, &iter->seq);
4500 /*
4501 * If we overflow the seq_file buffer, then it will
4502 * ask us for this data again at start up.
4503 * Use that instead.
4504 * ret is 0 if seq_file write succeeded.
4505 * -1 otherwise.
4506 */
4507 iter->leftover = ret;
4508 }
4509
4510 return 0;
4511}
4512
4513/*
4514 * Should be used after trace_array_get(), trace_types_lock
4515 * ensures that i_cdev was already initialized.
4516 */
4517static inline int tracing_get_cpu(struct inode *inode)
4518{
4519 if (inode->i_cdev) /* See trace_create_cpu_file() */
4520 return (long)inode->i_cdev - 1;
4521 return RING_BUFFER_ALL_CPUS;
4522}
4523
4524static const struct seq_operations tracer_seq_ops = {
4525 .start = s_start,
4526 .next = s_next,
4527 .stop = s_stop,
4528 .show = s_show,
4529};
4530
4531/*
4532 * Note, as iter itself can be allocated and freed in different
4533 * ways, this function is only used to free its content, and not
4534 * the iterator itself. The only requirement to all the allocations
4535 * is that it must zero all fields (kzalloc), as freeing works with
4536 * ethier allocated content or NULL.
4537 */
4538static void free_trace_iter_content(struct trace_iterator *iter)
4539{
4540 /* The fmt is either NULL, allocated or points to static_fmt_buf */
4541 if (iter->fmt != static_fmt_buf)
4542 kfree(iter->fmt);
4543
4544 kfree(iter->temp);
4545 kfree(iter->buffer_iter);
4546 mutex_destroy(&iter->mutex);
4547 free_cpumask_var(iter->started);
4548}
4549
4550static struct trace_iterator *
4551__tracing_open(struct inode *inode, struct file *file, bool snapshot)
4552{
4553 struct trace_array *tr = inode->i_private;
4554 struct trace_iterator *iter;
4555 int cpu;
4556
4557 if (tracing_disabled)
4558 return ERR_PTR(-ENODEV);
4559
4560 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4561 if (!iter)
4562 return ERR_PTR(-ENOMEM);
4563
4564 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4565 GFP_KERNEL);
4566 if (!iter->buffer_iter)
4567 goto release;
4568
4569 /*
4570 * trace_find_next_entry() may need to save off iter->ent.
4571 * It will place it into the iter->temp buffer. As most
4572 * events are less than 128, allocate a buffer of that size.
4573 * If one is greater, then trace_find_next_entry() will
4574 * allocate a new buffer to adjust for the bigger iter->ent.
4575 * It's not critical if it fails to get allocated here.
4576 */
4577 iter->temp = kmalloc(128, GFP_KERNEL);
4578 if (iter->temp)
4579 iter->temp_size = 128;
4580
4581 /*
4582 * trace_event_printf() may need to modify given format
4583 * string to replace %p with %px so that it shows real address
4584 * instead of hash value. However, that is only for the event
4585 * tracing, other tracer may not need. Defer the allocation
4586 * until it is needed.
4587 */
4588 iter->fmt = NULL;
4589 iter->fmt_size = 0;
4590
4591 mutex_lock(&trace_types_lock);
4592 iter->trace = tr->current_trace;
4593
4594 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4595 goto fail;
4596
4597 iter->tr = tr;
4598
4599#ifdef CONFIG_TRACER_MAX_TRACE
4600 /* Currently only the top directory has a snapshot */
4601 if (tr->current_trace->print_max || snapshot)
4602 iter->array_buffer = &tr->max_buffer;
4603 else
4604#endif
4605 iter->array_buffer = &tr->array_buffer;
4606 iter->snapshot = snapshot;
4607 iter->pos = -1;
4608 iter->cpu_file = tracing_get_cpu(inode);
4609 mutex_init(&iter->mutex);
4610
4611 /* Notify the tracer early; before we stop tracing. */
4612 if (iter->trace->open)
4613 iter->trace->open(iter);
4614
4615 /* Annotate start of buffers if we had overruns */
4616 if (ring_buffer_overruns(iter->array_buffer->buffer))
4617 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4618
4619 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4620 if (trace_clocks[tr->clock_id].in_ns)
4621 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4622
4623 /*
4624 * If pause-on-trace is enabled, then stop the trace while
4625 * dumping, unless this is the "snapshot" file
4626 */
4627 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4628 tracing_stop_tr(tr);
4629
4630 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4631 for_each_tracing_cpu(cpu) {
4632 iter->buffer_iter[cpu] =
4633 ring_buffer_read_prepare(iter->array_buffer->buffer,
4634 cpu, GFP_KERNEL);
4635 }
4636 ring_buffer_read_prepare_sync();
4637 for_each_tracing_cpu(cpu) {
4638 ring_buffer_read_start(iter->buffer_iter[cpu]);
4639 tracing_iter_reset(iter, cpu);
4640 }
4641 } else {
4642 cpu = iter->cpu_file;
4643 iter->buffer_iter[cpu] =
4644 ring_buffer_read_prepare(iter->array_buffer->buffer,
4645 cpu, GFP_KERNEL);
4646 ring_buffer_read_prepare_sync();
4647 ring_buffer_read_start(iter->buffer_iter[cpu]);
4648 tracing_iter_reset(iter, cpu);
4649 }
4650
4651 mutex_unlock(&trace_types_lock);
4652
4653 return iter;
4654
4655 fail:
4656 mutex_unlock(&trace_types_lock);
4657 free_trace_iter_content(iter);
4658release:
4659 seq_release_private(inode, file);
4660 return ERR_PTR(-ENOMEM);
4661}
4662
4663int tracing_open_generic(struct inode *inode, struct file *filp)
4664{
4665 int ret;
4666
4667 ret = tracing_check_open_get_tr(NULL);
4668 if (ret)
4669 return ret;
4670
4671 filp->private_data = inode->i_private;
4672 return 0;
4673}
4674
4675bool tracing_is_disabled(void)
4676{
4677 return (tracing_disabled) ? true: false;
4678}
4679
4680/*
4681 * Open and update trace_array ref count.
4682 * Must have the current trace_array passed to it.
4683 */
4684int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4685{
4686 struct trace_array *tr = inode->i_private;
4687 int ret;
4688
4689 ret = tracing_check_open_get_tr(tr);
4690 if (ret)
4691 return ret;
4692
4693 filp->private_data = inode->i_private;
4694
4695 return 0;
4696}
4697
4698/*
4699 * The private pointer of the inode is the trace_event_file.
4700 * Update the tr ref count associated to it.
4701 */
4702int tracing_open_file_tr(struct inode *inode, struct file *filp)
4703{
4704 struct trace_event_file *file = inode->i_private;
4705 int ret;
4706
4707 ret = tracing_check_open_get_tr(file->tr);
4708 if (ret)
4709 return ret;
4710
4711 mutex_lock(&event_mutex);
4712
4713 /* Fail if the file is marked for removal */
4714 if (file->flags & EVENT_FILE_FL_FREED) {
4715 trace_array_put(file->tr);
4716 ret = -ENODEV;
4717 } else {
4718 event_file_get(file);
4719 }
4720
4721 mutex_unlock(&event_mutex);
4722 if (ret)
4723 return ret;
4724
4725 filp->private_data = inode->i_private;
4726
4727 return 0;
4728}
4729
4730int tracing_release_file_tr(struct inode *inode, struct file *filp)
4731{
4732 struct trace_event_file *file = inode->i_private;
4733
4734 trace_array_put(file->tr);
4735 event_file_put(file);
4736
4737 return 0;
4738}
4739
4740int tracing_single_release_file_tr(struct inode *inode, struct file *filp)
4741{
4742 tracing_release_file_tr(inode, filp);
4743 return single_release(inode, filp);
4744}
4745
4746static int tracing_mark_open(struct inode *inode, struct file *filp)
4747{
4748 stream_open(inode, filp);
4749 return tracing_open_generic_tr(inode, filp);
4750}
4751
4752static int tracing_release(struct inode *inode, struct file *file)
4753{
4754 struct trace_array *tr = inode->i_private;
4755 struct seq_file *m = file->private_data;
4756 struct trace_iterator *iter;
4757 int cpu;
4758
4759 if (!(file->f_mode & FMODE_READ)) {
4760 trace_array_put(tr);
4761 return 0;
4762 }
4763
4764 /* Writes do not use seq_file */
4765 iter = m->private;
4766 mutex_lock(&trace_types_lock);
4767
4768 for_each_tracing_cpu(cpu) {
4769 if (iter->buffer_iter[cpu])
4770 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4771 }
4772
4773 if (iter->trace && iter->trace->close)
4774 iter->trace->close(iter);
4775
4776 if (!iter->snapshot && tr->stop_count)
4777 /* reenable tracing if it was previously enabled */
4778 tracing_start_tr(tr);
4779
4780 __trace_array_put(tr);
4781
4782 mutex_unlock(&trace_types_lock);
4783
4784 free_trace_iter_content(iter);
4785 seq_release_private(inode, file);
4786
4787 return 0;
4788}
4789
4790int tracing_release_generic_tr(struct inode *inode, struct file *file)
4791{
4792 struct trace_array *tr = inode->i_private;
4793
4794 trace_array_put(tr);
4795 return 0;
4796}
4797
4798static int tracing_single_release_tr(struct inode *inode, struct file *file)
4799{
4800 struct trace_array *tr = inode->i_private;
4801
4802 trace_array_put(tr);
4803
4804 return single_release(inode, file);
4805}
4806
4807static int tracing_open(struct inode *inode, struct file *file)
4808{
4809 struct trace_array *tr = inode->i_private;
4810 struct trace_iterator *iter;
4811 int ret;
4812
4813 ret = tracing_check_open_get_tr(tr);
4814 if (ret)
4815 return ret;
4816
4817 /* If this file was open for write, then erase contents */
4818 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4819 int cpu = tracing_get_cpu(inode);
4820 struct array_buffer *trace_buf = &tr->array_buffer;
4821
4822#ifdef CONFIG_TRACER_MAX_TRACE
4823 if (tr->current_trace->print_max)
4824 trace_buf = &tr->max_buffer;
4825#endif
4826
4827 if (cpu == RING_BUFFER_ALL_CPUS)
4828 tracing_reset_online_cpus(trace_buf);
4829 else
4830 tracing_reset_cpu(trace_buf, cpu);
4831 }
4832
4833 if (file->f_mode & FMODE_READ) {
4834 iter = __tracing_open(inode, file, false);
4835 if (IS_ERR(iter))
4836 ret = PTR_ERR(iter);
4837 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4838 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4839 }
4840
4841 if (ret < 0)
4842 trace_array_put(tr);
4843
4844 return ret;
4845}
4846
4847/*
4848 * Some tracers are not suitable for instance buffers.
4849 * A tracer is always available for the global array (toplevel)
4850 * or if it explicitly states that it is.
4851 */
4852static bool
4853trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4854{
4855#ifdef CONFIG_TRACER_SNAPSHOT
4856 /* arrays with mapped buffer range do not have snapshots */
4857 if (tr->range_addr_start && t->use_max_tr)
4858 return false;
4859#endif
4860 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4861}
4862
4863/* Find the next tracer that this trace array may use */
4864static struct tracer *
4865get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4866{
4867 while (t && !trace_ok_for_array(t, tr))
4868 t = t->next;
4869
4870 return t;
4871}
4872
4873static void *
4874t_next(struct seq_file *m, void *v, loff_t *pos)
4875{
4876 struct trace_array *tr = m->private;
4877 struct tracer *t = v;
4878
4879 (*pos)++;
4880
4881 if (t)
4882 t = get_tracer_for_array(tr, t->next);
4883
4884 return t;
4885}
4886
4887static void *t_start(struct seq_file *m, loff_t *pos)
4888{
4889 struct trace_array *tr = m->private;
4890 struct tracer *t;
4891 loff_t l = 0;
4892
4893 mutex_lock(&trace_types_lock);
4894
4895 t = get_tracer_for_array(tr, trace_types);
4896 for (; t && l < *pos; t = t_next(m, t, &l))
4897 ;
4898
4899 return t;
4900}
4901
4902static void t_stop(struct seq_file *m, void *p)
4903{
4904 mutex_unlock(&trace_types_lock);
4905}
4906
4907static int t_show(struct seq_file *m, void *v)
4908{
4909 struct tracer *t = v;
4910
4911 if (!t)
4912 return 0;
4913
4914 seq_puts(m, t->name);
4915 if (t->next)
4916 seq_putc(m, ' ');
4917 else
4918 seq_putc(m, '\n');
4919
4920 return 0;
4921}
4922
4923static const struct seq_operations show_traces_seq_ops = {
4924 .start = t_start,
4925 .next = t_next,
4926 .stop = t_stop,
4927 .show = t_show,
4928};
4929
4930static int show_traces_open(struct inode *inode, struct file *file)
4931{
4932 struct trace_array *tr = inode->i_private;
4933 struct seq_file *m;
4934 int ret;
4935
4936 ret = tracing_check_open_get_tr(tr);
4937 if (ret)
4938 return ret;
4939
4940 ret = seq_open(file, &show_traces_seq_ops);
4941 if (ret) {
4942 trace_array_put(tr);
4943 return ret;
4944 }
4945
4946 m = file->private_data;
4947 m->private = tr;
4948
4949 return 0;
4950}
4951
4952static int tracing_seq_release(struct inode *inode, struct file *file)
4953{
4954 struct trace_array *tr = inode->i_private;
4955
4956 trace_array_put(tr);
4957 return seq_release(inode, file);
4958}
4959
4960static ssize_t
4961tracing_write_stub(struct file *filp, const char __user *ubuf,
4962 size_t count, loff_t *ppos)
4963{
4964 return count;
4965}
4966
4967loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
4968{
4969 int ret;
4970
4971 if (file->f_mode & FMODE_READ)
4972 ret = seq_lseek(file, offset, whence);
4973 else
4974 file->f_pos = ret = 0;
4975
4976 return ret;
4977}
4978
4979static const struct file_operations tracing_fops = {
4980 .open = tracing_open,
4981 .read = seq_read,
4982 .read_iter = seq_read_iter,
4983 .splice_read = copy_splice_read,
4984 .write = tracing_write_stub,
4985 .llseek = tracing_lseek,
4986 .release = tracing_release,
4987};
4988
4989static const struct file_operations show_traces_fops = {
4990 .open = show_traces_open,
4991 .read = seq_read,
4992 .llseek = seq_lseek,
4993 .release = tracing_seq_release,
4994};
4995
4996static ssize_t
4997tracing_cpumask_read(struct file *filp, char __user *ubuf,
4998 size_t count, loff_t *ppos)
4999{
5000 struct trace_array *tr = file_inode(filp)->i_private;
5001 char *mask_str;
5002 int len;
5003
5004 len = snprintf(NULL, 0, "%*pb\n",
5005 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5006 mask_str = kmalloc(len, GFP_KERNEL);
5007 if (!mask_str)
5008 return -ENOMEM;
5009
5010 len = snprintf(mask_str, len, "%*pb\n",
5011 cpumask_pr_args(tr->tracing_cpumask));
5012 if (len >= count) {
5013 count = -EINVAL;
5014 goto out_err;
5015 }
5016 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5017
5018out_err:
5019 kfree(mask_str);
5020
5021 return count;
5022}
5023
5024int tracing_set_cpumask(struct trace_array *tr,
5025 cpumask_var_t tracing_cpumask_new)
5026{
5027 int cpu;
5028
5029 if (!tr)
5030 return -EINVAL;
5031
5032 local_irq_disable();
5033 arch_spin_lock(&tr->max_lock);
5034 for_each_tracing_cpu(cpu) {
5035 /*
5036 * Increase/decrease the disabled counter if we are
5037 * about to flip a bit in the cpumask:
5038 */
5039 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5040 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5041 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5042 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5043#ifdef CONFIG_TRACER_MAX_TRACE
5044 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5045#endif
5046 }
5047 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5048 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5049 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5050 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5051#ifdef CONFIG_TRACER_MAX_TRACE
5052 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5053#endif
5054 }
5055 }
5056 arch_spin_unlock(&tr->max_lock);
5057 local_irq_enable();
5058
5059 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5060
5061 return 0;
5062}
5063
5064static ssize_t
5065tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5066 size_t count, loff_t *ppos)
5067{
5068 struct trace_array *tr = file_inode(filp)->i_private;
5069 cpumask_var_t tracing_cpumask_new;
5070 int err;
5071
5072 if (count == 0 || count > KMALLOC_MAX_SIZE)
5073 return -EINVAL;
5074
5075 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5076 return -ENOMEM;
5077
5078 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5079 if (err)
5080 goto err_free;
5081
5082 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5083 if (err)
5084 goto err_free;
5085
5086 free_cpumask_var(tracing_cpumask_new);
5087
5088 return count;
5089
5090err_free:
5091 free_cpumask_var(tracing_cpumask_new);
5092
5093 return err;
5094}
5095
5096static const struct file_operations tracing_cpumask_fops = {
5097 .open = tracing_open_generic_tr,
5098 .read = tracing_cpumask_read,
5099 .write = tracing_cpumask_write,
5100 .release = tracing_release_generic_tr,
5101 .llseek = generic_file_llseek,
5102};
5103
5104static int tracing_trace_options_show(struct seq_file *m, void *v)
5105{
5106 struct tracer_opt *trace_opts;
5107 struct trace_array *tr = m->private;
5108 u32 tracer_flags;
5109 int i;
5110
5111 guard(mutex)(&trace_types_lock);
5112
5113 tracer_flags = tr->current_trace->flags->val;
5114 trace_opts = tr->current_trace->flags->opts;
5115
5116 for (i = 0; trace_options[i]; i++) {
5117 if (tr->trace_flags & (1 << i))
5118 seq_printf(m, "%s\n", trace_options[i]);
5119 else
5120 seq_printf(m, "no%s\n", trace_options[i]);
5121 }
5122
5123 for (i = 0; trace_opts[i].name; i++) {
5124 if (tracer_flags & trace_opts[i].bit)
5125 seq_printf(m, "%s\n", trace_opts[i].name);
5126 else
5127 seq_printf(m, "no%s\n", trace_opts[i].name);
5128 }
5129
5130 return 0;
5131}
5132
5133static int __set_tracer_option(struct trace_array *tr,
5134 struct tracer_flags *tracer_flags,
5135 struct tracer_opt *opts, int neg)
5136{
5137 struct tracer *trace = tracer_flags->trace;
5138 int ret;
5139
5140 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5141 if (ret)
5142 return ret;
5143
5144 if (neg)
5145 tracer_flags->val &= ~opts->bit;
5146 else
5147 tracer_flags->val |= opts->bit;
5148 return 0;
5149}
5150
5151/* Try to assign a tracer specific option */
5152static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5153{
5154 struct tracer *trace = tr->current_trace;
5155 struct tracer_flags *tracer_flags = trace->flags;
5156 struct tracer_opt *opts = NULL;
5157 int i;
5158
5159 for (i = 0; tracer_flags->opts[i].name; i++) {
5160 opts = &tracer_flags->opts[i];
5161
5162 if (strcmp(cmp, opts->name) == 0)
5163 return __set_tracer_option(tr, trace->flags, opts, neg);
5164 }
5165
5166 return -EINVAL;
5167}
5168
5169/* Some tracers require overwrite to stay enabled */
5170int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5171{
5172 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5173 return -1;
5174
5175 return 0;
5176}
5177
5178int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5179{
5180 if ((mask == TRACE_ITER_RECORD_TGID) ||
5181 (mask == TRACE_ITER_RECORD_CMD) ||
5182 (mask == TRACE_ITER_TRACE_PRINTK))
5183 lockdep_assert_held(&event_mutex);
5184
5185 /* do nothing if flag is already set */
5186 if (!!(tr->trace_flags & mask) == !!enabled)
5187 return 0;
5188
5189 /* Give the tracer a chance to approve the change */
5190 if (tr->current_trace->flag_changed)
5191 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5192 return -EINVAL;
5193
5194 if (mask == TRACE_ITER_TRACE_PRINTK) {
5195 if (enabled) {
5196 update_printk_trace(tr);
5197 } else {
5198 /*
5199 * The global_trace cannot clear this.
5200 * It's flag only gets cleared if another instance sets it.
5201 */
5202 if (printk_trace == &global_trace)
5203 return -EINVAL;
5204 /*
5205 * An instance must always have it set.
5206 * by default, that's the global_trace instane.
5207 */
5208 if (printk_trace == tr)
5209 update_printk_trace(&global_trace);
5210 }
5211 }
5212
5213 if (enabled)
5214 tr->trace_flags |= mask;
5215 else
5216 tr->trace_flags &= ~mask;
5217
5218 if (mask == TRACE_ITER_RECORD_CMD)
5219 trace_event_enable_cmd_record(enabled);
5220
5221 if (mask == TRACE_ITER_RECORD_TGID) {
5222
5223 if (trace_alloc_tgid_map() < 0) {
5224 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5225 return -ENOMEM;
5226 }
5227
5228 trace_event_enable_tgid_record(enabled);
5229 }
5230
5231 if (mask == TRACE_ITER_EVENT_FORK)
5232 trace_event_follow_fork(tr, enabled);
5233
5234 if (mask == TRACE_ITER_FUNC_FORK)
5235 ftrace_pid_follow_fork(tr, enabled);
5236
5237 if (mask == TRACE_ITER_OVERWRITE) {
5238 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5239#ifdef CONFIG_TRACER_MAX_TRACE
5240 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5241#endif
5242 }
5243
5244 if (mask == TRACE_ITER_PRINTK) {
5245 trace_printk_start_stop_comm(enabled);
5246 trace_printk_control(enabled);
5247 }
5248
5249 return 0;
5250}
5251
5252int trace_set_options(struct trace_array *tr, char *option)
5253{
5254 char *cmp;
5255 int neg = 0;
5256 int ret;
5257 size_t orig_len = strlen(option);
5258 int len;
5259
5260 cmp = strstrip(option);
5261
5262 len = str_has_prefix(cmp, "no");
5263 if (len)
5264 neg = 1;
5265
5266 cmp += len;
5267
5268 mutex_lock(&event_mutex);
5269 mutex_lock(&trace_types_lock);
5270
5271 ret = match_string(trace_options, -1, cmp);
5272 /* If no option could be set, test the specific tracer options */
5273 if (ret < 0)
5274 ret = set_tracer_option(tr, cmp, neg);
5275 else
5276 ret = set_tracer_flag(tr, 1 << ret, !neg);
5277
5278 mutex_unlock(&trace_types_lock);
5279 mutex_unlock(&event_mutex);
5280
5281 /*
5282 * If the first trailing whitespace is replaced with '\0' by strstrip,
5283 * turn it back into a space.
5284 */
5285 if (orig_len > strlen(option))
5286 option[strlen(option)] = ' ';
5287
5288 return ret;
5289}
5290
5291static void __init apply_trace_boot_options(void)
5292{
5293 char *buf = trace_boot_options_buf;
5294 char *option;
5295
5296 while (true) {
5297 option = strsep(&buf, ",");
5298
5299 if (!option)
5300 break;
5301
5302 if (*option)
5303 trace_set_options(&global_trace, option);
5304
5305 /* Put back the comma to allow this to be called again */
5306 if (buf)
5307 *(buf - 1) = ',';
5308 }
5309}
5310
5311static ssize_t
5312tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5313 size_t cnt, loff_t *ppos)
5314{
5315 struct seq_file *m = filp->private_data;
5316 struct trace_array *tr = m->private;
5317 char buf[64];
5318 int ret;
5319
5320 if (cnt >= sizeof(buf))
5321 return -EINVAL;
5322
5323 if (copy_from_user(buf, ubuf, cnt))
5324 return -EFAULT;
5325
5326 buf[cnt] = 0;
5327
5328 ret = trace_set_options(tr, buf);
5329 if (ret < 0)
5330 return ret;
5331
5332 *ppos += cnt;
5333
5334 return cnt;
5335}
5336
5337static int tracing_trace_options_open(struct inode *inode, struct file *file)
5338{
5339 struct trace_array *tr = inode->i_private;
5340 int ret;
5341
5342 ret = tracing_check_open_get_tr(tr);
5343 if (ret)
5344 return ret;
5345
5346 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5347 if (ret < 0)
5348 trace_array_put(tr);
5349
5350 return ret;
5351}
5352
5353static const struct file_operations tracing_iter_fops = {
5354 .open = tracing_trace_options_open,
5355 .read = seq_read,
5356 .llseek = seq_lseek,
5357 .release = tracing_single_release_tr,
5358 .write = tracing_trace_options_write,
5359};
5360
5361static const char readme_msg[] =
5362 "tracing mini-HOWTO:\n\n"
5363 "By default tracefs removes all OTH file permission bits.\n"
5364 "When mounting tracefs an optional group id can be specified\n"
5365 "which adds the group to every directory and file in tracefs:\n\n"
5366 "\t e.g. mount -t tracefs [-o [gid=<gid>]] nodev /sys/kernel/tracing\n\n"
5367 "# echo 0 > tracing_on : quick way to disable tracing\n"
5368 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5369 " Important files:\n"
5370 " trace\t\t\t- The static contents of the buffer\n"
5371 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5372 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5373 " current_tracer\t- function and latency tracers\n"
5374 " available_tracers\t- list of configured tracers for current_tracer\n"
5375 " error_log\t- error log for failed commands (that support it)\n"
5376 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5377 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5378 " trace_clock\t\t- change the clock used to order events\n"
5379 " local: Per cpu clock but may not be synced across CPUs\n"
5380 " global: Synced across CPUs but slows tracing down.\n"
5381 " counter: Not a clock, but just an increment\n"
5382 " uptime: Jiffy counter from time of boot\n"
5383 " perf: Same clock that perf events use\n"
5384#ifdef CONFIG_X86_64
5385 " x86-tsc: TSC cycle counter\n"
5386#endif
5387 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5388 " delta: Delta difference against a buffer-wide timestamp\n"
5389 " absolute: Absolute (standalone) timestamp\n"
5390 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5391 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5392 " tracing_cpumask\t- Limit which CPUs to trace\n"
5393 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5394 "\t\t\t Remove sub-buffer with rmdir\n"
5395 " trace_options\t\t- Set format or modify how tracing happens\n"
5396 "\t\t\t Disable an option by prefixing 'no' to the\n"
5397 "\t\t\t option name\n"
5398 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5399#ifdef CONFIG_DYNAMIC_FTRACE
5400 "\n available_filter_functions - list of functions that can be filtered on\n"
5401 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5402 "\t\t\t functions\n"
5403 "\t accepts: func_full_name or glob-matching-pattern\n"
5404 "\t modules: Can select a group via module\n"
5405 "\t Format: :mod:<module-name>\n"
5406 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5407 "\t triggers: a command to perform when function is hit\n"
5408 "\t Format: <function>:<trigger>[:count]\n"
5409 "\t trigger: traceon, traceoff\n"
5410 "\t\t enable_event:<system>:<event>\n"
5411 "\t\t disable_event:<system>:<event>\n"
5412#ifdef CONFIG_STACKTRACE
5413 "\t\t stacktrace\n"
5414#endif
5415#ifdef CONFIG_TRACER_SNAPSHOT
5416 "\t\t snapshot\n"
5417#endif
5418 "\t\t dump\n"
5419 "\t\t cpudump\n"
5420 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5421 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5422 "\t The first one will disable tracing every time do_fault is hit\n"
5423 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5424 "\t The first time do trap is hit and it disables tracing, the\n"
5425 "\t counter will decrement to 2. If tracing is already disabled,\n"
5426 "\t the counter will not decrement. It only decrements when the\n"
5427 "\t trigger did work\n"
5428 "\t To remove trigger without count:\n"
5429 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5430 "\t To remove trigger with a count:\n"
5431 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5432 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5433 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5434 "\t modules: Can select a group via module command :mod:\n"
5435 "\t Does not accept triggers\n"
5436#endif /* CONFIG_DYNAMIC_FTRACE */
5437#ifdef CONFIG_FUNCTION_TRACER
5438 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5439 "\t\t (function)\n"
5440 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5441 "\t\t (function)\n"
5442#endif
5443#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5444 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5445 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5446 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5447#endif
5448#ifdef CONFIG_TRACER_SNAPSHOT
5449 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5450 "\t\t\t snapshot buffer. Read the contents for more\n"
5451 "\t\t\t information\n"
5452#endif
5453#ifdef CONFIG_STACK_TRACER
5454 " stack_trace\t\t- Shows the max stack trace when active\n"
5455 " stack_max_size\t- Shows current max stack size that was traced\n"
5456 "\t\t\t Write into this file to reset the max size (trigger a\n"
5457 "\t\t\t new trace)\n"
5458#ifdef CONFIG_DYNAMIC_FTRACE
5459 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5460 "\t\t\t traces\n"
5461#endif
5462#endif /* CONFIG_STACK_TRACER */
5463#ifdef CONFIG_DYNAMIC_EVENTS
5464 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5465 "\t\t\t Write into this file to define/undefine new trace events.\n"
5466#endif
5467#ifdef CONFIG_KPROBE_EVENTS
5468 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5469 "\t\t\t Write into this file to define/undefine new trace events.\n"
5470#endif
5471#ifdef CONFIG_UPROBE_EVENTS
5472 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5473 "\t\t\t Write into this file to define/undefine new trace events.\n"
5474#endif
5475#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5476 defined(CONFIG_FPROBE_EVENTS)
5477 "\t accepts: event-definitions (one definition per line)\n"
5478#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5479 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5480 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5481#endif
5482#ifdef CONFIG_FPROBE_EVENTS
5483 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5484 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5485#endif
5486#ifdef CONFIG_HIST_TRIGGERS
5487 "\t s:[synthetic/]<event> <field> [<field>]\n"
5488#endif
5489 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5490 "\t -:[<group>/][<event>]\n"
5491#ifdef CONFIG_KPROBE_EVENTS
5492 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5493 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5494#endif
5495#ifdef CONFIG_UPROBE_EVENTS
5496 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5497#endif
5498 "\t args: <name>=fetcharg[:type]\n"
5499 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5500#ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5501 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5502#ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5503 "\t <argname>[->field[->field|.field...]],\n"
5504#endif
5505#else
5506 "\t $stack<index>, $stack, $retval, $comm,\n"
5507#endif
5508 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5509 "\t kernel return probes support: $retval, $arg<N>, $comm\n"
5510 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5511 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5512 "\t symstr, %pd/%pD, <type>\\[<array-size>\\]\n"
5513#ifdef CONFIG_HIST_TRIGGERS
5514 "\t field: <stype> <name>;\n"
5515 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5516 "\t [unsigned] char/int/long\n"
5517#endif
5518 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5519 "\t of the <attached-group>/<attached-event>.\n"
5520#endif
5521 " events/\t\t- Directory containing all trace event subsystems:\n"
5522 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5523 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5524 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5525 "\t\t\t events\n"
5526 " filter\t\t- If set, only events passing filter are traced\n"
5527 " events/<system>/<event>/\t- Directory containing control files for\n"
5528 "\t\t\t <event>:\n"
5529 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5530 " filter\t\t- If set, only events passing filter are traced\n"
5531 " trigger\t\t- If set, a command to perform when event is hit\n"
5532 "\t Format: <trigger>[:count][if <filter>]\n"
5533 "\t trigger: traceon, traceoff\n"
5534 "\t enable_event:<system>:<event>\n"
5535 "\t disable_event:<system>:<event>\n"
5536#ifdef CONFIG_HIST_TRIGGERS
5537 "\t enable_hist:<system>:<event>\n"
5538 "\t disable_hist:<system>:<event>\n"
5539#endif
5540#ifdef CONFIG_STACKTRACE
5541 "\t\t stacktrace\n"
5542#endif
5543#ifdef CONFIG_TRACER_SNAPSHOT
5544 "\t\t snapshot\n"
5545#endif
5546#ifdef CONFIG_HIST_TRIGGERS
5547 "\t\t hist (see below)\n"
5548#endif
5549 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5550 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5551 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5552 "\t events/block/block_unplug/trigger\n"
5553 "\t The first disables tracing every time block_unplug is hit.\n"
5554 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5555 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5556 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5557 "\t Like function triggers, the counter is only decremented if it\n"
5558 "\t enabled or disabled tracing.\n"
5559 "\t To remove a trigger without a count:\n"
5560 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5561 "\t To remove a trigger with a count:\n"
5562 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5563 "\t Filters can be ignored when removing a trigger.\n"
5564#ifdef CONFIG_HIST_TRIGGERS
5565 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5566 "\t Format: hist:keys=<field1[,field2,...]>\n"
5567 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5568 "\t [:values=<field1[,field2,...]>]\n"
5569 "\t [:sort=<field1[,field2,...]>]\n"
5570 "\t [:size=#entries]\n"
5571 "\t [:pause][:continue][:clear]\n"
5572 "\t [:name=histname1]\n"
5573 "\t [:nohitcount]\n"
5574 "\t [:<handler>.<action>]\n"
5575 "\t [if <filter>]\n\n"
5576 "\t Note, special fields can be used as well:\n"
5577 "\t common_timestamp - to record current timestamp\n"
5578 "\t common_cpu - to record the CPU the event happened on\n"
5579 "\n"
5580 "\t A hist trigger variable can be:\n"
5581 "\t - a reference to a field e.g. x=current_timestamp,\n"
5582 "\t - a reference to another variable e.g. y=$x,\n"
5583 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5584 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5585 "\n"
5586 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5587 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5588 "\t variable reference, field or numeric literal.\n"
5589 "\n"
5590 "\t When a matching event is hit, an entry is added to a hash\n"
5591 "\t table using the key(s) and value(s) named, and the value of a\n"
5592 "\t sum called 'hitcount' is incremented. Keys and values\n"
5593 "\t correspond to fields in the event's format description. Keys\n"
5594 "\t can be any field, or the special string 'common_stacktrace'.\n"
5595 "\t Compound keys consisting of up to two fields can be specified\n"
5596 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5597 "\t fields. Sort keys consisting of up to two fields can be\n"
5598 "\t specified using the 'sort' keyword. The sort direction can\n"
5599 "\t be modified by appending '.descending' or '.ascending' to a\n"
5600 "\t sort field. The 'size' parameter can be used to specify more\n"
5601 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5602 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5603 "\t its histogram data will be shared with other triggers of the\n"
5604 "\t same name, and trigger hits will update this common data.\n\n"
5605 "\t Reading the 'hist' file for the event will dump the hash\n"
5606 "\t table in its entirety to stdout. If there are multiple hist\n"
5607 "\t triggers attached to an event, there will be a table for each\n"
5608 "\t trigger in the output. The table displayed for a named\n"
5609 "\t trigger will be the same as any other instance having the\n"
5610 "\t same name. The default format used to display a given field\n"
5611 "\t can be modified by appending any of the following modifiers\n"
5612 "\t to the field name, as applicable:\n\n"
5613 "\t .hex display a number as a hex value\n"
5614 "\t .sym display an address as a symbol\n"
5615 "\t .sym-offset display an address as a symbol and offset\n"
5616 "\t .execname display a common_pid as a program name\n"
5617 "\t .syscall display a syscall id as a syscall name\n"
5618 "\t .log2 display log2 value rather than raw number\n"
5619 "\t .buckets=size display values in groups of size rather than raw number\n"
5620 "\t .usecs display a common_timestamp in microseconds\n"
5621 "\t .percent display a number of percentage value\n"
5622 "\t .graph display a bar-graph of a value\n\n"
5623 "\t The 'pause' parameter can be used to pause an existing hist\n"
5624 "\t trigger or to start a hist trigger but not log any events\n"
5625 "\t until told to do so. 'continue' can be used to start or\n"
5626 "\t restart a paused hist trigger.\n\n"
5627 "\t The 'clear' parameter will clear the contents of a running\n"
5628 "\t hist trigger and leave its current paused/active state\n"
5629 "\t unchanged.\n\n"
5630 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5631 "\t raw hitcount in the histogram.\n\n"
5632 "\t The enable_hist and disable_hist triggers can be used to\n"
5633 "\t have one event conditionally start and stop another event's\n"
5634 "\t already-attached hist trigger. The syntax is analogous to\n"
5635 "\t the enable_event and disable_event triggers.\n\n"
5636 "\t Hist trigger handlers and actions are executed whenever a\n"
5637 "\t a histogram entry is added or updated. They take the form:\n\n"
5638 "\t <handler>.<action>\n\n"
5639 "\t The available handlers are:\n\n"
5640 "\t onmatch(matching.event) - invoke on addition or update\n"
5641 "\t onmax(var) - invoke if var exceeds current max\n"
5642 "\t onchange(var) - invoke action if var changes\n\n"
5643 "\t The available actions are:\n\n"
5644 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5645 "\t save(field,...) - save current event fields\n"
5646#ifdef CONFIG_TRACER_SNAPSHOT
5647 "\t snapshot() - snapshot the trace buffer\n\n"
5648#endif
5649#ifdef CONFIG_SYNTH_EVENTS
5650 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5651 "\t Write into this file to define/undefine new synthetic events.\n"
5652 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5653#endif
5654#endif
5655;
5656
5657static ssize_t
5658tracing_readme_read(struct file *filp, char __user *ubuf,
5659 size_t cnt, loff_t *ppos)
5660{
5661 return simple_read_from_buffer(ubuf, cnt, ppos,
5662 readme_msg, strlen(readme_msg));
5663}
5664
5665static const struct file_operations tracing_readme_fops = {
5666 .open = tracing_open_generic,
5667 .read = tracing_readme_read,
5668 .llseek = generic_file_llseek,
5669};
5670
5671#ifdef CONFIG_TRACE_EVAL_MAP_FILE
5672static union trace_eval_map_item *
5673update_eval_map(union trace_eval_map_item *ptr)
5674{
5675 if (!ptr->map.eval_string) {
5676 if (ptr->tail.next) {
5677 ptr = ptr->tail.next;
5678 /* Set ptr to the next real item (skip head) */
5679 ptr++;
5680 } else
5681 return NULL;
5682 }
5683 return ptr;
5684}
5685
5686static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5687{
5688 union trace_eval_map_item *ptr = v;
5689
5690 /*
5691 * Paranoid! If ptr points to end, we don't want to increment past it.
5692 * This really should never happen.
5693 */
5694 (*pos)++;
5695 ptr = update_eval_map(ptr);
5696 if (WARN_ON_ONCE(!ptr))
5697 return NULL;
5698
5699 ptr++;
5700 ptr = update_eval_map(ptr);
5701
5702 return ptr;
5703}
5704
5705static void *eval_map_start(struct seq_file *m, loff_t *pos)
5706{
5707 union trace_eval_map_item *v;
5708 loff_t l = 0;
5709
5710 mutex_lock(&trace_eval_mutex);
5711
5712 v = trace_eval_maps;
5713 if (v)
5714 v++;
5715
5716 while (v && l < *pos) {
5717 v = eval_map_next(m, v, &l);
5718 }
5719
5720 return v;
5721}
5722
5723static void eval_map_stop(struct seq_file *m, void *v)
5724{
5725 mutex_unlock(&trace_eval_mutex);
5726}
5727
5728static int eval_map_show(struct seq_file *m, void *v)
5729{
5730 union trace_eval_map_item *ptr = v;
5731
5732 seq_printf(m, "%s %ld (%s)\n",
5733 ptr->map.eval_string, ptr->map.eval_value,
5734 ptr->map.system);
5735
5736 return 0;
5737}
5738
5739static const struct seq_operations tracing_eval_map_seq_ops = {
5740 .start = eval_map_start,
5741 .next = eval_map_next,
5742 .stop = eval_map_stop,
5743 .show = eval_map_show,
5744};
5745
5746static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5747{
5748 int ret;
5749
5750 ret = tracing_check_open_get_tr(NULL);
5751 if (ret)
5752 return ret;
5753
5754 return seq_open(filp, &tracing_eval_map_seq_ops);
5755}
5756
5757static const struct file_operations tracing_eval_map_fops = {
5758 .open = tracing_eval_map_open,
5759 .read = seq_read,
5760 .llseek = seq_lseek,
5761 .release = seq_release,
5762};
5763
5764static inline union trace_eval_map_item *
5765trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
5766{
5767 /* Return tail of array given the head */
5768 return ptr + ptr->head.length + 1;
5769}
5770
5771static void
5772trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
5773 int len)
5774{
5775 struct trace_eval_map **stop;
5776 struct trace_eval_map **map;
5777 union trace_eval_map_item *map_array;
5778 union trace_eval_map_item *ptr;
5779
5780 stop = start + len;
5781
5782 /*
5783 * The trace_eval_maps contains the map plus a head and tail item,
5784 * where the head holds the module and length of array, and the
5785 * tail holds a pointer to the next list.
5786 */
5787 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
5788 if (!map_array) {
5789 pr_warn("Unable to allocate trace eval mapping\n");
5790 return;
5791 }
5792
5793 guard(mutex)(&trace_eval_mutex);
5794
5795 if (!trace_eval_maps)
5796 trace_eval_maps = map_array;
5797 else {
5798 ptr = trace_eval_maps;
5799 for (;;) {
5800 ptr = trace_eval_jmp_to_tail(ptr);
5801 if (!ptr->tail.next)
5802 break;
5803 ptr = ptr->tail.next;
5804
5805 }
5806 ptr->tail.next = map_array;
5807 }
5808 map_array->head.mod = mod;
5809 map_array->head.length = len;
5810 map_array++;
5811
5812 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
5813 map_array->map = **map;
5814 map_array++;
5815 }
5816 memset(map_array, 0, sizeof(*map_array));
5817}
5818
5819static void trace_create_eval_file(struct dentry *d_tracer)
5820{
5821 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
5822 NULL, &tracing_eval_map_fops);
5823}
5824
5825#else /* CONFIG_TRACE_EVAL_MAP_FILE */
5826static inline void trace_create_eval_file(struct dentry *d_tracer) { }
5827static inline void trace_insert_eval_map_file(struct module *mod,
5828 struct trace_eval_map **start, int len) { }
5829#endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
5830
5831static void trace_insert_eval_map(struct module *mod,
5832 struct trace_eval_map **start, int len)
5833{
5834 struct trace_eval_map **map;
5835
5836 if (len <= 0)
5837 return;
5838
5839 map = start;
5840
5841 trace_event_eval_update(map, len);
5842
5843 trace_insert_eval_map_file(mod, start, len);
5844}
5845
5846static ssize_t
5847tracing_set_trace_read(struct file *filp, char __user *ubuf,
5848 size_t cnt, loff_t *ppos)
5849{
5850 struct trace_array *tr = filp->private_data;
5851 char buf[MAX_TRACER_SIZE+2];
5852 int r;
5853
5854 mutex_lock(&trace_types_lock);
5855 r = sprintf(buf, "%s\n", tr->current_trace->name);
5856 mutex_unlock(&trace_types_lock);
5857
5858 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5859}
5860
5861int tracer_init(struct tracer *t, struct trace_array *tr)
5862{
5863 tracing_reset_online_cpus(&tr->array_buffer);
5864 return t->init(tr);
5865}
5866
5867static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
5868{
5869 int cpu;
5870
5871 for_each_tracing_cpu(cpu)
5872 per_cpu_ptr(buf->data, cpu)->entries = val;
5873}
5874
5875static void update_buffer_entries(struct array_buffer *buf, int cpu)
5876{
5877 if (cpu == RING_BUFFER_ALL_CPUS) {
5878 set_buffer_entries(buf, ring_buffer_size(buf->buffer, 0));
5879 } else {
5880 per_cpu_ptr(buf->data, cpu)->entries = ring_buffer_size(buf->buffer, cpu);
5881 }
5882}
5883
5884#ifdef CONFIG_TRACER_MAX_TRACE
5885/* resize @tr's buffer to the size of @size_tr's entries */
5886static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
5887 struct array_buffer *size_buf, int cpu_id)
5888{
5889 int cpu, ret = 0;
5890
5891 if (cpu_id == RING_BUFFER_ALL_CPUS) {
5892 for_each_tracing_cpu(cpu) {
5893 ret = ring_buffer_resize(trace_buf->buffer,
5894 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
5895 if (ret < 0)
5896 break;
5897 per_cpu_ptr(trace_buf->data, cpu)->entries =
5898 per_cpu_ptr(size_buf->data, cpu)->entries;
5899 }
5900 } else {
5901 ret = ring_buffer_resize(trace_buf->buffer,
5902 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
5903 if (ret == 0)
5904 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
5905 per_cpu_ptr(size_buf->data, cpu_id)->entries;
5906 }
5907
5908 return ret;
5909}
5910#endif /* CONFIG_TRACER_MAX_TRACE */
5911
5912static int __tracing_resize_ring_buffer(struct trace_array *tr,
5913 unsigned long size, int cpu)
5914{
5915 int ret;
5916
5917 /*
5918 * If kernel or user changes the size of the ring buffer
5919 * we use the size that was given, and we can forget about
5920 * expanding it later.
5921 */
5922 trace_set_ring_buffer_expanded(tr);
5923
5924 /* May be called before buffers are initialized */
5925 if (!tr->array_buffer.buffer)
5926 return 0;
5927
5928 /* Do not allow tracing while resizing ring buffer */
5929 tracing_stop_tr(tr);
5930
5931 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
5932 if (ret < 0)
5933 goto out_start;
5934
5935#ifdef CONFIG_TRACER_MAX_TRACE
5936 if (!tr->allocated_snapshot)
5937 goto out;
5938
5939 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
5940 if (ret < 0) {
5941 int r = resize_buffer_duplicate_size(&tr->array_buffer,
5942 &tr->array_buffer, cpu);
5943 if (r < 0) {
5944 /*
5945 * AARGH! We are left with different
5946 * size max buffer!!!!
5947 * The max buffer is our "snapshot" buffer.
5948 * When a tracer needs a snapshot (one of the
5949 * latency tracers), it swaps the max buffer
5950 * with the saved snap shot. We succeeded to
5951 * update the size of the main buffer, but failed to
5952 * update the size of the max buffer. But when we tried
5953 * to reset the main buffer to the original size, we
5954 * failed there too. This is very unlikely to
5955 * happen, but if it does, warn and kill all
5956 * tracing.
5957 */
5958 WARN_ON(1);
5959 tracing_disabled = 1;
5960 }
5961 goto out_start;
5962 }
5963
5964 update_buffer_entries(&tr->max_buffer, cpu);
5965
5966 out:
5967#endif /* CONFIG_TRACER_MAX_TRACE */
5968
5969 update_buffer_entries(&tr->array_buffer, cpu);
5970 out_start:
5971 tracing_start_tr(tr);
5972 return ret;
5973}
5974
5975ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
5976 unsigned long size, int cpu_id)
5977{
5978 guard(mutex)(&trace_types_lock);
5979
5980 if (cpu_id != RING_BUFFER_ALL_CPUS) {
5981 /* make sure, this cpu is enabled in the mask */
5982 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask))
5983 return -EINVAL;
5984 }
5985
5986 return __tracing_resize_ring_buffer(tr, size, cpu_id);
5987}
5988
5989static void update_last_data(struct trace_array *tr)
5990{
5991 if (!tr->text_delta && !tr->data_delta)
5992 return;
5993
5994 /*
5995 * Need to clear all CPU buffers as there cannot be events
5996 * from the previous boot mixed with events with this boot
5997 * as that will cause a confusing trace. Need to clear all
5998 * CPU buffers, even for those that may currently be offline.
5999 */
6000 tracing_reset_all_cpus(&tr->array_buffer);
6001
6002 /* Using current data now */
6003 tr->text_delta = 0;
6004 tr->data_delta = 0;
6005}
6006
6007/**
6008 * tracing_update_buffers - used by tracing facility to expand ring buffers
6009 * @tr: The tracing instance
6010 *
6011 * To save on memory when the tracing is never used on a system with it
6012 * configured in. The ring buffers are set to a minimum size. But once
6013 * a user starts to use the tracing facility, then they need to grow
6014 * to their default size.
6015 *
6016 * This function is to be called when a tracer is about to be used.
6017 */
6018int tracing_update_buffers(struct trace_array *tr)
6019{
6020 int ret = 0;
6021
6022 mutex_lock(&trace_types_lock);
6023
6024 update_last_data(tr);
6025
6026 if (!tr->ring_buffer_expanded)
6027 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6028 RING_BUFFER_ALL_CPUS);
6029 mutex_unlock(&trace_types_lock);
6030
6031 return ret;
6032}
6033
6034struct trace_option_dentry;
6035
6036static void
6037create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6038
6039/*
6040 * Used to clear out the tracer before deletion of an instance.
6041 * Must have trace_types_lock held.
6042 */
6043static void tracing_set_nop(struct trace_array *tr)
6044{
6045 if (tr->current_trace == &nop_trace)
6046 return;
6047
6048 tr->current_trace->enabled--;
6049
6050 if (tr->current_trace->reset)
6051 tr->current_trace->reset(tr);
6052
6053 tr->current_trace = &nop_trace;
6054}
6055
6056static bool tracer_options_updated;
6057
6058static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6059{
6060 /* Only enable if the directory has been created already. */
6061 if (!tr->dir)
6062 return;
6063
6064 /* Only create trace option files after update_tracer_options finish */
6065 if (!tracer_options_updated)
6066 return;
6067
6068 create_trace_option_files(tr, t);
6069}
6070
6071int tracing_set_tracer(struct trace_array *tr, const char *buf)
6072{
6073 struct tracer *t;
6074#ifdef CONFIG_TRACER_MAX_TRACE
6075 bool had_max_tr;
6076#endif
6077 int ret;
6078
6079 guard(mutex)(&trace_types_lock);
6080
6081 update_last_data(tr);
6082
6083 if (!tr->ring_buffer_expanded) {
6084 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6085 RING_BUFFER_ALL_CPUS);
6086 if (ret < 0)
6087 return ret;
6088 ret = 0;
6089 }
6090
6091 for (t = trace_types; t; t = t->next) {
6092 if (strcmp(t->name, buf) == 0)
6093 break;
6094 }
6095 if (!t)
6096 return -EINVAL;
6097
6098 if (t == tr->current_trace)
6099 return 0;
6100
6101#ifdef CONFIG_TRACER_SNAPSHOT
6102 if (t->use_max_tr) {
6103 local_irq_disable();
6104 arch_spin_lock(&tr->max_lock);
6105 ret = tr->cond_snapshot ? -EBUSY : 0;
6106 arch_spin_unlock(&tr->max_lock);
6107 local_irq_enable();
6108 if (ret)
6109 return ret;
6110 }
6111#endif
6112 /* Some tracers won't work on kernel command line */
6113 if (system_state < SYSTEM_RUNNING && t->noboot) {
6114 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6115 t->name);
6116 return 0;
6117 }
6118
6119 /* Some tracers are only allowed for the top level buffer */
6120 if (!trace_ok_for_array(t, tr))
6121 return -EINVAL;
6122
6123 /* If trace pipe files are being read, we can't change the tracer */
6124 if (tr->trace_ref)
6125 return -EBUSY;
6126
6127 trace_branch_disable();
6128
6129 tr->current_trace->enabled--;
6130
6131 if (tr->current_trace->reset)
6132 tr->current_trace->reset(tr);
6133
6134#ifdef CONFIG_TRACER_MAX_TRACE
6135 had_max_tr = tr->current_trace->use_max_tr;
6136
6137 /* Current trace needs to be nop_trace before synchronize_rcu */
6138 tr->current_trace = &nop_trace;
6139
6140 if (had_max_tr && !t->use_max_tr) {
6141 /*
6142 * We need to make sure that the update_max_tr sees that
6143 * current_trace changed to nop_trace to keep it from
6144 * swapping the buffers after we resize it.
6145 * The update_max_tr is called from interrupts disabled
6146 * so a synchronized_sched() is sufficient.
6147 */
6148 synchronize_rcu();
6149 free_snapshot(tr);
6150 tracing_disarm_snapshot(tr);
6151 }
6152
6153 if (!had_max_tr && t->use_max_tr) {
6154 ret = tracing_arm_snapshot_locked(tr);
6155 if (ret)
6156 return ret;
6157 }
6158#else
6159 tr->current_trace = &nop_trace;
6160#endif
6161
6162 if (t->init) {
6163 ret = tracer_init(t, tr);
6164 if (ret) {
6165#ifdef CONFIG_TRACER_MAX_TRACE
6166 if (t->use_max_tr)
6167 tracing_disarm_snapshot(tr);
6168#endif
6169 return ret;
6170 }
6171 }
6172
6173 tr->current_trace = t;
6174 tr->current_trace->enabled++;
6175 trace_branch_enable(tr);
6176
6177 return 0;
6178}
6179
6180static ssize_t
6181tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6182 size_t cnt, loff_t *ppos)
6183{
6184 struct trace_array *tr = filp->private_data;
6185 char buf[MAX_TRACER_SIZE+1];
6186 char *name;
6187 size_t ret;
6188 int err;
6189
6190 ret = cnt;
6191
6192 if (cnt > MAX_TRACER_SIZE)
6193 cnt = MAX_TRACER_SIZE;
6194
6195 if (copy_from_user(buf, ubuf, cnt))
6196 return -EFAULT;
6197
6198 buf[cnt] = 0;
6199
6200 name = strim(buf);
6201
6202 err = tracing_set_tracer(tr, name);
6203 if (err)
6204 return err;
6205
6206 *ppos += ret;
6207
6208 return ret;
6209}
6210
6211static ssize_t
6212tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6213 size_t cnt, loff_t *ppos)
6214{
6215 char buf[64];
6216 int r;
6217
6218 r = snprintf(buf, sizeof(buf), "%ld\n",
6219 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6220 if (r > sizeof(buf))
6221 r = sizeof(buf);
6222 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6223}
6224
6225static ssize_t
6226tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6227 size_t cnt, loff_t *ppos)
6228{
6229 unsigned long val;
6230 int ret;
6231
6232 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6233 if (ret)
6234 return ret;
6235
6236 *ptr = val * 1000;
6237
6238 return cnt;
6239}
6240
6241static ssize_t
6242tracing_thresh_read(struct file *filp, char __user *ubuf,
6243 size_t cnt, loff_t *ppos)
6244{
6245 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6246}
6247
6248static ssize_t
6249tracing_thresh_write(struct file *filp, const char __user *ubuf,
6250 size_t cnt, loff_t *ppos)
6251{
6252 struct trace_array *tr = filp->private_data;
6253 int ret;
6254
6255 guard(mutex)(&trace_types_lock);
6256 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6257 if (ret < 0)
6258 return ret;
6259
6260 if (tr->current_trace->update_thresh) {
6261 ret = tr->current_trace->update_thresh(tr);
6262 if (ret < 0)
6263 return ret;
6264 }
6265
6266 return cnt;
6267}
6268
6269#ifdef CONFIG_TRACER_MAX_TRACE
6270
6271static ssize_t
6272tracing_max_lat_read(struct file *filp, char __user *ubuf,
6273 size_t cnt, loff_t *ppos)
6274{
6275 struct trace_array *tr = filp->private_data;
6276
6277 return tracing_nsecs_read(&tr->max_latency, ubuf, cnt, ppos);
6278}
6279
6280static ssize_t
6281tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6282 size_t cnt, loff_t *ppos)
6283{
6284 struct trace_array *tr = filp->private_data;
6285
6286 return tracing_nsecs_write(&tr->max_latency, ubuf, cnt, ppos);
6287}
6288
6289#endif
6290
6291static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6292{
6293 if (cpu == RING_BUFFER_ALL_CPUS) {
6294 if (cpumask_empty(tr->pipe_cpumask)) {
6295 cpumask_setall(tr->pipe_cpumask);
6296 return 0;
6297 }
6298 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6299 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6300 return 0;
6301 }
6302 return -EBUSY;
6303}
6304
6305static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6306{
6307 if (cpu == RING_BUFFER_ALL_CPUS) {
6308 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6309 cpumask_clear(tr->pipe_cpumask);
6310 } else {
6311 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6312 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6313 }
6314}
6315
6316static int tracing_open_pipe(struct inode *inode, struct file *filp)
6317{
6318 struct trace_array *tr = inode->i_private;
6319 struct trace_iterator *iter;
6320 int cpu;
6321 int ret;
6322
6323 ret = tracing_check_open_get_tr(tr);
6324 if (ret)
6325 return ret;
6326
6327 mutex_lock(&trace_types_lock);
6328 cpu = tracing_get_cpu(inode);
6329 ret = open_pipe_on_cpu(tr, cpu);
6330 if (ret)
6331 goto fail_pipe_on_cpu;
6332
6333 /* create a buffer to store the information to pass to userspace */
6334 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6335 if (!iter) {
6336 ret = -ENOMEM;
6337 goto fail_alloc_iter;
6338 }
6339
6340 trace_seq_init(&iter->seq);
6341 iter->trace = tr->current_trace;
6342
6343 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6344 ret = -ENOMEM;
6345 goto fail;
6346 }
6347
6348 /* trace pipe does not show start of buffer */
6349 cpumask_setall(iter->started);
6350
6351 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6352 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6353
6354 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6355 if (trace_clocks[tr->clock_id].in_ns)
6356 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6357
6358 iter->tr = tr;
6359 iter->array_buffer = &tr->array_buffer;
6360 iter->cpu_file = cpu;
6361 mutex_init(&iter->mutex);
6362 filp->private_data = iter;
6363
6364 if (iter->trace->pipe_open)
6365 iter->trace->pipe_open(iter);
6366
6367 nonseekable_open(inode, filp);
6368
6369 tr->trace_ref++;
6370
6371 mutex_unlock(&trace_types_lock);
6372 return ret;
6373
6374fail:
6375 kfree(iter);
6376fail_alloc_iter:
6377 close_pipe_on_cpu(tr, cpu);
6378fail_pipe_on_cpu:
6379 __trace_array_put(tr);
6380 mutex_unlock(&trace_types_lock);
6381 return ret;
6382}
6383
6384static int tracing_release_pipe(struct inode *inode, struct file *file)
6385{
6386 struct trace_iterator *iter = file->private_data;
6387 struct trace_array *tr = inode->i_private;
6388
6389 mutex_lock(&trace_types_lock);
6390
6391 tr->trace_ref--;
6392
6393 if (iter->trace->pipe_close)
6394 iter->trace->pipe_close(iter);
6395 close_pipe_on_cpu(tr, iter->cpu_file);
6396 mutex_unlock(&trace_types_lock);
6397
6398 free_trace_iter_content(iter);
6399 kfree(iter);
6400
6401 trace_array_put(tr);
6402
6403 return 0;
6404}
6405
6406static __poll_t
6407trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6408{
6409 struct trace_array *tr = iter->tr;
6410
6411 /* Iterators are static, they should be filled or empty */
6412 if (trace_buffer_iter(iter, iter->cpu_file))
6413 return EPOLLIN | EPOLLRDNORM;
6414
6415 if (tr->trace_flags & TRACE_ITER_BLOCK)
6416 /*
6417 * Always select as readable when in blocking mode
6418 */
6419 return EPOLLIN | EPOLLRDNORM;
6420 else
6421 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6422 filp, poll_table, iter->tr->buffer_percent);
6423}
6424
6425static __poll_t
6426tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6427{
6428 struct trace_iterator *iter = filp->private_data;
6429
6430 return trace_poll(iter, filp, poll_table);
6431}
6432
6433/* Must be called with iter->mutex held. */
6434static int tracing_wait_pipe(struct file *filp)
6435{
6436 struct trace_iterator *iter = filp->private_data;
6437 int ret;
6438
6439 while (trace_empty(iter)) {
6440
6441 if ((filp->f_flags & O_NONBLOCK)) {
6442 return -EAGAIN;
6443 }
6444
6445 /*
6446 * We block until we read something and tracing is disabled.
6447 * We still block if tracing is disabled, but we have never
6448 * read anything. This allows a user to cat this file, and
6449 * then enable tracing. But after we have read something,
6450 * we give an EOF when tracing is again disabled.
6451 *
6452 * iter->pos will be 0 if we haven't read anything.
6453 */
6454 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6455 break;
6456
6457 mutex_unlock(&iter->mutex);
6458
6459 ret = wait_on_pipe(iter, 0);
6460
6461 mutex_lock(&iter->mutex);
6462
6463 if (ret)
6464 return ret;
6465 }
6466
6467 return 1;
6468}
6469
6470/*
6471 * Consumer reader.
6472 */
6473static ssize_t
6474tracing_read_pipe(struct file *filp, char __user *ubuf,
6475 size_t cnt, loff_t *ppos)
6476{
6477 struct trace_iterator *iter = filp->private_data;
6478 ssize_t sret;
6479
6480 /*
6481 * Avoid more than one consumer on a single file descriptor
6482 * This is just a matter of traces coherency, the ring buffer itself
6483 * is protected.
6484 */
6485 guard(mutex)(&iter->mutex);
6486
6487 /* return any leftover data */
6488 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6489 if (sret != -EBUSY)
6490 return sret;
6491
6492 trace_seq_init(&iter->seq);
6493
6494 if (iter->trace->read) {
6495 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6496 if (sret)
6497 return sret;
6498 }
6499
6500waitagain:
6501 sret = tracing_wait_pipe(filp);
6502 if (sret <= 0)
6503 return sret;
6504
6505 /* stop when tracing is finished */
6506 if (trace_empty(iter))
6507 return 0;
6508
6509 if (cnt >= TRACE_SEQ_BUFFER_SIZE)
6510 cnt = TRACE_SEQ_BUFFER_SIZE - 1;
6511
6512 /* reset all but tr, trace, and overruns */
6513 trace_iterator_reset(iter);
6514 cpumask_clear(iter->started);
6515 trace_seq_init(&iter->seq);
6516
6517 trace_event_read_lock();
6518 trace_access_lock(iter->cpu_file);
6519 while (trace_find_next_entry_inc(iter) != NULL) {
6520 enum print_line_t ret;
6521 int save_len = iter->seq.seq.len;
6522
6523 ret = print_trace_line(iter);
6524 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6525 /*
6526 * If one print_trace_line() fills entire trace_seq in one shot,
6527 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6528 * In this case, we need to consume it, otherwise, loop will peek
6529 * this event next time, resulting in an infinite loop.
6530 */
6531 if (save_len == 0) {
6532 iter->seq.full = 0;
6533 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6534 trace_consume(iter);
6535 break;
6536 }
6537
6538 /* In other cases, don't print partial lines */
6539 iter->seq.seq.len = save_len;
6540 break;
6541 }
6542 if (ret != TRACE_TYPE_NO_CONSUME)
6543 trace_consume(iter);
6544
6545 if (trace_seq_used(&iter->seq) >= cnt)
6546 break;
6547
6548 /*
6549 * Setting the full flag means we reached the trace_seq buffer
6550 * size and we should leave by partial output condition above.
6551 * One of the trace_seq_* functions is not used properly.
6552 */
6553 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6554 iter->ent->type);
6555 }
6556 trace_access_unlock(iter->cpu_file);
6557 trace_event_read_unlock();
6558
6559 /* Now copy what we have to the user */
6560 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6561 if (iter->seq.readpos >= trace_seq_used(&iter->seq))
6562 trace_seq_init(&iter->seq);
6563
6564 /*
6565 * If there was nothing to send to user, in spite of consuming trace
6566 * entries, go back to wait for more entries.
6567 */
6568 if (sret == -EBUSY)
6569 goto waitagain;
6570
6571 return sret;
6572}
6573
6574static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6575 unsigned int idx)
6576{
6577 __free_page(spd->pages[idx]);
6578}
6579
6580static size_t
6581tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6582{
6583 size_t count;
6584 int save_len;
6585 int ret;
6586
6587 /* Seq buffer is page-sized, exactly what we need. */
6588 for (;;) {
6589 save_len = iter->seq.seq.len;
6590 ret = print_trace_line(iter);
6591
6592 if (trace_seq_has_overflowed(&iter->seq)) {
6593 iter->seq.seq.len = save_len;
6594 break;
6595 }
6596
6597 /*
6598 * This should not be hit, because it should only
6599 * be set if the iter->seq overflowed. But check it
6600 * anyway to be safe.
6601 */
6602 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6603 iter->seq.seq.len = save_len;
6604 break;
6605 }
6606
6607 count = trace_seq_used(&iter->seq) - save_len;
6608 if (rem < count) {
6609 rem = 0;
6610 iter->seq.seq.len = save_len;
6611 break;
6612 }
6613
6614 if (ret != TRACE_TYPE_NO_CONSUME)
6615 trace_consume(iter);
6616 rem -= count;
6617 if (!trace_find_next_entry_inc(iter)) {
6618 rem = 0;
6619 iter->ent = NULL;
6620 break;
6621 }
6622 }
6623
6624 return rem;
6625}
6626
6627static ssize_t tracing_splice_read_pipe(struct file *filp,
6628 loff_t *ppos,
6629 struct pipe_inode_info *pipe,
6630 size_t len,
6631 unsigned int flags)
6632{
6633 struct page *pages_def[PIPE_DEF_BUFFERS];
6634 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6635 struct trace_iterator *iter = filp->private_data;
6636 struct splice_pipe_desc spd = {
6637 .pages = pages_def,
6638 .partial = partial_def,
6639 .nr_pages = 0, /* This gets updated below. */
6640 .nr_pages_max = PIPE_DEF_BUFFERS,
6641 .ops = &default_pipe_buf_ops,
6642 .spd_release = tracing_spd_release_pipe,
6643 };
6644 ssize_t ret;
6645 size_t rem;
6646 unsigned int i;
6647
6648 if (splice_grow_spd(pipe, &spd))
6649 return -ENOMEM;
6650
6651 mutex_lock(&iter->mutex);
6652
6653 if (iter->trace->splice_read) {
6654 ret = iter->trace->splice_read(iter, filp,
6655 ppos, pipe, len, flags);
6656 if (ret)
6657 goto out_err;
6658 }
6659
6660 ret = tracing_wait_pipe(filp);
6661 if (ret <= 0)
6662 goto out_err;
6663
6664 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6665 ret = -EFAULT;
6666 goto out_err;
6667 }
6668
6669 trace_event_read_lock();
6670 trace_access_lock(iter->cpu_file);
6671
6672 /* Fill as many pages as possible. */
6673 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6674 spd.pages[i] = alloc_page(GFP_KERNEL);
6675 if (!spd.pages[i])
6676 break;
6677
6678 rem = tracing_fill_pipe_page(rem, iter);
6679
6680 /* Copy the data into the page, so we can start over. */
6681 ret = trace_seq_to_buffer(&iter->seq,
6682 page_address(spd.pages[i]),
6683 trace_seq_used(&iter->seq));
6684 if (ret < 0) {
6685 __free_page(spd.pages[i]);
6686 break;
6687 }
6688 spd.partial[i].offset = 0;
6689 spd.partial[i].len = trace_seq_used(&iter->seq);
6690
6691 trace_seq_init(&iter->seq);
6692 }
6693
6694 trace_access_unlock(iter->cpu_file);
6695 trace_event_read_unlock();
6696 mutex_unlock(&iter->mutex);
6697
6698 spd.nr_pages = i;
6699
6700 if (i)
6701 ret = splice_to_pipe(pipe, &spd);
6702 else
6703 ret = 0;
6704out:
6705 splice_shrink_spd(&spd);
6706 return ret;
6707
6708out_err:
6709 mutex_unlock(&iter->mutex);
6710 goto out;
6711}
6712
6713static ssize_t
6714tracing_entries_read(struct file *filp, char __user *ubuf,
6715 size_t cnt, loff_t *ppos)
6716{
6717 struct inode *inode = file_inode(filp);
6718 struct trace_array *tr = inode->i_private;
6719 int cpu = tracing_get_cpu(inode);
6720 char buf[64];
6721 int r = 0;
6722 ssize_t ret;
6723
6724 mutex_lock(&trace_types_lock);
6725
6726 if (cpu == RING_BUFFER_ALL_CPUS) {
6727 int cpu, buf_size_same;
6728 unsigned long size;
6729
6730 size = 0;
6731 buf_size_same = 1;
6732 /* check if all cpu sizes are same */
6733 for_each_tracing_cpu(cpu) {
6734 /* fill in the size from first enabled cpu */
6735 if (size == 0)
6736 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6737 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6738 buf_size_same = 0;
6739 break;
6740 }
6741 }
6742
6743 if (buf_size_same) {
6744 if (!tr->ring_buffer_expanded)
6745 r = sprintf(buf, "%lu (expanded: %lu)\n",
6746 size >> 10,
6747 trace_buf_size >> 10);
6748 else
6749 r = sprintf(buf, "%lu\n", size >> 10);
6750 } else
6751 r = sprintf(buf, "X\n");
6752 } else
6753 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6754
6755 mutex_unlock(&trace_types_lock);
6756
6757 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6758 return ret;
6759}
6760
6761static ssize_t
6762tracing_entries_write(struct file *filp, const char __user *ubuf,
6763 size_t cnt, loff_t *ppos)
6764{
6765 struct inode *inode = file_inode(filp);
6766 struct trace_array *tr = inode->i_private;
6767 unsigned long val;
6768 int ret;
6769
6770 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6771 if (ret)
6772 return ret;
6773
6774 /* must have at least 1 entry */
6775 if (!val)
6776 return -EINVAL;
6777
6778 /* value is in KB */
6779 val <<= 10;
6780 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6781 if (ret < 0)
6782 return ret;
6783
6784 *ppos += cnt;
6785
6786 return cnt;
6787}
6788
6789static ssize_t
6790tracing_total_entries_read(struct file *filp, char __user *ubuf,
6791 size_t cnt, loff_t *ppos)
6792{
6793 struct trace_array *tr = filp->private_data;
6794 char buf[64];
6795 int r, cpu;
6796 unsigned long size = 0, expanded_size = 0;
6797
6798 mutex_lock(&trace_types_lock);
6799 for_each_tracing_cpu(cpu) {
6800 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6801 if (!tr->ring_buffer_expanded)
6802 expanded_size += trace_buf_size >> 10;
6803 }
6804 if (tr->ring_buffer_expanded)
6805 r = sprintf(buf, "%lu\n", size);
6806 else
6807 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6808 mutex_unlock(&trace_types_lock);
6809
6810 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6811}
6812
6813static ssize_t
6814tracing_last_boot_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
6815{
6816 struct trace_array *tr = filp->private_data;
6817 struct seq_buf seq;
6818 char buf[64];
6819
6820 seq_buf_init(&seq, buf, 64);
6821
6822 seq_buf_printf(&seq, "text delta:\t%ld\n", tr->text_delta);
6823 seq_buf_printf(&seq, "data delta:\t%ld\n", tr->data_delta);
6824
6825 return simple_read_from_buffer(ubuf, cnt, ppos, buf, seq_buf_used(&seq));
6826}
6827
6828static int tracing_buffer_meta_open(struct inode *inode, struct file *filp)
6829{
6830 struct trace_array *tr = inode->i_private;
6831 int cpu = tracing_get_cpu(inode);
6832 int ret;
6833
6834 ret = tracing_check_open_get_tr(tr);
6835 if (ret)
6836 return ret;
6837
6838 ret = ring_buffer_meta_seq_init(filp, tr->array_buffer.buffer, cpu);
6839 if (ret < 0)
6840 __trace_array_put(tr);
6841 return ret;
6842}
6843
6844static ssize_t
6845tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
6846 size_t cnt, loff_t *ppos)
6847{
6848 /*
6849 * There is no need to read what the user has written, this function
6850 * is just to make sure that there is no error when "echo" is used
6851 */
6852
6853 *ppos += cnt;
6854
6855 return cnt;
6856}
6857
6858static int
6859tracing_free_buffer_release(struct inode *inode, struct file *filp)
6860{
6861 struct trace_array *tr = inode->i_private;
6862
6863 /* disable tracing ? */
6864 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
6865 tracer_tracing_off(tr);
6866 /* resize the ring buffer to 0 */
6867 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
6868
6869 trace_array_put(tr);
6870
6871 return 0;
6872}
6873
6874#define TRACE_MARKER_MAX_SIZE 4096
6875
6876static ssize_t
6877tracing_mark_write(struct file *filp, const char __user *ubuf,
6878 size_t cnt, loff_t *fpos)
6879{
6880 struct trace_array *tr = filp->private_data;
6881 struct ring_buffer_event *event;
6882 enum event_trigger_type tt = ETT_NONE;
6883 struct trace_buffer *buffer;
6884 struct print_entry *entry;
6885 int meta_size;
6886 ssize_t written;
6887 size_t size;
6888 int len;
6889
6890/* Used in tracing_mark_raw_write() as well */
6891#define FAULTED_STR "<faulted>"
6892#define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
6893
6894 if (tracing_disabled)
6895 return -EINVAL;
6896
6897 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6898 return -EINVAL;
6899
6900 if ((ssize_t)cnt < 0)
6901 return -EINVAL;
6902
6903 if (cnt > TRACE_MARKER_MAX_SIZE)
6904 cnt = TRACE_MARKER_MAX_SIZE;
6905
6906 meta_size = sizeof(*entry) + 2; /* add '\0' and possible '\n' */
6907 again:
6908 size = cnt + meta_size;
6909
6910 /* If less than "<faulted>", then make sure we can still add that */
6911 if (cnt < FAULTED_SIZE)
6912 size += FAULTED_SIZE - cnt;
6913
6914 buffer = tr->array_buffer.buffer;
6915 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
6916 tracing_gen_ctx());
6917 if (unlikely(!event)) {
6918 /*
6919 * If the size was greater than what was allowed, then
6920 * make it smaller and try again.
6921 */
6922 if (size > ring_buffer_max_event_size(buffer)) {
6923 /* cnt < FAULTED size should never be bigger than max */
6924 if (WARN_ON_ONCE(cnt < FAULTED_SIZE))
6925 return -EBADF;
6926 cnt = ring_buffer_max_event_size(buffer) - meta_size;
6927 /* The above should only happen once */
6928 if (WARN_ON_ONCE(cnt + meta_size == size))
6929 return -EBADF;
6930 goto again;
6931 }
6932
6933 /* Ring buffer disabled, return as if not open for write */
6934 return -EBADF;
6935 }
6936
6937 entry = ring_buffer_event_data(event);
6938 entry->ip = _THIS_IP_;
6939
6940 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
6941 if (len) {
6942 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6943 cnt = FAULTED_SIZE;
6944 written = -EFAULT;
6945 } else
6946 written = cnt;
6947
6948 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
6949 /* do not add \n before testing triggers, but add \0 */
6950 entry->buf[cnt] = '\0';
6951 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
6952 }
6953
6954 if (entry->buf[cnt - 1] != '\n') {
6955 entry->buf[cnt] = '\n';
6956 entry->buf[cnt + 1] = '\0';
6957 } else
6958 entry->buf[cnt] = '\0';
6959
6960 if (static_branch_unlikely(&trace_marker_exports_enabled))
6961 ftrace_exports(event, TRACE_EXPORT_MARKER);
6962 __buffer_unlock_commit(buffer, event);
6963
6964 if (tt)
6965 event_triggers_post_call(tr->trace_marker_file, tt);
6966
6967 return written;
6968}
6969
6970static ssize_t
6971tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
6972 size_t cnt, loff_t *fpos)
6973{
6974 struct trace_array *tr = filp->private_data;
6975 struct ring_buffer_event *event;
6976 struct trace_buffer *buffer;
6977 struct raw_data_entry *entry;
6978 ssize_t written;
6979 int size;
6980 int len;
6981
6982#define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
6983
6984 if (tracing_disabled)
6985 return -EINVAL;
6986
6987 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6988 return -EINVAL;
6989
6990 /* The marker must at least have a tag id */
6991 if (cnt < sizeof(unsigned int))
6992 return -EINVAL;
6993
6994 size = sizeof(*entry) + cnt;
6995 if (cnt < FAULT_SIZE_ID)
6996 size += FAULT_SIZE_ID - cnt;
6997
6998 buffer = tr->array_buffer.buffer;
6999
7000 if (size > ring_buffer_max_event_size(buffer))
7001 return -EINVAL;
7002
7003 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7004 tracing_gen_ctx());
7005 if (!event)
7006 /* Ring buffer disabled, return as if not open for write */
7007 return -EBADF;
7008
7009 entry = ring_buffer_event_data(event);
7010
7011 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7012 if (len) {
7013 entry->id = -1;
7014 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7015 written = -EFAULT;
7016 } else
7017 written = cnt;
7018
7019 __buffer_unlock_commit(buffer, event);
7020
7021 return written;
7022}
7023
7024static int tracing_clock_show(struct seq_file *m, void *v)
7025{
7026 struct trace_array *tr = m->private;
7027 int i;
7028
7029 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7030 seq_printf(m,
7031 "%s%s%s%s", i ? " " : "",
7032 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7033 i == tr->clock_id ? "]" : "");
7034 seq_putc(m, '\n');
7035
7036 return 0;
7037}
7038
7039int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7040{
7041 int i;
7042
7043 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7044 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7045 break;
7046 }
7047 if (i == ARRAY_SIZE(trace_clocks))
7048 return -EINVAL;
7049
7050 mutex_lock(&trace_types_lock);
7051
7052 tr->clock_id = i;
7053
7054 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7055
7056 /*
7057 * New clock may not be consistent with the previous clock.
7058 * Reset the buffer so that it doesn't have incomparable timestamps.
7059 */
7060 tracing_reset_online_cpus(&tr->array_buffer);
7061
7062#ifdef CONFIG_TRACER_MAX_TRACE
7063 if (tr->max_buffer.buffer)
7064 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7065 tracing_reset_online_cpus(&tr->max_buffer);
7066#endif
7067
7068 mutex_unlock(&trace_types_lock);
7069
7070 return 0;
7071}
7072
7073static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7074 size_t cnt, loff_t *fpos)
7075{
7076 struct seq_file *m = filp->private_data;
7077 struct trace_array *tr = m->private;
7078 char buf[64];
7079 const char *clockstr;
7080 int ret;
7081
7082 if (cnt >= sizeof(buf))
7083 return -EINVAL;
7084
7085 if (copy_from_user(buf, ubuf, cnt))
7086 return -EFAULT;
7087
7088 buf[cnt] = 0;
7089
7090 clockstr = strstrip(buf);
7091
7092 ret = tracing_set_clock(tr, clockstr);
7093 if (ret)
7094 return ret;
7095
7096 *fpos += cnt;
7097
7098 return cnt;
7099}
7100
7101static int tracing_clock_open(struct inode *inode, struct file *file)
7102{
7103 struct trace_array *tr = inode->i_private;
7104 int ret;
7105
7106 ret = tracing_check_open_get_tr(tr);
7107 if (ret)
7108 return ret;
7109
7110 ret = single_open(file, tracing_clock_show, inode->i_private);
7111 if (ret < 0)
7112 trace_array_put(tr);
7113
7114 return ret;
7115}
7116
7117static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7118{
7119 struct trace_array *tr = m->private;
7120
7121 mutex_lock(&trace_types_lock);
7122
7123 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7124 seq_puts(m, "delta [absolute]\n");
7125 else
7126 seq_puts(m, "[delta] absolute\n");
7127
7128 mutex_unlock(&trace_types_lock);
7129
7130 return 0;
7131}
7132
7133static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7134{
7135 struct trace_array *tr = inode->i_private;
7136 int ret;
7137
7138 ret = tracing_check_open_get_tr(tr);
7139 if (ret)
7140 return ret;
7141
7142 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7143 if (ret < 0)
7144 trace_array_put(tr);
7145
7146 return ret;
7147}
7148
7149u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7150{
7151 if (rbe == this_cpu_read(trace_buffered_event))
7152 return ring_buffer_time_stamp(buffer);
7153
7154 return ring_buffer_event_time_stamp(buffer, rbe);
7155}
7156
7157/*
7158 * Set or disable using the per CPU trace_buffer_event when possible.
7159 */
7160int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7161{
7162 guard(mutex)(&trace_types_lock);
7163
7164 if (set && tr->no_filter_buffering_ref++)
7165 return 0;
7166
7167 if (!set) {
7168 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref))
7169 return -EINVAL;
7170
7171 --tr->no_filter_buffering_ref;
7172 }
7173
7174 return 0;
7175}
7176
7177struct ftrace_buffer_info {
7178 struct trace_iterator iter;
7179 void *spare;
7180 unsigned int spare_cpu;
7181 unsigned int spare_size;
7182 unsigned int read;
7183};
7184
7185#ifdef CONFIG_TRACER_SNAPSHOT
7186static int tracing_snapshot_open(struct inode *inode, struct file *file)
7187{
7188 struct trace_array *tr = inode->i_private;
7189 struct trace_iterator *iter;
7190 struct seq_file *m;
7191 int ret;
7192
7193 ret = tracing_check_open_get_tr(tr);
7194 if (ret)
7195 return ret;
7196
7197 if (file->f_mode & FMODE_READ) {
7198 iter = __tracing_open(inode, file, true);
7199 if (IS_ERR(iter))
7200 ret = PTR_ERR(iter);
7201 } else {
7202 /* Writes still need the seq_file to hold the private data */
7203 ret = -ENOMEM;
7204 m = kzalloc(sizeof(*m), GFP_KERNEL);
7205 if (!m)
7206 goto out;
7207 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7208 if (!iter) {
7209 kfree(m);
7210 goto out;
7211 }
7212 ret = 0;
7213
7214 iter->tr = tr;
7215 iter->array_buffer = &tr->max_buffer;
7216 iter->cpu_file = tracing_get_cpu(inode);
7217 m->private = iter;
7218 file->private_data = m;
7219 }
7220out:
7221 if (ret < 0)
7222 trace_array_put(tr);
7223
7224 return ret;
7225}
7226
7227static void tracing_swap_cpu_buffer(void *tr)
7228{
7229 update_max_tr_single((struct trace_array *)tr, current, smp_processor_id());
7230}
7231
7232static ssize_t
7233tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7234 loff_t *ppos)
7235{
7236 struct seq_file *m = filp->private_data;
7237 struct trace_iterator *iter = m->private;
7238 struct trace_array *tr = iter->tr;
7239 unsigned long val;
7240 int ret;
7241
7242 ret = tracing_update_buffers(tr);
7243 if (ret < 0)
7244 return ret;
7245
7246 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7247 if (ret)
7248 return ret;
7249
7250 guard(mutex)(&trace_types_lock);
7251
7252 if (tr->current_trace->use_max_tr)
7253 return -EBUSY;
7254
7255 local_irq_disable();
7256 arch_spin_lock(&tr->max_lock);
7257 if (tr->cond_snapshot)
7258 ret = -EBUSY;
7259 arch_spin_unlock(&tr->max_lock);
7260 local_irq_enable();
7261 if (ret)
7262 return ret;
7263
7264 switch (val) {
7265 case 0:
7266 if (iter->cpu_file != RING_BUFFER_ALL_CPUS)
7267 return -EINVAL;
7268 if (tr->allocated_snapshot)
7269 free_snapshot(tr);
7270 break;
7271 case 1:
7272/* Only allow per-cpu swap if the ring buffer supports it */
7273#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7274 if (iter->cpu_file != RING_BUFFER_ALL_CPUS)
7275 return -EINVAL;
7276#endif
7277 if (tr->allocated_snapshot)
7278 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7279 &tr->array_buffer, iter->cpu_file);
7280
7281 ret = tracing_arm_snapshot_locked(tr);
7282 if (ret)
7283 return ret;
7284
7285 /* Now, we're going to swap */
7286 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
7287 local_irq_disable();
7288 update_max_tr(tr, current, smp_processor_id(), NULL);
7289 local_irq_enable();
7290 } else {
7291 smp_call_function_single(iter->cpu_file, tracing_swap_cpu_buffer,
7292 (void *)tr, 1);
7293 }
7294 tracing_disarm_snapshot(tr);
7295 break;
7296 default:
7297 if (tr->allocated_snapshot) {
7298 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7299 tracing_reset_online_cpus(&tr->max_buffer);
7300 else
7301 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7302 }
7303 break;
7304 }
7305
7306 if (ret >= 0) {
7307 *ppos += cnt;
7308 ret = cnt;
7309 }
7310
7311 return ret;
7312}
7313
7314static int tracing_snapshot_release(struct inode *inode, struct file *file)
7315{
7316 struct seq_file *m = file->private_data;
7317 int ret;
7318
7319 ret = tracing_release(inode, file);
7320
7321 if (file->f_mode & FMODE_READ)
7322 return ret;
7323
7324 /* If write only, the seq_file is just a stub */
7325 if (m)
7326 kfree(m->private);
7327 kfree(m);
7328
7329 return 0;
7330}
7331
7332static int tracing_buffers_open(struct inode *inode, struct file *filp);
7333static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7334 size_t count, loff_t *ppos);
7335static int tracing_buffers_release(struct inode *inode, struct file *file);
7336static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7337 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7338
7339static int snapshot_raw_open(struct inode *inode, struct file *filp)
7340{
7341 struct ftrace_buffer_info *info;
7342 int ret;
7343
7344 /* The following checks for tracefs lockdown */
7345 ret = tracing_buffers_open(inode, filp);
7346 if (ret < 0)
7347 return ret;
7348
7349 info = filp->private_data;
7350
7351 if (info->iter.trace->use_max_tr) {
7352 tracing_buffers_release(inode, filp);
7353 return -EBUSY;
7354 }
7355
7356 info->iter.snapshot = true;
7357 info->iter.array_buffer = &info->iter.tr->max_buffer;
7358
7359 return ret;
7360}
7361
7362#endif /* CONFIG_TRACER_SNAPSHOT */
7363
7364
7365static const struct file_operations tracing_thresh_fops = {
7366 .open = tracing_open_generic,
7367 .read = tracing_thresh_read,
7368 .write = tracing_thresh_write,
7369 .llseek = generic_file_llseek,
7370};
7371
7372#ifdef CONFIG_TRACER_MAX_TRACE
7373static const struct file_operations tracing_max_lat_fops = {
7374 .open = tracing_open_generic_tr,
7375 .read = tracing_max_lat_read,
7376 .write = tracing_max_lat_write,
7377 .llseek = generic_file_llseek,
7378 .release = tracing_release_generic_tr,
7379};
7380#endif
7381
7382static const struct file_operations set_tracer_fops = {
7383 .open = tracing_open_generic_tr,
7384 .read = tracing_set_trace_read,
7385 .write = tracing_set_trace_write,
7386 .llseek = generic_file_llseek,
7387 .release = tracing_release_generic_tr,
7388};
7389
7390static const struct file_operations tracing_pipe_fops = {
7391 .open = tracing_open_pipe,
7392 .poll = tracing_poll_pipe,
7393 .read = tracing_read_pipe,
7394 .splice_read = tracing_splice_read_pipe,
7395 .release = tracing_release_pipe,
7396};
7397
7398static const struct file_operations tracing_entries_fops = {
7399 .open = tracing_open_generic_tr,
7400 .read = tracing_entries_read,
7401 .write = tracing_entries_write,
7402 .llseek = generic_file_llseek,
7403 .release = tracing_release_generic_tr,
7404};
7405
7406static const struct file_operations tracing_buffer_meta_fops = {
7407 .open = tracing_buffer_meta_open,
7408 .read = seq_read,
7409 .llseek = seq_lseek,
7410 .release = tracing_seq_release,
7411};
7412
7413static const struct file_operations tracing_total_entries_fops = {
7414 .open = tracing_open_generic_tr,
7415 .read = tracing_total_entries_read,
7416 .llseek = generic_file_llseek,
7417 .release = tracing_release_generic_tr,
7418};
7419
7420static const struct file_operations tracing_free_buffer_fops = {
7421 .open = tracing_open_generic_tr,
7422 .write = tracing_free_buffer_write,
7423 .release = tracing_free_buffer_release,
7424};
7425
7426static const struct file_operations tracing_mark_fops = {
7427 .open = tracing_mark_open,
7428 .write = tracing_mark_write,
7429 .release = tracing_release_generic_tr,
7430};
7431
7432static const struct file_operations tracing_mark_raw_fops = {
7433 .open = tracing_mark_open,
7434 .write = tracing_mark_raw_write,
7435 .release = tracing_release_generic_tr,
7436};
7437
7438static const struct file_operations trace_clock_fops = {
7439 .open = tracing_clock_open,
7440 .read = seq_read,
7441 .llseek = seq_lseek,
7442 .release = tracing_single_release_tr,
7443 .write = tracing_clock_write,
7444};
7445
7446static const struct file_operations trace_time_stamp_mode_fops = {
7447 .open = tracing_time_stamp_mode_open,
7448 .read = seq_read,
7449 .llseek = seq_lseek,
7450 .release = tracing_single_release_tr,
7451};
7452
7453static const struct file_operations last_boot_fops = {
7454 .open = tracing_open_generic_tr,
7455 .read = tracing_last_boot_read,
7456 .llseek = generic_file_llseek,
7457 .release = tracing_release_generic_tr,
7458};
7459
7460#ifdef CONFIG_TRACER_SNAPSHOT
7461static const struct file_operations snapshot_fops = {
7462 .open = tracing_snapshot_open,
7463 .read = seq_read,
7464 .write = tracing_snapshot_write,
7465 .llseek = tracing_lseek,
7466 .release = tracing_snapshot_release,
7467};
7468
7469static const struct file_operations snapshot_raw_fops = {
7470 .open = snapshot_raw_open,
7471 .read = tracing_buffers_read,
7472 .release = tracing_buffers_release,
7473 .splice_read = tracing_buffers_splice_read,
7474};
7475
7476#endif /* CONFIG_TRACER_SNAPSHOT */
7477
7478/*
7479 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7480 * @filp: The active open file structure
7481 * @ubuf: The userspace provided buffer to read value into
7482 * @cnt: The maximum number of bytes to read
7483 * @ppos: The current "file" position
7484 *
7485 * This function implements the write interface for a struct trace_min_max_param.
7486 * The filp->private_data must point to a trace_min_max_param structure that
7487 * defines where to write the value, the min and the max acceptable values,
7488 * and a lock to protect the write.
7489 */
7490static ssize_t
7491trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7492{
7493 struct trace_min_max_param *param = filp->private_data;
7494 u64 val;
7495 int err;
7496
7497 if (!param)
7498 return -EFAULT;
7499
7500 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7501 if (err)
7502 return err;
7503
7504 if (param->lock)
7505 mutex_lock(param->lock);
7506
7507 if (param->min && val < *param->min)
7508 err = -EINVAL;
7509
7510 if (param->max && val > *param->max)
7511 err = -EINVAL;
7512
7513 if (!err)
7514 *param->val = val;
7515
7516 if (param->lock)
7517 mutex_unlock(param->lock);
7518
7519 if (err)
7520 return err;
7521
7522 return cnt;
7523}
7524
7525/*
7526 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7527 * @filp: The active open file structure
7528 * @ubuf: The userspace provided buffer to read value into
7529 * @cnt: The maximum number of bytes to read
7530 * @ppos: The current "file" position
7531 *
7532 * This function implements the read interface for a struct trace_min_max_param.
7533 * The filp->private_data must point to a trace_min_max_param struct with valid
7534 * data.
7535 */
7536static ssize_t
7537trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7538{
7539 struct trace_min_max_param *param = filp->private_data;
7540 char buf[U64_STR_SIZE];
7541 int len;
7542 u64 val;
7543
7544 if (!param)
7545 return -EFAULT;
7546
7547 val = *param->val;
7548
7549 if (cnt > sizeof(buf))
7550 cnt = sizeof(buf);
7551
7552 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7553
7554 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7555}
7556
7557const struct file_operations trace_min_max_fops = {
7558 .open = tracing_open_generic,
7559 .read = trace_min_max_read,
7560 .write = trace_min_max_write,
7561};
7562
7563#define TRACING_LOG_ERRS_MAX 8
7564#define TRACING_LOG_LOC_MAX 128
7565
7566#define CMD_PREFIX " Command: "
7567
7568struct err_info {
7569 const char **errs; /* ptr to loc-specific array of err strings */
7570 u8 type; /* index into errs -> specific err string */
7571 u16 pos; /* caret position */
7572 u64 ts;
7573};
7574
7575struct tracing_log_err {
7576 struct list_head list;
7577 struct err_info info;
7578 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7579 char *cmd; /* what caused err */
7580};
7581
7582static DEFINE_MUTEX(tracing_err_log_lock);
7583
7584static struct tracing_log_err *alloc_tracing_log_err(int len)
7585{
7586 struct tracing_log_err *err;
7587
7588 err = kzalloc(sizeof(*err), GFP_KERNEL);
7589 if (!err)
7590 return ERR_PTR(-ENOMEM);
7591
7592 err->cmd = kzalloc(len, GFP_KERNEL);
7593 if (!err->cmd) {
7594 kfree(err);
7595 return ERR_PTR(-ENOMEM);
7596 }
7597
7598 return err;
7599}
7600
7601static void free_tracing_log_err(struct tracing_log_err *err)
7602{
7603 kfree(err->cmd);
7604 kfree(err);
7605}
7606
7607static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7608 int len)
7609{
7610 struct tracing_log_err *err;
7611 char *cmd;
7612
7613 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7614 err = alloc_tracing_log_err(len);
7615 if (PTR_ERR(err) != -ENOMEM)
7616 tr->n_err_log_entries++;
7617
7618 return err;
7619 }
7620 cmd = kzalloc(len, GFP_KERNEL);
7621 if (!cmd)
7622 return ERR_PTR(-ENOMEM);
7623 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7624 kfree(err->cmd);
7625 err->cmd = cmd;
7626 list_del(&err->list);
7627
7628 return err;
7629}
7630
7631/**
7632 * err_pos - find the position of a string within a command for error careting
7633 * @cmd: The tracing command that caused the error
7634 * @str: The string to position the caret at within @cmd
7635 *
7636 * Finds the position of the first occurrence of @str within @cmd. The
7637 * return value can be passed to tracing_log_err() for caret placement
7638 * within @cmd.
7639 *
7640 * Returns the index within @cmd of the first occurrence of @str or 0
7641 * if @str was not found.
7642 */
7643unsigned int err_pos(char *cmd, const char *str)
7644{
7645 char *found;
7646
7647 if (WARN_ON(!strlen(cmd)))
7648 return 0;
7649
7650 found = strstr(cmd, str);
7651 if (found)
7652 return found - cmd;
7653
7654 return 0;
7655}
7656
7657/**
7658 * tracing_log_err - write an error to the tracing error log
7659 * @tr: The associated trace array for the error (NULL for top level array)
7660 * @loc: A string describing where the error occurred
7661 * @cmd: The tracing command that caused the error
7662 * @errs: The array of loc-specific static error strings
7663 * @type: The index into errs[], which produces the specific static err string
7664 * @pos: The position the caret should be placed in the cmd
7665 *
7666 * Writes an error into tracing/error_log of the form:
7667 *
7668 * <loc>: error: <text>
7669 * Command: <cmd>
7670 * ^
7671 *
7672 * tracing/error_log is a small log file containing the last
7673 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7674 * unless there has been a tracing error, and the error log can be
7675 * cleared and have its memory freed by writing the empty string in
7676 * truncation mode to it i.e. echo > tracing/error_log.
7677 *
7678 * NOTE: the @errs array along with the @type param are used to
7679 * produce a static error string - this string is not copied and saved
7680 * when the error is logged - only a pointer to it is saved. See
7681 * existing callers for examples of how static strings are typically
7682 * defined for use with tracing_log_err().
7683 */
7684void tracing_log_err(struct trace_array *tr,
7685 const char *loc, const char *cmd,
7686 const char **errs, u8 type, u16 pos)
7687{
7688 struct tracing_log_err *err;
7689 int len = 0;
7690
7691 if (!tr)
7692 tr = &global_trace;
7693
7694 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
7695
7696 guard(mutex)(&tracing_err_log_lock);
7697
7698 err = get_tracing_log_err(tr, len);
7699 if (PTR_ERR(err) == -ENOMEM)
7700 return;
7701
7702 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7703 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
7704
7705 err->info.errs = errs;
7706 err->info.type = type;
7707 err->info.pos = pos;
7708 err->info.ts = local_clock();
7709
7710 list_add_tail(&err->list, &tr->err_log);
7711}
7712
7713static void clear_tracing_err_log(struct trace_array *tr)
7714{
7715 struct tracing_log_err *err, *next;
7716
7717 mutex_lock(&tracing_err_log_lock);
7718 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7719 list_del(&err->list);
7720 free_tracing_log_err(err);
7721 }
7722
7723 tr->n_err_log_entries = 0;
7724 mutex_unlock(&tracing_err_log_lock);
7725}
7726
7727static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7728{
7729 struct trace_array *tr = m->private;
7730
7731 mutex_lock(&tracing_err_log_lock);
7732
7733 return seq_list_start(&tr->err_log, *pos);
7734}
7735
7736static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7737{
7738 struct trace_array *tr = m->private;
7739
7740 return seq_list_next(v, &tr->err_log, pos);
7741}
7742
7743static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7744{
7745 mutex_unlock(&tracing_err_log_lock);
7746}
7747
7748static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
7749{
7750 u16 i;
7751
7752 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7753 seq_putc(m, ' ');
7754 for (i = 0; i < pos; i++)
7755 seq_putc(m, ' ');
7756 seq_puts(m, "^\n");
7757}
7758
7759static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7760{
7761 struct tracing_log_err *err = v;
7762
7763 if (err) {
7764 const char *err_text = err->info.errs[err->info.type];
7765 u64 sec = err->info.ts;
7766 u32 nsec;
7767
7768 nsec = do_div(sec, NSEC_PER_SEC);
7769 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7770 err->loc, err_text);
7771 seq_printf(m, "%s", err->cmd);
7772 tracing_err_log_show_pos(m, err->info.pos);
7773 }
7774
7775 return 0;
7776}
7777
7778static const struct seq_operations tracing_err_log_seq_ops = {
7779 .start = tracing_err_log_seq_start,
7780 .next = tracing_err_log_seq_next,
7781 .stop = tracing_err_log_seq_stop,
7782 .show = tracing_err_log_seq_show
7783};
7784
7785static int tracing_err_log_open(struct inode *inode, struct file *file)
7786{
7787 struct trace_array *tr = inode->i_private;
7788 int ret = 0;
7789
7790 ret = tracing_check_open_get_tr(tr);
7791 if (ret)
7792 return ret;
7793
7794 /* If this file was opened for write, then erase contents */
7795 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7796 clear_tracing_err_log(tr);
7797
7798 if (file->f_mode & FMODE_READ) {
7799 ret = seq_open(file, &tracing_err_log_seq_ops);
7800 if (!ret) {
7801 struct seq_file *m = file->private_data;
7802 m->private = tr;
7803 } else {
7804 trace_array_put(tr);
7805 }
7806 }
7807 return ret;
7808}
7809
7810static ssize_t tracing_err_log_write(struct file *file,
7811 const char __user *buffer,
7812 size_t count, loff_t *ppos)
7813{
7814 return count;
7815}
7816
7817static int tracing_err_log_release(struct inode *inode, struct file *file)
7818{
7819 struct trace_array *tr = inode->i_private;
7820
7821 trace_array_put(tr);
7822
7823 if (file->f_mode & FMODE_READ)
7824 seq_release(inode, file);
7825
7826 return 0;
7827}
7828
7829static const struct file_operations tracing_err_log_fops = {
7830 .open = tracing_err_log_open,
7831 .write = tracing_err_log_write,
7832 .read = seq_read,
7833 .llseek = tracing_lseek,
7834 .release = tracing_err_log_release,
7835};
7836
7837static int tracing_buffers_open(struct inode *inode, struct file *filp)
7838{
7839 struct trace_array *tr = inode->i_private;
7840 struct ftrace_buffer_info *info;
7841 int ret;
7842
7843 ret = tracing_check_open_get_tr(tr);
7844 if (ret)
7845 return ret;
7846
7847 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7848 if (!info) {
7849 trace_array_put(tr);
7850 return -ENOMEM;
7851 }
7852
7853 mutex_lock(&trace_types_lock);
7854
7855 info->iter.tr = tr;
7856 info->iter.cpu_file = tracing_get_cpu(inode);
7857 info->iter.trace = tr->current_trace;
7858 info->iter.array_buffer = &tr->array_buffer;
7859 info->spare = NULL;
7860 /* Force reading ring buffer for first read */
7861 info->read = (unsigned int)-1;
7862
7863 filp->private_data = info;
7864
7865 tr->trace_ref++;
7866
7867 mutex_unlock(&trace_types_lock);
7868
7869 ret = nonseekable_open(inode, filp);
7870 if (ret < 0)
7871 trace_array_put(tr);
7872
7873 return ret;
7874}
7875
7876static __poll_t
7877tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7878{
7879 struct ftrace_buffer_info *info = filp->private_data;
7880 struct trace_iterator *iter = &info->iter;
7881
7882 return trace_poll(iter, filp, poll_table);
7883}
7884
7885static ssize_t
7886tracing_buffers_read(struct file *filp, char __user *ubuf,
7887 size_t count, loff_t *ppos)
7888{
7889 struct ftrace_buffer_info *info = filp->private_data;
7890 struct trace_iterator *iter = &info->iter;
7891 void *trace_data;
7892 int page_size;
7893 ssize_t ret = 0;
7894 ssize_t size;
7895
7896 if (!count)
7897 return 0;
7898
7899#ifdef CONFIG_TRACER_MAX_TRACE
7900 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7901 return -EBUSY;
7902#endif
7903
7904 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
7905
7906 /* Make sure the spare matches the current sub buffer size */
7907 if (info->spare) {
7908 if (page_size != info->spare_size) {
7909 ring_buffer_free_read_page(iter->array_buffer->buffer,
7910 info->spare_cpu, info->spare);
7911 info->spare = NULL;
7912 }
7913 }
7914
7915 if (!info->spare) {
7916 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
7917 iter->cpu_file);
7918 if (IS_ERR(info->spare)) {
7919 ret = PTR_ERR(info->spare);
7920 info->spare = NULL;
7921 } else {
7922 info->spare_cpu = iter->cpu_file;
7923 info->spare_size = page_size;
7924 }
7925 }
7926 if (!info->spare)
7927 return ret;
7928
7929 /* Do we have previous read data to read? */
7930 if (info->read < page_size)
7931 goto read;
7932
7933 again:
7934 trace_access_lock(iter->cpu_file);
7935 ret = ring_buffer_read_page(iter->array_buffer->buffer,
7936 info->spare,
7937 count,
7938 iter->cpu_file, 0);
7939 trace_access_unlock(iter->cpu_file);
7940
7941 if (ret < 0) {
7942 if (trace_empty(iter) && !iter->closed) {
7943 if ((filp->f_flags & O_NONBLOCK))
7944 return -EAGAIN;
7945
7946 ret = wait_on_pipe(iter, 0);
7947 if (ret)
7948 return ret;
7949
7950 goto again;
7951 }
7952 return 0;
7953 }
7954
7955 info->read = 0;
7956 read:
7957 size = page_size - info->read;
7958 if (size > count)
7959 size = count;
7960 trace_data = ring_buffer_read_page_data(info->spare);
7961 ret = copy_to_user(ubuf, trace_data + info->read, size);
7962 if (ret == size)
7963 return -EFAULT;
7964
7965 size -= ret;
7966
7967 *ppos += size;
7968 info->read += size;
7969
7970 return size;
7971}
7972
7973static int tracing_buffers_flush(struct file *file, fl_owner_t id)
7974{
7975 struct ftrace_buffer_info *info = file->private_data;
7976 struct trace_iterator *iter = &info->iter;
7977
7978 iter->closed = true;
7979 /* Make sure the waiters see the new wait_index */
7980 (void)atomic_fetch_inc_release(&iter->wait_index);
7981
7982 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
7983
7984 return 0;
7985}
7986
7987static int tracing_buffers_release(struct inode *inode, struct file *file)
7988{
7989 struct ftrace_buffer_info *info = file->private_data;
7990 struct trace_iterator *iter = &info->iter;
7991
7992 mutex_lock(&trace_types_lock);
7993
7994 iter->tr->trace_ref--;
7995
7996 __trace_array_put(iter->tr);
7997
7998 if (info->spare)
7999 ring_buffer_free_read_page(iter->array_buffer->buffer,
8000 info->spare_cpu, info->spare);
8001 kvfree(info);
8002
8003 mutex_unlock(&trace_types_lock);
8004
8005 return 0;
8006}
8007
8008struct buffer_ref {
8009 struct trace_buffer *buffer;
8010 void *page;
8011 int cpu;
8012 refcount_t refcount;
8013};
8014
8015static void buffer_ref_release(struct buffer_ref *ref)
8016{
8017 if (!refcount_dec_and_test(&ref->refcount))
8018 return;
8019 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8020 kfree(ref);
8021}
8022
8023static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8024 struct pipe_buffer *buf)
8025{
8026 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8027
8028 buffer_ref_release(ref);
8029 buf->private = 0;
8030}
8031
8032static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8033 struct pipe_buffer *buf)
8034{
8035 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8036
8037 if (refcount_read(&ref->refcount) > INT_MAX/2)
8038 return false;
8039
8040 refcount_inc(&ref->refcount);
8041 return true;
8042}
8043
8044/* Pipe buffer operations for a buffer. */
8045static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8046 .release = buffer_pipe_buf_release,
8047 .get = buffer_pipe_buf_get,
8048};
8049
8050/*
8051 * Callback from splice_to_pipe(), if we need to release some pages
8052 * at the end of the spd in case we error'ed out in filling the pipe.
8053 */
8054static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8055{
8056 struct buffer_ref *ref =
8057 (struct buffer_ref *)spd->partial[i].private;
8058
8059 buffer_ref_release(ref);
8060 spd->partial[i].private = 0;
8061}
8062
8063static ssize_t
8064tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8065 struct pipe_inode_info *pipe, size_t len,
8066 unsigned int flags)
8067{
8068 struct ftrace_buffer_info *info = file->private_data;
8069 struct trace_iterator *iter = &info->iter;
8070 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8071 struct page *pages_def[PIPE_DEF_BUFFERS];
8072 struct splice_pipe_desc spd = {
8073 .pages = pages_def,
8074 .partial = partial_def,
8075 .nr_pages_max = PIPE_DEF_BUFFERS,
8076 .ops = &buffer_pipe_buf_ops,
8077 .spd_release = buffer_spd_release,
8078 };
8079 struct buffer_ref *ref;
8080 bool woken = false;
8081 int page_size;
8082 int entries, i;
8083 ssize_t ret = 0;
8084
8085#ifdef CONFIG_TRACER_MAX_TRACE
8086 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8087 return -EBUSY;
8088#endif
8089
8090 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8091 if (*ppos & (page_size - 1))
8092 return -EINVAL;
8093
8094 if (len & (page_size - 1)) {
8095 if (len < page_size)
8096 return -EINVAL;
8097 len &= (~(page_size - 1));
8098 }
8099
8100 if (splice_grow_spd(pipe, &spd))
8101 return -ENOMEM;
8102
8103 again:
8104 trace_access_lock(iter->cpu_file);
8105 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8106
8107 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= page_size) {
8108 struct page *page;
8109 int r;
8110
8111 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8112 if (!ref) {
8113 ret = -ENOMEM;
8114 break;
8115 }
8116
8117 refcount_set(&ref->refcount, 1);
8118 ref->buffer = iter->array_buffer->buffer;
8119 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8120 if (IS_ERR(ref->page)) {
8121 ret = PTR_ERR(ref->page);
8122 ref->page = NULL;
8123 kfree(ref);
8124 break;
8125 }
8126 ref->cpu = iter->cpu_file;
8127
8128 r = ring_buffer_read_page(ref->buffer, ref->page,
8129 len, iter->cpu_file, 1);
8130 if (r < 0) {
8131 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8132 ref->page);
8133 kfree(ref);
8134 break;
8135 }
8136
8137 page = virt_to_page(ring_buffer_read_page_data(ref->page));
8138
8139 spd.pages[i] = page;
8140 spd.partial[i].len = page_size;
8141 spd.partial[i].offset = 0;
8142 spd.partial[i].private = (unsigned long)ref;
8143 spd.nr_pages++;
8144 *ppos += page_size;
8145
8146 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8147 }
8148
8149 trace_access_unlock(iter->cpu_file);
8150 spd.nr_pages = i;
8151
8152 /* did we read anything? */
8153 if (!spd.nr_pages) {
8154
8155 if (ret)
8156 goto out;
8157
8158 if (woken)
8159 goto out;
8160
8161 ret = -EAGAIN;
8162 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8163 goto out;
8164
8165 ret = wait_on_pipe(iter, iter->snapshot ? 0 : iter->tr->buffer_percent);
8166 if (ret)
8167 goto out;
8168
8169 /* No need to wait after waking up when tracing is off */
8170 if (!tracer_tracing_is_on(iter->tr))
8171 goto out;
8172
8173 /* Iterate one more time to collect any new data then exit */
8174 woken = true;
8175
8176 goto again;
8177 }
8178
8179 ret = splice_to_pipe(pipe, &spd);
8180out:
8181 splice_shrink_spd(&spd);
8182
8183 return ret;
8184}
8185
8186static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8187{
8188 struct ftrace_buffer_info *info = file->private_data;
8189 struct trace_iterator *iter = &info->iter;
8190 int err;
8191
8192 if (cmd == TRACE_MMAP_IOCTL_GET_READER) {
8193 if (!(file->f_flags & O_NONBLOCK)) {
8194 err = ring_buffer_wait(iter->array_buffer->buffer,
8195 iter->cpu_file,
8196 iter->tr->buffer_percent,
8197 NULL, NULL);
8198 if (err)
8199 return err;
8200 }
8201
8202 return ring_buffer_map_get_reader(iter->array_buffer->buffer,
8203 iter->cpu_file);
8204 } else if (cmd) {
8205 return -ENOTTY;
8206 }
8207
8208 /*
8209 * An ioctl call with cmd 0 to the ring buffer file will wake up all
8210 * waiters
8211 */
8212 mutex_lock(&trace_types_lock);
8213
8214 /* Make sure the waiters see the new wait_index */
8215 (void)atomic_fetch_inc_release(&iter->wait_index);
8216
8217 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8218
8219 mutex_unlock(&trace_types_lock);
8220 return 0;
8221}
8222
8223#ifdef CONFIG_TRACER_MAX_TRACE
8224static int get_snapshot_map(struct trace_array *tr)
8225{
8226 int err = 0;
8227
8228 /*
8229 * Called with mmap_lock held. lockdep would be unhappy if we would now
8230 * take trace_types_lock. Instead use the specific
8231 * snapshot_trigger_lock.
8232 */
8233 spin_lock(&tr->snapshot_trigger_lock);
8234
8235 if (tr->snapshot || tr->mapped == UINT_MAX)
8236 err = -EBUSY;
8237 else
8238 tr->mapped++;
8239
8240 spin_unlock(&tr->snapshot_trigger_lock);
8241
8242 /* Wait for update_max_tr() to observe iter->tr->mapped */
8243 if (tr->mapped == 1)
8244 synchronize_rcu();
8245
8246 return err;
8247
8248}
8249static void put_snapshot_map(struct trace_array *tr)
8250{
8251 spin_lock(&tr->snapshot_trigger_lock);
8252 if (!WARN_ON(!tr->mapped))
8253 tr->mapped--;
8254 spin_unlock(&tr->snapshot_trigger_lock);
8255}
8256#else
8257static inline int get_snapshot_map(struct trace_array *tr) { return 0; }
8258static inline void put_snapshot_map(struct trace_array *tr) { }
8259#endif
8260
8261static void tracing_buffers_mmap_close(struct vm_area_struct *vma)
8262{
8263 struct ftrace_buffer_info *info = vma->vm_file->private_data;
8264 struct trace_iterator *iter = &info->iter;
8265
8266 WARN_ON(ring_buffer_unmap(iter->array_buffer->buffer, iter->cpu_file));
8267 put_snapshot_map(iter->tr);
8268}
8269
8270static const struct vm_operations_struct tracing_buffers_vmops = {
8271 .close = tracing_buffers_mmap_close,
8272};
8273
8274static int tracing_buffers_mmap(struct file *filp, struct vm_area_struct *vma)
8275{
8276 struct ftrace_buffer_info *info = filp->private_data;
8277 struct trace_iterator *iter = &info->iter;
8278 int ret = 0;
8279
8280 /* Currently the boot mapped buffer is not supported for mmap */
8281 if (iter->tr->flags & TRACE_ARRAY_FL_BOOT)
8282 return -ENODEV;
8283
8284 ret = get_snapshot_map(iter->tr);
8285 if (ret)
8286 return ret;
8287
8288 ret = ring_buffer_map(iter->array_buffer->buffer, iter->cpu_file, vma);
8289 if (ret)
8290 put_snapshot_map(iter->tr);
8291
8292 vma->vm_ops = &tracing_buffers_vmops;
8293
8294 return ret;
8295}
8296
8297static const struct file_operations tracing_buffers_fops = {
8298 .open = tracing_buffers_open,
8299 .read = tracing_buffers_read,
8300 .poll = tracing_buffers_poll,
8301 .release = tracing_buffers_release,
8302 .flush = tracing_buffers_flush,
8303 .splice_read = tracing_buffers_splice_read,
8304 .unlocked_ioctl = tracing_buffers_ioctl,
8305 .mmap = tracing_buffers_mmap,
8306};
8307
8308static ssize_t
8309tracing_stats_read(struct file *filp, char __user *ubuf,
8310 size_t count, loff_t *ppos)
8311{
8312 struct inode *inode = file_inode(filp);
8313 struct trace_array *tr = inode->i_private;
8314 struct array_buffer *trace_buf = &tr->array_buffer;
8315 int cpu = tracing_get_cpu(inode);
8316 struct trace_seq *s;
8317 unsigned long cnt;
8318 unsigned long long t;
8319 unsigned long usec_rem;
8320
8321 s = kmalloc(sizeof(*s), GFP_KERNEL);
8322 if (!s)
8323 return -ENOMEM;
8324
8325 trace_seq_init(s);
8326
8327 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8328 trace_seq_printf(s, "entries: %ld\n", cnt);
8329
8330 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8331 trace_seq_printf(s, "overrun: %ld\n", cnt);
8332
8333 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8334 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8335
8336 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8337 trace_seq_printf(s, "bytes: %ld\n", cnt);
8338
8339 if (trace_clocks[tr->clock_id].in_ns) {
8340 /* local or global for trace_clock */
8341 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8342 usec_rem = do_div(t, USEC_PER_SEC);
8343 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8344 t, usec_rem);
8345
8346 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8347 usec_rem = do_div(t, USEC_PER_SEC);
8348 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8349 } else {
8350 /* counter or tsc mode for trace_clock */
8351 trace_seq_printf(s, "oldest event ts: %llu\n",
8352 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8353
8354 trace_seq_printf(s, "now ts: %llu\n",
8355 ring_buffer_time_stamp(trace_buf->buffer));
8356 }
8357
8358 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8359 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8360
8361 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8362 trace_seq_printf(s, "read events: %ld\n", cnt);
8363
8364 count = simple_read_from_buffer(ubuf, count, ppos,
8365 s->buffer, trace_seq_used(s));
8366
8367 kfree(s);
8368
8369 return count;
8370}
8371
8372static const struct file_operations tracing_stats_fops = {
8373 .open = tracing_open_generic_tr,
8374 .read = tracing_stats_read,
8375 .llseek = generic_file_llseek,
8376 .release = tracing_release_generic_tr,
8377};
8378
8379#ifdef CONFIG_DYNAMIC_FTRACE
8380
8381static ssize_t
8382tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8383 size_t cnt, loff_t *ppos)
8384{
8385 ssize_t ret;
8386 char *buf;
8387 int r;
8388
8389 /* 512 should be plenty to hold the amount needed */
8390#define DYN_INFO_BUF_SIZE 512
8391
8392 buf = kmalloc(DYN_INFO_BUF_SIZE, GFP_KERNEL);
8393 if (!buf)
8394 return -ENOMEM;
8395
8396 r = scnprintf(buf, DYN_INFO_BUF_SIZE,
8397 "%ld pages:%ld groups: %ld\n"
8398 "ftrace boot update time = %llu (ns)\n"
8399 "ftrace module total update time = %llu (ns)\n",
8400 ftrace_update_tot_cnt,
8401 ftrace_number_of_pages,
8402 ftrace_number_of_groups,
8403 ftrace_update_time,
8404 ftrace_total_mod_time);
8405
8406 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8407 kfree(buf);
8408 return ret;
8409}
8410
8411static const struct file_operations tracing_dyn_info_fops = {
8412 .open = tracing_open_generic,
8413 .read = tracing_read_dyn_info,
8414 .llseek = generic_file_llseek,
8415};
8416#endif /* CONFIG_DYNAMIC_FTRACE */
8417
8418#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8419static void
8420ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8421 struct trace_array *tr, struct ftrace_probe_ops *ops,
8422 void *data)
8423{
8424 tracing_snapshot_instance(tr);
8425}
8426
8427static void
8428ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8429 struct trace_array *tr, struct ftrace_probe_ops *ops,
8430 void *data)
8431{
8432 struct ftrace_func_mapper *mapper = data;
8433 long *count = NULL;
8434
8435 if (mapper)
8436 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8437
8438 if (count) {
8439
8440 if (*count <= 0)
8441 return;
8442
8443 (*count)--;
8444 }
8445
8446 tracing_snapshot_instance(tr);
8447}
8448
8449static int
8450ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8451 struct ftrace_probe_ops *ops, void *data)
8452{
8453 struct ftrace_func_mapper *mapper = data;
8454 long *count = NULL;
8455
8456 seq_printf(m, "%ps:", (void *)ip);
8457
8458 seq_puts(m, "snapshot");
8459
8460 if (mapper)
8461 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8462
8463 if (count)
8464 seq_printf(m, ":count=%ld\n", *count);
8465 else
8466 seq_puts(m, ":unlimited\n");
8467
8468 return 0;
8469}
8470
8471static int
8472ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8473 unsigned long ip, void *init_data, void **data)
8474{
8475 struct ftrace_func_mapper *mapper = *data;
8476
8477 if (!mapper) {
8478 mapper = allocate_ftrace_func_mapper();
8479 if (!mapper)
8480 return -ENOMEM;
8481 *data = mapper;
8482 }
8483
8484 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8485}
8486
8487static void
8488ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8489 unsigned long ip, void *data)
8490{
8491 struct ftrace_func_mapper *mapper = data;
8492
8493 if (!ip) {
8494 if (!mapper)
8495 return;
8496 free_ftrace_func_mapper(mapper, NULL);
8497 return;
8498 }
8499
8500 ftrace_func_mapper_remove_ip(mapper, ip);
8501}
8502
8503static struct ftrace_probe_ops snapshot_probe_ops = {
8504 .func = ftrace_snapshot,
8505 .print = ftrace_snapshot_print,
8506};
8507
8508static struct ftrace_probe_ops snapshot_count_probe_ops = {
8509 .func = ftrace_count_snapshot,
8510 .print = ftrace_snapshot_print,
8511 .init = ftrace_snapshot_init,
8512 .free = ftrace_snapshot_free,
8513};
8514
8515static int
8516ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8517 char *glob, char *cmd, char *param, int enable)
8518{
8519 struct ftrace_probe_ops *ops;
8520 void *count = (void *)-1;
8521 char *number;
8522 int ret;
8523
8524 if (!tr)
8525 return -ENODEV;
8526
8527 /* hash funcs only work with set_ftrace_filter */
8528 if (!enable)
8529 return -EINVAL;
8530
8531 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8532
8533 if (glob[0] == '!') {
8534 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
8535 if (!ret)
8536 tracing_disarm_snapshot(tr);
8537
8538 return ret;
8539 }
8540
8541 if (!param)
8542 goto out_reg;
8543
8544 number = strsep(¶m, ":");
8545
8546 if (!strlen(number))
8547 goto out_reg;
8548
8549 /*
8550 * We use the callback data field (which is a pointer)
8551 * as our counter.
8552 */
8553 ret = kstrtoul(number, 0, (unsigned long *)&count);
8554 if (ret)
8555 return ret;
8556
8557 out_reg:
8558 ret = tracing_arm_snapshot(tr);
8559 if (ret < 0)
8560 goto out;
8561
8562 ret = register_ftrace_function_probe(glob, tr, ops, count);
8563 if (ret < 0)
8564 tracing_disarm_snapshot(tr);
8565 out:
8566 return ret < 0 ? ret : 0;
8567}
8568
8569static struct ftrace_func_command ftrace_snapshot_cmd = {
8570 .name = "snapshot",
8571 .func = ftrace_trace_snapshot_callback,
8572};
8573
8574static __init int register_snapshot_cmd(void)
8575{
8576 return register_ftrace_command(&ftrace_snapshot_cmd);
8577}
8578#else
8579static inline __init int register_snapshot_cmd(void) { return 0; }
8580#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8581
8582static struct dentry *tracing_get_dentry(struct trace_array *tr)
8583{
8584 if (WARN_ON(!tr->dir))
8585 return ERR_PTR(-ENODEV);
8586
8587 /* Top directory uses NULL as the parent */
8588 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8589 return NULL;
8590
8591 /* All sub buffers have a descriptor */
8592 return tr->dir;
8593}
8594
8595static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8596{
8597 struct dentry *d_tracer;
8598
8599 if (tr->percpu_dir)
8600 return tr->percpu_dir;
8601
8602 d_tracer = tracing_get_dentry(tr);
8603 if (IS_ERR(d_tracer))
8604 return NULL;
8605
8606 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8607
8608 MEM_FAIL(!tr->percpu_dir,
8609 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8610
8611 return tr->percpu_dir;
8612}
8613
8614static struct dentry *
8615trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8616 void *data, long cpu, const struct file_operations *fops)
8617{
8618 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8619
8620 if (ret) /* See tracing_get_cpu() */
8621 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8622 return ret;
8623}
8624
8625static void
8626tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8627{
8628 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8629 struct dentry *d_cpu;
8630 char cpu_dir[30]; /* 30 characters should be more than enough */
8631
8632 if (!d_percpu)
8633 return;
8634
8635 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8636 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8637 if (!d_cpu) {
8638 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8639 return;
8640 }
8641
8642 /* per cpu trace_pipe */
8643 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8644 tr, cpu, &tracing_pipe_fops);
8645
8646 /* per cpu trace */
8647 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8648 tr, cpu, &tracing_fops);
8649
8650 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8651 tr, cpu, &tracing_buffers_fops);
8652
8653 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8654 tr, cpu, &tracing_stats_fops);
8655
8656 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8657 tr, cpu, &tracing_entries_fops);
8658
8659 if (tr->range_addr_start)
8660 trace_create_cpu_file("buffer_meta", TRACE_MODE_READ, d_cpu,
8661 tr, cpu, &tracing_buffer_meta_fops);
8662#ifdef CONFIG_TRACER_SNAPSHOT
8663 if (!tr->range_addr_start) {
8664 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8665 tr, cpu, &snapshot_fops);
8666
8667 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8668 tr, cpu, &snapshot_raw_fops);
8669 }
8670#endif
8671}
8672
8673#ifdef CONFIG_FTRACE_SELFTEST
8674/* Let selftest have access to static functions in this file */
8675#include "trace_selftest.c"
8676#endif
8677
8678static ssize_t
8679trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8680 loff_t *ppos)
8681{
8682 struct trace_option_dentry *topt = filp->private_data;
8683 char *buf;
8684
8685 if (topt->flags->val & topt->opt->bit)
8686 buf = "1\n";
8687 else
8688 buf = "0\n";
8689
8690 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8691}
8692
8693static ssize_t
8694trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8695 loff_t *ppos)
8696{
8697 struct trace_option_dentry *topt = filp->private_data;
8698 unsigned long val;
8699 int ret;
8700
8701 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8702 if (ret)
8703 return ret;
8704
8705 if (val != 0 && val != 1)
8706 return -EINVAL;
8707
8708 if (!!(topt->flags->val & topt->opt->bit) != val) {
8709 mutex_lock(&trace_types_lock);
8710 ret = __set_tracer_option(topt->tr, topt->flags,
8711 topt->opt, !val);
8712 mutex_unlock(&trace_types_lock);
8713 if (ret)
8714 return ret;
8715 }
8716
8717 *ppos += cnt;
8718
8719 return cnt;
8720}
8721
8722static int tracing_open_options(struct inode *inode, struct file *filp)
8723{
8724 struct trace_option_dentry *topt = inode->i_private;
8725 int ret;
8726
8727 ret = tracing_check_open_get_tr(topt->tr);
8728 if (ret)
8729 return ret;
8730
8731 filp->private_data = inode->i_private;
8732 return 0;
8733}
8734
8735static int tracing_release_options(struct inode *inode, struct file *file)
8736{
8737 struct trace_option_dentry *topt = file->private_data;
8738
8739 trace_array_put(topt->tr);
8740 return 0;
8741}
8742
8743static const struct file_operations trace_options_fops = {
8744 .open = tracing_open_options,
8745 .read = trace_options_read,
8746 .write = trace_options_write,
8747 .llseek = generic_file_llseek,
8748 .release = tracing_release_options,
8749};
8750
8751/*
8752 * In order to pass in both the trace_array descriptor as well as the index
8753 * to the flag that the trace option file represents, the trace_array
8754 * has a character array of trace_flags_index[], which holds the index
8755 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8756 * The address of this character array is passed to the flag option file
8757 * read/write callbacks.
8758 *
8759 * In order to extract both the index and the trace_array descriptor,
8760 * get_tr_index() uses the following algorithm.
8761 *
8762 * idx = *ptr;
8763 *
8764 * As the pointer itself contains the address of the index (remember
8765 * index[1] == 1).
8766 *
8767 * Then to get the trace_array descriptor, by subtracting that index
8768 * from the ptr, we get to the start of the index itself.
8769 *
8770 * ptr - idx == &index[0]
8771 *
8772 * Then a simple container_of() from that pointer gets us to the
8773 * trace_array descriptor.
8774 */
8775static void get_tr_index(void *data, struct trace_array **ptr,
8776 unsigned int *pindex)
8777{
8778 *pindex = *(unsigned char *)data;
8779
8780 *ptr = container_of(data - *pindex, struct trace_array,
8781 trace_flags_index);
8782}
8783
8784static ssize_t
8785trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8786 loff_t *ppos)
8787{
8788 void *tr_index = filp->private_data;
8789 struct trace_array *tr;
8790 unsigned int index;
8791 char *buf;
8792
8793 get_tr_index(tr_index, &tr, &index);
8794
8795 if (tr->trace_flags & (1 << index))
8796 buf = "1\n";
8797 else
8798 buf = "0\n";
8799
8800 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8801}
8802
8803static ssize_t
8804trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8805 loff_t *ppos)
8806{
8807 void *tr_index = filp->private_data;
8808 struct trace_array *tr;
8809 unsigned int index;
8810 unsigned long val;
8811 int ret;
8812
8813 get_tr_index(tr_index, &tr, &index);
8814
8815 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8816 if (ret)
8817 return ret;
8818
8819 if (val != 0 && val != 1)
8820 return -EINVAL;
8821
8822 mutex_lock(&event_mutex);
8823 mutex_lock(&trace_types_lock);
8824 ret = set_tracer_flag(tr, 1 << index, val);
8825 mutex_unlock(&trace_types_lock);
8826 mutex_unlock(&event_mutex);
8827
8828 if (ret < 0)
8829 return ret;
8830
8831 *ppos += cnt;
8832
8833 return cnt;
8834}
8835
8836static const struct file_operations trace_options_core_fops = {
8837 .open = tracing_open_generic,
8838 .read = trace_options_core_read,
8839 .write = trace_options_core_write,
8840 .llseek = generic_file_llseek,
8841};
8842
8843struct dentry *trace_create_file(const char *name,
8844 umode_t mode,
8845 struct dentry *parent,
8846 void *data,
8847 const struct file_operations *fops)
8848{
8849 struct dentry *ret;
8850
8851 ret = tracefs_create_file(name, mode, parent, data, fops);
8852 if (!ret)
8853 pr_warn("Could not create tracefs '%s' entry\n", name);
8854
8855 return ret;
8856}
8857
8858
8859static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8860{
8861 struct dentry *d_tracer;
8862
8863 if (tr->options)
8864 return tr->options;
8865
8866 d_tracer = tracing_get_dentry(tr);
8867 if (IS_ERR(d_tracer))
8868 return NULL;
8869
8870 tr->options = tracefs_create_dir("options", d_tracer);
8871 if (!tr->options) {
8872 pr_warn("Could not create tracefs directory 'options'\n");
8873 return NULL;
8874 }
8875
8876 return tr->options;
8877}
8878
8879static void
8880create_trace_option_file(struct trace_array *tr,
8881 struct trace_option_dentry *topt,
8882 struct tracer_flags *flags,
8883 struct tracer_opt *opt)
8884{
8885 struct dentry *t_options;
8886
8887 t_options = trace_options_init_dentry(tr);
8888 if (!t_options)
8889 return;
8890
8891 topt->flags = flags;
8892 topt->opt = opt;
8893 topt->tr = tr;
8894
8895 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
8896 t_options, topt, &trace_options_fops);
8897
8898}
8899
8900static void
8901create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8902{
8903 struct trace_option_dentry *topts;
8904 struct trace_options *tr_topts;
8905 struct tracer_flags *flags;
8906 struct tracer_opt *opts;
8907 int cnt;
8908 int i;
8909
8910 if (!tracer)
8911 return;
8912
8913 flags = tracer->flags;
8914
8915 if (!flags || !flags->opts)
8916 return;
8917
8918 /*
8919 * If this is an instance, only create flags for tracers
8920 * the instance may have.
8921 */
8922 if (!trace_ok_for_array(tracer, tr))
8923 return;
8924
8925 for (i = 0; i < tr->nr_topts; i++) {
8926 /* Make sure there's no duplicate flags. */
8927 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8928 return;
8929 }
8930
8931 opts = flags->opts;
8932
8933 for (cnt = 0; opts[cnt].name; cnt++)
8934 ;
8935
8936 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8937 if (!topts)
8938 return;
8939
8940 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8941 GFP_KERNEL);
8942 if (!tr_topts) {
8943 kfree(topts);
8944 return;
8945 }
8946
8947 tr->topts = tr_topts;
8948 tr->topts[tr->nr_topts].tracer = tracer;
8949 tr->topts[tr->nr_topts].topts = topts;
8950 tr->nr_topts++;
8951
8952 for (cnt = 0; opts[cnt].name; cnt++) {
8953 create_trace_option_file(tr, &topts[cnt], flags,
8954 &opts[cnt]);
8955 MEM_FAIL(topts[cnt].entry == NULL,
8956 "Failed to create trace option: %s",
8957 opts[cnt].name);
8958 }
8959}
8960
8961static struct dentry *
8962create_trace_option_core_file(struct trace_array *tr,
8963 const char *option, long index)
8964{
8965 struct dentry *t_options;
8966
8967 t_options = trace_options_init_dentry(tr);
8968 if (!t_options)
8969 return NULL;
8970
8971 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
8972 (void *)&tr->trace_flags_index[index],
8973 &trace_options_core_fops);
8974}
8975
8976static void create_trace_options_dir(struct trace_array *tr)
8977{
8978 struct dentry *t_options;
8979 bool top_level = tr == &global_trace;
8980 int i;
8981
8982 t_options = trace_options_init_dentry(tr);
8983 if (!t_options)
8984 return;
8985
8986 for (i = 0; trace_options[i]; i++) {
8987 if (top_level ||
8988 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8989 create_trace_option_core_file(tr, trace_options[i], i);
8990 }
8991}
8992
8993static ssize_t
8994rb_simple_read(struct file *filp, char __user *ubuf,
8995 size_t cnt, loff_t *ppos)
8996{
8997 struct trace_array *tr = filp->private_data;
8998 char buf[64];
8999 int r;
9000
9001 r = tracer_tracing_is_on(tr);
9002 r = sprintf(buf, "%d\n", r);
9003
9004 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9005}
9006
9007static ssize_t
9008rb_simple_write(struct file *filp, const char __user *ubuf,
9009 size_t cnt, loff_t *ppos)
9010{
9011 struct trace_array *tr = filp->private_data;
9012 struct trace_buffer *buffer = tr->array_buffer.buffer;
9013 unsigned long val;
9014 int ret;
9015
9016 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9017 if (ret)
9018 return ret;
9019
9020 if (buffer) {
9021 mutex_lock(&trace_types_lock);
9022 if (!!val == tracer_tracing_is_on(tr)) {
9023 val = 0; /* do nothing */
9024 } else if (val) {
9025 tracer_tracing_on(tr);
9026 if (tr->current_trace->start)
9027 tr->current_trace->start(tr);
9028 } else {
9029 tracer_tracing_off(tr);
9030 if (tr->current_trace->stop)
9031 tr->current_trace->stop(tr);
9032 /* Wake up any waiters */
9033 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9034 }
9035 mutex_unlock(&trace_types_lock);
9036 }
9037
9038 (*ppos)++;
9039
9040 return cnt;
9041}
9042
9043static const struct file_operations rb_simple_fops = {
9044 .open = tracing_open_generic_tr,
9045 .read = rb_simple_read,
9046 .write = rb_simple_write,
9047 .release = tracing_release_generic_tr,
9048 .llseek = default_llseek,
9049};
9050
9051static ssize_t
9052buffer_percent_read(struct file *filp, char __user *ubuf,
9053 size_t cnt, loff_t *ppos)
9054{
9055 struct trace_array *tr = filp->private_data;
9056 char buf[64];
9057 int r;
9058
9059 r = tr->buffer_percent;
9060 r = sprintf(buf, "%d\n", r);
9061
9062 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9063}
9064
9065static ssize_t
9066buffer_percent_write(struct file *filp, const char __user *ubuf,
9067 size_t cnt, loff_t *ppos)
9068{
9069 struct trace_array *tr = filp->private_data;
9070 unsigned long val;
9071 int ret;
9072
9073 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9074 if (ret)
9075 return ret;
9076
9077 if (val > 100)
9078 return -EINVAL;
9079
9080 tr->buffer_percent = val;
9081
9082 (*ppos)++;
9083
9084 return cnt;
9085}
9086
9087static const struct file_operations buffer_percent_fops = {
9088 .open = tracing_open_generic_tr,
9089 .read = buffer_percent_read,
9090 .write = buffer_percent_write,
9091 .release = tracing_release_generic_tr,
9092 .llseek = default_llseek,
9093};
9094
9095static ssize_t
9096buffer_subbuf_size_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
9097{
9098 struct trace_array *tr = filp->private_data;
9099 size_t size;
9100 char buf[64];
9101 int order;
9102 int r;
9103
9104 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9105 size = (PAGE_SIZE << order) / 1024;
9106
9107 r = sprintf(buf, "%zd\n", size);
9108
9109 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9110}
9111
9112static ssize_t
9113buffer_subbuf_size_write(struct file *filp, const char __user *ubuf,
9114 size_t cnt, loff_t *ppos)
9115{
9116 struct trace_array *tr = filp->private_data;
9117 unsigned long val;
9118 int old_order;
9119 int order;
9120 int pages;
9121 int ret;
9122
9123 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9124 if (ret)
9125 return ret;
9126
9127 val *= 1024; /* value passed in is in KB */
9128
9129 pages = DIV_ROUND_UP(val, PAGE_SIZE);
9130 order = fls(pages - 1);
9131
9132 /* limit between 1 and 128 system pages */
9133 if (order < 0 || order > 7)
9134 return -EINVAL;
9135
9136 /* Do not allow tracing while changing the order of the ring buffer */
9137 tracing_stop_tr(tr);
9138
9139 old_order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9140 if (old_order == order)
9141 goto out;
9142
9143 ret = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, order);
9144 if (ret)
9145 goto out;
9146
9147#ifdef CONFIG_TRACER_MAX_TRACE
9148
9149 if (!tr->allocated_snapshot)
9150 goto out_max;
9151
9152 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
9153 if (ret) {
9154 /* Put back the old order */
9155 cnt = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, old_order);
9156 if (WARN_ON_ONCE(cnt)) {
9157 /*
9158 * AARGH! We are left with different orders!
9159 * The max buffer is our "snapshot" buffer.
9160 * When a tracer needs a snapshot (one of the
9161 * latency tracers), it swaps the max buffer
9162 * with the saved snap shot. We succeeded to
9163 * update the order of the main buffer, but failed to
9164 * update the order of the max buffer. But when we tried
9165 * to reset the main buffer to the original size, we
9166 * failed there too. This is very unlikely to
9167 * happen, but if it does, warn and kill all
9168 * tracing.
9169 */
9170 tracing_disabled = 1;
9171 }
9172 goto out;
9173 }
9174 out_max:
9175#endif
9176 (*ppos)++;
9177 out:
9178 if (ret)
9179 cnt = ret;
9180 tracing_start_tr(tr);
9181 return cnt;
9182}
9183
9184static const struct file_operations buffer_subbuf_size_fops = {
9185 .open = tracing_open_generic_tr,
9186 .read = buffer_subbuf_size_read,
9187 .write = buffer_subbuf_size_write,
9188 .release = tracing_release_generic_tr,
9189 .llseek = default_llseek,
9190};
9191
9192static struct dentry *trace_instance_dir;
9193
9194static void
9195init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9196
9197static int
9198allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9199{
9200 enum ring_buffer_flags rb_flags;
9201
9202 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9203
9204 buf->tr = tr;
9205
9206 if (tr->range_addr_start && tr->range_addr_size) {
9207 buf->buffer = ring_buffer_alloc_range(size, rb_flags, 0,
9208 tr->range_addr_start,
9209 tr->range_addr_size);
9210
9211 ring_buffer_last_boot_delta(buf->buffer,
9212 &tr->text_delta, &tr->data_delta);
9213 /*
9214 * This is basically the same as a mapped buffer,
9215 * with the same restrictions.
9216 */
9217 tr->mapped++;
9218 } else {
9219 buf->buffer = ring_buffer_alloc(size, rb_flags);
9220 }
9221 if (!buf->buffer)
9222 return -ENOMEM;
9223
9224 buf->data = alloc_percpu(struct trace_array_cpu);
9225 if (!buf->data) {
9226 ring_buffer_free(buf->buffer);
9227 buf->buffer = NULL;
9228 return -ENOMEM;
9229 }
9230
9231 /* Allocate the first page for all buffers */
9232 set_buffer_entries(&tr->array_buffer,
9233 ring_buffer_size(tr->array_buffer.buffer, 0));
9234
9235 return 0;
9236}
9237
9238static void free_trace_buffer(struct array_buffer *buf)
9239{
9240 if (buf->buffer) {
9241 ring_buffer_free(buf->buffer);
9242 buf->buffer = NULL;
9243 free_percpu(buf->data);
9244 buf->data = NULL;
9245 }
9246}
9247
9248static int allocate_trace_buffers(struct trace_array *tr, int size)
9249{
9250 int ret;
9251
9252 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9253 if (ret)
9254 return ret;
9255
9256#ifdef CONFIG_TRACER_MAX_TRACE
9257 /* Fix mapped buffer trace arrays do not have snapshot buffers */
9258 if (tr->range_addr_start)
9259 return 0;
9260
9261 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9262 allocate_snapshot ? size : 1);
9263 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9264 free_trace_buffer(&tr->array_buffer);
9265 return -ENOMEM;
9266 }
9267 tr->allocated_snapshot = allocate_snapshot;
9268
9269 allocate_snapshot = false;
9270#endif
9271
9272 return 0;
9273}
9274
9275static void free_trace_buffers(struct trace_array *tr)
9276{
9277 if (!tr)
9278 return;
9279
9280 free_trace_buffer(&tr->array_buffer);
9281
9282#ifdef CONFIG_TRACER_MAX_TRACE
9283 free_trace_buffer(&tr->max_buffer);
9284#endif
9285}
9286
9287static void init_trace_flags_index(struct trace_array *tr)
9288{
9289 int i;
9290
9291 /* Used by the trace options files */
9292 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9293 tr->trace_flags_index[i] = i;
9294}
9295
9296static void __update_tracer_options(struct trace_array *tr)
9297{
9298 struct tracer *t;
9299
9300 for (t = trace_types; t; t = t->next)
9301 add_tracer_options(tr, t);
9302}
9303
9304static void update_tracer_options(struct trace_array *tr)
9305{
9306 mutex_lock(&trace_types_lock);
9307 tracer_options_updated = true;
9308 __update_tracer_options(tr);
9309 mutex_unlock(&trace_types_lock);
9310}
9311
9312/* Must have trace_types_lock held */
9313struct trace_array *trace_array_find(const char *instance)
9314{
9315 struct trace_array *tr, *found = NULL;
9316
9317 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9318 if (tr->name && strcmp(tr->name, instance) == 0) {
9319 found = tr;
9320 break;
9321 }
9322 }
9323
9324 return found;
9325}
9326
9327struct trace_array *trace_array_find_get(const char *instance)
9328{
9329 struct trace_array *tr;
9330
9331 mutex_lock(&trace_types_lock);
9332 tr = trace_array_find(instance);
9333 if (tr)
9334 tr->ref++;
9335 mutex_unlock(&trace_types_lock);
9336
9337 return tr;
9338}
9339
9340static int trace_array_create_dir(struct trace_array *tr)
9341{
9342 int ret;
9343
9344 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9345 if (!tr->dir)
9346 return -EINVAL;
9347
9348 ret = event_trace_add_tracer(tr->dir, tr);
9349 if (ret) {
9350 tracefs_remove(tr->dir);
9351 return ret;
9352 }
9353
9354 init_tracer_tracefs(tr, tr->dir);
9355 __update_tracer_options(tr);
9356
9357 return ret;
9358}
9359
9360static struct trace_array *
9361trace_array_create_systems(const char *name, const char *systems,
9362 unsigned long range_addr_start,
9363 unsigned long range_addr_size)
9364{
9365 struct trace_array *tr;
9366 int ret;
9367
9368 ret = -ENOMEM;
9369 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9370 if (!tr)
9371 return ERR_PTR(ret);
9372
9373 tr->name = kstrdup(name, GFP_KERNEL);
9374 if (!tr->name)
9375 goto out_free_tr;
9376
9377 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9378 goto out_free_tr;
9379
9380 if (!zalloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9381 goto out_free_tr;
9382
9383 if (systems) {
9384 tr->system_names = kstrdup_const(systems, GFP_KERNEL);
9385 if (!tr->system_names)
9386 goto out_free_tr;
9387 }
9388
9389 /* Only for boot up memory mapped ring buffers */
9390 tr->range_addr_start = range_addr_start;
9391 tr->range_addr_size = range_addr_size;
9392
9393 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9394
9395 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9396
9397 raw_spin_lock_init(&tr->start_lock);
9398
9399 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9400#ifdef CONFIG_TRACER_MAX_TRACE
9401 spin_lock_init(&tr->snapshot_trigger_lock);
9402#endif
9403 tr->current_trace = &nop_trace;
9404
9405 INIT_LIST_HEAD(&tr->systems);
9406 INIT_LIST_HEAD(&tr->events);
9407 INIT_LIST_HEAD(&tr->hist_vars);
9408 INIT_LIST_HEAD(&tr->err_log);
9409
9410 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9411 goto out_free_tr;
9412
9413 /* The ring buffer is defaultly expanded */
9414 trace_set_ring_buffer_expanded(tr);
9415
9416 if (ftrace_allocate_ftrace_ops(tr) < 0)
9417 goto out_free_tr;
9418
9419 ftrace_init_trace_array(tr);
9420
9421 init_trace_flags_index(tr);
9422
9423 if (trace_instance_dir) {
9424 ret = trace_array_create_dir(tr);
9425 if (ret)
9426 goto out_free_tr;
9427 } else
9428 __trace_early_add_events(tr);
9429
9430 list_add(&tr->list, &ftrace_trace_arrays);
9431
9432 tr->ref++;
9433
9434 return tr;
9435
9436 out_free_tr:
9437 ftrace_free_ftrace_ops(tr);
9438 free_trace_buffers(tr);
9439 free_cpumask_var(tr->pipe_cpumask);
9440 free_cpumask_var(tr->tracing_cpumask);
9441 kfree_const(tr->system_names);
9442 kfree(tr->name);
9443 kfree(tr);
9444
9445 return ERR_PTR(ret);
9446}
9447
9448static struct trace_array *trace_array_create(const char *name)
9449{
9450 return trace_array_create_systems(name, NULL, 0, 0);
9451}
9452
9453static int instance_mkdir(const char *name)
9454{
9455 struct trace_array *tr;
9456 int ret;
9457
9458 guard(mutex)(&event_mutex);
9459 guard(mutex)(&trace_types_lock);
9460
9461 ret = -EEXIST;
9462 if (trace_array_find(name))
9463 return -EEXIST;
9464
9465 tr = trace_array_create(name);
9466
9467 ret = PTR_ERR_OR_ZERO(tr);
9468
9469 return ret;
9470}
9471
9472static u64 map_pages(u64 start, u64 size)
9473{
9474 struct page **pages;
9475 phys_addr_t page_start;
9476 unsigned int page_count;
9477 unsigned int i;
9478 void *vaddr;
9479
9480 page_count = DIV_ROUND_UP(size, PAGE_SIZE);
9481
9482 page_start = start;
9483 pages = kmalloc_array(page_count, sizeof(struct page *), GFP_KERNEL);
9484 if (!pages)
9485 return 0;
9486
9487 for (i = 0; i < page_count; i++) {
9488 phys_addr_t addr = page_start + i * PAGE_SIZE;
9489 pages[i] = pfn_to_page(addr >> PAGE_SHIFT);
9490 }
9491 vaddr = vmap(pages, page_count, VM_MAP, PAGE_KERNEL);
9492 kfree(pages);
9493
9494 return (u64)(unsigned long)vaddr;
9495}
9496
9497/**
9498 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9499 * @name: The name of the trace array to be looked up/created.
9500 * @systems: A list of systems to create event directories for (NULL for all)
9501 *
9502 * Returns pointer to trace array with given name.
9503 * NULL, if it cannot be created.
9504 *
9505 * NOTE: This function increments the reference counter associated with the
9506 * trace array returned. This makes sure it cannot be freed while in use.
9507 * Use trace_array_put() once the trace array is no longer needed.
9508 * If the trace_array is to be freed, trace_array_destroy() needs to
9509 * be called after the trace_array_put(), or simply let user space delete
9510 * it from the tracefs instances directory. But until the
9511 * trace_array_put() is called, user space can not delete it.
9512 *
9513 */
9514struct trace_array *trace_array_get_by_name(const char *name, const char *systems)
9515{
9516 struct trace_array *tr;
9517
9518 guard(mutex)(&event_mutex);
9519 guard(mutex)(&trace_types_lock);
9520
9521 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9522 if (tr->name && strcmp(tr->name, name) == 0) {
9523 tr->ref++;
9524 return tr;
9525 }
9526 }
9527
9528 tr = trace_array_create_systems(name, systems, 0, 0);
9529
9530 if (IS_ERR(tr))
9531 tr = NULL;
9532 else
9533 tr->ref++;
9534
9535 return tr;
9536}
9537EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9538
9539static int __remove_instance(struct trace_array *tr)
9540{
9541 int i;
9542
9543 /* Reference counter for a newly created trace array = 1. */
9544 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9545 return -EBUSY;
9546
9547 list_del(&tr->list);
9548
9549 /* Disable all the flags that were enabled coming in */
9550 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9551 if ((1 << i) & ZEROED_TRACE_FLAGS)
9552 set_tracer_flag(tr, 1 << i, 0);
9553 }
9554
9555 if (printk_trace == tr)
9556 update_printk_trace(&global_trace);
9557
9558 tracing_set_nop(tr);
9559 clear_ftrace_function_probes(tr);
9560 event_trace_del_tracer(tr);
9561 ftrace_clear_pids(tr);
9562 ftrace_destroy_function_files(tr);
9563 tracefs_remove(tr->dir);
9564 free_percpu(tr->last_func_repeats);
9565 free_trace_buffers(tr);
9566 clear_tracing_err_log(tr);
9567
9568 for (i = 0; i < tr->nr_topts; i++) {
9569 kfree(tr->topts[i].topts);
9570 }
9571 kfree(tr->topts);
9572
9573 free_cpumask_var(tr->pipe_cpumask);
9574 free_cpumask_var(tr->tracing_cpumask);
9575 kfree_const(tr->system_names);
9576 kfree(tr->name);
9577 kfree(tr);
9578
9579 return 0;
9580}
9581
9582int trace_array_destroy(struct trace_array *this_tr)
9583{
9584 struct trace_array *tr;
9585
9586 if (!this_tr)
9587 return -EINVAL;
9588
9589 guard(mutex)(&event_mutex);
9590 guard(mutex)(&trace_types_lock);
9591
9592
9593 /* Making sure trace array exists before destroying it. */
9594 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9595 if (tr == this_tr)
9596 return __remove_instance(tr);
9597 }
9598
9599 return -ENODEV;
9600}
9601EXPORT_SYMBOL_GPL(trace_array_destroy);
9602
9603static int instance_rmdir(const char *name)
9604{
9605 struct trace_array *tr;
9606
9607 guard(mutex)(&event_mutex);
9608 guard(mutex)(&trace_types_lock);
9609
9610 tr = trace_array_find(name);
9611 if (!tr)
9612 return -ENODEV;
9613
9614 return __remove_instance(tr);
9615}
9616
9617static __init void create_trace_instances(struct dentry *d_tracer)
9618{
9619 struct trace_array *tr;
9620
9621 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9622 instance_mkdir,
9623 instance_rmdir);
9624 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9625 return;
9626
9627 guard(mutex)(&event_mutex);
9628 guard(mutex)(&trace_types_lock);
9629
9630 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9631 if (!tr->name)
9632 continue;
9633 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9634 "Failed to create instance directory\n"))
9635 return;
9636 }
9637}
9638
9639static void
9640init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9641{
9642 int cpu;
9643
9644 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9645 tr, &show_traces_fops);
9646
9647 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9648 tr, &set_tracer_fops);
9649
9650 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9651 tr, &tracing_cpumask_fops);
9652
9653 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9654 tr, &tracing_iter_fops);
9655
9656 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9657 tr, &tracing_fops);
9658
9659 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9660 tr, &tracing_pipe_fops);
9661
9662 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9663 tr, &tracing_entries_fops);
9664
9665 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9666 tr, &tracing_total_entries_fops);
9667
9668 trace_create_file("free_buffer", 0200, d_tracer,
9669 tr, &tracing_free_buffer_fops);
9670
9671 trace_create_file("trace_marker", 0220, d_tracer,
9672 tr, &tracing_mark_fops);
9673
9674 tr->trace_marker_file = __find_event_file(tr, "ftrace", "print");
9675
9676 trace_create_file("trace_marker_raw", 0220, d_tracer,
9677 tr, &tracing_mark_raw_fops);
9678
9679 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9680 &trace_clock_fops);
9681
9682 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9683 tr, &rb_simple_fops);
9684
9685 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9686 &trace_time_stamp_mode_fops);
9687
9688 tr->buffer_percent = 50;
9689
9690 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9691 tr, &buffer_percent_fops);
9692
9693 trace_create_file("buffer_subbuf_size_kb", TRACE_MODE_WRITE, d_tracer,
9694 tr, &buffer_subbuf_size_fops);
9695
9696 create_trace_options_dir(tr);
9697
9698#ifdef CONFIG_TRACER_MAX_TRACE
9699 trace_create_maxlat_file(tr, d_tracer);
9700#endif
9701
9702 if (ftrace_create_function_files(tr, d_tracer))
9703 MEM_FAIL(1, "Could not allocate function filter files");
9704
9705 if (tr->range_addr_start) {
9706 trace_create_file("last_boot_info", TRACE_MODE_READ, d_tracer,
9707 tr, &last_boot_fops);
9708#ifdef CONFIG_TRACER_SNAPSHOT
9709 } else {
9710 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9711 tr, &snapshot_fops);
9712#endif
9713 }
9714
9715 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9716 tr, &tracing_err_log_fops);
9717
9718 for_each_tracing_cpu(cpu)
9719 tracing_init_tracefs_percpu(tr, cpu);
9720
9721 ftrace_init_tracefs(tr, d_tracer);
9722}
9723
9724static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9725{
9726 struct vfsmount *mnt;
9727 struct file_system_type *type;
9728
9729 /*
9730 * To maintain backward compatibility for tools that mount
9731 * debugfs to get to the tracing facility, tracefs is automatically
9732 * mounted to the debugfs/tracing directory.
9733 */
9734 type = get_fs_type("tracefs");
9735 if (!type)
9736 return NULL;
9737 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9738 put_filesystem(type);
9739 if (IS_ERR(mnt))
9740 return NULL;
9741 mntget(mnt);
9742
9743 return mnt;
9744}
9745
9746/**
9747 * tracing_init_dentry - initialize top level trace array
9748 *
9749 * This is called when creating files or directories in the tracing
9750 * directory. It is called via fs_initcall() by any of the boot up code
9751 * and expects to return the dentry of the top level tracing directory.
9752 */
9753int tracing_init_dentry(void)
9754{
9755 struct trace_array *tr = &global_trace;
9756
9757 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9758 pr_warn("Tracing disabled due to lockdown\n");
9759 return -EPERM;
9760 }
9761
9762 /* The top level trace array uses NULL as parent */
9763 if (tr->dir)
9764 return 0;
9765
9766 if (WARN_ON(!tracefs_initialized()))
9767 return -ENODEV;
9768
9769 /*
9770 * As there may still be users that expect the tracing
9771 * files to exist in debugfs/tracing, we must automount
9772 * the tracefs file system there, so older tools still
9773 * work with the newer kernel.
9774 */
9775 tr->dir = debugfs_create_automount("tracing", NULL,
9776 trace_automount, NULL);
9777
9778 return 0;
9779}
9780
9781extern struct trace_eval_map *__start_ftrace_eval_maps[];
9782extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9783
9784static struct workqueue_struct *eval_map_wq __initdata;
9785static struct work_struct eval_map_work __initdata;
9786static struct work_struct tracerfs_init_work __initdata;
9787
9788static void __init eval_map_work_func(struct work_struct *work)
9789{
9790 int len;
9791
9792 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9793 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9794}
9795
9796static int __init trace_eval_init(void)
9797{
9798 INIT_WORK(&eval_map_work, eval_map_work_func);
9799
9800 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9801 if (!eval_map_wq) {
9802 pr_err("Unable to allocate eval_map_wq\n");
9803 /* Do work here */
9804 eval_map_work_func(&eval_map_work);
9805 return -ENOMEM;
9806 }
9807
9808 queue_work(eval_map_wq, &eval_map_work);
9809 return 0;
9810}
9811
9812subsys_initcall(trace_eval_init);
9813
9814static int __init trace_eval_sync(void)
9815{
9816 /* Make sure the eval map updates are finished */
9817 if (eval_map_wq)
9818 destroy_workqueue(eval_map_wq);
9819 return 0;
9820}
9821
9822late_initcall_sync(trace_eval_sync);
9823
9824
9825#ifdef CONFIG_MODULES
9826static void trace_module_add_evals(struct module *mod)
9827{
9828 if (!mod->num_trace_evals)
9829 return;
9830
9831 /*
9832 * Modules with bad taint do not have events created, do
9833 * not bother with enums either.
9834 */
9835 if (trace_module_has_bad_taint(mod))
9836 return;
9837
9838 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9839}
9840
9841#ifdef CONFIG_TRACE_EVAL_MAP_FILE
9842static void trace_module_remove_evals(struct module *mod)
9843{
9844 union trace_eval_map_item *map;
9845 union trace_eval_map_item **last = &trace_eval_maps;
9846
9847 if (!mod->num_trace_evals)
9848 return;
9849
9850 guard(mutex)(&trace_eval_mutex);
9851
9852 map = trace_eval_maps;
9853
9854 while (map) {
9855 if (map->head.mod == mod)
9856 break;
9857 map = trace_eval_jmp_to_tail(map);
9858 last = &map->tail.next;
9859 map = map->tail.next;
9860 }
9861 if (!map)
9862 return;
9863
9864 *last = trace_eval_jmp_to_tail(map)->tail.next;
9865 kfree(map);
9866}
9867#else
9868static inline void trace_module_remove_evals(struct module *mod) { }
9869#endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9870
9871static int trace_module_notify(struct notifier_block *self,
9872 unsigned long val, void *data)
9873{
9874 struct module *mod = data;
9875
9876 switch (val) {
9877 case MODULE_STATE_COMING:
9878 trace_module_add_evals(mod);
9879 break;
9880 case MODULE_STATE_GOING:
9881 trace_module_remove_evals(mod);
9882 break;
9883 }
9884
9885 return NOTIFY_OK;
9886}
9887
9888static struct notifier_block trace_module_nb = {
9889 .notifier_call = trace_module_notify,
9890 .priority = 0,
9891};
9892#endif /* CONFIG_MODULES */
9893
9894static __init void tracer_init_tracefs_work_func(struct work_struct *work)
9895{
9896
9897 event_trace_init();
9898
9899 init_tracer_tracefs(&global_trace, NULL);
9900 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9901
9902 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9903 &global_trace, &tracing_thresh_fops);
9904
9905 trace_create_file("README", TRACE_MODE_READ, NULL,
9906 NULL, &tracing_readme_fops);
9907
9908 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9909 NULL, &tracing_saved_cmdlines_fops);
9910
9911 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9912 NULL, &tracing_saved_cmdlines_size_fops);
9913
9914 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9915 NULL, &tracing_saved_tgids_fops);
9916
9917 trace_create_eval_file(NULL);
9918
9919#ifdef CONFIG_MODULES
9920 register_module_notifier(&trace_module_nb);
9921#endif
9922
9923#ifdef CONFIG_DYNAMIC_FTRACE
9924 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
9925 NULL, &tracing_dyn_info_fops);
9926#endif
9927
9928 create_trace_instances(NULL);
9929
9930 update_tracer_options(&global_trace);
9931}
9932
9933static __init int tracer_init_tracefs(void)
9934{
9935 int ret;
9936
9937 trace_access_lock_init();
9938
9939 ret = tracing_init_dentry();
9940 if (ret)
9941 return 0;
9942
9943 if (eval_map_wq) {
9944 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
9945 queue_work(eval_map_wq, &tracerfs_init_work);
9946 } else {
9947 tracer_init_tracefs_work_func(NULL);
9948 }
9949
9950 rv_init_interface();
9951
9952 return 0;
9953}
9954
9955fs_initcall(tracer_init_tracefs);
9956
9957static int trace_die_panic_handler(struct notifier_block *self,
9958 unsigned long ev, void *unused);
9959
9960static struct notifier_block trace_panic_notifier = {
9961 .notifier_call = trace_die_panic_handler,
9962 .priority = INT_MAX - 1,
9963};
9964
9965static struct notifier_block trace_die_notifier = {
9966 .notifier_call = trace_die_panic_handler,
9967 .priority = INT_MAX - 1,
9968};
9969
9970/*
9971 * The idea is to execute the following die/panic callback early, in order
9972 * to avoid showing irrelevant information in the trace (like other panic
9973 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
9974 * warnings get disabled (to prevent potential log flooding).
9975 */
9976static int trace_die_panic_handler(struct notifier_block *self,
9977 unsigned long ev, void *unused)
9978{
9979 if (!ftrace_dump_on_oops_enabled())
9980 return NOTIFY_DONE;
9981
9982 /* The die notifier requires DIE_OOPS to trigger */
9983 if (self == &trace_die_notifier && ev != DIE_OOPS)
9984 return NOTIFY_DONE;
9985
9986 ftrace_dump(DUMP_PARAM);
9987
9988 return NOTIFY_DONE;
9989}
9990
9991/*
9992 * printk is set to max of 1024, we really don't need it that big.
9993 * Nothing should be printing 1000 characters anyway.
9994 */
9995#define TRACE_MAX_PRINT 1000
9996
9997/*
9998 * Define here KERN_TRACE so that we have one place to modify
9999 * it if we decide to change what log level the ftrace dump
10000 * should be at.
10001 */
10002#define KERN_TRACE KERN_EMERG
10003
10004void
10005trace_printk_seq(struct trace_seq *s)
10006{
10007 /* Probably should print a warning here. */
10008 if (s->seq.len >= TRACE_MAX_PRINT)
10009 s->seq.len = TRACE_MAX_PRINT;
10010
10011 /*
10012 * More paranoid code. Although the buffer size is set to
10013 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
10014 * an extra layer of protection.
10015 */
10016 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
10017 s->seq.len = s->seq.size - 1;
10018
10019 /* should be zero ended, but we are paranoid. */
10020 s->buffer[s->seq.len] = 0;
10021
10022 printk(KERN_TRACE "%s", s->buffer);
10023
10024 trace_seq_init(s);
10025}
10026
10027static void trace_init_iter(struct trace_iterator *iter, struct trace_array *tr)
10028{
10029 iter->tr = tr;
10030 iter->trace = iter->tr->current_trace;
10031 iter->cpu_file = RING_BUFFER_ALL_CPUS;
10032 iter->array_buffer = &tr->array_buffer;
10033
10034 if (iter->trace && iter->trace->open)
10035 iter->trace->open(iter);
10036
10037 /* Annotate start of buffers if we had overruns */
10038 if (ring_buffer_overruns(iter->array_buffer->buffer))
10039 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10040
10041 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10042 if (trace_clocks[iter->tr->clock_id].in_ns)
10043 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10044
10045 /* Can not use kmalloc for iter.temp and iter.fmt */
10046 iter->temp = static_temp_buf;
10047 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10048 iter->fmt = static_fmt_buf;
10049 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10050}
10051
10052void trace_init_global_iter(struct trace_iterator *iter)
10053{
10054 trace_init_iter(iter, &global_trace);
10055}
10056
10057static void ftrace_dump_one(struct trace_array *tr, enum ftrace_dump_mode dump_mode)
10058{
10059 /* use static because iter can be a bit big for the stack */
10060 static struct trace_iterator iter;
10061 unsigned int old_userobj;
10062 unsigned long flags;
10063 int cnt = 0, cpu;
10064
10065 /*
10066 * Always turn off tracing when we dump.
10067 * We don't need to show trace output of what happens
10068 * between multiple crashes.
10069 *
10070 * If the user does a sysrq-z, then they can re-enable
10071 * tracing with echo 1 > tracing_on.
10072 */
10073 tracer_tracing_off(tr);
10074
10075 local_irq_save(flags);
10076
10077 /* Simulate the iterator */
10078 trace_init_iter(&iter, tr);
10079
10080 for_each_tracing_cpu(cpu) {
10081 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10082 }
10083
10084 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10085
10086 /* don't look at user memory in panic mode */
10087 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10088
10089 if (dump_mode == DUMP_ORIG)
10090 iter.cpu_file = raw_smp_processor_id();
10091 else
10092 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10093
10094 if (tr == &global_trace)
10095 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10096 else
10097 printk(KERN_TRACE "Dumping ftrace instance %s buffer:\n", tr->name);
10098
10099 /* Did function tracer already get disabled? */
10100 if (ftrace_is_dead()) {
10101 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10102 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10103 }
10104
10105 /*
10106 * We need to stop all tracing on all CPUS to read
10107 * the next buffer. This is a bit expensive, but is
10108 * not done often. We fill all what we can read,
10109 * and then release the locks again.
10110 */
10111
10112 while (!trace_empty(&iter)) {
10113
10114 if (!cnt)
10115 printk(KERN_TRACE "---------------------------------\n");
10116
10117 cnt++;
10118
10119 trace_iterator_reset(&iter);
10120 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10121
10122 if (trace_find_next_entry_inc(&iter) != NULL) {
10123 int ret;
10124
10125 ret = print_trace_line(&iter);
10126 if (ret != TRACE_TYPE_NO_CONSUME)
10127 trace_consume(&iter);
10128 }
10129 touch_nmi_watchdog();
10130
10131 trace_printk_seq(&iter.seq);
10132 }
10133
10134 if (!cnt)
10135 printk(KERN_TRACE " (ftrace buffer empty)\n");
10136 else
10137 printk(KERN_TRACE "---------------------------------\n");
10138
10139 tr->trace_flags |= old_userobj;
10140
10141 for_each_tracing_cpu(cpu) {
10142 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10143 }
10144 local_irq_restore(flags);
10145}
10146
10147static void ftrace_dump_by_param(void)
10148{
10149 bool first_param = true;
10150 char dump_param[MAX_TRACER_SIZE];
10151 char *buf, *token, *inst_name;
10152 struct trace_array *tr;
10153
10154 strscpy(dump_param, ftrace_dump_on_oops, MAX_TRACER_SIZE);
10155 buf = dump_param;
10156
10157 while ((token = strsep(&buf, ",")) != NULL) {
10158 if (first_param) {
10159 first_param = false;
10160 if (!strcmp("0", token))
10161 continue;
10162 else if (!strcmp("1", token)) {
10163 ftrace_dump_one(&global_trace, DUMP_ALL);
10164 continue;
10165 }
10166 else if (!strcmp("2", token) ||
10167 !strcmp("orig_cpu", token)) {
10168 ftrace_dump_one(&global_trace, DUMP_ORIG);
10169 continue;
10170 }
10171 }
10172
10173 inst_name = strsep(&token, "=");
10174 tr = trace_array_find(inst_name);
10175 if (!tr) {
10176 printk(KERN_TRACE "Instance %s not found\n", inst_name);
10177 continue;
10178 }
10179
10180 if (token && (!strcmp("2", token) ||
10181 !strcmp("orig_cpu", token)))
10182 ftrace_dump_one(tr, DUMP_ORIG);
10183 else
10184 ftrace_dump_one(tr, DUMP_ALL);
10185 }
10186}
10187
10188void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10189{
10190 static atomic_t dump_running;
10191
10192 /* Only allow one dump user at a time. */
10193 if (atomic_inc_return(&dump_running) != 1) {
10194 atomic_dec(&dump_running);
10195 return;
10196 }
10197
10198 switch (oops_dump_mode) {
10199 case DUMP_ALL:
10200 ftrace_dump_one(&global_trace, DUMP_ALL);
10201 break;
10202 case DUMP_ORIG:
10203 ftrace_dump_one(&global_trace, DUMP_ORIG);
10204 break;
10205 case DUMP_PARAM:
10206 ftrace_dump_by_param();
10207 break;
10208 case DUMP_NONE:
10209 break;
10210 default:
10211 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10212 ftrace_dump_one(&global_trace, DUMP_ALL);
10213 }
10214
10215 atomic_dec(&dump_running);
10216}
10217EXPORT_SYMBOL_GPL(ftrace_dump);
10218
10219#define WRITE_BUFSIZE 4096
10220
10221ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10222 size_t count, loff_t *ppos,
10223 int (*createfn)(const char *))
10224{
10225 char *kbuf, *buf, *tmp;
10226 int ret = 0;
10227 size_t done = 0;
10228 size_t size;
10229
10230 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10231 if (!kbuf)
10232 return -ENOMEM;
10233
10234 while (done < count) {
10235 size = count - done;
10236
10237 if (size >= WRITE_BUFSIZE)
10238 size = WRITE_BUFSIZE - 1;
10239
10240 if (copy_from_user(kbuf, buffer + done, size)) {
10241 ret = -EFAULT;
10242 goto out;
10243 }
10244 kbuf[size] = '\0';
10245 buf = kbuf;
10246 do {
10247 tmp = strchr(buf, '\n');
10248 if (tmp) {
10249 *tmp = '\0';
10250 size = tmp - buf + 1;
10251 } else {
10252 size = strlen(buf);
10253 if (done + size < count) {
10254 if (buf != kbuf)
10255 break;
10256 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10257 pr_warn("Line length is too long: Should be less than %d\n",
10258 WRITE_BUFSIZE - 2);
10259 ret = -EINVAL;
10260 goto out;
10261 }
10262 }
10263 done += size;
10264
10265 /* Remove comments */
10266 tmp = strchr(buf, '#');
10267
10268 if (tmp)
10269 *tmp = '\0';
10270
10271 ret = createfn(buf);
10272 if (ret)
10273 goto out;
10274 buf += size;
10275
10276 } while (done < count);
10277 }
10278 ret = done;
10279
10280out:
10281 kfree(kbuf);
10282
10283 return ret;
10284}
10285
10286#ifdef CONFIG_TRACER_MAX_TRACE
10287__init static bool tr_needs_alloc_snapshot(const char *name)
10288{
10289 char *test;
10290 int len = strlen(name);
10291 bool ret;
10292
10293 if (!boot_snapshot_index)
10294 return false;
10295
10296 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10297 boot_snapshot_info[len] == '\t')
10298 return true;
10299
10300 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10301 if (!test)
10302 return false;
10303
10304 sprintf(test, "\t%s\t", name);
10305 ret = strstr(boot_snapshot_info, test) == NULL;
10306 kfree(test);
10307 return ret;
10308}
10309
10310__init static void do_allocate_snapshot(const char *name)
10311{
10312 if (!tr_needs_alloc_snapshot(name))
10313 return;
10314
10315 /*
10316 * When allocate_snapshot is set, the next call to
10317 * allocate_trace_buffers() (called by trace_array_get_by_name())
10318 * will allocate the snapshot buffer. That will alse clear
10319 * this flag.
10320 */
10321 allocate_snapshot = true;
10322}
10323#else
10324static inline void do_allocate_snapshot(const char *name) { }
10325#endif
10326
10327__init static void enable_instances(void)
10328{
10329 struct trace_array *tr;
10330 char *curr_str;
10331 char *name;
10332 char *str;
10333 char *tok;
10334
10335 /* A tab is always appended */
10336 boot_instance_info[boot_instance_index - 1] = '\0';
10337 str = boot_instance_info;
10338
10339 while ((curr_str = strsep(&str, "\t"))) {
10340 phys_addr_t start = 0;
10341 phys_addr_t size = 0;
10342 unsigned long addr = 0;
10343 bool traceprintk = false;
10344 bool traceoff = false;
10345 char *flag_delim;
10346 char *addr_delim;
10347
10348 tok = strsep(&curr_str, ",");
10349
10350 flag_delim = strchr(tok, '^');
10351 addr_delim = strchr(tok, '@');
10352
10353 if (addr_delim)
10354 *addr_delim++ = '\0';
10355
10356 if (flag_delim)
10357 *flag_delim++ = '\0';
10358
10359 name = tok;
10360
10361 if (flag_delim) {
10362 char *flag;
10363
10364 while ((flag = strsep(&flag_delim, "^"))) {
10365 if (strcmp(flag, "traceoff") == 0) {
10366 traceoff = true;
10367 } else if ((strcmp(flag, "printk") == 0) ||
10368 (strcmp(flag, "traceprintk") == 0) ||
10369 (strcmp(flag, "trace_printk") == 0)) {
10370 traceprintk = true;
10371 } else {
10372 pr_info("Tracing: Invalid instance flag '%s' for %s\n",
10373 flag, name);
10374 }
10375 }
10376 }
10377
10378 tok = addr_delim;
10379 if (tok && isdigit(*tok)) {
10380 start = memparse(tok, &tok);
10381 if (!start) {
10382 pr_warn("Tracing: Invalid boot instance address for %s\n",
10383 name);
10384 continue;
10385 }
10386 if (*tok != ':') {
10387 pr_warn("Tracing: No size specified for instance %s\n", name);
10388 continue;
10389 }
10390 tok++;
10391 size = memparse(tok, &tok);
10392 if (!size) {
10393 pr_warn("Tracing: Invalid boot instance size for %s\n",
10394 name);
10395 continue;
10396 }
10397 } else if (tok) {
10398 if (!reserve_mem_find_by_name(tok, &start, &size)) {
10399 start = 0;
10400 pr_warn("Failed to map boot instance %s to %s\n", name, tok);
10401 continue;
10402 }
10403 }
10404
10405 if (start) {
10406 addr = map_pages(start, size);
10407 if (addr) {
10408 pr_info("Tracing: mapped boot instance %s at physical memory %pa of size 0x%lx\n",
10409 name, &start, (unsigned long)size);
10410 } else {
10411 pr_warn("Tracing: Failed to map boot instance %s\n", name);
10412 continue;
10413 }
10414 } else {
10415 /* Only non mapped buffers have snapshot buffers */
10416 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10417 do_allocate_snapshot(name);
10418 }
10419
10420 tr = trace_array_create_systems(name, NULL, addr, size);
10421 if (IS_ERR(tr)) {
10422 pr_warn("Tracing: Failed to create instance buffer %s\n", curr_str);
10423 continue;
10424 }
10425
10426 if (traceoff)
10427 tracer_tracing_off(tr);
10428
10429 if (traceprintk)
10430 update_printk_trace(tr);
10431
10432 /*
10433 * If start is set, then this is a mapped buffer, and
10434 * cannot be deleted by user space, so keep the reference
10435 * to it.
10436 */
10437 if (start) {
10438 tr->flags |= TRACE_ARRAY_FL_BOOT;
10439 tr->ref++;
10440 }
10441
10442 while ((tok = strsep(&curr_str, ","))) {
10443 early_enable_events(tr, tok, true);
10444 }
10445 }
10446}
10447
10448__init static int tracer_alloc_buffers(void)
10449{
10450 int ring_buf_size;
10451 int ret = -ENOMEM;
10452
10453
10454 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10455 pr_warn("Tracing disabled due to lockdown\n");
10456 return -EPERM;
10457 }
10458
10459 /*
10460 * Make sure we don't accidentally add more trace options
10461 * than we have bits for.
10462 */
10463 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10464
10465 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10466 goto out;
10467
10468 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10469 goto out_free_buffer_mask;
10470
10471 /* Only allocate trace_printk buffers if a trace_printk exists */
10472 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10473 /* Must be called before global_trace.buffer is allocated */
10474 trace_printk_init_buffers();
10475
10476 /* To save memory, keep the ring buffer size to its minimum */
10477 if (global_trace.ring_buffer_expanded)
10478 ring_buf_size = trace_buf_size;
10479 else
10480 ring_buf_size = 1;
10481
10482 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10483 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10484
10485 raw_spin_lock_init(&global_trace.start_lock);
10486
10487 /*
10488 * The prepare callbacks allocates some memory for the ring buffer. We
10489 * don't free the buffer if the CPU goes down. If we were to free
10490 * the buffer, then the user would lose any trace that was in the
10491 * buffer. The memory will be removed once the "instance" is removed.
10492 */
10493 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10494 "trace/RB:prepare", trace_rb_cpu_prepare,
10495 NULL);
10496 if (ret < 0)
10497 goto out_free_cpumask;
10498 /* Used for event triggers */
10499 ret = -ENOMEM;
10500 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10501 if (!temp_buffer)
10502 goto out_rm_hp_state;
10503
10504 if (trace_create_savedcmd() < 0)
10505 goto out_free_temp_buffer;
10506
10507 if (!zalloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10508 goto out_free_savedcmd;
10509
10510 /* TODO: make the number of buffers hot pluggable with CPUS */
10511 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10512 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10513 goto out_free_pipe_cpumask;
10514 }
10515 if (global_trace.buffer_disabled)
10516 tracing_off();
10517
10518 if (trace_boot_clock) {
10519 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10520 if (ret < 0)
10521 pr_warn("Trace clock %s not defined, going back to default\n",
10522 trace_boot_clock);
10523 }
10524
10525 /*
10526 * register_tracer() might reference current_trace, so it
10527 * needs to be set before we register anything. This is
10528 * just a bootstrap of current_trace anyway.
10529 */
10530 global_trace.current_trace = &nop_trace;
10531
10532 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10533#ifdef CONFIG_TRACER_MAX_TRACE
10534 spin_lock_init(&global_trace.snapshot_trigger_lock);
10535#endif
10536 ftrace_init_global_array_ops(&global_trace);
10537
10538 init_trace_flags_index(&global_trace);
10539
10540 register_tracer(&nop_trace);
10541
10542 /* Function tracing may start here (via kernel command line) */
10543 init_function_trace();
10544
10545 /* All seems OK, enable tracing */
10546 tracing_disabled = 0;
10547
10548 atomic_notifier_chain_register(&panic_notifier_list,
10549 &trace_panic_notifier);
10550
10551 register_die_notifier(&trace_die_notifier);
10552
10553 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10554
10555 INIT_LIST_HEAD(&global_trace.systems);
10556 INIT_LIST_HEAD(&global_trace.events);
10557 INIT_LIST_HEAD(&global_trace.hist_vars);
10558 INIT_LIST_HEAD(&global_trace.err_log);
10559 list_add(&global_trace.list, &ftrace_trace_arrays);
10560
10561 apply_trace_boot_options();
10562
10563 register_snapshot_cmd();
10564
10565 return 0;
10566
10567out_free_pipe_cpumask:
10568 free_cpumask_var(global_trace.pipe_cpumask);
10569out_free_savedcmd:
10570 trace_free_saved_cmdlines_buffer();
10571out_free_temp_buffer:
10572 ring_buffer_free(temp_buffer);
10573out_rm_hp_state:
10574 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10575out_free_cpumask:
10576 free_cpumask_var(global_trace.tracing_cpumask);
10577out_free_buffer_mask:
10578 free_cpumask_var(tracing_buffer_mask);
10579out:
10580 return ret;
10581}
10582
10583void __init ftrace_boot_snapshot(void)
10584{
10585#ifdef CONFIG_TRACER_MAX_TRACE
10586 struct trace_array *tr;
10587
10588 if (!snapshot_at_boot)
10589 return;
10590
10591 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10592 if (!tr->allocated_snapshot)
10593 continue;
10594
10595 tracing_snapshot_instance(tr);
10596 trace_array_puts(tr, "** Boot snapshot taken **\n");
10597 }
10598#endif
10599}
10600
10601void __init early_trace_init(void)
10602{
10603 if (tracepoint_printk) {
10604 tracepoint_print_iter =
10605 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10606 if (MEM_FAIL(!tracepoint_print_iter,
10607 "Failed to allocate trace iterator\n"))
10608 tracepoint_printk = 0;
10609 else
10610 static_key_enable(&tracepoint_printk_key.key);
10611 }
10612 tracer_alloc_buffers();
10613
10614 init_events();
10615}
10616
10617void __init trace_init(void)
10618{
10619 trace_event_init();
10620
10621 if (boot_instance_index)
10622 enable_instances();
10623}
10624
10625__init static void clear_boot_tracer(void)
10626{
10627 /*
10628 * The default tracer at boot buffer is an init section.
10629 * This function is called in lateinit. If we did not
10630 * find the boot tracer, then clear it out, to prevent
10631 * later registration from accessing the buffer that is
10632 * about to be freed.
10633 */
10634 if (!default_bootup_tracer)
10635 return;
10636
10637 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10638 default_bootup_tracer);
10639 default_bootup_tracer = NULL;
10640}
10641
10642#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10643__init static void tracing_set_default_clock(void)
10644{
10645 /* sched_clock_stable() is determined in late_initcall */
10646 if (!trace_boot_clock && !sched_clock_stable()) {
10647 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10648 pr_warn("Can not set tracing clock due to lockdown\n");
10649 return;
10650 }
10651
10652 printk(KERN_WARNING
10653 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10654 "If you want to keep using the local clock, then add:\n"
10655 " \"trace_clock=local\"\n"
10656 "on the kernel command line\n");
10657 tracing_set_clock(&global_trace, "global");
10658 }
10659}
10660#else
10661static inline void tracing_set_default_clock(void) { }
10662#endif
10663
10664__init static int late_trace_init(void)
10665{
10666 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10667 static_key_disable(&tracepoint_printk_key.key);
10668 tracepoint_printk = 0;
10669 }
10670
10671 tracing_set_default_clock();
10672 clear_boot_tracer();
10673 return 0;
10674}
10675
10676late_initcall_sync(late_trace_init);
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/pagemap.h>
24#include <linux/hardirq.h>
25#include <linux/linkage.h>
26#include <linux/uaccess.h>
27#include <linux/kprobes.h>
28#include <linux/ftrace.h>
29#include <linux/module.h>
30#include <linux/percpu.h>
31#include <linux/splice.h>
32#include <linux/kdebug.h>
33#include <linux/string.h>
34#include <linux/rwsem.h>
35#include <linux/slab.h>
36#include <linux/ctype.h>
37#include <linux/init.h>
38#include <linux/poll.h>
39#include <linux/nmi.h>
40#include <linux/fs.h>
41#include <linux/sched/rt.h>
42
43#include "trace.h"
44#include "trace_output.h"
45
46/*
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
49 */
50bool ring_buffer_expanded;
51
52/*
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
58 */
59static bool __read_mostly tracing_selftest_running;
60
61/*
62 * If a tracer is running, we do not want to run SELFTEST.
63 */
64bool __read_mostly tracing_selftest_disabled;
65
66/* For tracers that don't implement custom flags */
67static struct tracer_opt dummy_tracer_opt[] = {
68 { }
69};
70
71static struct tracer_flags dummy_tracer_flags = {
72 .val = 0,
73 .opts = dummy_tracer_opt
74};
75
76static int
77dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
78{
79 return 0;
80}
81
82/*
83 * To prevent the comm cache from being overwritten when no
84 * tracing is active, only save the comm when a trace event
85 * occurred.
86 */
87static DEFINE_PER_CPU(bool, trace_cmdline_save);
88
89/*
90 * Kill all tracing for good (never come back).
91 * It is initialized to 1 but will turn to zero if the initialization
92 * of the tracer is successful. But that is the only place that sets
93 * this back to zero.
94 */
95static int tracing_disabled = 1;
96
97DEFINE_PER_CPU(int, ftrace_cpu_disabled);
98
99cpumask_var_t __read_mostly tracing_buffer_mask;
100
101/*
102 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
103 *
104 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
105 * is set, then ftrace_dump is called. This will output the contents
106 * of the ftrace buffers to the console. This is very useful for
107 * capturing traces that lead to crashes and outputing it to a
108 * serial console.
109 *
110 * It is default off, but you can enable it with either specifying
111 * "ftrace_dump_on_oops" in the kernel command line, or setting
112 * /proc/sys/kernel/ftrace_dump_on_oops
113 * Set 1 if you want to dump buffers of all CPUs
114 * Set 2 if you want to dump the buffer of the CPU that triggered oops
115 */
116
117enum ftrace_dump_mode ftrace_dump_on_oops;
118
119/* When set, tracing will stop when a WARN*() is hit */
120int __disable_trace_on_warning;
121
122static int tracing_set_tracer(struct trace_array *tr, const char *buf);
123
124#define MAX_TRACER_SIZE 100
125static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
126static char *default_bootup_tracer;
127
128static bool allocate_snapshot;
129
130static int __init set_cmdline_ftrace(char *str)
131{
132 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
133 default_bootup_tracer = bootup_tracer_buf;
134 /* We are using ftrace early, expand it */
135 ring_buffer_expanded = true;
136 return 1;
137}
138__setup("ftrace=", set_cmdline_ftrace);
139
140static int __init set_ftrace_dump_on_oops(char *str)
141{
142 if (*str++ != '=' || !*str) {
143 ftrace_dump_on_oops = DUMP_ALL;
144 return 1;
145 }
146
147 if (!strcmp("orig_cpu", str)) {
148 ftrace_dump_on_oops = DUMP_ORIG;
149 return 1;
150 }
151
152 return 0;
153}
154__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
155
156static int __init stop_trace_on_warning(char *str)
157{
158 __disable_trace_on_warning = 1;
159 return 1;
160}
161__setup("traceoff_on_warning=", stop_trace_on_warning);
162
163static int __init boot_alloc_snapshot(char *str)
164{
165 allocate_snapshot = true;
166 /* We also need the main ring buffer expanded */
167 ring_buffer_expanded = true;
168 return 1;
169}
170__setup("alloc_snapshot", boot_alloc_snapshot);
171
172
173static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
174static char *trace_boot_options __initdata;
175
176static int __init set_trace_boot_options(char *str)
177{
178 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
179 trace_boot_options = trace_boot_options_buf;
180 return 0;
181}
182__setup("trace_options=", set_trace_boot_options);
183
184static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
185static char *trace_boot_clock __initdata;
186
187static int __init set_trace_boot_clock(char *str)
188{
189 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
190 trace_boot_clock = trace_boot_clock_buf;
191 return 0;
192}
193__setup("trace_clock=", set_trace_boot_clock);
194
195
196unsigned long long ns2usecs(cycle_t nsec)
197{
198 nsec += 500;
199 do_div(nsec, 1000);
200 return nsec;
201}
202
203/*
204 * The global_trace is the descriptor that holds the tracing
205 * buffers for the live tracing. For each CPU, it contains
206 * a link list of pages that will store trace entries. The
207 * page descriptor of the pages in the memory is used to hold
208 * the link list by linking the lru item in the page descriptor
209 * to each of the pages in the buffer per CPU.
210 *
211 * For each active CPU there is a data field that holds the
212 * pages for the buffer for that CPU. Each CPU has the same number
213 * of pages allocated for its buffer.
214 */
215static struct trace_array global_trace;
216
217LIST_HEAD(ftrace_trace_arrays);
218
219int trace_array_get(struct trace_array *this_tr)
220{
221 struct trace_array *tr;
222 int ret = -ENODEV;
223
224 mutex_lock(&trace_types_lock);
225 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
226 if (tr == this_tr) {
227 tr->ref++;
228 ret = 0;
229 break;
230 }
231 }
232 mutex_unlock(&trace_types_lock);
233
234 return ret;
235}
236
237static void __trace_array_put(struct trace_array *this_tr)
238{
239 WARN_ON(!this_tr->ref);
240 this_tr->ref--;
241}
242
243void trace_array_put(struct trace_array *this_tr)
244{
245 mutex_lock(&trace_types_lock);
246 __trace_array_put(this_tr);
247 mutex_unlock(&trace_types_lock);
248}
249
250int filter_check_discard(struct ftrace_event_file *file, void *rec,
251 struct ring_buffer *buffer,
252 struct ring_buffer_event *event)
253{
254 if (unlikely(file->flags & FTRACE_EVENT_FL_FILTERED) &&
255 !filter_match_preds(file->filter, rec)) {
256 ring_buffer_discard_commit(buffer, event);
257 return 1;
258 }
259
260 return 0;
261}
262EXPORT_SYMBOL_GPL(filter_check_discard);
263
264int call_filter_check_discard(struct ftrace_event_call *call, void *rec,
265 struct ring_buffer *buffer,
266 struct ring_buffer_event *event)
267{
268 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
269 !filter_match_preds(call->filter, rec)) {
270 ring_buffer_discard_commit(buffer, event);
271 return 1;
272 }
273
274 return 0;
275}
276EXPORT_SYMBOL_GPL(call_filter_check_discard);
277
278cycle_t buffer_ftrace_now(struct trace_buffer *buf, int cpu)
279{
280 u64 ts;
281
282 /* Early boot up does not have a buffer yet */
283 if (!buf->buffer)
284 return trace_clock_local();
285
286 ts = ring_buffer_time_stamp(buf->buffer, cpu);
287 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
288
289 return ts;
290}
291
292cycle_t ftrace_now(int cpu)
293{
294 return buffer_ftrace_now(&global_trace.trace_buffer, cpu);
295}
296
297/**
298 * tracing_is_enabled - Show if global_trace has been disabled
299 *
300 * Shows if the global trace has been enabled or not. It uses the
301 * mirror flag "buffer_disabled" to be used in fast paths such as for
302 * the irqsoff tracer. But it may be inaccurate due to races. If you
303 * need to know the accurate state, use tracing_is_on() which is a little
304 * slower, but accurate.
305 */
306int tracing_is_enabled(void)
307{
308 /*
309 * For quick access (irqsoff uses this in fast path), just
310 * return the mirror variable of the state of the ring buffer.
311 * It's a little racy, but we don't really care.
312 */
313 smp_rmb();
314 return !global_trace.buffer_disabled;
315}
316
317/*
318 * trace_buf_size is the size in bytes that is allocated
319 * for a buffer. Note, the number of bytes is always rounded
320 * to page size.
321 *
322 * This number is purposely set to a low number of 16384.
323 * If the dump on oops happens, it will be much appreciated
324 * to not have to wait for all that output. Anyway this can be
325 * boot time and run time configurable.
326 */
327#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
328
329static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
330
331/* trace_types holds a link list of available tracers. */
332static struct tracer *trace_types __read_mostly;
333
334/*
335 * trace_types_lock is used to protect the trace_types list.
336 */
337DEFINE_MUTEX(trace_types_lock);
338
339/*
340 * serialize the access of the ring buffer
341 *
342 * ring buffer serializes readers, but it is low level protection.
343 * The validity of the events (which returns by ring_buffer_peek() ..etc)
344 * are not protected by ring buffer.
345 *
346 * The content of events may become garbage if we allow other process consumes
347 * these events concurrently:
348 * A) the page of the consumed events may become a normal page
349 * (not reader page) in ring buffer, and this page will be rewrited
350 * by events producer.
351 * B) The page of the consumed events may become a page for splice_read,
352 * and this page will be returned to system.
353 *
354 * These primitives allow multi process access to different cpu ring buffer
355 * concurrently.
356 *
357 * These primitives don't distinguish read-only and read-consume access.
358 * Multi read-only access are also serialized.
359 */
360
361#ifdef CONFIG_SMP
362static DECLARE_RWSEM(all_cpu_access_lock);
363static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
364
365static inline void trace_access_lock(int cpu)
366{
367 if (cpu == RING_BUFFER_ALL_CPUS) {
368 /* gain it for accessing the whole ring buffer. */
369 down_write(&all_cpu_access_lock);
370 } else {
371 /* gain it for accessing a cpu ring buffer. */
372
373 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
374 down_read(&all_cpu_access_lock);
375
376 /* Secondly block other access to this @cpu ring buffer. */
377 mutex_lock(&per_cpu(cpu_access_lock, cpu));
378 }
379}
380
381static inline void trace_access_unlock(int cpu)
382{
383 if (cpu == RING_BUFFER_ALL_CPUS) {
384 up_write(&all_cpu_access_lock);
385 } else {
386 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
387 up_read(&all_cpu_access_lock);
388 }
389}
390
391static inline void trace_access_lock_init(void)
392{
393 int cpu;
394
395 for_each_possible_cpu(cpu)
396 mutex_init(&per_cpu(cpu_access_lock, cpu));
397}
398
399#else
400
401static DEFINE_MUTEX(access_lock);
402
403static inline void trace_access_lock(int cpu)
404{
405 (void)cpu;
406 mutex_lock(&access_lock);
407}
408
409static inline void trace_access_unlock(int cpu)
410{
411 (void)cpu;
412 mutex_unlock(&access_lock);
413}
414
415static inline void trace_access_lock_init(void)
416{
417}
418
419#endif
420
421/* trace_flags holds trace_options default values */
422unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
423 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
424 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
425 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | TRACE_ITER_FUNCTION;
426
427static void tracer_tracing_on(struct trace_array *tr)
428{
429 if (tr->trace_buffer.buffer)
430 ring_buffer_record_on(tr->trace_buffer.buffer);
431 /*
432 * This flag is looked at when buffers haven't been allocated
433 * yet, or by some tracers (like irqsoff), that just want to
434 * know if the ring buffer has been disabled, but it can handle
435 * races of where it gets disabled but we still do a record.
436 * As the check is in the fast path of the tracers, it is more
437 * important to be fast than accurate.
438 */
439 tr->buffer_disabled = 0;
440 /* Make the flag seen by readers */
441 smp_wmb();
442}
443
444/**
445 * tracing_on - enable tracing buffers
446 *
447 * This function enables tracing buffers that may have been
448 * disabled with tracing_off.
449 */
450void tracing_on(void)
451{
452 tracer_tracing_on(&global_trace);
453}
454EXPORT_SYMBOL_GPL(tracing_on);
455
456/**
457 * __trace_puts - write a constant string into the trace buffer.
458 * @ip: The address of the caller
459 * @str: The constant string to write
460 * @size: The size of the string.
461 */
462int __trace_puts(unsigned long ip, const char *str, int size)
463{
464 struct ring_buffer_event *event;
465 struct ring_buffer *buffer;
466 struct print_entry *entry;
467 unsigned long irq_flags;
468 int alloc;
469
470 if (unlikely(tracing_selftest_running || tracing_disabled))
471 return 0;
472
473 alloc = sizeof(*entry) + size + 2; /* possible \n added */
474
475 local_save_flags(irq_flags);
476 buffer = global_trace.trace_buffer.buffer;
477 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
478 irq_flags, preempt_count());
479 if (!event)
480 return 0;
481
482 entry = ring_buffer_event_data(event);
483 entry->ip = ip;
484
485 memcpy(&entry->buf, str, size);
486
487 /* Add a newline if necessary */
488 if (entry->buf[size - 1] != '\n') {
489 entry->buf[size] = '\n';
490 entry->buf[size + 1] = '\0';
491 } else
492 entry->buf[size] = '\0';
493
494 __buffer_unlock_commit(buffer, event);
495
496 return size;
497}
498EXPORT_SYMBOL_GPL(__trace_puts);
499
500/**
501 * __trace_bputs - write the pointer to a constant string into trace buffer
502 * @ip: The address of the caller
503 * @str: The constant string to write to the buffer to
504 */
505int __trace_bputs(unsigned long ip, const char *str)
506{
507 struct ring_buffer_event *event;
508 struct ring_buffer *buffer;
509 struct bputs_entry *entry;
510 unsigned long irq_flags;
511 int size = sizeof(struct bputs_entry);
512
513 if (unlikely(tracing_selftest_running || tracing_disabled))
514 return 0;
515
516 local_save_flags(irq_flags);
517 buffer = global_trace.trace_buffer.buffer;
518 event = trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
519 irq_flags, preempt_count());
520 if (!event)
521 return 0;
522
523 entry = ring_buffer_event_data(event);
524 entry->ip = ip;
525 entry->str = str;
526
527 __buffer_unlock_commit(buffer, event);
528
529 return 1;
530}
531EXPORT_SYMBOL_GPL(__trace_bputs);
532
533#ifdef CONFIG_TRACER_SNAPSHOT
534/**
535 * trace_snapshot - take a snapshot of the current buffer.
536 *
537 * This causes a swap between the snapshot buffer and the current live
538 * tracing buffer. You can use this to take snapshots of the live
539 * trace when some condition is triggered, but continue to trace.
540 *
541 * Note, make sure to allocate the snapshot with either
542 * a tracing_snapshot_alloc(), or by doing it manually
543 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
544 *
545 * If the snapshot buffer is not allocated, it will stop tracing.
546 * Basically making a permanent snapshot.
547 */
548void tracing_snapshot(void)
549{
550 struct trace_array *tr = &global_trace;
551 struct tracer *tracer = tr->current_trace;
552 unsigned long flags;
553
554 if (in_nmi()) {
555 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
556 internal_trace_puts("*** snapshot is being ignored ***\n");
557 return;
558 }
559
560 if (!tr->allocated_snapshot) {
561 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
562 internal_trace_puts("*** stopping trace here! ***\n");
563 tracing_off();
564 return;
565 }
566
567 /* Note, snapshot can not be used when the tracer uses it */
568 if (tracer->use_max_tr) {
569 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
570 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
571 return;
572 }
573
574 local_irq_save(flags);
575 update_max_tr(tr, current, smp_processor_id());
576 local_irq_restore(flags);
577}
578EXPORT_SYMBOL_GPL(tracing_snapshot);
579
580static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
581 struct trace_buffer *size_buf, int cpu_id);
582static void set_buffer_entries(struct trace_buffer *buf, unsigned long val);
583
584static int alloc_snapshot(struct trace_array *tr)
585{
586 int ret;
587
588 if (!tr->allocated_snapshot) {
589
590 /* allocate spare buffer */
591 ret = resize_buffer_duplicate_size(&tr->max_buffer,
592 &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
593 if (ret < 0)
594 return ret;
595
596 tr->allocated_snapshot = true;
597 }
598
599 return 0;
600}
601
602void free_snapshot(struct trace_array *tr)
603{
604 /*
605 * We don't free the ring buffer. instead, resize it because
606 * The max_tr ring buffer has some state (e.g. ring->clock) and
607 * we want preserve it.
608 */
609 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
610 set_buffer_entries(&tr->max_buffer, 1);
611 tracing_reset_online_cpus(&tr->max_buffer);
612 tr->allocated_snapshot = false;
613}
614
615/**
616 * tracing_alloc_snapshot - allocate snapshot buffer.
617 *
618 * This only allocates the snapshot buffer if it isn't already
619 * allocated - it doesn't also take a snapshot.
620 *
621 * This is meant to be used in cases where the snapshot buffer needs
622 * to be set up for events that can't sleep but need to be able to
623 * trigger a snapshot.
624 */
625int tracing_alloc_snapshot(void)
626{
627 struct trace_array *tr = &global_trace;
628 int ret;
629
630 ret = alloc_snapshot(tr);
631 WARN_ON(ret < 0);
632
633 return ret;
634}
635EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
636
637/**
638 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
639 *
640 * This is similar to trace_snapshot(), but it will allocate the
641 * snapshot buffer if it isn't already allocated. Use this only
642 * where it is safe to sleep, as the allocation may sleep.
643 *
644 * This causes a swap between the snapshot buffer and the current live
645 * tracing buffer. You can use this to take snapshots of the live
646 * trace when some condition is triggered, but continue to trace.
647 */
648void tracing_snapshot_alloc(void)
649{
650 int ret;
651
652 ret = tracing_alloc_snapshot();
653 if (ret < 0)
654 return;
655
656 tracing_snapshot();
657}
658EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
659#else
660void tracing_snapshot(void)
661{
662 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
663}
664EXPORT_SYMBOL_GPL(tracing_snapshot);
665int tracing_alloc_snapshot(void)
666{
667 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
668 return -ENODEV;
669}
670EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
671void tracing_snapshot_alloc(void)
672{
673 /* Give warning */
674 tracing_snapshot();
675}
676EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
677#endif /* CONFIG_TRACER_SNAPSHOT */
678
679static void tracer_tracing_off(struct trace_array *tr)
680{
681 if (tr->trace_buffer.buffer)
682 ring_buffer_record_off(tr->trace_buffer.buffer);
683 /*
684 * This flag is looked at when buffers haven't been allocated
685 * yet, or by some tracers (like irqsoff), that just want to
686 * know if the ring buffer has been disabled, but it can handle
687 * races of where it gets disabled but we still do a record.
688 * As the check is in the fast path of the tracers, it is more
689 * important to be fast than accurate.
690 */
691 tr->buffer_disabled = 1;
692 /* Make the flag seen by readers */
693 smp_wmb();
694}
695
696/**
697 * tracing_off - turn off tracing buffers
698 *
699 * This function stops the tracing buffers from recording data.
700 * It does not disable any overhead the tracers themselves may
701 * be causing. This function simply causes all recording to
702 * the ring buffers to fail.
703 */
704void tracing_off(void)
705{
706 tracer_tracing_off(&global_trace);
707}
708EXPORT_SYMBOL_GPL(tracing_off);
709
710void disable_trace_on_warning(void)
711{
712 if (__disable_trace_on_warning)
713 tracing_off();
714}
715
716/**
717 * tracer_tracing_is_on - show real state of ring buffer enabled
718 * @tr : the trace array to know if ring buffer is enabled
719 *
720 * Shows real state of the ring buffer if it is enabled or not.
721 */
722static int tracer_tracing_is_on(struct trace_array *tr)
723{
724 if (tr->trace_buffer.buffer)
725 return ring_buffer_record_is_on(tr->trace_buffer.buffer);
726 return !tr->buffer_disabled;
727}
728
729/**
730 * tracing_is_on - show state of ring buffers enabled
731 */
732int tracing_is_on(void)
733{
734 return tracer_tracing_is_on(&global_trace);
735}
736EXPORT_SYMBOL_GPL(tracing_is_on);
737
738static int __init set_buf_size(char *str)
739{
740 unsigned long buf_size;
741
742 if (!str)
743 return 0;
744 buf_size = memparse(str, &str);
745 /* nr_entries can not be zero */
746 if (buf_size == 0)
747 return 0;
748 trace_buf_size = buf_size;
749 return 1;
750}
751__setup("trace_buf_size=", set_buf_size);
752
753static int __init set_tracing_thresh(char *str)
754{
755 unsigned long threshold;
756 int ret;
757
758 if (!str)
759 return 0;
760 ret = kstrtoul(str, 0, &threshold);
761 if (ret < 0)
762 return 0;
763 tracing_thresh = threshold * 1000;
764 return 1;
765}
766__setup("tracing_thresh=", set_tracing_thresh);
767
768unsigned long nsecs_to_usecs(unsigned long nsecs)
769{
770 return nsecs / 1000;
771}
772
773/* These must match the bit postions in trace_iterator_flags */
774static const char *trace_options[] = {
775 "print-parent",
776 "sym-offset",
777 "sym-addr",
778 "verbose",
779 "raw",
780 "hex",
781 "bin",
782 "block",
783 "stacktrace",
784 "trace_printk",
785 "ftrace_preempt",
786 "branch",
787 "annotate",
788 "userstacktrace",
789 "sym-userobj",
790 "printk-msg-only",
791 "context-info",
792 "latency-format",
793 "sleep-time",
794 "graph-time",
795 "record-cmd",
796 "overwrite",
797 "disable_on_free",
798 "irq-info",
799 "markers",
800 "function-trace",
801 NULL
802};
803
804static struct {
805 u64 (*func)(void);
806 const char *name;
807 int in_ns; /* is this clock in nanoseconds? */
808} trace_clocks[] = {
809 { trace_clock_local, "local", 1 },
810 { trace_clock_global, "global", 1 },
811 { trace_clock_counter, "counter", 0 },
812 { trace_clock_jiffies, "uptime", 1 },
813 { trace_clock, "perf", 1 },
814 ARCH_TRACE_CLOCKS
815};
816
817/*
818 * trace_parser_get_init - gets the buffer for trace parser
819 */
820int trace_parser_get_init(struct trace_parser *parser, int size)
821{
822 memset(parser, 0, sizeof(*parser));
823
824 parser->buffer = kmalloc(size, GFP_KERNEL);
825 if (!parser->buffer)
826 return 1;
827
828 parser->size = size;
829 return 0;
830}
831
832/*
833 * trace_parser_put - frees the buffer for trace parser
834 */
835void trace_parser_put(struct trace_parser *parser)
836{
837 kfree(parser->buffer);
838}
839
840/*
841 * trace_get_user - reads the user input string separated by space
842 * (matched by isspace(ch))
843 *
844 * For each string found the 'struct trace_parser' is updated,
845 * and the function returns.
846 *
847 * Returns number of bytes read.
848 *
849 * See kernel/trace/trace.h for 'struct trace_parser' details.
850 */
851int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
852 size_t cnt, loff_t *ppos)
853{
854 char ch;
855 size_t read = 0;
856 ssize_t ret;
857
858 if (!*ppos)
859 trace_parser_clear(parser);
860
861 ret = get_user(ch, ubuf++);
862 if (ret)
863 goto out;
864
865 read++;
866 cnt--;
867
868 /*
869 * The parser is not finished with the last write,
870 * continue reading the user input without skipping spaces.
871 */
872 if (!parser->cont) {
873 /* skip white space */
874 while (cnt && isspace(ch)) {
875 ret = get_user(ch, ubuf++);
876 if (ret)
877 goto out;
878 read++;
879 cnt--;
880 }
881
882 /* only spaces were written */
883 if (isspace(ch)) {
884 *ppos += read;
885 ret = read;
886 goto out;
887 }
888
889 parser->idx = 0;
890 }
891
892 /* read the non-space input */
893 while (cnt && !isspace(ch)) {
894 if (parser->idx < parser->size - 1)
895 parser->buffer[parser->idx++] = ch;
896 else {
897 ret = -EINVAL;
898 goto out;
899 }
900 ret = get_user(ch, ubuf++);
901 if (ret)
902 goto out;
903 read++;
904 cnt--;
905 }
906
907 /* We either got finished input or we have to wait for another call. */
908 if (isspace(ch)) {
909 parser->buffer[parser->idx] = 0;
910 parser->cont = false;
911 } else if (parser->idx < parser->size - 1) {
912 parser->cont = true;
913 parser->buffer[parser->idx++] = ch;
914 } else {
915 ret = -EINVAL;
916 goto out;
917 }
918
919 *ppos += read;
920 ret = read;
921
922out:
923 return ret;
924}
925
926ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
927{
928 int len;
929 int ret;
930
931 if (!cnt)
932 return 0;
933
934 if (s->len <= s->readpos)
935 return -EBUSY;
936
937 len = s->len - s->readpos;
938 if (cnt > len)
939 cnt = len;
940 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
941 if (ret == cnt)
942 return -EFAULT;
943
944 cnt -= ret;
945
946 s->readpos += cnt;
947 return cnt;
948}
949
950static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
951{
952 int len;
953
954 if (s->len <= s->readpos)
955 return -EBUSY;
956
957 len = s->len - s->readpos;
958 if (cnt > len)
959 cnt = len;
960 memcpy(buf, s->buffer + s->readpos, cnt);
961
962 s->readpos += cnt;
963 return cnt;
964}
965
966/*
967 * ftrace_max_lock is used to protect the swapping of buffers
968 * when taking a max snapshot. The buffers themselves are
969 * protected by per_cpu spinlocks. But the action of the swap
970 * needs its own lock.
971 *
972 * This is defined as a arch_spinlock_t in order to help
973 * with performance when lockdep debugging is enabled.
974 *
975 * It is also used in other places outside the update_max_tr
976 * so it needs to be defined outside of the
977 * CONFIG_TRACER_MAX_TRACE.
978 */
979static arch_spinlock_t ftrace_max_lock =
980 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
981
982unsigned long __read_mostly tracing_thresh;
983
984#ifdef CONFIG_TRACER_MAX_TRACE
985unsigned long __read_mostly tracing_max_latency;
986
987/*
988 * Copy the new maximum trace into the separate maximum-trace
989 * structure. (this way the maximum trace is permanently saved,
990 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
991 */
992static void
993__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
994{
995 struct trace_buffer *trace_buf = &tr->trace_buffer;
996 struct trace_buffer *max_buf = &tr->max_buffer;
997 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
998 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
999
1000 max_buf->cpu = cpu;
1001 max_buf->time_start = data->preempt_timestamp;
1002
1003 max_data->saved_latency = tracing_max_latency;
1004 max_data->critical_start = data->critical_start;
1005 max_data->critical_end = data->critical_end;
1006
1007 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1008 max_data->pid = tsk->pid;
1009 /*
1010 * If tsk == current, then use current_uid(), as that does not use
1011 * RCU. The irq tracer can be called out of RCU scope.
1012 */
1013 if (tsk == current)
1014 max_data->uid = current_uid();
1015 else
1016 max_data->uid = task_uid(tsk);
1017
1018 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1019 max_data->policy = tsk->policy;
1020 max_data->rt_priority = tsk->rt_priority;
1021
1022 /* record this tasks comm */
1023 tracing_record_cmdline(tsk);
1024}
1025
1026/**
1027 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1028 * @tr: tracer
1029 * @tsk: the task with the latency
1030 * @cpu: The cpu that initiated the trace.
1031 *
1032 * Flip the buffers between the @tr and the max_tr and record information
1033 * about which task was the cause of this latency.
1034 */
1035void
1036update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1037{
1038 struct ring_buffer *buf;
1039
1040 if (tr->stop_count)
1041 return;
1042
1043 WARN_ON_ONCE(!irqs_disabled());
1044
1045 if (!tr->allocated_snapshot) {
1046 /* Only the nop tracer should hit this when disabling */
1047 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1048 return;
1049 }
1050
1051 arch_spin_lock(&ftrace_max_lock);
1052
1053 buf = tr->trace_buffer.buffer;
1054 tr->trace_buffer.buffer = tr->max_buffer.buffer;
1055 tr->max_buffer.buffer = buf;
1056
1057 __update_max_tr(tr, tsk, cpu);
1058 arch_spin_unlock(&ftrace_max_lock);
1059}
1060
1061/**
1062 * update_max_tr_single - only copy one trace over, and reset the rest
1063 * @tr - tracer
1064 * @tsk - task with the latency
1065 * @cpu - the cpu of the buffer to copy.
1066 *
1067 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1068 */
1069void
1070update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1071{
1072 int ret;
1073
1074 if (tr->stop_count)
1075 return;
1076
1077 WARN_ON_ONCE(!irqs_disabled());
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(&ftrace_max_lock);
1085
1086 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1087
1088 if (ret == -EBUSY) {
1089 /*
1090 * We failed to swap the buffer due to a commit taking
1091 * place on this CPU. We fail to record, but we reset
1092 * the max trace buffer (no one writes directly to it)
1093 * and flag that it failed.
1094 */
1095 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1096 "Failed to swap buffers due to commit in progress\n");
1097 }
1098
1099 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1100
1101 __update_max_tr(tr, tsk, cpu);
1102 arch_spin_unlock(&ftrace_max_lock);
1103}
1104#endif /* CONFIG_TRACER_MAX_TRACE */
1105
1106static void default_wait_pipe(struct trace_iterator *iter)
1107{
1108 /* Iterators are static, they should be filled or empty */
1109 if (trace_buffer_iter(iter, iter->cpu_file))
1110 return;
1111
1112 ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
1113}
1114
1115#ifdef CONFIG_FTRACE_STARTUP_TEST
1116static int run_tracer_selftest(struct tracer *type)
1117{
1118 struct trace_array *tr = &global_trace;
1119 struct tracer *saved_tracer = tr->current_trace;
1120 int ret;
1121
1122 if (!type->selftest || tracing_selftest_disabled)
1123 return 0;
1124
1125 /*
1126 * Run a selftest on this tracer.
1127 * Here we reset the trace buffer, and set the current
1128 * tracer to be this tracer. The tracer can then run some
1129 * internal tracing to verify that everything is in order.
1130 * If we fail, we do not register this tracer.
1131 */
1132 tracing_reset_online_cpus(&tr->trace_buffer);
1133
1134 tr->current_trace = type;
1135
1136#ifdef CONFIG_TRACER_MAX_TRACE
1137 if (type->use_max_tr) {
1138 /* If we expanded the buffers, make sure the max is expanded too */
1139 if (ring_buffer_expanded)
1140 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1141 RING_BUFFER_ALL_CPUS);
1142 tr->allocated_snapshot = true;
1143 }
1144#endif
1145
1146 /* the test is responsible for initializing and enabling */
1147 pr_info("Testing tracer %s: ", type->name);
1148 ret = type->selftest(type, tr);
1149 /* the test is responsible for resetting too */
1150 tr->current_trace = saved_tracer;
1151 if (ret) {
1152 printk(KERN_CONT "FAILED!\n");
1153 /* Add the warning after printing 'FAILED' */
1154 WARN_ON(1);
1155 return -1;
1156 }
1157 /* Only reset on passing, to avoid touching corrupted buffers */
1158 tracing_reset_online_cpus(&tr->trace_buffer);
1159
1160#ifdef CONFIG_TRACER_MAX_TRACE
1161 if (type->use_max_tr) {
1162 tr->allocated_snapshot = false;
1163
1164 /* Shrink the max buffer again */
1165 if (ring_buffer_expanded)
1166 ring_buffer_resize(tr->max_buffer.buffer, 1,
1167 RING_BUFFER_ALL_CPUS);
1168 }
1169#endif
1170
1171 printk(KERN_CONT "PASSED\n");
1172 return 0;
1173}
1174#else
1175static inline int run_tracer_selftest(struct tracer *type)
1176{
1177 return 0;
1178}
1179#endif /* CONFIG_FTRACE_STARTUP_TEST */
1180
1181/**
1182 * register_tracer - register a tracer with the ftrace system.
1183 * @type - the plugin for the tracer
1184 *
1185 * Register a new plugin tracer.
1186 */
1187int register_tracer(struct tracer *type)
1188{
1189 struct tracer *t;
1190 int ret = 0;
1191
1192 if (!type->name) {
1193 pr_info("Tracer must have a name\n");
1194 return -1;
1195 }
1196
1197 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1198 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1199 return -1;
1200 }
1201
1202 mutex_lock(&trace_types_lock);
1203
1204 tracing_selftest_running = true;
1205
1206 for (t = trace_types; t; t = t->next) {
1207 if (strcmp(type->name, t->name) == 0) {
1208 /* already found */
1209 pr_info("Tracer %s already registered\n",
1210 type->name);
1211 ret = -1;
1212 goto out;
1213 }
1214 }
1215
1216 if (!type->set_flag)
1217 type->set_flag = &dummy_set_flag;
1218 if (!type->flags)
1219 type->flags = &dummy_tracer_flags;
1220 else
1221 if (!type->flags->opts)
1222 type->flags->opts = dummy_tracer_opt;
1223 if (!type->wait_pipe)
1224 type->wait_pipe = default_wait_pipe;
1225
1226 ret = run_tracer_selftest(type);
1227 if (ret < 0)
1228 goto out;
1229
1230 type->next = trace_types;
1231 trace_types = type;
1232
1233 out:
1234 tracing_selftest_running = false;
1235 mutex_unlock(&trace_types_lock);
1236
1237 if (ret || !default_bootup_tracer)
1238 goto out_unlock;
1239
1240 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1241 goto out_unlock;
1242
1243 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1244 /* Do we want this tracer to start on bootup? */
1245 tracing_set_tracer(&global_trace, type->name);
1246 default_bootup_tracer = NULL;
1247 /* disable other selftests, since this will break it. */
1248 tracing_selftest_disabled = true;
1249#ifdef CONFIG_FTRACE_STARTUP_TEST
1250 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1251 type->name);
1252#endif
1253
1254 out_unlock:
1255 return ret;
1256}
1257
1258void tracing_reset(struct trace_buffer *buf, int cpu)
1259{
1260 struct ring_buffer *buffer = buf->buffer;
1261
1262 if (!buffer)
1263 return;
1264
1265 ring_buffer_record_disable(buffer);
1266
1267 /* Make sure all commits have finished */
1268 synchronize_sched();
1269 ring_buffer_reset_cpu(buffer, cpu);
1270
1271 ring_buffer_record_enable(buffer);
1272}
1273
1274void tracing_reset_online_cpus(struct trace_buffer *buf)
1275{
1276 struct ring_buffer *buffer = buf->buffer;
1277 int cpu;
1278
1279 if (!buffer)
1280 return;
1281
1282 ring_buffer_record_disable(buffer);
1283
1284 /* Make sure all commits have finished */
1285 synchronize_sched();
1286
1287 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
1288
1289 for_each_online_cpu(cpu)
1290 ring_buffer_reset_cpu(buffer, cpu);
1291
1292 ring_buffer_record_enable(buffer);
1293}
1294
1295/* Must have trace_types_lock held */
1296void tracing_reset_all_online_cpus(void)
1297{
1298 struct trace_array *tr;
1299
1300 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1301 tracing_reset_online_cpus(&tr->trace_buffer);
1302#ifdef CONFIG_TRACER_MAX_TRACE
1303 tracing_reset_online_cpus(&tr->max_buffer);
1304#endif
1305 }
1306}
1307
1308#define SAVED_CMDLINES 128
1309#define NO_CMDLINE_MAP UINT_MAX
1310static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1311static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
1312static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
1313static int cmdline_idx;
1314static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1315
1316/* temporary disable recording */
1317static atomic_t trace_record_cmdline_disabled __read_mostly;
1318
1319static void trace_init_cmdlines(void)
1320{
1321 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
1322 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
1323 cmdline_idx = 0;
1324}
1325
1326int is_tracing_stopped(void)
1327{
1328 return global_trace.stop_count;
1329}
1330
1331/**
1332 * tracing_start - quick start of the tracer
1333 *
1334 * If tracing is enabled but was stopped by tracing_stop,
1335 * this will start the tracer back up.
1336 */
1337void tracing_start(void)
1338{
1339 struct ring_buffer *buffer;
1340 unsigned long flags;
1341
1342 if (tracing_disabled)
1343 return;
1344
1345 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1346 if (--global_trace.stop_count) {
1347 if (global_trace.stop_count < 0) {
1348 /* Someone screwed up their debugging */
1349 WARN_ON_ONCE(1);
1350 global_trace.stop_count = 0;
1351 }
1352 goto out;
1353 }
1354
1355 /* Prevent the buffers from switching */
1356 arch_spin_lock(&ftrace_max_lock);
1357
1358 buffer = global_trace.trace_buffer.buffer;
1359 if (buffer)
1360 ring_buffer_record_enable(buffer);
1361
1362#ifdef CONFIG_TRACER_MAX_TRACE
1363 buffer = global_trace.max_buffer.buffer;
1364 if (buffer)
1365 ring_buffer_record_enable(buffer);
1366#endif
1367
1368 arch_spin_unlock(&ftrace_max_lock);
1369
1370 ftrace_start();
1371 out:
1372 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1373}
1374
1375static void tracing_start_tr(struct trace_array *tr)
1376{
1377 struct ring_buffer *buffer;
1378 unsigned long flags;
1379
1380 if (tracing_disabled)
1381 return;
1382
1383 /* If global, we need to also start the max tracer */
1384 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1385 return tracing_start();
1386
1387 raw_spin_lock_irqsave(&tr->start_lock, flags);
1388
1389 if (--tr->stop_count) {
1390 if (tr->stop_count < 0) {
1391 /* Someone screwed up their debugging */
1392 WARN_ON_ONCE(1);
1393 tr->stop_count = 0;
1394 }
1395 goto out;
1396 }
1397
1398 buffer = tr->trace_buffer.buffer;
1399 if (buffer)
1400 ring_buffer_record_enable(buffer);
1401
1402 out:
1403 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1404}
1405
1406/**
1407 * tracing_stop - quick stop of the tracer
1408 *
1409 * Light weight way to stop tracing. Use in conjunction with
1410 * tracing_start.
1411 */
1412void tracing_stop(void)
1413{
1414 struct ring_buffer *buffer;
1415 unsigned long flags;
1416
1417 ftrace_stop();
1418 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1419 if (global_trace.stop_count++)
1420 goto out;
1421
1422 /* Prevent the buffers from switching */
1423 arch_spin_lock(&ftrace_max_lock);
1424
1425 buffer = global_trace.trace_buffer.buffer;
1426 if (buffer)
1427 ring_buffer_record_disable(buffer);
1428
1429#ifdef CONFIG_TRACER_MAX_TRACE
1430 buffer = global_trace.max_buffer.buffer;
1431 if (buffer)
1432 ring_buffer_record_disable(buffer);
1433#endif
1434
1435 arch_spin_unlock(&ftrace_max_lock);
1436
1437 out:
1438 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1439}
1440
1441static void tracing_stop_tr(struct trace_array *tr)
1442{
1443 struct ring_buffer *buffer;
1444 unsigned long flags;
1445
1446 /* If global, we need to also stop the max tracer */
1447 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1448 return tracing_stop();
1449
1450 raw_spin_lock_irqsave(&tr->start_lock, flags);
1451 if (tr->stop_count++)
1452 goto out;
1453
1454 buffer = tr->trace_buffer.buffer;
1455 if (buffer)
1456 ring_buffer_record_disable(buffer);
1457
1458 out:
1459 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1460}
1461
1462void trace_stop_cmdline_recording(void);
1463
1464static void trace_save_cmdline(struct task_struct *tsk)
1465{
1466 unsigned pid, idx;
1467
1468 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1469 return;
1470
1471 /*
1472 * It's not the end of the world if we don't get
1473 * the lock, but we also don't want to spin
1474 * nor do we want to disable interrupts,
1475 * so if we miss here, then better luck next time.
1476 */
1477 if (!arch_spin_trylock(&trace_cmdline_lock))
1478 return;
1479
1480 idx = map_pid_to_cmdline[tsk->pid];
1481 if (idx == NO_CMDLINE_MAP) {
1482 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1483
1484 /*
1485 * Check whether the cmdline buffer at idx has a pid
1486 * mapped. We are going to overwrite that entry so we
1487 * need to clear the map_pid_to_cmdline. Otherwise we
1488 * would read the new comm for the old pid.
1489 */
1490 pid = map_cmdline_to_pid[idx];
1491 if (pid != NO_CMDLINE_MAP)
1492 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1493
1494 map_cmdline_to_pid[idx] = tsk->pid;
1495 map_pid_to_cmdline[tsk->pid] = idx;
1496
1497 cmdline_idx = idx;
1498 }
1499
1500 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1501
1502 arch_spin_unlock(&trace_cmdline_lock);
1503}
1504
1505void trace_find_cmdline(int pid, char comm[])
1506{
1507 unsigned map;
1508
1509 if (!pid) {
1510 strcpy(comm, "<idle>");
1511 return;
1512 }
1513
1514 if (WARN_ON_ONCE(pid < 0)) {
1515 strcpy(comm, "<XXX>");
1516 return;
1517 }
1518
1519 if (pid > PID_MAX_DEFAULT) {
1520 strcpy(comm, "<...>");
1521 return;
1522 }
1523
1524 preempt_disable();
1525 arch_spin_lock(&trace_cmdline_lock);
1526 map = map_pid_to_cmdline[pid];
1527 if (map != NO_CMDLINE_MAP)
1528 strcpy(comm, saved_cmdlines[map]);
1529 else
1530 strcpy(comm, "<...>");
1531
1532 arch_spin_unlock(&trace_cmdline_lock);
1533 preempt_enable();
1534}
1535
1536void tracing_record_cmdline(struct task_struct *tsk)
1537{
1538 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1539 return;
1540
1541 if (!__this_cpu_read(trace_cmdline_save))
1542 return;
1543
1544 __this_cpu_write(trace_cmdline_save, false);
1545
1546 trace_save_cmdline(tsk);
1547}
1548
1549void
1550tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1551 int pc)
1552{
1553 struct task_struct *tsk = current;
1554
1555 entry->preempt_count = pc & 0xff;
1556 entry->pid = (tsk) ? tsk->pid : 0;
1557 entry->flags =
1558#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1559 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1560#else
1561 TRACE_FLAG_IRQS_NOSUPPORT |
1562#endif
1563 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1564 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1565 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
1566 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
1567}
1568EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1569
1570struct ring_buffer_event *
1571trace_buffer_lock_reserve(struct ring_buffer *buffer,
1572 int type,
1573 unsigned long len,
1574 unsigned long flags, int pc)
1575{
1576 struct ring_buffer_event *event;
1577
1578 event = ring_buffer_lock_reserve(buffer, len);
1579 if (event != NULL) {
1580 struct trace_entry *ent = ring_buffer_event_data(event);
1581
1582 tracing_generic_entry_update(ent, flags, pc);
1583 ent->type = type;
1584 }
1585
1586 return event;
1587}
1588
1589void
1590__buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1591{
1592 __this_cpu_write(trace_cmdline_save, true);
1593 ring_buffer_unlock_commit(buffer, event);
1594}
1595
1596static inline void
1597__trace_buffer_unlock_commit(struct ring_buffer *buffer,
1598 struct ring_buffer_event *event,
1599 unsigned long flags, int pc)
1600{
1601 __buffer_unlock_commit(buffer, event);
1602
1603 ftrace_trace_stack(buffer, flags, 6, pc);
1604 ftrace_trace_userstack(buffer, flags, pc);
1605}
1606
1607void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1608 struct ring_buffer_event *event,
1609 unsigned long flags, int pc)
1610{
1611 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1612}
1613EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1614
1615static struct ring_buffer *temp_buffer;
1616
1617struct ring_buffer_event *
1618trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1619 struct ftrace_event_file *ftrace_file,
1620 int type, unsigned long len,
1621 unsigned long flags, int pc)
1622{
1623 struct ring_buffer_event *entry;
1624
1625 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1626 entry = trace_buffer_lock_reserve(*current_rb,
1627 type, len, flags, pc);
1628 /*
1629 * If tracing is off, but we have triggers enabled
1630 * we still need to look at the event data. Use the temp_buffer
1631 * to store the trace event for the tigger to use. It's recusive
1632 * safe and will not be recorded anywhere.
1633 */
1634 if (!entry && ftrace_file->flags & FTRACE_EVENT_FL_TRIGGER_COND) {
1635 *current_rb = temp_buffer;
1636 entry = trace_buffer_lock_reserve(*current_rb,
1637 type, len, flags, pc);
1638 }
1639 return entry;
1640}
1641EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1642
1643struct ring_buffer_event *
1644trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1645 int type, unsigned long len,
1646 unsigned long flags, int pc)
1647{
1648 *current_rb = global_trace.trace_buffer.buffer;
1649 return trace_buffer_lock_reserve(*current_rb,
1650 type, len, flags, pc);
1651}
1652EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1653
1654void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1655 struct ring_buffer_event *event,
1656 unsigned long flags, int pc)
1657{
1658 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1659}
1660EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1661
1662void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1663 struct ring_buffer_event *event,
1664 unsigned long flags, int pc,
1665 struct pt_regs *regs)
1666{
1667 __buffer_unlock_commit(buffer, event);
1668
1669 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1670 ftrace_trace_userstack(buffer, flags, pc);
1671}
1672EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1673
1674void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1675 struct ring_buffer_event *event)
1676{
1677 ring_buffer_discard_commit(buffer, event);
1678}
1679EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1680
1681void
1682trace_function(struct trace_array *tr,
1683 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1684 int pc)
1685{
1686 struct ftrace_event_call *call = &event_function;
1687 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1688 struct ring_buffer_event *event;
1689 struct ftrace_entry *entry;
1690
1691 /* If we are reading the ring buffer, don't trace */
1692 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1693 return;
1694
1695 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1696 flags, pc);
1697 if (!event)
1698 return;
1699 entry = ring_buffer_event_data(event);
1700 entry->ip = ip;
1701 entry->parent_ip = parent_ip;
1702
1703 if (!call_filter_check_discard(call, entry, buffer, event))
1704 __buffer_unlock_commit(buffer, event);
1705}
1706
1707#ifdef CONFIG_STACKTRACE
1708
1709#define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1710struct ftrace_stack {
1711 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1712};
1713
1714static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1715static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1716
1717static void __ftrace_trace_stack(struct ring_buffer *buffer,
1718 unsigned long flags,
1719 int skip, int pc, struct pt_regs *regs)
1720{
1721 struct ftrace_event_call *call = &event_kernel_stack;
1722 struct ring_buffer_event *event;
1723 struct stack_entry *entry;
1724 struct stack_trace trace;
1725 int use_stack;
1726 int size = FTRACE_STACK_ENTRIES;
1727
1728 trace.nr_entries = 0;
1729 trace.skip = skip;
1730
1731 /*
1732 * Since events can happen in NMIs there's no safe way to
1733 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1734 * or NMI comes in, it will just have to use the default
1735 * FTRACE_STACK_SIZE.
1736 */
1737 preempt_disable_notrace();
1738
1739 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1740 /*
1741 * We don't need any atomic variables, just a barrier.
1742 * If an interrupt comes in, we don't care, because it would
1743 * have exited and put the counter back to what we want.
1744 * We just need a barrier to keep gcc from moving things
1745 * around.
1746 */
1747 barrier();
1748 if (use_stack == 1) {
1749 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1750 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1751
1752 if (regs)
1753 save_stack_trace_regs(regs, &trace);
1754 else
1755 save_stack_trace(&trace);
1756
1757 if (trace.nr_entries > size)
1758 size = trace.nr_entries;
1759 } else
1760 /* From now on, use_stack is a boolean */
1761 use_stack = 0;
1762
1763 size *= sizeof(unsigned long);
1764
1765 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1766 sizeof(*entry) + size, flags, pc);
1767 if (!event)
1768 goto out;
1769 entry = ring_buffer_event_data(event);
1770
1771 memset(&entry->caller, 0, size);
1772
1773 if (use_stack)
1774 memcpy(&entry->caller, trace.entries,
1775 trace.nr_entries * sizeof(unsigned long));
1776 else {
1777 trace.max_entries = FTRACE_STACK_ENTRIES;
1778 trace.entries = entry->caller;
1779 if (regs)
1780 save_stack_trace_regs(regs, &trace);
1781 else
1782 save_stack_trace(&trace);
1783 }
1784
1785 entry->size = trace.nr_entries;
1786
1787 if (!call_filter_check_discard(call, entry, buffer, event))
1788 __buffer_unlock_commit(buffer, event);
1789
1790 out:
1791 /* Again, don't let gcc optimize things here */
1792 barrier();
1793 __this_cpu_dec(ftrace_stack_reserve);
1794 preempt_enable_notrace();
1795
1796}
1797
1798void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1799 int skip, int pc, struct pt_regs *regs)
1800{
1801 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1802 return;
1803
1804 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1805}
1806
1807void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1808 int skip, int pc)
1809{
1810 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1811 return;
1812
1813 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1814}
1815
1816void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1817 int pc)
1818{
1819 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1820}
1821
1822/**
1823 * trace_dump_stack - record a stack back trace in the trace buffer
1824 * @skip: Number of functions to skip (helper handlers)
1825 */
1826void trace_dump_stack(int skip)
1827{
1828 unsigned long flags;
1829
1830 if (tracing_disabled || tracing_selftest_running)
1831 return;
1832
1833 local_save_flags(flags);
1834
1835 /*
1836 * Skip 3 more, seems to get us at the caller of
1837 * this function.
1838 */
1839 skip += 3;
1840 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1841 flags, skip, preempt_count(), NULL);
1842}
1843
1844static DEFINE_PER_CPU(int, user_stack_count);
1845
1846void
1847ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1848{
1849 struct ftrace_event_call *call = &event_user_stack;
1850 struct ring_buffer_event *event;
1851 struct userstack_entry *entry;
1852 struct stack_trace trace;
1853
1854 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1855 return;
1856
1857 /*
1858 * NMIs can not handle page faults, even with fix ups.
1859 * The save user stack can (and often does) fault.
1860 */
1861 if (unlikely(in_nmi()))
1862 return;
1863
1864 /*
1865 * prevent recursion, since the user stack tracing may
1866 * trigger other kernel events.
1867 */
1868 preempt_disable();
1869 if (__this_cpu_read(user_stack_count))
1870 goto out;
1871
1872 __this_cpu_inc(user_stack_count);
1873
1874 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1875 sizeof(*entry), flags, pc);
1876 if (!event)
1877 goto out_drop_count;
1878 entry = ring_buffer_event_data(event);
1879
1880 entry->tgid = current->tgid;
1881 memset(&entry->caller, 0, sizeof(entry->caller));
1882
1883 trace.nr_entries = 0;
1884 trace.max_entries = FTRACE_STACK_ENTRIES;
1885 trace.skip = 0;
1886 trace.entries = entry->caller;
1887
1888 save_stack_trace_user(&trace);
1889 if (!call_filter_check_discard(call, entry, buffer, event))
1890 __buffer_unlock_commit(buffer, event);
1891
1892 out_drop_count:
1893 __this_cpu_dec(user_stack_count);
1894 out:
1895 preempt_enable();
1896}
1897
1898#ifdef UNUSED
1899static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1900{
1901 ftrace_trace_userstack(tr, flags, preempt_count());
1902}
1903#endif /* UNUSED */
1904
1905#endif /* CONFIG_STACKTRACE */
1906
1907/* created for use with alloc_percpu */
1908struct trace_buffer_struct {
1909 char buffer[TRACE_BUF_SIZE];
1910};
1911
1912static struct trace_buffer_struct *trace_percpu_buffer;
1913static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1914static struct trace_buffer_struct *trace_percpu_irq_buffer;
1915static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1916
1917/*
1918 * The buffer used is dependent on the context. There is a per cpu
1919 * buffer for normal context, softirq contex, hard irq context and
1920 * for NMI context. Thise allows for lockless recording.
1921 *
1922 * Note, if the buffers failed to be allocated, then this returns NULL
1923 */
1924static char *get_trace_buf(void)
1925{
1926 struct trace_buffer_struct *percpu_buffer;
1927
1928 /*
1929 * If we have allocated per cpu buffers, then we do not
1930 * need to do any locking.
1931 */
1932 if (in_nmi())
1933 percpu_buffer = trace_percpu_nmi_buffer;
1934 else if (in_irq())
1935 percpu_buffer = trace_percpu_irq_buffer;
1936 else if (in_softirq())
1937 percpu_buffer = trace_percpu_sirq_buffer;
1938 else
1939 percpu_buffer = trace_percpu_buffer;
1940
1941 if (!percpu_buffer)
1942 return NULL;
1943
1944 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1945}
1946
1947static int alloc_percpu_trace_buffer(void)
1948{
1949 struct trace_buffer_struct *buffers;
1950 struct trace_buffer_struct *sirq_buffers;
1951 struct trace_buffer_struct *irq_buffers;
1952 struct trace_buffer_struct *nmi_buffers;
1953
1954 buffers = alloc_percpu(struct trace_buffer_struct);
1955 if (!buffers)
1956 goto err_warn;
1957
1958 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1959 if (!sirq_buffers)
1960 goto err_sirq;
1961
1962 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1963 if (!irq_buffers)
1964 goto err_irq;
1965
1966 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1967 if (!nmi_buffers)
1968 goto err_nmi;
1969
1970 trace_percpu_buffer = buffers;
1971 trace_percpu_sirq_buffer = sirq_buffers;
1972 trace_percpu_irq_buffer = irq_buffers;
1973 trace_percpu_nmi_buffer = nmi_buffers;
1974
1975 return 0;
1976
1977 err_nmi:
1978 free_percpu(irq_buffers);
1979 err_irq:
1980 free_percpu(sirq_buffers);
1981 err_sirq:
1982 free_percpu(buffers);
1983 err_warn:
1984 WARN(1, "Could not allocate percpu trace_printk buffer");
1985 return -ENOMEM;
1986}
1987
1988static int buffers_allocated;
1989
1990void trace_printk_init_buffers(void)
1991{
1992 if (buffers_allocated)
1993 return;
1994
1995 if (alloc_percpu_trace_buffer())
1996 return;
1997
1998 pr_info("ftrace: Allocated trace_printk buffers\n");
1999
2000 /* Expand the buffers to set size */
2001 tracing_update_buffers();
2002
2003 buffers_allocated = 1;
2004
2005 /*
2006 * trace_printk_init_buffers() can be called by modules.
2007 * If that happens, then we need to start cmdline recording
2008 * directly here. If the global_trace.buffer is already
2009 * allocated here, then this was called by module code.
2010 */
2011 if (global_trace.trace_buffer.buffer)
2012 tracing_start_cmdline_record();
2013}
2014
2015void trace_printk_start_comm(void)
2016{
2017 /* Start tracing comms if trace printk is set */
2018 if (!buffers_allocated)
2019 return;
2020 tracing_start_cmdline_record();
2021}
2022
2023static void trace_printk_start_stop_comm(int enabled)
2024{
2025 if (!buffers_allocated)
2026 return;
2027
2028 if (enabled)
2029 tracing_start_cmdline_record();
2030 else
2031 tracing_stop_cmdline_record();
2032}
2033
2034/**
2035 * trace_vbprintk - write binary msg to tracing buffer
2036 *
2037 */
2038int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
2039{
2040 struct ftrace_event_call *call = &event_bprint;
2041 struct ring_buffer_event *event;
2042 struct ring_buffer *buffer;
2043 struct trace_array *tr = &global_trace;
2044 struct bprint_entry *entry;
2045 unsigned long flags;
2046 char *tbuffer;
2047 int len = 0, size, pc;
2048
2049 if (unlikely(tracing_selftest_running || tracing_disabled))
2050 return 0;
2051
2052 /* Don't pollute graph traces with trace_vprintk internals */
2053 pause_graph_tracing();
2054
2055 pc = preempt_count();
2056 preempt_disable_notrace();
2057
2058 tbuffer = get_trace_buf();
2059 if (!tbuffer) {
2060 len = 0;
2061 goto out;
2062 }
2063
2064 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
2065
2066 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
2067 goto out;
2068
2069 local_save_flags(flags);
2070 size = sizeof(*entry) + sizeof(u32) * len;
2071 buffer = tr->trace_buffer.buffer;
2072 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2073 flags, pc);
2074 if (!event)
2075 goto out;
2076 entry = ring_buffer_event_data(event);
2077 entry->ip = ip;
2078 entry->fmt = fmt;
2079
2080 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2081 if (!call_filter_check_discard(call, entry, buffer, event)) {
2082 __buffer_unlock_commit(buffer, event);
2083 ftrace_trace_stack(buffer, flags, 6, pc);
2084 }
2085
2086out:
2087 preempt_enable_notrace();
2088 unpause_graph_tracing();
2089
2090 return len;
2091}
2092EXPORT_SYMBOL_GPL(trace_vbprintk);
2093
2094static int
2095__trace_array_vprintk(struct ring_buffer *buffer,
2096 unsigned long ip, const char *fmt, va_list args)
2097{
2098 struct ftrace_event_call *call = &event_print;
2099 struct ring_buffer_event *event;
2100 int len = 0, size, pc;
2101 struct print_entry *entry;
2102 unsigned long flags;
2103 char *tbuffer;
2104
2105 if (tracing_disabled || tracing_selftest_running)
2106 return 0;
2107
2108 /* Don't pollute graph traces with trace_vprintk internals */
2109 pause_graph_tracing();
2110
2111 pc = preempt_count();
2112 preempt_disable_notrace();
2113
2114
2115 tbuffer = get_trace_buf();
2116 if (!tbuffer) {
2117 len = 0;
2118 goto out;
2119 }
2120
2121 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2122 if (len > TRACE_BUF_SIZE)
2123 goto out;
2124
2125 local_save_flags(flags);
2126 size = sizeof(*entry) + len + 1;
2127 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2128 flags, pc);
2129 if (!event)
2130 goto out;
2131 entry = ring_buffer_event_data(event);
2132 entry->ip = ip;
2133
2134 memcpy(&entry->buf, tbuffer, len);
2135 entry->buf[len] = '\0';
2136 if (!call_filter_check_discard(call, entry, buffer, event)) {
2137 __buffer_unlock_commit(buffer, event);
2138 ftrace_trace_stack(buffer, flags, 6, pc);
2139 }
2140 out:
2141 preempt_enable_notrace();
2142 unpause_graph_tracing();
2143
2144 return len;
2145}
2146
2147int trace_array_vprintk(struct trace_array *tr,
2148 unsigned long ip, const char *fmt, va_list args)
2149{
2150 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2151}
2152
2153int trace_array_printk(struct trace_array *tr,
2154 unsigned long ip, const char *fmt, ...)
2155{
2156 int ret;
2157 va_list ap;
2158
2159 if (!(trace_flags & TRACE_ITER_PRINTK))
2160 return 0;
2161
2162 va_start(ap, fmt);
2163 ret = trace_array_vprintk(tr, ip, fmt, ap);
2164 va_end(ap);
2165 return ret;
2166}
2167
2168int trace_array_printk_buf(struct ring_buffer *buffer,
2169 unsigned long ip, const char *fmt, ...)
2170{
2171 int ret;
2172 va_list ap;
2173
2174 if (!(trace_flags & TRACE_ITER_PRINTK))
2175 return 0;
2176
2177 va_start(ap, fmt);
2178 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2179 va_end(ap);
2180 return ret;
2181}
2182
2183int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2184{
2185 return trace_array_vprintk(&global_trace, ip, fmt, args);
2186}
2187EXPORT_SYMBOL_GPL(trace_vprintk);
2188
2189static void trace_iterator_increment(struct trace_iterator *iter)
2190{
2191 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2192
2193 iter->idx++;
2194 if (buf_iter)
2195 ring_buffer_read(buf_iter, NULL);
2196}
2197
2198static struct trace_entry *
2199peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2200 unsigned long *lost_events)
2201{
2202 struct ring_buffer_event *event;
2203 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2204
2205 if (buf_iter)
2206 event = ring_buffer_iter_peek(buf_iter, ts);
2207 else
2208 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2209 lost_events);
2210
2211 if (event) {
2212 iter->ent_size = ring_buffer_event_length(event);
2213 return ring_buffer_event_data(event);
2214 }
2215 iter->ent_size = 0;
2216 return NULL;
2217}
2218
2219static struct trace_entry *
2220__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2221 unsigned long *missing_events, u64 *ent_ts)
2222{
2223 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2224 struct trace_entry *ent, *next = NULL;
2225 unsigned long lost_events = 0, next_lost = 0;
2226 int cpu_file = iter->cpu_file;
2227 u64 next_ts = 0, ts;
2228 int next_cpu = -1;
2229 int next_size = 0;
2230 int cpu;
2231
2232 /*
2233 * If we are in a per_cpu trace file, don't bother by iterating over
2234 * all cpu and peek directly.
2235 */
2236 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2237 if (ring_buffer_empty_cpu(buffer, cpu_file))
2238 return NULL;
2239 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2240 if (ent_cpu)
2241 *ent_cpu = cpu_file;
2242
2243 return ent;
2244 }
2245
2246 for_each_tracing_cpu(cpu) {
2247
2248 if (ring_buffer_empty_cpu(buffer, cpu))
2249 continue;
2250
2251 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2252
2253 /*
2254 * Pick the entry with the smallest timestamp:
2255 */
2256 if (ent && (!next || ts < next_ts)) {
2257 next = ent;
2258 next_cpu = cpu;
2259 next_ts = ts;
2260 next_lost = lost_events;
2261 next_size = iter->ent_size;
2262 }
2263 }
2264
2265 iter->ent_size = next_size;
2266
2267 if (ent_cpu)
2268 *ent_cpu = next_cpu;
2269
2270 if (ent_ts)
2271 *ent_ts = next_ts;
2272
2273 if (missing_events)
2274 *missing_events = next_lost;
2275
2276 return next;
2277}
2278
2279/* Find the next real entry, without updating the iterator itself */
2280struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2281 int *ent_cpu, u64 *ent_ts)
2282{
2283 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2284}
2285
2286/* Find the next real entry, and increment the iterator to the next entry */
2287void *trace_find_next_entry_inc(struct trace_iterator *iter)
2288{
2289 iter->ent = __find_next_entry(iter, &iter->cpu,
2290 &iter->lost_events, &iter->ts);
2291
2292 if (iter->ent)
2293 trace_iterator_increment(iter);
2294
2295 return iter->ent ? iter : NULL;
2296}
2297
2298static void trace_consume(struct trace_iterator *iter)
2299{
2300 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2301 &iter->lost_events);
2302}
2303
2304static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2305{
2306 struct trace_iterator *iter = m->private;
2307 int i = (int)*pos;
2308 void *ent;
2309
2310 WARN_ON_ONCE(iter->leftover);
2311
2312 (*pos)++;
2313
2314 /* can't go backwards */
2315 if (iter->idx > i)
2316 return NULL;
2317
2318 if (iter->idx < 0)
2319 ent = trace_find_next_entry_inc(iter);
2320 else
2321 ent = iter;
2322
2323 while (ent && iter->idx < i)
2324 ent = trace_find_next_entry_inc(iter);
2325
2326 iter->pos = *pos;
2327
2328 return ent;
2329}
2330
2331void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2332{
2333 struct ring_buffer_event *event;
2334 struct ring_buffer_iter *buf_iter;
2335 unsigned long entries = 0;
2336 u64 ts;
2337
2338 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2339
2340 buf_iter = trace_buffer_iter(iter, cpu);
2341 if (!buf_iter)
2342 return;
2343
2344 ring_buffer_iter_reset(buf_iter);
2345
2346 /*
2347 * We could have the case with the max latency tracers
2348 * that a reset never took place on a cpu. This is evident
2349 * by the timestamp being before the start of the buffer.
2350 */
2351 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2352 if (ts >= iter->trace_buffer->time_start)
2353 break;
2354 entries++;
2355 ring_buffer_read(buf_iter, NULL);
2356 }
2357
2358 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2359}
2360
2361/*
2362 * The current tracer is copied to avoid a global locking
2363 * all around.
2364 */
2365static void *s_start(struct seq_file *m, loff_t *pos)
2366{
2367 struct trace_iterator *iter = m->private;
2368 struct trace_array *tr = iter->tr;
2369 int cpu_file = iter->cpu_file;
2370 void *p = NULL;
2371 loff_t l = 0;
2372 int cpu;
2373
2374 /*
2375 * copy the tracer to avoid using a global lock all around.
2376 * iter->trace is a copy of current_trace, the pointer to the
2377 * name may be used instead of a strcmp(), as iter->trace->name
2378 * will point to the same string as current_trace->name.
2379 */
2380 mutex_lock(&trace_types_lock);
2381 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2382 *iter->trace = *tr->current_trace;
2383 mutex_unlock(&trace_types_lock);
2384
2385#ifdef CONFIG_TRACER_MAX_TRACE
2386 if (iter->snapshot && iter->trace->use_max_tr)
2387 return ERR_PTR(-EBUSY);
2388#endif
2389
2390 if (!iter->snapshot)
2391 atomic_inc(&trace_record_cmdline_disabled);
2392
2393 if (*pos != iter->pos) {
2394 iter->ent = NULL;
2395 iter->cpu = 0;
2396 iter->idx = -1;
2397
2398 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2399 for_each_tracing_cpu(cpu)
2400 tracing_iter_reset(iter, cpu);
2401 } else
2402 tracing_iter_reset(iter, cpu_file);
2403
2404 iter->leftover = 0;
2405 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2406 ;
2407
2408 } else {
2409 /*
2410 * If we overflowed the seq_file before, then we want
2411 * to just reuse the trace_seq buffer again.
2412 */
2413 if (iter->leftover)
2414 p = iter;
2415 else {
2416 l = *pos - 1;
2417 p = s_next(m, p, &l);
2418 }
2419 }
2420
2421 trace_event_read_lock();
2422 trace_access_lock(cpu_file);
2423 return p;
2424}
2425
2426static void s_stop(struct seq_file *m, void *p)
2427{
2428 struct trace_iterator *iter = m->private;
2429
2430#ifdef CONFIG_TRACER_MAX_TRACE
2431 if (iter->snapshot && iter->trace->use_max_tr)
2432 return;
2433#endif
2434
2435 if (!iter->snapshot)
2436 atomic_dec(&trace_record_cmdline_disabled);
2437
2438 trace_access_unlock(iter->cpu_file);
2439 trace_event_read_unlock();
2440}
2441
2442static void
2443get_total_entries(struct trace_buffer *buf,
2444 unsigned long *total, unsigned long *entries)
2445{
2446 unsigned long count;
2447 int cpu;
2448
2449 *total = 0;
2450 *entries = 0;
2451
2452 for_each_tracing_cpu(cpu) {
2453 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2454 /*
2455 * If this buffer has skipped entries, then we hold all
2456 * entries for the trace and we need to ignore the
2457 * ones before the time stamp.
2458 */
2459 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2460 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2461 /* total is the same as the entries */
2462 *total += count;
2463 } else
2464 *total += count +
2465 ring_buffer_overrun_cpu(buf->buffer, cpu);
2466 *entries += count;
2467 }
2468}
2469
2470static void print_lat_help_header(struct seq_file *m)
2471{
2472 seq_puts(m, "# _------=> CPU# \n");
2473 seq_puts(m, "# / _-----=> irqs-off \n");
2474 seq_puts(m, "# | / _----=> need-resched \n");
2475 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2476 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2477 seq_puts(m, "# |||| / delay \n");
2478 seq_puts(m, "# cmd pid ||||| time | caller \n");
2479 seq_puts(m, "# \\ / ||||| \\ | / \n");
2480}
2481
2482static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2483{
2484 unsigned long total;
2485 unsigned long entries;
2486
2487 get_total_entries(buf, &total, &entries);
2488 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2489 entries, total, num_online_cpus());
2490 seq_puts(m, "#\n");
2491}
2492
2493static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2494{
2495 print_event_info(buf, m);
2496 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2497 seq_puts(m, "# | | | | |\n");
2498}
2499
2500static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2501{
2502 print_event_info(buf, m);
2503 seq_puts(m, "# _-----=> irqs-off\n");
2504 seq_puts(m, "# / _----=> need-resched\n");
2505 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2506 seq_puts(m, "# || / _--=> preempt-depth\n");
2507 seq_puts(m, "# ||| / delay\n");
2508 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2509 seq_puts(m, "# | | | |||| | |\n");
2510}
2511
2512void
2513print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2514{
2515 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2516 struct trace_buffer *buf = iter->trace_buffer;
2517 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2518 struct tracer *type = iter->trace;
2519 unsigned long entries;
2520 unsigned long total;
2521 const char *name = "preemption";
2522
2523 name = type->name;
2524
2525 get_total_entries(buf, &total, &entries);
2526
2527 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2528 name, UTS_RELEASE);
2529 seq_puts(m, "# -----------------------------------"
2530 "---------------------------------\n");
2531 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2532 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2533 nsecs_to_usecs(data->saved_latency),
2534 entries,
2535 total,
2536 buf->cpu,
2537#if defined(CONFIG_PREEMPT_NONE)
2538 "server",
2539#elif defined(CONFIG_PREEMPT_VOLUNTARY)
2540 "desktop",
2541#elif defined(CONFIG_PREEMPT)
2542 "preempt",
2543#else
2544 "unknown",
2545#endif
2546 /* These are reserved for later use */
2547 0, 0, 0, 0);
2548#ifdef CONFIG_SMP
2549 seq_printf(m, " #P:%d)\n", num_online_cpus());
2550#else
2551 seq_puts(m, ")\n");
2552#endif
2553 seq_puts(m, "# -----------------\n");
2554 seq_printf(m, "# | task: %.16s-%d "
2555 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2556 data->comm, data->pid,
2557 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2558 data->policy, data->rt_priority);
2559 seq_puts(m, "# -----------------\n");
2560
2561 if (data->critical_start) {
2562 seq_puts(m, "# => started at: ");
2563 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2564 trace_print_seq(m, &iter->seq);
2565 seq_puts(m, "\n# => ended at: ");
2566 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2567 trace_print_seq(m, &iter->seq);
2568 seq_puts(m, "\n#\n");
2569 }
2570
2571 seq_puts(m, "#\n");
2572}
2573
2574static void test_cpu_buff_start(struct trace_iterator *iter)
2575{
2576 struct trace_seq *s = &iter->seq;
2577
2578 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2579 return;
2580
2581 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2582 return;
2583
2584 if (cpumask_test_cpu(iter->cpu, iter->started))
2585 return;
2586
2587 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2588 return;
2589
2590 cpumask_set_cpu(iter->cpu, iter->started);
2591
2592 /* Don't print started cpu buffer for the first entry of the trace */
2593 if (iter->idx > 1)
2594 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2595 iter->cpu);
2596}
2597
2598static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2599{
2600 struct trace_seq *s = &iter->seq;
2601 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2602 struct trace_entry *entry;
2603 struct trace_event *event;
2604
2605 entry = iter->ent;
2606
2607 test_cpu_buff_start(iter);
2608
2609 event = ftrace_find_event(entry->type);
2610
2611 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2612 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2613 if (!trace_print_lat_context(iter))
2614 goto partial;
2615 } else {
2616 if (!trace_print_context(iter))
2617 goto partial;
2618 }
2619 }
2620
2621 if (event)
2622 return event->funcs->trace(iter, sym_flags, event);
2623
2624 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2625 goto partial;
2626
2627 return TRACE_TYPE_HANDLED;
2628partial:
2629 return TRACE_TYPE_PARTIAL_LINE;
2630}
2631
2632static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2633{
2634 struct trace_seq *s = &iter->seq;
2635 struct trace_entry *entry;
2636 struct trace_event *event;
2637
2638 entry = iter->ent;
2639
2640 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2641 if (!trace_seq_printf(s, "%d %d %llu ",
2642 entry->pid, iter->cpu, iter->ts))
2643 goto partial;
2644 }
2645
2646 event = ftrace_find_event(entry->type);
2647 if (event)
2648 return event->funcs->raw(iter, 0, event);
2649
2650 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2651 goto partial;
2652
2653 return TRACE_TYPE_HANDLED;
2654partial:
2655 return TRACE_TYPE_PARTIAL_LINE;
2656}
2657
2658static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2659{
2660 struct trace_seq *s = &iter->seq;
2661 unsigned char newline = '\n';
2662 struct trace_entry *entry;
2663 struct trace_event *event;
2664
2665 entry = iter->ent;
2666
2667 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2668 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2669 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2670 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2671 }
2672
2673 event = ftrace_find_event(entry->type);
2674 if (event) {
2675 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2676 if (ret != TRACE_TYPE_HANDLED)
2677 return ret;
2678 }
2679
2680 SEQ_PUT_FIELD_RET(s, newline);
2681
2682 return TRACE_TYPE_HANDLED;
2683}
2684
2685static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2686{
2687 struct trace_seq *s = &iter->seq;
2688 struct trace_entry *entry;
2689 struct trace_event *event;
2690
2691 entry = iter->ent;
2692
2693 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2694 SEQ_PUT_FIELD_RET(s, entry->pid);
2695 SEQ_PUT_FIELD_RET(s, iter->cpu);
2696 SEQ_PUT_FIELD_RET(s, iter->ts);
2697 }
2698
2699 event = ftrace_find_event(entry->type);
2700 return event ? event->funcs->binary(iter, 0, event) :
2701 TRACE_TYPE_HANDLED;
2702}
2703
2704int trace_empty(struct trace_iterator *iter)
2705{
2706 struct ring_buffer_iter *buf_iter;
2707 int cpu;
2708
2709 /* If we are looking at one CPU buffer, only check that one */
2710 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2711 cpu = iter->cpu_file;
2712 buf_iter = trace_buffer_iter(iter, cpu);
2713 if (buf_iter) {
2714 if (!ring_buffer_iter_empty(buf_iter))
2715 return 0;
2716 } else {
2717 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2718 return 0;
2719 }
2720 return 1;
2721 }
2722
2723 for_each_tracing_cpu(cpu) {
2724 buf_iter = trace_buffer_iter(iter, cpu);
2725 if (buf_iter) {
2726 if (!ring_buffer_iter_empty(buf_iter))
2727 return 0;
2728 } else {
2729 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2730 return 0;
2731 }
2732 }
2733
2734 return 1;
2735}
2736
2737/* Called with trace_event_read_lock() held. */
2738enum print_line_t print_trace_line(struct trace_iterator *iter)
2739{
2740 enum print_line_t ret;
2741
2742 if (iter->lost_events &&
2743 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2744 iter->cpu, iter->lost_events))
2745 return TRACE_TYPE_PARTIAL_LINE;
2746
2747 if (iter->trace && iter->trace->print_line) {
2748 ret = iter->trace->print_line(iter);
2749 if (ret != TRACE_TYPE_UNHANDLED)
2750 return ret;
2751 }
2752
2753 if (iter->ent->type == TRACE_BPUTS &&
2754 trace_flags & TRACE_ITER_PRINTK &&
2755 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2756 return trace_print_bputs_msg_only(iter);
2757
2758 if (iter->ent->type == TRACE_BPRINT &&
2759 trace_flags & TRACE_ITER_PRINTK &&
2760 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2761 return trace_print_bprintk_msg_only(iter);
2762
2763 if (iter->ent->type == TRACE_PRINT &&
2764 trace_flags & TRACE_ITER_PRINTK &&
2765 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2766 return trace_print_printk_msg_only(iter);
2767
2768 if (trace_flags & TRACE_ITER_BIN)
2769 return print_bin_fmt(iter);
2770
2771 if (trace_flags & TRACE_ITER_HEX)
2772 return print_hex_fmt(iter);
2773
2774 if (trace_flags & TRACE_ITER_RAW)
2775 return print_raw_fmt(iter);
2776
2777 return print_trace_fmt(iter);
2778}
2779
2780void trace_latency_header(struct seq_file *m)
2781{
2782 struct trace_iterator *iter = m->private;
2783
2784 /* print nothing if the buffers are empty */
2785 if (trace_empty(iter))
2786 return;
2787
2788 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2789 print_trace_header(m, iter);
2790
2791 if (!(trace_flags & TRACE_ITER_VERBOSE))
2792 print_lat_help_header(m);
2793}
2794
2795void trace_default_header(struct seq_file *m)
2796{
2797 struct trace_iterator *iter = m->private;
2798
2799 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2800 return;
2801
2802 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2803 /* print nothing if the buffers are empty */
2804 if (trace_empty(iter))
2805 return;
2806 print_trace_header(m, iter);
2807 if (!(trace_flags & TRACE_ITER_VERBOSE))
2808 print_lat_help_header(m);
2809 } else {
2810 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2811 if (trace_flags & TRACE_ITER_IRQ_INFO)
2812 print_func_help_header_irq(iter->trace_buffer, m);
2813 else
2814 print_func_help_header(iter->trace_buffer, m);
2815 }
2816 }
2817}
2818
2819static void test_ftrace_alive(struct seq_file *m)
2820{
2821 if (!ftrace_is_dead())
2822 return;
2823 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2824 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2825}
2826
2827#ifdef CONFIG_TRACER_MAX_TRACE
2828static void show_snapshot_main_help(struct seq_file *m)
2829{
2830 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2831 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2832 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2833 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n");
2834 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2835 seq_printf(m, "# is not a '0' or '1')\n");
2836}
2837
2838static void show_snapshot_percpu_help(struct seq_file *m)
2839{
2840 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2841#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2842 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2843 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2844#else
2845 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2846 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2847#endif
2848 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2849 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2850 seq_printf(m, "# is not a '0' or '1')\n");
2851}
2852
2853static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2854{
2855 if (iter->tr->allocated_snapshot)
2856 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2857 else
2858 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2859
2860 seq_printf(m, "# Snapshot commands:\n");
2861 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2862 show_snapshot_main_help(m);
2863 else
2864 show_snapshot_percpu_help(m);
2865}
2866#else
2867/* Should never be called */
2868static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2869#endif
2870
2871static int s_show(struct seq_file *m, void *v)
2872{
2873 struct trace_iterator *iter = v;
2874 int ret;
2875
2876 if (iter->ent == NULL) {
2877 if (iter->tr) {
2878 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2879 seq_puts(m, "#\n");
2880 test_ftrace_alive(m);
2881 }
2882 if (iter->snapshot && trace_empty(iter))
2883 print_snapshot_help(m, iter);
2884 else if (iter->trace && iter->trace->print_header)
2885 iter->trace->print_header(m);
2886 else
2887 trace_default_header(m);
2888
2889 } else if (iter->leftover) {
2890 /*
2891 * If we filled the seq_file buffer earlier, we
2892 * want to just show it now.
2893 */
2894 ret = trace_print_seq(m, &iter->seq);
2895
2896 /* ret should this time be zero, but you never know */
2897 iter->leftover = ret;
2898
2899 } else {
2900 print_trace_line(iter);
2901 ret = trace_print_seq(m, &iter->seq);
2902 /*
2903 * If we overflow the seq_file buffer, then it will
2904 * ask us for this data again at start up.
2905 * Use that instead.
2906 * ret is 0 if seq_file write succeeded.
2907 * -1 otherwise.
2908 */
2909 iter->leftover = ret;
2910 }
2911
2912 return 0;
2913}
2914
2915/*
2916 * Should be used after trace_array_get(), trace_types_lock
2917 * ensures that i_cdev was already initialized.
2918 */
2919static inline int tracing_get_cpu(struct inode *inode)
2920{
2921 if (inode->i_cdev) /* See trace_create_cpu_file() */
2922 return (long)inode->i_cdev - 1;
2923 return RING_BUFFER_ALL_CPUS;
2924}
2925
2926static const struct seq_operations tracer_seq_ops = {
2927 .start = s_start,
2928 .next = s_next,
2929 .stop = s_stop,
2930 .show = s_show,
2931};
2932
2933static struct trace_iterator *
2934__tracing_open(struct inode *inode, struct file *file, bool snapshot)
2935{
2936 struct trace_array *tr = inode->i_private;
2937 struct trace_iterator *iter;
2938 int cpu;
2939
2940 if (tracing_disabled)
2941 return ERR_PTR(-ENODEV);
2942
2943 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2944 if (!iter)
2945 return ERR_PTR(-ENOMEM);
2946
2947 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2948 GFP_KERNEL);
2949 if (!iter->buffer_iter)
2950 goto release;
2951
2952 /*
2953 * We make a copy of the current tracer to avoid concurrent
2954 * changes on it while we are reading.
2955 */
2956 mutex_lock(&trace_types_lock);
2957 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2958 if (!iter->trace)
2959 goto fail;
2960
2961 *iter->trace = *tr->current_trace;
2962
2963 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2964 goto fail;
2965
2966 iter->tr = tr;
2967
2968#ifdef CONFIG_TRACER_MAX_TRACE
2969 /* Currently only the top directory has a snapshot */
2970 if (tr->current_trace->print_max || snapshot)
2971 iter->trace_buffer = &tr->max_buffer;
2972 else
2973#endif
2974 iter->trace_buffer = &tr->trace_buffer;
2975 iter->snapshot = snapshot;
2976 iter->pos = -1;
2977 iter->cpu_file = tracing_get_cpu(inode);
2978 mutex_init(&iter->mutex);
2979
2980 /* Notify the tracer early; before we stop tracing. */
2981 if (iter->trace && iter->trace->open)
2982 iter->trace->open(iter);
2983
2984 /* Annotate start of buffers if we had overruns */
2985 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2986 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2987
2988 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2989 if (trace_clocks[tr->clock_id].in_ns)
2990 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2991
2992 /* stop the trace while dumping if we are not opening "snapshot" */
2993 if (!iter->snapshot)
2994 tracing_stop_tr(tr);
2995
2996 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2997 for_each_tracing_cpu(cpu) {
2998 iter->buffer_iter[cpu] =
2999 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
3000 }
3001 ring_buffer_read_prepare_sync();
3002 for_each_tracing_cpu(cpu) {
3003 ring_buffer_read_start(iter->buffer_iter[cpu]);
3004 tracing_iter_reset(iter, cpu);
3005 }
3006 } else {
3007 cpu = iter->cpu_file;
3008 iter->buffer_iter[cpu] =
3009 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
3010 ring_buffer_read_prepare_sync();
3011 ring_buffer_read_start(iter->buffer_iter[cpu]);
3012 tracing_iter_reset(iter, cpu);
3013 }
3014
3015 mutex_unlock(&trace_types_lock);
3016
3017 return iter;
3018
3019 fail:
3020 mutex_unlock(&trace_types_lock);
3021 kfree(iter->trace);
3022 kfree(iter->buffer_iter);
3023release:
3024 seq_release_private(inode, file);
3025 return ERR_PTR(-ENOMEM);
3026}
3027
3028int tracing_open_generic(struct inode *inode, struct file *filp)
3029{
3030 if (tracing_disabled)
3031 return -ENODEV;
3032
3033 filp->private_data = inode->i_private;
3034 return 0;
3035}
3036
3037bool tracing_is_disabled(void)
3038{
3039 return (tracing_disabled) ? true: false;
3040}
3041
3042/*
3043 * Open and update trace_array ref count.
3044 * Must have the current trace_array passed to it.
3045 */
3046static int tracing_open_generic_tr(struct inode *inode, struct file *filp)
3047{
3048 struct trace_array *tr = inode->i_private;
3049
3050 if (tracing_disabled)
3051 return -ENODEV;
3052
3053 if (trace_array_get(tr) < 0)
3054 return -ENODEV;
3055
3056 filp->private_data = inode->i_private;
3057
3058 return 0;
3059}
3060
3061static int tracing_release(struct inode *inode, struct file *file)
3062{
3063 struct trace_array *tr = inode->i_private;
3064 struct seq_file *m = file->private_data;
3065 struct trace_iterator *iter;
3066 int cpu;
3067
3068 if (!(file->f_mode & FMODE_READ)) {
3069 trace_array_put(tr);
3070 return 0;
3071 }
3072
3073 /* Writes do not use seq_file */
3074 iter = m->private;
3075 mutex_lock(&trace_types_lock);
3076
3077 for_each_tracing_cpu(cpu) {
3078 if (iter->buffer_iter[cpu])
3079 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3080 }
3081
3082 if (iter->trace && iter->trace->close)
3083 iter->trace->close(iter);
3084
3085 if (!iter->snapshot)
3086 /* reenable tracing if it was previously enabled */
3087 tracing_start_tr(tr);
3088
3089 __trace_array_put(tr);
3090
3091 mutex_unlock(&trace_types_lock);
3092
3093 mutex_destroy(&iter->mutex);
3094 free_cpumask_var(iter->started);
3095 kfree(iter->trace);
3096 kfree(iter->buffer_iter);
3097 seq_release_private(inode, file);
3098
3099 return 0;
3100}
3101
3102static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3103{
3104 struct trace_array *tr = inode->i_private;
3105
3106 trace_array_put(tr);
3107 return 0;
3108}
3109
3110static int tracing_single_release_tr(struct inode *inode, struct file *file)
3111{
3112 struct trace_array *tr = inode->i_private;
3113
3114 trace_array_put(tr);
3115
3116 return single_release(inode, file);
3117}
3118
3119static int tracing_open(struct inode *inode, struct file *file)
3120{
3121 struct trace_array *tr = inode->i_private;
3122 struct trace_iterator *iter;
3123 int ret = 0;
3124
3125 if (trace_array_get(tr) < 0)
3126 return -ENODEV;
3127
3128 /* If this file was open for write, then erase contents */
3129 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
3130 int cpu = tracing_get_cpu(inode);
3131
3132 if (cpu == RING_BUFFER_ALL_CPUS)
3133 tracing_reset_online_cpus(&tr->trace_buffer);
3134 else
3135 tracing_reset(&tr->trace_buffer, cpu);
3136 }
3137
3138 if (file->f_mode & FMODE_READ) {
3139 iter = __tracing_open(inode, file, false);
3140 if (IS_ERR(iter))
3141 ret = PTR_ERR(iter);
3142 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
3143 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3144 }
3145
3146 if (ret < 0)
3147 trace_array_put(tr);
3148
3149 return ret;
3150}
3151
3152/*
3153 * Some tracers are not suitable for instance buffers.
3154 * A tracer is always available for the global array (toplevel)
3155 * or if it explicitly states that it is.
3156 */
3157static bool
3158trace_ok_for_array(struct tracer *t, struct trace_array *tr)
3159{
3160 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
3161}
3162
3163/* Find the next tracer that this trace array may use */
3164static struct tracer *
3165get_tracer_for_array(struct trace_array *tr, struct tracer *t)
3166{
3167 while (t && !trace_ok_for_array(t, tr))
3168 t = t->next;
3169
3170 return t;
3171}
3172
3173static void *
3174t_next(struct seq_file *m, void *v, loff_t *pos)
3175{
3176 struct trace_array *tr = m->private;
3177 struct tracer *t = v;
3178
3179 (*pos)++;
3180
3181 if (t)
3182 t = get_tracer_for_array(tr, t->next);
3183
3184 return t;
3185}
3186
3187static void *t_start(struct seq_file *m, loff_t *pos)
3188{
3189 struct trace_array *tr = m->private;
3190 struct tracer *t;
3191 loff_t l = 0;
3192
3193 mutex_lock(&trace_types_lock);
3194
3195 t = get_tracer_for_array(tr, trace_types);
3196 for (; t && l < *pos; t = t_next(m, t, &l))
3197 ;
3198
3199 return t;
3200}
3201
3202static void t_stop(struct seq_file *m, void *p)
3203{
3204 mutex_unlock(&trace_types_lock);
3205}
3206
3207static int t_show(struct seq_file *m, void *v)
3208{
3209 struct tracer *t = v;
3210
3211 if (!t)
3212 return 0;
3213
3214 seq_printf(m, "%s", t->name);
3215 if (t->next)
3216 seq_putc(m, ' ');
3217 else
3218 seq_putc(m, '\n');
3219
3220 return 0;
3221}
3222
3223static const struct seq_operations show_traces_seq_ops = {
3224 .start = t_start,
3225 .next = t_next,
3226 .stop = t_stop,
3227 .show = t_show,
3228};
3229
3230static int show_traces_open(struct inode *inode, struct file *file)
3231{
3232 struct trace_array *tr = inode->i_private;
3233 struct seq_file *m;
3234 int ret;
3235
3236 if (tracing_disabled)
3237 return -ENODEV;
3238
3239 ret = seq_open(file, &show_traces_seq_ops);
3240 if (ret)
3241 return ret;
3242
3243 m = file->private_data;
3244 m->private = tr;
3245
3246 return 0;
3247}
3248
3249static ssize_t
3250tracing_write_stub(struct file *filp, const char __user *ubuf,
3251 size_t count, loff_t *ppos)
3252{
3253 return count;
3254}
3255
3256loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
3257{
3258 int ret;
3259
3260 if (file->f_mode & FMODE_READ)
3261 ret = seq_lseek(file, offset, whence);
3262 else
3263 file->f_pos = ret = 0;
3264
3265 return ret;
3266}
3267
3268static const struct file_operations tracing_fops = {
3269 .open = tracing_open,
3270 .read = seq_read,
3271 .write = tracing_write_stub,
3272 .llseek = tracing_lseek,
3273 .release = tracing_release,
3274};
3275
3276static const struct file_operations show_traces_fops = {
3277 .open = show_traces_open,
3278 .read = seq_read,
3279 .release = seq_release,
3280 .llseek = seq_lseek,
3281};
3282
3283/*
3284 * The tracer itself will not take this lock, but still we want
3285 * to provide a consistent cpumask to user-space:
3286 */
3287static DEFINE_MUTEX(tracing_cpumask_update_lock);
3288
3289/*
3290 * Temporary storage for the character representation of the
3291 * CPU bitmask (and one more byte for the newline):
3292 */
3293static char mask_str[NR_CPUS + 1];
3294
3295static ssize_t
3296tracing_cpumask_read(struct file *filp, char __user *ubuf,
3297 size_t count, loff_t *ppos)
3298{
3299 struct trace_array *tr = file_inode(filp)->i_private;
3300 int len;
3301
3302 mutex_lock(&tracing_cpumask_update_lock);
3303
3304 len = cpumask_scnprintf(mask_str, count, tr->tracing_cpumask);
3305 if (count - len < 2) {
3306 count = -EINVAL;
3307 goto out_err;
3308 }
3309 len += sprintf(mask_str + len, "\n");
3310 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3311
3312out_err:
3313 mutex_unlock(&tracing_cpumask_update_lock);
3314
3315 return count;
3316}
3317
3318static ssize_t
3319tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3320 size_t count, loff_t *ppos)
3321{
3322 struct trace_array *tr = file_inode(filp)->i_private;
3323 cpumask_var_t tracing_cpumask_new;
3324 int err, cpu;
3325
3326 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3327 return -ENOMEM;
3328
3329 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3330 if (err)
3331 goto err_unlock;
3332
3333 mutex_lock(&tracing_cpumask_update_lock);
3334
3335 local_irq_disable();
3336 arch_spin_lock(&ftrace_max_lock);
3337 for_each_tracing_cpu(cpu) {
3338 /*
3339 * Increase/decrease the disabled counter if we are
3340 * about to flip a bit in the cpumask:
3341 */
3342 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3343 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3344 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3345 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3346 }
3347 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3348 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3349 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3350 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3351 }
3352 }
3353 arch_spin_unlock(&ftrace_max_lock);
3354 local_irq_enable();
3355
3356 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
3357
3358 mutex_unlock(&tracing_cpumask_update_lock);
3359 free_cpumask_var(tracing_cpumask_new);
3360
3361 return count;
3362
3363err_unlock:
3364 free_cpumask_var(tracing_cpumask_new);
3365
3366 return err;
3367}
3368
3369static const struct file_operations tracing_cpumask_fops = {
3370 .open = tracing_open_generic_tr,
3371 .read = tracing_cpumask_read,
3372 .write = tracing_cpumask_write,
3373 .release = tracing_release_generic_tr,
3374 .llseek = generic_file_llseek,
3375};
3376
3377static int tracing_trace_options_show(struct seq_file *m, void *v)
3378{
3379 struct tracer_opt *trace_opts;
3380 struct trace_array *tr = m->private;
3381 u32 tracer_flags;
3382 int i;
3383
3384 mutex_lock(&trace_types_lock);
3385 tracer_flags = tr->current_trace->flags->val;
3386 trace_opts = tr->current_trace->flags->opts;
3387
3388 for (i = 0; trace_options[i]; i++) {
3389 if (trace_flags & (1 << i))
3390 seq_printf(m, "%s\n", trace_options[i]);
3391 else
3392 seq_printf(m, "no%s\n", trace_options[i]);
3393 }
3394
3395 for (i = 0; trace_opts[i].name; i++) {
3396 if (tracer_flags & trace_opts[i].bit)
3397 seq_printf(m, "%s\n", trace_opts[i].name);
3398 else
3399 seq_printf(m, "no%s\n", trace_opts[i].name);
3400 }
3401 mutex_unlock(&trace_types_lock);
3402
3403 return 0;
3404}
3405
3406static int __set_tracer_option(struct trace_array *tr,
3407 struct tracer_flags *tracer_flags,
3408 struct tracer_opt *opts, int neg)
3409{
3410 struct tracer *trace = tr->current_trace;
3411 int ret;
3412
3413 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
3414 if (ret)
3415 return ret;
3416
3417 if (neg)
3418 tracer_flags->val &= ~opts->bit;
3419 else
3420 tracer_flags->val |= opts->bit;
3421 return 0;
3422}
3423
3424/* Try to assign a tracer specific option */
3425static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
3426{
3427 struct tracer *trace = tr->current_trace;
3428 struct tracer_flags *tracer_flags = trace->flags;
3429 struct tracer_opt *opts = NULL;
3430 int i;
3431
3432 for (i = 0; tracer_flags->opts[i].name; i++) {
3433 opts = &tracer_flags->opts[i];
3434
3435 if (strcmp(cmp, opts->name) == 0)
3436 return __set_tracer_option(tr, trace->flags, opts, neg);
3437 }
3438
3439 return -EINVAL;
3440}
3441
3442/* Some tracers require overwrite to stay enabled */
3443int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3444{
3445 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3446 return -1;
3447
3448 return 0;
3449}
3450
3451int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3452{
3453 /* do nothing if flag is already set */
3454 if (!!(trace_flags & mask) == !!enabled)
3455 return 0;
3456
3457 /* Give the tracer a chance to approve the change */
3458 if (tr->current_trace->flag_changed)
3459 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
3460 return -EINVAL;
3461
3462 if (enabled)
3463 trace_flags |= mask;
3464 else
3465 trace_flags &= ~mask;
3466
3467 if (mask == TRACE_ITER_RECORD_CMD)
3468 trace_event_enable_cmd_record(enabled);
3469
3470 if (mask == TRACE_ITER_OVERWRITE) {
3471 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3472#ifdef CONFIG_TRACER_MAX_TRACE
3473 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3474#endif
3475 }
3476
3477 if (mask == TRACE_ITER_PRINTK)
3478 trace_printk_start_stop_comm(enabled);
3479
3480 return 0;
3481}
3482
3483static int trace_set_options(struct trace_array *tr, char *option)
3484{
3485 char *cmp;
3486 int neg = 0;
3487 int ret = -ENODEV;
3488 int i;
3489
3490 cmp = strstrip(option);
3491
3492 if (strncmp(cmp, "no", 2) == 0) {
3493 neg = 1;
3494 cmp += 2;
3495 }
3496
3497 mutex_lock(&trace_types_lock);
3498
3499 for (i = 0; trace_options[i]; i++) {
3500 if (strcmp(cmp, trace_options[i]) == 0) {
3501 ret = set_tracer_flag(tr, 1 << i, !neg);
3502 break;
3503 }
3504 }
3505
3506 /* If no option could be set, test the specific tracer options */
3507 if (!trace_options[i])
3508 ret = set_tracer_option(tr, cmp, neg);
3509
3510 mutex_unlock(&trace_types_lock);
3511
3512 return ret;
3513}
3514
3515static ssize_t
3516tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3517 size_t cnt, loff_t *ppos)
3518{
3519 struct seq_file *m = filp->private_data;
3520 struct trace_array *tr = m->private;
3521 char buf[64];
3522 int ret;
3523
3524 if (cnt >= sizeof(buf))
3525 return -EINVAL;
3526
3527 if (copy_from_user(&buf, ubuf, cnt))
3528 return -EFAULT;
3529
3530 buf[cnt] = 0;
3531
3532 ret = trace_set_options(tr, buf);
3533 if (ret < 0)
3534 return ret;
3535
3536 *ppos += cnt;
3537
3538 return cnt;
3539}
3540
3541static int tracing_trace_options_open(struct inode *inode, struct file *file)
3542{
3543 struct trace_array *tr = inode->i_private;
3544 int ret;
3545
3546 if (tracing_disabled)
3547 return -ENODEV;
3548
3549 if (trace_array_get(tr) < 0)
3550 return -ENODEV;
3551
3552 ret = single_open(file, tracing_trace_options_show, inode->i_private);
3553 if (ret < 0)
3554 trace_array_put(tr);
3555
3556 return ret;
3557}
3558
3559static const struct file_operations tracing_iter_fops = {
3560 .open = tracing_trace_options_open,
3561 .read = seq_read,
3562 .llseek = seq_lseek,
3563 .release = tracing_single_release_tr,
3564 .write = tracing_trace_options_write,
3565};
3566
3567static const char readme_msg[] =
3568 "tracing mini-HOWTO:\n\n"
3569 "# echo 0 > tracing_on : quick way to disable tracing\n"
3570 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3571 " Important files:\n"
3572 " trace\t\t\t- The static contents of the buffer\n"
3573 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3574 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3575 " current_tracer\t- function and latency tracers\n"
3576 " available_tracers\t- list of configured tracers for current_tracer\n"
3577 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3578 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3579 " trace_clock\t\t-change the clock used to order events\n"
3580 " local: Per cpu clock but may not be synced across CPUs\n"
3581 " global: Synced across CPUs but slows tracing down.\n"
3582 " counter: Not a clock, but just an increment\n"
3583 " uptime: Jiffy counter from time of boot\n"
3584 " perf: Same clock that perf events use\n"
3585#ifdef CONFIG_X86_64
3586 " x86-tsc: TSC cycle counter\n"
3587#endif
3588 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3589 " tracing_cpumask\t- Limit which CPUs to trace\n"
3590 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3591 "\t\t\t Remove sub-buffer with rmdir\n"
3592 " trace_options\t\t- Set format or modify how tracing happens\n"
3593 "\t\t\t Disable an option by adding a suffix 'no' to the\n"
3594 "\t\t\t option name\n"
3595#ifdef CONFIG_DYNAMIC_FTRACE
3596 "\n available_filter_functions - list of functions that can be filtered on\n"
3597 " set_ftrace_filter\t- echo function name in here to only trace these\n"
3598 "\t\t\t functions\n"
3599 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3600 "\t modules: Can select a group via module\n"
3601 "\t Format: :mod:<module-name>\n"
3602 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
3603 "\t triggers: a command to perform when function is hit\n"
3604 "\t Format: <function>:<trigger>[:count]\n"
3605 "\t trigger: traceon, traceoff\n"
3606 "\t\t enable_event:<system>:<event>\n"
3607 "\t\t disable_event:<system>:<event>\n"
3608#ifdef CONFIG_STACKTRACE
3609 "\t\t stacktrace\n"
3610#endif
3611#ifdef CONFIG_TRACER_SNAPSHOT
3612 "\t\t snapshot\n"
3613#endif
3614 "\t\t dump\n"
3615 "\t\t cpudump\n"
3616 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
3617 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
3618 "\t The first one will disable tracing every time do_fault is hit\n"
3619 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
3620 "\t The first time do trap is hit and it disables tracing, the\n"
3621 "\t counter will decrement to 2. If tracing is already disabled,\n"
3622 "\t the counter will not decrement. It only decrements when the\n"
3623 "\t trigger did work\n"
3624 "\t To remove trigger without count:\n"
3625 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
3626 "\t To remove trigger with a count:\n"
3627 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3628 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3629 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3630 "\t modules: Can select a group via module command :mod:\n"
3631 "\t Does not accept triggers\n"
3632#endif /* CONFIG_DYNAMIC_FTRACE */
3633#ifdef CONFIG_FUNCTION_TRACER
3634 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
3635 "\t\t (function)\n"
3636#endif
3637#ifdef CONFIG_FUNCTION_GRAPH_TRACER
3638 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3639 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3640#endif
3641#ifdef CONFIG_TRACER_SNAPSHOT
3642 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
3643 "\t\t\t snapshot buffer. Read the contents for more\n"
3644 "\t\t\t information\n"
3645#endif
3646#ifdef CONFIG_STACK_TRACER
3647 " stack_trace\t\t- Shows the max stack trace when active\n"
3648 " stack_max_size\t- Shows current max stack size that was traced\n"
3649 "\t\t\t Write into this file to reset the max size (trigger a\n"
3650 "\t\t\t new trace)\n"
3651#ifdef CONFIG_DYNAMIC_FTRACE
3652 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
3653 "\t\t\t traces\n"
3654#endif
3655#endif /* CONFIG_STACK_TRACER */
3656 " events/\t\t- Directory containing all trace event subsystems:\n"
3657 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
3658 " events/<system>/\t- Directory containing all trace events for <system>:\n"
3659 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
3660 "\t\t\t events\n"
3661 " filter\t\t- If set, only events passing filter are traced\n"
3662 " events/<system>/<event>/\t- Directory containing control files for\n"
3663 "\t\t\t <event>:\n"
3664 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
3665 " filter\t\t- If set, only events passing filter are traced\n"
3666 " trigger\t\t- If set, a command to perform when event is hit\n"
3667 "\t Format: <trigger>[:count][if <filter>]\n"
3668 "\t trigger: traceon, traceoff\n"
3669 "\t enable_event:<system>:<event>\n"
3670 "\t disable_event:<system>:<event>\n"
3671#ifdef CONFIG_STACKTRACE
3672 "\t\t stacktrace\n"
3673#endif
3674#ifdef CONFIG_TRACER_SNAPSHOT
3675 "\t\t snapshot\n"
3676#endif
3677 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
3678 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
3679 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
3680 "\t events/block/block_unplug/trigger\n"
3681 "\t The first disables tracing every time block_unplug is hit.\n"
3682 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
3683 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
3684 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
3685 "\t Like function triggers, the counter is only decremented if it\n"
3686 "\t enabled or disabled tracing.\n"
3687 "\t To remove a trigger without a count:\n"
3688 "\t echo '!<trigger> > <system>/<event>/trigger\n"
3689 "\t To remove a trigger with a count:\n"
3690 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
3691 "\t Filters can be ignored when removing a trigger.\n"
3692;
3693
3694static ssize_t
3695tracing_readme_read(struct file *filp, char __user *ubuf,
3696 size_t cnt, loff_t *ppos)
3697{
3698 return simple_read_from_buffer(ubuf, cnt, ppos,
3699 readme_msg, strlen(readme_msg));
3700}
3701
3702static const struct file_operations tracing_readme_fops = {
3703 .open = tracing_open_generic,
3704 .read = tracing_readme_read,
3705 .llseek = generic_file_llseek,
3706};
3707
3708static ssize_t
3709tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3710 size_t cnt, loff_t *ppos)
3711{
3712 char *buf_comm;
3713 char *file_buf;
3714 char *buf;
3715 int len = 0;
3716 int pid;
3717 int i;
3718
3719 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3720 if (!file_buf)
3721 return -ENOMEM;
3722
3723 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3724 if (!buf_comm) {
3725 kfree(file_buf);
3726 return -ENOMEM;
3727 }
3728
3729 buf = file_buf;
3730
3731 for (i = 0; i < SAVED_CMDLINES; i++) {
3732 int r;
3733
3734 pid = map_cmdline_to_pid[i];
3735 if (pid == -1 || pid == NO_CMDLINE_MAP)
3736 continue;
3737
3738 trace_find_cmdline(pid, buf_comm);
3739 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3740 buf += r;
3741 len += r;
3742 }
3743
3744 len = simple_read_from_buffer(ubuf, cnt, ppos,
3745 file_buf, len);
3746
3747 kfree(file_buf);
3748 kfree(buf_comm);
3749
3750 return len;
3751}
3752
3753static const struct file_operations tracing_saved_cmdlines_fops = {
3754 .open = tracing_open_generic,
3755 .read = tracing_saved_cmdlines_read,
3756 .llseek = generic_file_llseek,
3757};
3758
3759static ssize_t
3760tracing_set_trace_read(struct file *filp, char __user *ubuf,
3761 size_t cnt, loff_t *ppos)
3762{
3763 struct trace_array *tr = filp->private_data;
3764 char buf[MAX_TRACER_SIZE+2];
3765 int r;
3766
3767 mutex_lock(&trace_types_lock);
3768 r = sprintf(buf, "%s\n", tr->current_trace->name);
3769 mutex_unlock(&trace_types_lock);
3770
3771 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3772}
3773
3774int tracer_init(struct tracer *t, struct trace_array *tr)
3775{
3776 tracing_reset_online_cpus(&tr->trace_buffer);
3777 return t->init(tr);
3778}
3779
3780static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3781{
3782 int cpu;
3783
3784 for_each_tracing_cpu(cpu)
3785 per_cpu_ptr(buf->data, cpu)->entries = val;
3786}
3787
3788#ifdef CONFIG_TRACER_MAX_TRACE
3789/* resize @tr's buffer to the size of @size_tr's entries */
3790static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3791 struct trace_buffer *size_buf, int cpu_id)
3792{
3793 int cpu, ret = 0;
3794
3795 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3796 for_each_tracing_cpu(cpu) {
3797 ret = ring_buffer_resize(trace_buf->buffer,
3798 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3799 if (ret < 0)
3800 break;
3801 per_cpu_ptr(trace_buf->data, cpu)->entries =
3802 per_cpu_ptr(size_buf->data, cpu)->entries;
3803 }
3804 } else {
3805 ret = ring_buffer_resize(trace_buf->buffer,
3806 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3807 if (ret == 0)
3808 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3809 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3810 }
3811
3812 return ret;
3813}
3814#endif /* CONFIG_TRACER_MAX_TRACE */
3815
3816static int __tracing_resize_ring_buffer(struct trace_array *tr,
3817 unsigned long size, int cpu)
3818{
3819 int ret;
3820
3821 /*
3822 * If kernel or user changes the size of the ring buffer
3823 * we use the size that was given, and we can forget about
3824 * expanding it later.
3825 */
3826 ring_buffer_expanded = true;
3827
3828 /* May be called before buffers are initialized */
3829 if (!tr->trace_buffer.buffer)
3830 return 0;
3831
3832 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3833 if (ret < 0)
3834 return ret;
3835
3836#ifdef CONFIG_TRACER_MAX_TRACE
3837 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3838 !tr->current_trace->use_max_tr)
3839 goto out;
3840
3841 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3842 if (ret < 0) {
3843 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3844 &tr->trace_buffer, cpu);
3845 if (r < 0) {
3846 /*
3847 * AARGH! We are left with different
3848 * size max buffer!!!!
3849 * The max buffer is our "snapshot" buffer.
3850 * When a tracer needs a snapshot (one of the
3851 * latency tracers), it swaps the max buffer
3852 * with the saved snap shot. We succeeded to
3853 * update the size of the main buffer, but failed to
3854 * update the size of the max buffer. But when we tried
3855 * to reset the main buffer to the original size, we
3856 * failed there too. This is very unlikely to
3857 * happen, but if it does, warn and kill all
3858 * tracing.
3859 */
3860 WARN_ON(1);
3861 tracing_disabled = 1;
3862 }
3863 return ret;
3864 }
3865
3866 if (cpu == RING_BUFFER_ALL_CPUS)
3867 set_buffer_entries(&tr->max_buffer, size);
3868 else
3869 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3870
3871 out:
3872#endif /* CONFIG_TRACER_MAX_TRACE */
3873
3874 if (cpu == RING_BUFFER_ALL_CPUS)
3875 set_buffer_entries(&tr->trace_buffer, size);
3876 else
3877 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3878
3879 return ret;
3880}
3881
3882static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3883 unsigned long size, int cpu_id)
3884{
3885 int ret = size;
3886
3887 mutex_lock(&trace_types_lock);
3888
3889 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3890 /* make sure, this cpu is enabled in the mask */
3891 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3892 ret = -EINVAL;
3893 goto out;
3894 }
3895 }
3896
3897 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3898 if (ret < 0)
3899 ret = -ENOMEM;
3900
3901out:
3902 mutex_unlock(&trace_types_lock);
3903
3904 return ret;
3905}
3906
3907
3908/**
3909 * tracing_update_buffers - used by tracing facility to expand ring buffers
3910 *
3911 * To save on memory when the tracing is never used on a system with it
3912 * configured in. The ring buffers are set to a minimum size. But once
3913 * a user starts to use the tracing facility, then they need to grow
3914 * to their default size.
3915 *
3916 * This function is to be called when a tracer is about to be used.
3917 */
3918int tracing_update_buffers(void)
3919{
3920 int ret = 0;
3921
3922 mutex_lock(&trace_types_lock);
3923 if (!ring_buffer_expanded)
3924 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3925 RING_BUFFER_ALL_CPUS);
3926 mutex_unlock(&trace_types_lock);
3927
3928 return ret;
3929}
3930
3931struct trace_option_dentry;
3932
3933static struct trace_option_dentry *
3934create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3935
3936static void
3937destroy_trace_option_files(struct trace_option_dentry *topts);
3938
3939/*
3940 * Used to clear out the tracer before deletion of an instance.
3941 * Must have trace_types_lock held.
3942 */
3943static void tracing_set_nop(struct trace_array *tr)
3944{
3945 if (tr->current_trace == &nop_trace)
3946 return;
3947
3948 tr->current_trace->enabled--;
3949
3950 if (tr->current_trace->reset)
3951 tr->current_trace->reset(tr);
3952
3953 tr->current_trace = &nop_trace;
3954}
3955
3956static int tracing_set_tracer(struct trace_array *tr, const char *buf)
3957{
3958 static struct trace_option_dentry *topts;
3959 struct tracer *t;
3960#ifdef CONFIG_TRACER_MAX_TRACE
3961 bool had_max_tr;
3962#endif
3963 int ret = 0;
3964
3965 mutex_lock(&trace_types_lock);
3966
3967 if (!ring_buffer_expanded) {
3968 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3969 RING_BUFFER_ALL_CPUS);
3970 if (ret < 0)
3971 goto out;
3972 ret = 0;
3973 }
3974
3975 for (t = trace_types; t; t = t->next) {
3976 if (strcmp(t->name, buf) == 0)
3977 break;
3978 }
3979 if (!t) {
3980 ret = -EINVAL;
3981 goto out;
3982 }
3983 if (t == tr->current_trace)
3984 goto out;
3985
3986 /* Some tracers are only allowed for the top level buffer */
3987 if (!trace_ok_for_array(t, tr)) {
3988 ret = -EINVAL;
3989 goto out;
3990 }
3991
3992 trace_branch_disable();
3993
3994 tr->current_trace->enabled--;
3995
3996 if (tr->current_trace->reset)
3997 tr->current_trace->reset(tr);
3998
3999 /* Current trace needs to be nop_trace before synchronize_sched */
4000 tr->current_trace = &nop_trace;
4001
4002#ifdef CONFIG_TRACER_MAX_TRACE
4003 had_max_tr = tr->allocated_snapshot;
4004
4005 if (had_max_tr && !t->use_max_tr) {
4006 /*
4007 * We need to make sure that the update_max_tr sees that
4008 * current_trace changed to nop_trace to keep it from
4009 * swapping the buffers after we resize it.
4010 * The update_max_tr is called from interrupts disabled
4011 * so a synchronized_sched() is sufficient.
4012 */
4013 synchronize_sched();
4014 free_snapshot(tr);
4015 }
4016#endif
4017 /* Currently, only the top instance has options */
4018 if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
4019 destroy_trace_option_files(topts);
4020 topts = create_trace_option_files(tr, t);
4021 }
4022
4023#ifdef CONFIG_TRACER_MAX_TRACE
4024 if (t->use_max_tr && !had_max_tr) {
4025 ret = alloc_snapshot(tr);
4026 if (ret < 0)
4027 goto out;
4028 }
4029#endif
4030
4031 if (t->init) {
4032 ret = tracer_init(t, tr);
4033 if (ret)
4034 goto out;
4035 }
4036
4037 tr->current_trace = t;
4038 tr->current_trace->enabled++;
4039 trace_branch_enable(tr);
4040 out:
4041 mutex_unlock(&trace_types_lock);
4042
4043 return ret;
4044}
4045
4046static ssize_t
4047tracing_set_trace_write(struct file *filp, const char __user *ubuf,
4048 size_t cnt, loff_t *ppos)
4049{
4050 struct trace_array *tr = filp->private_data;
4051 char buf[MAX_TRACER_SIZE+1];
4052 int i;
4053 size_t ret;
4054 int err;
4055
4056 ret = cnt;
4057
4058 if (cnt > MAX_TRACER_SIZE)
4059 cnt = MAX_TRACER_SIZE;
4060
4061 if (copy_from_user(&buf, ubuf, cnt))
4062 return -EFAULT;
4063
4064 buf[cnt] = 0;
4065
4066 /* strip ending whitespace. */
4067 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
4068 buf[i] = 0;
4069
4070 err = tracing_set_tracer(tr, buf);
4071 if (err)
4072 return err;
4073
4074 *ppos += ret;
4075
4076 return ret;
4077}
4078
4079static ssize_t
4080tracing_max_lat_read(struct file *filp, char __user *ubuf,
4081 size_t cnt, loff_t *ppos)
4082{
4083 unsigned long *ptr = filp->private_data;
4084 char buf[64];
4085 int r;
4086
4087 r = snprintf(buf, sizeof(buf), "%ld\n",
4088 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
4089 if (r > sizeof(buf))
4090 r = sizeof(buf);
4091 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4092}
4093
4094static ssize_t
4095tracing_max_lat_write(struct file *filp, const char __user *ubuf,
4096 size_t cnt, loff_t *ppos)
4097{
4098 unsigned long *ptr = filp->private_data;
4099 unsigned long val;
4100 int ret;
4101
4102 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4103 if (ret)
4104 return ret;
4105
4106 *ptr = val * 1000;
4107
4108 return cnt;
4109}
4110
4111static int tracing_open_pipe(struct inode *inode, struct file *filp)
4112{
4113 struct trace_array *tr = inode->i_private;
4114 struct trace_iterator *iter;
4115 int ret = 0;
4116
4117 if (tracing_disabled)
4118 return -ENODEV;
4119
4120 if (trace_array_get(tr) < 0)
4121 return -ENODEV;
4122
4123 mutex_lock(&trace_types_lock);
4124
4125 /* create a buffer to store the information to pass to userspace */
4126 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4127 if (!iter) {
4128 ret = -ENOMEM;
4129 __trace_array_put(tr);
4130 goto out;
4131 }
4132
4133 /*
4134 * We make a copy of the current tracer to avoid concurrent
4135 * changes on it while we are reading.
4136 */
4137 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
4138 if (!iter->trace) {
4139 ret = -ENOMEM;
4140 goto fail;
4141 }
4142 *iter->trace = *tr->current_trace;
4143
4144 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
4145 ret = -ENOMEM;
4146 goto fail;
4147 }
4148
4149 /* trace pipe does not show start of buffer */
4150 cpumask_setall(iter->started);
4151
4152 if (trace_flags & TRACE_ITER_LATENCY_FMT)
4153 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4154
4155 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4156 if (trace_clocks[tr->clock_id].in_ns)
4157 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4158
4159 iter->tr = tr;
4160 iter->trace_buffer = &tr->trace_buffer;
4161 iter->cpu_file = tracing_get_cpu(inode);
4162 mutex_init(&iter->mutex);
4163 filp->private_data = iter;
4164
4165 if (iter->trace->pipe_open)
4166 iter->trace->pipe_open(iter);
4167
4168 nonseekable_open(inode, filp);
4169out:
4170 mutex_unlock(&trace_types_lock);
4171 return ret;
4172
4173fail:
4174 kfree(iter->trace);
4175 kfree(iter);
4176 __trace_array_put(tr);
4177 mutex_unlock(&trace_types_lock);
4178 return ret;
4179}
4180
4181static int tracing_release_pipe(struct inode *inode, struct file *file)
4182{
4183 struct trace_iterator *iter = file->private_data;
4184 struct trace_array *tr = inode->i_private;
4185
4186 mutex_lock(&trace_types_lock);
4187
4188 if (iter->trace->pipe_close)
4189 iter->trace->pipe_close(iter);
4190
4191 mutex_unlock(&trace_types_lock);
4192
4193 free_cpumask_var(iter->started);
4194 mutex_destroy(&iter->mutex);
4195 kfree(iter->trace);
4196 kfree(iter);
4197
4198 trace_array_put(tr);
4199
4200 return 0;
4201}
4202
4203static unsigned int
4204trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4205{
4206 /* Iterators are static, they should be filled or empty */
4207 if (trace_buffer_iter(iter, iter->cpu_file))
4208 return POLLIN | POLLRDNORM;
4209
4210 if (trace_flags & TRACE_ITER_BLOCK)
4211 /*
4212 * Always select as readable when in blocking mode
4213 */
4214 return POLLIN | POLLRDNORM;
4215 else
4216 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4217 filp, poll_table);
4218}
4219
4220static unsigned int
4221tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4222{
4223 struct trace_iterator *iter = filp->private_data;
4224
4225 return trace_poll(iter, filp, poll_table);
4226}
4227
4228/*
4229 * This is a make-shift waitqueue.
4230 * A tracer might use this callback on some rare cases:
4231 *
4232 * 1) the current tracer might hold the runqueue lock when it wakes up
4233 * a reader, hence a deadlock (sched, function, and function graph tracers)
4234 * 2) the function tracers, trace all functions, we don't want
4235 * the overhead of calling wake_up and friends
4236 * (and tracing them too)
4237 *
4238 * Anyway, this is really very primitive wakeup.
4239 */
4240void poll_wait_pipe(struct trace_iterator *iter)
4241{
4242 set_current_state(TASK_INTERRUPTIBLE);
4243 /* sleep for 100 msecs, and try again. */
4244 schedule_timeout(HZ / 10);
4245}
4246
4247/* Must be called with trace_types_lock mutex held. */
4248static int tracing_wait_pipe(struct file *filp)
4249{
4250 struct trace_iterator *iter = filp->private_data;
4251
4252 while (trace_empty(iter)) {
4253
4254 if ((filp->f_flags & O_NONBLOCK)) {
4255 return -EAGAIN;
4256 }
4257
4258 mutex_unlock(&iter->mutex);
4259
4260 iter->trace->wait_pipe(iter);
4261
4262 mutex_lock(&iter->mutex);
4263
4264 if (signal_pending(current))
4265 return -EINTR;
4266
4267 /*
4268 * We block until we read something and tracing is disabled.
4269 * We still block if tracing is disabled, but we have never
4270 * read anything. This allows a user to cat this file, and
4271 * then enable tracing. But after we have read something,
4272 * we give an EOF when tracing is again disabled.
4273 *
4274 * iter->pos will be 0 if we haven't read anything.
4275 */
4276 if (!tracing_is_on() && iter->pos)
4277 break;
4278 }
4279
4280 return 1;
4281}
4282
4283/*
4284 * Consumer reader.
4285 */
4286static ssize_t
4287tracing_read_pipe(struct file *filp, char __user *ubuf,
4288 size_t cnt, loff_t *ppos)
4289{
4290 struct trace_iterator *iter = filp->private_data;
4291 struct trace_array *tr = iter->tr;
4292 ssize_t sret;
4293
4294 /* return any leftover data */
4295 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4296 if (sret != -EBUSY)
4297 return sret;
4298
4299 trace_seq_init(&iter->seq);
4300
4301 /* copy the tracer to avoid using a global lock all around */
4302 mutex_lock(&trace_types_lock);
4303 if (unlikely(iter->trace->name != tr->current_trace->name))
4304 *iter->trace = *tr->current_trace;
4305 mutex_unlock(&trace_types_lock);
4306
4307 /*
4308 * Avoid more than one consumer on a single file descriptor
4309 * This is just a matter of traces coherency, the ring buffer itself
4310 * is protected.
4311 */
4312 mutex_lock(&iter->mutex);
4313 if (iter->trace->read) {
4314 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4315 if (sret)
4316 goto out;
4317 }
4318
4319waitagain:
4320 sret = tracing_wait_pipe(filp);
4321 if (sret <= 0)
4322 goto out;
4323
4324 /* stop when tracing is finished */
4325 if (trace_empty(iter)) {
4326 sret = 0;
4327 goto out;
4328 }
4329
4330 if (cnt >= PAGE_SIZE)
4331 cnt = PAGE_SIZE - 1;
4332
4333 /* reset all but tr, trace, and overruns */
4334 memset(&iter->seq, 0,
4335 sizeof(struct trace_iterator) -
4336 offsetof(struct trace_iterator, seq));
4337 cpumask_clear(iter->started);
4338 iter->pos = -1;
4339
4340 trace_event_read_lock();
4341 trace_access_lock(iter->cpu_file);
4342 while (trace_find_next_entry_inc(iter) != NULL) {
4343 enum print_line_t ret;
4344 int len = iter->seq.len;
4345
4346 ret = print_trace_line(iter);
4347 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4348 /* don't print partial lines */
4349 iter->seq.len = len;
4350 break;
4351 }
4352 if (ret != TRACE_TYPE_NO_CONSUME)
4353 trace_consume(iter);
4354
4355 if (iter->seq.len >= cnt)
4356 break;
4357
4358 /*
4359 * Setting the full flag means we reached the trace_seq buffer
4360 * size and we should leave by partial output condition above.
4361 * One of the trace_seq_* functions is not used properly.
4362 */
4363 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4364 iter->ent->type);
4365 }
4366 trace_access_unlock(iter->cpu_file);
4367 trace_event_read_unlock();
4368
4369 /* Now copy what we have to the user */
4370 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4371 if (iter->seq.readpos >= iter->seq.len)
4372 trace_seq_init(&iter->seq);
4373
4374 /*
4375 * If there was nothing to send to user, in spite of consuming trace
4376 * entries, go back to wait for more entries.
4377 */
4378 if (sret == -EBUSY)
4379 goto waitagain;
4380
4381out:
4382 mutex_unlock(&iter->mutex);
4383
4384 return sret;
4385}
4386
4387static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4388 unsigned int idx)
4389{
4390 __free_page(spd->pages[idx]);
4391}
4392
4393static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4394 .can_merge = 0,
4395 .confirm = generic_pipe_buf_confirm,
4396 .release = generic_pipe_buf_release,
4397 .steal = generic_pipe_buf_steal,
4398 .get = generic_pipe_buf_get,
4399};
4400
4401static size_t
4402tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4403{
4404 size_t count;
4405 int ret;
4406
4407 /* Seq buffer is page-sized, exactly what we need. */
4408 for (;;) {
4409 count = iter->seq.len;
4410 ret = print_trace_line(iter);
4411 count = iter->seq.len - count;
4412 if (rem < count) {
4413 rem = 0;
4414 iter->seq.len -= count;
4415 break;
4416 }
4417 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4418 iter->seq.len -= count;
4419 break;
4420 }
4421
4422 if (ret != TRACE_TYPE_NO_CONSUME)
4423 trace_consume(iter);
4424 rem -= count;
4425 if (!trace_find_next_entry_inc(iter)) {
4426 rem = 0;
4427 iter->ent = NULL;
4428 break;
4429 }
4430 }
4431
4432 return rem;
4433}
4434
4435static ssize_t tracing_splice_read_pipe(struct file *filp,
4436 loff_t *ppos,
4437 struct pipe_inode_info *pipe,
4438 size_t len,
4439 unsigned int flags)
4440{
4441 struct page *pages_def[PIPE_DEF_BUFFERS];
4442 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4443 struct trace_iterator *iter = filp->private_data;
4444 struct splice_pipe_desc spd = {
4445 .pages = pages_def,
4446 .partial = partial_def,
4447 .nr_pages = 0, /* This gets updated below. */
4448 .nr_pages_max = PIPE_DEF_BUFFERS,
4449 .flags = flags,
4450 .ops = &tracing_pipe_buf_ops,
4451 .spd_release = tracing_spd_release_pipe,
4452 };
4453 struct trace_array *tr = iter->tr;
4454 ssize_t ret;
4455 size_t rem;
4456 unsigned int i;
4457
4458 if (splice_grow_spd(pipe, &spd))
4459 return -ENOMEM;
4460
4461 /* copy the tracer to avoid using a global lock all around */
4462 mutex_lock(&trace_types_lock);
4463 if (unlikely(iter->trace->name != tr->current_trace->name))
4464 *iter->trace = *tr->current_trace;
4465 mutex_unlock(&trace_types_lock);
4466
4467 mutex_lock(&iter->mutex);
4468
4469 if (iter->trace->splice_read) {
4470 ret = iter->trace->splice_read(iter, filp,
4471 ppos, pipe, len, flags);
4472 if (ret)
4473 goto out_err;
4474 }
4475
4476 ret = tracing_wait_pipe(filp);
4477 if (ret <= 0)
4478 goto out_err;
4479
4480 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4481 ret = -EFAULT;
4482 goto out_err;
4483 }
4484
4485 trace_event_read_lock();
4486 trace_access_lock(iter->cpu_file);
4487
4488 /* Fill as many pages as possible. */
4489 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
4490 spd.pages[i] = alloc_page(GFP_KERNEL);
4491 if (!spd.pages[i])
4492 break;
4493
4494 rem = tracing_fill_pipe_page(rem, iter);
4495
4496 /* Copy the data into the page, so we can start over. */
4497 ret = trace_seq_to_buffer(&iter->seq,
4498 page_address(spd.pages[i]),
4499 iter->seq.len);
4500 if (ret < 0) {
4501 __free_page(spd.pages[i]);
4502 break;
4503 }
4504 spd.partial[i].offset = 0;
4505 spd.partial[i].len = iter->seq.len;
4506
4507 trace_seq_init(&iter->seq);
4508 }
4509
4510 trace_access_unlock(iter->cpu_file);
4511 trace_event_read_unlock();
4512 mutex_unlock(&iter->mutex);
4513
4514 spd.nr_pages = i;
4515
4516 ret = splice_to_pipe(pipe, &spd);
4517out:
4518 splice_shrink_spd(&spd);
4519 return ret;
4520
4521out_err:
4522 mutex_unlock(&iter->mutex);
4523 goto out;
4524}
4525
4526static ssize_t
4527tracing_entries_read(struct file *filp, char __user *ubuf,
4528 size_t cnt, loff_t *ppos)
4529{
4530 struct inode *inode = file_inode(filp);
4531 struct trace_array *tr = inode->i_private;
4532 int cpu = tracing_get_cpu(inode);
4533 char buf[64];
4534 int r = 0;
4535 ssize_t ret;
4536
4537 mutex_lock(&trace_types_lock);
4538
4539 if (cpu == RING_BUFFER_ALL_CPUS) {
4540 int cpu, buf_size_same;
4541 unsigned long size;
4542
4543 size = 0;
4544 buf_size_same = 1;
4545 /* check if all cpu sizes are same */
4546 for_each_tracing_cpu(cpu) {
4547 /* fill in the size from first enabled cpu */
4548 if (size == 0)
4549 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4550 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4551 buf_size_same = 0;
4552 break;
4553 }
4554 }
4555
4556 if (buf_size_same) {
4557 if (!ring_buffer_expanded)
4558 r = sprintf(buf, "%lu (expanded: %lu)\n",
4559 size >> 10,
4560 trace_buf_size >> 10);
4561 else
4562 r = sprintf(buf, "%lu\n", size >> 10);
4563 } else
4564 r = sprintf(buf, "X\n");
4565 } else
4566 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
4567
4568 mutex_unlock(&trace_types_lock);
4569
4570 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4571 return ret;
4572}
4573
4574static ssize_t
4575tracing_entries_write(struct file *filp, const char __user *ubuf,
4576 size_t cnt, loff_t *ppos)
4577{
4578 struct inode *inode = file_inode(filp);
4579 struct trace_array *tr = inode->i_private;
4580 unsigned long val;
4581 int ret;
4582
4583 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4584 if (ret)
4585 return ret;
4586
4587 /* must have at least 1 entry */
4588 if (!val)
4589 return -EINVAL;
4590
4591 /* value is in KB */
4592 val <<= 10;
4593 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
4594 if (ret < 0)
4595 return ret;
4596
4597 *ppos += cnt;
4598
4599 return cnt;
4600}
4601
4602static ssize_t
4603tracing_total_entries_read(struct file *filp, char __user *ubuf,
4604 size_t cnt, loff_t *ppos)
4605{
4606 struct trace_array *tr = filp->private_data;
4607 char buf[64];
4608 int r, cpu;
4609 unsigned long size = 0, expanded_size = 0;
4610
4611 mutex_lock(&trace_types_lock);
4612 for_each_tracing_cpu(cpu) {
4613 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4614 if (!ring_buffer_expanded)
4615 expanded_size += trace_buf_size >> 10;
4616 }
4617 if (ring_buffer_expanded)
4618 r = sprintf(buf, "%lu\n", size);
4619 else
4620 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4621 mutex_unlock(&trace_types_lock);
4622
4623 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4624}
4625
4626static ssize_t
4627tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4628 size_t cnt, loff_t *ppos)
4629{
4630 /*
4631 * There is no need to read what the user has written, this function
4632 * is just to make sure that there is no error when "echo" is used
4633 */
4634
4635 *ppos += cnt;
4636
4637 return cnt;
4638}
4639
4640static int
4641tracing_free_buffer_release(struct inode *inode, struct file *filp)
4642{
4643 struct trace_array *tr = inode->i_private;
4644
4645 /* disable tracing ? */
4646 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4647 tracer_tracing_off(tr);
4648 /* resize the ring buffer to 0 */
4649 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4650
4651 trace_array_put(tr);
4652
4653 return 0;
4654}
4655
4656static ssize_t
4657tracing_mark_write(struct file *filp, const char __user *ubuf,
4658 size_t cnt, loff_t *fpos)
4659{
4660 unsigned long addr = (unsigned long)ubuf;
4661 struct trace_array *tr = filp->private_data;
4662 struct ring_buffer_event *event;
4663 struct ring_buffer *buffer;
4664 struct print_entry *entry;
4665 unsigned long irq_flags;
4666 struct page *pages[2];
4667 void *map_page[2];
4668 int nr_pages = 1;
4669 ssize_t written;
4670 int offset;
4671 int size;
4672 int len;
4673 int ret;
4674 int i;
4675
4676 if (tracing_disabled)
4677 return -EINVAL;
4678
4679 if (!(trace_flags & TRACE_ITER_MARKERS))
4680 return -EINVAL;
4681
4682 if (cnt > TRACE_BUF_SIZE)
4683 cnt = TRACE_BUF_SIZE;
4684
4685 /*
4686 * Userspace is injecting traces into the kernel trace buffer.
4687 * We want to be as non intrusive as possible.
4688 * To do so, we do not want to allocate any special buffers
4689 * or take any locks, but instead write the userspace data
4690 * straight into the ring buffer.
4691 *
4692 * First we need to pin the userspace buffer into memory,
4693 * which, most likely it is, because it just referenced it.
4694 * But there's no guarantee that it is. By using get_user_pages_fast()
4695 * and kmap_atomic/kunmap_atomic() we can get access to the
4696 * pages directly. We then write the data directly into the
4697 * ring buffer.
4698 */
4699 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4700
4701 /* check if we cross pages */
4702 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4703 nr_pages = 2;
4704
4705 offset = addr & (PAGE_SIZE - 1);
4706 addr &= PAGE_MASK;
4707
4708 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4709 if (ret < nr_pages) {
4710 while (--ret >= 0)
4711 put_page(pages[ret]);
4712 written = -EFAULT;
4713 goto out;
4714 }
4715
4716 for (i = 0; i < nr_pages; i++)
4717 map_page[i] = kmap_atomic(pages[i]);
4718
4719 local_save_flags(irq_flags);
4720 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4721 buffer = tr->trace_buffer.buffer;
4722 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4723 irq_flags, preempt_count());
4724 if (!event) {
4725 /* Ring buffer disabled, return as if not open for write */
4726 written = -EBADF;
4727 goto out_unlock;
4728 }
4729
4730 entry = ring_buffer_event_data(event);
4731 entry->ip = _THIS_IP_;
4732
4733 if (nr_pages == 2) {
4734 len = PAGE_SIZE - offset;
4735 memcpy(&entry->buf, map_page[0] + offset, len);
4736 memcpy(&entry->buf[len], map_page[1], cnt - len);
4737 } else
4738 memcpy(&entry->buf, map_page[0] + offset, cnt);
4739
4740 if (entry->buf[cnt - 1] != '\n') {
4741 entry->buf[cnt] = '\n';
4742 entry->buf[cnt + 1] = '\0';
4743 } else
4744 entry->buf[cnt] = '\0';
4745
4746 __buffer_unlock_commit(buffer, event);
4747
4748 written = cnt;
4749
4750 *fpos += written;
4751
4752 out_unlock:
4753 for (i = 0; i < nr_pages; i++){
4754 kunmap_atomic(map_page[i]);
4755 put_page(pages[i]);
4756 }
4757 out:
4758 return written;
4759}
4760
4761static int tracing_clock_show(struct seq_file *m, void *v)
4762{
4763 struct trace_array *tr = m->private;
4764 int i;
4765
4766 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4767 seq_printf(m,
4768 "%s%s%s%s", i ? " " : "",
4769 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4770 i == tr->clock_id ? "]" : "");
4771 seq_putc(m, '\n');
4772
4773 return 0;
4774}
4775
4776static int tracing_set_clock(struct trace_array *tr, const char *clockstr)
4777{
4778 int i;
4779
4780 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4781 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4782 break;
4783 }
4784 if (i == ARRAY_SIZE(trace_clocks))
4785 return -EINVAL;
4786
4787 mutex_lock(&trace_types_lock);
4788
4789 tr->clock_id = i;
4790
4791 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4792
4793 /*
4794 * New clock may not be consistent with the previous clock.
4795 * Reset the buffer so that it doesn't have incomparable timestamps.
4796 */
4797 tracing_reset_online_cpus(&tr->trace_buffer);
4798
4799#ifdef CONFIG_TRACER_MAX_TRACE
4800 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4801 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4802 tracing_reset_online_cpus(&tr->max_buffer);
4803#endif
4804
4805 mutex_unlock(&trace_types_lock);
4806
4807 return 0;
4808}
4809
4810static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4811 size_t cnt, loff_t *fpos)
4812{
4813 struct seq_file *m = filp->private_data;
4814 struct trace_array *tr = m->private;
4815 char buf[64];
4816 const char *clockstr;
4817 int ret;
4818
4819 if (cnt >= sizeof(buf))
4820 return -EINVAL;
4821
4822 if (copy_from_user(&buf, ubuf, cnt))
4823 return -EFAULT;
4824
4825 buf[cnt] = 0;
4826
4827 clockstr = strstrip(buf);
4828
4829 ret = tracing_set_clock(tr, clockstr);
4830 if (ret)
4831 return ret;
4832
4833 *fpos += cnt;
4834
4835 return cnt;
4836}
4837
4838static int tracing_clock_open(struct inode *inode, struct file *file)
4839{
4840 struct trace_array *tr = inode->i_private;
4841 int ret;
4842
4843 if (tracing_disabled)
4844 return -ENODEV;
4845
4846 if (trace_array_get(tr))
4847 return -ENODEV;
4848
4849 ret = single_open(file, tracing_clock_show, inode->i_private);
4850 if (ret < 0)
4851 trace_array_put(tr);
4852
4853 return ret;
4854}
4855
4856struct ftrace_buffer_info {
4857 struct trace_iterator iter;
4858 void *spare;
4859 unsigned int read;
4860};
4861
4862#ifdef CONFIG_TRACER_SNAPSHOT
4863static int tracing_snapshot_open(struct inode *inode, struct file *file)
4864{
4865 struct trace_array *tr = inode->i_private;
4866 struct trace_iterator *iter;
4867 struct seq_file *m;
4868 int ret = 0;
4869
4870 if (trace_array_get(tr) < 0)
4871 return -ENODEV;
4872
4873 if (file->f_mode & FMODE_READ) {
4874 iter = __tracing_open(inode, file, true);
4875 if (IS_ERR(iter))
4876 ret = PTR_ERR(iter);
4877 } else {
4878 /* Writes still need the seq_file to hold the private data */
4879 ret = -ENOMEM;
4880 m = kzalloc(sizeof(*m), GFP_KERNEL);
4881 if (!m)
4882 goto out;
4883 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4884 if (!iter) {
4885 kfree(m);
4886 goto out;
4887 }
4888 ret = 0;
4889
4890 iter->tr = tr;
4891 iter->trace_buffer = &tr->max_buffer;
4892 iter->cpu_file = tracing_get_cpu(inode);
4893 m->private = iter;
4894 file->private_data = m;
4895 }
4896out:
4897 if (ret < 0)
4898 trace_array_put(tr);
4899
4900 return ret;
4901}
4902
4903static ssize_t
4904tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4905 loff_t *ppos)
4906{
4907 struct seq_file *m = filp->private_data;
4908 struct trace_iterator *iter = m->private;
4909 struct trace_array *tr = iter->tr;
4910 unsigned long val;
4911 int ret;
4912
4913 ret = tracing_update_buffers();
4914 if (ret < 0)
4915 return ret;
4916
4917 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4918 if (ret)
4919 return ret;
4920
4921 mutex_lock(&trace_types_lock);
4922
4923 if (tr->current_trace->use_max_tr) {
4924 ret = -EBUSY;
4925 goto out;
4926 }
4927
4928 switch (val) {
4929 case 0:
4930 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4931 ret = -EINVAL;
4932 break;
4933 }
4934 if (tr->allocated_snapshot)
4935 free_snapshot(tr);
4936 break;
4937 case 1:
4938/* Only allow per-cpu swap if the ring buffer supports it */
4939#ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4940 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4941 ret = -EINVAL;
4942 break;
4943 }
4944#endif
4945 if (!tr->allocated_snapshot) {
4946 ret = alloc_snapshot(tr);
4947 if (ret < 0)
4948 break;
4949 }
4950 local_irq_disable();
4951 /* Now, we're going to swap */
4952 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4953 update_max_tr(tr, current, smp_processor_id());
4954 else
4955 update_max_tr_single(tr, current, iter->cpu_file);
4956 local_irq_enable();
4957 break;
4958 default:
4959 if (tr->allocated_snapshot) {
4960 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4961 tracing_reset_online_cpus(&tr->max_buffer);
4962 else
4963 tracing_reset(&tr->max_buffer, iter->cpu_file);
4964 }
4965 break;
4966 }
4967
4968 if (ret >= 0) {
4969 *ppos += cnt;
4970 ret = cnt;
4971 }
4972out:
4973 mutex_unlock(&trace_types_lock);
4974 return ret;
4975}
4976
4977static int tracing_snapshot_release(struct inode *inode, struct file *file)
4978{
4979 struct seq_file *m = file->private_data;
4980 int ret;
4981
4982 ret = tracing_release(inode, file);
4983
4984 if (file->f_mode & FMODE_READ)
4985 return ret;
4986
4987 /* If write only, the seq_file is just a stub */
4988 if (m)
4989 kfree(m->private);
4990 kfree(m);
4991
4992 return 0;
4993}
4994
4995static int tracing_buffers_open(struct inode *inode, struct file *filp);
4996static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4997 size_t count, loff_t *ppos);
4998static int tracing_buffers_release(struct inode *inode, struct file *file);
4999static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5000 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
5001
5002static int snapshot_raw_open(struct inode *inode, struct file *filp)
5003{
5004 struct ftrace_buffer_info *info;
5005 int ret;
5006
5007 ret = tracing_buffers_open(inode, filp);
5008 if (ret < 0)
5009 return ret;
5010
5011 info = filp->private_data;
5012
5013 if (info->iter.trace->use_max_tr) {
5014 tracing_buffers_release(inode, filp);
5015 return -EBUSY;
5016 }
5017
5018 info->iter.snapshot = true;
5019 info->iter.trace_buffer = &info->iter.tr->max_buffer;
5020
5021 return ret;
5022}
5023
5024#endif /* CONFIG_TRACER_SNAPSHOT */
5025
5026
5027static const struct file_operations tracing_max_lat_fops = {
5028 .open = tracing_open_generic,
5029 .read = tracing_max_lat_read,
5030 .write = tracing_max_lat_write,
5031 .llseek = generic_file_llseek,
5032};
5033
5034static const struct file_operations set_tracer_fops = {
5035 .open = tracing_open_generic,
5036 .read = tracing_set_trace_read,
5037 .write = tracing_set_trace_write,
5038 .llseek = generic_file_llseek,
5039};
5040
5041static const struct file_operations tracing_pipe_fops = {
5042 .open = tracing_open_pipe,
5043 .poll = tracing_poll_pipe,
5044 .read = tracing_read_pipe,
5045 .splice_read = tracing_splice_read_pipe,
5046 .release = tracing_release_pipe,
5047 .llseek = no_llseek,
5048};
5049
5050static const struct file_operations tracing_entries_fops = {
5051 .open = tracing_open_generic_tr,
5052 .read = tracing_entries_read,
5053 .write = tracing_entries_write,
5054 .llseek = generic_file_llseek,
5055 .release = tracing_release_generic_tr,
5056};
5057
5058static const struct file_operations tracing_total_entries_fops = {
5059 .open = tracing_open_generic_tr,
5060 .read = tracing_total_entries_read,
5061 .llseek = generic_file_llseek,
5062 .release = tracing_release_generic_tr,
5063};
5064
5065static const struct file_operations tracing_free_buffer_fops = {
5066 .open = tracing_open_generic_tr,
5067 .write = tracing_free_buffer_write,
5068 .release = tracing_free_buffer_release,
5069};
5070
5071static const struct file_operations tracing_mark_fops = {
5072 .open = tracing_open_generic_tr,
5073 .write = tracing_mark_write,
5074 .llseek = generic_file_llseek,
5075 .release = tracing_release_generic_tr,
5076};
5077
5078static const struct file_operations trace_clock_fops = {
5079 .open = tracing_clock_open,
5080 .read = seq_read,
5081 .llseek = seq_lseek,
5082 .release = tracing_single_release_tr,
5083 .write = tracing_clock_write,
5084};
5085
5086#ifdef CONFIG_TRACER_SNAPSHOT
5087static const struct file_operations snapshot_fops = {
5088 .open = tracing_snapshot_open,
5089 .read = seq_read,
5090 .write = tracing_snapshot_write,
5091 .llseek = tracing_lseek,
5092 .release = tracing_snapshot_release,
5093};
5094
5095static const struct file_operations snapshot_raw_fops = {
5096 .open = snapshot_raw_open,
5097 .read = tracing_buffers_read,
5098 .release = tracing_buffers_release,
5099 .splice_read = tracing_buffers_splice_read,
5100 .llseek = no_llseek,
5101};
5102
5103#endif /* CONFIG_TRACER_SNAPSHOT */
5104
5105static int tracing_buffers_open(struct inode *inode, struct file *filp)
5106{
5107 struct trace_array *tr = inode->i_private;
5108 struct ftrace_buffer_info *info;
5109 int ret;
5110
5111 if (tracing_disabled)
5112 return -ENODEV;
5113
5114 if (trace_array_get(tr) < 0)
5115 return -ENODEV;
5116
5117 info = kzalloc(sizeof(*info), GFP_KERNEL);
5118 if (!info) {
5119 trace_array_put(tr);
5120 return -ENOMEM;
5121 }
5122
5123 mutex_lock(&trace_types_lock);
5124
5125 info->iter.tr = tr;
5126 info->iter.cpu_file = tracing_get_cpu(inode);
5127 info->iter.trace = tr->current_trace;
5128 info->iter.trace_buffer = &tr->trace_buffer;
5129 info->spare = NULL;
5130 /* Force reading ring buffer for first read */
5131 info->read = (unsigned int)-1;
5132
5133 filp->private_data = info;
5134
5135 mutex_unlock(&trace_types_lock);
5136
5137 ret = nonseekable_open(inode, filp);
5138 if (ret < 0)
5139 trace_array_put(tr);
5140
5141 return ret;
5142}
5143
5144static unsigned int
5145tracing_buffers_poll(struct file *filp, poll_table *poll_table)
5146{
5147 struct ftrace_buffer_info *info = filp->private_data;
5148 struct trace_iterator *iter = &info->iter;
5149
5150 return trace_poll(iter, filp, poll_table);
5151}
5152
5153static ssize_t
5154tracing_buffers_read(struct file *filp, char __user *ubuf,
5155 size_t count, loff_t *ppos)
5156{
5157 struct ftrace_buffer_info *info = filp->private_data;
5158 struct trace_iterator *iter = &info->iter;
5159 ssize_t ret;
5160 ssize_t size;
5161
5162 if (!count)
5163 return 0;
5164
5165 mutex_lock(&trace_types_lock);
5166
5167#ifdef CONFIG_TRACER_MAX_TRACE
5168 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5169 size = -EBUSY;
5170 goto out_unlock;
5171 }
5172#endif
5173
5174 if (!info->spare)
5175 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
5176 iter->cpu_file);
5177 size = -ENOMEM;
5178 if (!info->spare)
5179 goto out_unlock;
5180
5181 /* Do we have previous read data to read? */
5182 if (info->read < PAGE_SIZE)
5183 goto read;
5184
5185 again:
5186 trace_access_lock(iter->cpu_file);
5187 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5188 &info->spare,
5189 count,
5190 iter->cpu_file, 0);
5191 trace_access_unlock(iter->cpu_file);
5192
5193 if (ret < 0) {
5194 if (trace_empty(iter)) {
5195 if ((filp->f_flags & O_NONBLOCK)) {
5196 size = -EAGAIN;
5197 goto out_unlock;
5198 }
5199 mutex_unlock(&trace_types_lock);
5200 iter->trace->wait_pipe(iter);
5201 mutex_lock(&trace_types_lock);
5202 if (signal_pending(current)) {
5203 size = -EINTR;
5204 goto out_unlock;
5205 }
5206 goto again;
5207 }
5208 size = 0;
5209 goto out_unlock;
5210 }
5211
5212 info->read = 0;
5213 read:
5214 size = PAGE_SIZE - info->read;
5215 if (size > count)
5216 size = count;
5217
5218 ret = copy_to_user(ubuf, info->spare + info->read, size);
5219 if (ret == size) {
5220 size = -EFAULT;
5221 goto out_unlock;
5222 }
5223 size -= ret;
5224
5225 *ppos += size;
5226 info->read += size;
5227
5228 out_unlock:
5229 mutex_unlock(&trace_types_lock);
5230
5231 return size;
5232}
5233
5234static int tracing_buffers_release(struct inode *inode, struct file *file)
5235{
5236 struct ftrace_buffer_info *info = file->private_data;
5237 struct trace_iterator *iter = &info->iter;
5238
5239 mutex_lock(&trace_types_lock);
5240
5241 __trace_array_put(iter->tr);
5242
5243 if (info->spare)
5244 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5245 kfree(info);
5246
5247 mutex_unlock(&trace_types_lock);
5248
5249 return 0;
5250}
5251
5252struct buffer_ref {
5253 struct ring_buffer *buffer;
5254 void *page;
5255 int ref;
5256};
5257
5258static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5259 struct pipe_buffer *buf)
5260{
5261 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5262
5263 if (--ref->ref)
5264 return;
5265
5266 ring_buffer_free_read_page(ref->buffer, ref->page);
5267 kfree(ref);
5268 buf->private = 0;
5269}
5270
5271static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5272 struct pipe_buffer *buf)
5273{
5274 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5275
5276 ref->ref++;
5277}
5278
5279/* Pipe buffer operations for a buffer. */
5280static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5281 .can_merge = 0,
5282 .confirm = generic_pipe_buf_confirm,
5283 .release = buffer_pipe_buf_release,
5284 .steal = generic_pipe_buf_steal,
5285 .get = buffer_pipe_buf_get,
5286};
5287
5288/*
5289 * Callback from splice_to_pipe(), if we need to release some pages
5290 * at the end of the spd in case we error'ed out in filling the pipe.
5291 */
5292static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5293{
5294 struct buffer_ref *ref =
5295 (struct buffer_ref *)spd->partial[i].private;
5296
5297 if (--ref->ref)
5298 return;
5299
5300 ring_buffer_free_read_page(ref->buffer, ref->page);
5301 kfree(ref);
5302 spd->partial[i].private = 0;
5303}
5304
5305static ssize_t
5306tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5307 struct pipe_inode_info *pipe, size_t len,
5308 unsigned int flags)
5309{
5310 struct ftrace_buffer_info *info = file->private_data;
5311 struct trace_iterator *iter = &info->iter;
5312 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5313 struct page *pages_def[PIPE_DEF_BUFFERS];
5314 struct splice_pipe_desc spd = {
5315 .pages = pages_def,
5316 .partial = partial_def,
5317 .nr_pages_max = PIPE_DEF_BUFFERS,
5318 .flags = flags,
5319 .ops = &buffer_pipe_buf_ops,
5320 .spd_release = buffer_spd_release,
5321 };
5322 struct buffer_ref *ref;
5323 int entries, size, i;
5324 ssize_t ret;
5325
5326 mutex_lock(&trace_types_lock);
5327
5328#ifdef CONFIG_TRACER_MAX_TRACE
5329 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5330 ret = -EBUSY;
5331 goto out;
5332 }
5333#endif
5334
5335 if (splice_grow_spd(pipe, &spd)) {
5336 ret = -ENOMEM;
5337 goto out;
5338 }
5339
5340 if (*ppos & (PAGE_SIZE - 1)) {
5341 ret = -EINVAL;
5342 goto out;
5343 }
5344
5345 if (len & (PAGE_SIZE - 1)) {
5346 if (len < PAGE_SIZE) {
5347 ret = -EINVAL;
5348 goto out;
5349 }
5350 len &= PAGE_MASK;
5351 }
5352
5353 again:
5354 trace_access_lock(iter->cpu_file);
5355 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5356
5357 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
5358 struct page *page;
5359 int r;
5360
5361 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5362 if (!ref)
5363 break;
5364
5365 ref->ref = 1;
5366 ref->buffer = iter->trace_buffer->buffer;
5367 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5368 if (!ref->page) {
5369 kfree(ref);
5370 break;
5371 }
5372
5373 r = ring_buffer_read_page(ref->buffer, &ref->page,
5374 len, iter->cpu_file, 1);
5375 if (r < 0) {
5376 ring_buffer_free_read_page(ref->buffer, ref->page);
5377 kfree(ref);
5378 break;
5379 }
5380
5381 /*
5382 * zero out any left over data, this is going to
5383 * user land.
5384 */
5385 size = ring_buffer_page_len(ref->page);
5386 if (size < PAGE_SIZE)
5387 memset(ref->page + size, 0, PAGE_SIZE - size);
5388
5389 page = virt_to_page(ref->page);
5390
5391 spd.pages[i] = page;
5392 spd.partial[i].len = PAGE_SIZE;
5393 spd.partial[i].offset = 0;
5394 spd.partial[i].private = (unsigned long)ref;
5395 spd.nr_pages++;
5396 *ppos += PAGE_SIZE;
5397
5398 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5399 }
5400
5401 trace_access_unlock(iter->cpu_file);
5402 spd.nr_pages = i;
5403
5404 /* did we read anything? */
5405 if (!spd.nr_pages) {
5406 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5407 ret = -EAGAIN;
5408 goto out;
5409 }
5410 mutex_unlock(&trace_types_lock);
5411 iter->trace->wait_pipe(iter);
5412 mutex_lock(&trace_types_lock);
5413 if (signal_pending(current)) {
5414 ret = -EINTR;
5415 goto out;
5416 }
5417 goto again;
5418 }
5419
5420 ret = splice_to_pipe(pipe, &spd);
5421 splice_shrink_spd(&spd);
5422out:
5423 mutex_unlock(&trace_types_lock);
5424
5425 return ret;
5426}
5427
5428static const struct file_operations tracing_buffers_fops = {
5429 .open = tracing_buffers_open,
5430 .read = tracing_buffers_read,
5431 .poll = tracing_buffers_poll,
5432 .release = tracing_buffers_release,
5433 .splice_read = tracing_buffers_splice_read,
5434 .llseek = no_llseek,
5435};
5436
5437static ssize_t
5438tracing_stats_read(struct file *filp, char __user *ubuf,
5439 size_t count, loff_t *ppos)
5440{
5441 struct inode *inode = file_inode(filp);
5442 struct trace_array *tr = inode->i_private;
5443 struct trace_buffer *trace_buf = &tr->trace_buffer;
5444 int cpu = tracing_get_cpu(inode);
5445 struct trace_seq *s;
5446 unsigned long cnt;
5447 unsigned long long t;
5448 unsigned long usec_rem;
5449
5450 s = kmalloc(sizeof(*s), GFP_KERNEL);
5451 if (!s)
5452 return -ENOMEM;
5453
5454 trace_seq_init(s);
5455
5456 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5457 trace_seq_printf(s, "entries: %ld\n", cnt);
5458
5459 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5460 trace_seq_printf(s, "overrun: %ld\n", cnt);
5461
5462 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5463 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5464
5465 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5466 trace_seq_printf(s, "bytes: %ld\n", cnt);
5467
5468 if (trace_clocks[tr->clock_id].in_ns) {
5469 /* local or global for trace_clock */
5470 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5471 usec_rem = do_div(t, USEC_PER_SEC);
5472 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5473 t, usec_rem);
5474
5475 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5476 usec_rem = do_div(t, USEC_PER_SEC);
5477 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5478 } else {
5479 /* counter or tsc mode for trace_clock */
5480 trace_seq_printf(s, "oldest event ts: %llu\n",
5481 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5482
5483 trace_seq_printf(s, "now ts: %llu\n",
5484 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5485 }
5486
5487 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5488 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5489
5490 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5491 trace_seq_printf(s, "read events: %ld\n", cnt);
5492
5493 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5494
5495 kfree(s);
5496
5497 return count;
5498}
5499
5500static const struct file_operations tracing_stats_fops = {
5501 .open = tracing_open_generic_tr,
5502 .read = tracing_stats_read,
5503 .llseek = generic_file_llseek,
5504 .release = tracing_release_generic_tr,
5505};
5506
5507#ifdef CONFIG_DYNAMIC_FTRACE
5508
5509int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5510{
5511 return 0;
5512}
5513
5514static ssize_t
5515tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5516 size_t cnt, loff_t *ppos)
5517{
5518 static char ftrace_dyn_info_buffer[1024];
5519 static DEFINE_MUTEX(dyn_info_mutex);
5520 unsigned long *p = filp->private_data;
5521 char *buf = ftrace_dyn_info_buffer;
5522 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5523 int r;
5524
5525 mutex_lock(&dyn_info_mutex);
5526 r = sprintf(buf, "%ld ", *p);
5527
5528 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5529 buf[r++] = '\n';
5530
5531 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5532
5533 mutex_unlock(&dyn_info_mutex);
5534
5535 return r;
5536}
5537
5538static const struct file_operations tracing_dyn_info_fops = {
5539 .open = tracing_open_generic,
5540 .read = tracing_read_dyn_info,
5541 .llseek = generic_file_llseek,
5542};
5543#endif /* CONFIG_DYNAMIC_FTRACE */
5544
5545#if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5546static void
5547ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5548{
5549 tracing_snapshot();
5550}
5551
5552static void
5553ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5554{
5555 unsigned long *count = (long *)data;
5556
5557 if (!*count)
5558 return;
5559
5560 if (*count != -1)
5561 (*count)--;
5562
5563 tracing_snapshot();
5564}
5565
5566static int
5567ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5568 struct ftrace_probe_ops *ops, void *data)
5569{
5570 long count = (long)data;
5571
5572 seq_printf(m, "%ps:", (void *)ip);
5573
5574 seq_printf(m, "snapshot");
5575
5576 if (count == -1)
5577 seq_printf(m, ":unlimited\n");
5578 else
5579 seq_printf(m, ":count=%ld\n", count);
5580
5581 return 0;
5582}
5583
5584static struct ftrace_probe_ops snapshot_probe_ops = {
5585 .func = ftrace_snapshot,
5586 .print = ftrace_snapshot_print,
5587};
5588
5589static struct ftrace_probe_ops snapshot_count_probe_ops = {
5590 .func = ftrace_count_snapshot,
5591 .print = ftrace_snapshot_print,
5592};
5593
5594static int
5595ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5596 char *glob, char *cmd, char *param, int enable)
5597{
5598 struct ftrace_probe_ops *ops;
5599 void *count = (void *)-1;
5600 char *number;
5601 int ret;
5602
5603 /* hash funcs only work with set_ftrace_filter */
5604 if (!enable)
5605 return -EINVAL;
5606
5607 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5608
5609 if (glob[0] == '!') {
5610 unregister_ftrace_function_probe_func(glob+1, ops);
5611 return 0;
5612 }
5613
5614 if (!param)
5615 goto out_reg;
5616
5617 number = strsep(¶m, ":");
5618
5619 if (!strlen(number))
5620 goto out_reg;
5621
5622 /*
5623 * We use the callback data field (which is a pointer)
5624 * as our counter.
5625 */
5626 ret = kstrtoul(number, 0, (unsigned long *)&count);
5627 if (ret)
5628 return ret;
5629
5630 out_reg:
5631 ret = register_ftrace_function_probe(glob, ops, count);
5632
5633 if (ret >= 0)
5634 alloc_snapshot(&global_trace);
5635
5636 return ret < 0 ? ret : 0;
5637}
5638
5639static struct ftrace_func_command ftrace_snapshot_cmd = {
5640 .name = "snapshot",
5641 .func = ftrace_trace_snapshot_callback,
5642};
5643
5644static __init int register_snapshot_cmd(void)
5645{
5646 return register_ftrace_command(&ftrace_snapshot_cmd);
5647}
5648#else
5649static inline __init int register_snapshot_cmd(void) { return 0; }
5650#endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5651
5652struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5653{
5654 if (tr->dir)
5655 return tr->dir;
5656
5657 if (!debugfs_initialized())
5658 return NULL;
5659
5660 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5661 tr->dir = debugfs_create_dir("tracing", NULL);
5662
5663 if (!tr->dir)
5664 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5665
5666 return tr->dir;
5667}
5668
5669struct dentry *tracing_init_dentry(void)
5670{
5671 return tracing_init_dentry_tr(&global_trace);
5672}
5673
5674static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5675{
5676 struct dentry *d_tracer;
5677
5678 if (tr->percpu_dir)
5679 return tr->percpu_dir;
5680
5681 d_tracer = tracing_init_dentry_tr(tr);
5682 if (!d_tracer)
5683 return NULL;
5684
5685 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5686
5687 WARN_ONCE(!tr->percpu_dir,
5688 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5689
5690 return tr->percpu_dir;
5691}
5692
5693static struct dentry *
5694trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
5695 void *data, long cpu, const struct file_operations *fops)
5696{
5697 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
5698
5699 if (ret) /* See tracing_get_cpu() */
5700 ret->d_inode->i_cdev = (void *)(cpu + 1);
5701 return ret;
5702}
5703
5704static void
5705tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5706{
5707 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5708 struct dentry *d_cpu;
5709 char cpu_dir[30]; /* 30 characters should be more than enough */
5710
5711 if (!d_percpu)
5712 return;
5713
5714 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5715 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5716 if (!d_cpu) {
5717 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5718 return;
5719 }
5720
5721 /* per cpu trace_pipe */
5722 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
5723 tr, cpu, &tracing_pipe_fops);
5724
5725 /* per cpu trace */
5726 trace_create_cpu_file("trace", 0644, d_cpu,
5727 tr, cpu, &tracing_fops);
5728
5729 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
5730 tr, cpu, &tracing_buffers_fops);
5731
5732 trace_create_cpu_file("stats", 0444, d_cpu,
5733 tr, cpu, &tracing_stats_fops);
5734
5735 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
5736 tr, cpu, &tracing_entries_fops);
5737
5738#ifdef CONFIG_TRACER_SNAPSHOT
5739 trace_create_cpu_file("snapshot", 0644, d_cpu,
5740 tr, cpu, &snapshot_fops);
5741
5742 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
5743 tr, cpu, &snapshot_raw_fops);
5744#endif
5745}
5746
5747#ifdef CONFIG_FTRACE_SELFTEST
5748/* Let selftest have access to static functions in this file */
5749#include "trace_selftest.c"
5750#endif
5751
5752struct trace_option_dentry {
5753 struct tracer_opt *opt;
5754 struct tracer_flags *flags;
5755 struct trace_array *tr;
5756 struct dentry *entry;
5757};
5758
5759static ssize_t
5760trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5761 loff_t *ppos)
5762{
5763 struct trace_option_dentry *topt = filp->private_data;
5764 char *buf;
5765
5766 if (topt->flags->val & topt->opt->bit)
5767 buf = "1\n";
5768 else
5769 buf = "0\n";
5770
5771 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5772}
5773
5774static ssize_t
5775trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5776 loff_t *ppos)
5777{
5778 struct trace_option_dentry *topt = filp->private_data;
5779 unsigned long val;
5780 int ret;
5781
5782 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5783 if (ret)
5784 return ret;
5785
5786 if (val != 0 && val != 1)
5787 return -EINVAL;
5788
5789 if (!!(topt->flags->val & topt->opt->bit) != val) {
5790 mutex_lock(&trace_types_lock);
5791 ret = __set_tracer_option(topt->tr, topt->flags,
5792 topt->opt, !val);
5793 mutex_unlock(&trace_types_lock);
5794 if (ret)
5795 return ret;
5796 }
5797
5798 *ppos += cnt;
5799
5800 return cnt;
5801}
5802
5803
5804static const struct file_operations trace_options_fops = {
5805 .open = tracing_open_generic,
5806 .read = trace_options_read,
5807 .write = trace_options_write,
5808 .llseek = generic_file_llseek,
5809};
5810
5811static ssize_t
5812trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5813 loff_t *ppos)
5814{
5815 long index = (long)filp->private_data;
5816 char *buf;
5817
5818 if (trace_flags & (1 << index))
5819 buf = "1\n";
5820 else
5821 buf = "0\n";
5822
5823 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5824}
5825
5826static ssize_t
5827trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5828 loff_t *ppos)
5829{
5830 struct trace_array *tr = &global_trace;
5831 long index = (long)filp->private_data;
5832 unsigned long val;
5833 int ret;
5834
5835 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5836 if (ret)
5837 return ret;
5838
5839 if (val != 0 && val != 1)
5840 return -EINVAL;
5841
5842 mutex_lock(&trace_types_lock);
5843 ret = set_tracer_flag(tr, 1 << index, val);
5844 mutex_unlock(&trace_types_lock);
5845
5846 if (ret < 0)
5847 return ret;
5848
5849 *ppos += cnt;
5850
5851 return cnt;
5852}
5853
5854static const struct file_operations trace_options_core_fops = {
5855 .open = tracing_open_generic,
5856 .read = trace_options_core_read,
5857 .write = trace_options_core_write,
5858 .llseek = generic_file_llseek,
5859};
5860
5861struct dentry *trace_create_file(const char *name,
5862 umode_t mode,
5863 struct dentry *parent,
5864 void *data,
5865 const struct file_operations *fops)
5866{
5867 struct dentry *ret;
5868
5869 ret = debugfs_create_file(name, mode, parent, data, fops);
5870 if (!ret)
5871 pr_warning("Could not create debugfs '%s' entry\n", name);
5872
5873 return ret;
5874}
5875
5876
5877static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5878{
5879 struct dentry *d_tracer;
5880
5881 if (tr->options)
5882 return tr->options;
5883
5884 d_tracer = tracing_init_dentry_tr(tr);
5885 if (!d_tracer)
5886 return NULL;
5887
5888 tr->options = debugfs_create_dir("options", d_tracer);
5889 if (!tr->options) {
5890 pr_warning("Could not create debugfs directory 'options'\n");
5891 return NULL;
5892 }
5893
5894 return tr->options;
5895}
5896
5897static void
5898create_trace_option_file(struct trace_array *tr,
5899 struct trace_option_dentry *topt,
5900 struct tracer_flags *flags,
5901 struct tracer_opt *opt)
5902{
5903 struct dentry *t_options;
5904
5905 t_options = trace_options_init_dentry(tr);
5906 if (!t_options)
5907 return;
5908
5909 topt->flags = flags;
5910 topt->opt = opt;
5911 topt->tr = tr;
5912
5913 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5914 &trace_options_fops);
5915
5916}
5917
5918static struct trace_option_dentry *
5919create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5920{
5921 struct trace_option_dentry *topts;
5922 struct tracer_flags *flags;
5923 struct tracer_opt *opts;
5924 int cnt;
5925
5926 if (!tracer)
5927 return NULL;
5928
5929 flags = tracer->flags;
5930
5931 if (!flags || !flags->opts)
5932 return NULL;
5933
5934 opts = flags->opts;
5935
5936 for (cnt = 0; opts[cnt].name; cnt++)
5937 ;
5938
5939 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5940 if (!topts)
5941 return NULL;
5942
5943 for (cnt = 0; opts[cnt].name; cnt++)
5944 create_trace_option_file(tr, &topts[cnt], flags,
5945 &opts[cnt]);
5946
5947 return topts;
5948}
5949
5950static void
5951destroy_trace_option_files(struct trace_option_dentry *topts)
5952{
5953 int cnt;
5954
5955 if (!topts)
5956 return;
5957
5958 for (cnt = 0; topts[cnt].opt; cnt++) {
5959 if (topts[cnt].entry)
5960 debugfs_remove(topts[cnt].entry);
5961 }
5962
5963 kfree(topts);
5964}
5965
5966static struct dentry *
5967create_trace_option_core_file(struct trace_array *tr,
5968 const char *option, long index)
5969{
5970 struct dentry *t_options;
5971
5972 t_options = trace_options_init_dentry(tr);
5973 if (!t_options)
5974 return NULL;
5975
5976 return trace_create_file(option, 0644, t_options, (void *)index,
5977 &trace_options_core_fops);
5978}
5979
5980static __init void create_trace_options_dir(struct trace_array *tr)
5981{
5982 struct dentry *t_options;
5983 int i;
5984
5985 t_options = trace_options_init_dentry(tr);
5986 if (!t_options)
5987 return;
5988
5989 for (i = 0; trace_options[i]; i++)
5990 create_trace_option_core_file(tr, trace_options[i], i);
5991}
5992
5993static ssize_t
5994rb_simple_read(struct file *filp, char __user *ubuf,
5995 size_t cnt, loff_t *ppos)
5996{
5997 struct trace_array *tr = filp->private_data;
5998 char buf[64];
5999 int r;
6000
6001 r = tracer_tracing_is_on(tr);
6002 r = sprintf(buf, "%d\n", r);
6003
6004 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6005}
6006
6007static ssize_t
6008rb_simple_write(struct file *filp, const char __user *ubuf,
6009 size_t cnt, loff_t *ppos)
6010{
6011 struct trace_array *tr = filp->private_data;
6012 struct ring_buffer *buffer = tr->trace_buffer.buffer;
6013 unsigned long val;
6014 int ret;
6015
6016 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6017 if (ret)
6018 return ret;
6019
6020 if (buffer) {
6021 mutex_lock(&trace_types_lock);
6022 if (val) {
6023 tracer_tracing_on(tr);
6024 if (tr->current_trace->start)
6025 tr->current_trace->start(tr);
6026 } else {
6027 tracer_tracing_off(tr);
6028 if (tr->current_trace->stop)
6029 tr->current_trace->stop(tr);
6030 }
6031 mutex_unlock(&trace_types_lock);
6032 }
6033
6034 (*ppos)++;
6035
6036 return cnt;
6037}
6038
6039static const struct file_operations rb_simple_fops = {
6040 .open = tracing_open_generic_tr,
6041 .read = rb_simple_read,
6042 .write = rb_simple_write,
6043 .release = tracing_release_generic_tr,
6044 .llseek = default_llseek,
6045};
6046
6047struct dentry *trace_instance_dir;
6048
6049static void
6050init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
6051
6052static int
6053allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
6054{
6055 enum ring_buffer_flags rb_flags;
6056
6057 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
6058
6059 buf->tr = tr;
6060
6061 buf->buffer = ring_buffer_alloc(size, rb_flags);
6062 if (!buf->buffer)
6063 return -ENOMEM;
6064
6065 buf->data = alloc_percpu(struct trace_array_cpu);
6066 if (!buf->data) {
6067 ring_buffer_free(buf->buffer);
6068 return -ENOMEM;
6069 }
6070
6071 /* Allocate the first page for all buffers */
6072 set_buffer_entries(&tr->trace_buffer,
6073 ring_buffer_size(tr->trace_buffer.buffer, 0));
6074
6075 return 0;
6076}
6077
6078static int allocate_trace_buffers(struct trace_array *tr, int size)
6079{
6080 int ret;
6081
6082 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
6083 if (ret)
6084 return ret;
6085
6086#ifdef CONFIG_TRACER_MAX_TRACE
6087 ret = allocate_trace_buffer(tr, &tr->max_buffer,
6088 allocate_snapshot ? size : 1);
6089 if (WARN_ON(ret)) {
6090 ring_buffer_free(tr->trace_buffer.buffer);
6091 free_percpu(tr->trace_buffer.data);
6092 return -ENOMEM;
6093 }
6094 tr->allocated_snapshot = allocate_snapshot;
6095
6096 /*
6097 * Only the top level trace array gets its snapshot allocated
6098 * from the kernel command line.
6099 */
6100 allocate_snapshot = false;
6101#endif
6102 return 0;
6103}
6104
6105static int new_instance_create(const char *name)
6106{
6107 struct trace_array *tr;
6108 int ret;
6109
6110 mutex_lock(&trace_types_lock);
6111
6112 ret = -EEXIST;
6113 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6114 if (tr->name && strcmp(tr->name, name) == 0)
6115 goto out_unlock;
6116 }
6117
6118 ret = -ENOMEM;
6119 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
6120 if (!tr)
6121 goto out_unlock;
6122
6123 tr->name = kstrdup(name, GFP_KERNEL);
6124 if (!tr->name)
6125 goto out_free_tr;
6126
6127 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
6128 goto out_free_tr;
6129
6130 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
6131
6132 raw_spin_lock_init(&tr->start_lock);
6133
6134 tr->current_trace = &nop_trace;
6135
6136 INIT_LIST_HEAD(&tr->systems);
6137 INIT_LIST_HEAD(&tr->events);
6138
6139 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
6140 goto out_free_tr;
6141
6142 tr->dir = debugfs_create_dir(name, trace_instance_dir);
6143 if (!tr->dir)
6144 goto out_free_tr;
6145
6146 ret = event_trace_add_tracer(tr->dir, tr);
6147 if (ret) {
6148 debugfs_remove_recursive(tr->dir);
6149 goto out_free_tr;
6150 }
6151
6152 init_tracer_debugfs(tr, tr->dir);
6153
6154 list_add(&tr->list, &ftrace_trace_arrays);
6155
6156 mutex_unlock(&trace_types_lock);
6157
6158 return 0;
6159
6160 out_free_tr:
6161 if (tr->trace_buffer.buffer)
6162 ring_buffer_free(tr->trace_buffer.buffer);
6163 free_cpumask_var(tr->tracing_cpumask);
6164 kfree(tr->name);
6165 kfree(tr);
6166
6167 out_unlock:
6168 mutex_unlock(&trace_types_lock);
6169
6170 return ret;
6171
6172}
6173
6174static int instance_delete(const char *name)
6175{
6176 struct trace_array *tr;
6177 int found = 0;
6178 int ret;
6179
6180 mutex_lock(&trace_types_lock);
6181
6182 ret = -ENODEV;
6183 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6184 if (tr->name && strcmp(tr->name, name) == 0) {
6185 found = 1;
6186 break;
6187 }
6188 }
6189 if (!found)
6190 goto out_unlock;
6191
6192 ret = -EBUSY;
6193 if (tr->ref)
6194 goto out_unlock;
6195
6196 list_del(&tr->list);
6197
6198 tracing_set_nop(tr);
6199 event_trace_del_tracer(tr);
6200 ftrace_destroy_function_files(tr);
6201 debugfs_remove_recursive(tr->dir);
6202 free_percpu(tr->trace_buffer.data);
6203 ring_buffer_free(tr->trace_buffer.buffer);
6204
6205 kfree(tr->name);
6206 kfree(tr);
6207
6208 ret = 0;
6209
6210 out_unlock:
6211 mutex_unlock(&trace_types_lock);
6212
6213 return ret;
6214}
6215
6216static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
6217{
6218 struct dentry *parent;
6219 int ret;
6220
6221 /* Paranoid: Make sure the parent is the "instances" directory */
6222 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6223 if (WARN_ON_ONCE(parent != trace_instance_dir))
6224 return -ENOENT;
6225
6226 /*
6227 * The inode mutex is locked, but debugfs_create_dir() will also
6228 * take the mutex. As the instances directory can not be destroyed
6229 * or changed in any other way, it is safe to unlock it, and
6230 * let the dentry try. If two users try to make the same dir at
6231 * the same time, then the new_instance_create() will determine the
6232 * winner.
6233 */
6234 mutex_unlock(&inode->i_mutex);
6235
6236 ret = new_instance_create(dentry->d_iname);
6237
6238 mutex_lock(&inode->i_mutex);
6239
6240 return ret;
6241}
6242
6243static int instance_rmdir(struct inode *inode, struct dentry *dentry)
6244{
6245 struct dentry *parent;
6246 int ret;
6247
6248 /* Paranoid: Make sure the parent is the "instances" directory */
6249 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6250 if (WARN_ON_ONCE(parent != trace_instance_dir))
6251 return -ENOENT;
6252
6253 /* The caller did a dget() on dentry */
6254 mutex_unlock(&dentry->d_inode->i_mutex);
6255
6256 /*
6257 * The inode mutex is locked, but debugfs_create_dir() will also
6258 * take the mutex. As the instances directory can not be destroyed
6259 * or changed in any other way, it is safe to unlock it, and
6260 * let the dentry try. If two users try to make the same dir at
6261 * the same time, then the instance_delete() will determine the
6262 * winner.
6263 */
6264 mutex_unlock(&inode->i_mutex);
6265
6266 ret = instance_delete(dentry->d_iname);
6267
6268 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
6269 mutex_lock(&dentry->d_inode->i_mutex);
6270
6271 return ret;
6272}
6273
6274static const struct inode_operations instance_dir_inode_operations = {
6275 .lookup = simple_lookup,
6276 .mkdir = instance_mkdir,
6277 .rmdir = instance_rmdir,
6278};
6279
6280static __init void create_trace_instances(struct dentry *d_tracer)
6281{
6282 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
6283 if (WARN_ON(!trace_instance_dir))
6284 return;
6285
6286 /* Hijack the dir inode operations, to allow mkdir */
6287 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
6288}
6289
6290static void
6291init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
6292{
6293 int cpu;
6294
6295 trace_create_file("available_tracers", 0444, d_tracer,
6296 tr, &show_traces_fops);
6297
6298 trace_create_file("current_tracer", 0644, d_tracer,
6299 tr, &set_tracer_fops);
6300
6301 trace_create_file("tracing_cpumask", 0644, d_tracer,
6302 tr, &tracing_cpumask_fops);
6303
6304 trace_create_file("trace_options", 0644, d_tracer,
6305 tr, &tracing_iter_fops);
6306
6307 trace_create_file("trace", 0644, d_tracer,
6308 tr, &tracing_fops);
6309
6310 trace_create_file("trace_pipe", 0444, d_tracer,
6311 tr, &tracing_pipe_fops);
6312
6313 trace_create_file("buffer_size_kb", 0644, d_tracer,
6314 tr, &tracing_entries_fops);
6315
6316 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6317 tr, &tracing_total_entries_fops);
6318
6319 trace_create_file("free_buffer", 0200, d_tracer,
6320 tr, &tracing_free_buffer_fops);
6321
6322 trace_create_file("trace_marker", 0220, d_tracer,
6323 tr, &tracing_mark_fops);
6324
6325 trace_create_file("trace_clock", 0644, d_tracer, tr,
6326 &trace_clock_fops);
6327
6328 trace_create_file("tracing_on", 0644, d_tracer,
6329 tr, &rb_simple_fops);
6330
6331 if (ftrace_create_function_files(tr, d_tracer))
6332 WARN(1, "Could not allocate function filter files");
6333
6334#ifdef CONFIG_TRACER_SNAPSHOT
6335 trace_create_file("snapshot", 0644, d_tracer,
6336 tr, &snapshot_fops);
6337#endif
6338
6339 for_each_tracing_cpu(cpu)
6340 tracing_init_debugfs_percpu(tr, cpu);
6341
6342}
6343
6344static __init int tracer_init_debugfs(void)
6345{
6346 struct dentry *d_tracer;
6347
6348 trace_access_lock_init();
6349
6350 d_tracer = tracing_init_dentry();
6351 if (!d_tracer)
6352 return 0;
6353
6354 init_tracer_debugfs(&global_trace, d_tracer);
6355
6356#ifdef CONFIG_TRACER_MAX_TRACE
6357 trace_create_file("tracing_max_latency", 0644, d_tracer,
6358 &tracing_max_latency, &tracing_max_lat_fops);
6359#endif
6360
6361 trace_create_file("tracing_thresh", 0644, d_tracer,
6362 &tracing_thresh, &tracing_max_lat_fops);
6363
6364 trace_create_file("README", 0444, d_tracer,
6365 NULL, &tracing_readme_fops);
6366
6367 trace_create_file("saved_cmdlines", 0444, d_tracer,
6368 NULL, &tracing_saved_cmdlines_fops);
6369
6370#ifdef CONFIG_DYNAMIC_FTRACE
6371 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6372 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6373#endif
6374
6375 create_trace_instances(d_tracer);
6376
6377 create_trace_options_dir(&global_trace);
6378
6379 return 0;
6380}
6381
6382static int trace_panic_handler(struct notifier_block *this,
6383 unsigned long event, void *unused)
6384{
6385 if (ftrace_dump_on_oops)
6386 ftrace_dump(ftrace_dump_on_oops);
6387 return NOTIFY_OK;
6388}
6389
6390static struct notifier_block trace_panic_notifier = {
6391 .notifier_call = trace_panic_handler,
6392 .next = NULL,
6393 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6394};
6395
6396static int trace_die_handler(struct notifier_block *self,
6397 unsigned long val,
6398 void *data)
6399{
6400 switch (val) {
6401 case DIE_OOPS:
6402 if (ftrace_dump_on_oops)
6403 ftrace_dump(ftrace_dump_on_oops);
6404 break;
6405 default:
6406 break;
6407 }
6408 return NOTIFY_OK;
6409}
6410
6411static struct notifier_block trace_die_notifier = {
6412 .notifier_call = trace_die_handler,
6413 .priority = 200
6414};
6415
6416/*
6417 * printk is set to max of 1024, we really don't need it that big.
6418 * Nothing should be printing 1000 characters anyway.
6419 */
6420#define TRACE_MAX_PRINT 1000
6421
6422/*
6423 * Define here KERN_TRACE so that we have one place to modify
6424 * it if we decide to change what log level the ftrace dump
6425 * should be at.
6426 */
6427#define KERN_TRACE KERN_EMERG
6428
6429void
6430trace_printk_seq(struct trace_seq *s)
6431{
6432 /* Probably should print a warning here. */
6433 if (s->len >= TRACE_MAX_PRINT)
6434 s->len = TRACE_MAX_PRINT;
6435
6436 /* should be zero ended, but we are paranoid. */
6437 s->buffer[s->len] = 0;
6438
6439 printk(KERN_TRACE "%s", s->buffer);
6440
6441 trace_seq_init(s);
6442}
6443
6444void trace_init_global_iter(struct trace_iterator *iter)
6445{
6446 iter->tr = &global_trace;
6447 iter->trace = iter->tr->current_trace;
6448 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6449 iter->trace_buffer = &global_trace.trace_buffer;
6450
6451 if (iter->trace && iter->trace->open)
6452 iter->trace->open(iter);
6453
6454 /* Annotate start of buffers if we had overruns */
6455 if (ring_buffer_overruns(iter->trace_buffer->buffer))
6456 iter->iter_flags |= TRACE_FILE_ANNOTATE;
6457
6458 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6459 if (trace_clocks[iter->tr->clock_id].in_ns)
6460 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6461}
6462
6463void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6464{
6465 /* use static because iter can be a bit big for the stack */
6466 static struct trace_iterator iter;
6467 static atomic_t dump_running;
6468 unsigned int old_userobj;
6469 unsigned long flags;
6470 int cnt = 0, cpu;
6471
6472 /* Only allow one dump user at a time. */
6473 if (atomic_inc_return(&dump_running) != 1) {
6474 atomic_dec(&dump_running);
6475 return;
6476 }
6477
6478 /*
6479 * Always turn off tracing when we dump.
6480 * We don't need to show trace output of what happens
6481 * between multiple crashes.
6482 *
6483 * If the user does a sysrq-z, then they can re-enable
6484 * tracing with echo 1 > tracing_on.
6485 */
6486 tracing_off();
6487
6488 local_irq_save(flags);
6489
6490 /* Simulate the iterator */
6491 trace_init_global_iter(&iter);
6492
6493 for_each_tracing_cpu(cpu) {
6494 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6495 }
6496
6497 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6498
6499 /* don't look at user memory in panic mode */
6500 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6501
6502 switch (oops_dump_mode) {
6503 case DUMP_ALL:
6504 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6505 break;
6506 case DUMP_ORIG:
6507 iter.cpu_file = raw_smp_processor_id();
6508 break;
6509 case DUMP_NONE:
6510 goto out_enable;
6511 default:
6512 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6513 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6514 }
6515
6516 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6517
6518 /* Did function tracer already get disabled? */
6519 if (ftrace_is_dead()) {
6520 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6521 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6522 }
6523
6524 /*
6525 * We need to stop all tracing on all CPUS to read the
6526 * the next buffer. This is a bit expensive, but is
6527 * not done often. We fill all what we can read,
6528 * and then release the locks again.
6529 */
6530
6531 while (!trace_empty(&iter)) {
6532
6533 if (!cnt)
6534 printk(KERN_TRACE "---------------------------------\n");
6535
6536 cnt++;
6537
6538 /* reset all but tr, trace, and overruns */
6539 memset(&iter.seq, 0,
6540 sizeof(struct trace_iterator) -
6541 offsetof(struct trace_iterator, seq));
6542 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6543 iter.pos = -1;
6544
6545 if (trace_find_next_entry_inc(&iter) != NULL) {
6546 int ret;
6547
6548 ret = print_trace_line(&iter);
6549 if (ret != TRACE_TYPE_NO_CONSUME)
6550 trace_consume(&iter);
6551 }
6552 touch_nmi_watchdog();
6553
6554 trace_printk_seq(&iter.seq);
6555 }
6556
6557 if (!cnt)
6558 printk(KERN_TRACE " (ftrace buffer empty)\n");
6559 else
6560 printk(KERN_TRACE "---------------------------------\n");
6561
6562 out_enable:
6563 trace_flags |= old_userobj;
6564
6565 for_each_tracing_cpu(cpu) {
6566 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6567 }
6568 atomic_dec(&dump_running);
6569 local_irq_restore(flags);
6570}
6571EXPORT_SYMBOL_GPL(ftrace_dump);
6572
6573__init static int tracer_alloc_buffers(void)
6574{
6575 int ring_buf_size;
6576 int ret = -ENOMEM;
6577
6578
6579 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6580 goto out;
6581
6582 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
6583 goto out_free_buffer_mask;
6584
6585 /* Only allocate trace_printk buffers if a trace_printk exists */
6586 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6587 /* Must be called before global_trace.buffer is allocated */
6588 trace_printk_init_buffers();
6589
6590 /* To save memory, keep the ring buffer size to its minimum */
6591 if (ring_buffer_expanded)
6592 ring_buf_size = trace_buf_size;
6593 else
6594 ring_buf_size = 1;
6595
6596 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6597 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
6598
6599 raw_spin_lock_init(&global_trace.start_lock);
6600
6601 /* Used for event triggers */
6602 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
6603 if (!temp_buffer)
6604 goto out_free_cpumask;
6605
6606 /* TODO: make the number of buffers hot pluggable with CPUS */
6607 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6608 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6609 WARN_ON(1);
6610 goto out_free_temp_buffer;
6611 }
6612
6613 if (global_trace.buffer_disabled)
6614 tracing_off();
6615
6616 trace_init_cmdlines();
6617
6618 if (trace_boot_clock) {
6619 ret = tracing_set_clock(&global_trace, trace_boot_clock);
6620 if (ret < 0)
6621 pr_warning("Trace clock %s not defined, going back to default\n",
6622 trace_boot_clock);
6623 }
6624
6625 /*
6626 * register_tracer() might reference current_trace, so it
6627 * needs to be set before we register anything. This is
6628 * just a bootstrap of current_trace anyway.
6629 */
6630 global_trace.current_trace = &nop_trace;
6631
6632 register_tracer(&nop_trace);
6633
6634 /* All seems OK, enable tracing */
6635 tracing_disabled = 0;
6636
6637 atomic_notifier_chain_register(&panic_notifier_list,
6638 &trace_panic_notifier);
6639
6640 register_die_notifier(&trace_die_notifier);
6641
6642 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6643
6644 INIT_LIST_HEAD(&global_trace.systems);
6645 INIT_LIST_HEAD(&global_trace.events);
6646 list_add(&global_trace.list, &ftrace_trace_arrays);
6647
6648 while (trace_boot_options) {
6649 char *option;
6650
6651 option = strsep(&trace_boot_options, ",");
6652 trace_set_options(&global_trace, option);
6653 }
6654
6655 register_snapshot_cmd();
6656
6657 return 0;
6658
6659out_free_temp_buffer:
6660 ring_buffer_free(temp_buffer);
6661out_free_cpumask:
6662 free_percpu(global_trace.trace_buffer.data);
6663#ifdef CONFIG_TRACER_MAX_TRACE
6664 free_percpu(global_trace.max_buffer.data);
6665#endif
6666 free_cpumask_var(global_trace.tracing_cpumask);
6667out_free_buffer_mask:
6668 free_cpumask_var(tracing_buffer_mask);
6669out:
6670 return ret;
6671}
6672
6673__init static int clear_boot_tracer(void)
6674{
6675 /*
6676 * The default tracer at boot buffer is an init section.
6677 * This function is called in lateinit. If we did not
6678 * find the boot tracer, then clear it out, to prevent
6679 * later registration from accessing the buffer that is
6680 * about to be freed.
6681 */
6682 if (!default_bootup_tracer)
6683 return 0;
6684
6685 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6686 default_bootup_tracer);
6687 default_bootup_tracer = NULL;
6688
6689 return 0;
6690}
6691
6692early_initcall(tracer_alloc_buffers);
6693fs_initcall(tracer_init_debugfs);
6694late_initcall(clear_boot_tracer);