1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * ring buffer based function tracer
4 *
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 *
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 *
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
14 */
15 #include <linux/ring_buffer.h>
16 #include <linux/utsname.h>
17 #include <linux/stacktrace.h>
18 #include <linux/writeback.h>
19 #include <linux/kallsyms.h>
20 #include <linux/security.h>
21 #include <linux/seq_file.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/tracefs.h>
25 #include <linux/pagemap.h>
26 #include <linux/hardirq.h>
27 #include <linux/linkage.h>
28 #include <linux/uaccess.h>
29 #include <linux/vmalloc.h>
30 #include <linux/ftrace.h>
31 #include <linux/module.h>
32 #include <linux/percpu.h>
33 #include <linux/splice.h>
34 #include <linux/kdebug.h>
35 #include <linux/string.h>
36 #include <linux/mount.h>
37 #include <linux/rwsem.h>
38 #include <linux/slab.h>
39 #include <linux/ctype.h>
40 #include <linux/init.h>
41 #include <linux/panic_notifier.h>
42 #include <linux/poll.h>
43 #include <linux/nmi.h>
44 #include <linux/fs.h>
45 #include <linux/trace.h>
46 #include <linux/sched/clock.h>
47 #include <linux/sched/rt.h>
48 #include <linux/fsnotify.h>
49 #include <linux/irq_work.h>
50 #include <linux/workqueue.h>
51
52 #include <asm/setup.h> /* COMMAND_LINE_SIZE */
53
54 #include "trace.h"
55 #include "trace_output.h"
56
57 #ifdef CONFIG_FTRACE_STARTUP_TEST
58 /*
59 * We need to change this state when a selftest is running.
60 * A selftest will lurk into the ring-buffer to count the
61 * entries inserted during the selftest although some concurrent
62 * insertions into the ring-buffer such as trace_printk could occurred
63 * at the same time, giving false positive or negative results.
64 */
65 static bool __read_mostly tracing_selftest_running;
66
67 /*
68 * If boot-time tracing including tracers/events via kernel cmdline
69 * is running, we do not want to run SELFTEST.
70 */
71 bool __read_mostly tracing_selftest_disabled;
72
disable_tracing_selftest(const char * reason)73 void __init disable_tracing_selftest(const char *reason)
74 {
75 if (!tracing_selftest_disabled) {
76 tracing_selftest_disabled = true;
77 pr_info("Ftrace startup test is disabled due to %s\n", reason);
78 }
79 }
80 #else
81 #define tracing_selftest_running 0
82 #define tracing_selftest_disabled 0
83 #endif
84
85 /* Pipe tracepoints to printk */
86 static struct trace_iterator *tracepoint_print_iter;
87 int tracepoint_printk;
88 static bool tracepoint_printk_stop_on_boot __initdata;
89 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
90
91 /* For tracers that don't implement custom flags */
92 static struct tracer_opt dummy_tracer_opt[] = {
93 { }
94 };
95
96 static int
dummy_set_flag(struct trace_array * tr,u32 old_flags,u32 bit,int set)97 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
98 {
99 return 0;
100 }
101
102 /*
103 * To prevent the comm cache from being overwritten when no
104 * tracing is active, only save the comm when a trace event
105 * occurred.
106 */
107 DEFINE_PER_CPU(bool, trace_taskinfo_save);
108
109 /*
110 * Kill all tracing for good (never come back).
111 * It is initialized to 1 but will turn to zero if the initialization
112 * of the tracer is successful. But that is the only place that sets
113 * this back to zero.
114 */
115 static int tracing_disabled = 1;
116
117 cpumask_var_t __read_mostly tracing_buffer_mask;
118
119 /*
120 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
121 *
122 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
123 * is set, then ftrace_dump is called. This will output the contents
124 * of the ftrace buffers to the console. This is very useful for
125 * capturing traces that lead to crashes and outputing it to a
126 * serial console.
127 *
128 * It is default off, but you can enable it with either specifying
129 * "ftrace_dump_on_oops" in the kernel command line, or setting
130 * /proc/sys/kernel/ftrace_dump_on_oops
131 * Set 1 if you want to dump buffers of all CPUs
132 * Set 2 if you want to dump the buffer of the CPU that triggered oops
133 * Set instance name if you want to dump the specific trace instance
134 * Multiple instance dump is also supported, and instances are seperated
135 * by commas.
136 */
137 /* Set to string format zero to disable by default */
138 char ftrace_dump_on_oops[MAX_TRACER_SIZE] = "0";
139
140 /* When set, tracing will stop when a WARN*() is hit */
141 int __disable_trace_on_warning;
142
143 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
144 /* Map of enums to their values, for "eval_map" file */
145 struct trace_eval_map_head {
146 struct module *mod;
147 unsigned long length;
148 };
149
150 union trace_eval_map_item;
151
152 struct trace_eval_map_tail {
153 /*
154 * "end" is first and points to NULL as it must be different
155 * than "mod" or "eval_string"
156 */
157 union trace_eval_map_item *next;
158 const char *end; /* points to NULL */
159 };
160
161 static DEFINE_MUTEX(trace_eval_mutex);
162
163 /*
164 * The trace_eval_maps are saved in an array with two extra elements,
165 * one at the beginning, and one at the end. The beginning item contains
166 * the count of the saved maps (head.length), and the module they
167 * belong to if not built in (head.mod). The ending item contains a
168 * pointer to the next array of saved eval_map items.
169 */
170 union trace_eval_map_item {
171 struct trace_eval_map map;
172 struct trace_eval_map_head head;
173 struct trace_eval_map_tail tail;
174 };
175
176 static union trace_eval_map_item *trace_eval_maps;
177 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
178
179 int tracing_set_tracer(struct trace_array *tr, const char *buf);
180 static void ftrace_trace_userstack(struct trace_array *tr,
181 struct trace_buffer *buffer,
182 unsigned int trace_ctx);
183
184 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
185 static char *default_bootup_tracer;
186
187 static bool allocate_snapshot;
188 static bool snapshot_at_boot;
189
190 static char boot_instance_info[COMMAND_LINE_SIZE] __initdata;
191 static int boot_instance_index;
192
193 static char boot_snapshot_info[COMMAND_LINE_SIZE] __initdata;
194 static int boot_snapshot_index;
195
set_cmdline_ftrace(char * str)196 static int __init set_cmdline_ftrace(char *str)
197 {
198 strscpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
199 default_bootup_tracer = bootup_tracer_buf;
200 /* We are using ftrace early, expand it */
201 trace_set_ring_buffer_expanded(NULL);
202 return 1;
203 }
204 __setup("ftrace=", set_cmdline_ftrace);
205
ftrace_dump_on_oops_enabled(void)206 int ftrace_dump_on_oops_enabled(void)
207 {
208 if (!strcmp("0", ftrace_dump_on_oops))
209 return 0;
210 else
211 return 1;
212 }
213
set_ftrace_dump_on_oops(char * str)214 static int __init set_ftrace_dump_on_oops(char *str)
215 {
216 if (!*str) {
217 strscpy(ftrace_dump_on_oops, "1", MAX_TRACER_SIZE);
218 return 1;
219 }
220
221 if (*str == ',') {
222 strscpy(ftrace_dump_on_oops, "1", MAX_TRACER_SIZE);
223 strscpy(ftrace_dump_on_oops + 1, str, MAX_TRACER_SIZE - 1);
224 return 1;
225 }
226
227 if (*str++ == '=') {
228 strscpy(ftrace_dump_on_oops, str, MAX_TRACER_SIZE);
229 return 1;
230 }
231
232 return 0;
233 }
234 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
235
stop_trace_on_warning(char * str)236 static int __init stop_trace_on_warning(char *str)
237 {
238 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
239 __disable_trace_on_warning = 1;
240 return 1;
241 }
242 __setup("traceoff_on_warning", stop_trace_on_warning);
243
boot_alloc_snapshot(char * str)244 static int __init boot_alloc_snapshot(char *str)
245 {
246 char *slot = boot_snapshot_info + boot_snapshot_index;
247 int left = sizeof(boot_snapshot_info) - boot_snapshot_index;
248 int ret;
249
250 if (str[0] == '=') {
251 str++;
252 if (strlen(str) >= left)
253 return -1;
254
255 ret = snprintf(slot, left, "%s\t", str);
256 boot_snapshot_index += ret;
257 } else {
258 allocate_snapshot = true;
259 /* We also need the main ring buffer expanded */
260 trace_set_ring_buffer_expanded(NULL);
261 }
262 return 1;
263 }
264 __setup("alloc_snapshot", boot_alloc_snapshot);
265
266
boot_snapshot(char * str)267 static int __init boot_snapshot(char *str)
268 {
269 snapshot_at_boot = true;
270 boot_alloc_snapshot(str);
271 return 1;
272 }
273 __setup("ftrace_boot_snapshot", boot_snapshot);
274
275
boot_instance(char * str)276 static int __init boot_instance(char *str)
277 {
278 char *slot = boot_instance_info + boot_instance_index;
279 int left = sizeof(boot_instance_info) - boot_instance_index;
280 int ret;
281
282 if (strlen(str) >= left)
283 return -1;
284
285 ret = snprintf(slot, left, "%s\t", str);
286 boot_instance_index += ret;
287
288 return 1;
289 }
290 __setup("trace_instance=", boot_instance);
291
292
293 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
294
set_trace_boot_options(char * str)295 static int __init set_trace_boot_options(char *str)
296 {
297 strscpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
298 return 1;
299 }
300 __setup("trace_options=", set_trace_boot_options);
301
302 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
303 static char *trace_boot_clock __initdata;
304
set_trace_boot_clock(char * str)305 static int __init set_trace_boot_clock(char *str)
306 {
307 strscpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
308 trace_boot_clock = trace_boot_clock_buf;
309 return 1;
310 }
311 __setup("trace_clock=", set_trace_boot_clock);
312
set_tracepoint_printk(char * str)313 static int __init set_tracepoint_printk(char *str)
314 {
315 /* Ignore the "tp_printk_stop_on_boot" param */
316 if (*str == '_')
317 return 0;
318
319 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
320 tracepoint_printk = 1;
321 return 1;
322 }
323 __setup("tp_printk", set_tracepoint_printk);
324
set_tracepoint_printk_stop(char * str)325 static int __init set_tracepoint_printk_stop(char *str)
326 {
327 tracepoint_printk_stop_on_boot = true;
328 return 1;
329 }
330 __setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
331
ns2usecs(u64 nsec)332 unsigned long long ns2usecs(u64 nsec)
333 {
334 nsec += 500;
335 do_div(nsec, 1000);
336 return nsec;
337 }
338
339 static void
trace_process_export(struct trace_export * export,struct ring_buffer_event * event,int flag)340 trace_process_export(struct trace_export *export,
341 struct ring_buffer_event *event, int flag)
342 {
343 struct trace_entry *entry;
344 unsigned int size = 0;
345
346 if (export->flags & flag) {
347 entry = ring_buffer_event_data(event);
348 size = ring_buffer_event_length(event);
349 export->write(export, entry, size);
350 }
351 }
352
353 static DEFINE_MUTEX(ftrace_export_lock);
354
355 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
356
357 static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
358 static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
359 static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
360
ftrace_exports_enable(struct trace_export * export)361 static inline void ftrace_exports_enable(struct trace_export *export)
362 {
363 if (export->flags & TRACE_EXPORT_FUNCTION)
364 static_branch_inc(&trace_function_exports_enabled);
365
366 if (export->flags & TRACE_EXPORT_EVENT)
367 static_branch_inc(&trace_event_exports_enabled);
368
369 if (export->flags & TRACE_EXPORT_MARKER)
370 static_branch_inc(&trace_marker_exports_enabled);
371 }
372
ftrace_exports_disable(struct trace_export * export)373 static inline void ftrace_exports_disable(struct trace_export *export)
374 {
375 if (export->flags & TRACE_EXPORT_FUNCTION)
376 static_branch_dec(&trace_function_exports_enabled);
377
378 if (export->flags & TRACE_EXPORT_EVENT)
379 static_branch_dec(&trace_event_exports_enabled);
380
381 if (export->flags & TRACE_EXPORT_MARKER)
382 static_branch_dec(&trace_marker_exports_enabled);
383 }
384
ftrace_exports(struct ring_buffer_event * event,int flag)385 static void ftrace_exports(struct ring_buffer_event *event, int flag)
386 {
387 struct trace_export *export;
388
389 preempt_disable_notrace();
390
391 export = rcu_dereference_raw_check(ftrace_exports_list);
392 while (export) {
393 trace_process_export(export, event, flag);
394 export = rcu_dereference_raw_check(export->next);
395 }
396
397 preempt_enable_notrace();
398 }
399
400 static inline void
add_trace_export(struct trace_export ** list,struct trace_export * export)401 add_trace_export(struct trace_export **list, struct trace_export *export)
402 {
403 rcu_assign_pointer(export->next, *list);
404 /*
405 * We are entering export into the list but another
406 * CPU might be walking that list. We need to make sure
407 * the export->next pointer is valid before another CPU sees
408 * the export pointer included into the list.
409 */
410 rcu_assign_pointer(*list, export);
411 }
412
413 static inline int
rm_trace_export(struct trace_export ** list,struct trace_export * export)414 rm_trace_export(struct trace_export **list, struct trace_export *export)
415 {
416 struct trace_export **p;
417
418 for (p = list; *p != NULL; p = &(*p)->next)
419 if (*p == export)
420 break;
421
422 if (*p != export)
423 return -1;
424
425 rcu_assign_pointer(*p, (*p)->next);
426
427 return 0;
428 }
429
430 static inline void
add_ftrace_export(struct trace_export ** list,struct trace_export * export)431 add_ftrace_export(struct trace_export **list, struct trace_export *export)
432 {
433 ftrace_exports_enable(export);
434
435 add_trace_export(list, export);
436 }
437
438 static inline int
rm_ftrace_export(struct trace_export ** list,struct trace_export * export)439 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
440 {
441 int ret;
442
443 ret = rm_trace_export(list, export);
444 ftrace_exports_disable(export);
445
446 return ret;
447 }
448
register_ftrace_export(struct trace_export * export)449 int register_ftrace_export(struct trace_export *export)
450 {
451 if (WARN_ON_ONCE(!export->write))
452 return -1;
453
454 mutex_lock(&ftrace_export_lock);
455
456 add_ftrace_export(&ftrace_exports_list, export);
457
458 mutex_unlock(&ftrace_export_lock);
459
460 return 0;
461 }
462 EXPORT_SYMBOL_GPL(register_ftrace_export);
463
unregister_ftrace_export(struct trace_export * export)464 int unregister_ftrace_export(struct trace_export *export)
465 {
466 int ret;
467
468 mutex_lock(&ftrace_export_lock);
469
470 ret = rm_ftrace_export(&ftrace_exports_list, export);
471
472 mutex_unlock(&ftrace_export_lock);
473
474 return ret;
475 }
476 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
477
478 /* trace_flags holds trace_options default values */
479 #define TRACE_DEFAULT_FLAGS \
480 (FUNCTION_DEFAULT_FLAGS | \
481 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
482 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
483 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
484 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
485 TRACE_ITER_HASH_PTR)
486
487 /* trace_options that are only supported by global_trace */
488 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
489 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
490
491 /* trace_flags that are default zero for instances */
492 #define ZEROED_TRACE_FLAGS \
493 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
494
495 /*
496 * The global_trace is the descriptor that holds the top-level tracing
497 * buffers for the live tracing.
498 */
499 static struct trace_array global_trace = {
500 .trace_flags = TRACE_DEFAULT_FLAGS,
501 };
502
trace_set_ring_buffer_expanded(struct trace_array * tr)503 void trace_set_ring_buffer_expanded(struct trace_array *tr)
504 {
505 if (!tr)
506 tr = &global_trace;
507 tr->ring_buffer_expanded = true;
508 }
509
510 LIST_HEAD(ftrace_trace_arrays);
511
trace_array_get(struct trace_array * this_tr)512 int trace_array_get(struct trace_array *this_tr)
513 {
514 struct trace_array *tr;
515 int ret = -ENODEV;
516
517 mutex_lock(&trace_types_lock);
518 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
519 if (tr == this_tr) {
520 tr->ref++;
521 ret = 0;
522 break;
523 }
524 }
525 mutex_unlock(&trace_types_lock);
526
527 return ret;
528 }
529
__trace_array_put(struct trace_array * this_tr)530 static void __trace_array_put(struct trace_array *this_tr)
531 {
532 WARN_ON(!this_tr->ref);
533 this_tr->ref--;
534 }
535
536 /**
537 * trace_array_put - Decrement the reference counter for this trace array.
538 * @this_tr : pointer to the trace array
539 *
540 * NOTE: Use this when we no longer need the trace array returned by
541 * trace_array_get_by_name(). This ensures the trace array can be later
542 * destroyed.
543 *
544 */
trace_array_put(struct trace_array * this_tr)545 void trace_array_put(struct trace_array *this_tr)
546 {
547 if (!this_tr)
548 return;
549
550 mutex_lock(&trace_types_lock);
551 __trace_array_put(this_tr);
552 mutex_unlock(&trace_types_lock);
553 }
554 EXPORT_SYMBOL_GPL(trace_array_put);
555
tracing_check_open_get_tr(struct trace_array * tr)556 int tracing_check_open_get_tr(struct trace_array *tr)
557 {
558 int ret;
559
560 ret = security_locked_down(LOCKDOWN_TRACEFS);
561 if (ret)
562 return ret;
563
564 if (tracing_disabled)
565 return -ENODEV;
566
567 if (tr && trace_array_get(tr) < 0)
568 return -ENODEV;
569
570 return 0;
571 }
572
call_filter_check_discard(struct trace_event_call * call,void * rec,struct trace_buffer * buffer,struct ring_buffer_event * event)573 int call_filter_check_discard(struct trace_event_call *call, void *rec,
574 struct trace_buffer *buffer,
575 struct ring_buffer_event *event)
576 {
577 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
578 !filter_match_preds(call->filter, rec)) {
579 __trace_event_discard_commit(buffer, event);
580 return 1;
581 }
582
583 return 0;
584 }
585
586 /**
587 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
588 * @filtered_pids: The list of pids to check
589 * @search_pid: The PID to find in @filtered_pids
590 *
591 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
592 */
593 bool
trace_find_filtered_pid(struct trace_pid_list * filtered_pids,pid_t search_pid)594 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
595 {
596 return trace_pid_list_is_set(filtered_pids, search_pid);
597 }
598
599 /**
600 * trace_ignore_this_task - should a task be ignored for tracing
601 * @filtered_pids: The list of pids to check
602 * @filtered_no_pids: The list of pids not to be traced
603 * @task: The task that should be ignored if not filtered
604 *
605 * Checks if @task should be traced or not from @filtered_pids.
606 * Returns true if @task should *NOT* be traced.
607 * Returns false if @task should be traced.
608 */
609 bool
trace_ignore_this_task(struct trace_pid_list * filtered_pids,struct trace_pid_list * filtered_no_pids,struct task_struct * task)610 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
611 struct trace_pid_list *filtered_no_pids,
612 struct task_struct *task)
613 {
614 /*
615 * If filtered_no_pids is not empty, and the task's pid is listed
616 * in filtered_no_pids, then return true.
617 * Otherwise, if filtered_pids is empty, that means we can
618 * trace all tasks. If it has content, then only trace pids
619 * within filtered_pids.
620 */
621
622 return (filtered_pids &&
623 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
624 (filtered_no_pids &&
625 trace_find_filtered_pid(filtered_no_pids, task->pid));
626 }
627
628 /**
629 * trace_filter_add_remove_task - Add or remove a task from a pid_list
630 * @pid_list: The list to modify
631 * @self: The current task for fork or NULL for exit
632 * @task: The task to add or remove
633 *
634 * If adding a task, if @self is defined, the task is only added if @self
635 * is also included in @pid_list. This happens on fork and tasks should
636 * only be added when the parent is listed. If @self is NULL, then the
637 * @task pid will be removed from the list, which would happen on exit
638 * of a task.
639 */
trace_filter_add_remove_task(struct trace_pid_list * pid_list,struct task_struct * self,struct task_struct * task)640 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
641 struct task_struct *self,
642 struct task_struct *task)
643 {
644 if (!pid_list)
645 return;
646
647 /* For forks, we only add if the forking task is listed */
648 if (self) {
649 if (!trace_find_filtered_pid(pid_list, self->pid))
650 return;
651 }
652
653 /* "self" is set for forks, and NULL for exits */
654 if (self)
655 trace_pid_list_set(pid_list, task->pid);
656 else
657 trace_pid_list_clear(pid_list, task->pid);
658 }
659
660 /**
661 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
662 * @pid_list: The pid list to show
663 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
664 * @pos: The position of the file
665 *
666 * This is used by the seq_file "next" operation to iterate the pids
667 * listed in a trace_pid_list structure.
668 *
669 * Returns the pid+1 as we want to display pid of zero, but NULL would
670 * stop the iteration.
671 */
trace_pid_next(struct trace_pid_list * pid_list,void * v,loff_t * pos)672 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
673 {
674 long pid = (unsigned long)v;
675 unsigned int next;
676
677 (*pos)++;
678
679 /* pid already is +1 of the actual previous bit */
680 if (trace_pid_list_next(pid_list, pid, &next) < 0)
681 return NULL;
682
683 pid = next;
684
685 /* Return pid + 1 to allow zero to be represented */
686 return (void *)(pid + 1);
687 }
688
689 /**
690 * trace_pid_start - Used for seq_file to start reading pid lists
691 * @pid_list: The pid list to show
692 * @pos: The position of the file
693 *
694 * This is used by seq_file "start" operation to start the iteration
695 * of listing pids.
696 *
697 * Returns the pid+1 as we want to display pid of zero, but NULL would
698 * stop the iteration.
699 */
trace_pid_start(struct trace_pid_list * pid_list,loff_t * pos)700 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
701 {
702 unsigned long pid;
703 unsigned int first;
704 loff_t l = 0;
705
706 if (trace_pid_list_first(pid_list, &first) < 0)
707 return NULL;
708
709 pid = first;
710
711 /* Return pid + 1 so that zero can be the exit value */
712 for (pid++; pid && l < *pos;
713 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
714 ;
715 return (void *)pid;
716 }
717
718 /**
719 * trace_pid_show - show the current pid in seq_file processing
720 * @m: The seq_file structure to write into
721 * @v: A void pointer of the pid (+1) value to display
722 *
723 * Can be directly used by seq_file operations to display the current
724 * pid value.
725 */
trace_pid_show(struct seq_file * m,void * v)726 int trace_pid_show(struct seq_file *m, void *v)
727 {
728 unsigned long pid = (unsigned long)v - 1;
729
730 seq_printf(m, "%lu\n", pid);
731 return 0;
732 }
733
734 /* 128 should be much more than enough */
735 #define PID_BUF_SIZE 127
736
trace_pid_write(struct trace_pid_list * filtered_pids,struct trace_pid_list ** new_pid_list,const char __user * ubuf,size_t cnt)737 int trace_pid_write(struct trace_pid_list *filtered_pids,
738 struct trace_pid_list **new_pid_list,
739 const char __user *ubuf, size_t cnt)
740 {
741 struct trace_pid_list *pid_list;
742 struct trace_parser parser;
743 unsigned long val;
744 int nr_pids = 0;
745 ssize_t read = 0;
746 ssize_t ret;
747 loff_t pos;
748 pid_t pid;
749
750 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
751 return -ENOMEM;
752
753 /*
754 * Always recreate a new array. The write is an all or nothing
755 * operation. Always create a new array when adding new pids by
756 * the user. If the operation fails, then the current list is
757 * not modified.
758 */
759 pid_list = trace_pid_list_alloc();
760 if (!pid_list) {
761 trace_parser_put(&parser);
762 return -ENOMEM;
763 }
764
765 if (filtered_pids) {
766 /* copy the current bits to the new max */
767 ret = trace_pid_list_first(filtered_pids, &pid);
768 while (!ret) {
769 trace_pid_list_set(pid_list, pid);
770 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
771 nr_pids++;
772 }
773 }
774
775 ret = 0;
776 while (cnt > 0) {
777
778 pos = 0;
779
780 ret = trace_get_user(&parser, ubuf, cnt, &pos);
781 if (ret < 0)
782 break;
783
784 read += ret;
785 ubuf += ret;
786 cnt -= ret;
787
788 if (!trace_parser_loaded(&parser))
789 break;
790
791 ret = -EINVAL;
792 if (kstrtoul(parser.buffer, 0, &val))
793 break;
794
795 pid = (pid_t)val;
796
797 if (trace_pid_list_set(pid_list, pid) < 0) {
798 ret = -1;
799 break;
800 }
801 nr_pids++;
802
803 trace_parser_clear(&parser);
804 ret = 0;
805 }
806 trace_parser_put(&parser);
807
808 if (ret < 0) {
809 trace_pid_list_free(pid_list);
810 return ret;
811 }
812
813 if (!nr_pids) {
814 /* Cleared the list of pids */
815 trace_pid_list_free(pid_list);
816 pid_list = NULL;
817 }
818
819 *new_pid_list = pid_list;
820
821 return read;
822 }
823
buffer_ftrace_now(struct array_buffer * buf,int cpu)824 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
825 {
826 u64 ts;
827
828 /* Early boot up does not have a buffer yet */
829 if (!buf->buffer)
830 return trace_clock_local();
831
832 ts = ring_buffer_time_stamp(buf->buffer);
833 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
834
835 return ts;
836 }
837
ftrace_now(int cpu)838 u64 ftrace_now(int cpu)
839 {
840 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
841 }
842
843 /**
844 * tracing_is_enabled - Show if global_trace has been enabled
845 *
846 * Shows if the global trace has been enabled or not. It uses the
847 * mirror flag "buffer_disabled" to be used in fast paths such as for
848 * the irqsoff tracer. But it may be inaccurate due to races. If you
849 * need to know the accurate state, use tracing_is_on() which is a little
850 * slower, but accurate.
851 */
tracing_is_enabled(void)852 int tracing_is_enabled(void)
853 {
854 /*
855 * For quick access (irqsoff uses this in fast path), just
856 * return the mirror variable of the state of the ring buffer.
857 * It's a little racy, but we don't really care.
858 */
859 smp_rmb();
860 return !global_trace.buffer_disabled;
861 }
862
863 /*
864 * trace_buf_size is the size in bytes that is allocated
865 * for a buffer. Note, the number of bytes is always rounded
866 * to page size.
867 *
868 * This number is purposely set to a low number of 16384.
869 * If the dump on oops happens, it will be much appreciated
870 * to not have to wait for all that output. Anyway this can be
871 * boot time and run time configurable.
872 */
873 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
874
875 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
876
877 /* trace_types holds a link list of available tracers. */
878 static struct tracer *trace_types __read_mostly;
879
880 /*
881 * trace_types_lock is used to protect the trace_types list.
882 */
883 DEFINE_MUTEX(trace_types_lock);
884
885 /*
886 * serialize the access of the ring buffer
887 *
888 * ring buffer serializes readers, but it is low level protection.
889 * The validity of the events (which returns by ring_buffer_peek() ..etc)
890 * are not protected by ring buffer.
891 *
892 * The content of events may become garbage if we allow other process consumes
893 * these events concurrently:
894 * A) the page of the consumed events may become a normal page
895 * (not reader page) in ring buffer, and this page will be rewritten
896 * by events producer.
897 * B) The page of the consumed events may become a page for splice_read,
898 * and this page will be returned to system.
899 *
900 * These primitives allow multi process access to different cpu ring buffer
901 * concurrently.
902 *
903 * These primitives don't distinguish read-only and read-consume access.
904 * Multi read-only access are also serialized.
905 */
906
907 #ifdef CONFIG_SMP
908 static DECLARE_RWSEM(all_cpu_access_lock);
909 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
910
trace_access_lock(int cpu)911 static inline void trace_access_lock(int cpu)
912 {
913 if (cpu == RING_BUFFER_ALL_CPUS) {
914 /* gain it for accessing the whole ring buffer. */
915 down_write(&all_cpu_access_lock);
916 } else {
917 /* gain it for accessing a cpu ring buffer. */
918
919 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
920 down_read(&all_cpu_access_lock);
921
922 /* Secondly block other access to this @cpu ring buffer. */
923 mutex_lock(&per_cpu(cpu_access_lock, cpu));
924 }
925 }
926
trace_access_unlock(int cpu)927 static inline void trace_access_unlock(int cpu)
928 {
929 if (cpu == RING_BUFFER_ALL_CPUS) {
930 up_write(&all_cpu_access_lock);
931 } else {
932 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
933 up_read(&all_cpu_access_lock);
934 }
935 }
936
trace_access_lock_init(void)937 static inline void trace_access_lock_init(void)
938 {
939 int cpu;
940
941 for_each_possible_cpu(cpu)
942 mutex_init(&per_cpu(cpu_access_lock, cpu));
943 }
944
945 #else
946
947 static DEFINE_MUTEX(access_lock);
948
trace_access_lock(int cpu)949 static inline void trace_access_lock(int cpu)
950 {
951 (void)cpu;
952 mutex_lock(&access_lock);
953 }
954
trace_access_unlock(int cpu)955 static inline void trace_access_unlock(int cpu)
956 {
957 (void)cpu;
958 mutex_unlock(&access_lock);
959 }
960
trace_access_lock_init(void)961 static inline void trace_access_lock_init(void)
962 {
963 }
964
965 #endif
966
967 #ifdef CONFIG_STACKTRACE
968 static void __ftrace_trace_stack(struct trace_buffer *buffer,
969 unsigned int trace_ctx,
970 int skip, struct pt_regs *regs);
971 static inline void ftrace_trace_stack(struct trace_array *tr,
972 struct trace_buffer *buffer,
973 unsigned int trace_ctx,
974 int skip, struct pt_regs *regs);
975
976 #else
__ftrace_trace_stack(struct trace_buffer * buffer,unsigned int trace_ctx,int skip,struct pt_regs * regs)977 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
978 unsigned int trace_ctx,
979 int skip, struct pt_regs *regs)
980 {
981 }
ftrace_trace_stack(struct trace_array * tr,struct trace_buffer * buffer,unsigned long trace_ctx,int skip,struct pt_regs * regs)982 static inline void ftrace_trace_stack(struct trace_array *tr,
983 struct trace_buffer *buffer,
984 unsigned long trace_ctx,
985 int skip, struct pt_regs *regs)
986 {
987 }
988
989 #endif
990
991 static __always_inline void
trace_event_setup(struct ring_buffer_event * event,int type,unsigned int trace_ctx)992 trace_event_setup(struct ring_buffer_event *event,
993 int type, unsigned int trace_ctx)
994 {
995 struct trace_entry *ent = ring_buffer_event_data(event);
996
997 tracing_generic_entry_update(ent, type, trace_ctx);
998 }
999
1000 static __always_inline struct ring_buffer_event *
__trace_buffer_lock_reserve(struct trace_buffer * buffer,int type,unsigned long len,unsigned int trace_ctx)1001 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
1002 int type,
1003 unsigned long len,
1004 unsigned int trace_ctx)
1005 {
1006 struct ring_buffer_event *event;
1007
1008 event = ring_buffer_lock_reserve(buffer, len);
1009 if (event != NULL)
1010 trace_event_setup(event, type, trace_ctx);
1011
1012 return event;
1013 }
1014
tracer_tracing_on(struct trace_array * tr)1015 void tracer_tracing_on(struct trace_array *tr)
1016 {
1017 if (tr->array_buffer.buffer)
1018 ring_buffer_record_on(tr->array_buffer.buffer);
1019 /*
1020 * This flag is looked at when buffers haven't been allocated
1021 * yet, or by some tracers (like irqsoff), that just want to
1022 * know if the ring buffer has been disabled, but it can handle
1023 * races of where it gets disabled but we still do a record.
1024 * As the check is in the fast path of the tracers, it is more
1025 * important to be fast than accurate.
1026 */
1027 tr->buffer_disabled = 0;
1028 /* Make the flag seen by readers */
1029 smp_wmb();
1030 }
1031
1032 /**
1033 * tracing_on - enable tracing buffers
1034 *
1035 * This function enables tracing buffers that may have been
1036 * disabled with tracing_off.
1037 */
tracing_on(void)1038 void tracing_on(void)
1039 {
1040 tracer_tracing_on(&global_trace);
1041 }
1042 EXPORT_SYMBOL_GPL(tracing_on);
1043
1044
1045 static __always_inline void
__buffer_unlock_commit(struct trace_buffer * buffer,struct ring_buffer_event * event)1046 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
1047 {
1048 __this_cpu_write(trace_taskinfo_save, true);
1049
1050 /* If this is the temp buffer, we need to commit fully */
1051 if (this_cpu_read(trace_buffered_event) == event) {
1052 /* Length is in event->array[0] */
1053 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1054 /* Release the temp buffer */
1055 this_cpu_dec(trace_buffered_event_cnt);
1056 /* ring_buffer_unlock_commit() enables preemption */
1057 preempt_enable_notrace();
1058 } else
1059 ring_buffer_unlock_commit(buffer);
1060 }
1061
__trace_array_puts(struct trace_array * tr,unsigned long ip,const char * str,int size)1062 int __trace_array_puts(struct trace_array *tr, unsigned long ip,
1063 const char *str, int size)
1064 {
1065 struct ring_buffer_event *event;
1066 struct trace_buffer *buffer;
1067 struct print_entry *entry;
1068 unsigned int trace_ctx;
1069 int alloc;
1070
1071 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1072 return 0;
1073
1074 if (unlikely(tracing_selftest_running && tr == &global_trace))
1075 return 0;
1076
1077 if (unlikely(tracing_disabled))
1078 return 0;
1079
1080 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1081
1082 trace_ctx = tracing_gen_ctx();
1083 buffer = tr->array_buffer.buffer;
1084 ring_buffer_nest_start(buffer);
1085 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1086 trace_ctx);
1087 if (!event) {
1088 size = 0;
1089 goto out;
1090 }
1091
1092 entry = ring_buffer_event_data(event);
1093 entry->ip = ip;
1094
1095 memcpy(&entry->buf, str, size);
1096
1097 /* Add a newline if necessary */
1098 if (entry->buf[size - 1] != '\n') {
1099 entry->buf[size] = '\n';
1100 entry->buf[size + 1] = '\0';
1101 } else
1102 entry->buf[size] = '\0';
1103
1104 __buffer_unlock_commit(buffer, event);
1105 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1106 out:
1107 ring_buffer_nest_end(buffer);
1108 return size;
1109 }
1110 EXPORT_SYMBOL_GPL(__trace_array_puts);
1111
1112 /**
1113 * __trace_puts - write a constant string into the trace buffer.
1114 * @ip: The address of the caller
1115 * @str: The constant string to write
1116 * @size: The size of the string.
1117 */
__trace_puts(unsigned long ip,const char * str,int size)1118 int __trace_puts(unsigned long ip, const char *str, int size)
1119 {
1120 return __trace_array_puts(&global_trace, ip, str, size);
1121 }
1122 EXPORT_SYMBOL_GPL(__trace_puts);
1123
1124 /**
1125 * __trace_bputs - write the pointer to a constant string into trace buffer
1126 * @ip: The address of the caller
1127 * @str: The constant string to write to the buffer to
1128 */
__trace_bputs(unsigned long ip,const char * str)1129 int __trace_bputs(unsigned long ip, const char *str)
1130 {
1131 struct ring_buffer_event *event;
1132 struct trace_buffer *buffer;
1133 struct bputs_entry *entry;
1134 unsigned int trace_ctx;
1135 int size = sizeof(struct bputs_entry);
1136 int ret = 0;
1137
1138 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1139 return 0;
1140
1141 if (unlikely(tracing_selftest_running || tracing_disabled))
1142 return 0;
1143
1144 trace_ctx = tracing_gen_ctx();
1145 buffer = global_trace.array_buffer.buffer;
1146
1147 ring_buffer_nest_start(buffer);
1148 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1149 trace_ctx);
1150 if (!event)
1151 goto out;
1152
1153 entry = ring_buffer_event_data(event);
1154 entry->ip = ip;
1155 entry->str = str;
1156
1157 __buffer_unlock_commit(buffer, event);
1158 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1159
1160 ret = 1;
1161 out:
1162 ring_buffer_nest_end(buffer);
1163 return ret;
1164 }
1165 EXPORT_SYMBOL_GPL(__trace_bputs);
1166
1167 #ifdef CONFIG_TRACER_SNAPSHOT
tracing_snapshot_instance_cond(struct trace_array * tr,void * cond_data)1168 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1169 void *cond_data)
1170 {
1171 struct tracer *tracer = tr->current_trace;
1172 unsigned long flags;
1173
1174 if (in_nmi()) {
1175 trace_array_puts(tr, "*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1176 trace_array_puts(tr, "*** snapshot is being ignored ***\n");
1177 return;
1178 }
1179
1180 if (!tr->allocated_snapshot) {
1181 trace_array_puts(tr, "*** SNAPSHOT NOT ALLOCATED ***\n");
1182 trace_array_puts(tr, "*** stopping trace here! ***\n");
1183 tracer_tracing_off(tr);
1184 return;
1185 }
1186
1187 /* Note, snapshot can not be used when the tracer uses it */
1188 if (tracer->use_max_tr) {
1189 trace_array_puts(tr, "*** LATENCY TRACER ACTIVE ***\n");
1190 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1191 return;
1192 }
1193
1194 if (tr->mapped) {
1195 trace_array_puts(tr, "*** BUFFER MEMORY MAPPED ***\n");
1196 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1197 return;
1198 }
1199
1200 local_irq_save(flags);
1201 update_max_tr(tr, current, smp_processor_id(), cond_data);
1202 local_irq_restore(flags);
1203 }
1204
tracing_snapshot_instance(struct trace_array * tr)1205 void tracing_snapshot_instance(struct trace_array *tr)
1206 {
1207 tracing_snapshot_instance_cond(tr, NULL);
1208 }
1209
1210 /**
1211 * tracing_snapshot - take a snapshot of the current buffer.
1212 *
1213 * This causes a swap between the snapshot buffer and the current live
1214 * tracing buffer. You can use this to take snapshots of the live
1215 * trace when some condition is triggered, but continue to trace.
1216 *
1217 * Note, make sure to allocate the snapshot with either
1218 * a tracing_snapshot_alloc(), or by doing it manually
1219 * with: echo 1 > /sys/kernel/tracing/snapshot
1220 *
1221 * If the snapshot buffer is not allocated, it will stop tracing.
1222 * Basically making a permanent snapshot.
1223 */
tracing_snapshot(void)1224 void tracing_snapshot(void)
1225 {
1226 struct trace_array *tr = &global_trace;
1227
1228 tracing_snapshot_instance(tr);
1229 }
1230 EXPORT_SYMBOL_GPL(tracing_snapshot);
1231
1232 /**
1233 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1234 * @tr: The tracing instance to snapshot
1235 * @cond_data: The data to be tested conditionally, and possibly saved
1236 *
1237 * This is the same as tracing_snapshot() except that the snapshot is
1238 * conditional - the snapshot will only happen if the
1239 * cond_snapshot.update() implementation receiving the cond_data
1240 * returns true, which means that the trace array's cond_snapshot
1241 * update() operation used the cond_data to determine whether the
1242 * snapshot should be taken, and if it was, presumably saved it along
1243 * with the snapshot.
1244 */
tracing_snapshot_cond(struct trace_array * tr,void * cond_data)1245 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1246 {
1247 tracing_snapshot_instance_cond(tr, cond_data);
1248 }
1249 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1250
1251 /**
1252 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1253 * @tr: The tracing instance
1254 *
1255 * When the user enables a conditional snapshot using
1256 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1257 * with the snapshot. This accessor is used to retrieve it.
1258 *
1259 * Should not be called from cond_snapshot.update(), since it takes
1260 * the tr->max_lock lock, which the code calling
1261 * cond_snapshot.update() has already done.
1262 *
1263 * Returns the cond_data associated with the trace array's snapshot.
1264 */
tracing_cond_snapshot_data(struct trace_array * tr)1265 void *tracing_cond_snapshot_data(struct trace_array *tr)
1266 {
1267 void *cond_data = NULL;
1268
1269 local_irq_disable();
1270 arch_spin_lock(&tr->max_lock);
1271
1272 if (tr->cond_snapshot)
1273 cond_data = tr->cond_snapshot->cond_data;
1274
1275 arch_spin_unlock(&tr->max_lock);
1276 local_irq_enable();
1277
1278 return cond_data;
1279 }
1280 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1281
1282 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1283 struct array_buffer *size_buf, int cpu_id);
1284 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1285
tracing_alloc_snapshot_instance(struct trace_array * tr)1286 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1287 {
1288 int order;
1289 int ret;
1290
1291 if (!tr->allocated_snapshot) {
1292
1293 /* Make the snapshot buffer have the same order as main buffer */
1294 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
1295 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
1296 if (ret < 0)
1297 return ret;
1298
1299 /* allocate spare buffer */
1300 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1301 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1302 if (ret < 0)
1303 return ret;
1304
1305 tr->allocated_snapshot = true;
1306 }
1307
1308 return 0;
1309 }
1310
free_snapshot(struct trace_array * tr)1311 static void free_snapshot(struct trace_array *tr)
1312 {
1313 /*
1314 * We don't free the ring buffer. instead, resize it because
1315 * The max_tr ring buffer has some state (e.g. ring->clock) and
1316 * we want preserve it.
1317 */
1318 ring_buffer_subbuf_order_set(tr->max_buffer.buffer, 0);
1319 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1320 set_buffer_entries(&tr->max_buffer, 1);
1321 tracing_reset_online_cpus(&tr->max_buffer);
1322 tr->allocated_snapshot = false;
1323 }
1324
tracing_arm_snapshot_locked(struct trace_array * tr)1325 static int tracing_arm_snapshot_locked(struct trace_array *tr)
1326 {
1327 int ret;
1328
1329 lockdep_assert_held(&trace_types_lock);
1330
1331 spin_lock(&tr->snapshot_trigger_lock);
1332 if (tr->snapshot == UINT_MAX || tr->mapped) {
1333 spin_unlock(&tr->snapshot_trigger_lock);
1334 return -EBUSY;
1335 }
1336
1337 tr->snapshot++;
1338 spin_unlock(&tr->snapshot_trigger_lock);
1339
1340 ret = tracing_alloc_snapshot_instance(tr);
1341 if (ret) {
1342 spin_lock(&tr->snapshot_trigger_lock);
1343 tr->snapshot--;
1344 spin_unlock(&tr->snapshot_trigger_lock);
1345 }
1346
1347 return ret;
1348 }
1349
tracing_arm_snapshot(struct trace_array * tr)1350 int tracing_arm_snapshot(struct trace_array *tr)
1351 {
1352 int ret;
1353
1354 mutex_lock(&trace_types_lock);
1355 ret = tracing_arm_snapshot_locked(tr);
1356 mutex_unlock(&trace_types_lock);
1357
1358 return ret;
1359 }
1360
tracing_disarm_snapshot(struct trace_array * tr)1361 void tracing_disarm_snapshot(struct trace_array *tr)
1362 {
1363 spin_lock(&tr->snapshot_trigger_lock);
1364 if (!WARN_ON(!tr->snapshot))
1365 tr->snapshot--;
1366 spin_unlock(&tr->snapshot_trigger_lock);
1367 }
1368
1369 /**
1370 * tracing_alloc_snapshot - allocate snapshot buffer.
1371 *
1372 * This only allocates the snapshot buffer if it isn't already
1373 * allocated - it doesn't also take a snapshot.
1374 *
1375 * This is meant to be used in cases where the snapshot buffer needs
1376 * to be set up for events that can't sleep but need to be able to
1377 * trigger a snapshot.
1378 */
tracing_alloc_snapshot(void)1379 int tracing_alloc_snapshot(void)
1380 {
1381 struct trace_array *tr = &global_trace;
1382 int ret;
1383
1384 ret = tracing_alloc_snapshot_instance(tr);
1385 WARN_ON(ret < 0);
1386
1387 return ret;
1388 }
1389 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1390
1391 /**
1392 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1393 *
1394 * This is similar to tracing_snapshot(), but it will allocate the
1395 * snapshot buffer if it isn't already allocated. Use this only
1396 * where it is safe to sleep, as the allocation may sleep.
1397 *
1398 * This causes a swap between the snapshot buffer and the current live
1399 * tracing buffer. You can use this to take snapshots of the live
1400 * trace when some condition is triggered, but continue to trace.
1401 */
tracing_snapshot_alloc(void)1402 void tracing_snapshot_alloc(void)
1403 {
1404 int ret;
1405
1406 ret = tracing_alloc_snapshot();
1407 if (ret < 0)
1408 return;
1409
1410 tracing_snapshot();
1411 }
1412 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1413
1414 /**
1415 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1416 * @tr: The tracing instance
1417 * @cond_data: User data to associate with the snapshot
1418 * @update: Implementation of the cond_snapshot update function
1419 *
1420 * Check whether the conditional snapshot for the given instance has
1421 * already been enabled, or if the current tracer is already using a
1422 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1423 * save the cond_data and update function inside.
1424 *
1425 * Returns 0 if successful, error otherwise.
1426 */
tracing_snapshot_cond_enable(struct trace_array * tr,void * cond_data,cond_update_fn_t update)1427 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1428 cond_update_fn_t update)
1429 {
1430 struct cond_snapshot *cond_snapshot;
1431 int ret = 0;
1432
1433 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1434 if (!cond_snapshot)
1435 return -ENOMEM;
1436
1437 cond_snapshot->cond_data = cond_data;
1438 cond_snapshot->update = update;
1439
1440 mutex_lock(&trace_types_lock);
1441
1442 if (tr->current_trace->use_max_tr) {
1443 ret = -EBUSY;
1444 goto fail_unlock;
1445 }
1446
1447 /*
1448 * The cond_snapshot can only change to NULL without the
1449 * trace_types_lock. We don't care if we race with it going
1450 * to NULL, but we want to make sure that it's not set to
1451 * something other than NULL when we get here, which we can
1452 * do safely with only holding the trace_types_lock and not
1453 * having to take the max_lock.
1454 */
1455 if (tr->cond_snapshot) {
1456 ret = -EBUSY;
1457 goto fail_unlock;
1458 }
1459
1460 ret = tracing_arm_snapshot_locked(tr);
1461 if (ret)
1462 goto fail_unlock;
1463
1464 local_irq_disable();
1465 arch_spin_lock(&tr->max_lock);
1466 tr->cond_snapshot = cond_snapshot;
1467 arch_spin_unlock(&tr->max_lock);
1468 local_irq_enable();
1469
1470 mutex_unlock(&trace_types_lock);
1471
1472 return ret;
1473
1474 fail_unlock:
1475 mutex_unlock(&trace_types_lock);
1476 kfree(cond_snapshot);
1477 return ret;
1478 }
1479 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1480
1481 /**
1482 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1483 * @tr: The tracing instance
1484 *
1485 * Check whether the conditional snapshot for the given instance is
1486 * enabled; if so, free the cond_snapshot associated with it,
1487 * otherwise return -EINVAL.
1488 *
1489 * Returns 0 if successful, error otherwise.
1490 */
tracing_snapshot_cond_disable(struct trace_array * tr)1491 int tracing_snapshot_cond_disable(struct trace_array *tr)
1492 {
1493 int ret = 0;
1494
1495 local_irq_disable();
1496 arch_spin_lock(&tr->max_lock);
1497
1498 if (!tr->cond_snapshot)
1499 ret = -EINVAL;
1500 else {
1501 kfree(tr->cond_snapshot);
1502 tr->cond_snapshot = NULL;
1503 }
1504
1505 arch_spin_unlock(&tr->max_lock);
1506 local_irq_enable();
1507
1508 tracing_disarm_snapshot(tr);
1509
1510 return ret;
1511 }
1512 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1513 #else
tracing_snapshot(void)1514 void tracing_snapshot(void)
1515 {
1516 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1517 }
1518 EXPORT_SYMBOL_GPL(tracing_snapshot);
tracing_snapshot_cond(struct trace_array * tr,void * cond_data)1519 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1520 {
1521 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1522 }
1523 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
tracing_alloc_snapshot(void)1524 int tracing_alloc_snapshot(void)
1525 {
1526 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1527 return -ENODEV;
1528 }
1529 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
tracing_snapshot_alloc(void)1530 void tracing_snapshot_alloc(void)
1531 {
1532 /* Give warning */
1533 tracing_snapshot();
1534 }
1535 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
tracing_cond_snapshot_data(struct trace_array * tr)1536 void *tracing_cond_snapshot_data(struct trace_array *tr)
1537 {
1538 return NULL;
1539 }
1540 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
tracing_snapshot_cond_enable(struct trace_array * tr,void * cond_data,cond_update_fn_t update)1541 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1542 {
1543 return -ENODEV;
1544 }
1545 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
tracing_snapshot_cond_disable(struct trace_array * tr)1546 int tracing_snapshot_cond_disable(struct trace_array *tr)
1547 {
1548 return false;
1549 }
1550 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1551 #define free_snapshot(tr) do { } while (0)
1552 #define tracing_arm_snapshot_locked(tr) ({ -EBUSY; })
1553 #endif /* CONFIG_TRACER_SNAPSHOT */
1554
tracer_tracing_off(struct trace_array * tr)1555 void tracer_tracing_off(struct trace_array *tr)
1556 {
1557 if (tr->array_buffer.buffer)
1558 ring_buffer_record_off(tr->array_buffer.buffer);
1559 /*
1560 * This flag is looked at when buffers haven't been allocated
1561 * yet, or by some tracers (like irqsoff), that just want to
1562 * know if the ring buffer has been disabled, but it can handle
1563 * races of where it gets disabled but we still do a record.
1564 * As the check is in the fast path of the tracers, it is more
1565 * important to be fast than accurate.
1566 */
1567 tr->buffer_disabled = 1;
1568 /* Make the flag seen by readers */
1569 smp_wmb();
1570 }
1571
1572 /**
1573 * tracing_off - turn off tracing buffers
1574 *
1575 * This function stops the tracing buffers from recording data.
1576 * It does not disable any overhead the tracers themselves may
1577 * be causing. This function simply causes all recording to
1578 * the ring buffers to fail.
1579 */
tracing_off(void)1580 void tracing_off(void)
1581 {
1582 tracer_tracing_off(&global_trace);
1583 }
1584 EXPORT_SYMBOL_GPL(tracing_off);
1585
disable_trace_on_warning(void)1586 void disable_trace_on_warning(void)
1587 {
1588 if (__disable_trace_on_warning) {
1589 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1590 "Disabling tracing due to warning\n");
1591 tracing_off();
1592 }
1593 }
1594
1595 /**
1596 * tracer_tracing_is_on - show real state of ring buffer enabled
1597 * @tr : the trace array to know if ring buffer is enabled
1598 *
1599 * Shows real state of the ring buffer if it is enabled or not.
1600 */
tracer_tracing_is_on(struct trace_array * tr)1601 bool tracer_tracing_is_on(struct trace_array *tr)
1602 {
1603 if (tr->array_buffer.buffer)
1604 return ring_buffer_record_is_set_on(tr->array_buffer.buffer);
1605 return !tr->buffer_disabled;
1606 }
1607
1608 /**
1609 * tracing_is_on - show state of ring buffers enabled
1610 */
tracing_is_on(void)1611 int tracing_is_on(void)
1612 {
1613 return tracer_tracing_is_on(&global_trace);
1614 }
1615 EXPORT_SYMBOL_GPL(tracing_is_on);
1616
set_buf_size(char * str)1617 static int __init set_buf_size(char *str)
1618 {
1619 unsigned long buf_size;
1620
1621 if (!str)
1622 return 0;
1623 buf_size = memparse(str, &str);
1624 /*
1625 * nr_entries can not be zero and the startup
1626 * tests require some buffer space. Therefore
1627 * ensure we have at least 4096 bytes of buffer.
1628 */
1629 trace_buf_size = max(4096UL, buf_size);
1630 return 1;
1631 }
1632 __setup("trace_buf_size=", set_buf_size);
1633
set_tracing_thresh(char * str)1634 static int __init set_tracing_thresh(char *str)
1635 {
1636 unsigned long threshold;
1637 int ret;
1638
1639 if (!str)
1640 return 0;
1641 ret = kstrtoul(str, 0, &threshold);
1642 if (ret < 0)
1643 return 0;
1644 tracing_thresh = threshold * 1000;
1645 return 1;
1646 }
1647 __setup("tracing_thresh=", set_tracing_thresh);
1648
nsecs_to_usecs(unsigned long nsecs)1649 unsigned long nsecs_to_usecs(unsigned long nsecs)
1650 {
1651 return nsecs / 1000;
1652 }
1653
1654 /*
1655 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1656 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1657 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1658 * of strings in the order that the evals (enum) were defined.
1659 */
1660 #undef C
1661 #define C(a, b) b
1662
1663 /* These must match the bit positions in trace_iterator_flags */
1664 static const char *trace_options[] = {
1665 TRACE_FLAGS
1666 NULL
1667 };
1668
1669 static struct {
1670 u64 (*func)(void);
1671 const char *name;
1672 int in_ns; /* is this clock in nanoseconds? */
1673 } trace_clocks[] = {
1674 { trace_clock_local, "local", 1 },
1675 { trace_clock_global, "global", 1 },
1676 { trace_clock_counter, "counter", 0 },
1677 { trace_clock_jiffies, "uptime", 0 },
1678 { trace_clock, "perf", 1 },
1679 { ktime_get_mono_fast_ns, "mono", 1 },
1680 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1681 { ktime_get_boot_fast_ns, "boot", 1 },
1682 { ktime_get_tai_fast_ns, "tai", 1 },
1683 ARCH_TRACE_CLOCKS
1684 };
1685
trace_clock_in_ns(struct trace_array * tr)1686 bool trace_clock_in_ns(struct trace_array *tr)
1687 {
1688 if (trace_clocks[tr->clock_id].in_ns)
1689 return true;
1690
1691 return false;
1692 }
1693
1694 /*
1695 * trace_parser_get_init - gets the buffer for trace parser
1696 */
trace_parser_get_init(struct trace_parser * parser,int size)1697 int trace_parser_get_init(struct trace_parser *parser, int size)
1698 {
1699 memset(parser, 0, sizeof(*parser));
1700
1701 parser->buffer = kmalloc(size, GFP_KERNEL);
1702 if (!parser->buffer)
1703 return 1;
1704
1705 parser->size = size;
1706 return 0;
1707 }
1708
1709 /*
1710 * trace_parser_put - frees the buffer for trace parser
1711 */
trace_parser_put(struct trace_parser * parser)1712 void trace_parser_put(struct trace_parser *parser)
1713 {
1714 kfree(parser->buffer);
1715 parser->buffer = NULL;
1716 }
1717
1718 /*
1719 * trace_get_user - reads the user input string separated by space
1720 * (matched by isspace(ch))
1721 *
1722 * For each string found the 'struct trace_parser' is updated,
1723 * and the function returns.
1724 *
1725 * Returns number of bytes read.
1726 *
1727 * See kernel/trace/trace.h for 'struct trace_parser' details.
1728 */
trace_get_user(struct trace_parser * parser,const char __user * ubuf,size_t cnt,loff_t * ppos)1729 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1730 size_t cnt, loff_t *ppos)
1731 {
1732 char ch;
1733 size_t read = 0;
1734 ssize_t ret;
1735
1736 if (!*ppos)
1737 trace_parser_clear(parser);
1738
1739 ret = get_user(ch, ubuf++);
1740 if (ret)
1741 goto out;
1742
1743 read++;
1744 cnt--;
1745
1746 /*
1747 * The parser is not finished with the last write,
1748 * continue reading the user input without skipping spaces.
1749 */
1750 if (!parser->cont) {
1751 /* skip white space */
1752 while (cnt && isspace(ch)) {
1753 ret = get_user(ch, ubuf++);
1754 if (ret)
1755 goto out;
1756 read++;
1757 cnt--;
1758 }
1759
1760 parser->idx = 0;
1761
1762 /* only spaces were written */
1763 if (isspace(ch) || !ch) {
1764 *ppos += read;
1765 ret = read;
1766 goto out;
1767 }
1768 }
1769
1770 /* read the non-space input */
1771 while (cnt && !isspace(ch) && ch) {
1772 if (parser->idx < parser->size - 1)
1773 parser->buffer[parser->idx++] = ch;
1774 else {
1775 ret = -EINVAL;
1776 goto out;
1777 }
1778 ret = get_user(ch, ubuf++);
1779 if (ret)
1780 goto out;
1781 read++;
1782 cnt--;
1783 }
1784
1785 /* We either got finished input or we have to wait for another call. */
1786 if (isspace(ch) || !ch) {
1787 parser->buffer[parser->idx] = 0;
1788 parser->cont = false;
1789 } else if (parser->idx < parser->size - 1) {
1790 parser->cont = true;
1791 parser->buffer[parser->idx++] = ch;
1792 /* Make sure the parsed string always terminates with '\0'. */
1793 parser->buffer[parser->idx] = 0;
1794 } else {
1795 ret = -EINVAL;
1796 goto out;
1797 }
1798
1799 *ppos += read;
1800 ret = read;
1801
1802 out:
1803 return ret;
1804 }
1805
1806 /* TODO add a seq_buf_to_buffer() */
trace_seq_to_buffer(struct trace_seq * s,void * buf,size_t cnt)1807 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1808 {
1809 int len;
1810
1811 if (trace_seq_used(s) <= s->readpos)
1812 return -EBUSY;
1813
1814 len = trace_seq_used(s) - s->readpos;
1815 if (cnt > len)
1816 cnt = len;
1817 memcpy(buf, s->buffer + s->readpos, cnt);
1818
1819 s->readpos += cnt;
1820 return cnt;
1821 }
1822
1823 unsigned long __read_mostly tracing_thresh;
1824
1825 #ifdef CONFIG_TRACER_MAX_TRACE
1826 static const struct file_operations tracing_max_lat_fops;
1827
1828 #ifdef LATENCY_FS_NOTIFY
1829
1830 static struct workqueue_struct *fsnotify_wq;
1831
latency_fsnotify_workfn(struct work_struct * work)1832 static void latency_fsnotify_workfn(struct work_struct *work)
1833 {
1834 struct trace_array *tr = container_of(work, struct trace_array,
1835 fsnotify_work);
1836 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1837 }
1838
latency_fsnotify_workfn_irq(struct irq_work * iwork)1839 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1840 {
1841 struct trace_array *tr = container_of(iwork, struct trace_array,
1842 fsnotify_irqwork);
1843 queue_work(fsnotify_wq, &tr->fsnotify_work);
1844 }
1845
trace_create_maxlat_file(struct trace_array * tr,struct dentry * d_tracer)1846 static void trace_create_maxlat_file(struct trace_array *tr,
1847 struct dentry *d_tracer)
1848 {
1849 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1850 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1851 tr->d_max_latency = trace_create_file("tracing_max_latency",
1852 TRACE_MODE_WRITE,
1853 d_tracer, tr,
1854 &tracing_max_lat_fops);
1855 }
1856
latency_fsnotify_init(void)1857 __init static int latency_fsnotify_init(void)
1858 {
1859 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1860 WQ_UNBOUND | WQ_HIGHPRI, 0);
1861 if (!fsnotify_wq) {
1862 pr_err("Unable to allocate tr_max_lat_wq\n");
1863 return -ENOMEM;
1864 }
1865 return 0;
1866 }
1867
1868 late_initcall_sync(latency_fsnotify_init);
1869
latency_fsnotify(struct trace_array * tr)1870 void latency_fsnotify(struct trace_array *tr)
1871 {
1872 if (!fsnotify_wq)
1873 return;
1874 /*
1875 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1876 * possible that we are called from __schedule() or do_idle(), which
1877 * could cause a deadlock.
1878 */
1879 irq_work_queue(&tr->fsnotify_irqwork);
1880 }
1881
1882 #else /* !LATENCY_FS_NOTIFY */
1883
1884 #define trace_create_maxlat_file(tr, d_tracer) \
1885 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1886 d_tracer, tr, &tracing_max_lat_fops)
1887
1888 #endif
1889
1890 /*
1891 * Copy the new maximum trace into the separate maximum-trace
1892 * structure. (this way the maximum trace is permanently saved,
1893 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1894 */
1895 static void
__update_max_tr(struct trace_array * tr,struct task_struct * tsk,int cpu)1896 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1897 {
1898 struct array_buffer *trace_buf = &tr->array_buffer;
1899 struct array_buffer *max_buf = &tr->max_buffer;
1900 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1901 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1902
1903 max_buf->cpu = cpu;
1904 max_buf->time_start = data->preempt_timestamp;
1905
1906 max_data->saved_latency = tr->max_latency;
1907 max_data->critical_start = data->critical_start;
1908 max_data->critical_end = data->critical_end;
1909
1910 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1911 max_data->pid = tsk->pid;
1912 /*
1913 * If tsk == current, then use current_uid(), as that does not use
1914 * RCU. The irq tracer can be called out of RCU scope.
1915 */
1916 if (tsk == current)
1917 max_data->uid = current_uid();
1918 else
1919 max_data->uid = task_uid(tsk);
1920
1921 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1922 max_data->policy = tsk->policy;
1923 max_data->rt_priority = tsk->rt_priority;
1924
1925 /* record this tasks comm */
1926 tracing_record_cmdline(tsk);
1927 latency_fsnotify(tr);
1928 }
1929
1930 /**
1931 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1932 * @tr: tracer
1933 * @tsk: the task with the latency
1934 * @cpu: The cpu that initiated the trace.
1935 * @cond_data: User data associated with a conditional snapshot
1936 *
1937 * Flip the buffers between the @tr and the max_tr and record information
1938 * about which task was the cause of this latency.
1939 */
1940 void
update_max_tr(struct trace_array * tr,struct task_struct * tsk,int cpu,void * cond_data)1941 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1942 void *cond_data)
1943 {
1944 if (tr->stop_count)
1945 return;
1946
1947 WARN_ON_ONCE(!irqs_disabled());
1948
1949 if (!tr->allocated_snapshot) {
1950 /* Only the nop tracer should hit this when disabling */
1951 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1952 return;
1953 }
1954
1955 arch_spin_lock(&tr->max_lock);
1956
1957 /* Inherit the recordable setting from array_buffer */
1958 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1959 ring_buffer_record_on(tr->max_buffer.buffer);
1960 else
1961 ring_buffer_record_off(tr->max_buffer.buffer);
1962
1963 #ifdef CONFIG_TRACER_SNAPSHOT
1964 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1965 arch_spin_unlock(&tr->max_lock);
1966 return;
1967 }
1968 #endif
1969 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1970
1971 __update_max_tr(tr, tsk, cpu);
1972
1973 arch_spin_unlock(&tr->max_lock);
1974
1975 /* Any waiters on the old snapshot buffer need to wake up */
1976 ring_buffer_wake_waiters(tr->array_buffer.buffer, RING_BUFFER_ALL_CPUS);
1977 }
1978
1979 /**
1980 * update_max_tr_single - only copy one trace over, and reset the rest
1981 * @tr: tracer
1982 * @tsk: task with the latency
1983 * @cpu: the cpu of the buffer to copy.
1984 *
1985 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1986 */
1987 void
update_max_tr_single(struct trace_array * tr,struct task_struct * tsk,int cpu)1988 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1989 {
1990 int ret;
1991
1992 if (tr->stop_count)
1993 return;
1994
1995 WARN_ON_ONCE(!irqs_disabled());
1996 if (!tr->allocated_snapshot) {
1997 /* Only the nop tracer should hit this when disabling */
1998 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1999 return;
2000 }
2001
2002 arch_spin_lock(&tr->max_lock);
2003
2004 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
2005
2006 if (ret == -EBUSY) {
2007 /*
2008 * We failed to swap the buffer due to a commit taking
2009 * place on this CPU. We fail to record, but we reset
2010 * the max trace buffer (no one writes directly to it)
2011 * and flag that it failed.
2012 * Another reason is resize is in progress.
2013 */
2014 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
2015 "Failed to swap buffers due to commit or resize in progress\n");
2016 }
2017
2018 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
2019
2020 __update_max_tr(tr, tsk, cpu);
2021 arch_spin_unlock(&tr->max_lock);
2022 }
2023
2024 #endif /* CONFIG_TRACER_MAX_TRACE */
2025
2026 struct pipe_wait {
2027 struct trace_iterator *iter;
2028 int wait_index;
2029 };
2030
wait_pipe_cond(void * data)2031 static bool wait_pipe_cond(void *data)
2032 {
2033 struct pipe_wait *pwait = data;
2034 struct trace_iterator *iter = pwait->iter;
2035
2036 if (atomic_read_acquire(&iter->wait_index) != pwait->wait_index)
2037 return true;
2038
2039 return iter->closed;
2040 }
2041
wait_on_pipe(struct trace_iterator * iter,int full)2042 static int wait_on_pipe(struct trace_iterator *iter, int full)
2043 {
2044 struct pipe_wait pwait;
2045 int ret;
2046
2047 /* Iterators are static, they should be filled or empty */
2048 if (trace_buffer_iter(iter, iter->cpu_file))
2049 return 0;
2050
2051 pwait.wait_index = atomic_read_acquire(&iter->wait_index);
2052 pwait.iter = iter;
2053
2054 ret = ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file, full,
2055 wait_pipe_cond, &pwait);
2056
2057 #ifdef CONFIG_TRACER_MAX_TRACE
2058 /*
2059 * Make sure this is still the snapshot buffer, as if a snapshot were
2060 * to happen, this would now be the main buffer.
2061 */
2062 if (iter->snapshot)
2063 iter->array_buffer = &iter->tr->max_buffer;
2064 #endif
2065 return ret;
2066 }
2067
2068 #ifdef CONFIG_FTRACE_STARTUP_TEST
2069 static bool selftests_can_run;
2070
2071 struct trace_selftests {
2072 struct list_head list;
2073 struct tracer *type;
2074 };
2075
2076 static LIST_HEAD(postponed_selftests);
2077
save_selftest(struct tracer * type)2078 static int save_selftest(struct tracer *type)
2079 {
2080 struct trace_selftests *selftest;
2081
2082 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
2083 if (!selftest)
2084 return -ENOMEM;
2085
2086 selftest->type = type;
2087 list_add(&selftest->list, &postponed_selftests);
2088 return 0;
2089 }
2090
run_tracer_selftest(struct tracer * type)2091 static int run_tracer_selftest(struct tracer *type)
2092 {
2093 struct trace_array *tr = &global_trace;
2094 struct tracer *saved_tracer = tr->current_trace;
2095 int ret;
2096
2097 if (!type->selftest || tracing_selftest_disabled)
2098 return 0;
2099
2100 /*
2101 * If a tracer registers early in boot up (before scheduling is
2102 * initialized and such), then do not run its selftests yet.
2103 * Instead, run it a little later in the boot process.
2104 */
2105 if (!selftests_can_run)
2106 return save_selftest(type);
2107
2108 if (!tracing_is_on()) {
2109 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2110 type->name);
2111 return 0;
2112 }
2113
2114 /*
2115 * Run a selftest on this tracer.
2116 * Here we reset the trace buffer, and set the current
2117 * tracer to be this tracer. The tracer can then run some
2118 * internal tracing to verify that everything is in order.
2119 * If we fail, we do not register this tracer.
2120 */
2121 tracing_reset_online_cpus(&tr->array_buffer);
2122
2123 tr->current_trace = type;
2124
2125 #ifdef CONFIG_TRACER_MAX_TRACE
2126 if (type->use_max_tr) {
2127 /* If we expanded the buffers, make sure the max is expanded too */
2128 if (tr->ring_buffer_expanded)
2129 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2130 RING_BUFFER_ALL_CPUS);
2131 tr->allocated_snapshot = true;
2132 }
2133 #endif
2134
2135 /* the test is responsible for initializing and enabling */
2136 pr_info("Testing tracer %s: ", type->name);
2137 ret = type->selftest(type, tr);
2138 /* the test is responsible for resetting too */
2139 tr->current_trace = saved_tracer;
2140 if (ret) {
2141 printk(KERN_CONT "FAILED!\n");
2142 /* Add the warning after printing 'FAILED' */
2143 WARN_ON(1);
2144 return -1;
2145 }
2146 /* Only reset on passing, to avoid touching corrupted buffers */
2147 tracing_reset_online_cpus(&tr->array_buffer);
2148
2149 #ifdef CONFIG_TRACER_MAX_TRACE
2150 if (type->use_max_tr) {
2151 tr->allocated_snapshot = false;
2152
2153 /* Shrink the max buffer again */
2154 if (tr->ring_buffer_expanded)
2155 ring_buffer_resize(tr->max_buffer.buffer, 1,
2156 RING_BUFFER_ALL_CPUS);
2157 }
2158 #endif
2159
2160 printk(KERN_CONT "PASSED\n");
2161 return 0;
2162 }
2163
do_run_tracer_selftest(struct tracer * type)2164 static int do_run_tracer_selftest(struct tracer *type)
2165 {
2166 int ret;
2167
2168 /*
2169 * Tests can take a long time, especially if they are run one after the
2170 * other, as does happen during bootup when all the tracers are
2171 * registered. This could cause the soft lockup watchdog to trigger.
2172 */
2173 cond_resched();
2174
2175 tracing_selftest_running = true;
2176 ret = run_tracer_selftest(type);
2177 tracing_selftest_running = false;
2178
2179 return ret;
2180 }
2181
init_trace_selftests(void)2182 static __init int init_trace_selftests(void)
2183 {
2184 struct trace_selftests *p, *n;
2185 struct tracer *t, **last;
2186 int ret;
2187
2188 selftests_can_run = true;
2189
2190 mutex_lock(&trace_types_lock);
2191
2192 if (list_empty(&postponed_selftests))
2193 goto out;
2194
2195 pr_info("Running postponed tracer tests:\n");
2196
2197 tracing_selftest_running = true;
2198 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2199 /* This loop can take minutes when sanitizers are enabled, so
2200 * lets make sure we allow RCU processing.
2201 */
2202 cond_resched();
2203 ret = run_tracer_selftest(p->type);
2204 /* If the test fails, then warn and remove from available_tracers */
2205 if (ret < 0) {
2206 WARN(1, "tracer: %s failed selftest, disabling\n",
2207 p->type->name);
2208 last = &trace_types;
2209 for (t = trace_types; t; t = t->next) {
2210 if (t == p->type) {
2211 *last = t->next;
2212 break;
2213 }
2214 last = &t->next;
2215 }
2216 }
2217 list_del(&p->list);
2218 kfree(p);
2219 }
2220 tracing_selftest_running = false;
2221
2222 out:
2223 mutex_unlock(&trace_types_lock);
2224
2225 return 0;
2226 }
2227 core_initcall(init_trace_selftests);
2228 #else
do_run_tracer_selftest(struct tracer * type)2229 static inline int do_run_tracer_selftest(struct tracer *type)
2230 {
2231 return 0;
2232 }
2233 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2234
2235 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2236
2237 static 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 */
register_tracer(struct tracer * type)2245 int __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
tracing_reset_cpu(struct array_buffer * buf,int cpu)2327 static 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
tracing_reset_online_cpus(struct array_buffer * buf)2343 void 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
2362 /* Must have trace_types_lock held */
tracing_reset_all_online_cpus_unlocked(void)2363 void tracing_reset_all_online_cpus_unlocked(void)
2364 {
2365 struct trace_array *tr;
2366
2367 lockdep_assert_held(&trace_types_lock);
2368
2369 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2370 if (!tr->clear_trace)
2371 continue;
2372 tr->clear_trace = false;
2373 tracing_reset_online_cpus(&tr->array_buffer);
2374 #ifdef CONFIG_TRACER_MAX_TRACE
2375 tracing_reset_online_cpus(&tr->max_buffer);
2376 #endif
2377 }
2378 }
2379
tracing_reset_all_online_cpus(void)2380 void tracing_reset_all_online_cpus(void)
2381 {
2382 mutex_lock(&trace_types_lock);
2383 tracing_reset_all_online_cpus_unlocked();
2384 mutex_unlock(&trace_types_lock);
2385 }
2386
is_tracing_stopped(void)2387 int is_tracing_stopped(void)
2388 {
2389 return global_trace.stop_count;
2390 }
2391
tracing_start_tr(struct trace_array * tr)2392 static void tracing_start_tr(struct trace_array *tr)
2393 {
2394 struct trace_buffer *buffer;
2395 unsigned long flags;
2396
2397 if (tracing_disabled)
2398 return;
2399
2400 raw_spin_lock_irqsave(&tr->start_lock, flags);
2401 if (--tr->stop_count) {
2402 if (WARN_ON_ONCE(tr->stop_count < 0)) {
2403 /* Someone screwed up their debugging */
2404 tr->stop_count = 0;
2405 }
2406 goto out;
2407 }
2408
2409 /* Prevent the buffers from switching */
2410 arch_spin_lock(&tr->max_lock);
2411
2412 buffer = tr->array_buffer.buffer;
2413 if (buffer)
2414 ring_buffer_record_enable(buffer);
2415
2416 #ifdef CONFIG_TRACER_MAX_TRACE
2417 buffer = tr->max_buffer.buffer;
2418 if (buffer)
2419 ring_buffer_record_enable(buffer);
2420 #endif
2421
2422 arch_spin_unlock(&tr->max_lock);
2423
2424 out:
2425 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2426 }
2427
2428 /**
2429 * tracing_start - quick start of the tracer
2430 *
2431 * If tracing is enabled but was stopped by tracing_stop,
2432 * this will start the tracer back up.
2433 */
tracing_start(void)2434 void tracing_start(void)
2435
2436 {
2437 return tracing_start_tr(&global_trace);
2438 }
2439
tracing_stop_tr(struct trace_array * tr)2440 static void tracing_stop_tr(struct trace_array *tr)
2441 {
2442 struct trace_buffer *buffer;
2443 unsigned long flags;
2444
2445 raw_spin_lock_irqsave(&tr->start_lock, flags);
2446 if (tr->stop_count++)
2447 goto out;
2448
2449 /* Prevent the buffers from switching */
2450 arch_spin_lock(&tr->max_lock);
2451
2452 buffer = tr->array_buffer.buffer;
2453 if (buffer)
2454 ring_buffer_record_disable(buffer);
2455
2456 #ifdef CONFIG_TRACER_MAX_TRACE
2457 buffer = tr->max_buffer.buffer;
2458 if (buffer)
2459 ring_buffer_record_disable(buffer);
2460 #endif
2461
2462 arch_spin_unlock(&tr->max_lock);
2463
2464 out:
2465 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2466 }
2467
2468 /**
2469 * tracing_stop - quick stop of the tracer
2470 *
2471 * Light weight way to stop tracing. Use in conjunction with
2472 * tracing_start.
2473 */
tracing_stop(void)2474 void tracing_stop(void)
2475 {
2476 return tracing_stop_tr(&global_trace);
2477 }
2478
2479 /*
2480 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2481 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2482 * simplifies those functions and keeps them in sync.
2483 */
trace_handle_return(struct trace_seq * s)2484 enum print_line_t trace_handle_return(struct trace_seq *s)
2485 {
2486 return trace_seq_has_overflowed(s) ?
2487 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2488 }
2489 EXPORT_SYMBOL_GPL(trace_handle_return);
2490
migration_disable_value(void)2491 static unsigned short migration_disable_value(void)
2492 {
2493 #if defined(CONFIG_SMP)
2494 return current->migration_disabled;
2495 #else
2496 return 0;
2497 #endif
2498 }
2499
tracing_gen_ctx_irq_test(unsigned int irqs_status)2500 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2501 {
2502 unsigned int trace_flags = irqs_status;
2503 unsigned int pc;
2504
2505 pc = preempt_count();
2506
2507 if (pc & NMI_MASK)
2508 trace_flags |= TRACE_FLAG_NMI;
2509 if (pc & HARDIRQ_MASK)
2510 trace_flags |= TRACE_FLAG_HARDIRQ;
2511 if (in_serving_softirq())
2512 trace_flags |= TRACE_FLAG_SOFTIRQ;
2513 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2514 trace_flags |= TRACE_FLAG_BH_OFF;
2515
2516 if (tif_need_resched())
2517 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2518 if (test_preempt_need_resched())
2519 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2520 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2521 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2522 }
2523
2524 struct ring_buffer_event *
trace_buffer_lock_reserve(struct trace_buffer * buffer,int type,unsigned long len,unsigned int trace_ctx)2525 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2526 int type,
2527 unsigned long len,
2528 unsigned int trace_ctx)
2529 {
2530 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2531 }
2532
2533 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2534 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2535 static int trace_buffered_event_ref;
2536
2537 /**
2538 * trace_buffered_event_enable - enable buffering events
2539 *
2540 * When events are being filtered, it is quicker to use a temporary
2541 * buffer to write the event data into if there's a likely chance
2542 * that it will not be committed. The discard of the ring buffer
2543 * is not as fast as committing, and is much slower than copying
2544 * a commit.
2545 *
2546 * When an event is to be filtered, allocate per cpu buffers to
2547 * write the event data into, and if the event is filtered and discarded
2548 * it is simply dropped, otherwise, the entire data is to be committed
2549 * in one shot.
2550 */
trace_buffered_event_enable(void)2551 void trace_buffered_event_enable(void)
2552 {
2553 struct ring_buffer_event *event;
2554 struct page *page;
2555 int cpu;
2556
2557 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2558
2559 if (trace_buffered_event_ref++)
2560 return;
2561
2562 for_each_tracing_cpu(cpu) {
2563 page = alloc_pages_node(cpu_to_node(cpu),
2564 GFP_KERNEL | __GFP_NORETRY, 0);
2565 /* This is just an optimization and can handle failures */
2566 if (!page) {
2567 pr_err("Failed to allocate event buffer\n");
2568 break;
2569 }
2570
2571 event = page_address(page);
2572 memset(event, 0, sizeof(*event));
2573
2574 per_cpu(trace_buffered_event, cpu) = event;
2575
2576 preempt_disable();
2577 if (cpu == smp_processor_id() &&
2578 __this_cpu_read(trace_buffered_event) !=
2579 per_cpu(trace_buffered_event, cpu))
2580 WARN_ON_ONCE(1);
2581 preempt_enable();
2582 }
2583 }
2584
enable_trace_buffered_event(void * data)2585 static void enable_trace_buffered_event(void *data)
2586 {
2587 /* Probably not needed, but do it anyway */
2588 smp_rmb();
2589 this_cpu_dec(trace_buffered_event_cnt);
2590 }
2591
disable_trace_buffered_event(void * data)2592 static void disable_trace_buffered_event(void *data)
2593 {
2594 this_cpu_inc(trace_buffered_event_cnt);
2595 }
2596
2597 /**
2598 * trace_buffered_event_disable - disable buffering events
2599 *
2600 * When a filter is removed, it is faster to not use the buffered
2601 * events, and to commit directly into the ring buffer. Free up
2602 * the temp buffers when there are no more users. This requires
2603 * special synchronization with current events.
2604 */
trace_buffered_event_disable(void)2605 void trace_buffered_event_disable(void)
2606 {
2607 int cpu;
2608
2609 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2610
2611 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2612 return;
2613
2614 if (--trace_buffered_event_ref)
2615 return;
2616
2617 /* For each CPU, set the buffer as used. */
2618 on_each_cpu_mask(tracing_buffer_mask, disable_trace_buffered_event,
2619 NULL, true);
2620
2621 /* Wait for all current users to finish */
2622 synchronize_rcu();
2623
2624 for_each_tracing_cpu(cpu) {
2625 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2626 per_cpu(trace_buffered_event, cpu) = NULL;
2627 }
2628
2629 /*
2630 * Wait for all CPUs that potentially started checking if they can use
2631 * their event buffer only after the previous synchronize_rcu() call and
2632 * they still read a valid pointer from trace_buffered_event. It must be
2633 * ensured they don't see cleared trace_buffered_event_cnt else they
2634 * could wrongly decide to use the pointed-to buffer which is now freed.
2635 */
2636 synchronize_rcu();
2637
2638 /* For each CPU, relinquish the buffer */
2639 on_each_cpu_mask(tracing_buffer_mask, enable_trace_buffered_event, NULL,
2640 true);
2641 }
2642
2643 static struct trace_buffer *temp_buffer;
2644
2645 struct ring_buffer_event *
trace_event_buffer_lock_reserve(struct trace_buffer ** current_rb,struct trace_event_file * trace_file,int type,unsigned long len,unsigned int trace_ctx)2646 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2647 struct trace_event_file *trace_file,
2648 int type, unsigned long len,
2649 unsigned int trace_ctx)
2650 {
2651 struct ring_buffer_event *entry;
2652 struct trace_array *tr = trace_file->tr;
2653 int val;
2654
2655 *current_rb = tr->array_buffer.buffer;
2656
2657 if (!tr->no_filter_buffering_ref &&
2658 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2659 preempt_disable_notrace();
2660 /*
2661 * Filtering is on, so try to use the per cpu buffer first.
2662 * This buffer will simulate a ring_buffer_event,
2663 * where the type_len is zero and the array[0] will
2664 * hold the full length.
2665 * (see include/linux/ring-buffer.h for details on
2666 * how the ring_buffer_event is structured).
2667 *
2668 * Using a temp buffer during filtering and copying it
2669 * on a matched filter is quicker than writing directly
2670 * into the ring buffer and then discarding it when
2671 * it doesn't match. That is because the discard
2672 * requires several atomic operations to get right.
2673 * Copying on match and doing nothing on a failed match
2674 * is still quicker than no copy on match, but having
2675 * to discard out of the ring buffer on a failed match.
2676 */
2677 if ((entry = __this_cpu_read(trace_buffered_event))) {
2678 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2679
2680 val = this_cpu_inc_return(trace_buffered_event_cnt);
2681
2682 /*
2683 * Preemption is disabled, but interrupts and NMIs
2684 * can still come in now. If that happens after
2685 * the above increment, then it will have to go
2686 * back to the old method of allocating the event
2687 * on the ring buffer, and if the filter fails, it
2688 * will have to call ring_buffer_discard_commit()
2689 * to remove it.
2690 *
2691 * Need to also check the unlikely case that the
2692 * length is bigger than the temp buffer size.
2693 * If that happens, then the reserve is pretty much
2694 * guaranteed to fail, as the ring buffer currently
2695 * only allows events less than a page. But that may
2696 * change in the future, so let the ring buffer reserve
2697 * handle the failure in that case.
2698 */
2699 if (val == 1 && likely(len <= max_len)) {
2700 trace_event_setup(entry, type, trace_ctx);
2701 entry->array[0] = len;
2702 /* Return with preemption disabled */
2703 return entry;
2704 }
2705 this_cpu_dec(trace_buffered_event_cnt);
2706 }
2707 /* __trace_buffer_lock_reserve() disables preemption */
2708 preempt_enable_notrace();
2709 }
2710
2711 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2712 trace_ctx);
2713 /*
2714 * If tracing is off, but we have triggers enabled
2715 * we still need to look at the event data. Use the temp_buffer
2716 * to store the trace event for the trigger to use. It's recursive
2717 * safe and will not be recorded anywhere.
2718 */
2719 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2720 *current_rb = temp_buffer;
2721 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2722 trace_ctx);
2723 }
2724 return entry;
2725 }
2726 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2727
2728 static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2729 static DEFINE_MUTEX(tracepoint_printk_mutex);
2730
output_printk(struct trace_event_buffer * fbuffer)2731 static void output_printk(struct trace_event_buffer *fbuffer)
2732 {
2733 struct trace_event_call *event_call;
2734 struct trace_event_file *file;
2735 struct trace_event *event;
2736 unsigned long flags;
2737 struct trace_iterator *iter = tracepoint_print_iter;
2738
2739 /* We should never get here if iter is NULL */
2740 if (WARN_ON_ONCE(!iter))
2741 return;
2742
2743 event_call = fbuffer->trace_file->event_call;
2744 if (!event_call || !event_call->event.funcs ||
2745 !event_call->event.funcs->trace)
2746 return;
2747
2748 file = fbuffer->trace_file;
2749 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2750 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2751 !filter_match_preds(file->filter, fbuffer->entry)))
2752 return;
2753
2754 event = &fbuffer->trace_file->event_call->event;
2755
2756 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2757 trace_seq_init(&iter->seq);
2758 iter->ent = fbuffer->entry;
2759 event_call->event.funcs->trace(iter, 0, event);
2760 trace_seq_putc(&iter->seq, 0);
2761 printk("%s", iter->seq.buffer);
2762
2763 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2764 }
2765
tracepoint_printk_sysctl(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)2766 int tracepoint_printk_sysctl(const struct ctl_table *table, int write,
2767 void *buffer, size_t *lenp,
2768 loff_t *ppos)
2769 {
2770 int save_tracepoint_printk;
2771 int ret;
2772
2773 mutex_lock(&tracepoint_printk_mutex);
2774 save_tracepoint_printk = tracepoint_printk;
2775
2776 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2777
2778 /*
2779 * This will force exiting early, as tracepoint_printk
2780 * is always zero when tracepoint_printk_iter is not allocated
2781 */
2782 if (!tracepoint_print_iter)
2783 tracepoint_printk = 0;
2784
2785 if (save_tracepoint_printk == tracepoint_printk)
2786 goto out;
2787
2788 if (tracepoint_printk)
2789 static_key_enable(&tracepoint_printk_key.key);
2790 else
2791 static_key_disable(&tracepoint_printk_key.key);
2792
2793 out:
2794 mutex_unlock(&tracepoint_printk_mutex);
2795
2796 return ret;
2797 }
2798
trace_event_buffer_commit(struct trace_event_buffer * fbuffer)2799 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2800 {
2801 enum event_trigger_type tt = ETT_NONE;
2802 struct trace_event_file *file = fbuffer->trace_file;
2803
2804 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2805 fbuffer->entry, &tt))
2806 goto discard;
2807
2808 if (static_key_false(&tracepoint_printk_key.key))
2809 output_printk(fbuffer);
2810
2811 if (static_branch_unlikely(&trace_event_exports_enabled))
2812 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2813
2814 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2815 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
2816
2817 discard:
2818 if (tt)
2819 event_triggers_post_call(file, tt);
2820
2821 }
2822 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2823
2824 /*
2825 * Skip 3:
2826 *
2827 * trace_buffer_unlock_commit_regs()
2828 * trace_event_buffer_commit()
2829 * trace_event_raw_event_xxx()
2830 */
2831 # define STACK_SKIP 3
2832
trace_buffer_unlock_commit_regs(struct trace_array * tr,struct trace_buffer * buffer,struct ring_buffer_event * event,unsigned int trace_ctx,struct pt_regs * regs)2833 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2834 struct trace_buffer *buffer,
2835 struct ring_buffer_event *event,
2836 unsigned int trace_ctx,
2837 struct pt_regs *regs)
2838 {
2839 __buffer_unlock_commit(buffer, event);
2840
2841 /*
2842 * If regs is not set, then skip the necessary functions.
2843 * Note, we can still get here via blktrace, wakeup tracer
2844 * and mmiotrace, but that's ok if they lose a function or
2845 * two. They are not that meaningful.
2846 */
2847 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
2848 ftrace_trace_userstack(tr, buffer, trace_ctx);
2849 }
2850
2851 /*
2852 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2853 */
2854 void
trace_buffer_unlock_commit_nostack(struct trace_buffer * buffer,struct ring_buffer_event * event)2855 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2856 struct ring_buffer_event *event)
2857 {
2858 __buffer_unlock_commit(buffer, event);
2859 }
2860
2861 void
trace_function(struct trace_array * tr,unsigned long ip,unsigned long parent_ip,unsigned int trace_ctx)2862 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
2863 parent_ip, unsigned int trace_ctx)
2864 {
2865 struct trace_event_call *call = &event_function;
2866 struct trace_buffer *buffer = tr->array_buffer.buffer;
2867 struct ring_buffer_event *event;
2868 struct ftrace_entry *entry;
2869
2870 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2871 trace_ctx);
2872 if (!event)
2873 return;
2874 entry = ring_buffer_event_data(event);
2875 entry->ip = ip;
2876 entry->parent_ip = parent_ip;
2877
2878 if (!call_filter_check_discard(call, entry, buffer, event)) {
2879 if (static_branch_unlikely(&trace_function_exports_enabled))
2880 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
2881 __buffer_unlock_commit(buffer, event);
2882 }
2883 }
2884
2885 #ifdef CONFIG_STACKTRACE
2886
2887 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2888 #define FTRACE_KSTACK_NESTING 4
2889
2890 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
2891
2892 struct ftrace_stack {
2893 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2894 };
2895
2896
2897 struct ftrace_stacks {
2898 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
2899 };
2900
2901 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
2902 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2903
__ftrace_trace_stack(struct trace_buffer * buffer,unsigned int trace_ctx,int skip,struct pt_regs * regs)2904 static void __ftrace_trace_stack(struct trace_buffer *buffer,
2905 unsigned int trace_ctx,
2906 int skip, struct pt_regs *regs)
2907 {
2908 struct trace_event_call *call = &event_kernel_stack;
2909 struct ring_buffer_event *event;
2910 unsigned int size, nr_entries;
2911 struct ftrace_stack *fstack;
2912 struct stack_entry *entry;
2913 int stackidx;
2914
2915 /*
2916 * Add one, for this function and the call to save_stack_trace()
2917 * If regs is set, then these functions will not be in the way.
2918 */
2919 #ifndef CONFIG_UNWINDER_ORC
2920 if (!regs)
2921 skip++;
2922 #endif
2923
2924 preempt_disable_notrace();
2925
2926 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
2927
2928 /* This should never happen. If it does, yell once and skip */
2929 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
2930 goto out;
2931
2932 /*
2933 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
2934 * interrupt will either see the value pre increment or post
2935 * increment. If the interrupt happens pre increment it will have
2936 * restored the counter when it returns. We just need a barrier to
2937 * keep gcc from moving things around.
2938 */
2939 barrier();
2940
2941 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
2942 size = ARRAY_SIZE(fstack->calls);
2943
2944 if (regs) {
2945 nr_entries = stack_trace_save_regs(regs, fstack->calls,
2946 size, skip);
2947 } else {
2948 nr_entries = stack_trace_save(fstack->calls, size, skip);
2949 }
2950
2951 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2952 struct_size(entry, caller, nr_entries),
2953 trace_ctx);
2954 if (!event)
2955 goto out;
2956 entry = ring_buffer_event_data(event);
2957
2958 entry->size = nr_entries;
2959 memcpy(&entry->caller, fstack->calls,
2960 flex_array_size(entry, caller, nr_entries));
2961
2962 if (!call_filter_check_discard(call, entry, buffer, event))
2963 __buffer_unlock_commit(buffer, event);
2964
2965 out:
2966 /* Again, don't let gcc optimize things here */
2967 barrier();
2968 __this_cpu_dec(ftrace_stack_reserve);
2969 preempt_enable_notrace();
2970
2971 }
2972
ftrace_trace_stack(struct trace_array * tr,struct trace_buffer * buffer,unsigned int trace_ctx,int skip,struct pt_regs * regs)2973 static inline void ftrace_trace_stack(struct trace_array *tr,
2974 struct trace_buffer *buffer,
2975 unsigned int trace_ctx,
2976 int skip, struct pt_regs *regs)
2977 {
2978 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
2979 return;
2980
2981 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
2982 }
2983
__trace_stack(struct trace_array * tr,unsigned int trace_ctx,int skip)2984 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
2985 int skip)
2986 {
2987 struct trace_buffer *buffer = tr->array_buffer.buffer;
2988
2989 if (rcu_is_watching()) {
2990 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
2991 return;
2992 }
2993
2994 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
2995 return;
2996
2997 /*
2998 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
2999 * but if the above rcu_is_watching() failed, then the NMI
3000 * triggered someplace critical, and ct_irq_enter() should
3001 * not be called from NMI.
3002 */
3003 if (unlikely(in_nmi()))
3004 return;
3005
3006 ct_irq_enter_irqson();
3007 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3008 ct_irq_exit_irqson();
3009 }
3010
3011 /**
3012 * trace_dump_stack - record a stack back trace in the trace buffer
3013 * @skip: Number of functions to skip (helper handlers)
3014 */
trace_dump_stack(int skip)3015 void trace_dump_stack(int skip)
3016 {
3017 if (tracing_disabled || tracing_selftest_running)
3018 return;
3019
3020 #ifndef CONFIG_UNWINDER_ORC
3021 /* Skip 1 to skip this function. */
3022 skip++;
3023 #endif
3024 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3025 tracing_gen_ctx(), skip, NULL);
3026 }
3027 EXPORT_SYMBOL_GPL(trace_dump_stack);
3028
3029 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3030 static DEFINE_PER_CPU(int, user_stack_count);
3031
3032 static void
ftrace_trace_userstack(struct trace_array * tr,struct trace_buffer * buffer,unsigned int trace_ctx)3033 ftrace_trace_userstack(struct trace_array *tr,
3034 struct trace_buffer *buffer, unsigned int trace_ctx)
3035 {
3036 struct trace_event_call *call = &event_user_stack;
3037 struct ring_buffer_event *event;
3038 struct userstack_entry *entry;
3039
3040 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3041 return;
3042
3043 /*
3044 * NMIs can not handle page faults, even with fix ups.
3045 * The save user stack can (and often does) fault.
3046 */
3047 if (unlikely(in_nmi()))
3048 return;
3049
3050 /*
3051 * prevent recursion, since the user stack tracing may
3052 * trigger other kernel events.
3053 */
3054 preempt_disable();
3055 if (__this_cpu_read(user_stack_count))
3056 goto out;
3057
3058 __this_cpu_inc(user_stack_count);
3059
3060 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3061 sizeof(*entry), trace_ctx);
3062 if (!event)
3063 goto out_drop_count;
3064 entry = ring_buffer_event_data(event);
3065
3066 entry->tgid = current->tgid;
3067 memset(&entry->caller, 0, sizeof(entry->caller));
3068
3069 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3070 if (!call_filter_check_discard(call, entry, buffer, event))
3071 __buffer_unlock_commit(buffer, event);
3072
3073 out_drop_count:
3074 __this_cpu_dec(user_stack_count);
3075 out:
3076 preempt_enable();
3077 }
3078 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
ftrace_trace_userstack(struct trace_array * tr,struct trace_buffer * buffer,unsigned int trace_ctx)3079 static void ftrace_trace_userstack(struct trace_array *tr,
3080 struct trace_buffer *buffer,
3081 unsigned int trace_ctx)
3082 {
3083 }
3084 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3085
3086 #endif /* CONFIG_STACKTRACE */
3087
3088 static inline void
func_repeats_set_delta_ts(struct func_repeats_entry * entry,unsigned long long delta)3089 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3090 unsigned long long delta)
3091 {
3092 entry->bottom_delta_ts = delta & U32_MAX;
3093 entry->top_delta_ts = (delta >> 32);
3094 }
3095
trace_last_func_repeats(struct trace_array * tr,struct trace_func_repeats * last_info,unsigned int trace_ctx)3096 void trace_last_func_repeats(struct trace_array *tr,
3097 struct trace_func_repeats *last_info,
3098 unsigned int trace_ctx)
3099 {
3100 struct trace_buffer *buffer = tr->array_buffer.buffer;
3101 struct func_repeats_entry *entry;
3102 struct ring_buffer_event *event;
3103 u64 delta;
3104
3105 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3106 sizeof(*entry), trace_ctx);
3107 if (!event)
3108 return;
3109
3110 delta = ring_buffer_event_time_stamp(buffer, event) -
3111 last_info->ts_last_call;
3112
3113 entry = ring_buffer_event_data(event);
3114 entry->ip = last_info->ip;
3115 entry->parent_ip = last_info->parent_ip;
3116 entry->count = last_info->count;
3117 func_repeats_set_delta_ts(entry, delta);
3118
3119 __buffer_unlock_commit(buffer, event);
3120 }
3121
3122 /* created for use with alloc_percpu */
3123 struct trace_buffer_struct {
3124 int nesting;
3125 char buffer[4][TRACE_BUF_SIZE];
3126 };
3127
3128 static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3129
3130 /*
3131 * This allows for lockless recording. If we're nested too deeply, then
3132 * this returns NULL.
3133 */
get_trace_buf(void)3134 static char *get_trace_buf(void)
3135 {
3136 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3137
3138 if (!trace_percpu_buffer || buffer->nesting >= 4)
3139 return NULL;
3140
3141 buffer->nesting++;
3142
3143 /* Interrupts must see nesting incremented before we use the buffer */
3144 barrier();
3145 return &buffer->buffer[buffer->nesting - 1][0];
3146 }
3147
put_trace_buf(void)3148 static void put_trace_buf(void)
3149 {
3150 /* Don't let the decrement of nesting leak before this */
3151 barrier();
3152 this_cpu_dec(trace_percpu_buffer->nesting);
3153 }
3154
alloc_percpu_trace_buffer(void)3155 static int alloc_percpu_trace_buffer(void)
3156 {
3157 struct trace_buffer_struct __percpu *buffers;
3158
3159 if (trace_percpu_buffer)
3160 return 0;
3161
3162 buffers = alloc_percpu(struct trace_buffer_struct);
3163 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3164 return -ENOMEM;
3165
3166 trace_percpu_buffer = buffers;
3167 return 0;
3168 }
3169
3170 static int buffers_allocated;
3171
trace_printk_init_buffers(void)3172 void trace_printk_init_buffers(void)
3173 {
3174 if (buffers_allocated)
3175 return;
3176
3177 if (alloc_percpu_trace_buffer())
3178 return;
3179
3180 /* trace_printk() is for debug use only. Don't use it in production. */
3181
3182 pr_warn("\n");
3183 pr_warn("**********************************************************\n");
3184 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3185 pr_warn("** **\n");
3186 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3187 pr_warn("** **\n");
3188 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3189 pr_warn("** unsafe for production use. **\n");
3190 pr_warn("** **\n");
3191 pr_warn("** If you see this message and you are not debugging **\n");
3192 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3193 pr_warn("** **\n");
3194 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3195 pr_warn("**********************************************************\n");
3196
3197 /* Expand the buffers to set size */
3198 tracing_update_buffers(&global_trace);
3199
3200 buffers_allocated = 1;
3201
3202 /*
3203 * trace_printk_init_buffers() can be called by modules.
3204 * If that happens, then we need to start cmdline recording
3205 * directly here. If the global_trace.buffer is already
3206 * allocated here, then this was called by module code.
3207 */
3208 if (global_trace.array_buffer.buffer)
3209 tracing_start_cmdline_record();
3210 }
3211 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3212
trace_printk_start_comm(void)3213 void trace_printk_start_comm(void)
3214 {
3215 /* Start tracing comms if trace printk is set */
3216 if (!buffers_allocated)
3217 return;
3218 tracing_start_cmdline_record();
3219 }
3220
trace_printk_start_stop_comm(int enabled)3221 static void trace_printk_start_stop_comm(int enabled)
3222 {
3223 if (!buffers_allocated)
3224 return;
3225
3226 if (enabled)
3227 tracing_start_cmdline_record();
3228 else
3229 tracing_stop_cmdline_record();
3230 }
3231
3232 /**
3233 * trace_vbprintk - write binary msg to tracing buffer
3234 * @ip: The address of the caller
3235 * @fmt: The string format to write to the buffer
3236 * @args: Arguments for @fmt
3237 */
trace_vbprintk(unsigned long ip,const char * fmt,va_list args)3238 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3239 {
3240 struct trace_event_call *call = &event_bprint;
3241 struct ring_buffer_event *event;
3242 struct trace_buffer *buffer;
3243 struct trace_array *tr = &global_trace;
3244 struct bprint_entry *entry;
3245 unsigned int trace_ctx;
3246 char *tbuffer;
3247 int len = 0, size;
3248
3249 if (unlikely(tracing_selftest_running || tracing_disabled))
3250 return 0;
3251
3252 /* Don't pollute graph traces with trace_vprintk internals */
3253 pause_graph_tracing();
3254
3255 trace_ctx = tracing_gen_ctx();
3256 preempt_disable_notrace();
3257
3258 tbuffer = get_trace_buf();
3259 if (!tbuffer) {
3260 len = 0;
3261 goto out_nobuffer;
3262 }
3263
3264 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3265
3266 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3267 goto out_put;
3268
3269 size = sizeof(*entry) + sizeof(u32) * len;
3270 buffer = tr->array_buffer.buffer;
3271 ring_buffer_nest_start(buffer);
3272 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3273 trace_ctx);
3274 if (!event)
3275 goto out;
3276 entry = ring_buffer_event_data(event);
3277 entry->ip = ip;
3278 entry->fmt = fmt;
3279
3280 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3281 if (!call_filter_check_discard(call, entry, buffer, event)) {
3282 __buffer_unlock_commit(buffer, event);
3283 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3284 }
3285
3286 out:
3287 ring_buffer_nest_end(buffer);
3288 out_put:
3289 put_trace_buf();
3290
3291 out_nobuffer:
3292 preempt_enable_notrace();
3293 unpause_graph_tracing();
3294
3295 return len;
3296 }
3297 EXPORT_SYMBOL_GPL(trace_vbprintk);
3298
3299 __printf(3, 0)
3300 static int
__trace_array_vprintk(struct trace_buffer * buffer,unsigned long ip,const char * fmt,va_list args)3301 __trace_array_vprintk(struct trace_buffer *buffer,
3302 unsigned long ip, const char *fmt, va_list args)
3303 {
3304 struct trace_event_call *call = &event_print;
3305 struct ring_buffer_event *event;
3306 int len = 0, size;
3307 struct print_entry *entry;
3308 unsigned int trace_ctx;
3309 char *tbuffer;
3310
3311 if (tracing_disabled)
3312 return 0;
3313
3314 /* Don't pollute graph traces with trace_vprintk internals */
3315 pause_graph_tracing();
3316
3317 trace_ctx = tracing_gen_ctx();
3318 preempt_disable_notrace();
3319
3320
3321 tbuffer = get_trace_buf();
3322 if (!tbuffer) {
3323 len = 0;
3324 goto out_nobuffer;
3325 }
3326
3327 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3328
3329 size = sizeof(*entry) + len + 1;
3330 ring_buffer_nest_start(buffer);
3331 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3332 trace_ctx);
3333 if (!event)
3334 goto out;
3335 entry = ring_buffer_event_data(event);
3336 entry->ip = ip;
3337
3338 memcpy(&entry->buf, tbuffer, len + 1);
3339 if (!call_filter_check_discard(call, entry, buffer, event)) {
3340 __buffer_unlock_commit(buffer, event);
3341 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3342 }
3343
3344 out:
3345 ring_buffer_nest_end(buffer);
3346 put_trace_buf();
3347
3348 out_nobuffer:
3349 preempt_enable_notrace();
3350 unpause_graph_tracing();
3351
3352 return len;
3353 }
3354
3355 __printf(3, 0)
trace_array_vprintk(struct trace_array * tr,unsigned long ip,const char * fmt,va_list args)3356 int trace_array_vprintk(struct trace_array *tr,
3357 unsigned long ip, const char *fmt, va_list args)
3358 {
3359 if (tracing_selftest_running && tr == &global_trace)
3360 return 0;
3361
3362 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3363 }
3364
3365 /**
3366 * trace_array_printk - Print a message to a specific instance
3367 * @tr: The instance trace_array descriptor
3368 * @ip: The instruction pointer that this is called from.
3369 * @fmt: The format to print (printf format)
3370 *
3371 * If a subsystem sets up its own instance, they have the right to
3372 * printk strings into their tracing instance buffer using this
3373 * function. Note, this function will not write into the top level
3374 * buffer (use trace_printk() for that), as writing into the top level
3375 * buffer should only have events that can be individually disabled.
3376 * trace_printk() is only used for debugging a kernel, and should not
3377 * be ever incorporated in normal use.
3378 *
3379 * trace_array_printk() can be used, as it will not add noise to the
3380 * top level tracing buffer.
3381 *
3382 * Note, trace_array_init_printk() must be called on @tr before this
3383 * can be used.
3384 */
3385 __printf(3, 0)
trace_array_printk(struct trace_array * tr,unsigned long ip,const char * fmt,...)3386 int trace_array_printk(struct trace_array *tr,
3387 unsigned long ip, const char *fmt, ...)
3388 {
3389 int ret;
3390 va_list ap;
3391
3392 if (!tr)
3393 return -ENOENT;
3394
3395 /* This is only allowed for created instances */
3396 if (tr == &global_trace)
3397 return 0;
3398
3399 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3400 return 0;
3401
3402 va_start(ap, fmt);
3403 ret = trace_array_vprintk(tr, ip, fmt, ap);
3404 va_end(ap);
3405 return ret;
3406 }
3407 EXPORT_SYMBOL_GPL(trace_array_printk);
3408
3409 /**
3410 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3411 * @tr: The trace array to initialize the buffers for
3412 *
3413 * As trace_array_printk() only writes into instances, they are OK to
3414 * have in the kernel (unlike trace_printk()). This needs to be called
3415 * before trace_array_printk() can be used on a trace_array.
3416 */
trace_array_init_printk(struct trace_array * tr)3417 int trace_array_init_printk(struct trace_array *tr)
3418 {
3419 if (!tr)
3420 return -ENOENT;
3421
3422 /* This is only allowed for created instances */
3423 if (tr == &global_trace)
3424 return -EINVAL;
3425
3426 return alloc_percpu_trace_buffer();
3427 }
3428 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3429
3430 __printf(3, 4)
trace_array_printk_buf(struct trace_buffer * buffer,unsigned long ip,const char * fmt,...)3431 int trace_array_printk_buf(struct trace_buffer *buffer,
3432 unsigned long ip, const char *fmt, ...)
3433 {
3434 int ret;
3435 va_list ap;
3436
3437 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3438 return 0;
3439
3440 va_start(ap, fmt);
3441 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3442 va_end(ap);
3443 return ret;
3444 }
3445
3446 __printf(2, 0)
trace_vprintk(unsigned long ip,const char * fmt,va_list args)3447 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3448 {
3449 return trace_array_vprintk(&global_trace, ip, fmt, args);
3450 }
3451 EXPORT_SYMBOL_GPL(trace_vprintk);
3452
trace_iterator_increment(struct trace_iterator * iter)3453 static void trace_iterator_increment(struct trace_iterator *iter)
3454 {
3455 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3456
3457 iter->idx++;
3458 if (buf_iter)
3459 ring_buffer_iter_advance(buf_iter);
3460 }
3461
3462 static struct trace_entry *
peek_next_entry(struct trace_iterator * iter,int cpu,u64 * ts,unsigned long * lost_events)3463 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3464 unsigned long *lost_events)
3465 {
3466 struct ring_buffer_event *event;
3467 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3468
3469 if (buf_iter) {
3470 event = ring_buffer_iter_peek(buf_iter, ts);
3471 if (lost_events)
3472 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3473 (unsigned long)-1 : 0;
3474 } else {
3475 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3476 lost_events);
3477 }
3478
3479 if (event) {
3480 iter->ent_size = ring_buffer_event_length(event);
3481 return ring_buffer_event_data(event);
3482 }
3483 iter->ent_size = 0;
3484 return NULL;
3485 }
3486
3487 static struct trace_entry *
__find_next_entry(struct trace_iterator * iter,int * ent_cpu,unsigned long * missing_events,u64 * ent_ts)3488 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3489 unsigned long *missing_events, u64 *ent_ts)
3490 {
3491 struct trace_buffer *buffer = iter->array_buffer->buffer;
3492 struct trace_entry *ent, *next = NULL;
3493 unsigned long lost_events = 0, next_lost = 0;
3494 int cpu_file = iter->cpu_file;
3495 u64 next_ts = 0, ts;
3496 int next_cpu = -1;
3497 int next_size = 0;
3498 int cpu;
3499
3500 /*
3501 * If we are in a per_cpu trace file, don't bother by iterating over
3502 * all cpu and peek directly.
3503 */
3504 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3505 if (ring_buffer_empty_cpu(buffer, cpu_file))
3506 return NULL;
3507 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3508 if (ent_cpu)
3509 *ent_cpu = cpu_file;
3510
3511 return ent;
3512 }
3513
3514 for_each_tracing_cpu(cpu) {
3515
3516 if (ring_buffer_empty_cpu(buffer, cpu))
3517 continue;
3518
3519 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3520
3521 /*
3522 * Pick the entry with the smallest timestamp:
3523 */
3524 if (ent && (!next || ts < next_ts)) {
3525 next = ent;
3526 next_cpu = cpu;
3527 next_ts = ts;
3528 next_lost = lost_events;
3529 next_size = iter->ent_size;
3530 }
3531 }
3532
3533 iter->ent_size = next_size;
3534
3535 if (ent_cpu)
3536 *ent_cpu = next_cpu;
3537
3538 if (ent_ts)
3539 *ent_ts = next_ts;
3540
3541 if (missing_events)
3542 *missing_events = next_lost;
3543
3544 return next;
3545 }
3546
3547 #define STATIC_FMT_BUF_SIZE 128
3548 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3549
trace_iter_expand_format(struct trace_iterator * iter)3550 char *trace_iter_expand_format(struct trace_iterator *iter)
3551 {
3552 char *tmp;
3553
3554 /*
3555 * iter->tr is NULL when used with tp_printk, which makes
3556 * this get called where it is not safe to call krealloc().
3557 */
3558 if (!iter->tr || iter->fmt == static_fmt_buf)
3559 return NULL;
3560
3561 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3562 GFP_KERNEL);
3563 if (tmp) {
3564 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3565 iter->fmt = tmp;
3566 }
3567
3568 return tmp;
3569 }
3570
3571 /* Returns true if the string is safe to dereference from an event */
trace_safe_str(struct trace_iterator * iter,const char * str,bool star,int len)3572 static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3573 bool star, int len)
3574 {
3575 unsigned long addr = (unsigned long)str;
3576 struct trace_event *trace_event;
3577 struct trace_event_call *event;
3578
3579 /* Ignore strings with no length */
3580 if (star && !len)
3581 return true;
3582
3583 /* OK if part of the event data */
3584 if ((addr >= (unsigned long)iter->ent) &&
3585 (addr < (unsigned long)iter->ent + iter->ent_size))
3586 return true;
3587
3588 /* OK if part of the temp seq buffer */
3589 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3590 (addr < (unsigned long)iter->tmp_seq.buffer + TRACE_SEQ_BUFFER_SIZE))
3591 return true;
3592
3593 /* Core rodata can not be freed */
3594 if (is_kernel_rodata(addr))
3595 return true;
3596
3597 if (trace_is_tracepoint_string(str))
3598 return true;
3599
3600 /*
3601 * Now this could be a module event, referencing core module
3602 * data, which is OK.
3603 */
3604 if (!iter->ent)
3605 return false;
3606
3607 trace_event = ftrace_find_event(iter->ent->type);
3608 if (!trace_event)
3609 return false;
3610
3611 event = container_of(trace_event, struct trace_event_call, event);
3612 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3613 return false;
3614
3615 /* Would rather have rodata, but this will suffice */
3616 if (within_module_core(addr, event->module))
3617 return true;
3618
3619 return false;
3620 }
3621
3622 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3623
test_can_verify_check(const char * fmt,...)3624 static int test_can_verify_check(const char *fmt, ...)
3625 {
3626 char buf[16];
3627 va_list ap;
3628 int ret;
3629
3630 /*
3631 * The verifier is dependent on vsnprintf() modifies the va_list
3632 * passed to it, where it is sent as a reference. Some architectures
3633 * (like x86_32) passes it by value, which means that vsnprintf()
3634 * does not modify the va_list passed to it, and the verifier
3635 * would then need to be able to understand all the values that
3636 * vsnprintf can use. If it is passed by value, then the verifier
3637 * is disabled.
3638 */
3639 va_start(ap, fmt);
3640 vsnprintf(buf, 16, "%d", ap);
3641 ret = va_arg(ap, int);
3642 va_end(ap);
3643
3644 return ret;
3645 }
3646
test_can_verify(void)3647 static void test_can_verify(void)
3648 {
3649 if (!test_can_verify_check("%d %d", 0, 1)) {
3650 pr_info("trace event string verifier disabled\n");
3651 static_branch_inc(&trace_no_verify);
3652 }
3653 }
3654
3655 /**
3656 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3657 * @iter: The iterator that holds the seq buffer and the event being printed
3658 * @fmt: The format used to print the event
3659 * @ap: The va_list holding the data to print from @fmt.
3660 *
3661 * This writes the data into the @iter->seq buffer using the data from
3662 * @fmt and @ap. If the format has a %s, then the source of the string
3663 * is examined to make sure it is safe to print, otherwise it will
3664 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3665 * pointer.
3666 */
trace_check_vprintf(struct trace_iterator * iter,const char * fmt,va_list ap)3667 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3668 va_list ap)
3669 {
3670 const char *p = fmt;
3671 const char *str;
3672 int i, j;
3673
3674 if (WARN_ON_ONCE(!fmt))
3675 return;
3676
3677 if (static_branch_unlikely(&trace_no_verify))
3678 goto print;
3679
3680 /* Don't bother checking when doing a ftrace_dump() */
3681 if (iter->fmt == static_fmt_buf)
3682 goto print;
3683
3684 while (*p) {
3685 bool star = false;
3686 int len = 0;
3687
3688 j = 0;
3689
3690 /* We only care about %s and variants */
3691 for (i = 0; p[i]; i++) {
3692 if (i + 1 >= iter->fmt_size) {
3693 /*
3694 * If we can't expand the copy buffer,
3695 * just print it.
3696 */
3697 if (!trace_iter_expand_format(iter))
3698 goto print;
3699 }
3700
3701 if (p[i] == '\\' && p[i+1]) {
3702 i++;
3703 continue;
3704 }
3705 if (p[i] == '%') {
3706 /* Need to test cases like %08.*s */
3707 for (j = 1; p[i+j]; j++) {
3708 if (isdigit(p[i+j]) ||
3709 p[i+j] == '.')
3710 continue;
3711 if (p[i+j] == '*') {
3712 star = true;
3713 continue;
3714 }
3715 break;
3716 }
3717 if (p[i+j] == 's')
3718 break;
3719 star = false;
3720 }
3721 j = 0;
3722 }
3723 /* If no %s found then just print normally */
3724 if (!p[i])
3725 break;
3726
3727 /* Copy up to the %s, and print that */
3728 strncpy(iter->fmt, p, i);
3729 iter->fmt[i] = '\0';
3730 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3731
3732 /*
3733 * If iter->seq is full, the above call no longer guarantees
3734 * that ap is in sync with fmt processing, and further calls
3735 * to va_arg() can return wrong positional arguments.
3736 *
3737 * Ensure that ap is no longer used in this case.
3738 */
3739 if (iter->seq.full) {
3740 p = "";
3741 break;
3742 }
3743
3744 if (star)
3745 len = va_arg(ap, int);
3746
3747 /* The ap now points to the string data of the %s */
3748 str = va_arg(ap, const char *);
3749
3750 /*
3751 * If you hit this warning, it is likely that the
3752 * trace event in question used %s on a string that
3753 * was saved at the time of the event, but may not be
3754 * around when the trace is read. Use __string(),
3755 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3756 * instead. See samples/trace_events/trace-events-sample.h
3757 * for reference.
3758 */
3759 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3760 "fmt: '%s' current_buffer: '%s'",
3761 fmt, seq_buf_str(&iter->seq.seq))) {
3762 int ret;
3763
3764 /* Try to safely read the string */
3765 if (star) {
3766 if (len + 1 > iter->fmt_size)
3767 len = iter->fmt_size - 1;
3768 if (len < 0)
3769 len = 0;
3770 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3771 iter->fmt[len] = 0;
3772 star = false;
3773 } else {
3774 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3775 iter->fmt_size);
3776 }
3777 if (ret < 0)
3778 trace_seq_printf(&iter->seq, "(0x%px)", str);
3779 else
3780 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3781 str, iter->fmt);
3782 str = "[UNSAFE-MEMORY]";
3783 strcpy(iter->fmt, "%s");
3784 } else {
3785 strncpy(iter->fmt, p + i, j + 1);
3786 iter->fmt[j+1] = '\0';
3787 }
3788 if (star)
3789 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3790 else
3791 trace_seq_printf(&iter->seq, iter->fmt, str);
3792
3793 p += i + j + 1;
3794 }
3795 print:
3796 if (*p)
3797 trace_seq_vprintf(&iter->seq, p, ap);
3798 }
3799
trace_event_format(struct trace_iterator * iter,const char * fmt)3800 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3801 {
3802 const char *p, *new_fmt;
3803 char *q;
3804
3805 if (WARN_ON_ONCE(!fmt))
3806 return fmt;
3807
3808 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
3809 return fmt;
3810
3811 p = fmt;
3812 new_fmt = q = iter->fmt;
3813 while (*p) {
3814 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
3815 if (!trace_iter_expand_format(iter))
3816 return fmt;
3817
3818 q += iter->fmt - new_fmt;
3819 new_fmt = iter->fmt;
3820 }
3821
3822 *q++ = *p++;
3823
3824 /* Replace %p with %px */
3825 if (p[-1] == '%') {
3826 if (p[0] == '%') {
3827 *q++ = *p++;
3828 } else if (p[0] == 'p' && !isalnum(p[1])) {
3829 *q++ = *p++;
3830 *q++ = 'x';
3831 }
3832 }
3833 }
3834 *q = '\0';
3835
3836 return new_fmt;
3837 }
3838
3839 #define STATIC_TEMP_BUF_SIZE 128
3840 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3841
3842 /* Find the next real entry, without updating the iterator itself */
trace_find_next_entry(struct trace_iterator * iter,int * ent_cpu,u64 * ent_ts)3843 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3844 int *ent_cpu, u64 *ent_ts)
3845 {
3846 /* __find_next_entry will reset ent_size */
3847 int ent_size = iter->ent_size;
3848 struct trace_entry *entry;
3849
3850 /*
3851 * If called from ftrace_dump(), then the iter->temp buffer
3852 * will be the static_temp_buf and not created from kmalloc.
3853 * If the entry size is greater than the buffer, we can
3854 * not save it. Just return NULL in that case. This is only
3855 * used to add markers when two consecutive events' time
3856 * stamps have a large delta. See trace_print_lat_context()
3857 */
3858 if (iter->temp == static_temp_buf &&
3859 STATIC_TEMP_BUF_SIZE < ent_size)
3860 return NULL;
3861
3862 /*
3863 * The __find_next_entry() may call peek_next_entry(), which may
3864 * call ring_buffer_peek() that may make the contents of iter->ent
3865 * undefined. Need to copy iter->ent now.
3866 */
3867 if (iter->ent && iter->ent != iter->temp) {
3868 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3869 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3870 void *temp;
3871 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3872 if (!temp)
3873 return NULL;
3874 kfree(iter->temp);
3875 iter->temp = temp;
3876 iter->temp_size = iter->ent_size;
3877 }
3878 memcpy(iter->temp, iter->ent, iter->ent_size);
3879 iter->ent = iter->temp;
3880 }
3881 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3882 /* Put back the original ent_size */
3883 iter->ent_size = ent_size;
3884
3885 return entry;
3886 }
3887
3888 /* Find the next real entry, and increment the iterator to the next entry */
trace_find_next_entry_inc(struct trace_iterator * iter)3889 void *trace_find_next_entry_inc(struct trace_iterator *iter)
3890 {
3891 iter->ent = __find_next_entry(iter, &iter->cpu,
3892 &iter->lost_events, &iter->ts);
3893
3894 if (iter->ent)
3895 trace_iterator_increment(iter);
3896
3897 return iter->ent ? iter : NULL;
3898 }
3899
trace_consume(struct trace_iterator * iter)3900 static void trace_consume(struct trace_iterator *iter)
3901 {
3902 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3903 &iter->lost_events);
3904 }
3905
s_next(struct seq_file * m,void * v,loff_t * pos)3906 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3907 {
3908 struct trace_iterator *iter = m->private;
3909 int i = (int)*pos;
3910 void *ent;
3911
3912 WARN_ON_ONCE(iter->leftover);
3913
3914 (*pos)++;
3915
3916 /* can't go backwards */
3917 if (iter->idx > i)
3918 return NULL;
3919
3920 if (iter->idx < 0)
3921 ent = trace_find_next_entry_inc(iter);
3922 else
3923 ent = iter;
3924
3925 while (ent && iter->idx < i)
3926 ent = trace_find_next_entry_inc(iter);
3927
3928 iter->pos = *pos;
3929
3930 return ent;
3931 }
3932
tracing_iter_reset(struct trace_iterator * iter,int cpu)3933 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3934 {
3935 struct ring_buffer_iter *buf_iter;
3936 unsigned long entries = 0;
3937 u64 ts;
3938
3939 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
3940
3941 buf_iter = trace_buffer_iter(iter, cpu);
3942 if (!buf_iter)
3943 return;
3944
3945 ring_buffer_iter_reset(buf_iter);
3946
3947 /*
3948 * We could have the case with the max latency tracers
3949 * that a reset never took place on a cpu. This is evident
3950 * by the timestamp being before the start of the buffer.
3951 */
3952 while (ring_buffer_iter_peek(buf_iter, &ts)) {
3953 if (ts >= iter->array_buffer->time_start)
3954 break;
3955 entries++;
3956 ring_buffer_iter_advance(buf_iter);
3957 /* This could be a big loop */
3958 cond_resched();
3959 }
3960
3961 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
3962 }
3963
3964 /*
3965 * The current tracer is copied to avoid a global locking
3966 * all around.
3967 */
s_start(struct seq_file * m,loff_t * pos)3968 static void *s_start(struct seq_file *m, loff_t *pos)
3969 {
3970 struct trace_iterator *iter = m->private;
3971 struct trace_array *tr = iter->tr;
3972 int cpu_file = iter->cpu_file;
3973 void *p = NULL;
3974 loff_t l = 0;
3975 int cpu;
3976
3977 mutex_lock(&trace_types_lock);
3978 if (unlikely(tr->current_trace != iter->trace)) {
3979 /* Close iter->trace before switching to the new current tracer */
3980 if (iter->trace->close)
3981 iter->trace->close(iter);
3982 iter->trace = tr->current_trace;
3983 /* Reopen the new current tracer */
3984 if (iter->trace->open)
3985 iter->trace->open(iter);
3986 }
3987 mutex_unlock(&trace_types_lock);
3988
3989 #ifdef CONFIG_TRACER_MAX_TRACE
3990 if (iter->snapshot && iter->trace->use_max_tr)
3991 return ERR_PTR(-EBUSY);
3992 #endif
3993
3994 if (*pos != iter->pos) {
3995 iter->ent = NULL;
3996 iter->cpu = 0;
3997 iter->idx = -1;
3998
3999 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4000 for_each_tracing_cpu(cpu)
4001 tracing_iter_reset(iter, cpu);
4002 } else
4003 tracing_iter_reset(iter, cpu_file);
4004
4005 iter->leftover = 0;
4006 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4007 ;
4008
4009 } else {
4010 /*
4011 * If we overflowed the seq_file before, then we want
4012 * to just reuse the trace_seq buffer again.
4013 */
4014 if (iter->leftover)
4015 p = iter;
4016 else {
4017 l = *pos - 1;
4018 p = s_next(m, p, &l);
4019 }
4020 }
4021
4022 trace_event_read_lock();
4023 trace_access_lock(cpu_file);
4024 return p;
4025 }
4026
s_stop(struct seq_file * m,void * p)4027 static void s_stop(struct seq_file *m, void *p)
4028 {
4029 struct trace_iterator *iter = m->private;
4030
4031 #ifdef CONFIG_TRACER_MAX_TRACE
4032 if (iter->snapshot && iter->trace->use_max_tr)
4033 return;
4034 #endif
4035
4036 trace_access_unlock(iter->cpu_file);
4037 trace_event_read_unlock();
4038 }
4039
4040 static void
get_total_entries_cpu(struct array_buffer * buf,unsigned long * total,unsigned long * entries,int cpu)4041 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4042 unsigned long *entries, int cpu)
4043 {
4044 unsigned long count;
4045
4046 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4047 /*
4048 * If this buffer has skipped entries, then we hold all
4049 * entries for the trace and we need to ignore the
4050 * ones before the time stamp.
4051 */
4052 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4053 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4054 /* total is the same as the entries */
4055 *total = count;
4056 } else
4057 *total = count +
4058 ring_buffer_overrun_cpu(buf->buffer, cpu);
4059 *entries = count;
4060 }
4061
4062 static void
get_total_entries(struct array_buffer * buf,unsigned long * total,unsigned long * entries)4063 get_total_entries(struct array_buffer *buf,
4064 unsigned long *total, unsigned long *entries)
4065 {
4066 unsigned long t, e;
4067 int cpu;
4068
4069 *total = 0;
4070 *entries = 0;
4071
4072 for_each_tracing_cpu(cpu) {
4073 get_total_entries_cpu(buf, &t, &e, cpu);
4074 *total += t;
4075 *entries += e;
4076 }
4077 }
4078
trace_total_entries_cpu(struct trace_array * tr,int cpu)4079 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4080 {
4081 unsigned long total, entries;
4082
4083 if (!tr)
4084 tr = &global_trace;
4085
4086 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4087
4088 return entries;
4089 }
4090
trace_total_entries(struct trace_array * tr)4091 unsigned long trace_total_entries(struct trace_array *tr)
4092 {
4093 unsigned long total, entries;
4094
4095 if (!tr)
4096 tr = &global_trace;
4097
4098 get_total_entries(&tr->array_buffer, &total, &entries);
4099
4100 return entries;
4101 }
4102
print_lat_help_header(struct seq_file * m)4103 static void print_lat_help_header(struct seq_file *m)
4104 {
4105 seq_puts(m, "# _------=> CPU# \n"
4106 "# / _-----=> irqs-off/BH-disabled\n"
4107 "# | / _----=> need-resched \n"
4108 "# || / _---=> hardirq/softirq \n"
4109 "# ||| / _--=> preempt-depth \n"
4110 "# |||| / _-=> migrate-disable \n"
4111 "# ||||| / delay \n"
4112 "# cmd pid |||||| time | caller \n"
4113 "# \\ / |||||| \\ | / \n");
4114 }
4115
print_event_info(struct array_buffer * buf,struct seq_file * m)4116 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4117 {
4118 unsigned long total;
4119 unsigned long entries;
4120
4121 get_total_entries(buf, &total, &entries);
4122 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4123 entries, total, num_online_cpus());
4124 seq_puts(m, "#\n");
4125 }
4126
print_func_help_header(struct array_buffer * buf,struct seq_file * m,unsigned int flags)4127 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4128 unsigned int flags)
4129 {
4130 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4131
4132 print_event_info(buf, m);
4133
4134 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4135 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4136 }
4137
print_func_help_header_irq(struct array_buffer * buf,struct seq_file * m,unsigned int flags)4138 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4139 unsigned int flags)
4140 {
4141 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4142 static const char space[] = " ";
4143 int prec = tgid ? 12 : 2;
4144
4145 print_event_info(buf, m);
4146
4147 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4148 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4149 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4150 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4151 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4152 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4153 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4154 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4155 }
4156
4157 void
print_trace_header(struct seq_file * m,struct trace_iterator * iter)4158 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4159 {
4160 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4161 struct array_buffer *buf = iter->array_buffer;
4162 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4163 struct tracer *type = iter->trace;
4164 unsigned long entries;
4165 unsigned long total;
4166 const char *name = type->name;
4167
4168 get_total_entries(buf, &total, &entries);
4169
4170 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4171 name, init_utsname()->release);
4172 seq_puts(m, "# -----------------------------------"
4173 "---------------------------------\n");
4174 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4175 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4176 nsecs_to_usecs(data->saved_latency),
4177 entries,
4178 total,
4179 buf->cpu,
4180 preempt_model_none() ? "server" :
4181 preempt_model_voluntary() ? "desktop" :
4182 preempt_model_full() ? "preempt" :
4183 preempt_model_rt() ? "preempt_rt" :
4184 "unknown",
4185 /* These are reserved for later use */
4186 0, 0, 0, 0);
4187 #ifdef CONFIG_SMP
4188 seq_printf(m, " #P:%d)\n", num_online_cpus());
4189 #else
4190 seq_puts(m, ")\n");
4191 #endif
4192 seq_puts(m, "# -----------------\n");
4193 seq_printf(m, "# | task: %.16s-%d "
4194 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4195 data->comm, data->pid,
4196 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4197 data->policy, data->rt_priority);
4198 seq_puts(m, "# -----------------\n");
4199
4200 if (data->critical_start) {
4201 seq_puts(m, "# => started at: ");
4202 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4203 trace_print_seq(m, &iter->seq);
4204 seq_puts(m, "\n# => ended at: ");
4205 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4206 trace_print_seq(m, &iter->seq);
4207 seq_puts(m, "\n#\n");
4208 }
4209
4210 seq_puts(m, "#\n");
4211 }
4212
test_cpu_buff_start(struct trace_iterator * iter)4213 static void test_cpu_buff_start(struct trace_iterator *iter)
4214 {
4215 struct trace_seq *s = &iter->seq;
4216 struct trace_array *tr = iter->tr;
4217
4218 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4219 return;
4220
4221 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4222 return;
4223
4224 if (cpumask_available(iter->started) &&
4225 cpumask_test_cpu(iter->cpu, iter->started))
4226 return;
4227
4228 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4229 return;
4230
4231 if (cpumask_available(iter->started))
4232 cpumask_set_cpu(iter->cpu, iter->started);
4233
4234 /* Don't print started cpu buffer for the first entry of the trace */
4235 if (iter->idx > 1)
4236 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4237 iter->cpu);
4238 }
4239
print_trace_fmt(struct trace_iterator * iter)4240 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4241 {
4242 struct trace_array *tr = iter->tr;
4243 struct trace_seq *s = &iter->seq;
4244 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4245 struct trace_entry *entry;
4246 struct trace_event *event;
4247
4248 entry = iter->ent;
4249
4250 test_cpu_buff_start(iter);
4251
4252 event = ftrace_find_event(entry->type);
4253
4254 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4255 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4256 trace_print_lat_context(iter);
4257 else
4258 trace_print_context(iter);
4259 }
4260
4261 if (trace_seq_has_overflowed(s))
4262 return TRACE_TYPE_PARTIAL_LINE;
4263
4264 if (event) {
4265 if (tr->trace_flags & TRACE_ITER_FIELDS)
4266 return print_event_fields(iter, event);
4267 return event->funcs->trace(iter, sym_flags, event);
4268 }
4269
4270 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4271
4272 return trace_handle_return(s);
4273 }
4274
print_raw_fmt(struct trace_iterator * iter)4275 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4276 {
4277 struct trace_array *tr = iter->tr;
4278 struct trace_seq *s = &iter->seq;
4279 struct trace_entry *entry;
4280 struct trace_event *event;
4281
4282 entry = iter->ent;
4283
4284 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4285 trace_seq_printf(s, "%d %d %llu ",
4286 entry->pid, iter->cpu, iter->ts);
4287
4288 if (trace_seq_has_overflowed(s))
4289 return TRACE_TYPE_PARTIAL_LINE;
4290
4291 event = ftrace_find_event(entry->type);
4292 if (event)
4293 return event->funcs->raw(iter, 0, event);
4294
4295 trace_seq_printf(s, "%d ?\n", entry->type);
4296
4297 return trace_handle_return(s);
4298 }
4299
print_hex_fmt(struct trace_iterator * iter)4300 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4301 {
4302 struct trace_array *tr = iter->tr;
4303 struct trace_seq *s = &iter->seq;
4304 unsigned char newline = '\n';
4305 struct trace_entry *entry;
4306 struct trace_event *event;
4307
4308 entry = iter->ent;
4309
4310 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4311 SEQ_PUT_HEX_FIELD(s, entry->pid);
4312 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4313 SEQ_PUT_HEX_FIELD(s, iter->ts);
4314 if (trace_seq_has_overflowed(s))
4315 return TRACE_TYPE_PARTIAL_LINE;
4316 }
4317
4318 event = ftrace_find_event(entry->type);
4319 if (event) {
4320 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4321 if (ret != TRACE_TYPE_HANDLED)
4322 return ret;
4323 }
4324
4325 SEQ_PUT_FIELD(s, newline);
4326
4327 return trace_handle_return(s);
4328 }
4329
print_bin_fmt(struct trace_iterator * iter)4330 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4331 {
4332 struct trace_array *tr = iter->tr;
4333 struct trace_seq *s = &iter->seq;
4334 struct trace_entry *entry;
4335 struct trace_event *event;
4336
4337 entry = iter->ent;
4338
4339 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4340 SEQ_PUT_FIELD(s, entry->pid);
4341 SEQ_PUT_FIELD(s, iter->cpu);
4342 SEQ_PUT_FIELD(s, iter->ts);
4343 if (trace_seq_has_overflowed(s))
4344 return TRACE_TYPE_PARTIAL_LINE;
4345 }
4346
4347 event = ftrace_find_event(entry->type);
4348 return event ? event->funcs->binary(iter, 0, event) :
4349 TRACE_TYPE_HANDLED;
4350 }
4351
trace_empty(struct trace_iterator * iter)4352 int trace_empty(struct trace_iterator *iter)
4353 {
4354 struct ring_buffer_iter *buf_iter;
4355 int cpu;
4356
4357 /* If we are looking at one CPU buffer, only check that one */
4358 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4359 cpu = iter->cpu_file;
4360 buf_iter = trace_buffer_iter(iter, cpu);
4361 if (buf_iter) {
4362 if (!ring_buffer_iter_empty(buf_iter))
4363 return 0;
4364 } else {
4365 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4366 return 0;
4367 }
4368 return 1;
4369 }
4370
4371 for_each_tracing_cpu(cpu) {
4372 buf_iter = trace_buffer_iter(iter, cpu);
4373 if (buf_iter) {
4374 if (!ring_buffer_iter_empty(buf_iter))
4375 return 0;
4376 } else {
4377 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4378 return 0;
4379 }
4380 }
4381
4382 return 1;
4383 }
4384
4385 /* Called with trace_event_read_lock() held. */
print_trace_line(struct trace_iterator * iter)4386 enum print_line_t print_trace_line(struct trace_iterator *iter)
4387 {
4388 struct trace_array *tr = iter->tr;
4389 unsigned long trace_flags = tr->trace_flags;
4390 enum print_line_t ret;
4391
4392 if (iter->lost_events) {
4393 if (iter->lost_events == (unsigned long)-1)
4394 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4395 iter->cpu);
4396 else
4397 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4398 iter->cpu, iter->lost_events);
4399 if (trace_seq_has_overflowed(&iter->seq))
4400 return TRACE_TYPE_PARTIAL_LINE;
4401 }
4402
4403 if (iter->trace && iter->trace->print_line) {
4404 ret = iter->trace->print_line(iter);
4405 if (ret != TRACE_TYPE_UNHANDLED)
4406 return ret;
4407 }
4408
4409 if (iter->ent->type == TRACE_BPUTS &&
4410 trace_flags & TRACE_ITER_PRINTK &&
4411 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4412 return trace_print_bputs_msg_only(iter);
4413
4414 if (iter->ent->type == TRACE_BPRINT &&
4415 trace_flags & TRACE_ITER_PRINTK &&
4416 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4417 return trace_print_bprintk_msg_only(iter);
4418
4419 if (iter->ent->type == TRACE_PRINT &&
4420 trace_flags & TRACE_ITER_PRINTK &&
4421 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4422 return trace_print_printk_msg_only(iter);
4423
4424 if (trace_flags & TRACE_ITER_BIN)
4425 return print_bin_fmt(iter);
4426
4427 if (trace_flags & TRACE_ITER_HEX)
4428 return print_hex_fmt(iter);
4429
4430 if (trace_flags & TRACE_ITER_RAW)
4431 return print_raw_fmt(iter);
4432
4433 return print_trace_fmt(iter);
4434 }
4435
trace_latency_header(struct seq_file * m)4436 void trace_latency_header(struct seq_file *m)
4437 {
4438 struct trace_iterator *iter = m->private;
4439 struct trace_array *tr = iter->tr;
4440
4441 /* print nothing if the buffers are empty */
4442 if (trace_empty(iter))
4443 return;
4444
4445 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4446 print_trace_header(m, iter);
4447
4448 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4449 print_lat_help_header(m);
4450 }
4451
trace_default_header(struct seq_file * m)4452 void trace_default_header(struct seq_file *m)
4453 {
4454 struct trace_iterator *iter = m->private;
4455 struct trace_array *tr = iter->tr;
4456 unsigned long trace_flags = tr->trace_flags;
4457
4458 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4459 return;
4460
4461 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4462 /* print nothing if the buffers are empty */
4463 if (trace_empty(iter))
4464 return;
4465 print_trace_header(m, iter);
4466 if (!(trace_flags & TRACE_ITER_VERBOSE))
4467 print_lat_help_header(m);
4468 } else {
4469 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4470 if (trace_flags & TRACE_ITER_IRQ_INFO)
4471 print_func_help_header_irq(iter->array_buffer,
4472 m, trace_flags);
4473 else
4474 print_func_help_header(iter->array_buffer, m,
4475 trace_flags);
4476 }
4477 }
4478 }
4479
test_ftrace_alive(struct seq_file * m)4480 static void test_ftrace_alive(struct seq_file *m)
4481 {
4482 if (!ftrace_is_dead())
4483 return;
4484 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4485 "# MAY BE MISSING FUNCTION EVENTS\n");
4486 }
4487
4488 #ifdef CONFIG_TRACER_MAX_TRACE
show_snapshot_main_help(struct seq_file * m)4489 static void show_snapshot_main_help(struct seq_file *m)
4490 {
4491 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4492 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4493 "# Takes a snapshot of the main buffer.\n"
4494 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4495 "# (Doesn't have to be '2' works with any number that\n"
4496 "# is not a '0' or '1')\n");
4497 }
4498
show_snapshot_percpu_help(struct seq_file * m)4499 static void show_snapshot_percpu_help(struct seq_file *m)
4500 {
4501 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4502 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4503 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4504 "# Takes a snapshot of the main buffer for this cpu.\n");
4505 #else
4506 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4507 "# Must use main snapshot file to allocate.\n");
4508 #endif
4509 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4510 "# (Doesn't have to be '2' works with any number that\n"
4511 "# is not a '0' or '1')\n");
4512 }
4513
print_snapshot_help(struct seq_file * m,struct trace_iterator * iter)4514 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4515 {
4516 if (iter->tr->allocated_snapshot)
4517 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4518 else
4519 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4520
4521 seq_puts(m, "# Snapshot commands:\n");
4522 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4523 show_snapshot_main_help(m);
4524 else
4525 show_snapshot_percpu_help(m);
4526 }
4527 #else
4528 /* Should never be called */
print_snapshot_help(struct seq_file * m,struct trace_iterator * iter)4529 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4530 #endif
4531
s_show(struct seq_file * m,void * v)4532 static int s_show(struct seq_file *m, void *v)
4533 {
4534 struct trace_iterator *iter = v;
4535 int ret;
4536
4537 if (iter->ent == NULL) {
4538 if (iter->tr) {
4539 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4540 seq_puts(m, "#\n");
4541 test_ftrace_alive(m);
4542 }
4543 if (iter->snapshot && trace_empty(iter))
4544 print_snapshot_help(m, iter);
4545 else if (iter->trace && iter->trace->print_header)
4546 iter->trace->print_header(m);
4547 else
4548 trace_default_header(m);
4549
4550 } else if (iter->leftover) {
4551 /*
4552 * If we filled the seq_file buffer earlier, we
4553 * want to just show it now.
4554 */
4555 ret = trace_print_seq(m, &iter->seq);
4556
4557 /* ret should this time be zero, but you never know */
4558 iter->leftover = ret;
4559
4560 } else {
4561 ret = print_trace_line(iter);
4562 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4563 iter->seq.full = 0;
4564 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
4565 }
4566 ret = trace_print_seq(m, &iter->seq);
4567 /*
4568 * If we overflow the seq_file buffer, then it will
4569 * ask us for this data again at start up.
4570 * Use that instead.
4571 * ret is 0 if seq_file write succeeded.
4572 * -1 otherwise.
4573 */
4574 iter->leftover = ret;
4575 }
4576
4577 return 0;
4578 }
4579
4580 /*
4581 * Should be used after trace_array_get(), trace_types_lock
4582 * ensures that i_cdev was already initialized.
4583 */
tracing_get_cpu(struct inode * inode)4584 static inline int tracing_get_cpu(struct inode *inode)
4585 {
4586 if (inode->i_cdev) /* See trace_create_cpu_file() */
4587 return (long)inode->i_cdev - 1;
4588 return RING_BUFFER_ALL_CPUS;
4589 }
4590
4591 static const struct seq_operations tracer_seq_ops = {
4592 .start = s_start,
4593 .next = s_next,
4594 .stop = s_stop,
4595 .show = s_show,
4596 };
4597
4598 /*
4599 * Note, as iter itself can be allocated and freed in different
4600 * ways, this function is only used to free its content, and not
4601 * the iterator itself. The only requirement to all the allocations
4602 * is that it must zero all fields (kzalloc), as freeing works with
4603 * ethier allocated content or NULL.
4604 */
free_trace_iter_content(struct trace_iterator * iter)4605 static void free_trace_iter_content(struct trace_iterator *iter)
4606 {
4607 /* The fmt is either NULL, allocated or points to static_fmt_buf */
4608 if (iter->fmt != static_fmt_buf)
4609 kfree(iter->fmt);
4610
4611 kfree(iter->temp);
4612 kfree(iter->buffer_iter);
4613 mutex_destroy(&iter->mutex);
4614 free_cpumask_var(iter->started);
4615 }
4616
4617 static struct trace_iterator *
__tracing_open(struct inode * inode,struct file * file,bool snapshot)4618 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4619 {
4620 struct trace_array *tr = inode->i_private;
4621 struct trace_iterator *iter;
4622 int cpu;
4623
4624 if (tracing_disabled)
4625 return ERR_PTR(-ENODEV);
4626
4627 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4628 if (!iter)
4629 return ERR_PTR(-ENOMEM);
4630
4631 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4632 GFP_KERNEL);
4633 if (!iter->buffer_iter)
4634 goto release;
4635
4636 /*
4637 * trace_find_next_entry() may need to save off iter->ent.
4638 * It will place it into the iter->temp buffer. As most
4639 * events are less than 128, allocate a buffer of that size.
4640 * If one is greater, then trace_find_next_entry() will
4641 * allocate a new buffer to adjust for the bigger iter->ent.
4642 * It's not critical if it fails to get allocated here.
4643 */
4644 iter->temp = kmalloc(128, GFP_KERNEL);
4645 if (iter->temp)
4646 iter->temp_size = 128;
4647
4648 /*
4649 * trace_event_printf() may need to modify given format
4650 * string to replace %p with %px so that it shows real address
4651 * instead of hash value. However, that is only for the event
4652 * tracing, other tracer may not need. Defer the allocation
4653 * until it is needed.
4654 */
4655 iter->fmt = NULL;
4656 iter->fmt_size = 0;
4657
4658 mutex_lock(&trace_types_lock);
4659 iter->trace = tr->current_trace;
4660
4661 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4662 goto fail;
4663
4664 iter->tr = tr;
4665
4666 #ifdef CONFIG_TRACER_MAX_TRACE
4667 /* Currently only the top directory has a snapshot */
4668 if (tr->current_trace->print_max || snapshot)
4669 iter->array_buffer = &tr->max_buffer;
4670 else
4671 #endif
4672 iter->array_buffer = &tr->array_buffer;
4673 iter->snapshot = snapshot;
4674 iter->pos = -1;
4675 iter->cpu_file = tracing_get_cpu(inode);
4676 mutex_init(&iter->mutex);
4677
4678 /* Notify the tracer early; before we stop tracing. */
4679 if (iter->trace->open)
4680 iter->trace->open(iter);
4681
4682 /* Annotate start of buffers if we had overruns */
4683 if (ring_buffer_overruns(iter->array_buffer->buffer))
4684 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4685
4686 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4687 if (trace_clocks[tr->clock_id].in_ns)
4688 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4689
4690 /*
4691 * If pause-on-trace is enabled, then stop the trace while
4692 * dumping, unless this is the "snapshot" file
4693 */
4694 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4695 tracing_stop_tr(tr);
4696
4697 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4698 for_each_tracing_cpu(cpu) {
4699 iter->buffer_iter[cpu] =
4700 ring_buffer_read_prepare(iter->array_buffer->buffer,
4701 cpu, GFP_KERNEL);
4702 }
4703 ring_buffer_read_prepare_sync();
4704 for_each_tracing_cpu(cpu) {
4705 ring_buffer_read_start(iter->buffer_iter[cpu]);
4706 tracing_iter_reset(iter, cpu);
4707 }
4708 } else {
4709 cpu = iter->cpu_file;
4710 iter->buffer_iter[cpu] =
4711 ring_buffer_read_prepare(iter->array_buffer->buffer,
4712 cpu, GFP_KERNEL);
4713 ring_buffer_read_prepare_sync();
4714 ring_buffer_read_start(iter->buffer_iter[cpu]);
4715 tracing_iter_reset(iter, cpu);
4716 }
4717
4718 mutex_unlock(&trace_types_lock);
4719
4720 return iter;
4721
4722 fail:
4723 mutex_unlock(&trace_types_lock);
4724 free_trace_iter_content(iter);
4725 release:
4726 seq_release_private(inode, file);
4727 return ERR_PTR(-ENOMEM);
4728 }
4729
tracing_open_generic(struct inode * inode,struct file * filp)4730 int tracing_open_generic(struct inode *inode, struct file *filp)
4731 {
4732 int ret;
4733
4734 ret = tracing_check_open_get_tr(NULL);
4735 if (ret)
4736 return ret;
4737
4738 filp->private_data = inode->i_private;
4739 return 0;
4740 }
4741
tracing_is_disabled(void)4742 bool tracing_is_disabled(void)
4743 {
4744 return (tracing_disabled) ? true: false;
4745 }
4746
4747 /*
4748 * Open and update trace_array ref count.
4749 * Must have the current trace_array passed to it.
4750 */
tracing_open_generic_tr(struct inode * inode,struct file * filp)4751 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4752 {
4753 struct trace_array *tr = inode->i_private;
4754 int ret;
4755
4756 ret = tracing_check_open_get_tr(tr);
4757 if (ret)
4758 return ret;
4759
4760 filp->private_data = inode->i_private;
4761
4762 return 0;
4763 }
4764
4765 /*
4766 * The private pointer of the inode is the trace_event_file.
4767 * Update the tr ref count associated to it.
4768 */
tracing_open_file_tr(struct inode * inode,struct file * filp)4769 int tracing_open_file_tr(struct inode *inode, struct file *filp)
4770 {
4771 struct trace_event_file *file = inode->i_private;
4772 int ret;
4773
4774 ret = tracing_check_open_get_tr(file->tr);
4775 if (ret)
4776 return ret;
4777
4778 mutex_lock(&event_mutex);
4779
4780 /* Fail if the file is marked for removal */
4781 if (file->flags & EVENT_FILE_FL_FREED) {
4782 trace_array_put(file->tr);
4783 ret = -ENODEV;
4784 } else {
4785 event_file_get(file);
4786 }
4787
4788 mutex_unlock(&event_mutex);
4789 if (ret)
4790 return ret;
4791
4792 filp->private_data = inode->i_private;
4793
4794 return 0;
4795 }
4796
tracing_release_file_tr(struct inode * inode,struct file * filp)4797 int tracing_release_file_tr(struct inode *inode, struct file *filp)
4798 {
4799 struct trace_event_file *file = inode->i_private;
4800
4801 trace_array_put(file->tr);
4802 event_file_put(file);
4803
4804 return 0;
4805 }
4806
tracing_single_release_file_tr(struct inode * inode,struct file * filp)4807 int tracing_single_release_file_tr(struct inode *inode, struct file *filp)
4808 {
4809 tracing_release_file_tr(inode, filp);
4810 return single_release(inode, filp);
4811 }
4812
tracing_mark_open(struct inode * inode,struct file * filp)4813 static int tracing_mark_open(struct inode *inode, struct file *filp)
4814 {
4815 stream_open(inode, filp);
4816 return tracing_open_generic_tr(inode, filp);
4817 }
4818
tracing_release(struct inode * inode,struct file * file)4819 static int tracing_release(struct inode *inode, struct file *file)
4820 {
4821 struct trace_array *tr = inode->i_private;
4822 struct seq_file *m = file->private_data;
4823 struct trace_iterator *iter;
4824 int cpu;
4825
4826 if (!(file->f_mode & FMODE_READ)) {
4827 trace_array_put(tr);
4828 return 0;
4829 }
4830
4831 /* Writes do not use seq_file */
4832 iter = m->private;
4833 mutex_lock(&trace_types_lock);
4834
4835 for_each_tracing_cpu(cpu) {
4836 if (iter->buffer_iter[cpu])
4837 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4838 }
4839
4840 if (iter->trace && iter->trace->close)
4841 iter->trace->close(iter);
4842
4843 if (!iter->snapshot && tr->stop_count)
4844 /* reenable tracing if it was previously enabled */
4845 tracing_start_tr(tr);
4846
4847 __trace_array_put(tr);
4848
4849 mutex_unlock(&trace_types_lock);
4850
4851 free_trace_iter_content(iter);
4852 seq_release_private(inode, file);
4853
4854 return 0;
4855 }
4856
tracing_release_generic_tr(struct inode * inode,struct file * file)4857 int tracing_release_generic_tr(struct inode *inode, struct file *file)
4858 {
4859 struct trace_array *tr = inode->i_private;
4860
4861 trace_array_put(tr);
4862 return 0;
4863 }
4864
tracing_single_release_tr(struct inode * inode,struct file * file)4865 static int tracing_single_release_tr(struct inode *inode, struct file *file)
4866 {
4867 struct trace_array *tr = inode->i_private;
4868
4869 trace_array_put(tr);
4870
4871 return single_release(inode, file);
4872 }
4873
tracing_open(struct inode * inode,struct file * file)4874 static int tracing_open(struct inode *inode, struct file *file)
4875 {
4876 struct trace_array *tr = inode->i_private;
4877 struct trace_iterator *iter;
4878 int ret;
4879
4880 ret = tracing_check_open_get_tr(tr);
4881 if (ret)
4882 return ret;
4883
4884 /* If this file was open for write, then erase contents */
4885 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4886 int cpu = tracing_get_cpu(inode);
4887 struct array_buffer *trace_buf = &tr->array_buffer;
4888
4889 #ifdef CONFIG_TRACER_MAX_TRACE
4890 if (tr->current_trace->print_max)
4891 trace_buf = &tr->max_buffer;
4892 #endif
4893
4894 if (cpu == RING_BUFFER_ALL_CPUS)
4895 tracing_reset_online_cpus(trace_buf);
4896 else
4897 tracing_reset_cpu(trace_buf, cpu);
4898 }
4899
4900 if (file->f_mode & FMODE_READ) {
4901 iter = __tracing_open(inode, file, false);
4902 if (IS_ERR(iter))
4903 ret = PTR_ERR(iter);
4904 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4905 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4906 }
4907
4908 if (ret < 0)
4909 trace_array_put(tr);
4910
4911 return ret;
4912 }
4913
4914 /*
4915 * Some tracers are not suitable for instance buffers.
4916 * A tracer is always available for the global array (toplevel)
4917 * or if it explicitly states that it is.
4918 */
4919 static bool
trace_ok_for_array(struct tracer * t,struct trace_array * tr)4920 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4921 {
4922 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4923 }
4924
4925 /* Find the next tracer that this trace array may use */
4926 static struct tracer *
get_tracer_for_array(struct trace_array * tr,struct tracer * t)4927 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4928 {
4929 while (t && !trace_ok_for_array(t, tr))
4930 t = t->next;
4931
4932 return t;
4933 }
4934
4935 static void *
t_next(struct seq_file * m,void * v,loff_t * pos)4936 t_next(struct seq_file *m, void *v, loff_t *pos)
4937 {
4938 struct trace_array *tr = m->private;
4939 struct tracer *t = v;
4940
4941 (*pos)++;
4942
4943 if (t)
4944 t = get_tracer_for_array(tr, t->next);
4945
4946 return t;
4947 }
4948
t_start(struct seq_file * m,loff_t * pos)4949 static void *t_start(struct seq_file *m, loff_t *pos)
4950 {
4951 struct trace_array *tr = m->private;
4952 struct tracer *t;
4953 loff_t l = 0;
4954
4955 mutex_lock(&trace_types_lock);
4956
4957 t = get_tracer_for_array(tr, trace_types);
4958 for (; t && l < *pos; t = t_next(m, t, &l))
4959 ;
4960
4961 return t;
4962 }
4963
t_stop(struct seq_file * m,void * p)4964 static void t_stop(struct seq_file *m, void *p)
4965 {
4966 mutex_unlock(&trace_types_lock);
4967 }
4968
t_show(struct seq_file * m,void * v)4969 static int t_show(struct seq_file *m, void *v)
4970 {
4971 struct tracer *t = v;
4972
4973 if (!t)
4974 return 0;
4975
4976 seq_puts(m, t->name);
4977 if (t->next)
4978 seq_putc(m, ' ');
4979 else
4980 seq_putc(m, '\n');
4981
4982 return 0;
4983 }
4984
4985 static const struct seq_operations show_traces_seq_ops = {
4986 .start = t_start,
4987 .next = t_next,
4988 .stop = t_stop,
4989 .show = t_show,
4990 };
4991
show_traces_open(struct inode * inode,struct file * file)4992 static int show_traces_open(struct inode *inode, struct file *file)
4993 {
4994 struct trace_array *tr = inode->i_private;
4995 struct seq_file *m;
4996 int ret;
4997
4998 ret = tracing_check_open_get_tr(tr);
4999 if (ret)
5000 return ret;
5001
5002 ret = seq_open(file, &show_traces_seq_ops);
5003 if (ret) {
5004 trace_array_put(tr);
5005 return ret;
5006 }
5007
5008 m = file->private_data;
5009 m->private = tr;
5010
5011 return 0;
5012 }
5013
show_traces_release(struct inode * inode,struct file * file)5014 static int show_traces_release(struct inode *inode, struct file *file)
5015 {
5016 struct trace_array *tr = inode->i_private;
5017
5018 trace_array_put(tr);
5019 return seq_release(inode, file);
5020 }
5021
5022 static ssize_t
tracing_write_stub(struct file * filp,const char __user * ubuf,size_t count,loff_t * ppos)5023 tracing_write_stub(struct file *filp, const char __user *ubuf,
5024 size_t count, loff_t *ppos)
5025 {
5026 return count;
5027 }
5028
tracing_lseek(struct file * file,loff_t offset,int whence)5029 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5030 {
5031 int ret;
5032
5033 if (file->f_mode & FMODE_READ)
5034 ret = seq_lseek(file, offset, whence);
5035 else
5036 file->f_pos = ret = 0;
5037
5038 return ret;
5039 }
5040
5041 static const struct file_operations tracing_fops = {
5042 .open = tracing_open,
5043 .read = seq_read,
5044 .read_iter = seq_read_iter,
5045 .splice_read = copy_splice_read,
5046 .write = tracing_write_stub,
5047 .llseek = tracing_lseek,
5048 .release = tracing_release,
5049 };
5050
5051 static const struct file_operations show_traces_fops = {
5052 .open = show_traces_open,
5053 .read = seq_read,
5054 .llseek = seq_lseek,
5055 .release = show_traces_release,
5056 };
5057
5058 static ssize_t
tracing_cpumask_read(struct file * filp,char __user * ubuf,size_t count,loff_t * ppos)5059 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5060 size_t count, loff_t *ppos)
5061 {
5062 struct trace_array *tr = file_inode(filp)->i_private;
5063 char *mask_str;
5064 int len;
5065
5066 len = snprintf(NULL, 0, "%*pb\n",
5067 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5068 mask_str = kmalloc(len, GFP_KERNEL);
5069 if (!mask_str)
5070 return -ENOMEM;
5071
5072 len = snprintf(mask_str, len, "%*pb\n",
5073 cpumask_pr_args(tr->tracing_cpumask));
5074 if (len >= count) {
5075 count = -EINVAL;
5076 goto out_err;
5077 }
5078 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5079
5080 out_err:
5081 kfree(mask_str);
5082
5083 return count;
5084 }
5085
tracing_set_cpumask(struct trace_array * tr,cpumask_var_t tracing_cpumask_new)5086 int tracing_set_cpumask(struct trace_array *tr,
5087 cpumask_var_t tracing_cpumask_new)
5088 {
5089 int cpu;
5090
5091 if (!tr)
5092 return -EINVAL;
5093
5094 local_irq_disable();
5095 arch_spin_lock(&tr->max_lock);
5096 for_each_tracing_cpu(cpu) {
5097 /*
5098 * Increase/decrease the disabled counter if we are
5099 * about to flip a bit in the cpumask:
5100 */
5101 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5102 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5103 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5104 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5105 #ifdef CONFIG_TRACER_MAX_TRACE
5106 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5107 #endif
5108 }
5109 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5110 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5111 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5112 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5113 #ifdef CONFIG_TRACER_MAX_TRACE
5114 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5115 #endif
5116 }
5117 }
5118 arch_spin_unlock(&tr->max_lock);
5119 local_irq_enable();
5120
5121 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5122
5123 return 0;
5124 }
5125
5126 static ssize_t
tracing_cpumask_write(struct file * filp,const char __user * ubuf,size_t count,loff_t * ppos)5127 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5128 size_t count, loff_t *ppos)
5129 {
5130 struct trace_array *tr = file_inode(filp)->i_private;
5131 cpumask_var_t tracing_cpumask_new;
5132 int err;
5133
5134 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5135 return -ENOMEM;
5136
5137 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5138 if (err)
5139 goto err_free;
5140
5141 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5142 if (err)
5143 goto err_free;
5144
5145 free_cpumask_var(tracing_cpumask_new);
5146
5147 return count;
5148
5149 err_free:
5150 free_cpumask_var(tracing_cpumask_new);
5151
5152 return err;
5153 }
5154
5155 static const struct file_operations tracing_cpumask_fops = {
5156 .open = tracing_open_generic_tr,
5157 .read = tracing_cpumask_read,
5158 .write = tracing_cpumask_write,
5159 .release = tracing_release_generic_tr,
5160 .llseek = generic_file_llseek,
5161 };
5162
tracing_trace_options_show(struct seq_file * m,void * v)5163 static int tracing_trace_options_show(struct seq_file *m, void *v)
5164 {
5165 struct tracer_opt *trace_opts;
5166 struct trace_array *tr = m->private;
5167 u32 tracer_flags;
5168 int i;
5169
5170 mutex_lock(&trace_types_lock);
5171 tracer_flags = tr->current_trace->flags->val;
5172 trace_opts = tr->current_trace->flags->opts;
5173
5174 for (i = 0; trace_options[i]; i++) {
5175 if (tr->trace_flags & (1 << i))
5176 seq_printf(m, "%s\n", trace_options[i]);
5177 else
5178 seq_printf(m, "no%s\n", trace_options[i]);
5179 }
5180
5181 for (i = 0; trace_opts[i].name; i++) {
5182 if (tracer_flags & trace_opts[i].bit)
5183 seq_printf(m, "%s\n", trace_opts[i].name);
5184 else
5185 seq_printf(m, "no%s\n", trace_opts[i].name);
5186 }
5187 mutex_unlock(&trace_types_lock);
5188
5189 return 0;
5190 }
5191
__set_tracer_option(struct trace_array * tr,struct tracer_flags * tracer_flags,struct tracer_opt * opts,int neg)5192 static int __set_tracer_option(struct trace_array *tr,
5193 struct tracer_flags *tracer_flags,
5194 struct tracer_opt *opts, int neg)
5195 {
5196 struct tracer *trace = tracer_flags->trace;
5197 int ret;
5198
5199 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5200 if (ret)
5201 return ret;
5202
5203 if (neg)
5204 tracer_flags->val &= ~opts->bit;
5205 else
5206 tracer_flags->val |= opts->bit;
5207 return 0;
5208 }
5209
5210 /* Try to assign a tracer specific option */
set_tracer_option(struct trace_array * tr,char * cmp,int neg)5211 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5212 {
5213 struct tracer *trace = tr->current_trace;
5214 struct tracer_flags *tracer_flags = trace->flags;
5215 struct tracer_opt *opts = NULL;
5216 int i;
5217
5218 for (i = 0; tracer_flags->opts[i].name; i++) {
5219 opts = &tracer_flags->opts[i];
5220
5221 if (strcmp(cmp, opts->name) == 0)
5222 return __set_tracer_option(tr, trace->flags, opts, neg);
5223 }
5224
5225 return -EINVAL;
5226 }
5227
5228 /* Some tracers require overwrite to stay enabled */
trace_keep_overwrite(struct tracer * tracer,u32 mask,int set)5229 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5230 {
5231 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5232 return -1;
5233
5234 return 0;
5235 }
5236
set_tracer_flag(struct trace_array * tr,unsigned int mask,int enabled)5237 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5238 {
5239 if ((mask == TRACE_ITER_RECORD_TGID) ||
5240 (mask == TRACE_ITER_RECORD_CMD))
5241 lockdep_assert_held(&event_mutex);
5242
5243 /* do nothing if flag is already set */
5244 if (!!(tr->trace_flags & mask) == !!enabled)
5245 return 0;
5246
5247 /* Give the tracer a chance to approve the change */
5248 if (tr->current_trace->flag_changed)
5249 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5250 return -EINVAL;
5251
5252 if (enabled)
5253 tr->trace_flags |= mask;
5254 else
5255 tr->trace_flags &= ~mask;
5256
5257 if (mask == TRACE_ITER_RECORD_CMD)
5258 trace_event_enable_cmd_record(enabled);
5259
5260 if (mask == TRACE_ITER_RECORD_TGID) {
5261
5262 if (trace_alloc_tgid_map() < 0) {
5263 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5264 return -ENOMEM;
5265 }
5266
5267 trace_event_enable_tgid_record(enabled);
5268 }
5269
5270 if (mask == TRACE_ITER_EVENT_FORK)
5271 trace_event_follow_fork(tr, enabled);
5272
5273 if (mask == TRACE_ITER_FUNC_FORK)
5274 ftrace_pid_follow_fork(tr, enabled);
5275
5276 if (mask == TRACE_ITER_OVERWRITE) {
5277 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5278 #ifdef CONFIG_TRACER_MAX_TRACE
5279 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5280 #endif
5281 }
5282
5283 if (mask == TRACE_ITER_PRINTK) {
5284 trace_printk_start_stop_comm(enabled);
5285 trace_printk_control(enabled);
5286 }
5287
5288 return 0;
5289 }
5290
trace_set_options(struct trace_array * tr,char * option)5291 int trace_set_options(struct trace_array *tr, char *option)
5292 {
5293 char *cmp;
5294 int neg = 0;
5295 int ret;
5296 size_t orig_len = strlen(option);
5297 int len;
5298
5299 cmp = strstrip(option);
5300
5301 len = str_has_prefix(cmp, "no");
5302 if (len)
5303 neg = 1;
5304
5305 cmp += len;
5306
5307 mutex_lock(&event_mutex);
5308 mutex_lock(&trace_types_lock);
5309
5310 ret = match_string(trace_options, -1, cmp);
5311 /* If no option could be set, test the specific tracer options */
5312 if (ret < 0)
5313 ret = set_tracer_option(tr, cmp, neg);
5314 else
5315 ret = set_tracer_flag(tr, 1 << ret, !neg);
5316
5317 mutex_unlock(&trace_types_lock);
5318 mutex_unlock(&event_mutex);
5319
5320 /*
5321 * If the first trailing whitespace is replaced with '\0' by strstrip,
5322 * turn it back into a space.
5323 */
5324 if (orig_len > strlen(option))
5325 option[strlen(option)] = ' ';
5326
5327 return ret;
5328 }
5329
apply_trace_boot_options(void)5330 static void __init apply_trace_boot_options(void)
5331 {
5332 char *buf = trace_boot_options_buf;
5333 char *option;
5334
5335 while (true) {
5336 option = strsep(&buf, ",");
5337
5338 if (!option)
5339 break;
5340
5341 if (*option)
5342 trace_set_options(&global_trace, option);
5343
5344 /* Put back the comma to allow this to be called again */
5345 if (buf)
5346 *(buf - 1) = ',';
5347 }
5348 }
5349
5350 static ssize_t
tracing_trace_options_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)5351 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5352 size_t cnt, loff_t *ppos)
5353 {
5354 struct seq_file *m = filp->private_data;
5355 struct trace_array *tr = m->private;
5356 char buf[64];
5357 int ret;
5358
5359 if (cnt >= sizeof(buf))
5360 return -EINVAL;
5361
5362 if (copy_from_user(buf, ubuf, cnt))
5363 return -EFAULT;
5364
5365 buf[cnt] = 0;
5366
5367 ret = trace_set_options(tr, buf);
5368 if (ret < 0)
5369 return ret;
5370
5371 *ppos += cnt;
5372
5373 return cnt;
5374 }
5375
tracing_trace_options_open(struct inode * inode,struct file * file)5376 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5377 {
5378 struct trace_array *tr = inode->i_private;
5379 int ret;
5380
5381 ret = tracing_check_open_get_tr(tr);
5382 if (ret)
5383 return ret;
5384
5385 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5386 if (ret < 0)
5387 trace_array_put(tr);
5388
5389 return ret;
5390 }
5391
5392 static const struct file_operations tracing_iter_fops = {
5393 .open = tracing_trace_options_open,
5394 .read = seq_read,
5395 .llseek = seq_lseek,
5396 .release = tracing_single_release_tr,
5397 .write = tracing_trace_options_write,
5398 };
5399
5400 static const char readme_msg[] =
5401 "tracing mini-HOWTO:\n\n"
5402 "# echo 0 > tracing_on : quick way to disable tracing\n"
5403 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5404 " Important files:\n"
5405 " trace\t\t\t- The static contents of the buffer\n"
5406 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5407 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5408 " current_tracer\t- function and latency tracers\n"
5409 " available_tracers\t- list of configured tracers for current_tracer\n"
5410 " error_log\t- error log for failed commands (that support it)\n"
5411 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5412 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5413 " trace_clock\t\t- change the clock used to order events\n"
5414 " local: Per cpu clock but may not be synced across CPUs\n"
5415 " global: Synced across CPUs but slows tracing down.\n"
5416 " counter: Not a clock, but just an increment\n"
5417 " uptime: Jiffy counter from time of boot\n"
5418 " perf: Same clock that perf events use\n"
5419 #ifdef CONFIG_X86_64
5420 " x86-tsc: TSC cycle counter\n"
5421 #endif
5422 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5423 " delta: Delta difference against a buffer-wide timestamp\n"
5424 " absolute: Absolute (standalone) timestamp\n"
5425 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5426 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5427 " tracing_cpumask\t- Limit which CPUs to trace\n"
5428 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5429 "\t\t\t Remove sub-buffer with rmdir\n"
5430 " trace_options\t\t- Set format or modify how tracing happens\n"
5431 "\t\t\t Disable an option by prefixing 'no' to the\n"
5432 "\t\t\t option name\n"
5433 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5434 #ifdef CONFIG_DYNAMIC_FTRACE
5435 "\n available_filter_functions - list of functions that can be filtered on\n"
5436 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5437 "\t\t\t functions\n"
5438 "\t accepts: func_full_name or glob-matching-pattern\n"
5439 "\t modules: Can select a group via module\n"
5440 "\t Format: :mod:<module-name>\n"
5441 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5442 "\t triggers: a command to perform when function is hit\n"
5443 "\t Format: <function>:<trigger>[:count]\n"
5444 "\t trigger: traceon, traceoff\n"
5445 "\t\t enable_event:<system>:<event>\n"
5446 "\t\t disable_event:<system>:<event>\n"
5447 #ifdef CONFIG_STACKTRACE
5448 "\t\t stacktrace\n"
5449 #endif
5450 #ifdef CONFIG_TRACER_SNAPSHOT
5451 "\t\t snapshot\n"
5452 #endif
5453 "\t\t dump\n"
5454 "\t\t cpudump\n"
5455 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5456 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5457 "\t The first one will disable tracing every time do_fault is hit\n"
5458 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5459 "\t The first time do trap is hit and it disables tracing, the\n"
5460 "\t counter will decrement to 2. If tracing is already disabled,\n"
5461 "\t the counter will not decrement. It only decrements when the\n"
5462 "\t trigger did work\n"
5463 "\t To remove trigger without count:\n"
5464 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5465 "\t To remove trigger with a count:\n"
5466 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5467 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5468 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5469 "\t modules: Can select a group via module command :mod:\n"
5470 "\t Does not accept triggers\n"
5471 #endif /* CONFIG_DYNAMIC_FTRACE */
5472 #ifdef CONFIG_FUNCTION_TRACER
5473 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5474 "\t\t (function)\n"
5475 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5476 "\t\t (function)\n"
5477 #endif
5478 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5479 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5480 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5481 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5482 #endif
5483 #ifdef CONFIG_TRACER_SNAPSHOT
5484 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5485 "\t\t\t snapshot buffer. Read the contents for more\n"
5486 "\t\t\t information\n"
5487 #endif
5488 #ifdef CONFIG_STACK_TRACER
5489 " stack_trace\t\t- Shows the max stack trace when active\n"
5490 " stack_max_size\t- Shows current max stack size that was traced\n"
5491 "\t\t\t Write into this file to reset the max size (trigger a\n"
5492 "\t\t\t new trace)\n"
5493 #ifdef CONFIG_DYNAMIC_FTRACE
5494 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5495 "\t\t\t traces\n"
5496 #endif
5497 #endif /* CONFIG_STACK_TRACER */
5498 #ifdef CONFIG_DYNAMIC_EVENTS
5499 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5500 "\t\t\t Write into this file to define/undefine new trace events.\n"
5501 #endif
5502 #ifdef CONFIG_KPROBE_EVENTS
5503 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5504 "\t\t\t Write into this file to define/undefine new trace events.\n"
5505 #endif
5506 #ifdef CONFIG_UPROBE_EVENTS
5507 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5508 "\t\t\t Write into this file to define/undefine new trace events.\n"
5509 #endif
5510 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5511 defined(CONFIG_FPROBE_EVENTS)
5512 "\t accepts: event-definitions (one definition per line)\n"
5513 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5514 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5515 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5516 #endif
5517 #ifdef CONFIG_FPROBE_EVENTS
5518 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5519 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5520 #endif
5521 #ifdef CONFIG_HIST_TRIGGERS
5522 "\t s:[synthetic/]<event> <field> [<field>]\n"
5523 #endif
5524 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5525 "\t -:[<group>/][<event>]\n"
5526 #ifdef CONFIG_KPROBE_EVENTS
5527 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5528 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5529 #endif
5530 #ifdef CONFIG_UPROBE_EVENTS
5531 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5532 #endif
5533 "\t args: <name>=fetcharg[:type]\n"
5534 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5535 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5536 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5537 #ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5538 "\t <argname>[->field[->field|.field...]],\n"
5539 #endif
5540 #else
5541 "\t $stack<index>, $stack, $retval, $comm,\n"
5542 #endif
5543 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5544 "\t kernel return probes support: $retval, $arg<N>, $comm\n"
5545 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5546 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5547 "\t symstr, %pd/%pD, <type>\\[<array-size>\\]\n"
5548 #ifdef CONFIG_HIST_TRIGGERS
5549 "\t field: <stype> <name>;\n"
5550 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5551 "\t [unsigned] char/int/long\n"
5552 #endif
5553 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5554 "\t of the <attached-group>/<attached-event>.\n"
5555 #endif
5556 " events/\t\t- Directory containing all trace event subsystems:\n"
5557 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5558 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5559 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5560 "\t\t\t events\n"
5561 " filter\t\t- If set, only events passing filter are traced\n"
5562 " events/<system>/<event>/\t- Directory containing control files for\n"
5563 "\t\t\t <event>:\n"
5564 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5565 " filter\t\t- If set, only events passing filter are traced\n"
5566 " trigger\t\t- If set, a command to perform when event is hit\n"
5567 "\t Format: <trigger>[:count][if <filter>]\n"
5568 "\t trigger: traceon, traceoff\n"
5569 "\t enable_event:<system>:<event>\n"
5570 "\t disable_event:<system>:<event>\n"
5571 #ifdef CONFIG_HIST_TRIGGERS
5572 "\t enable_hist:<system>:<event>\n"
5573 "\t disable_hist:<system>:<event>\n"
5574 #endif
5575 #ifdef CONFIG_STACKTRACE
5576 "\t\t stacktrace\n"
5577 #endif
5578 #ifdef CONFIG_TRACER_SNAPSHOT
5579 "\t\t snapshot\n"
5580 #endif
5581 #ifdef CONFIG_HIST_TRIGGERS
5582 "\t\t hist (see below)\n"
5583 #endif
5584 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5585 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5586 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5587 "\t events/block/block_unplug/trigger\n"
5588 "\t The first disables tracing every time block_unplug is hit.\n"
5589 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5590 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5591 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5592 "\t Like function triggers, the counter is only decremented if it\n"
5593 "\t enabled or disabled tracing.\n"
5594 "\t To remove a trigger without a count:\n"
5595 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5596 "\t To remove a trigger with a count:\n"
5597 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5598 "\t Filters can be ignored when removing a trigger.\n"
5599 #ifdef CONFIG_HIST_TRIGGERS
5600 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5601 "\t Format: hist:keys=<field1[,field2,...]>\n"
5602 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5603 "\t [:values=<field1[,field2,...]>]\n"
5604 "\t [:sort=<field1[,field2,...]>]\n"
5605 "\t [:size=#entries]\n"
5606 "\t [:pause][:continue][:clear]\n"
5607 "\t [:name=histname1]\n"
5608 "\t [:nohitcount]\n"
5609 "\t [:<handler>.<action>]\n"
5610 "\t [if <filter>]\n\n"
5611 "\t Note, special fields can be used as well:\n"
5612 "\t common_timestamp - to record current timestamp\n"
5613 "\t common_cpu - to record the CPU the event happened on\n"
5614 "\n"
5615 "\t A hist trigger variable can be:\n"
5616 "\t - a reference to a field e.g. x=current_timestamp,\n"
5617 "\t - a reference to another variable e.g. y=$x,\n"
5618 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5619 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5620 "\n"
5621 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5622 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5623 "\t variable reference, field or numeric literal.\n"
5624 "\n"
5625 "\t When a matching event is hit, an entry is added to a hash\n"
5626 "\t table using the key(s) and value(s) named, and the value of a\n"
5627 "\t sum called 'hitcount' is incremented. Keys and values\n"
5628 "\t correspond to fields in the event's format description. Keys\n"
5629 "\t can be any field, or the special string 'common_stacktrace'.\n"
5630 "\t Compound keys consisting of up to two fields can be specified\n"
5631 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5632 "\t fields. Sort keys consisting of up to two fields can be\n"
5633 "\t specified using the 'sort' keyword. The sort direction can\n"
5634 "\t be modified by appending '.descending' or '.ascending' to a\n"
5635 "\t sort field. The 'size' parameter can be used to specify more\n"
5636 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5637 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5638 "\t its histogram data will be shared with other triggers of the\n"
5639 "\t same name, and trigger hits will update this common data.\n\n"
5640 "\t Reading the 'hist' file for the event will dump the hash\n"
5641 "\t table in its entirety to stdout. If there are multiple hist\n"
5642 "\t triggers attached to an event, there will be a table for each\n"
5643 "\t trigger in the output. The table displayed for a named\n"
5644 "\t trigger will be the same as any other instance having the\n"
5645 "\t same name. The default format used to display a given field\n"
5646 "\t can be modified by appending any of the following modifiers\n"
5647 "\t to the field name, as applicable:\n\n"
5648 "\t .hex display a number as a hex value\n"
5649 "\t .sym display an address as a symbol\n"
5650 "\t .sym-offset display an address as a symbol and offset\n"
5651 "\t .execname display a common_pid as a program name\n"
5652 "\t .syscall display a syscall id as a syscall name\n"
5653 "\t .log2 display log2 value rather than raw number\n"
5654 "\t .buckets=size display values in groups of size rather than raw number\n"
5655 "\t .usecs display a common_timestamp in microseconds\n"
5656 "\t .percent display a number of percentage value\n"
5657 "\t .graph display a bar-graph of a value\n\n"
5658 "\t The 'pause' parameter can be used to pause an existing hist\n"
5659 "\t trigger or to start a hist trigger but not log any events\n"
5660 "\t until told to do so. 'continue' can be used to start or\n"
5661 "\t restart a paused hist trigger.\n\n"
5662 "\t The 'clear' parameter will clear the contents of a running\n"
5663 "\t hist trigger and leave its current paused/active state\n"
5664 "\t unchanged.\n\n"
5665 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5666 "\t raw hitcount in the histogram.\n\n"
5667 "\t The enable_hist and disable_hist triggers can be used to\n"
5668 "\t have one event conditionally start and stop another event's\n"
5669 "\t already-attached hist trigger. The syntax is analogous to\n"
5670 "\t the enable_event and disable_event triggers.\n\n"
5671 "\t Hist trigger handlers and actions are executed whenever a\n"
5672 "\t a histogram entry is added or updated. They take the form:\n\n"
5673 "\t <handler>.<action>\n\n"
5674 "\t The available handlers are:\n\n"
5675 "\t onmatch(matching.event) - invoke on addition or update\n"
5676 "\t onmax(var) - invoke if var exceeds current max\n"
5677 "\t onchange(var) - invoke action if var changes\n\n"
5678 "\t The available actions are:\n\n"
5679 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5680 "\t save(field,...) - save current event fields\n"
5681 #ifdef CONFIG_TRACER_SNAPSHOT
5682 "\t snapshot() - snapshot the trace buffer\n\n"
5683 #endif
5684 #ifdef CONFIG_SYNTH_EVENTS
5685 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5686 "\t Write into this file to define/undefine new synthetic events.\n"
5687 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5688 #endif
5689 #endif
5690 ;
5691
5692 static ssize_t
tracing_readme_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)5693 tracing_readme_read(struct file *filp, char __user *ubuf,
5694 size_t cnt, loff_t *ppos)
5695 {
5696 return simple_read_from_buffer(ubuf, cnt, ppos,
5697 readme_msg, strlen(readme_msg));
5698 }
5699
5700 static const struct file_operations tracing_readme_fops = {
5701 .open = tracing_open_generic,
5702 .read = tracing_readme_read,
5703 .llseek = generic_file_llseek,
5704 };
5705
5706 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
5707 static union trace_eval_map_item *
update_eval_map(union trace_eval_map_item * ptr)5708 update_eval_map(union trace_eval_map_item *ptr)
5709 {
5710 if (!ptr->map.eval_string) {
5711 if (ptr->tail.next) {
5712 ptr = ptr->tail.next;
5713 /* Set ptr to the next real item (skip head) */
5714 ptr++;
5715 } else
5716 return NULL;
5717 }
5718 return ptr;
5719 }
5720
eval_map_next(struct seq_file * m,void * v,loff_t * pos)5721 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5722 {
5723 union trace_eval_map_item *ptr = v;
5724
5725 /*
5726 * Paranoid! If ptr points to end, we don't want to increment past it.
5727 * This really should never happen.
5728 */
5729 (*pos)++;
5730 ptr = update_eval_map(ptr);
5731 if (WARN_ON_ONCE(!ptr))
5732 return NULL;
5733
5734 ptr++;
5735 ptr = update_eval_map(ptr);
5736
5737 return ptr;
5738 }
5739
eval_map_start(struct seq_file * m,loff_t * pos)5740 static void *eval_map_start(struct seq_file *m, loff_t *pos)
5741 {
5742 union trace_eval_map_item *v;
5743 loff_t l = 0;
5744
5745 mutex_lock(&trace_eval_mutex);
5746
5747 v = trace_eval_maps;
5748 if (v)
5749 v++;
5750
5751 while (v && l < *pos) {
5752 v = eval_map_next(m, v, &l);
5753 }
5754
5755 return v;
5756 }
5757
eval_map_stop(struct seq_file * m,void * v)5758 static void eval_map_stop(struct seq_file *m, void *v)
5759 {
5760 mutex_unlock(&trace_eval_mutex);
5761 }
5762
eval_map_show(struct seq_file * m,void * v)5763 static int eval_map_show(struct seq_file *m, void *v)
5764 {
5765 union trace_eval_map_item *ptr = v;
5766
5767 seq_printf(m, "%s %ld (%s)\n",
5768 ptr->map.eval_string, ptr->map.eval_value,
5769 ptr->map.system);
5770
5771 return 0;
5772 }
5773
5774 static const struct seq_operations tracing_eval_map_seq_ops = {
5775 .start = eval_map_start,
5776 .next = eval_map_next,
5777 .stop = eval_map_stop,
5778 .show = eval_map_show,
5779 };
5780
tracing_eval_map_open(struct inode * inode,struct file * filp)5781 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5782 {
5783 int ret;
5784
5785 ret = tracing_check_open_get_tr(NULL);
5786 if (ret)
5787 return ret;
5788
5789 return seq_open(filp, &tracing_eval_map_seq_ops);
5790 }
5791
5792 static const struct file_operations tracing_eval_map_fops = {
5793 .open = tracing_eval_map_open,
5794 .read = seq_read,
5795 .llseek = seq_lseek,
5796 .release = seq_release,
5797 };
5798
5799 static inline union trace_eval_map_item *
trace_eval_jmp_to_tail(union trace_eval_map_item * ptr)5800 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
5801 {
5802 /* Return tail of array given the head */
5803 return ptr + ptr->head.length + 1;
5804 }
5805
5806 static void
trace_insert_eval_map_file(struct module * mod,struct trace_eval_map ** start,int len)5807 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
5808 int len)
5809 {
5810 struct trace_eval_map **stop;
5811 struct trace_eval_map **map;
5812 union trace_eval_map_item *map_array;
5813 union trace_eval_map_item *ptr;
5814
5815 stop = start + len;
5816
5817 /*
5818 * The trace_eval_maps contains the map plus a head and tail item,
5819 * where the head holds the module and length of array, and the
5820 * tail holds a pointer to the next list.
5821 */
5822 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
5823 if (!map_array) {
5824 pr_warn("Unable to allocate trace eval mapping\n");
5825 return;
5826 }
5827
5828 mutex_lock(&trace_eval_mutex);
5829
5830 if (!trace_eval_maps)
5831 trace_eval_maps = map_array;
5832 else {
5833 ptr = trace_eval_maps;
5834 for (;;) {
5835 ptr = trace_eval_jmp_to_tail(ptr);
5836 if (!ptr->tail.next)
5837 break;
5838 ptr = ptr->tail.next;
5839
5840 }
5841 ptr->tail.next = map_array;
5842 }
5843 map_array->head.mod = mod;
5844 map_array->head.length = len;
5845 map_array++;
5846
5847 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
5848 map_array->map = **map;
5849 map_array++;
5850 }
5851 memset(map_array, 0, sizeof(*map_array));
5852
5853 mutex_unlock(&trace_eval_mutex);
5854 }
5855
trace_create_eval_file(struct dentry * d_tracer)5856 static void trace_create_eval_file(struct dentry *d_tracer)
5857 {
5858 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
5859 NULL, &tracing_eval_map_fops);
5860 }
5861
5862 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
trace_create_eval_file(struct dentry * d_tracer)5863 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
trace_insert_eval_map_file(struct module * mod,struct trace_eval_map ** start,int len)5864 static inline void trace_insert_eval_map_file(struct module *mod,
5865 struct trace_eval_map **start, int len) { }
5866 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
5867
trace_insert_eval_map(struct module * mod,struct trace_eval_map ** start,int len)5868 static void trace_insert_eval_map(struct module *mod,
5869 struct trace_eval_map **start, int len)
5870 {
5871 struct trace_eval_map **map;
5872
5873 if (len <= 0)
5874 return;
5875
5876 map = start;
5877
5878 trace_event_eval_update(map, len);
5879
5880 trace_insert_eval_map_file(mod, start, len);
5881 }
5882
5883 static ssize_t
tracing_set_trace_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)5884 tracing_set_trace_read(struct file *filp, char __user *ubuf,
5885 size_t cnt, loff_t *ppos)
5886 {
5887 struct trace_array *tr = filp->private_data;
5888 char buf[MAX_TRACER_SIZE+2];
5889 int r;
5890
5891 mutex_lock(&trace_types_lock);
5892 r = sprintf(buf, "%s\n", tr->current_trace->name);
5893 mutex_unlock(&trace_types_lock);
5894
5895 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5896 }
5897
tracer_init(struct tracer * t,struct trace_array * tr)5898 int tracer_init(struct tracer *t, struct trace_array *tr)
5899 {
5900 tracing_reset_online_cpus(&tr->array_buffer);
5901 return t->init(tr);
5902 }
5903
set_buffer_entries(struct array_buffer * buf,unsigned long val)5904 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
5905 {
5906 int cpu;
5907
5908 for_each_tracing_cpu(cpu)
5909 per_cpu_ptr(buf->data, cpu)->entries = val;
5910 }
5911
update_buffer_entries(struct array_buffer * buf,int cpu)5912 static void update_buffer_entries(struct array_buffer *buf, int cpu)
5913 {
5914 if (cpu == RING_BUFFER_ALL_CPUS) {
5915 set_buffer_entries(buf, ring_buffer_size(buf->buffer, 0));
5916 } else {
5917 per_cpu_ptr(buf->data, cpu)->entries = ring_buffer_size(buf->buffer, cpu);
5918 }
5919 }
5920
5921 #ifdef CONFIG_TRACER_MAX_TRACE
5922 /* resize @tr's buffer to the size of @size_tr's entries */
resize_buffer_duplicate_size(struct array_buffer * trace_buf,struct array_buffer * size_buf,int cpu_id)5923 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
5924 struct array_buffer *size_buf, int cpu_id)
5925 {
5926 int cpu, ret = 0;
5927
5928 if (cpu_id == RING_BUFFER_ALL_CPUS) {
5929 for_each_tracing_cpu(cpu) {
5930 ret = ring_buffer_resize(trace_buf->buffer,
5931 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
5932 if (ret < 0)
5933 break;
5934 per_cpu_ptr(trace_buf->data, cpu)->entries =
5935 per_cpu_ptr(size_buf->data, cpu)->entries;
5936 }
5937 } else {
5938 ret = ring_buffer_resize(trace_buf->buffer,
5939 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
5940 if (ret == 0)
5941 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
5942 per_cpu_ptr(size_buf->data, cpu_id)->entries;
5943 }
5944
5945 return ret;
5946 }
5947 #endif /* CONFIG_TRACER_MAX_TRACE */
5948
__tracing_resize_ring_buffer(struct trace_array * tr,unsigned long size,int cpu)5949 static int __tracing_resize_ring_buffer(struct trace_array *tr,
5950 unsigned long size, int cpu)
5951 {
5952 int ret;
5953
5954 /*
5955 * If kernel or user changes the size of the ring buffer
5956 * we use the size that was given, and we can forget about
5957 * expanding it later.
5958 */
5959 trace_set_ring_buffer_expanded(tr);
5960
5961 /* May be called before buffers are initialized */
5962 if (!tr->array_buffer.buffer)
5963 return 0;
5964
5965 /* Do not allow tracing while resizing ring buffer */
5966 tracing_stop_tr(tr);
5967
5968 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
5969 if (ret < 0)
5970 goto out_start;
5971
5972 #ifdef CONFIG_TRACER_MAX_TRACE
5973 if (!tr->allocated_snapshot)
5974 goto out;
5975
5976 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
5977 if (ret < 0) {
5978 int r = resize_buffer_duplicate_size(&tr->array_buffer,
5979 &tr->array_buffer, cpu);
5980 if (r < 0) {
5981 /*
5982 * AARGH! We are left with different
5983 * size max buffer!!!!
5984 * The max buffer is our "snapshot" buffer.
5985 * When a tracer needs a snapshot (one of the
5986 * latency tracers), it swaps the max buffer
5987 * with the saved snap shot. We succeeded to
5988 * update the size of the main buffer, but failed to
5989 * update the size of the max buffer. But when we tried
5990 * to reset the main buffer to the original size, we
5991 * failed there too. This is very unlikely to
5992 * happen, but if it does, warn and kill all
5993 * tracing.
5994 */
5995 WARN_ON(1);
5996 tracing_disabled = 1;
5997 }
5998 goto out_start;
5999 }
6000
6001 update_buffer_entries(&tr->max_buffer, cpu);
6002
6003 out:
6004 #endif /* CONFIG_TRACER_MAX_TRACE */
6005
6006 update_buffer_entries(&tr->array_buffer, cpu);
6007 out_start:
6008 tracing_start_tr(tr);
6009 return ret;
6010 }
6011
tracing_resize_ring_buffer(struct trace_array * tr,unsigned long size,int cpu_id)6012 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6013 unsigned long size, int cpu_id)
6014 {
6015 int ret;
6016
6017 mutex_lock(&trace_types_lock);
6018
6019 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6020 /* make sure, this cpu is enabled in the mask */
6021 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6022 ret = -EINVAL;
6023 goto out;
6024 }
6025 }
6026
6027 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6028 if (ret < 0)
6029 ret = -ENOMEM;
6030
6031 out:
6032 mutex_unlock(&trace_types_lock);
6033
6034 return ret;
6035 }
6036
6037
6038 /**
6039 * tracing_update_buffers - used by tracing facility to expand ring buffers
6040 * @tr: The tracing instance
6041 *
6042 * To save on memory when the tracing is never used on a system with it
6043 * configured in. The ring buffers are set to a minimum size. But once
6044 * a user starts to use the tracing facility, then they need to grow
6045 * to their default size.
6046 *
6047 * This function is to be called when a tracer is about to be used.
6048 */
tracing_update_buffers(struct trace_array * tr)6049 int tracing_update_buffers(struct trace_array *tr)
6050 {
6051 int ret = 0;
6052
6053 mutex_lock(&trace_types_lock);
6054 if (!tr->ring_buffer_expanded)
6055 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6056 RING_BUFFER_ALL_CPUS);
6057 mutex_unlock(&trace_types_lock);
6058
6059 return ret;
6060 }
6061
6062 struct trace_option_dentry;
6063
6064 static void
6065 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6066
6067 /*
6068 * Used to clear out the tracer before deletion of an instance.
6069 * Must have trace_types_lock held.
6070 */
tracing_set_nop(struct trace_array * tr)6071 static void tracing_set_nop(struct trace_array *tr)
6072 {
6073 if (tr->current_trace == &nop_trace)
6074 return;
6075
6076 tr->current_trace->enabled--;
6077
6078 if (tr->current_trace->reset)
6079 tr->current_trace->reset(tr);
6080
6081 tr->current_trace = &nop_trace;
6082 }
6083
6084 static bool tracer_options_updated;
6085
add_tracer_options(struct trace_array * tr,struct tracer * t)6086 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6087 {
6088 /* Only enable if the directory has been created already. */
6089 if (!tr->dir)
6090 return;
6091
6092 /* Only create trace option files after update_tracer_options finish */
6093 if (!tracer_options_updated)
6094 return;
6095
6096 create_trace_option_files(tr, t);
6097 }
6098
tracing_set_tracer(struct trace_array * tr,const char * buf)6099 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6100 {
6101 struct tracer *t;
6102 #ifdef CONFIG_TRACER_MAX_TRACE
6103 bool had_max_tr;
6104 #endif
6105 int ret = 0;
6106
6107 mutex_lock(&trace_types_lock);
6108
6109 if (!tr->ring_buffer_expanded) {
6110 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6111 RING_BUFFER_ALL_CPUS);
6112 if (ret < 0)
6113 goto out;
6114 ret = 0;
6115 }
6116
6117 for (t = trace_types; t; t = t->next) {
6118 if (strcmp(t->name, buf) == 0)
6119 break;
6120 }
6121 if (!t) {
6122 ret = -EINVAL;
6123 goto out;
6124 }
6125 if (t == tr->current_trace)
6126 goto out;
6127
6128 #ifdef CONFIG_TRACER_SNAPSHOT
6129 if (t->use_max_tr) {
6130 local_irq_disable();
6131 arch_spin_lock(&tr->max_lock);
6132 if (tr->cond_snapshot)
6133 ret = -EBUSY;
6134 arch_spin_unlock(&tr->max_lock);
6135 local_irq_enable();
6136 if (ret)
6137 goto out;
6138 }
6139 #endif
6140 /* Some tracers won't work on kernel command line */
6141 if (system_state < SYSTEM_RUNNING && t->noboot) {
6142 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6143 t->name);
6144 goto out;
6145 }
6146
6147 /* Some tracers are only allowed for the top level buffer */
6148 if (!trace_ok_for_array(t, tr)) {
6149 ret = -EINVAL;
6150 goto out;
6151 }
6152
6153 /* If trace pipe files are being read, we can't change the tracer */
6154 if (tr->trace_ref) {
6155 ret = -EBUSY;
6156 goto out;
6157 }
6158
6159 trace_branch_disable();
6160
6161 tr->current_trace->enabled--;
6162
6163 if (tr->current_trace->reset)
6164 tr->current_trace->reset(tr);
6165
6166 #ifdef CONFIG_TRACER_MAX_TRACE
6167 had_max_tr = tr->current_trace->use_max_tr;
6168
6169 /* Current trace needs to be nop_trace before synchronize_rcu */
6170 tr->current_trace = &nop_trace;
6171
6172 if (had_max_tr && !t->use_max_tr) {
6173 /*
6174 * We need to make sure that the update_max_tr sees that
6175 * current_trace changed to nop_trace to keep it from
6176 * swapping the buffers after we resize it.
6177 * The update_max_tr is called from interrupts disabled
6178 * so a synchronized_sched() is sufficient.
6179 */
6180 synchronize_rcu();
6181 free_snapshot(tr);
6182 tracing_disarm_snapshot(tr);
6183 }
6184
6185 if (!had_max_tr && t->use_max_tr) {
6186 ret = tracing_arm_snapshot_locked(tr);
6187 if (ret)
6188 goto out;
6189 }
6190 #else
6191 tr->current_trace = &nop_trace;
6192 #endif
6193
6194 if (t->init) {
6195 ret = tracer_init(t, tr);
6196 if (ret) {
6197 #ifdef CONFIG_TRACER_MAX_TRACE
6198 if (t->use_max_tr)
6199 tracing_disarm_snapshot(tr);
6200 #endif
6201 goto out;
6202 }
6203 }
6204
6205 tr->current_trace = t;
6206 tr->current_trace->enabled++;
6207 trace_branch_enable(tr);
6208 out:
6209 mutex_unlock(&trace_types_lock);
6210
6211 return ret;
6212 }
6213
6214 static ssize_t
tracing_set_trace_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)6215 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6216 size_t cnt, loff_t *ppos)
6217 {
6218 struct trace_array *tr = filp->private_data;
6219 char buf[MAX_TRACER_SIZE+1];
6220 char *name;
6221 size_t ret;
6222 int err;
6223
6224 ret = cnt;
6225
6226 if (cnt > MAX_TRACER_SIZE)
6227 cnt = MAX_TRACER_SIZE;
6228
6229 if (copy_from_user(buf, ubuf, cnt))
6230 return -EFAULT;
6231
6232 buf[cnt] = 0;
6233
6234 name = strim(buf);
6235
6236 err = tracing_set_tracer(tr, name);
6237 if (err)
6238 return err;
6239
6240 *ppos += ret;
6241
6242 return ret;
6243 }
6244
6245 static ssize_t
tracing_nsecs_read(unsigned long * ptr,char __user * ubuf,size_t cnt,loff_t * ppos)6246 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6247 size_t cnt, loff_t *ppos)
6248 {
6249 char buf[64];
6250 int r;
6251
6252 r = snprintf(buf, sizeof(buf), "%ld\n",
6253 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6254 if (r > sizeof(buf))
6255 r = sizeof(buf);
6256 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6257 }
6258
6259 static ssize_t
tracing_nsecs_write(unsigned long * ptr,const char __user * ubuf,size_t cnt,loff_t * ppos)6260 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6261 size_t cnt, loff_t *ppos)
6262 {
6263 unsigned long val;
6264 int ret;
6265
6266 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6267 if (ret)
6268 return ret;
6269
6270 *ptr = val * 1000;
6271
6272 return cnt;
6273 }
6274
6275 static ssize_t
tracing_thresh_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)6276 tracing_thresh_read(struct file *filp, char __user *ubuf,
6277 size_t cnt, loff_t *ppos)
6278 {
6279 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6280 }
6281
6282 static ssize_t
tracing_thresh_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)6283 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6284 size_t cnt, loff_t *ppos)
6285 {
6286 struct trace_array *tr = filp->private_data;
6287 int ret;
6288
6289 mutex_lock(&trace_types_lock);
6290 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6291 if (ret < 0)
6292 goto out;
6293
6294 if (tr->current_trace->update_thresh) {
6295 ret = tr->current_trace->update_thresh(tr);
6296 if (ret < 0)
6297 goto out;
6298 }
6299
6300 ret = cnt;
6301 out:
6302 mutex_unlock(&trace_types_lock);
6303
6304 return ret;
6305 }
6306
6307 #ifdef CONFIG_TRACER_MAX_TRACE
6308
6309 static ssize_t
tracing_max_lat_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)6310 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6311 size_t cnt, loff_t *ppos)
6312 {
6313 struct trace_array *tr = filp->private_data;
6314
6315 return tracing_nsecs_read(&tr->max_latency, ubuf, cnt, ppos);
6316 }
6317
6318 static ssize_t
tracing_max_lat_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)6319 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6320 size_t cnt, loff_t *ppos)
6321 {
6322 struct trace_array *tr = filp->private_data;
6323
6324 return tracing_nsecs_write(&tr->max_latency, ubuf, cnt, ppos);
6325 }
6326
6327 #endif
6328
open_pipe_on_cpu(struct trace_array * tr,int cpu)6329 static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6330 {
6331 if (cpu == RING_BUFFER_ALL_CPUS) {
6332 if (cpumask_empty(tr->pipe_cpumask)) {
6333 cpumask_setall(tr->pipe_cpumask);
6334 return 0;
6335 }
6336 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6337 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6338 return 0;
6339 }
6340 return -EBUSY;
6341 }
6342
close_pipe_on_cpu(struct trace_array * tr,int cpu)6343 static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6344 {
6345 if (cpu == RING_BUFFER_ALL_CPUS) {
6346 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6347 cpumask_clear(tr->pipe_cpumask);
6348 } else {
6349 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6350 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6351 }
6352 }
6353
tracing_open_pipe(struct inode * inode,struct file * filp)6354 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6355 {
6356 struct trace_array *tr = inode->i_private;
6357 struct trace_iterator *iter;
6358 int cpu;
6359 int ret;
6360
6361 ret = tracing_check_open_get_tr(tr);
6362 if (ret)
6363 return ret;
6364
6365 mutex_lock(&trace_types_lock);
6366 cpu = tracing_get_cpu(inode);
6367 ret = open_pipe_on_cpu(tr, cpu);
6368 if (ret)
6369 goto fail_pipe_on_cpu;
6370
6371 /* create a buffer to store the information to pass to userspace */
6372 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6373 if (!iter) {
6374 ret = -ENOMEM;
6375 goto fail_alloc_iter;
6376 }
6377
6378 trace_seq_init(&iter->seq);
6379 iter->trace = tr->current_trace;
6380
6381 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6382 ret = -ENOMEM;
6383 goto fail;
6384 }
6385
6386 /* trace pipe does not show start of buffer */
6387 cpumask_setall(iter->started);
6388
6389 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6390 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6391
6392 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6393 if (trace_clocks[tr->clock_id].in_ns)
6394 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6395
6396 iter->tr = tr;
6397 iter->array_buffer = &tr->array_buffer;
6398 iter->cpu_file = cpu;
6399 mutex_init(&iter->mutex);
6400 filp->private_data = iter;
6401
6402 if (iter->trace->pipe_open)
6403 iter->trace->pipe_open(iter);
6404
6405 nonseekable_open(inode, filp);
6406
6407 tr->trace_ref++;
6408
6409 mutex_unlock(&trace_types_lock);
6410 return ret;
6411
6412 fail:
6413 kfree(iter);
6414 fail_alloc_iter:
6415 close_pipe_on_cpu(tr, cpu);
6416 fail_pipe_on_cpu:
6417 __trace_array_put(tr);
6418 mutex_unlock(&trace_types_lock);
6419 return ret;
6420 }
6421
tracing_release_pipe(struct inode * inode,struct file * file)6422 static int tracing_release_pipe(struct inode *inode, struct file *file)
6423 {
6424 struct trace_iterator *iter = file->private_data;
6425 struct trace_array *tr = inode->i_private;
6426
6427 mutex_lock(&trace_types_lock);
6428
6429 tr->trace_ref--;
6430
6431 if (iter->trace->pipe_close)
6432 iter->trace->pipe_close(iter);
6433 close_pipe_on_cpu(tr, iter->cpu_file);
6434 mutex_unlock(&trace_types_lock);
6435
6436 free_trace_iter_content(iter);
6437 kfree(iter);
6438
6439 trace_array_put(tr);
6440
6441 return 0;
6442 }
6443
6444 static __poll_t
trace_poll(struct trace_iterator * iter,struct file * filp,poll_table * poll_table)6445 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6446 {
6447 struct trace_array *tr = iter->tr;
6448
6449 /* Iterators are static, they should be filled or empty */
6450 if (trace_buffer_iter(iter, iter->cpu_file))
6451 return EPOLLIN | EPOLLRDNORM;
6452
6453 if (tr->trace_flags & TRACE_ITER_BLOCK)
6454 /*
6455 * Always select as readable when in blocking mode
6456 */
6457 return EPOLLIN | EPOLLRDNORM;
6458 else
6459 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6460 filp, poll_table, iter->tr->buffer_percent);
6461 }
6462
6463 static __poll_t
tracing_poll_pipe(struct file * filp,poll_table * poll_table)6464 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6465 {
6466 struct trace_iterator *iter = filp->private_data;
6467
6468 return trace_poll(iter, filp, poll_table);
6469 }
6470
6471 /* Must be called with iter->mutex held. */
tracing_wait_pipe(struct file * filp)6472 static int tracing_wait_pipe(struct file *filp)
6473 {
6474 struct trace_iterator *iter = filp->private_data;
6475 int ret;
6476
6477 while (trace_empty(iter)) {
6478
6479 if ((filp->f_flags & O_NONBLOCK)) {
6480 return -EAGAIN;
6481 }
6482
6483 /*
6484 * We block until we read something and tracing is disabled.
6485 * We still block if tracing is disabled, but we have never
6486 * read anything. This allows a user to cat this file, and
6487 * then enable tracing. But after we have read something,
6488 * we give an EOF when tracing is again disabled.
6489 *
6490 * iter->pos will be 0 if we haven't read anything.
6491 */
6492 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6493 break;
6494
6495 mutex_unlock(&iter->mutex);
6496
6497 ret = wait_on_pipe(iter, 0);
6498
6499 mutex_lock(&iter->mutex);
6500
6501 if (ret)
6502 return ret;
6503 }
6504
6505 return 1;
6506 }
6507
6508 /*
6509 * Consumer reader.
6510 */
6511 static ssize_t
tracing_read_pipe(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)6512 tracing_read_pipe(struct file *filp, char __user *ubuf,
6513 size_t cnt, loff_t *ppos)
6514 {
6515 struct trace_iterator *iter = filp->private_data;
6516 ssize_t sret;
6517
6518 /*
6519 * Avoid more than one consumer on a single file descriptor
6520 * This is just a matter of traces coherency, the ring buffer itself
6521 * is protected.
6522 */
6523 mutex_lock(&iter->mutex);
6524
6525 /* return any leftover data */
6526 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6527 if (sret != -EBUSY)
6528 goto out;
6529
6530 trace_seq_init(&iter->seq);
6531
6532 if (iter->trace->read) {
6533 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6534 if (sret)
6535 goto out;
6536 }
6537
6538 waitagain:
6539 sret = tracing_wait_pipe(filp);
6540 if (sret <= 0)
6541 goto out;
6542
6543 /* stop when tracing is finished */
6544 if (trace_empty(iter)) {
6545 sret = 0;
6546 goto out;
6547 }
6548
6549 if (cnt >= TRACE_SEQ_BUFFER_SIZE)
6550 cnt = TRACE_SEQ_BUFFER_SIZE - 1;
6551
6552 /* reset all but tr, trace, and overruns */
6553 trace_iterator_reset(iter);
6554 cpumask_clear(iter->started);
6555 trace_seq_init(&iter->seq);
6556
6557 trace_event_read_lock();
6558 trace_access_lock(iter->cpu_file);
6559 while (trace_find_next_entry_inc(iter) != NULL) {
6560 enum print_line_t ret;
6561 int save_len = iter->seq.seq.len;
6562
6563 ret = print_trace_line(iter);
6564 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6565 /*
6566 * If one print_trace_line() fills entire trace_seq in one shot,
6567 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6568 * In this case, we need to consume it, otherwise, loop will peek
6569 * this event next time, resulting in an infinite loop.
6570 */
6571 if (save_len == 0) {
6572 iter->seq.full = 0;
6573 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6574 trace_consume(iter);
6575 break;
6576 }
6577
6578 /* In other cases, don't print partial lines */
6579 iter->seq.seq.len = save_len;
6580 break;
6581 }
6582 if (ret != TRACE_TYPE_NO_CONSUME)
6583 trace_consume(iter);
6584
6585 if (trace_seq_used(&iter->seq) >= cnt)
6586 break;
6587
6588 /*
6589 * Setting the full flag means we reached the trace_seq buffer
6590 * size and we should leave by partial output condition above.
6591 * One of the trace_seq_* functions is not used properly.
6592 */
6593 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6594 iter->ent->type);
6595 }
6596 trace_access_unlock(iter->cpu_file);
6597 trace_event_read_unlock();
6598
6599 /* Now copy what we have to the user */
6600 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6601 if (iter->seq.readpos >= trace_seq_used(&iter->seq))
6602 trace_seq_init(&iter->seq);
6603
6604 /*
6605 * If there was nothing to send to user, in spite of consuming trace
6606 * entries, go back to wait for more entries.
6607 */
6608 if (sret == -EBUSY)
6609 goto waitagain;
6610
6611 out:
6612 mutex_unlock(&iter->mutex);
6613
6614 return sret;
6615 }
6616
tracing_spd_release_pipe(struct splice_pipe_desc * spd,unsigned int idx)6617 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6618 unsigned int idx)
6619 {
6620 __free_page(spd->pages[idx]);
6621 }
6622
6623 static size_t
tracing_fill_pipe_page(size_t rem,struct trace_iterator * iter)6624 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6625 {
6626 size_t count;
6627 int save_len;
6628 int ret;
6629
6630 /* Seq buffer is page-sized, exactly what we need. */
6631 for (;;) {
6632 save_len = iter->seq.seq.len;
6633 ret = print_trace_line(iter);
6634
6635 if (trace_seq_has_overflowed(&iter->seq)) {
6636 iter->seq.seq.len = save_len;
6637 break;
6638 }
6639
6640 /*
6641 * This should not be hit, because it should only
6642 * be set if the iter->seq overflowed. But check it
6643 * anyway to be safe.
6644 */
6645 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6646 iter->seq.seq.len = save_len;
6647 break;
6648 }
6649
6650 count = trace_seq_used(&iter->seq) - save_len;
6651 if (rem < count) {
6652 rem = 0;
6653 iter->seq.seq.len = save_len;
6654 break;
6655 }
6656
6657 if (ret != TRACE_TYPE_NO_CONSUME)
6658 trace_consume(iter);
6659 rem -= count;
6660 if (!trace_find_next_entry_inc(iter)) {
6661 rem = 0;
6662 iter->ent = NULL;
6663 break;
6664 }
6665 }
6666
6667 return rem;
6668 }
6669
tracing_splice_read_pipe(struct file * filp,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)6670 static ssize_t tracing_splice_read_pipe(struct file *filp,
6671 loff_t *ppos,
6672 struct pipe_inode_info *pipe,
6673 size_t len,
6674 unsigned int flags)
6675 {
6676 struct page *pages_def[PIPE_DEF_BUFFERS];
6677 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6678 struct trace_iterator *iter = filp->private_data;
6679 struct splice_pipe_desc spd = {
6680 .pages = pages_def,
6681 .partial = partial_def,
6682 .nr_pages = 0, /* This gets updated below. */
6683 .nr_pages_max = PIPE_DEF_BUFFERS,
6684 .ops = &default_pipe_buf_ops,
6685 .spd_release = tracing_spd_release_pipe,
6686 };
6687 ssize_t ret;
6688 size_t rem;
6689 unsigned int i;
6690
6691 if (splice_grow_spd(pipe, &spd))
6692 return -ENOMEM;
6693
6694 mutex_lock(&iter->mutex);
6695
6696 if (iter->trace->splice_read) {
6697 ret = iter->trace->splice_read(iter, filp,
6698 ppos, pipe, len, flags);
6699 if (ret)
6700 goto out_err;
6701 }
6702
6703 ret = tracing_wait_pipe(filp);
6704 if (ret <= 0)
6705 goto out_err;
6706
6707 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6708 ret = -EFAULT;
6709 goto out_err;
6710 }
6711
6712 trace_event_read_lock();
6713 trace_access_lock(iter->cpu_file);
6714
6715 /* Fill as many pages as possible. */
6716 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6717 spd.pages[i] = alloc_page(GFP_KERNEL);
6718 if (!spd.pages[i])
6719 break;
6720
6721 rem = tracing_fill_pipe_page(rem, iter);
6722
6723 /* Copy the data into the page, so we can start over. */
6724 ret = trace_seq_to_buffer(&iter->seq,
6725 page_address(spd.pages[i]),
6726 trace_seq_used(&iter->seq));
6727 if (ret < 0) {
6728 __free_page(spd.pages[i]);
6729 break;
6730 }
6731 spd.partial[i].offset = 0;
6732 spd.partial[i].len = trace_seq_used(&iter->seq);
6733
6734 trace_seq_init(&iter->seq);
6735 }
6736
6737 trace_access_unlock(iter->cpu_file);
6738 trace_event_read_unlock();
6739 mutex_unlock(&iter->mutex);
6740
6741 spd.nr_pages = i;
6742
6743 if (i)
6744 ret = splice_to_pipe(pipe, &spd);
6745 else
6746 ret = 0;
6747 out:
6748 splice_shrink_spd(&spd);
6749 return ret;
6750
6751 out_err:
6752 mutex_unlock(&iter->mutex);
6753 goto out;
6754 }
6755
6756 static ssize_t
tracing_entries_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)6757 tracing_entries_read(struct file *filp, char __user *ubuf,
6758 size_t cnt, loff_t *ppos)
6759 {
6760 struct inode *inode = file_inode(filp);
6761 struct trace_array *tr = inode->i_private;
6762 int cpu = tracing_get_cpu(inode);
6763 char buf[64];
6764 int r = 0;
6765 ssize_t ret;
6766
6767 mutex_lock(&trace_types_lock);
6768
6769 if (cpu == RING_BUFFER_ALL_CPUS) {
6770 int cpu, buf_size_same;
6771 unsigned long size;
6772
6773 size = 0;
6774 buf_size_same = 1;
6775 /* check if all cpu sizes are same */
6776 for_each_tracing_cpu(cpu) {
6777 /* fill in the size from first enabled cpu */
6778 if (size == 0)
6779 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6780 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6781 buf_size_same = 0;
6782 break;
6783 }
6784 }
6785
6786 if (buf_size_same) {
6787 if (!tr->ring_buffer_expanded)
6788 r = sprintf(buf, "%lu (expanded: %lu)\n",
6789 size >> 10,
6790 trace_buf_size >> 10);
6791 else
6792 r = sprintf(buf, "%lu\n", size >> 10);
6793 } else
6794 r = sprintf(buf, "X\n");
6795 } else
6796 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6797
6798 mutex_unlock(&trace_types_lock);
6799
6800 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6801 return ret;
6802 }
6803
6804 static ssize_t
tracing_entries_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)6805 tracing_entries_write(struct file *filp, const char __user *ubuf,
6806 size_t cnt, loff_t *ppos)
6807 {
6808 struct inode *inode = file_inode(filp);
6809 struct trace_array *tr = inode->i_private;
6810 unsigned long val;
6811 int ret;
6812
6813 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6814 if (ret)
6815 return ret;
6816
6817 /* must have at least 1 entry */
6818 if (!val)
6819 return -EINVAL;
6820
6821 /* value is in KB */
6822 val <<= 10;
6823 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6824 if (ret < 0)
6825 return ret;
6826
6827 *ppos += cnt;
6828
6829 return cnt;
6830 }
6831
6832 static ssize_t
tracing_total_entries_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)6833 tracing_total_entries_read(struct file *filp, char __user *ubuf,
6834 size_t cnt, loff_t *ppos)
6835 {
6836 struct trace_array *tr = filp->private_data;
6837 char buf[64];
6838 int r, cpu;
6839 unsigned long size = 0, expanded_size = 0;
6840
6841 mutex_lock(&trace_types_lock);
6842 for_each_tracing_cpu(cpu) {
6843 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6844 if (!tr->ring_buffer_expanded)
6845 expanded_size += trace_buf_size >> 10;
6846 }
6847 if (tr->ring_buffer_expanded)
6848 r = sprintf(buf, "%lu\n", size);
6849 else
6850 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6851 mutex_unlock(&trace_types_lock);
6852
6853 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6854 }
6855
6856 static ssize_t
tracing_free_buffer_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)6857 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
6858 size_t cnt, loff_t *ppos)
6859 {
6860 /*
6861 * There is no need to read what the user has written, this function
6862 * is just to make sure that there is no error when "echo" is used
6863 */
6864
6865 *ppos += cnt;
6866
6867 return cnt;
6868 }
6869
6870 static int
tracing_free_buffer_release(struct inode * inode,struct file * filp)6871 tracing_free_buffer_release(struct inode *inode, struct file *filp)
6872 {
6873 struct trace_array *tr = inode->i_private;
6874
6875 /* disable tracing ? */
6876 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
6877 tracer_tracing_off(tr);
6878 /* resize the ring buffer to 0 */
6879 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
6880
6881 trace_array_put(tr);
6882
6883 return 0;
6884 }
6885
6886 #define TRACE_MARKER_MAX_SIZE 4096
6887
6888 static ssize_t
tracing_mark_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * fpos)6889 tracing_mark_write(struct file *filp, const char __user *ubuf,
6890 size_t cnt, loff_t *fpos)
6891 {
6892 struct trace_array *tr = filp->private_data;
6893 struct ring_buffer_event *event;
6894 enum event_trigger_type tt = ETT_NONE;
6895 struct trace_buffer *buffer;
6896 struct print_entry *entry;
6897 int meta_size;
6898 ssize_t written;
6899 size_t size;
6900 int len;
6901
6902 /* Used in tracing_mark_raw_write() as well */
6903 #define FAULTED_STR "<faulted>"
6904 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
6905
6906 if (tracing_disabled)
6907 return -EINVAL;
6908
6909 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6910 return -EINVAL;
6911
6912 if ((ssize_t)cnt < 0)
6913 return -EINVAL;
6914
6915 if (cnt > TRACE_MARKER_MAX_SIZE)
6916 cnt = TRACE_MARKER_MAX_SIZE;
6917
6918 meta_size = sizeof(*entry) + 2; /* add '\0' and possible '\n' */
6919 again:
6920 size = cnt + meta_size;
6921
6922 /* If less than "<faulted>", then make sure we can still add that */
6923 if (cnt < FAULTED_SIZE)
6924 size += FAULTED_SIZE - cnt;
6925
6926 buffer = tr->array_buffer.buffer;
6927 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
6928 tracing_gen_ctx());
6929 if (unlikely(!event)) {
6930 /*
6931 * If the size was greater than what was allowed, then
6932 * make it smaller and try again.
6933 */
6934 if (size > ring_buffer_max_event_size(buffer)) {
6935 /* cnt < FAULTED size should never be bigger than max */
6936 if (WARN_ON_ONCE(cnt < FAULTED_SIZE))
6937 return -EBADF;
6938 cnt = ring_buffer_max_event_size(buffer) - meta_size;
6939 /* The above should only happen once */
6940 if (WARN_ON_ONCE(cnt + meta_size == size))
6941 return -EBADF;
6942 goto again;
6943 }
6944
6945 /* Ring buffer disabled, return as if not open for write */
6946 return -EBADF;
6947 }
6948
6949 entry = ring_buffer_event_data(event);
6950 entry->ip = _THIS_IP_;
6951
6952 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
6953 if (len) {
6954 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6955 cnt = FAULTED_SIZE;
6956 written = -EFAULT;
6957 } else
6958 written = cnt;
6959
6960 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
6961 /* do not add \n before testing triggers, but add \0 */
6962 entry->buf[cnt] = '\0';
6963 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
6964 }
6965
6966 if (entry->buf[cnt - 1] != '\n') {
6967 entry->buf[cnt] = '\n';
6968 entry->buf[cnt + 1] = '\0';
6969 } else
6970 entry->buf[cnt] = '\0';
6971
6972 if (static_branch_unlikely(&trace_marker_exports_enabled))
6973 ftrace_exports(event, TRACE_EXPORT_MARKER);
6974 __buffer_unlock_commit(buffer, event);
6975
6976 if (tt)
6977 event_triggers_post_call(tr->trace_marker_file, tt);
6978
6979 return written;
6980 }
6981
6982 static ssize_t
tracing_mark_raw_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * fpos)6983 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
6984 size_t cnt, loff_t *fpos)
6985 {
6986 struct trace_array *tr = filp->private_data;
6987 struct ring_buffer_event *event;
6988 struct trace_buffer *buffer;
6989 struct raw_data_entry *entry;
6990 ssize_t written;
6991 int size;
6992 int len;
6993
6994 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
6995
6996 if (tracing_disabled)
6997 return -EINVAL;
6998
6999 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7000 return -EINVAL;
7001
7002 /* The marker must at least have a tag id */
7003 if (cnt < sizeof(unsigned int))
7004 return -EINVAL;
7005
7006 size = sizeof(*entry) + cnt;
7007 if (cnt < FAULT_SIZE_ID)
7008 size += FAULT_SIZE_ID - cnt;
7009
7010 buffer = tr->array_buffer.buffer;
7011
7012 if (size > ring_buffer_max_event_size(buffer))
7013 return -EINVAL;
7014
7015 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7016 tracing_gen_ctx());
7017 if (!event)
7018 /* Ring buffer disabled, return as if not open for write */
7019 return -EBADF;
7020
7021 entry = ring_buffer_event_data(event);
7022
7023 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7024 if (len) {
7025 entry->id = -1;
7026 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7027 written = -EFAULT;
7028 } else
7029 written = cnt;
7030
7031 __buffer_unlock_commit(buffer, event);
7032
7033 return written;
7034 }
7035
tracing_clock_show(struct seq_file * m,void * v)7036 static int tracing_clock_show(struct seq_file *m, void *v)
7037 {
7038 struct trace_array *tr = m->private;
7039 int i;
7040
7041 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7042 seq_printf(m,
7043 "%s%s%s%s", i ? " " : "",
7044 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7045 i == tr->clock_id ? "]" : "");
7046 seq_putc(m, '\n');
7047
7048 return 0;
7049 }
7050
tracing_set_clock(struct trace_array * tr,const char * clockstr)7051 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7052 {
7053 int i;
7054
7055 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7056 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7057 break;
7058 }
7059 if (i == ARRAY_SIZE(trace_clocks))
7060 return -EINVAL;
7061
7062 mutex_lock(&trace_types_lock);
7063
7064 tr->clock_id = i;
7065
7066 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7067
7068 /*
7069 * New clock may not be consistent with the previous clock.
7070 * Reset the buffer so that it doesn't have incomparable timestamps.
7071 */
7072 tracing_reset_online_cpus(&tr->array_buffer);
7073
7074 #ifdef CONFIG_TRACER_MAX_TRACE
7075 if (tr->max_buffer.buffer)
7076 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7077 tracing_reset_online_cpus(&tr->max_buffer);
7078 #endif
7079
7080 mutex_unlock(&trace_types_lock);
7081
7082 return 0;
7083 }
7084
tracing_clock_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * fpos)7085 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7086 size_t cnt, loff_t *fpos)
7087 {
7088 struct seq_file *m = filp->private_data;
7089 struct trace_array *tr = m->private;
7090 char buf[64];
7091 const char *clockstr;
7092 int ret;
7093
7094 if (cnt >= sizeof(buf))
7095 return -EINVAL;
7096
7097 if (copy_from_user(buf, ubuf, cnt))
7098 return -EFAULT;
7099
7100 buf[cnt] = 0;
7101
7102 clockstr = strstrip(buf);
7103
7104 ret = tracing_set_clock(tr, clockstr);
7105 if (ret)
7106 return ret;
7107
7108 *fpos += cnt;
7109
7110 return cnt;
7111 }
7112
tracing_clock_open(struct inode * inode,struct file * file)7113 static int tracing_clock_open(struct inode *inode, struct file *file)
7114 {
7115 struct trace_array *tr = inode->i_private;
7116 int ret;
7117
7118 ret = tracing_check_open_get_tr(tr);
7119 if (ret)
7120 return ret;
7121
7122 ret = single_open(file, tracing_clock_show, inode->i_private);
7123 if (ret < 0)
7124 trace_array_put(tr);
7125
7126 return ret;
7127 }
7128
tracing_time_stamp_mode_show(struct seq_file * m,void * v)7129 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7130 {
7131 struct trace_array *tr = m->private;
7132
7133 mutex_lock(&trace_types_lock);
7134
7135 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7136 seq_puts(m, "delta [absolute]\n");
7137 else
7138 seq_puts(m, "[delta] absolute\n");
7139
7140 mutex_unlock(&trace_types_lock);
7141
7142 return 0;
7143 }
7144
tracing_time_stamp_mode_open(struct inode * inode,struct file * file)7145 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7146 {
7147 struct trace_array *tr = inode->i_private;
7148 int ret;
7149
7150 ret = tracing_check_open_get_tr(tr);
7151 if (ret)
7152 return ret;
7153
7154 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7155 if (ret < 0)
7156 trace_array_put(tr);
7157
7158 return ret;
7159 }
7160
tracing_event_time_stamp(struct trace_buffer * buffer,struct ring_buffer_event * rbe)7161 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7162 {
7163 if (rbe == this_cpu_read(trace_buffered_event))
7164 return ring_buffer_time_stamp(buffer);
7165
7166 return ring_buffer_event_time_stamp(buffer, rbe);
7167 }
7168
7169 /*
7170 * Set or disable using the per CPU trace_buffer_event when possible.
7171 */
tracing_set_filter_buffering(struct trace_array * tr,bool set)7172 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7173 {
7174 int ret = 0;
7175
7176 mutex_lock(&trace_types_lock);
7177
7178 if (set && tr->no_filter_buffering_ref++)
7179 goto out;
7180
7181 if (!set) {
7182 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7183 ret = -EINVAL;
7184 goto out;
7185 }
7186
7187 --tr->no_filter_buffering_ref;
7188 }
7189 out:
7190 mutex_unlock(&trace_types_lock);
7191
7192 return ret;
7193 }
7194
7195 struct ftrace_buffer_info {
7196 struct trace_iterator iter;
7197 void *spare;
7198 unsigned int spare_cpu;
7199 unsigned int spare_size;
7200 unsigned int read;
7201 };
7202
7203 #ifdef CONFIG_TRACER_SNAPSHOT
tracing_snapshot_open(struct inode * inode,struct file * file)7204 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7205 {
7206 struct trace_array *tr = inode->i_private;
7207 struct trace_iterator *iter;
7208 struct seq_file *m;
7209 int ret;
7210
7211 ret = tracing_check_open_get_tr(tr);
7212 if (ret)
7213 return ret;
7214
7215 if (file->f_mode & FMODE_READ) {
7216 iter = __tracing_open(inode, file, true);
7217 if (IS_ERR(iter))
7218 ret = PTR_ERR(iter);
7219 } else {
7220 /* Writes still need the seq_file to hold the private data */
7221 ret = -ENOMEM;
7222 m = kzalloc(sizeof(*m), GFP_KERNEL);
7223 if (!m)
7224 goto out;
7225 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7226 if (!iter) {
7227 kfree(m);
7228 goto out;
7229 }
7230 ret = 0;
7231
7232 iter->tr = tr;
7233 iter->array_buffer = &tr->max_buffer;
7234 iter->cpu_file = tracing_get_cpu(inode);
7235 m->private = iter;
7236 file->private_data = m;
7237 }
7238 out:
7239 if (ret < 0)
7240 trace_array_put(tr);
7241
7242 return ret;
7243 }
7244
tracing_swap_cpu_buffer(void * tr)7245 static void tracing_swap_cpu_buffer(void *tr)
7246 {
7247 update_max_tr_single((struct trace_array *)tr, current, smp_processor_id());
7248 }
7249
7250 static ssize_t
tracing_snapshot_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)7251 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7252 loff_t *ppos)
7253 {
7254 struct seq_file *m = filp->private_data;
7255 struct trace_iterator *iter = m->private;
7256 struct trace_array *tr = iter->tr;
7257 unsigned long val;
7258 int ret;
7259
7260 ret = tracing_update_buffers(tr);
7261 if (ret < 0)
7262 return ret;
7263
7264 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7265 if (ret)
7266 return ret;
7267
7268 mutex_lock(&trace_types_lock);
7269
7270 if (tr->current_trace->use_max_tr) {
7271 ret = -EBUSY;
7272 goto out;
7273 }
7274
7275 local_irq_disable();
7276 arch_spin_lock(&tr->max_lock);
7277 if (tr->cond_snapshot)
7278 ret = -EBUSY;
7279 arch_spin_unlock(&tr->max_lock);
7280 local_irq_enable();
7281 if (ret)
7282 goto out;
7283
7284 switch (val) {
7285 case 0:
7286 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7287 ret = -EINVAL;
7288 break;
7289 }
7290 if (tr->allocated_snapshot)
7291 free_snapshot(tr);
7292 break;
7293 case 1:
7294 /* Only allow per-cpu swap if the ring buffer supports it */
7295 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7296 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7297 ret = -EINVAL;
7298 break;
7299 }
7300 #endif
7301 if (tr->allocated_snapshot)
7302 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7303 &tr->array_buffer, iter->cpu_file);
7304
7305 ret = tracing_arm_snapshot_locked(tr);
7306 if (ret)
7307 break;
7308
7309 /* Now, we're going to swap */
7310 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
7311 local_irq_disable();
7312 update_max_tr(tr, current, smp_processor_id(), NULL);
7313 local_irq_enable();
7314 } else {
7315 smp_call_function_single(iter->cpu_file, tracing_swap_cpu_buffer,
7316 (void *)tr, 1);
7317 }
7318 tracing_disarm_snapshot(tr);
7319 break;
7320 default:
7321 if (tr->allocated_snapshot) {
7322 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7323 tracing_reset_online_cpus(&tr->max_buffer);
7324 else
7325 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7326 }
7327 break;
7328 }
7329
7330 if (ret >= 0) {
7331 *ppos += cnt;
7332 ret = cnt;
7333 }
7334 out:
7335 mutex_unlock(&trace_types_lock);
7336 return ret;
7337 }
7338
tracing_snapshot_release(struct inode * inode,struct file * file)7339 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7340 {
7341 struct seq_file *m = file->private_data;
7342 int ret;
7343
7344 ret = tracing_release(inode, file);
7345
7346 if (file->f_mode & FMODE_READ)
7347 return ret;
7348
7349 /* If write only, the seq_file is just a stub */
7350 if (m)
7351 kfree(m->private);
7352 kfree(m);
7353
7354 return 0;
7355 }
7356
7357 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7358 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7359 size_t count, loff_t *ppos);
7360 static int tracing_buffers_release(struct inode *inode, struct file *file);
7361 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7362 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7363
snapshot_raw_open(struct inode * inode,struct file * filp)7364 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7365 {
7366 struct ftrace_buffer_info *info;
7367 int ret;
7368
7369 /* The following checks for tracefs lockdown */
7370 ret = tracing_buffers_open(inode, filp);
7371 if (ret < 0)
7372 return ret;
7373
7374 info = filp->private_data;
7375
7376 if (info->iter.trace->use_max_tr) {
7377 tracing_buffers_release(inode, filp);
7378 return -EBUSY;
7379 }
7380
7381 info->iter.snapshot = true;
7382 info->iter.array_buffer = &info->iter.tr->max_buffer;
7383
7384 return ret;
7385 }
7386
7387 #endif /* CONFIG_TRACER_SNAPSHOT */
7388
7389
7390 static const struct file_operations tracing_thresh_fops = {
7391 .open = tracing_open_generic,
7392 .read = tracing_thresh_read,
7393 .write = tracing_thresh_write,
7394 .llseek = generic_file_llseek,
7395 };
7396
7397 #ifdef CONFIG_TRACER_MAX_TRACE
7398 static const struct file_operations tracing_max_lat_fops = {
7399 .open = tracing_open_generic_tr,
7400 .read = tracing_max_lat_read,
7401 .write = tracing_max_lat_write,
7402 .llseek = generic_file_llseek,
7403 .release = tracing_release_generic_tr,
7404 };
7405 #endif
7406
7407 static const struct file_operations set_tracer_fops = {
7408 .open = tracing_open_generic_tr,
7409 .read = tracing_set_trace_read,
7410 .write = tracing_set_trace_write,
7411 .llseek = generic_file_llseek,
7412 .release = tracing_release_generic_tr,
7413 };
7414
7415 static const struct file_operations tracing_pipe_fops = {
7416 .open = tracing_open_pipe,
7417 .poll = tracing_poll_pipe,
7418 .read = tracing_read_pipe,
7419 .splice_read = tracing_splice_read_pipe,
7420 .release = tracing_release_pipe,
7421 .llseek = no_llseek,
7422 };
7423
7424 static const struct file_operations tracing_entries_fops = {
7425 .open = tracing_open_generic_tr,
7426 .read = tracing_entries_read,
7427 .write = tracing_entries_write,
7428 .llseek = generic_file_llseek,
7429 .release = tracing_release_generic_tr,
7430 };
7431
7432 static const struct file_operations tracing_total_entries_fops = {
7433 .open = tracing_open_generic_tr,
7434 .read = tracing_total_entries_read,
7435 .llseek = generic_file_llseek,
7436 .release = tracing_release_generic_tr,
7437 };
7438
7439 static const struct file_operations tracing_free_buffer_fops = {
7440 .open = tracing_open_generic_tr,
7441 .write = tracing_free_buffer_write,
7442 .release = tracing_free_buffer_release,
7443 };
7444
7445 static const struct file_operations tracing_mark_fops = {
7446 .open = tracing_mark_open,
7447 .write = tracing_mark_write,
7448 .release = tracing_release_generic_tr,
7449 };
7450
7451 static const struct file_operations tracing_mark_raw_fops = {
7452 .open = tracing_mark_open,
7453 .write = tracing_mark_raw_write,
7454 .release = tracing_release_generic_tr,
7455 };
7456
7457 static const struct file_operations trace_clock_fops = {
7458 .open = tracing_clock_open,
7459 .read = seq_read,
7460 .llseek = seq_lseek,
7461 .release = tracing_single_release_tr,
7462 .write = tracing_clock_write,
7463 };
7464
7465 static const struct file_operations trace_time_stamp_mode_fops = {
7466 .open = tracing_time_stamp_mode_open,
7467 .read = seq_read,
7468 .llseek = seq_lseek,
7469 .release = tracing_single_release_tr,
7470 };
7471
7472 #ifdef CONFIG_TRACER_SNAPSHOT
7473 static const struct file_operations snapshot_fops = {
7474 .open = tracing_snapshot_open,
7475 .read = seq_read,
7476 .write = tracing_snapshot_write,
7477 .llseek = tracing_lseek,
7478 .release = tracing_snapshot_release,
7479 };
7480
7481 static const struct file_operations snapshot_raw_fops = {
7482 .open = snapshot_raw_open,
7483 .read = tracing_buffers_read,
7484 .release = tracing_buffers_release,
7485 .splice_read = tracing_buffers_splice_read,
7486 .llseek = no_llseek,
7487 };
7488
7489 #endif /* CONFIG_TRACER_SNAPSHOT */
7490
7491 /*
7492 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7493 * @filp: The active open file structure
7494 * @ubuf: The userspace provided buffer to read value into
7495 * @cnt: The maximum number of bytes to read
7496 * @ppos: The current "file" position
7497 *
7498 * This function implements the write interface for a struct trace_min_max_param.
7499 * The filp->private_data must point to a trace_min_max_param structure that
7500 * defines where to write the value, the min and the max acceptable values,
7501 * and a lock to protect the write.
7502 */
7503 static ssize_t
trace_min_max_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)7504 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7505 {
7506 struct trace_min_max_param *param = filp->private_data;
7507 u64 val;
7508 int err;
7509
7510 if (!param)
7511 return -EFAULT;
7512
7513 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7514 if (err)
7515 return err;
7516
7517 if (param->lock)
7518 mutex_lock(param->lock);
7519
7520 if (param->min && val < *param->min)
7521 err = -EINVAL;
7522
7523 if (param->max && val > *param->max)
7524 err = -EINVAL;
7525
7526 if (!err)
7527 *param->val = val;
7528
7529 if (param->lock)
7530 mutex_unlock(param->lock);
7531
7532 if (err)
7533 return err;
7534
7535 return cnt;
7536 }
7537
7538 /*
7539 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7540 * @filp: The active open file structure
7541 * @ubuf: The userspace provided buffer to read value into
7542 * @cnt: The maximum number of bytes to read
7543 * @ppos: The current "file" position
7544 *
7545 * This function implements the read interface for a struct trace_min_max_param.
7546 * The filp->private_data must point to a trace_min_max_param struct with valid
7547 * data.
7548 */
7549 static ssize_t
trace_min_max_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)7550 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7551 {
7552 struct trace_min_max_param *param = filp->private_data;
7553 char buf[U64_STR_SIZE];
7554 int len;
7555 u64 val;
7556
7557 if (!param)
7558 return -EFAULT;
7559
7560 val = *param->val;
7561
7562 if (cnt > sizeof(buf))
7563 cnt = sizeof(buf);
7564
7565 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7566
7567 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7568 }
7569
7570 const struct file_operations trace_min_max_fops = {
7571 .open = tracing_open_generic,
7572 .read = trace_min_max_read,
7573 .write = trace_min_max_write,
7574 };
7575
7576 #define TRACING_LOG_ERRS_MAX 8
7577 #define TRACING_LOG_LOC_MAX 128
7578
7579 #define CMD_PREFIX " Command: "
7580
7581 struct err_info {
7582 const char **errs; /* ptr to loc-specific array of err strings */
7583 u8 type; /* index into errs -> specific err string */
7584 u16 pos; /* caret position */
7585 u64 ts;
7586 };
7587
7588 struct tracing_log_err {
7589 struct list_head list;
7590 struct err_info info;
7591 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7592 char *cmd; /* what caused err */
7593 };
7594
7595 static DEFINE_MUTEX(tracing_err_log_lock);
7596
alloc_tracing_log_err(int len)7597 static struct tracing_log_err *alloc_tracing_log_err(int len)
7598 {
7599 struct tracing_log_err *err;
7600
7601 err = kzalloc(sizeof(*err), GFP_KERNEL);
7602 if (!err)
7603 return ERR_PTR(-ENOMEM);
7604
7605 err->cmd = kzalloc(len, GFP_KERNEL);
7606 if (!err->cmd) {
7607 kfree(err);
7608 return ERR_PTR(-ENOMEM);
7609 }
7610
7611 return err;
7612 }
7613
free_tracing_log_err(struct tracing_log_err * err)7614 static void free_tracing_log_err(struct tracing_log_err *err)
7615 {
7616 kfree(err->cmd);
7617 kfree(err);
7618 }
7619
get_tracing_log_err(struct trace_array * tr,int len)7620 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7621 int len)
7622 {
7623 struct tracing_log_err *err;
7624 char *cmd;
7625
7626 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7627 err = alloc_tracing_log_err(len);
7628 if (PTR_ERR(err) != -ENOMEM)
7629 tr->n_err_log_entries++;
7630
7631 return err;
7632 }
7633 cmd = kzalloc(len, GFP_KERNEL);
7634 if (!cmd)
7635 return ERR_PTR(-ENOMEM);
7636 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7637 kfree(err->cmd);
7638 err->cmd = cmd;
7639 list_del(&err->list);
7640
7641 return err;
7642 }
7643
7644 /**
7645 * err_pos - find the position of a string within a command for error careting
7646 * @cmd: The tracing command that caused the error
7647 * @str: The string to position the caret at within @cmd
7648 *
7649 * Finds the position of the first occurrence of @str within @cmd. The
7650 * return value can be passed to tracing_log_err() for caret placement
7651 * within @cmd.
7652 *
7653 * Returns the index within @cmd of the first occurrence of @str or 0
7654 * if @str was not found.
7655 */
err_pos(char * cmd,const char * str)7656 unsigned int err_pos(char *cmd, const char *str)
7657 {
7658 char *found;
7659
7660 if (WARN_ON(!strlen(cmd)))
7661 return 0;
7662
7663 found = strstr(cmd, str);
7664 if (found)
7665 return found - cmd;
7666
7667 return 0;
7668 }
7669
7670 /**
7671 * tracing_log_err - write an error to the tracing error log
7672 * @tr: The associated trace array for the error (NULL for top level array)
7673 * @loc: A string describing where the error occurred
7674 * @cmd: The tracing command that caused the error
7675 * @errs: The array of loc-specific static error strings
7676 * @type: The index into errs[], which produces the specific static err string
7677 * @pos: The position the caret should be placed in the cmd
7678 *
7679 * Writes an error into tracing/error_log of the form:
7680 *
7681 * <loc>: error: <text>
7682 * Command: <cmd>
7683 * ^
7684 *
7685 * tracing/error_log is a small log file containing the last
7686 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7687 * unless there has been a tracing error, and the error log can be
7688 * cleared and have its memory freed by writing the empty string in
7689 * truncation mode to it i.e. echo > tracing/error_log.
7690 *
7691 * NOTE: the @errs array along with the @type param are used to
7692 * produce a static error string - this string is not copied and saved
7693 * when the error is logged - only a pointer to it is saved. See
7694 * existing callers for examples of how static strings are typically
7695 * defined for use with tracing_log_err().
7696 */
tracing_log_err(struct trace_array * tr,const char * loc,const char * cmd,const char ** errs,u8 type,u16 pos)7697 void tracing_log_err(struct trace_array *tr,
7698 const char *loc, const char *cmd,
7699 const char **errs, u8 type, u16 pos)
7700 {
7701 struct tracing_log_err *err;
7702 int len = 0;
7703
7704 if (!tr)
7705 tr = &global_trace;
7706
7707 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
7708
7709 mutex_lock(&tracing_err_log_lock);
7710 err = get_tracing_log_err(tr, len);
7711 if (PTR_ERR(err) == -ENOMEM) {
7712 mutex_unlock(&tracing_err_log_lock);
7713 return;
7714 }
7715
7716 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7717 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
7718
7719 err->info.errs = errs;
7720 err->info.type = type;
7721 err->info.pos = pos;
7722 err->info.ts = local_clock();
7723
7724 list_add_tail(&err->list, &tr->err_log);
7725 mutex_unlock(&tracing_err_log_lock);
7726 }
7727
clear_tracing_err_log(struct trace_array * tr)7728 static void clear_tracing_err_log(struct trace_array *tr)
7729 {
7730 struct tracing_log_err *err, *next;
7731
7732 mutex_lock(&tracing_err_log_lock);
7733 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7734 list_del(&err->list);
7735 free_tracing_log_err(err);
7736 }
7737
7738 tr->n_err_log_entries = 0;
7739 mutex_unlock(&tracing_err_log_lock);
7740 }
7741
tracing_err_log_seq_start(struct seq_file * m,loff_t * pos)7742 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7743 {
7744 struct trace_array *tr = m->private;
7745
7746 mutex_lock(&tracing_err_log_lock);
7747
7748 return seq_list_start(&tr->err_log, *pos);
7749 }
7750
tracing_err_log_seq_next(struct seq_file * m,void * v,loff_t * pos)7751 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7752 {
7753 struct trace_array *tr = m->private;
7754
7755 return seq_list_next(v, &tr->err_log, pos);
7756 }
7757
tracing_err_log_seq_stop(struct seq_file * m,void * v)7758 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7759 {
7760 mutex_unlock(&tracing_err_log_lock);
7761 }
7762
tracing_err_log_show_pos(struct seq_file * m,u16 pos)7763 static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
7764 {
7765 u16 i;
7766
7767 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7768 seq_putc(m, ' ');
7769 for (i = 0; i < pos; i++)
7770 seq_putc(m, ' ');
7771 seq_puts(m, "^\n");
7772 }
7773
tracing_err_log_seq_show(struct seq_file * m,void * v)7774 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7775 {
7776 struct tracing_log_err *err = v;
7777
7778 if (err) {
7779 const char *err_text = err->info.errs[err->info.type];
7780 u64 sec = err->info.ts;
7781 u32 nsec;
7782
7783 nsec = do_div(sec, NSEC_PER_SEC);
7784 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7785 err->loc, err_text);
7786 seq_printf(m, "%s", err->cmd);
7787 tracing_err_log_show_pos(m, err->info.pos);
7788 }
7789
7790 return 0;
7791 }
7792
7793 static const struct seq_operations tracing_err_log_seq_ops = {
7794 .start = tracing_err_log_seq_start,
7795 .next = tracing_err_log_seq_next,
7796 .stop = tracing_err_log_seq_stop,
7797 .show = tracing_err_log_seq_show
7798 };
7799
tracing_err_log_open(struct inode * inode,struct file * file)7800 static int tracing_err_log_open(struct inode *inode, struct file *file)
7801 {
7802 struct trace_array *tr = inode->i_private;
7803 int ret = 0;
7804
7805 ret = tracing_check_open_get_tr(tr);
7806 if (ret)
7807 return ret;
7808
7809 /* If this file was opened for write, then erase contents */
7810 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7811 clear_tracing_err_log(tr);
7812
7813 if (file->f_mode & FMODE_READ) {
7814 ret = seq_open(file, &tracing_err_log_seq_ops);
7815 if (!ret) {
7816 struct seq_file *m = file->private_data;
7817 m->private = tr;
7818 } else {
7819 trace_array_put(tr);
7820 }
7821 }
7822 return ret;
7823 }
7824
tracing_err_log_write(struct file * file,const char __user * buffer,size_t count,loff_t * ppos)7825 static ssize_t tracing_err_log_write(struct file *file,
7826 const char __user *buffer,
7827 size_t count, loff_t *ppos)
7828 {
7829 return count;
7830 }
7831
tracing_err_log_release(struct inode * inode,struct file * file)7832 static int tracing_err_log_release(struct inode *inode, struct file *file)
7833 {
7834 struct trace_array *tr = inode->i_private;
7835
7836 trace_array_put(tr);
7837
7838 if (file->f_mode & FMODE_READ)
7839 seq_release(inode, file);
7840
7841 return 0;
7842 }
7843
7844 static const struct file_operations tracing_err_log_fops = {
7845 .open = tracing_err_log_open,
7846 .write = tracing_err_log_write,
7847 .read = seq_read,
7848 .llseek = tracing_lseek,
7849 .release = tracing_err_log_release,
7850 };
7851
tracing_buffers_open(struct inode * inode,struct file * filp)7852 static int tracing_buffers_open(struct inode *inode, struct file *filp)
7853 {
7854 struct trace_array *tr = inode->i_private;
7855 struct ftrace_buffer_info *info;
7856 int ret;
7857
7858 ret = tracing_check_open_get_tr(tr);
7859 if (ret)
7860 return ret;
7861
7862 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7863 if (!info) {
7864 trace_array_put(tr);
7865 return -ENOMEM;
7866 }
7867
7868 mutex_lock(&trace_types_lock);
7869
7870 info->iter.tr = tr;
7871 info->iter.cpu_file = tracing_get_cpu(inode);
7872 info->iter.trace = tr->current_trace;
7873 info->iter.array_buffer = &tr->array_buffer;
7874 info->spare = NULL;
7875 /* Force reading ring buffer for first read */
7876 info->read = (unsigned int)-1;
7877
7878 filp->private_data = info;
7879
7880 tr->trace_ref++;
7881
7882 mutex_unlock(&trace_types_lock);
7883
7884 ret = nonseekable_open(inode, filp);
7885 if (ret < 0)
7886 trace_array_put(tr);
7887
7888 return ret;
7889 }
7890
7891 static __poll_t
tracing_buffers_poll(struct file * filp,poll_table * poll_table)7892 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7893 {
7894 struct ftrace_buffer_info *info = filp->private_data;
7895 struct trace_iterator *iter = &info->iter;
7896
7897 return trace_poll(iter, filp, poll_table);
7898 }
7899
7900 static ssize_t
tracing_buffers_read(struct file * filp,char __user * ubuf,size_t count,loff_t * ppos)7901 tracing_buffers_read(struct file *filp, char __user *ubuf,
7902 size_t count, loff_t *ppos)
7903 {
7904 struct ftrace_buffer_info *info = filp->private_data;
7905 struct trace_iterator *iter = &info->iter;
7906 void *trace_data;
7907 int page_size;
7908 ssize_t ret = 0;
7909 ssize_t size;
7910
7911 if (!count)
7912 return 0;
7913
7914 #ifdef CONFIG_TRACER_MAX_TRACE
7915 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7916 return -EBUSY;
7917 #endif
7918
7919 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
7920
7921 /* Make sure the spare matches the current sub buffer size */
7922 if (info->spare) {
7923 if (page_size != info->spare_size) {
7924 ring_buffer_free_read_page(iter->array_buffer->buffer,
7925 info->spare_cpu, info->spare);
7926 info->spare = NULL;
7927 }
7928 }
7929
7930 if (!info->spare) {
7931 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
7932 iter->cpu_file);
7933 if (IS_ERR(info->spare)) {
7934 ret = PTR_ERR(info->spare);
7935 info->spare = NULL;
7936 } else {
7937 info->spare_cpu = iter->cpu_file;
7938 info->spare_size = page_size;
7939 }
7940 }
7941 if (!info->spare)
7942 return ret;
7943
7944 /* Do we have previous read data to read? */
7945 if (info->read < page_size)
7946 goto read;
7947
7948 again:
7949 trace_access_lock(iter->cpu_file);
7950 ret = ring_buffer_read_page(iter->array_buffer->buffer,
7951 info->spare,
7952 count,
7953 iter->cpu_file, 0);
7954 trace_access_unlock(iter->cpu_file);
7955
7956 if (ret < 0) {
7957 if (trace_empty(iter) && !iter->closed) {
7958 if ((filp->f_flags & O_NONBLOCK))
7959 return -EAGAIN;
7960
7961 ret = wait_on_pipe(iter, 0);
7962 if (ret)
7963 return ret;
7964
7965 goto again;
7966 }
7967 return 0;
7968 }
7969
7970 info->read = 0;
7971 read:
7972 size = page_size - info->read;
7973 if (size > count)
7974 size = count;
7975 trace_data = ring_buffer_read_page_data(info->spare);
7976 ret = copy_to_user(ubuf, trace_data + info->read, size);
7977 if (ret == size)
7978 return -EFAULT;
7979
7980 size -= ret;
7981
7982 *ppos += size;
7983 info->read += size;
7984
7985 return size;
7986 }
7987
tracing_buffers_flush(struct file * file,fl_owner_t id)7988 static int tracing_buffers_flush(struct file *file, fl_owner_t id)
7989 {
7990 struct ftrace_buffer_info *info = file->private_data;
7991 struct trace_iterator *iter = &info->iter;
7992
7993 iter->closed = true;
7994 /* Make sure the waiters see the new wait_index */
7995 (void)atomic_fetch_inc_release(&iter->wait_index);
7996
7997 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
7998
7999 return 0;
8000 }
8001
tracing_buffers_release(struct inode * inode,struct file * file)8002 static int tracing_buffers_release(struct inode *inode, struct file *file)
8003 {
8004 struct ftrace_buffer_info *info = file->private_data;
8005 struct trace_iterator *iter = &info->iter;
8006
8007 mutex_lock(&trace_types_lock);
8008
8009 iter->tr->trace_ref--;
8010
8011 __trace_array_put(iter->tr);
8012
8013 if (info->spare)
8014 ring_buffer_free_read_page(iter->array_buffer->buffer,
8015 info->spare_cpu, info->spare);
8016 kvfree(info);
8017
8018 mutex_unlock(&trace_types_lock);
8019
8020 return 0;
8021 }
8022
8023 struct buffer_ref {
8024 struct trace_buffer *buffer;
8025 void *page;
8026 int cpu;
8027 refcount_t refcount;
8028 };
8029
buffer_ref_release(struct buffer_ref * ref)8030 static void buffer_ref_release(struct buffer_ref *ref)
8031 {
8032 if (!refcount_dec_and_test(&ref->refcount))
8033 return;
8034 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8035 kfree(ref);
8036 }
8037
buffer_pipe_buf_release(struct pipe_inode_info * pipe,struct pipe_buffer * buf)8038 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8039 struct pipe_buffer *buf)
8040 {
8041 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8042
8043 buffer_ref_release(ref);
8044 buf->private = 0;
8045 }
8046
buffer_pipe_buf_get(struct pipe_inode_info * pipe,struct pipe_buffer * buf)8047 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8048 struct pipe_buffer *buf)
8049 {
8050 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8051
8052 if (refcount_read(&ref->refcount) > INT_MAX/2)
8053 return false;
8054
8055 refcount_inc(&ref->refcount);
8056 return true;
8057 }
8058
8059 /* Pipe buffer operations for a buffer. */
8060 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8061 .release = buffer_pipe_buf_release,
8062 .get = buffer_pipe_buf_get,
8063 };
8064
8065 /*
8066 * Callback from splice_to_pipe(), if we need to release some pages
8067 * at the end of the spd in case we error'ed out in filling the pipe.
8068 */
buffer_spd_release(struct splice_pipe_desc * spd,unsigned int i)8069 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8070 {
8071 struct buffer_ref *ref =
8072 (struct buffer_ref *)spd->partial[i].private;
8073
8074 buffer_ref_release(ref);
8075 spd->partial[i].private = 0;
8076 }
8077
8078 static ssize_t
tracing_buffers_splice_read(struct file * file,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)8079 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8080 struct pipe_inode_info *pipe, size_t len,
8081 unsigned int flags)
8082 {
8083 struct ftrace_buffer_info *info = file->private_data;
8084 struct trace_iterator *iter = &info->iter;
8085 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8086 struct page *pages_def[PIPE_DEF_BUFFERS];
8087 struct splice_pipe_desc spd = {
8088 .pages = pages_def,
8089 .partial = partial_def,
8090 .nr_pages_max = PIPE_DEF_BUFFERS,
8091 .ops = &buffer_pipe_buf_ops,
8092 .spd_release = buffer_spd_release,
8093 };
8094 struct buffer_ref *ref;
8095 bool woken = false;
8096 int page_size;
8097 int entries, i;
8098 ssize_t ret = 0;
8099
8100 #ifdef CONFIG_TRACER_MAX_TRACE
8101 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8102 return -EBUSY;
8103 #endif
8104
8105 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8106 if (*ppos & (page_size - 1))
8107 return -EINVAL;
8108
8109 if (len & (page_size - 1)) {
8110 if (len < page_size)
8111 return -EINVAL;
8112 len &= (~(page_size - 1));
8113 }
8114
8115 if (splice_grow_spd(pipe, &spd))
8116 return -ENOMEM;
8117
8118 again:
8119 trace_access_lock(iter->cpu_file);
8120 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8121
8122 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= page_size) {
8123 struct page *page;
8124 int r;
8125
8126 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8127 if (!ref) {
8128 ret = -ENOMEM;
8129 break;
8130 }
8131
8132 refcount_set(&ref->refcount, 1);
8133 ref->buffer = iter->array_buffer->buffer;
8134 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8135 if (IS_ERR(ref->page)) {
8136 ret = PTR_ERR(ref->page);
8137 ref->page = NULL;
8138 kfree(ref);
8139 break;
8140 }
8141 ref->cpu = iter->cpu_file;
8142
8143 r = ring_buffer_read_page(ref->buffer, ref->page,
8144 len, iter->cpu_file, 1);
8145 if (r < 0) {
8146 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8147 ref->page);
8148 kfree(ref);
8149 break;
8150 }
8151
8152 page = virt_to_page(ring_buffer_read_page_data(ref->page));
8153
8154 spd.pages[i] = page;
8155 spd.partial[i].len = page_size;
8156 spd.partial[i].offset = 0;
8157 spd.partial[i].private = (unsigned long)ref;
8158 spd.nr_pages++;
8159 *ppos += page_size;
8160
8161 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8162 }
8163
8164 trace_access_unlock(iter->cpu_file);
8165 spd.nr_pages = i;
8166
8167 /* did we read anything? */
8168 if (!spd.nr_pages) {
8169
8170 if (ret)
8171 goto out;
8172
8173 if (woken)
8174 goto out;
8175
8176 ret = -EAGAIN;
8177 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8178 goto out;
8179
8180 ret = wait_on_pipe(iter, iter->snapshot ? 0 : iter->tr->buffer_percent);
8181 if (ret)
8182 goto out;
8183
8184 /* No need to wait after waking up when tracing is off */
8185 if (!tracer_tracing_is_on(iter->tr))
8186 goto out;
8187
8188 /* Iterate one more time to collect any new data then exit */
8189 woken = true;
8190
8191 goto again;
8192 }
8193
8194 ret = splice_to_pipe(pipe, &spd);
8195 out:
8196 splice_shrink_spd(&spd);
8197
8198 return ret;
8199 }
8200
tracing_buffers_ioctl(struct file * file,unsigned int cmd,unsigned long arg)8201 static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8202 {
8203 struct ftrace_buffer_info *info = file->private_data;
8204 struct trace_iterator *iter = &info->iter;
8205 int err;
8206
8207 if (cmd == TRACE_MMAP_IOCTL_GET_READER) {
8208 if (!(file->f_flags & O_NONBLOCK)) {
8209 err = ring_buffer_wait(iter->array_buffer->buffer,
8210 iter->cpu_file,
8211 iter->tr->buffer_percent,
8212 NULL, NULL);
8213 if (err)
8214 return err;
8215 }
8216
8217 return ring_buffer_map_get_reader(iter->array_buffer->buffer,
8218 iter->cpu_file);
8219 } else if (cmd) {
8220 return -ENOTTY;
8221 }
8222
8223 /*
8224 * An ioctl call with cmd 0 to the ring buffer file will wake up all
8225 * waiters
8226 */
8227 mutex_lock(&trace_types_lock);
8228
8229 /* Make sure the waiters see the new wait_index */
8230 (void)atomic_fetch_inc_release(&iter->wait_index);
8231
8232 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8233
8234 mutex_unlock(&trace_types_lock);
8235 return 0;
8236 }
8237
8238 #ifdef CONFIG_TRACER_MAX_TRACE
get_snapshot_map(struct trace_array * tr)8239 static int get_snapshot_map(struct trace_array *tr)
8240 {
8241 int err = 0;
8242
8243 /*
8244 * Called with mmap_lock held. lockdep would be unhappy if we would now
8245 * take trace_types_lock. Instead use the specific
8246 * snapshot_trigger_lock.
8247 */
8248 spin_lock(&tr->snapshot_trigger_lock);
8249
8250 if (tr->snapshot || tr->mapped == UINT_MAX)
8251 err = -EBUSY;
8252 else
8253 tr->mapped++;
8254
8255 spin_unlock(&tr->snapshot_trigger_lock);
8256
8257 /* Wait for update_max_tr() to observe iter->tr->mapped */
8258 if (tr->mapped == 1)
8259 synchronize_rcu();
8260
8261 return err;
8262
8263 }
put_snapshot_map(struct trace_array * tr)8264 static void put_snapshot_map(struct trace_array *tr)
8265 {
8266 spin_lock(&tr->snapshot_trigger_lock);
8267 if (!WARN_ON(!tr->mapped))
8268 tr->mapped--;
8269 spin_unlock(&tr->snapshot_trigger_lock);
8270 }
8271 #else
get_snapshot_map(struct trace_array * tr)8272 static inline int get_snapshot_map(struct trace_array *tr) { return 0; }
put_snapshot_map(struct trace_array * tr)8273 static inline void put_snapshot_map(struct trace_array *tr) { }
8274 #endif
8275
tracing_buffers_mmap_close(struct vm_area_struct * vma)8276 static void tracing_buffers_mmap_close(struct vm_area_struct *vma)
8277 {
8278 struct ftrace_buffer_info *info = vma->vm_file->private_data;
8279 struct trace_iterator *iter = &info->iter;
8280
8281 WARN_ON(ring_buffer_unmap(iter->array_buffer->buffer, iter->cpu_file));
8282 put_snapshot_map(iter->tr);
8283 }
8284
8285 static const struct vm_operations_struct tracing_buffers_vmops = {
8286 .close = tracing_buffers_mmap_close,
8287 };
8288
tracing_buffers_mmap(struct file * filp,struct vm_area_struct * vma)8289 static int tracing_buffers_mmap(struct file *filp, struct vm_area_struct *vma)
8290 {
8291 struct ftrace_buffer_info *info = filp->private_data;
8292 struct trace_iterator *iter = &info->iter;
8293 int ret = 0;
8294
8295 ret = get_snapshot_map(iter->tr);
8296 if (ret)
8297 return ret;
8298
8299 ret = ring_buffer_map(iter->array_buffer->buffer, iter->cpu_file, vma);
8300 if (ret)
8301 put_snapshot_map(iter->tr);
8302
8303 vma->vm_ops = &tracing_buffers_vmops;
8304
8305 return ret;
8306 }
8307
8308 static const struct file_operations tracing_buffers_fops = {
8309 .open = tracing_buffers_open,
8310 .read = tracing_buffers_read,
8311 .poll = tracing_buffers_poll,
8312 .release = tracing_buffers_release,
8313 .flush = tracing_buffers_flush,
8314 .splice_read = tracing_buffers_splice_read,
8315 .unlocked_ioctl = tracing_buffers_ioctl,
8316 .llseek = no_llseek,
8317 .mmap = tracing_buffers_mmap,
8318 };
8319
8320 static ssize_t
tracing_stats_read(struct file * filp,char __user * ubuf,size_t count,loff_t * ppos)8321 tracing_stats_read(struct file *filp, char __user *ubuf,
8322 size_t count, loff_t *ppos)
8323 {
8324 struct inode *inode = file_inode(filp);
8325 struct trace_array *tr = inode->i_private;
8326 struct array_buffer *trace_buf = &tr->array_buffer;
8327 int cpu = tracing_get_cpu(inode);
8328 struct trace_seq *s;
8329 unsigned long cnt;
8330 unsigned long long t;
8331 unsigned long usec_rem;
8332
8333 s = kmalloc(sizeof(*s), GFP_KERNEL);
8334 if (!s)
8335 return -ENOMEM;
8336
8337 trace_seq_init(s);
8338
8339 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8340 trace_seq_printf(s, "entries: %ld\n", cnt);
8341
8342 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8343 trace_seq_printf(s, "overrun: %ld\n", cnt);
8344
8345 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8346 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8347
8348 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8349 trace_seq_printf(s, "bytes: %ld\n", cnt);
8350
8351 if (trace_clocks[tr->clock_id].in_ns) {
8352 /* local or global for trace_clock */
8353 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8354 usec_rem = do_div(t, USEC_PER_SEC);
8355 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8356 t, usec_rem);
8357
8358 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8359 usec_rem = do_div(t, USEC_PER_SEC);
8360 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8361 } else {
8362 /* counter or tsc mode for trace_clock */
8363 trace_seq_printf(s, "oldest event ts: %llu\n",
8364 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8365
8366 trace_seq_printf(s, "now ts: %llu\n",
8367 ring_buffer_time_stamp(trace_buf->buffer));
8368 }
8369
8370 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8371 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8372
8373 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8374 trace_seq_printf(s, "read events: %ld\n", cnt);
8375
8376 count = simple_read_from_buffer(ubuf, count, ppos,
8377 s->buffer, trace_seq_used(s));
8378
8379 kfree(s);
8380
8381 return count;
8382 }
8383
8384 static const struct file_operations tracing_stats_fops = {
8385 .open = tracing_open_generic_tr,
8386 .read = tracing_stats_read,
8387 .llseek = generic_file_llseek,
8388 .release = tracing_release_generic_tr,
8389 };
8390
8391 #ifdef CONFIG_DYNAMIC_FTRACE
8392
8393 static ssize_t
tracing_read_dyn_info(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)8394 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8395 size_t cnt, loff_t *ppos)
8396 {
8397 ssize_t ret;
8398 char *buf;
8399 int r;
8400
8401 /* 256 should be plenty to hold the amount needed */
8402 buf = kmalloc(256, GFP_KERNEL);
8403 if (!buf)
8404 return -ENOMEM;
8405
8406 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8407 ftrace_update_tot_cnt,
8408 ftrace_number_of_pages,
8409 ftrace_number_of_groups);
8410
8411 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8412 kfree(buf);
8413 return ret;
8414 }
8415
8416 static const struct file_operations tracing_dyn_info_fops = {
8417 .open = tracing_open_generic,
8418 .read = tracing_read_dyn_info,
8419 .llseek = generic_file_llseek,
8420 };
8421 #endif /* CONFIG_DYNAMIC_FTRACE */
8422
8423 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8424 static void
ftrace_snapshot(unsigned long ip,unsigned long parent_ip,struct trace_array * tr,struct ftrace_probe_ops * ops,void * data)8425 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8426 struct trace_array *tr, struct ftrace_probe_ops *ops,
8427 void *data)
8428 {
8429 tracing_snapshot_instance(tr);
8430 }
8431
8432 static void
ftrace_count_snapshot(unsigned long ip,unsigned long parent_ip,struct trace_array * tr,struct ftrace_probe_ops * ops,void * data)8433 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8434 struct trace_array *tr, struct ftrace_probe_ops *ops,
8435 void *data)
8436 {
8437 struct ftrace_func_mapper *mapper = data;
8438 long *count = NULL;
8439
8440 if (mapper)
8441 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8442
8443 if (count) {
8444
8445 if (*count <= 0)
8446 return;
8447
8448 (*count)--;
8449 }
8450
8451 tracing_snapshot_instance(tr);
8452 }
8453
8454 static int
ftrace_snapshot_print(struct seq_file * m,unsigned long ip,struct ftrace_probe_ops * ops,void * data)8455 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8456 struct ftrace_probe_ops *ops, void *data)
8457 {
8458 struct ftrace_func_mapper *mapper = data;
8459 long *count = NULL;
8460
8461 seq_printf(m, "%ps:", (void *)ip);
8462
8463 seq_puts(m, "snapshot");
8464
8465 if (mapper)
8466 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8467
8468 if (count)
8469 seq_printf(m, ":count=%ld\n", *count);
8470 else
8471 seq_puts(m, ":unlimited\n");
8472
8473 return 0;
8474 }
8475
8476 static int
ftrace_snapshot_init(struct ftrace_probe_ops * ops,struct trace_array * tr,unsigned long ip,void * init_data,void ** data)8477 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8478 unsigned long ip, void *init_data, void **data)
8479 {
8480 struct ftrace_func_mapper *mapper = *data;
8481
8482 if (!mapper) {
8483 mapper = allocate_ftrace_func_mapper();
8484 if (!mapper)
8485 return -ENOMEM;
8486 *data = mapper;
8487 }
8488
8489 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8490 }
8491
8492 static void
ftrace_snapshot_free(struct ftrace_probe_ops * ops,struct trace_array * tr,unsigned long ip,void * data)8493 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8494 unsigned long ip, void *data)
8495 {
8496 struct ftrace_func_mapper *mapper = data;
8497
8498 if (!ip) {
8499 if (!mapper)
8500 return;
8501 free_ftrace_func_mapper(mapper, NULL);
8502 return;
8503 }
8504
8505 ftrace_func_mapper_remove_ip(mapper, ip);
8506 }
8507
8508 static struct ftrace_probe_ops snapshot_probe_ops = {
8509 .func = ftrace_snapshot,
8510 .print = ftrace_snapshot_print,
8511 };
8512
8513 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8514 .func = ftrace_count_snapshot,
8515 .print = ftrace_snapshot_print,
8516 .init = ftrace_snapshot_init,
8517 .free = ftrace_snapshot_free,
8518 };
8519
8520 static int
ftrace_trace_snapshot_callback(struct trace_array * tr,struct ftrace_hash * hash,char * glob,char * cmd,char * param,int enable)8521 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8522 char *glob, char *cmd, char *param, int enable)
8523 {
8524 struct ftrace_probe_ops *ops;
8525 void *count = (void *)-1;
8526 char *number;
8527 int ret;
8528
8529 if (!tr)
8530 return -ENODEV;
8531
8532 /* hash funcs only work with set_ftrace_filter */
8533 if (!enable)
8534 return -EINVAL;
8535
8536 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8537
8538 if (glob[0] == '!') {
8539 ret = unregister_ftrace_function_probe_func(glob+1, tr, ops);
8540 if (!ret)
8541 tracing_disarm_snapshot(tr);
8542
8543 return ret;
8544 }
8545
8546 if (!param)
8547 goto out_reg;
8548
8549 number = strsep(¶m, ":");
8550
8551 if (!strlen(number))
8552 goto out_reg;
8553
8554 /*
8555 * We use the callback data field (which is a pointer)
8556 * as our counter.
8557 */
8558 ret = kstrtoul(number, 0, (unsigned long *)&count);
8559 if (ret)
8560 return ret;
8561
8562 out_reg:
8563 ret = tracing_arm_snapshot(tr);
8564 if (ret < 0)
8565 goto out;
8566
8567 ret = register_ftrace_function_probe(glob, tr, ops, count);
8568 if (ret < 0)
8569 tracing_disarm_snapshot(tr);
8570 out:
8571 return ret < 0 ? ret : 0;
8572 }
8573
8574 static struct ftrace_func_command ftrace_snapshot_cmd = {
8575 .name = "snapshot",
8576 .func = ftrace_trace_snapshot_callback,
8577 };
8578
register_snapshot_cmd(void)8579 static __init int register_snapshot_cmd(void)
8580 {
8581 return register_ftrace_command(&ftrace_snapshot_cmd);
8582 }
8583 #else
register_snapshot_cmd(void)8584 static inline __init int register_snapshot_cmd(void) { return 0; }
8585 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8586
tracing_get_dentry(struct trace_array * tr)8587 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8588 {
8589 if (WARN_ON(!tr->dir))
8590 return ERR_PTR(-ENODEV);
8591
8592 /* Top directory uses NULL as the parent */
8593 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8594 return NULL;
8595
8596 /* All sub buffers have a descriptor */
8597 return tr->dir;
8598 }
8599
tracing_dentry_percpu(struct trace_array * tr,int cpu)8600 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8601 {
8602 struct dentry *d_tracer;
8603
8604 if (tr->percpu_dir)
8605 return tr->percpu_dir;
8606
8607 d_tracer = tracing_get_dentry(tr);
8608 if (IS_ERR(d_tracer))
8609 return NULL;
8610
8611 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8612
8613 MEM_FAIL(!tr->percpu_dir,
8614 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8615
8616 return tr->percpu_dir;
8617 }
8618
8619 static struct dentry *
trace_create_cpu_file(const char * name,umode_t mode,struct dentry * parent,void * data,long cpu,const struct file_operations * fops)8620 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8621 void *data, long cpu, const struct file_operations *fops)
8622 {
8623 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8624
8625 if (ret) /* See tracing_get_cpu() */
8626 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8627 return ret;
8628 }
8629
8630 static void
tracing_init_tracefs_percpu(struct trace_array * tr,long cpu)8631 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8632 {
8633 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8634 struct dentry *d_cpu;
8635 char cpu_dir[30]; /* 30 characters should be more than enough */
8636
8637 if (!d_percpu)
8638 return;
8639
8640 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8641 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8642 if (!d_cpu) {
8643 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8644 return;
8645 }
8646
8647 /* per cpu trace_pipe */
8648 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8649 tr, cpu, &tracing_pipe_fops);
8650
8651 /* per cpu trace */
8652 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8653 tr, cpu, &tracing_fops);
8654
8655 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8656 tr, cpu, &tracing_buffers_fops);
8657
8658 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8659 tr, cpu, &tracing_stats_fops);
8660
8661 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8662 tr, cpu, &tracing_entries_fops);
8663
8664 #ifdef CONFIG_TRACER_SNAPSHOT
8665 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8666 tr, cpu, &snapshot_fops);
8667
8668 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8669 tr, cpu, &snapshot_raw_fops);
8670 #endif
8671 }
8672
8673 #ifdef CONFIG_FTRACE_SELFTEST
8674 /* Let selftest have access to static functions in this file */
8675 #include "trace_selftest.c"
8676 #endif
8677
8678 static ssize_t
trace_options_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)8679 trace_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
8693 static ssize_t
trace_options_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)8694 trace_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
tracing_open_options(struct inode * inode,struct file * filp)8722 static 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
tracing_release_options(struct inode * inode,struct file * file)8735 static 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
8743 static 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 */
get_tr_index(void * data,struct trace_array ** ptr,unsigned int * pindex)8775 static 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
8784 static ssize_t
trace_options_core_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)8785 trace_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
8803 static ssize_t
trace_options_core_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)8804 trace_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
8836 static 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
trace_create_file(const char * name,umode_t mode,struct dentry * parent,void * data,const struct file_operations * fops)8843 struct 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
trace_options_init_dentry(struct trace_array * tr)8859 static 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
8879 static void
create_trace_option_file(struct trace_array * tr,struct trace_option_dentry * topt,struct tracer_flags * flags,struct tracer_opt * opt)8880 create_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
8900 static void
create_trace_option_files(struct trace_array * tr,struct tracer * tracer)8901 create_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
8961 static struct dentry *
create_trace_option_core_file(struct trace_array * tr,const char * option,long index)8962 create_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
create_trace_options_dir(struct trace_array * tr)8976 static 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
8993 static ssize_t
rb_simple_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)8994 rb_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
9007 static ssize_t
rb_simple_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)9008 rb_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
9043 static 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
9051 static ssize_t
buffer_percent_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)9052 buffer_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
9065 static ssize_t
buffer_percent_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)9066 buffer_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
9087 static 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
9095 static ssize_t
buffer_subbuf_size_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)9096 buffer_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
9112 static ssize_t
buffer_subbuf_size_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)9113 buffer_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
9184 static 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
9192 static struct dentry *trace_instance_dir;
9193
9194 static void
9195 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9196
9197 static int
allocate_trace_buffer(struct trace_array * tr,struct array_buffer * buf,int size)9198 allocate_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 buf->buffer = ring_buffer_alloc(size, rb_flags);
9207 if (!buf->buffer)
9208 return -ENOMEM;
9209
9210 buf->data = alloc_percpu(struct trace_array_cpu);
9211 if (!buf->data) {
9212 ring_buffer_free(buf->buffer);
9213 buf->buffer = NULL;
9214 return -ENOMEM;
9215 }
9216
9217 /* Allocate the first page for all buffers */
9218 set_buffer_entries(&tr->array_buffer,
9219 ring_buffer_size(tr->array_buffer.buffer, 0));
9220
9221 return 0;
9222 }
9223
free_trace_buffer(struct array_buffer * buf)9224 static void free_trace_buffer(struct array_buffer *buf)
9225 {
9226 if (buf->buffer) {
9227 ring_buffer_free(buf->buffer);
9228 buf->buffer = NULL;
9229 free_percpu(buf->data);
9230 buf->data = NULL;
9231 }
9232 }
9233
allocate_trace_buffers(struct trace_array * tr,int size)9234 static int allocate_trace_buffers(struct trace_array *tr, int size)
9235 {
9236 int ret;
9237
9238 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9239 if (ret)
9240 return ret;
9241
9242 #ifdef CONFIG_TRACER_MAX_TRACE
9243 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9244 allocate_snapshot ? size : 1);
9245 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9246 free_trace_buffer(&tr->array_buffer);
9247 return -ENOMEM;
9248 }
9249 tr->allocated_snapshot = allocate_snapshot;
9250
9251 allocate_snapshot = false;
9252 #endif
9253
9254 return 0;
9255 }
9256
free_trace_buffers(struct trace_array * tr)9257 static void free_trace_buffers(struct trace_array *tr)
9258 {
9259 if (!tr)
9260 return;
9261
9262 free_trace_buffer(&tr->array_buffer);
9263
9264 #ifdef CONFIG_TRACER_MAX_TRACE
9265 free_trace_buffer(&tr->max_buffer);
9266 #endif
9267 }
9268
init_trace_flags_index(struct trace_array * tr)9269 static void init_trace_flags_index(struct trace_array *tr)
9270 {
9271 int i;
9272
9273 /* Used by the trace options files */
9274 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9275 tr->trace_flags_index[i] = i;
9276 }
9277
__update_tracer_options(struct trace_array * tr)9278 static void __update_tracer_options(struct trace_array *tr)
9279 {
9280 struct tracer *t;
9281
9282 for (t = trace_types; t; t = t->next)
9283 add_tracer_options(tr, t);
9284 }
9285
update_tracer_options(struct trace_array * tr)9286 static void update_tracer_options(struct trace_array *tr)
9287 {
9288 mutex_lock(&trace_types_lock);
9289 tracer_options_updated = true;
9290 __update_tracer_options(tr);
9291 mutex_unlock(&trace_types_lock);
9292 }
9293
9294 /* Must have trace_types_lock held */
trace_array_find(const char * instance)9295 struct trace_array *trace_array_find(const char *instance)
9296 {
9297 struct trace_array *tr, *found = NULL;
9298
9299 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9300 if (tr->name && strcmp(tr->name, instance) == 0) {
9301 found = tr;
9302 break;
9303 }
9304 }
9305
9306 return found;
9307 }
9308
trace_array_find_get(const char * instance)9309 struct trace_array *trace_array_find_get(const char *instance)
9310 {
9311 struct trace_array *tr;
9312
9313 mutex_lock(&trace_types_lock);
9314 tr = trace_array_find(instance);
9315 if (tr)
9316 tr->ref++;
9317 mutex_unlock(&trace_types_lock);
9318
9319 return tr;
9320 }
9321
trace_array_create_dir(struct trace_array * tr)9322 static int trace_array_create_dir(struct trace_array *tr)
9323 {
9324 int ret;
9325
9326 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9327 if (!tr->dir)
9328 return -EINVAL;
9329
9330 ret = event_trace_add_tracer(tr->dir, tr);
9331 if (ret) {
9332 tracefs_remove(tr->dir);
9333 return ret;
9334 }
9335
9336 init_tracer_tracefs(tr, tr->dir);
9337 __update_tracer_options(tr);
9338
9339 return ret;
9340 }
9341
9342 static struct trace_array *
trace_array_create_systems(const char * name,const char * systems)9343 trace_array_create_systems(const char *name, const char *systems)
9344 {
9345 struct trace_array *tr;
9346 int ret;
9347
9348 ret = -ENOMEM;
9349 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9350 if (!tr)
9351 return ERR_PTR(ret);
9352
9353 tr->name = kstrdup(name, GFP_KERNEL);
9354 if (!tr->name)
9355 goto out_free_tr;
9356
9357 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9358 goto out_free_tr;
9359
9360 if (!zalloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9361 goto out_free_tr;
9362
9363 if (systems) {
9364 tr->system_names = kstrdup_const(systems, GFP_KERNEL);
9365 if (!tr->system_names)
9366 goto out_free_tr;
9367 }
9368
9369 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9370
9371 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9372
9373 raw_spin_lock_init(&tr->start_lock);
9374
9375 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9376 #ifdef CONFIG_TRACER_MAX_TRACE
9377 spin_lock_init(&tr->snapshot_trigger_lock);
9378 #endif
9379 tr->current_trace = &nop_trace;
9380
9381 INIT_LIST_HEAD(&tr->systems);
9382 INIT_LIST_HEAD(&tr->events);
9383 INIT_LIST_HEAD(&tr->hist_vars);
9384 INIT_LIST_HEAD(&tr->err_log);
9385
9386 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9387 goto out_free_tr;
9388
9389 /* The ring buffer is defaultly expanded */
9390 trace_set_ring_buffer_expanded(tr);
9391
9392 if (ftrace_allocate_ftrace_ops(tr) < 0)
9393 goto out_free_tr;
9394
9395 ftrace_init_trace_array(tr);
9396
9397 init_trace_flags_index(tr);
9398
9399 if (trace_instance_dir) {
9400 ret = trace_array_create_dir(tr);
9401 if (ret)
9402 goto out_free_tr;
9403 } else
9404 __trace_early_add_events(tr);
9405
9406 list_add(&tr->list, &ftrace_trace_arrays);
9407
9408 tr->ref++;
9409
9410 return tr;
9411
9412 out_free_tr:
9413 ftrace_free_ftrace_ops(tr);
9414 free_trace_buffers(tr);
9415 free_cpumask_var(tr->pipe_cpumask);
9416 free_cpumask_var(tr->tracing_cpumask);
9417 kfree_const(tr->system_names);
9418 kfree(tr->name);
9419 kfree(tr);
9420
9421 return ERR_PTR(ret);
9422 }
9423
trace_array_create(const char * name)9424 static struct trace_array *trace_array_create(const char *name)
9425 {
9426 return trace_array_create_systems(name, NULL);
9427 }
9428
instance_mkdir(const char * name)9429 static int instance_mkdir(const char *name)
9430 {
9431 struct trace_array *tr;
9432 int ret;
9433
9434 mutex_lock(&event_mutex);
9435 mutex_lock(&trace_types_lock);
9436
9437 ret = -EEXIST;
9438 if (trace_array_find(name))
9439 goto out_unlock;
9440
9441 tr = trace_array_create(name);
9442
9443 ret = PTR_ERR_OR_ZERO(tr);
9444
9445 out_unlock:
9446 mutex_unlock(&trace_types_lock);
9447 mutex_unlock(&event_mutex);
9448 return ret;
9449 }
9450
9451 /**
9452 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9453 * @name: The name of the trace array to be looked up/created.
9454 * @systems: A list of systems to create event directories for (NULL for all)
9455 *
9456 * Returns pointer to trace array with given name.
9457 * NULL, if it cannot be created.
9458 *
9459 * NOTE: This function increments the reference counter associated with the
9460 * trace array returned. This makes sure it cannot be freed while in use.
9461 * Use trace_array_put() once the trace array is no longer needed.
9462 * If the trace_array is to be freed, trace_array_destroy() needs to
9463 * be called after the trace_array_put(), or simply let user space delete
9464 * it from the tracefs instances directory. But until the
9465 * trace_array_put() is called, user space can not delete it.
9466 *
9467 */
trace_array_get_by_name(const char * name,const char * systems)9468 struct trace_array *trace_array_get_by_name(const char *name, const char *systems)
9469 {
9470 struct trace_array *tr;
9471
9472 mutex_lock(&event_mutex);
9473 mutex_lock(&trace_types_lock);
9474
9475 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9476 if (tr->name && strcmp(tr->name, name) == 0)
9477 goto out_unlock;
9478 }
9479
9480 tr = trace_array_create_systems(name, systems);
9481
9482 if (IS_ERR(tr))
9483 tr = NULL;
9484 out_unlock:
9485 if (tr)
9486 tr->ref++;
9487
9488 mutex_unlock(&trace_types_lock);
9489 mutex_unlock(&event_mutex);
9490 return tr;
9491 }
9492 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9493
__remove_instance(struct trace_array * tr)9494 static int __remove_instance(struct trace_array *tr)
9495 {
9496 int i;
9497
9498 /* Reference counter for a newly created trace array = 1. */
9499 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9500 return -EBUSY;
9501
9502 list_del(&tr->list);
9503
9504 /* Disable all the flags that were enabled coming in */
9505 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9506 if ((1 << i) & ZEROED_TRACE_FLAGS)
9507 set_tracer_flag(tr, 1 << i, 0);
9508 }
9509
9510 tracing_set_nop(tr);
9511 clear_ftrace_function_probes(tr);
9512 event_trace_del_tracer(tr);
9513 ftrace_clear_pids(tr);
9514 ftrace_destroy_function_files(tr);
9515 tracefs_remove(tr->dir);
9516 free_percpu(tr->last_func_repeats);
9517 free_trace_buffers(tr);
9518 clear_tracing_err_log(tr);
9519
9520 for (i = 0; i < tr->nr_topts; i++) {
9521 kfree(tr->topts[i].topts);
9522 }
9523 kfree(tr->topts);
9524
9525 free_cpumask_var(tr->pipe_cpumask);
9526 free_cpumask_var(tr->tracing_cpumask);
9527 kfree_const(tr->system_names);
9528 kfree(tr->name);
9529 kfree(tr);
9530
9531 return 0;
9532 }
9533
trace_array_destroy(struct trace_array * this_tr)9534 int trace_array_destroy(struct trace_array *this_tr)
9535 {
9536 struct trace_array *tr;
9537 int ret;
9538
9539 if (!this_tr)
9540 return -EINVAL;
9541
9542 mutex_lock(&event_mutex);
9543 mutex_lock(&trace_types_lock);
9544
9545 ret = -ENODEV;
9546
9547 /* Making sure trace array exists before destroying it. */
9548 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9549 if (tr == this_tr) {
9550 ret = __remove_instance(tr);
9551 break;
9552 }
9553 }
9554
9555 mutex_unlock(&trace_types_lock);
9556 mutex_unlock(&event_mutex);
9557
9558 return ret;
9559 }
9560 EXPORT_SYMBOL_GPL(trace_array_destroy);
9561
instance_rmdir(const char * name)9562 static int instance_rmdir(const char *name)
9563 {
9564 struct trace_array *tr;
9565 int ret;
9566
9567 mutex_lock(&event_mutex);
9568 mutex_lock(&trace_types_lock);
9569
9570 ret = -ENODEV;
9571 tr = trace_array_find(name);
9572 if (tr)
9573 ret = __remove_instance(tr);
9574
9575 mutex_unlock(&trace_types_lock);
9576 mutex_unlock(&event_mutex);
9577
9578 return ret;
9579 }
9580
create_trace_instances(struct dentry * d_tracer)9581 static __init void create_trace_instances(struct dentry *d_tracer)
9582 {
9583 struct trace_array *tr;
9584
9585 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9586 instance_mkdir,
9587 instance_rmdir);
9588 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9589 return;
9590
9591 mutex_lock(&event_mutex);
9592 mutex_lock(&trace_types_lock);
9593
9594 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9595 if (!tr->name)
9596 continue;
9597 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9598 "Failed to create instance directory\n"))
9599 break;
9600 }
9601
9602 mutex_unlock(&trace_types_lock);
9603 mutex_unlock(&event_mutex);
9604 }
9605
9606 static void
init_tracer_tracefs(struct trace_array * tr,struct dentry * d_tracer)9607 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9608 {
9609 int cpu;
9610
9611 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9612 tr, &show_traces_fops);
9613
9614 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9615 tr, &set_tracer_fops);
9616
9617 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9618 tr, &tracing_cpumask_fops);
9619
9620 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9621 tr, &tracing_iter_fops);
9622
9623 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9624 tr, &tracing_fops);
9625
9626 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9627 tr, &tracing_pipe_fops);
9628
9629 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9630 tr, &tracing_entries_fops);
9631
9632 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9633 tr, &tracing_total_entries_fops);
9634
9635 trace_create_file("free_buffer", 0200, d_tracer,
9636 tr, &tracing_free_buffer_fops);
9637
9638 trace_create_file("trace_marker", 0220, d_tracer,
9639 tr, &tracing_mark_fops);
9640
9641 tr->trace_marker_file = __find_event_file(tr, "ftrace", "print");
9642
9643 trace_create_file("trace_marker_raw", 0220, d_tracer,
9644 tr, &tracing_mark_raw_fops);
9645
9646 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9647 &trace_clock_fops);
9648
9649 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9650 tr, &rb_simple_fops);
9651
9652 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9653 &trace_time_stamp_mode_fops);
9654
9655 tr->buffer_percent = 50;
9656
9657 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9658 tr, &buffer_percent_fops);
9659
9660 trace_create_file("buffer_subbuf_size_kb", TRACE_MODE_WRITE, d_tracer,
9661 tr, &buffer_subbuf_size_fops);
9662
9663 create_trace_options_dir(tr);
9664
9665 #ifdef CONFIG_TRACER_MAX_TRACE
9666 trace_create_maxlat_file(tr, d_tracer);
9667 #endif
9668
9669 if (ftrace_create_function_files(tr, d_tracer))
9670 MEM_FAIL(1, "Could not allocate function filter files");
9671
9672 #ifdef CONFIG_TRACER_SNAPSHOT
9673 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9674 tr, &snapshot_fops);
9675 #endif
9676
9677 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9678 tr, &tracing_err_log_fops);
9679
9680 for_each_tracing_cpu(cpu)
9681 tracing_init_tracefs_percpu(tr, cpu);
9682
9683 ftrace_init_tracefs(tr, d_tracer);
9684 }
9685
trace_automount(struct dentry * mntpt,void * ingore)9686 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9687 {
9688 struct vfsmount *mnt;
9689 struct file_system_type *type;
9690
9691 /*
9692 * To maintain backward compatibility for tools that mount
9693 * debugfs to get to the tracing facility, tracefs is automatically
9694 * mounted to the debugfs/tracing directory.
9695 */
9696 type = get_fs_type("tracefs");
9697 if (!type)
9698 return NULL;
9699 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9700 put_filesystem(type);
9701 if (IS_ERR(mnt))
9702 return NULL;
9703 mntget(mnt);
9704
9705 return mnt;
9706 }
9707
9708 /**
9709 * tracing_init_dentry - initialize top level trace array
9710 *
9711 * This is called when creating files or directories in the tracing
9712 * directory. It is called via fs_initcall() by any of the boot up code
9713 * and expects to return the dentry of the top level tracing directory.
9714 */
tracing_init_dentry(void)9715 int tracing_init_dentry(void)
9716 {
9717 struct trace_array *tr = &global_trace;
9718
9719 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9720 pr_warn("Tracing disabled due to lockdown\n");
9721 return -EPERM;
9722 }
9723
9724 /* The top level trace array uses NULL as parent */
9725 if (tr->dir)
9726 return 0;
9727
9728 if (WARN_ON(!tracefs_initialized()))
9729 return -ENODEV;
9730
9731 /*
9732 * As there may still be users that expect the tracing
9733 * files to exist in debugfs/tracing, we must automount
9734 * the tracefs file system there, so older tools still
9735 * work with the newer kernel.
9736 */
9737 tr->dir = debugfs_create_automount("tracing", NULL,
9738 trace_automount, NULL);
9739
9740 return 0;
9741 }
9742
9743 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9744 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9745
9746 static struct workqueue_struct *eval_map_wq __initdata;
9747 static struct work_struct eval_map_work __initdata;
9748 static struct work_struct tracerfs_init_work __initdata;
9749
eval_map_work_func(struct work_struct * work)9750 static void __init eval_map_work_func(struct work_struct *work)
9751 {
9752 int len;
9753
9754 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9755 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9756 }
9757
trace_eval_init(void)9758 static int __init trace_eval_init(void)
9759 {
9760 INIT_WORK(&eval_map_work, eval_map_work_func);
9761
9762 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9763 if (!eval_map_wq) {
9764 pr_err("Unable to allocate eval_map_wq\n");
9765 /* Do work here */
9766 eval_map_work_func(&eval_map_work);
9767 return -ENOMEM;
9768 }
9769
9770 queue_work(eval_map_wq, &eval_map_work);
9771 return 0;
9772 }
9773
9774 subsys_initcall(trace_eval_init);
9775
trace_eval_sync(void)9776 static int __init trace_eval_sync(void)
9777 {
9778 /* Make sure the eval map updates are finished */
9779 if (eval_map_wq)
9780 destroy_workqueue(eval_map_wq);
9781 return 0;
9782 }
9783
9784 late_initcall_sync(trace_eval_sync);
9785
9786
9787 #ifdef CONFIG_MODULES
trace_module_add_evals(struct module * mod)9788 static void trace_module_add_evals(struct module *mod)
9789 {
9790 if (!mod->num_trace_evals)
9791 return;
9792
9793 /*
9794 * Modules with bad taint do not have events created, do
9795 * not bother with enums either.
9796 */
9797 if (trace_module_has_bad_taint(mod))
9798 return;
9799
9800 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9801 }
9802
9803 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
trace_module_remove_evals(struct module * mod)9804 static void trace_module_remove_evals(struct module *mod)
9805 {
9806 union trace_eval_map_item *map;
9807 union trace_eval_map_item **last = &trace_eval_maps;
9808
9809 if (!mod->num_trace_evals)
9810 return;
9811
9812 mutex_lock(&trace_eval_mutex);
9813
9814 map = trace_eval_maps;
9815
9816 while (map) {
9817 if (map->head.mod == mod)
9818 break;
9819 map = trace_eval_jmp_to_tail(map);
9820 last = &map->tail.next;
9821 map = map->tail.next;
9822 }
9823 if (!map)
9824 goto out;
9825
9826 *last = trace_eval_jmp_to_tail(map)->tail.next;
9827 kfree(map);
9828 out:
9829 mutex_unlock(&trace_eval_mutex);
9830 }
9831 #else
trace_module_remove_evals(struct module * mod)9832 static inline void trace_module_remove_evals(struct module *mod) { }
9833 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9834
trace_module_notify(struct notifier_block * self,unsigned long val,void * data)9835 static int trace_module_notify(struct notifier_block *self,
9836 unsigned long val, void *data)
9837 {
9838 struct module *mod = data;
9839
9840 switch (val) {
9841 case MODULE_STATE_COMING:
9842 trace_module_add_evals(mod);
9843 break;
9844 case MODULE_STATE_GOING:
9845 trace_module_remove_evals(mod);
9846 break;
9847 }
9848
9849 return NOTIFY_OK;
9850 }
9851
9852 static struct notifier_block trace_module_nb = {
9853 .notifier_call = trace_module_notify,
9854 .priority = 0,
9855 };
9856 #endif /* CONFIG_MODULES */
9857
tracer_init_tracefs_work_func(struct work_struct * work)9858 static __init void tracer_init_tracefs_work_func(struct work_struct *work)
9859 {
9860
9861 event_trace_init();
9862
9863 init_tracer_tracefs(&global_trace, NULL);
9864 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9865
9866 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9867 &global_trace, &tracing_thresh_fops);
9868
9869 trace_create_file("README", TRACE_MODE_READ, NULL,
9870 NULL, &tracing_readme_fops);
9871
9872 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9873 NULL, &tracing_saved_cmdlines_fops);
9874
9875 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9876 NULL, &tracing_saved_cmdlines_size_fops);
9877
9878 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9879 NULL, &tracing_saved_tgids_fops);
9880
9881 trace_create_eval_file(NULL);
9882
9883 #ifdef CONFIG_MODULES
9884 register_module_notifier(&trace_module_nb);
9885 #endif
9886
9887 #ifdef CONFIG_DYNAMIC_FTRACE
9888 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
9889 NULL, &tracing_dyn_info_fops);
9890 #endif
9891
9892 create_trace_instances(NULL);
9893
9894 update_tracer_options(&global_trace);
9895 }
9896
tracer_init_tracefs(void)9897 static __init int tracer_init_tracefs(void)
9898 {
9899 int ret;
9900
9901 trace_access_lock_init();
9902
9903 ret = tracing_init_dentry();
9904 if (ret)
9905 return 0;
9906
9907 if (eval_map_wq) {
9908 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
9909 queue_work(eval_map_wq, &tracerfs_init_work);
9910 } else {
9911 tracer_init_tracefs_work_func(NULL);
9912 }
9913
9914 rv_init_interface();
9915
9916 return 0;
9917 }
9918
9919 fs_initcall(tracer_init_tracefs);
9920
9921 static int trace_die_panic_handler(struct notifier_block *self,
9922 unsigned long ev, void *unused);
9923
9924 static struct notifier_block trace_panic_notifier = {
9925 .notifier_call = trace_die_panic_handler,
9926 .priority = INT_MAX - 1,
9927 };
9928
9929 static struct notifier_block trace_die_notifier = {
9930 .notifier_call = trace_die_panic_handler,
9931 .priority = INT_MAX - 1,
9932 };
9933
9934 /*
9935 * The idea is to execute the following die/panic callback early, in order
9936 * to avoid showing irrelevant information in the trace (like other panic
9937 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
9938 * warnings get disabled (to prevent potential log flooding).
9939 */
trace_die_panic_handler(struct notifier_block * self,unsigned long ev,void * unused)9940 static int trace_die_panic_handler(struct notifier_block *self,
9941 unsigned long ev, void *unused)
9942 {
9943 if (!ftrace_dump_on_oops_enabled())
9944 return NOTIFY_DONE;
9945
9946 /* The die notifier requires DIE_OOPS to trigger */
9947 if (self == &trace_die_notifier && ev != DIE_OOPS)
9948 return NOTIFY_DONE;
9949
9950 ftrace_dump(DUMP_PARAM);
9951
9952 return NOTIFY_DONE;
9953 }
9954
9955 /*
9956 * printk is set to max of 1024, we really don't need it that big.
9957 * Nothing should be printing 1000 characters anyway.
9958 */
9959 #define TRACE_MAX_PRINT 1000
9960
9961 /*
9962 * Define here KERN_TRACE so that we have one place to modify
9963 * it if we decide to change what log level the ftrace dump
9964 * should be at.
9965 */
9966 #define KERN_TRACE KERN_EMERG
9967
9968 void
trace_printk_seq(struct trace_seq * s)9969 trace_printk_seq(struct trace_seq *s)
9970 {
9971 /* Probably should print a warning here. */
9972 if (s->seq.len >= TRACE_MAX_PRINT)
9973 s->seq.len = TRACE_MAX_PRINT;
9974
9975 /*
9976 * More paranoid code. Although the buffer size is set to
9977 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9978 * an extra layer of protection.
9979 */
9980 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9981 s->seq.len = s->seq.size - 1;
9982
9983 /* should be zero ended, but we are paranoid. */
9984 s->buffer[s->seq.len] = 0;
9985
9986 printk(KERN_TRACE "%s", s->buffer);
9987
9988 trace_seq_init(s);
9989 }
9990
trace_init_iter(struct trace_iterator * iter,struct trace_array * tr)9991 static void trace_init_iter(struct trace_iterator *iter, struct trace_array *tr)
9992 {
9993 iter->tr = tr;
9994 iter->trace = iter->tr->current_trace;
9995 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9996 iter->array_buffer = &tr->array_buffer;
9997
9998 if (iter->trace && iter->trace->open)
9999 iter->trace->open(iter);
10000
10001 /* Annotate start of buffers if we had overruns */
10002 if (ring_buffer_overruns(iter->array_buffer->buffer))
10003 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10004
10005 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10006 if (trace_clocks[iter->tr->clock_id].in_ns)
10007 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10008
10009 /* Can not use kmalloc for iter.temp and iter.fmt */
10010 iter->temp = static_temp_buf;
10011 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10012 iter->fmt = static_fmt_buf;
10013 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10014 }
10015
trace_init_global_iter(struct trace_iterator * iter)10016 void trace_init_global_iter(struct trace_iterator *iter)
10017 {
10018 trace_init_iter(iter, &global_trace);
10019 }
10020
ftrace_dump_one(struct trace_array * tr,enum ftrace_dump_mode dump_mode)10021 static void ftrace_dump_one(struct trace_array *tr, enum ftrace_dump_mode dump_mode)
10022 {
10023 /* use static because iter can be a bit big for the stack */
10024 static struct trace_iterator iter;
10025 unsigned int old_userobj;
10026 unsigned long flags;
10027 int cnt = 0, cpu;
10028
10029 /*
10030 * Always turn off tracing when we dump.
10031 * We don't need to show trace output of what happens
10032 * between multiple crashes.
10033 *
10034 * If the user does a sysrq-z, then they can re-enable
10035 * tracing with echo 1 > tracing_on.
10036 */
10037 tracer_tracing_off(tr);
10038
10039 local_irq_save(flags);
10040
10041 /* Simulate the iterator */
10042 trace_init_iter(&iter, tr);
10043
10044 for_each_tracing_cpu(cpu) {
10045 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10046 }
10047
10048 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10049
10050 /* don't look at user memory in panic mode */
10051 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10052
10053 if (dump_mode == DUMP_ORIG)
10054 iter.cpu_file = raw_smp_processor_id();
10055 else
10056 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10057
10058 if (tr == &global_trace)
10059 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10060 else
10061 printk(KERN_TRACE "Dumping ftrace instance %s buffer:\n", tr->name);
10062
10063 /* Did function tracer already get disabled? */
10064 if (ftrace_is_dead()) {
10065 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10066 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10067 }
10068
10069 /*
10070 * We need to stop all tracing on all CPUS to read
10071 * the next buffer. This is a bit expensive, but is
10072 * not done often. We fill all what we can read,
10073 * and then release the locks again.
10074 */
10075
10076 while (!trace_empty(&iter)) {
10077
10078 if (!cnt)
10079 printk(KERN_TRACE "---------------------------------\n");
10080
10081 cnt++;
10082
10083 trace_iterator_reset(&iter);
10084 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10085
10086 if (trace_find_next_entry_inc(&iter) != NULL) {
10087 int ret;
10088
10089 ret = print_trace_line(&iter);
10090 if (ret != TRACE_TYPE_NO_CONSUME)
10091 trace_consume(&iter);
10092 }
10093 touch_nmi_watchdog();
10094
10095 trace_printk_seq(&iter.seq);
10096 }
10097
10098 if (!cnt)
10099 printk(KERN_TRACE " (ftrace buffer empty)\n");
10100 else
10101 printk(KERN_TRACE "---------------------------------\n");
10102
10103 tr->trace_flags |= old_userobj;
10104
10105 for_each_tracing_cpu(cpu) {
10106 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10107 }
10108 local_irq_restore(flags);
10109 }
10110
ftrace_dump_by_param(void)10111 static void ftrace_dump_by_param(void)
10112 {
10113 bool first_param = true;
10114 char dump_param[MAX_TRACER_SIZE];
10115 char *buf, *token, *inst_name;
10116 struct trace_array *tr;
10117
10118 strscpy(dump_param, ftrace_dump_on_oops, MAX_TRACER_SIZE);
10119 buf = dump_param;
10120
10121 while ((token = strsep(&buf, ",")) != NULL) {
10122 if (first_param) {
10123 first_param = false;
10124 if (!strcmp("0", token))
10125 continue;
10126 else if (!strcmp("1", token)) {
10127 ftrace_dump_one(&global_trace, DUMP_ALL);
10128 continue;
10129 }
10130 else if (!strcmp("2", token) ||
10131 !strcmp("orig_cpu", token)) {
10132 ftrace_dump_one(&global_trace, DUMP_ORIG);
10133 continue;
10134 }
10135 }
10136
10137 inst_name = strsep(&token, "=");
10138 tr = trace_array_find(inst_name);
10139 if (!tr) {
10140 printk(KERN_TRACE "Instance %s not found\n", inst_name);
10141 continue;
10142 }
10143
10144 if (token && (!strcmp("2", token) ||
10145 !strcmp("orig_cpu", token)))
10146 ftrace_dump_one(tr, DUMP_ORIG);
10147 else
10148 ftrace_dump_one(tr, DUMP_ALL);
10149 }
10150 }
10151
ftrace_dump(enum ftrace_dump_mode oops_dump_mode)10152 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10153 {
10154 static atomic_t dump_running;
10155
10156 /* Only allow one dump user at a time. */
10157 if (atomic_inc_return(&dump_running) != 1) {
10158 atomic_dec(&dump_running);
10159 return;
10160 }
10161
10162 switch (oops_dump_mode) {
10163 case DUMP_ALL:
10164 ftrace_dump_one(&global_trace, DUMP_ALL);
10165 break;
10166 case DUMP_ORIG:
10167 ftrace_dump_one(&global_trace, DUMP_ORIG);
10168 break;
10169 case DUMP_PARAM:
10170 ftrace_dump_by_param();
10171 break;
10172 case DUMP_NONE:
10173 break;
10174 default:
10175 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10176 ftrace_dump_one(&global_trace, DUMP_ALL);
10177 }
10178
10179 atomic_dec(&dump_running);
10180 }
10181 EXPORT_SYMBOL_GPL(ftrace_dump);
10182
10183 #define WRITE_BUFSIZE 4096
10184
trace_parse_run_command(struct file * file,const char __user * buffer,size_t count,loff_t * ppos,int (* createfn)(const char *))10185 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10186 size_t count, loff_t *ppos,
10187 int (*createfn)(const char *))
10188 {
10189 char *kbuf, *buf, *tmp;
10190 int ret = 0;
10191 size_t done = 0;
10192 size_t size;
10193
10194 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10195 if (!kbuf)
10196 return -ENOMEM;
10197
10198 while (done < count) {
10199 size = count - done;
10200
10201 if (size >= WRITE_BUFSIZE)
10202 size = WRITE_BUFSIZE - 1;
10203
10204 if (copy_from_user(kbuf, buffer + done, size)) {
10205 ret = -EFAULT;
10206 goto out;
10207 }
10208 kbuf[size] = '\0';
10209 buf = kbuf;
10210 do {
10211 tmp = strchr(buf, '\n');
10212 if (tmp) {
10213 *tmp = '\0';
10214 size = tmp - buf + 1;
10215 } else {
10216 size = strlen(buf);
10217 if (done + size < count) {
10218 if (buf != kbuf)
10219 break;
10220 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10221 pr_warn("Line length is too long: Should be less than %d\n",
10222 WRITE_BUFSIZE - 2);
10223 ret = -EINVAL;
10224 goto out;
10225 }
10226 }
10227 done += size;
10228
10229 /* Remove comments */
10230 tmp = strchr(buf, '#');
10231
10232 if (tmp)
10233 *tmp = '\0';
10234
10235 ret = createfn(buf);
10236 if (ret)
10237 goto out;
10238 buf += size;
10239
10240 } while (done < count);
10241 }
10242 ret = done;
10243
10244 out:
10245 kfree(kbuf);
10246
10247 return ret;
10248 }
10249
10250 #ifdef CONFIG_TRACER_MAX_TRACE
tr_needs_alloc_snapshot(const char * name)10251 __init static bool tr_needs_alloc_snapshot(const char *name)
10252 {
10253 char *test;
10254 int len = strlen(name);
10255 bool ret;
10256
10257 if (!boot_snapshot_index)
10258 return false;
10259
10260 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10261 boot_snapshot_info[len] == '\t')
10262 return true;
10263
10264 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10265 if (!test)
10266 return false;
10267
10268 sprintf(test, "\t%s\t", name);
10269 ret = strstr(boot_snapshot_info, test) == NULL;
10270 kfree(test);
10271 return ret;
10272 }
10273
do_allocate_snapshot(const char * name)10274 __init static void do_allocate_snapshot(const char *name)
10275 {
10276 if (!tr_needs_alloc_snapshot(name))
10277 return;
10278
10279 /*
10280 * When allocate_snapshot is set, the next call to
10281 * allocate_trace_buffers() (called by trace_array_get_by_name())
10282 * will allocate the snapshot buffer. That will alse clear
10283 * this flag.
10284 */
10285 allocate_snapshot = true;
10286 }
10287 #else
do_allocate_snapshot(const char * name)10288 static inline void do_allocate_snapshot(const char *name) { }
10289 #endif
10290
enable_instances(void)10291 __init static void enable_instances(void)
10292 {
10293 struct trace_array *tr;
10294 char *curr_str;
10295 char *str;
10296 char *tok;
10297
10298 /* A tab is always appended */
10299 boot_instance_info[boot_instance_index - 1] = '\0';
10300 str = boot_instance_info;
10301
10302 while ((curr_str = strsep(&str, "\t"))) {
10303
10304 tok = strsep(&curr_str, ",");
10305
10306 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10307 do_allocate_snapshot(tok);
10308
10309 tr = trace_array_get_by_name(tok, NULL);
10310 if (!tr) {
10311 pr_warn("Failed to create instance buffer %s\n", curr_str);
10312 continue;
10313 }
10314 /* Allow user space to delete it */
10315 trace_array_put(tr);
10316
10317 while ((tok = strsep(&curr_str, ","))) {
10318 early_enable_events(tr, tok, true);
10319 }
10320 }
10321 }
10322
tracer_alloc_buffers(void)10323 __init static int tracer_alloc_buffers(void)
10324 {
10325 int ring_buf_size;
10326 int ret = -ENOMEM;
10327
10328
10329 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10330 pr_warn("Tracing disabled due to lockdown\n");
10331 return -EPERM;
10332 }
10333
10334 /*
10335 * Make sure we don't accidentally add more trace options
10336 * than we have bits for.
10337 */
10338 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10339
10340 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10341 goto out;
10342
10343 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10344 goto out_free_buffer_mask;
10345
10346 /* Only allocate trace_printk buffers if a trace_printk exists */
10347 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10348 /* Must be called before global_trace.buffer is allocated */
10349 trace_printk_init_buffers();
10350
10351 /* To save memory, keep the ring buffer size to its minimum */
10352 if (global_trace.ring_buffer_expanded)
10353 ring_buf_size = trace_buf_size;
10354 else
10355 ring_buf_size = 1;
10356
10357 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10358 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10359
10360 raw_spin_lock_init(&global_trace.start_lock);
10361
10362 /*
10363 * The prepare callbacks allocates some memory for the ring buffer. We
10364 * don't free the buffer if the CPU goes down. If we were to free
10365 * the buffer, then the user would lose any trace that was in the
10366 * buffer. The memory will be removed once the "instance" is removed.
10367 */
10368 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10369 "trace/RB:prepare", trace_rb_cpu_prepare,
10370 NULL);
10371 if (ret < 0)
10372 goto out_free_cpumask;
10373 /* Used for event triggers */
10374 ret = -ENOMEM;
10375 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10376 if (!temp_buffer)
10377 goto out_rm_hp_state;
10378
10379 if (trace_create_savedcmd() < 0)
10380 goto out_free_temp_buffer;
10381
10382 if (!zalloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10383 goto out_free_savedcmd;
10384
10385 /* TODO: make the number of buffers hot pluggable with CPUS */
10386 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10387 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10388 goto out_free_pipe_cpumask;
10389 }
10390 if (global_trace.buffer_disabled)
10391 tracing_off();
10392
10393 if (trace_boot_clock) {
10394 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10395 if (ret < 0)
10396 pr_warn("Trace clock %s not defined, going back to default\n",
10397 trace_boot_clock);
10398 }
10399
10400 /*
10401 * register_tracer() might reference current_trace, so it
10402 * needs to be set before we register anything. This is
10403 * just a bootstrap of current_trace anyway.
10404 */
10405 global_trace.current_trace = &nop_trace;
10406
10407 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10408 #ifdef CONFIG_TRACER_MAX_TRACE
10409 spin_lock_init(&global_trace.snapshot_trigger_lock);
10410 #endif
10411 ftrace_init_global_array_ops(&global_trace);
10412
10413 init_trace_flags_index(&global_trace);
10414
10415 register_tracer(&nop_trace);
10416
10417 /* Function tracing may start here (via kernel command line) */
10418 init_function_trace();
10419
10420 /* All seems OK, enable tracing */
10421 tracing_disabled = 0;
10422
10423 atomic_notifier_chain_register(&panic_notifier_list,
10424 &trace_panic_notifier);
10425
10426 register_die_notifier(&trace_die_notifier);
10427
10428 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10429
10430 INIT_LIST_HEAD(&global_trace.systems);
10431 INIT_LIST_HEAD(&global_trace.events);
10432 INIT_LIST_HEAD(&global_trace.hist_vars);
10433 INIT_LIST_HEAD(&global_trace.err_log);
10434 list_add(&global_trace.list, &ftrace_trace_arrays);
10435
10436 apply_trace_boot_options();
10437
10438 register_snapshot_cmd();
10439
10440 test_can_verify();
10441
10442 return 0;
10443
10444 out_free_pipe_cpumask:
10445 free_cpumask_var(global_trace.pipe_cpumask);
10446 out_free_savedcmd:
10447 trace_free_saved_cmdlines_buffer();
10448 out_free_temp_buffer:
10449 ring_buffer_free(temp_buffer);
10450 out_rm_hp_state:
10451 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10452 out_free_cpumask:
10453 free_cpumask_var(global_trace.tracing_cpumask);
10454 out_free_buffer_mask:
10455 free_cpumask_var(tracing_buffer_mask);
10456 out:
10457 return ret;
10458 }
10459
ftrace_boot_snapshot(void)10460 void __init ftrace_boot_snapshot(void)
10461 {
10462 #ifdef CONFIG_TRACER_MAX_TRACE
10463 struct trace_array *tr;
10464
10465 if (!snapshot_at_boot)
10466 return;
10467
10468 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10469 if (!tr->allocated_snapshot)
10470 continue;
10471
10472 tracing_snapshot_instance(tr);
10473 trace_array_puts(tr, "** Boot snapshot taken **\n");
10474 }
10475 #endif
10476 }
10477
early_trace_init(void)10478 void __init early_trace_init(void)
10479 {
10480 if (tracepoint_printk) {
10481 tracepoint_print_iter =
10482 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10483 if (MEM_FAIL(!tracepoint_print_iter,
10484 "Failed to allocate trace iterator\n"))
10485 tracepoint_printk = 0;
10486 else
10487 static_key_enable(&tracepoint_printk_key.key);
10488 }
10489 tracer_alloc_buffers();
10490
10491 init_events();
10492 }
10493
trace_init(void)10494 void __init trace_init(void)
10495 {
10496 trace_event_init();
10497
10498 if (boot_instance_index)
10499 enable_instances();
10500 }
10501
clear_boot_tracer(void)10502 __init static void clear_boot_tracer(void)
10503 {
10504 /*
10505 * The default tracer at boot buffer is an init section.
10506 * This function is called in lateinit. If we did not
10507 * find the boot tracer, then clear it out, to prevent
10508 * later registration from accessing the buffer that is
10509 * about to be freed.
10510 */
10511 if (!default_bootup_tracer)
10512 return;
10513
10514 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10515 default_bootup_tracer);
10516 default_bootup_tracer = NULL;
10517 }
10518
10519 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
tracing_set_default_clock(void)10520 __init static void tracing_set_default_clock(void)
10521 {
10522 /* sched_clock_stable() is determined in late_initcall */
10523 if (!trace_boot_clock && !sched_clock_stable()) {
10524 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10525 pr_warn("Can not set tracing clock due to lockdown\n");
10526 return;
10527 }
10528
10529 printk(KERN_WARNING
10530 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10531 "If you want to keep using the local clock, then add:\n"
10532 " \"trace_clock=local\"\n"
10533 "on the kernel command line\n");
10534 tracing_set_clock(&global_trace, "global");
10535 }
10536 }
10537 #else
tracing_set_default_clock(void)10538 static inline void tracing_set_default_clock(void) { }
10539 #endif
10540
late_trace_init(void)10541 __init static int late_trace_init(void)
10542 {
10543 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10544 static_key_disable(&tracepoint_printk_key.key);
10545 tracepoint_printk = 0;
10546 }
10547
10548 tracing_set_default_clock();
10549 clear_boot_tracer();
10550 return 0;
10551 }
10552
10553 late_initcall_sync(late_trace_init);
10554