]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * ring buffer based function tracer | |
3 | * | |
4 | * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com> | |
5 | * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com> | |
6 | * | |
7 | * Originally taken from the RT patch by: | |
8 | * Arnaldo Carvalho de Melo <acme@redhat.com> | |
9 | * | |
10 | * Based on code from the latency_tracer, that is: | |
11 | * Copyright (C) 2004-2006 Ingo Molnar | |
12 | * Copyright (C) 2004 Nadia Yvette Chambers | |
13 | */ | |
14 | #include <linux/ring_buffer.h> | |
15 | #include <generated/utsrelease.h> | |
16 | #include <linux/stacktrace.h> | |
17 | #include <linux/writeback.h> | |
18 | #include <linux/kallsyms.h> | |
19 | #include <linux/seq_file.h> | |
20 | #include <linux/notifier.h> | |
21 | #include <linux/irqflags.h> | |
22 | #include <linux/debugfs.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/hardirq.h> | |
25 | #include <linux/linkage.h> | |
26 | #include <linux/uaccess.h> | |
27 | #include <linux/kprobes.h> | |
28 | #include <linux/ftrace.h> | |
29 | #include <linux/module.h> | |
30 | #include <linux/percpu.h> | |
31 | #include <linux/splice.h> | |
32 | #include <linux/kdebug.h> | |
33 | #include <linux/string.h> | |
34 | #include <linux/rwsem.h> | |
35 | #include <linux/slab.h> | |
36 | #include <linux/ctype.h> | |
37 | #include <linux/init.h> | |
38 | #include <linux/poll.h> | |
39 | #include <linux/nmi.h> | |
40 | #include <linux/fs.h> | |
41 | #include <linux/sched/rt.h> | |
42 | ||
43 | #include "trace.h" | |
44 | #include "trace_output.h" | |
45 | ||
46 | /* | |
47 | * On boot up, the ring buffer is set to the minimum size, so that | |
48 | * we do not waste memory on systems that are not using tracing. | |
49 | */ | |
50 | bool ring_buffer_expanded; | |
51 | ||
52 | /* | |
53 | * We need to change this state when a selftest is running. | |
54 | * A selftest will lurk into the ring-buffer to count the | |
55 | * entries inserted during the selftest although some concurrent | |
56 | * insertions into the ring-buffer such as trace_printk could occurred | |
57 | * at the same time, giving false positive or negative results. | |
58 | */ | |
59 | static bool __read_mostly tracing_selftest_running; | |
60 | ||
61 | /* | |
62 | * If a tracer is running, we do not want to run SELFTEST. | |
63 | */ | |
64 | bool __read_mostly tracing_selftest_disabled; | |
65 | ||
66 | /* For tracers that don't implement custom flags */ | |
67 | static struct tracer_opt dummy_tracer_opt[] = { | |
68 | { } | |
69 | }; | |
70 | ||
71 | static struct tracer_flags dummy_tracer_flags = { | |
72 | .val = 0, | |
73 | .opts = dummy_tracer_opt | |
74 | }; | |
75 | ||
76 | static int | |
77 | dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set) | |
78 | { | |
79 | return 0; | |
80 | } | |
81 | ||
82 | /* | |
83 | * To prevent the comm cache from being overwritten when no | |
84 | * tracing is active, only save the comm when a trace event | |
85 | * occurred. | |
86 | */ | |
87 | static DEFINE_PER_CPU(bool, trace_cmdline_save); | |
88 | ||
89 | /* | |
90 | * Kill all tracing for good (never come back). | |
91 | * It is initialized to 1 but will turn to zero if the initialization | |
92 | * of the tracer is successful. But that is the only place that sets | |
93 | * this back to zero. | |
94 | */ | |
95 | static int tracing_disabled = 1; | |
96 | ||
97 | DEFINE_PER_CPU(int, ftrace_cpu_disabled); | |
98 | ||
99 | cpumask_var_t __read_mostly tracing_buffer_mask; | |
100 | ||
101 | /* | |
102 | * ftrace_dump_on_oops - variable to dump ftrace buffer on oops | |
103 | * | |
104 | * If there is an oops (or kernel panic) and the ftrace_dump_on_oops | |
105 | * is set, then ftrace_dump is called. This will output the contents | |
106 | * of the ftrace buffers to the console. This is very useful for | |
107 | * capturing traces that lead to crashes and outputing it to a | |
108 | * serial console. | |
109 | * | |
110 | * It is default off, but you can enable it with either specifying | |
111 | * "ftrace_dump_on_oops" in the kernel command line, or setting | |
112 | * /proc/sys/kernel/ftrace_dump_on_oops | |
113 | * Set 1 if you want to dump buffers of all CPUs | |
114 | * Set 2 if you want to dump the buffer of the CPU that triggered oops | |
115 | */ | |
116 | ||
117 | enum ftrace_dump_mode ftrace_dump_on_oops; | |
118 | ||
119 | /* When set, tracing will stop when a WARN*() is hit */ | |
120 | int __disable_trace_on_warning; | |
121 | ||
122 | static int tracing_set_tracer(struct trace_array *tr, const char *buf); | |
123 | ||
124 | #define MAX_TRACER_SIZE 100 | |
125 | static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata; | |
126 | static char *default_bootup_tracer; | |
127 | ||
128 | static bool allocate_snapshot; | |
129 | ||
130 | static int __init set_cmdline_ftrace(char *str) | |
131 | { | |
132 | strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE); | |
133 | default_bootup_tracer = bootup_tracer_buf; | |
134 | /* We are using ftrace early, expand it */ | |
135 | ring_buffer_expanded = true; | |
136 | return 1; | |
137 | } | |
138 | __setup("ftrace=", set_cmdline_ftrace); | |
139 | ||
140 | static int __init set_ftrace_dump_on_oops(char *str) | |
141 | { | |
142 | if (*str++ != '=' || !*str) { | |
143 | ftrace_dump_on_oops = DUMP_ALL; | |
144 | return 1; | |
145 | } | |
146 | ||
147 | if (!strcmp("orig_cpu", str)) { | |
148 | ftrace_dump_on_oops = DUMP_ORIG; | |
149 | return 1; | |
150 | } | |
151 | ||
152 | return 0; | |
153 | } | |
154 | __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops); | |
155 | ||
156 | static int __init stop_trace_on_warning(char *str) | |
157 | { | |
158 | __disable_trace_on_warning = 1; | |
159 | return 1; | |
160 | } | |
161 | __setup("traceoff_on_warning=", stop_trace_on_warning); | |
162 | ||
163 | static int __init boot_alloc_snapshot(char *str) | |
164 | { | |
165 | allocate_snapshot = true; | |
166 | /* We also need the main ring buffer expanded */ | |
167 | ring_buffer_expanded = true; | |
168 | return 1; | |
169 | } | |
170 | __setup("alloc_snapshot", boot_alloc_snapshot); | |
171 | ||
172 | ||
173 | static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata; | |
174 | static char *trace_boot_options __initdata; | |
175 | ||
176 | static int __init set_trace_boot_options(char *str) | |
177 | { | |
178 | strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE); | |
179 | trace_boot_options = trace_boot_options_buf; | |
180 | return 0; | |
181 | } | |
182 | __setup("trace_options=", set_trace_boot_options); | |
183 | ||
184 | static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata; | |
185 | static char *trace_boot_clock __initdata; | |
186 | ||
187 | static int __init set_trace_boot_clock(char *str) | |
188 | { | |
189 | strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE); | |
190 | trace_boot_clock = trace_boot_clock_buf; | |
191 | return 0; | |
192 | } | |
193 | __setup("trace_clock=", set_trace_boot_clock); | |
194 | ||
195 | ||
196 | unsigned long long ns2usecs(cycle_t nsec) | |
197 | { | |
198 | nsec += 500; | |
199 | do_div(nsec, 1000); | |
200 | return nsec; | |
201 | } | |
202 | ||
203 | /* | |
204 | * The global_trace is the descriptor that holds the tracing | |
205 | * buffers for the live tracing. For each CPU, it contains | |
206 | * a link list of pages that will store trace entries. The | |
207 | * page descriptor of the pages in the memory is used to hold | |
208 | * the link list by linking the lru item in the page descriptor | |
209 | * to each of the pages in the buffer per CPU. | |
210 | * | |
211 | * For each active CPU there is a data field that holds the | |
212 | * pages for the buffer for that CPU. Each CPU has the same number | |
213 | * of pages allocated for its buffer. | |
214 | */ | |
215 | static struct trace_array global_trace; | |
216 | ||
217 | LIST_HEAD(ftrace_trace_arrays); | |
218 | ||
219 | int trace_array_get(struct trace_array *this_tr) | |
220 | { | |
221 | struct trace_array *tr; | |
222 | int ret = -ENODEV; | |
223 | ||
224 | mutex_lock(&trace_types_lock); | |
225 | list_for_each_entry(tr, &ftrace_trace_arrays, list) { | |
226 | if (tr == this_tr) { | |
227 | tr->ref++; | |
228 | ret = 0; | |
229 | break; | |
230 | } | |
231 | } | |
232 | mutex_unlock(&trace_types_lock); | |
233 | ||
234 | return ret; | |
235 | } | |
236 | ||
237 | static void __trace_array_put(struct trace_array *this_tr) | |
238 | { | |
239 | WARN_ON(!this_tr->ref); | |
240 | this_tr->ref--; | |
241 | } | |
242 | ||
243 | void trace_array_put(struct trace_array *this_tr) | |
244 | { | |
245 | mutex_lock(&trace_types_lock); | |
246 | __trace_array_put(this_tr); | |
247 | mutex_unlock(&trace_types_lock); | |
248 | } | |
249 | ||
250 | int filter_check_discard(struct ftrace_event_file *file, void *rec, | |
251 | struct ring_buffer *buffer, | |
252 | struct ring_buffer_event *event) | |
253 | { | |
254 | if (unlikely(file->flags & FTRACE_EVENT_FL_FILTERED) && | |
255 | !filter_match_preds(file->filter, rec)) { | |
256 | ring_buffer_discard_commit(buffer, event); | |
257 | return 1; | |
258 | } | |
259 | ||
260 | return 0; | |
261 | } | |
262 | EXPORT_SYMBOL_GPL(filter_check_discard); | |
263 | ||
264 | int call_filter_check_discard(struct ftrace_event_call *call, void *rec, | |
265 | struct ring_buffer *buffer, | |
266 | struct ring_buffer_event *event) | |
267 | { | |
268 | if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) && | |
269 | !filter_match_preds(call->filter, rec)) { | |
270 | ring_buffer_discard_commit(buffer, event); | |
271 | return 1; | |
272 | } | |
273 | ||
274 | return 0; | |
275 | } | |
276 | EXPORT_SYMBOL_GPL(call_filter_check_discard); | |
277 | ||
278 | static cycle_t buffer_ftrace_now(struct trace_buffer *buf, int cpu) | |
279 | { | |
280 | u64 ts; | |
281 | ||
282 | /* Early boot up does not have a buffer yet */ | |
283 | if (!buf->buffer) | |
284 | return trace_clock_local(); | |
285 | ||
286 | ts = ring_buffer_time_stamp(buf->buffer, cpu); | |
287 | ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts); | |
288 | ||
289 | return ts; | |
290 | } | |
291 | ||
292 | cycle_t ftrace_now(int cpu) | |
293 | { | |
294 | return buffer_ftrace_now(&global_trace.trace_buffer, cpu); | |
295 | } | |
296 | ||
297 | /** | |
298 | * tracing_is_enabled - Show if global_trace has been disabled | |
299 | * | |
300 | * Shows if the global trace has been enabled or not. It uses the | |
301 | * mirror flag "buffer_disabled" to be used in fast paths such as for | |
302 | * the irqsoff tracer. But it may be inaccurate due to races. If you | |
303 | * need to know the accurate state, use tracing_is_on() which is a little | |
304 | * slower, but accurate. | |
305 | */ | |
306 | int tracing_is_enabled(void) | |
307 | { | |
308 | /* | |
309 | * For quick access (irqsoff uses this in fast path), just | |
310 | * return the mirror variable of the state of the ring buffer. | |
311 | * It's a little racy, but we don't really care. | |
312 | */ | |
313 | smp_rmb(); | |
314 | return !global_trace.buffer_disabled; | |
315 | } | |
316 | ||
317 | /* | |
318 | * trace_buf_size is the size in bytes that is allocated | |
319 | * for a buffer. Note, the number of bytes is always rounded | |
320 | * to page size. | |
321 | * | |
322 | * This number is purposely set to a low number of 16384. | |
323 | * If the dump on oops happens, it will be much appreciated | |
324 | * to not have to wait for all that output. Anyway this can be | |
325 | * boot time and run time configurable. | |
326 | */ | |
327 | #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */ | |
328 | ||
329 | static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT; | |
330 | ||
331 | /* trace_types holds a link list of available tracers. */ | |
332 | static struct tracer *trace_types __read_mostly; | |
333 | ||
334 | /* | |
335 | * trace_types_lock is used to protect the trace_types list. | |
336 | */ | |
337 | DEFINE_MUTEX(trace_types_lock); | |
338 | ||
339 | /* | |
340 | * serialize the access of the ring buffer | |
341 | * | |
342 | * ring buffer serializes readers, but it is low level protection. | |
343 | * The validity of the events (which returns by ring_buffer_peek() ..etc) | |
344 | * are not protected by ring buffer. | |
345 | * | |
346 | * The content of events may become garbage if we allow other process consumes | |
347 | * these events concurrently: | |
348 | * A) the page of the consumed events may become a normal page | |
349 | * (not reader page) in ring buffer, and this page will be rewrited | |
350 | * by events producer. | |
351 | * B) The page of the consumed events may become a page for splice_read, | |
352 | * and this page will be returned to system. | |
353 | * | |
354 | * These primitives allow multi process access to different cpu ring buffer | |
355 | * concurrently. | |
356 | * | |
357 | * These primitives don't distinguish read-only and read-consume access. | |
358 | * Multi read-only access are also serialized. | |
359 | */ | |
360 | ||
361 | #ifdef CONFIG_SMP | |
362 | static DECLARE_RWSEM(all_cpu_access_lock); | |
363 | static DEFINE_PER_CPU(struct mutex, cpu_access_lock); | |
364 | ||
365 | static inline void trace_access_lock(int cpu) | |
366 | { | |
367 | if (cpu == RING_BUFFER_ALL_CPUS) { | |
368 | /* gain it for accessing the whole ring buffer. */ | |
369 | down_write(&all_cpu_access_lock); | |
370 | } else { | |
371 | /* gain it for accessing a cpu ring buffer. */ | |
372 | ||
373 | /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */ | |
374 | down_read(&all_cpu_access_lock); | |
375 | ||
376 | /* Secondly block other access to this @cpu ring buffer. */ | |
377 | mutex_lock(&per_cpu(cpu_access_lock, cpu)); | |
378 | } | |
379 | } | |
380 | ||
381 | static inline void trace_access_unlock(int cpu) | |
382 | { | |
383 | if (cpu == RING_BUFFER_ALL_CPUS) { | |
384 | up_write(&all_cpu_access_lock); | |
385 | } else { | |
386 | mutex_unlock(&per_cpu(cpu_access_lock, cpu)); | |
387 | up_read(&all_cpu_access_lock); | |
388 | } | |
389 | } | |
390 | ||
391 | static inline void trace_access_lock_init(void) | |
392 | { | |
393 | int cpu; | |
394 | ||
395 | for_each_possible_cpu(cpu) | |
396 | mutex_init(&per_cpu(cpu_access_lock, cpu)); | |
397 | } | |
398 | ||
399 | #else | |
400 | ||
401 | static DEFINE_MUTEX(access_lock); | |
402 | ||
403 | static inline void trace_access_lock(int cpu) | |
404 | { | |
405 | (void)cpu; | |
406 | mutex_lock(&access_lock); | |
407 | } | |
408 | ||
409 | static inline void trace_access_unlock(int cpu) | |
410 | { | |
411 | (void)cpu; | |
412 | mutex_unlock(&access_lock); | |
413 | } | |
414 | ||
415 | static inline void trace_access_lock_init(void) | |
416 | { | |
417 | } | |
418 | ||
419 | #endif | |
420 | ||
421 | /* trace_flags holds trace_options default values */ | |
422 | unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | | |
423 | TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME | | |
424 | TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | | |
425 | TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | TRACE_ITER_FUNCTION; | |
426 | ||
427 | static void tracer_tracing_on(struct trace_array *tr) | |
428 | { | |
429 | if (tr->trace_buffer.buffer) | |
430 | ring_buffer_record_on(tr->trace_buffer.buffer); | |
431 | /* | |
432 | * This flag is looked at when buffers haven't been allocated | |
433 | * yet, or by some tracers (like irqsoff), that just want to | |
434 | * know if the ring buffer has been disabled, but it can handle | |
435 | * races of where it gets disabled but we still do a record. | |
436 | * As the check is in the fast path of the tracers, it is more | |
437 | * important to be fast than accurate. | |
438 | */ | |
439 | tr->buffer_disabled = 0; | |
440 | /* Make the flag seen by readers */ | |
441 | smp_wmb(); | |
442 | } | |
443 | ||
444 | /** | |
445 | * tracing_on - enable tracing buffers | |
446 | * | |
447 | * This function enables tracing buffers that may have been | |
448 | * disabled with tracing_off. | |
449 | */ | |
450 | void tracing_on(void) | |
451 | { | |
452 | tracer_tracing_on(&global_trace); | |
453 | } | |
454 | EXPORT_SYMBOL_GPL(tracing_on); | |
455 | ||
456 | /** | |
457 | * __trace_puts - write a constant string into the trace buffer. | |
458 | * @ip: The address of the caller | |
459 | * @str: The constant string to write | |
460 | * @size: The size of the string. | |
461 | */ | |
462 | int __trace_puts(unsigned long ip, const char *str, int size) | |
463 | { | |
464 | struct ring_buffer_event *event; | |
465 | struct ring_buffer *buffer; | |
466 | struct print_entry *entry; | |
467 | unsigned long irq_flags; | |
468 | int alloc; | |
469 | int pc; | |
470 | ||
471 | pc = preempt_count(); | |
472 | ||
473 | if (unlikely(tracing_selftest_running || tracing_disabled)) | |
474 | return 0; | |
475 | ||
476 | alloc = sizeof(*entry) + size + 2; /* possible \n added */ | |
477 | ||
478 | local_save_flags(irq_flags); | |
479 | buffer = global_trace.trace_buffer.buffer; | |
480 | event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc, | |
481 | irq_flags, pc); | |
482 | if (!event) | |
483 | return 0; | |
484 | ||
485 | entry = ring_buffer_event_data(event); | |
486 | entry->ip = ip; | |
487 | ||
488 | memcpy(&entry->buf, str, size); | |
489 | ||
490 | /* Add a newline if necessary */ | |
491 | if (entry->buf[size - 1] != '\n') { | |
492 | entry->buf[size] = '\n'; | |
493 | entry->buf[size + 1] = '\0'; | |
494 | } else | |
495 | entry->buf[size] = '\0'; | |
496 | ||
497 | __buffer_unlock_commit(buffer, event); | |
498 | ftrace_trace_stack(buffer, irq_flags, 4, pc); | |
499 | ||
500 | return size; | |
501 | } | |
502 | EXPORT_SYMBOL_GPL(__trace_puts); | |
503 | ||
504 | /** | |
505 | * __trace_bputs - write the pointer to a constant string into trace buffer | |
506 | * @ip: The address of the caller | |
507 | * @str: The constant string to write to the buffer to | |
508 | */ | |
509 | int __trace_bputs(unsigned long ip, const char *str) | |
510 | { | |
511 | struct ring_buffer_event *event; | |
512 | struct ring_buffer *buffer; | |
513 | struct bputs_entry *entry; | |
514 | unsigned long irq_flags; | |
515 | int size = sizeof(struct bputs_entry); | |
516 | int pc; | |
517 | ||
518 | pc = preempt_count(); | |
519 | ||
520 | if (unlikely(tracing_selftest_running || tracing_disabled)) | |
521 | return 0; | |
522 | ||
523 | local_save_flags(irq_flags); | |
524 | buffer = global_trace.trace_buffer.buffer; | |
525 | event = trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size, | |
526 | irq_flags, pc); | |
527 | if (!event) | |
528 | return 0; | |
529 | ||
530 | entry = ring_buffer_event_data(event); | |
531 | entry->ip = ip; | |
532 | entry->str = str; | |
533 | ||
534 | __buffer_unlock_commit(buffer, event); | |
535 | ftrace_trace_stack(buffer, irq_flags, 4, pc); | |
536 | ||
537 | return 1; | |
538 | } | |
539 | EXPORT_SYMBOL_GPL(__trace_bputs); | |
540 | ||
541 | #ifdef CONFIG_TRACER_SNAPSHOT | |
542 | /** | |
543 | * trace_snapshot - take a snapshot of the current buffer. | |
544 | * | |
545 | * This causes a swap between the snapshot buffer and the current live | |
546 | * tracing buffer. You can use this to take snapshots of the live | |
547 | * trace when some condition is triggered, but continue to trace. | |
548 | * | |
549 | * Note, make sure to allocate the snapshot with either | |
550 | * a tracing_snapshot_alloc(), or by doing it manually | |
551 | * with: echo 1 > /sys/kernel/debug/tracing/snapshot | |
552 | * | |
553 | * If the snapshot buffer is not allocated, it will stop tracing. | |
554 | * Basically making a permanent snapshot. | |
555 | */ | |
556 | void tracing_snapshot(void) | |
557 | { | |
558 | struct trace_array *tr = &global_trace; | |
559 | struct tracer *tracer = tr->current_trace; | |
560 | unsigned long flags; | |
561 | ||
562 | if (in_nmi()) { | |
563 | internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n"); | |
564 | internal_trace_puts("*** snapshot is being ignored ***\n"); | |
565 | return; | |
566 | } | |
567 | ||
568 | if (!tr->allocated_snapshot) { | |
569 | internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n"); | |
570 | internal_trace_puts("*** stopping trace here! ***\n"); | |
571 | tracing_off(); | |
572 | return; | |
573 | } | |
574 | ||
575 | /* Note, snapshot can not be used when the tracer uses it */ | |
576 | if (tracer->use_max_tr) { | |
577 | internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n"); | |
578 | internal_trace_puts("*** Can not use snapshot (sorry) ***\n"); | |
579 | return; | |
580 | } | |
581 | ||
582 | local_irq_save(flags); | |
583 | update_max_tr(tr, current, smp_processor_id()); | |
584 | local_irq_restore(flags); | |
585 | } | |
586 | EXPORT_SYMBOL_GPL(tracing_snapshot); | |
587 | ||
588 | static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf, | |
589 | struct trace_buffer *size_buf, int cpu_id); | |
590 | static void set_buffer_entries(struct trace_buffer *buf, unsigned long val); | |
591 | ||
592 | static int alloc_snapshot(struct trace_array *tr) | |
593 | { | |
594 | int ret; | |
595 | ||
596 | if (!tr->allocated_snapshot) { | |
597 | ||
598 | /* allocate spare buffer */ | |
599 | ret = resize_buffer_duplicate_size(&tr->max_buffer, | |
600 | &tr->trace_buffer, RING_BUFFER_ALL_CPUS); | |
601 | if (ret < 0) | |
602 | return ret; | |
603 | ||
604 | tr->allocated_snapshot = true; | |
605 | } | |
606 | ||
607 | return 0; | |
608 | } | |
609 | ||
610 | static void free_snapshot(struct trace_array *tr) | |
611 | { | |
612 | /* | |
613 | * We don't free the ring buffer. instead, resize it because | |
614 | * The max_tr ring buffer has some state (e.g. ring->clock) and | |
615 | * we want preserve it. | |
616 | */ | |
617 | ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS); | |
618 | set_buffer_entries(&tr->max_buffer, 1); | |
619 | tracing_reset_online_cpus(&tr->max_buffer); | |
620 | tr->allocated_snapshot = false; | |
621 | } | |
622 | ||
623 | /** | |
624 | * tracing_alloc_snapshot - allocate snapshot buffer. | |
625 | * | |
626 | * This only allocates the snapshot buffer if it isn't already | |
627 | * allocated - it doesn't also take a snapshot. | |
628 | * | |
629 | * This is meant to be used in cases where the snapshot buffer needs | |
630 | * to be set up for events that can't sleep but need to be able to | |
631 | * trigger a snapshot. | |
632 | */ | |
633 | int tracing_alloc_snapshot(void) | |
634 | { | |
635 | struct trace_array *tr = &global_trace; | |
636 | int ret; | |
637 | ||
638 | ret = alloc_snapshot(tr); | |
639 | WARN_ON(ret < 0); | |
640 | ||
641 | return ret; | |
642 | } | |
643 | EXPORT_SYMBOL_GPL(tracing_alloc_snapshot); | |
644 | ||
645 | /** | |
646 | * trace_snapshot_alloc - allocate and take a snapshot of the current buffer. | |
647 | * | |
648 | * This is similar to trace_snapshot(), but it will allocate the | |
649 | * snapshot buffer if it isn't already allocated. Use this only | |
650 | * where it is safe to sleep, as the allocation may sleep. | |
651 | * | |
652 | * This causes a swap between the snapshot buffer and the current live | |
653 | * tracing buffer. You can use this to take snapshots of the live | |
654 | * trace when some condition is triggered, but continue to trace. | |
655 | */ | |
656 | void tracing_snapshot_alloc(void) | |
657 | { | |
658 | int ret; | |
659 | ||
660 | ret = tracing_alloc_snapshot(); | |
661 | if (ret < 0) | |
662 | return; | |
663 | ||
664 | tracing_snapshot(); | |
665 | } | |
666 | EXPORT_SYMBOL_GPL(tracing_snapshot_alloc); | |
667 | #else | |
668 | void tracing_snapshot(void) | |
669 | { | |
670 | WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used"); | |
671 | } | |
672 | EXPORT_SYMBOL_GPL(tracing_snapshot); | |
673 | int tracing_alloc_snapshot(void) | |
674 | { | |
675 | WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used"); | |
676 | return -ENODEV; | |
677 | } | |
678 | EXPORT_SYMBOL_GPL(tracing_alloc_snapshot); | |
679 | void tracing_snapshot_alloc(void) | |
680 | { | |
681 | /* Give warning */ | |
682 | tracing_snapshot(); | |
683 | } | |
684 | EXPORT_SYMBOL_GPL(tracing_snapshot_alloc); | |
685 | #endif /* CONFIG_TRACER_SNAPSHOT */ | |
686 | ||
687 | static void tracer_tracing_off(struct trace_array *tr) | |
688 | { | |
689 | if (tr->trace_buffer.buffer) | |
690 | ring_buffer_record_off(tr->trace_buffer.buffer); | |
691 | /* | |
692 | * This flag is looked at when buffers haven't been allocated | |
693 | * yet, or by some tracers (like irqsoff), that just want to | |
694 | * know if the ring buffer has been disabled, but it can handle | |
695 | * races of where it gets disabled but we still do a record. | |
696 | * As the check is in the fast path of the tracers, it is more | |
697 | * important to be fast than accurate. | |
698 | */ | |
699 | tr->buffer_disabled = 1; | |
700 | /* Make the flag seen by readers */ | |
701 | smp_wmb(); | |
702 | } | |
703 | ||
704 | /** | |
705 | * tracing_off - turn off tracing buffers | |
706 | * | |
707 | * This function stops the tracing buffers from recording data. | |
708 | * It does not disable any overhead the tracers themselves may | |
709 | * be causing. This function simply causes all recording to | |
710 | * the ring buffers to fail. | |
711 | */ | |
712 | void tracing_off(void) | |
713 | { | |
714 | tracer_tracing_off(&global_trace); | |
715 | } | |
716 | EXPORT_SYMBOL_GPL(tracing_off); | |
717 | ||
718 | void disable_trace_on_warning(void) | |
719 | { | |
720 | if (__disable_trace_on_warning) | |
721 | tracing_off(); | |
722 | } | |
723 | ||
724 | /** | |
725 | * tracer_tracing_is_on - show real state of ring buffer enabled | |
726 | * @tr : the trace array to know if ring buffer is enabled | |
727 | * | |
728 | * Shows real state of the ring buffer if it is enabled or not. | |
729 | */ | |
730 | static int tracer_tracing_is_on(struct trace_array *tr) | |
731 | { | |
732 | if (tr->trace_buffer.buffer) | |
733 | return ring_buffer_record_is_on(tr->trace_buffer.buffer); | |
734 | return !tr->buffer_disabled; | |
735 | } | |
736 | ||
737 | /** | |
738 | * tracing_is_on - show state of ring buffers enabled | |
739 | */ | |
740 | int tracing_is_on(void) | |
741 | { | |
742 | return tracer_tracing_is_on(&global_trace); | |
743 | } | |
744 | EXPORT_SYMBOL_GPL(tracing_is_on); | |
745 | ||
746 | static int __init set_buf_size(char *str) | |
747 | { | |
748 | unsigned long buf_size; | |
749 | ||
750 | if (!str) | |
751 | return 0; | |
752 | buf_size = memparse(str, &str); | |
753 | /* nr_entries can not be zero */ | |
754 | if (buf_size == 0) | |
755 | return 0; | |
756 | trace_buf_size = buf_size; | |
757 | return 1; | |
758 | } | |
759 | __setup("trace_buf_size=", set_buf_size); | |
760 | ||
761 | static int __init set_tracing_thresh(char *str) | |
762 | { | |
763 | unsigned long threshold; | |
764 | int ret; | |
765 | ||
766 | if (!str) | |
767 | return 0; | |
768 | ret = kstrtoul(str, 0, &threshold); | |
769 | if (ret < 0) | |
770 | return 0; | |
771 | tracing_thresh = threshold * 1000; | |
772 | return 1; | |
773 | } | |
774 | __setup("tracing_thresh=", set_tracing_thresh); | |
775 | ||
776 | unsigned long nsecs_to_usecs(unsigned long nsecs) | |
777 | { | |
778 | return nsecs / 1000; | |
779 | } | |
780 | ||
781 | /* These must match the bit postions in trace_iterator_flags */ | |
782 | static const char *trace_options[] = { | |
783 | "print-parent", | |
784 | "sym-offset", | |
785 | "sym-addr", | |
786 | "verbose", | |
787 | "raw", | |
788 | "hex", | |
789 | "bin", | |
790 | "block", | |
791 | "stacktrace", | |
792 | "trace_printk", | |
793 | "ftrace_preempt", | |
794 | "branch", | |
795 | "annotate", | |
796 | "userstacktrace", | |
797 | "sym-userobj", | |
798 | "printk-msg-only", | |
799 | "context-info", | |
800 | "latency-format", | |
801 | "sleep-time", | |
802 | "graph-time", | |
803 | "record-cmd", | |
804 | "overwrite", | |
805 | "disable_on_free", | |
806 | "irq-info", | |
807 | "markers", | |
808 | "function-trace", | |
809 | NULL | |
810 | }; | |
811 | ||
812 | static struct { | |
813 | u64 (*func)(void); | |
814 | const char *name; | |
815 | int in_ns; /* is this clock in nanoseconds? */ | |
816 | } trace_clocks[] = { | |
817 | { trace_clock_local, "local", 1 }, | |
818 | { trace_clock_global, "global", 1 }, | |
819 | { trace_clock_counter, "counter", 0 }, | |
820 | { trace_clock_jiffies, "uptime", 1 }, | |
821 | { trace_clock, "perf", 1 }, | |
822 | ARCH_TRACE_CLOCKS | |
823 | }; | |
824 | ||
825 | /* | |
826 | * trace_parser_get_init - gets the buffer for trace parser | |
827 | */ | |
828 | int trace_parser_get_init(struct trace_parser *parser, int size) | |
829 | { | |
830 | memset(parser, 0, sizeof(*parser)); | |
831 | ||
832 | parser->buffer = kmalloc(size, GFP_KERNEL); | |
833 | if (!parser->buffer) | |
834 | return 1; | |
835 | ||
836 | parser->size = size; | |
837 | return 0; | |
838 | } | |
839 | ||
840 | /* | |
841 | * trace_parser_put - frees the buffer for trace parser | |
842 | */ | |
843 | void trace_parser_put(struct trace_parser *parser) | |
844 | { | |
845 | kfree(parser->buffer); | |
846 | } | |
847 | ||
848 | /* | |
849 | * trace_get_user - reads the user input string separated by space | |
850 | * (matched by isspace(ch)) | |
851 | * | |
852 | * For each string found the 'struct trace_parser' is updated, | |
853 | * and the function returns. | |
854 | * | |
855 | * Returns number of bytes read. | |
856 | * | |
857 | * See kernel/trace/trace.h for 'struct trace_parser' details. | |
858 | */ | |
859 | int trace_get_user(struct trace_parser *parser, const char __user *ubuf, | |
860 | size_t cnt, loff_t *ppos) | |
861 | { | |
862 | char ch; | |
863 | size_t read = 0; | |
864 | ssize_t ret; | |
865 | ||
866 | if (!*ppos) | |
867 | trace_parser_clear(parser); | |
868 | ||
869 | ret = get_user(ch, ubuf++); | |
870 | if (ret) | |
871 | goto out; | |
872 | ||
873 | read++; | |
874 | cnt--; | |
875 | ||
876 | /* | |
877 | * The parser is not finished with the last write, | |
878 | * continue reading the user input without skipping spaces. | |
879 | */ | |
880 | if (!parser->cont) { | |
881 | /* skip white space */ | |
882 | while (cnt && isspace(ch)) { | |
883 | ret = get_user(ch, ubuf++); | |
884 | if (ret) | |
885 | goto out; | |
886 | read++; | |
887 | cnt--; | |
888 | } | |
889 | ||
890 | /* only spaces were written */ | |
891 | if (isspace(ch)) { | |
892 | *ppos += read; | |
893 | ret = read; | |
894 | goto out; | |
895 | } | |
896 | ||
897 | parser->idx = 0; | |
898 | } | |
899 | ||
900 | /* read the non-space input */ | |
901 | while (cnt && !isspace(ch)) { | |
902 | if (parser->idx < parser->size - 1) | |
903 | parser->buffer[parser->idx++] = ch; | |
904 | else { | |
905 | ret = -EINVAL; | |
906 | goto out; | |
907 | } | |
908 | ret = get_user(ch, ubuf++); | |
909 | if (ret) | |
910 | goto out; | |
911 | read++; | |
912 | cnt--; | |
913 | } | |
914 | ||
915 | /* We either got finished input or we have to wait for another call. */ | |
916 | if (isspace(ch)) { | |
917 | parser->buffer[parser->idx] = 0; | |
918 | parser->cont = false; | |
919 | } else if (parser->idx < parser->size - 1) { | |
920 | parser->cont = true; | |
921 | parser->buffer[parser->idx++] = ch; | |
922 | } else { | |
923 | ret = -EINVAL; | |
924 | goto out; | |
925 | } | |
926 | ||
927 | *ppos += read; | |
928 | ret = read; | |
929 | ||
930 | out: | |
931 | return ret; | |
932 | } | |
933 | ||
934 | ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt) | |
935 | { | |
936 | int len; | |
937 | int ret; | |
938 | ||
939 | if (!cnt) | |
940 | return 0; | |
941 | ||
942 | if (s->len <= s->readpos) | |
943 | return -EBUSY; | |
944 | ||
945 | len = s->len - s->readpos; | |
946 | if (cnt > len) | |
947 | cnt = len; | |
948 | ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt); | |
949 | if (ret == cnt) | |
950 | return -EFAULT; | |
951 | ||
952 | cnt -= ret; | |
953 | ||
954 | s->readpos += cnt; | |
955 | return cnt; | |
956 | } | |
957 | ||
958 | static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt) | |
959 | { | |
960 | int len; | |
961 | ||
962 | if (s->len <= s->readpos) | |
963 | return -EBUSY; | |
964 | ||
965 | len = s->len - s->readpos; | |
966 | if (cnt > len) | |
967 | cnt = len; | |
968 | memcpy(buf, s->buffer + s->readpos, cnt); | |
969 | ||
970 | s->readpos += cnt; | |
971 | return cnt; | |
972 | } | |
973 | ||
974 | unsigned long __read_mostly tracing_thresh; | |
975 | ||
976 | #ifdef CONFIG_TRACER_MAX_TRACE | |
977 | /* | |
978 | * Copy the new maximum trace into the separate maximum-trace | |
979 | * structure. (this way the maximum trace is permanently saved, | |
980 | * for later retrieval via /sys/kernel/debug/tracing/latency_trace) | |
981 | */ | |
982 | static void | |
983 | __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) | |
984 | { | |
985 | struct trace_buffer *trace_buf = &tr->trace_buffer; | |
986 | struct trace_buffer *max_buf = &tr->max_buffer; | |
987 | struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu); | |
988 | struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu); | |
989 | ||
990 | max_buf->cpu = cpu; | |
991 | max_buf->time_start = data->preempt_timestamp; | |
992 | ||
993 | max_data->saved_latency = tr->max_latency; | |
994 | max_data->critical_start = data->critical_start; | |
995 | max_data->critical_end = data->critical_end; | |
996 | ||
997 | memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN); | |
998 | max_data->pid = tsk->pid; | |
999 | /* | |
1000 | * If tsk == current, then use current_uid(), as that does not use | |
1001 | * RCU. The irq tracer can be called out of RCU scope. | |
1002 | */ | |
1003 | if (tsk == current) | |
1004 | max_data->uid = current_uid(); | |
1005 | else | |
1006 | max_data->uid = task_uid(tsk); | |
1007 | ||
1008 | max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO; | |
1009 | max_data->policy = tsk->policy; | |
1010 | max_data->rt_priority = tsk->rt_priority; | |
1011 | ||
1012 | /* record this tasks comm */ | |
1013 | tracing_record_cmdline(tsk); | |
1014 | } | |
1015 | ||
1016 | /** | |
1017 | * update_max_tr - snapshot all trace buffers from global_trace to max_tr | |
1018 | * @tr: tracer | |
1019 | * @tsk: the task with the latency | |
1020 | * @cpu: The cpu that initiated the trace. | |
1021 | * | |
1022 | * Flip the buffers between the @tr and the max_tr and record information | |
1023 | * about which task was the cause of this latency. | |
1024 | */ | |
1025 | void | |
1026 | update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) | |
1027 | { | |
1028 | struct ring_buffer *buf; | |
1029 | ||
1030 | if (tr->stop_count) | |
1031 | return; | |
1032 | ||
1033 | WARN_ON_ONCE(!irqs_disabled()); | |
1034 | ||
1035 | if (!tr->allocated_snapshot) { | |
1036 | /* Only the nop tracer should hit this when disabling */ | |
1037 | WARN_ON_ONCE(tr->current_trace != &nop_trace); | |
1038 | return; | |
1039 | } | |
1040 | ||
1041 | arch_spin_lock(&tr->max_lock); | |
1042 | ||
1043 | buf = tr->trace_buffer.buffer; | |
1044 | tr->trace_buffer.buffer = tr->max_buffer.buffer; | |
1045 | tr->max_buffer.buffer = buf; | |
1046 | ||
1047 | __update_max_tr(tr, tsk, cpu); | |
1048 | arch_spin_unlock(&tr->max_lock); | |
1049 | } | |
1050 | ||
1051 | /** | |
1052 | * update_max_tr_single - only copy one trace over, and reset the rest | |
1053 | * @tr - tracer | |
1054 | * @tsk - task with the latency | |
1055 | * @cpu - the cpu of the buffer to copy. | |
1056 | * | |
1057 | * Flip the trace of a single CPU buffer between the @tr and the max_tr. | |
1058 | */ | |
1059 | void | |
1060 | update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu) | |
1061 | { | |
1062 | int ret; | |
1063 | ||
1064 | if (tr->stop_count) | |
1065 | return; | |
1066 | ||
1067 | WARN_ON_ONCE(!irqs_disabled()); | |
1068 | if (!tr->allocated_snapshot) { | |
1069 | /* Only the nop tracer should hit this when disabling */ | |
1070 | WARN_ON_ONCE(tr->current_trace != &nop_trace); | |
1071 | return; | |
1072 | } | |
1073 | ||
1074 | arch_spin_lock(&tr->max_lock); | |
1075 | ||
1076 | ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu); | |
1077 | ||
1078 | if (ret == -EBUSY) { | |
1079 | /* | |
1080 | * We failed to swap the buffer due to a commit taking | |
1081 | * place on this CPU. We fail to record, but we reset | |
1082 | * the max trace buffer (no one writes directly to it) | |
1083 | * and flag that it failed. | |
1084 | */ | |
1085 | trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_, | |
1086 | "Failed to swap buffers due to commit in progress\n"); | |
1087 | } | |
1088 | ||
1089 | WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY); | |
1090 | ||
1091 | __update_max_tr(tr, tsk, cpu); | |
1092 | arch_spin_unlock(&tr->max_lock); | |
1093 | } | |
1094 | #endif /* CONFIG_TRACER_MAX_TRACE */ | |
1095 | ||
1096 | static int wait_on_pipe(struct trace_iterator *iter) | |
1097 | { | |
1098 | /* Iterators are static, they should be filled or empty */ | |
1099 | if (trace_buffer_iter(iter, iter->cpu_file)) | |
1100 | return 0; | |
1101 | ||
1102 | return ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file); | |
1103 | } | |
1104 | ||
1105 | #ifdef CONFIG_FTRACE_STARTUP_TEST | |
1106 | static int run_tracer_selftest(struct tracer *type) | |
1107 | { | |
1108 | struct trace_array *tr = &global_trace; | |
1109 | struct tracer *saved_tracer = tr->current_trace; | |
1110 | int ret; | |
1111 | ||
1112 | if (!type->selftest || tracing_selftest_disabled) | |
1113 | return 0; | |
1114 | ||
1115 | /* | |
1116 | * Run a selftest on this tracer. | |
1117 | * Here we reset the trace buffer, and set the current | |
1118 | * tracer to be this tracer. The tracer can then run some | |
1119 | * internal tracing to verify that everything is in order. | |
1120 | * If we fail, we do not register this tracer. | |
1121 | */ | |
1122 | tracing_reset_online_cpus(&tr->trace_buffer); | |
1123 | ||
1124 | tr->current_trace = type; | |
1125 | ||
1126 | #ifdef CONFIG_TRACER_MAX_TRACE | |
1127 | if (type->use_max_tr) { | |
1128 | /* If we expanded the buffers, make sure the max is expanded too */ | |
1129 | if (ring_buffer_expanded) | |
1130 | ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size, | |
1131 | RING_BUFFER_ALL_CPUS); | |
1132 | tr->allocated_snapshot = true; | |
1133 | } | |
1134 | #endif | |
1135 | ||
1136 | /* the test is responsible for initializing and enabling */ | |
1137 | pr_info("Testing tracer %s: ", type->name); | |
1138 | ret = type->selftest(type, tr); | |
1139 | /* the test is responsible for resetting too */ | |
1140 | tr->current_trace = saved_tracer; | |
1141 | if (ret) { | |
1142 | printk(KERN_CONT "FAILED!\n"); | |
1143 | /* Add the warning after printing 'FAILED' */ | |
1144 | WARN_ON(1); | |
1145 | return -1; | |
1146 | } | |
1147 | /* Only reset on passing, to avoid touching corrupted buffers */ | |
1148 | tracing_reset_online_cpus(&tr->trace_buffer); | |
1149 | ||
1150 | #ifdef CONFIG_TRACER_MAX_TRACE | |
1151 | if (type->use_max_tr) { | |
1152 | tr->allocated_snapshot = false; | |
1153 | ||
1154 | /* Shrink the max buffer again */ | |
1155 | if (ring_buffer_expanded) | |
1156 | ring_buffer_resize(tr->max_buffer.buffer, 1, | |
1157 | RING_BUFFER_ALL_CPUS); | |
1158 | } | |
1159 | #endif | |
1160 | ||
1161 | printk(KERN_CONT "PASSED\n"); | |
1162 | return 0; | |
1163 | } | |
1164 | #else | |
1165 | static inline int run_tracer_selftest(struct tracer *type) | |
1166 | { | |
1167 | return 0; | |
1168 | } | |
1169 | #endif /* CONFIG_FTRACE_STARTUP_TEST */ | |
1170 | ||
1171 | /** | |
1172 | * register_tracer - register a tracer with the ftrace system. | |
1173 | * @type - the plugin for the tracer | |
1174 | * | |
1175 | * Register a new plugin tracer. | |
1176 | */ | |
1177 | int register_tracer(struct tracer *type) | |
1178 | { | |
1179 | struct tracer *t; | |
1180 | int ret = 0; | |
1181 | ||
1182 | if (!type->name) { | |
1183 | pr_info("Tracer must have a name\n"); | |
1184 | return -1; | |
1185 | } | |
1186 | ||
1187 | if (strlen(type->name) >= MAX_TRACER_SIZE) { | |
1188 | pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE); | |
1189 | return -1; | |
1190 | } | |
1191 | ||
1192 | mutex_lock(&trace_types_lock); | |
1193 | ||
1194 | tracing_selftest_running = true; | |
1195 | ||
1196 | for (t = trace_types; t; t = t->next) { | |
1197 | if (strcmp(type->name, t->name) == 0) { | |
1198 | /* already found */ | |
1199 | pr_info("Tracer %s already registered\n", | |
1200 | type->name); | |
1201 | ret = -1; | |
1202 | goto out; | |
1203 | } | |
1204 | } | |
1205 | ||
1206 | if (!type->set_flag) | |
1207 | type->set_flag = &dummy_set_flag; | |
1208 | if (!type->flags) | |
1209 | type->flags = &dummy_tracer_flags; | |
1210 | else | |
1211 | if (!type->flags->opts) | |
1212 | type->flags->opts = dummy_tracer_opt; | |
1213 | ||
1214 | ret = run_tracer_selftest(type); | |
1215 | if (ret < 0) | |
1216 | goto out; | |
1217 | ||
1218 | type->next = trace_types; | |
1219 | trace_types = type; | |
1220 | ||
1221 | out: | |
1222 | tracing_selftest_running = false; | |
1223 | mutex_unlock(&trace_types_lock); | |
1224 | ||
1225 | if (ret || !default_bootup_tracer) | |
1226 | goto out_unlock; | |
1227 | ||
1228 | if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE)) | |
1229 | goto out_unlock; | |
1230 | ||
1231 | printk(KERN_INFO "Starting tracer '%s'\n", type->name); | |
1232 | /* Do we want this tracer to start on bootup? */ | |
1233 | tracing_set_tracer(&global_trace, type->name); | |
1234 | default_bootup_tracer = NULL; | |
1235 | /* disable other selftests, since this will break it. */ | |
1236 | tracing_selftest_disabled = true; | |
1237 | #ifdef CONFIG_FTRACE_STARTUP_TEST | |
1238 | printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n", | |
1239 | type->name); | |
1240 | #endif | |
1241 | ||
1242 | out_unlock: | |
1243 | return ret; | |
1244 | } | |
1245 | ||
1246 | void tracing_reset(struct trace_buffer *buf, int cpu) | |
1247 | { | |
1248 | struct ring_buffer *buffer = buf->buffer; | |
1249 | ||
1250 | if (!buffer) | |
1251 | return; | |
1252 | ||
1253 | ring_buffer_record_disable(buffer); | |
1254 | ||
1255 | /* Make sure all commits have finished */ | |
1256 | synchronize_sched(); | |
1257 | ring_buffer_reset_cpu(buffer, cpu); | |
1258 | ||
1259 | ring_buffer_record_enable(buffer); | |
1260 | } | |
1261 | ||
1262 | void tracing_reset_online_cpus(struct trace_buffer *buf) | |
1263 | { | |
1264 | struct ring_buffer *buffer = buf->buffer; | |
1265 | int cpu; | |
1266 | ||
1267 | if (!buffer) | |
1268 | return; | |
1269 | ||
1270 | ring_buffer_record_disable(buffer); | |
1271 | ||
1272 | /* Make sure all commits have finished */ | |
1273 | synchronize_sched(); | |
1274 | ||
1275 | buf->time_start = buffer_ftrace_now(buf, buf->cpu); | |
1276 | ||
1277 | for_each_online_cpu(cpu) | |
1278 | ring_buffer_reset_cpu(buffer, cpu); | |
1279 | ||
1280 | ring_buffer_record_enable(buffer); | |
1281 | } | |
1282 | ||
1283 | /* Must have trace_types_lock held */ | |
1284 | void tracing_reset_all_online_cpus(void) | |
1285 | { | |
1286 | struct trace_array *tr; | |
1287 | ||
1288 | list_for_each_entry(tr, &ftrace_trace_arrays, list) { | |
1289 | tracing_reset_online_cpus(&tr->trace_buffer); | |
1290 | #ifdef CONFIG_TRACER_MAX_TRACE | |
1291 | tracing_reset_online_cpus(&tr->max_buffer); | |
1292 | #endif | |
1293 | } | |
1294 | } | |
1295 | ||
1296 | #define SAVED_CMDLINES_DEFAULT 128 | |
1297 | #define NO_CMDLINE_MAP UINT_MAX | |
1298 | static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED; | |
1299 | struct saved_cmdlines_buffer { | |
1300 | unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1]; | |
1301 | unsigned *map_cmdline_to_pid; | |
1302 | unsigned cmdline_num; | |
1303 | int cmdline_idx; | |
1304 | char *saved_cmdlines; | |
1305 | }; | |
1306 | static struct saved_cmdlines_buffer *savedcmd; | |
1307 | ||
1308 | /* temporary disable recording */ | |
1309 | static atomic_t trace_record_cmdline_disabled __read_mostly; | |
1310 | ||
1311 | static inline char *get_saved_cmdlines(int idx) | |
1312 | { | |
1313 | return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN]; | |
1314 | } | |
1315 | ||
1316 | static inline void set_cmdline(int idx, const char *cmdline) | |
1317 | { | |
1318 | memcpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN); | |
1319 | } | |
1320 | ||
1321 | static int allocate_cmdlines_buffer(unsigned int val, | |
1322 | struct saved_cmdlines_buffer *s) | |
1323 | { | |
1324 | s->map_cmdline_to_pid = kmalloc(val * sizeof(*s->map_cmdline_to_pid), | |
1325 | GFP_KERNEL); | |
1326 | if (!s->map_cmdline_to_pid) | |
1327 | return -ENOMEM; | |
1328 | ||
1329 | s->saved_cmdlines = kmalloc(val * TASK_COMM_LEN, GFP_KERNEL); | |
1330 | if (!s->saved_cmdlines) { | |
1331 | kfree(s->map_cmdline_to_pid); | |
1332 | return -ENOMEM; | |
1333 | } | |
1334 | ||
1335 | s->cmdline_idx = 0; | |
1336 | s->cmdline_num = val; | |
1337 | memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP, | |
1338 | sizeof(s->map_pid_to_cmdline)); | |
1339 | memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP, | |
1340 | val * sizeof(*s->map_cmdline_to_pid)); | |
1341 | ||
1342 | return 0; | |
1343 | } | |
1344 | ||
1345 | static int trace_create_savedcmd(void) | |
1346 | { | |
1347 | int ret; | |
1348 | ||
1349 | savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL); | |
1350 | if (!savedcmd) | |
1351 | return -ENOMEM; | |
1352 | ||
1353 | ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd); | |
1354 | if (ret < 0) { | |
1355 | kfree(savedcmd); | |
1356 | savedcmd = NULL; | |
1357 | return -ENOMEM; | |
1358 | } | |
1359 | ||
1360 | return 0; | |
1361 | } | |
1362 | ||
1363 | int is_tracing_stopped(void) | |
1364 | { | |
1365 | return global_trace.stop_count; | |
1366 | } | |
1367 | ||
1368 | /** | |
1369 | * tracing_start - quick start of the tracer | |
1370 | * | |
1371 | * If tracing is enabled but was stopped by tracing_stop, | |
1372 | * this will start the tracer back up. | |
1373 | */ | |
1374 | void tracing_start(void) | |
1375 | { | |
1376 | struct ring_buffer *buffer; | |
1377 | unsigned long flags; | |
1378 | ||
1379 | if (tracing_disabled) | |
1380 | return; | |
1381 | ||
1382 | raw_spin_lock_irqsave(&global_trace.start_lock, flags); | |
1383 | if (--global_trace.stop_count) { | |
1384 | if (global_trace.stop_count < 0) { | |
1385 | /* Someone screwed up their debugging */ | |
1386 | WARN_ON_ONCE(1); | |
1387 | global_trace.stop_count = 0; | |
1388 | } | |
1389 | goto out; | |
1390 | } | |
1391 | ||
1392 | /* Prevent the buffers from switching */ | |
1393 | arch_spin_lock(&global_trace.max_lock); | |
1394 | ||
1395 | buffer = global_trace.trace_buffer.buffer; | |
1396 | if (buffer) | |
1397 | ring_buffer_record_enable(buffer); | |
1398 | ||
1399 | #ifdef CONFIG_TRACER_MAX_TRACE | |
1400 | buffer = global_trace.max_buffer.buffer; | |
1401 | if (buffer) | |
1402 | ring_buffer_record_enable(buffer); | |
1403 | #endif | |
1404 | ||
1405 | arch_spin_unlock(&global_trace.max_lock); | |
1406 | ||
1407 | out: | |
1408 | raw_spin_unlock_irqrestore(&global_trace.start_lock, flags); | |
1409 | } | |
1410 | ||
1411 | static void tracing_start_tr(struct trace_array *tr) | |
1412 | { | |
1413 | struct ring_buffer *buffer; | |
1414 | unsigned long flags; | |
1415 | ||
1416 | if (tracing_disabled) | |
1417 | return; | |
1418 | ||
1419 | /* If global, we need to also start the max tracer */ | |
1420 | if (tr->flags & TRACE_ARRAY_FL_GLOBAL) | |
1421 | return tracing_start(); | |
1422 | ||
1423 | raw_spin_lock_irqsave(&tr->start_lock, flags); | |
1424 | ||
1425 | if (--tr->stop_count) { | |
1426 | if (tr->stop_count < 0) { | |
1427 | /* Someone screwed up their debugging */ | |
1428 | WARN_ON_ONCE(1); | |
1429 | tr->stop_count = 0; | |
1430 | } | |
1431 | goto out; | |
1432 | } | |
1433 | ||
1434 | buffer = tr->trace_buffer.buffer; | |
1435 | if (buffer) | |
1436 | ring_buffer_record_enable(buffer); | |
1437 | ||
1438 | out: | |
1439 | raw_spin_unlock_irqrestore(&tr->start_lock, flags); | |
1440 | } | |
1441 | ||
1442 | /** | |
1443 | * tracing_stop - quick stop of the tracer | |
1444 | * | |
1445 | * Light weight way to stop tracing. Use in conjunction with | |
1446 | * tracing_start. | |
1447 | */ | |
1448 | void tracing_stop(void) | |
1449 | { | |
1450 | struct ring_buffer *buffer; | |
1451 | unsigned long flags; | |
1452 | ||
1453 | raw_spin_lock_irqsave(&global_trace.start_lock, flags); | |
1454 | if (global_trace.stop_count++) | |
1455 | goto out; | |
1456 | ||
1457 | /* Prevent the buffers from switching */ | |
1458 | arch_spin_lock(&global_trace.max_lock); | |
1459 | ||
1460 | buffer = global_trace.trace_buffer.buffer; | |
1461 | if (buffer) | |
1462 | ring_buffer_record_disable(buffer); | |
1463 | ||
1464 | #ifdef CONFIG_TRACER_MAX_TRACE | |
1465 | buffer = global_trace.max_buffer.buffer; | |
1466 | if (buffer) | |
1467 | ring_buffer_record_disable(buffer); | |
1468 | #endif | |
1469 | ||
1470 | arch_spin_unlock(&global_trace.max_lock); | |
1471 | ||
1472 | out: | |
1473 | raw_spin_unlock_irqrestore(&global_trace.start_lock, flags); | |
1474 | } | |
1475 | ||
1476 | static void tracing_stop_tr(struct trace_array *tr) | |
1477 | { | |
1478 | struct ring_buffer *buffer; | |
1479 | unsigned long flags; | |
1480 | ||
1481 | /* If global, we need to also stop the max tracer */ | |
1482 | if (tr->flags & TRACE_ARRAY_FL_GLOBAL) | |
1483 | return tracing_stop(); | |
1484 | ||
1485 | raw_spin_lock_irqsave(&tr->start_lock, flags); | |
1486 | if (tr->stop_count++) | |
1487 | goto out; | |
1488 | ||
1489 | buffer = tr->trace_buffer.buffer; | |
1490 | if (buffer) | |
1491 | ring_buffer_record_disable(buffer); | |
1492 | ||
1493 | out: | |
1494 | raw_spin_unlock_irqrestore(&tr->start_lock, flags); | |
1495 | } | |
1496 | ||
1497 | void trace_stop_cmdline_recording(void); | |
1498 | ||
1499 | static int trace_save_cmdline(struct task_struct *tsk) | |
1500 | { | |
1501 | unsigned pid, idx; | |
1502 | ||
1503 | if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT)) | |
1504 | return 0; | |
1505 | ||
1506 | /* | |
1507 | * It's not the end of the world if we don't get | |
1508 | * the lock, but we also don't want to spin | |
1509 | * nor do we want to disable interrupts, | |
1510 | * so if we miss here, then better luck next time. | |
1511 | */ | |
1512 | if (!arch_spin_trylock(&trace_cmdline_lock)) | |
1513 | return 0; | |
1514 | ||
1515 | idx = savedcmd->map_pid_to_cmdline[tsk->pid]; | |
1516 | if (idx == NO_CMDLINE_MAP) { | |
1517 | idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num; | |
1518 | ||
1519 | /* | |
1520 | * Check whether the cmdline buffer at idx has a pid | |
1521 | * mapped. We are going to overwrite that entry so we | |
1522 | * need to clear the map_pid_to_cmdline. Otherwise we | |
1523 | * would read the new comm for the old pid. | |
1524 | */ | |
1525 | pid = savedcmd->map_cmdline_to_pid[idx]; | |
1526 | if (pid != NO_CMDLINE_MAP) | |
1527 | savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP; | |
1528 | ||
1529 | savedcmd->map_cmdline_to_pid[idx] = tsk->pid; | |
1530 | savedcmd->map_pid_to_cmdline[tsk->pid] = idx; | |
1531 | ||
1532 | savedcmd->cmdline_idx = idx; | |
1533 | } | |
1534 | ||
1535 | set_cmdline(idx, tsk->comm); | |
1536 | ||
1537 | arch_spin_unlock(&trace_cmdline_lock); | |
1538 | ||
1539 | return 1; | |
1540 | } | |
1541 | ||
1542 | static void __trace_find_cmdline(int pid, char comm[]) | |
1543 | { | |
1544 | unsigned map; | |
1545 | ||
1546 | if (!pid) { | |
1547 | strcpy(comm, "<idle>"); | |
1548 | return; | |
1549 | } | |
1550 | ||
1551 | if (WARN_ON_ONCE(pid < 0)) { | |
1552 | strcpy(comm, "<XXX>"); | |
1553 | return; | |
1554 | } | |
1555 | ||
1556 | if (pid > PID_MAX_DEFAULT) { | |
1557 | strcpy(comm, "<...>"); | |
1558 | return; | |
1559 | } | |
1560 | ||
1561 | map = savedcmd->map_pid_to_cmdline[pid]; | |
1562 | if (map != NO_CMDLINE_MAP) | |
1563 | strcpy(comm, get_saved_cmdlines(map)); | |
1564 | else | |
1565 | strcpy(comm, "<...>"); | |
1566 | } | |
1567 | ||
1568 | void trace_find_cmdline(int pid, char comm[]) | |
1569 | { | |
1570 | preempt_disable(); | |
1571 | arch_spin_lock(&trace_cmdline_lock); | |
1572 | ||
1573 | __trace_find_cmdline(pid, comm); | |
1574 | ||
1575 | arch_spin_unlock(&trace_cmdline_lock); | |
1576 | preempt_enable(); | |
1577 | } | |
1578 | ||
1579 | void tracing_record_cmdline(struct task_struct *tsk) | |
1580 | { | |
1581 | if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on()) | |
1582 | return; | |
1583 | ||
1584 | if (!__this_cpu_read(trace_cmdline_save)) | |
1585 | return; | |
1586 | ||
1587 | if (trace_save_cmdline(tsk)) | |
1588 | __this_cpu_write(trace_cmdline_save, false); | |
1589 | } | |
1590 | ||
1591 | void | |
1592 | tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags, | |
1593 | int pc) | |
1594 | { | |
1595 | struct task_struct *tsk = current; | |
1596 | ||
1597 | entry->preempt_count = pc & 0xff; | |
1598 | entry->pid = (tsk) ? tsk->pid : 0; | |
1599 | entry->flags = | |
1600 | #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT | |
1601 | (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) | | |
1602 | #else | |
1603 | TRACE_FLAG_IRQS_NOSUPPORT | | |
1604 | #endif | |
1605 | ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) | | |
1606 | ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) | | |
1607 | (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) | | |
1608 | (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0); | |
1609 | } | |
1610 | EXPORT_SYMBOL_GPL(tracing_generic_entry_update); | |
1611 | ||
1612 | struct ring_buffer_event * | |
1613 | trace_buffer_lock_reserve(struct ring_buffer *buffer, | |
1614 | int type, | |
1615 | unsigned long len, | |
1616 | unsigned long flags, int pc) | |
1617 | { | |
1618 | struct ring_buffer_event *event; | |
1619 | ||
1620 | event = ring_buffer_lock_reserve(buffer, len); | |
1621 | if (event != NULL) { | |
1622 | struct trace_entry *ent = ring_buffer_event_data(event); | |
1623 | ||
1624 | tracing_generic_entry_update(ent, flags, pc); | |
1625 | ent->type = type; | |
1626 | } | |
1627 | ||
1628 | return event; | |
1629 | } | |
1630 | ||
1631 | void | |
1632 | __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event) | |
1633 | { | |
1634 | __this_cpu_write(trace_cmdline_save, true); | |
1635 | ring_buffer_unlock_commit(buffer, event); | |
1636 | } | |
1637 | ||
1638 | static inline void | |
1639 | __trace_buffer_unlock_commit(struct ring_buffer *buffer, | |
1640 | struct ring_buffer_event *event, | |
1641 | unsigned long flags, int pc) | |
1642 | { | |
1643 | __buffer_unlock_commit(buffer, event); | |
1644 | ||
1645 | ftrace_trace_stack(buffer, flags, 6, pc); | |
1646 | ftrace_trace_userstack(buffer, flags, pc); | |
1647 | } | |
1648 | ||
1649 | void trace_buffer_unlock_commit(struct ring_buffer *buffer, | |
1650 | struct ring_buffer_event *event, | |
1651 | unsigned long flags, int pc) | |
1652 | { | |
1653 | __trace_buffer_unlock_commit(buffer, event, flags, pc); | |
1654 | } | |
1655 | EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit); | |
1656 | ||
1657 | static struct ring_buffer *temp_buffer; | |
1658 | ||
1659 | struct ring_buffer_event * | |
1660 | trace_event_buffer_lock_reserve(struct ring_buffer **current_rb, | |
1661 | struct ftrace_event_file *ftrace_file, | |
1662 | int type, unsigned long len, | |
1663 | unsigned long flags, int pc) | |
1664 | { | |
1665 | struct ring_buffer_event *entry; | |
1666 | ||
1667 | *current_rb = ftrace_file->tr->trace_buffer.buffer; | |
1668 | entry = trace_buffer_lock_reserve(*current_rb, | |
1669 | type, len, flags, pc); | |
1670 | /* | |
1671 | * If tracing is off, but we have triggers enabled | |
1672 | * we still need to look at the event data. Use the temp_buffer | |
1673 | * to store the trace event for the tigger to use. It's recusive | |
1674 | * safe and will not be recorded anywhere. | |
1675 | */ | |
1676 | if (!entry && ftrace_file->flags & FTRACE_EVENT_FL_TRIGGER_COND) { | |
1677 | *current_rb = temp_buffer; | |
1678 | entry = trace_buffer_lock_reserve(*current_rb, | |
1679 | type, len, flags, pc); | |
1680 | } | |
1681 | return entry; | |
1682 | } | |
1683 | EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve); | |
1684 | ||
1685 | struct ring_buffer_event * | |
1686 | trace_current_buffer_lock_reserve(struct ring_buffer **current_rb, | |
1687 | int type, unsigned long len, | |
1688 | unsigned long flags, int pc) | |
1689 | { | |
1690 | *current_rb = global_trace.trace_buffer.buffer; | |
1691 | return trace_buffer_lock_reserve(*current_rb, | |
1692 | type, len, flags, pc); | |
1693 | } | |
1694 | EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve); | |
1695 | ||
1696 | void trace_current_buffer_unlock_commit(struct ring_buffer *buffer, | |
1697 | struct ring_buffer_event *event, | |
1698 | unsigned long flags, int pc) | |
1699 | { | |
1700 | __trace_buffer_unlock_commit(buffer, event, flags, pc); | |
1701 | } | |
1702 | EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit); | |
1703 | ||
1704 | void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer, | |
1705 | struct ring_buffer_event *event, | |
1706 | unsigned long flags, int pc, | |
1707 | struct pt_regs *regs) | |
1708 | { | |
1709 | __buffer_unlock_commit(buffer, event); | |
1710 | ||
1711 | ftrace_trace_stack_regs(buffer, flags, 0, pc, regs); | |
1712 | ftrace_trace_userstack(buffer, flags, pc); | |
1713 | } | |
1714 | EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs); | |
1715 | ||
1716 | void trace_current_buffer_discard_commit(struct ring_buffer *buffer, | |
1717 | struct ring_buffer_event *event) | |
1718 | { | |
1719 | ring_buffer_discard_commit(buffer, event); | |
1720 | } | |
1721 | EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit); | |
1722 | ||
1723 | void | |
1724 | trace_function(struct trace_array *tr, | |
1725 | unsigned long ip, unsigned long parent_ip, unsigned long flags, | |
1726 | int pc) | |
1727 | { | |
1728 | struct ftrace_event_call *call = &event_function; | |
1729 | struct ring_buffer *buffer = tr->trace_buffer.buffer; | |
1730 | struct ring_buffer_event *event; | |
1731 | struct ftrace_entry *entry; | |
1732 | ||
1733 | /* If we are reading the ring buffer, don't trace */ | |
1734 | if (unlikely(__this_cpu_read(ftrace_cpu_disabled))) | |
1735 | return; | |
1736 | ||
1737 | event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry), | |
1738 | flags, pc); | |
1739 | if (!event) | |
1740 | return; | |
1741 | entry = ring_buffer_event_data(event); | |
1742 | entry->ip = ip; | |
1743 | entry->parent_ip = parent_ip; | |
1744 | ||
1745 | if (!call_filter_check_discard(call, entry, buffer, event)) | |
1746 | __buffer_unlock_commit(buffer, event); | |
1747 | } | |
1748 | ||
1749 | #ifdef CONFIG_STACKTRACE | |
1750 | ||
1751 | #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long)) | |
1752 | struct ftrace_stack { | |
1753 | unsigned long calls[FTRACE_STACK_MAX_ENTRIES]; | |
1754 | }; | |
1755 | ||
1756 | static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack); | |
1757 | static DEFINE_PER_CPU(int, ftrace_stack_reserve); | |
1758 | ||
1759 | static void __ftrace_trace_stack(struct ring_buffer *buffer, | |
1760 | unsigned long flags, | |
1761 | int skip, int pc, struct pt_regs *regs) | |
1762 | { | |
1763 | struct ftrace_event_call *call = &event_kernel_stack; | |
1764 | struct ring_buffer_event *event; | |
1765 | struct stack_entry *entry; | |
1766 | struct stack_trace trace; | |
1767 | int use_stack; | |
1768 | int size = FTRACE_STACK_ENTRIES; | |
1769 | ||
1770 | trace.nr_entries = 0; | |
1771 | trace.skip = skip; | |
1772 | ||
1773 | /* | |
1774 | * Since events can happen in NMIs there's no safe way to | |
1775 | * use the per cpu ftrace_stacks. We reserve it and if an interrupt | |
1776 | * or NMI comes in, it will just have to use the default | |
1777 | * FTRACE_STACK_SIZE. | |
1778 | */ | |
1779 | preempt_disable_notrace(); | |
1780 | ||
1781 | use_stack = __this_cpu_inc_return(ftrace_stack_reserve); | |
1782 | /* | |
1783 | * We don't need any atomic variables, just a barrier. | |
1784 | * If an interrupt comes in, we don't care, because it would | |
1785 | * have exited and put the counter back to what we want. | |
1786 | * We just need a barrier to keep gcc from moving things | |
1787 | * around. | |
1788 | */ | |
1789 | barrier(); | |
1790 | if (use_stack == 1) { | |
1791 | trace.entries = this_cpu_ptr(ftrace_stack.calls); | |
1792 | trace.max_entries = FTRACE_STACK_MAX_ENTRIES; | |
1793 | ||
1794 | if (regs) | |
1795 | save_stack_trace_regs(regs, &trace); | |
1796 | else | |
1797 | save_stack_trace(&trace); | |
1798 | ||
1799 | if (trace.nr_entries > size) | |
1800 | size = trace.nr_entries; | |
1801 | } else | |
1802 | /* From now on, use_stack is a boolean */ | |
1803 | use_stack = 0; | |
1804 | ||
1805 | size *= sizeof(unsigned long); | |
1806 | ||
1807 | event = trace_buffer_lock_reserve(buffer, TRACE_STACK, | |
1808 | sizeof(*entry) + size, flags, pc); | |
1809 | if (!event) | |
1810 | goto out; | |
1811 | entry = ring_buffer_event_data(event); | |
1812 | ||
1813 | memset(&entry->caller, 0, size); | |
1814 | ||
1815 | if (use_stack) | |
1816 | memcpy(&entry->caller, trace.entries, | |
1817 | trace.nr_entries * sizeof(unsigned long)); | |
1818 | else { | |
1819 | trace.max_entries = FTRACE_STACK_ENTRIES; | |
1820 | trace.entries = entry->caller; | |
1821 | if (regs) | |
1822 | save_stack_trace_regs(regs, &trace); | |
1823 | else | |
1824 | save_stack_trace(&trace); | |
1825 | } | |
1826 | ||
1827 | entry->size = trace.nr_entries; | |
1828 | ||
1829 | if (!call_filter_check_discard(call, entry, buffer, event)) | |
1830 | __buffer_unlock_commit(buffer, event); | |
1831 | ||
1832 | out: | |
1833 | /* Again, don't let gcc optimize things here */ | |
1834 | barrier(); | |
1835 | __this_cpu_dec(ftrace_stack_reserve); | |
1836 | preempt_enable_notrace(); | |
1837 | ||
1838 | } | |
1839 | ||
1840 | void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags, | |
1841 | int skip, int pc, struct pt_regs *regs) | |
1842 | { | |
1843 | if (!(trace_flags & TRACE_ITER_STACKTRACE)) | |
1844 | return; | |
1845 | ||
1846 | __ftrace_trace_stack(buffer, flags, skip, pc, regs); | |
1847 | } | |
1848 | ||
1849 | void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags, | |
1850 | int skip, int pc) | |
1851 | { | |
1852 | if (!(trace_flags & TRACE_ITER_STACKTRACE)) | |
1853 | return; | |
1854 | ||
1855 | __ftrace_trace_stack(buffer, flags, skip, pc, NULL); | |
1856 | } | |
1857 | ||
1858 | void __trace_stack(struct trace_array *tr, unsigned long flags, int skip, | |
1859 | int pc) | |
1860 | { | |
1861 | __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL); | |
1862 | } | |
1863 | ||
1864 | /** | |
1865 | * trace_dump_stack - record a stack back trace in the trace buffer | |
1866 | * @skip: Number of functions to skip (helper handlers) | |
1867 | */ | |
1868 | void trace_dump_stack(int skip) | |
1869 | { | |
1870 | unsigned long flags; | |
1871 | ||
1872 | if (tracing_disabled || tracing_selftest_running) | |
1873 | return; | |
1874 | ||
1875 | local_save_flags(flags); | |
1876 | ||
1877 | /* | |
1878 | * Skip 3 more, seems to get us at the caller of | |
1879 | * this function. | |
1880 | */ | |
1881 | skip += 3; | |
1882 | __ftrace_trace_stack(global_trace.trace_buffer.buffer, | |
1883 | flags, skip, preempt_count(), NULL); | |
1884 | } | |
1885 | ||
1886 | static DEFINE_PER_CPU(int, user_stack_count); | |
1887 | ||
1888 | void | |
1889 | ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc) | |
1890 | { | |
1891 | struct ftrace_event_call *call = &event_user_stack; | |
1892 | struct ring_buffer_event *event; | |
1893 | struct userstack_entry *entry; | |
1894 | struct stack_trace trace; | |
1895 | ||
1896 | if (!(trace_flags & TRACE_ITER_USERSTACKTRACE)) | |
1897 | return; | |
1898 | ||
1899 | /* | |
1900 | * NMIs can not handle page faults, even with fix ups. | |
1901 | * The save user stack can (and often does) fault. | |
1902 | */ | |
1903 | if (unlikely(in_nmi())) | |
1904 | return; | |
1905 | ||
1906 | /* | |
1907 | * prevent recursion, since the user stack tracing may | |
1908 | * trigger other kernel events. | |
1909 | */ | |
1910 | preempt_disable(); | |
1911 | if (__this_cpu_read(user_stack_count)) | |
1912 | goto out; | |
1913 | ||
1914 | __this_cpu_inc(user_stack_count); | |
1915 | ||
1916 | event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK, | |
1917 | sizeof(*entry), flags, pc); | |
1918 | if (!event) | |
1919 | goto out_drop_count; | |
1920 | entry = ring_buffer_event_data(event); | |
1921 | ||
1922 | entry->tgid = current->tgid; | |
1923 | memset(&entry->caller, 0, sizeof(entry->caller)); | |
1924 | ||
1925 | trace.nr_entries = 0; | |
1926 | trace.max_entries = FTRACE_STACK_ENTRIES; | |
1927 | trace.skip = 0; | |
1928 | trace.entries = entry->caller; | |
1929 | ||
1930 | save_stack_trace_user(&trace); | |
1931 | if (!call_filter_check_discard(call, entry, buffer, event)) | |
1932 | __buffer_unlock_commit(buffer, event); | |
1933 | ||
1934 | out_drop_count: | |
1935 | __this_cpu_dec(user_stack_count); | |
1936 | out: | |
1937 | preempt_enable(); | |
1938 | } | |
1939 | ||
1940 | #ifdef UNUSED | |
1941 | static void __trace_userstack(struct trace_array *tr, unsigned long flags) | |
1942 | { | |
1943 | ftrace_trace_userstack(tr, flags, preempt_count()); | |
1944 | } | |
1945 | #endif /* UNUSED */ | |
1946 | ||
1947 | #endif /* CONFIG_STACKTRACE */ | |
1948 | ||
1949 | /* created for use with alloc_percpu */ | |
1950 | struct trace_buffer_struct { | |
1951 | char buffer[TRACE_BUF_SIZE]; | |
1952 | }; | |
1953 | ||
1954 | static struct trace_buffer_struct *trace_percpu_buffer; | |
1955 | static struct trace_buffer_struct *trace_percpu_sirq_buffer; | |
1956 | static struct trace_buffer_struct *trace_percpu_irq_buffer; | |
1957 | static struct trace_buffer_struct *trace_percpu_nmi_buffer; | |
1958 | ||
1959 | /* | |
1960 | * The buffer used is dependent on the context. There is a per cpu | |
1961 | * buffer for normal context, softirq contex, hard irq context and | |
1962 | * for NMI context. Thise allows for lockless recording. | |
1963 | * | |
1964 | * Note, if the buffers failed to be allocated, then this returns NULL | |
1965 | */ | |
1966 | static char *get_trace_buf(void) | |
1967 | { | |
1968 | struct trace_buffer_struct *percpu_buffer; | |
1969 | ||
1970 | /* | |
1971 | * If we have allocated per cpu buffers, then we do not | |
1972 | * need to do any locking. | |
1973 | */ | |
1974 | if (in_nmi()) | |
1975 | percpu_buffer = trace_percpu_nmi_buffer; | |
1976 | else if (in_irq()) | |
1977 | percpu_buffer = trace_percpu_irq_buffer; | |
1978 | else if (in_softirq()) | |
1979 | percpu_buffer = trace_percpu_sirq_buffer; | |
1980 | else | |
1981 | percpu_buffer = trace_percpu_buffer; | |
1982 | ||
1983 | if (!percpu_buffer) | |
1984 | return NULL; | |
1985 | ||
1986 | return this_cpu_ptr(&percpu_buffer->buffer[0]); | |
1987 | } | |
1988 | ||
1989 | static int alloc_percpu_trace_buffer(void) | |
1990 | { | |
1991 | struct trace_buffer_struct *buffers; | |
1992 | struct trace_buffer_struct *sirq_buffers; | |
1993 | struct trace_buffer_struct *irq_buffers; | |
1994 | struct trace_buffer_struct *nmi_buffers; | |
1995 | ||
1996 | buffers = alloc_percpu(struct trace_buffer_struct); | |
1997 | if (!buffers) | |
1998 | goto err_warn; | |
1999 | ||
2000 | sirq_buffers = alloc_percpu(struct trace_buffer_struct); | |
2001 | if (!sirq_buffers) | |
2002 | goto err_sirq; | |
2003 | ||
2004 | irq_buffers = alloc_percpu(struct trace_buffer_struct); | |
2005 | if (!irq_buffers) | |
2006 | goto err_irq; | |
2007 | ||
2008 | nmi_buffers = alloc_percpu(struct trace_buffer_struct); | |
2009 | if (!nmi_buffers) | |
2010 | goto err_nmi; | |
2011 | ||
2012 | trace_percpu_buffer = buffers; | |
2013 | trace_percpu_sirq_buffer = sirq_buffers; | |
2014 | trace_percpu_irq_buffer = irq_buffers; | |
2015 | trace_percpu_nmi_buffer = nmi_buffers; | |
2016 | ||
2017 | return 0; | |
2018 | ||
2019 | err_nmi: | |
2020 | free_percpu(irq_buffers); | |
2021 | err_irq: | |
2022 | free_percpu(sirq_buffers); | |
2023 | err_sirq: | |
2024 | free_percpu(buffers); | |
2025 | err_warn: | |
2026 | WARN(1, "Could not allocate percpu trace_printk buffer"); | |
2027 | return -ENOMEM; | |
2028 | } | |
2029 | ||
2030 | static int buffers_allocated; | |
2031 | ||
2032 | void trace_printk_init_buffers(void) | |
2033 | { | |
2034 | if (buffers_allocated) | |
2035 | return; | |
2036 | ||
2037 | if (alloc_percpu_trace_buffer()) | |
2038 | return; | |
2039 | ||
2040 | /* trace_printk() is for debug use only. Don't use it in production. */ | |
2041 | ||
2042 | pr_warning("\n**********************************************************\n"); | |
2043 | pr_warning("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n"); | |
2044 | pr_warning("** **\n"); | |
2045 | pr_warning("** trace_printk() being used. Allocating extra memory. **\n"); | |
2046 | pr_warning("** **\n"); | |
2047 | pr_warning("** This means that this is a DEBUG kernel and it is **\n"); | |
2048 | pr_warning("** unsafe for produciton use. **\n"); | |
2049 | pr_warning("** **\n"); | |
2050 | pr_warning("** If you see this message and you are not debugging **\n"); | |
2051 | pr_warning("** the kernel, report this immediately to your vendor! **\n"); | |
2052 | pr_warning("** **\n"); | |
2053 | pr_warning("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n"); | |
2054 | pr_warning("**********************************************************\n"); | |
2055 | ||
2056 | /* Expand the buffers to set size */ | |
2057 | tracing_update_buffers(); | |
2058 | ||
2059 | buffers_allocated = 1; | |
2060 | ||
2061 | /* | |
2062 | * trace_printk_init_buffers() can be called by modules. | |
2063 | * If that happens, then we need to start cmdline recording | |
2064 | * directly here. If the global_trace.buffer is already | |
2065 | * allocated here, then this was called by module code. | |
2066 | */ | |
2067 | if (global_trace.trace_buffer.buffer) | |
2068 | tracing_start_cmdline_record(); | |
2069 | } | |
2070 | ||
2071 | void trace_printk_start_comm(void) | |
2072 | { | |
2073 | /* Start tracing comms if trace printk is set */ | |
2074 | if (!buffers_allocated) | |
2075 | return; | |
2076 | tracing_start_cmdline_record(); | |
2077 | } | |
2078 | ||
2079 | static void trace_printk_start_stop_comm(int enabled) | |
2080 | { | |
2081 | if (!buffers_allocated) | |
2082 | return; | |
2083 | ||
2084 | if (enabled) | |
2085 | tracing_start_cmdline_record(); | |
2086 | else | |
2087 | tracing_stop_cmdline_record(); | |
2088 | } | |
2089 | ||
2090 | /** | |
2091 | * trace_vbprintk - write binary msg to tracing buffer | |
2092 | * | |
2093 | */ | |
2094 | int trace_vbprintk(unsigned long ip, const char *fmt, va_list args) | |
2095 | { | |
2096 | struct ftrace_event_call *call = &event_bprint; | |
2097 | struct ring_buffer_event *event; | |
2098 | struct ring_buffer *buffer; | |
2099 | struct trace_array *tr = &global_trace; | |
2100 | struct bprint_entry *entry; | |
2101 | unsigned long flags; | |
2102 | char *tbuffer; | |
2103 | int len = 0, size, pc; | |
2104 | ||
2105 | if (unlikely(tracing_selftest_running || tracing_disabled)) | |
2106 | return 0; | |
2107 | ||
2108 | /* Don't pollute graph traces with trace_vprintk internals */ | |
2109 | pause_graph_tracing(); | |
2110 | ||
2111 | pc = preempt_count(); | |
2112 | preempt_disable_notrace(); | |
2113 | ||
2114 | tbuffer = get_trace_buf(); | |
2115 | if (!tbuffer) { | |
2116 | len = 0; | |
2117 | goto out; | |
2118 | } | |
2119 | ||
2120 | len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args); | |
2121 | ||
2122 | if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0) | |
2123 | goto out; | |
2124 | ||
2125 | local_save_flags(flags); | |
2126 | size = sizeof(*entry) + sizeof(u32) * len; | |
2127 | buffer = tr->trace_buffer.buffer; | |
2128 | event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size, | |
2129 | flags, pc); | |
2130 | if (!event) | |
2131 | goto out; | |
2132 | entry = ring_buffer_event_data(event); | |
2133 | entry->ip = ip; | |
2134 | entry->fmt = fmt; | |
2135 | ||
2136 | memcpy(entry->buf, tbuffer, sizeof(u32) * len); | |
2137 | if (!call_filter_check_discard(call, entry, buffer, event)) { | |
2138 | __buffer_unlock_commit(buffer, event); | |
2139 | ftrace_trace_stack(buffer, flags, 6, pc); | |
2140 | } | |
2141 | ||
2142 | out: | |
2143 | preempt_enable_notrace(); | |
2144 | unpause_graph_tracing(); | |
2145 | ||
2146 | return len; | |
2147 | } | |
2148 | EXPORT_SYMBOL_GPL(trace_vbprintk); | |
2149 | ||
2150 | static int | |
2151 | __trace_array_vprintk(struct ring_buffer *buffer, | |
2152 | unsigned long ip, const char *fmt, va_list args) | |
2153 | { | |
2154 | struct ftrace_event_call *call = &event_print; | |
2155 | struct ring_buffer_event *event; | |
2156 | int len = 0, size, pc; | |
2157 | struct print_entry *entry; | |
2158 | unsigned long flags; | |
2159 | char *tbuffer; | |
2160 | ||
2161 | if (tracing_disabled || tracing_selftest_running) | |
2162 | return 0; | |
2163 | ||
2164 | /* Don't pollute graph traces with trace_vprintk internals */ | |
2165 | pause_graph_tracing(); | |
2166 | ||
2167 | pc = preempt_count(); | |
2168 | preempt_disable_notrace(); | |
2169 | ||
2170 | ||
2171 | tbuffer = get_trace_buf(); | |
2172 | if (!tbuffer) { | |
2173 | len = 0; | |
2174 | goto out; | |
2175 | } | |
2176 | ||
2177 | len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args); | |
2178 | if (len > TRACE_BUF_SIZE) | |
2179 | goto out; | |
2180 | ||
2181 | local_save_flags(flags); | |
2182 | size = sizeof(*entry) + len + 1; | |
2183 | event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size, | |
2184 | flags, pc); | |
2185 | if (!event) | |
2186 | goto out; | |
2187 | entry = ring_buffer_event_data(event); | |
2188 | entry->ip = ip; | |
2189 | ||
2190 | memcpy(&entry->buf, tbuffer, len); | |
2191 | entry->buf[len] = '\0'; | |
2192 | if (!call_filter_check_discard(call, entry, buffer, event)) { | |
2193 | __buffer_unlock_commit(buffer, event); | |
2194 | ftrace_trace_stack(buffer, flags, 6, pc); | |
2195 | } | |
2196 | out: | |
2197 | preempt_enable_notrace(); | |
2198 | unpause_graph_tracing(); | |
2199 | ||
2200 | return len; | |
2201 | } | |
2202 | ||
2203 | int trace_array_vprintk(struct trace_array *tr, | |
2204 | unsigned long ip, const char *fmt, va_list args) | |
2205 | { | |
2206 | return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args); | |
2207 | } | |
2208 | ||
2209 | int trace_array_printk(struct trace_array *tr, | |
2210 | unsigned long ip, const char *fmt, ...) | |
2211 | { | |
2212 | int ret; | |
2213 | va_list ap; | |
2214 | ||
2215 | if (!(trace_flags & TRACE_ITER_PRINTK)) | |
2216 | return 0; | |
2217 | ||
2218 | va_start(ap, fmt); | |
2219 | ret = trace_array_vprintk(tr, ip, fmt, ap); | |
2220 | va_end(ap); | |
2221 | return ret; | |
2222 | } | |
2223 | ||
2224 | int trace_array_printk_buf(struct ring_buffer *buffer, | |
2225 | unsigned long ip, const char *fmt, ...) | |
2226 | { | |
2227 | int ret; | |
2228 | va_list ap; | |
2229 | ||
2230 | if (!(trace_flags & TRACE_ITER_PRINTK)) | |
2231 | return 0; | |
2232 | ||
2233 | va_start(ap, fmt); | |
2234 | ret = __trace_array_vprintk(buffer, ip, fmt, ap); | |
2235 | va_end(ap); | |
2236 | return ret; | |
2237 | } | |
2238 | ||
2239 | int trace_vprintk(unsigned long ip, const char *fmt, va_list args) | |
2240 | { | |
2241 | return trace_array_vprintk(&global_trace, ip, fmt, args); | |
2242 | } | |
2243 | EXPORT_SYMBOL_GPL(trace_vprintk); | |
2244 | ||
2245 | static void trace_iterator_increment(struct trace_iterator *iter) | |
2246 | { | |
2247 | struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu); | |
2248 | ||
2249 | iter->idx++; | |
2250 | if (buf_iter) | |
2251 | ring_buffer_read(buf_iter, NULL); | |
2252 | } | |
2253 | ||
2254 | static struct trace_entry * | |
2255 | peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts, | |
2256 | unsigned long *lost_events) | |
2257 | { | |
2258 | struct ring_buffer_event *event; | |
2259 | struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu); | |
2260 | ||
2261 | if (buf_iter) | |
2262 | event = ring_buffer_iter_peek(buf_iter, ts); | |
2263 | else | |
2264 | event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts, | |
2265 | lost_events); | |
2266 | ||
2267 | if (event) { | |
2268 | iter->ent_size = ring_buffer_event_length(event); | |
2269 | return ring_buffer_event_data(event); | |
2270 | } | |
2271 | iter->ent_size = 0; | |
2272 | return NULL; | |
2273 | } | |
2274 | ||
2275 | static struct trace_entry * | |
2276 | __find_next_entry(struct trace_iterator *iter, int *ent_cpu, | |
2277 | unsigned long *missing_events, u64 *ent_ts) | |
2278 | { | |
2279 | struct ring_buffer *buffer = iter->trace_buffer->buffer; | |
2280 | struct trace_entry *ent, *next = NULL; | |
2281 | unsigned long lost_events = 0, next_lost = 0; | |
2282 | int cpu_file = iter->cpu_file; | |
2283 | u64 next_ts = 0, ts; | |
2284 | int next_cpu = -1; | |
2285 | int next_size = 0; | |
2286 | int cpu; | |
2287 | ||
2288 | /* | |
2289 | * If we are in a per_cpu trace file, don't bother by iterating over | |
2290 | * all cpu and peek directly. | |
2291 | */ | |
2292 | if (cpu_file > RING_BUFFER_ALL_CPUS) { | |
2293 | if (ring_buffer_empty_cpu(buffer, cpu_file)) | |
2294 | return NULL; | |
2295 | ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events); | |
2296 | if (ent_cpu) | |
2297 | *ent_cpu = cpu_file; | |
2298 | ||
2299 | return ent; | |
2300 | } | |
2301 | ||
2302 | for_each_tracing_cpu(cpu) { | |
2303 | ||
2304 | if (ring_buffer_empty_cpu(buffer, cpu)) | |
2305 | continue; | |
2306 | ||
2307 | ent = peek_next_entry(iter, cpu, &ts, &lost_events); | |
2308 | ||
2309 | /* | |
2310 | * Pick the entry with the smallest timestamp: | |
2311 | */ | |
2312 | if (ent && (!next || ts < next_ts)) { | |
2313 | next = ent; | |
2314 | next_cpu = cpu; | |
2315 | next_ts = ts; | |
2316 | next_lost = lost_events; | |
2317 | next_size = iter->ent_size; | |
2318 | } | |
2319 | } | |
2320 | ||
2321 | iter->ent_size = next_size; | |
2322 | ||
2323 | if (ent_cpu) | |
2324 | *ent_cpu = next_cpu; | |
2325 | ||
2326 | if (ent_ts) | |
2327 | *ent_ts = next_ts; | |
2328 | ||
2329 | if (missing_events) | |
2330 | *missing_events = next_lost; | |
2331 | ||
2332 | return next; | |
2333 | } | |
2334 | ||
2335 | /* Find the next real entry, without updating the iterator itself */ | |
2336 | struct trace_entry *trace_find_next_entry(struct trace_iterator *iter, | |
2337 | int *ent_cpu, u64 *ent_ts) | |
2338 | { | |
2339 | return __find_next_entry(iter, ent_cpu, NULL, ent_ts); | |
2340 | } | |
2341 | ||
2342 | /* Find the next real entry, and increment the iterator to the next entry */ | |
2343 | void *trace_find_next_entry_inc(struct trace_iterator *iter) | |
2344 | { | |
2345 | iter->ent = __find_next_entry(iter, &iter->cpu, | |
2346 | &iter->lost_events, &iter->ts); | |
2347 | ||
2348 | if (iter->ent) | |
2349 | trace_iterator_increment(iter); | |
2350 | ||
2351 | return iter->ent ? iter : NULL; | |
2352 | } | |
2353 | ||
2354 | static void trace_consume(struct trace_iterator *iter) | |
2355 | { | |
2356 | ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts, | |
2357 | &iter->lost_events); | |
2358 | } | |
2359 | ||
2360 | static void *s_next(struct seq_file *m, void *v, loff_t *pos) | |
2361 | { | |
2362 | struct trace_iterator *iter = m->private; | |
2363 | int i = (int)*pos; | |
2364 | void *ent; | |
2365 | ||
2366 | WARN_ON_ONCE(iter->leftover); | |
2367 | ||
2368 | (*pos)++; | |
2369 | ||
2370 | /* can't go backwards */ | |
2371 | if (iter->idx > i) | |
2372 | return NULL; | |
2373 | ||
2374 | if (iter->idx < 0) | |
2375 | ent = trace_find_next_entry_inc(iter); | |
2376 | else | |
2377 | ent = iter; | |
2378 | ||
2379 | while (ent && iter->idx < i) | |
2380 | ent = trace_find_next_entry_inc(iter); | |
2381 | ||
2382 | iter->pos = *pos; | |
2383 | ||
2384 | return ent; | |
2385 | } | |
2386 | ||
2387 | void tracing_iter_reset(struct trace_iterator *iter, int cpu) | |
2388 | { | |
2389 | struct ring_buffer_event *event; | |
2390 | struct ring_buffer_iter *buf_iter; | |
2391 | unsigned long entries = 0; | |
2392 | u64 ts; | |
2393 | ||
2394 | per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0; | |
2395 | ||
2396 | buf_iter = trace_buffer_iter(iter, cpu); | |
2397 | if (!buf_iter) | |
2398 | return; | |
2399 | ||
2400 | ring_buffer_iter_reset(buf_iter); | |
2401 | ||
2402 | /* | |
2403 | * We could have the case with the max latency tracers | |
2404 | * that a reset never took place on a cpu. This is evident | |
2405 | * by the timestamp being before the start of the buffer. | |
2406 | */ | |
2407 | while ((event = ring_buffer_iter_peek(buf_iter, &ts))) { | |
2408 | if (ts >= iter->trace_buffer->time_start) | |
2409 | break; | |
2410 | entries++; | |
2411 | ring_buffer_read(buf_iter, NULL); | |
2412 | } | |
2413 | ||
2414 | per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries; | |
2415 | } | |
2416 | ||
2417 | /* | |
2418 | * The current tracer is copied to avoid a global locking | |
2419 | * all around. | |
2420 | */ | |
2421 | static void *s_start(struct seq_file *m, loff_t *pos) | |
2422 | { | |
2423 | struct trace_iterator *iter = m->private; | |
2424 | struct trace_array *tr = iter->tr; | |
2425 | int cpu_file = iter->cpu_file; | |
2426 | void *p = NULL; | |
2427 | loff_t l = 0; | |
2428 | int cpu; | |
2429 | ||
2430 | /* | |
2431 | * copy the tracer to avoid using a global lock all around. | |
2432 | * iter->trace is a copy of current_trace, the pointer to the | |
2433 | * name may be used instead of a strcmp(), as iter->trace->name | |
2434 | * will point to the same string as current_trace->name. | |
2435 | */ | |
2436 | mutex_lock(&trace_types_lock); | |
2437 | if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name)) | |
2438 | *iter->trace = *tr->current_trace; | |
2439 | mutex_unlock(&trace_types_lock); | |
2440 | ||
2441 | #ifdef CONFIG_TRACER_MAX_TRACE | |
2442 | if (iter->snapshot && iter->trace->use_max_tr) | |
2443 | return ERR_PTR(-EBUSY); | |
2444 | #endif | |
2445 | ||
2446 | if (!iter->snapshot) | |
2447 | atomic_inc(&trace_record_cmdline_disabled); | |
2448 | ||
2449 | if (*pos != iter->pos) { | |
2450 | iter->ent = NULL; | |
2451 | iter->cpu = 0; | |
2452 | iter->idx = -1; | |
2453 | ||
2454 | if (cpu_file == RING_BUFFER_ALL_CPUS) { | |
2455 | for_each_tracing_cpu(cpu) | |
2456 | tracing_iter_reset(iter, cpu); | |
2457 | } else | |
2458 | tracing_iter_reset(iter, cpu_file); | |
2459 | ||
2460 | iter->leftover = 0; | |
2461 | for (p = iter; p && l < *pos; p = s_next(m, p, &l)) | |
2462 | ; | |
2463 | ||
2464 | } else { | |
2465 | /* | |
2466 | * If we overflowed the seq_file before, then we want | |
2467 | * to just reuse the trace_seq buffer again. | |
2468 | */ | |
2469 | if (iter->leftover) | |
2470 | p = iter; | |
2471 | else { | |
2472 | l = *pos - 1; | |
2473 | p = s_next(m, p, &l); | |
2474 | } | |
2475 | } | |
2476 | ||
2477 | trace_event_read_lock(); | |
2478 | trace_access_lock(cpu_file); | |
2479 | return p; | |
2480 | } | |
2481 | ||
2482 | static void s_stop(struct seq_file *m, void *p) | |
2483 | { | |
2484 | struct trace_iterator *iter = m->private; | |
2485 | ||
2486 | #ifdef CONFIG_TRACER_MAX_TRACE | |
2487 | if (iter->snapshot && iter->trace->use_max_tr) | |
2488 | return; | |
2489 | #endif | |
2490 | ||
2491 | if (!iter->snapshot) | |
2492 | atomic_dec(&trace_record_cmdline_disabled); | |
2493 | ||
2494 | trace_access_unlock(iter->cpu_file); | |
2495 | trace_event_read_unlock(); | |
2496 | } | |
2497 | ||
2498 | static void | |
2499 | get_total_entries(struct trace_buffer *buf, | |
2500 | unsigned long *total, unsigned long *entries) | |
2501 | { | |
2502 | unsigned long count; | |
2503 | int cpu; | |
2504 | ||
2505 | *total = 0; | |
2506 | *entries = 0; | |
2507 | ||
2508 | for_each_tracing_cpu(cpu) { | |
2509 | count = ring_buffer_entries_cpu(buf->buffer, cpu); | |
2510 | /* | |
2511 | * If this buffer has skipped entries, then we hold all | |
2512 | * entries for the trace and we need to ignore the | |
2513 | * ones before the time stamp. | |
2514 | */ | |
2515 | if (per_cpu_ptr(buf->data, cpu)->skipped_entries) { | |
2516 | count -= per_cpu_ptr(buf->data, cpu)->skipped_entries; | |
2517 | /* total is the same as the entries */ | |
2518 | *total += count; | |
2519 | } else | |
2520 | *total += count + | |
2521 | ring_buffer_overrun_cpu(buf->buffer, cpu); | |
2522 | *entries += count; | |
2523 | } | |
2524 | } | |
2525 | ||
2526 | static void print_lat_help_header(struct seq_file *m) | |
2527 | { | |
2528 | seq_puts(m, "# _------=> CPU# \n"); | |
2529 | seq_puts(m, "# / _-----=> irqs-off \n"); | |
2530 | seq_puts(m, "# | / _----=> need-resched \n"); | |
2531 | seq_puts(m, "# || / _---=> hardirq/softirq \n"); | |
2532 | seq_puts(m, "# ||| / _--=> preempt-depth \n"); | |
2533 | seq_puts(m, "# |||| / delay \n"); | |
2534 | seq_puts(m, "# cmd pid ||||| time | caller \n"); | |
2535 | seq_puts(m, "# \\ / ||||| \\ | / \n"); | |
2536 | } | |
2537 | ||
2538 | static void print_event_info(struct trace_buffer *buf, struct seq_file *m) | |
2539 | { | |
2540 | unsigned long total; | |
2541 | unsigned long entries; | |
2542 | ||
2543 | get_total_entries(buf, &total, &entries); | |
2544 | seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n", | |
2545 | entries, total, num_online_cpus()); | |
2546 | seq_puts(m, "#\n"); | |
2547 | } | |
2548 | ||
2549 | static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m) | |
2550 | { | |
2551 | print_event_info(buf, m); | |
2552 | seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n"); | |
2553 | seq_puts(m, "# | | | | |\n"); | |
2554 | } | |
2555 | ||
2556 | static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m) | |
2557 | { | |
2558 | print_event_info(buf, m); | |
2559 | seq_puts(m, "# _-----=> irqs-off\n"); | |
2560 | seq_puts(m, "# / _----=> need-resched\n"); | |
2561 | seq_puts(m, "# | / _---=> hardirq/softirq\n"); | |
2562 | seq_puts(m, "# || / _--=> preempt-depth\n"); | |
2563 | seq_puts(m, "# ||| / delay\n"); | |
2564 | seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n"); | |
2565 | seq_puts(m, "# | | | |||| | |\n"); | |
2566 | } | |
2567 | ||
2568 | void | |
2569 | print_trace_header(struct seq_file *m, struct trace_iterator *iter) | |
2570 | { | |
2571 | unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK); | |
2572 | struct trace_buffer *buf = iter->trace_buffer; | |
2573 | struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu); | |
2574 | struct tracer *type = iter->trace; | |
2575 | unsigned long entries; | |
2576 | unsigned long total; | |
2577 | const char *name = "preemption"; | |
2578 | ||
2579 | name = type->name; | |
2580 | ||
2581 | get_total_entries(buf, &total, &entries); | |
2582 | ||
2583 | seq_printf(m, "# %s latency trace v1.1.5 on %s\n", | |
2584 | name, UTS_RELEASE); | |
2585 | seq_puts(m, "# -----------------------------------" | |
2586 | "---------------------------------\n"); | |
2587 | seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |" | |
2588 | " (M:%s VP:%d, KP:%d, SP:%d HP:%d", | |
2589 | nsecs_to_usecs(data->saved_latency), | |
2590 | entries, | |
2591 | total, | |
2592 | buf->cpu, | |
2593 | #if defined(CONFIG_PREEMPT_NONE) | |
2594 | "server", | |
2595 | #elif defined(CONFIG_PREEMPT_VOLUNTARY) | |
2596 | "desktop", | |
2597 | #elif defined(CONFIG_PREEMPT) | |
2598 | "preempt", | |
2599 | #else | |
2600 | "unknown", | |
2601 | #endif | |
2602 | /* These are reserved for later use */ | |
2603 | 0, 0, 0, 0); | |
2604 | #ifdef CONFIG_SMP | |
2605 | seq_printf(m, " #P:%d)\n", num_online_cpus()); | |
2606 | #else | |
2607 | seq_puts(m, ")\n"); | |
2608 | #endif | |
2609 | seq_puts(m, "# -----------------\n"); | |
2610 | seq_printf(m, "# | task: %.16s-%d " | |
2611 | "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n", | |
2612 | data->comm, data->pid, | |
2613 | from_kuid_munged(seq_user_ns(m), data->uid), data->nice, | |
2614 | data->policy, data->rt_priority); | |
2615 | seq_puts(m, "# -----------------\n"); | |
2616 | ||
2617 | if (data->critical_start) { | |
2618 | seq_puts(m, "# => started at: "); | |
2619 | seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags); | |
2620 | trace_print_seq(m, &iter->seq); | |
2621 | seq_puts(m, "\n# => ended at: "); | |
2622 | seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags); | |
2623 | trace_print_seq(m, &iter->seq); | |
2624 | seq_puts(m, "\n#\n"); | |
2625 | } | |
2626 | ||
2627 | seq_puts(m, "#\n"); | |
2628 | } | |
2629 | ||
2630 | static void test_cpu_buff_start(struct trace_iterator *iter) | |
2631 | { | |
2632 | struct trace_seq *s = &iter->seq; | |
2633 | ||
2634 | if (!(trace_flags & TRACE_ITER_ANNOTATE)) | |
2635 | return; | |
2636 | ||
2637 | if (!(iter->iter_flags & TRACE_FILE_ANNOTATE)) | |
2638 | return; | |
2639 | ||
2640 | if (cpumask_test_cpu(iter->cpu, iter->started)) | |
2641 | return; | |
2642 | ||
2643 | if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries) | |
2644 | return; | |
2645 | ||
2646 | cpumask_set_cpu(iter->cpu, iter->started); | |
2647 | ||
2648 | /* Don't print started cpu buffer for the first entry of the trace */ | |
2649 | if (iter->idx > 1) | |
2650 | trace_seq_printf(s, "##### CPU %u buffer started ####\n", | |
2651 | iter->cpu); | |
2652 | } | |
2653 | ||
2654 | static enum print_line_t print_trace_fmt(struct trace_iterator *iter) | |
2655 | { | |
2656 | struct trace_seq *s = &iter->seq; | |
2657 | unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK); | |
2658 | struct trace_entry *entry; | |
2659 | struct trace_event *event; | |
2660 | ||
2661 | entry = iter->ent; | |
2662 | ||
2663 | test_cpu_buff_start(iter); | |
2664 | ||
2665 | event = ftrace_find_event(entry->type); | |
2666 | ||
2667 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
2668 | if (iter->iter_flags & TRACE_FILE_LAT_FMT) { | |
2669 | if (!trace_print_lat_context(iter)) | |
2670 | goto partial; | |
2671 | } else { | |
2672 | if (!trace_print_context(iter)) | |
2673 | goto partial; | |
2674 | } | |
2675 | } | |
2676 | ||
2677 | if (event) | |
2678 | return event->funcs->trace(iter, sym_flags, event); | |
2679 | ||
2680 | if (!trace_seq_printf(s, "Unknown type %d\n", entry->type)) | |
2681 | goto partial; | |
2682 | ||
2683 | return TRACE_TYPE_HANDLED; | |
2684 | partial: | |
2685 | return TRACE_TYPE_PARTIAL_LINE; | |
2686 | } | |
2687 | ||
2688 | static enum print_line_t print_raw_fmt(struct trace_iterator *iter) | |
2689 | { | |
2690 | struct trace_seq *s = &iter->seq; | |
2691 | struct trace_entry *entry; | |
2692 | struct trace_event *event; | |
2693 | ||
2694 | entry = iter->ent; | |
2695 | ||
2696 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
2697 | if (!trace_seq_printf(s, "%d %d %llu ", | |
2698 | entry->pid, iter->cpu, iter->ts)) | |
2699 | goto partial; | |
2700 | } | |
2701 | ||
2702 | event = ftrace_find_event(entry->type); | |
2703 | if (event) | |
2704 | return event->funcs->raw(iter, 0, event); | |
2705 | ||
2706 | if (!trace_seq_printf(s, "%d ?\n", entry->type)) | |
2707 | goto partial; | |
2708 | ||
2709 | return TRACE_TYPE_HANDLED; | |
2710 | partial: | |
2711 | return TRACE_TYPE_PARTIAL_LINE; | |
2712 | } | |
2713 | ||
2714 | static enum print_line_t print_hex_fmt(struct trace_iterator *iter) | |
2715 | { | |
2716 | struct trace_seq *s = &iter->seq; | |
2717 | unsigned char newline = '\n'; | |
2718 | struct trace_entry *entry; | |
2719 | struct trace_event *event; | |
2720 | ||
2721 | entry = iter->ent; | |
2722 | ||
2723 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
2724 | SEQ_PUT_HEX_FIELD_RET(s, entry->pid); | |
2725 | SEQ_PUT_HEX_FIELD_RET(s, iter->cpu); | |
2726 | SEQ_PUT_HEX_FIELD_RET(s, iter->ts); | |
2727 | } | |
2728 | ||
2729 | event = ftrace_find_event(entry->type); | |
2730 | if (event) { | |
2731 | enum print_line_t ret = event->funcs->hex(iter, 0, event); | |
2732 | if (ret != TRACE_TYPE_HANDLED) | |
2733 | return ret; | |
2734 | } | |
2735 | ||
2736 | SEQ_PUT_FIELD_RET(s, newline); | |
2737 | ||
2738 | return TRACE_TYPE_HANDLED; | |
2739 | } | |
2740 | ||
2741 | static enum print_line_t print_bin_fmt(struct trace_iterator *iter) | |
2742 | { | |
2743 | struct trace_seq *s = &iter->seq; | |
2744 | struct trace_entry *entry; | |
2745 | struct trace_event *event; | |
2746 | ||
2747 | entry = iter->ent; | |
2748 | ||
2749 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
2750 | SEQ_PUT_FIELD_RET(s, entry->pid); | |
2751 | SEQ_PUT_FIELD_RET(s, iter->cpu); | |
2752 | SEQ_PUT_FIELD_RET(s, iter->ts); | |
2753 | } | |
2754 | ||
2755 | event = ftrace_find_event(entry->type); | |
2756 | return event ? event->funcs->binary(iter, 0, event) : | |
2757 | TRACE_TYPE_HANDLED; | |
2758 | } | |
2759 | ||
2760 | int trace_empty(struct trace_iterator *iter) | |
2761 | { | |
2762 | struct ring_buffer_iter *buf_iter; | |
2763 | int cpu; | |
2764 | ||
2765 | /* If we are looking at one CPU buffer, only check that one */ | |
2766 | if (iter->cpu_file != RING_BUFFER_ALL_CPUS) { | |
2767 | cpu = iter->cpu_file; | |
2768 | buf_iter = trace_buffer_iter(iter, cpu); | |
2769 | if (buf_iter) { | |
2770 | if (!ring_buffer_iter_empty(buf_iter)) | |
2771 | return 0; | |
2772 | } else { | |
2773 | if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu)) | |
2774 | return 0; | |
2775 | } | |
2776 | return 1; | |
2777 | } | |
2778 | ||
2779 | for_each_tracing_cpu(cpu) { | |
2780 | buf_iter = trace_buffer_iter(iter, cpu); | |
2781 | if (buf_iter) { | |
2782 | if (!ring_buffer_iter_empty(buf_iter)) | |
2783 | return 0; | |
2784 | } else { | |
2785 | if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu)) | |
2786 | return 0; | |
2787 | } | |
2788 | } | |
2789 | ||
2790 | return 1; | |
2791 | } | |
2792 | ||
2793 | /* Called with trace_event_read_lock() held. */ | |
2794 | enum print_line_t print_trace_line(struct trace_iterator *iter) | |
2795 | { | |
2796 | enum print_line_t ret; | |
2797 | ||
2798 | if (iter->lost_events && | |
2799 | !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n", | |
2800 | iter->cpu, iter->lost_events)) | |
2801 | return TRACE_TYPE_PARTIAL_LINE; | |
2802 | ||
2803 | if (iter->trace && iter->trace->print_line) { | |
2804 | ret = iter->trace->print_line(iter); | |
2805 | if (ret != TRACE_TYPE_UNHANDLED) | |
2806 | return ret; | |
2807 | } | |
2808 | ||
2809 | if (iter->ent->type == TRACE_BPUTS && | |
2810 | trace_flags & TRACE_ITER_PRINTK && | |
2811 | trace_flags & TRACE_ITER_PRINTK_MSGONLY) | |
2812 | return trace_print_bputs_msg_only(iter); | |
2813 | ||
2814 | if (iter->ent->type == TRACE_BPRINT && | |
2815 | trace_flags & TRACE_ITER_PRINTK && | |
2816 | trace_flags & TRACE_ITER_PRINTK_MSGONLY) | |
2817 | return trace_print_bprintk_msg_only(iter); | |
2818 | ||
2819 | if (iter->ent->type == TRACE_PRINT && | |
2820 | trace_flags & TRACE_ITER_PRINTK && | |
2821 | trace_flags & TRACE_ITER_PRINTK_MSGONLY) | |
2822 | return trace_print_printk_msg_only(iter); | |
2823 | ||
2824 | if (trace_flags & TRACE_ITER_BIN) | |
2825 | return print_bin_fmt(iter); | |
2826 | ||
2827 | if (trace_flags & TRACE_ITER_HEX) | |
2828 | return print_hex_fmt(iter); | |
2829 | ||
2830 | if (trace_flags & TRACE_ITER_RAW) | |
2831 | return print_raw_fmt(iter); | |
2832 | ||
2833 | return print_trace_fmt(iter); | |
2834 | } | |
2835 | ||
2836 | void trace_latency_header(struct seq_file *m) | |
2837 | { | |
2838 | struct trace_iterator *iter = m->private; | |
2839 | ||
2840 | /* print nothing if the buffers are empty */ | |
2841 | if (trace_empty(iter)) | |
2842 | return; | |
2843 | ||
2844 | if (iter->iter_flags & TRACE_FILE_LAT_FMT) | |
2845 | print_trace_header(m, iter); | |
2846 | ||
2847 | if (!(trace_flags & TRACE_ITER_VERBOSE)) | |
2848 | print_lat_help_header(m); | |
2849 | } | |
2850 | ||
2851 | void trace_default_header(struct seq_file *m) | |
2852 | { | |
2853 | struct trace_iterator *iter = m->private; | |
2854 | ||
2855 | if (!(trace_flags & TRACE_ITER_CONTEXT_INFO)) | |
2856 | return; | |
2857 | ||
2858 | if (iter->iter_flags & TRACE_FILE_LAT_FMT) { | |
2859 | /* print nothing if the buffers are empty */ | |
2860 | if (trace_empty(iter)) | |
2861 | return; | |
2862 | print_trace_header(m, iter); | |
2863 | if (!(trace_flags & TRACE_ITER_VERBOSE)) | |
2864 | print_lat_help_header(m); | |
2865 | } else { | |
2866 | if (!(trace_flags & TRACE_ITER_VERBOSE)) { | |
2867 | if (trace_flags & TRACE_ITER_IRQ_INFO) | |
2868 | print_func_help_header_irq(iter->trace_buffer, m); | |
2869 | else | |
2870 | print_func_help_header(iter->trace_buffer, m); | |
2871 | } | |
2872 | } | |
2873 | } | |
2874 | ||
2875 | static void test_ftrace_alive(struct seq_file *m) | |
2876 | { | |
2877 | if (!ftrace_is_dead()) | |
2878 | return; | |
2879 | seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"); | |
2880 | seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n"); | |
2881 | } | |
2882 | ||
2883 | #ifdef CONFIG_TRACER_MAX_TRACE | |
2884 | static void show_snapshot_main_help(struct seq_file *m) | |
2885 | { | |
2886 | seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"); | |
2887 | seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"); | |
2888 | seq_printf(m, "# Takes a snapshot of the main buffer.\n"); | |
2889 | seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"); | |
2890 | seq_printf(m, "# (Doesn't have to be '2' works with any number that\n"); | |
2891 | seq_printf(m, "# is not a '0' or '1')\n"); | |
2892 | } | |
2893 | ||
2894 | static void show_snapshot_percpu_help(struct seq_file *m) | |
2895 | { | |
2896 | seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n"); | |
2897 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP | |
2898 | seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"); | |
2899 | seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n"); | |
2900 | #else | |
2901 | seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n"); | |
2902 | seq_printf(m, "# Must use main snapshot file to allocate.\n"); | |
2903 | #endif | |
2904 | seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"); | |
2905 | seq_printf(m, "# (Doesn't have to be '2' works with any number that\n"); | |
2906 | seq_printf(m, "# is not a '0' or '1')\n"); | |
2907 | } | |
2908 | ||
2909 | static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) | |
2910 | { | |
2911 | if (iter->tr->allocated_snapshot) | |
2912 | seq_printf(m, "#\n# * Snapshot is allocated *\n#\n"); | |
2913 | else | |
2914 | seq_printf(m, "#\n# * Snapshot is freed *\n#\n"); | |
2915 | ||
2916 | seq_printf(m, "# Snapshot commands:\n"); | |
2917 | if (iter->cpu_file == RING_BUFFER_ALL_CPUS) | |
2918 | show_snapshot_main_help(m); | |
2919 | else | |
2920 | show_snapshot_percpu_help(m); | |
2921 | } | |
2922 | #else | |
2923 | /* Should never be called */ | |
2924 | static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { } | |
2925 | #endif | |
2926 | ||
2927 | static int s_show(struct seq_file *m, void *v) | |
2928 | { | |
2929 | struct trace_iterator *iter = v; | |
2930 | int ret; | |
2931 | ||
2932 | if (iter->ent == NULL) { | |
2933 | if (iter->tr) { | |
2934 | seq_printf(m, "# tracer: %s\n", iter->trace->name); | |
2935 | seq_puts(m, "#\n"); | |
2936 | test_ftrace_alive(m); | |
2937 | } | |
2938 | if (iter->snapshot && trace_empty(iter)) | |
2939 | print_snapshot_help(m, iter); | |
2940 | else if (iter->trace && iter->trace->print_header) | |
2941 | iter->trace->print_header(m); | |
2942 | else | |
2943 | trace_default_header(m); | |
2944 | ||
2945 | } else if (iter->leftover) { | |
2946 | /* | |
2947 | * If we filled the seq_file buffer earlier, we | |
2948 | * want to just show it now. | |
2949 | */ | |
2950 | ret = trace_print_seq(m, &iter->seq); | |
2951 | ||
2952 | /* ret should this time be zero, but you never know */ | |
2953 | iter->leftover = ret; | |
2954 | ||
2955 | } else { | |
2956 | print_trace_line(iter); | |
2957 | ret = trace_print_seq(m, &iter->seq); | |
2958 | /* | |
2959 | * If we overflow the seq_file buffer, then it will | |
2960 | * ask us for this data again at start up. | |
2961 | * Use that instead. | |
2962 | * ret is 0 if seq_file write succeeded. | |
2963 | * -1 otherwise. | |
2964 | */ | |
2965 | iter->leftover = ret; | |
2966 | } | |
2967 | ||
2968 | return 0; | |
2969 | } | |
2970 | ||
2971 | /* | |
2972 | * Should be used after trace_array_get(), trace_types_lock | |
2973 | * ensures that i_cdev was already initialized. | |
2974 | */ | |
2975 | static inline int tracing_get_cpu(struct inode *inode) | |
2976 | { | |
2977 | if (inode->i_cdev) /* See trace_create_cpu_file() */ | |
2978 | return (long)inode->i_cdev - 1; | |
2979 | return RING_BUFFER_ALL_CPUS; | |
2980 | } | |
2981 | ||
2982 | static const struct seq_operations tracer_seq_ops = { | |
2983 | .start = s_start, | |
2984 | .next = s_next, | |
2985 | .stop = s_stop, | |
2986 | .show = s_show, | |
2987 | }; | |
2988 | ||
2989 | static struct trace_iterator * | |
2990 | __tracing_open(struct inode *inode, struct file *file, bool snapshot) | |
2991 | { | |
2992 | struct trace_array *tr = inode->i_private; | |
2993 | struct trace_iterator *iter; | |
2994 | int cpu; | |
2995 | ||
2996 | if (tracing_disabled) | |
2997 | return ERR_PTR(-ENODEV); | |
2998 | ||
2999 | iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter)); | |
3000 | if (!iter) | |
3001 | return ERR_PTR(-ENOMEM); | |
3002 | ||
3003 | iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(), | |
3004 | GFP_KERNEL); | |
3005 | if (!iter->buffer_iter) | |
3006 | goto release; | |
3007 | ||
3008 | /* | |
3009 | * We make a copy of the current tracer to avoid concurrent | |
3010 | * changes on it while we are reading. | |
3011 | */ | |
3012 | mutex_lock(&trace_types_lock); | |
3013 | iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL); | |
3014 | if (!iter->trace) | |
3015 | goto fail; | |
3016 | ||
3017 | *iter->trace = *tr->current_trace; | |
3018 | ||
3019 | if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL)) | |
3020 | goto fail; | |
3021 | ||
3022 | iter->tr = tr; | |
3023 | ||
3024 | #ifdef CONFIG_TRACER_MAX_TRACE | |
3025 | /* Currently only the top directory has a snapshot */ | |
3026 | if (tr->current_trace->print_max || snapshot) | |
3027 | iter->trace_buffer = &tr->max_buffer; | |
3028 | else | |
3029 | #endif | |
3030 | iter->trace_buffer = &tr->trace_buffer; | |
3031 | iter->snapshot = snapshot; | |
3032 | iter->pos = -1; | |
3033 | iter->cpu_file = tracing_get_cpu(inode); | |
3034 | mutex_init(&iter->mutex); | |
3035 | ||
3036 | /* Notify the tracer early; before we stop tracing. */ | |
3037 | if (iter->trace && iter->trace->open) | |
3038 | iter->trace->open(iter); | |
3039 | ||
3040 | /* Annotate start of buffers if we had overruns */ | |
3041 | if (ring_buffer_overruns(iter->trace_buffer->buffer)) | |
3042 | iter->iter_flags |= TRACE_FILE_ANNOTATE; | |
3043 | ||
3044 | /* Output in nanoseconds only if we are using a clock in nanoseconds. */ | |
3045 | if (trace_clocks[tr->clock_id].in_ns) | |
3046 | iter->iter_flags |= TRACE_FILE_TIME_IN_NS; | |
3047 | ||
3048 | /* stop the trace while dumping if we are not opening "snapshot" */ | |
3049 | if (!iter->snapshot) | |
3050 | tracing_stop_tr(tr); | |
3051 | ||
3052 | if (iter->cpu_file == RING_BUFFER_ALL_CPUS) { | |
3053 | for_each_tracing_cpu(cpu) { | |
3054 | iter->buffer_iter[cpu] = | |
3055 | ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu); | |
3056 | } | |
3057 | ring_buffer_read_prepare_sync(); | |
3058 | for_each_tracing_cpu(cpu) { | |
3059 | ring_buffer_read_start(iter->buffer_iter[cpu]); | |
3060 | tracing_iter_reset(iter, cpu); | |
3061 | } | |
3062 | } else { | |
3063 | cpu = iter->cpu_file; | |
3064 | iter->buffer_iter[cpu] = | |
3065 | ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu); | |
3066 | ring_buffer_read_prepare_sync(); | |
3067 | ring_buffer_read_start(iter->buffer_iter[cpu]); | |
3068 | tracing_iter_reset(iter, cpu); | |
3069 | } | |
3070 | ||
3071 | mutex_unlock(&trace_types_lock); | |
3072 | ||
3073 | return iter; | |
3074 | ||
3075 | fail: | |
3076 | mutex_unlock(&trace_types_lock); | |
3077 | kfree(iter->trace); | |
3078 | kfree(iter->buffer_iter); | |
3079 | release: | |
3080 | seq_release_private(inode, file); | |
3081 | return ERR_PTR(-ENOMEM); | |
3082 | } | |
3083 | ||
3084 | int tracing_open_generic(struct inode *inode, struct file *filp) | |
3085 | { | |
3086 | if (tracing_disabled) | |
3087 | return -ENODEV; | |
3088 | ||
3089 | filp->private_data = inode->i_private; | |
3090 | return 0; | |
3091 | } | |
3092 | ||
3093 | bool tracing_is_disabled(void) | |
3094 | { | |
3095 | return (tracing_disabled) ? true: false; | |
3096 | } | |
3097 | ||
3098 | /* | |
3099 | * Open and update trace_array ref count. | |
3100 | * Must have the current trace_array passed to it. | |
3101 | */ | |
3102 | static int tracing_open_generic_tr(struct inode *inode, struct file *filp) | |
3103 | { | |
3104 | struct trace_array *tr = inode->i_private; | |
3105 | ||
3106 | if (tracing_disabled) | |
3107 | return -ENODEV; | |
3108 | ||
3109 | if (trace_array_get(tr) < 0) | |
3110 | return -ENODEV; | |
3111 | ||
3112 | filp->private_data = inode->i_private; | |
3113 | ||
3114 | return 0; | |
3115 | } | |
3116 | ||
3117 | static int tracing_release(struct inode *inode, struct file *file) | |
3118 | { | |
3119 | struct trace_array *tr = inode->i_private; | |
3120 | struct seq_file *m = file->private_data; | |
3121 | struct trace_iterator *iter; | |
3122 | int cpu; | |
3123 | ||
3124 | if (!(file->f_mode & FMODE_READ)) { | |
3125 | trace_array_put(tr); | |
3126 | return 0; | |
3127 | } | |
3128 | ||
3129 | /* Writes do not use seq_file */ | |
3130 | iter = m->private; | |
3131 | mutex_lock(&trace_types_lock); | |
3132 | ||
3133 | for_each_tracing_cpu(cpu) { | |
3134 | if (iter->buffer_iter[cpu]) | |
3135 | ring_buffer_read_finish(iter->buffer_iter[cpu]); | |
3136 | } | |
3137 | ||
3138 | if (iter->trace && iter->trace->close) | |
3139 | iter->trace->close(iter); | |
3140 | ||
3141 | if (!iter->snapshot) | |
3142 | /* reenable tracing if it was previously enabled */ | |
3143 | tracing_start_tr(tr); | |
3144 | ||
3145 | __trace_array_put(tr); | |
3146 | ||
3147 | mutex_unlock(&trace_types_lock); | |
3148 | ||
3149 | mutex_destroy(&iter->mutex); | |
3150 | free_cpumask_var(iter->started); | |
3151 | kfree(iter->trace); | |
3152 | kfree(iter->buffer_iter); | |
3153 | seq_release_private(inode, file); | |
3154 | ||
3155 | return 0; | |
3156 | } | |
3157 | ||
3158 | static int tracing_release_generic_tr(struct inode *inode, struct file *file) | |
3159 | { | |
3160 | struct trace_array *tr = inode->i_private; | |
3161 | ||
3162 | trace_array_put(tr); | |
3163 | return 0; | |
3164 | } | |
3165 | ||
3166 | static int tracing_single_release_tr(struct inode *inode, struct file *file) | |
3167 | { | |
3168 | struct trace_array *tr = inode->i_private; | |
3169 | ||
3170 | trace_array_put(tr); | |
3171 | ||
3172 | return single_release(inode, file); | |
3173 | } | |
3174 | ||
3175 | static int tracing_open(struct inode *inode, struct file *file) | |
3176 | { | |
3177 | struct trace_array *tr = inode->i_private; | |
3178 | struct trace_iterator *iter; | |
3179 | int ret = 0; | |
3180 | ||
3181 | if (trace_array_get(tr) < 0) | |
3182 | return -ENODEV; | |
3183 | ||
3184 | /* If this file was open for write, then erase contents */ | |
3185 | if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) { | |
3186 | int cpu = tracing_get_cpu(inode); | |
3187 | ||
3188 | if (cpu == RING_BUFFER_ALL_CPUS) | |
3189 | tracing_reset_online_cpus(&tr->trace_buffer); | |
3190 | else | |
3191 | tracing_reset(&tr->trace_buffer, cpu); | |
3192 | } | |
3193 | ||
3194 | if (file->f_mode & FMODE_READ) { | |
3195 | iter = __tracing_open(inode, file, false); | |
3196 | if (IS_ERR(iter)) | |
3197 | ret = PTR_ERR(iter); | |
3198 | else if (trace_flags & TRACE_ITER_LATENCY_FMT) | |
3199 | iter->iter_flags |= TRACE_FILE_LAT_FMT; | |
3200 | } | |
3201 | ||
3202 | if (ret < 0) | |
3203 | trace_array_put(tr); | |
3204 | ||
3205 | return ret; | |
3206 | } | |
3207 | ||
3208 | /* | |
3209 | * Some tracers are not suitable for instance buffers. | |
3210 | * A tracer is always available for the global array (toplevel) | |
3211 | * or if it explicitly states that it is. | |
3212 | */ | |
3213 | static bool | |
3214 | trace_ok_for_array(struct tracer *t, struct trace_array *tr) | |
3215 | { | |
3216 | return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances; | |
3217 | } | |
3218 | ||
3219 | /* Find the next tracer that this trace array may use */ | |
3220 | static struct tracer * | |
3221 | get_tracer_for_array(struct trace_array *tr, struct tracer *t) | |
3222 | { | |
3223 | while (t && !trace_ok_for_array(t, tr)) | |
3224 | t = t->next; | |
3225 | ||
3226 | return t; | |
3227 | } | |
3228 | ||
3229 | static void * | |
3230 | t_next(struct seq_file *m, void *v, loff_t *pos) | |
3231 | { | |
3232 | struct trace_array *tr = m->private; | |
3233 | struct tracer *t = v; | |
3234 | ||
3235 | (*pos)++; | |
3236 | ||
3237 | if (t) | |
3238 | t = get_tracer_for_array(tr, t->next); | |
3239 | ||
3240 | return t; | |
3241 | } | |
3242 | ||
3243 | static void *t_start(struct seq_file *m, loff_t *pos) | |
3244 | { | |
3245 | struct trace_array *tr = m->private; | |
3246 | struct tracer *t; | |
3247 | loff_t l = 0; | |
3248 | ||
3249 | mutex_lock(&trace_types_lock); | |
3250 | ||
3251 | t = get_tracer_for_array(tr, trace_types); | |
3252 | for (; t && l < *pos; t = t_next(m, t, &l)) | |
3253 | ; | |
3254 | ||
3255 | return t; | |
3256 | } | |
3257 | ||
3258 | static void t_stop(struct seq_file *m, void *p) | |
3259 | { | |
3260 | mutex_unlock(&trace_types_lock); | |
3261 | } | |
3262 | ||
3263 | static int t_show(struct seq_file *m, void *v) | |
3264 | { | |
3265 | struct tracer *t = v; | |
3266 | ||
3267 | if (!t) | |
3268 | return 0; | |
3269 | ||
3270 | seq_printf(m, "%s", t->name); | |
3271 | if (t->next) | |
3272 | seq_putc(m, ' '); | |
3273 | else | |
3274 | seq_putc(m, '\n'); | |
3275 | ||
3276 | return 0; | |
3277 | } | |
3278 | ||
3279 | static const struct seq_operations show_traces_seq_ops = { | |
3280 | .start = t_start, | |
3281 | .next = t_next, | |
3282 | .stop = t_stop, | |
3283 | .show = t_show, | |
3284 | }; | |
3285 | ||
3286 | static int show_traces_open(struct inode *inode, struct file *file) | |
3287 | { | |
3288 | struct trace_array *tr = inode->i_private; | |
3289 | struct seq_file *m; | |
3290 | int ret; | |
3291 | ||
3292 | if (tracing_disabled) | |
3293 | return -ENODEV; | |
3294 | ||
3295 | ret = seq_open(file, &show_traces_seq_ops); | |
3296 | if (ret) | |
3297 | return ret; | |
3298 | ||
3299 | m = file->private_data; | |
3300 | m->private = tr; | |
3301 | ||
3302 | return 0; | |
3303 | } | |
3304 | ||
3305 | static ssize_t | |
3306 | tracing_write_stub(struct file *filp, const char __user *ubuf, | |
3307 | size_t count, loff_t *ppos) | |
3308 | { | |
3309 | return count; | |
3310 | } | |
3311 | ||
3312 | loff_t tracing_lseek(struct file *file, loff_t offset, int whence) | |
3313 | { | |
3314 | int ret; | |
3315 | ||
3316 | if (file->f_mode & FMODE_READ) | |
3317 | ret = seq_lseek(file, offset, whence); | |
3318 | else | |
3319 | file->f_pos = ret = 0; | |
3320 | ||
3321 | return ret; | |
3322 | } | |
3323 | ||
3324 | static const struct file_operations tracing_fops = { | |
3325 | .open = tracing_open, | |
3326 | .read = seq_read, | |
3327 | .write = tracing_write_stub, | |
3328 | .llseek = tracing_lseek, | |
3329 | .release = tracing_release, | |
3330 | }; | |
3331 | ||
3332 | static const struct file_operations show_traces_fops = { | |
3333 | .open = show_traces_open, | |
3334 | .read = seq_read, | |
3335 | .release = seq_release, | |
3336 | .llseek = seq_lseek, | |
3337 | }; | |
3338 | ||
3339 | /* | |
3340 | * The tracer itself will not take this lock, but still we want | |
3341 | * to provide a consistent cpumask to user-space: | |
3342 | */ | |
3343 | static DEFINE_MUTEX(tracing_cpumask_update_lock); | |
3344 | ||
3345 | /* | |
3346 | * Temporary storage for the character representation of the | |
3347 | * CPU bitmask (and one more byte for the newline): | |
3348 | */ | |
3349 | static char mask_str[NR_CPUS + 1]; | |
3350 | ||
3351 | static ssize_t | |
3352 | tracing_cpumask_read(struct file *filp, char __user *ubuf, | |
3353 | size_t count, loff_t *ppos) | |
3354 | { | |
3355 | struct trace_array *tr = file_inode(filp)->i_private; | |
3356 | int len; | |
3357 | ||
3358 | mutex_lock(&tracing_cpumask_update_lock); | |
3359 | ||
3360 | len = cpumask_scnprintf(mask_str, count, tr->tracing_cpumask); | |
3361 | if (count - len < 2) { | |
3362 | count = -EINVAL; | |
3363 | goto out_err; | |
3364 | } | |
3365 | len += sprintf(mask_str + len, "\n"); | |
3366 | count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1); | |
3367 | ||
3368 | out_err: | |
3369 | mutex_unlock(&tracing_cpumask_update_lock); | |
3370 | ||
3371 | return count; | |
3372 | } | |
3373 | ||
3374 | static ssize_t | |
3375 | tracing_cpumask_write(struct file *filp, const char __user *ubuf, | |
3376 | size_t count, loff_t *ppos) | |
3377 | { | |
3378 | struct trace_array *tr = file_inode(filp)->i_private; | |
3379 | cpumask_var_t tracing_cpumask_new; | |
3380 | int err, cpu; | |
3381 | ||
3382 | if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL)) | |
3383 | return -ENOMEM; | |
3384 | ||
3385 | err = cpumask_parse_user(ubuf, count, tracing_cpumask_new); | |
3386 | if (err) | |
3387 | goto err_unlock; | |
3388 | ||
3389 | mutex_lock(&tracing_cpumask_update_lock); | |
3390 | ||
3391 | local_irq_disable(); | |
3392 | arch_spin_lock(&tr->max_lock); | |
3393 | for_each_tracing_cpu(cpu) { | |
3394 | /* | |
3395 | * Increase/decrease the disabled counter if we are | |
3396 | * about to flip a bit in the cpumask: | |
3397 | */ | |
3398 | if (cpumask_test_cpu(cpu, tr->tracing_cpumask) && | |
3399 | !cpumask_test_cpu(cpu, tracing_cpumask_new)) { | |
3400 | atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled); | |
3401 | ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu); | |
3402 | } | |
3403 | if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) && | |
3404 | cpumask_test_cpu(cpu, tracing_cpumask_new)) { | |
3405 | atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled); | |
3406 | ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu); | |
3407 | } | |
3408 | } | |
3409 | arch_spin_unlock(&tr->max_lock); | |
3410 | local_irq_enable(); | |
3411 | ||
3412 | cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new); | |
3413 | ||
3414 | mutex_unlock(&tracing_cpumask_update_lock); | |
3415 | free_cpumask_var(tracing_cpumask_new); | |
3416 | ||
3417 | return count; | |
3418 | ||
3419 | err_unlock: | |
3420 | free_cpumask_var(tracing_cpumask_new); | |
3421 | ||
3422 | return err; | |
3423 | } | |
3424 | ||
3425 | static const struct file_operations tracing_cpumask_fops = { | |
3426 | .open = tracing_open_generic_tr, | |
3427 | .read = tracing_cpumask_read, | |
3428 | .write = tracing_cpumask_write, | |
3429 | .release = tracing_release_generic_tr, | |
3430 | .llseek = generic_file_llseek, | |
3431 | }; | |
3432 | ||
3433 | static int tracing_trace_options_show(struct seq_file *m, void *v) | |
3434 | { | |
3435 | struct tracer_opt *trace_opts; | |
3436 | struct trace_array *tr = m->private; | |
3437 | u32 tracer_flags; | |
3438 | int i; | |
3439 | ||
3440 | mutex_lock(&trace_types_lock); | |
3441 | tracer_flags = tr->current_trace->flags->val; | |
3442 | trace_opts = tr->current_trace->flags->opts; | |
3443 | ||
3444 | for (i = 0; trace_options[i]; i++) { | |
3445 | if (trace_flags & (1 << i)) | |
3446 | seq_printf(m, "%s\n", trace_options[i]); | |
3447 | else | |
3448 | seq_printf(m, "no%s\n", trace_options[i]); | |
3449 | } | |
3450 | ||
3451 | for (i = 0; trace_opts[i].name; i++) { | |
3452 | if (tracer_flags & trace_opts[i].bit) | |
3453 | seq_printf(m, "%s\n", trace_opts[i].name); | |
3454 | else | |
3455 | seq_printf(m, "no%s\n", trace_opts[i].name); | |
3456 | } | |
3457 | mutex_unlock(&trace_types_lock); | |
3458 | ||
3459 | return 0; | |
3460 | } | |
3461 | ||
3462 | static int __set_tracer_option(struct trace_array *tr, | |
3463 | struct tracer_flags *tracer_flags, | |
3464 | struct tracer_opt *opts, int neg) | |
3465 | { | |
3466 | struct tracer *trace = tr->current_trace; | |
3467 | int ret; | |
3468 | ||
3469 | ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg); | |
3470 | if (ret) | |
3471 | return ret; | |
3472 | ||
3473 | if (neg) | |
3474 | tracer_flags->val &= ~opts->bit; | |
3475 | else | |
3476 | tracer_flags->val |= opts->bit; | |
3477 | return 0; | |
3478 | } | |
3479 | ||
3480 | /* Try to assign a tracer specific option */ | |
3481 | static int set_tracer_option(struct trace_array *tr, char *cmp, int neg) | |
3482 | { | |
3483 | struct tracer *trace = tr->current_trace; | |
3484 | struct tracer_flags *tracer_flags = trace->flags; | |
3485 | struct tracer_opt *opts = NULL; | |
3486 | int i; | |
3487 | ||
3488 | for (i = 0; tracer_flags->opts[i].name; i++) { | |
3489 | opts = &tracer_flags->opts[i]; | |
3490 | ||
3491 | if (strcmp(cmp, opts->name) == 0) | |
3492 | return __set_tracer_option(tr, trace->flags, opts, neg); | |
3493 | } | |
3494 | ||
3495 | return -EINVAL; | |
3496 | } | |
3497 | ||
3498 | /* Some tracers require overwrite to stay enabled */ | |
3499 | int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set) | |
3500 | { | |
3501 | if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set) | |
3502 | return -1; | |
3503 | ||
3504 | return 0; | |
3505 | } | |
3506 | ||
3507 | int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled) | |
3508 | { | |
3509 | /* do nothing if flag is already set */ | |
3510 | if (!!(trace_flags & mask) == !!enabled) | |
3511 | return 0; | |
3512 | ||
3513 | /* Give the tracer a chance to approve the change */ | |
3514 | if (tr->current_trace->flag_changed) | |
3515 | if (tr->current_trace->flag_changed(tr, mask, !!enabled)) | |
3516 | return -EINVAL; | |
3517 | ||
3518 | if (enabled) | |
3519 | trace_flags |= mask; | |
3520 | else | |
3521 | trace_flags &= ~mask; | |
3522 | ||
3523 | if (mask == TRACE_ITER_RECORD_CMD) | |
3524 | trace_event_enable_cmd_record(enabled); | |
3525 | ||
3526 | if (mask == TRACE_ITER_OVERWRITE) { | |
3527 | ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled); | |
3528 | #ifdef CONFIG_TRACER_MAX_TRACE | |
3529 | ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled); | |
3530 | #endif | |
3531 | } | |
3532 | ||
3533 | if (mask == TRACE_ITER_PRINTK) | |
3534 | trace_printk_start_stop_comm(enabled); | |
3535 | ||
3536 | return 0; | |
3537 | } | |
3538 | ||
3539 | static int trace_set_options(struct trace_array *tr, char *option) | |
3540 | { | |
3541 | char *cmp; | |
3542 | int neg = 0; | |
3543 | int ret = -ENODEV; | |
3544 | int i; | |
3545 | ||
3546 | cmp = strstrip(option); | |
3547 | ||
3548 | if (strncmp(cmp, "no", 2) == 0) { | |
3549 | neg = 1; | |
3550 | cmp += 2; | |
3551 | } | |
3552 | ||
3553 | mutex_lock(&trace_types_lock); | |
3554 | ||
3555 | for (i = 0; trace_options[i]; i++) { | |
3556 | if (strcmp(cmp, trace_options[i]) == 0) { | |
3557 | ret = set_tracer_flag(tr, 1 << i, !neg); | |
3558 | break; | |
3559 | } | |
3560 | } | |
3561 | ||
3562 | /* If no option could be set, test the specific tracer options */ | |
3563 | if (!trace_options[i]) | |
3564 | ret = set_tracer_option(tr, cmp, neg); | |
3565 | ||
3566 | mutex_unlock(&trace_types_lock); | |
3567 | ||
3568 | return ret; | |
3569 | } | |
3570 | ||
3571 | static ssize_t | |
3572 | tracing_trace_options_write(struct file *filp, const char __user *ubuf, | |
3573 | size_t cnt, loff_t *ppos) | |
3574 | { | |
3575 | struct seq_file *m = filp->private_data; | |
3576 | struct trace_array *tr = m->private; | |
3577 | char buf[64]; | |
3578 | int ret; | |
3579 | ||
3580 | if (cnt >= sizeof(buf)) | |
3581 | return -EINVAL; | |
3582 | ||
3583 | if (copy_from_user(&buf, ubuf, cnt)) | |
3584 | return -EFAULT; | |
3585 | ||
3586 | buf[cnt] = 0; | |
3587 | ||
3588 | ret = trace_set_options(tr, buf); | |
3589 | if (ret < 0) | |
3590 | return ret; | |
3591 | ||
3592 | *ppos += cnt; | |
3593 | ||
3594 | return cnt; | |
3595 | } | |
3596 | ||
3597 | static int tracing_trace_options_open(struct inode *inode, struct file *file) | |
3598 | { | |
3599 | struct trace_array *tr = inode->i_private; | |
3600 | int ret; | |
3601 | ||
3602 | if (tracing_disabled) | |
3603 | return -ENODEV; | |
3604 | ||
3605 | if (trace_array_get(tr) < 0) | |
3606 | return -ENODEV; | |
3607 | ||
3608 | ret = single_open(file, tracing_trace_options_show, inode->i_private); | |
3609 | if (ret < 0) | |
3610 | trace_array_put(tr); | |
3611 | ||
3612 | return ret; | |
3613 | } | |
3614 | ||
3615 | static const struct file_operations tracing_iter_fops = { | |
3616 | .open = tracing_trace_options_open, | |
3617 | .read = seq_read, | |
3618 | .llseek = seq_lseek, | |
3619 | .release = tracing_single_release_tr, | |
3620 | .write = tracing_trace_options_write, | |
3621 | }; | |
3622 | ||
3623 | static const char readme_msg[] = | |
3624 | "tracing mini-HOWTO:\n\n" | |
3625 | "# echo 0 > tracing_on : quick way to disable tracing\n" | |
3626 | "# echo 1 > tracing_on : quick way to re-enable tracing\n\n" | |
3627 | " Important files:\n" | |
3628 | " trace\t\t\t- The static contents of the buffer\n" | |
3629 | "\t\t\t To clear the buffer write into this file: echo > trace\n" | |
3630 | " trace_pipe\t\t- A consuming read to see the contents of the buffer\n" | |
3631 | " current_tracer\t- function and latency tracers\n" | |
3632 | " available_tracers\t- list of configured tracers for current_tracer\n" | |
3633 | " buffer_size_kb\t- view and modify size of per cpu buffer\n" | |
3634 | " buffer_total_size_kb - view total size of all cpu buffers\n\n" | |
3635 | " trace_clock\t\t-change the clock used to order events\n" | |
3636 | " local: Per cpu clock but may not be synced across CPUs\n" | |
3637 | " global: Synced across CPUs but slows tracing down.\n" | |
3638 | " counter: Not a clock, but just an increment\n" | |
3639 | " uptime: Jiffy counter from time of boot\n" | |
3640 | " perf: Same clock that perf events use\n" | |
3641 | #ifdef CONFIG_X86_64 | |
3642 | " x86-tsc: TSC cycle counter\n" | |
3643 | #endif | |
3644 | "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n" | |
3645 | " tracing_cpumask\t- Limit which CPUs to trace\n" | |
3646 | " instances\t\t- Make sub-buffers with: mkdir instances/foo\n" | |
3647 | "\t\t\t Remove sub-buffer with rmdir\n" | |
3648 | " trace_options\t\t- Set format or modify how tracing happens\n" | |
3649 | "\t\t\t Disable an option by adding a suffix 'no' to the\n" | |
3650 | "\t\t\t option name\n" | |
3651 | " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n" | |
3652 | #ifdef CONFIG_DYNAMIC_FTRACE | |
3653 | "\n available_filter_functions - list of functions that can be filtered on\n" | |
3654 | " set_ftrace_filter\t- echo function name in here to only trace these\n" | |
3655 | "\t\t\t functions\n" | |
3656 | "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n" | |
3657 | "\t modules: Can select a group via module\n" | |
3658 | "\t Format: :mod:<module-name>\n" | |
3659 | "\t example: echo :mod:ext3 > set_ftrace_filter\n" | |
3660 | "\t triggers: a command to perform when function is hit\n" | |
3661 | "\t Format: <function>:<trigger>[:count]\n" | |
3662 | "\t trigger: traceon, traceoff\n" | |
3663 | "\t\t enable_event:<system>:<event>\n" | |
3664 | "\t\t disable_event:<system>:<event>\n" | |
3665 | #ifdef CONFIG_STACKTRACE | |
3666 | "\t\t stacktrace\n" | |
3667 | #endif | |
3668 | #ifdef CONFIG_TRACER_SNAPSHOT | |
3669 | "\t\t snapshot\n" | |
3670 | #endif | |
3671 | "\t\t dump\n" | |
3672 | "\t\t cpudump\n" | |
3673 | "\t example: echo do_fault:traceoff > set_ftrace_filter\n" | |
3674 | "\t echo do_trap:traceoff:3 > set_ftrace_filter\n" | |
3675 | "\t The first one will disable tracing every time do_fault is hit\n" | |
3676 | "\t The second will disable tracing at most 3 times when do_trap is hit\n" | |
3677 | "\t The first time do trap is hit and it disables tracing, the\n" | |
3678 | "\t counter will decrement to 2. If tracing is already disabled,\n" | |
3679 | "\t the counter will not decrement. It only decrements when the\n" | |
3680 | "\t trigger did work\n" | |
3681 | "\t To remove trigger without count:\n" | |
3682 | "\t echo '!<function>:<trigger> > set_ftrace_filter\n" | |
3683 | "\t To remove trigger with a count:\n" | |
3684 | "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n" | |
3685 | " set_ftrace_notrace\t- echo function name in here to never trace.\n" | |
3686 | "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n" | |
3687 | "\t modules: Can select a group via module command :mod:\n" | |
3688 | "\t Does not accept triggers\n" | |
3689 | #endif /* CONFIG_DYNAMIC_FTRACE */ | |
3690 | #ifdef CONFIG_FUNCTION_TRACER | |
3691 | " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n" | |
3692 | "\t\t (function)\n" | |
3693 | #endif | |
3694 | #ifdef CONFIG_FUNCTION_GRAPH_TRACER | |
3695 | " set_graph_function\t- Trace the nested calls of a function (function_graph)\n" | |
3696 | " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n" | |
3697 | #endif | |
3698 | #ifdef CONFIG_TRACER_SNAPSHOT | |
3699 | "\n snapshot\t\t- Like 'trace' but shows the content of the static\n" | |
3700 | "\t\t\t snapshot buffer. Read the contents for more\n" | |
3701 | "\t\t\t information\n" | |
3702 | #endif | |
3703 | #ifdef CONFIG_STACK_TRACER | |
3704 | " stack_trace\t\t- Shows the max stack trace when active\n" | |
3705 | " stack_max_size\t- Shows current max stack size that was traced\n" | |
3706 | "\t\t\t Write into this file to reset the max size (trigger a\n" | |
3707 | "\t\t\t new trace)\n" | |
3708 | #ifdef CONFIG_DYNAMIC_FTRACE | |
3709 | " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n" | |
3710 | "\t\t\t traces\n" | |
3711 | #endif | |
3712 | #endif /* CONFIG_STACK_TRACER */ | |
3713 | " events/\t\t- Directory containing all trace event subsystems:\n" | |
3714 | " enable\t\t- Write 0/1 to enable/disable tracing of all events\n" | |
3715 | " events/<system>/\t- Directory containing all trace events for <system>:\n" | |
3716 | " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n" | |
3717 | "\t\t\t events\n" | |
3718 | " filter\t\t- If set, only events passing filter are traced\n" | |
3719 | " events/<system>/<event>/\t- Directory containing control files for\n" | |
3720 | "\t\t\t <event>:\n" | |
3721 | " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n" | |
3722 | " filter\t\t- If set, only events passing filter are traced\n" | |
3723 | " trigger\t\t- If set, a command to perform when event is hit\n" | |
3724 | "\t Format: <trigger>[:count][if <filter>]\n" | |
3725 | "\t trigger: traceon, traceoff\n" | |
3726 | "\t enable_event:<system>:<event>\n" | |
3727 | "\t disable_event:<system>:<event>\n" | |
3728 | #ifdef CONFIG_STACKTRACE | |
3729 | "\t\t stacktrace\n" | |
3730 | #endif | |
3731 | #ifdef CONFIG_TRACER_SNAPSHOT | |
3732 | "\t\t snapshot\n" | |
3733 | #endif | |
3734 | "\t example: echo traceoff > events/block/block_unplug/trigger\n" | |
3735 | "\t echo traceoff:3 > events/block/block_unplug/trigger\n" | |
3736 | "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n" | |
3737 | "\t events/block/block_unplug/trigger\n" | |
3738 | "\t The first disables tracing every time block_unplug is hit.\n" | |
3739 | "\t The second disables tracing the first 3 times block_unplug is hit.\n" | |
3740 | "\t The third enables the kmalloc event the first 3 times block_unplug\n" | |
3741 | "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n" | |
3742 | "\t Like function triggers, the counter is only decremented if it\n" | |
3743 | "\t enabled or disabled tracing.\n" | |
3744 | "\t To remove a trigger without a count:\n" | |
3745 | "\t echo '!<trigger> > <system>/<event>/trigger\n" | |
3746 | "\t To remove a trigger with a count:\n" | |
3747 | "\t echo '!<trigger>:0 > <system>/<event>/trigger\n" | |
3748 | "\t Filters can be ignored when removing a trigger.\n" | |
3749 | ; | |
3750 | ||
3751 | static ssize_t | |
3752 | tracing_readme_read(struct file *filp, char __user *ubuf, | |
3753 | size_t cnt, loff_t *ppos) | |
3754 | { | |
3755 | return simple_read_from_buffer(ubuf, cnt, ppos, | |
3756 | readme_msg, strlen(readme_msg)); | |
3757 | } | |
3758 | ||
3759 | static const struct file_operations tracing_readme_fops = { | |
3760 | .open = tracing_open_generic, | |
3761 | .read = tracing_readme_read, | |
3762 | .llseek = generic_file_llseek, | |
3763 | }; | |
3764 | ||
3765 | static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos) | |
3766 | { | |
3767 | unsigned int *ptr = v; | |
3768 | ||
3769 | if (*pos || m->count) | |
3770 | ptr++; | |
3771 | ||
3772 | (*pos)++; | |
3773 | ||
3774 | for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num]; | |
3775 | ptr++) { | |
3776 | if (*ptr == -1 || *ptr == NO_CMDLINE_MAP) | |
3777 | continue; | |
3778 | ||
3779 | return ptr; | |
3780 | } | |
3781 | ||
3782 | return NULL; | |
3783 | } | |
3784 | ||
3785 | static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos) | |
3786 | { | |
3787 | void *v; | |
3788 | loff_t l = 0; | |
3789 | ||
3790 | preempt_disable(); | |
3791 | arch_spin_lock(&trace_cmdline_lock); | |
3792 | ||
3793 | v = &savedcmd->map_cmdline_to_pid[0]; | |
3794 | while (l <= *pos) { | |
3795 | v = saved_cmdlines_next(m, v, &l); | |
3796 | if (!v) | |
3797 | return NULL; | |
3798 | } | |
3799 | ||
3800 | return v; | |
3801 | } | |
3802 | ||
3803 | static void saved_cmdlines_stop(struct seq_file *m, void *v) | |
3804 | { | |
3805 | arch_spin_unlock(&trace_cmdline_lock); | |
3806 | preempt_enable(); | |
3807 | } | |
3808 | ||
3809 | static int saved_cmdlines_show(struct seq_file *m, void *v) | |
3810 | { | |
3811 | char buf[TASK_COMM_LEN]; | |
3812 | unsigned int *pid = v; | |
3813 | ||
3814 | __trace_find_cmdline(*pid, buf); | |
3815 | seq_printf(m, "%d %s\n", *pid, buf); | |
3816 | return 0; | |
3817 | } | |
3818 | ||
3819 | static const struct seq_operations tracing_saved_cmdlines_seq_ops = { | |
3820 | .start = saved_cmdlines_start, | |
3821 | .next = saved_cmdlines_next, | |
3822 | .stop = saved_cmdlines_stop, | |
3823 | .show = saved_cmdlines_show, | |
3824 | }; | |
3825 | ||
3826 | static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp) | |
3827 | { | |
3828 | if (tracing_disabled) | |
3829 | return -ENODEV; | |
3830 | ||
3831 | return seq_open(filp, &tracing_saved_cmdlines_seq_ops); | |
3832 | } | |
3833 | ||
3834 | static const struct file_operations tracing_saved_cmdlines_fops = { | |
3835 | .open = tracing_saved_cmdlines_open, | |
3836 | .read = seq_read, | |
3837 | .llseek = seq_lseek, | |
3838 | .release = seq_release, | |
3839 | }; | |
3840 | ||
3841 | static ssize_t | |
3842 | tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf, | |
3843 | size_t cnt, loff_t *ppos) | |
3844 | { | |
3845 | char buf[64]; | |
3846 | int r; | |
3847 | ||
3848 | arch_spin_lock(&trace_cmdline_lock); | |
3849 | r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num); | |
3850 | arch_spin_unlock(&trace_cmdline_lock); | |
3851 | ||
3852 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3853 | } | |
3854 | ||
3855 | static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s) | |
3856 | { | |
3857 | kfree(s->saved_cmdlines); | |
3858 | kfree(s->map_cmdline_to_pid); | |
3859 | kfree(s); | |
3860 | } | |
3861 | ||
3862 | static int tracing_resize_saved_cmdlines(unsigned int val) | |
3863 | { | |
3864 | struct saved_cmdlines_buffer *s, *savedcmd_temp; | |
3865 | ||
3866 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
3867 | if (!s) | |
3868 | return -ENOMEM; | |
3869 | ||
3870 | if (allocate_cmdlines_buffer(val, s) < 0) { | |
3871 | kfree(s); | |
3872 | return -ENOMEM; | |
3873 | } | |
3874 | ||
3875 | arch_spin_lock(&trace_cmdline_lock); | |
3876 | savedcmd_temp = savedcmd; | |
3877 | savedcmd = s; | |
3878 | arch_spin_unlock(&trace_cmdline_lock); | |
3879 | free_saved_cmdlines_buffer(savedcmd_temp); | |
3880 | ||
3881 | return 0; | |
3882 | } | |
3883 | ||
3884 | static ssize_t | |
3885 | tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf, | |
3886 | size_t cnt, loff_t *ppos) | |
3887 | { | |
3888 | unsigned long val; | |
3889 | int ret; | |
3890 | ||
3891 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
3892 | if (ret) | |
3893 | return ret; | |
3894 | ||
3895 | /* must have at least 1 entry or less than PID_MAX_DEFAULT */ | |
3896 | if (!val || val > PID_MAX_DEFAULT) | |
3897 | return -EINVAL; | |
3898 | ||
3899 | ret = tracing_resize_saved_cmdlines((unsigned int)val); | |
3900 | if (ret < 0) | |
3901 | return ret; | |
3902 | ||
3903 | *ppos += cnt; | |
3904 | ||
3905 | return cnt; | |
3906 | } | |
3907 | ||
3908 | static const struct file_operations tracing_saved_cmdlines_size_fops = { | |
3909 | .open = tracing_open_generic, | |
3910 | .read = tracing_saved_cmdlines_size_read, | |
3911 | .write = tracing_saved_cmdlines_size_write, | |
3912 | }; | |
3913 | ||
3914 | static ssize_t | |
3915 | tracing_set_trace_read(struct file *filp, char __user *ubuf, | |
3916 | size_t cnt, loff_t *ppos) | |
3917 | { | |
3918 | struct trace_array *tr = filp->private_data; | |
3919 | char buf[MAX_TRACER_SIZE+2]; | |
3920 | int r; | |
3921 | ||
3922 | mutex_lock(&trace_types_lock); | |
3923 | r = sprintf(buf, "%s\n", tr->current_trace->name); | |
3924 | mutex_unlock(&trace_types_lock); | |
3925 | ||
3926 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3927 | } | |
3928 | ||
3929 | int tracer_init(struct tracer *t, struct trace_array *tr) | |
3930 | { | |
3931 | tracing_reset_online_cpus(&tr->trace_buffer); | |
3932 | return t->init(tr); | |
3933 | } | |
3934 | ||
3935 | static void set_buffer_entries(struct trace_buffer *buf, unsigned long val) | |
3936 | { | |
3937 | int cpu; | |
3938 | ||
3939 | for_each_tracing_cpu(cpu) | |
3940 | per_cpu_ptr(buf->data, cpu)->entries = val; | |
3941 | } | |
3942 | ||
3943 | #ifdef CONFIG_TRACER_MAX_TRACE | |
3944 | /* resize @tr's buffer to the size of @size_tr's entries */ | |
3945 | static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf, | |
3946 | struct trace_buffer *size_buf, int cpu_id) | |
3947 | { | |
3948 | int cpu, ret = 0; | |
3949 | ||
3950 | if (cpu_id == RING_BUFFER_ALL_CPUS) { | |
3951 | for_each_tracing_cpu(cpu) { | |
3952 | ret = ring_buffer_resize(trace_buf->buffer, | |
3953 | per_cpu_ptr(size_buf->data, cpu)->entries, cpu); | |
3954 | if (ret < 0) | |
3955 | break; | |
3956 | per_cpu_ptr(trace_buf->data, cpu)->entries = | |
3957 | per_cpu_ptr(size_buf->data, cpu)->entries; | |
3958 | } | |
3959 | } else { | |
3960 | ret = ring_buffer_resize(trace_buf->buffer, | |
3961 | per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id); | |
3962 | if (ret == 0) | |
3963 | per_cpu_ptr(trace_buf->data, cpu_id)->entries = | |
3964 | per_cpu_ptr(size_buf->data, cpu_id)->entries; | |
3965 | } | |
3966 | ||
3967 | return ret; | |
3968 | } | |
3969 | #endif /* CONFIG_TRACER_MAX_TRACE */ | |
3970 | ||
3971 | static int __tracing_resize_ring_buffer(struct trace_array *tr, | |
3972 | unsigned long size, int cpu) | |
3973 | { | |
3974 | int ret; | |
3975 | ||
3976 | /* | |
3977 | * If kernel or user changes the size of the ring buffer | |
3978 | * we use the size that was given, and we can forget about | |
3979 | * expanding it later. | |
3980 | */ | |
3981 | ring_buffer_expanded = true; | |
3982 | ||
3983 | /* May be called before buffers are initialized */ | |
3984 | if (!tr->trace_buffer.buffer) | |
3985 | return 0; | |
3986 | ||
3987 | ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu); | |
3988 | if (ret < 0) | |
3989 | return ret; | |
3990 | ||
3991 | #ifdef CONFIG_TRACER_MAX_TRACE | |
3992 | if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) || | |
3993 | !tr->current_trace->use_max_tr) | |
3994 | goto out; | |
3995 | ||
3996 | ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu); | |
3997 | if (ret < 0) { | |
3998 | int r = resize_buffer_duplicate_size(&tr->trace_buffer, | |
3999 | &tr->trace_buffer, cpu); | |
4000 | if (r < 0) { | |
4001 | /* | |
4002 | * AARGH! We are left with different | |
4003 | * size max buffer!!!! | |
4004 | * The max buffer is our "snapshot" buffer. | |
4005 | * When a tracer needs a snapshot (one of the | |
4006 | * latency tracers), it swaps the max buffer | |
4007 | * with the saved snap shot. We succeeded to | |
4008 | * update the size of the main buffer, but failed to | |
4009 | * update the size of the max buffer. But when we tried | |
4010 | * to reset the main buffer to the original size, we | |
4011 | * failed there too. This is very unlikely to | |
4012 | * happen, but if it does, warn and kill all | |
4013 | * tracing. | |
4014 | */ | |
4015 | WARN_ON(1); | |
4016 | tracing_disabled = 1; | |
4017 | } | |
4018 | return ret; | |
4019 | } | |
4020 | ||
4021 | if (cpu == RING_BUFFER_ALL_CPUS) | |
4022 | set_buffer_entries(&tr->max_buffer, size); | |
4023 | else | |
4024 | per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size; | |
4025 | ||
4026 | out: | |
4027 | #endif /* CONFIG_TRACER_MAX_TRACE */ | |
4028 | ||
4029 | if (cpu == RING_BUFFER_ALL_CPUS) | |
4030 | set_buffer_entries(&tr->trace_buffer, size); | |
4031 | else | |
4032 | per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size; | |
4033 | ||
4034 | return ret; | |
4035 | } | |
4036 | ||
4037 | static ssize_t tracing_resize_ring_buffer(struct trace_array *tr, | |
4038 | unsigned long size, int cpu_id) | |
4039 | { | |
4040 | int ret = size; | |
4041 | ||
4042 | mutex_lock(&trace_types_lock); | |
4043 | ||
4044 | if (cpu_id != RING_BUFFER_ALL_CPUS) { | |
4045 | /* make sure, this cpu is enabled in the mask */ | |
4046 | if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) { | |
4047 | ret = -EINVAL; | |
4048 | goto out; | |
4049 | } | |
4050 | } | |
4051 | ||
4052 | ret = __tracing_resize_ring_buffer(tr, size, cpu_id); | |
4053 | if (ret < 0) | |
4054 | ret = -ENOMEM; | |
4055 | ||
4056 | out: | |
4057 | mutex_unlock(&trace_types_lock); | |
4058 | ||
4059 | return ret; | |
4060 | } | |
4061 | ||
4062 | ||
4063 | /** | |
4064 | * tracing_update_buffers - used by tracing facility to expand ring buffers | |
4065 | * | |
4066 | * To save on memory when the tracing is never used on a system with it | |
4067 | * configured in. The ring buffers are set to a minimum size. But once | |
4068 | * a user starts to use the tracing facility, then they need to grow | |
4069 | * to their default size. | |
4070 | * | |
4071 | * This function is to be called when a tracer is about to be used. | |
4072 | */ | |
4073 | int tracing_update_buffers(void) | |
4074 | { | |
4075 | int ret = 0; | |
4076 | ||
4077 | mutex_lock(&trace_types_lock); | |
4078 | if (!ring_buffer_expanded) | |
4079 | ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size, | |
4080 | RING_BUFFER_ALL_CPUS); | |
4081 | mutex_unlock(&trace_types_lock); | |
4082 | ||
4083 | return ret; | |
4084 | } | |
4085 | ||
4086 | struct trace_option_dentry; | |
4087 | ||
4088 | static struct trace_option_dentry * | |
4089 | create_trace_option_files(struct trace_array *tr, struct tracer *tracer); | |
4090 | ||
4091 | static void | |
4092 | destroy_trace_option_files(struct trace_option_dentry *topts); | |
4093 | ||
4094 | /* | |
4095 | * Used to clear out the tracer before deletion of an instance. | |
4096 | * Must have trace_types_lock held. | |
4097 | */ | |
4098 | static void tracing_set_nop(struct trace_array *tr) | |
4099 | { | |
4100 | if (tr->current_trace == &nop_trace) | |
4101 | return; | |
4102 | ||
4103 | tr->current_trace->enabled--; | |
4104 | ||
4105 | if (tr->current_trace->reset) | |
4106 | tr->current_trace->reset(tr); | |
4107 | ||
4108 | tr->current_trace = &nop_trace; | |
4109 | } | |
4110 | ||
4111 | static int tracing_set_tracer(struct trace_array *tr, const char *buf) | |
4112 | { | |
4113 | static struct trace_option_dentry *topts; | |
4114 | struct tracer *t; | |
4115 | #ifdef CONFIG_TRACER_MAX_TRACE | |
4116 | bool had_max_tr; | |
4117 | #endif | |
4118 | int ret = 0; | |
4119 | ||
4120 | mutex_lock(&trace_types_lock); | |
4121 | ||
4122 | if (!ring_buffer_expanded) { | |
4123 | ret = __tracing_resize_ring_buffer(tr, trace_buf_size, | |
4124 | RING_BUFFER_ALL_CPUS); | |
4125 | if (ret < 0) | |
4126 | goto out; | |
4127 | ret = 0; | |
4128 | } | |
4129 | ||
4130 | for (t = trace_types; t; t = t->next) { | |
4131 | if (strcmp(t->name, buf) == 0) | |
4132 | break; | |
4133 | } | |
4134 | if (!t) { | |
4135 | ret = -EINVAL; | |
4136 | goto out; | |
4137 | } | |
4138 | if (t == tr->current_trace) | |
4139 | goto out; | |
4140 | ||
4141 | /* Some tracers are only allowed for the top level buffer */ | |
4142 | if (!trace_ok_for_array(t, tr)) { | |
4143 | ret = -EINVAL; | |
4144 | goto out; | |
4145 | } | |
4146 | ||
4147 | trace_branch_disable(); | |
4148 | ||
4149 | tr->current_trace->enabled--; | |
4150 | ||
4151 | if (tr->current_trace->reset) | |
4152 | tr->current_trace->reset(tr); | |
4153 | ||
4154 | /* Current trace needs to be nop_trace before synchronize_sched */ | |
4155 | tr->current_trace = &nop_trace; | |
4156 | ||
4157 | #ifdef CONFIG_TRACER_MAX_TRACE | |
4158 | had_max_tr = tr->allocated_snapshot; | |
4159 | ||
4160 | if (had_max_tr && !t->use_max_tr) { | |
4161 | /* | |
4162 | * We need to make sure that the update_max_tr sees that | |
4163 | * current_trace changed to nop_trace to keep it from | |
4164 | * swapping the buffers after we resize it. | |
4165 | * The update_max_tr is called from interrupts disabled | |
4166 | * so a synchronized_sched() is sufficient. | |
4167 | */ | |
4168 | synchronize_sched(); | |
4169 | free_snapshot(tr); | |
4170 | } | |
4171 | #endif | |
4172 | /* Currently, only the top instance has options */ | |
4173 | if (tr->flags & TRACE_ARRAY_FL_GLOBAL) { | |
4174 | destroy_trace_option_files(topts); | |
4175 | topts = create_trace_option_files(tr, t); | |
4176 | } | |
4177 | ||
4178 | #ifdef CONFIG_TRACER_MAX_TRACE | |
4179 | if (t->use_max_tr && !had_max_tr) { | |
4180 | ret = alloc_snapshot(tr); | |
4181 | if (ret < 0) | |
4182 | goto out; | |
4183 | } | |
4184 | #endif | |
4185 | ||
4186 | if (t->init) { | |
4187 | ret = tracer_init(t, tr); | |
4188 | if (ret) | |
4189 | goto out; | |
4190 | } | |
4191 | ||
4192 | tr->current_trace = t; | |
4193 | tr->current_trace->enabled++; | |
4194 | trace_branch_enable(tr); | |
4195 | out: | |
4196 | mutex_unlock(&trace_types_lock); | |
4197 | ||
4198 | return ret; | |
4199 | } | |
4200 | ||
4201 | static ssize_t | |
4202 | tracing_set_trace_write(struct file *filp, const char __user *ubuf, | |
4203 | size_t cnt, loff_t *ppos) | |
4204 | { | |
4205 | struct trace_array *tr = filp->private_data; | |
4206 | char buf[MAX_TRACER_SIZE+1]; | |
4207 | int i; | |
4208 | size_t ret; | |
4209 | int err; | |
4210 | ||
4211 | ret = cnt; | |
4212 | ||
4213 | if (cnt > MAX_TRACER_SIZE) | |
4214 | cnt = MAX_TRACER_SIZE; | |
4215 | ||
4216 | if (copy_from_user(&buf, ubuf, cnt)) | |
4217 | return -EFAULT; | |
4218 | ||
4219 | buf[cnt] = 0; | |
4220 | ||
4221 | /* strip ending whitespace. */ | |
4222 | for (i = cnt - 1; i > 0 && isspace(buf[i]); i--) | |
4223 | buf[i] = 0; | |
4224 | ||
4225 | err = tracing_set_tracer(tr, buf); | |
4226 | if (err) | |
4227 | return err; | |
4228 | ||
4229 | *ppos += ret; | |
4230 | ||
4231 | return ret; | |
4232 | } | |
4233 | ||
4234 | static ssize_t | |
4235 | tracing_max_lat_read(struct file *filp, char __user *ubuf, | |
4236 | size_t cnt, loff_t *ppos) | |
4237 | { | |
4238 | unsigned long *ptr = filp->private_data; | |
4239 | char buf[64]; | |
4240 | int r; | |
4241 | ||
4242 | r = snprintf(buf, sizeof(buf), "%ld\n", | |
4243 | *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr)); | |
4244 | if (r > sizeof(buf)) | |
4245 | r = sizeof(buf); | |
4246 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
4247 | } | |
4248 | ||
4249 | static ssize_t | |
4250 | tracing_max_lat_write(struct file *filp, const char __user *ubuf, | |
4251 | size_t cnt, loff_t *ppos) | |
4252 | { | |
4253 | unsigned long *ptr = filp->private_data; | |
4254 | unsigned long val; | |
4255 | int ret; | |
4256 | ||
4257 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
4258 | if (ret) | |
4259 | return ret; | |
4260 | ||
4261 | *ptr = val * 1000; | |
4262 | ||
4263 | return cnt; | |
4264 | } | |
4265 | ||
4266 | static int tracing_open_pipe(struct inode *inode, struct file *filp) | |
4267 | { | |
4268 | struct trace_array *tr = inode->i_private; | |
4269 | struct trace_iterator *iter; | |
4270 | int ret = 0; | |
4271 | ||
4272 | if (tracing_disabled) | |
4273 | return -ENODEV; | |
4274 | ||
4275 | if (trace_array_get(tr) < 0) | |
4276 | return -ENODEV; | |
4277 | ||
4278 | mutex_lock(&trace_types_lock); | |
4279 | ||
4280 | /* create a buffer to store the information to pass to userspace */ | |
4281 | iter = kzalloc(sizeof(*iter), GFP_KERNEL); | |
4282 | if (!iter) { | |
4283 | ret = -ENOMEM; | |
4284 | __trace_array_put(tr); | |
4285 | goto out; | |
4286 | } | |
4287 | ||
4288 | /* | |
4289 | * We make a copy of the current tracer to avoid concurrent | |
4290 | * changes on it while we are reading. | |
4291 | */ | |
4292 | iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL); | |
4293 | if (!iter->trace) { | |
4294 | ret = -ENOMEM; | |
4295 | goto fail; | |
4296 | } | |
4297 | *iter->trace = *tr->current_trace; | |
4298 | ||
4299 | if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) { | |
4300 | ret = -ENOMEM; | |
4301 | goto fail; | |
4302 | } | |
4303 | ||
4304 | /* trace pipe does not show start of buffer */ | |
4305 | cpumask_setall(iter->started); | |
4306 | ||
4307 | if (trace_flags & TRACE_ITER_LATENCY_FMT) | |
4308 | iter->iter_flags |= TRACE_FILE_LAT_FMT; | |
4309 | ||
4310 | /* Output in nanoseconds only if we are using a clock in nanoseconds. */ | |
4311 | if (trace_clocks[tr->clock_id].in_ns) | |
4312 | iter->iter_flags |= TRACE_FILE_TIME_IN_NS; | |
4313 | ||
4314 | iter->tr = tr; | |
4315 | iter->trace_buffer = &tr->trace_buffer; | |
4316 | iter->cpu_file = tracing_get_cpu(inode); | |
4317 | mutex_init(&iter->mutex); | |
4318 | filp->private_data = iter; | |
4319 | ||
4320 | if (iter->trace->pipe_open) | |
4321 | iter->trace->pipe_open(iter); | |
4322 | ||
4323 | nonseekable_open(inode, filp); | |
4324 | out: | |
4325 | mutex_unlock(&trace_types_lock); | |
4326 | return ret; | |
4327 | ||
4328 | fail: | |
4329 | kfree(iter->trace); | |
4330 | kfree(iter); | |
4331 | __trace_array_put(tr); | |
4332 | mutex_unlock(&trace_types_lock); | |
4333 | return ret; | |
4334 | } | |
4335 | ||
4336 | static int tracing_release_pipe(struct inode *inode, struct file *file) | |
4337 | { | |
4338 | struct trace_iterator *iter = file->private_data; | |
4339 | struct trace_array *tr = inode->i_private; | |
4340 | ||
4341 | mutex_lock(&trace_types_lock); | |
4342 | ||
4343 | if (iter->trace->pipe_close) | |
4344 | iter->trace->pipe_close(iter); | |
4345 | ||
4346 | mutex_unlock(&trace_types_lock); | |
4347 | ||
4348 | free_cpumask_var(iter->started); | |
4349 | mutex_destroy(&iter->mutex); | |
4350 | kfree(iter->trace); | |
4351 | kfree(iter); | |
4352 | ||
4353 | trace_array_put(tr); | |
4354 | ||
4355 | return 0; | |
4356 | } | |
4357 | ||
4358 | static unsigned int | |
4359 | trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table) | |
4360 | { | |
4361 | /* Iterators are static, they should be filled or empty */ | |
4362 | if (trace_buffer_iter(iter, iter->cpu_file)) | |
4363 | return POLLIN | POLLRDNORM; | |
4364 | ||
4365 | if (trace_flags & TRACE_ITER_BLOCK) | |
4366 | /* | |
4367 | * Always select as readable when in blocking mode | |
4368 | */ | |
4369 | return POLLIN | POLLRDNORM; | |
4370 | else | |
4371 | return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file, | |
4372 | filp, poll_table); | |
4373 | } | |
4374 | ||
4375 | static unsigned int | |
4376 | tracing_poll_pipe(struct file *filp, poll_table *poll_table) | |
4377 | { | |
4378 | struct trace_iterator *iter = filp->private_data; | |
4379 | ||
4380 | return trace_poll(iter, filp, poll_table); | |
4381 | } | |
4382 | ||
4383 | /* Must be called with trace_types_lock mutex held. */ | |
4384 | static int tracing_wait_pipe(struct file *filp) | |
4385 | { | |
4386 | struct trace_iterator *iter = filp->private_data; | |
4387 | int ret; | |
4388 | ||
4389 | while (trace_empty(iter)) { | |
4390 | ||
4391 | if ((filp->f_flags & O_NONBLOCK)) { | |
4392 | return -EAGAIN; | |
4393 | } | |
4394 | ||
4395 | /* | |
4396 | * We block until we read something and tracing is disabled. | |
4397 | * We still block if tracing is disabled, but we have never | |
4398 | * read anything. This allows a user to cat this file, and | |
4399 | * then enable tracing. But after we have read something, | |
4400 | * we give an EOF when tracing is again disabled. | |
4401 | * | |
4402 | * iter->pos will be 0 if we haven't read anything. | |
4403 | */ | |
4404 | if (!tracing_is_on() && iter->pos) | |
4405 | break; | |
4406 | ||
4407 | mutex_unlock(&iter->mutex); | |
4408 | ||
4409 | ret = wait_on_pipe(iter); | |
4410 | ||
4411 | mutex_lock(&iter->mutex); | |
4412 | ||
4413 | if (ret) | |
4414 | return ret; | |
4415 | ||
4416 | if (signal_pending(current)) | |
4417 | return -EINTR; | |
4418 | } | |
4419 | ||
4420 | return 1; | |
4421 | } | |
4422 | ||
4423 | /* | |
4424 | * Consumer reader. | |
4425 | */ | |
4426 | static ssize_t | |
4427 | tracing_read_pipe(struct file *filp, char __user *ubuf, | |
4428 | size_t cnt, loff_t *ppos) | |
4429 | { | |
4430 | struct trace_iterator *iter = filp->private_data; | |
4431 | struct trace_array *tr = iter->tr; | |
4432 | ssize_t sret; | |
4433 | ||
4434 | /* return any leftover data */ | |
4435 | sret = trace_seq_to_user(&iter->seq, ubuf, cnt); | |
4436 | if (sret != -EBUSY) | |
4437 | return sret; | |
4438 | ||
4439 | trace_seq_init(&iter->seq); | |
4440 | ||
4441 | /* copy the tracer to avoid using a global lock all around */ | |
4442 | mutex_lock(&trace_types_lock); | |
4443 | if (unlikely(iter->trace->name != tr->current_trace->name)) | |
4444 | *iter->trace = *tr->current_trace; | |
4445 | mutex_unlock(&trace_types_lock); | |
4446 | ||
4447 | /* | |
4448 | * Avoid more than one consumer on a single file descriptor | |
4449 | * This is just a matter of traces coherency, the ring buffer itself | |
4450 | * is protected. | |
4451 | */ | |
4452 | mutex_lock(&iter->mutex); | |
4453 | if (iter->trace->read) { | |
4454 | sret = iter->trace->read(iter, filp, ubuf, cnt, ppos); | |
4455 | if (sret) | |
4456 | goto out; | |
4457 | } | |
4458 | ||
4459 | waitagain: | |
4460 | sret = tracing_wait_pipe(filp); | |
4461 | if (sret <= 0) | |
4462 | goto out; | |
4463 | ||
4464 | /* stop when tracing is finished */ | |
4465 | if (trace_empty(iter)) { | |
4466 | sret = 0; | |
4467 | goto out; | |
4468 | } | |
4469 | ||
4470 | if (cnt >= PAGE_SIZE) | |
4471 | cnt = PAGE_SIZE - 1; | |
4472 | ||
4473 | /* reset all but tr, trace, and overruns */ | |
4474 | memset(&iter->seq, 0, | |
4475 | sizeof(struct trace_iterator) - | |
4476 | offsetof(struct trace_iterator, seq)); | |
4477 | cpumask_clear(iter->started); | |
4478 | iter->pos = -1; | |
4479 | ||
4480 | trace_event_read_lock(); | |
4481 | trace_access_lock(iter->cpu_file); | |
4482 | while (trace_find_next_entry_inc(iter) != NULL) { | |
4483 | enum print_line_t ret; | |
4484 | int len = iter->seq.len; | |
4485 | ||
4486 | ret = print_trace_line(iter); | |
4487 | if (ret == TRACE_TYPE_PARTIAL_LINE) { | |
4488 | /* don't print partial lines */ | |
4489 | iter->seq.len = len; | |
4490 | break; | |
4491 | } | |
4492 | if (ret != TRACE_TYPE_NO_CONSUME) | |
4493 | trace_consume(iter); | |
4494 | ||
4495 | if (iter->seq.len >= cnt) | |
4496 | break; | |
4497 | ||
4498 | /* | |
4499 | * Setting the full flag means we reached the trace_seq buffer | |
4500 | * size and we should leave by partial output condition above. | |
4501 | * One of the trace_seq_* functions is not used properly. | |
4502 | */ | |
4503 | WARN_ONCE(iter->seq.full, "full flag set for trace type %d", | |
4504 | iter->ent->type); | |
4505 | } | |
4506 | trace_access_unlock(iter->cpu_file); | |
4507 | trace_event_read_unlock(); | |
4508 | ||
4509 | /* Now copy what we have to the user */ | |
4510 | sret = trace_seq_to_user(&iter->seq, ubuf, cnt); | |
4511 | if (iter->seq.readpos >= iter->seq.len) | |
4512 | trace_seq_init(&iter->seq); | |
4513 | ||
4514 | /* | |
4515 | * If there was nothing to send to user, in spite of consuming trace | |
4516 | * entries, go back to wait for more entries. | |
4517 | */ | |
4518 | if (sret == -EBUSY) | |
4519 | goto waitagain; | |
4520 | ||
4521 | out: | |
4522 | mutex_unlock(&iter->mutex); | |
4523 | ||
4524 | return sret; | |
4525 | } | |
4526 | ||
4527 | static void tracing_spd_release_pipe(struct splice_pipe_desc *spd, | |
4528 | unsigned int idx) | |
4529 | { | |
4530 | __free_page(spd->pages[idx]); | |
4531 | } | |
4532 | ||
4533 | static const struct pipe_buf_operations tracing_pipe_buf_ops = { | |
4534 | .can_merge = 0, | |
4535 | .confirm = generic_pipe_buf_confirm, | |
4536 | .release = generic_pipe_buf_release, | |
4537 | .steal = generic_pipe_buf_steal, | |
4538 | .get = generic_pipe_buf_get, | |
4539 | }; | |
4540 | ||
4541 | static size_t | |
4542 | tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter) | |
4543 | { | |
4544 | size_t count; | |
4545 | int ret; | |
4546 | ||
4547 | /* Seq buffer is page-sized, exactly what we need. */ | |
4548 | for (;;) { | |
4549 | count = iter->seq.len; | |
4550 | ret = print_trace_line(iter); | |
4551 | count = iter->seq.len - count; | |
4552 | if (rem < count) { | |
4553 | rem = 0; | |
4554 | iter->seq.len -= count; | |
4555 | break; | |
4556 | } | |
4557 | if (ret == TRACE_TYPE_PARTIAL_LINE) { | |
4558 | iter->seq.len -= count; | |
4559 | break; | |
4560 | } | |
4561 | ||
4562 | if (ret != TRACE_TYPE_NO_CONSUME) | |
4563 | trace_consume(iter); | |
4564 | rem -= count; | |
4565 | if (!trace_find_next_entry_inc(iter)) { | |
4566 | rem = 0; | |
4567 | iter->ent = NULL; | |
4568 | break; | |
4569 | } | |
4570 | } | |
4571 | ||
4572 | return rem; | |
4573 | } | |
4574 | ||
4575 | static ssize_t tracing_splice_read_pipe(struct file *filp, | |
4576 | loff_t *ppos, | |
4577 | struct pipe_inode_info *pipe, | |
4578 | size_t len, | |
4579 | unsigned int flags) | |
4580 | { | |
4581 | struct page *pages_def[PIPE_DEF_BUFFERS]; | |
4582 | struct partial_page partial_def[PIPE_DEF_BUFFERS]; | |
4583 | struct trace_iterator *iter = filp->private_data; | |
4584 | struct splice_pipe_desc spd = { | |
4585 | .pages = pages_def, | |
4586 | .partial = partial_def, | |
4587 | .nr_pages = 0, /* This gets updated below. */ | |
4588 | .nr_pages_max = PIPE_DEF_BUFFERS, | |
4589 | .flags = flags, | |
4590 | .ops = &tracing_pipe_buf_ops, | |
4591 | .spd_release = tracing_spd_release_pipe, | |
4592 | }; | |
4593 | struct trace_array *tr = iter->tr; | |
4594 | ssize_t ret; | |
4595 | size_t rem; | |
4596 | unsigned int i; | |
4597 | ||
4598 | if (splice_grow_spd(pipe, &spd)) | |
4599 | return -ENOMEM; | |
4600 | ||
4601 | /* copy the tracer to avoid using a global lock all around */ | |
4602 | mutex_lock(&trace_types_lock); | |
4603 | if (unlikely(iter->trace->name != tr->current_trace->name)) | |
4604 | *iter->trace = *tr->current_trace; | |
4605 | mutex_unlock(&trace_types_lock); | |
4606 | ||
4607 | mutex_lock(&iter->mutex); | |
4608 | ||
4609 | if (iter->trace->splice_read) { | |
4610 | ret = iter->trace->splice_read(iter, filp, | |
4611 | ppos, pipe, len, flags); | |
4612 | if (ret) | |
4613 | goto out_err; | |
4614 | } | |
4615 | ||
4616 | ret = tracing_wait_pipe(filp); | |
4617 | if (ret <= 0) | |
4618 | goto out_err; | |
4619 | ||
4620 | if (!iter->ent && !trace_find_next_entry_inc(iter)) { | |
4621 | ret = -EFAULT; | |
4622 | goto out_err; | |
4623 | } | |
4624 | ||
4625 | trace_event_read_lock(); | |
4626 | trace_access_lock(iter->cpu_file); | |
4627 | ||
4628 | /* Fill as many pages as possible. */ | |
4629 | for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) { | |
4630 | spd.pages[i] = alloc_page(GFP_KERNEL); | |
4631 | if (!spd.pages[i]) | |
4632 | break; | |
4633 | ||
4634 | rem = tracing_fill_pipe_page(rem, iter); | |
4635 | ||
4636 | /* Copy the data into the page, so we can start over. */ | |
4637 | ret = trace_seq_to_buffer(&iter->seq, | |
4638 | page_address(spd.pages[i]), | |
4639 | iter->seq.len); | |
4640 | if (ret < 0) { | |
4641 | __free_page(spd.pages[i]); | |
4642 | break; | |
4643 | } | |
4644 | spd.partial[i].offset = 0; | |
4645 | spd.partial[i].len = iter->seq.len; | |
4646 | ||
4647 | trace_seq_init(&iter->seq); | |
4648 | } | |
4649 | ||
4650 | trace_access_unlock(iter->cpu_file); | |
4651 | trace_event_read_unlock(); | |
4652 | mutex_unlock(&iter->mutex); | |
4653 | ||
4654 | spd.nr_pages = i; | |
4655 | ||
4656 | ret = splice_to_pipe(pipe, &spd); | |
4657 | out: | |
4658 | splice_shrink_spd(&spd); | |
4659 | return ret; | |
4660 | ||
4661 | out_err: | |
4662 | mutex_unlock(&iter->mutex); | |
4663 | goto out; | |
4664 | } | |
4665 | ||
4666 | static ssize_t | |
4667 | tracing_entries_read(struct file *filp, char __user *ubuf, | |
4668 | size_t cnt, loff_t *ppos) | |
4669 | { | |
4670 | struct inode *inode = file_inode(filp); | |
4671 | struct trace_array *tr = inode->i_private; | |
4672 | int cpu = tracing_get_cpu(inode); | |
4673 | char buf[64]; | |
4674 | int r = 0; | |
4675 | ssize_t ret; | |
4676 | ||
4677 | mutex_lock(&trace_types_lock); | |
4678 | ||
4679 | if (cpu == RING_BUFFER_ALL_CPUS) { | |
4680 | int cpu, buf_size_same; | |
4681 | unsigned long size; | |
4682 | ||
4683 | size = 0; | |
4684 | buf_size_same = 1; | |
4685 | /* check if all cpu sizes are same */ | |
4686 | for_each_tracing_cpu(cpu) { | |
4687 | /* fill in the size from first enabled cpu */ | |
4688 | if (size == 0) | |
4689 | size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries; | |
4690 | if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) { | |
4691 | buf_size_same = 0; | |
4692 | break; | |
4693 | } | |
4694 | } | |
4695 | ||
4696 | if (buf_size_same) { | |
4697 | if (!ring_buffer_expanded) | |
4698 | r = sprintf(buf, "%lu (expanded: %lu)\n", | |
4699 | size >> 10, | |
4700 | trace_buf_size >> 10); | |
4701 | else | |
4702 | r = sprintf(buf, "%lu\n", size >> 10); | |
4703 | } else | |
4704 | r = sprintf(buf, "X\n"); | |
4705 | } else | |
4706 | r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10); | |
4707 | ||
4708 | mutex_unlock(&trace_types_lock); | |
4709 | ||
4710 | ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
4711 | return ret; | |
4712 | } | |
4713 | ||
4714 | static ssize_t | |
4715 | tracing_entries_write(struct file *filp, const char __user *ubuf, | |
4716 | size_t cnt, loff_t *ppos) | |
4717 | { | |
4718 | struct inode *inode = file_inode(filp); | |
4719 | struct trace_array *tr = inode->i_private; | |
4720 | unsigned long val; | |
4721 | int ret; | |
4722 | ||
4723 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
4724 | if (ret) | |
4725 | return ret; | |
4726 | ||
4727 | /* must have at least 1 entry */ | |
4728 | if (!val) | |
4729 | return -EINVAL; | |
4730 | ||
4731 | /* value is in KB */ | |
4732 | val <<= 10; | |
4733 | ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode)); | |
4734 | if (ret < 0) | |
4735 | return ret; | |
4736 | ||
4737 | *ppos += cnt; | |
4738 | ||
4739 | return cnt; | |
4740 | } | |
4741 | ||
4742 | static ssize_t | |
4743 | tracing_total_entries_read(struct file *filp, char __user *ubuf, | |
4744 | size_t cnt, loff_t *ppos) | |
4745 | { | |
4746 | struct trace_array *tr = filp->private_data; | |
4747 | char buf[64]; | |
4748 | int r, cpu; | |
4749 | unsigned long size = 0, expanded_size = 0; | |
4750 | ||
4751 | mutex_lock(&trace_types_lock); | |
4752 | for_each_tracing_cpu(cpu) { | |
4753 | size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10; | |
4754 | if (!ring_buffer_expanded) | |
4755 | expanded_size += trace_buf_size >> 10; | |
4756 | } | |
4757 | if (ring_buffer_expanded) | |
4758 | r = sprintf(buf, "%lu\n", size); | |
4759 | else | |
4760 | r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size); | |
4761 | mutex_unlock(&trace_types_lock); | |
4762 | ||
4763 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
4764 | } | |
4765 | ||
4766 | static ssize_t | |
4767 | tracing_free_buffer_write(struct file *filp, const char __user *ubuf, | |
4768 | size_t cnt, loff_t *ppos) | |
4769 | { | |
4770 | /* | |
4771 | * There is no need to read what the user has written, this function | |
4772 | * is just to make sure that there is no error when "echo" is used | |
4773 | */ | |
4774 | ||
4775 | *ppos += cnt; | |
4776 | ||
4777 | return cnt; | |
4778 | } | |
4779 | ||
4780 | static int | |
4781 | tracing_free_buffer_release(struct inode *inode, struct file *filp) | |
4782 | { | |
4783 | struct trace_array *tr = inode->i_private; | |
4784 | ||
4785 | /* disable tracing ? */ | |
4786 | if (trace_flags & TRACE_ITER_STOP_ON_FREE) | |
4787 | tracer_tracing_off(tr); | |
4788 | /* resize the ring buffer to 0 */ | |
4789 | tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS); | |
4790 | ||
4791 | trace_array_put(tr); | |
4792 | ||
4793 | return 0; | |
4794 | } | |
4795 | ||
4796 | static ssize_t | |
4797 | tracing_mark_write(struct file *filp, const char __user *ubuf, | |
4798 | size_t cnt, loff_t *fpos) | |
4799 | { | |
4800 | unsigned long addr = (unsigned long)ubuf; | |
4801 | struct trace_array *tr = filp->private_data; | |
4802 | struct ring_buffer_event *event; | |
4803 | struct ring_buffer *buffer; | |
4804 | struct print_entry *entry; | |
4805 | unsigned long irq_flags; | |
4806 | struct page *pages[2]; | |
4807 | void *map_page[2]; | |
4808 | int nr_pages = 1; | |
4809 | ssize_t written; | |
4810 | int offset; | |
4811 | int size; | |
4812 | int len; | |
4813 | int ret; | |
4814 | int i; | |
4815 | ||
4816 | if (tracing_disabled) | |
4817 | return -EINVAL; | |
4818 | ||
4819 | if (!(trace_flags & TRACE_ITER_MARKERS)) | |
4820 | return -EINVAL; | |
4821 | ||
4822 | if (cnt > TRACE_BUF_SIZE) | |
4823 | cnt = TRACE_BUF_SIZE; | |
4824 | ||
4825 | /* | |
4826 | * Userspace is injecting traces into the kernel trace buffer. | |
4827 | * We want to be as non intrusive as possible. | |
4828 | * To do so, we do not want to allocate any special buffers | |
4829 | * or take any locks, but instead write the userspace data | |
4830 | * straight into the ring buffer. | |
4831 | * | |
4832 | * First we need to pin the userspace buffer into memory, | |
4833 | * which, most likely it is, because it just referenced it. | |
4834 | * But there's no guarantee that it is. By using get_user_pages_fast() | |
4835 | * and kmap_atomic/kunmap_atomic() we can get access to the | |
4836 | * pages directly. We then write the data directly into the | |
4837 | * ring buffer. | |
4838 | */ | |
4839 | BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE); | |
4840 | ||
4841 | /* check if we cross pages */ | |
4842 | if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK)) | |
4843 | nr_pages = 2; | |
4844 | ||
4845 | offset = addr & (PAGE_SIZE - 1); | |
4846 | addr &= PAGE_MASK; | |
4847 | ||
4848 | ret = get_user_pages_fast(addr, nr_pages, 0, pages); | |
4849 | if (ret < nr_pages) { | |
4850 | while (--ret >= 0) | |
4851 | put_page(pages[ret]); | |
4852 | written = -EFAULT; | |
4853 | goto out; | |
4854 | } | |
4855 | ||
4856 | for (i = 0; i < nr_pages; i++) | |
4857 | map_page[i] = kmap_atomic(pages[i]); | |
4858 | ||
4859 | local_save_flags(irq_flags); | |
4860 | size = sizeof(*entry) + cnt + 2; /* possible \n added */ | |
4861 | buffer = tr->trace_buffer.buffer; | |
4862 | event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size, | |
4863 | irq_flags, preempt_count()); | |
4864 | if (!event) { | |
4865 | /* Ring buffer disabled, return as if not open for write */ | |
4866 | written = -EBADF; | |
4867 | goto out_unlock; | |
4868 | } | |
4869 | ||
4870 | entry = ring_buffer_event_data(event); | |
4871 | entry->ip = _THIS_IP_; | |
4872 | ||
4873 | if (nr_pages == 2) { | |
4874 | len = PAGE_SIZE - offset; | |
4875 | memcpy(&entry->buf, map_page[0] + offset, len); | |
4876 | memcpy(&entry->buf[len], map_page[1], cnt - len); | |
4877 | } else | |
4878 | memcpy(&entry->buf, map_page[0] + offset, cnt); | |
4879 | ||
4880 | if (entry->buf[cnt - 1] != '\n') { | |
4881 | entry->buf[cnt] = '\n'; | |
4882 | entry->buf[cnt + 1] = '\0'; | |
4883 | } else | |
4884 | entry->buf[cnt] = '\0'; | |
4885 | ||
4886 | __buffer_unlock_commit(buffer, event); | |
4887 | ||
4888 | written = cnt; | |
4889 | ||
4890 | *fpos += written; | |
4891 | ||
4892 | out_unlock: | |
4893 | for (i = 0; i < nr_pages; i++){ | |
4894 | kunmap_atomic(map_page[i]); | |
4895 | put_page(pages[i]); | |
4896 | } | |
4897 | out: | |
4898 | return written; | |
4899 | } | |
4900 | ||
4901 | static int tracing_clock_show(struct seq_file *m, void *v) | |
4902 | { | |
4903 | struct trace_array *tr = m->private; | |
4904 | int i; | |
4905 | ||
4906 | for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) | |
4907 | seq_printf(m, | |
4908 | "%s%s%s%s", i ? " " : "", | |
4909 | i == tr->clock_id ? "[" : "", trace_clocks[i].name, | |
4910 | i == tr->clock_id ? "]" : ""); | |
4911 | seq_putc(m, '\n'); | |
4912 | ||
4913 | return 0; | |
4914 | } | |
4915 | ||
4916 | static int tracing_set_clock(struct trace_array *tr, const char *clockstr) | |
4917 | { | |
4918 | int i; | |
4919 | ||
4920 | for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) { | |
4921 | if (strcmp(trace_clocks[i].name, clockstr) == 0) | |
4922 | break; | |
4923 | } | |
4924 | if (i == ARRAY_SIZE(trace_clocks)) | |
4925 | return -EINVAL; | |
4926 | ||
4927 | mutex_lock(&trace_types_lock); | |
4928 | ||
4929 | tr->clock_id = i; | |
4930 | ||
4931 | ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func); | |
4932 | ||
4933 | /* | |
4934 | * New clock may not be consistent with the previous clock. | |
4935 | * Reset the buffer so that it doesn't have incomparable timestamps. | |
4936 | */ | |
4937 | tracing_reset_online_cpus(&tr->trace_buffer); | |
4938 | ||
4939 | #ifdef CONFIG_TRACER_MAX_TRACE | |
4940 | if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer) | |
4941 | ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func); | |
4942 | tracing_reset_online_cpus(&tr->max_buffer); | |
4943 | #endif | |
4944 | ||
4945 | mutex_unlock(&trace_types_lock); | |
4946 | ||
4947 | return 0; | |
4948 | } | |
4949 | ||
4950 | static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf, | |
4951 | size_t cnt, loff_t *fpos) | |
4952 | { | |
4953 | struct seq_file *m = filp->private_data; | |
4954 | struct trace_array *tr = m->private; | |
4955 | char buf[64]; | |
4956 | const char *clockstr; | |
4957 | int ret; | |
4958 | ||
4959 | if (cnt >= sizeof(buf)) | |
4960 | return -EINVAL; | |
4961 | ||
4962 | if (copy_from_user(&buf, ubuf, cnt)) | |
4963 | return -EFAULT; | |
4964 | ||
4965 | buf[cnt] = 0; | |
4966 | ||
4967 | clockstr = strstrip(buf); | |
4968 | ||
4969 | ret = tracing_set_clock(tr, clockstr); | |
4970 | if (ret) | |
4971 | return ret; | |
4972 | ||
4973 | *fpos += cnt; | |
4974 | ||
4975 | return cnt; | |
4976 | } | |
4977 | ||
4978 | static int tracing_clock_open(struct inode *inode, struct file *file) | |
4979 | { | |
4980 | struct trace_array *tr = inode->i_private; | |
4981 | int ret; | |
4982 | ||
4983 | if (tracing_disabled) | |
4984 | return -ENODEV; | |
4985 | ||
4986 | if (trace_array_get(tr)) | |
4987 | return -ENODEV; | |
4988 | ||
4989 | ret = single_open(file, tracing_clock_show, inode->i_private); | |
4990 | if (ret < 0) | |
4991 | trace_array_put(tr); | |
4992 | ||
4993 | return ret; | |
4994 | } | |
4995 | ||
4996 | struct ftrace_buffer_info { | |
4997 | struct trace_iterator iter; | |
4998 | void *spare; | |
4999 | unsigned int read; | |
5000 | }; | |
5001 | ||
5002 | #ifdef CONFIG_TRACER_SNAPSHOT | |
5003 | static int tracing_snapshot_open(struct inode *inode, struct file *file) | |
5004 | { | |
5005 | struct trace_array *tr = inode->i_private; | |
5006 | struct trace_iterator *iter; | |
5007 | struct seq_file *m; | |
5008 | int ret = 0; | |
5009 | ||
5010 | if (trace_array_get(tr) < 0) | |
5011 | return -ENODEV; | |
5012 | ||
5013 | if (file->f_mode & FMODE_READ) { | |
5014 | iter = __tracing_open(inode, file, true); | |
5015 | if (IS_ERR(iter)) | |
5016 | ret = PTR_ERR(iter); | |
5017 | } else { | |
5018 | /* Writes still need the seq_file to hold the private data */ | |
5019 | ret = -ENOMEM; | |
5020 | m = kzalloc(sizeof(*m), GFP_KERNEL); | |
5021 | if (!m) | |
5022 | goto out; | |
5023 | iter = kzalloc(sizeof(*iter), GFP_KERNEL); | |
5024 | if (!iter) { | |
5025 | kfree(m); | |
5026 | goto out; | |
5027 | } | |
5028 | ret = 0; | |
5029 | ||
5030 | iter->tr = tr; | |
5031 | iter->trace_buffer = &tr->max_buffer; | |
5032 | iter->cpu_file = tracing_get_cpu(inode); | |
5033 | m->private = iter; | |
5034 | file->private_data = m; | |
5035 | } | |
5036 | out: | |
5037 | if (ret < 0) | |
5038 | trace_array_put(tr); | |
5039 | ||
5040 | return ret; | |
5041 | } | |
5042 | ||
5043 | static ssize_t | |
5044 | tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt, | |
5045 | loff_t *ppos) | |
5046 | { | |
5047 | struct seq_file *m = filp->private_data; | |
5048 | struct trace_iterator *iter = m->private; | |
5049 | struct trace_array *tr = iter->tr; | |
5050 | unsigned long val; | |
5051 | int ret; | |
5052 | ||
5053 | ret = tracing_update_buffers(); | |
5054 | if (ret < 0) | |
5055 | return ret; | |
5056 | ||
5057 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
5058 | if (ret) | |
5059 | return ret; | |
5060 | ||
5061 | mutex_lock(&trace_types_lock); | |
5062 | ||
5063 | if (tr->current_trace->use_max_tr) { | |
5064 | ret = -EBUSY; | |
5065 | goto out; | |
5066 | } | |
5067 | ||
5068 | switch (val) { | |
5069 | case 0: | |
5070 | if (iter->cpu_file != RING_BUFFER_ALL_CPUS) { | |
5071 | ret = -EINVAL; | |
5072 | break; | |
5073 | } | |
5074 | if (tr->allocated_snapshot) | |
5075 | free_snapshot(tr); | |
5076 | break; | |
5077 | case 1: | |
5078 | /* Only allow per-cpu swap if the ring buffer supports it */ | |
5079 | #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP | |
5080 | if (iter->cpu_file != RING_BUFFER_ALL_CPUS) { | |
5081 | ret = -EINVAL; | |
5082 | break; | |
5083 | } | |
5084 | #endif | |
5085 | if (!tr->allocated_snapshot) { | |
5086 | ret = alloc_snapshot(tr); | |
5087 | if (ret < 0) | |
5088 | break; | |
5089 | } | |
5090 | local_irq_disable(); | |
5091 | /* Now, we're going to swap */ | |
5092 | if (iter->cpu_file == RING_BUFFER_ALL_CPUS) | |
5093 | update_max_tr(tr, current, smp_processor_id()); | |
5094 | else | |
5095 | update_max_tr_single(tr, current, iter->cpu_file); | |
5096 | local_irq_enable(); | |
5097 | break; | |
5098 | default: | |
5099 | if (tr->allocated_snapshot) { | |
5100 | if (iter->cpu_file == RING_BUFFER_ALL_CPUS) | |
5101 | tracing_reset_online_cpus(&tr->max_buffer); | |
5102 | else | |
5103 | tracing_reset(&tr->max_buffer, iter->cpu_file); | |
5104 | } | |
5105 | break; | |
5106 | } | |
5107 | ||
5108 | if (ret >= 0) { | |
5109 | *ppos += cnt; | |
5110 | ret = cnt; | |
5111 | } | |
5112 | out: | |
5113 | mutex_unlock(&trace_types_lock); | |
5114 | return ret; | |
5115 | } | |
5116 | ||
5117 | static int tracing_snapshot_release(struct inode *inode, struct file *file) | |
5118 | { | |
5119 | struct seq_file *m = file->private_data; | |
5120 | int ret; | |
5121 | ||
5122 | ret = tracing_release(inode, file); | |
5123 | ||
5124 | if (file->f_mode & FMODE_READ) | |
5125 | return ret; | |
5126 | ||
5127 | /* If write only, the seq_file is just a stub */ | |
5128 | if (m) | |
5129 | kfree(m->private); | |
5130 | kfree(m); | |
5131 | ||
5132 | return 0; | |
5133 | } | |
5134 | ||
5135 | static int tracing_buffers_open(struct inode *inode, struct file *filp); | |
5136 | static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf, | |
5137 | size_t count, loff_t *ppos); | |
5138 | static int tracing_buffers_release(struct inode *inode, struct file *file); | |
5139 | static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos, | |
5140 | struct pipe_inode_info *pipe, size_t len, unsigned int flags); | |
5141 | ||
5142 | static int snapshot_raw_open(struct inode *inode, struct file *filp) | |
5143 | { | |
5144 | struct ftrace_buffer_info *info; | |
5145 | int ret; | |
5146 | ||
5147 | ret = tracing_buffers_open(inode, filp); | |
5148 | if (ret < 0) | |
5149 | return ret; | |
5150 | ||
5151 | info = filp->private_data; | |
5152 | ||
5153 | if (info->iter.trace->use_max_tr) { | |
5154 | tracing_buffers_release(inode, filp); | |
5155 | return -EBUSY; | |
5156 | } | |
5157 | ||
5158 | info->iter.snapshot = true; | |
5159 | info->iter.trace_buffer = &info->iter.tr->max_buffer; | |
5160 | ||
5161 | return ret; | |
5162 | } | |
5163 | ||
5164 | #endif /* CONFIG_TRACER_SNAPSHOT */ | |
5165 | ||
5166 | ||
5167 | static const struct file_operations tracing_max_lat_fops = { | |
5168 | .open = tracing_open_generic, | |
5169 | .read = tracing_max_lat_read, | |
5170 | .write = tracing_max_lat_write, | |
5171 | .llseek = generic_file_llseek, | |
5172 | }; | |
5173 | ||
5174 | static const struct file_operations set_tracer_fops = { | |
5175 | .open = tracing_open_generic, | |
5176 | .read = tracing_set_trace_read, | |
5177 | .write = tracing_set_trace_write, | |
5178 | .llseek = generic_file_llseek, | |
5179 | }; | |
5180 | ||
5181 | static const struct file_operations tracing_pipe_fops = { | |
5182 | .open = tracing_open_pipe, | |
5183 | .poll = tracing_poll_pipe, | |
5184 | .read = tracing_read_pipe, | |
5185 | .splice_read = tracing_splice_read_pipe, | |
5186 | .release = tracing_release_pipe, | |
5187 | .llseek = no_llseek, | |
5188 | }; | |
5189 | ||
5190 | static const struct file_operations tracing_entries_fops = { | |
5191 | .open = tracing_open_generic_tr, | |
5192 | .read = tracing_entries_read, | |
5193 | .write = tracing_entries_write, | |
5194 | .llseek = generic_file_llseek, | |
5195 | .release = tracing_release_generic_tr, | |
5196 | }; | |
5197 | ||
5198 | static const struct file_operations tracing_total_entries_fops = { | |
5199 | .open = tracing_open_generic_tr, | |
5200 | .read = tracing_total_entries_read, | |
5201 | .llseek = generic_file_llseek, | |
5202 | .release = tracing_release_generic_tr, | |
5203 | }; | |
5204 | ||
5205 | static const struct file_operations tracing_free_buffer_fops = { | |
5206 | .open = tracing_open_generic_tr, | |
5207 | .write = tracing_free_buffer_write, | |
5208 | .release = tracing_free_buffer_release, | |
5209 | }; | |
5210 | ||
5211 | static const struct file_operations tracing_mark_fops = { | |
5212 | .open = tracing_open_generic_tr, | |
5213 | .write = tracing_mark_write, | |
5214 | .llseek = generic_file_llseek, | |
5215 | .release = tracing_release_generic_tr, | |
5216 | }; | |
5217 | ||
5218 | static const struct file_operations trace_clock_fops = { | |
5219 | .open = tracing_clock_open, | |
5220 | .read = seq_read, | |
5221 | .llseek = seq_lseek, | |
5222 | .release = tracing_single_release_tr, | |
5223 | .write = tracing_clock_write, | |
5224 | }; | |
5225 | ||
5226 | #ifdef CONFIG_TRACER_SNAPSHOT | |
5227 | static const struct file_operations snapshot_fops = { | |
5228 | .open = tracing_snapshot_open, | |
5229 | .read = seq_read, | |
5230 | .write = tracing_snapshot_write, | |
5231 | .llseek = tracing_lseek, | |
5232 | .release = tracing_snapshot_release, | |
5233 | }; | |
5234 | ||
5235 | static const struct file_operations snapshot_raw_fops = { | |
5236 | .open = snapshot_raw_open, | |
5237 | .read = tracing_buffers_read, | |
5238 | .release = tracing_buffers_release, | |
5239 | .splice_read = tracing_buffers_splice_read, | |
5240 | .llseek = no_llseek, | |
5241 | }; | |
5242 | ||
5243 | #endif /* CONFIG_TRACER_SNAPSHOT */ | |
5244 | ||
5245 | static int tracing_buffers_open(struct inode *inode, struct file *filp) | |
5246 | { | |
5247 | struct trace_array *tr = inode->i_private; | |
5248 | struct ftrace_buffer_info *info; | |
5249 | int ret; | |
5250 | ||
5251 | if (tracing_disabled) | |
5252 | return -ENODEV; | |
5253 | ||
5254 | if (trace_array_get(tr) < 0) | |
5255 | return -ENODEV; | |
5256 | ||
5257 | info = kzalloc(sizeof(*info), GFP_KERNEL); | |
5258 | if (!info) { | |
5259 | trace_array_put(tr); | |
5260 | return -ENOMEM; | |
5261 | } | |
5262 | ||
5263 | mutex_lock(&trace_types_lock); | |
5264 | ||
5265 | info->iter.tr = tr; | |
5266 | info->iter.cpu_file = tracing_get_cpu(inode); | |
5267 | info->iter.trace = tr->current_trace; | |
5268 | info->iter.trace_buffer = &tr->trace_buffer; | |
5269 | info->spare = NULL; | |
5270 | /* Force reading ring buffer for first read */ | |
5271 | info->read = (unsigned int)-1; | |
5272 | ||
5273 | filp->private_data = info; | |
5274 | ||
5275 | mutex_unlock(&trace_types_lock); | |
5276 | ||
5277 | ret = nonseekable_open(inode, filp); | |
5278 | if (ret < 0) | |
5279 | trace_array_put(tr); | |
5280 | ||
5281 | return ret; | |
5282 | } | |
5283 | ||
5284 | static unsigned int | |
5285 | tracing_buffers_poll(struct file *filp, poll_table *poll_table) | |
5286 | { | |
5287 | struct ftrace_buffer_info *info = filp->private_data; | |
5288 | struct trace_iterator *iter = &info->iter; | |
5289 | ||
5290 | return trace_poll(iter, filp, poll_table); | |
5291 | } | |
5292 | ||
5293 | static ssize_t | |
5294 | tracing_buffers_read(struct file *filp, char __user *ubuf, | |
5295 | size_t count, loff_t *ppos) | |
5296 | { | |
5297 | struct ftrace_buffer_info *info = filp->private_data; | |
5298 | struct trace_iterator *iter = &info->iter; | |
5299 | ssize_t ret; | |
5300 | ssize_t size; | |
5301 | ||
5302 | if (!count) | |
5303 | return 0; | |
5304 | ||
5305 | mutex_lock(&trace_types_lock); | |
5306 | ||
5307 | #ifdef CONFIG_TRACER_MAX_TRACE | |
5308 | if (iter->snapshot && iter->tr->current_trace->use_max_tr) { | |
5309 | size = -EBUSY; | |
5310 | goto out_unlock; | |
5311 | } | |
5312 | #endif | |
5313 | ||
5314 | if (!info->spare) | |
5315 | info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer, | |
5316 | iter->cpu_file); | |
5317 | size = -ENOMEM; | |
5318 | if (!info->spare) | |
5319 | goto out_unlock; | |
5320 | ||
5321 | /* Do we have previous read data to read? */ | |
5322 | if (info->read < PAGE_SIZE) | |
5323 | goto read; | |
5324 | ||
5325 | again: | |
5326 | trace_access_lock(iter->cpu_file); | |
5327 | ret = ring_buffer_read_page(iter->trace_buffer->buffer, | |
5328 | &info->spare, | |
5329 | count, | |
5330 | iter->cpu_file, 0); | |
5331 | trace_access_unlock(iter->cpu_file); | |
5332 | ||
5333 | if (ret < 0) { | |
5334 | if (trace_empty(iter)) { | |
5335 | if ((filp->f_flags & O_NONBLOCK)) { | |
5336 | size = -EAGAIN; | |
5337 | goto out_unlock; | |
5338 | } | |
5339 | mutex_unlock(&trace_types_lock); | |
5340 | ret = wait_on_pipe(iter); | |
5341 | mutex_lock(&trace_types_lock); | |
5342 | if (ret) { | |
5343 | size = ret; | |
5344 | goto out_unlock; | |
5345 | } | |
5346 | if (signal_pending(current)) { | |
5347 | size = -EINTR; | |
5348 | goto out_unlock; | |
5349 | } | |
5350 | goto again; | |
5351 | } | |
5352 | size = 0; | |
5353 | goto out_unlock; | |
5354 | } | |
5355 | ||
5356 | info->read = 0; | |
5357 | read: | |
5358 | size = PAGE_SIZE - info->read; | |
5359 | if (size > count) | |
5360 | size = count; | |
5361 | ||
5362 | ret = copy_to_user(ubuf, info->spare + info->read, size); | |
5363 | if (ret == size) { | |
5364 | size = -EFAULT; | |
5365 | goto out_unlock; | |
5366 | } | |
5367 | size -= ret; | |
5368 | ||
5369 | *ppos += size; | |
5370 | info->read += size; | |
5371 | ||
5372 | out_unlock: | |
5373 | mutex_unlock(&trace_types_lock); | |
5374 | ||
5375 | return size; | |
5376 | } | |
5377 | ||
5378 | static int tracing_buffers_release(struct inode *inode, struct file *file) | |
5379 | { | |
5380 | struct ftrace_buffer_info *info = file->private_data; | |
5381 | struct trace_iterator *iter = &info->iter; | |
5382 | ||
5383 | mutex_lock(&trace_types_lock); | |
5384 | ||
5385 | __trace_array_put(iter->tr); | |
5386 | ||
5387 | if (info->spare) | |
5388 | ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare); | |
5389 | kfree(info); | |
5390 | ||
5391 | mutex_unlock(&trace_types_lock); | |
5392 | ||
5393 | return 0; | |
5394 | } | |
5395 | ||
5396 | struct buffer_ref { | |
5397 | struct ring_buffer *buffer; | |
5398 | void *page; | |
5399 | int ref; | |
5400 | }; | |
5401 | ||
5402 | static void buffer_pipe_buf_release(struct pipe_inode_info *pipe, | |
5403 | struct pipe_buffer *buf) | |
5404 | { | |
5405 | struct buffer_ref *ref = (struct buffer_ref *)buf->private; | |
5406 | ||
5407 | if (--ref->ref) | |
5408 | return; | |
5409 | ||
5410 | ring_buffer_free_read_page(ref->buffer, ref->page); | |
5411 | kfree(ref); | |
5412 | buf->private = 0; | |
5413 | } | |
5414 | ||
5415 | static void buffer_pipe_buf_get(struct pipe_inode_info *pipe, | |
5416 | struct pipe_buffer *buf) | |
5417 | { | |
5418 | struct buffer_ref *ref = (struct buffer_ref *)buf->private; | |
5419 | ||
5420 | ref->ref++; | |
5421 | } | |
5422 | ||
5423 | /* Pipe buffer operations for a buffer. */ | |
5424 | static const struct pipe_buf_operations buffer_pipe_buf_ops = { | |
5425 | .can_merge = 0, | |
5426 | .confirm = generic_pipe_buf_confirm, | |
5427 | .release = buffer_pipe_buf_release, | |
5428 | .steal = generic_pipe_buf_steal, | |
5429 | .get = buffer_pipe_buf_get, | |
5430 | }; | |
5431 | ||
5432 | /* | |
5433 | * Callback from splice_to_pipe(), if we need to release some pages | |
5434 | * at the end of the spd in case we error'ed out in filling the pipe. | |
5435 | */ | |
5436 | static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i) | |
5437 | { | |
5438 | struct buffer_ref *ref = | |
5439 | (struct buffer_ref *)spd->partial[i].private; | |
5440 | ||
5441 | if (--ref->ref) | |
5442 | return; | |
5443 | ||
5444 | ring_buffer_free_read_page(ref->buffer, ref->page); | |
5445 | kfree(ref); | |
5446 | spd->partial[i].private = 0; | |
5447 | } | |
5448 | ||
5449 | static ssize_t | |
5450 | tracing_buffers_splice_read(struct file *file, loff_t *ppos, | |
5451 | struct pipe_inode_info *pipe, size_t len, | |
5452 | unsigned int flags) | |
5453 | { | |
5454 | struct ftrace_buffer_info *info = file->private_data; | |
5455 | struct trace_iterator *iter = &info->iter; | |
5456 | struct partial_page partial_def[PIPE_DEF_BUFFERS]; | |
5457 | struct page *pages_def[PIPE_DEF_BUFFERS]; | |
5458 | struct splice_pipe_desc spd = { | |
5459 | .pages = pages_def, | |
5460 | .partial = partial_def, | |
5461 | .nr_pages_max = PIPE_DEF_BUFFERS, | |
5462 | .flags = flags, | |
5463 | .ops = &buffer_pipe_buf_ops, | |
5464 | .spd_release = buffer_spd_release, | |
5465 | }; | |
5466 | struct buffer_ref *ref; | |
5467 | int entries, size, i; | |
5468 | ssize_t ret; | |
5469 | ||
5470 | mutex_lock(&trace_types_lock); | |
5471 | ||
5472 | #ifdef CONFIG_TRACER_MAX_TRACE | |
5473 | if (iter->snapshot && iter->tr->current_trace->use_max_tr) { | |
5474 | ret = -EBUSY; | |
5475 | goto out; | |
5476 | } | |
5477 | #endif | |
5478 | ||
5479 | if (splice_grow_spd(pipe, &spd)) { | |
5480 | ret = -ENOMEM; | |
5481 | goto out; | |
5482 | } | |
5483 | ||
5484 | if (*ppos & (PAGE_SIZE - 1)) { | |
5485 | ret = -EINVAL; | |
5486 | goto out; | |
5487 | } | |
5488 | ||
5489 | if (len & (PAGE_SIZE - 1)) { | |
5490 | if (len < PAGE_SIZE) { | |
5491 | ret = -EINVAL; | |
5492 | goto out; | |
5493 | } | |
5494 | len &= PAGE_MASK; | |
5495 | } | |
5496 | ||
5497 | again: | |
5498 | trace_access_lock(iter->cpu_file); | |
5499 | entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file); | |
5500 | ||
5501 | for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) { | |
5502 | struct page *page; | |
5503 | int r; | |
5504 | ||
5505 | ref = kzalloc(sizeof(*ref), GFP_KERNEL); | |
5506 | if (!ref) | |
5507 | break; | |
5508 | ||
5509 | ref->ref = 1; | |
5510 | ref->buffer = iter->trace_buffer->buffer; | |
5511 | ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file); | |
5512 | if (!ref->page) { | |
5513 | kfree(ref); | |
5514 | break; | |
5515 | } | |
5516 | ||
5517 | r = ring_buffer_read_page(ref->buffer, &ref->page, | |
5518 | len, iter->cpu_file, 1); | |
5519 | if (r < 0) { | |
5520 | ring_buffer_free_read_page(ref->buffer, ref->page); | |
5521 | kfree(ref); | |
5522 | break; | |
5523 | } | |
5524 | ||
5525 | /* | |
5526 | * zero out any left over data, this is going to | |
5527 | * user land. | |
5528 | */ | |
5529 | size = ring_buffer_page_len(ref->page); | |
5530 | if (size < PAGE_SIZE) | |
5531 | memset(ref->page + size, 0, PAGE_SIZE - size); | |
5532 | ||
5533 | page = virt_to_page(ref->page); | |
5534 | ||
5535 | spd.pages[i] = page; | |
5536 | spd.partial[i].len = PAGE_SIZE; | |
5537 | spd.partial[i].offset = 0; | |
5538 | spd.partial[i].private = (unsigned long)ref; | |
5539 | spd.nr_pages++; | |
5540 | *ppos += PAGE_SIZE; | |
5541 | ||
5542 | entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file); | |
5543 | } | |
5544 | ||
5545 | trace_access_unlock(iter->cpu_file); | |
5546 | spd.nr_pages = i; | |
5547 | ||
5548 | /* did we read anything? */ | |
5549 | if (!spd.nr_pages) { | |
5550 | if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) { | |
5551 | ret = -EAGAIN; | |
5552 | goto out; | |
5553 | } | |
5554 | mutex_unlock(&trace_types_lock); | |
5555 | ret = wait_on_pipe(iter); | |
5556 | mutex_lock(&trace_types_lock); | |
5557 | if (ret) | |
5558 | goto out; | |
5559 | if (signal_pending(current)) { | |
5560 | ret = -EINTR; | |
5561 | goto out; | |
5562 | } | |
5563 | goto again; | |
5564 | } | |
5565 | ||
5566 | ret = splice_to_pipe(pipe, &spd); | |
5567 | splice_shrink_spd(&spd); | |
5568 | out: | |
5569 | mutex_unlock(&trace_types_lock); | |
5570 | ||
5571 | return ret; | |
5572 | } | |
5573 | ||
5574 | static const struct file_operations tracing_buffers_fops = { | |
5575 | .open = tracing_buffers_open, | |
5576 | .read = tracing_buffers_read, | |
5577 | .poll = tracing_buffers_poll, | |
5578 | .release = tracing_buffers_release, | |
5579 | .splice_read = tracing_buffers_splice_read, | |
5580 | .llseek = no_llseek, | |
5581 | }; | |
5582 | ||
5583 | static ssize_t | |
5584 | tracing_stats_read(struct file *filp, char __user *ubuf, | |
5585 | size_t count, loff_t *ppos) | |
5586 | { | |
5587 | struct inode *inode = file_inode(filp); | |
5588 | struct trace_array *tr = inode->i_private; | |
5589 | struct trace_buffer *trace_buf = &tr->trace_buffer; | |
5590 | int cpu = tracing_get_cpu(inode); | |
5591 | struct trace_seq *s; | |
5592 | unsigned long cnt; | |
5593 | unsigned long long t; | |
5594 | unsigned long usec_rem; | |
5595 | ||
5596 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
5597 | if (!s) | |
5598 | return -ENOMEM; | |
5599 | ||
5600 | trace_seq_init(s); | |
5601 | ||
5602 | cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu); | |
5603 | trace_seq_printf(s, "entries: %ld\n", cnt); | |
5604 | ||
5605 | cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu); | |
5606 | trace_seq_printf(s, "overrun: %ld\n", cnt); | |
5607 | ||
5608 | cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu); | |
5609 | trace_seq_printf(s, "commit overrun: %ld\n", cnt); | |
5610 | ||
5611 | cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu); | |
5612 | trace_seq_printf(s, "bytes: %ld\n", cnt); | |
5613 | ||
5614 | if (trace_clocks[tr->clock_id].in_ns) { | |
5615 | /* local or global for trace_clock */ | |
5616 | t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu)); | |
5617 | usec_rem = do_div(t, USEC_PER_SEC); | |
5618 | trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n", | |
5619 | t, usec_rem); | |
5620 | ||
5621 | t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu)); | |
5622 | usec_rem = do_div(t, USEC_PER_SEC); | |
5623 | trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem); | |
5624 | } else { | |
5625 | /* counter or tsc mode for trace_clock */ | |
5626 | trace_seq_printf(s, "oldest event ts: %llu\n", | |
5627 | ring_buffer_oldest_event_ts(trace_buf->buffer, cpu)); | |
5628 | ||
5629 | trace_seq_printf(s, "now ts: %llu\n", | |
5630 | ring_buffer_time_stamp(trace_buf->buffer, cpu)); | |
5631 | } | |
5632 | ||
5633 | cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu); | |
5634 | trace_seq_printf(s, "dropped events: %ld\n", cnt); | |
5635 | ||
5636 | cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu); | |
5637 | trace_seq_printf(s, "read events: %ld\n", cnt); | |
5638 | ||
5639 | count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len); | |
5640 | ||
5641 | kfree(s); | |
5642 | ||
5643 | return count; | |
5644 | } | |
5645 | ||
5646 | static const struct file_operations tracing_stats_fops = { | |
5647 | .open = tracing_open_generic_tr, | |
5648 | .read = tracing_stats_read, | |
5649 | .llseek = generic_file_llseek, | |
5650 | .release = tracing_release_generic_tr, | |
5651 | }; | |
5652 | ||
5653 | #ifdef CONFIG_DYNAMIC_FTRACE | |
5654 | ||
5655 | int __weak ftrace_arch_read_dyn_info(char *buf, int size) | |
5656 | { | |
5657 | return 0; | |
5658 | } | |
5659 | ||
5660 | static ssize_t | |
5661 | tracing_read_dyn_info(struct file *filp, char __user *ubuf, | |
5662 | size_t cnt, loff_t *ppos) | |
5663 | { | |
5664 | static char ftrace_dyn_info_buffer[1024]; | |
5665 | static DEFINE_MUTEX(dyn_info_mutex); | |
5666 | unsigned long *p = filp->private_data; | |
5667 | char *buf = ftrace_dyn_info_buffer; | |
5668 | int size = ARRAY_SIZE(ftrace_dyn_info_buffer); | |
5669 | int r; | |
5670 | ||
5671 | mutex_lock(&dyn_info_mutex); | |
5672 | r = sprintf(buf, "%ld ", *p); | |
5673 | ||
5674 | r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r); | |
5675 | buf[r++] = '\n'; | |
5676 | ||
5677 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
5678 | ||
5679 | mutex_unlock(&dyn_info_mutex); | |
5680 | ||
5681 | return r; | |
5682 | } | |
5683 | ||
5684 | static const struct file_operations tracing_dyn_info_fops = { | |
5685 | .open = tracing_open_generic, | |
5686 | .read = tracing_read_dyn_info, | |
5687 | .llseek = generic_file_llseek, | |
5688 | }; | |
5689 | #endif /* CONFIG_DYNAMIC_FTRACE */ | |
5690 | ||
5691 | #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) | |
5692 | static void | |
5693 | ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data) | |
5694 | { | |
5695 | tracing_snapshot(); | |
5696 | } | |
5697 | ||
5698 | static void | |
5699 | ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data) | |
5700 | { | |
5701 | unsigned long *count = (long *)data; | |
5702 | ||
5703 | if (!*count) | |
5704 | return; | |
5705 | ||
5706 | if (*count != -1) | |
5707 | (*count)--; | |
5708 | ||
5709 | tracing_snapshot(); | |
5710 | } | |
5711 | ||
5712 | static int | |
5713 | ftrace_snapshot_print(struct seq_file *m, unsigned long ip, | |
5714 | struct ftrace_probe_ops *ops, void *data) | |
5715 | { | |
5716 | long count = (long)data; | |
5717 | ||
5718 | seq_printf(m, "%ps:", (void *)ip); | |
5719 | ||
5720 | seq_printf(m, "snapshot"); | |
5721 | ||
5722 | if (count == -1) | |
5723 | seq_printf(m, ":unlimited\n"); | |
5724 | else | |
5725 | seq_printf(m, ":count=%ld\n", count); | |
5726 | ||
5727 | return 0; | |
5728 | } | |
5729 | ||
5730 | static struct ftrace_probe_ops snapshot_probe_ops = { | |
5731 | .func = ftrace_snapshot, | |
5732 | .print = ftrace_snapshot_print, | |
5733 | }; | |
5734 | ||
5735 | static struct ftrace_probe_ops snapshot_count_probe_ops = { | |
5736 | .func = ftrace_count_snapshot, | |
5737 | .print = ftrace_snapshot_print, | |
5738 | }; | |
5739 | ||
5740 | static int | |
5741 | ftrace_trace_snapshot_callback(struct ftrace_hash *hash, | |
5742 | char *glob, char *cmd, char *param, int enable) | |
5743 | { | |
5744 | struct ftrace_probe_ops *ops; | |
5745 | void *count = (void *)-1; | |
5746 | char *number; | |
5747 | int ret; | |
5748 | ||
5749 | /* hash funcs only work with set_ftrace_filter */ | |
5750 | if (!enable) | |
5751 | return -EINVAL; | |
5752 | ||
5753 | ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops; | |
5754 | ||
5755 | if (glob[0] == '!') { | |
5756 | unregister_ftrace_function_probe_func(glob+1, ops); | |
5757 | return 0; | |
5758 | } | |
5759 | ||
5760 | if (!param) | |
5761 | goto out_reg; | |
5762 | ||
5763 | number = strsep(¶m, ":"); | |
5764 | ||
5765 | if (!strlen(number)) | |
5766 | goto out_reg; | |
5767 | ||
5768 | /* | |
5769 | * We use the callback data field (which is a pointer) | |
5770 | * as our counter. | |
5771 | */ | |
5772 | ret = kstrtoul(number, 0, (unsigned long *)&count); | |
5773 | if (ret) | |
5774 | return ret; | |
5775 | ||
5776 | out_reg: | |
5777 | ret = register_ftrace_function_probe(glob, ops, count); | |
5778 | ||
5779 | if (ret >= 0) | |
5780 | alloc_snapshot(&global_trace); | |
5781 | ||
5782 | return ret < 0 ? ret : 0; | |
5783 | } | |
5784 | ||
5785 | static struct ftrace_func_command ftrace_snapshot_cmd = { | |
5786 | .name = "snapshot", | |
5787 | .func = ftrace_trace_snapshot_callback, | |
5788 | }; | |
5789 | ||
5790 | static __init int register_snapshot_cmd(void) | |
5791 | { | |
5792 | return register_ftrace_command(&ftrace_snapshot_cmd); | |
5793 | } | |
5794 | #else | |
5795 | static inline __init int register_snapshot_cmd(void) { return 0; } | |
5796 | #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */ | |
5797 | ||
5798 | struct dentry *tracing_init_dentry_tr(struct trace_array *tr) | |
5799 | { | |
5800 | if (tr->dir) | |
5801 | return tr->dir; | |
5802 | ||
5803 | if (!debugfs_initialized()) | |
5804 | return NULL; | |
5805 | ||
5806 | if (tr->flags & TRACE_ARRAY_FL_GLOBAL) | |
5807 | tr->dir = debugfs_create_dir("tracing", NULL); | |
5808 | ||
5809 | if (!tr->dir) | |
5810 | pr_warn_once("Could not create debugfs directory 'tracing'\n"); | |
5811 | ||
5812 | return tr->dir; | |
5813 | } | |
5814 | ||
5815 | struct dentry *tracing_init_dentry(void) | |
5816 | { | |
5817 | return tracing_init_dentry_tr(&global_trace); | |
5818 | } | |
5819 | ||
5820 | static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu) | |
5821 | { | |
5822 | struct dentry *d_tracer; | |
5823 | ||
5824 | if (tr->percpu_dir) | |
5825 | return tr->percpu_dir; | |
5826 | ||
5827 | d_tracer = tracing_init_dentry_tr(tr); | |
5828 | if (!d_tracer) | |
5829 | return NULL; | |
5830 | ||
5831 | tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer); | |
5832 | ||
5833 | WARN_ONCE(!tr->percpu_dir, | |
5834 | "Could not create debugfs directory 'per_cpu/%d'\n", cpu); | |
5835 | ||
5836 | return tr->percpu_dir; | |
5837 | } | |
5838 | ||
5839 | static struct dentry * | |
5840 | trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent, | |
5841 | void *data, long cpu, const struct file_operations *fops) | |
5842 | { | |
5843 | struct dentry *ret = trace_create_file(name, mode, parent, data, fops); | |
5844 | ||
5845 | if (ret) /* See tracing_get_cpu() */ | |
5846 | ret->d_inode->i_cdev = (void *)(cpu + 1); | |
5847 | return ret; | |
5848 | } | |
5849 | ||
5850 | static void | |
5851 | tracing_init_debugfs_percpu(struct trace_array *tr, long cpu) | |
5852 | { | |
5853 | struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu); | |
5854 | struct dentry *d_cpu; | |
5855 | char cpu_dir[30]; /* 30 characters should be more than enough */ | |
5856 | ||
5857 | if (!d_percpu) | |
5858 | return; | |
5859 | ||
5860 | snprintf(cpu_dir, 30, "cpu%ld", cpu); | |
5861 | d_cpu = debugfs_create_dir(cpu_dir, d_percpu); | |
5862 | if (!d_cpu) { | |
5863 | pr_warning("Could not create debugfs '%s' entry\n", cpu_dir); | |
5864 | return; | |
5865 | } | |
5866 | ||
5867 | /* per cpu trace_pipe */ | |
5868 | trace_create_cpu_file("trace_pipe", 0444, d_cpu, | |
5869 | tr, cpu, &tracing_pipe_fops); | |
5870 | ||
5871 | /* per cpu trace */ | |
5872 | trace_create_cpu_file("trace", 0644, d_cpu, | |
5873 | tr, cpu, &tracing_fops); | |
5874 | ||
5875 | trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu, | |
5876 | tr, cpu, &tracing_buffers_fops); | |
5877 | ||
5878 | trace_create_cpu_file("stats", 0444, d_cpu, | |
5879 | tr, cpu, &tracing_stats_fops); | |
5880 | ||
5881 | trace_create_cpu_file("buffer_size_kb", 0444, d_cpu, | |
5882 | tr, cpu, &tracing_entries_fops); | |
5883 | ||
5884 | #ifdef CONFIG_TRACER_SNAPSHOT | |
5885 | trace_create_cpu_file("snapshot", 0644, d_cpu, | |
5886 | tr, cpu, &snapshot_fops); | |
5887 | ||
5888 | trace_create_cpu_file("snapshot_raw", 0444, d_cpu, | |
5889 | tr, cpu, &snapshot_raw_fops); | |
5890 | #endif | |
5891 | } | |
5892 | ||
5893 | #ifdef CONFIG_FTRACE_SELFTEST | |
5894 | /* Let selftest have access to static functions in this file */ | |
5895 | #include "trace_selftest.c" | |
5896 | #endif | |
5897 | ||
5898 | struct trace_option_dentry { | |
5899 | struct tracer_opt *opt; | |
5900 | struct tracer_flags *flags; | |
5901 | struct trace_array *tr; | |
5902 | struct dentry *entry; | |
5903 | }; | |
5904 | ||
5905 | static ssize_t | |
5906 | trace_options_read(struct file *filp, char __user *ubuf, size_t cnt, | |
5907 | loff_t *ppos) | |
5908 | { | |
5909 | struct trace_option_dentry *topt = filp->private_data; | |
5910 | char *buf; | |
5911 | ||
5912 | if (topt->flags->val & topt->opt->bit) | |
5913 | buf = "1\n"; | |
5914 | else | |
5915 | buf = "0\n"; | |
5916 | ||
5917 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2); | |
5918 | } | |
5919 | ||
5920 | static ssize_t | |
5921 | trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt, | |
5922 | loff_t *ppos) | |
5923 | { | |
5924 | struct trace_option_dentry *topt = filp->private_data; | |
5925 | unsigned long val; | |
5926 | int ret; | |
5927 | ||
5928 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
5929 | if (ret) | |
5930 | return ret; | |
5931 | ||
5932 | if (val != 0 && val != 1) | |
5933 | return -EINVAL; | |
5934 | ||
5935 | if (!!(topt->flags->val & topt->opt->bit) != val) { | |
5936 | mutex_lock(&trace_types_lock); | |
5937 | ret = __set_tracer_option(topt->tr, topt->flags, | |
5938 | topt->opt, !val); | |
5939 | mutex_unlock(&trace_types_lock); | |
5940 | if (ret) | |
5941 | return ret; | |
5942 | } | |
5943 | ||
5944 | *ppos += cnt; | |
5945 | ||
5946 | return cnt; | |
5947 | } | |
5948 | ||
5949 | ||
5950 | static const struct file_operations trace_options_fops = { | |
5951 | .open = tracing_open_generic, | |
5952 | .read = trace_options_read, | |
5953 | .write = trace_options_write, | |
5954 | .llseek = generic_file_llseek, | |
5955 | }; | |
5956 | ||
5957 | static ssize_t | |
5958 | trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt, | |
5959 | loff_t *ppos) | |
5960 | { | |
5961 | long index = (long)filp->private_data; | |
5962 | char *buf; | |
5963 | ||
5964 | if (trace_flags & (1 << index)) | |
5965 | buf = "1\n"; | |
5966 | else | |
5967 | buf = "0\n"; | |
5968 | ||
5969 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2); | |
5970 | } | |
5971 | ||
5972 | static ssize_t | |
5973 | trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt, | |
5974 | loff_t *ppos) | |
5975 | { | |
5976 | struct trace_array *tr = &global_trace; | |
5977 | long index = (long)filp->private_data; | |
5978 | unsigned long val; | |
5979 | int ret; | |
5980 | ||
5981 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
5982 | if (ret) | |
5983 | return ret; | |
5984 | ||
5985 | if (val != 0 && val != 1) | |
5986 | return -EINVAL; | |
5987 | ||
5988 | mutex_lock(&trace_types_lock); | |
5989 | ret = set_tracer_flag(tr, 1 << index, val); | |
5990 | mutex_unlock(&trace_types_lock); | |
5991 | ||
5992 | if (ret < 0) | |
5993 | return ret; | |
5994 | ||
5995 | *ppos += cnt; | |
5996 | ||
5997 | return cnt; | |
5998 | } | |
5999 | ||
6000 | static const struct file_operations trace_options_core_fops = { | |
6001 | .open = tracing_open_generic, | |
6002 | .read = trace_options_core_read, | |
6003 | .write = trace_options_core_write, | |
6004 | .llseek = generic_file_llseek, | |
6005 | }; | |
6006 | ||
6007 | struct dentry *trace_create_file(const char *name, | |
6008 | umode_t mode, | |
6009 | struct dentry *parent, | |
6010 | void *data, | |
6011 | const struct file_operations *fops) | |
6012 | { | |
6013 | struct dentry *ret; | |
6014 | ||
6015 | ret = debugfs_create_file(name, mode, parent, data, fops); | |
6016 | if (!ret) | |
6017 | pr_warning("Could not create debugfs '%s' entry\n", name); | |
6018 | ||
6019 | return ret; | |
6020 | } | |
6021 | ||
6022 | ||
6023 | static struct dentry *trace_options_init_dentry(struct trace_array *tr) | |
6024 | { | |
6025 | struct dentry *d_tracer; | |
6026 | ||
6027 | if (tr->options) | |
6028 | return tr->options; | |
6029 | ||
6030 | d_tracer = tracing_init_dentry_tr(tr); | |
6031 | if (!d_tracer) | |
6032 | return NULL; | |
6033 | ||
6034 | tr->options = debugfs_create_dir("options", d_tracer); | |
6035 | if (!tr->options) { | |
6036 | pr_warning("Could not create debugfs directory 'options'\n"); | |
6037 | return NULL; | |
6038 | } | |
6039 | ||
6040 | return tr->options; | |
6041 | } | |
6042 | ||
6043 | static void | |
6044 | create_trace_option_file(struct trace_array *tr, | |
6045 | struct trace_option_dentry *topt, | |
6046 | struct tracer_flags *flags, | |
6047 | struct tracer_opt *opt) | |
6048 | { | |
6049 | struct dentry *t_options; | |
6050 | ||
6051 | t_options = trace_options_init_dentry(tr); | |
6052 | if (!t_options) | |
6053 | return; | |
6054 | ||
6055 | topt->flags = flags; | |
6056 | topt->opt = opt; | |
6057 | topt->tr = tr; | |
6058 | ||
6059 | topt->entry = trace_create_file(opt->name, 0644, t_options, topt, | |
6060 | &trace_options_fops); | |
6061 | ||
6062 | } | |
6063 | ||
6064 | static struct trace_option_dentry * | |
6065 | create_trace_option_files(struct trace_array *tr, struct tracer *tracer) | |
6066 | { | |
6067 | struct trace_option_dentry *topts; | |
6068 | struct tracer_flags *flags; | |
6069 | struct tracer_opt *opts; | |
6070 | int cnt; | |
6071 | ||
6072 | if (!tracer) | |
6073 | return NULL; | |
6074 | ||
6075 | flags = tracer->flags; | |
6076 | ||
6077 | if (!flags || !flags->opts) | |
6078 | return NULL; | |
6079 | ||
6080 | opts = flags->opts; | |
6081 | ||
6082 | for (cnt = 0; opts[cnt].name; cnt++) | |
6083 | ; | |
6084 | ||
6085 | topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL); | |
6086 | if (!topts) | |
6087 | return NULL; | |
6088 | ||
6089 | for (cnt = 0; opts[cnt].name; cnt++) | |
6090 | create_trace_option_file(tr, &topts[cnt], flags, | |
6091 | &opts[cnt]); | |
6092 | ||
6093 | return topts; | |
6094 | } | |
6095 | ||
6096 | static void | |
6097 | destroy_trace_option_files(struct trace_option_dentry *topts) | |
6098 | { | |
6099 | int cnt; | |
6100 | ||
6101 | if (!topts) | |
6102 | return; | |
6103 | ||
6104 | for (cnt = 0; topts[cnt].opt; cnt++) { | |
6105 | if (topts[cnt].entry) | |
6106 | debugfs_remove(topts[cnt].entry); | |
6107 | } | |
6108 | ||
6109 | kfree(topts); | |
6110 | } | |
6111 | ||
6112 | static struct dentry * | |
6113 | create_trace_option_core_file(struct trace_array *tr, | |
6114 | const char *option, long index) | |
6115 | { | |
6116 | struct dentry *t_options; | |
6117 | ||
6118 | t_options = trace_options_init_dentry(tr); | |
6119 | if (!t_options) | |
6120 | return NULL; | |
6121 | ||
6122 | return trace_create_file(option, 0644, t_options, (void *)index, | |
6123 | &trace_options_core_fops); | |
6124 | } | |
6125 | ||
6126 | static __init void create_trace_options_dir(struct trace_array *tr) | |
6127 | { | |
6128 | struct dentry *t_options; | |
6129 | int i; | |
6130 | ||
6131 | t_options = trace_options_init_dentry(tr); | |
6132 | if (!t_options) | |
6133 | return; | |
6134 | ||
6135 | for (i = 0; trace_options[i]; i++) | |
6136 | create_trace_option_core_file(tr, trace_options[i], i); | |
6137 | } | |
6138 | ||
6139 | static ssize_t | |
6140 | rb_simple_read(struct file *filp, char __user *ubuf, | |
6141 | size_t cnt, loff_t *ppos) | |
6142 | { | |
6143 | struct trace_array *tr = filp->private_data; | |
6144 | char buf[64]; | |
6145 | int r; | |
6146 | ||
6147 | r = tracer_tracing_is_on(tr); | |
6148 | r = sprintf(buf, "%d\n", r); | |
6149 | ||
6150 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
6151 | } | |
6152 | ||
6153 | static ssize_t | |
6154 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
6155 | size_t cnt, loff_t *ppos) | |
6156 | { | |
6157 | struct trace_array *tr = filp->private_data; | |
6158 | struct ring_buffer *buffer = tr->trace_buffer.buffer; | |
6159 | unsigned long val; | |
6160 | int ret; | |
6161 | ||
6162 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
6163 | if (ret) | |
6164 | return ret; | |
6165 | ||
6166 | if (buffer) { | |
6167 | mutex_lock(&trace_types_lock); | |
6168 | if (val) { | |
6169 | tracer_tracing_on(tr); | |
6170 | if (tr->current_trace->start) | |
6171 | tr->current_trace->start(tr); | |
6172 | } else { | |
6173 | tracer_tracing_off(tr); | |
6174 | if (tr->current_trace->stop) | |
6175 | tr->current_trace->stop(tr); | |
6176 | } | |
6177 | mutex_unlock(&trace_types_lock); | |
6178 | } | |
6179 | ||
6180 | (*ppos)++; | |
6181 | ||
6182 | return cnt; | |
6183 | } | |
6184 | ||
6185 | static const struct file_operations rb_simple_fops = { | |
6186 | .open = tracing_open_generic_tr, | |
6187 | .read = rb_simple_read, | |
6188 | .write = rb_simple_write, | |
6189 | .release = tracing_release_generic_tr, | |
6190 | .llseek = default_llseek, | |
6191 | }; | |
6192 | ||
6193 | struct dentry *trace_instance_dir; | |
6194 | ||
6195 | static void | |
6196 | init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer); | |
6197 | ||
6198 | static int | |
6199 | allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size) | |
6200 | { | |
6201 | enum ring_buffer_flags rb_flags; | |
6202 | ||
6203 | rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0; | |
6204 | ||
6205 | buf->tr = tr; | |
6206 | ||
6207 | buf->buffer = ring_buffer_alloc(size, rb_flags); | |
6208 | if (!buf->buffer) | |
6209 | return -ENOMEM; | |
6210 | ||
6211 | buf->data = alloc_percpu(struct trace_array_cpu); | |
6212 | if (!buf->data) { | |
6213 | ring_buffer_free(buf->buffer); | |
6214 | return -ENOMEM; | |
6215 | } | |
6216 | ||
6217 | /* Allocate the first page for all buffers */ | |
6218 | set_buffer_entries(&tr->trace_buffer, | |
6219 | ring_buffer_size(tr->trace_buffer.buffer, 0)); | |
6220 | ||
6221 | return 0; | |
6222 | } | |
6223 | ||
6224 | static int allocate_trace_buffers(struct trace_array *tr, int size) | |
6225 | { | |
6226 | int ret; | |
6227 | ||
6228 | ret = allocate_trace_buffer(tr, &tr->trace_buffer, size); | |
6229 | if (ret) | |
6230 | return ret; | |
6231 | ||
6232 | #ifdef CONFIG_TRACER_MAX_TRACE | |
6233 | ret = allocate_trace_buffer(tr, &tr->max_buffer, | |
6234 | allocate_snapshot ? size : 1); | |
6235 | if (WARN_ON(ret)) { | |
6236 | ring_buffer_free(tr->trace_buffer.buffer); | |
6237 | free_percpu(tr->trace_buffer.data); | |
6238 | return -ENOMEM; | |
6239 | } | |
6240 | tr->allocated_snapshot = allocate_snapshot; | |
6241 | ||
6242 | /* | |
6243 | * Only the top level trace array gets its snapshot allocated | |
6244 | * from the kernel command line. | |
6245 | */ | |
6246 | allocate_snapshot = false; | |
6247 | #endif | |
6248 | return 0; | |
6249 | } | |
6250 | ||
6251 | static void free_trace_buffer(struct trace_buffer *buf) | |
6252 | { | |
6253 | if (buf->buffer) { | |
6254 | ring_buffer_free(buf->buffer); | |
6255 | buf->buffer = NULL; | |
6256 | free_percpu(buf->data); | |
6257 | buf->data = NULL; | |
6258 | } | |
6259 | } | |
6260 | ||
6261 | static void free_trace_buffers(struct trace_array *tr) | |
6262 | { | |
6263 | if (!tr) | |
6264 | return; | |
6265 | ||
6266 | free_trace_buffer(&tr->trace_buffer); | |
6267 | ||
6268 | #ifdef CONFIG_TRACER_MAX_TRACE | |
6269 | free_trace_buffer(&tr->max_buffer); | |
6270 | #endif | |
6271 | } | |
6272 | ||
6273 | static int new_instance_create(const char *name) | |
6274 | { | |
6275 | struct trace_array *tr; | |
6276 | int ret; | |
6277 | ||
6278 | mutex_lock(&trace_types_lock); | |
6279 | ||
6280 | ret = -EEXIST; | |
6281 | list_for_each_entry(tr, &ftrace_trace_arrays, list) { | |
6282 | if (tr->name && strcmp(tr->name, name) == 0) | |
6283 | goto out_unlock; | |
6284 | } | |
6285 | ||
6286 | ret = -ENOMEM; | |
6287 | tr = kzalloc(sizeof(*tr), GFP_KERNEL); | |
6288 | if (!tr) | |
6289 | goto out_unlock; | |
6290 | ||
6291 | tr->name = kstrdup(name, GFP_KERNEL); | |
6292 | if (!tr->name) | |
6293 | goto out_free_tr; | |
6294 | ||
6295 | if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL)) | |
6296 | goto out_free_tr; | |
6297 | ||
6298 | cpumask_copy(tr->tracing_cpumask, cpu_all_mask); | |
6299 | ||
6300 | raw_spin_lock_init(&tr->start_lock); | |
6301 | ||
6302 | tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; | |
6303 | ||
6304 | tr->current_trace = &nop_trace; | |
6305 | ||
6306 | INIT_LIST_HEAD(&tr->systems); | |
6307 | INIT_LIST_HEAD(&tr->events); | |
6308 | ||
6309 | if (allocate_trace_buffers(tr, trace_buf_size) < 0) | |
6310 | goto out_free_tr; | |
6311 | ||
6312 | tr->dir = debugfs_create_dir(name, trace_instance_dir); | |
6313 | if (!tr->dir) | |
6314 | goto out_free_tr; | |
6315 | ||
6316 | ret = event_trace_add_tracer(tr->dir, tr); | |
6317 | if (ret) { | |
6318 | debugfs_remove_recursive(tr->dir); | |
6319 | goto out_free_tr; | |
6320 | } | |
6321 | ||
6322 | init_tracer_debugfs(tr, tr->dir); | |
6323 | ||
6324 | list_add(&tr->list, &ftrace_trace_arrays); | |
6325 | ||
6326 | mutex_unlock(&trace_types_lock); | |
6327 | ||
6328 | return 0; | |
6329 | ||
6330 | out_free_tr: | |
6331 | free_trace_buffers(tr); | |
6332 | free_cpumask_var(tr->tracing_cpumask); | |
6333 | kfree(tr->name); | |
6334 | kfree(tr); | |
6335 | ||
6336 | out_unlock: | |
6337 | mutex_unlock(&trace_types_lock); | |
6338 | ||
6339 | return ret; | |
6340 | ||
6341 | } | |
6342 | ||
6343 | static int instance_delete(const char *name) | |
6344 | { | |
6345 | struct trace_array *tr; | |
6346 | int found = 0; | |
6347 | int ret; | |
6348 | ||
6349 | mutex_lock(&trace_types_lock); | |
6350 | ||
6351 | ret = -ENODEV; | |
6352 | list_for_each_entry(tr, &ftrace_trace_arrays, list) { | |
6353 | if (tr->name && strcmp(tr->name, name) == 0) { | |
6354 | found = 1; | |
6355 | break; | |
6356 | } | |
6357 | } | |
6358 | if (!found) | |
6359 | goto out_unlock; | |
6360 | ||
6361 | ret = -EBUSY; | |
6362 | if (tr->ref) | |
6363 | goto out_unlock; | |
6364 | ||
6365 | list_del(&tr->list); | |
6366 | ||
6367 | tracing_set_nop(tr); | |
6368 | event_trace_del_tracer(tr); | |
6369 | ftrace_destroy_function_files(tr); | |
6370 | debugfs_remove_recursive(tr->dir); | |
6371 | free_trace_buffers(tr); | |
6372 | ||
6373 | kfree(tr->name); | |
6374 | kfree(tr); | |
6375 | ||
6376 | ret = 0; | |
6377 | ||
6378 | out_unlock: | |
6379 | mutex_unlock(&trace_types_lock); | |
6380 | ||
6381 | return ret; | |
6382 | } | |
6383 | ||
6384 | static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode) | |
6385 | { | |
6386 | struct dentry *parent; | |
6387 | int ret; | |
6388 | ||
6389 | /* Paranoid: Make sure the parent is the "instances" directory */ | |
6390 | parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); | |
6391 | if (WARN_ON_ONCE(parent != trace_instance_dir)) | |
6392 | return -ENOENT; | |
6393 | ||
6394 | /* | |
6395 | * The inode mutex is locked, but debugfs_create_dir() will also | |
6396 | * take the mutex. As the instances directory can not be destroyed | |
6397 | * or changed in any other way, it is safe to unlock it, and | |
6398 | * let the dentry try. If two users try to make the same dir at | |
6399 | * the same time, then the new_instance_create() will determine the | |
6400 | * winner. | |
6401 | */ | |
6402 | mutex_unlock(&inode->i_mutex); | |
6403 | ||
6404 | ret = new_instance_create(dentry->d_iname); | |
6405 | ||
6406 | mutex_lock(&inode->i_mutex); | |
6407 | ||
6408 | return ret; | |
6409 | } | |
6410 | ||
6411 | static int instance_rmdir(struct inode *inode, struct dentry *dentry) | |
6412 | { | |
6413 | struct dentry *parent; | |
6414 | int ret; | |
6415 | ||
6416 | /* Paranoid: Make sure the parent is the "instances" directory */ | |
6417 | parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); | |
6418 | if (WARN_ON_ONCE(parent != trace_instance_dir)) | |
6419 | return -ENOENT; | |
6420 | ||
6421 | /* The caller did a dget() on dentry */ | |
6422 | mutex_unlock(&dentry->d_inode->i_mutex); | |
6423 | ||
6424 | /* | |
6425 | * The inode mutex is locked, but debugfs_create_dir() will also | |
6426 | * take the mutex. As the instances directory can not be destroyed | |
6427 | * or changed in any other way, it is safe to unlock it, and | |
6428 | * let the dentry try. If two users try to make the same dir at | |
6429 | * the same time, then the instance_delete() will determine the | |
6430 | * winner. | |
6431 | */ | |
6432 | mutex_unlock(&inode->i_mutex); | |
6433 | ||
6434 | ret = instance_delete(dentry->d_iname); | |
6435 | ||
6436 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT); | |
6437 | mutex_lock(&dentry->d_inode->i_mutex); | |
6438 | ||
6439 | return ret; | |
6440 | } | |
6441 | ||
6442 | static const struct inode_operations instance_dir_inode_operations = { | |
6443 | .lookup = simple_lookup, | |
6444 | .mkdir = instance_mkdir, | |
6445 | .rmdir = instance_rmdir, | |
6446 | }; | |
6447 | ||
6448 | static __init void create_trace_instances(struct dentry *d_tracer) | |
6449 | { | |
6450 | trace_instance_dir = debugfs_create_dir("instances", d_tracer); | |
6451 | if (WARN_ON(!trace_instance_dir)) | |
6452 | return; | |
6453 | ||
6454 | /* Hijack the dir inode operations, to allow mkdir */ | |
6455 | trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations; | |
6456 | } | |
6457 | ||
6458 | static void | |
6459 | init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer) | |
6460 | { | |
6461 | int cpu; | |
6462 | ||
6463 | trace_create_file("available_tracers", 0444, d_tracer, | |
6464 | tr, &show_traces_fops); | |
6465 | ||
6466 | trace_create_file("current_tracer", 0644, d_tracer, | |
6467 | tr, &set_tracer_fops); | |
6468 | ||
6469 | trace_create_file("tracing_cpumask", 0644, d_tracer, | |
6470 | tr, &tracing_cpumask_fops); | |
6471 | ||
6472 | trace_create_file("trace_options", 0644, d_tracer, | |
6473 | tr, &tracing_iter_fops); | |
6474 | ||
6475 | trace_create_file("trace", 0644, d_tracer, | |
6476 | tr, &tracing_fops); | |
6477 | ||
6478 | trace_create_file("trace_pipe", 0444, d_tracer, | |
6479 | tr, &tracing_pipe_fops); | |
6480 | ||
6481 | trace_create_file("buffer_size_kb", 0644, d_tracer, | |
6482 | tr, &tracing_entries_fops); | |
6483 | ||
6484 | trace_create_file("buffer_total_size_kb", 0444, d_tracer, | |
6485 | tr, &tracing_total_entries_fops); | |
6486 | ||
6487 | trace_create_file("free_buffer", 0200, d_tracer, | |
6488 | tr, &tracing_free_buffer_fops); | |
6489 | ||
6490 | trace_create_file("trace_marker", 0220, d_tracer, | |
6491 | tr, &tracing_mark_fops); | |
6492 | ||
6493 | trace_create_file("trace_clock", 0644, d_tracer, tr, | |
6494 | &trace_clock_fops); | |
6495 | ||
6496 | trace_create_file("tracing_on", 0644, d_tracer, | |
6497 | tr, &rb_simple_fops); | |
6498 | ||
6499 | #ifdef CONFIG_TRACER_MAX_TRACE | |
6500 | trace_create_file("tracing_max_latency", 0644, d_tracer, | |
6501 | &tr->max_latency, &tracing_max_lat_fops); | |
6502 | #endif | |
6503 | ||
6504 | if (ftrace_create_function_files(tr, d_tracer)) | |
6505 | WARN(1, "Could not allocate function filter files"); | |
6506 | ||
6507 | #ifdef CONFIG_TRACER_SNAPSHOT | |
6508 | trace_create_file("snapshot", 0644, d_tracer, | |
6509 | tr, &snapshot_fops); | |
6510 | #endif | |
6511 | ||
6512 | for_each_tracing_cpu(cpu) | |
6513 | tracing_init_debugfs_percpu(tr, cpu); | |
6514 | ||
6515 | } | |
6516 | ||
6517 | static __init int tracer_init_debugfs(void) | |
6518 | { | |
6519 | struct dentry *d_tracer; | |
6520 | ||
6521 | trace_access_lock_init(); | |
6522 | ||
6523 | d_tracer = tracing_init_dentry(); | |
6524 | if (!d_tracer) | |
6525 | return 0; | |
6526 | ||
6527 | init_tracer_debugfs(&global_trace, d_tracer); | |
6528 | ||
6529 | trace_create_file("tracing_thresh", 0644, d_tracer, | |
6530 | &tracing_thresh, &tracing_max_lat_fops); | |
6531 | ||
6532 | trace_create_file("README", 0444, d_tracer, | |
6533 | NULL, &tracing_readme_fops); | |
6534 | ||
6535 | trace_create_file("saved_cmdlines", 0444, d_tracer, | |
6536 | NULL, &tracing_saved_cmdlines_fops); | |
6537 | ||
6538 | trace_create_file("saved_cmdlines_size", 0644, d_tracer, | |
6539 | NULL, &tracing_saved_cmdlines_size_fops); | |
6540 | ||
6541 | #ifdef CONFIG_DYNAMIC_FTRACE | |
6542 | trace_create_file("dyn_ftrace_total_info", 0444, d_tracer, | |
6543 | &ftrace_update_tot_cnt, &tracing_dyn_info_fops); | |
6544 | #endif | |
6545 | ||
6546 | create_trace_instances(d_tracer); | |
6547 | ||
6548 | create_trace_options_dir(&global_trace); | |
6549 | ||
6550 | return 0; | |
6551 | } | |
6552 | ||
6553 | static int trace_panic_handler(struct notifier_block *this, | |
6554 | unsigned long event, void *unused) | |
6555 | { | |
6556 | if (ftrace_dump_on_oops) | |
6557 | ftrace_dump(ftrace_dump_on_oops); | |
6558 | return NOTIFY_OK; | |
6559 | } | |
6560 | ||
6561 | static struct notifier_block trace_panic_notifier = { | |
6562 | .notifier_call = trace_panic_handler, | |
6563 | .next = NULL, | |
6564 | .priority = 150 /* priority: INT_MAX >= x >= 0 */ | |
6565 | }; | |
6566 | ||
6567 | static int trace_die_handler(struct notifier_block *self, | |
6568 | unsigned long val, | |
6569 | void *data) | |
6570 | { | |
6571 | switch (val) { | |
6572 | case DIE_OOPS: | |
6573 | if (ftrace_dump_on_oops) | |
6574 | ftrace_dump(ftrace_dump_on_oops); | |
6575 | break; | |
6576 | default: | |
6577 | break; | |
6578 | } | |
6579 | return NOTIFY_OK; | |
6580 | } | |
6581 | ||
6582 | static struct notifier_block trace_die_notifier = { | |
6583 | .notifier_call = trace_die_handler, | |
6584 | .priority = 200 | |
6585 | }; | |
6586 | ||
6587 | /* | |
6588 | * printk is set to max of 1024, we really don't need it that big. | |
6589 | * Nothing should be printing 1000 characters anyway. | |
6590 | */ | |
6591 | #define TRACE_MAX_PRINT 1000 | |
6592 | ||
6593 | /* | |
6594 | * Define here KERN_TRACE so that we have one place to modify | |
6595 | * it if we decide to change what log level the ftrace dump | |
6596 | * should be at. | |
6597 | */ | |
6598 | #define KERN_TRACE KERN_EMERG | |
6599 | ||
6600 | void | |
6601 | trace_printk_seq(struct trace_seq *s) | |
6602 | { | |
6603 | /* Probably should print a warning here. */ | |
6604 | if (s->len >= TRACE_MAX_PRINT) | |
6605 | s->len = TRACE_MAX_PRINT; | |
6606 | ||
6607 | /* should be zero ended, but we are paranoid. */ | |
6608 | s->buffer[s->len] = 0; | |
6609 | ||
6610 | printk(KERN_TRACE "%s", s->buffer); | |
6611 | ||
6612 | trace_seq_init(s); | |
6613 | } | |
6614 | ||
6615 | void trace_init_global_iter(struct trace_iterator *iter) | |
6616 | { | |
6617 | iter->tr = &global_trace; | |
6618 | iter->trace = iter->tr->current_trace; | |
6619 | iter->cpu_file = RING_BUFFER_ALL_CPUS; | |
6620 | iter->trace_buffer = &global_trace.trace_buffer; | |
6621 | ||
6622 | if (iter->trace && iter->trace->open) | |
6623 | iter->trace->open(iter); | |
6624 | ||
6625 | /* Annotate start of buffers if we had overruns */ | |
6626 | if (ring_buffer_overruns(iter->trace_buffer->buffer)) | |
6627 | iter->iter_flags |= TRACE_FILE_ANNOTATE; | |
6628 | ||
6629 | /* Output in nanoseconds only if we are using a clock in nanoseconds. */ | |
6630 | if (trace_clocks[iter->tr->clock_id].in_ns) | |
6631 | iter->iter_flags |= TRACE_FILE_TIME_IN_NS; | |
6632 | } | |
6633 | ||
6634 | void ftrace_dump(enum ftrace_dump_mode oops_dump_mode) | |
6635 | { | |
6636 | /* use static because iter can be a bit big for the stack */ | |
6637 | static struct trace_iterator iter; | |
6638 | static atomic_t dump_running; | |
6639 | unsigned int old_userobj; | |
6640 | unsigned long flags; | |
6641 | int cnt = 0, cpu; | |
6642 | ||
6643 | /* Only allow one dump user at a time. */ | |
6644 | if (atomic_inc_return(&dump_running) != 1) { | |
6645 | atomic_dec(&dump_running); | |
6646 | return; | |
6647 | } | |
6648 | ||
6649 | /* | |
6650 | * Always turn off tracing when we dump. | |
6651 | * We don't need to show trace output of what happens | |
6652 | * between multiple crashes. | |
6653 | * | |
6654 | * If the user does a sysrq-z, then they can re-enable | |
6655 | * tracing with echo 1 > tracing_on. | |
6656 | */ | |
6657 | tracing_off(); | |
6658 | ||
6659 | local_irq_save(flags); | |
6660 | ||
6661 | /* Simulate the iterator */ | |
6662 | trace_init_global_iter(&iter); | |
6663 | ||
6664 | for_each_tracing_cpu(cpu) { | |
6665 | atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled); | |
6666 | } | |
6667 | ||
6668 | old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ; | |
6669 | ||
6670 | /* don't look at user memory in panic mode */ | |
6671 | trace_flags &= ~TRACE_ITER_SYM_USEROBJ; | |
6672 | ||
6673 | switch (oops_dump_mode) { | |
6674 | case DUMP_ALL: | |
6675 | iter.cpu_file = RING_BUFFER_ALL_CPUS; | |
6676 | break; | |
6677 | case DUMP_ORIG: | |
6678 | iter.cpu_file = raw_smp_processor_id(); | |
6679 | break; | |
6680 | case DUMP_NONE: | |
6681 | goto out_enable; | |
6682 | default: | |
6683 | printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n"); | |
6684 | iter.cpu_file = RING_BUFFER_ALL_CPUS; | |
6685 | } | |
6686 | ||
6687 | printk(KERN_TRACE "Dumping ftrace buffer:\n"); | |
6688 | ||
6689 | /* Did function tracer already get disabled? */ | |
6690 | if (ftrace_is_dead()) { | |
6691 | printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n"); | |
6692 | printk("# MAY BE MISSING FUNCTION EVENTS\n"); | |
6693 | } | |
6694 | ||
6695 | /* | |
6696 | * We need to stop all tracing on all CPUS to read the | |
6697 | * the next buffer. This is a bit expensive, but is | |
6698 | * not done often. We fill all what we can read, | |
6699 | * and then release the locks again. | |
6700 | */ | |
6701 | ||
6702 | while (!trace_empty(&iter)) { | |
6703 | ||
6704 | if (!cnt) | |
6705 | printk(KERN_TRACE "---------------------------------\n"); | |
6706 | ||
6707 | cnt++; | |
6708 | ||
6709 | /* reset all but tr, trace, and overruns */ | |
6710 | memset(&iter.seq, 0, | |
6711 | sizeof(struct trace_iterator) - | |
6712 | offsetof(struct trace_iterator, seq)); | |
6713 | iter.iter_flags |= TRACE_FILE_LAT_FMT; | |
6714 | iter.pos = -1; | |
6715 | ||
6716 | if (trace_find_next_entry_inc(&iter) != NULL) { | |
6717 | int ret; | |
6718 | ||
6719 | ret = print_trace_line(&iter); | |
6720 | if (ret != TRACE_TYPE_NO_CONSUME) | |
6721 | trace_consume(&iter); | |
6722 | } | |
6723 | touch_nmi_watchdog(); | |
6724 | ||
6725 | trace_printk_seq(&iter.seq); | |
6726 | } | |
6727 | ||
6728 | if (!cnt) | |
6729 | printk(KERN_TRACE " (ftrace buffer empty)\n"); | |
6730 | else | |
6731 | printk(KERN_TRACE "---------------------------------\n"); | |
6732 | ||
6733 | out_enable: | |
6734 | trace_flags |= old_userobj; | |
6735 | ||
6736 | for_each_tracing_cpu(cpu) { | |
6737 | atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled); | |
6738 | } | |
6739 | atomic_dec(&dump_running); | |
6740 | local_irq_restore(flags); | |
6741 | } | |
6742 | EXPORT_SYMBOL_GPL(ftrace_dump); | |
6743 | ||
6744 | __init static int tracer_alloc_buffers(void) | |
6745 | { | |
6746 | int ring_buf_size; | |
6747 | int ret = -ENOMEM; | |
6748 | ||
6749 | ||
6750 | if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL)) | |
6751 | goto out; | |
6752 | ||
6753 | if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL)) | |
6754 | goto out_free_buffer_mask; | |
6755 | ||
6756 | /* Only allocate trace_printk buffers if a trace_printk exists */ | |
6757 | if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt) | |
6758 | /* Must be called before global_trace.buffer is allocated */ | |
6759 | trace_printk_init_buffers(); | |
6760 | ||
6761 | /* To save memory, keep the ring buffer size to its minimum */ | |
6762 | if (ring_buffer_expanded) | |
6763 | ring_buf_size = trace_buf_size; | |
6764 | else | |
6765 | ring_buf_size = 1; | |
6766 | ||
6767 | cpumask_copy(tracing_buffer_mask, cpu_possible_mask); | |
6768 | cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask); | |
6769 | ||
6770 | raw_spin_lock_init(&global_trace.start_lock); | |
6771 | ||
6772 | /* Used for event triggers */ | |
6773 | temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE); | |
6774 | if (!temp_buffer) | |
6775 | goto out_free_cpumask; | |
6776 | ||
6777 | if (trace_create_savedcmd() < 0) | |
6778 | goto out_free_temp_buffer; | |
6779 | ||
6780 | /* TODO: make the number of buffers hot pluggable with CPUS */ | |
6781 | if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) { | |
6782 | printk(KERN_ERR "tracer: failed to allocate ring buffer!\n"); | |
6783 | WARN_ON(1); | |
6784 | goto out_free_savedcmd; | |
6785 | } | |
6786 | ||
6787 | if (global_trace.buffer_disabled) | |
6788 | tracing_off(); | |
6789 | ||
6790 | if (trace_boot_clock) { | |
6791 | ret = tracing_set_clock(&global_trace, trace_boot_clock); | |
6792 | if (ret < 0) | |
6793 | pr_warning("Trace clock %s not defined, going back to default\n", | |
6794 | trace_boot_clock); | |
6795 | } | |
6796 | ||
6797 | /* | |
6798 | * register_tracer() might reference current_trace, so it | |
6799 | * needs to be set before we register anything. This is | |
6800 | * just a bootstrap of current_trace anyway. | |
6801 | */ | |
6802 | global_trace.current_trace = &nop_trace; | |
6803 | ||
6804 | global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; | |
6805 | ||
6806 | ftrace_init_global_array_ops(&global_trace); | |
6807 | ||
6808 | register_tracer(&nop_trace); | |
6809 | ||
6810 | /* All seems OK, enable tracing */ | |
6811 | tracing_disabled = 0; | |
6812 | ||
6813 | atomic_notifier_chain_register(&panic_notifier_list, | |
6814 | &trace_panic_notifier); | |
6815 | ||
6816 | register_die_notifier(&trace_die_notifier); | |
6817 | ||
6818 | global_trace.flags = TRACE_ARRAY_FL_GLOBAL; | |
6819 | ||
6820 | INIT_LIST_HEAD(&global_trace.systems); | |
6821 | INIT_LIST_HEAD(&global_trace.events); | |
6822 | list_add(&global_trace.list, &ftrace_trace_arrays); | |
6823 | ||
6824 | while (trace_boot_options) { | |
6825 | char *option; | |
6826 | ||
6827 | option = strsep(&trace_boot_options, ","); | |
6828 | trace_set_options(&global_trace, option); | |
6829 | } | |
6830 | ||
6831 | register_snapshot_cmd(); | |
6832 | ||
6833 | return 0; | |
6834 | ||
6835 | out_free_savedcmd: | |
6836 | free_saved_cmdlines_buffer(savedcmd); | |
6837 | out_free_temp_buffer: | |
6838 | ring_buffer_free(temp_buffer); | |
6839 | out_free_cpumask: | |
6840 | free_cpumask_var(global_trace.tracing_cpumask); | |
6841 | out_free_buffer_mask: | |
6842 | free_cpumask_var(tracing_buffer_mask); | |
6843 | out: | |
6844 | return ret; | |
6845 | } | |
6846 | ||
6847 | __init static int clear_boot_tracer(void) | |
6848 | { | |
6849 | /* | |
6850 | * The default tracer at boot buffer is an init section. | |
6851 | * This function is called in lateinit. If we did not | |
6852 | * find the boot tracer, then clear it out, to prevent | |
6853 | * later registration from accessing the buffer that is | |
6854 | * about to be freed. | |
6855 | */ | |
6856 | if (!default_bootup_tracer) | |
6857 | return 0; | |
6858 | ||
6859 | printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n", | |
6860 | default_bootup_tracer); | |
6861 | default_bootup_tracer = NULL; | |
6862 | ||
6863 | return 0; | |
6864 | } | |
6865 | ||
6866 | early_initcall(tracer_alloc_buffers); | |
6867 | fs_initcall(tracer_init_debugfs); | |
6868 | late_initcall(clear_boot_tracer); |