]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * ring buffer based function tracer | |
3 | * | |
4 | * Copyright (C) 2007-2008 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 William Lee Irwin III | |
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 | ||
42 | #include "trace.h" | |
43 | #include "trace_output.h" | |
44 | ||
45 | /* | |
46 | * On boot up, the ring buffer is set to the minimum size, so that | |
47 | * we do not waste memory on systems that are not using tracing. | |
48 | */ | |
49 | int ring_buffer_expanded; | |
50 | ||
51 | /* | |
52 | * We need to change this state when a selftest is running. | |
53 | * A selftest will lurk into the ring-buffer to count the | |
54 | * entries inserted during the selftest although some concurrent | |
55 | * insertions into the ring-buffer such as trace_printk could occurred | |
56 | * at the same time, giving false positive or negative results. | |
57 | */ | |
58 | static bool __read_mostly tracing_selftest_running; | |
59 | ||
60 | /* | |
61 | * If a tracer is running, we do not want to run SELFTEST. | |
62 | */ | |
63 | bool __read_mostly tracing_selftest_disabled; | |
64 | ||
65 | /* For tracers that don't implement custom flags */ | |
66 | static struct tracer_opt dummy_tracer_opt[] = { | |
67 | { } | |
68 | }; | |
69 | ||
70 | static struct tracer_flags dummy_tracer_flags = { | |
71 | .val = 0, | |
72 | .opts = dummy_tracer_opt | |
73 | }; | |
74 | ||
75 | static int dummy_set_flag(u32 old_flags, u32 bit, int set) | |
76 | { | |
77 | return 0; | |
78 | } | |
79 | ||
80 | /* | |
81 | * Kill all tracing for good (never come back). | |
82 | * It is initialized to 1 but will turn to zero if the initialization | |
83 | * of the tracer is successful. But that is the only place that sets | |
84 | * this back to zero. | |
85 | */ | |
86 | static int tracing_disabled = 1; | |
87 | ||
88 | DEFINE_PER_CPU(int, ftrace_cpu_disabled); | |
89 | ||
90 | cpumask_var_t __read_mostly tracing_buffer_mask; | |
91 | ||
92 | /* | |
93 | * ftrace_dump_on_oops - variable to dump ftrace buffer on oops | |
94 | * | |
95 | * If there is an oops (or kernel panic) and the ftrace_dump_on_oops | |
96 | * is set, then ftrace_dump is called. This will output the contents | |
97 | * of the ftrace buffers to the console. This is very useful for | |
98 | * capturing traces that lead to crashes and outputing it to a | |
99 | * serial console. | |
100 | * | |
101 | * It is default off, but you can enable it with either specifying | |
102 | * "ftrace_dump_on_oops" in the kernel command line, or setting | |
103 | * /proc/sys/kernel/ftrace_dump_on_oops | |
104 | * Set 1 if you want to dump buffers of all CPUs | |
105 | * Set 2 if you want to dump the buffer of the CPU that triggered oops | |
106 | */ | |
107 | ||
108 | enum ftrace_dump_mode ftrace_dump_on_oops; | |
109 | ||
110 | static int tracing_set_tracer(const char *buf); | |
111 | ||
112 | #define MAX_TRACER_SIZE 100 | |
113 | static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata; | |
114 | static char *default_bootup_tracer; | |
115 | ||
116 | static int __init set_cmdline_ftrace(char *str) | |
117 | { | |
118 | strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE); | |
119 | default_bootup_tracer = bootup_tracer_buf; | |
120 | /* We are using ftrace early, expand it */ | |
121 | ring_buffer_expanded = 1; | |
122 | return 1; | |
123 | } | |
124 | __setup("ftrace=", set_cmdline_ftrace); | |
125 | ||
126 | static int __init set_ftrace_dump_on_oops(char *str) | |
127 | { | |
128 | if (*str++ != '=' || !*str) { | |
129 | ftrace_dump_on_oops = DUMP_ALL; | |
130 | return 1; | |
131 | } | |
132 | ||
133 | if (!strcmp("orig_cpu", str)) { | |
134 | ftrace_dump_on_oops = DUMP_ORIG; | |
135 | return 1; | |
136 | } | |
137 | ||
138 | return 0; | |
139 | } | |
140 | __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops); | |
141 | ||
142 | unsigned long long ns2usecs(cycle_t nsec) | |
143 | { | |
144 | nsec += 500; | |
145 | do_div(nsec, 1000); | |
146 | return nsec; | |
147 | } | |
148 | ||
149 | /* | |
150 | * The global_trace is the descriptor that holds the tracing | |
151 | * buffers for the live tracing. For each CPU, it contains | |
152 | * a link list of pages that will store trace entries. The | |
153 | * page descriptor of the pages in the memory is used to hold | |
154 | * the link list by linking the lru item in the page descriptor | |
155 | * to each of the pages in the buffer per CPU. | |
156 | * | |
157 | * For each active CPU there is a data field that holds the | |
158 | * pages for the buffer for that CPU. Each CPU has the same number | |
159 | * of pages allocated for its buffer. | |
160 | */ | |
161 | static struct trace_array global_trace; | |
162 | ||
163 | static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu); | |
164 | ||
165 | int filter_current_check_discard(struct ring_buffer *buffer, | |
166 | struct ftrace_event_call *call, void *rec, | |
167 | struct ring_buffer_event *event) | |
168 | { | |
169 | return filter_check_discard(call, rec, buffer, event); | |
170 | } | |
171 | EXPORT_SYMBOL_GPL(filter_current_check_discard); | |
172 | ||
173 | cycle_t ftrace_now(int cpu) | |
174 | { | |
175 | u64 ts; | |
176 | ||
177 | /* Early boot up does not have a buffer yet */ | |
178 | if (!global_trace.buffer) | |
179 | return trace_clock_local(); | |
180 | ||
181 | ts = ring_buffer_time_stamp(global_trace.buffer, cpu); | |
182 | ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts); | |
183 | ||
184 | return ts; | |
185 | } | |
186 | ||
187 | /* | |
188 | * The max_tr is used to snapshot the global_trace when a maximum | |
189 | * latency is reached. Some tracers will use this to store a maximum | |
190 | * trace while it continues examining live traces. | |
191 | * | |
192 | * The buffers for the max_tr are set up the same as the global_trace. | |
193 | * When a snapshot is taken, the link list of the max_tr is swapped | |
194 | * with the link list of the global_trace and the buffers are reset for | |
195 | * the global_trace so the tracing can continue. | |
196 | */ | |
197 | static struct trace_array max_tr; | |
198 | ||
199 | static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data); | |
200 | ||
201 | /* tracer_enabled is used to toggle activation of a tracer */ | |
202 | static int tracer_enabled = 1; | |
203 | ||
204 | /** | |
205 | * tracing_is_enabled - return tracer_enabled status | |
206 | * | |
207 | * This function is used by other tracers to know the status | |
208 | * of the tracer_enabled flag. Tracers may use this function | |
209 | * to know if it should enable their features when starting | |
210 | * up. See irqsoff tracer for an example (start_irqsoff_tracer). | |
211 | */ | |
212 | int tracing_is_enabled(void) | |
213 | { | |
214 | return tracer_enabled; | |
215 | } | |
216 | ||
217 | /* | |
218 | * trace_buf_size is the size in bytes that is allocated | |
219 | * for a buffer. Note, the number of bytes is always rounded | |
220 | * to page size. | |
221 | * | |
222 | * This number is purposely set to a low number of 16384. | |
223 | * If the dump on oops happens, it will be much appreciated | |
224 | * to not have to wait for all that output. Anyway this can be | |
225 | * boot time and run time configurable. | |
226 | */ | |
227 | #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */ | |
228 | ||
229 | static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT; | |
230 | ||
231 | /* trace_types holds a link list of available tracers. */ | |
232 | static struct tracer *trace_types __read_mostly; | |
233 | ||
234 | /* current_trace points to the tracer that is currently active */ | |
235 | static struct tracer *current_trace __read_mostly; | |
236 | ||
237 | /* | |
238 | * trace_types_lock is used to protect the trace_types list. | |
239 | */ | |
240 | static DEFINE_MUTEX(trace_types_lock); | |
241 | ||
242 | /* | |
243 | * serialize the access of the ring buffer | |
244 | * | |
245 | * ring buffer serializes readers, but it is low level protection. | |
246 | * The validity of the events (which returns by ring_buffer_peek() ..etc) | |
247 | * are not protected by ring buffer. | |
248 | * | |
249 | * The content of events may become garbage if we allow other process consumes | |
250 | * these events concurrently: | |
251 | * A) the page of the consumed events may become a normal page | |
252 | * (not reader page) in ring buffer, and this page will be rewrited | |
253 | * by events producer. | |
254 | * B) The page of the consumed events may become a page for splice_read, | |
255 | * and this page will be returned to system. | |
256 | * | |
257 | * These primitives allow multi process access to different cpu ring buffer | |
258 | * concurrently. | |
259 | * | |
260 | * These primitives don't distinguish read-only and read-consume access. | |
261 | * Multi read-only access are also serialized. | |
262 | */ | |
263 | ||
264 | #ifdef CONFIG_SMP | |
265 | static DECLARE_RWSEM(all_cpu_access_lock); | |
266 | static DEFINE_PER_CPU(struct mutex, cpu_access_lock); | |
267 | ||
268 | static inline void trace_access_lock(int cpu) | |
269 | { | |
270 | if (cpu == TRACE_PIPE_ALL_CPU) { | |
271 | /* gain it for accessing the whole ring buffer. */ | |
272 | down_write(&all_cpu_access_lock); | |
273 | } else { | |
274 | /* gain it for accessing a cpu ring buffer. */ | |
275 | ||
276 | /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */ | |
277 | down_read(&all_cpu_access_lock); | |
278 | ||
279 | /* Secondly block other access to this @cpu ring buffer. */ | |
280 | mutex_lock(&per_cpu(cpu_access_lock, cpu)); | |
281 | } | |
282 | } | |
283 | ||
284 | static inline void trace_access_unlock(int cpu) | |
285 | { | |
286 | if (cpu == TRACE_PIPE_ALL_CPU) { | |
287 | up_write(&all_cpu_access_lock); | |
288 | } else { | |
289 | mutex_unlock(&per_cpu(cpu_access_lock, cpu)); | |
290 | up_read(&all_cpu_access_lock); | |
291 | } | |
292 | } | |
293 | ||
294 | static inline void trace_access_lock_init(void) | |
295 | { | |
296 | int cpu; | |
297 | ||
298 | for_each_possible_cpu(cpu) | |
299 | mutex_init(&per_cpu(cpu_access_lock, cpu)); | |
300 | } | |
301 | ||
302 | #else | |
303 | ||
304 | static DEFINE_MUTEX(access_lock); | |
305 | ||
306 | static inline void trace_access_lock(int cpu) | |
307 | { | |
308 | (void)cpu; | |
309 | mutex_lock(&access_lock); | |
310 | } | |
311 | ||
312 | static inline void trace_access_unlock(int cpu) | |
313 | { | |
314 | (void)cpu; | |
315 | mutex_unlock(&access_lock); | |
316 | } | |
317 | ||
318 | static inline void trace_access_lock_init(void) | |
319 | { | |
320 | } | |
321 | ||
322 | #endif | |
323 | ||
324 | /* trace_wait is a waitqueue for tasks blocked on trace_poll */ | |
325 | static DECLARE_WAIT_QUEUE_HEAD(trace_wait); | |
326 | ||
327 | /* trace_flags holds trace_options default values */ | |
328 | unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | | |
329 | TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME | | |
330 | TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | | |
331 | TRACE_ITER_IRQ_INFO; | |
332 | ||
333 | static int trace_stop_count; | |
334 | static DEFINE_RAW_SPINLOCK(tracing_start_lock); | |
335 | ||
336 | static void wakeup_work_handler(struct work_struct *work) | |
337 | { | |
338 | wake_up(&trace_wait); | |
339 | } | |
340 | ||
341 | static DECLARE_DELAYED_WORK(wakeup_work, wakeup_work_handler); | |
342 | ||
343 | /** | |
344 | * tracing_on - enable tracing buffers | |
345 | * | |
346 | * This function enables tracing buffers that may have been | |
347 | * disabled with tracing_off. | |
348 | */ | |
349 | void tracing_on(void) | |
350 | { | |
351 | if (global_trace.buffer) | |
352 | ring_buffer_record_on(global_trace.buffer); | |
353 | /* | |
354 | * This flag is only looked at when buffers haven't been | |
355 | * allocated yet. We don't really care about the race | |
356 | * between setting this flag and actually turning | |
357 | * on the buffer. | |
358 | */ | |
359 | global_trace.buffer_disabled = 0; | |
360 | } | |
361 | EXPORT_SYMBOL_GPL(tracing_on); | |
362 | ||
363 | /** | |
364 | * tracing_off - turn off tracing buffers | |
365 | * | |
366 | * This function stops the tracing buffers from recording data. | |
367 | * It does not disable any overhead the tracers themselves may | |
368 | * be causing. This function simply causes all recording to | |
369 | * the ring buffers to fail. | |
370 | */ | |
371 | void tracing_off(void) | |
372 | { | |
373 | if (global_trace.buffer) | |
374 | ring_buffer_record_off(global_trace.buffer); | |
375 | /* | |
376 | * This flag is only looked at when buffers haven't been | |
377 | * allocated yet. We don't really care about the race | |
378 | * between setting this flag and actually turning | |
379 | * on the buffer. | |
380 | */ | |
381 | global_trace.buffer_disabled = 1; | |
382 | } | |
383 | EXPORT_SYMBOL_GPL(tracing_off); | |
384 | ||
385 | /** | |
386 | * tracing_is_on - show state of ring buffers enabled | |
387 | */ | |
388 | int tracing_is_on(void) | |
389 | { | |
390 | if (global_trace.buffer) | |
391 | return ring_buffer_record_is_on(global_trace.buffer); | |
392 | return !global_trace.buffer_disabled; | |
393 | } | |
394 | EXPORT_SYMBOL_GPL(tracing_is_on); | |
395 | ||
396 | /** | |
397 | * trace_wake_up - wake up tasks waiting for trace input | |
398 | * | |
399 | * Schedules a delayed work to wake up any task that is blocked on the | |
400 | * trace_wait queue. These is used with trace_poll for tasks polling the | |
401 | * trace. | |
402 | */ | |
403 | void trace_wake_up(void) | |
404 | { | |
405 | const unsigned long delay = msecs_to_jiffies(2); | |
406 | ||
407 | if (trace_flags & TRACE_ITER_BLOCK) | |
408 | return; | |
409 | schedule_delayed_work(&wakeup_work, delay); | |
410 | } | |
411 | ||
412 | static int __init set_buf_size(char *str) | |
413 | { | |
414 | unsigned long buf_size; | |
415 | ||
416 | if (!str) | |
417 | return 0; | |
418 | buf_size = memparse(str, &str); | |
419 | /* nr_entries can not be zero */ | |
420 | if (buf_size == 0) | |
421 | return 0; | |
422 | trace_buf_size = buf_size; | |
423 | return 1; | |
424 | } | |
425 | __setup("trace_buf_size=", set_buf_size); | |
426 | ||
427 | static int __init set_tracing_thresh(char *str) | |
428 | { | |
429 | unsigned long threshhold; | |
430 | int ret; | |
431 | ||
432 | if (!str) | |
433 | return 0; | |
434 | ret = strict_strtoul(str, 0, &threshhold); | |
435 | if (ret < 0) | |
436 | return 0; | |
437 | tracing_thresh = threshhold * 1000; | |
438 | return 1; | |
439 | } | |
440 | __setup("tracing_thresh=", set_tracing_thresh); | |
441 | ||
442 | unsigned long nsecs_to_usecs(unsigned long nsecs) | |
443 | { | |
444 | return nsecs / 1000; | |
445 | } | |
446 | ||
447 | /* These must match the bit postions in trace_iterator_flags */ | |
448 | static const char *trace_options[] = { | |
449 | "print-parent", | |
450 | "sym-offset", | |
451 | "sym-addr", | |
452 | "verbose", | |
453 | "raw", | |
454 | "hex", | |
455 | "bin", | |
456 | "block", | |
457 | "stacktrace", | |
458 | "trace_printk", | |
459 | "ftrace_preempt", | |
460 | "branch", | |
461 | "annotate", | |
462 | "userstacktrace", | |
463 | "sym-userobj", | |
464 | "printk-msg-only", | |
465 | "context-info", | |
466 | "latency-format", | |
467 | "sleep-time", | |
468 | "graph-time", | |
469 | "record-cmd", | |
470 | "overwrite", | |
471 | "disable_on_free", | |
472 | "irq-info", | |
473 | NULL | |
474 | }; | |
475 | ||
476 | static struct { | |
477 | u64 (*func)(void); | |
478 | const char *name; | |
479 | } trace_clocks[] = { | |
480 | { trace_clock_local, "local" }, | |
481 | { trace_clock_global, "global" }, | |
482 | { trace_clock_counter, "counter" }, | |
483 | }; | |
484 | ||
485 | int trace_clock_id; | |
486 | ||
487 | /* | |
488 | * trace_parser_get_init - gets the buffer for trace parser | |
489 | */ | |
490 | int trace_parser_get_init(struct trace_parser *parser, int size) | |
491 | { | |
492 | memset(parser, 0, sizeof(*parser)); | |
493 | ||
494 | parser->buffer = kmalloc(size, GFP_KERNEL); | |
495 | if (!parser->buffer) | |
496 | return 1; | |
497 | ||
498 | parser->size = size; | |
499 | return 0; | |
500 | } | |
501 | ||
502 | /* | |
503 | * trace_parser_put - frees the buffer for trace parser | |
504 | */ | |
505 | void trace_parser_put(struct trace_parser *parser) | |
506 | { | |
507 | kfree(parser->buffer); | |
508 | } | |
509 | ||
510 | /* | |
511 | * trace_get_user - reads the user input string separated by space | |
512 | * (matched by isspace(ch)) | |
513 | * | |
514 | * For each string found the 'struct trace_parser' is updated, | |
515 | * and the function returns. | |
516 | * | |
517 | * Returns number of bytes read. | |
518 | * | |
519 | * See kernel/trace/trace.h for 'struct trace_parser' details. | |
520 | */ | |
521 | int trace_get_user(struct trace_parser *parser, const char __user *ubuf, | |
522 | size_t cnt, loff_t *ppos) | |
523 | { | |
524 | char ch; | |
525 | size_t read = 0; | |
526 | ssize_t ret; | |
527 | ||
528 | if (!*ppos) | |
529 | trace_parser_clear(parser); | |
530 | ||
531 | ret = get_user(ch, ubuf++); | |
532 | if (ret) | |
533 | goto out; | |
534 | ||
535 | read++; | |
536 | cnt--; | |
537 | ||
538 | /* | |
539 | * The parser is not finished with the last write, | |
540 | * continue reading the user input without skipping spaces. | |
541 | */ | |
542 | if (!parser->cont) { | |
543 | /* skip white space */ | |
544 | while (cnt && isspace(ch)) { | |
545 | ret = get_user(ch, ubuf++); | |
546 | if (ret) | |
547 | goto out; | |
548 | read++; | |
549 | cnt--; | |
550 | } | |
551 | ||
552 | /* only spaces were written */ | |
553 | if (isspace(ch)) { | |
554 | *ppos += read; | |
555 | ret = read; | |
556 | goto out; | |
557 | } | |
558 | ||
559 | parser->idx = 0; | |
560 | } | |
561 | ||
562 | /* read the non-space input */ | |
563 | while (cnt && !isspace(ch)) { | |
564 | if (parser->idx < parser->size - 1) | |
565 | parser->buffer[parser->idx++] = ch; | |
566 | else { | |
567 | ret = -EINVAL; | |
568 | goto out; | |
569 | } | |
570 | ret = get_user(ch, ubuf++); | |
571 | if (ret) | |
572 | goto out; | |
573 | read++; | |
574 | cnt--; | |
575 | } | |
576 | ||
577 | /* We either got finished input or we have to wait for another call. */ | |
578 | if (isspace(ch)) { | |
579 | parser->buffer[parser->idx] = 0; | |
580 | parser->cont = false; | |
581 | } else { | |
582 | parser->cont = true; | |
583 | parser->buffer[parser->idx++] = ch; | |
584 | } | |
585 | ||
586 | *ppos += read; | |
587 | ret = read; | |
588 | ||
589 | out: | |
590 | return ret; | |
591 | } | |
592 | ||
593 | ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt) | |
594 | { | |
595 | int len; | |
596 | int ret; | |
597 | ||
598 | if (!cnt) | |
599 | return 0; | |
600 | ||
601 | if (s->len <= s->readpos) | |
602 | return -EBUSY; | |
603 | ||
604 | len = s->len - s->readpos; | |
605 | if (cnt > len) | |
606 | cnt = len; | |
607 | ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt); | |
608 | if (ret == cnt) | |
609 | return -EFAULT; | |
610 | ||
611 | cnt -= ret; | |
612 | ||
613 | s->readpos += cnt; | |
614 | return cnt; | |
615 | } | |
616 | ||
617 | static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt) | |
618 | { | |
619 | int len; | |
620 | ||
621 | if (s->len <= s->readpos) | |
622 | return -EBUSY; | |
623 | ||
624 | len = s->len - s->readpos; | |
625 | if (cnt > len) | |
626 | cnt = len; | |
627 | memcpy(buf, s->buffer + s->readpos, cnt); | |
628 | ||
629 | s->readpos += cnt; | |
630 | return cnt; | |
631 | } | |
632 | ||
633 | /* | |
634 | * ftrace_max_lock is used to protect the swapping of buffers | |
635 | * when taking a max snapshot. The buffers themselves are | |
636 | * protected by per_cpu spinlocks. But the action of the swap | |
637 | * needs its own lock. | |
638 | * | |
639 | * This is defined as a arch_spinlock_t in order to help | |
640 | * with performance when lockdep debugging is enabled. | |
641 | * | |
642 | * It is also used in other places outside the update_max_tr | |
643 | * so it needs to be defined outside of the | |
644 | * CONFIG_TRACER_MAX_TRACE. | |
645 | */ | |
646 | static arch_spinlock_t ftrace_max_lock = | |
647 | (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; | |
648 | ||
649 | unsigned long __read_mostly tracing_thresh; | |
650 | ||
651 | #ifdef CONFIG_TRACER_MAX_TRACE | |
652 | unsigned long __read_mostly tracing_max_latency; | |
653 | ||
654 | /* | |
655 | * Copy the new maximum trace into the separate maximum-trace | |
656 | * structure. (this way the maximum trace is permanently saved, | |
657 | * for later retrieval via /sys/kernel/debug/tracing/latency_trace) | |
658 | */ | |
659 | static void | |
660 | __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) | |
661 | { | |
662 | struct trace_array_cpu *data = tr->data[cpu]; | |
663 | struct trace_array_cpu *max_data; | |
664 | ||
665 | max_tr.cpu = cpu; | |
666 | max_tr.time_start = data->preempt_timestamp; | |
667 | ||
668 | max_data = max_tr.data[cpu]; | |
669 | max_data->saved_latency = tracing_max_latency; | |
670 | max_data->critical_start = data->critical_start; | |
671 | max_data->critical_end = data->critical_end; | |
672 | ||
673 | memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN); | |
674 | max_data->pid = tsk->pid; | |
675 | max_data->uid = task_uid(tsk); | |
676 | max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO; | |
677 | max_data->policy = tsk->policy; | |
678 | max_data->rt_priority = tsk->rt_priority; | |
679 | ||
680 | /* record this tasks comm */ | |
681 | tracing_record_cmdline(tsk); | |
682 | } | |
683 | ||
684 | /** | |
685 | * update_max_tr - snapshot all trace buffers from global_trace to max_tr | |
686 | * @tr: tracer | |
687 | * @tsk: the task with the latency | |
688 | * @cpu: The cpu that initiated the trace. | |
689 | * | |
690 | * Flip the buffers between the @tr and the max_tr and record information | |
691 | * about which task was the cause of this latency. | |
692 | */ | |
693 | void | |
694 | update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) | |
695 | { | |
696 | struct ring_buffer *buf = tr->buffer; | |
697 | ||
698 | if (trace_stop_count) | |
699 | return; | |
700 | ||
701 | WARN_ON_ONCE(!irqs_disabled()); | |
702 | if (!current_trace->use_max_tr) { | |
703 | WARN_ON_ONCE(1); | |
704 | return; | |
705 | } | |
706 | arch_spin_lock(&ftrace_max_lock); | |
707 | ||
708 | tr->buffer = max_tr.buffer; | |
709 | max_tr.buffer = buf; | |
710 | ||
711 | __update_max_tr(tr, tsk, cpu); | |
712 | arch_spin_unlock(&ftrace_max_lock); | |
713 | } | |
714 | ||
715 | /** | |
716 | * update_max_tr_single - only copy one trace over, and reset the rest | |
717 | * @tr - tracer | |
718 | * @tsk - task with the latency | |
719 | * @cpu - the cpu of the buffer to copy. | |
720 | * | |
721 | * Flip the trace of a single CPU buffer between the @tr and the max_tr. | |
722 | */ | |
723 | void | |
724 | update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu) | |
725 | { | |
726 | int ret; | |
727 | ||
728 | if (trace_stop_count) | |
729 | return; | |
730 | ||
731 | WARN_ON_ONCE(!irqs_disabled()); | |
732 | if (!current_trace->use_max_tr) { | |
733 | WARN_ON_ONCE(1); | |
734 | return; | |
735 | } | |
736 | ||
737 | arch_spin_lock(&ftrace_max_lock); | |
738 | ||
739 | ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu); | |
740 | ||
741 | if (ret == -EBUSY) { | |
742 | /* | |
743 | * We failed to swap the buffer due to a commit taking | |
744 | * place on this CPU. We fail to record, but we reset | |
745 | * the max trace buffer (no one writes directly to it) | |
746 | * and flag that it failed. | |
747 | */ | |
748 | trace_array_printk(&max_tr, _THIS_IP_, | |
749 | "Failed to swap buffers due to commit in progress\n"); | |
750 | } | |
751 | ||
752 | WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY); | |
753 | ||
754 | __update_max_tr(tr, tsk, cpu); | |
755 | arch_spin_unlock(&ftrace_max_lock); | |
756 | } | |
757 | #endif /* CONFIG_TRACER_MAX_TRACE */ | |
758 | ||
759 | /** | |
760 | * register_tracer - register a tracer with the ftrace system. | |
761 | * @type - the plugin for the tracer | |
762 | * | |
763 | * Register a new plugin tracer. | |
764 | */ | |
765 | int register_tracer(struct tracer *type) | |
766 | { | |
767 | struct tracer *t; | |
768 | int ret = 0; | |
769 | ||
770 | if (!type->name) { | |
771 | pr_info("Tracer must have a name\n"); | |
772 | return -1; | |
773 | } | |
774 | ||
775 | if (strlen(type->name) >= MAX_TRACER_SIZE) { | |
776 | pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE); | |
777 | return -1; | |
778 | } | |
779 | ||
780 | mutex_lock(&trace_types_lock); | |
781 | ||
782 | tracing_selftest_running = true; | |
783 | ||
784 | for (t = trace_types; t; t = t->next) { | |
785 | if (strcmp(type->name, t->name) == 0) { | |
786 | /* already found */ | |
787 | pr_info("Tracer %s already registered\n", | |
788 | type->name); | |
789 | ret = -1; | |
790 | goto out; | |
791 | } | |
792 | } | |
793 | ||
794 | if (!type->set_flag) | |
795 | type->set_flag = &dummy_set_flag; | |
796 | if (!type->flags) | |
797 | type->flags = &dummy_tracer_flags; | |
798 | else | |
799 | if (!type->flags->opts) | |
800 | type->flags->opts = dummy_tracer_opt; | |
801 | if (!type->wait_pipe) | |
802 | type->wait_pipe = default_wait_pipe; | |
803 | ||
804 | ||
805 | #ifdef CONFIG_FTRACE_STARTUP_TEST | |
806 | if (type->selftest && !tracing_selftest_disabled) { | |
807 | struct tracer *saved_tracer = current_trace; | |
808 | struct trace_array *tr = &global_trace; | |
809 | ||
810 | /* | |
811 | * Run a selftest on this tracer. | |
812 | * Here we reset the trace buffer, and set the current | |
813 | * tracer to be this tracer. The tracer can then run some | |
814 | * internal tracing to verify that everything is in order. | |
815 | * If we fail, we do not register this tracer. | |
816 | */ | |
817 | tracing_reset_online_cpus(tr); | |
818 | ||
819 | current_trace = type; | |
820 | ||
821 | /* If we expanded the buffers, make sure the max is expanded too */ | |
822 | if (ring_buffer_expanded && type->use_max_tr) | |
823 | ring_buffer_resize(max_tr.buffer, trace_buf_size, | |
824 | RING_BUFFER_ALL_CPUS); | |
825 | ||
826 | /* the test is responsible for initializing and enabling */ | |
827 | pr_info("Testing tracer %s: ", type->name); | |
828 | ret = type->selftest(type, tr); | |
829 | /* the test is responsible for resetting too */ | |
830 | current_trace = saved_tracer; | |
831 | if (ret) { | |
832 | printk(KERN_CONT "FAILED!\n"); | |
833 | /* Add the warning after printing 'FAILED' */ | |
834 | WARN_ON(1); | |
835 | goto out; | |
836 | } | |
837 | /* Only reset on passing, to avoid touching corrupted buffers */ | |
838 | tracing_reset_online_cpus(tr); | |
839 | ||
840 | /* Shrink the max buffer again */ | |
841 | if (ring_buffer_expanded && type->use_max_tr) | |
842 | ring_buffer_resize(max_tr.buffer, 1, | |
843 | RING_BUFFER_ALL_CPUS); | |
844 | ||
845 | printk(KERN_CONT "PASSED\n"); | |
846 | } | |
847 | #endif | |
848 | ||
849 | type->next = trace_types; | |
850 | trace_types = type; | |
851 | ||
852 | out: | |
853 | tracing_selftest_running = false; | |
854 | mutex_unlock(&trace_types_lock); | |
855 | ||
856 | if (ret || !default_bootup_tracer) | |
857 | goto out_unlock; | |
858 | ||
859 | if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE)) | |
860 | goto out_unlock; | |
861 | ||
862 | printk(KERN_INFO "Starting tracer '%s'\n", type->name); | |
863 | /* Do we want this tracer to start on bootup? */ | |
864 | tracing_set_tracer(type->name); | |
865 | default_bootup_tracer = NULL; | |
866 | /* disable other selftests, since this will break it. */ | |
867 | tracing_selftest_disabled = 1; | |
868 | #ifdef CONFIG_FTRACE_STARTUP_TEST | |
869 | printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n", | |
870 | type->name); | |
871 | #endif | |
872 | ||
873 | out_unlock: | |
874 | return ret; | |
875 | } | |
876 | ||
877 | void unregister_tracer(struct tracer *type) | |
878 | { | |
879 | struct tracer **t; | |
880 | ||
881 | mutex_lock(&trace_types_lock); | |
882 | for (t = &trace_types; *t; t = &(*t)->next) { | |
883 | if (*t == type) | |
884 | goto found; | |
885 | } | |
886 | pr_info("Tracer %s not registered\n", type->name); | |
887 | goto out; | |
888 | ||
889 | found: | |
890 | *t = (*t)->next; | |
891 | ||
892 | if (type == current_trace && tracer_enabled) { | |
893 | tracer_enabled = 0; | |
894 | tracing_stop(); | |
895 | if (current_trace->stop) | |
896 | current_trace->stop(&global_trace); | |
897 | current_trace = &nop_trace; | |
898 | } | |
899 | out: | |
900 | mutex_unlock(&trace_types_lock); | |
901 | } | |
902 | ||
903 | void tracing_reset(struct trace_array *tr, int cpu) | |
904 | { | |
905 | struct ring_buffer *buffer = tr->buffer; | |
906 | ||
907 | ring_buffer_record_disable(buffer); | |
908 | ||
909 | /* Make sure all commits have finished */ | |
910 | synchronize_sched(); | |
911 | ring_buffer_reset_cpu(buffer, cpu); | |
912 | ||
913 | ring_buffer_record_enable(buffer); | |
914 | } | |
915 | ||
916 | void tracing_reset_online_cpus(struct trace_array *tr) | |
917 | { | |
918 | struct ring_buffer *buffer = tr->buffer; | |
919 | int cpu; | |
920 | ||
921 | ring_buffer_record_disable(buffer); | |
922 | ||
923 | /* Make sure all commits have finished */ | |
924 | synchronize_sched(); | |
925 | ||
926 | tr->time_start = ftrace_now(tr->cpu); | |
927 | ||
928 | for_each_online_cpu(cpu) | |
929 | ring_buffer_reset_cpu(buffer, cpu); | |
930 | ||
931 | ring_buffer_record_enable(buffer); | |
932 | } | |
933 | ||
934 | void tracing_reset_current(int cpu) | |
935 | { | |
936 | tracing_reset(&global_trace, cpu); | |
937 | } | |
938 | ||
939 | void tracing_reset_current_online_cpus(void) | |
940 | { | |
941 | tracing_reset_online_cpus(&global_trace); | |
942 | } | |
943 | ||
944 | #define SAVED_CMDLINES 128 | |
945 | #define NO_CMDLINE_MAP UINT_MAX | |
946 | static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1]; | |
947 | static unsigned map_cmdline_to_pid[SAVED_CMDLINES]; | |
948 | static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN]; | |
949 | static int cmdline_idx; | |
950 | static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED; | |
951 | ||
952 | /* temporary disable recording */ | |
953 | static atomic_t trace_record_cmdline_disabled __read_mostly; | |
954 | ||
955 | static void trace_init_cmdlines(void) | |
956 | { | |
957 | memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline)); | |
958 | memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid)); | |
959 | cmdline_idx = 0; | |
960 | } | |
961 | ||
962 | int is_tracing_stopped(void) | |
963 | { | |
964 | return trace_stop_count; | |
965 | } | |
966 | ||
967 | /** | |
968 | * ftrace_off_permanent - disable all ftrace code permanently | |
969 | * | |
970 | * This should only be called when a serious anomally has | |
971 | * been detected. This will turn off the function tracing, | |
972 | * ring buffers, and other tracing utilites. It takes no | |
973 | * locks and can be called from any context. | |
974 | */ | |
975 | void ftrace_off_permanent(void) | |
976 | { | |
977 | tracing_disabled = 1; | |
978 | ftrace_stop(); | |
979 | tracing_off_permanent(); | |
980 | } | |
981 | ||
982 | /** | |
983 | * tracing_start - quick start of the tracer | |
984 | * | |
985 | * If tracing is enabled but was stopped by tracing_stop, | |
986 | * this will start the tracer back up. | |
987 | */ | |
988 | void tracing_start(void) | |
989 | { | |
990 | struct ring_buffer *buffer; | |
991 | unsigned long flags; | |
992 | ||
993 | if (tracing_disabled) | |
994 | return; | |
995 | ||
996 | raw_spin_lock_irqsave(&tracing_start_lock, flags); | |
997 | if (--trace_stop_count) { | |
998 | if (trace_stop_count < 0) { | |
999 | /* Someone screwed up their debugging */ | |
1000 | WARN_ON_ONCE(1); | |
1001 | trace_stop_count = 0; | |
1002 | } | |
1003 | goto out; | |
1004 | } | |
1005 | ||
1006 | /* Prevent the buffers from switching */ | |
1007 | arch_spin_lock(&ftrace_max_lock); | |
1008 | ||
1009 | buffer = global_trace.buffer; | |
1010 | if (buffer) | |
1011 | ring_buffer_record_enable(buffer); | |
1012 | ||
1013 | buffer = max_tr.buffer; | |
1014 | if (buffer) | |
1015 | ring_buffer_record_enable(buffer); | |
1016 | ||
1017 | arch_spin_unlock(&ftrace_max_lock); | |
1018 | ||
1019 | ftrace_start(); | |
1020 | out: | |
1021 | raw_spin_unlock_irqrestore(&tracing_start_lock, flags); | |
1022 | } | |
1023 | ||
1024 | /** | |
1025 | * tracing_stop - quick stop of the tracer | |
1026 | * | |
1027 | * Light weight way to stop tracing. Use in conjunction with | |
1028 | * tracing_start. | |
1029 | */ | |
1030 | void tracing_stop(void) | |
1031 | { | |
1032 | struct ring_buffer *buffer; | |
1033 | unsigned long flags; | |
1034 | ||
1035 | ftrace_stop(); | |
1036 | raw_spin_lock_irqsave(&tracing_start_lock, flags); | |
1037 | if (trace_stop_count++) | |
1038 | goto out; | |
1039 | ||
1040 | /* Prevent the buffers from switching */ | |
1041 | arch_spin_lock(&ftrace_max_lock); | |
1042 | ||
1043 | buffer = global_trace.buffer; | |
1044 | if (buffer) | |
1045 | ring_buffer_record_disable(buffer); | |
1046 | ||
1047 | buffer = max_tr.buffer; | |
1048 | if (buffer) | |
1049 | ring_buffer_record_disable(buffer); | |
1050 | ||
1051 | arch_spin_unlock(&ftrace_max_lock); | |
1052 | ||
1053 | out: | |
1054 | raw_spin_unlock_irqrestore(&tracing_start_lock, flags); | |
1055 | } | |
1056 | ||
1057 | void trace_stop_cmdline_recording(void); | |
1058 | ||
1059 | static void trace_save_cmdline(struct task_struct *tsk) | |
1060 | { | |
1061 | unsigned pid, idx; | |
1062 | ||
1063 | if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT)) | |
1064 | return; | |
1065 | ||
1066 | /* | |
1067 | * It's not the end of the world if we don't get | |
1068 | * the lock, but we also don't want to spin | |
1069 | * nor do we want to disable interrupts, | |
1070 | * so if we miss here, then better luck next time. | |
1071 | */ | |
1072 | if (!arch_spin_trylock(&trace_cmdline_lock)) | |
1073 | return; | |
1074 | ||
1075 | idx = map_pid_to_cmdline[tsk->pid]; | |
1076 | if (idx == NO_CMDLINE_MAP) { | |
1077 | idx = (cmdline_idx + 1) % SAVED_CMDLINES; | |
1078 | ||
1079 | /* | |
1080 | * Check whether the cmdline buffer at idx has a pid | |
1081 | * mapped. We are going to overwrite that entry so we | |
1082 | * need to clear the map_pid_to_cmdline. Otherwise we | |
1083 | * would read the new comm for the old pid. | |
1084 | */ | |
1085 | pid = map_cmdline_to_pid[idx]; | |
1086 | if (pid != NO_CMDLINE_MAP) | |
1087 | map_pid_to_cmdline[pid] = NO_CMDLINE_MAP; | |
1088 | ||
1089 | map_cmdline_to_pid[idx] = tsk->pid; | |
1090 | map_pid_to_cmdline[tsk->pid] = idx; | |
1091 | ||
1092 | cmdline_idx = idx; | |
1093 | } | |
1094 | ||
1095 | memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN); | |
1096 | ||
1097 | arch_spin_unlock(&trace_cmdline_lock); | |
1098 | } | |
1099 | ||
1100 | void trace_find_cmdline(int pid, char comm[]) | |
1101 | { | |
1102 | unsigned map; | |
1103 | ||
1104 | if (!pid) { | |
1105 | strcpy(comm, "<idle>"); | |
1106 | return; | |
1107 | } | |
1108 | ||
1109 | if (WARN_ON_ONCE(pid < 0)) { | |
1110 | strcpy(comm, "<XXX>"); | |
1111 | return; | |
1112 | } | |
1113 | ||
1114 | if (pid > PID_MAX_DEFAULT) { | |
1115 | strcpy(comm, "<...>"); | |
1116 | return; | |
1117 | } | |
1118 | ||
1119 | preempt_disable(); | |
1120 | arch_spin_lock(&trace_cmdline_lock); | |
1121 | map = map_pid_to_cmdline[pid]; | |
1122 | if (map != NO_CMDLINE_MAP) | |
1123 | strcpy(comm, saved_cmdlines[map]); | |
1124 | else | |
1125 | strcpy(comm, "<...>"); | |
1126 | ||
1127 | arch_spin_unlock(&trace_cmdline_lock); | |
1128 | preempt_enable(); | |
1129 | } | |
1130 | ||
1131 | void tracing_record_cmdline(struct task_struct *tsk) | |
1132 | { | |
1133 | if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled || | |
1134 | !tracing_is_on()) | |
1135 | return; | |
1136 | ||
1137 | trace_save_cmdline(tsk); | |
1138 | } | |
1139 | ||
1140 | void | |
1141 | tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags, | |
1142 | int pc) | |
1143 | { | |
1144 | struct task_struct *tsk = current; | |
1145 | ||
1146 | entry->preempt_count = pc & 0xff; | |
1147 | entry->pid = (tsk) ? tsk->pid : 0; | |
1148 | entry->padding = 0; | |
1149 | entry->flags = | |
1150 | #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT | |
1151 | (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) | | |
1152 | #else | |
1153 | TRACE_FLAG_IRQS_NOSUPPORT | | |
1154 | #endif | |
1155 | ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) | | |
1156 | ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) | | |
1157 | (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0); | |
1158 | } | |
1159 | EXPORT_SYMBOL_GPL(tracing_generic_entry_update); | |
1160 | ||
1161 | struct ring_buffer_event * | |
1162 | trace_buffer_lock_reserve(struct ring_buffer *buffer, | |
1163 | int type, | |
1164 | unsigned long len, | |
1165 | unsigned long flags, int pc) | |
1166 | { | |
1167 | struct ring_buffer_event *event; | |
1168 | ||
1169 | event = ring_buffer_lock_reserve(buffer, len); | |
1170 | if (event != NULL) { | |
1171 | struct trace_entry *ent = ring_buffer_event_data(event); | |
1172 | ||
1173 | tracing_generic_entry_update(ent, flags, pc); | |
1174 | ent->type = type; | |
1175 | } | |
1176 | ||
1177 | return event; | |
1178 | } | |
1179 | ||
1180 | static inline void | |
1181 | __trace_buffer_unlock_commit(struct ring_buffer *buffer, | |
1182 | struct ring_buffer_event *event, | |
1183 | unsigned long flags, int pc, | |
1184 | int wake) | |
1185 | { | |
1186 | ring_buffer_unlock_commit(buffer, event); | |
1187 | ||
1188 | ftrace_trace_stack(buffer, flags, 6, pc); | |
1189 | ftrace_trace_userstack(buffer, flags, pc); | |
1190 | ||
1191 | if (wake) | |
1192 | trace_wake_up(); | |
1193 | } | |
1194 | ||
1195 | void trace_buffer_unlock_commit(struct ring_buffer *buffer, | |
1196 | struct ring_buffer_event *event, | |
1197 | unsigned long flags, int pc) | |
1198 | { | |
1199 | __trace_buffer_unlock_commit(buffer, event, flags, pc, 1); | |
1200 | } | |
1201 | ||
1202 | struct ring_buffer_event * | |
1203 | trace_current_buffer_lock_reserve(struct ring_buffer **current_rb, | |
1204 | int type, unsigned long len, | |
1205 | unsigned long flags, int pc) | |
1206 | { | |
1207 | *current_rb = global_trace.buffer; | |
1208 | return trace_buffer_lock_reserve(*current_rb, | |
1209 | type, len, flags, pc); | |
1210 | } | |
1211 | EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve); | |
1212 | ||
1213 | void trace_current_buffer_unlock_commit(struct ring_buffer *buffer, | |
1214 | struct ring_buffer_event *event, | |
1215 | unsigned long flags, int pc) | |
1216 | { | |
1217 | __trace_buffer_unlock_commit(buffer, event, flags, pc, 1); | |
1218 | } | |
1219 | EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit); | |
1220 | ||
1221 | void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer, | |
1222 | struct ring_buffer_event *event, | |
1223 | unsigned long flags, int pc) | |
1224 | { | |
1225 | __trace_buffer_unlock_commit(buffer, event, flags, pc, 0); | |
1226 | } | |
1227 | EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit); | |
1228 | ||
1229 | void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer *buffer, | |
1230 | struct ring_buffer_event *event, | |
1231 | unsigned long flags, int pc, | |
1232 | struct pt_regs *regs) | |
1233 | { | |
1234 | ring_buffer_unlock_commit(buffer, event); | |
1235 | ||
1236 | ftrace_trace_stack_regs(buffer, flags, 0, pc, regs); | |
1237 | ftrace_trace_userstack(buffer, flags, pc); | |
1238 | } | |
1239 | EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs); | |
1240 | ||
1241 | void trace_current_buffer_discard_commit(struct ring_buffer *buffer, | |
1242 | struct ring_buffer_event *event) | |
1243 | { | |
1244 | ring_buffer_discard_commit(buffer, event); | |
1245 | } | |
1246 | EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit); | |
1247 | ||
1248 | void | |
1249 | trace_function(struct trace_array *tr, | |
1250 | unsigned long ip, unsigned long parent_ip, unsigned long flags, | |
1251 | int pc) | |
1252 | { | |
1253 | struct ftrace_event_call *call = &event_function; | |
1254 | struct ring_buffer *buffer = tr->buffer; | |
1255 | struct ring_buffer_event *event; | |
1256 | struct ftrace_entry *entry; | |
1257 | ||
1258 | /* If we are reading the ring buffer, don't trace */ | |
1259 | if (unlikely(__this_cpu_read(ftrace_cpu_disabled))) | |
1260 | return; | |
1261 | ||
1262 | event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry), | |
1263 | flags, pc); | |
1264 | if (!event) | |
1265 | return; | |
1266 | entry = ring_buffer_event_data(event); | |
1267 | entry->ip = ip; | |
1268 | entry->parent_ip = parent_ip; | |
1269 | ||
1270 | if (!filter_check_discard(call, entry, buffer, event)) | |
1271 | ring_buffer_unlock_commit(buffer, event); | |
1272 | } | |
1273 | ||
1274 | void | |
1275 | ftrace(struct trace_array *tr, struct trace_array_cpu *data, | |
1276 | unsigned long ip, unsigned long parent_ip, unsigned long flags, | |
1277 | int pc) | |
1278 | { | |
1279 | if (likely(!atomic_read(&data->disabled))) | |
1280 | trace_function(tr, ip, parent_ip, flags, pc); | |
1281 | } | |
1282 | ||
1283 | #ifdef CONFIG_STACKTRACE | |
1284 | ||
1285 | #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long)) | |
1286 | struct ftrace_stack { | |
1287 | unsigned long calls[FTRACE_STACK_MAX_ENTRIES]; | |
1288 | }; | |
1289 | ||
1290 | static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack); | |
1291 | static DEFINE_PER_CPU(int, ftrace_stack_reserve); | |
1292 | ||
1293 | static void __ftrace_trace_stack(struct ring_buffer *buffer, | |
1294 | unsigned long flags, | |
1295 | int skip, int pc, struct pt_regs *regs) | |
1296 | { | |
1297 | struct ftrace_event_call *call = &event_kernel_stack; | |
1298 | struct ring_buffer_event *event; | |
1299 | struct stack_entry *entry; | |
1300 | struct stack_trace trace; | |
1301 | int use_stack; | |
1302 | int size = FTRACE_STACK_ENTRIES; | |
1303 | ||
1304 | trace.nr_entries = 0; | |
1305 | trace.skip = skip; | |
1306 | ||
1307 | /* | |
1308 | * Since events can happen in NMIs there's no safe way to | |
1309 | * use the per cpu ftrace_stacks. We reserve it and if an interrupt | |
1310 | * or NMI comes in, it will just have to use the default | |
1311 | * FTRACE_STACK_SIZE. | |
1312 | */ | |
1313 | preempt_disable_notrace(); | |
1314 | ||
1315 | use_stack = ++__get_cpu_var(ftrace_stack_reserve); | |
1316 | /* | |
1317 | * We don't need any atomic variables, just a barrier. | |
1318 | * If an interrupt comes in, we don't care, because it would | |
1319 | * have exited and put the counter back to what we want. | |
1320 | * We just need a barrier to keep gcc from moving things | |
1321 | * around. | |
1322 | */ | |
1323 | barrier(); | |
1324 | if (use_stack == 1) { | |
1325 | trace.entries = &__get_cpu_var(ftrace_stack).calls[0]; | |
1326 | trace.max_entries = FTRACE_STACK_MAX_ENTRIES; | |
1327 | ||
1328 | if (regs) | |
1329 | save_stack_trace_regs(regs, &trace); | |
1330 | else | |
1331 | save_stack_trace(&trace); | |
1332 | ||
1333 | if (trace.nr_entries > size) | |
1334 | size = trace.nr_entries; | |
1335 | } else | |
1336 | /* From now on, use_stack is a boolean */ | |
1337 | use_stack = 0; | |
1338 | ||
1339 | size *= sizeof(unsigned long); | |
1340 | ||
1341 | event = trace_buffer_lock_reserve(buffer, TRACE_STACK, | |
1342 | sizeof(*entry) + size, flags, pc); | |
1343 | if (!event) | |
1344 | goto out; | |
1345 | entry = ring_buffer_event_data(event); | |
1346 | ||
1347 | memset(&entry->caller, 0, size); | |
1348 | ||
1349 | if (use_stack) | |
1350 | memcpy(&entry->caller, trace.entries, | |
1351 | trace.nr_entries * sizeof(unsigned long)); | |
1352 | else { | |
1353 | trace.max_entries = FTRACE_STACK_ENTRIES; | |
1354 | trace.entries = entry->caller; | |
1355 | if (regs) | |
1356 | save_stack_trace_regs(regs, &trace); | |
1357 | else | |
1358 | save_stack_trace(&trace); | |
1359 | } | |
1360 | ||
1361 | entry->size = trace.nr_entries; | |
1362 | ||
1363 | if (!filter_check_discard(call, entry, buffer, event)) | |
1364 | ring_buffer_unlock_commit(buffer, event); | |
1365 | ||
1366 | out: | |
1367 | /* Again, don't let gcc optimize things here */ | |
1368 | barrier(); | |
1369 | __get_cpu_var(ftrace_stack_reserve)--; | |
1370 | preempt_enable_notrace(); | |
1371 | ||
1372 | } | |
1373 | ||
1374 | void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags, | |
1375 | int skip, int pc, struct pt_regs *regs) | |
1376 | { | |
1377 | if (!(trace_flags & TRACE_ITER_STACKTRACE)) | |
1378 | return; | |
1379 | ||
1380 | __ftrace_trace_stack(buffer, flags, skip, pc, regs); | |
1381 | } | |
1382 | ||
1383 | void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags, | |
1384 | int skip, int pc) | |
1385 | { | |
1386 | if (!(trace_flags & TRACE_ITER_STACKTRACE)) | |
1387 | return; | |
1388 | ||
1389 | __ftrace_trace_stack(buffer, flags, skip, pc, NULL); | |
1390 | } | |
1391 | ||
1392 | void __trace_stack(struct trace_array *tr, unsigned long flags, int skip, | |
1393 | int pc) | |
1394 | { | |
1395 | __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL); | |
1396 | } | |
1397 | ||
1398 | /** | |
1399 | * trace_dump_stack - record a stack back trace in the trace buffer | |
1400 | */ | |
1401 | void trace_dump_stack(void) | |
1402 | { | |
1403 | unsigned long flags; | |
1404 | ||
1405 | if (tracing_disabled || tracing_selftest_running) | |
1406 | return; | |
1407 | ||
1408 | local_save_flags(flags); | |
1409 | ||
1410 | /* skipping 3 traces, seems to get us at the caller of this function */ | |
1411 | __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL); | |
1412 | } | |
1413 | ||
1414 | static DEFINE_PER_CPU(int, user_stack_count); | |
1415 | ||
1416 | void | |
1417 | ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc) | |
1418 | { | |
1419 | struct ftrace_event_call *call = &event_user_stack; | |
1420 | struct ring_buffer_event *event; | |
1421 | struct userstack_entry *entry; | |
1422 | struct stack_trace trace; | |
1423 | ||
1424 | if (!(trace_flags & TRACE_ITER_USERSTACKTRACE)) | |
1425 | return; | |
1426 | ||
1427 | /* | |
1428 | * NMIs can not handle page faults, even with fix ups. | |
1429 | * The save user stack can (and often does) fault. | |
1430 | */ | |
1431 | if (unlikely(in_nmi())) | |
1432 | return; | |
1433 | ||
1434 | /* | |
1435 | * prevent recursion, since the user stack tracing may | |
1436 | * trigger other kernel events. | |
1437 | */ | |
1438 | preempt_disable(); | |
1439 | if (__this_cpu_read(user_stack_count)) | |
1440 | goto out; | |
1441 | ||
1442 | __this_cpu_inc(user_stack_count); | |
1443 | ||
1444 | event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK, | |
1445 | sizeof(*entry), flags, pc); | |
1446 | if (!event) | |
1447 | goto out_drop_count; | |
1448 | entry = ring_buffer_event_data(event); | |
1449 | ||
1450 | entry->tgid = current->tgid; | |
1451 | memset(&entry->caller, 0, sizeof(entry->caller)); | |
1452 | ||
1453 | trace.nr_entries = 0; | |
1454 | trace.max_entries = FTRACE_STACK_ENTRIES; | |
1455 | trace.skip = 0; | |
1456 | trace.entries = entry->caller; | |
1457 | ||
1458 | save_stack_trace_user(&trace); | |
1459 | if (!filter_check_discard(call, entry, buffer, event)) | |
1460 | ring_buffer_unlock_commit(buffer, event); | |
1461 | ||
1462 | out_drop_count: | |
1463 | __this_cpu_dec(user_stack_count); | |
1464 | out: | |
1465 | preempt_enable(); | |
1466 | } | |
1467 | ||
1468 | #ifdef UNUSED | |
1469 | static void __trace_userstack(struct trace_array *tr, unsigned long flags) | |
1470 | { | |
1471 | ftrace_trace_userstack(tr, flags, preempt_count()); | |
1472 | } | |
1473 | #endif /* UNUSED */ | |
1474 | ||
1475 | #endif /* CONFIG_STACKTRACE */ | |
1476 | ||
1477 | /* created for use with alloc_percpu */ | |
1478 | struct trace_buffer_struct { | |
1479 | char buffer[TRACE_BUF_SIZE]; | |
1480 | }; | |
1481 | ||
1482 | static struct trace_buffer_struct *trace_percpu_buffer; | |
1483 | static struct trace_buffer_struct *trace_percpu_sirq_buffer; | |
1484 | static struct trace_buffer_struct *trace_percpu_irq_buffer; | |
1485 | static struct trace_buffer_struct *trace_percpu_nmi_buffer; | |
1486 | ||
1487 | /* | |
1488 | * The buffer used is dependent on the context. There is a per cpu | |
1489 | * buffer for normal context, softirq contex, hard irq context and | |
1490 | * for NMI context. Thise allows for lockless recording. | |
1491 | * | |
1492 | * Note, if the buffers failed to be allocated, then this returns NULL | |
1493 | */ | |
1494 | static char *get_trace_buf(void) | |
1495 | { | |
1496 | struct trace_buffer_struct *percpu_buffer; | |
1497 | struct trace_buffer_struct *buffer; | |
1498 | ||
1499 | /* | |
1500 | * If we have allocated per cpu buffers, then we do not | |
1501 | * need to do any locking. | |
1502 | */ | |
1503 | if (in_nmi()) | |
1504 | percpu_buffer = trace_percpu_nmi_buffer; | |
1505 | else if (in_irq()) | |
1506 | percpu_buffer = trace_percpu_irq_buffer; | |
1507 | else if (in_softirq()) | |
1508 | percpu_buffer = trace_percpu_sirq_buffer; | |
1509 | else | |
1510 | percpu_buffer = trace_percpu_buffer; | |
1511 | ||
1512 | if (!percpu_buffer) | |
1513 | return NULL; | |
1514 | ||
1515 | buffer = per_cpu_ptr(percpu_buffer, smp_processor_id()); | |
1516 | ||
1517 | return buffer->buffer; | |
1518 | } | |
1519 | ||
1520 | static int alloc_percpu_trace_buffer(void) | |
1521 | { | |
1522 | struct trace_buffer_struct *buffers; | |
1523 | struct trace_buffer_struct *sirq_buffers; | |
1524 | struct trace_buffer_struct *irq_buffers; | |
1525 | struct trace_buffer_struct *nmi_buffers; | |
1526 | ||
1527 | buffers = alloc_percpu(struct trace_buffer_struct); | |
1528 | if (!buffers) | |
1529 | goto err_warn; | |
1530 | ||
1531 | sirq_buffers = alloc_percpu(struct trace_buffer_struct); | |
1532 | if (!sirq_buffers) | |
1533 | goto err_sirq; | |
1534 | ||
1535 | irq_buffers = alloc_percpu(struct trace_buffer_struct); | |
1536 | if (!irq_buffers) | |
1537 | goto err_irq; | |
1538 | ||
1539 | nmi_buffers = alloc_percpu(struct trace_buffer_struct); | |
1540 | if (!nmi_buffers) | |
1541 | goto err_nmi; | |
1542 | ||
1543 | trace_percpu_buffer = buffers; | |
1544 | trace_percpu_sirq_buffer = sirq_buffers; | |
1545 | trace_percpu_irq_buffer = irq_buffers; | |
1546 | trace_percpu_nmi_buffer = nmi_buffers; | |
1547 | ||
1548 | return 0; | |
1549 | ||
1550 | err_nmi: | |
1551 | free_percpu(irq_buffers); | |
1552 | err_irq: | |
1553 | free_percpu(sirq_buffers); | |
1554 | err_sirq: | |
1555 | free_percpu(buffers); | |
1556 | err_warn: | |
1557 | WARN(1, "Could not allocate percpu trace_printk buffer"); | |
1558 | return -ENOMEM; | |
1559 | } | |
1560 | ||
1561 | void trace_printk_init_buffers(void) | |
1562 | { | |
1563 | static int buffers_allocated; | |
1564 | ||
1565 | if (buffers_allocated) | |
1566 | return; | |
1567 | ||
1568 | if (alloc_percpu_trace_buffer()) | |
1569 | return; | |
1570 | ||
1571 | pr_info("ftrace: Allocated trace_printk buffers\n"); | |
1572 | ||
1573 | buffers_allocated = 1; | |
1574 | } | |
1575 | ||
1576 | /** | |
1577 | * trace_vbprintk - write binary msg to tracing buffer | |
1578 | * | |
1579 | */ | |
1580 | int trace_vbprintk(unsigned long ip, const char *fmt, va_list args) | |
1581 | { | |
1582 | struct ftrace_event_call *call = &event_bprint; | |
1583 | struct ring_buffer_event *event; | |
1584 | struct ring_buffer *buffer; | |
1585 | struct trace_array *tr = &global_trace; | |
1586 | struct bprint_entry *entry; | |
1587 | unsigned long flags; | |
1588 | char *tbuffer; | |
1589 | int len = 0, size, pc; | |
1590 | ||
1591 | if (unlikely(tracing_selftest_running || tracing_disabled)) | |
1592 | return 0; | |
1593 | ||
1594 | /* Don't pollute graph traces with trace_vprintk internals */ | |
1595 | pause_graph_tracing(); | |
1596 | ||
1597 | pc = preempt_count(); | |
1598 | preempt_disable_notrace(); | |
1599 | ||
1600 | tbuffer = get_trace_buf(); | |
1601 | if (!tbuffer) { | |
1602 | len = 0; | |
1603 | goto out; | |
1604 | } | |
1605 | ||
1606 | len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args); | |
1607 | ||
1608 | if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0) | |
1609 | goto out; | |
1610 | ||
1611 | local_save_flags(flags); | |
1612 | size = sizeof(*entry) + sizeof(u32) * len; | |
1613 | buffer = tr->buffer; | |
1614 | event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size, | |
1615 | flags, pc); | |
1616 | if (!event) | |
1617 | goto out; | |
1618 | entry = ring_buffer_event_data(event); | |
1619 | entry->ip = ip; | |
1620 | entry->fmt = fmt; | |
1621 | ||
1622 | memcpy(entry->buf, tbuffer, sizeof(u32) * len); | |
1623 | if (!filter_check_discard(call, entry, buffer, event)) { | |
1624 | ring_buffer_unlock_commit(buffer, event); | |
1625 | ftrace_trace_stack(buffer, flags, 6, pc); | |
1626 | } | |
1627 | ||
1628 | out: | |
1629 | preempt_enable_notrace(); | |
1630 | unpause_graph_tracing(); | |
1631 | ||
1632 | return len; | |
1633 | } | |
1634 | EXPORT_SYMBOL_GPL(trace_vbprintk); | |
1635 | ||
1636 | int trace_array_printk(struct trace_array *tr, | |
1637 | unsigned long ip, const char *fmt, ...) | |
1638 | { | |
1639 | int ret; | |
1640 | va_list ap; | |
1641 | ||
1642 | if (!(trace_flags & TRACE_ITER_PRINTK)) | |
1643 | return 0; | |
1644 | ||
1645 | va_start(ap, fmt); | |
1646 | ret = trace_array_vprintk(tr, ip, fmt, ap); | |
1647 | va_end(ap); | |
1648 | return ret; | |
1649 | } | |
1650 | ||
1651 | int trace_array_vprintk(struct trace_array *tr, | |
1652 | unsigned long ip, const char *fmt, va_list args) | |
1653 | { | |
1654 | struct ftrace_event_call *call = &event_print; | |
1655 | struct ring_buffer_event *event; | |
1656 | struct ring_buffer *buffer; | |
1657 | int len = 0, size, pc; | |
1658 | struct print_entry *entry; | |
1659 | unsigned long flags; | |
1660 | char *tbuffer; | |
1661 | ||
1662 | if (tracing_disabled || tracing_selftest_running) | |
1663 | return 0; | |
1664 | ||
1665 | /* Don't pollute graph traces with trace_vprintk internals */ | |
1666 | pause_graph_tracing(); | |
1667 | ||
1668 | pc = preempt_count(); | |
1669 | preempt_disable_notrace(); | |
1670 | ||
1671 | ||
1672 | tbuffer = get_trace_buf(); | |
1673 | if (!tbuffer) { | |
1674 | len = 0; | |
1675 | goto out; | |
1676 | } | |
1677 | ||
1678 | len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args); | |
1679 | if (len > TRACE_BUF_SIZE) | |
1680 | goto out; | |
1681 | ||
1682 | local_save_flags(flags); | |
1683 | size = sizeof(*entry) + len + 1; | |
1684 | buffer = tr->buffer; | |
1685 | event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size, | |
1686 | flags, pc); | |
1687 | if (!event) | |
1688 | goto out; | |
1689 | entry = ring_buffer_event_data(event); | |
1690 | entry->ip = ip; | |
1691 | ||
1692 | memcpy(&entry->buf, tbuffer, len); | |
1693 | entry->buf[len] = '\0'; | |
1694 | if (!filter_check_discard(call, entry, buffer, event)) { | |
1695 | ring_buffer_unlock_commit(buffer, event); | |
1696 | ftrace_trace_stack(buffer, flags, 6, pc); | |
1697 | } | |
1698 | out: | |
1699 | preempt_enable_notrace(); | |
1700 | unpause_graph_tracing(); | |
1701 | ||
1702 | return len; | |
1703 | } | |
1704 | ||
1705 | int trace_vprintk(unsigned long ip, const char *fmt, va_list args) | |
1706 | { | |
1707 | return trace_array_vprintk(&global_trace, ip, fmt, args); | |
1708 | } | |
1709 | EXPORT_SYMBOL_GPL(trace_vprintk); | |
1710 | ||
1711 | static void trace_iterator_increment(struct trace_iterator *iter) | |
1712 | { | |
1713 | struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu); | |
1714 | ||
1715 | iter->idx++; | |
1716 | if (buf_iter) | |
1717 | ring_buffer_read(buf_iter, NULL); | |
1718 | } | |
1719 | ||
1720 | static struct trace_entry * | |
1721 | peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts, | |
1722 | unsigned long *lost_events) | |
1723 | { | |
1724 | struct ring_buffer_event *event; | |
1725 | struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu); | |
1726 | ||
1727 | if (buf_iter) | |
1728 | event = ring_buffer_iter_peek(buf_iter, ts); | |
1729 | else | |
1730 | event = ring_buffer_peek(iter->tr->buffer, cpu, ts, | |
1731 | lost_events); | |
1732 | ||
1733 | if (event) { | |
1734 | iter->ent_size = ring_buffer_event_length(event); | |
1735 | return ring_buffer_event_data(event); | |
1736 | } | |
1737 | iter->ent_size = 0; | |
1738 | return NULL; | |
1739 | } | |
1740 | ||
1741 | static struct trace_entry * | |
1742 | __find_next_entry(struct trace_iterator *iter, int *ent_cpu, | |
1743 | unsigned long *missing_events, u64 *ent_ts) | |
1744 | { | |
1745 | struct ring_buffer *buffer = iter->tr->buffer; | |
1746 | struct trace_entry *ent, *next = NULL; | |
1747 | unsigned long lost_events = 0, next_lost = 0; | |
1748 | int cpu_file = iter->cpu_file; | |
1749 | u64 next_ts = 0, ts; | |
1750 | int next_cpu = -1; | |
1751 | int next_size = 0; | |
1752 | int cpu; | |
1753 | ||
1754 | /* | |
1755 | * If we are in a per_cpu trace file, don't bother by iterating over | |
1756 | * all cpu and peek directly. | |
1757 | */ | |
1758 | if (cpu_file > TRACE_PIPE_ALL_CPU) { | |
1759 | if (ring_buffer_empty_cpu(buffer, cpu_file)) | |
1760 | return NULL; | |
1761 | ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events); | |
1762 | if (ent_cpu) | |
1763 | *ent_cpu = cpu_file; | |
1764 | ||
1765 | return ent; | |
1766 | } | |
1767 | ||
1768 | for_each_tracing_cpu(cpu) { | |
1769 | ||
1770 | if (ring_buffer_empty_cpu(buffer, cpu)) | |
1771 | continue; | |
1772 | ||
1773 | ent = peek_next_entry(iter, cpu, &ts, &lost_events); | |
1774 | ||
1775 | /* | |
1776 | * Pick the entry with the smallest timestamp: | |
1777 | */ | |
1778 | if (ent && (!next || ts < next_ts)) { | |
1779 | next = ent; | |
1780 | next_cpu = cpu; | |
1781 | next_ts = ts; | |
1782 | next_lost = lost_events; | |
1783 | next_size = iter->ent_size; | |
1784 | } | |
1785 | } | |
1786 | ||
1787 | iter->ent_size = next_size; | |
1788 | ||
1789 | if (ent_cpu) | |
1790 | *ent_cpu = next_cpu; | |
1791 | ||
1792 | if (ent_ts) | |
1793 | *ent_ts = next_ts; | |
1794 | ||
1795 | if (missing_events) | |
1796 | *missing_events = next_lost; | |
1797 | ||
1798 | return next; | |
1799 | } | |
1800 | ||
1801 | /* Find the next real entry, without updating the iterator itself */ | |
1802 | struct trace_entry *trace_find_next_entry(struct trace_iterator *iter, | |
1803 | int *ent_cpu, u64 *ent_ts) | |
1804 | { | |
1805 | return __find_next_entry(iter, ent_cpu, NULL, ent_ts); | |
1806 | } | |
1807 | ||
1808 | /* Find the next real entry, and increment the iterator to the next entry */ | |
1809 | void *trace_find_next_entry_inc(struct trace_iterator *iter) | |
1810 | { | |
1811 | iter->ent = __find_next_entry(iter, &iter->cpu, | |
1812 | &iter->lost_events, &iter->ts); | |
1813 | ||
1814 | if (iter->ent) | |
1815 | trace_iterator_increment(iter); | |
1816 | ||
1817 | return iter->ent ? iter : NULL; | |
1818 | } | |
1819 | ||
1820 | static void trace_consume(struct trace_iterator *iter) | |
1821 | { | |
1822 | ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts, | |
1823 | &iter->lost_events); | |
1824 | } | |
1825 | ||
1826 | static void *s_next(struct seq_file *m, void *v, loff_t *pos) | |
1827 | { | |
1828 | struct trace_iterator *iter = m->private; | |
1829 | int i = (int)*pos; | |
1830 | void *ent; | |
1831 | ||
1832 | WARN_ON_ONCE(iter->leftover); | |
1833 | ||
1834 | (*pos)++; | |
1835 | ||
1836 | /* can't go backwards */ | |
1837 | if (iter->idx > i) | |
1838 | return NULL; | |
1839 | ||
1840 | if (iter->idx < 0) | |
1841 | ent = trace_find_next_entry_inc(iter); | |
1842 | else | |
1843 | ent = iter; | |
1844 | ||
1845 | while (ent && iter->idx < i) | |
1846 | ent = trace_find_next_entry_inc(iter); | |
1847 | ||
1848 | iter->pos = *pos; | |
1849 | ||
1850 | return ent; | |
1851 | } | |
1852 | ||
1853 | void tracing_iter_reset(struct trace_iterator *iter, int cpu) | |
1854 | { | |
1855 | struct trace_array *tr = iter->tr; | |
1856 | struct ring_buffer_event *event; | |
1857 | struct ring_buffer_iter *buf_iter; | |
1858 | unsigned long entries = 0; | |
1859 | u64 ts; | |
1860 | ||
1861 | tr->data[cpu]->skipped_entries = 0; | |
1862 | ||
1863 | buf_iter = trace_buffer_iter(iter, cpu); | |
1864 | if (!buf_iter) | |
1865 | return; | |
1866 | ||
1867 | ring_buffer_iter_reset(buf_iter); | |
1868 | ||
1869 | /* | |
1870 | * We could have the case with the max latency tracers | |
1871 | * that a reset never took place on a cpu. This is evident | |
1872 | * by the timestamp being before the start of the buffer. | |
1873 | */ | |
1874 | while ((event = ring_buffer_iter_peek(buf_iter, &ts))) { | |
1875 | if (ts >= iter->tr->time_start) | |
1876 | break; | |
1877 | entries++; | |
1878 | ring_buffer_read(buf_iter, NULL); | |
1879 | } | |
1880 | ||
1881 | tr->data[cpu]->skipped_entries = entries; | |
1882 | } | |
1883 | ||
1884 | /* | |
1885 | * The current tracer is copied to avoid a global locking | |
1886 | * all around. | |
1887 | */ | |
1888 | static void *s_start(struct seq_file *m, loff_t *pos) | |
1889 | { | |
1890 | struct trace_iterator *iter = m->private; | |
1891 | static struct tracer *old_tracer; | |
1892 | int cpu_file = iter->cpu_file; | |
1893 | void *p = NULL; | |
1894 | loff_t l = 0; | |
1895 | int cpu; | |
1896 | ||
1897 | /* copy the tracer to avoid using a global lock all around */ | |
1898 | mutex_lock(&trace_types_lock); | |
1899 | if (unlikely(old_tracer != current_trace && current_trace)) { | |
1900 | old_tracer = current_trace; | |
1901 | *iter->trace = *current_trace; | |
1902 | } | |
1903 | mutex_unlock(&trace_types_lock); | |
1904 | ||
1905 | atomic_inc(&trace_record_cmdline_disabled); | |
1906 | ||
1907 | if (*pos != iter->pos) { | |
1908 | iter->ent = NULL; | |
1909 | iter->cpu = 0; | |
1910 | iter->idx = -1; | |
1911 | ||
1912 | if (cpu_file == TRACE_PIPE_ALL_CPU) { | |
1913 | for_each_tracing_cpu(cpu) | |
1914 | tracing_iter_reset(iter, cpu); | |
1915 | } else | |
1916 | tracing_iter_reset(iter, cpu_file); | |
1917 | ||
1918 | iter->leftover = 0; | |
1919 | for (p = iter; p && l < *pos; p = s_next(m, p, &l)) | |
1920 | ; | |
1921 | ||
1922 | } else { | |
1923 | /* | |
1924 | * If we overflowed the seq_file before, then we want | |
1925 | * to just reuse the trace_seq buffer again. | |
1926 | */ | |
1927 | if (iter->leftover) | |
1928 | p = iter; | |
1929 | else { | |
1930 | l = *pos - 1; | |
1931 | p = s_next(m, p, &l); | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | trace_event_read_lock(); | |
1936 | trace_access_lock(cpu_file); | |
1937 | return p; | |
1938 | } | |
1939 | ||
1940 | static void s_stop(struct seq_file *m, void *p) | |
1941 | { | |
1942 | struct trace_iterator *iter = m->private; | |
1943 | ||
1944 | atomic_dec(&trace_record_cmdline_disabled); | |
1945 | trace_access_unlock(iter->cpu_file); | |
1946 | trace_event_read_unlock(); | |
1947 | } | |
1948 | ||
1949 | static void | |
1950 | get_total_entries(struct trace_array *tr, unsigned long *total, unsigned long *entries) | |
1951 | { | |
1952 | unsigned long count; | |
1953 | int cpu; | |
1954 | ||
1955 | *total = 0; | |
1956 | *entries = 0; | |
1957 | ||
1958 | for_each_tracing_cpu(cpu) { | |
1959 | count = ring_buffer_entries_cpu(tr->buffer, cpu); | |
1960 | /* | |
1961 | * If this buffer has skipped entries, then we hold all | |
1962 | * entries for the trace and we need to ignore the | |
1963 | * ones before the time stamp. | |
1964 | */ | |
1965 | if (tr->data[cpu]->skipped_entries) { | |
1966 | count -= tr->data[cpu]->skipped_entries; | |
1967 | /* total is the same as the entries */ | |
1968 | *total += count; | |
1969 | } else | |
1970 | *total += count + | |
1971 | ring_buffer_overrun_cpu(tr->buffer, cpu); | |
1972 | *entries += count; | |
1973 | } | |
1974 | } | |
1975 | ||
1976 | static void print_lat_help_header(struct seq_file *m) | |
1977 | { | |
1978 | seq_puts(m, "# _------=> CPU# \n"); | |
1979 | seq_puts(m, "# / _-----=> irqs-off \n"); | |
1980 | seq_puts(m, "# | / _----=> need-resched \n"); | |
1981 | seq_puts(m, "# || / _---=> hardirq/softirq \n"); | |
1982 | seq_puts(m, "# ||| / _--=> preempt-depth \n"); | |
1983 | seq_puts(m, "# |||| / delay \n"); | |
1984 | seq_puts(m, "# cmd pid ||||| time | caller \n"); | |
1985 | seq_puts(m, "# \\ / ||||| \\ | / \n"); | |
1986 | } | |
1987 | ||
1988 | static void print_event_info(struct trace_array *tr, struct seq_file *m) | |
1989 | { | |
1990 | unsigned long total; | |
1991 | unsigned long entries; | |
1992 | ||
1993 | get_total_entries(tr, &total, &entries); | |
1994 | seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n", | |
1995 | entries, total, num_online_cpus()); | |
1996 | seq_puts(m, "#\n"); | |
1997 | } | |
1998 | ||
1999 | static void print_func_help_header(struct trace_array *tr, struct seq_file *m) | |
2000 | { | |
2001 | print_event_info(tr, m); | |
2002 | seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n"); | |
2003 | seq_puts(m, "# | | | | |\n"); | |
2004 | } | |
2005 | ||
2006 | static void print_func_help_header_irq(struct trace_array *tr, struct seq_file *m) | |
2007 | { | |
2008 | print_event_info(tr, m); | |
2009 | seq_puts(m, "# _-----=> irqs-off\n"); | |
2010 | seq_puts(m, "# / _----=> need-resched\n"); | |
2011 | seq_puts(m, "# | / _---=> hardirq/softirq\n"); | |
2012 | seq_puts(m, "# || / _--=> preempt-depth\n"); | |
2013 | seq_puts(m, "# ||| / delay\n"); | |
2014 | seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n"); | |
2015 | seq_puts(m, "# | | | |||| | |\n"); | |
2016 | } | |
2017 | ||
2018 | void | |
2019 | print_trace_header(struct seq_file *m, struct trace_iterator *iter) | |
2020 | { | |
2021 | unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK); | |
2022 | struct trace_array *tr = iter->tr; | |
2023 | struct trace_array_cpu *data = tr->data[tr->cpu]; | |
2024 | struct tracer *type = current_trace; | |
2025 | unsigned long entries; | |
2026 | unsigned long total; | |
2027 | const char *name = "preemption"; | |
2028 | ||
2029 | if (type) | |
2030 | name = type->name; | |
2031 | ||
2032 | get_total_entries(tr, &total, &entries); | |
2033 | ||
2034 | seq_printf(m, "# %s latency trace v1.1.5 on %s\n", | |
2035 | name, UTS_RELEASE); | |
2036 | seq_puts(m, "# -----------------------------------" | |
2037 | "---------------------------------\n"); | |
2038 | seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |" | |
2039 | " (M:%s VP:%d, KP:%d, SP:%d HP:%d", | |
2040 | nsecs_to_usecs(data->saved_latency), | |
2041 | entries, | |
2042 | total, | |
2043 | tr->cpu, | |
2044 | #if defined(CONFIG_PREEMPT_NONE) | |
2045 | "server", | |
2046 | #elif defined(CONFIG_PREEMPT_VOLUNTARY) | |
2047 | "desktop", | |
2048 | #elif defined(CONFIG_PREEMPT) | |
2049 | "preempt", | |
2050 | #else | |
2051 | "unknown", | |
2052 | #endif | |
2053 | /* These are reserved for later use */ | |
2054 | 0, 0, 0, 0); | |
2055 | #ifdef CONFIG_SMP | |
2056 | seq_printf(m, " #P:%d)\n", num_online_cpus()); | |
2057 | #else | |
2058 | seq_puts(m, ")\n"); | |
2059 | #endif | |
2060 | seq_puts(m, "# -----------------\n"); | |
2061 | seq_printf(m, "# | task: %.16s-%d " | |
2062 | "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n", | |
2063 | data->comm, data->pid, data->uid, data->nice, | |
2064 | data->policy, data->rt_priority); | |
2065 | seq_puts(m, "# -----------------\n"); | |
2066 | ||
2067 | if (data->critical_start) { | |
2068 | seq_puts(m, "# => started at: "); | |
2069 | seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags); | |
2070 | trace_print_seq(m, &iter->seq); | |
2071 | seq_puts(m, "\n# => ended at: "); | |
2072 | seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags); | |
2073 | trace_print_seq(m, &iter->seq); | |
2074 | seq_puts(m, "\n#\n"); | |
2075 | } | |
2076 | ||
2077 | seq_puts(m, "#\n"); | |
2078 | } | |
2079 | ||
2080 | static void test_cpu_buff_start(struct trace_iterator *iter) | |
2081 | { | |
2082 | struct trace_seq *s = &iter->seq; | |
2083 | ||
2084 | if (!(trace_flags & TRACE_ITER_ANNOTATE)) | |
2085 | return; | |
2086 | ||
2087 | if (!(iter->iter_flags & TRACE_FILE_ANNOTATE)) | |
2088 | return; | |
2089 | ||
2090 | if (cpumask_test_cpu(iter->cpu, iter->started)) | |
2091 | return; | |
2092 | ||
2093 | if (iter->tr->data[iter->cpu]->skipped_entries) | |
2094 | return; | |
2095 | ||
2096 | cpumask_set_cpu(iter->cpu, iter->started); | |
2097 | ||
2098 | /* Don't print started cpu buffer for the first entry of the trace */ | |
2099 | if (iter->idx > 1) | |
2100 | trace_seq_printf(s, "##### CPU %u buffer started ####\n", | |
2101 | iter->cpu); | |
2102 | } | |
2103 | ||
2104 | static enum print_line_t print_trace_fmt(struct trace_iterator *iter) | |
2105 | { | |
2106 | struct trace_seq *s = &iter->seq; | |
2107 | unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK); | |
2108 | struct trace_entry *entry; | |
2109 | struct trace_event *event; | |
2110 | ||
2111 | entry = iter->ent; | |
2112 | ||
2113 | test_cpu_buff_start(iter); | |
2114 | ||
2115 | event = ftrace_find_event(entry->type); | |
2116 | ||
2117 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
2118 | if (iter->iter_flags & TRACE_FILE_LAT_FMT) { | |
2119 | if (!trace_print_lat_context(iter)) | |
2120 | goto partial; | |
2121 | } else { | |
2122 | if (!trace_print_context(iter)) | |
2123 | goto partial; | |
2124 | } | |
2125 | } | |
2126 | ||
2127 | if (event) | |
2128 | return event->funcs->trace(iter, sym_flags, event); | |
2129 | ||
2130 | if (!trace_seq_printf(s, "Unknown type %d\n", entry->type)) | |
2131 | goto partial; | |
2132 | ||
2133 | return TRACE_TYPE_HANDLED; | |
2134 | partial: | |
2135 | return TRACE_TYPE_PARTIAL_LINE; | |
2136 | } | |
2137 | ||
2138 | static enum print_line_t print_raw_fmt(struct trace_iterator *iter) | |
2139 | { | |
2140 | struct trace_seq *s = &iter->seq; | |
2141 | struct trace_entry *entry; | |
2142 | struct trace_event *event; | |
2143 | ||
2144 | entry = iter->ent; | |
2145 | ||
2146 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
2147 | if (!trace_seq_printf(s, "%d %d %llu ", | |
2148 | entry->pid, iter->cpu, iter->ts)) | |
2149 | goto partial; | |
2150 | } | |
2151 | ||
2152 | event = ftrace_find_event(entry->type); | |
2153 | if (event) | |
2154 | return event->funcs->raw(iter, 0, event); | |
2155 | ||
2156 | if (!trace_seq_printf(s, "%d ?\n", entry->type)) | |
2157 | goto partial; | |
2158 | ||
2159 | return TRACE_TYPE_HANDLED; | |
2160 | partial: | |
2161 | return TRACE_TYPE_PARTIAL_LINE; | |
2162 | } | |
2163 | ||
2164 | static enum print_line_t print_hex_fmt(struct trace_iterator *iter) | |
2165 | { | |
2166 | struct trace_seq *s = &iter->seq; | |
2167 | unsigned char newline = '\n'; | |
2168 | struct trace_entry *entry; | |
2169 | struct trace_event *event; | |
2170 | ||
2171 | entry = iter->ent; | |
2172 | ||
2173 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
2174 | SEQ_PUT_HEX_FIELD_RET(s, entry->pid); | |
2175 | SEQ_PUT_HEX_FIELD_RET(s, iter->cpu); | |
2176 | SEQ_PUT_HEX_FIELD_RET(s, iter->ts); | |
2177 | } | |
2178 | ||
2179 | event = ftrace_find_event(entry->type); | |
2180 | if (event) { | |
2181 | enum print_line_t ret = event->funcs->hex(iter, 0, event); | |
2182 | if (ret != TRACE_TYPE_HANDLED) | |
2183 | return ret; | |
2184 | } | |
2185 | ||
2186 | SEQ_PUT_FIELD_RET(s, newline); | |
2187 | ||
2188 | return TRACE_TYPE_HANDLED; | |
2189 | } | |
2190 | ||
2191 | static enum print_line_t print_bin_fmt(struct trace_iterator *iter) | |
2192 | { | |
2193 | struct trace_seq *s = &iter->seq; | |
2194 | struct trace_entry *entry; | |
2195 | struct trace_event *event; | |
2196 | ||
2197 | entry = iter->ent; | |
2198 | ||
2199 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
2200 | SEQ_PUT_FIELD_RET(s, entry->pid); | |
2201 | SEQ_PUT_FIELD_RET(s, iter->cpu); | |
2202 | SEQ_PUT_FIELD_RET(s, iter->ts); | |
2203 | } | |
2204 | ||
2205 | event = ftrace_find_event(entry->type); | |
2206 | return event ? event->funcs->binary(iter, 0, event) : | |
2207 | TRACE_TYPE_HANDLED; | |
2208 | } | |
2209 | ||
2210 | int trace_empty(struct trace_iterator *iter) | |
2211 | { | |
2212 | struct ring_buffer_iter *buf_iter; | |
2213 | int cpu; | |
2214 | ||
2215 | /* If we are looking at one CPU buffer, only check that one */ | |
2216 | if (iter->cpu_file != TRACE_PIPE_ALL_CPU) { | |
2217 | cpu = iter->cpu_file; | |
2218 | buf_iter = trace_buffer_iter(iter, cpu); | |
2219 | if (buf_iter) { | |
2220 | if (!ring_buffer_iter_empty(buf_iter)) | |
2221 | return 0; | |
2222 | } else { | |
2223 | if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu)) | |
2224 | return 0; | |
2225 | } | |
2226 | return 1; | |
2227 | } | |
2228 | ||
2229 | for_each_tracing_cpu(cpu) { | |
2230 | buf_iter = trace_buffer_iter(iter, cpu); | |
2231 | if (buf_iter) { | |
2232 | if (!ring_buffer_iter_empty(buf_iter)) | |
2233 | return 0; | |
2234 | } else { | |
2235 | if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu)) | |
2236 | return 0; | |
2237 | } | |
2238 | } | |
2239 | ||
2240 | return 1; | |
2241 | } | |
2242 | ||
2243 | /* Called with trace_event_read_lock() held. */ | |
2244 | enum print_line_t print_trace_line(struct trace_iterator *iter) | |
2245 | { | |
2246 | enum print_line_t ret; | |
2247 | ||
2248 | if (iter->lost_events && | |
2249 | !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n", | |
2250 | iter->cpu, iter->lost_events)) | |
2251 | return TRACE_TYPE_PARTIAL_LINE; | |
2252 | ||
2253 | if (iter->trace && iter->trace->print_line) { | |
2254 | ret = iter->trace->print_line(iter); | |
2255 | if (ret != TRACE_TYPE_UNHANDLED) | |
2256 | return ret; | |
2257 | } | |
2258 | ||
2259 | if (iter->ent->type == TRACE_BPRINT && | |
2260 | trace_flags & TRACE_ITER_PRINTK && | |
2261 | trace_flags & TRACE_ITER_PRINTK_MSGONLY) | |
2262 | return trace_print_bprintk_msg_only(iter); | |
2263 | ||
2264 | if (iter->ent->type == TRACE_PRINT && | |
2265 | trace_flags & TRACE_ITER_PRINTK && | |
2266 | trace_flags & TRACE_ITER_PRINTK_MSGONLY) | |
2267 | return trace_print_printk_msg_only(iter); | |
2268 | ||
2269 | if (trace_flags & TRACE_ITER_BIN) | |
2270 | return print_bin_fmt(iter); | |
2271 | ||
2272 | if (trace_flags & TRACE_ITER_HEX) | |
2273 | return print_hex_fmt(iter); | |
2274 | ||
2275 | if (trace_flags & TRACE_ITER_RAW) | |
2276 | return print_raw_fmt(iter); | |
2277 | ||
2278 | return print_trace_fmt(iter); | |
2279 | } | |
2280 | ||
2281 | void trace_latency_header(struct seq_file *m) | |
2282 | { | |
2283 | struct trace_iterator *iter = m->private; | |
2284 | ||
2285 | /* print nothing if the buffers are empty */ | |
2286 | if (trace_empty(iter)) | |
2287 | return; | |
2288 | ||
2289 | if (iter->iter_flags & TRACE_FILE_LAT_FMT) | |
2290 | print_trace_header(m, iter); | |
2291 | ||
2292 | if (!(trace_flags & TRACE_ITER_VERBOSE)) | |
2293 | print_lat_help_header(m); | |
2294 | } | |
2295 | ||
2296 | void trace_default_header(struct seq_file *m) | |
2297 | { | |
2298 | struct trace_iterator *iter = m->private; | |
2299 | ||
2300 | if (!(trace_flags & TRACE_ITER_CONTEXT_INFO)) | |
2301 | return; | |
2302 | ||
2303 | if (iter->iter_flags & TRACE_FILE_LAT_FMT) { | |
2304 | /* print nothing if the buffers are empty */ | |
2305 | if (trace_empty(iter)) | |
2306 | return; | |
2307 | print_trace_header(m, iter); | |
2308 | if (!(trace_flags & TRACE_ITER_VERBOSE)) | |
2309 | print_lat_help_header(m); | |
2310 | } else { | |
2311 | if (!(trace_flags & TRACE_ITER_VERBOSE)) { | |
2312 | if (trace_flags & TRACE_ITER_IRQ_INFO) | |
2313 | print_func_help_header_irq(iter->tr, m); | |
2314 | else | |
2315 | print_func_help_header(iter->tr, m); | |
2316 | } | |
2317 | } | |
2318 | } | |
2319 | ||
2320 | static void test_ftrace_alive(struct seq_file *m) | |
2321 | { | |
2322 | if (!ftrace_is_dead()) | |
2323 | return; | |
2324 | seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"); | |
2325 | seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n"); | |
2326 | } | |
2327 | ||
2328 | static int s_show(struct seq_file *m, void *v) | |
2329 | { | |
2330 | struct trace_iterator *iter = v; | |
2331 | int ret; | |
2332 | ||
2333 | if (iter->ent == NULL) { | |
2334 | if (iter->tr) { | |
2335 | seq_printf(m, "# tracer: %s\n", iter->trace->name); | |
2336 | seq_puts(m, "#\n"); | |
2337 | test_ftrace_alive(m); | |
2338 | } | |
2339 | if (iter->trace && iter->trace->print_header) | |
2340 | iter->trace->print_header(m); | |
2341 | else | |
2342 | trace_default_header(m); | |
2343 | ||
2344 | } else if (iter->leftover) { | |
2345 | /* | |
2346 | * If we filled the seq_file buffer earlier, we | |
2347 | * want to just show it now. | |
2348 | */ | |
2349 | ret = trace_print_seq(m, &iter->seq); | |
2350 | ||
2351 | /* ret should this time be zero, but you never know */ | |
2352 | iter->leftover = ret; | |
2353 | ||
2354 | } else { | |
2355 | print_trace_line(iter); | |
2356 | ret = trace_print_seq(m, &iter->seq); | |
2357 | /* | |
2358 | * If we overflow the seq_file buffer, then it will | |
2359 | * ask us for this data again at start up. | |
2360 | * Use that instead. | |
2361 | * ret is 0 if seq_file write succeeded. | |
2362 | * -1 otherwise. | |
2363 | */ | |
2364 | iter->leftover = ret; | |
2365 | } | |
2366 | ||
2367 | return 0; | |
2368 | } | |
2369 | ||
2370 | static const struct seq_operations tracer_seq_ops = { | |
2371 | .start = s_start, | |
2372 | .next = s_next, | |
2373 | .stop = s_stop, | |
2374 | .show = s_show, | |
2375 | }; | |
2376 | ||
2377 | static struct trace_iterator * | |
2378 | __tracing_open(struct inode *inode, struct file *file) | |
2379 | { | |
2380 | long cpu_file = (long) inode->i_private; | |
2381 | struct trace_iterator *iter; | |
2382 | int cpu; | |
2383 | ||
2384 | if (tracing_disabled) | |
2385 | return ERR_PTR(-ENODEV); | |
2386 | ||
2387 | iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter)); | |
2388 | if (!iter) | |
2389 | return ERR_PTR(-ENOMEM); | |
2390 | ||
2391 | iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(), | |
2392 | GFP_KERNEL); | |
2393 | if (!iter->buffer_iter) | |
2394 | goto release; | |
2395 | ||
2396 | /* | |
2397 | * We make a copy of the current tracer to avoid concurrent | |
2398 | * changes on it while we are reading. | |
2399 | */ | |
2400 | mutex_lock(&trace_types_lock); | |
2401 | iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL); | |
2402 | if (!iter->trace) | |
2403 | goto fail; | |
2404 | ||
2405 | if (current_trace) | |
2406 | *iter->trace = *current_trace; | |
2407 | ||
2408 | if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL)) | |
2409 | goto fail; | |
2410 | ||
2411 | if (current_trace && current_trace->print_max) | |
2412 | iter->tr = &max_tr; | |
2413 | else | |
2414 | iter->tr = &global_trace; | |
2415 | iter->pos = -1; | |
2416 | mutex_init(&iter->mutex); | |
2417 | iter->cpu_file = cpu_file; | |
2418 | ||
2419 | /* Notify the tracer early; before we stop tracing. */ | |
2420 | if (iter->trace && iter->trace->open) | |
2421 | iter->trace->open(iter); | |
2422 | ||
2423 | /* Annotate start of buffers if we had overruns */ | |
2424 | if (ring_buffer_overruns(iter->tr->buffer)) | |
2425 | iter->iter_flags |= TRACE_FILE_ANNOTATE; | |
2426 | ||
2427 | /* stop the trace while dumping */ | |
2428 | tracing_stop(); | |
2429 | ||
2430 | if (iter->cpu_file == TRACE_PIPE_ALL_CPU) { | |
2431 | for_each_tracing_cpu(cpu) { | |
2432 | iter->buffer_iter[cpu] = | |
2433 | ring_buffer_read_prepare(iter->tr->buffer, cpu); | |
2434 | } | |
2435 | ring_buffer_read_prepare_sync(); | |
2436 | for_each_tracing_cpu(cpu) { | |
2437 | ring_buffer_read_start(iter->buffer_iter[cpu]); | |
2438 | tracing_iter_reset(iter, cpu); | |
2439 | } | |
2440 | } else { | |
2441 | cpu = iter->cpu_file; | |
2442 | iter->buffer_iter[cpu] = | |
2443 | ring_buffer_read_prepare(iter->tr->buffer, cpu); | |
2444 | ring_buffer_read_prepare_sync(); | |
2445 | ring_buffer_read_start(iter->buffer_iter[cpu]); | |
2446 | tracing_iter_reset(iter, cpu); | |
2447 | } | |
2448 | ||
2449 | mutex_unlock(&trace_types_lock); | |
2450 | ||
2451 | return iter; | |
2452 | ||
2453 | fail: | |
2454 | mutex_unlock(&trace_types_lock); | |
2455 | kfree(iter->trace); | |
2456 | kfree(iter->buffer_iter); | |
2457 | release: | |
2458 | seq_release_private(inode, file); | |
2459 | return ERR_PTR(-ENOMEM); | |
2460 | } | |
2461 | ||
2462 | int tracing_open_generic(struct inode *inode, struct file *filp) | |
2463 | { | |
2464 | if (tracing_disabled) | |
2465 | return -ENODEV; | |
2466 | ||
2467 | filp->private_data = inode->i_private; | |
2468 | return 0; | |
2469 | } | |
2470 | ||
2471 | static int tracing_release(struct inode *inode, struct file *file) | |
2472 | { | |
2473 | struct seq_file *m = file->private_data; | |
2474 | struct trace_iterator *iter; | |
2475 | int cpu; | |
2476 | ||
2477 | if (!(file->f_mode & FMODE_READ)) | |
2478 | return 0; | |
2479 | ||
2480 | iter = m->private; | |
2481 | ||
2482 | mutex_lock(&trace_types_lock); | |
2483 | for_each_tracing_cpu(cpu) { | |
2484 | if (iter->buffer_iter[cpu]) | |
2485 | ring_buffer_read_finish(iter->buffer_iter[cpu]); | |
2486 | } | |
2487 | ||
2488 | if (iter->trace && iter->trace->close) | |
2489 | iter->trace->close(iter); | |
2490 | ||
2491 | /* reenable tracing if it was previously enabled */ | |
2492 | tracing_start(); | |
2493 | mutex_unlock(&trace_types_lock); | |
2494 | ||
2495 | mutex_destroy(&iter->mutex); | |
2496 | free_cpumask_var(iter->started); | |
2497 | kfree(iter->trace); | |
2498 | kfree(iter->buffer_iter); | |
2499 | seq_release_private(inode, file); | |
2500 | return 0; | |
2501 | } | |
2502 | ||
2503 | static int tracing_open(struct inode *inode, struct file *file) | |
2504 | { | |
2505 | struct trace_iterator *iter; | |
2506 | int ret = 0; | |
2507 | ||
2508 | /* If this file was open for write, then erase contents */ | |
2509 | if ((file->f_mode & FMODE_WRITE) && | |
2510 | (file->f_flags & O_TRUNC)) { | |
2511 | long cpu = (long) inode->i_private; | |
2512 | ||
2513 | if (cpu == TRACE_PIPE_ALL_CPU) | |
2514 | tracing_reset_online_cpus(&global_trace); | |
2515 | else | |
2516 | tracing_reset(&global_trace, cpu); | |
2517 | } | |
2518 | ||
2519 | if (file->f_mode & FMODE_READ) { | |
2520 | iter = __tracing_open(inode, file); | |
2521 | if (IS_ERR(iter)) | |
2522 | ret = PTR_ERR(iter); | |
2523 | else if (trace_flags & TRACE_ITER_LATENCY_FMT) | |
2524 | iter->iter_flags |= TRACE_FILE_LAT_FMT; | |
2525 | } | |
2526 | return ret; | |
2527 | } | |
2528 | ||
2529 | static void * | |
2530 | t_next(struct seq_file *m, void *v, loff_t *pos) | |
2531 | { | |
2532 | struct tracer *t = v; | |
2533 | ||
2534 | (*pos)++; | |
2535 | ||
2536 | if (t) | |
2537 | t = t->next; | |
2538 | ||
2539 | return t; | |
2540 | } | |
2541 | ||
2542 | static void *t_start(struct seq_file *m, loff_t *pos) | |
2543 | { | |
2544 | struct tracer *t; | |
2545 | loff_t l = 0; | |
2546 | ||
2547 | mutex_lock(&trace_types_lock); | |
2548 | for (t = trace_types; t && l < *pos; t = t_next(m, t, &l)) | |
2549 | ; | |
2550 | ||
2551 | return t; | |
2552 | } | |
2553 | ||
2554 | static void t_stop(struct seq_file *m, void *p) | |
2555 | { | |
2556 | mutex_unlock(&trace_types_lock); | |
2557 | } | |
2558 | ||
2559 | static int t_show(struct seq_file *m, void *v) | |
2560 | { | |
2561 | struct tracer *t = v; | |
2562 | ||
2563 | if (!t) | |
2564 | return 0; | |
2565 | ||
2566 | seq_printf(m, "%s", t->name); | |
2567 | if (t->next) | |
2568 | seq_putc(m, ' '); | |
2569 | else | |
2570 | seq_putc(m, '\n'); | |
2571 | ||
2572 | return 0; | |
2573 | } | |
2574 | ||
2575 | static const struct seq_operations show_traces_seq_ops = { | |
2576 | .start = t_start, | |
2577 | .next = t_next, | |
2578 | .stop = t_stop, | |
2579 | .show = t_show, | |
2580 | }; | |
2581 | ||
2582 | static int show_traces_open(struct inode *inode, struct file *file) | |
2583 | { | |
2584 | if (tracing_disabled) | |
2585 | return -ENODEV; | |
2586 | ||
2587 | return seq_open(file, &show_traces_seq_ops); | |
2588 | } | |
2589 | ||
2590 | static ssize_t | |
2591 | tracing_write_stub(struct file *filp, const char __user *ubuf, | |
2592 | size_t count, loff_t *ppos) | |
2593 | { | |
2594 | return count; | |
2595 | } | |
2596 | ||
2597 | static loff_t tracing_seek(struct file *file, loff_t offset, int origin) | |
2598 | { | |
2599 | if (file->f_mode & FMODE_READ) | |
2600 | return seq_lseek(file, offset, origin); | |
2601 | else | |
2602 | return 0; | |
2603 | } | |
2604 | ||
2605 | static const struct file_operations tracing_fops = { | |
2606 | .open = tracing_open, | |
2607 | .read = seq_read, | |
2608 | .write = tracing_write_stub, | |
2609 | .llseek = tracing_seek, | |
2610 | .release = tracing_release, | |
2611 | }; | |
2612 | ||
2613 | static const struct file_operations show_traces_fops = { | |
2614 | .open = show_traces_open, | |
2615 | .read = seq_read, | |
2616 | .release = seq_release, | |
2617 | .llseek = seq_lseek, | |
2618 | }; | |
2619 | ||
2620 | /* | |
2621 | * Only trace on a CPU if the bitmask is set: | |
2622 | */ | |
2623 | static cpumask_var_t tracing_cpumask; | |
2624 | ||
2625 | /* | |
2626 | * The tracer itself will not take this lock, but still we want | |
2627 | * to provide a consistent cpumask to user-space: | |
2628 | */ | |
2629 | static DEFINE_MUTEX(tracing_cpumask_update_lock); | |
2630 | ||
2631 | /* | |
2632 | * Temporary storage for the character representation of the | |
2633 | * CPU bitmask (and one more byte for the newline): | |
2634 | */ | |
2635 | static char mask_str[NR_CPUS + 1]; | |
2636 | ||
2637 | static ssize_t | |
2638 | tracing_cpumask_read(struct file *filp, char __user *ubuf, | |
2639 | size_t count, loff_t *ppos) | |
2640 | { | |
2641 | int len; | |
2642 | ||
2643 | mutex_lock(&tracing_cpumask_update_lock); | |
2644 | ||
2645 | len = cpumask_scnprintf(mask_str, count, tracing_cpumask); | |
2646 | if (count - len < 2) { | |
2647 | count = -EINVAL; | |
2648 | goto out_err; | |
2649 | } | |
2650 | len += sprintf(mask_str + len, "\n"); | |
2651 | count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1); | |
2652 | ||
2653 | out_err: | |
2654 | mutex_unlock(&tracing_cpumask_update_lock); | |
2655 | ||
2656 | return count; | |
2657 | } | |
2658 | ||
2659 | static ssize_t | |
2660 | tracing_cpumask_write(struct file *filp, const char __user *ubuf, | |
2661 | size_t count, loff_t *ppos) | |
2662 | { | |
2663 | int err, cpu; | |
2664 | cpumask_var_t tracing_cpumask_new; | |
2665 | ||
2666 | if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL)) | |
2667 | return -ENOMEM; | |
2668 | ||
2669 | err = cpumask_parse_user(ubuf, count, tracing_cpumask_new); | |
2670 | if (err) | |
2671 | goto err_unlock; | |
2672 | ||
2673 | mutex_lock(&tracing_cpumask_update_lock); | |
2674 | ||
2675 | local_irq_disable(); | |
2676 | arch_spin_lock(&ftrace_max_lock); | |
2677 | for_each_tracing_cpu(cpu) { | |
2678 | /* | |
2679 | * Increase/decrease the disabled counter if we are | |
2680 | * about to flip a bit in the cpumask: | |
2681 | */ | |
2682 | if (cpumask_test_cpu(cpu, tracing_cpumask) && | |
2683 | !cpumask_test_cpu(cpu, tracing_cpumask_new)) { | |
2684 | atomic_inc(&global_trace.data[cpu]->disabled); | |
2685 | ring_buffer_record_disable_cpu(global_trace.buffer, cpu); | |
2686 | } | |
2687 | if (!cpumask_test_cpu(cpu, tracing_cpumask) && | |
2688 | cpumask_test_cpu(cpu, tracing_cpumask_new)) { | |
2689 | atomic_dec(&global_trace.data[cpu]->disabled); | |
2690 | ring_buffer_record_enable_cpu(global_trace.buffer, cpu); | |
2691 | } | |
2692 | } | |
2693 | arch_spin_unlock(&ftrace_max_lock); | |
2694 | local_irq_enable(); | |
2695 | ||
2696 | cpumask_copy(tracing_cpumask, tracing_cpumask_new); | |
2697 | ||
2698 | mutex_unlock(&tracing_cpumask_update_lock); | |
2699 | free_cpumask_var(tracing_cpumask_new); | |
2700 | ||
2701 | return count; | |
2702 | ||
2703 | err_unlock: | |
2704 | free_cpumask_var(tracing_cpumask_new); | |
2705 | ||
2706 | return err; | |
2707 | } | |
2708 | ||
2709 | static const struct file_operations tracing_cpumask_fops = { | |
2710 | .open = tracing_open_generic, | |
2711 | .read = tracing_cpumask_read, | |
2712 | .write = tracing_cpumask_write, | |
2713 | .llseek = generic_file_llseek, | |
2714 | }; | |
2715 | ||
2716 | static int tracing_trace_options_show(struct seq_file *m, void *v) | |
2717 | { | |
2718 | struct tracer_opt *trace_opts; | |
2719 | u32 tracer_flags; | |
2720 | int i; | |
2721 | ||
2722 | mutex_lock(&trace_types_lock); | |
2723 | tracer_flags = current_trace->flags->val; | |
2724 | trace_opts = current_trace->flags->opts; | |
2725 | ||
2726 | for (i = 0; trace_options[i]; i++) { | |
2727 | if (trace_flags & (1 << i)) | |
2728 | seq_printf(m, "%s\n", trace_options[i]); | |
2729 | else | |
2730 | seq_printf(m, "no%s\n", trace_options[i]); | |
2731 | } | |
2732 | ||
2733 | for (i = 0; trace_opts[i].name; i++) { | |
2734 | if (tracer_flags & trace_opts[i].bit) | |
2735 | seq_printf(m, "%s\n", trace_opts[i].name); | |
2736 | else | |
2737 | seq_printf(m, "no%s\n", trace_opts[i].name); | |
2738 | } | |
2739 | mutex_unlock(&trace_types_lock); | |
2740 | ||
2741 | return 0; | |
2742 | } | |
2743 | ||
2744 | static int __set_tracer_option(struct tracer *trace, | |
2745 | struct tracer_flags *tracer_flags, | |
2746 | struct tracer_opt *opts, int neg) | |
2747 | { | |
2748 | int ret; | |
2749 | ||
2750 | ret = trace->set_flag(tracer_flags->val, opts->bit, !neg); | |
2751 | if (ret) | |
2752 | return ret; | |
2753 | ||
2754 | if (neg) | |
2755 | tracer_flags->val &= ~opts->bit; | |
2756 | else | |
2757 | tracer_flags->val |= opts->bit; | |
2758 | return 0; | |
2759 | } | |
2760 | ||
2761 | /* Try to assign a tracer specific option */ | |
2762 | static int set_tracer_option(struct tracer *trace, char *cmp, int neg) | |
2763 | { | |
2764 | struct tracer_flags *tracer_flags = trace->flags; | |
2765 | struct tracer_opt *opts = NULL; | |
2766 | int i; | |
2767 | ||
2768 | for (i = 0; tracer_flags->opts[i].name; i++) { | |
2769 | opts = &tracer_flags->opts[i]; | |
2770 | ||
2771 | if (strcmp(cmp, opts->name) == 0) | |
2772 | return __set_tracer_option(trace, trace->flags, | |
2773 | opts, neg); | |
2774 | } | |
2775 | ||
2776 | return -EINVAL; | |
2777 | } | |
2778 | ||
2779 | static void set_tracer_flags(unsigned int mask, int enabled) | |
2780 | { | |
2781 | /* do nothing if flag is already set */ | |
2782 | if (!!(trace_flags & mask) == !!enabled) | |
2783 | return; | |
2784 | ||
2785 | if (enabled) | |
2786 | trace_flags |= mask; | |
2787 | else | |
2788 | trace_flags &= ~mask; | |
2789 | ||
2790 | if (mask == TRACE_ITER_RECORD_CMD) | |
2791 | trace_event_enable_cmd_record(enabled); | |
2792 | ||
2793 | if (mask == TRACE_ITER_OVERWRITE) | |
2794 | ring_buffer_change_overwrite(global_trace.buffer, enabled); | |
2795 | } | |
2796 | ||
2797 | static ssize_t | |
2798 | tracing_trace_options_write(struct file *filp, const char __user *ubuf, | |
2799 | size_t cnt, loff_t *ppos) | |
2800 | { | |
2801 | char buf[64]; | |
2802 | char *cmp; | |
2803 | int neg = 0; | |
2804 | int ret; | |
2805 | int i; | |
2806 | ||
2807 | if (cnt >= sizeof(buf)) | |
2808 | return -EINVAL; | |
2809 | ||
2810 | if (copy_from_user(&buf, ubuf, cnt)) | |
2811 | return -EFAULT; | |
2812 | ||
2813 | buf[cnt] = 0; | |
2814 | cmp = strstrip(buf); | |
2815 | ||
2816 | if (strncmp(cmp, "no", 2) == 0) { | |
2817 | neg = 1; | |
2818 | cmp += 2; | |
2819 | } | |
2820 | ||
2821 | for (i = 0; trace_options[i]; i++) { | |
2822 | if (strcmp(cmp, trace_options[i]) == 0) { | |
2823 | set_tracer_flags(1 << i, !neg); | |
2824 | break; | |
2825 | } | |
2826 | } | |
2827 | ||
2828 | /* If no option could be set, test the specific tracer options */ | |
2829 | if (!trace_options[i]) { | |
2830 | mutex_lock(&trace_types_lock); | |
2831 | ret = set_tracer_option(current_trace, cmp, neg); | |
2832 | mutex_unlock(&trace_types_lock); | |
2833 | if (ret) | |
2834 | return ret; | |
2835 | } | |
2836 | ||
2837 | *ppos += cnt; | |
2838 | ||
2839 | return cnt; | |
2840 | } | |
2841 | ||
2842 | static int tracing_trace_options_open(struct inode *inode, struct file *file) | |
2843 | { | |
2844 | if (tracing_disabled) | |
2845 | return -ENODEV; | |
2846 | return single_open(file, tracing_trace_options_show, NULL); | |
2847 | } | |
2848 | ||
2849 | static const struct file_operations tracing_iter_fops = { | |
2850 | .open = tracing_trace_options_open, | |
2851 | .read = seq_read, | |
2852 | .llseek = seq_lseek, | |
2853 | .release = single_release, | |
2854 | .write = tracing_trace_options_write, | |
2855 | }; | |
2856 | ||
2857 | static const char readme_msg[] = | |
2858 | "tracing mini-HOWTO:\n\n" | |
2859 | "# mount -t debugfs nodev /sys/kernel/debug\n\n" | |
2860 | "# cat /sys/kernel/debug/tracing/available_tracers\n" | |
2861 | "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n" | |
2862 | "# cat /sys/kernel/debug/tracing/current_tracer\n" | |
2863 | "nop\n" | |
2864 | "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n" | |
2865 | "# cat /sys/kernel/debug/tracing/current_tracer\n" | |
2866 | "wakeup\n" | |
2867 | "# cat /sys/kernel/debug/tracing/trace_options\n" | |
2868 | "noprint-parent nosym-offset nosym-addr noverbose\n" | |
2869 | "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n" | |
2870 | "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n" | |
2871 | "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n" | |
2872 | "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n" | |
2873 | ; | |
2874 | ||
2875 | static ssize_t | |
2876 | tracing_readme_read(struct file *filp, char __user *ubuf, | |
2877 | size_t cnt, loff_t *ppos) | |
2878 | { | |
2879 | return simple_read_from_buffer(ubuf, cnt, ppos, | |
2880 | readme_msg, strlen(readme_msg)); | |
2881 | } | |
2882 | ||
2883 | static const struct file_operations tracing_readme_fops = { | |
2884 | .open = tracing_open_generic, | |
2885 | .read = tracing_readme_read, | |
2886 | .llseek = generic_file_llseek, | |
2887 | }; | |
2888 | ||
2889 | static ssize_t | |
2890 | tracing_saved_cmdlines_read(struct file *file, char __user *ubuf, | |
2891 | size_t cnt, loff_t *ppos) | |
2892 | { | |
2893 | char *buf_comm; | |
2894 | char *file_buf; | |
2895 | char *buf; | |
2896 | int len = 0; | |
2897 | int pid; | |
2898 | int i; | |
2899 | ||
2900 | file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL); | |
2901 | if (!file_buf) | |
2902 | return -ENOMEM; | |
2903 | ||
2904 | buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL); | |
2905 | if (!buf_comm) { | |
2906 | kfree(file_buf); | |
2907 | return -ENOMEM; | |
2908 | } | |
2909 | ||
2910 | buf = file_buf; | |
2911 | ||
2912 | for (i = 0; i < SAVED_CMDLINES; i++) { | |
2913 | int r; | |
2914 | ||
2915 | pid = map_cmdline_to_pid[i]; | |
2916 | if (pid == -1 || pid == NO_CMDLINE_MAP) | |
2917 | continue; | |
2918 | ||
2919 | trace_find_cmdline(pid, buf_comm); | |
2920 | r = sprintf(buf, "%d %s\n", pid, buf_comm); | |
2921 | buf += r; | |
2922 | len += r; | |
2923 | } | |
2924 | ||
2925 | len = simple_read_from_buffer(ubuf, cnt, ppos, | |
2926 | file_buf, len); | |
2927 | ||
2928 | kfree(file_buf); | |
2929 | kfree(buf_comm); | |
2930 | ||
2931 | return len; | |
2932 | } | |
2933 | ||
2934 | static const struct file_operations tracing_saved_cmdlines_fops = { | |
2935 | .open = tracing_open_generic, | |
2936 | .read = tracing_saved_cmdlines_read, | |
2937 | .llseek = generic_file_llseek, | |
2938 | }; | |
2939 | ||
2940 | static ssize_t | |
2941 | tracing_ctrl_read(struct file *filp, char __user *ubuf, | |
2942 | size_t cnt, loff_t *ppos) | |
2943 | { | |
2944 | char buf[64]; | |
2945 | int r; | |
2946 | ||
2947 | r = sprintf(buf, "%u\n", tracer_enabled); | |
2948 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
2949 | } | |
2950 | ||
2951 | static ssize_t | |
2952 | tracing_ctrl_write(struct file *filp, const char __user *ubuf, | |
2953 | size_t cnt, loff_t *ppos) | |
2954 | { | |
2955 | struct trace_array *tr = filp->private_data; | |
2956 | unsigned long val; | |
2957 | int ret; | |
2958 | ||
2959 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
2960 | if (ret) | |
2961 | return ret; | |
2962 | ||
2963 | val = !!val; | |
2964 | ||
2965 | mutex_lock(&trace_types_lock); | |
2966 | if (tracer_enabled ^ val) { | |
2967 | ||
2968 | /* Only need to warn if this is used to change the state */ | |
2969 | WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on"); | |
2970 | ||
2971 | if (val) { | |
2972 | tracer_enabled = 1; | |
2973 | if (current_trace->start) | |
2974 | current_trace->start(tr); | |
2975 | tracing_start(); | |
2976 | } else { | |
2977 | tracer_enabled = 0; | |
2978 | tracing_stop(); | |
2979 | if (current_trace->stop) | |
2980 | current_trace->stop(tr); | |
2981 | } | |
2982 | } | |
2983 | mutex_unlock(&trace_types_lock); | |
2984 | ||
2985 | *ppos += cnt; | |
2986 | ||
2987 | return cnt; | |
2988 | } | |
2989 | ||
2990 | static ssize_t | |
2991 | tracing_set_trace_read(struct file *filp, char __user *ubuf, | |
2992 | size_t cnt, loff_t *ppos) | |
2993 | { | |
2994 | char buf[MAX_TRACER_SIZE+2]; | |
2995 | int r; | |
2996 | ||
2997 | mutex_lock(&trace_types_lock); | |
2998 | if (current_trace) | |
2999 | r = sprintf(buf, "%s\n", current_trace->name); | |
3000 | else | |
3001 | r = sprintf(buf, "\n"); | |
3002 | mutex_unlock(&trace_types_lock); | |
3003 | ||
3004 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3005 | } | |
3006 | ||
3007 | int tracer_init(struct tracer *t, struct trace_array *tr) | |
3008 | { | |
3009 | tracing_reset_online_cpus(tr); | |
3010 | return t->init(tr); | |
3011 | } | |
3012 | ||
3013 | static void set_buffer_entries(struct trace_array *tr, unsigned long val) | |
3014 | { | |
3015 | int cpu; | |
3016 | for_each_tracing_cpu(cpu) | |
3017 | tr->data[cpu]->entries = val; | |
3018 | } | |
3019 | ||
3020 | static int __tracing_resize_ring_buffer(unsigned long size, int cpu) | |
3021 | { | |
3022 | int ret; | |
3023 | ||
3024 | /* | |
3025 | * If kernel or user changes the size of the ring buffer | |
3026 | * we use the size that was given, and we can forget about | |
3027 | * expanding it later. | |
3028 | */ | |
3029 | ring_buffer_expanded = 1; | |
3030 | ||
3031 | ret = ring_buffer_resize(global_trace.buffer, size, cpu); | |
3032 | if (ret < 0) | |
3033 | return ret; | |
3034 | ||
3035 | if (!current_trace->use_max_tr) | |
3036 | goto out; | |
3037 | ||
3038 | ret = ring_buffer_resize(max_tr.buffer, size, cpu); | |
3039 | if (ret < 0) { | |
3040 | int r = 0; | |
3041 | ||
3042 | if (cpu == RING_BUFFER_ALL_CPUS) { | |
3043 | int i; | |
3044 | for_each_tracing_cpu(i) { | |
3045 | r = ring_buffer_resize(global_trace.buffer, | |
3046 | global_trace.data[i]->entries, | |
3047 | i); | |
3048 | if (r < 0) | |
3049 | break; | |
3050 | } | |
3051 | } else { | |
3052 | r = ring_buffer_resize(global_trace.buffer, | |
3053 | global_trace.data[cpu]->entries, | |
3054 | cpu); | |
3055 | } | |
3056 | ||
3057 | if (r < 0) { | |
3058 | /* | |
3059 | * AARGH! We are left with different | |
3060 | * size max buffer!!!! | |
3061 | * The max buffer is our "snapshot" buffer. | |
3062 | * When a tracer needs a snapshot (one of the | |
3063 | * latency tracers), it swaps the max buffer | |
3064 | * with the saved snap shot. We succeeded to | |
3065 | * update the size of the main buffer, but failed to | |
3066 | * update the size of the max buffer. But when we tried | |
3067 | * to reset the main buffer to the original size, we | |
3068 | * failed there too. This is very unlikely to | |
3069 | * happen, but if it does, warn and kill all | |
3070 | * tracing. | |
3071 | */ | |
3072 | WARN_ON(1); | |
3073 | tracing_disabled = 1; | |
3074 | } | |
3075 | return ret; | |
3076 | } | |
3077 | ||
3078 | if (cpu == RING_BUFFER_ALL_CPUS) | |
3079 | set_buffer_entries(&max_tr, size); | |
3080 | else | |
3081 | max_tr.data[cpu]->entries = size; | |
3082 | ||
3083 | out: | |
3084 | if (cpu == RING_BUFFER_ALL_CPUS) | |
3085 | set_buffer_entries(&global_trace, size); | |
3086 | else | |
3087 | global_trace.data[cpu]->entries = size; | |
3088 | ||
3089 | return ret; | |
3090 | } | |
3091 | ||
3092 | static ssize_t tracing_resize_ring_buffer(unsigned long size, int cpu_id) | |
3093 | { | |
3094 | int ret = size; | |
3095 | ||
3096 | mutex_lock(&trace_types_lock); | |
3097 | ||
3098 | if (cpu_id != RING_BUFFER_ALL_CPUS) { | |
3099 | /* make sure, this cpu is enabled in the mask */ | |
3100 | if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) { | |
3101 | ret = -EINVAL; | |
3102 | goto out; | |
3103 | } | |
3104 | } | |
3105 | ||
3106 | ret = __tracing_resize_ring_buffer(size, cpu_id); | |
3107 | if (ret < 0) | |
3108 | ret = -ENOMEM; | |
3109 | ||
3110 | out: | |
3111 | mutex_unlock(&trace_types_lock); | |
3112 | ||
3113 | return ret; | |
3114 | } | |
3115 | ||
3116 | ||
3117 | /** | |
3118 | * tracing_update_buffers - used by tracing facility to expand ring buffers | |
3119 | * | |
3120 | * To save on memory when the tracing is never used on a system with it | |
3121 | * configured in. The ring buffers are set to a minimum size. But once | |
3122 | * a user starts to use the tracing facility, then they need to grow | |
3123 | * to their default size. | |
3124 | * | |
3125 | * This function is to be called when a tracer is about to be used. | |
3126 | */ | |
3127 | int tracing_update_buffers(void) | |
3128 | { | |
3129 | int ret = 0; | |
3130 | ||
3131 | mutex_lock(&trace_types_lock); | |
3132 | if (!ring_buffer_expanded) | |
3133 | ret = __tracing_resize_ring_buffer(trace_buf_size, | |
3134 | RING_BUFFER_ALL_CPUS); | |
3135 | mutex_unlock(&trace_types_lock); | |
3136 | ||
3137 | return ret; | |
3138 | } | |
3139 | ||
3140 | struct trace_option_dentry; | |
3141 | ||
3142 | static struct trace_option_dentry * | |
3143 | create_trace_option_files(struct tracer *tracer); | |
3144 | ||
3145 | static void | |
3146 | destroy_trace_option_files(struct trace_option_dentry *topts); | |
3147 | ||
3148 | static int tracing_set_tracer(const char *buf) | |
3149 | { | |
3150 | static struct trace_option_dentry *topts; | |
3151 | struct trace_array *tr = &global_trace; | |
3152 | struct tracer *t; | |
3153 | int ret = 0; | |
3154 | ||
3155 | mutex_lock(&trace_types_lock); | |
3156 | ||
3157 | if (!ring_buffer_expanded) { | |
3158 | ret = __tracing_resize_ring_buffer(trace_buf_size, | |
3159 | RING_BUFFER_ALL_CPUS); | |
3160 | if (ret < 0) | |
3161 | goto out; | |
3162 | ret = 0; | |
3163 | } | |
3164 | ||
3165 | for (t = trace_types; t; t = t->next) { | |
3166 | if (strcmp(t->name, buf) == 0) | |
3167 | break; | |
3168 | } | |
3169 | if (!t) { | |
3170 | ret = -EINVAL; | |
3171 | goto out; | |
3172 | } | |
3173 | if (t == current_trace) | |
3174 | goto out; | |
3175 | ||
3176 | trace_branch_disable(); | |
3177 | if (current_trace && current_trace->reset) | |
3178 | current_trace->reset(tr); | |
3179 | if (current_trace && current_trace->use_max_tr) { | |
3180 | /* | |
3181 | * We don't free the ring buffer. instead, resize it because | |
3182 | * The max_tr ring buffer has some state (e.g. ring->clock) and | |
3183 | * we want preserve it. | |
3184 | */ | |
3185 | ring_buffer_resize(max_tr.buffer, 1, RING_BUFFER_ALL_CPUS); | |
3186 | set_buffer_entries(&max_tr, 1); | |
3187 | } | |
3188 | destroy_trace_option_files(topts); | |
3189 | ||
3190 | current_trace = &nop_trace; | |
3191 | ||
3192 | topts = create_trace_option_files(t); | |
3193 | if (t->use_max_tr) { | |
3194 | int cpu; | |
3195 | /* we need to make per cpu buffer sizes equivalent */ | |
3196 | for_each_tracing_cpu(cpu) { | |
3197 | ret = ring_buffer_resize(max_tr.buffer, | |
3198 | global_trace.data[cpu]->entries, | |
3199 | cpu); | |
3200 | if (ret < 0) | |
3201 | goto out; | |
3202 | max_tr.data[cpu]->entries = | |
3203 | global_trace.data[cpu]->entries; | |
3204 | } | |
3205 | } | |
3206 | ||
3207 | if (t->init) { | |
3208 | ret = tracer_init(t, tr); | |
3209 | if (ret) | |
3210 | goto out; | |
3211 | } | |
3212 | ||
3213 | current_trace = t; | |
3214 | trace_branch_enable(tr); | |
3215 | out: | |
3216 | mutex_unlock(&trace_types_lock); | |
3217 | ||
3218 | return ret; | |
3219 | } | |
3220 | ||
3221 | static ssize_t | |
3222 | tracing_set_trace_write(struct file *filp, const char __user *ubuf, | |
3223 | size_t cnt, loff_t *ppos) | |
3224 | { | |
3225 | char buf[MAX_TRACER_SIZE+1]; | |
3226 | int i; | |
3227 | size_t ret; | |
3228 | int err; | |
3229 | ||
3230 | ret = cnt; | |
3231 | ||
3232 | if (cnt > MAX_TRACER_SIZE) | |
3233 | cnt = MAX_TRACER_SIZE; | |
3234 | ||
3235 | if (copy_from_user(&buf, ubuf, cnt)) | |
3236 | return -EFAULT; | |
3237 | ||
3238 | buf[cnt] = 0; | |
3239 | ||
3240 | /* strip ending whitespace. */ | |
3241 | for (i = cnt - 1; i > 0 && isspace(buf[i]); i--) | |
3242 | buf[i] = 0; | |
3243 | ||
3244 | err = tracing_set_tracer(buf); | |
3245 | if (err) | |
3246 | return err; | |
3247 | ||
3248 | *ppos += ret; | |
3249 | ||
3250 | return ret; | |
3251 | } | |
3252 | ||
3253 | static ssize_t | |
3254 | tracing_max_lat_read(struct file *filp, char __user *ubuf, | |
3255 | size_t cnt, loff_t *ppos) | |
3256 | { | |
3257 | unsigned long *ptr = filp->private_data; | |
3258 | char buf[64]; | |
3259 | int r; | |
3260 | ||
3261 | r = snprintf(buf, sizeof(buf), "%ld\n", | |
3262 | *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr)); | |
3263 | if (r > sizeof(buf)) | |
3264 | r = sizeof(buf); | |
3265 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3266 | } | |
3267 | ||
3268 | static ssize_t | |
3269 | tracing_max_lat_write(struct file *filp, const char __user *ubuf, | |
3270 | size_t cnt, loff_t *ppos) | |
3271 | { | |
3272 | unsigned long *ptr = filp->private_data; | |
3273 | unsigned long val; | |
3274 | int ret; | |
3275 | ||
3276 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
3277 | if (ret) | |
3278 | return ret; | |
3279 | ||
3280 | *ptr = val * 1000; | |
3281 | ||
3282 | return cnt; | |
3283 | } | |
3284 | ||
3285 | static int tracing_open_pipe(struct inode *inode, struct file *filp) | |
3286 | { | |
3287 | long cpu_file = (long) inode->i_private; | |
3288 | struct trace_iterator *iter; | |
3289 | int ret = 0; | |
3290 | ||
3291 | if (tracing_disabled) | |
3292 | return -ENODEV; | |
3293 | ||
3294 | mutex_lock(&trace_types_lock); | |
3295 | ||
3296 | /* create a buffer to store the information to pass to userspace */ | |
3297 | iter = kzalloc(sizeof(*iter), GFP_KERNEL); | |
3298 | if (!iter) { | |
3299 | ret = -ENOMEM; | |
3300 | goto out; | |
3301 | } | |
3302 | ||
3303 | /* | |
3304 | * We make a copy of the current tracer to avoid concurrent | |
3305 | * changes on it while we are reading. | |
3306 | */ | |
3307 | iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL); | |
3308 | if (!iter->trace) { | |
3309 | ret = -ENOMEM; | |
3310 | goto fail; | |
3311 | } | |
3312 | if (current_trace) | |
3313 | *iter->trace = *current_trace; | |
3314 | ||
3315 | if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) { | |
3316 | ret = -ENOMEM; | |
3317 | goto fail; | |
3318 | } | |
3319 | ||
3320 | /* trace pipe does not show start of buffer */ | |
3321 | cpumask_setall(iter->started); | |
3322 | ||
3323 | if (trace_flags & TRACE_ITER_LATENCY_FMT) | |
3324 | iter->iter_flags |= TRACE_FILE_LAT_FMT; | |
3325 | ||
3326 | iter->cpu_file = cpu_file; | |
3327 | iter->tr = &global_trace; | |
3328 | mutex_init(&iter->mutex); | |
3329 | filp->private_data = iter; | |
3330 | ||
3331 | if (iter->trace->pipe_open) | |
3332 | iter->trace->pipe_open(iter); | |
3333 | ||
3334 | nonseekable_open(inode, filp); | |
3335 | out: | |
3336 | mutex_unlock(&trace_types_lock); | |
3337 | return ret; | |
3338 | ||
3339 | fail: | |
3340 | kfree(iter->trace); | |
3341 | kfree(iter); | |
3342 | mutex_unlock(&trace_types_lock); | |
3343 | return ret; | |
3344 | } | |
3345 | ||
3346 | static int tracing_release_pipe(struct inode *inode, struct file *file) | |
3347 | { | |
3348 | struct trace_iterator *iter = file->private_data; | |
3349 | ||
3350 | mutex_lock(&trace_types_lock); | |
3351 | ||
3352 | if (iter->trace->pipe_close) | |
3353 | iter->trace->pipe_close(iter); | |
3354 | ||
3355 | mutex_unlock(&trace_types_lock); | |
3356 | ||
3357 | free_cpumask_var(iter->started); | |
3358 | mutex_destroy(&iter->mutex); | |
3359 | kfree(iter->trace); | |
3360 | kfree(iter); | |
3361 | ||
3362 | return 0; | |
3363 | } | |
3364 | ||
3365 | static unsigned int | |
3366 | tracing_poll_pipe(struct file *filp, poll_table *poll_table) | |
3367 | { | |
3368 | struct trace_iterator *iter = filp->private_data; | |
3369 | ||
3370 | if (trace_flags & TRACE_ITER_BLOCK) { | |
3371 | /* | |
3372 | * Always select as readable when in blocking mode | |
3373 | */ | |
3374 | return POLLIN | POLLRDNORM; | |
3375 | } else { | |
3376 | if (!trace_empty(iter)) | |
3377 | return POLLIN | POLLRDNORM; | |
3378 | poll_wait(filp, &trace_wait, poll_table); | |
3379 | if (!trace_empty(iter)) | |
3380 | return POLLIN | POLLRDNORM; | |
3381 | ||
3382 | return 0; | |
3383 | } | |
3384 | } | |
3385 | ||
3386 | ||
3387 | void default_wait_pipe(struct trace_iterator *iter) | |
3388 | { | |
3389 | DEFINE_WAIT(wait); | |
3390 | ||
3391 | prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE); | |
3392 | ||
3393 | if (trace_empty(iter)) | |
3394 | schedule(); | |
3395 | ||
3396 | finish_wait(&trace_wait, &wait); | |
3397 | } | |
3398 | ||
3399 | /* | |
3400 | * This is a make-shift waitqueue. | |
3401 | * A tracer might use this callback on some rare cases: | |
3402 | * | |
3403 | * 1) the current tracer might hold the runqueue lock when it wakes up | |
3404 | * a reader, hence a deadlock (sched, function, and function graph tracers) | |
3405 | * 2) the function tracers, trace all functions, we don't want | |
3406 | * the overhead of calling wake_up and friends | |
3407 | * (and tracing them too) | |
3408 | * | |
3409 | * Anyway, this is really very primitive wakeup. | |
3410 | */ | |
3411 | void poll_wait_pipe(struct trace_iterator *iter) | |
3412 | { | |
3413 | set_current_state(TASK_INTERRUPTIBLE); | |
3414 | /* sleep for 100 msecs, and try again. */ | |
3415 | schedule_timeout(HZ / 10); | |
3416 | } | |
3417 | ||
3418 | /* Must be called with trace_types_lock mutex held. */ | |
3419 | static int tracing_wait_pipe(struct file *filp) | |
3420 | { | |
3421 | struct trace_iterator *iter = filp->private_data; | |
3422 | ||
3423 | while (trace_empty(iter)) { | |
3424 | ||
3425 | if ((filp->f_flags & O_NONBLOCK)) { | |
3426 | return -EAGAIN; | |
3427 | } | |
3428 | ||
3429 | mutex_unlock(&iter->mutex); | |
3430 | ||
3431 | iter->trace->wait_pipe(iter); | |
3432 | ||
3433 | mutex_lock(&iter->mutex); | |
3434 | ||
3435 | if (signal_pending(current)) | |
3436 | return -EINTR; | |
3437 | ||
3438 | /* | |
3439 | * We block until we read something and tracing is disabled. | |
3440 | * We still block if tracing is disabled, but we have never | |
3441 | * read anything. This allows a user to cat this file, and | |
3442 | * then enable tracing. But after we have read something, | |
3443 | * we give an EOF when tracing is again disabled. | |
3444 | * | |
3445 | * iter->pos will be 0 if we haven't read anything. | |
3446 | */ | |
3447 | if (!tracer_enabled && iter->pos) | |
3448 | break; | |
3449 | } | |
3450 | ||
3451 | return 1; | |
3452 | } | |
3453 | ||
3454 | /* | |
3455 | * Consumer reader. | |
3456 | */ | |
3457 | static ssize_t | |
3458 | tracing_read_pipe(struct file *filp, char __user *ubuf, | |
3459 | size_t cnt, loff_t *ppos) | |
3460 | { | |
3461 | struct trace_iterator *iter = filp->private_data; | |
3462 | static struct tracer *old_tracer; | |
3463 | ssize_t sret; | |
3464 | ||
3465 | /* return any leftover data */ | |
3466 | sret = trace_seq_to_user(&iter->seq, ubuf, cnt); | |
3467 | if (sret != -EBUSY) | |
3468 | return sret; | |
3469 | ||
3470 | trace_seq_init(&iter->seq); | |
3471 | ||
3472 | /* copy the tracer to avoid using a global lock all around */ | |
3473 | mutex_lock(&trace_types_lock); | |
3474 | if (unlikely(old_tracer != current_trace && current_trace)) { | |
3475 | old_tracer = current_trace; | |
3476 | *iter->trace = *current_trace; | |
3477 | } | |
3478 | mutex_unlock(&trace_types_lock); | |
3479 | ||
3480 | /* | |
3481 | * Avoid more than one consumer on a single file descriptor | |
3482 | * This is just a matter of traces coherency, the ring buffer itself | |
3483 | * is protected. | |
3484 | */ | |
3485 | mutex_lock(&iter->mutex); | |
3486 | if (iter->trace->read) { | |
3487 | sret = iter->trace->read(iter, filp, ubuf, cnt, ppos); | |
3488 | if (sret) | |
3489 | goto out; | |
3490 | } | |
3491 | ||
3492 | waitagain: | |
3493 | sret = tracing_wait_pipe(filp); | |
3494 | if (sret <= 0) | |
3495 | goto out; | |
3496 | ||
3497 | /* stop when tracing is finished */ | |
3498 | if (trace_empty(iter)) { | |
3499 | sret = 0; | |
3500 | goto out; | |
3501 | } | |
3502 | ||
3503 | if (cnt >= PAGE_SIZE) | |
3504 | cnt = PAGE_SIZE - 1; | |
3505 | ||
3506 | /* reset all but tr, trace, and overruns */ | |
3507 | memset(&iter->seq, 0, | |
3508 | sizeof(struct trace_iterator) - | |
3509 | offsetof(struct trace_iterator, seq)); | |
3510 | iter->pos = -1; | |
3511 | ||
3512 | trace_event_read_lock(); | |
3513 | trace_access_lock(iter->cpu_file); | |
3514 | while (trace_find_next_entry_inc(iter) != NULL) { | |
3515 | enum print_line_t ret; | |
3516 | int len = iter->seq.len; | |
3517 | ||
3518 | ret = print_trace_line(iter); | |
3519 | if (ret == TRACE_TYPE_PARTIAL_LINE) { | |
3520 | /* don't print partial lines */ | |
3521 | iter->seq.len = len; | |
3522 | break; | |
3523 | } | |
3524 | if (ret != TRACE_TYPE_NO_CONSUME) | |
3525 | trace_consume(iter); | |
3526 | ||
3527 | if (iter->seq.len >= cnt) | |
3528 | break; | |
3529 | ||
3530 | /* | |
3531 | * Setting the full flag means we reached the trace_seq buffer | |
3532 | * size and we should leave by partial output condition above. | |
3533 | * One of the trace_seq_* functions is not used properly. | |
3534 | */ | |
3535 | WARN_ONCE(iter->seq.full, "full flag set for trace type %d", | |
3536 | iter->ent->type); | |
3537 | } | |
3538 | trace_access_unlock(iter->cpu_file); | |
3539 | trace_event_read_unlock(); | |
3540 | ||
3541 | /* Now copy what we have to the user */ | |
3542 | sret = trace_seq_to_user(&iter->seq, ubuf, cnt); | |
3543 | if (iter->seq.readpos >= iter->seq.len) | |
3544 | trace_seq_init(&iter->seq); | |
3545 | ||
3546 | /* | |
3547 | * If there was nothing to send to user, in spite of consuming trace | |
3548 | * entries, go back to wait for more entries. | |
3549 | */ | |
3550 | if (sret == -EBUSY) | |
3551 | goto waitagain; | |
3552 | ||
3553 | out: | |
3554 | mutex_unlock(&iter->mutex); | |
3555 | ||
3556 | return sret; | |
3557 | } | |
3558 | ||
3559 | static void tracing_pipe_buf_release(struct pipe_inode_info *pipe, | |
3560 | struct pipe_buffer *buf) | |
3561 | { | |
3562 | __free_page(buf->page); | |
3563 | } | |
3564 | ||
3565 | static void tracing_spd_release_pipe(struct splice_pipe_desc *spd, | |
3566 | unsigned int idx) | |
3567 | { | |
3568 | __free_page(spd->pages[idx]); | |
3569 | } | |
3570 | ||
3571 | static const struct pipe_buf_operations tracing_pipe_buf_ops = { | |
3572 | .can_merge = 0, | |
3573 | .map = generic_pipe_buf_map, | |
3574 | .unmap = generic_pipe_buf_unmap, | |
3575 | .confirm = generic_pipe_buf_confirm, | |
3576 | .release = tracing_pipe_buf_release, | |
3577 | .steal = generic_pipe_buf_steal, | |
3578 | .get = generic_pipe_buf_get, | |
3579 | }; | |
3580 | ||
3581 | static size_t | |
3582 | tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter) | |
3583 | { | |
3584 | size_t count; | |
3585 | int ret; | |
3586 | ||
3587 | /* Seq buffer is page-sized, exactly what we need. */ | |
3588 | for (;;) { | |
3589 | count = iter->seq.len; | |
3590 | ret = print_trace_line(iter); | |
3591 | count = iter->seq.len - count; | |
3592 | if (rem < count) { | |
3593 | rem = 0; | |
3594 | iter->seq.len -= count; | |
3595 | break; | |
3596 | } | |
3597 | if (ret == TRACE_TYPE_PARTIAL_LINE) { | |
3598 | iter->seq.len -= count; | |
3599 | break; | |
3600 | } | |
3601 | ||
3602 | if (ret != TRACE_TYPE_NO_CONSUME) | |
3603 | trace_consume(iter); | |
3604 | rem -= count; | |
3605 | if (!trace_find_next_entry_inc(iter)) { | |
3606 | rem = 0; | |
3607 | iter->ent = NULL; | |
3608 | break; | |
3609 | } | |
3610 | } | |
3611 | ||
3612 | return rem; | |
3613 | } | |
3614 | ||
3615 | static ssize_t tracing_splice_read_pipe(struct file *filp, | |
3616 | loff_t *ppos, | |
3617 | struct pipe_inode_info *pipe, | |
3618 | size_t len, | |
3619 | unsigned int flags) | |
3620 | { | |
3621 | struct page *pages_def[PIPE_DEF_BUFFERS]; | |
3622 | struct partial_page partial_def[PIPE_DEF_BUFFERS]; | |
3623 | struct trace_iterator *iter = filp->private_data; | |
3624 | struct splice_pipe_desc spd = { | |
3625 | .pages = pages_def, | |
3626 | .partial = partial_def, | |
3627 | .nr_pages = 0, /* This gets updated below. */ | |
3628 | .nr_pages_max = PIPE_DEF_BUFFERS, | |
3629 | .flags = flags, | |
3630 | .ops = &tracing_pipe_buf_ops, | |
3631 | .spd_release = tracing_spd_release_pipe, | |
3632 | }; | |
3633 | static struct tracer *old_tracer; | |
3634 | ssize_t ret; | |
3635 | size_t rem; | |
3636 | unsigned int i; | |
3637 | ||
3638 | if (splice_grow_spd(pipe, &spd)) | |
3639 | return -ENOMEM; | |
3640 | ||
3641 | /* copy the tracer to avoid using a global lock all around */ | |
3642 | mutex_lock(&trace_types_lock); | |
3643 | if (unlikely(old_tracer != current_trace && current_trace)) { | |
3644 | old_tracer = current_trace; | |
3645 | *iter->trace = *current_trace; | |
3646 | } | |
3647 | mutex_unlock(&trace_types_lock); | |
3648 | ||
3649 | mutex_lock(&iter->mutex); | |
3650 | ||
3651 | if (iter->trace->splice_read) { | |
3652 | ret = iter->trace->splice_read(iter, filp, | |
3653 | ppos, pipe, len, flags); | |
3654 | if (ret) | |
3655 | goto out_err; | |
3656 | } | |
3657 | ||
3658 | ret = tracing_wait_pipe(filp); | |
3659 | if (ret <= 0) | |
3660 | goto out_err; | |
3661 | ||
3662 | if (!iter->ent && !trace_find_next_entry_inc(iter)) { | |
3663 | ret = -EFAULT; | |
3664 | goto out_err; | |
3665 | } | |
3666 | ||
3667 | trace_event_read_lock(); | |
3668 | trace_access_lock(iter->cpu_file); | |
3669 | ||
3670 | /* Fill as many pages as possible. */ | |
3671 | for (i = 0, rem = len; i < pipe->buffers && rem; i++) { | |
3672 | spd.pages[i] = alloc_page(GFP_KERNEL); | |
3673 | if (!spd.pages[i]) | |
3674 | break; | |
3675 | ||
3676 | rem = tracing_fill_pipe_page(rem, iter); | |
3677 | ||
3678 | /* Copy the data into the page, so we can start over. */ | |
3679 | ret = trace_seq_to_buffer(&iter->seq, | |
3680 | page_address(spd.pages[i]), | |
3681 | iter->seq.len); | |
3682 | if (ret < 0) { | |
3683 | __free_page(spd.pages[i]); | |
3684 | break; | |
3685 | } | |
3686 | spd.partial[i].offset = 0; | |
3687 | spd.partial[i].len = iter->seq.len; | |
3688 | ||
3689 | trace_seq_init(&iter->seq); | |
3690 | } | |
3691 | ||
3692 | trace_access_unlock(iter->cpu_file); | |
3693 | trace_event_read_unlock(); | |
3694 | mutex_unlock(&iter->mutex); | |
3695 | ||
3696 | spd.nr_pages = i; | |
3697 | ||
3698 | ret = splice_to_pipe(pipe, &spd); | |
3699 | out: | |
3700 | splice_shrink_spd(&spd); | |
3701 | return ret; | |
3702 | ||
3703 | out_err: | |
3704 | mutex_unlock(&iter->mutex); | |
3705 | goto out; | |
3706 | } | |
3707 | ||
3708 | struct ftrace_entries_info { | |
3709 | struct trace_array *tr; | |
3710 | int cpu; | |
3711 | }; | |
3712 | ||
3713 | static int tracing_entries_open(struct inode *inode, struct file *filp) | |
3714 | { | |
3715 | struct ftrace_entries_info *info; | |
3716 | ||
3717 | if (tracing_disabled) | |
3718 | return -ENODEV; | |
3719 | ||
3720 | info = kzalloc(sizeof(*info), GFP_KERNEL); | |
3721 | if (!info) | |
3722 | return -ENOMEM; | |
3723 | ||
3724 | info->tr = &global_trace; | |
3725 | info->cpu = (unsigned long)inode->i_private; | |
3726 | ||
3727 | filp->private_data = info; | |
3728 | ||
3729 | return 0; | |
3730 | } | |
3731 | ||
3732 | static ssize_t | |
3733 | tracing_entries_read(struct file *filp, char __user *ubuf, | |
3734 | size_t cnt, loff_t *ppos) | |
3735 | { | |
3736 | struct ftrace_entries_info *info = filp->private_data; | |
3737 | struct trace_array *tr = info->tr; | |
3738 | char buf[64]; | |
3739 | int r = 0; | |
3740 | ssize_t ret; | |
3741 | ||
3742 | mutex_lock(&trace_types_lock); | |
3743 | ||
3744 | if (info->cpu == RING_BUFFER_ALL_CPUS) { | |
3745 | int cpu, buf_size_same; | |
3746 | unsigned long size; | |
3747 | ||
3748 | size = 0; | |
3749 | buf_size_same = 1; | |
3750 | /* check if all cpu sizes are same */ | |
3751 | for_each_tracing_cpu(cpu) { | |
3752 | /* fill in the size from first enabled cpu */ | |
3753 | if (size == 0) | |
3754 | size = tr->data[cpu]->entries; | |
3755 | if (size != tr->data[cpu]->entries) { | |
3756 | buf_size_same = 0; | |
3757 | break; | |
3758 | } | |
3759 | } | |
3760 | ||
3761 | if (buf_size_same) { | |
3762 | if (!ring_buffer_expanded) | |
3763 | r = sprintf(buf, "%lu (expanded: %lu)\n", | |
3764 | size >> 10, | |
3765 | trace_buf_size >> 10); | |
3766 | else | |
3767 | r = sprintf(buf, "%lu\n", size >> 10); | |
3768 | } else | |
3769 | r = sprintf(buf, "X\n"); | |
3770 | } else | |
3771 | r = sprintf(buf, "%lu\n", tr->data[info->cpu]->entries >> 10); | |
3772 | ||
3773 | mutex_unlock(&trace_types_lock); | |
3774 | ||
3775 | ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3776 | return ret; | |
3777 | } | |
3778 | ||
3779 | static ssize_t | |
3780 | tracing_entries_write(struct file *filp, const char __user *ubuf, | |
3781 | size_t cnt, loff_t *ppos) | |
3782 | { | |
3783 | struct ftrace_entries_info *info = filp->private_data; | |
3784 | unsigned long val; | |
3785 | int ret; | |
3786 | ||
3787 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
3788 | if (ret) | |
3789 | return ret; | |
3790 | ||
3791 | /* must have at least 1 entry */ | |
3792 | if (!val) | |
3793 | return -EINVAL; | |
3794 | ||
3795 | /* value is in KB */ | |
3796 | val <<= 10; | |
3797 | ||
3798 | ret = tracing_resize_ring_buffer(val, info->cpu); | |
3799 | if (ret < 0) | |
3800 | return ret; | |
3801 | ||
3802 | *ppos += cnt; | |
3803 | ||
3804 | return cnt; | |
3805 | } | |
3806 | ||
3807 | static int | |
3808 | tracing_entries_release(struct inode *inode, struct file *filp) | |
3809 | { | |
3810 | struct ftrace_entries_info *info = filp->private_data; | |
3811 | ||
3812 | kfree(info); | |
3813 | ||
3814 | return 0; | |
3815 | } | |
3816 | ||
3817 | static ssize_t | |
3818 | tracing_total_entries_read(struct file *filp, char __user *ubuf, | |
3819 | size_t cnt, loff_t *ppos) | |
3820 | { | |
3821 | struct trace_array *tr = filp->private_data; | |
3822 | char buf[64]; | |
3823 | int r, cpu; | |
3824 | unsigned long size = 0, expanded_size = 0; | |
3825 | ||
3826 | mutex_lock(&trace_types_lock); | |
3827 | for_each_tracing_cpu(cpu) { | |
3828 | size += tr->data[cpu]->entries >> 10; | |
3829 | if (!ring_buffer_expanded) | |
3830 | expanded_size += trace_buf_size >> 10; | |
3831 | } | |
3832 | if (ring_buffer_expanded) | |
3833 | r = sprintf(buf, "%lu\n", size); | |
3834 | else | |
3835 | r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size); | |
3836 | mutex_unlock(&trace_types_lock); | |
3837 | ||
3838 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3839 | } | |
3840 | ||
3841 | static ssize_t | |
3842 | tracing_free_buffer_write(struct file *filp, const char __user *ubuf, | |
3843 | size_t cnt, loff_t *ppos) | |
3844 | { | |
3845 | /* | |
3846 | * There is no need to read what the user has written, this function | |
3847 | * is just to make sure that there is no error when "echo" is used | |
3848 | */ | |
3849 | ||
3850 | *ppos += cnt; | |
3851 | ||
3852 | return cnt; | |
3853 | } | |
3854 | ||
3855 | static int | |
3856 | tracing_free_buffer_release(struct inode *inode, struct file *filp) | |
3857 | { | |
3858 | /* disable tracing ? */ | |
3859 | if (trace_flags & TRACE_ITER_STOP_ON_FREE) | |
3860 | tracing_off(); | |
3861 | /* resize the ring buffer to 0 */ | |
3862 | tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS); | |
3863 | ||
3864 | return 0; | |
3865 | } | |
3866 | ||
3867 | static ssize_t | |
3868 | tracing_mark_write(struct file *filp, const char __user *ubuf, | |
3869 | size_t cnt, loff_t *fpos) | |
3870 | { | |
3871 | unsigned long addr = (unsigned long)ubuf; | |
3872 | struct ring_buffer_event *event; | |
3873 | struct ring_buffer *buffer; | |
3874 | struct print_entry *entry; | |
3875 | unsigned long irq_flags; | |
3876 | struct page *pages[2]; | |
3877 | void *map_page[2]; | |
3878 | int nr_pages = 1; | |
3879 | ssize_t written; | |
3880 | int offset; | |
3881 | int size; | |
3882 | int len; | |
3883 | int ret; | |
3884 | int i; | |
3885 | ||
3886 | if (tracing_disabled) | |
3887 | return -EINVAL; | |
3888 | ||
3889 | if (cnt > TRACE_BUF_SIZE) | |
3890 | cnt = TRACE_BUF_SIZE; | |
3891 | ||
3892 | /* | |
3893 | * Userspace is injecting traces into the kernel trace buffer. | |
3894 | * We want to be as non intrusive as possible. | |
3895 | * To do so, we do not want to allocate any special buffers | |
3896 | * or take any locks, but instead write the userspace data | |
3897 | * straight into the ring buffer. | |
3898 | * | |
3899 | * First we need to pin the userspace buffer into memory, | |
3900 | * which, most likely it is, because it just referenced it. | |
3901 | * But there's no guarantee that it is. By using get_user_pages_fast() | |
3902 | * and kmap_atomic/kunmap_atomic() we can get access to the | |
3903 | * pages directly. We then write the data directly into the | |
3904 | * ring buffer. | |
3905 | */ | |
3906 | BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE); | |
3907 | ||
3908 | /* check if we cross pages */ | |
3909 | if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK)) | |
3910 | nr_pages = 2; | |
3911 | ||
3912 | offset = addr & (PAGE_SIZE - 1); | |
3913 | addr &= PAGE_MASK; | |
3914 | ||
3915 | ret = get_user_pages_fast(addr, nr_pages, 0, pages); | |
3916 | if (ret < nr_pages) { | |
3917 | while (--ret >= 0) | |
3918 | put_page(pages[ret]); | |
3919 | written = -EFAULT; | |
3920 | goto out; | |
3921 | } | |
3922 | ||
3923 | for (i = 0; i < nr_pages; i++) | |
3924 | map_page[i] = kmap_atomic(pages[i]); | |
3925 | ||
3926 | local_save_flags(irq_flags); | |
3927 | size = sizeof(*entry) + cnt + 2; /* possible \n added */ | |
3928 | buffer = global_trace.buffer; | |
3929 | event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size, | |
3930 | irq_flags, preempt_count()); | |
3931 | if (!event) { | |
3932 | /* Ring buffer disabled, return as if not open for write */ | |
3933 | written = -EBADF; | |
3934 | goto out_unlock; | |
3935 | } | |
3936 | ||
3937 | entry = ring_buffer_event_data(event); | |
3938 | entry->ip = _THIS_IP_; | |
3939 | ||
3940 | if (nr_pages == 2) { | |
3941 | len = PAGE_SIZE - offset; | |
3942 | memcpy(&entry->buf, map_page[0] + offset, len); | |
3943 | memcpy(&entry->buf[len], map_page[1], cnt - len); | |
3944 | } else | |
3945 | memcpy(&entry->buf, map_page[0] + offset, cnt); | |
3946 | ||
3947 | if (entry->buf[cnt - 1] != '\n') { | |
3948 | entry->buf[cnt] = '\n'; | |
3949 | entry->buf[cnt + 1] = '\0'; | |
3950 | } else | |
3951 | entry->buf[cnt] = '\0'; | |
3952 | ||
3953 | ring_buffer_unlock_commit(buffer, event); | |
3954 | ||
3955 | written = cnt; | |
3956 | ||
3957 | *fpos += written; | |
3958 | ||
3959 | out_unlock: | |
3960 | for (i = 0; i < nr_pages; i++){ | |
3961 | kunmap_atomic(map_page[i]); | |
3962 | put_page(pages[i]); | |
3963 | } | |
3964 | out: | |
3965 | return written; | |
3966 | } | |
3967 | ||
3968 | static int tracing_clock_show(struct seq_file *m, void *v) | |
3969 | { | |
3970 | int i; | |
3971 | ||
3972 | for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) | |
3973 | seq_printf(m, | |
3974 | "%s%s%s%s", i ? " " : "", | |
3975 | i == trace_clock_id ? "[" : "", trace_clocks[i].name, | |
3976 | i == trace_clock_id ? "]" : ""); | |
3977 | seq_putc(m, '\n'); | |
3978 | ||
3979 | return 0; | |
3980 | } | |
3981 | ||
3982 | static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf, | |
3983 | size_t cnt, loff_t *fpos) | |
3984 | { | |
3985 | char buf[64]; | |
3986 | const char *clockstr; | |
3987 | int i; | |
3988 | ||
3989 | if (cnt >= sizeof(buf)) | |
3990 | return -EINVAL; | |
3991 | ||
3992 | if (copy_from_user(&buf, ubuf, cnt)) | |
3993 | return -EFAULT; | |
3994 | ||
3995 | buf[cnt] = 0; | |
3996 | ||
3997 | clockstr = strstrip(buf); | |
3998 | ||
3999 | for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) { | |
4000 | if (strcmp(trace_clocks[i].name, clockstr) == 0) | |
4001 | break; | |
4002 | } | |
4003 | if (i == ARRAY_SIZE(trace_clocks)) | |
4004 | return -EINVAL; | |
4005 | ||
4006 | trace_clock_id = i; | |
4007 | ||
4008 | mutex_lock(&trace_types_lock); | |
4009 | ||
4010 | ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func); | |
4011 | if (max_tr.buffer) | |
4012 | ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func); | |
4013 | ||
4014 | mutex_unlock(&trace_types_lock); | |
4015 | ||
4016 | *fpos += cnt; | |
4017 | ||
4018 | return cnt; | |
4019 | } | |
4020 | ||
4021 | static int tracing_clock_open(struct inode *inode, struct file *file) | |
4022 | { | |
4023 | if (tracing_disabled) | |
4024 | return -ENODEV; | |
4025 | return single_open(file, tracing_clock_show, NULL); | |
4026 | } | |
4027 | ||
4028 | static const struct file_operations tracing_max_lat_fops = { | |
4029 | .open = tracing_open_generic, | |
4030 | .read = tracing_max_lat_read, | |
4031 | .write = tracing_max_lat_write, | |
4032 | .llseek = generic_file_llseek, | |
4033 | }; | |
4034 | ||
4035 | static const struct file_operations tracing_ctrl_fops = { | |
4036 | .open = tracing_open_generic, | |
4037 | .read = tracing_ctrl_read, | |
4038 | .write = tracing_ctrl_write, | |
4039 | .llseek = generic_file_llseek, | |
4040 | }; | |
4041 | ||
4042 | static const struct file_operations set_tracer_fops = { | |
4043 | .open = tracing_open_generic, | |
4044 | .read = tracing_set_trace_read, | |
4045 | .write = tracing_set_trace_write, | |
4046 | .llseek = generic_file_llseek, | |
4047 | }; | |
4048 | ||
4049 | static const struct file_operations tracing_pipe_fops = { | |
4050 | .open = tracing_open_pipe, | |
4051 | .poll = tracing_poll_pipe, | |
4052 | .read = tracing_read_pipe, | |
4053 | .splice_read = tracing_splice_read_pipe, | |
4054 | .release = tracing_release_pipe, | |
4055 | .llseek = no_llseek, | |
4056 | }; | |
4057 | ||
4058 | static const struct file_operations tracing_entries_fops = { | |
4059 | .open = tracing_entries_open, | |
4060 | .read = tracing_entries_read, | |
4061 | .write = tracing_entries_write, | |
4062 | .release = tracing_entries_release, | |
4063 | .llseek = generic_file_llseek, | |
4064 | }; | |
4065 | ||
4066 | static const struct file_operations tracing_total_entries_fops = { | |
4067 | .open = tracing_open_generic, | |
4068 | .read = tracing_total_entries_read, | |
4069 | .llseek = generic_file_llseek, | |
4070 | }; | |
4071 | ||
4072 | static const struct file_operations tracing_free_buffer_fops = { | |
4073 | .write = tracing_free_buffer_write, | |
4074 | .release = tracing_free_buffer_release, | |
4075 | }; | |
4076 | ||
4077 | static const struct file_operations tracing_mark_fops = { | |
4078 | .open = tracing_open_generic, | |
4079 | .write = tracing_mark_write, | |
4080 | .llseek = generic_file_llseek, | |
4081 | }; | |
4082 | ||
4083 | static const struct file_operations trace_clock_fops = { | |
4084 | .open = tracing_clock_open, | |
4085 | .read = seq_read, | |
4086 | .llseek = seq_lseek, | |
4087 | .release = single_release, | |
4088 | .write = tracing_clock_write, | |
4089 | }; | |
4090 | ||
4091 | struct ftrace_buffer_info { | |
4092 | struct trace_array *tr; | |
4093 | void *spare; | |
4094 | int cpu; | |
4095 | unsigned int read; | |
4096 | }; | |
4097 | ||
4098 | static int tracing_buffers_open(struct inode *inode, struct file *filp) | |
4099 | { | |
4100 | int cpu = (int)(long)inode->i_private; | |
4101 | struct ftrace_buffer_info *info; | |
4102 | ||
4103 | if (tracing_disabled) | |
4104 | return -ENODEV; | |
4105 | ||
4106 | info = kzalloc(sizeof(*info), GFP_KERNEL); | |
4107 | if (!info) | |
4108 | return -ENOMEM; | |
4109 | ||
4110 | info->tr = &global_trace; | |
4111 | info->cpu = cpu; | |
4112 | info->spare = NULL; | |
4113 | /* Force reading ring buffer for first read */ | |
4114 | info->read = (unsigned int)-1; | |
4115 | ||
4116 | filp->private_data = info; | |
4117 | ||
4118 | return nonseekable_open(inode, filp); | |
4119 | } | |
4120 | ||
4121 | static ssize_t | |
4122 | tracing_buffers_read(struct file *filp, char __user *ubuf, | |
4123 | size_t count, loff_t *ppos) | |
4124 | { | |
4125 | struct ftrace_buffer_info *info = filp->private_data; | |
4126 | ssize_t ret; | |
4127 | size_t size; | |
4128 | ||
4129 | if (!count) | |
4130 | return 0; | |
4131 | ||
4132 | if (!info->spare) | |
4133 | info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu); | |
4134 | if (!info->spare) | |
4135 | return -ENOMEM; | |
4136 | ||
4137 | /* Do we have previous read data to read? */ | |
4138 | if (info->read < PAGE_SIZE) | |
4139 | goto read; | |
4140 | ||
4141 | trace_access_lock(info->cpu); | |
4142 | ret = ring_buffer_read_page(info->tr->buffer, | |
4143 | &info->spare, | |
4144 | count, | |
4145 | info->cpu, 0); | |
4146 | trace_access_unlock(info->cpu); | |
4147 | if (ret < 0) | |
4148 | return 0; | |
4149 | ||
4150 | info->read = 0; | |
4151 | ||
4152 | read: | |
4153 | size = PAGE_SIZE - info->read; | |
4154 | if (size > count) | |
4155 | size = count; | |
4156 | ||
4157 | ret = copy_to_user(ubuf, info->spare + info->read, size); | |
4158 | if (ret == size) | |
4159 | return -EFAULT; | |
4160 | size -= ret; | |
4161 | ||
4162 | *ppos += size; | |
4163 | info->read += size; | |
4164 | ||
4165 | return size; | |
4166 | } | |
4167 | ||
4168 | static int tracing_buffers_release(struct inode *inode, struct file *file) | |
4169 | { | |
4170 | struct ftrace_buffer_info *info = file->private_data; | |
4171 | ||
4172 | if (info->spare) | |
4173 | ring_buffer_free_read_page(info->tr->buffer, info->spare); | |
4174 | kfree(info); | |
4175 | ||
4176 | return 0; | |
4177 | } | |
4178 | ||
4179 | struct buffer_ref { | |
4180 | struct ring_buffer *buffer; | |
4181 | void *page; | |
4182 | int ref; | |
4183 | }; | |
4184 | ||
4185 | static void buffer_pipe_buf_release(struct pipe_inode_info *pipe, | |
4186 | struct pipe_buffer *buf) | |
4187 | { | |
4188 | struct buffer_ref *ref = (struct buffer_ref *)buf->private; | |
4189 | ||
4190 | if (--ref->ref) | |
4191 | return; | |
4192 | ||
4193 | ring_buffer_free_read_page(ref->buffer, ref->page); | |
4194 | kfree(ref); | |
4195 | buf->private = 0; | |
4196 | } | |
4197 | ||
4198 | static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe, | |
4199 | struct pipe_buffer *buf) | |
4200 | { | |
4201 | return 1; | |
4202 | } | |
4203 | ||
4204 | static void buffer_pipe_buf_get(struct pipe_inode_info *pipe, | |
4205 | struct pipe_buffer *buf) | |
4206 | { | |
4207 | struct buffer_ref *ref = (struct buffer_ref *)buf->private; | |
4208 | ||
4209 | ref->ref++; | |
4210 | } | |
4211 | ||
4212 | /* Pipe buffer operations for a buffer. */ | |
4213 | static const struct pipe_buf_operations buffer_pipe_buf_ops = { | |
4214 | .can_merge = 0, | |
4215 | .map = generic_pipe_buf_map, | |
4216 | .unmap = generic_pipe_buf_unmap, | |
4217 | .confirm = generic_pipe_buf_confirm, | |
4218 | .release = buffer_pipe_buf_release, | |
4219 | .steal = buffer_pipe_buf_steal, | |
4220 | .get = buffer_pipe_buf_get, | |
4221 | }; | |
4222 | ||
4223 | /* | |
4224 | * Callback from splice_to_pipe(), if we need to release some pages | |
4225 | * at the end of the spd in case we error'ed out in filling the pipe. | |
4226 | */ | |
4227 | static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i) | |
4228 | { | |
4229 | struct buffer_ref *ref = | |
4230 | (struct buffer_ref *)spd->partial[i].private; | |
4231 | ||
4232 | if (--ref->ref) | |
4233 | return; | |
4234 | ||
4235 | ring_buffer_free_read_page(ref->buffer, ref->page); | |
4236 | kfree(ref); | |
4237 | spd->partial[i].private = 0; | |
4238 | } | |
4239 | ||
4240 | static ssize_t | |
4241 | tracing_buffers_splice_read(struct file *file, loff_t *ppos, | |
4242 | struct pipe_inode_info *pipe, size_t len, | |
4243 | unsigned int flags) | |
4244 | { | |
4245 | struct ftrace_buffer_info *info = file->private_data; | |
4246 | struct partial_page partial_def[PIPE_DEF_BUFFERS]; | |
4247 | struct page *pages_def[PIPE_DEF_BUFFERS]; | |
4248 | struct splice_pipe_desc spd = { | |
4249 | .pages = pages_def, | |
4250 | .partial = partial_def, | |
4251 | .nr_pages_max = PIPE_DEF_BUFFERS, | |
4252 | .flags = flags, | |
4253 | .ops = &buffer_pipe_buf_ops, | |
4254 | .spd_release = buffer_spd_release, | |
4255 | }; | |
4256 | struct buffer_ref *ref; | |
4257 | int entries, size, i; | |
4258 | size_t ret; | |
4259 | ||
4260 | if (splice_grow_spd(pipe, &spd)) | |
4261 | return -ENOMEM; | |
4262 | ||
4263 | if (*ppos & (PAGE_SIZE - 1)) { | |
4264 | WARN_ONCE(1, "Ftrace: previous read must page-align\n"); | |
4265 | ret = -EINVAL; | |
4266 | goto out; | |
4267 | } | |
4268 | ||
4269 | if (len & (PAGE_SIZE - 1)) { | |
4270 | WARN_ONCE(1, "Ftrace: splice_read should page-align\n"); | |
4271 | if (len < PAGE_SIZE) { | |
4272 | ret = -EINVAL; | |
4273 | goto out; | |
4274 | } | |
4275 | len &= PAGE_MASK; | |
4276 | } | |
4277 | ||
4278 | trace_access_lock(info->cpu); | |
4279 | entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu); | |
4280 | ||
4281 | for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) { | |
4282 | struct page *page; | |
4283 | int r; | |
4284 | ||
4285 | ref = kzalloc(sizeof(*ref), GFP_KERNEL); | |
4286 | if (!ref) | |
4287 | break; | |
4288 | ||
4289 | ref->ref = 1; | |
4290 | ref->buffer = info->tr->buffer; | |
4291 | ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu); | |
4292 | if (!ref->page) { | |
4293 | kfree(ref); | |
4294 | break; | |
4295 | } | |
4296 | ||
4297 | r = ring_buffer_read_page(ref->buffer, &ref->page, | |
4298 | len, info->cpu, 1); | |
4299 | if (r < 0) { | |
4300 | ring_buffer_free_read_page(ref->buffer, ref->page); | |
4301 | kfree(ref); | |
4302 | break; | |
4303 | } | |
4304 | ||
4305 | /* | |
4306 | * zero out any left over data, this is going to | |
4307 | * user land. | |
4308 | */ | |
4309 | size = ring_buffer_page_len(ref->page); | |
4310 | if (size < PAGE_SIZE) | |
4311 | memset(ref->page + size, 0, PAGE_SIZE - size); | |
4312 | ||
4313 | page = virt_to_page(ref->page); | |
4314 | ||
4315 | spd.pages[i] = page; | |
4316 | spd.partial[i].len = PAGE_SIZE; | |
4317 | spd.partial[i].offset = 0; | |
4318 | spd.partial[i].private = (unsigned long)ref; | |
4319 | spd.nr_pages++; | |
4320 | *ppos += PAGE_SIZE; | |
4321 | ||
4322 | entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu); | |
4323 | } | |
4324 | ||
4325 | trace_access_unlock(info->cpu); | |
4326 | spd.nr_pages = i; | |
4327 | ||
4328 | /* did we read anything? */ | |
4329 | if (!spd.nr_pages) { | |
4330 | if (flags & SPLICE_F_NONBLOCK) | |
4331 | ret = -EAGAIN; | |
4332 | else | |
4333 | ret = 0; | |
4334 | /* TODO: block */ | |
4335 | goto out; | |
4336 | } | |
4337 | ||
4338 | ret = splice_to_pipe(pipe, &spd); | |
4339 | splice_shrink_spd(&spd); | |
4340 | out: | |
4341 | return ret; | |
4342 | } | |
4343 | ||
4344 | static const struct file_operations tracing_buffers_fops = { | |
4345 | .open = tracing_buffers_open, | |
4346 | .read = tracing_buffers_read, | |
4347 | .release = tracing_buffers_release, | |
4348 | .splice_read = tracing_buffers_splice_read, | |
4349 | .llseek = no_llseek, | |
4350 | }; | |
4351 | ||
4352 | static ssize_t | |
4353 | tracing_stats_read(struct file *filp, char __user *ubuf, | |
4354 | size_t count, loff_t *ppos) | |
4355 | { | |
4356 | unsigned long cpu = (unsigned long)filp->private_data; | |
4357 | struct trace_array *tr = &global_trace; | |
4358 | struct trace_seq *s; | |
4359 | unsigned long cnt; | |
4360 | unsigned long long t; | |
4361 | unsigned long usec_rem; | |
4362 | ||
4363 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
4364 | if (!s) | |
4365 | return -ENOMEM; | |
4366 | ||
4367 | trace_seq_init(s); | |
4368 | ||
4369 | cnt = ring_buffer_entries_cpu(tr->buffer, cpu); | |
4370 | trace_seq_printf(s, "entries: %ld\n", cnt); | |
4371 | ||
4372 | cnt = ring_buffer_overrun_cpu(tr->buffer, cpu); | |
4373 | trace_seq_printf(s, "overrun: %ld\n", cnt); | |
4374 | ||
4375 | cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu); | |
4376 | trace_seq_printf(s, "commit overrun: %ld\n", cnt); | |
4377 | ||
4378 | cnt = ring_buffer_bytes_cpu(tr->buffer, cpu); | |
4379 | trace_seq_printf(s, "bytes: %ld\n", cnt); | |
4380 | ||
4381 | t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu)); | |
4382 | usec_rem = do_div(t, USEC_PER_SEC); | |
4383 | trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n", t, usec_rem); | |
4384 | ||
4385 | t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu)); | |
4386 | usec_rem = do_div(t, USEC_PER_SEC); | |
4387 | trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem); | |
4388 | ||
4389 | count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len); | |
4390 | ||
4391 | kfree(s); | |
4392 | ||
4393 | return count; | |
4394 | } | |
4395 | ||
4396 | static const struct file_operations tracing_stats_fops = { | |
4397 | .open = tracing_open_generic, | |
4398 | .read = tracing_stats_read, | |
4399 | .llseek = generic_file_llseek, | |
4400 | }; | |
4401 | ||
4402 | #ifdef CONFIG_DYNAMIC_FTRACE | |
4403 | ||
4404 | int __weak ftrace_arch_read_dyn_info(char *buf, int size) | |
4405 | { | |
4406 | return 0; | |
4407 | } | |
4408 | ||
4409 | static ssize_t | |
4410 | tracing_read_dyn_info(struct file *filp, char __user *ubuf, | |
4411 | size_t cnt, loff_t *ppos) | |
4412 | { | |
4413 | static char ftrace_dyn_info_buffer[1024]; | |
4414 | static DEFINE_MUTEX(dyn_info_mutex); | |
4415 | unsigned long *p = filp->private_data; | |
4416 | char *buf = ftrace_dyn_info_buffer; | |
4417 | int size = ARRAY_SIZE(ftrace_dyn_info_buffer); | |
4418 | int r; | |
4419 | ||
4420 | mutex_lock(&dyn_info_mutex); | |
4421 | r = sprintf(buf, "%ld ", *p); | |
4422 | ||
4423 | r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r); | |
4424 | buf[r++] = '\n'; | |
4425 | ||
4426 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
4427 | ||
4428 | mutex_unlock(&dyn_info_mutex); | |
4429 | ||
4430 | return r; | |
4431 | } | |
4432 | ||
4433 | static const struct file_operations tracing_dyn_info_fops = { | |
4434 | .open = tracing_open_generic, | |
4435 | .read = tracing_read_dyn_info, | |
4436 | .llseek = generic_file_llseek, | |
4437 | }; | |
4438 | #endif | |
4439 | ||
4440 | static struct dentry *d_tracer; | |
4441 | ||
4442 | struct dentry *tracing_init_dentry(void) | |
4443 | { | |
4444 | static int once; | |
4445 | ||
4446 | if (d_tracer) | |
4447 | return d_tracer; | |
4448 | ||
4449 | if (!debugfs_initialized()) | |
4450 | return NULL; | |
4451 | ||
4452 | d_tracer = debugfs_create_dir("tracing", NULL); | |
4453 | ||
4454 | if (!d_tracer && !once) { | |
4455 | once = 1; | |
4456 | pr_warning("Could not create debugfs directory 'tracing'\n"); | |
4457 | return NULL; | |
4458 | } | |
4459 | ||
4460 | return d_tracer; | |
4461 | } | |
4462 | ||
4463 | static struct dentry *d_percpu; | |
4464 | ||
4465 | struct dentry *tracing_dentry_percpu(void) | |
4466 | { | |
4467 | static int once; | |
4468 | struct dentry *d_tracer; | |
4469 | ||
4470 | if (d_percpu) | |
4471 | return d_percpu; | |
4472 | ||
4473 | d_tracer = tracing_init_dentry(); | |
4474 | ||
4475 | if (!d_tracer) | |
4476 | return NULL; | |
4477 | ||
4478 | d_percpu = debugfs_create_dir("per_cpu", d_tracer); | |
4479 | ||
4480 | if (!d_percpu && !once) { | |
4481 | once = 1; | |
4482 | pr_warning("Could not create debugfs directory 'per_cpu'\n"); | |
4483 | return NULL; | |
4484 | } | |
4485 | ||
4486 | return d_percpu; | |
4487 | } | |
4488 | ||
4489 | static void tracing_init_debugfs_percpu(long cpu) | |
4490 | { | |
4491 | struct dentry *d_percpu = tracing_dentry_percpu(); | |
4492 | struct dentry *d_cpu; | |
4493 | char cpu_dir[30]; /* 30 characters should be more than enough */ | |
4494 | ||
4495 | if (!d_percpu) | |
4496 | return; | |
4497 | ||
4498 | snprintf(cpu_dir, 30, "cpu%ld", cpu); | |
4499 | d_cpu = debugfs_create_dir(cpu_dir, d_percpu); | |
4500 | if (!d_cpu) { | |
4501 | pr_warning("Could not create debugfs '%s' entry\n", cpu_dir); | |
4502 | return; | |
4503 | } | |
4504 | ||
4505 | /* per cpu trace_pipe */ | |
4506 | trace_create_file("trace_pipe", 0444, d_cpu, | |
4507 | (void *) cpu, &tracing_pipe_fops); | |
4508 | ||
4509 | /* per cpu trace */ | |
4510 | trace_create_file("trace", 0644, d_cpu, | |
4511 | (void *) cpu, &tracing_fops); | |
4512 | ||
4513 | trace_create_file("trace_pipe_raw", 0444, d_cpu, | |
4514 | (void *) cpu, &tracing_buffers_fops); | |
4515 | ||
4516 | trace_create_file("stats", 0444, d_cpu, | |
4517 | (void *) cpu, &tracing_stats_fops); | |
4518 | ||
4519 | trace_create_file("buffer_size_kb", 0444, d_cpu, | |
4520 | (void *) cpu, &tracing_entries_fops); | |
4521 | } | |
4522 | ||
4523 | #ifdef CONFIG_FTRACE_SELFTEST | |
4524 | /* Let selftest have access to static functions in this file */ | |
4525 | #include "trace_selftest.c" | |
4526 | #endif | |
4527 | ||
4528 | struct trace_option_dentry { | |
4529 | struct tracer_opt *opt; | |
4530 | struct tracer_flags *flags; | |
4531 | struct dentry *entry; | |
4532 | }; | |
4533 | ||
4534 | static ssize_t | |
4535 | trace_options_read(struct file *filp, char __user *ubuf, size_t cnt, | |
4536 | loff_t *ppos) | |
4537 | { | |
4538 | struct trace_option_dentry *topt = filp->private_data; | |
4539 | char *buf; | |
4540 | ||
4541 | if (topt->flags->val & topt->opt->bit) | |
4542 | buf = "1\n"; | |
4543 | else | |
4544 | buf = "0\n"; | |
4545 | ||
4546 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2); | |
4547 | } | |
4548 | ||
4549 | static ssize_t | |
4550 | trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt, | |
4551 | loff_t *ppos) | |
4552 | { | |
4553 | struct trace_option_dentry *topt = filp->private_data; | |
4554 | unsigned long val; | |
4555 | int ret; | |
4556 | ||
4557 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
4558 | if (ret) | |
4559 | return ret; | |
4560 | ||
4561 | if (val != 0 && val != 1) | |
4562 | return -EINVAL; | |
4563 | ||
4564 | if (!!(topt->flags->val & topt->opt->bit) != val) { | |
4565 | mutex_lock(&trace_types_lock); | |
4566 | ret = __set_tracer_option(current_trace, topt->flags, | |
4567 | topt->opt, !val); | |
4568 | mutex_unlock(&trace_types_lock); | |
4569 | if (ret) | |
4570 | return ret; | |
4571 | } | |
4572 | ||
4573 | *ppos += cnt; | |
4574 | ||
4575 | return cnt; | |
4576 | } | |
4577 | ||
4578 | ||
4579 | static const struct file_operations trace_options_fops = { | |
4580 | .open = tracing_open_generic, | |
4581 | .read = trace_options_read, | |
4582 | .write = trace_options_write, | |
4583 | .llseek = generic_file_llseek, | |
4584 | }; | |
4585 | ||
4586 | static ssize_t | |
4587 | trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt, | |
4588 | loff_t *ppos) | |
4589 | { | |
4590 | long index = (long)filp->private_data; | |
4591 | char *buf; | |
4592 | ||
4593 | if (trace_flags & (1 << index)) | |
4594 | buf = "1\n"; | |
4595 | else | |
4596 | buf = "0\n"; | |
4597 | ||
4598 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2); | |
4599 | } | |
4600 | ||
4601 | static ssize_t | |
4602 | trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt, | |
4603 | loff_t *ppos) | |
4604 | { | |
4605 | long index = (long)filp->private_data; | |
4606 | unsigned long val; | |
4607 | int ret; | |
4608 | ||
4609 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
4610 | if (ret) | |
4611 | return ret; | |
4612 | ||
4613 | if (val != 0 && val != 1) | |
4614 | return -EINVAL; | |
4615 | set_tracer_flags(1 << index, val); | |
4616 | ||
4617 | *ppos += cnt; | |
4618 | ||
4619 | return cnt; | |
4620 | } | |
4621 | ||
4622 | static const struct file_operations trace_options_core_fops = { | |
4623 | .open = tracing_open_generic, | |
4624 | .read = trace_options_core_read, | |
4625 | .write = trace_options_core_write, | |
4626 | .llseek = generic_file_llseek, | |
4627 | }; | |
4628 | ||
4629 | struct dentry *trace_create_file(const char *name, | |
4630 | umode_t mode, | |
4631 | struct dentry *parent, | |
4632 | void *data, | |
4633 | const struct file_operations *fops) | |
4634 | { | |
4635 | struct dentry *ret; | |
4636 | ||
4637 | ret = debugfs_create_file(name, mode, parent, data, fops); | |
4638 | if (!ret) | |
4639 | pr_warning("Could not create debugfs '%s' entry\n", name); | |
4640 | ||
4641 | return ret; | |
4642 | } | |
4643 | ||
4644 | ||
4645 | static struct dentry *trace_options_init_dentry(void) | |
4646 | { | |
4647 | struct dentry *d_tracer; | |
4648 | static struct dentry *t_options; | |
4649 | ||
4650 | if (t_options) | |
4651 | return t_options; | |
4652 | ||
4653 | d_tracer = tracing_init_dentry(); | |
4654 | if (!d_tracer) | |
4655 | return NULL; | |
4656 | ||
4657 | t_options = debugfs_create_dir("options", d_tracer); | |
4658 | if (!t_options) { | |
4659 | pr_warning("Could not create debugfs directory 'options'\n"); | |
4660 | return NULL; | |
4661 | } | |
4662 | ||
4663 | return t_options; | |
4664 | } | |
4665 | ||
4666 | static void | |
4667 | create_trace_option_file(struct trace_option_dentry *topt, | |
4668 | struct tracer_flags *flags, | |
4669 | struct tracer_opt *opt) | |
4670 | { | |
4671 | struct dentry *t_options; | |
4672 | ||
4673 | t_options = trace_options_init_dentry(); | |
4674 | if (!t_options) | |
4675 | return; | |
4676 | ||
4677 | topt->flags = flags; | |
4678 | topt->opt = opt; | |
4679 | ||
4680 | topt->entry = trace_create_file(opt->name, 0644, t_options, topt, | |
4681 | &trace_options_fops); | |
4682 | ||
4683 | } | |
4684 | ||
4685 | static struct trace_option_dentry * | |
4686 | create_trace_option_files(struct tracer *tracer) | |
4687 | { | |
4688 | struct trace_option_dentry *topts; | |
4689 | struct tracer_flags *flags; | |
4690 | struct tracer_opt *opts; | |
4691 | int cnt; | |
4692 | ||
4693 | if (!tracer) | |
4694 | return NULL; | |
4695 | ||
4696 | flags = tracer->flags; | |
4697 | ||
4698 | if (!flags || !flags->opts) | |
4699 | return NULL; | |
4700 | ||
4701 | opts = flags->opts; | |
4702 | ||
4703 | for (cnt = 0; opts[cnt].name; cnt++) | |
4704 | ; | |
4705 | ||
4706 | topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL); | |
4707 | if (!topts) | |
4708 | return NULL; | |
4709 | ||
4710 | for (cnt = 0; opts[cnt].name; cnt++) | |
4711 | create_trace_option_file(&topts[cnt], flags, | |
4712 | &opts[cnt]); | |
4713 | ||
4714 | return topts; | |
4715 | } | |
4716 | ||
4717 | static void | |
4718 | destroy_trace_option_files(struct trace_option_dentry *topts) | |
4719 | { | |
4720 | int cnt; | |
4721 | ||
4722 | if (!topts) | |
4723 | return; | |
4724 | ||
4725 | for (cnt = 0; topts[cnt].opt; cnt++) { | |
4726 | if (topts[cnt].entry) | |
4727 | debugfs_remove(topts[cnt].entry); | |
4728 | } | |
4729 | ||
4730 | kfree(topts); | |
4731 | } | |
4732 | ||
4733 | static struct dentry * | |
4734 | create_trace_option_core_file(const char *option, long index) | |
4735 | { | |
4736 | struct dentry *t_options; | |
4737 | ||
4738 | t_options = trace_options_init_dentry(); | |
4739 | if (!t_options) | |
4740 | return NULL; | |
4741 | ||
4742 | return trace_create_file(option, 0644, t_options, (void *)index, | |
4743 | &trace_options_core_fops); | |
4744 | } | |
4745 | ||
4746 | static __init void create_trace_options_dir(void) | |
4747 | { | |
4748 | struct dentry *t_options; | |
4749 | int i; | |
4750 | ||
4751 | t_options = trace_options_init_dentry(); | |
4752 | if (!t_options) | |
4753 | return; | |
4754 | ||
4755 | for (i = 0; trace_options[i]; i++) | |
4756 | create_trace_option_core_file(trace_options[i], i); | |
4757 | } | |
4758 | ||
4759 | static ssize_t | |
4760 | rb_simple_read(struct file *filp, char __user *ubuf, | |
4761 | size_t cnt, loff_t *ppos) | |
4762 | { | |
4763 | struct trace_array *tr = filp->private_data; | |
4764 | struct ring_buffer *buffer = tr->buffer; | |
4765 | char buf[64]; | |
4766 | int r; | |
4767 | ||
4768 | if (buffer) | |
4769 | r = ring_buffer_record_is_on(buffer); | |
4770 | else | |
4771 | r = 0; | |
4772 | ||
4773 | r = sprintf(buf, "%d\n", r); | |
4774 | ||
4775 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
4776 | } | |
4777 | ||
4778 | static ssize_t | |
4779 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
4780 | size_t cnt, loff_t *ppos) | |
4781 | { | |
4782 | struct trace_array *tr = filp->private_data; | |
4783 | struct ring_buffer *buffer = tr->buffer; | |
4784 | unsigned long val; | |
4785 | int ret; | |
4786 | ||
4787 | ret = kstrtoul_from_user(ubuf, cnt, 10, &val); | |
4788 | if (ret) | |
4789 | return ret; | |
4790 | ||
4791 | if (buffer) { | |
4792 | if (val) | |
4793 | ring_buffer_record_on(buffer); | |
4794 | else | |
4795 | ring_buffer_record_off(buffer); | |
4796 | } | |
4797 | ||
4798 | (*ppos)++; | |
4799 | ||
4800 | return cnt; | |
4801 | } | |
4802 | ||
4803 | static const struct file_operations rb_simple_fops = { | |
4804 | .open = tracing_open_generic, | |
4805 | .read = rb_simple_read, | |
4806 | .write = rb_simple_write, | |
4807 | .llseek = default_llseek, | |
4808 | }; | |
4809 | ||
4810 | static __init int tracer_init_debugfs(void) | |
4811 | { | |
4812 | struct dentry *d_tracer; | |
4813 | int cpu; | |
4814 | ||
4815 | trace_access_lock_init(); | |
4816 | ||
4817 | d_tracer = tracing_init_dentry(); | |
4818 | ||
4819 | trace_create_file("tracing_enabled", 0644, d_tracer, | |
4820 | &global_trace, &tracing_ctrl_fops); | |
4821 | ||
4822 | trace_create_file("trace_options", 0644, d_tracer, | |
4823 | NULL, &tracing_iter_fops); | |
4824 | ||
4825 | trace_create_file("tracing_cpumask", 0644, d_tracer, | |
4826 | NULL, &tracing_cpumask_fops); | |
4827 | ||
4828 | trace_create_file("trace", 0644, d_tracer, | |
4829 | (void *) TRACE_PIPE_ALL_CPU, &tracing_fops); | |
4830 | ||
4831 | trace_create_file("available_tracers", 0444, d_tracer, | |
4832 | &global_trace, &show_traces_fops); | |
4833 | ||
4834 | trace_create_file("current_tracer", 0644, d_tracer, | |
4835 | &global_trace, &set_tracer_fops); | |
4836 | ||
4837 | #ifdef CONFIG_TRACER_MAX_TRACE | |
4838 | trace_create_file("tracing_max_latency", 0644, d_tracer, | |
4839 | &tracing_max_latency, &tracing_max_lat_fops); | |
4840 | #endif | |
4841 | ||
4842 | trace_create_file("tracing_thresh", 0644, d_tracer, | |
4843 | &tracing_thresh, &tracing_max_lat_fops); | |
4844 | ||
4845 | trace_create_file("README", 0444, d_tracer, | |
4846 | NULL, &tracing_readme_fops); | |
4847 | ||
4848 | trace_create_file("trace_pipe", 0444, d_tracer, | |
4849 | (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops); | |
4850 | ||
4851 | trace_create_file("buffer_size_kb", 0644, d_tracer, | |
4852 | (void *) RING_BUFFER_ALL_CPUS, &tracing_entries_fops); | |
4853 | ||
4854 | trace_create_file("buffer_total_size_kb", 0444, d_tracer, | |
4855 | &global_trace, &tracing_total_entries_fops); | |
4856 | ||
4857 | trace_create_file("free_buffer", 0644, d_tracer, | |
4858 | &global_trace, &tracing_free_buffer_fops); | |
4859 | ||
4860 | trace_create_file("trace_marker", 0220, d_tracer, | |
4861 | NULL, &tracing_mark_fops); | |
4862 | ||
4863 | trace_create_file("saved_cmdlines", 0444, d_tracer, | |
4864 | NULL, &tracing_saved_cmdlines_fops); | |
4865 | ||
4866 | trace_create_file("trace_clock", 0644, d_tracer, NULL, | |
4867 | &trace_clock_fops); | |
4868 | ||
4869 | trace_create_file("tracing_on", 0644, d_tracer, | |
4870 | &global_trace, &rb_simple_fops); | |
4871 | ||
4872 | #ifdef CONFIG_DYNAMIC_FTRACE | |
4873 | trace_create_file("dyn_ftrace_total_info", 0444, d_tracer, | |
4874 | &ftrace_update_tot_cnt, &tracing_dyn_info_fops); | |
4875 | #endif | |
4876 | ||
4877 | create_trace_options_dir(); | |
4878 | ||
4879 | for_each_tracing_cpu(cpu) | |
4880 | tracing_init_debugfs_percpu(cpu); | |
4881 | ||
4882 | return 0; | |
4883 | } | |
4884 | ||
4885 | static int trace_panic_handler(struct notifier_block *this, | |
4886 | unsigned long event, void *unused) | |
4887 | { | |
4888 | if (ftrace_dump_on_oops) | |
4889 | ftrace_dump(ftrace_dump_on_oops); | |
4890 | return NOTIFY_OK; | |
4891 | } | |
4892 | ||
4893 | static struct notifier_block trace_panic_notifier = { | |
4894 | .notifier_call = trace_panic_handler, | |
4895 | .next = NULL, | |
4896 | .priority = 150 /* priority: INT_MAX >= x >= 0 */ | |
4897 | }; | |
4898 | ||
4899 | static int trace_die_handler(struct notifier_block *self, | |
4900 | unsigned long val, | |
4901 | void *data) | |
4902 | { | |
4903 | switch (val) { | |
4904 | case DIE_OOPS: | |
4905 | if (ftrace_dump_on_oops) | |
4906 | ftrace_dump(ftrace_dump_on_oops); | |
4907 | break; | |
4908 | default: | |
4909 | break; | |
4910 | } | |
4911 | return NOTIFY_OK; | |
4912 | } | |
4913 | ||
4914 | static struct notifier_block trace_die_notifier = { | |
4915 | .notifier_call = trace_die_handler, | |
4916 | .priority = 200 | |
4917 | }; | |
4918 | ||
4919 | /* | |
4920 | * printk is set to max of 1024, we really don't need it that big. | |
4921 | * Nothing should be printing 1000 characters anyway. | |
4922 | */ | |
4923 | #define TRACE_MAX_PRINT 1000 | |
4924 | ||
4925 | /* | |
4926 | * Define here KERN_TRACE so that we have one place to modify | |
4927 | * it if we decide to change what log level the ftrace dump | |
4928 | * should be at. | |
4929 | */ | |
4930 | #define KERN_TRACE KERN_EMERG | |
4931 | ||
4932 | void | |
4933 | trace_printk_seq(struct trace_seq *s) | |
4934 | { | |
4935 | /* Probably should print a warning here. */ | |
4936 | if (s->len >= 1000) | |
4937 | s->len = 1000; | |
4938 | ||
4939 | /* should be zero ended, but we are paranoid. */ | |
4940 | s->buffer[s->len] = 0; | |
4941 | ||
4942 | printk(KERN_TRACE "%s", s->buffer); | |
4943 | ||
4944 | trace_seq_init(s); | |
4945 | } | |
4946 | ||
4947 | void trace_init_global_iter(struct trace_iterator *iter) | |
4948 | { | |
4949 | iter->tr = &global_trace; | |
4950 | iter->trace = current_trace; | |
4951 | iter->cpu_file = TRACE_PIPE_ALL_CPU; | |
4952 | } | |
4953 | ||
4954 | static void | |
4955 | __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode) | |
4956 | { | |
4957 | static arch_spinlock_t ftrace_dump_lock = | |
4958 | (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; | |
4959 | /* use static because iter can be a bit big for the stack */ | |
4960 | static struct trace_iterator iter; | |
4961 | unsigned int old_userobj; | |
4962 | static int dump_ran; | |
4963 | unsigned long flags; | |
4964 | int cnt = 0, cpu; | |
4965 | ||
4966 | /* only one dump */ | |
4967 | local_irq_save(flags); | |
4968 | arch_spin_lock(&ftrace_dump_lock); | |
4969 | if (dump_ran) | |
4970 | goto out; | |
4971 | ||
4972 | dump_ran = 1; | |
4973 | ||
4974 | tracing_off(); | |
4975 | ||
4976 | /* Did function tracer already get disabled? */ | |
4977 | if (ftrace_is_dead()) { | |
4978 | printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n"); | |
4979 | printk("# MAY BE MISSING FUNCTION EVENTS\n"); | |
4980 | } | |
4981 | ||
4982 | if (disable_tracing) | |
4983 | ftrace_kill(); | |
4984 | ||
4985 | trace_init_global_iter(&iter); | |
4986 | ||
4987 | for_each_tracing_cpu(cpu) { | |
4988 | atomic_inc(&iter.tr->data[cpu]->disabled); | |
4989 | } | |
4990 | ||
4991 | old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ; | |
4992 | ||
4993 | /* don't look at user memory in panic mode */ | |
4994 | trace_flags &= ~TRACE_ITER_SYM_USEROBJ; | |
4995 | ||
4996 | /* Simulate the iterator */ | |
4997 | iter.tr = &global_trace; | |
4998 | iter.trace = current_trace; | |
4999 | ||
5000 | switch (oops_dump_mode) { | |
5001 | case DUMP_ALL: | |
5002 | iter.cpu_file = TRACE_PIPE_ALL_CPU; | |
5003 | break; | |
5004 | case DUMP_ORIG: | |
5005 | iter.cpu_file = raw_smp_processor_id(); | |
5006 | break; | |
5007 | case DUMP_NONE: | |
5008 | goto out_enable; | |
5009 | default: | |
5010 | printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n"); | |
5011 | iter.cpu_file = TRACE_PIPE_ALL_CPU; | |
5012 | } | |
5013 | ||
5014 | printk(KERN_TRACE "Dumping ftrace buffer:\n"); | |
5015 | ||
5016 | /* | |
5017 | * We need to stop all tracing on all CPUS to read the | |
5018 | * the next buffer. This is a bit expensive, but is | |
5019 | * not done often. We fill all what we can read, | |
5020 | * and then release the locks again. | |
5021 | */ | |
5022 | ||
5023 | while (!trace_empty(&iter)) { | |
5024 | ||
5025 | if (!cnt) | |
5026 | printk(KERN_TRACE "---------------------------------\n"); | |
5027 | ||
5028 | cnt++; | |
5029 | ||
5030 | /* reset all but tr, trace, and overruns */ | |
5031 | memset(&iter.seq, 0, | |
5032 | sizeof(struct trace_iterator) - | |
5033 | offsetof(struct trace_iterator, seq)); | |
5034 | iter.iter_flags |= TRACE_FILE_LAT_FMT; | |
5035 | iter.pos = -1; | |
5036 | ||
5037 | if (trace_find_next_entry_inc(&iter) != NULL) { | |
5038 | int ret; | |
5039 | ||
5040 | ret = print_trace_line(&iter); | |
5041 | if (ret != TRACE_TYPE_NO_CONSUME) | |
5042 | trace_consume(&iter); | |
5043 | } | |
5044 | touch_nmi_watchdog(); | |
5045 | ||
5046 | trace_printk_seq(&iter.seq); | |
5047 | } | |
5048 | ||
5049 | if (!cnt) | |
5050 | printk(KERN_TRACE " (ftrace buffer empty)\n"); | |
5051 | else | |
5052 | printk(KERN_TRACE "---------------------------------\n"); | |
5053 | ||
5054 | out_enable: | |
5055 | /* Re-enable tracing if requested */ | |
5056 | if (!disable_tracing) { | |
5057 | trace_flags |= old_userobj; | |
5058 | ||
5059 | for_each_tracing_cpu(cpu) { | |
5060 | atomic_dec(&iter.tr->data[cpu]->disabled); | |
5061 | } | |
5062 | tracing_on(); | |
5063 | } | |
5064 | ||
5065 | out: | |
5066 | arch_spin_unlock(&ftrace_dump_lock); | |
5067 | local_irq_restore(flags); | |
5068 | } | |
5069 | ||
5070 | /* By default: disable tracing after the dump */ | |
5071 | void ftrace_dump(enum ftrace_dump_mode oops_dump_mode) | |
5072 | { | |
5073 | __ftrace_dump(true, oops_dump_mode); | |
5074 | } | |
5075 | EXPORT_SYMBOL_GPL(ftrace_dump); | |
5076 | ||
5077 | __init static int tracer_alloc_buffers(void) | |
5078 | { | |
5079 | int ring_buf_size; | |
5080 | enum ring_buffer_flags rb_flags; | |
5081 | int i; | |
5082 | int ret = -ENOMEM; | |
5083 | ||
5084 | ||
5085 | if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL)) | |
5086 | goto out; | |
5087 | ||
5088 | if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL)) | |
5089 | goto out_free_buffer_mask; | |
5090 | ||
5091 | /* Only allocate trace_printk buffers if a trace_printk exists */ | |
5092 | if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt) | |
5093 | trace_printk_init_buffers(); | |
5094 | ||
5095 | /* To save memory, keep the ring buffer size to its minimum */ | |
5096 | if (ring_buffer_expanded) | |
5097 | ring_buf_size = trace_buf_size; | |
5098 | else | |
5099 | ring_buf_size = 1; | |
5100 | ||
5101 | rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0; | |
5102 | ||
5103 | cpumask_copy(tracing_buffer_mask, cpu_possible_mask); | |
5104 | cpumask_copy(tracing_cpumask, cpu_all_mask); | |
5105 | ||
5106 | /* TODO: make the number of buffers hot pluggable with CPUS */ | |
5107 | global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags); | |
5108 | if (!global_trace.buffer) { | |
5109 | printk(KERN_ERR "tracer: failed to allocate ring buffer!\n"); | |
5110 | WARN_ON(1); | |
5111 | goto out_free_cpumask; | |
5112 | } | |
5113 | if (global_trace.buffer_disabled) | |
5114 | tracing_off(); | |
5115 | ||
5116 | ||
5117 | #ifdef CONFIG_TRACER_MAX_TRACE | |
5118 | max_tr.buffer = ring_buffer_alloc(1, rb_flags); | |
5119 | if (!max_tr.buffer) { | |
5120 | printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n"); | |
5121 | WARN_ON(1); | |
5122 | ring_buffer_free(global_trace.buffer); | |
5123 | goto out_free_cpumask; | |
5124 | } | |
5125 | #endif | |
5126 | ||
5127 | /* Allocate the first page for all buffers */ | |
5128 | for_each_tracing_cpu(i) { | |
5129 | global_trace.data[i] = &per_cpu(global_trace_cpu, i); | |
5130 | max_tr.data[i] = &per_cpu(max_tr_data, i); | |
5131 | } | |
5132 | ||
5133 | set_buffer_entries(&global_trace, | |
5134 | ring_buffer_size(global_trace.buffer, 0)); | |
5135 | #ifdef CONFIG_TRACER_MAX_TRACE | |
5136 | set_buffer_entries(&max_tr, 1); | |
5137 | #endif | |
5138 | ||
5139 | trace_init_cmdlines(); | |
5140 | ||
5141 | register_tracer(&nop_trace); | |
5142 | current_trace = &nop_trace; | |
5143 | /* All seems OK, enable tracing */ | |
5144 | tracing_disabled = 0; | |
5145 | ||
5146 | atomic_notifier_chain_register(&panic_notifier_list, | |
5147 | &trace_panic_notifier); | |
5148 | ||
5149 | register_die_notifier(&trace_die_notifier); | |
5150 | ||
5151 | return 0; | |
5152 | ||
5153 | out_free_cpumask: | |
5154 | free_cpumask_var(tracing_cpumask); | |
5155 | out_free_buffer_mask: | |
5156 | free_cpumask_var(tracing_buffer_mask); | |
5157 | out: | |
5158 | return ret; | |
5159 | } | |
5160 | ||
5161 | __init static int clear_boot_tracer(void) | |
5162 | { | |
5163 | /* | |
5164 | * The default tracer at boot buffer is an init section. | |
5165 | * This function is called in lateinit. If we did not | |
5166 | * find the boot tracer, then clear it out, to prevent | |
5167 | * later registration from accessing the buffer that is | |
5168 | * about to be freed. | |
5169 | */ | |
5170 | if (!default_bootup_tracer) | |
5171 | return 0; | |
5172 | ||
5173 | printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n", | |
5174 | default_bootup_tracer); | |
5175 | default_bootup_tracer = NULL; | |
5176 | ||
5177 | return 0; | |
5178 | } | |
5179 | ||
5180 | early_initcall(tracer_alloc_buffers); | |
5181 | fs_initcall(tracer_init_debugfs); | |
5182 | late_initcall(clear_boot_tracer); |