]>
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 <linux/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/ctype.h> | |
34 | #include <linux/init.h> | |
35 | #include <linux/poll.h> | |
36 | #include <linux/gfp.h> | |
37 | #include <linux/fs.h> | |
38 | ||
39 | #include "trace.h" | |
40 | #include "trace_output.h" | |
41 | ||
42 | #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE) | |
43 | ||
44 | unsigned long __read_mostly tracing_max_latency; | |
45 | unsigned long __read_mostly tracing_thresh; | |
46 | ||
47 | /* | |
48 | * On boot up, the ring buffer is set to the minimum size, so that | |
49 | * we do not waste memory on systems that are not using tracing. | |
50 | */ | |
51 | static int ring_buffer_expanded; | |
52 | ||
53 | /* | |
54 | * We need to change this state when a selftest is running. | |
55 | * A selftest will lurk into the ring-buffer to count the | |
56 | * entries inserted during the selftest although some concurrent | |
57 | * insertions into the ring-buffer such as trace_printk could occurred | |
58 | * at the same time, giving false positive or negative results. | |
59 | */ | |
60 | static bool __read_mostly tracing_selftest_running; | |
61 | ||
62 | /* | |
63 | * If a tracer is running, we do not want to run SELFTEST. | |
64 | */ | |
65 | static bool __read_mostly tracing_selftest_disabled; | |
66 | ||
67 | /* For tracers that don't implement custom flags */ | |
68 | static struct tracer_opt dummy_tracer_opt[] = { | |
69 | { } | |
70 | }; | |
71 | ||
72 | static struct tracer_flags dummy_tracer_flags = { | |
73 | .val = 0, | |
74 | .opts = dummy_tracer_opt | |
75 | }; | |
76 | ||
77 | static int dummy_set_flag(u32 old_flags, u32 bit, int set) | |
78 | { | |
79 | return 0; | |
80 | } | |
81 | ||
82 | /* | |
83 | * Kill all tracing for good (never come back). | |
84 | * It is initialized to 1 but will turn to zero if the initialization | |
85 | * of the tracer is successful. But that is the only place that sets | |
86 | * this back to zero. | |
87 | */ | |
88 | static int tracing_disabled = 1; | |
89 | ||
90 | static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled); | |
91 | ||
92 | static inline void ftrace_disable_cpu(void) | |
93 | { | |
94 | preempt_disable(); | |
95 | local_inc(&__get_cpu_var(ftrace_cpu_disabled)); | |
96 | } | |
97 | ||
98 | static inline void ftrace_enable_cpu(void) | |
99 | { | |
100 | local_dec(&__get_cpu_var(ftrace_cpu_disabled)); | |
101 | preempt_enable(); | |
102 | } | |
103 | ||
104 | static cpumask_var_t __read_mostly tracing_buffer_mask; | |
105 | ||
106 | /* Define which cpu buffers are currently read in trace_pipe */ | |
107 | static cpumask_var_t tracing_reader_cpumask; | |
108 | ||
109 | #define for_each_tracing_cpu(cpu) \ | |
110 | for_each_cpu(cpu, tracing_buffer_mask) | |
111 | ||
112 | /* | |
113 | * ftrace_dump_on_oops - variable to dump ftrace buffer on oops | |
114 | * | |
115 | * If there is an oops (or kernel panic) and the ftrace_dump_on_oops | |
116 | * is set, then ftrace_dump is called. This will output the contents | |
117 | * of the ftrace buffers to the console. This is very useful for | |
118 | * capturing traces that lead to crashes and outputing it to a | |
119 | * serial console. | |
120 | * | |
121 | * It is default off, but you can enable it with either specifying | |
122 | * "ftrace_dump_on_oops" in the kernel command line, or setting | |
123 | * /proc/sys/kernel/ftrace_dump_on_oops to true. | |
124 | */ | |
125 | int ftrace_dump_on_oops; | |
126 | ||
127 | static int tracing_set_tracer(const char *buf); | |
128 | ||
129 | #define BOOTUP_TRACER_SIZE 100 | |
130 | static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata; | |
131 | static char *default_bootup_tracer; | |
132 | ||
133 | static int __init set_ftrace(char *str) | |
134 | { | |
135 | strncpy(bootup_tracer_buf, str, BOOTUP_TRACER_SIZE); | |
136 | default_bootup_tracer = bootup_tracer_buf; | |
137 | /* We are using ftrace early, expand it */ | |
138 | ring_buffer_expanded = 1; | |
139 | return 1; | |
140 | } | |
141 | __setup("ftrace=", set_ftrace); | |
142 | ||
143 | static int __init set_ftrace_dump_on_oops(char *str) | |
144 | { | |
145 | ftrace_dump_on_oops = 1; | |
146 | return 1; | |
147 | } | |
148 | __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops); | |
149 | ||
150 | long | |
151 | ns2usecs(cycle_t nsec) | |
152 | { | |
153 | nsec += 500; | |
154 | do_div(nsec, 1000); | |
155 | return nsec; | |
156 | } | |
157 | ||
158 | /* | |
159 | * The global_trace is the descriptor that holds the tracing | |
160 | * buffers for the live tracing. For each CPU, it contains | |
161 | * a link list of pages that will store trace entries. The | |
162 | * page descriptor of the pages in the memory is used to hold | |
163 | * the link list by linking the lru item in the page descriptor | |
164 | * to each of the pages in the buffer per CPU. | |
165 | * | |
166 | * For each active CPU there is a data field that holds the | |
167 | * pages for the buffer for that CPU. Each CPU has the same number | |
168 | * of pages allocated for its buffer. | |
169 | */ | |
170 | static struct trace_array global_trace; | |
171 | ||
172 | static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu); | |
173 | ||
174 | cycle_t ftrace_now(int cpu) | |
175 | { | |
176 | u64 ts; | |
177 | ||
178 | /* Early boot up does not have a buffer yet */ | |
179 | if (!global_trace.buffer) | |
180 | return trace_clock_local(); | |
181 | ||
182 | ts = ring_buffer_time_stamp(global_trace.buffer, cpu); | |
183 | ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts); | |
184 | ||
185 | return ts; | |
186 | } | |
187 | ||
188 | /* | |
189 | * The max_tr is used to snapshot the global_trace when a maximum | |
190 | * latency is reached. Some tracers will use this to store a maximum | |
191 | * trace while it continues examining live traces. | |
192 | * | |
193 | * The buffers for the max_tr are set up the same as the global_trace. | |
194 | * When a snapshot is taken, the link list of the max_tr is swapped | |
195 | * with the link list of the global_trace and the buffers are reset for | |
196 | * the global_trace so the tracing can continue. | |
197 | */ | |
198 | static struct trace_array max_tr; | |
199 | ||
200 | static DEFINE_PER_CPU(struct trace_array_cpu, max_data); | |
201 | ||
202 | /* tracer_enabled is used to toggle activation of a tracer */ | |
203 | static int tracer_enabled = 1; | |
204 | ||
205 | /** | |
206 | * tracing_is_enabled - return tracer_enabled status | |
207 | * | |
208 | * This function is used by other tracers to know the status | |
209 | * of the tracer_enabled flag. Tracers may use this function | |
210 | * to know if it should enable their features when starting | |
211 | * up. See irqsoff tracer for an example (start_irqsoff_tracer). | |
212 | */ | |
213 | int tracing_is_enabled(void) | |
214 | { | |
215 | return tracer_enabled; | |
216 | } | |
217 | ||
218 | /* | |
219 | * trace_buf_size is the size in bytes that is allocated | |
220 | * for a buffer. Note, the number of bytes is always rounded | |
221 | * to page size. | |
222 | * | |
223 | * This number is purposely set to a low number of 16384. | |
224 | * If the dump on oops happens, it will be much appreciated | |
225 | * to not have to wait for all that output. Anyway this can be | |
226 | * boot time and run time configurable. | |
227 | */ | |
228 | #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */ | |
229 | ||
230 | static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT; | |
231 | ||
232 | /* trace_types holds a link list of available tracers. */ | |
233 | static struct tracer *trace_types __read_mostly; | |
234 | ||
235 | /* current_trace points to the tracer that is currently active */ | |
236 | static struct tracer *current_trace __read_mostly; | |
237 | ||
238 | /* | |
239 | * max_tracer_type_len is used to simplify the allocating of | |
240 | * buffers to read userspace tracer names. We keep track of | |
241 | * the longest tracer name registered. | |
242 | */ | |
243 | static int max_tracer_type_len; | |
244 | ||
245 | /* | |
246 | * trace_types_lock is used to protect the trace_types list. | |
247 | * This lock is also used to keep user access serialized. | |
248 | * Accesses from userspace will grab this lock while userspace | |
249 | * activities happen inside the kernel. | |
250 | */ | |
251 | static DEFINE_MUTEX(trace_types_lock); | |
252 | ||
253 | /* trace_wait is a waitqueue for tasks blocked on trace_poll */ | |
254 | static DECLARE_WAIT_QUEUE_HEAD(trace_wait); | |
255 | ||
256 | /* trace_flags holds trace_options default values */ | |
257 | unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | | |
258 | TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO; | |
259 | ||
260 | /** | |
261 | * trace_wake_up - wake up tasks waiting for trace input | |
262 | * | |
263 | * Simply wakes up any task that is blocked on the trace_wait | |
264 | * queue. These is used with trace_poll for tasks polling the trace. | |
265 | */ | |
266 | void trace_wake_up(void) | |
267 | { | |
268 | /* | |
269 | * The runqueue_is_locked() can fail, but this is the best we | |
270 | * have for now: | |
271 | */ | |
272 | if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked()) | |
273 | wake_up(&trace_wait); | |
274 | } | |
275 | ||
276 | static int __init set_buf_size(char *str) | |
277 | { | |
278 | unsigned long buf_size; | |
279 | int ret; | |
280 | ||
281 | if (!str) | |
282 | return 0; | |
283 | ret = strict_strtoul(str, 0, &buf_size); | |
284 | /* nr_entries can not be zero */ | |
285 | if (ret < 0 || buf_size == 0) | |
286 | return 0; | |
287 | trace_buf_size = buf_size; | |
288 | return 1; | |
289 | } | |
290 | __setup("trace_buf_size=", set_buf_size); | |
291 | ||
292 | unsigned long nsecs_to_usecs(unsigned long nsecs) | |
293 | { | |
294 | return nsecs / 1000; | |
295 | } | |
296 | ||
297 | /* These must match the bit postions in trace_iterator_flags */ | |
298 | static const char *trace_options[] = { | |
299 | "print-parent", | |
300 | "sym-offset", | |
301 | "sym-addr", | |
302 | "verbose", | |
303 | "raw", | |
304 | "hex", | |
305 | "bin", | |
306 | "block", | |
307 | "stacktrace", | |
308 | "sched-tree", | |
309 | "trace_printk", | |
310 | "ftrace_preempt", | |
311 | "branch", | |
312 | "annotate", | |
313 | "userstacktrace", | |
314 | "sym-userobj", | |
315 | "printk-msg-only", | |
316 | "context-info", | |
317 | "latency-format", | |
318 | "global-clock", | |
319 | NULL | |
320 | }; | |
321 | ||
322 | /* | |
323 | * ftrace_max_lock is used to protect the swapping of buffers | |
324 | * when taking a max snapshot. The buffers themselves are | |
325 | * protected by per_cpu spinlocks. But the action of the swap | |
326 | * needs its own lock. | |
327 | * | |
328 | * This is defined as a raw_spinlock_t in order to help | |
329 | * with performance when lockdep debugging is enabled. | |
330 | */ | |
331 | static raw_spinlock_t ftrace_max_lock = | |
332 | (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; | |
333 | ||
334 | /* | |
335 | * Copy the new maximum trace into the separate maximum-trace | |
336 | * structure. (this way the maximum trace is permanently saved, | |
337 | * for later retrieval via /debugfs/tracing/latency_trace) | |
338 | */ | |
339 | static void | |
340 | __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) | |
341 | { | |
342 | struct trace_array_cpu *data = tr->data[cpu]; | |
343 | ||
344 | max_tr.cpu = cpu; | |
345 | max_tr.time_start = data->preempt_timestamp; | |
346 | ||
347 | data = max_tr.data[cpu]; | |
348 | data->saved_latency = tracing_max_latency; | |
349 | ||
350 | memcpy(data->comm, tsk->comm, TASK_COMM_LEN); | |
351 | data->pid = tsk->pid; | |
352 | data->uid = task_uid(tsk); | |
353 | data->nice = tsk->static_prio - 20 - MAX_RT_PRIO; | |
354 | data->policy = tsk->policy; | |
355 | data->rt_priority = tsk->rt_priority; | |
356 | ||
357 | /* record this tasks comm */ | |
358 | tracing_record_cmdline(tsk); | |
359 | } | |
360 | ||
361 | ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt) | |
362 | { | |
363 | int len; | |
364 | int ret; | |
365 | ||
366 | if (!cnt) | |
367 | return 0; | |
368 | ||
369 | if (s->len <= s->readpos) | |
370 | return -EBUSY; | |
371 | ||
372 | len = s->len - s->readpos; | |
373 | if (cnt > len) | |
374 | cnt = len; | |
375 | ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt); | |
376 | if (ret == cnt) | |
377 | return -EFAULT; | |
378 | ||
379 | cnt -= ret; | |
380 | ||
381 | s->readpos += cnt; | |
382 | return cnt; | |
383 | } | |
384 | ||
385 | ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt) | |
386 | { | |
387 | int len; | |
388 | void *ret; | |
389 | ||
390 | if (s->len <= s->readpos) | |
391 | return -EBUSY; | |
392 | ||
393 | len = s->len - s->readpos; | |
394 | if (cnt > len) | |
395 | cnt = len; | |
396 | ret = memcpy(buf, s->buffer + s->readpos, cnt); | |
397 | if (!ret) | |
398 | return -EFAULT; | |
399 | ||
400 | s->readpos += cnt; | |
401 | return cnt; | |
402 | } | |
403 | ||
404 | static void | |
405 | trace_print_seq(struct seq_file *m, struct trace_seq *s) | |
406 | { | |
407 | int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len; | |
408 | ||
409 | s->buffer[len] = 0; | |
410 | seq_puts(m, s->buffer); | |
411 | ||
412 | trace_seq_init(s); | |
413 | } | |
414 | ||
415 | /** | |
416 | * update_max_tr - snapshot all trace buffers from global_trace to max_tr | |
417 | * @tr: tracer | |
418 | * @tsk: the task with the latency | |
419 | * @cpu: The cpu that initiated the trace. | |
420 | * | |
421 | * Flip the buffers between the @tr and the max_tr and record information | |
422 | * about which task was the cause of this latency. | |
423 | */ | |
424 | void | |
425 | update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu) | |
426 | { | |
427 | struct ring_buffer *buf = tr->buffer; | |
428 | ||
429 | WARN_ON_ONCE(!irqs_disabled()); | |
430 | __raw_spin_lock(&ftrace_max_lock); | |
431 | ||
432 | tr->buffer = max_tr.buffer; | |
433 | max_tr.buffer = buf; | |
434 | ||
435 | ftrace_disable_cpu(); | |
436 | ring_buffer_reset(tr->buffer); | |
437 | ftrace_enable_cpu(); | |
438 | ||
439 | __update_max_tr(tr, tsk, cpu); | |
440 | __raw_spin_unlock(&ftrace_max_lock); | |
441 | } | |
442 | ||
443 | /** | |
444 | * update_max_tr_single - only copy one trace over, and reset the rest | |
445 | * @tr - tracer | |
446 | * @tsk - task with the latency | |
447 | * @cpu - the cpu of the buffer to copy. | |
448 | * | |
449 | * Flip the trace of a single CPU buffer between the @tr and the max_tr. | |
450 | */ | |
451 | void | |
452 | update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu) | |
453 | { | |
454 | int ret; | |
455 | ||
456 | WARN_ON_ONCE(!irqs_disabled()); | |
457 | __raw_spin_lock(&ftrace_max_lock); | |
458 | ||
459 | ftrace_disable_cpu(); | |
460 | ||
461 | ring_buffer_reset(max_tr.buffer); | |
462 | ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu); | |
463 | ||
464 | ftrace_enable_cpu(); | |
465 | ||
466 | WARN_ON_ONCE(ret && ret != -EAGAIN); | |
467 | ||
468 | __update_max_tr(tr, tsk, cpu); | |
469 | __raw_spin_unlock(&ftrace_max_lock); | |
470 | } | |
471 | ||
472 | /** | |
473 | * register_tracer - register a tracer with the ftrace system. | |
474 | * @type - the plugin for the tracer | |
475 | * | |
476 | * Register a new plugin tracer. | |
477 | */ | |
478 | int register_tracer(struct tracer *type) | |
479 | __releases(kernel_lock) | |
480 | __acquires(kernel_lock) | |
481 | { | |
482 | struct tracer *t; | |
483 | int len; | |
484 | int ret = 0; | |
485 | ||
486 | if (!type->name) { | |
487 | pr_info("Tracer must have a name\n"); | |
488 | return -1; | |
489 | } | |
490 | ||
491 | /* | |
492 | * When this gets called we hold the BKL which means that | |
493 | * preemption is disabled. Various trace selftests however | |
494 | * need to disable and enable preemption for successful tests. | |
495 | * So we drop the BKL here and grab it after the tests again. | |
496 | */ | |
497 | unlock_kernel(); | |
498 | mutex_lock(&trace_types_lock); | |
499 | ||
500 | tracing_selftest_running = true; | |
501 | ||
502 | for (t = trace_types; t; t = t->next) { | |
503 | if (strcmp(type->name, t->name) == 0) { | |
504 | /* already found */ | |
505 | pr_info("Trace %s already registered\n", | |
506 | type->name); | |
507 | ret = -1; | |
508 | goto out; | |
509 | } | |
510 | } | |
511 | ||
512 | if (!type->set_flag) | |
513 | type->set_flag = &dummy_set_flag; | |
514 | if (!type->flags) | |
515 | type->flags = &dummy_tracer_flags; | |
516 | else | |
517 | if (!type->flags->opts) | |
518 | type->flags->opts = dummy_tracer_opt; | |
519 | if (!type->wait_pipe) | |
520 | type->wait_pipe = default_wait_pipe; | |
521 | ||
522 | ||
523 | #ifdef CONFIG_FTRACE_STARTUP_TEST | |
524 | if (type->selftest && !tracing_selftest_disabled) { | |
525 | struct tracer *saved_tracer = current_trace; | |
526 | struct trace_array *tr = &global_trace; | |
527 | int i; | |
528 | ||
529 | /* | |
530 | * Run a selftest on this tracer. | |
531 | * Here we reset the trace buffer, and set the current | |
532 | * tracer to be this tracer. The tracer can then run some | |
533 | * internal tracing to verify that everything is in order. | |
534 | * If we fail, we do not register this tracer. | |
535 | */ | |
536 | for_each_tracing_cpu(i) | |
537 | tracing_reset(tr, i); | |
538 | ||
539 | current_trace = type; | |
540 | /* the test is responsible for initializing and enabling */ | |
541 | pr_info("Testing tracer %s: ", type->name); | |
542 | ret = type->selftest(type, tr); | |
543 | /* the test is responsible for resetting too */ | |
544 | current_trace = saved_tracer; | |
545 | if (ret) { | |
546 | printk(KERN_CONT "FAILED!\n"); | |
547 | goto out; | |
548 | } | |
549 | /* Only reset on passing, to avoid touching corrupted buffers */ | |
550 | for_each_tracing_cpu(i) | |
551 | tracing_reset(tr, i); | |
552 | ||
553 | printk(KERN_CONT "PASSED\n"); | |
554 | } | |
555 | #endif | |
556 | ||
557 | type->next = trace_types; | |
558 | trace_types = type; | |
559 | len = strlen(type->name); | |
560 | if (len > max_tracer_type_len) | |
561 | max_tracer_type_len = len; | |
562 | ||
563 | out: | |
564 | tracing_selftest_running = false; | |
565 | mutex_unlock(&trace_types_lock); | |
566 | ||
567 | if (ret || !default_bootup_tracer) | |
568 | goto out_unlock; | |
569 | ||
570 | if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE)) | |
571 | goto out_unlock; | |
572 | ||
573 | printk(KERN_INFO "Starting tracer '%s'\n", type->name); | |
574 | /* Do we want this tracer to start on bootup? */ | |
575 | tracing_set_tracer(type->name); | |
576 | default_bootup_tracer = NULL; | |
577 | /* disable other selftests, since this will break it. */ | |
578 | tracing_selftest_disabled = 1; | |
579 | #ifdef CONFIG_FTRACE_STARTUP_TEST | |
580 | printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n", | |
581 | type->name); | |
582 | #endif | |
583 | ||
584 | out_unlock: | |
585 | lock_kernel(); | |
586 | return ret; | |
587 | } | |
588 | ||
589 | void unregister_tracer(struct tracer *type) | |
590 | { | |
591 | struct tracer **t; | |
592 | int len; | |
593 | ||
594 | mutex_lock(&trace_types_lock); | |
595 | for (t = &trace_types; *t; t = &(*t)->next) { | |
596 | if (*t == type) | |
597 | goto found; | |
598 | } | |
599 | pr_info("Trace %s not registered\n", type->name); | |
600 | goto out; | |
601 | ||
602 | found: | |
603 | *t = (*t)->next; | |
604 | ||
605 | if (type == current_trace && tracer_enabled) { | |
606 | tracer_enabled = 0; | |
607 | tracing_stop(); | |
608 | if (current_trace->stop) | |
609 | current_trace->stop(&global_trace); | |
610 | current_trace = &nop_trace; | |
611 | } | |
612 | ||
613 | if (strlen(type->name) != max_tracer_type_len) | |
614 | goto out; | |
615 | ||
616 | max_tracer_type_len = 0; | |
617 | for (t = &trace_types; *t; t = &(*t)->next) { | |
618 | len = strlen((*t)->name); | |
619 | if (len > max_tracer_type_len) | |
620 | max_tracer_type_len = len; | |
621 | } | |
622 | out: | |
623 | mutex_unlock(&trace_types_lock); | |
624 | } | |
625 | ||
626 | void tracing_reset(struct trace_array *tr, int cpu) | |
627 | { | |
628 | ftrace_disable_cpu(); | |
629 | ring_buffer_reset_cpu(tr->buffer, cpu); | |
630 | ftrace_enable_cpu(); | |
631 | } | |
632 | ||
633 | void tracing_reset_online_cpus(struct trace_array *tr) | |
634 | { | |
635 | int cpu; | |
636 | ||
637 | tr->time_start = ftrace_now(tr->cpu); | |
638 | ||
639 | for_each_online_cpu(cpu) | |
640 | tracing_reset(tr, cpu); | |
641 | } | |
642 | ||
643 | #define SAVED_CMDLINES 128 | |
644 | static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1]; | |
645 | static unsigned map_cmdline_to_pid[SAVED_CMDLINES]; | |
646 | static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN]; | |
647 | static int cmdline_idx; | |
648 | static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED; | |
649 | ||
650 | /* temporary disable recording */ | |
651 | static atomic_t trace_record_cmdline_disabled __read_mostly; | |
652 | ||
653 | static void trace_init_cmdlines(void) | |
654 | { | |
655 | memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline)); | |
656 | memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid)); | |
657 | cmdline_idx = 0; | |
658 | } | |
659 | ||
660 | static int trace_stop_count; | |
661 | static DEFINE_SPINLOCK(tracing_start_lock); | |
662 | ||
663 | /** | |
664 | * ftrace_off_permanent - disable all ftrace code permanently | |
665 | * | |
666 | * This should only be called when a serious anomally has | |
667 | * been detected. This will turn off the function tracing, | |
668 | * ring buffers, and other tracing utilites. It takes no | |
669 | * locks and can be called from any context. | |
670 | */ | |
671 | void ftrace_off_permanent(void) | |
672 | { | |
673 | tracing_disabled = 1; | |
674 | ftrace_stop(); | |
675 | tracing_off_permanent(); | |
676 | } | |
677 | ||
678 | /** | |
679 | * tracing_start - quick start of the tracer | |
680 | * | |
681 | * If tracing is enabled but was stopped by tracing_stop, | |
682 | * this will start the tracer back up. | |
683 | */ | |
684 | void tracing_start(void) | |
685 | { | |
686 | struct ring_buffer *buffer; | |
687 | unsigned long flags; | |
688 | ||
689 | if (tracing_disabled) | |
690 | return; | |
691 | ||
692 | spin_lock_irqsave(&tracing_start_lock, flags); | |
693 | if (--trace_stop_count) { | |
694 | if (trace_stop_count < 0) { | |
695 | /* Someone screwed up their debugging */ | |
696 | WARN_ON_ONCE(1); | |
697 | trace_stop_count = 0; | |
698 | } | |
699 | goto out; | |
700 | } | |
701 | ||
702 | ||
703 | buffer = global_trace.buffer; | |
704 | if (buffer) | |
705 | ring_buffer_record_enable(buffer); | |
706 | ||
707 | buffer = max_tr.buffer; | |
708 | if (buffer) | |
709 | ring_buffer_record_enable(buffer); | |
710 | ||
711 | ftrace_start(); | |
712 | out: | |
713 | spin_unlock_irqrestore(&tracing_start_lock, flags); | |
714 | } | |
715 | ||
716 | /** | |
717 | * tracing_stop - quick stop of the tracer | |
718 | * | |
719 | * Light weight way to stop tracing. Use in conjunction with | |
720 | * tracing_start. | |
721 | */ | |
722 | void tracing_stop(void) | |
723 | { | |
724 | struct ring_buffer *buffer; | |
725 | unsigned long flags; | |
726 | ||
727 | ftrace_stop(); | |
728 | spin_lock_irqsave(&tracing_start_lock, flags); | |
729 | if (trace_stop_count++) | |
730 | goto out; | |
731 | ||
732 | buffer = global_trace.buffer; | |
733 | if (buffer) | |
734 | ring_buffer_record_disable(buffer); | |
735 | ||
736 | buffer = max_tr.buffer; | |
737 | if (buffer) | |
738 | ring_buffer_record_disable(buffer); | |
739 | ||
740 | out: | |
741 | spin_unlock_irqrestore(&tracing_start_lock, flags); | |
742 | } | |
743 | ||
744 | void trace_stop_cmdline_recording(void); | |
745 | ||
746 | static void trace_save_cmdline(struct task_struct *tsk) | |
747 | { | |
748 | unsigned map; | |
749 | unsigned idx; | |
750 | ||
751 | if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT)) | |
752 | return; | |
753 | ||
754 | /* | |
755 | * It's not the end of the world if we don't get | |
756 | * the lock, but we also don't want to spin | |
757 | * nor do we want to disable interrupts, | |
758 | * so if we miss here, then better luck next time. | |
759 | */ | |
760 | if (!__raw_spin_trylock(&trace_cmdline_lock)) | |
761 | return; | |
762 | ||
763 | idx = map_pid_to_cmdline[tsk->pid]; | |
764 | if (idx >= SAVED_CMDLINES) { | |
765 | idx = (cmdline_idx + 1) % SAVED_CMDLINES; | |
766 | ||
767 | map = map_cmdline_to_pid[idx]; | |
768 | if (map <= PID_MAX_DEFAULT) | |
769 | map_pid_to_cmdline[map] = (unsigned)-1; | |
770 | ||
771 | map_pid_to_cmdline[tsk->pid] = idx; | |
772 | ||
773 | cmdline_idx = idx; | |
774 | } | |
775 | ||
776 | memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN); | |
777 | ||
778 | __raw_spin_unlock(&trace_cmdline_lock); | |
779 | } | |
780 | ||
781 | void trace_find_cmdline(int pid, char comm[]) | |
782 | { | |
783 | unsigned map; | |
784 | ||
785 | if (!pid) { | |
786 | strcpy(comm, "<idle>"); | |
787 | return; | |
788 | } | |
789 | ||
790 | if (pid > PID_MAX_DEFAULT) { | |
791 | strcpy(comm, "<...>"); | |
792 | return; | |
793 | } | |
794 | ||
795 | __raw_spin_lock(&trace_cmdline_lock); | |
796 | map = map_pid_to_cmdline[pid]; | |
797 | if (map >= SAVED_CMDLINES) | |
798 | goto out; | |
799 | ||
800 | strcpy(comm, saved_cmdlines[map]); | |
801 | ||
802 | out: | |
803 | __raw_spin_unlock(&trace_cmdline_lock); | |
804 | } | |
805 | ||
806 | void tracing_record_cmdline(struct task_struct *tsk) | |
807 | { | |
808 | if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on()) | |
809 | return; | |
810 | ||
811 | trace_save_cmdline(tsk); | |
812 | } | |
813 | ||
814 | void | |
815 | tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags, | |
816 | int pc) | |
817 | { | |
818 | struct task_struct *tsk = current; | |
819 | ||
820 | entry->preempt_count = pc & 0xff; | |
821 | entry->pid = (tsk) ? tsk->pid : 0; | |
822 | entry->tgid = (tsk) ? tsk->tgid : 0; | |
823 | entry->flags = | |
824 | #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT | |
825 | (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) | | |
826 | #else | |
827 | TRACE_FLAG_IRQS_NOSUPPORT | | |
828 | #endif | |
829 | ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) | | |
830 | ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) | | |
831 | (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0); | |
832 | } | |
833 | ||
834 | struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr, | |
835 | unsigned char type, | |
836 | unsigned long len, | |
837 | unsigned long flags, int pc) | |
838 | { | |
839 | struct ring_buffer_event *event; | |
840 | ||
841 | event = ring_buffer_lock_reserve(tr->buffer, len); | |
842 | if (event != NULL) { | |
843 | struct trace_entry *ent = ring_buffer_event_data(event); | |
844 | ||
845 | tracing_generic_entry_update(ent, flags, pc); | |
846 | ent->type = type; | |
847 | } | |
848 | ||
849 | return event; | |
850 | } | |
851 | static void ftrace_trace_stack(struct trace_array *tr, | |
852 | unsigned long flags, int skip, int pc); | |
853 | static void ftrace_trace_userstack(struct trace_array *tr, | |
854 | unsigned long flags, int pc); | |
855 | ||
856 | void trace_buffer_unlock_commit(struct trace_array *tr, | |
857 | struct ring_buffer_event *event, | |
858 | unsigned long flags, int pc) | |
859 | { | |
860 | ring_buffer_unlock_commit(tr->buffer, event); | |
861 | ||
862 | ftrace_trace_stack(tr, flags, 6, pc); | |
863 | ftrace_trace_userstack(tr, flags, pc); | |
864 | trace_wake_up(); | |
865 | } | |
866 | ||
867 | struct ring_buffer_event * | |
868 | trace_current_buffer_lock_reserve(unsigned char type, unsigned long len, | |
869 | unsigned long flags, int pc) | |
870 | { | |
871 | return trace_buffer_lock_reserve(&global_trace, | |
872 | type, len, flags, pc); | |
873 | } | |
874 | ||
875 | void trace_current_buffer_unlock_commit(struct ring_buffer_event *event, | |
876 | unsigned long flags, int pc) | |
877 | { | |
878 | return trace_buffer_unlock_commit(&global_trace, event, flags, pc); | |
879 | } | |
880 | ||
881 | void | |
882 | trace_function(struct trace_array *tr, | |
883 | unsigned long ip, unsigned long parent_ip, unsigned long flags, | |
884 | int pc) | |
885 | { | |
886 | struct ring_buffer_event *event; | |
887 | struct ftrace_entry *entry; | |
888 | ||
889 | /* If we are reading the ring buffer, don't trace */ | |
890 | if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled)))) | |
891 | return; | |
892 | ||
893 | event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry), | |
894 | flags, pc); | |
895 | if (!event) | |
896 | return; | |
897 | entry = ring_buffer_event_data(event); | |
898 | entry->ip = ip; | |
899 | entry->parent_ip = parent_ip; | |
900 | ring_buffer_unlock_commit(tr->buffer, event); | |
901 | } | |
902 | ||
903 | #ifdef CONFIG_FUNCTION_GRAPH_TRACER | |
904 | static void __trace_graph_entry(struct trace_array *tr, | |
905 | struct ftrace_graph_ent *trace, | |
906 | unsigned long flags, | |
907 | int pc) | |
908 | { | |
909 | struct ring_buffer_event *event; | |
910 | struct ftrace_graph_ent_entry *entry; | |
911 | ||
912 | if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled)))) | |
913 | return; | |
914 | ||
915 | event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT, | |
916 | sizeof(*entry), flags, pc); | |
917 | if (!event) | |
918 | return; | |
919 | entry = ring_buffer_event_data(event); | |
920 | entry->graph_ent = *trace; | |
921 | ring_buffer_unlock_commit(global_trace.buffer, event); | |
922 | } | |
923 | ||
924 | static void __trace_graph_return(struct trace_array *tr, | |
925 | struct ftrace_graph_ret *trace, | |
926 | unsigned long flags, | |
927 | int pc) | |
928 | { | |
929 | struct ring_buffer_event *event; | |
930 | struct ftrace_graph_ret_entry *entry; | |
931 | ||
932 | if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled)))) | |
933 | return; | |
934 | ||
935 | event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET, | |
936 | sizeof(*entry), flags, pc); | |
937 | if (!event) | |
938 | return; | |
939 | entry = ring_buffer_event_data(event); | |
940 | entry->ret = *trace; | |
941 | ring_buffer_unlock_commit(global_trace.buffer, event); | |
942 | } | |
943 | #endif | |
944 | ||
945 | void | |
946 | ftrace(struct trace_array *tr, struct trace_array_cpu *data, | |
947 | unsigned long ip, unsigned long parent_ip, unsigned long flags, | |
948 | int pc) | |
949 | { | |
950 | if (likely(!atomic_read(&data->disabled))) | |
951 | trace_function(tr, ip, parent_ip, flags, pc); | |
952 | } | |
953 | ||
954 | static void __ftrace_trace_stack(struct trace_array *tr, | |
955 | unsigned long flags, | |
956 | int skip, int pc) | |
957 | { | |
958 | #ifdef CONFIG_STACKTRACE | |
959 | struct ring_buffer_event *event; | |
960 | struct stack_entry *entry; | |
961 | struct stack_trace trace; | |
962 | ||
963 | event = trace_buffer_lock_reserve(tr, TRACE_STACK, | |
964 | sizeof(*entry), flags, pc); | |
965 | if (!event) | |
966 | return; | |
967 | entry = ring_buffer_event_data(event); | |
968 | memset(&entry->caller, 0, sizeof(entry->caller)); | |
969 | ||
970 | trace.nr_entries = 0; | |
971 | trace.max_entries = FTRACE_STACK_ENTRIES; | |
972 | trace.skip = skip; | |
973 | trace.entries = entry->caller; | |
974 | ||
975 | save_stack_trace(&trace); | |
976 | ring_buffer_unlock_commit(tr->buffer, event); | |
977 | #endif | |
978 | } | |
979 | ||
980 | static void ftrace_trace_stack(struct trace_array *tr, | |
981 | unsigned long flags, | |
982 | int skip, int pc) | |
983 | { | |
984 | if (!(trace_flags & TRACE_ITER_STACKTRACE)) | |
985 | return; | |
986 | ||
987 | __ftrace_trace_stack(tr, flags, skip, pc); | |
988 | } | |
989 | ||
990 | void __trace_stack(struct trace_array *tr, | |
991 | unsigned long flags, | |
992 | int skip, int pc) | |
993 | { | |
994 | __ftrace_trace_stack(tr, flags, skip, pc); | |
995 | } | |
996 | ||
997 | static void ftrace_trace_userstack(struct trace_array *tr, | |
998 | unsigned long flags, int pc) | |
999 | { | |
1000 | #ifdef CONFIG_STACKTRACE | |
1001 | struct ring_buffer_event *event; | |
1002 | struct userstack_entry *entry; | |
1003 | struct stack_trace trace; | |
1004 | ||
1005 | if (!(trace_flags & TRACE_ITER_USERSTACKTRACE)) | |
1006 | return; | |
1007 | ||
1008 | event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK, | |
1009 | sizeof(*entry), flags, pc); | |
1010 | if (!event) | |
1011 | return; | |
1012 | entry = ring_buffer_event_data(event); | |
1013 | ||
1014 | memset(&entry->caller, 0, sizeof(entry->caller)); | |
1015 | ||
1016 | trace.nr_entries = 0; | |
1017 | trace.max_entries = FTRACE_STACK_ENTRIES; | |
1018 | trace.skip = 0; | |
1019 | trace.entries = entry->caller; | |
1020 | ||
1021 | save_stack_trace_user(&trace); | |
1022 | ring_buffer_unlock_commit(tr->buffer, event); | |
1023 | #endif | |
1024 | } | |
1025 | ||
1026 | #ifdef UNUSED | |
1027 | static void __trace_userstack(struct trace_array *tr, unsigned long flags) | |
1028 | { | |
1029 | ftrace_trace_userstack(tr, flags, preempt_count()); | |
1030 | } | |
1031 | #endif /* UNUSED */ | |
1032 | ||
1033 | static void | |
1034 | ftrace_trace_special(void *__tr, | |
1035 | unsigned long arg1, unsigned long arg2, unsigned long arg3, | |
1036 | int pc) | |
1037 | { | |
1038 | struct ring_buffer_event *event; | |
1039 | struct trace_array *tr = __tr; | |
1040 | struct special_entry *entry; | |
1041 | ||
1042 | event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL, | |
1043 | sizeof(*entry), 0, pc); | |
1044 | if (!event) | |
1045 | return; | |
1046 | entry = ring_buffer_event_data(event); | |
1047 | entry->arg1 = arg1; | |
1048 | entry->arg2 = arg2; | |
1049 | entry->arg3 = arg3; | |
1050 | trace_buffer_unlock_commit(tr, event, 0, pc); | |
1051 | } | |
1052 | ||
1053 | void | |
1054 | __trace_special(void *__tr, void *__data, | |
1055 | unsigned long arg1, unsigned long arg2, unsigned long arg3) | |
1056 | { | |
1057 | ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count()); | |
1058 | } | |
1059 | ||
1060 | void | |
1061 | tracing_sched_switch_trace(struct trace_array *tr, | |
1062 | struct task_struct *prev, | |
1063 | struct task_struct *next, | |
1064 | unsigned long flags, int pc) | |
1065 | { | |
1066 | struct ring_buffer_event *event; | |
1067 | struct ctx_switch_entry *entry; | |
1068 | ||
1069 | event = trace_buffer_lock_reserve(tr, TRACE_CTX, | |
1070 | sizeof(*entry), flags, pc); | |
1071 | if (!event) | |
1072 | return; | |
1073 | entry = ring_buffer_event_data(event); | |
1074 | entry->prev_pid = prev->pid; | |
1075 | entry->prev_prio = prev->prio; | |
1076 | entry->prev_state = prev->state; | |
1077 | entry->next_pid = next->pid; | |
1078 | entry->next_prio = next->prio; | |
1079 | entry->next_state = next->state; | |
1080 | entry->next_cpu = task_cpu(next); | |
1081 | trace_buffer_unlock_commit(tr, event, flags, pc); | |
1082 | } | |
1083 | ||
1084 | void | |
1085 | tracing_sched_wakeup_trace(struct trace_array *tr, | |
1086 | struct task_struct *wakee, | |
1087 | struct task_struct *curr, | |
1088 | unsigned long flags, int pc) | |
1089 | { | |
1090 | struct ring_buffer_event *event; | |
1091 | struct ctx_switch_entry *entry; | |
1092 | ||
1093 | event = trace_buffer_lock_reserve(tr, TRACE_WAKE, | |
1094 | sizeof(*entry), flags, pc); | |
1095 | if (!event) | |
1096 | return; | |
1097 | entry = ring_buffer_event_data(event); | |
1098 | entry->prev_pid = curr->pid; | |
1099 | entry->prev_prio = curr->prio; | |
1100 | entry->prev_state = curr->state; | |
1101 | entry->next_pid = wakee->pid; | |
1102 | entry->next_prio = wakee->prio; | |
1103 | entry->next_state = wakee->state; | |
1104 | entry->next_cpu = task_cpu(wakee); | |
1105 | ||
1106 | ring_buffer_unlock_commit(tr->buffer, event); | |
1107 | ftrace_trace_stack(tr, flags, 6, pc); | |
1108 | ftrace_trace_userstack(tr, flags, pc); | |
1109 | } | |
1110 | ||
1111 | void | |
1112 | ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3) | |
1113 | { | |
1114 | struct trace_array *tr = &global_trace; | |
1115 | struct trace_array_cpu *data; | |
1116 | unsigned long flags; | |
1117 | int cpu; | |
1118 | int pc; | |
1119 | ||
1120 | if (tracing_disabled) | |
1121 | return; | |
1122 | ||
1123 | pc = preempt_count(); | |
1124 | local_irq_save(flags); | |
1125 | cpu = raw_smp_processor_id(); | |
1126 | data = tr->data[cpu]; | |
1127 | ||
1128 | if (likely(atomic_inc_return(&data->disabled) == 1)) | |
1129 | ftrace_trace_special(tr, arg1, arg2, arg3, pc); | |
1130 | ||
1131 | atomic_dec(&data->disabled); | |
1132 | local_irq_restore(flags); | |
1133 | } | |
1134 | ||
1135 | #ifdef CONFIG_FUNCTION_GRAPH_TRACER | |
1136 | int trace_graph_entry(struct ftrace_graph_ent *trace) | |
1137 | { | |
1138 | struct trace_array *tr = &global_trace; | |
1139 | struct trace_array_cpu *data; | |
1140 | unsigned long flags; | |
1141 | long disabled; | |
1142 | int cpu; | |
1143 | int pc; | |
1144 | ||
1145 | if (!ftrace_trace_task(current)) | |
1146 | return 0; | |
1147 | ||
1148 | if (!ftrace_graph_addr(trace->func)) | |
1149 | return 0; | |
1150 | ||
1151 | local_irq_save(flags); | |
1152 | cpu = raw_smp_processor_id(); | |
1153 | data = tr->data[cpu]; | |
1154 | disabled = atomic_inc_return(&data->disabled); | |
1155 | if (likely(disabled == 1)) { | |
1156 | pc = preempt_count(); | |
1157 | __trace_graph_entry(tr, trace, flags, pc); | |
1158 | } | |
1159 | /* Only do the atomic if it is not already set */ | |
1160 | if (!test_tsk_trace_graph(current)) | |
1161 | set_tsk_trace_graph(current); | |
1162 | atomic_dec(&data->disabled); | |
1163 | local_irq_restore(flags); | |
1164 | ||
1165 | return 1; | |
1166 | } | |
1167 | ||
1168 | void trace_graph_return(struct ftrace_graph_ret *trace) | |
1169 | { | |
1170 | struct trace_array *tr = &global_trace; | |
1171 | struct trace_array_cpu *data; | |
1172 | unsigned long flags; | |
1173 | long disabled; | |
1174 | int cpu; | |
1175 | int pc; | |
1176 | ||
1177 | local_irq_save(flags); | |
1178 | cpu = raw_smp_processor_id(); | |
1179 | data = tr->data[cpu]; | |
1180 | disabled = atomic_inc_return(&data->disabled); | |
1181 | if (likely(disabled == 1)) { | |
1182 | pc = preempt_count(); | |
1183 | __trace_graph_return(tr, trace, flags, pc); | |
1184 | } | |
1185 | if (!trace->depth) | |
1186 | clear_tsk_trace_graph(current); | |
1187 | atomic_dec(&data->disabled); | |
1188 | local_irq_restore(flags); | |
1189 | } | |
1190 | #endif /* CONFIG_FUNCTION_GRAPH_TRACER */ | |
1191 | ||
1192 | ||
1193 | /** | |
1194 | * trace_vbprintk - write binary msg to tracing buffer | |
1195 | * | |
1196 | */ | |
1197 | int trace_vbprintk(unsigned long ip, int depth, const char *fmt, va_list args) | |
1198 | { | |
1199 | static raw_spinlock_t trace_buf_lock = | |
1200 | (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; | |
1201 | static u32 trace_buf[TRACE_BUF_SIZE]; | |
1202 | ||
1203 | struct ring_buffer_event *event; | |
1204 | struct trace_array *tr = &global_trace; | |
1205 | struct trace_array_cpu *data; | |
1206 | struct bprint_entry *entry; | |
1207 | unsigned long flags; | |
1208 | int resched; | |
1209 | int cpu, len = 0, size, pc; | |
1210 | ||
1211 | if (unlikely(tracing_selftest_running || tracing_disabled)) | |
1212 | return 0; | |
1213 | ||
1214 | /* Don't pollute graph traces with trace_vprintk internals */ | |
1215 | pause_graph_tracing(); | |
1216 | ||
1217 | pc = preempt_count(); | |
1218 | resched = ftrace_preempt_disable(); | |
1219 | cpu = raw_smp_processor_id(); | |
1220 | data = tr->data[cpu]; | |
1221 | ||
1222 | if (unlikely(atomic_read(&data->disabled))) | |
1223 | goto out; | |
1224 | ||
1225 | /* Lockdep uses trace_printk for lock tracing */ | |
1226 | local_irq_save(flags); | |
1227 | __raw_spin_lock(&trace_buf_lock); | |
1228 | len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args); | |
1229 | ||
1230 | if (len > TRACE_BUF_SIZE || len < 0) | |
1231 | goto out_unlock; | |
1232 | ||
1233 | size = sizeof(*entry) + sizeof(u32) * len; | |
1234 | event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc); | |
1235 | if (!event) | |
1236 | goto out_unlock; | |
1237 | entry = ring_buffer_event_data(event); | |
1238 | entry->ip = ip; | |
1239 | entry->depth = depth; | |
1240 | entry->fmt = fmt; | |
1241 | ||
1242 | memcpy(entry->buf, trace_buf, sizeof(u32) * len); | |
1243 | ring_buffer_unlock_commit(tr->buffer, event); | |
1244 | ||
1245 | out_unlock: | |
1246 | __raw_spin_unlock(&trace_buf_lock); | |
1247 | local_irq_restore(flags); | |
1248 | ||
1249 | out: | |
1250 | ftrace_preempt_enable(resched); | |
1251 | unpause_graph_tracing(); | |
1252 | ||
1253 | return len; | |
1254 | } | |
1255 | EXPORT_SYMBOL_GPL(trace_vbprintk); | |
1256 | ||
1257 | int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args) | |
1258 | { | |
1259 | static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED; | |
1260 | static char trace_buf[TRACE_BUF_SIZE]; | |
1261 | ||
1262 | struct ring_buffer_event *event; | |
1263 | struct trace_array *tr = &global_trace; | |
1264 | struct trace_array_cpu *data; | |
1265 | int cpu, len = 0, size, pc; | |
1266 | struct print_entry *entry; | |
1267 | unsigned long irq_flags; | |
1268 | ||
1269 | if (tracing_disabled || tracing_selftest_running) | |
1270 | return 0; | |
1271 | ||
1272 | pc = preempt_count(); | |
1273 | preempt_disable_notrace(); | |
1274 | cpu = raw_smp_processor_id(); | |
1275 | data = tr->data[cpu]; | |
1276 | ||
1277 | if (unlikely(atomic_read(&data->disabled))) | |
1278 | goto out; | |
1279 | ||
1280 | pause_graph_tracing(); | |
1281 | raw_local_irq_save(irq_flags); | |
1282 | __raw_spin_lock(&trace_buf_lock); | |
1283 | len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args); | |
1284 | ||
1285 | len = min(len, TRACE_BUF_SIZE-1); | |
1286 | trace_buf[len] = 0; | |
1287 | ||
1288 | size = sizeof(*entry) + len + 1; | |
1289 | event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc); | |
1290 | if (!event) | |
1291 | goto out_unlock; | |
1292 | entry = ring_buffer_event_data(event); | |
1293 | entry->ip = ip; | |
1294 | entry->depth = depth; | |
1295 | ||
1296 | memcpy(&entry->buf, trace_buf, len); | |
1297 | entry->buf[len] = 0; | |
1298 | ring_buffer_unlock_commit(tr->buffer, event); | |
1299 | ||
1300 | out_unlock: | |
1301 | __raw_spin_unlock(&trace_buf_lock); | |
1302 | raw_local_irq_restore(irq_flags); | |
1303 | unpause_graph_tracing(); | |
1304 | out: | |
1305 | preempt_enable_notrace(); | |
1306 | ||
1307 | return len; | |
1308 | } | |
1309 | EXPORT_SYMBOL_GPL(trace_vprintk); | |
1310 | ||
1311 | enum trace_file_type { | |
1312 | TRACE_FILE_LAT_FMT = 1, | |
1313 | TRACE_FILE_ANNOTATE = 2, | |
1314 | }; | |
1315 | ||
1316 | static void trace_iterator_increment(struct trace_iterator *iter) | |
1317 | { | |
1318 | /* Don't allow ftrace to trace into the ring buffers */ | |
1319 | ftrace_disable_cpu(); | |
1320 | ||
1321 | iter->idx++; | |
1322 | if (iter->buffer_iter[iter->cpu]) | |
1323 | ring_buffer_read(iter->buffer_iter[iter->cpu], NULL); | |
1324 | ||
1325 | ftrace_enable_cpu(); | |
1326 | } | |
1327 | ||
1328 | static struct trace_entry * | |
1329 | peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts) | |
1330 | { | |
1331 | struct ring_buffer_event *event; | |
1332 | struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu]; | |
1333 | ||
1334 | /* Don't allow ftrace to trace into the ring buffers */ | |
1335 | ftrace_disable_cpu(); | |
1336 | ||
1337 | if (buf_iter) | |
1338 | event = ring_buffer_iter_peek(buf_iter, ts); | |
1339 | else | |
1340 | event = ring_buffer_peek(iter->tr->buffer, cpu, ts); | |
1341 | ||
1342 | ftrace_enable_cpu(); | |
1343 | ||
1344 | return event ? ring_buffer_event_data(event) : NULL; | |
1345 | } | |
1346 | ||
1347 | static struct trace_entry * | |
1348 | __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts) | |
1349 | { | |
1350 | struct ring_buffer *buffer = iter->tr->buffer; | |
1351 | struct trace_entry *ent, *next = NULL; | |
1352 | int cpu_file = iter->cpu_file; | |
1353 | u64 next_ts = 0, ts; | |
1354 | int next_cpu = -1; | |
1355 | int cpu; | |
1356 | ||
1357 | /* | |
1358 | * If we are in a per_cpu trace file, don't bother by iterating over | |
1359 | * all cpu and peek directly. | |
1360 | */ | |
1361 | if (cpu_file > TRACE_PIPE_ALL_CPU) { | |
1362 | if (ring_buffer_empty_cpu(buffer, cpu_file)) | |
1363 | return NULL; | |
1364 | ent = peek_next_entry(iter, cpu_file, ent_ts); | |
1365 | if (ent_cpu) | |
1366 | *ent_cpu = cpu_file; | |
1367 | ||
1368 | return ent; | |
1369 | } | |
1370 | ||
1371 | for_each_tracing_cpu(cpu) { | |
1372 | ||
1373 | if (ring_buffer_empty_cpu(buffer, cpu)) | |
1374 | continue; | |
1375 | ||
1376 | ent = peek_next_entry(iter, cpu, &ts); | |
1377 | ||
1378 | /* | |
1379 | * Pick the entry with the smallest timestamp: | |
1380 | */ | |
1381 | if (ent && (!next || ts < next_ts)) { | |
1382 | next = ent; | |
1383 | next_cpu = cpu; | |
1384 | next_ts = ts; | |
1385 | } | |
1386 | } | |
1387 | ||
1388 | if (ent_cpu) | |
1389 | *ent_cpu = next_cpu; | |
1390 | ||
1391 | if (ent_ts) | |
1392 | *ent_ts = next_ts; | |
1393 | ||
1394 | return next; | |
1395 | } | |
1396 | ||
1397 | /* Find the next real entry, without updating the iterator itself */ | |
1398 | struct trace_entry *trace_find_next_entry(struct trace_iterator *iter, | |
1399 | int *ent_cpu, u64 *ent_ts) | |
1400 | { | |
1401 | return __find_next_entry(iter, ent_cpu, ent_ts); | |
1402 | } | |
1403 | ||
1404 | /* Find the next real entry, and increment the iterator to the next entry */ | |
1405 | static void *find_next_entry_inc(struct trace_iterator *iter) | |
1406 | { | |
1407 | iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts); | |
1408 | ||
1409 | if (iter->ent) | |
1410 | trace_iterator_increment(iter); | |
1411 | ||
1412 | return iter->ent ? iter : NULL; | |
1413 | } | |
1414 | ||
1415 | static void trace_consume(struct trace_iterator *iter) | |
1416 | { | |
1417 | /* Don't allow ftrace to trace into the ring buffers */ | |
1418 | ftrace_disable_cpu(); | |
1419 | ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts); | |
1420 | ftrace_enable_cpu(); | |
1421 | } | |
1422 | ||
1423 | static void *s_next(struct seq_file *m, void *v, loff_t *pos) | |
1424 | { | |
1425 | struct trace_iterator *iter = m->private; | |
1426 | int i = (int)*pos; | |
1427 | void *ent; | |
1428 | ||
1429 | (*pos)++; | |
1430 | ||
1431 | /* can't go backwards */ | |
1432 | if (iter->idx > i) | |
1433 | return NULL; | |
1434 | ||
1435 | if (iter->idx < 0) | |
1436 | ent = find_next_entry_inc(iter); | |
1437 | else | |
1438 | ent = iter; | |
1439 | ||
1440 | while (ent && iter->idx < i) | |
1441 | ent = find_next_entry_inc(iter); | |
1442 | ||
1443 | iter->pos = *pos; | |
1444 | ||
1445 | return ent; | |
1446 | } | |
1447 | ||
1448 | /* | |
1449 | * No necessary locking here. The worst thing which can | |
1450 | * happen is loosing events consumed at the same time | |
1451 | * by a trace_pipe reader. | |
1452 | * Other than that, we don't risk to crash the ring buffer | |
1453 | * because it serializes the readers. | |
1454 | * | |
1455 | * The current tracer is copied to avoid a global locking | |
1456 | * all around. | |
1457 | */ | |
1458 | static void *s_start(struct seq_file *m, loff_t *pos) | |
1459 | { | |
1460 | struct trace_iterator *iter = m->private; | |
1461 | static struct tracer *old_tracer; | |
1462 | int cpu_file = iter->cpu_file; | |
1463 | void *p = NULL; | |
1464 | loff_t l = 0; | |
1465 | int cpu; | |
1466 | ||
1467 | /* copy the tracer to avoid using a global lock all around */ | |
1468 | mutex_lock(&trace_types_lock); | |
1469 | if (unlikely(old_tracer != current_trace && current_trace)) { | |
1470 | old_tracer = current_trace; | |
1471 | *iter->trace = *current_trace; | |
1472 | } | |
1473 | mutex_unlock(&trace_types_lock); | |
1474 | ||
1475 | atomic_inc(&trace_record_cmdline_disabled); | |
1476 | ||
1477 | if (*pos != iter->pos) { | |
1478 | iter->ent = NULL; | |
1479 | iter->cpu = 0; | |
1480 | iter->idx = -1; | |
1481 | ||
1482 | ftrace_disable_cpu(); | |
1483 | ||
1484 | if (cpu_file == TRACE_PIPE_ALL_CPU) { | |
1485 | for_each_tracing_cpu(cpu) | |
1486 | ring_buffer_iter_reset(iter->buffer_iter[cpu]); | |
1487 | } else | |
1488 | ring_buffer_iter_reset(iter->buffer_iter[cpu_file]); | |
1489 | ||
1490 | ||
1491 | ftrace_enable_cpu(); | |
1492 | ||
1493 | for (p = iter; p && l < *pos; p = s_next(m, p, &l)) | |
1494 | ; | |
1495 | ||
1496 | } else { | |
1497 | l = *pos - 1; | |
1498 | p = s_next(m, p, &l); | |
1499 | } | |
1500 | ||
1501 | return p; | |
1502 | } | |
1503 | ||
1504 | static void s_stop(struct seq_file *m, void *p) | |
1505 | { | |
1506 | atomic_dec(&trace_record_cmdline_disabled); | |
1507 | } | |
1508 | ||
1509 | static void print_lat_help_header(struct seq_file *m) | |
1510 | { | |
1511 | seq_puts(m, "# _------=> CPU# \n"); | |
1512 | seq_puts(m, "# / _-----=> irqs-off \n"); | |
1513 | seq_puts(m, "# | / _----=> need-resched \n"); | |
1514 | seq_puts(m, "# || / _---=> hardirq/softirq \n"); | |
1515 | seq_puts(m, "# ||| / _--=> preempt-depth \n"); | |
1516 | seq_puts(m, "# |||| / \n"); | |
1517 | seq_puts(m, "# ||||| delay \n"); | |
1518 | seq_puts(m, "# cmd pid ||||| time | caller \n"); | |
1519 | seq_puts(m, "# \\ / ||||| \\ | / \n"); | |
1520 | } | |
1521 | ||
1522 | static void print_func_help_header(struct seq_file *m) | |
1523 | { | |
1524 | seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n"); | |
1525 | seq_puts(m, "# | | | | |\n"); | |
1526 | } | |
1527 | ||
1528 | ||
1529 | static void | |
1530 | print_trace_header(struct seq_file *m, struct trace_iterator *iter) | |
1531 | { | |
1532 | unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK); | |
1533 | struct trace_array *tr = iter->tr; | |
1534 | struct trace_array_cpu *data = tr->data[tr->cpu]; | |
1535 | struct tracer *type = current_trace; | |
1536 | unsigned long total; | |
1537 | unsigned long entries; | |
1538 | const char *name = "preemption"; | |
1539 | ||
1540 | if (type) | |
1541 | name = type->name; | |
1542 | ||
1543 | entries = ring_buffer_entries(iter->tr->buffer); | |
1544 | total = entries + | |
1545 | ring_buffer_overruns(iter->tr->buffer); | |
1546 | ||
1547 | seq_printf(m, "# %s latency trace v1.1.5 on %s\n", | |
1548 | name, UTS_RELEASE); | |
1549 | seq_puts(m, "# -----------------------------------" | |
1550 | "---------------------------------\n"); | |
1551 | seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |" | |
1552 | " (M:%s VP:%d, KP:%d, SP:%d HP:%d", | |
1553 | nsecs_to_usecs(data->saved_latency), | |
1554 | entries, | |
1555 | total, | |
1556 | tr->cpu, | |
1557 | #if defined(CONFIG_PREEMPT_NONE) | |
1558 | "server", | |
1559 | #elif defined(CONFIG_PREEMPT_VOLUNTARY) | |
1560 | "desktop", | |
1561 | #elif defined(CONFIG_PREEMPT) | |
1562 | "preempt", | |
1563 | #else | |
1564 | "unknown", | |
1565 | #endif | |
1566 | /* These are reserved for later use */ | |
1567 | 0, 0, 0, 0); | |
1568 | #ifdef CONFIG_SMP | |
1569 | seq_printf(m, " #P:%d)\n", num_online_cpus()); | |
1570 | #else | |
1571 | seq_puts(m, ")\n"); | |
1572 | #endif | |
1573 | seq_puts(m, "# -----------------\n"); | |
1574 | seq_printf(m, "# | task: %.16s-%d " | |
1575 | "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n", | |
1576 | data->comm, data->pid, data->uid, data->nice, | |
1577 | data->policy, data->rt_priority); | |
1578 | seq_puts(m, "# -----------------\n"); | |
1579 | ||
1580 | if (data->critical_start) { | |
1581 | seq_puts(m, "# => started at: "); | |
1582 | seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags); | |
1583 | trace_print_seq(m, &iter->seq); | |
1584 | seq_puts(m, "\n# => ended at: "); | |
1585 | seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags); | |
1586 | trace_print_seq(m, &iter->seq); | |
1587 | seq_puts(m, "#\n"); | |
1588 | } | |
1589 | ||
1590 | seq_puts(m, "#\n"); | |
1591 | } | |
1592 | ||
1593 | static void test_cpu_buff_start(struct trace_iterator *iter) | |
1594 | { | |
1595 | struct trace_seq *s = &iter->seq; | |
1596 | ||
1597 | if (!(trace_flags & TRACE_ITER_ANNOTATE)) | |
1598 | return; | |
1599 | ||
1600 | if (!(iter->iter_flags & TRACE_FILE_ANNOTATE)) | |
1601 | return; | |
1602 | ||
1603 | if (cpumask_test_cpu(iter->cpu, iter->started)) | |
1604 | return; | |
1605 | ||
1606 | cpumask_set_cpu(iter->cpu, iter->started); | |
1607 | trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu); | |
1608 | } | |
1609 | ||
1610 | static enum print_line_t print_trace_fmt(struct trace_iterator *iter) | |
1611 | { | |
1612 | struct trace_seq *s = &iter->seq; | |
1613 | unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK); | |
1614 | struct trace_entry *entry; | |
1615 | struct trace_event *event; | |
1616 | ||
1617 | entry = iter->ent; | |
1618 | ||
1619 | test_cpu_buff_start(iter); | |
1620 | ||
1621 | event = ftrace_find_event(entry->type); | |
1622 | ||
1623 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
1624 | if (iter->iter_flags & TRACE_FILE_LAT_FMT) { | |
1625 | if (!trace_print_lat_context(iter)) | |
1626 | goto partial; | |
1627 | } else { | |
1628 | if (!trace_print_context(iter)) | |
1629 | goto partial; | |
1630 | } | |
1631 | } | |
1632 | ||
1633 | if (event) | |
1634 | return event->trace(iter, sym_flags); | |
1635 | ||
1636 | if (!trace_seq_printf(s, "Unknown type %d\n", entry->type)) | |
1637 | goto partial; | |
1638 | ||
1639 | return TRACE_TYPE_HANDLED; | |
1640 | partial: | |
1641 | return TRACE_TYPE_PARTIAL_LINE; | |
1642 | } | |
1643 | ||
1644 | static enum print_line_t print_raw_fmt(struct trace_iterator *iter) | |
1645 | { | |
1646 | struct trace_seq *s = &iter->seq; | |
1647 | struct trace_entry *entry; | |
1648 | struct trace_event *event; | |
1649 | ||
1650 | entry = iter->ent; | |
1651 | ||
1652 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
1653 | if (!trace_seq_printf(s, "%d %d %llu ", | |
1654 | entry->pid, iter->cpu, iter->ts)) | |
1655 | goto partial; | |
1656 | } | |
1657 | ||
1658 | event = ftrace_find_event(entry->type); | |
1659 | if (event) | |
1660 | return event->raw(iter, 0); | |
1661 | ||
1662 | if (!trace_seq_printf(s, "%d ?\n", entry->type)) | |
1663 | goto partial; | |
1664 | ||
1665 | return TRACE_TYPE_HANDLED; | |
1666 | partial: | |
1667 | return TRACE_TYPE_PARTIAL_LINE; | |
1668 | } | |
1669 | ||
1670 | static enum print_line_t print_hex_fmt(struct trace_iterator *iter) | |
1671 | { | |
1672 | struct trace_seq *s = &iter->seq; | |
1673 | unsigned char newline = '\n'; | |
1674 | struct trace_entry *entry; | |
1675 | struct trace_event *event; | |
1676 | ||
1677 | entry = iter->ent; | |
1678 | ||
1679 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
1680 | SEQ_PUT_HEX_FIELD_RET(s, entry->pid); | |
1681 | SEQ_PUT_HEX_FIELD_RET(s, iter->cpu); | |
1682 | SEQ_PUT_HEX_FIELD_RET(s, iter->ts); | |
1683 | } | |
1684 | ||
1685 | event = ftrace_find_event(entry->type); | |
1686 | if (event) { | |
1687 | enum print_line_t ret = event->hex(iter, 0); | |
1688 | if (ret != TRACE_TYPE_HANDLED) | |
1689 | return ret; | |
1690 | } | |
1691 | ||
1692 | SEQ_PUT_FIELD_RET(s, newline); | |
1693 | ||
1694 | return TRACE_TYPE_HANDLED; | |
1695 | } | |
1696 | ||
1697 | static enum print_line_t print_bprintk_msg_only(struct trace_iterator *iter) | |
1698 | { | |
1699 | struct trace_seq *s = &iter->seq; | |
1700 | struct trace_entry *entry = iter->ent; | |
1701 | struct bprint_entry *field; | |
1702 | int ret; | |
1703 | ||
1704 | trace_assign_type(field, entry); | |
1705 | ||
1706 | ret = trace_seq_bprintf(s, field->fmt, field->buf); | |
1707 | if (!ret) | |
1708 | return TRACE_TYPE_PARTIAL_LINE; | |
1709 | ||
1710 | return TRACE_TYPE_HANDLED; | |
1711 | } | |
1712 | ||
1713 | static enum print_line_t print_printk_msg_only(struct trace_iterator *iter) | |
1714 | { | |
1715 | struct trace_seq *s = &iter->seq; | |
1716 | struct trace_entry *entry = iter->ent; | |
1717 | struct print_entry *field; | |
1718 | int ret; | |
1719 | ||
1720 | trace_assign_type(field, entry); | |
1721 | ||
1722 | ret = trace_seq_printf(s, "%s", field->buf); | |
1723 | if (!ret) | |
1724 | return TRACE_TYPE_PARTIAL_LINE; | |
1725 | ||
1726 | return TRACE_TYPE_HANDLED; | |
1727 | } | |
1728 | ||
1729 | static enum print_line_t print_bin_fmt(struct trace_iterator *iter) | |
1730 | { | |
1731 | struct trace_seq *s = &iter->seq; | |
1732 | struct trace_entry *entry; | |
1733 | struct trace_event *event; | |
1734 | ||
1735 | entry = iter->ent; | |
1736 | ||
1737 | if (trace_flags & TRACE_ITER_CONTEXT_INFO) { | |
1738 | SEQ_PUT_FIELD_RET(s, entry->pid); | |
1739 | SEQ_PUT_FIELD_RET(s, iter->cpu); | |
1740 | SEQ_PUT_FIELD_RET(s, iter->ts); | |
1741 | } | |
1742 | ||
1743 | event = ftrace_find_event(entry->type); | |
1744 | return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED; | |
1745 | } | |
1746 | ||
1747 | static int trace_empty(struct trace_iterator *iter) | |
1748 | { | |
1749 | int cpu; | |
1750 | ||
1751 | /* If we are looking at one CPU buffer, only check that one */ | |
1752 | if (iter->cpu_file != TRACE_PIPE_ALL_CPU) { | |
1753 | cpu = iter->cpu_file; | |
1754 | if (iter->buffer_iter[cpu]) { | |
1755 | if (!ring_buffer_iter_empty(iter->buffer_iter[cpu])) | |
1756 | return 0; | |
1757 | } else { | |
1758 | if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu)) | |
1759 | return 0; | |
1760 | } | |
1761 | return 1; | |
1762 | } | |
1763 | ||
1764 | for_each_tracing_cpu(cpu) { | |
1765 | if (iter->buffer_iter[cpu]) { | |
1766 | if (!ring_buffer_iter_empty(iter->buffer_iter[cpu])) | |
1767 | return 0; | |
1768 | } else { | |
1769 | if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu)) | |
1770 | return 0; | |
1771 | } | |
1772 | } | |
1773 | ||
1774 | return 1; | |
1775 | } | |
1776 | ||
1777 | static enum print_line_t print_trace_line(struct trace_iterator *iter) | |
1778 | { | |
1779 | enum print_line_t ret; | |
1780 | ||
1781 | if (iter->trace && iter->trace->print_line) { | |
1782 | ret = iter->trace->print_line(iter); | |
1783 | if (ret != TRACE_TYPE_UNHANDLED) | |
1784 | return ret; | |
1785 | } | |
1786 | ||
1787 | if (iter->ent->type == TRACE_BPRINT && | |
1788 | trace_flags & TRACE_ITER_PRINTK && | |
1789 | trace_flags & TRACE_ITER_PRINTK_MSGONLY) | |
1790 | return print_bprintk_msg_only(iter); | |
1791 | ||
1792 | if (iter->ent->type == TRACE_PRINT && | |
1793 | trace_flags & TRACE_ITER_PRINTK && | |
1794 | trace_flags & TRACE_ITER_PRINTK_MSGONLY) | |
1795 | return print_printk_msg_only(iter); | |
1796 | ||
1797 | if (trace_flags & TRACE_ITER_BIN) | |
1798 | return print_bin_fmt(iter); | |
1799 | ||
1800 | if (trace_flags & TRACE_ITER_HEX) | |
1801 | return print_hex_fmt(iter); | |
1802 | ||
1803 | if (trace_flags & TRACE_ITER_RAW) | |
1804 | return print_raw_fmt(iter); | |
1805 | ||
1806 | return print_trace_fmt(iter); | |
1807 | } | |
1808 | ||
1809 | static int s_show(struct seq_file *m, void *v) | |
1810 | { | |
1811 | struct trace_iterator *iter = v; | |
1812 | ||
1813 | if (iter->ent == NULL) { | |
1814 | if (iter->tr) { | |
1815 | seq_printf(m, "# tracer: %s\n", iter->trace->name); | |
1816 | seq_puts(m, "#\n"); | |
1817 | } | |
1818 | if (iter->trace && iter->trace->print_header) | |
1819 | iter->trace->print_header(m); | |
1820 | else if (iter->iter_flags & TRACE_FILE_LAT_FMT) { | |
1821 | /* print nothing if the buffers are empty */ | |
1822 | if (trace_empty(iter)) | |
1823 | return 0; | |
1824 | print_trace_header(m, iter); | |
1825 | if (!(trace_flags & TRACE_ITER_VERBOSE)) | |
1826 | print_lat_help_header(m); | |
1827 | } else { | |
1828 | if (!(trace_flags & TRACE_ITER_VERBOSE)) | |
1829 | print_func_help_header(m); | |
1830 | } | |
1831 | } else { | |
1832 | print_trace_line(iter); | |
1833 | trace_print_seq(m, &iter->seq); | |
1834 | } | |
1835 | ||
1836 | return 0; | |
1837 | } | |
1838 | ||
1839 | static struct seq_operations tracer_seq_ops = { | |
1840 | .start = s_start, | |
1841 | .next = s_next, | |
1842 | .stop = s_stop, | |
1843 | .show = s_show, | |
1844 | }; | |
1845 | ||
1846 | static struct trace_iterator * | |
1847 | __tracing_open(struct inode *inode, struct file *file) | |
1848 | { | |
1849 | long cpu_file = (long) inode->i_private; | |
1850 | void *fail_ret = ERR_PTR(-ENOMEM); | |
1851 | struct trace_iterator *iter; | |
1852 | struct seq_file *m; | |
1853 | int cpu, ret; | |
1854 | ||
1855 | if (tracing_disabled) | |
1856 | return ERR_PTR(-ENODEV); | |
1857 | ||
1858 | iter = kzalloc(sizeof(*iter), GFP_KERNEL); | |
1859 | if (!iter) | |
1860 | return ERR_PTR(-ENOMEM); | |
1861 | ||
1862 | /* | |
1863 | * We make a copy of the current tracer to avoid concurrent | |
1864 | * changes on it while we are reading. | |
1865 | */ | |
1866 | mutex_lock(&trace_types_lock); | |
1867 | iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL); | |
1868 | if (!iter->trace) | |
1869 | goto fail; | |
1870 | ||
1871 | if (current_trace) | |
1872 | *iter->trace = *current_trace; | |
1873 | ||
1874 | if (current_trace && current_trace->print_max) | |
1875 | iter->tr = &max_tr; | |
1876 | else | |
1877 | iter->tr = &global_trace; | |
1878 | iter->pos = -1; | |
1879 | mutex_init(&iter->mutex); | |
1880 | iter->cpu_file = cpu_file; | |
1881 | ||
1882 | /* Notify the tracer early; before we stop tracing. */ | |
1883 | if (iter->trace && iter->trace->open) | |
1884 | iter->trace->open(iter); | |
1885 | ||
1886 | /* Annotate start of buffers if we had overruns */ | |
1887 | if (ring_buffer_overruns(iter->tr->buffer)) | |
1888 | iter->iter_flags |= TRACE_FILE_ANNOTATE; | |
1889 | ||
1890 | if (iter->cpu_file == TRACE_PIPE_ALL_CPU) { | |
1891 | for_each_tracing_cpu(cpu) { | |
1892 | ||
1893 | iter->buffer_iter[cpu] = | |
1894 | ring_buffer_read_start(iter->tr->buffer, cpu); | |
1895 | } | |
1896 | } else { | |
1897 | cpu = iter->cpu_file; | |
1898 | iter->buffer_iter[cpu] = | |
1899 | ring_buffer_read_start(iter->tr->buffer, cpu); | |
1900 | } | |
1901 | ||
1902 | /* TODO stop tracer */ | |
1903 | ret = seq_open(file, &tracer_seq_ops); | |
1904 | if (ret < 0) { | |
1905 | fail_ret = ERR_PTR(ret); | |
1906 | goto fail_buffer; | |
1907 | } | |
1908 | ||
1909 | m = file->private_data; | |
1910 | m->private = iter; | |
1911 | ||
1912 | /* stop the trace while dumping */ | |
1913 | tracing_stop(); | |
1914 | ||
1915 | mutex_unlock(&trace_types_lock); | |
1916 | ||
1917 | return iter; | |
1918 | ||
1919 | fail_buffer: | |
1920 | for_each_tracing_cpu(cpu) { | |
1921 | if (iter->buffer_iter[cpu]) | |
1922 | ring_buffer_read_finish(iter->buffer_iter[cpu]); | |
1923 | } | |
1924 | fail: | |
1925 | mutex_unlock(&trace_types_lock); | |
1926 | kfree(iter->trace); | |
1927 | kfree(iter); | |
1928 | ||
1929 | return fail_ret; | |
1930 | } | |
1931 | ||
1932 | int tracing_open_generic(struct inode *inode, struct file *filp) | |
1933 | { | |
1934 | if (tracing_disabled) | |
1935 | return -ENODEV; | |
1936 | ||
1937 | filp->private_data = inode->i_private; | |
1938 | return 0; | |
1939 | } | |
1940 | ||
1941 | static int tracing_release(struct inode *inode, struct file *file) | |
1942 | { | |
1943 | struct seq_file *m = (struct seq_file *)file->private_data; | |
1944 | struct trace_iterator *iter; | |
1945 | int cpu; | |
1946 | ||
1947 | if (!(file->f_mode & FMODE_READ)) | |
1948 | return 0; | |
1949 | ||
1950 | iter = m->private; | |
1951 | ||
1952 | mutex_lock(&trace_types_lock); | |
1953 | for_each_tracing_cpu(cpu) { | |
1954 | if (iter->buffer_iter[cpu]) | |
1955 | ring_buffer_read_finish(iter->buffer_iter[cpu]); | |
1956 | } | |
1957 | ||
1958 | if (iter->trace && iter->trace->close) | |
1959 | iter->trace->close(iter); | |
1960 | ||
1961 | /* reenable tracing if it was previously enabled */ | |
1962 | tracing_start(); | |
1963 | mutex_unlock(&trace_types_lock); | |
1964 | ||
1965 | seq_release(inode, file); | |
1966 | mutex_destroy(&iter->mutex); | |
1967 | kfree(iter->trace); | |
1968 | kfree(iter); | |
1969 | return 0; | |
1970 | } | |
1971 | ||
1972 | static int tracing_open(struct inode *inode, struct file *file) | |
1973 | { | |
1974 | struct trace_iterator *iter; | |
1975 | int ret = 0; | |
1976 | ||
1977 | /* If this file was open for write, then erase contents */ | |
1978 | if ((file->f_mode & FMODE_WRITE) && | |
1979 | !(file->f_flags & O_APPEND)) { | |
1980 | long cpu = (long) inode->i_private; | |
1981 | ||
1982 | if (cpu == TRACE_PIPE_ALL_CPU) | |
1983 | tracing_reset_online_cpus(&global_trace); | |
1984 | else | |
1985 | tracing_reset(&global_trace, cpu); | |
1986 | } | |
1987 | ||
1988 | if (file->f_mode & FMODE_READ) { | |
1989 | iter = __tracing_open(inode, file); | |
1990 | if (IS_ERR(iter)) | |
1991 | ret = PTR_ERR(iter); | |
1992 | else if (trace_flags & TRACE_ITER_LATENCY_FMT) | |
1993 | iter->iter_flags |= TRACE_FILE_LAT_FMT; | |
1994 | } | |
1995 | return ret; | |
1996 | } | |
1997 | ||
1998 | static void * | |
1999 | t_next(struct seq_file *m, void *v, loff_t *pos) | |
2000 | { | |
2001 | struct tracer *t = m->private; | |
2002 | ||
2003 | (*pos)++; | |
2004 | ||
2005 | if (t) | |
2006 | t = t->next; | |
2007 | ||
2008 | m->private = t; | |
2009 | ||
2010 | return t; | |
2011 | } | |
2012 | ||
2013 | static void *t_start(struct seq_file *m, loff_t *pos) | |
2014 | { | |
2015 | struct tracer *t = m->private; | |
2016 | loff_t l = 0; | |
2017 | ||
2018 | mutex_lock(&trace_types_lock); | |
2019 | for (; t && l < *pos; t = t_next(m, t, &l)) | |
2020 | ; | |
2021 | ||
2022 | return t; | |
2023 | } | |
2024 | ||
2025 | static void t_stop(struct seq_file *m, void *p) | |
2026 | { | |
2027 | mutex_unlock(&trace_types_lock); | |
2028 | } | |
2029 | ||
2030 | static int t_show(struct seq_file *m, void *v) | |
2031 | { | |
2032 | struct tracer *t = v; | |
2033 | ||
2034 | if (!t) | |
2035 | return 0; | |
2036 | ||
2037 | seq_printf(m, "%s", t->name); | |
2038 | if (t->next) | |
2039 | seq_putc(m, ' '); | |
2040 | else | |
2041 | seq_putc(m, '\n'); | |
2042 | ||
2043 | return 0; | |
2044 | } | |
2045 | ||
2046 | static struct seq_operations show_traces_seq_ops = { | |
2047 | .start = t_start, | |
2048 | .next = t_next, | |
2049 | .stop = t_stop, | |
2050 | .show = t_show, | |
2051 | }; | |
2052 | ||
2053 | static int show_traces_open(struct inode *inode, struct file *file) | |
2054 | { | |
2055 | int ret; | |
2056 | ||
2057 | if (tracing_disabled) | |
2058 | return -ENODEV; | |
2059 | ||
2060 | ret = seq_open(file, &show_traces_seq_ops); | |
2061 | if (!ret) { | |
2062 | struct seq_file *m = file->private_data; | |
2063 | m->private = trace_types; | |
2064 | } | |
2065 | ||
2066 | return ret; | |
2067 | } | |
2068 | ||
2069 | static ssize_t | |
2070 | tracing_write_stub(struct file *filp, const char __user *ubuf, | |
2071 | size_t count, loff_t *ppos) | |
2072 | { | |
2073 | return count; | |
2074 | } | |
2075 | ||
2076 | static const struct file_operations tracing_fops = { | |
2077 | .open = tracing_open, | |
2078 | .read = seq_read, | |
2079 | .write = tracing_write_stub, | |
2080 | .llseek = seq_lseek, | |
2081 | .release = tracing_release, | |
2082 | }; | |
2083 | ||
2084 | static const struct file_operations show_traces_fops = { | |
2085 | .open = show_traces_open, | |
2086 | .read = seq_read, | |
2087 | .release = seq_release, | |
2088 | }; | |
2089 | ||
2090 | /* | |
2091 | * Only trace on a CPU if the bitmask is set: | |
2092 | */ | |
2093 | static cpumask_var_t tracing_cpumask; | |
2094 | ||
2095 | /* | |
2096 | * The tracer itself will not take this lock, but still we want | |
2097 | * to provide a consistent cpumask to user-space: | |
2098 | */ | |
2099 | static DEFINE_MUTEX(tracing_cpumask_update_lock); | |
2100 | ||
2101 | /* | |
2102 | * Temporary storage for the character representation of the | |
2103 | * CPU bitmask (and one more byte for the newline): | |
2104 | */ | |
2105 | static char mask_str[NR_CPUS + 1]; | |
2106 | ||
2107 | static ssize_t | |
2108 | tracing_cpumask_read(struct file *filp, char __user *ubuf, | |
2109 | size_t count, loff_t *ppos) | |
2110 | { | |
2111 | int len; | |
2112 | ||
2113 | mutex_lock(&tracing_cpumask_update_lock); | |
2114 | ||
2115 | len = cpumask_scnprintf(mask_str, count, tracing_cpumask); | |
2116 | if (count - len < 2) { | |
2117 | count = -EINVAL; | |
2118 | goto out_err; | |
2119 | } | |
2120 | len += sprintf(mask_str + len, "\n"); | |
2121 | count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1); | |
2122 | ||
2123 | out_err: | |
2124 | mutex_unlock(&tracing_cpumask_update_lock); | |
2125 | ||
2126 | return count; | |
2127 | } | |
2128 | ||
2129 | static ssize_t | |
2130 | tracing_cpumask_write(struct file *filp, const char __user *ubuf, | |
2131 | size_t count, loff_t *ppos) | |
2132 | { | |
2133 | int err, cpu; | |
2134 | cpumask_var_t tracing_cpumask_new; | |
2135 | ||
2136 | if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL)) | |
2137 | return -ENOMEM; | |
2138 | ||
2139 | mutex_lock(&tracing_cpumask_update_lock); | |
2140 | err = cpumask_parse_user(ubuf, count, tracing_cpumask_new); | |
2141 | if (err) | |
2142 | goto err_unlock; | |
2143 | ||
2144 | local_irq_disable(); | |
2145 | __raw_spin_lock(&ftrace_max_lock); | |
2146 | for_each_tracing_cpu(cpu) { | |
2147 | /* | |
2148 | * Increase/decrease the disabled counter if we are | |
2149 | * about to flip a bit in the cpumask: | |
2150 | */ | |
2151 | if (cpumask_test_cpu(cpu, tracing_cpumask) && | |
2152 | !cpumask_test_cpu(cpu, tracing_cpumask_new)) { | |
2153 | atomic_inc(&global_trace.data[cpu]->disabled); | |
2154 | } | |
2155 | if (!cpumask_test_cpu(cpu, tracing_cpumask) && | |
2156 | cpumask_test_cpu(cpu, tracing_cpumask_new)) { | |
2157 | atomic_dec(&global_trace.data[cpu]->disabled); | |
2158 | } | |
2159 | } | |
2160 | __raw_spin_unlock(&ftrace_max_lock); | |
2161 | local_irq_enable(); | |
2162 | ||
2163 | cpumask_copy(tracing_cpumask, tracing_cpumask_new); | |
2164 | ||
2165 | mutex_unlock(&tracing_cpumask_update_lock); | |
2166 | free_cpumask_var(tracing_cpumask_new); | |
2167 | ||
2168 | return count; | |
2169 | ||
2170 | err_unlock: | |
2171 | mutex_unlock(&tracing_cpumask_update_lock); | |
2172 | free_cpumask_var(tracing_cpumask); | |
2173 | ||
2174 | return err; | |
2175 | } | |
2176 | ||
2177 | static const struct file_operations tracing_cpumask_fops = { | |
2178 | .open = tracing_open_generic, | |
2179 | .read = tracing_cpumask_read, | |
2180 | .write = tracing_cpumask_write, | |
2181 | }; | |
2182 | ||
2183 | static ssize_t | |
2184 | tracing_trace_options_read(struct file *filp, char __user *ubuf, | |
2185 | size_t cnt, loff_t *ppos) | |
2186 | { | |
2187 | struct tracer_opt *trace_opts; | |
2188 | u32 tracer_flags; | |
2189 | int len = 0; | |
2190 | char *buf; | |
2191 | int r = 0; | |
2192 | int i; | |
2193 | ||
2194 | ||
2195 | /* calculate max size */ | |
2196 | for (i = 0; trace_options[i]; i++) { | |
2197 | len += strlen(trace_options[i]); | |
2198 | len += 3; /* "no" and newline */ | |
2199 | } | |
2200 | ||
2201 | mutex_lock(&trace_types_lock); | |
2202 | tracer_flags = current_trace->flags->val; | |
2203 | trace_opts = current_trace->flags->opts; | |
2204 | ||
2205 | /* | |
2206 | * Increase the size with names of options specific | |
2207 | * of the current tracer. | |
2208 | */ | |
2209 | for (i = 0; trace_opts[i].name; i++) { | |
2210 | len += strlen(trace_opts[i].name); | |
2211 | len += 3; /* "no" and newline */ | |
2212 | } | |
2213 | ||
2214 | /* +2 for \n and \0 */ | |
2215 | buf = kmalloc(len + 2, GFP_KERNEL); | |
2216 | if (!buf) { | |
2217 | mutex_unlock(&trace_types_lock); | |
2218 | return -ENOMEM; | |
2219 | } | |
2220 | ||
2221 | for (i = 0; trace_options[i]; i++) { | |
2222 | if (trace_flags & (1 << i)) | |
2223 | r += sprintf(buf + r, "%s\n", trace_options[i]); | |
2224 | else | |
2225 | r += sprintf(buf + r, "no%s\n", trace_options[i]); | |
2226 | } | |
2227 | ||
2228 | for (i = 0; trace_opts[i].name; i++) { | |
2229 | if (tracer_flags & trace_opts[i].bit) | |
2230 | r += sprintf(buf + r, "%s\n", | |
2231 | trace_opts[i].name); | |
2232 | else | |
2233 | r += sprintf(buf + r, "no%s\n", | |
2234 | trace_opts[i].name); | |
2235 | } | |
2236 | mutex_unlock(&trace_types_lock); | |
2237 | ||
2238 | WARN_ON(r >= len + 2); | |
2239 | ||
2240 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
2241 | ||
2242 | kfree(buf); | |
2243 | return r; | |
2244 | } | |
2245 | ||
2246 | /* Try to assign a tracer specific option */ | |
2247 | static int set_tracer_option(struct tracer *trace, char *cmp, int neg) | |
2248 | { | |
2249 | struct tracer_flags *trace_flags = trace->flags; | |
2250 | struct tracer_opt *opts = NULL; | |
2251 | int ret = 0, i = 0; | |
2252 | int len; | |
2253 | ||
2254 | for (i = 0; trace_flags->opts[i].name; i++) { | |
2255 | opts = &trace_flags->opts[i]; | |
2256 | len = strlen(opts->name); | |
2257 | ||
2258 | if (strncmp(cmp, opts->name, len) == 0) { | |
2259 | ret = trace->set_flag(trace_flags->val, | |
2260 | opts->bit, !neg); | |
2261 | break; | |
2262 | } | |
2263 | } | |
2264 | /* Not found */ | |
2265 | if (!trace_flags->opts[i].name) | |
2266 | return -EINVAL; | |
2267 | ||
2268 | /* Refused to handle */ | |
2269 | if (ret) | |
2270 | return ret; | |
2271 | ||
2272 | if (neg) | |
2273 | trace_flags->val &= ~opts->bit; | |
2274 | else | |
2275 | trace_flags->val |= opts->bit; | |
2276 | ||
2277 | return 0; | |
2278 | } | |
2279 | ||
2280 | static void set_tracer_flags(unsigned int mask, int enabled) | |
2281 | { | |
2282 | /* do nothing if flag is already set */ | |
2283 | if (!!(trace_flags & mask) == !!enabled) | |
2284 | return; | |
2285 | ||
2286 | if (enabled) | |
2287 | trace_flags |= mask; | |
2288 | else | |
2289 | trace_flags &= ~mask; | |
2290 | ||
2291 | if (mask == TRACE_ITER_GLOBAL_CLK) { | |
2292 | u64 (*func)(void); | |
2293 | ||
2294 | if (enabled) | |
2295 | func = trace_clock_global; | |
2296 | else | |
2297 | func = trace_clock_local; | |
2298 | ||
2299 | mutex_lock(&trace_types_lock); | |
2300 | ring_buffer_set_clock(global_trace.buffer, func); | |
2301 | ||
2302 | if (max_tr.buffer) | |
2303 | ring_buffer_set_clock(max_tr.buffer, func); | |
2304 | mutex_unlock(&trace_types_lock); | |
2305 | } | |
2306 | } | |
2307 | ||
2308 | static ssize_t | |
2309 | tracing_trace_options_write(struct file *filp, const char __user *ubuf, | |
2310 | size_t cnt, loff_t *ppos) | |
2311 | { | |
2312 | char buf[64]; | |
2313 | char *cmp = buf; | |
2314 | int neg = 0; | |
2315 | int ret; | |
2316 | int i; | |
2317 | ||
2318 | if (cnt >= sizeof(buf)) | |
2319 | return -EINVAL; | |
2320 | ||
2321 | if (copy_from_user(&buf, ubuf, cnt)) | |
2322 | return -EFAULT; | |
2323 | ||
2324 | buf[cnt] = 0; | |
2325 | ||
2326 | if (strncmp(buf, "no", 2) == 0) { | |
2327 | neg = 1; | |
2328 | cmp += 2; | |
2329 | } | |
2330 | ||
2331 | for (i = 0; trace_options[i]; i++) { | |
2332 | int len = strlen(trace_options[i]); | |
2333 | ||
2334 | if (strncmp(cmp, trace_options[i], len) == 0) { | |
2335 | set_tracer_flags(1 << i, !neg); | |
2336 | break; | |
2337 | } | |
2338 | } | |
2339 | ||
2340 | /* If no option could be set, test the specific tracer options */ | |
2341 | if (!trace_options[i]) { | |
2342 | mutex_lock(&trace_types_lock); | |
2343 | ret = set_tracer_option(current_trace, cmp, neg); | |
2344 | mutex_unlock(&trace_types_lock); | |
2345 | if (ret) | |
2346 | return ret; | |
2347 | } | |
2348 | ||
2349 | filp->f_pos += cnt; | |
2350 | ||
2351 | return cnt; | |
2352 | } | |
2353 | ||
2354 | static const struct file_operations tracing_iter_fops = { | |
2355 | .open = tracing_open_generic, | |
2356 | .read = tracing_trace_options_read, | |
2357 | .write = tracing_trace_options_write, | |
2358 | }; | |
2359 | ||
2360 | static const char readme_msg[] = | |
2361 | "tracing mini-HOWTO:\n\n" | |
2362 | "# mkdir /debug\n" | |
2363 | "# mount -t debugfs nodev /debug\n\n" | |
2364 | "# cat /debug/tracing/available_tracers\n" | |
2365 | "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n" | |
2366 | "# cat /debug/tracing/current_tracer\n" | |
2367 | "none\n" | |
2368 | "# echo sched_switch > /debug/tracing/current_tracer\n" | |
2369 | "# cat /debug/tracing/current_tracer\n" | |
2370 | "sched_switch\n" | |
2371 | "# cat /debug/tracing/trace_options\n" | |
2372 | "noprint-parent nosym-offset nosym-addr noverbose\n" | |
2373 | "# echo print-parent > /debug/tracing/trace_options\n" | |
2374 | "# echo 1 > /debug/tracing/tracing_enabled\n" | |
2375 | "# cat /debug/tracing/trace > /tmp/trace.txt\n" | |
2376 | "echo 0 > /debug/tracing/tracing_enabled\n" | |
2377 | ; | |
2378 | ||
2379 | static ssize_t | |
2380 | tracing_readme_read(struct file *filp, char __user *ubuf, | |
2381 | size_t cnt, loff_t *ppos) | |
2382 | { | |
2383 | return simple_read_from_buffer(ubuf, cnt, ppos, | |
2384 | readme_msg, strlen(readme_msg)); | |
2385 | } | |
2386 | ||
2387 | static const struct file_operations tracing_readme_fops = { | |
2388 | .open = tracing_open_generic, | |
2389 | .read = tracing_readme_read, | |
2390 | }; | |
2391 | ||
2392 | static ssize_t | |
2393 | tracing_ctrl_read(struct file *filp, char __user *ubuf, | |
2394 | size_t cnt, loff_t *ppos) | |
2395 | { | |
2396 | char buf[64]; | |
2397 | int r; | |
2398 | ||
2399 | r = sprintf(buf, "%u\n", tracer_enabled); | |
2400 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
2401 | } | |
2402 | ||
2403 | static ssize_t | |
2404 | tracing_ctrl_write(struct file *filp, const char __user *ubuf, | |
2405 | size_t cnt, loff_t *ppos) | |
2406 | { | |
2407 | struct trace_array *tr = filp->private_data; | |
2408 | char buf[64]; | |
2409 | unsigned long val; | |
2410 | int ret; | |
2411 | ||
2412 | if (cnt >= sizeof(buf)) | |
2413 | return -EINVAL; | |
2414 | ||
2415 | if (copy_from_user(&buf, ubuf, cnt)) | |
2416 | return -EFAULT; | |
2417 | ||
2418 | buf[cnt] = 0; | |
2419 | ||
2420 | ret = strict_strtoul(buf, 10, &val); | |
2421 | if (ret < 0) | |
2422 | return ret; | |
2423 | ||
2424 | val = !!val; | |
2425 | ||
2426 | mutex_lock(&trace_types_lock); | |
2427 | if (tracer_enabled ^ val) { | |
2428 | if (val) { | |
2429 | tracer_enabled = 1; | |
2430 | if (current_trace->start) | |
2431 | current_trace->start(tr); | |
2432 | tracing_start(); | |
2433 | } else { | |
2434 | tracer_enabled = 0; | |
2435 | tracing_stop(); | |
2436 | if (current_trace->stop) | |
2437 | current_trace->stop(tr); | |
2438 | } | |
2439 | } | |
2440 | mutex_unlock(&trace_types_lock); | |
2441 | ||
2442 | filp->f_pos += cnt; | |
2443 | ||
2444 | return cnt; | |
2445 | } | |
2446 | ||
2447 | static ssize_t | |
2448 | tracing_set_trace_read(struct file *filp, char __user *ubuf, | |
2449 | size_t cnt, loff_t *ppos) | |
2450 | { | |
2451 | char buf[max_tracer_type_len+2]; | |
2452 | int r; | |
2453 | ||
2454 | mutex_lock(&trace_types_lock); | |
2455 | if (current_trace) | |
2456 | r = sprintf(buf, "%s\n", current_trace->name); | |
2457 | else | |
2458 | r = sprintf(buf, "\n"); | |
2459 | mutex_unlock(&trace_types_lock); | |
2460 | ||
2461 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
2462 | } | |
2463 | ||
2464 | int tracer_init(struct tracer *t, struct trace_array *tr) | |
2465 | { | |
2466 | tracing_reset_online_cpus(tr); | |
2467 | return t->init(tr); | |
2468 | } | |
2469 | ||
2470 | static int tracing_resize_ring_buffer(unsigned long size) | |
2471 | { | |
2472 | int ret; | |
2473 | ||
2474 | /* | |
2475 | * If kernel or user changes the size of the ring buffer | |
2476 | * we use the size that was given, and we can forget about | |
2477 | * expanding it later. | |
2478 | */ | |
2479 | ring_buffer_expanded = 1; | |
2480 | ||
2481 | ret = ring_buffer_resize(global_trace.buffer, size); | |
2482 | if (ret < 0) | |
2483 | return ret; | |
2484 | ||
2485 | ret = ring_buffer_resize(max_tr.buffer, size); | |
2486 | if (ret < 0) { | |
2487 | int r; | |
2488 | ||
2489 | r = ring_buffer_resize(global_trace.buffer, | |
2490 | global_trace.entries); | |
2491 | if (r < 0) { | |
2492 | /* | |
2493 | * AARGH! We are left with different | |
2494 | * size max buffer!!!! | |
2495 | * The max buffer is our "snapshot" buffer. | |
2496 | * When a tracer needs a snapshot (one of the | |
2497 | * latency tracers), it swaps the max buffer | |
2498 | * with the saved snap shot. We succeeded to | |
2499 | * update the size of the main buffer, but failed to | |
2500 | * update the size of the max buffer. But when we tried | |
2501 | * to reset the main buffer to the original size, we | |
2502 | * failed there too. This is very unlikely to | |
2503 | * happen, but if it does, warn and kill all | |
2504 | * tracing. | |
2505 | */ | |
2506 | WARN_ON(1); | |
2507 | tracing_disabled = 1; | |
2508 | } | |
2509 | return ret; | |
2510 | } | |
2511 | ||
2512 | global_trace.entries = size; | |
2513 | ||
2514 | return ret; | |
2515 | } | |
2516 | ||
2517 | /** | |
2518 | * tracing_update_buffers - used by tracing facility to expand ring buffers | |
2519 | * | |
2520 | * To save on memory when the tracing is never used on a system with it | |
2521 | * configured in. The ring buffers are set to a minimum size. But once | |
2522 | * a user starts to use the tracing facility, then they need to grow | |
2523 | * to their default size. | |
2524 | * | |
2525 | * This function is to be called when a tracer is about to be used. | |
2526 | */ | |
2527 | int tracing_update_buffers(void) | |
2528 | { | |
2529 | int ret = 0; | |
2530 | ||
2531 | mutex_lock(&trace_types_lock); | |
2532 | if (!ring_buffer_expanded) | |
2533 | ret = tracing_resize_ring_buffer(trace_buf_size); | |
2534 | mutex_unlock(&trace_types_lock); | |
2535 | ||
2536 | return ret; | |
2537 | } | |
2538 | ||
2539 | struct trace_option_dentry; | |
2540 | ||
2541 | static struct trace_option_dentry * | |
2542 | create_trace_option_files(struct tracer *tracer); | |
2543 | ||
2544 | static void | |
2545 | destroy_trace_option_files(struct trace_option_dentry *topts); | |
2546 | ||
2547 | static int tracing_set_tracer(const char *buf) | |
2548 | { | |
2549 | static struct trace_option_dentry *topts; | |
2550 | struct trace_array *tr = &global_trace; | |
2551 | struct tracer *t; | |
2552 | int ret = 0; | |
2553 | ||
2554 | mutex_lock(&trace_types_lock); | |
2555 | ||
2556 | if (!ring_buffer_expanded) { | |
2557 | ret = tracing_resize_ring_buffer(trace_buf_size); | |
2558 | if (ret < 0) | |
2559 | goto out; | |
2560 | ret = 0; | |
2561 | } | |
2562 | ||
2563 | for (t = trace_types; t; t = t->next) { | |
2564 | if (strcmp(t->name, buf) == 0) | |
2565 | break; | |
2566 | } | |
2567 | if (!t) { | |
2568 | ret = -EINVAL; | |
2569 | goto out; | |
2570 | } | |
2571 | if (t == current_trace) | |
2572 | goto out; | |
2573 | ||
2574 | trace_branch_disable(); | |
2575 | if (current_trace && current_trace->reset) | |
2576 | current_trace->reset(tr); | |
2577 | ||
2578 | destroy_trace_option_files(topts); | |
2579 | ||
2580 | current_trace = t; | |
2581 | ||
2582 | topts = create_trace_option_files(current_trace); | |
2583 | ||
2584 | if (t->init) { | |
2585 | ret = tracer_init(t, tr); | |
2586 | if (ret) | |
2587 | goto out; | |
2588 | } | |
2589 | ||
2590 | trace_branch_enable(tr); | |
2591 | out: | |
2592 | mutex_unlock(&trace_types_lock); | |
2593 | ||
2594 | return ret; | |
2595 | } | |
2596 | ||
2597 | static ssize_t | |
2598 | tracing_set_trace_write(struct file *filp, const char __user *ubuf, | |
2599 | size_t cnt, loff_t *ppos) | |
2600 | { | |
2601 | char buf[max_tracer_type_len+1]; | |
2602 | int i; | |
2603 | size_t ret; | |
2604 | int err; | |
2605 | ||
2606 | ret = cnt; | |
2607 | ||
2608 | if (cnt > max_tracer_type_len) | |
2609 | cnt = max_tracer_type_len; | |
2610 | ||
2611 | if (copy_from_user(&buf, ubuf, cnt)) | |
2612 | return -EFAULT; | |
2613 | ||
2614 | buf[cnt] = 0; | |
2615 | ||
2616 | /* strip ending whitespace. */ | |
2617 | for (i = cnt - 1; i > 0 && isspace(buf[i]); i--) | |
2618 | buf[i] = 0; | |
2619 | ||
2620 | err = tracing_set_tracer(buf); | |
2621 | if (err) | |
2622 | return err; | |
2623 | ||
2624 | filp->f_pos += ret; | |
2625 | ||
2626 | return ret; | |
2627 | } | |
2628 | ||
2629 | static ssize_t | |
2630 | tracing_max_lat_read(struct file *filp, char __user *ubuf, | |
2631 | size_t cnt, loff_t *ppos) | |
2632 | { | |
2633 | unsigned long *ptr = filp->private_data; | |
2634 | char buf[64]; | |
2635 | int r; | |
2636 | ||
2637 | r = snprintf(buf, sizeof(buf), "%ld\n", | |
2638 | *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr)); | |
2639 | if (r > sizeof(buf)) | |
2640 | r = sizeof(buf); | |
2641 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
2642 | } | |
2643 | ||
2644 | static ssize_t | |
2645 | tracing_max_lat_write(struct file *filp, const char __user *ubuf, | |
2646 | size_t cnt, loff_t *ppos) | |
2647 | { | |
2648 | unsigned long *ptr = filp->private_data; | |
2649 | char buf[64]; | |
2650 | unsigned long val; | |
2651 | int ret; | |
2652 | ||
2653 | if (cnt >= sizeof(buf)) | |
2654 | return -EINVAL; | |
2655 | ||
2656 | if (copy_from_user(&buf, ubuf, cnt)) | |
2657 | return -EFAULT; | |
2658 | ||
2659 | buf[cnt] = 0; | |
2660 | ||
2661 | ret = strict_strtoul(buf, 10, &val); | |
2662 | if (ret < 0) | |
2663 | return ret; | |
2664 | ||
2665 | *ptr = val * 1000; | |
2666 | ||
2667 | return cnt; | |
2668 | } | |
2669 | ||
2670 | static int tracing_open_pipe(struct inode *inode, struct file *filp) | |
2671 | { | |
2672 | long cpu_file = (long) inode->i_private; | |
2673 | struct trace_iterator *iter; | |
2674 | int ret = 0; | |
2675 | ||
2676 | if (tracing_disabled) | |
2677 | return -ENODEV; | |
2678 | ||
2679 | mutex_lock(&trace_types_lock); | |
2680 | ||
2681 | /* We only allow one reader per cpu */ | |
2682 | if (cpu_file == TRACE_PIPE_ALL_CPU) { | |
2683 | if (!cpumask_empty(tracing_reader_cpumask)) { | |
2684 | ret = -EBUSY; | |
2685 | goto out; | |
2686 | } | |
2687 | cpumask_setall(tracing_reader_cpumask); | |
2688 | } else { | |
2689 | if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask)) | |
2690 | cpumask_set_cpu(cpu_file, tracing_reader_cpumask); | |
2691 | else { | |
2692 | ret = -EBUSY; | |
2693 | goto out; | |
2694 | } | |
2695 | } | |
2696 | ||
2697 | /* create a buffer to store the information to pass to userspace */ | |
2698 | iter = kzalloc(sizeof(*iter), GFP_KERNEL); | |
2699 | if (!iter) { | |
2700 | ret = -ENOMEM; | |
2701 | goto out; | |
2702 | } | |
2703 | ||
2704 | /* | |
2705 | * We make a copy of the current tracer to avoid concurrent | |
2706 | * changes on it while we are reading. | |
2707 | */ | |
2708 | iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL); | |
2709 | if (!iter->trace) { | |
2710 | ret = -ENOMEM; | |
2711 | goto fail; | |
2712 | } | |
2713 | if (current_trace) | |
2714 | *iter->trace = *current_trace; | |
2715 | ||
2716 | if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) { | |
2717 | ret = -ENOMEM; | |
2718 | goto fail; | |
2719 | } | |
2720 | ||
2721 | /* trace pipe does not show start of buffer */ | |
2722 | cpumask_setall(iter->started); | |
2723 | ||
2724 | iter->cpu_file = cpu_file; | |
2725 | iter->tr = &global_trace; | |
2726 | mutex_init(&iter->mutex); | |
2727 | filp->private_data = iter; | |
2728 | ||
2729 | if (iter->trace->pipe_open) | |
2730 | iter->trace->pipe_open(iter); | |
2731 | ||
2732 | out: | |
2733 | mutex_unlock(&trace_types_lock); | |
2734 | return ret; | |
2735 | ||
2736 | fail: | |
2737 | kfree(iter->trace); | |
2738 | kfree(iter); | |
2739 | mutex_unlock(&trace_types_lock); | |
2740 | return ret; | |
2741 | } | |
2742 | ||
2743 | static int tracing_release_pipe(struct inode *inode, struct file *file) | |
2744 | { | |
2745 | struct trace_iterator *iter = file->private_data; | |
2746 | ||
2747 | mutex_lock(&trace_types_lock); | |
2748 | ||
2749 | if (iter->cpu_file == TRACE_PIPE_ALL_CPU) | |
2750 | cpumask_clear(tracing_reader_cpumask); | |
2751 | else | |
2752 | cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask); | |
2753 | ||
2754 | mutex_unlock(&trace_types_lock); | |
2755 | ||
2756 | free_cpumask_var(iter->started); | |
2757 | mutex_destroy(&iter->mutex); | |
2758 | kfree(iter->trace); | |
2759 | kfree(iter); | |
2760 | ||
2761 | return 0; | |
2762 | } | |
2763 | ||
2764 | static unsigned int | |
2765 | tracing_poll_pipe(struct file *filp, poll_table *poll_table) | |
2766 | { | |
2767 | struct trace_iterator *iter = filp->private_data; | |
2768 | ||
2769 | if (trace_flags & TRACE_ITER_BLOCK) { | |
2770 | /* | |
2771 | * Always select as readable when in blocking mode | |
2772 | */ | |
2773 | return POLLIN | POLLRDNORM; | |
2774 | } else { | |
2775 | if (!trace_empty(iter)) | |
2776 | return POLLIN | POLLRDNORM; | |
2777 | poll_wait(filp, &trace_wait, poll_table); | |
2778 | if (!trace_empty(iter)) | |
2779 | return POLLIN | POLLRDNORM; | |
2780 | ||
2781 | return 0; | |
2782 | } | |
2783 | } | |
2784 | ||
2785 | ||
2786 | void default_wait_pipe(struct trace_iterator *iter) | |
2787 | { | |
2788 | DEFINE_WAIT(wait); | |
2789 | ||
2790 | prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE); | |
2791 | ||
2792 | if (trace_empty(iter)) | |
2793 | schedule(); | |
2794 | ||
2795 | finish_wait(&trace_wait, &wait); | |
2796 | } | |
2797 | ||
2798 | /* | |
2799 | * This is a make-shift waitqueue. | |
2800 | * A tracer might use this callback on some rare cases: | |
2801 | * | |
2802 | * 1) the current tracer might hold the runqueue lock when it wakes up | |
2803 | * a reader, hence a deadlock (sched, function, and function graph tracers) | |
2804 | * 2) the function tracers, trace all functions, we don't want | |
2805 | * the overhead of calling wake_up and friends | |
2806 | * (and tracing them too) | |
2807 | * | |
2808 | * Anyway, this is really very primitive wakeup. | |
2809 | */ | |
2810 | void poll_wait_pipe(struct trace_iterator *iter) | |
2811 | { | |
2812 | set_current_state(TASK_INTERRUPTIBLE); | |
2813 | /* sleep for 100 msecs, and try again. */ | |
2814 | schedule_timeout(HZ / 10); | |
2815 | } | |
2816 | ||
2817 | /* Must be called with trace_types_lock mutex held. */ | |
2818 | static int tracing_wait_pipe(struct file *filp) | |
2819 | { | |
2820 | struct trace_iterator *iter = filp->private_data; | |
2821 | ||
2822 | while (trace_empty(iter)) { | |
2823 | ||
2824 | if ((filp->f_flags & O_NONBLOCK)) { | |
2825 | return -EAGAIN; | |
2826 | } | |
2827 | ||
2828 | mutex_unlock(&iter->mutex); | |
2829 | ||
2830 | iter->trace->wait_pipe(iter); | |
2831 | ||
2832 | mutex_lock(&iter->mutex); | |
2833 | ||
2834 | if (signal_pending(current)) | |
2835 | return -EINTR; | |
2836 | ||
2837 | /* | |
2838 | * We block until we read something and tracing is disabled. | |
2839 | * We still block if tracing is disabled, but we have never | |
2840 | * read anything. This allows a user to cat this file, and | |
2841 | * then enable tracing. But after we have read something, | |
2842 | * we give an EOF when tracing is again disabled. | |
2843 | * | |
2844 | * iter->pos will be 0 if we haven't read anything. | |
2845 | */ | |
2846 | if (!tracer_enabled && iter->pos) | |
2847 | break; | |
2848 | } | |
2849 | ||
2850 | return 1; | |
2851 | } | |
2852 | ||
2853 | /* | |
2854 | * Consumer reader. | |
2855 | */ | |
2856 | static ssize_t | |
2857 | tracing_read_pipe(struct file *filp, char __user *ubuf, | |
2858 | size_t cnt, loff_t *ppos) | |
2859 | { | |
2860 | struct trace_iterator *iter = filp->private_data; | |
2861 | static struct tracer *old_tracer; | |
2862 | ssize_t sret; | |
2863 | ||
2864 | /* return any leftover data */ | |
2865 | sret = trace_seq_to_user(&iter->seq, ubuf, cnt); | |
2866 | if (sret != -EBUSY) | |
2867 | return sret; | |
2868 | ||
2869 | trace_seq_init(&iter->seq); | |
2870 | ||
2871 | /* copy the tracer to avoid using a global lock all around */ | |
2872 | mutex_lock(&trace_types_lock); | |
2873 | if (unlikely(old_tracer != current_trace && current_trace)) { | |
2874 | old_tracer = current_trace; | |
2875 | *iter->trace = *current_trace; | |
2876 | } | |
2877 | mutex_unlock(&trace_types_lock); | |
2878 | ||
2879 | /* | |
2880 | * Avoid more than one consumer on a single file descriptor | |
2881 | * This is just a matter of traces coherency, the ring buffer itself | |
2882 | * is protected. | |
2883 | */ | |
2884 | mutex_lock(&iter->mutex); | |
2885 | if (iter->trace->read) { | |
2886 | sret = iter->trace->read(iter, filp, ubuf, cnt, ppos); | |
2887 | if (sret) | |
2888 | goto out; | |
2889 | } | |
2890 | ||
2891 | waitagain: | |
2892 | sret = tracing_wait_pipe(filp); | |
2893 | if (sret <= 0) | |
2894 | goto out; | |
2895 | ||
2896 | /* stop when tracing is finished */ | |
2897 | if (trace_empty(iter)) { | |
2898 | sret = 0; | |
2899 | goto out; | |
2900 | } | |
2901 | ||
2902 | if (cnt >= PAGE_SIZE) | |
2903 | cnt = PAGE_SIZE - 1; | |
2904 | ||
2905 | /* reset all but tr, trace, and overruns */ | |
2906 | memset(&iter->seq, 0, | |
2907 | sizeof(struct trace_iterator) - | |
2908 | offsetof(struct trace_iterator, seq)); | |
2909 | iter->pos = -1; | |
2910 | ||
2911 | while (find_next_entry_inc(iter) != NULL) { | |
2912 | enum print_line_t ret; | |
2913 | int len = iter->seq.len; | |
2914 | ||
2915 | ret = print_trace_line(iter); | |
2916 | if (ret == TRACE_TYPE_PARTIAL_LINE) { | |
2917 | /* don't print partial lines */ | |
2918 | iter->seq.len = len; | |
2919 | break; | |
2920 | } | |
2921 | if (ret != TRACE_TYPE_NO_CONSUME) | |
2922 | trace_consume(iter); | |
2923 | ||
2924 | if (iter->seq.len >= cnt) | |
2925 | break; | |
2926 | } | |
2927 | ||
2928 | /* Now copy what we have to the user */ | |
2929 | sret = trace_seq_to_user(&iter->seq, ubuf, cnt); | |
2930 | if (iter->seq.readpos >= iter->seq.len) | |
2931 | trace_seq_init(&iter->seq); | |
2932 | ||
2933 | /* | |
2934 | * If there was nothing to send to user, inspite of consuming trace | |
2935 | * entries, go back to wait for more entries. | |
2936 | */ | |
2937 | if (sret == -EBUSY) | |
2938 | goto waitagain; | |
2939 | ||
2940 | out: | |
2941 | mutex_unlock(&iter->mutex); | |
2942 | ||
2943 | return sret; | |
2944 | } | |
2945 | ||
2946 | static void tracing_pipe_buf_release(struct pipe_inode_info *pipe, | |
2947 | struct pipe_buffer *buf) | |
2948 | { | |
2949 | __free_page(buf->page); | |
2950 | } | |
2951 | ||
2952 | static void tracing_spd_release_pipe(struct splice_pipe_desc *spd, | |
2953 | unsigned int idx) | |
2954 | { | |
2955 | __free_page(spd->pages[idx]); | |
2956 | } | |
2957 | ||
2958 | static struct pipe_buf_operations tracing_pipe_buf_ops = { | |
2959 | .can_merge = 0, | |
2960 | .map = generic_pipe_buf_map, | |
2961 | .unmap = generic_pipe_buf_unmap, | |
2962 | .confirm = generic_pipe_buf_confirm, | |
2963 | .release = tracing_pipe_buf_release, | |
2964 | .steal = generic_pipe_buf_steal, | |
2965 | .get = generic_pipe_buf_get, | |
2966 | }; | |
2967 | ||
2968 | static size_t | |
2969 | tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter) | |
2970 | { | |
2971 | size_t count; | |
2972 | int ret; | |
2973 | ||
2974 | /* Seq buffer is page-sized, exactly what we need. */ | |
2975 | for (;;) { | |
2976 | count = iter->seq.len; | |
2977 | ret = print_trace_line(iter); | |
2978 | count = iter->seq.len - count; | |
2979 | if (rem < count) { | |
2980 | rem = 0; | |
2981 | iter->seq.len -= count; | |
2982 | break; | |
2983 | } | |
2984 | if (ret == TRACE_TYPE_PARTIAL_LINE) { | |
2985 | iter->seq.len -= count; | |
2986 | break; | |
2987 | } | |
2988 | ||
2989 | trace_consume(iter); | |
2990 | rem -= count; | |
2991 | if (!find_next_entry_inc(iter)) { | |
2992 | rem = 0; | |
2993 | iter->ent = NULL; | |
2994 | break; | |
2995 | } | |
2996 | } | |
2997 | ||
2998 | return rem; | |
2999 | } | |
3000 | ||
3001 | static ssize_t tracing_splice_read_pipe(struct file *filp, | |
3002 | loff_t *ppos, | |
3003 | struct pipe_inode_info *pipe, | |
3004 | size_t len, | |
3005 | unsigned int flags) | |
3006 | { | |
3007 | struct page *pages[PIPE_BUFFERS]; | |
3008 | struct partial_page partial[PIPE_BUFFERS]; | |
3009 | struct trace_iterator *iter = filp->private_data; | |
3010 | struct splice_pipe_desc spd = { | |
3011 | .pages = pages, | |
3012 | .partial = partial, | |
3013 | .nr_pages = 0, /* This gets updated below. */ | |
3014 | .flags = flags, | |
3015 | .ops = &tracing_pipe_buf_ops, | |
3016 | .spd_release = tracing_spd_release_pipe, | |
3017 | }; | |
3018 | static struct tracer *old_tracer; | |
3019 | ssize_t ret; | |
3020 | size_t rem; | |
3021 | unsigned int i; | |
3022 | ||
3023 | /* copy the tracer to avoid using a global lock all around */ | |
3024 | mutex_lock(&trace_types_lock); | |
3025 | if (unlikely(old_tracer != current_trace && current_trace)) { | |
3026 | old_tracer = current_trace; | |
3027 | *iter->trace = *current_trace; | |
3028 | } | |
3029 | mutex_unlock(&trace_types_lock); | |
3030 | ||
3031 | mutex_lock(&iter->mutex); | |
3032 | ||
3033 | if (iter->trace->splice_read) { | |
3034 | ret = iter->trace->splice_read(iter, filp, | |
3035 | ppos, pipe, len, flags); | |
3036 | if (ret) | |
3037 | goto out_err; | |
3038 | } | |
3039 | ||
3040 | ret = tracing_wait_pipe(filp); | |
3041 | if (ret <= 0) | |
3042 | goto out_err; | |
3043 | ||
3044 | if (!iter->ent && !find_next_entry_inc(iter)) { | |
3045 | ret = -EFAULT; | |
3046 | goto out_err; | |
3047 | } | |
3048 | ||
3049 | /* Fill as many pages as possible. */ | |
3050 | for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) { | |
3051 | pages[i] = alloc_page(GFP_KERNEL); | |
3052 | if (!pages[i]) | |
3053 | break; | |
3054 | ||
3055 | rem = tracing_fill_pipe_page(rem, iter); | |
3056 | ||
3057 | /* Copy the data into the page, so we can start over. */ | |
3058 | ret = trace_seq_to_buffer(&iter->seq, | |
3059 | page_address(pages[i]), | |
3060 | iter->seq.len); | |
3061 | if (ret < 0) { | |
3062 | __free_page(pages[i]); | |
3063 | break; | |
3064 | } | |
3065 | partial[i].offset = 0; | |
3066 | partial[i].len = iter->seq.len; | |
3067 | ||
3068 | trace_seq_init(&iter->seq); | |
3069 | } | |
3070 | ||
3071 | mutex_unlock(&iter->mutex); | |
3072 | ||
3073 | spd.nr_pages = i; | |
3074 | ||
3075 | return splice_to_pipe(pipe, &spd); | |
3076 | ||
3077 | out_err: | |
3078 | mutex_unlock(&iter->mutex); | |
3079 | ||
3080 | return ret; | |
3081 | } | |
3082 | ||
3083 | static ssize_t | |
3084 | tracing_entries_read(struct file *filp, char __user *ubuf, | |
3085 | size_t cnt, loff_t *ppos) | |
3086 | { | |
3087 | struct trace_array *tr = filp->private_data; | |
3088 | char buf[96]; | |
3089 | int r; | |
3090 | ||
3091 | mutex_lock(&trace_types_lock); | |
3092 | if (!ring_buffer_expanded) | |
3093 | r = sprintf(buf, "%lu (expanded: %lu)\n", | |
3094 | tr->entries >> 10, | |
3095 | trace_buf_size >> 10); | |
3096 | else | |
3097 | r = sprintf(buf, "%lu\n", tr->entries >> 10); | |
3098 | mutex_unlock(&trace_types_lock); | |
3099 | ||
3100 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3101 | } | |
3102 | ||
3103 | static ssize_t | |
3104 | tracing_entries_write(struct file *filp, const char __user *ubuf, | |
3105 | size_t cnt, loff_t *ppos) | |
3106 | { | |
3107 | unsigned long val; | |
3108 | char buf[64]; | |
3109 | int ret, cpu; | |
3110 | ||
3111 | if (cnt >= sizeof(buf)) | |
3112 | return -EINVAL; | |
3113 | ||
3114 | if (copy_from_user(&buf, ubuf, cnt)) | |
3115 | return -EFAULT; | |
3116 | ||
3117 | buf[cnt] = 0; | |
3118 | ||
3119 | ret = strict_strtoul(buf, 10, &val); | |
3120 | if (ret < 0) | |
3121 | return ret; | |
3122 | ||
3123 | /* must have at least 1 entry */ | |
3124 | if (!val) | |
3125 | return -EINVAL; | |
3126 | ||
3127 | mutex_lock(&trace_types_lock); | |
3128 | ||
3129 | tracing_stop(); | |
3130 | ||
3131 | /* disable all cpu buffers */ | |
3132 | for_each_tracing_cpu(cpu) { | |
3133 | if (global_trace.data[cpu]) | |
3134 | atomic_inc(&global_trace.data[cpu]->disabled); | |
3135 | if (max_tr.data[cpu]) | |
3136 | atomic_inc(&max_tr.data[cpu]->disabled); | |
3137 | } | |
3138 | ||
3139 | /* value is in KB */ | |
3140 | val <<= 10; | |
3141 | ||
3142 | if (val != global_trace.entries) { | |
3143 | ret = tracing_resize_ring_buffer(val); | |
3144 | if (ret < 0) { | |
3145 | cnt = ret; | |
3146 | goto out; | |
3147 | } | |
3148 | } | |
3149 | ||
3150 | filp->f_pos += cnt; | |
3151 | ||
3152 | /* If check pages failed, return ENOMEM */ | |
3153 | if (tracing_disabled) | |
3154 | cnt = -ENOMEM; | |
3155 | out: | |
3156 | for_each_tracing_cpu(cpu) { | |
3157 | if (global_trace.data[cpu]) | |
3158 | atomic_dec(&global_trace.data[cpu]->disabled); | |
3159 | if (max_tr.data[cpu]) | |
3160 | atomic_dec(&max_tr.data[cpu]->disabled); | |
3161 | } | |
3162 | ||
3163 | tracing_start(); | |
3164 | max_tr.entries = global_trace.entries; | |
3165 | mutex_unlock(&trace_types_lock); | |
3166 | ||
3167 | return cnt; | |
3168 | } | |
3169 | ||
3170 | static int mark_printk(const char *fmt, ...) | |
3171 | { | |
3172 | int ret; | |
3173 | va_list args; | |
3174 | va_start(args, fmt); | |
3175 | ret = trace_vprintk(0, -1, fmt, args); | |
3176 | va_end(args); | |
3177 | return ret; | |
3178 | } | |
3179 | ||
3180 | static ssize_t | |
3181 | tracing_mark_write(struct file *filp, const char __user *ubuf, | |
3182 | size_t cnt, loff_t *fpos) | |
3183 | { | |
3184 | char *buf; | |
3185 | char *end; | |
3186 | ||
3187 | if (tracing_disabled) | |
3188 | return -EINVAL; | |
3189 | ||
3190 | if (cnt > TRACE_BUF_SIZE) | |
3191 | cnt = TRACE_BUF_SIZE; | |
3192 | ||
3193 | buf = kmalloc(cnt + 1, GFP_KERNEL); | |
3194 | if (buf == NULL) | |
3195 | return -ENOMEM; | |
3196 | ||
3197 | if (copy_from_user(buf, ubuf, cnt)) { | |
3198 | kfree(buf); | |
3199 | return -EFAULT; | |
3200 | } | |
3201 | ||
3202 | /* Cut from the first nil or newline. */ | |
3203 | buf[cnt] = '\0'; | |
3204 | end = strchr(buf, '\n'); | |
3205 | if (end) | |
3206 | *end = '\0'; | |
3207 | ||
3208 | cnt = mark_printk("%s\n", buf); | |
3209 | kfree(buf); | |
3210 | *fpos += cnt; | |
3211 | ||
3212 | return cnt; | |
3213 | } | |
3214 | ||
3215 | static const struct file_operations tracing_max_lat_fops = { | |
3216 | .open = tracing_open_generic, | |
3217 | .read = tracing_max_lat_read, | |
3218 | .write = tracing_max_lat_write, | |
3219 | }; | |
3220 | ||
3221 | static const struct file_operations tracing_ctrl_fops = { | |
3222 | .open = tracing_open_generic, | |
3223 | .read = tracing_ctrl_read, | |
3224 | .write = tracing_ctrl_write, | |
3225 | }; | |
3226 | ||
3227 | static const struct file_operations set_tracer_fops = { | |
3228 | .open = tracing_open_generic, | |
3229 | .read = tracing_set_trace_read, | |
3230 | .write = tracing_set_trace_write, | |
3231 | }; | |
3232 | ||
3233 | static const struct file_operations tracing_pipe_fops = { | |
3234 | .open = tracing_open_pipe, | |
3235 | .poll = tracing_poll_pipe, | |
3236 | .read = tracing_read_pipe, | |
3237 | .splice_read = tracing_splice_read_pipe, | |
3238 | .release = tracing_release_pipe, | |
3239 | }; | |
3240 | ||
3241 | static const struct file_operations tracing_entries_fops = { | |
3242 | .open = tracing_open_generic, | |
3243 | .read = tracing_entries_read, | |
3244 | .write = tracing_entries_write, | |
3245 | }; | |
3246 | ||
3247 | static const struct file_operations tracing_mark_fops = { | |
3248 | .open = tracing_open_generic, | |
3249 | .write = tracing_mark_write, | |
3250 | }; | |
3251 | ||
3252 | struct ftrace_buffer_info { | |
3253 | struct trace_array *tr; | |
3254 | void *spare; | |
3255 | int cpu; | |
3256 | unsigned int read; | |
3257 | }; | |
3258 | ||
3259 | static int tracing_buffers_open(struct inode *inode, struct file *filp) | |
3260 | { | |
3261 | int cpu = (int)(long)inode->i_private; | |
3262 | struct ftrace_buffer_info *info; | |
3263 | ||
3264 | if (tracing_disabled) | |
3265 | return -ENODEV; | |
3266 | ||
3267 | info = kzalloc(sizeof(*info), GFP_KERNEL); | |
3268 | if (!info) | |
3269 | return -ENOMEM; | |
3270 | ||
3271 | info->tr = &global_trace; | |
3272 | info->cpu = cpu; | |
3273 | info->spare = ring_buffer_alloc_read_page(info->tr->buffer); | |
3274 | /* Force reading ring buffer for first read */ | |
3275 | info->read = (unsigned int)-1; | |
3276 | if (!info->spare) | |
3277 | goto out; | |
3278 | ||
3279 | filp->private_data = info; | |
3280 | ||
3281 | return 0; | |
3282 | ||
3283 | out: | |
3284 | kfree(info); | |
3285 | return -ENOMEM; | |
3286 | } | |
3287 | ||
3288 | static ssize_t | |
3289 | tracing_buffers_read(struct file *filp, char __user *ubuf, | |
3290 | size_t count, loff_t *ppos) | |
3291 | { | |
3292 | struct ftrace_buffer_info *info = filp->private_data; | |
3293 | unsigned int pos; | |
3294 | ssize_t ret; | |
3295 | size_t size; | |
3296 | ||
3297 | if (!count) | |
3298 | return 0; | |
3299 | ||
3300 | /* Do we have previous read data to read? */ | |
3301 | if (info->read < PAGE_SIZE) | |
3302 | goto read; | |
3303 | ||
3304 | info->read = 0; | |
3305 | ||
3306 | ret = ring_buffer_read_page(info->tr->buffer, | |
3307 | &info->spare, | |
3308 | count, | |
3309 | info->cpu, 0); | |
3310 | if (ret < 0) | |
3311 | return 0; | |
3312 | ||
3313 | pos = ring_buffer_page_len(info->spare); | |
3314 | ||
3315 | if (pos < PAGE_SIZE) | |
3316 | memset(info->spare + pos, 0, PAGE_SIZE - pos); | |
3317 | ||
3318 | read: | |
3319 | size = PAGE_SIZE - info->read; | |
3320 | if (size > count) | |
3321 | size = count; | |
3322 | ||
3323 | ret = copy_to_user(ubuf, info->spare + info->read, size); | |
3324 | if (ret == size) | |
3325 | return -EFAULT; | |
3326 | size -= ret; | |
3327 | ||
3328 | *ppos += size; | |
3329 | info->read += size; | |
3330 | ||
3331 | return size; | |
3332 | } | |
3333 | ||
3334 | static int tracing_buffers_release(struct inode *inode, struct file *file) | |
3335 | { | |
3336 | struct ftrace_buffer_info *info = file->private_data; | |
3337 | ||
3338 | ring_buffer_free_read_page(info->tr->buffer, info->spare); | |
3339 | kfree(info); | |
3340 | ||
3341 | return 0; | |
3342 | } | |
3343 | ||
3344 | struct buffer_ref { | |
3345 | struct ring_buffer *buffer; | |
3346 | void *page; | |
3347 | int ref; | |
3348 | }; | |
3349 | ||
3350 | static void buffer_pipe_buf_release(struct pipe_inode_info *pipe, | |
3351 | struct pipe_buffer *buf) | |
3352 | { | |
3353 | struct buffer_ref *ref = (struct buffer_ref *)buf->private; | |
3354 | ||
3355 | if (--ref->ref) | |
3356 | return; | |
3357 | ||
3358 | ring_buffer_free_read_page(ref->buffer, ref->page); | |
3359 | kfree(ref); | |
3360 | buf->private = 0; | |
3361 | } | |
3362 | ||
3363 | static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe, | |
3364 | struct pipe_buffer *buf) | |
3365 | { | |
3366 | return 1; | |
3367 | } | |
3368 | ||
3369 | static void buffer_pipe_buf_get(struct pipe_inode_info *pipe, | |
3370 | struct pipe_buffer *buf) | |
3371 | { | |
3372 | struct buffer_ref *ref = (struct buffer_ref *)buf->private; | |
3373 | ||
3374 | ref->ref++; | |
3375 | } | |
3376 | ||
3377 | /* Pipe buffer operations for a buffer. */ | |
3378 | static struct pipe_buf_operations buffer_pipe_buf_ops = { | |
3379 | .can_merge = 0, | |
3380 | .map = generic_pipe_buf_map, | |
3381 | .unmap = generic_pipe_buf_unmap, | |
3382 | .confirm = generic_pipe_buf_confirm, | |
3383 | .release = buffer_pipe_buf_release, | |
3384 | .steal = buffer_pipe_buf_steal, | |
3385 | .get = buffer_pipe_buf_get, | |
3386 | }; | |
3387 | ||
3388 | /* | |
3389 | * Callback from splice_to_pipe(), if we need to release some pages | |
3390 | * at the end of the spd in case we error'ed out in filling the pipe. | |
3391 | */ | |
3392 | static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i) | |
3393 | { | |
3394 | struct buffer_ref *ref = | |
3395 | (struct buffer_ref *)spd->partial[i].private; | |
3396 | ||
3397 | if (--ref->ref) | |
3398 | return; | |
3399 | ||
3400 | ring_buffer_free_read_page(ref->buffer, ref->page); | |
3401 | kfree(ref); | |
3402 | spd->partial[i].private = 0; | |
3403 | } | |
3404 | ||
3405 | static ssize_t | |
3406 | tracing_buffers_splice_read(struct file *file, loff_t *ppos, | |
3407 | struct pipe_inode_info *pipe, size_t len, | |
3408 | unsigned int flags) | |
3409 | { | |
3410 | struct ftrace_buffer_info *info = file->private_data; | |
3411 | struct partial_page partial[PIPE_BUFFERS]; | |
3412 | struct page *pages[PIPE_BUFFERS]; | |
3413 | struct splice_pipe_desc spd = { | |
3414 | .pages = pages, | |
3415 | .partial = partial, | |
3416 | .flags = flags, | |
3417 | .ops = &buffer_pipe_buf_ops, | |
3418 | .spd_release = buffer_spd_release, | |
3419 | }; | |
3420 | struct buffer_ref *ref; | |
3421 | int size, i; | |
3422 | size_t ret; | |
3423 | ||
3424 | /* | |
3425 | * We can't seek on a buffer input | |
3426 | */ | |
3427 | if (unlikely(*ppos)) | |
3428 | return -ESPIPE; | |
3429 | ||
3430 | ||
3431 | for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) { | |
3432 | struct page *page; | |
3433 | int r; | |
3434 | ||
3435 | ref = kzalloc(sizeof(*ref), GFP_KERNEL); | |
3436 | if (!ref) | |
3437 | break; | |
3438 | ||
3439 | ref->buffer = info->tr->buffer; | |
3440 | ref->page = ring_buffer_alloc_read_page(ref->buffer); | |
3441 | if (!ref->page) { | |
3442 | kfree(ref); | |
3443 | break; | |
3444 | } | |
3445 | ||
3446 | r = ring_buffer_read_page(ref->buffer, &ref->page, | |
3447 | len, info->cpu, 0); | |
3448 | if (r < 0) { | |
3449 | ring_buffer_free_read_page(ref->buffer, | |
3450 | ref->page); | |
3451 | kfree(ref); | |
3452 | break; | |
3453 | } | |
3454 | ||
3455 | /* | |
3456 | * zero out any left over data, this is going to | |
3457 | * user land. | |
3458 | */ | |
3459 | size = ring_buffer_page_len(ref->page); | |
3460 | if (size < PAGE_SIZE) | |
3461 | memset(ref->page + size, 0, PAGE_SIZE - size); | |
3462 | ||
3463 | page = virt_to_page(ref->page); | |
3464 | ||
3465 | spd.pages[i] = page; | |
3466 | spd.partial[i].len = PAGE_SIZE; | |
3467 | spd.partial[i].offset = 0; | |
3468 | spd.partial[i].private = (unsigned long)ref; | |
3469 | spd.nr_pages++; | |
3470 | } | |
3471 | ||
3472 | spd.nr_pages = i; | |
3473 | ||
3474 | /* did we read anything? */ | |
3475 | if (!spd.nr_pages) { | |
3476 | if (flags & SPLICE_F_NONBLOCK) | |
3477 | ret = -EAGAIN; | |
3478 | else | |
3479 | ret = 0; | |
3480 | /* TODO: block */ | |
3481 | return ret; | |
3482 | } | |
3483 | ||
3484 | ret = splice_to_pipe(pipe, &spd); | |
3485 | ||
3486 | return ret; | |
3487 | } | |
3488 | ||
3489 | static const struct file_operations tracing_buffers_fops = { | |
3490 | .open = tracing_buffers_open, | |
3491 | .read = tracing_buffers_read, | |
3492 | .release = tracing_buffers_release, | |
3493 | .splice_read = tracing_buffers_splice_read, | |
3494 | .llseek = no_llseek, | |
3495 | }; | |
3496 | ||
3497 | #ifdef CONFIG_DYNAMIC_FTRACE | |
3498 | ||
3499 | int __weak ftrace_arch_read_dyn_info(char *buf, int size) | |
3500 | { | |
3501 | return 0; | |
3502 | } | |
3503 | ||
3504 | static ssize_t | |
3505 | tracing_read_dyn_info(struct file *filp, char __user *ubuf, | |
3506 | size_t cnt, loff_t *ppos) | |
3507 | { | |
3508 | static char ftrace_dyn_info_buffer[1024]; | |
3509 | static DEFINE_MUTEX(dyn_info_mutex); | |
3510 | unsigned long *p = filp->private_data; | |
3511 | char *buf = ftrace_dyn_info_buffer; | |
3512 | int size = ARRAY_SIZE(ftrace_dyn_info_buffer); | |
3513 | int r; | |
3514 | ||
3515 | mutex_lock(&dyn_info_mutex); | |
3516 | r = sprintf(buf, "%ld ", *p); | |
3517 | ||
3518 | r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r); | |
3519 | buf[r++] = '\n'; | |
3520 | ||
3521 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3522 | ||
3523 | mutex_unlock(&dyn_info_mutex); | |
3524 | ||
3525 | return r; | |
3526 | } | |
3527 | ||
3528 | static const struct file_operations tracing_dyn_info_fops = { | |
3529 | .open = tracing_open_generic, | |
3530 | .read = tracing_read_dyn_info, | |
3531 | }; | |
3532 | #endif | |
3533 | ||
3534 | static struct dentry *d_tracer; | |
3535 | ||
3536 | struct dentry *tracing_init_dentry(void) | |
3537 | { | |
3538 | static int once; | |
3539 | ||
3540 | if (d_tracer) | |
3541 | return d_tracer; | |
3542 | ||
3543 | d_tracer = debugfs_create_dir("tracing", NULL); | |
3544 | ||
3545 | if (!d_tracer && !once) { | |
3546 | once = 1; | |
3547 | pr_warning("Could not create debugfs directory 'tracing'\n"); | |
3548 | return NULL; | |
3549 | } | |
3550 | ||
3551 | return d_tracer; | |
3552 | } | |
3553 | ||
3554 | static struct dentry *d_percpu; | |
3555 | ||
3556 | struct dentry *tracing_dentry_percpu(void) | |
3557 | { | |
3558 | static int once; | |
3559 | struct dentry *d_tracer; | |
3560 | ||
3561 | if (d_percpu) | |
3562 | return d_percpu; | |
3563 | ||
3564 | d_tracer = tracing_init_dentry(); | |
3565 | ||
3566 | if (!d_tracer) | |
3567 | return NULL; | |
3568 | ||
3569 | d_percpu = debugfs_create_dir("per_cpu", d_tracer); | |
3570 | ||
3571 | if (!d_percpu && !once) { | |
3572 | once = 1; | |
3573 | pr_warning("Could not create debugfs directory 'per_cpu'\n"); | |
3574 | return NULL; | |
3575 | } | |
3576 | ||
3577 | return d_percpu; | |
3578 | } | |
3579 | ||
3580 | static void tracing_init_debugfs_percpu(long cpu) | |
3581 | { | |
3582 | struct dentry *d_percpu = tracing_dentry_percpu(); | |
3583 | struct dentry *entry, *d_cpu; | |
3584 | /* strlen(cpu) + MAX(log10(cpu)) + '\0' */ | |
3585 | char cpu_dir[7]; | |
3586 | ||
3587 | if (cpu > 999 || cpu < 0) | |
3588 | return; | |
3589 | ||
3590 | sprintf(cpu_dir, "cpu%ld", cpu); | |
3591 | d_cpu = debugfs_create_dir(cpu_dir, d_percpu); | |
3592 | if (!d_cpu) { | |
3593 | pr_warning("Could not create debugfs '%s' entry\n", cpu_dir); | |
3594 | return; | |
3595 | } | |
3596 | ||
3597 | /* per cpu trace_pipe */ | |
3598 | entry = debugfs_create_file("trace_pipe", 0444, d_cpu, | |
3599 | (void *) cpu, &tracing_pipe_fops); | |
3600 | if (!entry) | |
3601 | pr_warning("Could not create debugfs 'trace_pipe' entry\n"); | |
3602 | ||
3603 | /* per cpu trace */ | |
3604 | entry = debugfs_create_file("trace", 0644, d_cpu, | |
3605 | (void *) cpu, &tracing_fops); | |
3606 | if (!entry) | |
3607 | pr_warning("Could not create debugfs 'trace' entry\n"); | |
3608 | ||
3609 | entry = debugfs_create_file("trace_pipe_raw", 0444, d_cpu, | |
3610 | (void *) cpu, &tracing_buffers_fops); | |
3611 | if (!entry) | |
3612 | pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n"); | |
3613 | } | |
3614 | ||
3615 | #ifdef CONFIG_FTRACE_SELFTEST | |
3616 | /* Let selftest have access to static functions in this file */ | |
3617 | #include "trace_selftest.c" | |
3618 | #endif | |
3619 | ||
3620 | struct trace_option_dentry { | |
3621 | struct tracer_opt *opt; | |
3622 | struct tracer_flags *flags; | |
3623 | struct dentry *entry; | |
3624 | }; | |
3625 | ||
3626 | static ssize_t | |
3627 | trace_options_read(struct file *filp, char __user *ubuf, size_t cnt, | |
3628 | loff_t *ppos) | |
3629 | { | |
3630 | struct trace_option_dentry *topt = filp->private_data; | |
3631 | char *buf; | |
3632 | ||
3633 | if (topt->flags->val & topt->opt->bit) | |
3634 | buf = "1\n"; | |
3635 | else | |
3636 | buf = "0\n"; | |
3637 | ||
3638 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2); | |
3639 | } | |
3640 | ||
3641 | static ssize_t | |
3642 | trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt, | |
3643 | loff_t *ppos) | |
3644 | { | |
3645 | struct trace_option_dentry *topt = filp->private_data; | |
3646 | unsigned long val; | |
3647 | char buf[64]; | |
3648 | int ret; | |
3649 | ||
3650 | if (cnt >= sizeof(buf)) | |
3651 | return -EINVAL; | |
3652 | ||
3653 | if (copy_from_user(&buf, ubuf, cnt)) | |
3654 | return -EFAULT; | |
3655 | ||
3656 | buf[cnt] = 0; | |
3657 | ||
3658 | ret = strict_strtoul(buf, 10, &val); | |
3659 | if (ret < 0) | |
3660 | return ret; | |
3661 | ||
3662 | ret = 0; | |
3663 | switch (val) { | |
3664 | case 0: | |
3665 | /* do nothing if already cleared */ | |
3666 | if (!(topt->flags->val & topt->opt->bit)) | |
3667 | break; | |
3668 | ||
3669 | mutex_lock(&trace_types_lock); | |
3670 | if (current_trace->set_flag) | |
3671 | ret = current_trace->set_flag(topt->flags->val, | |
3672 | topt->opt->bit, 0); | |
3673 | mutex_unlock(&trace_types_lock); | |
3674 | if (ret) | |
3675 | return ret; | |
3676 | topt->flags->val &= ~topt->opt->bit; | |
3677 | break; | |
3678 | case 1: | |
3679 | /* do nothing if already set */ | |
3680 | if (topt->flags->val & topt->opt->bit) | |
3681 | break; | |
3682 | ||
3683 | mutex_lock(&trace_types_lock); | |
3684 | if (current_trace->set_flag) | |
3685 | ret = current_trace->set_flag(topt->flags->val, | |
3686 | topt->opt->bit, 1); | |
3687 | mutex_unlock(&trace_types_lock); | |
3688 | if (ret) | |
3689 | return ret; | |
3690 | topt->flags->val |= topt->opt->bit; | |
3691 | break; | |
3692 | ||
3693 | default: | |
3694 | return -EINVAL; | |
3695 | } | |
3696 | ||
3697 | *ppos += cnt; | |
3698 | ||
3699 | return cnt; | |
3700 | } | |
3701 | ||
3702 | ||
3703 | static const struct file_operations trace_options_fops = { | |
3704 | .open = tracing_open_generic, | |
3705 | .read = trace_options_read, | |
3706 | .write = trace_options_write, | |
3707 | }; | |
3708 | ||
3709 | static ssize_t | |
3710 | trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt, | |
3711 | loff_t *ppos) | |
3712 | { | |
3713 | long index = (long)filp->private_data; | |
3714 | char *buf; | |
3715 | ||
3716 | if (trace_flags & (1 << index)) | |
3717 | buf = "1\n"; | |
3718 | else | |
3719 | buf = "0\n"; | |
3720 | ||
3721 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2); | |
3722 | } | |
3723 | ||
3724 | static ssize_t | |
3725 | trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt, | |
3726 | loff_t *ppos) | |
3727 | { | |
3728 | long index = (long)filp->private_data; | |
3729 | char buf[64]; | |
3730 | unsigned long val; | |
3731 | int ret; | |
3732 | ||
3733 | if (cnt >= sizeof(buf)) | |
3734 | return -EINVAL; | |
3735 | ||
3736 | if (copy_from_user(&buf, ubuf, cnt)) | |
3737 | return -EFAULT; | |
3738 | ||
3739 | buf[cnt] = 0; | |
3740 | ||
3741 | ret = strict_strtoul(buf, 10, &val); | |
3742 | if (ret < 0) | |
3743 | return ret; | |
3744 | ||
3745 | switch (val) { | |
3746 | case 0: | |
3747 | trace_flags &= ~(1 << index); | |
3748 | break; | |
3749 | case 1: | |
3750 | trace_flags |= 1 << index; | |
3751 | break; | |
3752 | ||
3753 | default: | |
3754 | return -EINVAL; | |
3755 | } | |
3756 | ||
3757 | *ppos += cnt; | |
3758 | ||
3759 | return cnt; | |
3760 | } | |
3761 | ||
3762 | static const struct file_operations trace_options_core_fops = { | |
3763 | .open = tracing_open_generic, | |
3764 | .read = trace_options_core_read, | |
3765 | .write = trace_options_core_write, | |
3766 | }; | |
3767 | ||
3768 | static struct dentry *trace_options_init_dentry(void) | |
3769 | { | |
3770 | struct dentry *d_tracer; | |
3771 | static struct dentry *t_options; | |
3772 | ||
3773 | if (t_options) | |
3774 | return t_options; | |
3775 | ||
3776 | d_tracer = tracing_init_dentry(); | |
3777 | if (!d_tracer) | |
3778 | return NULL; | |
3779 | ||
3780 | t_options = debugfs_create_dir("options", d_tracer); | |
3781 | if (!t_options) { | |
3782 | pr_warning("Could not create debugfs directory 'options'\n"); | |
3783 | return NULL; | |
3784 | } | |
3785 | ||
3786 | return t_options; | |
3787 | } | |
3788 | ||
3789 | static void | |
3790 | create_trace_option_file(struct trace_option_dentry *topt, | |
3791 | struct tracer_flags *flags, | |
3792 | struct tracer_opt *opt) | |
3793 | { | |
3794 | struct dentry *t_options; | |
3795 | struct dentry *entry; | |
3796 | ||
3797 | t_options = trace_options_init_dentry(); | |
3798 | if (!t_options) | |
3799 | return; | |
3800 | ||
3801 | topt->flags = flags; | |
3802 | topt->opt = opt; | |
3803 | ||
3804 | entry = debugfs_create_file(opt->name, 0644, t_options, topt, | |
3805 | &trace_options_fops); | |
3806 | ||
3807 | topt->entry = entry; | |
3808 | ||
3809 | } | |
3810 | ||
3811 | static struct trace_option_dentry * | |
3812 | create_trace_option_files(struct tracer *tracer) | |
3813 | { | |
3814 | struct trace_option_dentry *topts; | |
3815 | struct tracer_flags *flags; | |
3816 | struct tracer_opt *opts; | |
3817 | int cnt; | |
3818 | ||
3819 | if (!tracer) | |
3820 | return NULL; | |
3821 | ||
3822 | flags = tracer->flags; | |
3823 | ||
3824 | if (!flags || !flags->opts) | |
3825 | return NULL; | |
3826 | ||
3827 | opts = flags->opts; | |
3828 | ||
3829 | for (cnt = 0; opts[cnt].name; cnt++) | |
3830 | ; | |
3831 | ||
3832 | topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL); | |
3833 | if (!topts) | |
3834 | return NULL; | |
3835 | ||
3836 | for (cnt = 0; opts[cnt].name; cnt++) | |
3837 | create_trace_option_file(&topts[cnt], flags, | |
3838 | &opts[cnt]); | |
3839 | ||
3840 | return topts; | |
3841 | } | |
3842 | ||
3843 | static void | |
3844 | destroy_trace_option_files(struct trace_option_dentry *topts) | |
3845 | { | |
3846 | int cnt; | |
3847 | ||
3848 | if (!topts) | |
3849 | return; | |
3850 | ||
3851 | for (cnt = 0; topts[cnt].opt; cnt++) { | |
3852 | if (topts[cnt].entry) | |
3853 | debugfs_remove(topts[cnt].entry); | |
3854 | } | |
3855 | ||
3856 | kfree(topts); | |
3857 | } | |
3858 | ||
3859 | static struct dentry * | |
3860 | create_trace_option_core_file(const char *option, long index) | |
3861 | { | |
3862 | struct dentry *t_options; | |
3863 | struct dentry *entry; | |
3864 | ||
3865 | t_options = trace_options_init_dentry(); | |
3866 | if (!t_options) | |
3867 | return NULL; | |
3868 | ||
3869 | entry = debugfs_create_file(option, 0644, t_options, (void *)index, | |
3870 | &trace_options_core_fops); | |
3871 | ||
3872 | return entry; | |
3873 | } | |
3874 | ||
3875 | static __init void create_trace_options_dir(void) | |
3876 | { | |
3877 | struct dentry *t_options; | |
3878 | struct dentry *entry; | |
3879 | int i; | |
3880 | ||
3881 | t_options = trace_options_init_dentry(); | |
3882 | if (!t_options) | |
3883 | return; | |
3884 | ||
3885 | for (i = 0; trace_options[i]; i++) { | |
3886 | entry = create_trace_option_core_file(trace_options[i], i); | |
3887 | if (!entry) | |
3888 | pr_warning("Could not create debugfs %s entry\n", | |
3889 | trace_options[i]); | |
3890 | } | |
3891 | } | |
3892 | ||
3893 | static __init int tracer_init_debugfs(void) | |
3894 | { | |
3895 | struct dentry *d_tracer; | |
3896 | struct dentry *entry; | |
3897 | int cpu; | |
3898 | ||
3899 | d_tracer = tracing_init_dentry(); | |
3900 | ||
3901 | entry = debugfs_create_file("tracing_enabled", 0644, d_tracer, | |
3902 | &global_trace, &tracing_ctrl_fops); | |
3903 | if (!entry) | |
3904 | pr_warning("Could not create debugfs 'tracing_enabled' entry\n"); | |
3905 | ||
3906 | entry = debugfs_create_file("trace_options", 0644, d_tracer, | |
3907 | NULL, &tracing_iter_fops); | |
3908 | if (!entry) | |
3909 | pr_warning("Could not create debugfs 'trace_options' entry\n"); | |
3910 | ||
3911 | create_trace_options_dir(); | |
3912 | ||
3913 | entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer, | |
3914 | NULL, &tracing_cpumask_fops); | |
3915 | if (!entry) | |
3916 | pr_warning("Could not create debugfs 'tracing_cpumask' entry\n"); | |
3917 | ||
3918 | entry = debugfs_create_file("trace", 0644, d_tracer, | |
3919 | (void *) TRACE_PIPE_ALL_CPU, &tracing_fops); | |
3920 | if (!entry) | |
3921 | pr_warning("Could not create debugfs 'trace' entry\n"); | |
3922 | ||
3923 | entry = debugfs_create_file("available_tracers", 0444, d_tracer, | |
3924 | &global_trace, &show_traces_fops); | |
3925 | if (!entry) | |
3926 | pr_warning("Could not create debugfs 'available_tracers' entry\n"); | |
3927 | ||
3928 | entry = debugfs_create_file("current_tracer", 0444, d_tracer, | |
3929 | &global_trace, &set_tracer_fops); | |
3930 | if (!entry) | |
3931 | pr_warning("Could not create debugfs 'current_tracer' entry\n"); | |
3932 | ||
3933 | entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer, | |
3934 | &tracing_max_latency, | |
3935 | &tracing_max_lat_fops); | |
3936 | if (!entry) | |
3937 | pr_warning("Could not create debugfs " | |
3938 | "'tracing_max_latency' entry\n"); | |
3939 | ||
3940 | entry = debugfs_create_file("tracing_thresh", 0644, d_tracer, | |
3941 | &tracing_thresh, &tracing_max_lat_fops); | |
3942 | if (!entry) | |
3943 | pr_warning("Could not create debugfs " | |
3944 | "'tracing_thresh' entry\n"); | |
3945 | entry = debugfs_create_file("README", 0644, d_tracer, | |
3946 | NULL, &tracing_readme_fops); | |
3947 | if (!entry) | |
3948 | pr_warning("Could not create debugfs 'README' entry\n"); | |
3949 | ||
3950 | entry = debugfs_create_file("trace_pipe", 0444, d_tracer, | |
3951 | (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops); | |
3952 | if (!entry) | |
3953 | pr_warning("Could not create debugfs " | |
3954 | "'trace_pipe' entry\n"); | |
3955 | ||
3956 | entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer, | |
3957 | &global_trace, &tracing_entries_fops); | |
3958 | if (!entry) | |
3959 | pr_warning("Could not create debugfs " | |
3960 | "'buffer_size_kb' entry\n"); | |
3961 | ||
3962 | entry = debugfs_create_file("trace_marker", 0220, d_tracer, | |
3963 | NULL, &tracing_mark_fops); | |
3964 | if (!entry) | |
3965 | pr_warning("Could not create debugfs " | |
3966 | "'trace_marker' entry\n"); | |
3967 | ||
3968 | #ifdef CONFIG_DYNAMIC_FTRACE | |
3969 | entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer, | |
3970 | &ftrace_update_tot_cnt, | |
3971 | &tracing_dyn_info_fops); | |
3972 | if (!entry) | |
3973 | pr_warning("Could not create debugfs " | |
3974 | "'dyn_ftrace_total_info' entry\n"); | |
3975 | #endif | |
3976 | #ifdef CONFIG_SYSPROF_TRACER | |
3977 | init_tracer_sysprof_debugfs(d_tracer); | |
3978 | #endif | |
3979 | ||
3980 | for_each_tracing_cpu(cpu) | |
3981 | tracing_init_debugfs_percpu(cpu); | |
3982 | ||
3983 | return 0; | |
3984 | } | |
3985 | ||
3986 | static int trace_panic_handler(struct notifier_block *this, | |
3987 | unsigned long event, void *unused) | |
3988 | { | |
3989 | if (ftrace_dump_on_oops) | |
3990 | ftrace_dump(); | |
3991 | return NOTIFY_OK; | |
3992 | } | |
3993 | ||
3994 | static struct notifier_block trace_panic_notifier = { | |
3995 | .notifier_call = trace_panic_handler, | |
3996 | .next = NULL, | |
3997 | .priority = 150 /* priority: INT_MAX >= x >= 0 */ | |
3998 | }; | |
3999 | ||
4000 | static int trace_die_handler(struct notifier_block *self, | |
4001 | unsigned long val, | |
4002 | void *data) | |
4003 | { | |
4004 | switch (val) { | |
4005 | case DIE_OOPS: | |
4006 | if (ftrace_dump_on_oops) | |
4007 | ftrace_dump(); | |
4008 | break; | |
4009 | default: | |
4010 | break; | |
4011 | } | |
4012 | return NOTIFY_OK; | |
4013 | } | |
4014 | ||
4015 | static struct notifier_block trace_die_notifier = { | |
4016 | .notifier_call = trace_die_handler, | |
4017 | .priority = 200 | |
4018 | }; | |
4019 | ||
4020 | /* | |
4021 | * printk is set to max of 1024, we really don't need it that big. | |
4022 | * Nothing should be printing 1000 characters anyway. | |
4023 | */ | |
4024 | #define TRACE_MAX_PRINT 1000 | |
4025 | ||
4026 | /* | |
4027 | * Define here KERN_TRACE so that we have one place to modify | |
4028 | * it if we decide to change what log level the ftrace dump | |
4029 | * should be at. | |
4030 | */ | |
4031 | #define KERN_TRACE KERN_EMERG | |
4032 | ||
4033 | static void | |
4034 | trace_printk_seq(struct trace_seq *s) | |
4035 | { | |
4036 | /* Probably should print a warning here. */ | |
4037 | if (s->len >= 1000) | |
4038 | s->len = 1000; | |
4039 | ||
4040 | /* should be zero ended, but we are paranoid. */ | |
4041 | s->buffer[s->len] = 0; | |
4042 | ||
4043 | printk(KERN_TRACE "%s", s->buffer); | |
4044 | ||
4045 | trace_seq_init(s); | |
4046 | } | |
4047 | ||
4048 | void ftrace_dump(void) | |
4049 | { | |
4050 | static DEFINE_SPINLOCK(ftrace_dump_lock); | |
4051 | /* use static because iter can be a bit big for the stack */ | |
4052 | static struct trace_iterator iter; | |
4053 | static int dump_ran; | |
4054 | unsigned long flags; | |
4055 | int cnt = 0, cpu; | |
4056 | ||
4057 | /* only one dump */ | |
4058 | spin_lock_irqsave(&ftrace_dump_lock, flags); | |
4059 | if (dump_ran) | |
4060 | goto out; | |
4061 | ||
4062 | dump_ran = 1; | |
4063 | ||
4064 | /* No turning back! */ | |
4065 | tracing_off(); | |
4066 | ftrace_kill(); | |
4067 | ||
4068 | for_each_tracing_cpu(cpu) { | |
4069 | atomic_inc(&global_trace.data[cpu]->disabled); | |
4070 | } | |
4071 | ||
4072 | /* don't look at user memory in panic mode */ | |
4073 | trace_flags &= ~TRACE_ITER_SYM_USEROBJ; | |
4074 | ||
4075 | printk(KERN_TRACE "Dumping ftrace buffer:\n"); | |
4076 | ||
4077 | /* Simulate the iterator */ | |
4078 | iter.tr = &global_trace; | |
4079 | iter.trace = current_trace; | |
4080 | iter.cpu_file = TRACE_PIPE_ALL_CPU; | |
4081 | ||
4082 | /* | |
4083 | * We need to stop all tracing on all CPUS to read the | |
4084 | * the next buffer. This is a bit expensive, but is | |
4085 | * not done often. We fill all what we can read, | |
4086 | * and then release the locks again. | |
4087 | */ | |
4088 | ||
4089 | while (!trace_empty(&iter)) { | |
4090 | ||
4091 | if (!cnt) | |
4092 | printk(KERN_TRACE "---------------------------------\n"); | |
4093 | ||
4094 | cnt++; | |
4095 | ||
4096 | /* reset all but tr, trace, and overruns */ | |
4097 | memset(&iter.seq, 0, | |
4098 | sizeof(struct trace_iterator) - | |
4099 | offsetof(struct trace_iterator, seq)); | |
4100 | iter.iter_flags |= TRACE_FILE_LAT_FMT; | |
4101 | iter.pos = -1; | |
4102 | ||
4103 | if (find_next_entry_inc(&iter) != NULL) { | |
4104 | print_trace_line(&iter); | |
4105 | trace_consume(&iter); | |
4106 | } | |
4107 | ||
4108 | trace_printk_seq(&iter.seq); | |
4109 | } | |
4110 | ||
4111 | if (!cnt) | |
4112 | printk(KERN_TRACE " (ftrace buffer empty)\n"); | |
4113 | else | |
4114 | printk(KERN_TRACE "---------------------------------\n"); | |
4115 | ||
4116 | out: | |
4117 | spin_unlock_irqrestore(&ftrace_dump_lock, flags); | |
4118 | } | |
4119 | ||
4120 | __init static int tracer_alloc_buffers(void) | |
4121 | { | |
4122 | struct trace_array_cpu *data; | |
4123 | int ring_buf_size; | |
4124 | int i; | |
4125 | int ret = -ENOMEM; | |
4126 | ||
4127 | if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL)) | |
4128 | goto out; | |
4129 | ||
4130 | if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL)) | |
4131 | goto out_free_buffer_mask; | |
4132 | ||
4133 | if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL)) | |
4134 | goto out_free_tracing_cpumask; | |
4135 | ||
4136 | /* To save memory, keep the ring buffer size to its minimum */ | |
4137 | if (ring_buffer_expanded) | |
4138 | ring_buf_size = trace_buf_size; | |
4139 | else | |
4140 | ring_buf_size = 1; | |
4141 | ||
4142 | cpumask_copy(tracing_buffer_mask, cpu_possible_mask); | |
4143 | cpumask_copy(tracing_cpumask, cpu_all_mask); | |
4144 | cpumask_clear(tracing_reader_cpumask); | |
4145 | ||
4146 | /* TODO: make the number of buffers hot pluggable with CPUS */ | |
4147 | global_trace.buffer = ring_buffer_alloc(ring_buf_size, | |
4148 | TRACE_BUFFER_FLAGS); | |
4149 | if (!global_trace.buffer) { | |
4150 | printk(KERN_ERR "tracer: failed to allocate ring buffer!\n"); | |
4151 | WARN_ON(1); | |
4152 | goto out_free_cpumask; | |
4153 | } | |
4154 | global_trace.entries = ring_buffer_size(global_trace.buffer); | |
4155 | ||
4156 | ||
4157 | #ifdef CONFIG_TRACER_MAX_TRACE | |
4158 | max_tr.buffer = ring_buffer_alloc(ring_buf_size, | |
4159 | TRACE_BUFFER_FLAGS); | |
4160 | if (!max_tr.buffer) { | |
4161 | printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n"); | |
4162 | WARN_ON(1); | |
4163 | ring_buffer_free(global_trace.buffer); | |
4164 | goto out_free_cpumask; | |
4165 | } | |
4166 | max_tr.entries = ring_buffer_size(max_tr.buffer); | |
4167 | WARN_ON(max_tr.entries != global_trace.entries); | |
4168 | #endif | |
4169 | ||
4170 | /* Allocate the first page for all buffers */ | |
4171 | for_each_tracing_cpu(i) { | |
4172 | data = global_trace.data[i] = &per_cpu(global_trace_cpu, i); | |
4173 | max_tr.data[i] = &per_cpu(max_data, i); | |
4174 | } | |
4175 | ||
4176 | trace_init_cmdlines(); | |
4177 | ||
4178 | register_tracer(&nop_trace); | |
4179 | current_trace = &nop_trace; | |
4180 | #ifdef CONFIG_BOOT_TRACER | |
4181 | register_tracer(&boot_tracer); | |
4182 | #endif | |
4183 | /* All seems OK, enable tracing */ | |
4184 | tracing_disabled = 0; | |
4185 | ||
4186 | atomic_notifier_chain_register(&panic_notifier_list, | |
4187 | &trace_panic_notifier); | |
4188 | ||
4189 | register_die_notifier(&trace_die_notifier); | |
4190 | ||
4191 | return 0; | |
4192 | ||
4193 | out_free_cpumask: | |
4194 | free_cpumask_var(tracing_reader_cpumask); | |
4195 | out_free_tracing_cpumask: | |
4196 | free_cpumask_var(tracing_cpumask); | |
4197 | out_free_buffer_mask: | |
4198 | free_cpumask_var(tracing_buffer_mask); | |
4199 | out: | |
4200 | return ret; | |
4201 | } | |
4202 | ||
4203 | __init static int clear_boot_tracer(void) | |
4204 | { | |
4205 | /* | |
4206 | * The default tracer at boot buffer is an init section. | |
4207 | * This function is called in lateinit. If we did not | |
4208 | * find the boot tracer, then clear it out, to prevent | |
4209 | * later registration from accessing the buffer that is | |
4210 | * about to be freed. | |
4211 | */ | |
4212 | if (!default_bootup_tracer) | |
4213 | return 0; | |
4214 | ||
4215 | printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n", | |
4216 | default_bootup_tracer); | |
4217 | default_bootup_tracer = NULL; | |
4218 | ||
4219 | return 0; | |
4220 | } | |
4221 | ||
4222 | early_initcall(tracer_alloc_buffers); | |
4223 | fs_initcall(tracer_init_debugfs); | |
4224 | late_initcall(clear_boot_tracer); |