2c92b3d9ea3077df8cd5d272cc48335e66a64681
[sfrench/cifs-2.6.git] / kernel / trace / trace.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * ring buffer based function tracer
4  *
5  * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7  *
8  * Originally taken from the RT patch by:
9  *    Arnaldo Carvalho de Melo <acme@redhat.com>
10  *
11  * Based on code from the latency_tracer, that is:
12  *  Copyright (C) 2004-2006 Ingo Molnar
13  *  Copyright (C) 2004 Nadia Yvette Chambers
14  */
15 #include <linux/ring_buffer.h>
16 #include <generated/utsrelease.h>
17 #include <linux/stacktrace.h>
18 #include <linux/writeback.h>
19 #include <linux/kallsyms.h>
20 #include <linux/seq_file.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/tracefs.h>
25 #include <linux/pagemap.h>
26 #include <linux/hardirq.h>
27 #include <linux/linkage.h>
28 #include <linux/uaccess.h>
29 #include <linux/vmalloc.h>
30 #include <linux/ftrace.h>
31 #include <linux/module.h>
32 #include <linux/percpu.h>
33 #include <linux/splice.h>
34 #include <linux/kdebug.h>
35 #include <linux/string.h>
36 #include <linux/mount.h>
37 #include <linux/rwsem.h>
38 #include <linux/slab.h>
39 #include <linux/ctype.h>
40 #include <linux/init.h>
41 #include <linux/poll.h>
42 #include <linux/nmi.h>
43 #include <linux/fs.h>
44 #include <linux/trace.h>
45 #include <linux/sched/clock.h>
46 #include <linux/sched/rt.h>
47
48 #include "trace.h"
49 #include "trace_output.h"
50
51 /*
52  * On boot up, the ring buffer is set to the minimum size, so that
53  * we do not waste memory on systems that are not using tracing.
54  */
55 bool ring_buffer_expanded;
56
57 /*
58  * We need to change this state when a selftest is running.
59  * A selftest will lurk into the ring-buffer to count the
60  * entries inserted during the selftest although some concurrent
61  * insertions into the ring-buffer such as trace_printk could occurred
62  * at the same time, giving false positive or negative results.
63  */
64 static bool __read_mostly tracing_selftest_running;
65
66 /*
67  * If a tracer is running, we do not want to run SELFTEST.
68  */
69 bool __read_mostly tracing_selftest_disabled;
70
71 /* Pipe tracepoints to printk */
72 struct trace_iterator *tracepoint_print_iter;
73 int tracepoint_printk;
74 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
75
76 /* For tracers that don't implement custom flags */
77 static struct tracer_opt dummy_tracer_opt[] = {
78         { }
79 };
80
81 static int
82 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
83 {
84         return 0;
85 }
86
87 /*
88  * To prevent the comm cache from being overwritten when no
89  * tracing is active, only save the comm when a trace event
90  * occurred.
91  */
92 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
93
94 /*
95  * Kill all tracing for good (never come back).
96  * It is initialized to 1 but will turn to zero if the initialization
97  * of the tracer is successful. But that is the only place that sets
98  * this back to zero.
99  */
100 static int tracing_disabled = 1;
101
102 cpumask_var_t __read_mostly     tracing_buffer_mask;
103
104 /*
105  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
106  *
107  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
108  * is set, then ftrace_dump is called. This will output the contents
109  * of the ftrace buffers to the console.  This is very useful for
110  * capturing traces that lead to crashes and outputing it to a
111  * serial console.
112  *
113  * It is default off, but you can enable it with either specifying
114  * "ftrace_dump_on_oops" in the kernel command line, or setting
115  * /proc/sys/kernel/ftrace_dump_on_oops
116  * Set 1 if you want to dump buffers of all CPUs
117  * Set 2 if you want to dump the buffer of the CPU that triggered oops
118  */
119
120 enum ftrace_dump_mode ftrace_dump_on_oops;
121
122 /* When set, tracing will stop when a WARN*() is hit */
123 int __disable_trace_on_warning;
124
125 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
126 /* Map of enums to their values, for "eval_map" file */
127 struct trace_eval_map_head {
128         struct module                   *mod;
129         unsigned long                   length;
130 };
131
132 union trace_eval_map_item;
133
134 struct trace_eval_map_tail {
135         /*
136          * "end" is first and points to NULL as it must be different
137          * than "mod" or "eval_string"
138          */
139         union trace_eval_map_item       *next;
140         const char                      *end;   /* points to NULL */
141 };
142
143 static DEFINE_MUTEX(trace_eval_mutex);
144
145 /*
146  * The trace_eval_maps are saved in an array with two extra elements,
147  * one at the beginning, and one at the end. The beginning item contains
148  * the count of the saved maps (head.length), and the module they
149  * belong to if not built in (head.mod). The ending item contains a
150  * pointer to the next array of saved eval_map items.
151  */
152 union trace_eval_map_item {
153         struct trace_eval_map           map;
154         struct trace_eval_map_head      head;
155         struct trace_eval_map_tail      tail;
156 };
157
158 static union trace_eval_map_item *trace_eval_maps;
159 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
160
161 static int tracing_set_tracer(struct trace_array *tr, const char *buf);
162 static void ftrace_trace_userstack(struct ring_buffer *buffer,
163                                    unsigned long flags, int pc);
164
165 #define MAX_TRACER_SIZE         100
166 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
167 static char *default_bootup_tracer;
168
169 static bool allocate_snapshot;
170
171 static int __init set_cmdline_ftrace(char *str)
172 {
173         strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
174         default_bootup_tracer = bootup_tracer_buf;
175         /* We are using ftrace early, expand it */
176         ring_buffer_expanded = true;
177         return 1;
178 }
179 __setup("ftrace=", set_cmdline_ftrace);
180
181 static int __init set_ftrace_dump_on_oops(char *str)
182 {
183         if (*str++ != '=' || !*str) {
184                 ftrace_dump_on_oops = DUMP_ALL;
185                 return 1;
186         }
187
188         if (!strcmp("orig_cpu", str)) {
189                 ftrace_dump_on_oops = DUMP_ORIG;
190                 return 1;
191         }
192
193         return 0;
194 }
195 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
196
197 static int __init stop_trace_on_warning(char *str)
198 {
199         if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
200                 __disable_trace_on_warning = 1;
201         return 1;
202 }
203 __setup("traceoff_on_warning", stop_trace_on_warning);
204
205 static int __init boot_alloc_snapshot(char *str)
206 {
207         allocate_snapshot = true;
208         /* We also need the main ring buffer expanded */
209         ring_buffer_expanded = true;
210         return 1;
211 }
212 __setup("alloc_snapshot", boot_alloc_snapshot);
213
214
215 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
216
217 static int __init set_trace_boot_options(char *str)
218 {
219         strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
220         return 0;
221 }
222 __setup("trace_options=", set_trace_boot_options);
223
224 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
225 static char *trace_boot_clock __initdata;
226
227 static int __init set_trace_boot_clock(char *str)
228 {
229         strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
230         trace_boot_clock = trace_boot_clock_buf;
231         return 0;
232 }
233 __setup("trace_clock=", set_trace_boot_clock);
234
235 static int __init set_tracepoint_printk(char *str)
236 {
237         if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
238                 tracepoint_printk = 1;
239         return 1;
240 }
241 __setup("tp_printk", set_tracepoint_printk);
242
243 unsigned long long ns2usecs(u64 nsec)
244 {
245         nsec += 500;
246         do_div(nsec, 1000);
247         return nsec;
248 }
249
250 /* trace_flags holds trace_options default values */
251 #define TRACE_DEFAULT_FLAGS                                             \
252         (FUNCTION_DEFAULT_FLAGS |                                       \
253          TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |                  \
254          TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO |                \
255          TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |                 \
256          TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS)
257
258 /* trace_options that are only supported by global_trace */
259 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK |                      \
260                TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
261
262 /* trace_flags that are default zero for instances */
263 #define ZEROED_TRACE_FLAGS \
264         (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
265
266 /*
267  * The global_trace is the descriptor that holds the top-level tracing
268  * buffers for the live tracing.
269  */
270 static struct trace_array global_trace = {
271         .trace_flags = TRACE_DEFAULT_FLAGS,
272 };
273
274 LIST_HEAD(ftrace_trace_arrays);
275
276 int trace_array_get(struct trace_array *this_tr)
277 {
278         struct trace_array *tr;
279         int ret = -ENODEV;
280
281         mutex_lock(&trace_types_lock);
282         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
283                 if (tr == this_tr) {
284                         tr->ref++;
285                         ret = 0;
286                         break;
287                 }
288         }
289         mutex_unlock(&trace_types_lock);
290
291         return ret;
292 }
293
294 static void __trace_array_put(struct trace_array *this_tr)
295 {
296         WARN_ON(!this_tr->ref);
297         this_tr->ref--;
298 }
299
300 void trace_array_put(struct trace_array *this_tr)
301 {
302         mutex_lock(&trace_types_lock);
303         __trace_array_put(this_tr);
304         mutex_unlock(&trace_types_lock);
305 }
306
307 int call_filter_check_discard(struct trace_event_call *call, void *rec,
308                               struct ring_buffer *buffer,
309                               struct ring_buffer_event *event)
310 {
311         if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
312             !filter_match_preds(call->filter, rec)) {
313                 __trace_event_discard_commit(buffer, event);
314                 return 1;
315         }
316
317         return 0;
318 }
319
320 void trace_free_pid_list(struct trace_pid_list *pid_list)
321 {
322         vfree(pid_list->pids);
323         kfree(pid_list);
324 }
325
326 /**
327  * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
328  * @filtered_pids: The list of pids to check
329  * @search_pid: The PID to find in @filtered_pids
330  *
331  * Returns true if @search_pid is fonud in @filtered_pids, and false otherwis.
332  */
333 bool
334 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
335 {
336         /*
337          * If pid_max changed after filtered_pids was created, we
338          * by default ignore all pids greater than the previous pid_max.
339          */
340         if (search_pid >= filtered_pids->pid_max)
341                 return false;
342
343         return test_bit(search_pid, filtered_pids->pids);
344 }
345
346 /**
347  * trace_ignore_this_task - should a task be ignored for tracing
348  * @filtered_pids: The list of pids to check
349  * @task: The task that should be ignored if not filtered
350  *
351  * Checks if @task should be traced or not from @filtered_pids.
352  * Returns true if @task should *NOT* be traced.
353  * Returns false if @task should be traced.
354  */
355 bool
356 trace_ignore_this_task(struct trace_pid_list *filtered_pids, struct task_struct *task)
357 {
358         /*
359          * Return false, because if filtered_pids does not exist,
360          * all pids are good to trace.
361          */
362         if (!filtered_pids)
363                 return false;
364
365         return !trace_find_filtered_pid(filtered_pids, task->pid);
366 }
367
368 /**
369  * trace_pid_filter_add_remove_task - Add or remove a task from a pid_list
370  * @pid_list: The list to modify
371  * @self: The current task for fork or NULL for exit
372  * @task: The task to add or remove
373  *
374  * If adding a task, if @self is defined, the task is only added if @self
375  * is also included in @pid_list. This happens on fork and tasks should
376  * only be added when the parent is listed. If @self is NULL, then the
377  * @task pid will be removed from the list, which would happen on exit
378  * of a task.
379  */
380 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
381                                   struct task_struct *self,
382                                   struct task_struct *task)
383 {
384         if (!pid_list)
385                 return;
386
387         /* For forks, we only add if the forking task is listed */
388         if (self) {
389                 if (!trace_find_filtered_pid(pid_list, self->pid))
390                         return;
391         }
392
393         /* Sorry, but we don't support pid_max changing after setting */
394         if (task->pid >= pid_list->pid_max)
395                 return;
396
397         /* "self" is set for forks, and NULL for exits */
398         if (self)
399                 set_bit(task->pid, pid_list->pids);
400         else
401                 clear_bit(task->pid, pid_list->pids);
402 }
403
404 /**
405  * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
406  * @pid_list: The pid list to show
407  * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
408  * @pos: The position of the file
409  *
410  * This is used by the seq_file "next" operation to iterate the pids
411  * listed in a trace_pid_list structure.
412  *
413  * Returns the pid+1 as we want to display pid of zero, but NULL would
414  * stop the iteration.
415  */
416 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
417 {
418         unsigned long pid = (unsigned long)v;
419
420         (*pos)++;
421
422         /* pid already is +1 of the actual prevous bit */
423         pid = find_next_bit(pid_list->pids, pid_list->pid_max, pid);
424
425         /* Return pid + 1 to allow zero to be represented */
426         if (pid < pid_list->pid_max)
427                 return (void *)(pid + 1);
428
429         return NULL;
430 }
431
432 /**
433  * trace_pid_start - Used for seq_file to start reading pid lists
434  * @pid_list: The pid list to show
435  * @pos: The position of the file
436  *
437  * This is used by seq_file "start" operation to start the iteration
438  * of listing pids.
439  *
440  * Returns the pid+1 as we want to display pid of zero, but NULL would
441  * stop the iteration.
442  */
443 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
444 {
445         unsigned long pid;
446         loff_t l = 0;
447
448         pid = find_first_bit(pid_list->pids, pid_list->pid_max);
449         if (pid >= pid_list->pid_max)
450                 return NULL;
451
452         /* Return pid + 1 so that zero can be the exit value */
453         for (pid++; pid && l < *pos;
454              pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
455                 ;
456         return (void *)pid;
457 }
458
459 /**
460  * trace_pid_show - show the current pid in seq_file processing
461  * @m: The seq_file structure to write into
462  * @v: A void pointer of the pid (+1) value to display
463  *
464  * Can be directly used by seq_file operations to display the current
465  * pid value.
466  */
467 int trace_pid_show(struct seq_file *m, void *v)
468 {
469         unsigned long pid = (unsigned long)v - 1;
470
471         seq_printf(m, "%lu\n", pid);
472         return 0;
473 }
474
475 /* 128 should be much more than enough */
476 #define PID_BUF_SIZE            127
477
478 int trace_pid_write(struct trace_pid_list *filtered_pids,
479                     struct trace_pid_list **new_pid_list,
480                     const char __user *ubuf, size_t cnt)
481 {
482         struct trace_pid_list *pid_list;
483         struct trace_parser parser;
484         unsigned long val;
485         int nr_pids = 0;
486         ssize_t read = 0;
487         ssize_t ret = 0;
488         loff_t pos;
489         pid_t pid;
490
491         if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
492                 return -ENOMEM;
493
494         /*
495          * Always recreate a new array. The write is an all or nothing
496          * operation. Always create a new array when adding new pids by
497          * the user. If the operation fails, then the current list is
498          * not modified.
499          */
500         pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
501         if (!pid_list) {
502                 trace_parser_put(&parser);
503                 return -ENOMEM;
504         }
505
506         pid_list->pid_max = READ_ONCE(pid_max);
507
508         /* Only truncating will shrink pid_max */
509         if (filtered_pids && filtered_pids->pid_max > pid_list->pid_max)
510                 pid_list->pid_max = filtered_pids->pid_max;
511
512         pid_list->pids = vzalloc((pid_list->pid_max + 7) >> 3);
513         if (!pid_list->pids) {
514                 trace_parser_put(&parser);
515                 kfree(pid_list);
516                 return -ENOMEM;
517         }
518
519         if (filtered_pids) {
520                 /* copy the current bits to the new max */
521                 for_each_set_bit(pid, filtered_pids->pids,
522                                  filtered_pids->pid_max) {
523                         set_bit(pid, pid_list->pids);
524                         nr_pids++;
525                 }
526         }
527
528         while (cnt > 0) {
529
530                 pos = 0;
531
532                 ret = trace_get_user(&parser, ubuf, cnt, &pos);
533                 if (ret < 0 || !trace_parser_loaded(&parser))
534                         break;
535
536                 read += ret;
537                 ubuf += ret;
538                 cnt -= ret;
539
540                 ret = -EINVAL;
541                 if (kstrtoul(parser.buffer, 0, &val))
542                         break;
543                 if (val >= pid_list->pid_max)
544                         break;
545
546                 pid = (pid_t)val;
547
548                 set_bit(pid, pid_list->pids);
549                 nr_pids++;
550
551                 trace_parser_clear(&parser);
552                 ret = 0;
553         }
554         trace_parser_put(&parser);
555
556         if (ret < 0) {
557                 trace_free_pid_list(pid_list);
558                 return ret;
559         }
560
561         if (!nr_pids) {
562                 /* Cleared the list of pids */
563                 trace_free_pid_list(pid_list);
564                 read = ret;
565                 pid_list = NULL;
566         }
567
568         *new_pid_list = pid_list;
569
570         return read;
571 }
572
573 static u64 buffer_ftrace_now(struct trace_buffer *buf, int cpu)
574 {
575         u64 ts;
576
577         /* Early boot up does not have a buffer yet */
578         if (!buf->buffer)
579                 return trace_clock_local();
580
581         ts = ring_buffer_time_stamp(buf->buffer, cpu);
582         ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
583
584         return ts;
585 }
586
587 u64 ftrace_now(int cpu)
588 {
589         return buffer_ftrace_now(&global_trace.trace_buffer, cpu);
590 }
591
592 /**
593  * tracing_is_enabled - Show if global_trace has been disabled
594  *
595  * Shows if the global trace has been enabled or not. It uses the
596  * mirror flag "buffer_disabled" to be used in fast paths such as for
597  * the irqsoff tracer. But it may be inaccurate due to races. If you
598  * need to know the accurate state, use tracing_is_on() which is a little
599  * slower, but accurate.
600  */
601 int tracing_is_enabled(void)
602 {
603         /*
604          * For quick access (irqsoff uses this in fast path), just
605          * return the mirror variable of the state of the ring buffer.
606          * It's a little racy, but we don't really care.
607          */
608         smp_rmb();
609         return !global_trace.buffer_disabled;
610 }
611
612 /*
613  * trace_buf_size is the size in bytes that is allocated
614  * for a buffer. Note, the number of bytes is always rounded
615  * to page size.
616  *
617  * This number is purposely set to a low number of 16384.
618  * If the dump on oops happens, it will be much appreciated
619  * to not have to wait for all that output. Anyway this can be
620  * boot time and run time configurable.
621  */
622 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
623
624 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
625
626 /* trace_types holds a link list of available tracers. */
627 static struct tracer            *trace_types __read_mostly;
628
629 /*
630  * trace_types_lock is used to protect the trace_types list.
631  */
632 DEFINE_MUTEX(trace_types_lock);
633
634 /*
635  * serialize the access of the ring buffer
636  *
637  * ring buffer serializes readers, but it is low level protection.
638  * The validity of the events (which returns by ring_buffer_peek() ..etc)
639  * are not protected by ring buffer.
640  *
641  * The content of events may become garbage if we allow other process consumes
642  * these events concurrently:
643  *   A) the page of the consumed events may become a normal page
644  *      (not reader page) in ring buffer, and this page will be rewrited
645  *      by events producer.
646  *   B) The page of the consumed events may become a page for splice_read,
647  *      and this page will be returned to system.
648  *
649  * These primitives allow multi process access to different cpu ring buffer
650  * concurrently.
651  *
652  * These primitives don't distinguish read-only and read-consume access.
653  * Multi read-only access are also serialized.
654  */
655
656 #ifdef CONFIG_SMP
657 static DECLARE_RWSEM(all_cpu_access_lock);
658 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
659
660 static inline void trace_access_lock(int cpu)
661 {
662         if (cpu == RING_BUFFER_ALL_CPUS) {
663                 /* gain it for accessing the whole ring buffer. */
664                 down_write(&all_cpu_access_lock);
665         } else {
666                 /* gain it for accessing a cpu ring buffer. */
667
668                 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
669                 down_read(&all_cpu_access_lock);
670
671                 /* Secondly block other access to this @cpu ring buffer. */
672                 mutex_lock(&per_cpu(cpu_access_lock, cpu));
673         }
674 }
675
676 static inline void trace_access_unlock(int cpu)
677 {
678         if (cpu == RING_BUFFER_ALL_CPUS) {
679                 up_write(&all_cpu_access_lock);
680         } else {
681                 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
682                 up_read(&all_cpu_access_lock);
683         }
684 }
685
686 static inline void trace_access_lock_init(void)
687 {
688         int cpu;
689
690         for_each_possible_cpu(cpu)
691                 mutex_init(&per_cpu(cpu_access_lock, cpu));
692 }
693
694 #else
695
696 static DEFINE_MUTEX(access_lock);
697
698 static inline void trace_access_lock(int cpu)
699 {
700         (void)cpu;
701         mutex_lock(&access_lock);
702 }
703
704 static inline void trace_access_unlock(int cpu)
705 {
706         (void)cpu;
707         mutex_unlock(&access_lock);
708 }
709
710 static inline void trace_access_lock_init(void)
711 {
712 }
713
714 #endif
715
716 #ifdef CONFIG_STACKTRACE
717 static void __ftrace_trace_stack(struct ring_buffer *buffer,
718                                  unsigned long flags,
719                                  int skip, int pc, struct pt_regs *regs);
720 static inline void ftrace_trace_stack(struct trace_array *tr,
721                                       struct ring_buffer *buffer,
722                                       unsigned long flags,
723                                       int skip, int pc, struct pt_regs *regs);
724
725 #else
726 static inline void __ftrace_trace_stack(struct ring_buffer *buffer,
727                                         unsigned long flags,
728                                         int skip, int pc, struct pt_regs *regs)
729 {
730 }
731 static inline void ftrace_trace_stack(struct trace_array *tr,
732                                       struct ring_buffer *buffer,
733                                       unsigned long flags,
734                                       int skip, int pc, struct pt_regs *regs)
735 {
736 }
737
738 #endif
739
740 static __always_inline void
741 trace_event_setup(struct ring_buffer_event *event,
742                   int type, unsigned long flags, int pc)
743 {
744         struct trace_entry *ent = ring_buffer_event_data(event);
745
746         tracing_generic_entry_update(ent, flags, pc);
747         ent->type = type;
748 }
749
750 static __always_inline struct ring_buffer_event *
751 __trace_buffer_lock_reserve(struct ring_buffer *buffer,
752                           int type,
753                           unsigned long len,
754                           unsigned long flags, int pc)
755 {
756         struct ring_buffer_event *event;
757
758         event = ring_buffer_lock_reserve(buffer, len);
759         if (event != NULL)
760                 trace_event_setup(event, type, flags, pc);
761
762         return event;
763 }
764
765 void tracer_tracing_on(struct trace_array *tr)
766 {
767         if (tr->trace_buffer.buffer)
768                 ring_buffer_record_on(tr->trace_buffer.buffer);
769         /*
770          * This flag is looked at when buffers haven't been allocated
771          * yet, or by some tracers (like irqsoff), that just want to
772          * know if the ring buffer has been disabled, but it can handle
773          * races of where it gets disabled but we still do a record.
774          * As the check is in the fast path of the tracers, it is more
775          * important to be fast than accurate.
776          */
777         tr->buffer_disabled = 0;
778         /* Make the flag seen by readers */
779         smp_wmb();
780 }
781
782 /**
783  * tracing_on - enable tracing buffers
784  *
785  * This function enables tracing buffers that may have been
786  * disabled with tracing_off.
787  */
788 void tracing_on(void)
789 {
790         tracer_tracing_on(&global_trace);
791 }
792 EXPORT_SYMBOL_GPL(tracing_on);
793
794
795 static __always_inline void
796 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
797 {
798         __this_cpu_write(trace_taskinfo_save, true);
799
800         /* If this is the temp buffer, we need to commit fully */
801         if (this_cpu_read(trace_buffered_event) == event) {
802                 /* Length is in event->array[0] */
803                 ring_buffer_write(buffer, event->array[0], &event->array[1]);
804                 /* Release the temp buffer */
805                 this_cpu_dec(trace_buffered_event_cnt);
806         } else
807                 ring_buffer_unlock_commit(buffer, event);
808 }
809
810 /**
811  * __trace_puts - write a constant string into the trace buffer.
812  * @ip:    The address of the caller
813  * @str:   The constant string to write
814  * @size:  The size of the string.
815  */
816 int __trace_puts(unsigned long ip, const char *str, int size)
817 {
818         struct ring_buffer_event *event;
819         struct ring_buffer *buffer;
820         struct print_entry *entry;
821         unsigned long irq_flags;
822         int alloc;
823         int pc;
824
825         if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
826                 return 0;
827
828         pc = preempt_count();
829
830         if (unlikely(tracing_selftest_running || tracing_disabled))
831                 return 0;
832
833         alloc = sizeof(*entry) + size + 2; /* possible \n added */
834
835         local_save_flags(irq_flags);
836         buffer = global_trace.trace_buffer.buffer;
837         event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc, 
838                                             irq_flags, pc);
839         if (!event)
840                 return 0;
841
842         entry = ring_buffer_event_data(event);
843         entry->ip = ip;
844
845         memcpy(&entry->buf, str, size);
846
847         /* Add a newline if necessary */
848         if (entry->buf[size - 1] != '\n') {
849                 entry->buf[size] = '\n';
850                 entry->buf[size + 1] = '\0';
851         } else
852                 entry->buf[size] = '\0';
853
854         __buffer_unlock_commit(buffer, event);
855         ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
856
857         return size;
858 }
859 EXPORT_SYMBOL_GPL(__trace_puts);
860
861 /**
862  * __trace_bputs - write the pointer to a constant string into trace buffer
863  * @ip:    The address of the caller
864  * @str:   The constant string to write to the buffer to
865  */
866 int __trace_bputs(unsigned long ip, const char *str)
867 {
868         struct ring_buffer_event *event;
869         struct ring_buffer *buffer;
870         struct bputs_entry *entry;
871         unsigned long irq_flags;
872         int size = sizeof(struct bputs_entry);
873         int pc;
874
875         if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
876                 return 0;
877
878         pc = preempt_count();
879
880         if (unlikely(tracing_selftest_running || tracing_disabled))
881                 return 0;
882
883         local_save_flags(irq_flags);
884         buffer = global_trace.trace_buffer.buffer;
885         event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
886                                             irq_flags, pc);
887         if (!event)
888                 return 0;
889
890         entry = ring_buffer_event_data(event);
891         entry->ip                       = ip;
892         entry->str                      = str;
893
894         __buffer_unlock_commit(buffer, event);
895         ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
896
897         return 1;
898 }
899 EXPORT_SYMBOL_GPL(__trace_bputs);
900
901 #ifdef CONFIG_TRACER_SNAPSHOT
902 void tracing_snapshot_instance_cond(struct trace_array *tr, void *cond_data)
903 {
904         struct tracer *tracer = tr->current_trace;
905         unsigned long flags;
906
907         if (in_nmi()) {
908                 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
909                 internal_trace_puts("*** snapshot is being ignored        ***\n");
910                 return;
911         }
912
913         if (!tr->allocated_snapshot) {
914                 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
915                 internal_trace_puts("*** stopping trace here!   ***\n");
916                 tracing_off();
917                 return;
918         }
919
920         /* Note, snapshot can not be used when the tracer uses it */
921         if (tracer->use_max_tr) {
922                 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
923                 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
924                 return;
925         }
926
927         local_irq_save(flags);
928         update_max_tr(tr, current, smp_processor_id(), cond_data);
929         local_irq_restore(flags);
930 }
931
932 void tracing_snapshot_instance(struct trace_array *tr)
933 {
934         tracing_snapshot_instance_cond(tr, NULL);
935 }
936
937 /**
938  * tracing_snapshot - take a snapshot of the current buffer.
939  *
940  * This causes a swap between the snapshot buffer and the current live
941  * tracing buffer. You can use this to take snapshots of the live
942  * trace when some condition is triggered, but continue to trace.
943  *
944  * Note, make sure to allocate the snapshot with either
945  * a tracing_snapshot_alloc(), or by doing it manually
946  * with: echo 1 > /sys/kernel/debug/tracing/snapshot
947  *
948  * If the snapshot buffer is not allocated, it will stop tracing.
949  * Basically making a permanent snapshot.
950  */
951 void tracing_snapshot(void)
952 {
953         struct trace_array *tr = &global_trace;
954
955         tracing_snapshot_instance(tr);
956 }
957 EXPORT_SYMBOL_GPL(tracing_snapshot);
958
959 /**
960  * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
961  * @tr:         The tracing instance to snapshot
962  * @cond_data:  The data to be tested conditionally, and possibly saved
963  *
964  * This is the same as tracing_snapshot() except that the snapshot is
965  * conditional - the snapshot will only happen if the
966  * cond_snapshot.update() implementation receiving the cond_data
967  * returns true, which means that the trace array's cond_snapshot
968  * update() operation used the cond_data to determine whether the
969  * snapshot should be taken, and if it was, presumably saved it along
970  * with the snapshot.
971  */
972 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
973 {
974         tracing_snapshot_instance_cond(tr, cond_data);
975 }
976 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
977
978 /**
979  * tracing_snapshot_cond_data - get the user data associated with a snapshot
980  * @tr:         The tracing instance
981  *
982  * When the user enables a conditional snapshot using
983  * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
984  * with the snapshot.  This accessor is used to retrieve it.
985  *
986  * Should not be called from cond_snapshot.update(), since it takes
987  * the tr->max_lock lock, which the code calling
988  * cond_snapshot.update() has already done.
989  *
990  * Returns the cond_data associated with the trace array's snapshot.
991  */
992 void *tracing_cond_snapshot_data(struct trace_array *tr)
993 {
994         void *cond_data = NULL;
995
996         arch_spin_lock(&tr->max_lock);
997
998         if (tr->cond_snapshot)
999                 cond_data = tr->cond_snapshot->cond_data;
1000
1001         arch_spin_unlock(&tr->max_lock);
1002
1003         return cond_data;
1004 }
1005 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1006
1007 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
1008                                         struct trace_buffer *size_buf, int cpu_id);
1009 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val);
1010
1011 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1012 {
1013         int ret;
1014
1015         if (!tr->allocated_snapshot) {
1016
1017                 /* allocate spare buffer */
1018                 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1019                                    &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
1020                 if (ret < 0)
1021                         return ret;
1022
1023                 tr->allocated_snapshot = true;
1024         }
1025
1026         return 0;
1027 }
1028
1029 static void free_snapshot(struct trace_array *tr)
1030 {
1031         /*
1032          * We don't free the ring buffer. instead, resize it because
1033          * The max_tr ring buffer has some state (e.g. ring->clock) and
1034          * we want preserve it.
1035          */
1036         ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1037         set_buffer_entries(&tr->max_buffer, 1);
1038         tracing_reset_online_cpus(&tr->max_buffer);
1039         tr->allocated_snapshot = false;
1040 }
1041
1042 /**
1043  * tracing_alloc_snapshot - allocate snapshot buffer.
1044  *
1045  * This only allocates the snapshot buffer if it isn't already
1046  * allocated - it doesn't also take a snapshot.
1047  *
1048  * This is meant to be used in cases where the snapshot buffer needs
1049  * to be set up for events that can't sleep but need to be able to
1050  * trigger a snapshot.
1051  */
1052 int tracing_alloc_snapshot(void)
1053 {
1054         struct trace_array *tr = &global_trace;
1055         int ret;
1056
1057         ret = tracing_alloc_snapshot_instance(tr);
1058         WARN_ON(ret < 0);
1059
1060         return ret;
1061 }
1062 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1063
1064 /**
1065  * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1066  *
1067  * This is similar to tracing_snapshot(), but it will allocate the
1068  * snapshot buffer if it isn't already allocated. Use this only
1069  * where it is safe to sleep, as the allocation may sleep.
1070  *
1071  * This causes a swap between the snapshot buffer and the current live
1072  * tracing buffer. You can use this to take snapshots of the live
1073  * trace when some condition is triggered, but continue to trace.
1074  */
1075 void tracing_snapshot_alloc(void)
1076 {
1077         int ret;
1078
1079         ret = tracing_alloc_snapshot();
1080         if (ret < 0)
1081                 return;
1082
1083         tracing_snapshot();
1084 }
1085 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1086
1087 /**
1088  * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1089  * @tr:         The tracing instance
1090  * @cond_data:  User data to associate with the snapshot
1091  * @update:     Implementation of the cond_snapshot update function
1092  *
1093  * Check whether the conditional snapshot for the given instance has
1094  * already been enabled, or if the current tracer is already using a
1095  * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1096  * save the cond_data and update function inside.
1097  *
1098  * Returns 0 if successful, error otherwise.
1099  */
1100 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1101                                  cond_update_fn_t update)
1102 {
1103         struct cond_snapshot *cond_snapshot;
1104         int ret = 0;
1105
1106         cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1107         if (!cond_snapshot)
1108                 return -ENOMEM;
1109
1110         cond_snapshot->cond_data = cond_data;
1111         cond_snapshot->update = update;
1112
1113         mutex_lock(&trace_types_lock);
1114
1115         ret = tracing_alloc_snapshot_instance(tr);
1116         if (ret)
1117                 goto fail_unlock;
1118
1119         if (tr->current_trace->use_max_tr) {
1120                 ret = -EBUSY;
1121                 goto fail_unlock;
1122         }
1123
1124         /*
1125          * The cond_snapshot can only change to NULL without the
1126          * trace_types_lock. We don't care if we race with it going
1127          * to NULL, but we want to make sure that it's not set to
1128          * something other than NULL when we get here, which we can
1129          * do safely with only holding the trace_types_lock and not
1130          * having to take the max_lock.
1131          */
1132         if (tr->cond_snapshot) {
1133                 ret = -EBUSY;
1134                 goto fail_unlock;
1135         }
1136
1137         arch_spin_lock(&tr->max_lock);
1138         tr->cond_snapshot = cond_snapshot;
1139         arch_spin_unlock(&tr->max_lock);
1140
1141         mutex_unlock(&trace_types_lock);
1142
1143         return ret;
1144
1145  fail_unlock:
1146         mutex_unlock(&trace_types_lock);
1147         kfree(cond_snapshot);
1148         return ret;
1149 }
1150 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1151
1152 /**
1153  * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1154  * @tr:         The tracing instance
1155  *
1156  * Check whether the conditional snapshot for the given instance is
1157  * enabled; if so, free the cond_snapshot associated with it,
1158  * otherwise return -EINVAL.
1159  *
1160  * Returns 0 if successful, error otherwise.
1161  */
1162 int tracing_snapshot_cond_disable(struct trace_array *tr)
1163 {
1164         int ret = 0;
1165
1166         arch_spin_lock(&tr->max_lock);
1167
1168         if (!tr->cond_snapshot)
1169                 ret = -EINVAL;
1170         else {
1171                 kfree(tr->cond_snapshot);
1172                 tr->cond_snapshot = NULL;
1173         }
1174
1175         arch_spin_unlock(&tr->max_lock);
1176
1177         return ret;
1178 }
1179 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1180 #else
1181 void tracing_snapshot(void)
1182 {
1183         WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1184 }
1185 EXPORT_SYMBOL_GPL(tracing_snapshot);
1186 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1187 {
1188         WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1189 }
1190 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1191 int tracing_alloc_snapshot(void)
1192 {
1193         WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1194         return -ENODEV;
1195 }
1196 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1197 void tracing_snapshot_alloc(void)
1198 {
1199         /* Give warning */
1200         tracing_snapshot();
1201 }
1202 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1203 void *tracing_cond_snapshot_data(struct trace_array *tr)
1204 {
1205         return NULL;
1206 }
1207 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1208 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1209 {
1210         return -ENODEV;
1211 }
1212 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1213 int tracing_snapshot_cond_disable(struct trace_array *tr)
1214 {
1215         return false;
1216 }
1217 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1218 #endif /* CONFIG_TRACER_SNAPSHOT */
1219
1220 void tracer_tracing_off(struct trace_array *tr)
1221 {
1222         if (tr->trace_buffer.buffer)
1223                 ring_buffer_record_off(tr->trace_buffer.buffer);
1224         /*
1225          * This flag is looked at when buffers haven't been allocated
1226          * yet, or by some tracers (like irqsoff), that just want to
1227          * know if the ring buffer has been disabled, but it can handle
1228          * races of where it gets disabled but we still do a record.
1229          * As the check is in the fast path of the tracers, it is more
1230          * important to be fast than accurate.
1231          */
1232         tr->buffer_disabled = 1;
1233         /* Make the flag seen by readers */
1234         smp_wmb();
1235 }
1236
1237 /**
1238  * tracing_off - turn off tracing buffers
1239  *
1240  * This function stops the tracing buffers from recording data.
1241  * It does not disable any overhead the tracers themselves may
1242  * be causing. This function simply causes all recording to
1243  * the ring buffers to fail.
1244  */
1245 void tracing_off(void)
1246 {
1247         tracer_tracing_off(&global_trace);
1248 }
1249 EXPORT_SYMBOL_GPL(tracing_off);
1250
1251 void disable_trace_on_warning(void)
1252 {
1253         if (__disable_trace_on_warning)
1254                 tracing_off();
1255 }
1256
1257 /**
1258  * tracer_tracing_is_on - show real state of ring buffer enabled
1259  * @tr : the trace array to know if ring buffer is enabled
1260  *
1261  * Shows real state of the ring buffer if it is enabled or not.
1262  */
1263 bool tracer_tracing_is_on(struct trace_array *tr)
1264 {
1265         if (tr->trace_buffer.buffer)
1266                 return ring_buffer_record_is_on(tr->trace_buffer.buffer);
1267         return !tr->buffer_disabled;
1268 }
1269
1270 /**
1271  * tracing_is_on - show state of ring buffers enabled
1272  */
1273 int tracing_is_on(void)
1274 {
1275         return tracer_tracing_is_on(&global_trace);
1276 }
1277 EXPORT_SYMBOL_GPL(tracing_is_on);
1278
1279 static int __init set_buf_size(char *str)
1280 {
1281         unsigned long buf_size;
1282
1283         if (!str)
1284                 return 0;
1285         buf_size = memparse(str, &str);
1286         /* nr_entries can not be zero */
1287         if (buf_size == 0)
1288                 return 0;
1289         trace_buf_size = buf_size;
1290         return 1;
1291 }
1292 __setup("trace_buf_size=", set_buf_size);
1293
1294 static int __init set_tracing_thresh(char *str)
1295 {
1296         unsigned long threshold;
1297         int ret;
1298
1299         if (!str)
1300                 return 0;
1301         ret = kstrtoul(str, 0, &threshold);
1302         if (ret < 0)
1303                 return 0;
1304         tracing_thresh = threshold * 1000;
1305         return 1;
1306 }
1307 __setup("tracing_thresh=", set_tracing_thresh);
1308
1309 unsigned long nsecs_to_usecs(unsigned long nsecs)
1310 {
1311         return nsecs / 1000;
1312 }
1313
1314 /*
1315  * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1316  * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1317  * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1318  * of strings in the order that the evals (enum) were defined.
1319  */
1320 #undef C
1321 #define C(a, b) b
1322
1323 /* These must match the bit postions in trace_iterator_flags */
1324 static const char *trace_options[] = {
1325         TRACE_FLAGS
1326         NULL
1327 };
1328
1329 static struct {
1330         u64 (*func)(void);
1331         const char *name;
1332         int in_ns;              /* is this clock in nanoseconds? */
1333 } trace_clocks[] = {
1334         { trace_clock_local,            "local",        1 },
1335         { trace_clock_global,           "global",       1 },
1336         { trace_clock_counter,          "counter",      0 },
1337         { trace_clock_jiffies,          "uptime",       0 },
1338         { trace_clock,                  "perf",         1 },
1339         { ktime_get_mono_fast_ns,       "mono",         1 },
1340         { ktime_get_raw_fast_ns,        "mono_raw",     1 },
1341         { ktime_get_boot_fast_ns,       "boot",         1 },
1342         ARCH_TRACE_CLOCKS
1343 };
1344
1345 bool trace_clock_in_ns(struct trace_array *tr)
1346 {
1347         if (trace_clocks[tr->clock_id].in_ns)
1348                 return true;
1349
1350         return false;
1351 }
1352
1353 /*
1354  * trace_parser_get_init - gets the buffer for trace parser
1355  */
1356 int trace_parser_get_init(struct trace_parser *parser, int size)
1357 {
1358         memset(parser, 0, sizeof(*parser));
1359
1360         parser->buffer = kmalloc(size, GFP_KERNEL);
1361         if (!parser->buffer)
1362                 return 1;
1363
1364         parser->size = size;
1365         return 0;
1366 }
1367
1368 /*
1369  * trace_parser_put - frees the buffer for trace parser
1370  */
1371 void trace_parser_put(struct trace_parser *parser)
1372 {
1373         kfree(parser->buffer);
1374         parser->buffer = NULL;
1375 }
1376
1377 /*
1378  * trace_get_user - reads the user input string separated by  space
1379  * (matched by isspace(ch))
1380  *
1381  * For each string found the 'struct trace_parser' is updated,
1382  * and the function returns.
1383  *
1384  * Returns number of bytes read.
1385  *
1386  * See kernel/trace/trace.h for 'struct trace_parser' details.
1387  */
1388 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1389         size_t cnt, loff_t *ppos)
1390 {
1391         char ch;
1392         size_t read = 0;
1393         ssize_t ret;
1394
1395         if (!*ppos)
1396                 trace_parser_clear(parser);
1397
1398         ret = get_user(ch, ubuf++);
1399         if (ret)
1400                 goto out;
1401
1402         read++;
1403         cnt--;
1404
1405         /*
1406          * The parser is not finished with the last write,
1407          * continue reading the user input without skipping spaces.
1408          */
1409         if (!parser->cont) {
1410                 /* skip white space */
1411                 while (cnt && isspace(ch)) {
1412                         ret = get_user(ch, ubuf++);
1413                         if (ret)
1414                                 goto out;
1415                         read++;
1416                         cnt--;
1417                 }
1418
1419                 parser->idx = 0;
1420
1421                 /* only spaces were written */
1422                 if (isspace(ch) || !ch) {
1423                         *ppos += read;
1424                         ret = read;
1425                         goto out;
1426                 }
1427         }
1428
1429         /* read the non-space input */
1430         while (cnt && !isspace(ch) && ch) {
1431                 if (parser->idx < parser->size - 1)
1432                         parser->buffer[parser->idx++] = ch;
1433                 else {
1434                         ret = -EINVAL;
1435                         goto out;
1436                 }
1437                 ret = get_user(ch, ubuf++);
1438                 if (ret)
1439                         goto out;
1440                 read++;
1441                 cnt--;
1442         }
1443
1444         /* We either got finished input or we have to wait for another call. */
1445         if (isspace(ch) || !ch) {
1446                 parser->buffer[parser->idx] = 0;
1447                 parser->cont = false;
1448         } else if (parser->idx < parser->size - 1) {
1449                 parser->cont = true;
1450                 parser->buffer[parser->idx++] = ch;
1451                 /* Make sure the parsed string always terminates with '\0'. */
1452                 parser->buffer[parser->idx] = 0;
1453         } else {
1454                 ret = -EINVAL;
1455                 goto out;
1456         }
1457
1458         *ppos += read;
1459         ret = read;
1460
1461 out:
1462         return ret;
1463 }
1464
1465 /* TODO add a seq_buf_to_buffer() */
1466 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1467 {
1468         int len;
1469
1470         if (trace_seq_used(s) <= s->seq.readpos)
1471                 return -EBUSY;
1472
1473         len = trace_seq_used(s) - s->seq.readpos;
1474         if (cnt > len)
1475                 cnt = len;
1476         memcpy(buf, s->buffer + s->seq.readpos, cnt);
1477
1478         s->seq.readpos += cnt;
1479         return cnt;
1480 }
1481
1482 unsigned long __read_mostly     tracing_thresh;
1483
1484 #ifdef CONFIG_TRACER_MAX_TRACE
1485 /*
1486  * Copy the new maximum trace into the separate maximum-trace
1487  * structure. (this way the maximum trace is permanently saved,
1488  * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1489  */
1490 static void
1491 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1492 {
1493         struct trace_buffer *trace_buf = &tr->trace_buffer;
1494         struct trace_buffer *max_buf = &tr->max_buffer;
1495         struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1496         struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1497
1498         max_buf->cpu = cpu;
1499         max_buf->time_start = data->preempt_timestamp;
1500
1501         max_data->saved_latency = tr->max_latency;
1502         max_data->critical_start = data->critical_start;
1503         max_data->critical_end = data->critical_end;
1504
1505         strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1506         max_data->pid = tsk->pid;
1507         /*
1508          * If tsk == current, then use current_uid(), as that does not use
1509          * RCU. The irq tracer can be called out of RCU scope.
1510          */
1511         if (tsk == current)
1512                 max_data->uid = current_uid();
1513         else
1514                 max_data->uid = task_uid(tsk);
1515
1516         max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1517         max_data->policy = tsk->policy;
1518         max_data->rt_priority = tsk->rt_priority;
1519
1520         /* record this tasks comm */
1521         tracing_record_cmdline(tsk);
1522 }
1523
1524 /**
1525  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1526  * @tr: tracer
1527  * @tsk: the task with the latency
1528  * @cpu: The cpu that initiated the trace.
1529  * @cond_data: User data associated with a conditional snapshot
1530  *
1531  * Flip the buffers between the @tr and the max_tr and record information
1532  * about which task was the cause of this latency.
1533  */
1534 void
1535 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1536               void *cond_data)
1537 {
1538         if (tr->stop_count)
1539                 return;
1540
1541         WARN_ON_ONCE(!irqs_disabled());
1542
1543         if (!tr->allocated_snapshot) {
1544                 /* Only the nop tracer should hit this when disabling */
1545                 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1546                 return;
1547         }
1548
1549         arch_spin_lock(&tr->max_lock);
1550
1551         /* Inherit the recordable setting from trace_buffer */
1552         if (ring_buffer_record_is_set_on(tr->trace_buffer.buffer))
1553                 ring_buffer_record_on(tr->max_buffer.buffer);
1554         else
1555                 ring_buffer_record_off(tr->max_buffer.buffer);
1556
1557 #ifdef CONFIG_TRACER_SNAPSHOT
1558         if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data))
1559                 goto out_unlock;
1560 #endif
1561         swap(tr->trace_buffer.buffer, tr->max_buffer.buffer);
1562
1563         __update_max_tr(tr, tsk, cpu);
1564
1565  out_unlock:
1566         arch_spin_unlock(&tr->max_lock);
1567 }
1568
1569 /**
1570  * update_max_tr_single - only copy one trace over, and reset the rest
1571  * @tr - tracer
1572  * @tsk - task with the latency
1573  * @cpu - the cpu of the buffer to copy.
1574  *
1575  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1576  */
1577 void
1578 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1579 {
1580         int ret;
1581
1582         if (tr->stop_count)
1583                 return;
1584
1585         WARN_ON_ONCE(!irqs_disabled());
1586         if (!tr->allocated_snapshot) {
1587                 /* Only the nop tracer should hit this when disabling */
1588                 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1589                 return;
1590         }
1591
1592         arch_spin_lock(&tr->max_lock);
1593
1594         ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1595
1596         if (ret == -EBUSY) {
1597                 /*
1598                  * We failed to swap the buffer due to a commit taking
1599                  * place on this CPU. We fail to record, but we reset
1600                  * the max trace buffer (no one writes directly to it)
1601                  * and flag that it failed.
1602                  */
1603                 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1604                         "Failed to swap buffers due to commit in progress\n");
1605         }
1606
1607         WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1608
1609         __update_max_tr(tr, tsk, cpu);
1610         arch_spin_unlock(&tr->max_lock);
1611 }
1612 #endif /* CONFIG_TRACER_MAX_TRACE */
1613
1614 static int wait_on_pipe(struct trace_iterator *iter, int full)
1615 {
1616         /* Iterators are static, they should be filled or empty */
1617         if (trace_buffer_iter(iter, iter->cpu_file))
1618                 return 0;
1619
1620         return ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file,
1621                                 full);
1622 }
1623
1624 #ifdef CONFIG_FTRACE_STARTUP_TEST
1625 static bool selftests_can_run;
1626
1627 struct trace_selftests {
1628         struct list_head                list;
1629         struct tracer                   *type;
1630 };
1631
1632 static LIST_HEAD(postponed_selftests);
1633
1634 static int save_selftest(struct tracer *type)
1635 {
1636         struct trace_selftests *selftest;
1637
1638         selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1639         if (!selftest)
1640                 return -ENOMEM;
1641
1642         selftest->type = type;
1643         list_add(&selftest->list, &postponed_selftests);
1644         return 0;
1645 }
1646
1647 static int run_tracer_selftest(struct tracer *type)
1648 {
1649         struct trace_array *tr = &global_trace;
1650         struct tracer *saved_tracer = tr->current_trace;
1651         int ret;
1652
1653         if (!type->selftest || tracing_selftest_disabled)
1654                 return 0;
1655
1656         /*
1657          * If a tracer registers early in boot up (before scheduling is
1658          * initialized and such), then do not run its selftests yet.
1659          * Instead, run it a little later in the boot process.
1660          */
1661         if (!selftests_can_run)
1662                 return save_selftest(type);
1663
1664         /*
1665          * Run a selftest on this tracer.
1666          * Here we reset the trace buffer, and set the current
1667          * tracer to be this tracer. The tracer can then run some
1668          * internal tracing to verify that everything is in order.
1669          * If we fail, we do not register this tracer.
1670          */
1671         tracing_reset_online_cpus(&tr->trace_buffer);
1672
1673         tr->current_trace = type;
1674
1675 #ifdef CONFIG_TRACER_MAX_TRACE
1676         if (type->use_max_tr) {
1677                 /* If we expanded the buffers, make sure the max is expanded too */
1678                 if (ring_buffer_expanded)
1679                         ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1680                                            RING_BUFFER_ALL_CPUS);
1681                 tr->allocated_snapshot = true;
1682         }
1683 #endif
1684
1685         /* the test is responsible for initializing and enabling */
1686         pr_info("Testing tracer %s: ", type->name);
1687         ret = type->selftest(type, tr);
1688         /* the test is responsible for resetting too */
1689         tr->current_trace = saved_tracer;
1690         if (ret) {
1691                 printk(KERN_CONT "FAILED!\n");
1692                 /* Add the warning after printing 'FAILED' */
1693                 WARN_ON(1);
1694                 return -1;
1695         }
1696         /* Only reset on passing, to avoid touching corrupted buffers */
1697         tracing_reset_online_cpus(&tr->trace_buffer);
1698
1699 #ifdef CONFIG_TRACER_MAX_TRACE
1700         if (type->use_max_tr) {
1701                 tr->allocated_snapshot = false;
1702
1703                 /* Shrink the max buffer again */
1704                 if (ring_buffer_expanded)
1705                         ring_buffer_resize(tr->max_buffer.buffer, 1,
1706                                            RING_BUFFER_ALL_CPUS);
1707         }
1708 #endif
1709
1710         printk(KERN_CONT "PASSED\n");
1711         return 0;
1712 }
1713
1714 static __init int init_trace_selftests(void)
1715 {
1716         struct trace_selftests *p, *n;
1717         struct tracer *t, **last;
1718         int ret;
1719
1720         selftests_can_run = true;
1721
1722         mutex_lock(&trace_types_lock);
1723
1724         if (list_empty(&postponed_selftests))
1725                 goto out;
1726
1727         pr_info("Running postponed tracer tests:\n");
1728
1729         list_for_each_entry_safe(p, n, &postponed_selftests, list) {
1730                 /* This loop can take minutes when sanitizers are enabled, so
1731                  * lets make sure we allow RCU processing.
1732                  */
1733                 cond_resched();
1734                 ret = run_tracer_selftest(p->type);
1735                 /* If the test fails, then warn and remove from available_tracers */
1736                 if (ret < 0) {
1737                         WARN(1, "tracer: %s failed selftest, disabling\n",
1738                              p->type->name);
1739                         last = &trace_types;
1740                         for (t = trace_types; t; t = t->next) {
1741                                 if (t == p->type) {
1742                                         *last = t->next;
1743                                         break;
1744                                 }
1745                                 last = &t->next;
1746                         }
1747                 }
1748                 list_del(&p->list);
1749                 kfree(p);
1750         }
1751
1752  out:
1753         mutex_unlock(&trace_types_lock);
1754
1755         return 0;
1756 }
1757 core_initcall(init_trace_selftests);
1758 #else
1759 static inline int run_tracer_selftest(struct tracer *type)
1760 {
1761         return 0;
1762 }
1763 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1764
1765 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
1766
1767 static void __init apply_trace_boot_options(void);
1768
1769 /**
1770  * register_tracer - register a tracer with the ftrace system.
1771  * @type - the plugin for the tracer
1772  *
1773  * Register a new plugin tracer.
1774  */
1775 int __init register_tracer(struct tracer *type)
1776 {
1777         struct tracer *t;
1778         int ret = 0;
1779
1780         if (!type->name) {
1781                 pr_info("Tracer must have a name\n");
1782                 return -1;
1783         }
1784
1785         if (strlen(type->name) >= MAX_TRACER_SIZE) {
1786                 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1787                 return -1;
1788         }
1789
1790         mutex_lock(&trace_types_lock);
1791
1792         tracing_selftest_running = true;
1793
1794         for (t = trace_types; t; t = t->next) {
1795                 if (strcmp(type->name, t->name) == 0) {
1796                         /* already found */
1797                         pr_info("Tracer %s already registered\n",
1798                                 type->name);
1799                         ret = -1;
1800                         goto out;
1801                 }
1802         }
1803
1804         if (!type->set_flag)
1805                 type->set_flag = &dummy_set_flag;
1806         if (!type->flags) {
1807                 /*allocate a dummy tracer_flags*/
1808                 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
1809                 if (!type->flags) {
1810                         ret = -ENOMEM;
1811                         goto out;
1812                 }
1813                 type->flags->val = 0;
1814                 type->flags->opts = dummy_tracer_opt;
1815         } else
1816                 if (!type->flags->opts)
1817                         type->flags->opts = dummy_tracer_opt;
1818
1819         /* store the tracer for __set_tracer_option */
1820         type->flags->trace = type;
1821
1822         ret = run_tracer_selftest(type);
1823         if (ret < 0)
1824                 goto out;
1825
1826         type->next = trace_types;
1827         trace_types = type;
1828         add_tracer_options(&global_trace, type);
1829
1830  out:
1831         tracing_selftest_running = false;
1832         mutex_unlock(&trace_types_lock);
1833
1834         if (ret || !default_bootup_tracer)
1835                 goto out_unlock;
1836
1837         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1838                 goto out_unlock;
1839
1840         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1841         /* Do we want this tracer to start on bootup? */
1842         tracing_set_tracer(&global_trace, type->name);
1843         default_bootup_tracer = NULL;
1844
1845         apply_trace_boot_options();
1846
1847         /* disable other selftests, since this will break it. */
1848         tracing_selftest_disabled = true;
1849 #ifdef CONFIG_FTRACE_STARTUP_TEST
1850         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1851                type->name);
1852 #endif
1853
1854  out_unlock:
1855         return ret;
1856 }
1857
1858 void tracing_reset(struct trace_buffer *buf, int cpu)
1859 {
1860         struct ring_buffer *buffer = buf->buffer;
1861
1862         if (!buffer)
1863                 return;
1864
1865         ring_buffer_record_disable(buffer);
1866
1867         /* Make sure all commits have finished */
1868         synchronize_rcu();
1869         ring_buffer_reset_cpu(buffer, cpu);
1870
1871         ring_buffer_record_enable(buffer);
1872 }
1873
1874 void tracing_reset_online_cpus(struct trace_buffer *buf)
1875 {
1876         struct ring_buffer *buffer = buf->buffer;
1877         int cpu;
1878
1879         if (!buffer)
1880                 return;
1881
1882         ring_buffer_record_disable(buffer);
1883
1884         /* Make sure all commits have finished */
1885         synchronize_rcu();
1886
1887         buf->time_start = buffer_ftrace_now(buf, buf->cpu);
1888
1889         for_each_online_cpu(cpu)
1890                 ring_buffer_reset_cpu(buffer, cpu);
1891
1892         ring_buffer_record_enable(buffer);
1893 }
1894
1895 /* Must have trace_types_lock held */
1896 void tracing_reset_all_online_cpus(void)
1897 {
1898         struct trace_array *tr;
1899
1900         list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1901                 if (!tr->clear_trace)
1902                         continue;
1903                 tr->clear_trace = false;
1904                 tracing_reset_online_cpus(&tr->trace_buffer);
1905 #ifdef CONFIG_TRACER_MAX_TRACE
1906                 tracing_reset_online_cpus(&tr->max_buffer);
1907 #endif
1908         }
1909 }
1910
1911 static int *tgid_map;
1912
1913 #define SAVED_CMDLINES_DEFAULT 128
1914 #define NO_CMDLINE_MAP UINT_MAX
1915 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1916 struct saved_cmdlines_buffer {
1917         unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1918         unsigned *map_cmdline_to_pid;
1919         unsigned cmdline_num;
1920         int cmdline_idx;
1921         char *saved_cmdlines;
1922 };
1923 static struct saved_cmdlines_buffer *savedcmd;
1924
1925 /* temporary disable recording */
1926 static atomic_t trace_record_taskinfo_disabled __read_mostly;
1927
1928 static inline char *get_saved_cmdlines(int idx)
1929 {
1930         return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
1931 }
1932
1933 static inline void set_cmdline(int idx, const char *cmdline)
1934 {
1935         strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
1936 }
1937
1938 static int allocate_cmdlines_buffer(unsigned int val,
1939                                     struct saved_cmdlines_buffer *s)
1940 {
1941         s->map_cmdline_to_pid = kmalloc_array(val,
1942                                               sizeof(*s->map_cmdline_to_pid),
1943                                               GFP_KERNEL);
1944         if (!s->map_cmdline_to_pid)
1945                 return -ENOMEM;
1946
1947         s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
1948         if (!s->saved_cmdlines) {
1949                 kfree(s->map_cmdline_to_pid);
1950                 return -ENOMEM;
1951         }
1952
1953         s->cmdline_idx = 0;
1954         s->cmdline_num = val;
1955         memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
1956                sizeof(s->map_pid_to_cmdline));
1957         memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
1958                val * sizeof(*s->map_cmdline_to_pid));
1959
1960         return 0;
1961 }
1962
1963 static int trace_create_savedcmd(void)
1964 {
1965         int ret;
1966
1967         savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
1968         if (!savedcmd)
1969                 return -ENOMEM;
1970
1971         ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
1972         if (ret < 0) {
1973                 kfree(savedcmd);
1974                 savedcmd = NULL;
1975                 return -ENOMEM;
1976         }
1977
1978         return 0;
1979 }
1980
1981 int is_tracing_stopped(void)
1982 {
1983         return global_trace.stop_count;
1984 }
1985
1986 /**
1987  * tracing_start - quick start of the tracer
1988  *
1989  * If tracing is enabled but was stopped by tracing_stop,
1990  * this will start the tracer back up.
1991  */
1992 void tracing_start(void)
1993 {
1994         struct ring_buffer *buffer;
1995         unsigned long flags;
1996
1997         if (tracing_disabled)
1998                 return;
1999
2000         raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2001         if (--global_trace.stop_count) {
2002                 if (global_trace.stop_count < 0) {
2003                         /* Someone screwed up their debugging */
2004                         WARN_ON_ONCE(1);
2005                         global_trace.stop_count = 0;
2006                 }
2007                 goto out;
2008         }
2009
2010         /* Prevent the buffers from switching */
2011         arch_spin_lock(&global_trace.max_lock);
2012
2013         buffer = global_trace.trace_buffer.buffer;
2014         if (buffer)
2015                 ring_buffer_record_enable(buffer);
2016
2017 #ifdef CONFIG_TRACER_MAX_TRACE
2018         buffer = global_trace.max_buffer.buffer;
2019         if (buffer)
2020                 ring_buffer_record_enable(buffer);
2021 #endif
2022
2023         arch_spin_unlock(&global_trace.max_lock);
2024
2025  out:
2026         raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2027 }
2028
2029 static void tracing_start_tr(struct trace_array *tr)
2030 {
2031         struct ring_buffer *buffer;
2032         unsigned long flags;
2033
2034         if (tracing_disabled)
2035                 return;
2036
2037         /* If global, we need to also start the max tracer */
2038         if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2039                 return tracing_start();
2040
2041         raw_spin_lock_irqsave(&tr->start_lock, flags);
2042
2043         if (--tr->stop_count) {
2044                 if (tr->stop_count < 0) {
2045                         /* Someone screwed up their debugging */
2046                         WARN_ON_ONCE(1);
2047                         tr->stop_count = 0;
2048                 }
2049                 goto out;
2050         }
2051
2052         buffer = tr->trace_buffer.buffer;
2053         if (buffer)
2054                 ring_buffer_record_enable(buffer);
2055
2056  out:
2057         raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2058 }
2059
2060 /**
2061  * tracing_stop - quick stop of the tracer
2062  *
2063  * Light weight way to stop tracing. Use in conjunction with
2064  * tracing_start.
2065  */
2066 void tracing_stop(void)
2067 {
2068         struct ring_buffer *buffer;
2069         unsigned long flags;
2070
2071         raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2072         if (global_trace.stop_count++)
2073                 goto out;
2074
2075         /* Prevent the buffers from switching */
2076         arch_spin_lock(&global_trace.max_lock);
2077
2078         buffer = global_trace.trace_buffer.buffer;
2079         if (buffer)
2080                 ring_buffer_record_disable(buffer);
2081
2082 #ifdef CONFIG_TRACER_MAX_TRACE
2083         buffer = global_trace.max_buffer.buffer;
2084         if (buffer)
2085                 ring_buffer_record_disable(buffer);
2086 #endif
2087
2088         arch_spin_unlock(&global_trace.max_lock);
2089
2090  out:
2091         raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2092 }
2093
2094 static void tracing_stop_tr(struct trace_array *tr)
2095 {
2096         struct ring_buffer *buffer;
2097         unsigned long flags;
2098
2099         /* If global, we need to also stop the max tracer */
2100         if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2101                 return tracing_stop();
2102
2103         raw_spin_lock_irqsave(&tr->start_lock, flags);
2104         if (tr->stop_count++)
2105                 goto out;
2106
2107         buffer = tr->trace_buffer.buffer;
2108         if (buffer)
2109                 ring_buffer_record_disable(buffer);
2110
2111  out:
2112         raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2113 }
2114
2115 static int trace_save_cmdline(struct task_struct *tsk)
2116 {
2117         unsigned pid, idx;
2118
2119         /* treat recording of idle task as a success */
2120         if (!tsk->pid)
2121                 return 1;
2122
2123         if (unlikely(tsk->pid > PID_MAX_DEFAULT))
2124                 return 0;
2125
2126         /*
2127          * It's not the end of the world if we don't get
2128          * the lock, but we also don't want to spin
2129          * nor do we want to disable interrupts,
2130          * so if we miss here, then better luck next time.
2131          */
2132         if (!arch_spin_trylock(&trace_cmdline_lock))
2133                 return 0;
2134
2135         idx = savedcmd->map_pid_to_cmdline[tsk->pid];
2136         if (idx == NO_CMDLINE_MAP) {
2137                 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2138
2139                 /*
2140                  * Check whether the cmdline buffer at idx has a pid
2141                  * mapped. We are going to overwrite that entry so we
2142                  * need to clear the map_pid_to_cmdline. Otherwise we
2143                  * would read the new comm for the old pid.
2144                  */
2145                 pid = savedcmd->map_cmdline_to_pid[idx];
2146                 if (pid != NO_CMDLINE_MAP)
2147                         savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
2148
2149                 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2150                 savedcmd->map_pid_to_cmdline[tsk->pid] = idx;
2151
2152                 savedcmd->cmdline_idx = idx;
2153         }
2154
2155         set_cmdline(idx, tsk->comm);
2156
2157         arch_spin_unlock(&trace_cmdline_lock);
2158
2159         return 1;
2160 }
2161
2162 static void __trace_find_cmdline(int pid, char comm[])
2163 {
2164         unsigned map;
2165
2166         if (!pid) {
2167                 strcpy(comm, "<idle>");
2168                 return;
2169         }
2170
2171         if (WARN_ON_ONCE(pid < 0)) {
2172                 strcpy(comm, "<XXX>");
2173                 return;
2174         }
2175
2176         if (pid > PID_MAX_DEFAULT) {
2177                 strcpy(comm, "<...>");
2178                 return;
2179         }
2180
2181         map = savedcmd->map_pid_to_cmdline[pid];
2182         if (map != NO_CMDLINE_MAP)
2183                 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2184         else
2185                 strcpy(comm, "<...>");
2186 }
2187
2188 void trace_find_cmdline(int pid, char comm[])
2189 {
2190         preempt_disable();
2191         arch_spin_lock(&trace_cmdline_lock);
2192
2193         __trace_find_cmdline(pid, comm);
2194
2195         arch_spin_unlock(&trace_cmdline_lock);
2196         preempt_enable();
2197 }
2198
2199 int trace_find_tgid(int pid)
2200 {
2201         if (unlikely(!tgid_map || !pid || pid > PID_MAX_DEFAULT))
2202                 return 0;
2203
2204         return tgid_map[pid];
2205 }
2206
2207 static int trace_save_tgid(struct task_struct *tsk)
2208 {
2209         /* treat recording of idle task as a success */
2210         if (!tsk->pid)
2211                 return 1;
2212
2213         if (unlikely(!tgid_map || tsk->pid > PID_MAX_DEFAULT))
2214                 return 0;
2215
2216         tgid_map[tsk->pid] = tsk->tgid;
2217         return 1;
2218 }
2219
2220 static bool tracing_record_taskinfo_skip(int flags)
2221 {
2222         if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2223                 return true;
2224         if (atomic_read(&trace_record_taskinfo_disabled) || !tracing_is_on())
2225                 return true;
2226         if (!__this_cpu_read(trace_taskinfo_save))
2227                 return true;
2228         return false;
2229 }
2230
2231 /**
2232  * tracing_record_taskinfo - record the task info of a task
2233  *
2234  * @task  - task to record
2235  * @flags - TRACE_RECORD_CMDLINE for recording comm
2236  *        - TRACE_RECORD_TGID for recording tgid
2237  */
2238 void tracing_record_taskinfo(struct task_struct *task, int flags)
2239 {
2240         bool done;
2241
2242         if (tracing_record_taskinfo_skip(flags))
2243                 return;
2244
2245         /*
2246          * Record as much task information as possible. If some fail, continue
2247          * to try to record the others.
2248          */
2249         done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2250         done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2251
2252         /* If recording any information failed, retry again soon. */
2253         if (!done)
2254                 return;
2255
2256         __this_cpu_write(trace_taskinfo_save, false);
2257 }
2258
2259 /**
2260  * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2261  *
2262  * @prev - previous task during sched_switch
2263  * @next - next task during sched_switch
2264  * @flags - TRACE_RECORD_CMDLINE for recording comm
2265  *          TRACE_RECORD_TGID for recording tgid
2266  */
2267 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2268                                           struct task_struct *next, int flags)
2269 {
2270         bool done;
2271
2272         if (tracing_record_taskinfo_skip(flags))
2273                 return;
2274
2275         /*
2276          * Record as much task information as possible. If some fail, continue
2277          * to try to record the others.
2278          */
2279         done  = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2280         done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2281         done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2282         done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2283
2284         /* If recording any information failed, retry again soon. */
2285         if (!done)
2286                 return;
2287
2288         __this_cpu_write(trace_taskinfo_save, false);
2289 }
2290
2291 /* Helpers to record a specific task information */
2292 void tracing_record_cmdline(struct task_struct *task)
2293 {
2294         tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2295 }
2296
2297 void tracing_record_tgid(struct task_struct *task)
2298 {
2299         tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2300 }
2301
2302 /*
2303  * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2304  * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2305  * simplifies those functions and keeps them in sync.
2306  */
2307 enum print_line_t trace_handle_return(struct trace_seq *s)
2308 {
2309         return trace_seq_has_overflowed(s) ?
2310                 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2311 }
2312 EXPORT_SYMBOL_GPL(trace_handle_return);
2313
2314 void
2315 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
2316                              int pc)
2317 {
2318         struct task_struct *tsk = current;
2319
2320         entry->preempt_count            = pc & 0xff;
2321         entry->pid                      = (tsk) ? tsk->pid : 0;
2322         entry->flags =
2323 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
2324                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
2325 #else
2326                 TRACE_FLAG_IRQS_NOSUPPORT |
2327 #endif
2328                 ((pc & NMI_MASK    ) ? TRACE_FLAG_NMI     : 0) |
2329                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
2330                 ((pc & SOFTIRQ_OFFSET) ? TRACE_FLAG_SOFTIRQ : 0) |
2331                 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
2332                 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
2333 }
2334 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
2335
2336 struct ring_buffer_event *
2337 trace_buffer_lock_reserve(struct ring_buffer *buffer,
2338                           int type,
2339                           unsigned long len,
2340                           unsigned long flags, int pc)
2341 {
2342         return __trace_buffer_lock_reserve(buffer, type, len, flags, pc);
2343 }
2344
2345 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2346 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2347 static int trace_buffered_event_ref;
2348
2349 /**
2350  * trace_buffered_event_enable - enable buffering events
2351  *
2352  * When events are being filtered, it is quicker to use a temporary
2353  * buffer to write the event data into if there's a likely chance
2354  * that it will not be committed. The discard of the ring buffer
2355  * is not as fast as committing, and is much slower than copying
2356  * a commit.
2357  *
2358  * When an event is to be filtered, allocate per cpu buffers to
2359  * write the event data into, and if the event is filtered and discarded
2360  * it is simply dropped, otherwise, the entire data is to be committed
2361  * in one shot.
2362  */
2363 void trace_buffered_event_enable(void)
2364 {
2365         struct ring_buffer_event *event;
2366         struct page *page;
2367         int cpu;
2368
2369         WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2370
2371         if (trace_buffered_event_ref++)
2372                 return;
2373
2374         for_each_tracing_cpu(cpu) {
2375                 page = alloc_pages_node(cpu_to_node(cpu),
2376                                         GFP_KERNEL | __GFP_NORETRY, 0);
2377                 if (!page)
2378                         goto failed;
2379
2380                 event = page_address(page);
2381                 memset(event, 0, sizeof(*event));
2382
2383                 per_cpu(trace_buffered_event, cpu) = event;
2384
2385                 preempt_disable();
2386                 if (cpu == smp_processor_id() &&
2387                     this_cpu_read(trace_buffered_event) !=
2388                     per_cpu(trace_buffered_event, cpu))
2389                         WARN_ON_ONCE(1);
2390                 preempt_enable();
2391         }
2392
2393         return;
2394  failed:
2395         trace_buffered_event_disable();
2396 }
2397
2398 static void enable_trace_buffered_event(void *data)
2399 {
2400         /* Probably not needed, but do it anyway */
2401         smp_rmb();
2402         this_cpu_dec(trace_buffered_event_cnt);
2403 }
2404
2405 static void disable_trace_buffered_event(void *data)
2406 {
2407         this_cpu_inc(trace_buffered_event_cnt);
2408 }
2409
2410 /**
2411  * trace_buffered_event_disable - disable buffering events
2412  *
2413  * When a filter is removed, it is faster to not use the buffered
2414  * events, and to commit directly into the ring buffer. Free up
2415  * the temp buffers when there are no more users. This requires
2416  * special synchronization with current events.
2417  */
2418 void trace_buffered_event_disable(void)
2419 {
2420         int cpu;
2421
2422         WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2423
2424         if (WARN_ON_ONCE(!trace_buffered_event_ref))
2425                 return;
2426
2427         if (--trace_buffered_event_ref)
2428                 return;
2429
2430         preempt_disable();
2431         /* For each CPU, set the buffer as used. */
2432         smp_call_function_many(tracing_buffer_mask,
2433                                disable_trace_buffered_event, NULL, 1);
2434         preempt_enable();
2435
2436         /* Wait for all current users to finish */
2437         synchronize_rcu();
2438
2439         for_each_tracing_cpu(cpu) {
2440                 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2441                 per_cpu(trace_buffered_event, cpu) = NULL;
2442         }
2443         /*
2444          * Make sure trace_buffered_event is NULL before clearing
2445          * trace_buffered_event_cnt.
2446          */
2447         smp_wmb();
2448
2449         preempt_disable();
2450         /* Do the work on each cpu */
2451         smp_call_function_many(tracing_buffer_mask,
2452                                enable_trace_buffered_event, NULL, 1);
2453         preempt_enable();
2454 }
2455
2456 static struct ring_buffer *temp_buffer;
2457
2458 struct ring_buffer_event *
2459 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
2460                           struct trace_event_file *trace_file,
2461                           int type, unsigned long len,
2462                           unsigned long flags, int pc)
2463 {
2464         struct ring_buffer_event *entry;
2465         int val;
2466
2467         *current_rb = trace_file->tr->trace_buffer.buffer;
2468
2469         if (!ring_buffer_time_stamp_abs(*current_rb) && (trace_file->flags &
2470              (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED)) &&
2471             (entry = this_cpu_read(trace_buffered_event))) {
2472                 /* Try to use the per cpu buffer first */
2473                 val = this_cpu_inc_return(trace_buffered_event_cnt);
2474                 if (val == 1) {
2475                         trace_event_setup(entry, type, flags, pc);
2476                         entry->array[0] = len;
2477                         return entry;
2478                 }
2479                 this_cpu_dec(trace_buffered_event_cnt);
2480         }
2481
2482         entry = __trace_buffer_lock_reserve(*current_rb,
2483                                             type, len, flags, pc);
2484         /*
2485          * If tracing is off, but we have triggers enabled
2486          * we still need to look at the event data. Use the temp_buffer
2487          * to store the trace event for the tigger to use. It's recusive
2488          * safe and will not be recorded anywhere.
2489          */
2490         if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2491                 *current_rb = temp_buffer;
2492                 entry = __trace_buffer_lock_reserve(*current_rb,
2493                                                     type, len, flags, pc);
2494         }
2495         return entry;
2496 }
2497 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2498
2499 static DEFINE_SPINLOCK(tracepoint_iter_lock);
2500 static DEFINE_MUTEX(tracepoint_printk_mutex);
2501
2502 static void output_printk(struct trace_event_buffer *fbuffer)
2503 {
2504         struct trace_event_call *event_call;
2505         struct trace_event *event;
2506         unsigned long flags;
2507         struct trace_iterator *iter = tracepoint_print_iter;
2508
2509         /* We should never get here if iter is NULL */
2510         if (WARN_ON_ONCE(!iter))
2511                 return;
2512
2513         event_call = fbuffer->trace_file->event_call;
2514         if (!event_call || !event_call->event.funcs ||
2515             !event_call->event.funcs->trace)
2516                 return;
2517
2518         event = &fbuffer->trace_file->event_call->event;
2519
2520         spin_lock_irqsave(&tracepoint_iter_lock, flags);
2521         trace_seq_init(&iter->seq);
2522         iter->ent = fbuffer->entry;
2523         event_call->event.funcs->trace(iter, 0, event);
2524         trace_seq_putc(&iter->seq, 0);
2525         printk("%s", iter->seq.buffer);
2526
2527         spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2528 }
2529
2530 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2531                              void __user *buffer, size_t *lenp,
2532                              loff_t *ppos)
2533 {
2534         int save_tracepoint_printk;
2535         int ret;
2536
2537         mutex_lock(&tracepoint_printk_mutex);
2538         save_tracepoint_printk = tracepoint_printk;
2539
2540         ret = proc_dointvec(table, write, buffer, lenp, ppos);
2541
2542         /*
2543          * This will force exiting early, as tracepoint_printk
2544          * is always zero when tracepoint_printk_iter is not allocated
2545          */
2546         if (!tracepoint_print_iter)
2547                 tracepoint_printk = 0;
2548
2549         if (save_tracepoint_printk == tracepoint_printk)
2550                 goto out;
2551
2552         if (tracepoint_printk)
2553                 static_key_enable(&tracepoint_printk_key.key);
2554         else
2555                 static_key_disable(&tracepoint_printk_key.key);
2556
2557  out:
2558         mutex_unlock(&tracepoint_printk_mutex);
2559
2560         return ret;
2561 }
2562
2563 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2564 {
2565         if (static_key_false(&tracepoint_printk_key.key))
2566                 output_printk(fbuffer);
2567
2568         event_trigger_unlock_commit(fbuffer->trace_file, fbuffer->buffer,
2569                                     fbuffer->event, fbuffer->entry,
2570                                     fbuffer->flags, fbuffer->pc);
2571 }
2572 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2573
2574 /*
2575  * Skip 3:
2576  *
2577  *   trace_buffer_unlock_commit_regs()
2578  *   trace_event_buffer_commit()
2579  *   trace_event_raw_event_xxx()
2580  */
2581 # define STACK_SKIP 3
2582
2583 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2584                                      struct ring_buffer *buffer,
2585                                      struct ring_buffer_event *event,
2586                                      unsigned long flags, int pc,
2587                                      struct pt_regs *regs)
2588 {
2589         __buffer_unlock_commit(buffer, event);
2590
2591         /*
2592          * If regs is not set, then skip the necessary functions.
2593          * Note, we can still get here via blktrace, wakeup tracer
2594          * and mmiotrace, but that's ok if they lose a function or
2595          * two. They are not that meaningful.
2596          */
2597         ftrace_trace_stack(tr, buffer, flags, regs ? 0 : STACK_SKIP, pc, regs);
2598         ftrace_trace_userstack(buffer, flags, pc);
2599 }
2600
2601 /*
2602  * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2603  */
2604 void
2605 trace_buffer_unlock_commit_nostack(struct ring_buffer *buffer,
2606                                    struct ring_buffer_event *event)
2607 {
2608         __buffer_unlock_commit(buffer, event);
2609 }
2610
2611 static void
2612 trace_process_export(struct trace_export *export,
2613                struct ring_buffer_event *event)
2614 {
2615         struct trace_entry *entry;
2616         unsigned int size = 0;
2617
2618         entry = ring_buffer_event_data(event);
2619         size = ring_buffer_event_length(event);
2620         export->write(export, entry, size);
2621 }
2622
2623 static DEFINE_MUTEX(ftrace_export_lock);
2624
2625 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
2626
2627 static DEFINE_STATIC_KEY_FALSE(ftrace_exports_enabled);
2628
2629 static inline void ftrace_exports_enable(void)
2630 {
2631         static_branch_enable(&ftrace_exports_enabled);
2632 }
2633
2634 static inline void ftrace_exports_disable(void)
2635 {
2636         static_branch_disable(&ftrace_exports_enabled);
2637 }
2638
2639 static void ftrace_exports(struct ring_buffer_event *event)
2640 {
2641         struct trace_export *export;
2642
2643         preempt_disable_notrace();
2644
2645         export = rcu_dereference_raw_notrace(ftrace_exports_list);
2646         while (export) {
2647                 trace_process_export(export, event);
2648                 export = rcu_dereference_raw_notrace(export->next);
2649         }
2650
2651         preempt_enable_notrace();
2652 }
2653
2654 static inline void
2655 add_trace_export(struct trace_export **list, struct trace_export *export)
2656 {
2657         rcu_assign_pointer(export->next, *list);
2658         /*
2659          * We are entering export into the list but another
2660          * CPU might be walking that list. We need to make sure
2661          * the export->next pointer is valid before another CPU sees
2662          * the export pointer included into the list.
2663          */
2664         rcu_assign_pointer(*list, export);
2665 }
2666
2667 static inline int
2668 rm_trace_export(struct trace_export **list, struct trace_export *export)
2669 {
2670         struct trace_export **p;
2671
2672         for (p = list; *p != NULL; p = &(*p)->next)
2673                 if (*p == export)
2674                         break;
2675
2676         if (*p != export)
2677                 return -1;
2678
2679         rcu_assign_pointer(*p, (*p)->next);
2680
2681         return 0;
2682 }
2683
2684 static inline void
2685 add_ftrace_export(struct trace_export **list, struct trace_export *export)
2686 {
2687         if (*list == NULL)
2688                 ftrace_exports_enable();
2689
2690         add_trace_export(list, export);
2691 }
2692
2693 static inline int
2694 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
2695 {
2696         int ret;
2697
2698         ret = rm_trace_export(list, export);
2699         if (*list == NULL)
2700                 ftrace_exports_disable();
2701
2702         return ret;
2703 }
2704
2705 int register_ftrace_export(struct trace_export *export)
2706 {
2707         if (WARN_ON_ONCE(!export->write))
2708                 return -1;
2709
2710         mutex_lock(&ftrace_export_lock);
2711
2712         add_ftrace_export(&ftrace_exports_list, export);
2713
2714         mutex_unlock(&ftrace_export_lock);
2715
2716         return 0;
2717 }
2718 EXPORT_SYMBOL_GPL(register_ftrace_export);
2719
2720 int unregister_ftrace_export(struct trace_export *export)
2721 {
2722         int ret;
2723
2724         mutex_lock(&ftrace_export_lock);
2725
2726         ret = rm_ftrace_export(&ftrace_exports_list, export);
2727
2728         mutex_unlock(&ftrace_export_lock);
2729
2730         return ret;
2731 }
2732 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
2733
2734 void
2735 trace_function(struct trace_array *tr,
2736                unsigned long ip, unsigned long parent_ip, unsigned long flags,
2737                int pc)
2738 {
2739         struct trace_event_call *call = &event_function;
2740         struct ring_buffer *buffer = tr->trace_buffer.buffer;
2741         struct ring_buffer_event *event;
2742         struct ftrace_entry *entry;
2743
2744         event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2745                                             flags, pc);
2746         if (!event)
2747                 return;
2748         entry   = ring_buffer_event_data(event);
2749         entry->ip                       = ip;
2750         entry->parent_ip                = parent_ip;
2751
2752         if (!call_filter_check_discard(call, entry, buffer, event)) {
2753                 if (static_branch_unlikely(&ftrace_exports_enabled))
2754                         ftrace_exports(event);
2755                 __buffer_unlock_commit(buffer, event);
2756         }
2757 }
2758
2759 #ifdef CONFIG_STACKTRACE
2760
2761 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2762 #define FTRACE_KSTACK_NESTING   4
2763
2764 #define FTRACE_KSTACK_ENTRIES   (PAGE_SIZE / FTRACE_KSTACK_NESTING)
2765
2766 struct ftrace_stack {
2767         unsigned long           calls[FTRACE_KSTACK_ENTRIES];
2768 };
2769
2770
2771 struct ftrace_stacks {
2772         struct ftrace_stack     stacks[FTRACE_KSTACK_NESTING];
2773 };
2774
2775 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
2776 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2777
2778 static void __ftrace_trace_stack(struct ring_buffer *buffer,
2779                                  unsigned long flags,
2780                                  int skip, int pc, struct pt_regs *regs)
2781 {
2782         struct trace_event_call *call = &event_kernel_stack;
2783         struct ring_buffer_event *event;
2784         unsigned int size, nr_entries;
2785         struct ftrace_stack *fstack;
2786         struct stack_entry *entry;
2787         int stackidx;
2788
2789         /*
2790          * Add one, for this function and the call to save_stack_trace()
2791          * If regs is set, then these functions will not be in the way.
2792          */
2793 #ifndef CONFIG_UNWINDER_ORC
2794         if (!regs)
2795                 skip++;
2796 #endif
2797
2798         /*
2799          * Since events can happen in NMIs there's no safe way to
2800          * use the per cpu ftrace_stacks. We reserve it and if an interrupt
2801          * or NMI comes in, it will just have to use the default
2802          * FTRACE_STACK_SIZE.
2803          */
2804         preempt_disable_notrace();
2805
2806         stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
2807
2808         /* This should never happen. If it does, yell once and skip */
2809         if (WARN_ON_ONCE(stackidx > FTRACE_KSTACK_NESTING))
2810                 goto out;
2811
2812         /*
2813          * The above __this_cpu_inc_return() is 'atomic' cpu local. An
2814          * interrupt will either see the value pre increment or post
2815          * increment. If the interrupt happens pre increment it will have
2816          * restored the counter when it returns.  We just need a barrier to
2817          * keep gcc from moving things around.
2818          */
2819         barrier();
2820
2821         fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
2822         size = ARRAY_SIZE(fstack->calls);
2823
2824         if (regs) {
2825                 nr_entries = stack_trace_save_regs(regs, fstack->calls,
2826                                                    size, skip);
2827         } else {
2828                 nr_entries = stack_trace_save(fstack->calls, size, skip);
2829         }
2830
2831         size = nr_entries * sizeof(unsigned long);
2832         event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2833                                             sizeof(*entry) + size, flags, pc);
2834         if (!event)
2835                 goto out;
2836         entry = ring_buffer_event_data(event);
2837
2838         memcpy(&entry->caller, fstack->calls, size);
2839         entry->size = nr_entries;
2840
2841         if (!call_filter_check_discard(call, entry, buffer, event))
2842                 __buffer_unlock_commit(buffer, event);
2843
2844  out:
2845         /* Again, don't let gcc optimize things here */
2846         barrier();
2847         __this_cpu_dec(ftrace_stack_reserve);
2848         preempt_enable_notrace();
2849
2850 }
2851
2852 static inline void ftrace_trace_stack(struct trace_array *tr,
2853                                       struct ring_buffer *buffer,
2854                                       unsigned long flags,
2855                                       int skip, int pc, struct pt_regs *regs)
2856 {
2857         if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
2858                 return;
2859
2860         __ftrace_trace_stack(buffer, flags, skip, pc, regs);
2861 }
2862
2863 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
2864                    int pc)
2865 {
2866         struct ring_buffer *buffer = tr->trace_buffer.buffer;
2867
2868         if (rcu_is_watching()) {
2869                 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
2870                 return;
2871         }
2872
2873         /*
2874          * When an NMI triggers, RCU is enabled via rcu_nmi_enter(),
2875          * but if the above rcu_is_watching() failed, then the NMI
2876          * triggered someplace critical, and rcu_irq_enter() should
2877          * not be called from NMI.
2878          */
2879         if (unlikely(in_nmi()))
2880                 return;
2881
2882         rcu_irq_enter_irqson();
2883         __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
2884         rcu_irq_exit_irqson();
2885 }
2886
2887 /**
2888  * trace_dump_stack - record a stack back trace in the trace buffer
2889  * @skip: Number of functions to skip (helper handlers)
2890  */
2891 void trace_dump_stack(int skip)
2892 {
2893         unsigned long flags;
2894
2895         if (tracing_disabled || tracing_selftest_running)
2896                 return;
2897
2898         local_save_flags(flags);
2899
2900 #ifndef CONFIG_UNWINDER_ORC
2901         /* Skip 1 to skip this function. */
2902         skip++;
2903 #endif
2904         __ftrace_trace_stack(global_trace.trace_buffer.buffer,
2905                              flags, skip, preempt_count(), NULL);
2906 }
2907 EXPORT_SYMBOL_GPL(trace_dump_stack);
2908
2909 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
2910 static DEFINE_PER_CPU(int, user_stack_count);
2911
2912 static void
2913 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
2914 {
2915         struct trace_event_call *call = &event_user_stack;
2916         struct ring_buffer_event *event;
2917         struct userstack_entry *entry;
2918
2919         if (!(global_trace.trace_flags & TRACE_ITER_USERSTACKTRACE))
2920                 return;
2921
2922         /*
2923          * NMIs can not handle page faults, even with fix ups.
2924          * The save user stack can (and often does) fault.
2925          */
2926         if (unlikely(in_nmi()))
2927                 return;
2928
2929         /*
2930          * prevent recursion, since the user stack tracing may
2931          * trigger other kernel events.
2932          */
2933         preempt_disable();
2934         if (__this_cpu_read(user_stack_count))
2935                 goto out;
2936
2937         __this_cpu_inc(user_stack_count);
2938
2939         event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
2940                                             sizeof(*entry), flags, pc);
2941         if (!event)
2942                 goto out_drop_count;
2943         entry   = ring_buffer_event_data(event);
2944
2945         entry->tgid             = current->tgid;
2946         memset(&entry->caller, 0, sizeof(entry->caller));
2947
2948         stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
2949         if (!call_filter_check_discard(call, entry, buffer, event))
2950                 __buffer_unlock_commit(buffer, event);
2951
2952  out_drop_count:
2953         __this_cpu_dec(user_stack_count);
2954  out:
2955         preempt_enable();
2956 }
2957 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
2958 static void ftrace_trace_userstack(struct ring_buffer *buffer,
2959                                    unsigned long flags, int pc)
2960 {
2961 }
2962 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
2963
2964 #endif /* CONFIG_STACKTRACE */
2965
2966 /* created for use with alloc_percpu */
2967 struct trace_buffer_struct {
2968         int nesting;
2969         char buffer[4][TRACE_BUF_SIZE];
2970 };
2971
2972 static struct trace_buffer_struct *trace_percpu_buffer;
2973
2974 /*
2975  * Thise allows for lockless recording.  If we're nested too deeply, then
2976  * this returns NULL.
2977  */
2978 static char *get_trace_buf(void)
2979 {
2980         struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
2981
2982         if (!buffer || buffer->nesting >= 4)
2983                 return NULL;
2984
2985         buffer->nesting++;
2986
2987         /* Interrupts must see nesting incremented before we use the buffer */
2988         barrier();
2989         return &buffer->buffer[buffer->nesting][0];
2990 }
2991
2992 static void put_trace_buf(void)
2993 {
2994         /* Don't let the decrement of nesting leak before this */
2995         barrier();
2996         this_cpu_dec(trace_percpu_buffer->nesting);
2997 }
2998
2999 static int alloc_percpu_trace_buffer(void)
3000 {
3001         struct trace_buffer_struct *buffers;
3002
3003         buffers = alloc_percpu(struct trace_buffer_struct);
3004         if (WARN(!buffers, "Could not allocate percpu trace_printk buffer"))
3005                 return -ENOMEM;
3006
3007         trace_percpu_buffer = buffers;
3008         return 0;
3009 }
3010
3011 static int buffers_allocated;
3012
3013 void trace_printk_init_buffers(void)
3014 {
3015         if (buffers_allocated)
3016                 return;
3017
3018         if (alloc_percpu_trace_buffer())
3019                 return;
3020
3021         /* trace_printk() is for debug use only. Don't use it in production. */
3022
3023         pr_warn("\n");
3024         pr_warn("**********************************************************\n");
3025         pr_warn("**   NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE   **\n");
3026         pr_warn("**                                                      **\n");
3027         pr_warn("** trace_printk() being used. Allocating extra memory.  **\n");
3028         pr_warn("**                                                      **\n");
3029         pr_warn("** This means that this is a DEBUG kernel and it is     **\n");
3030         pr_warn("** unsafe for production use.                           **\n");
3031         pr_warn("**                                                      **\n");
3032         pr_warn("** If you see this message and you are not debugging    **\n");
3033         pr_warn("** the kernel, report this immediately to your vendor!  **\n");
3034         pr_warn("**                                                      **\n");
3035         pr_warn("**   NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE   **\n");
3036         pr_warn("**********************************************************\n");
3037
3038         /* Expand the buffers to set size */
3039         tracing_update_buffers();
3040
3041         buffers_allocated = 1;
3042
3043         /*
3044          * trace_printk_init_buffers() can be called by modules.
3045          * If that happens, then we need to start cmdline recording
3046          * directly here. If the global_trace.buffer is already
3047          * allocated here, then this was called by module code.
3048          */
3049         if (global_trace.trace_buffer.buffer)
3050                 tracing_start_cmdline_record();
3051 }
3052 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3053
3054 void trace_printk_start_comm(void)
3055 {
3056         /* Start tracing comms if trace printk is set */
3057         if (!buffers_allocated)
3058                 return;
3059         tracing_start_cmdline_record();
3060 }
3061
3062 static void trace_printk_start_stop_comm(int enabled)
3063 {
3064         if (!buffers_allocated)
3065                 return;
3066
3067         if (enabled)
3068                 tracing_start_cmdline_record();
3069         else
3070                 tracing_stop_cmdline_record();
3071 }
3072
3073 /**
3074  * trace_vbprintk - write binary msg to tracing buffer
3075  *
3076  */
3077 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3078 {
3079         struct trace_event_call *call = &event_bprint;
3080         struct ring_buffer_event *event;
3081         struct ring_buffer *buffer;
3082         struct trace_array *tr = &global_trace;
3083         struct bprint_entry *entry;
3084         unsigned long flags;
3085         char *tbuffer;
3086         int len = 0, size, pc;
3087
3088         if (unlikely(tracing_selftest_running || tracing_disabled))
3089                 return 0;
3090
3091         /* Don't pollute graph traces with trace_vprintk internals */
3092         pause_graph_tracing();
3093
3094         pc = preempt_count();
3095         preempt_disable_notrace();
3096
3097         tbuffer = get_trace_buf();
3098         if (!tbuffer) {
3099                 len = 0;
3100                 goto out_nobuffer;
3101         }
3102
3103         len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3104
3105         if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3106                 goto out;
3107
3108         local_save_flags(flags);
3109         size = sizeof(*entry) + sizeof(u32) * len;
3110         buffer = tr->trace_buffer.buffer;
3111         event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3112                                             flags, pc);
3113         if (!event)
3114                 goto out;
3115         entry = ring_buffer_event_data(event);
3116         entry->ip                       = ip;
3117         entry->fmt                      = fmt;
3118
3119         memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3120         if (!call_filter_check_discard(call, entry, buffer, event)) {
3121                 __buffer_unlock_commit(buffer, event);
3122                 ftrace_trace_stack(tr, buffer, flags, 6, pc, NULL);
3123         }
3124
3125 out:
3126         put_trace_buf();
3127
3128 out_nobuffer:
3129         preempt_enable_notrace();
3130         unpause_graph_tracing();
3131
3132         return len;
3133 }
3134 EXPORT_SYMBOL_GPL(trace_vbprintk);
3135
3136 __printf(3, 0)
3137 static int
3138 __trace_array_vprintk(struct ring_buffer *buffer,
3139                       unsigned long ip, const char *fmt, va_list args)
3140 {
3141         struct trace_event_call *call = &event_print;
3142         struct ring_buffer_event *event;
3143         int len = 0, size, pc;
3144         struct print_entry *entry;
3145         unsigned long flags;
3146         char *tbuffer;
3147
3148         if (tracing_disabled || tracing_selftest_running)
3149                 return 0;
3150
3151         /* Don't pollute graph traces with trace_vprintk internals */
3152         pause_graph_tracing();
3153
3154         pc = preempt_count();
3155         preempt_disable_notrace();
3156
3157
3158         tbuffer = get_trace_buf();
3159         if (!tbuffer) {
3160                 len = 0;
3161                 goto out_nobuffer;
3162         }
3163
3164         len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3165
3166         local_save_flags(flags);
3167         size = sizeof(*entry) + len + 1;
3168         event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3169                                             flags, pc);
3170         if (!event)
3171                 goto out;
3172         entry = ring_buffer_event_data(event);
3173         entry->ip = ip;
3174
3175         memcpy(&entry->buf, tbuffer, len + 1);
3176         if (!call_filter_check_discard(call, entry, buffer, event)) {
3177                 __buffer_unlock_commit(buffer, event);
3178                 ftrace_trace_stack(&global_trace, buffer, flags, 6, pc, NULL);
3179         }
3180
3181 out:
3182         put_trace_buf();
3183
3184 out_nobuffer:
3185         preempt_enable_notrace();
3186         unpause_graph_tracing();
3187
3188         return len;
3189 }
3190
3191 __printf(3, 0)
3192 int trace_array_vprintk(struct trace_array *tr,
3193                         unsigned long ip, const char *fmt, va_list args)
3194 {
3195         return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
3196 }
3197
3198 __printf(3, 0)
3199 int trace_array_printk(struct trace_array *tr,
3200                        unsigned long ip, const char *fmt, ...)
3201 {
3202         int ret;
3203         va_list ap;
3204
3205         if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3206                 return 0;
3207
3208         va_start(ap, fmt);
3209         ret = trace_array_vprintk(tr, ip, fmt, ap);
3210         va_end(ap);
3211         return ret;
3212 }
3213 EXPORT_SYMBOL_GPL(trace_array_printk);
3214
3215 __printf(3, 4)
3216 int trace_array_printk_buf(struct ring_buffer *buffer,
3217                            unsigned long ip, const char *fmt, ...)
3218 {
3219         int ret;
3220         va_list ap;
3221
3222         if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3223                 return 0;
3224
3225         va_start(ap, fmt);
3226         ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3227         va_end(ap);
3228         return ret;
3229 }
3230
3231 __printf(2, 0)
3232 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3233 {
3234         return trace_array_vprintk(&global_trace, ip, fmt, args);
3235 }
3236 EXPORT_SYMBOL_GPL(trace_vprintk);
3237
3238 static void trace_iterator_increment(struct trace_iterator *iter)
3239 {
3240         struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3241
3242         iter->idx++;
3243         if (buf_iter)
3244                 ring_buffer_read(buf_iter, NULL);
3245 }
3246
3247 static struct trace_entry *
3248 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3249                 unsigned long *lost_events)
3250 {
3251         struct ring_buffer_event *event;
3252         struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3253
3254         if (buf_iter)
3255                 event = ring_buffer_iter_peek(buf_iter, ts);
3256         else
3257                 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
3258                                          lost_events);
3259
3260         if (event) {
3261                 iter->ent_size = ring_buffer_event_length(event);
3262                 return ring_buffer_event_data(event);
3263         }
3264         iter->ent_size = 0;
3265         return NULL;
3266 }
3267
3268 static struct trace_entry *
3269 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3270                   unsigned long *missing_events, u64 *ent_ts)
3271 {
3272         struct ring_buffer *buffer = iter->trace_buffer->buffer;
3273         struct trace_entry *ent, *next = NULL;
3274         unsigned long lost_events = 0, next_lost = 0;
3275         int cpu_file = iter->cpu_file;
3276         u64 next_ts = 0, ts;
3277         int next_cpu = -1;
3278         int next_size = 0;
3279         int cpu;
3280
3281         /*
3282          * If we are in a per_cpu trace file, don't bother by iterating over
3283          * all cpu and peek directly.
3284          */
3285         if (cpu_file > RING_BUFFER_ALL_CPUS) {
3286                 if (ring_buffer_empty_cpu(buffer, cpu_file))
3287                         return NULL;
3288                 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3289                 if (ent_cpu)
3290                         *ent_cpu = cpu_file;
3291
3292                 return ent;
3293         }
3294
3295         for_each_tracing_cpu(cpu) {
3296
3297                 if (ring_buffer_empty_cpu(buffer, cpu))
3298                         continue;
3299
3300                 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3301
3302                 /*
3303                  * Pick the entry with the smallest timestamp:
3304                  */
3305                 if (ent && (!next || ts < next_ts)) {
3306                         next = ent;
3307                         next_cpu = cpu;
3308                         next_ts = ts;
3309                         next_lost = lost_events;
3310                         next_size = iter->ent_size;
3311                 }
3312         }
3313
3314         iter->ent_size = next_size;
3315
3316         if (ent_cpu)
3317                 *ent_cpu = next_cpu;
3318
3319         if (ent_ts)
3320                 *ent_ts = next_ts;
3321
3322         if (missing_events)
3323                 *missing_events = next_lost;
3324
3325         return next;
3326 }
3327
3328 /* Find the next real entry, without updating the iterator itself */
3329 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3330                                           int *ent_cpu, u64 *ent_ts)
3331 {
3332         return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3333 }
3334
3335 /* Find the next real entry, and increment the iterator to the next entry */
3336 void *trace_find_next_entry_inc(struct trace_iterator *iter)
3337 {
3338         iter->ent = __find_next_entry(iter, &iter->cpu,
3339                                       &iter->lost_events, &iter->ts);
3340
3341         if (iter->ent)
3342                 trace_iterator_increment(iter);
3343
3344         return iter->ent ? iter : NULL;
3345 }
3346
3347 static void trace_consume(struct trace_iterator *iter)
3348 {
3349         ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
3350                             &iter->lost_events);
3351 }
3352
3353 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3354 {
3355         struct trace_iterator *iter = m->private;
3356         int i = (int)*pos;
3357         void *ent;
3358
3359         WARN_ON_ONCE(iter->leftover);
3360
3361         (*pos)++;
3362
3363         /* can't go backwards */
3364         if (iter->idx > i)
3365                 return NULL;
3366
3367         if (iter->idx < 0)
3368                 ent = trace_find_next_entry_inc(iter);
3369         else
3370                 ent = iter;
3371
3372         while (ent && iter->idx < i)
3373                 ent = trace_find_next_entry_inc(iter);
3374
3375         iter->pos = *pos;
3376
3377         return ent;
3378 }
3379
3380 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3381 {
3382         struct ring_buffer_event *event;
3383         struct ring_buffer_iter *buf_iter;
3384         unsigned long entries = 0;
3385         u64 ts;
3386
3387         per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
3388
3389         buf_iter = trace_buffer_iter(iter, cpu);
3390         if (!buf_iter)
3391                 return;
3392
3393         ring_buffer_iter_reset(buf_iter);
3394
3395         /*
3396          * We could have the case with the max latency tracers
3397          * that a reset never took place on a cpu. This is evident
3398          * by the timestamp being before the start of the buffer.
3399          */
3400         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
3401                 if (ts >= iter->trace_buffer->time_start)
3402                         break;
3403                 entries++;
3404                 ring_buffer_read(buf_iter, NULL);
3405         }
3406
3407         per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
3408 }
3409
3410 /*
3411  * The current tracer is copied to avoid a global locking
3412  * all around.
3413  */
3414 static void *s_start(struct seq_file *m, loff_t *pos)
3415 {
3416         struct trace_iterator *iter = m->private;
3417         struct trace_array *tr = iter->tr;
3418         int cpu_file = iter->cpu_file;
3419         void *p = NULL;
3420         loff_t l = 0;
3421         int cpu;
3422
3423         /*
3424          * copy the tracer to avoid using a global lock all around.
3425          * iter->trace is a copy of current_trace, the pointer to the
3426          * name may be used instead of a strcmp(), as iter->trace->name
3427          * will point to the same string as current_trace->name.
3428          */
3429         mutex_lock(&trace_types_lock);
3430         if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
3431                 *iter->trace = *tr->current_trace;
3432         mutex_unlock(&trace_types_lock);
3433
3434 #ifdef CONFIG_TRACER_MAX_TRACE
3435         if (iter->snapshot && iter->trace->use_max_tr)
3436                 return ERR_PTR(-EBUSY);
3437 #endif
3438
3439         if (!iter->snapshot)
3440                 atomic_inc(&trace_record_taskinfo_disabled);
3441
3442         if (*pos != iter->pos) {
3443                 iter->ent = NULL;
3444                 iter->cpu = 0;
3445                 iter->idx = -1;
3446
3447                 if (cpu_file == RING_BUFFER_ALL_CPUS) {
3448                         for_each_tracing_cpu(cpu)
3449                                 tracing_iter_reset(iter, cpu);
3450                 } else
3451                         tracing_iter_reset(iter, cpu_file);
3452
3453                 iter->leftover = 0;
3454                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
3455                         ;
3456
3457         } else {
3458                 /*
3459                  * If we overflowed the seq_file before, then we want
3460                  * to just reuse the trace_seq buffer again.
3461                  */
3462                 if (iter->leftover)
3463                         p = iter;
3464                 else {
3465                         l = *pos - 1;
3466                         p = s_next(m, p, &l);
3467                 }
3468         }
3469
3470         trace_event_read_lock();
3471         trace_access_lock(cpu_file);
3472         return p;
3473 }
3474
3475 static void s_stop(struct seq_file *m, void *p)
3476 {
3477         struct trace_iterator *iter = m->private;
3478
3479 #ifdef CONFIG_TRACER_MAX_TRACE
3480         if (iter->snapshot && iter->trace->use_max_tr)
3481                 return;
3482 #endif
3483
3484         if (!iter->snapshot)
3485                 atomic_dec(&trace_record_taskinfo_disabled);
3486
3487         trace_access_unlock(iter->cpu_file);
3488         trace_event_read_unlock();
3489 }
3490
3491 static void
3492 get_total_entries_cpu(struct trace_buffer *buf, unsigned long *total,
3493                       unsigned long *entries, int cpu)
3494 {
3495         unsigned long count;
3496
3497         count = ring_buffer_entries_cpu(buf->buffer, cpu);
3498         /*
3499          * If this buffer has skipped entries, then we hold all
3500          * entries for the trace and we need to ignore the
3501          * ones before the time stamp.
3502          */
3503         if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
3504                 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
3505                 /* total is the same as the entries */
3506                 *total = count;
3507         } else
3508                 *total = count +
3509                         ring_buffer_overrun_cpu(buf->buffer, cpu);
3510         *entries = count;
3511 }
3512
3513 static void
3514 get_total_entries(struct trace_buffer *buf,
3515                   unsigned long *total, unsigned long *entries)
3516 {
3517         unsigned long t, e;
3518         int cpu;
3519
3520         *total = 0;
3521         *entries = 0;
3522
3523         for_each_tracing_cpu(cpu) {
3524                 get_total_entries_cpu(buf, &t, &e, cpu);
3525                 *total += t;
3526                 *entries += e;
3527         }
3528 }
3529
3530 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
3531 {
3532         unsigned long total, entries;
3533
3534         if (!tr)
3535                 tr = &global_trace;
3536
3537         get_total_entries_cpu(&tr->trace_buffer, &total, &entries, cpu);
3538
3539         return entries;
3540 }
3541
3542 unsigned long trace_total_entries(struct trace_array *tr)
3543 {
3544         unsigned long total, entries;
3545
3546         if (!tr)
3547                 tr = &global_trace;
3548
3549         get_total_entries(&tr->trace_buffer, &total, &entries);
3550
3551         return entries;
3552 }
3553
3554 static void print_lat_help_header(struct seq_file *m)
3555 {
3556         seq_puts(m, "#                  _------=> CPU#            \n"
3557                     "#                 / _-----=> irqs-off        \n"
3558                     "#                | / _----=> need-resched    \n"
3559                     "#                || / _---=> hardirq/softirq \n"
3560                     "#                ||| / _--=> preempt-depth   \n"
3561                     "#                |||| /     delay            \n"
3562                     "#  cmd     pid   ||||| time  |   caller      \n"
3563                     "#     \\   /      |||||  \\    |   /         \n");
3564 }
3565
3566 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
3567 {
3568         unsigned long total;
3569         unsigned long entries;
3570
3571         get_total_entries(buf, &total, &entries);
3572         seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu   #P:%d\n",
3573                    entries, total, num_online_cpus());
3574         seq_puts(m, "#\n");
3575 }
3576
3577 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m,
3578                                    unsigned int flags)
3579 {
3580         bool tgid = flags & TRACE_ITER_RECORD_TGID;
3581
3582         print_event_info(buf, m);
3583
3584         seq_printf(m, "#           TASK-PID   %s  CPU#   TIMESTAMP  FUNCTION\n", tgid ? "TGID     " : "");
3585         seq_printf(m, "#              | |     %s    |       |         |\n",      tgid ? "  |      " : "");
3586 }
3587
3588 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m,
3589                                        unsigned int flags)
3590 {
3591         bool tgid = flags & TRACE_ITER_RECORD_TGID;
3592         const char *space = "          ";
3593         int prec = tgid ? 10 : 2;
3594
3595         print_event_info(buf, m);
3596
3597         seq_printf(m, "#                          %.*s  _-----=> irqs-off\n", prec, space);
3598         seq_printf(m, "#                          %.*s / _----=> need-resched\n", prec, space);
3599         seq_printf(m, "#                          %.*s| / _---=> hardirq/softirq\n", prec, space);
3600         seq_printf(m, "#                          %.*s|| / _--=> preempt-depth\n", prec, space);
3601         seq_printf(m, "#                          %.*s||| /     delay\n", prec, space);
3602         seq_printf(m, "#           TASK-PID %.*sCPU#  ||||    TIMESTAMP  FUNCTION\n", prec, "   TGID   ");
3603         seq_printf(m, "#              | |   %.*s  |   ||||       |         |\n", prec, "     |    ");
3604 }
3605
3606 void
3607 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
3608 {
3609         unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
3610         struct trace_buffer *buf = iter->trace_buffer;
3611         struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
3612         struct tracer *type = iter->trace;
3613         unsigned long entries;
3614         unsigned long total;
3615         const char *name = "preemption";
3616
3617         name = type->name;
3618
3619         get_total_entries(buf, &total, &entries);
3620
3621         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
3622                    name, UTS_RELEASE);
3623         seq_puts(m, "# -----------------------------------"
3624                  "---------------------------------\n");
3625         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
3626                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
3627                    nsecs_to_usecs(data->saved_latency),
3628                    entries,
3629                    total,
3630                    buf->cpu,
3631 #if defined(CONFIG_PREEMPT_NONE)
3632                    "server",
3633 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
3634                    "desktop",
3635 #elif defined(CONFIG_PREEMPT)
3636                    "preempt",
3637 #else
3638                    "unknown",
3639 #endif
3640                    /* These are reserved for later use */
3641                    0, 0, 0, 0);
3642 #ifdef CONFIG_SMP
3643         seq_printf(m, " #P:%d)\n", num_online_cpus());
3644 #else
3645         seq_puts(m, ")\n");
3646 #endif
3647         seq_puts(m, "#    -----------------\n");
3648         seq_printf(m, "#    | task: %.16s-%d "
3649                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
3650                    data->comm, data->pid,
3651                    from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
3652                    data->policy, data->rt_priority);
3653         seq_puts(m, "#    -----------------\n");
3654
3655         if (data->critical_start) {
3656                 seq_puts(m, "#  => started at: ");
3657                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
3658                 trace_print_seq(m, &iter->seq);
3659                 seq_puts(m, "\n#  => ended at:   ");
3660                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
3661                 trace_print_seq(m, &iter->seq);
3662                 seq_puts(m, "\n#\n");
3663         }
3664
3665         seq_puts(m, "#\n");
3666 }
3667
3668 static void test_cpu_buff_start(struct trace_iterator *iter)
3669 {
3670         struct trace_seq *s = &iter->seq;
3671         struct trace_array *tr = iter->tr;
3672
3673         if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
3674                 return;
3675
3676         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
3677                 return;
3678
3679         if (cpumask_available(iter->started) &&
3680             cpumask_test_cpu(iter->cpu, iter->started))
3681                 return;
3682
3683         if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
3684                 return;
3685
3686         if (cpumask_available(iter->started))
3687                 cpumask_set_cpu(iter->cpu, iter->started);
3688
3689         /* Don't print started cpu buffer for the first entry of the trace */
3690         if (iter->idx > 1)
3691                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
3692                                 iter->cpu);
3693 }
3694
3695 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
3696 {
3697         struct trace_array *tr = iter->tr;
3698         struct trace_seq *s = &iter->seq;
3699         unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
3700         struct trace_entry *entry;
3701         struct trace_event *event;
3702
3703         entry = iter->ent;
3704
3705         test_cpu_buff_start(iter);
3706
3707         event = ftrace_find_event(entry->type);
3708
3709         if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3710                 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
3711                         trace_print_lat_context(iter);
3712                 else
3713                         trace_print_context(iter);
3714         }
3715
3716         if (trace_seq_has_overflowed(s))
3717                 return TRACE_TYPE_PARTIAL_LINE;
3718
3719         if (event)
3720                 return event->funcs->trace(iter, sym_flags, event);
3721
3722         trace_seq_printf(s, "Unknown type %d\n", entry->type);
3723
3724         return trace_handle_return(s);
3725 }
3726
3727 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
3728 {
3729         struct trace_array *tr = iter->tr;
3730         struct trace_seq *s = &iter->seq;
3731         struct trace_entry *entry;
3732         struct trace_event *event;
3733
3734         entry = iter->ent;
3735
3736         if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
3737                 trace_seq_printf(s, "%d %d %llu ",
3738                                  entry->pid, iter->cpu, iter->ts);
3739
3740         if (trace_seq_has_overflowed(s))
3741                 return TRACE_TYPE_PARTIAL_LINE;
3742
3743         event = ftrace_find_event(entry->type);
3744         if (event)
3745                 return event->funcs->raw(iter, 0, event);
3746
3747         trace_seq_printf(s, "%d ?\n", entry->type);
3748
3749         return trace_handle_return(s);
3750 }
3751
3752 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
3753 {
3754         struct trace_array *tr = iter->tr;
3755         struct trace_seq *s = &iter->seq;
3756         unsigned char newline = '\n';
3757         struct trace_entry *entry;
3758         struct trace_event *event;
3759
3760         entry = iter->ent;
3761
3762         if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3763                 SEQ_PUT_HEX_FIELD(s, entry->pid);
3764                 SEQ_PUT_HEX_FIELD(s, iter->cpu);
3765                 SEQ_PUT_HEX_FIELD(s, iter->ts);
3766                 if (trace_seq_has_overflowed(s))
3767                         return TRACE_TYPE_PARTIAL_LINE;
3768         }
3769
3770         event = ftrace_find_event(entry->type);
3771         if (event) {
3772                 enum print_line_t ret = event->funcs->hex(iter, 0, event);
3773                 if (ret != TRACE_TYPE_HANDLED)
3774                         return ret;
3775         }
3776
3777         SEQ_PUT_FIELD(s, newline);
3778
3779         return trace_handle_return(s);
3780 }
3781
3782 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
3783 {
3784         struct trace_array *tr = iter->tr;
3785         struct trace_seq *s = &iter->seq;
3786         struct trace_entry *entry;
3787         struct trace_event *event;
3788
3789         entry = iter->ent;
3790
3791         if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3792                 SEQ_PUT_FIELD(s, entry->pid);
3793                 SEQ_PUT_FIELD(s, iter->cpu);
3794                 SEQ_PUT_FIELD(s, iter->ts);
3795                 if (trace_seq_has_overflowed(s))
3796                         return TRACE_TYPE_PARTIAL_LINE;
3797         }
3798
3799         event = ftrace_find_event(entry->type);
3800         return event ? event->funcs->binary(iter, 0, event) :
3801                 TRACE_TYPE_HANDLED;
3802 }
3803
3804 int trace_empty(struct trace_iterator *iter)
3805 {
3806         struct ring_buffer_iter *buf_iter;
3807         int cpu;
3808
3809         /* If we are looking at one CPU buffer, only check that one */
3810         if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
3811                 cpu = iter->cpu_file;
3812                 buf_iter = trace_buffer_iter(iter, cpu);
3813                 if (buf_iter) {
3814                         if (!ring_buffer_iter_empty(buf_iter))
3815                                 return 0;
3816                 } else {
3817                         if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
3818                                 return 0;
3819                 }
3820                 return 1;
3821         }
3822
3823         for_each_tracing_cpu(cpu) {
3824                 buf_iter = trace_buffer_iter(iter, cpu);
3825                 if (buf_iter) {
3826                         if (!ring_buffer_iter_empty(buf_iter))
3827                                 return 0;
3828                 } else {
3829                         if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
3830                                 return 0;
3831                 }
3832         }
3833
3834         return 1;
3835 }
3836
3837 /*  Called with trace_event_read_lock() held. */
3838 enum print_line_t print_trace_line(struct trace_iterator *iter)
3839 {
3840         struct trace_array *tr = iter->tr;
3841         unsigned long trace_flags = tr->trace_flags;
3842         enum print_line_t ret;
3843
3844         if (iter->lost_events) {
3845                 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
3846                                  iter->cpu, iter->lost_events);
3847                 if (trace_seq_has_overflowed(&iter->seq))
3848                         return TRACE_TYPE_PARTIAL_LINE;
3849         }
3850
3851         if (iter->trace && iter->trace->print_line) {
3852                 ret = iter->trace->print_line(iter);
3853                 if (ret != TRACE_TYPE_UNHANDLED)
3854                         return ret;
3855         }
3856
3857         if (iter->ent->type == TRACE_BPUTS &&
3858                         trace_flags & TRACE_ITER_PRINTK &&
3859                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
3860                 return trace_print_bputs_msg_only(iter);
3861
3862         if (iter->ent->type == TRACE_BPRINT &&
3863                         trace_flags & TRACE_ITER_PRINTK &&
3864                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
3865                 return trace_print_bprintk_msg_only(iter);
3866
3867         if (iter->ent->type == TRACE_PRINT &&
3868                         trace_flags & TRACE_ITER_PRINTK &&
3869                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
3870                 return trace_print_printk_msg_only(iter);
3871
3872         if (trace_flags & TRACE_ITER_BIN)
3873                 return print_bin_fmt(iter);
3874
3875         if (trace_flags & TRACE_ITER_HEX)
3876                 return print_hex_fmt(iter);
3877
3878         if (trace_flags & TRACE_ITER_RAW)
3879                 return print_raw_fmt(iter);
3880
3881         return print_trace_fmt(iter);
3882 }
3883
3884 void trace_latency_header(struct seq_file *m)
3885 {
3886         struct trace_iterator *iter = m->private;
3887         struct trace_array *tr = iter->tr;
3888
3889         /* print nothing if the buffers are empty */
3890         if (trace_empty(iter))
3891                 return;
3892
3893         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
3894                 print_trace_header(m, iter);
3895
3896         if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
3897                 print_lat_help_header(m);
3898 }
3899
3900 void trace_default_header(struct seq_file *m)
3901 {
3902         struct trace_iterator *iter = m->private;
3903         struct trace_array *tr = iter->tr;
3904         unsigned long trace_flags = tr->trace_flags;
3905
3906         if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
3907                 return;
3908
3909         if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
3910                 /* print nothing if the buffers are empty */
3911                 if (trace_empty(iter))
3912                         return;
3913                 print_trace_header(m, iter);
3914                 if (!(trace_flags & TRACE_ITER_VERBOSE))
3915                         print_lat_help_header(m);
3916         } else {
3917                 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
3918                         if (trace_flags & TRACE_ITER_IRQ_INFO)
3919                                 print_func_help_header_irq(iter->trace_buffer,
3920                                                            m, trace_flags);
3921                         else
3922                                 print_func_help_header(iter->trace_buffer, m,
3923                                                        trace_flags);
3924                 }
3925         }
3926 }
3927
3928 static void test_ftrace_alive(struct seq_file *m)
3929 {
3930         if (!ftrace_is_dead())
3931                 return;
3932         seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
3933                     "#          MAY BE MISSING FUNCTION EVENTS\n");
3934 }
3935
3936 #ifdef CONFIG_TRACER_MAX_TRACE
3937 static void show_snapshot_main_help(struct seq_file *m)
3938 {
3939         seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
3940                     "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
3941                     "#                      Takes a snapshot of the main buffer.\n"
3942                     "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
3943                     "#                      (Doesn't have to be '2' works with any number that\n"
3944                     "#                       is not a '0' or '1')\n");
3945 }
3946
3947 static void show_snapshot_percpu_help(struct seq_file *m)
3948 {
3949         seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
3950 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
3951         seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
3952                     "#                      Takes a snapshot of the main buffer for this cpu.\n");
3953 #else
3954         seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
3955                     "#                     Must use main snapshot file to allocate.\n");
3956 #endif
3957         seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
3958                     "#                      (Doesn't have to be '2' works with any number that\n"
3959                     "#                       is not a '0' or '1')\n");
3960 }
3961
3962 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
3963 {
3964         if (iter->tr->allocated_snapshot)
3965                 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
3966         else
3967                 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
3968
3969         seq_puts(m, "# Snapshot commands:\n");
3970         if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
3971                 show_snapshot_main_help(m);
3972         else
3973                 show_snapshot_percpu_help(m);
3974 }
3975 #else
3976 /* Should never be called */
3977 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
3978 #endif
3979
3980 static int s_show(struct seq_file *m, void *v)
3981 {
3982         struct trace_iterator *iter = v;
3983         int ret;
3984
3985         if (iter->ent == NULL) {
3986                 if (iter->tr) {
3987                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
3988                         seq_puts(m, "#\n");
3989                         test_ftrace_alive(m);
3990                 }
3991                 if (iter->snapshot && trace_empty(iter))
3992                         print_snapshot_help(m, iter);
3993                 else if (iter->trace && iter->trace->print_header)
3994                         iter->trace->print_header(m);
3995                 else
3996                         trace_default_header(m);
3997
3998         } else if (iter->leftover) {
3999                 /*
4000                  * If we filled the seq_file buffer earlier, we
4001                  * want to just show it now.
4002                  */
4003                 ret = trace_print_seq(m, &iter->seq);
4004
4005                 /* ret should this time be zero, but you never know */
4006                 iter->leftover = ret;
4007
4008         } else {
4009                 print_trace_line(iter);
4010                 ret = trace_print_seq(m, &iter->seq);
4011                 /*
4012                  * If we overflow the seq_file buffer, then it will
4013                  * ask us for this data again at start up.
4014                  * Use that instead.
4015                  *  ret is 0 if seq_file write succeeded.
4016                  *        -1 otherwise.
4017                  */
4018                 iter->leftover = ret;
4019         }
4020
4021         return 0;
4022 }
4023
4024 /*
4025  * Should be used after trace_array_get(), trace_types_lock
4026  * ensures that i_cdev was already initialized.
4027  */
4028 static inline int tracing_get_cpu(struct inode *inode)
4029 {
4030         if (inode->i_cdev) /* See trace_create_cpu_file() */
4031                 return (long)inode->i_cdev - 1;
4032         return RING_BUFFER_ALL_CPUS;
4033 }
4034
4035 static const struct seq_operations tracer_seq_ops = {
4036         .start          = s_start,
4037         .next           = s_next,
4038         .stop           = s_stop,
4039         .show           = s_show,
4040 };
4041
4042 static struct trace_iterator *
4043 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4044 {