2 * ring buffer based function tracer
4 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 Nadia Yvette Chambers
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/tracefs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/vmalloc.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/mount.h>
36 #include <linux/rwsem.h>
37 #include <linux/slab.h>
38 #include <linux/ctype.h>
39 #include <linux/init.h>
40 #include <linux/poll.h>
41 #include <linux/nmi.h>
43 #include <linux/trace.h>
44 #include <linux/sched/rt.h>
47 #include "trace_output.h"
50 * On boot up, the ring buffer is set to the minimum size, so that
51 * we do not waste memory on systems that are not using tracing.
53 bool ring_buffer_expanded;
56 * We need to change this state when a selftest is running.
57 * A selftest will lurk into the ring-buffer to count the
58 * entries inserted during the selftest although some concurrent
59 * insertions into the ring-buffer such as trace_printk could occurred
60 * at the same time, giving false positive or negative results.
62 static bool __read_mostly tracing_selftest_running;
65 * If a tracer is running, we do not want to run SELFTEST.
67 bool __read_mostly tracing_selftest_disabled;
69 /* Pipe tracepoints to printk */
70 struct trace_iterator *tracepoint_print_iter;
71 int tracepoint_printk;
72 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
74 /* For tracers that don't implement custom flags */
75 static struct tracer_opt dummy_tracer_opt[] = {
80 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
86 * To prevent the comm cache from being overwritten when no
87 * tracing is active, only save the comm when a trace event
90 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
93 * Kill all tracing for good (never come back).
94 * It is initialized to 1 but will turn to zero if the initialization
95 * of the tracer is successful. But that is the only place that sets
98 static int tracing_disabled = 1;
100 cpumask_var_t __read_mostly tracing_buffer_mask;
103 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
105 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
106 * is set, then ftrace_dump is called. This will output the contents
107 * of the ftrace buffers to the console. This is very useful for
108 * capturing traces that lead to crashes and outputing it to a
111 * It is default off, but you can enable it with either specifying
112 * "ftrace_dump_on_oops" in the kernel command line, or setting
113 * /proc/sys/kernel/ftrace_dump_on_oops
114 * Set 1 if you want to dump buffers of all CPUs
115 * Set 2 if you want to dump the buffer of the CPU that triggered oops
118 enum ftrace_dump_mode ftrace_dump_on_oops;
120 /* When set, tracing will stop when a WARN*() is hit */
121 int __disable_trace_on_warning;
123 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
124 /* Map of enums to their values, for "eval_map" file */
125 struct trace_eval_map_head {
127 unsigned long length;
130 union trace_eval_map_item;
132 struct trace_eval_map_tail {
134 * "end" is first and points to NULL as it must be different
135 * than "mod" or "eval_string"
137 union trace_eval_map_item *next;
138 const char *end; /* points to NULL */
141 static DEFINE_MUTEX(trace_eval_mutex);
144 * The trace_eval_maps are saved in an array with two extra elements,
145 * one at the beginning, and one at the end. The beginning item contains
146 * the count of the saved maps (head.length), and the module they
147 * belong to if not built in (head.mod). The ending item contains a
148 * pointer to the next array of saved eval_map items.
150 union trace_eval_map_item {
151 struct trace_eval_map map;
152 struct trace_eval_map_head head;
153 struct trace_eval_map_tail tail;
156 static union trace_eval_map_item *trace_eval_maps;
157 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
159 static int tracing_set_tracer(struct trace_array *tr, const char *buf);
161 #define MAX_TRACER_SIZE 100
162 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
163 static char *default_bootup_tracer;
165 static bool allocate_snapshot;
167 static int __init set_cmdline_ftrace(char *str)
169 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
170 default_bootup_tracer = bootup_tracer_buf;
171 /* We are using ftrace early, expand it */
172 ring_buffer_expanded = true;
175 __setup("ftrace=", set_cmdline_ftrace);
177 static int __init set_ftrace_dump_on_oops(char *str)
179 if (*str++ != '=' || !*str) {
180 ftrace_dump_on_oops = DUMP_ALL;
184 if (!strcmp("orig_cpu", str)) {
185 ftrace_dump_on_oops = DUMP_ORIG;
191 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
193 static int __init stop_trace_on_warning(char *str)
195 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
196 __disable_trace_on_warning = 1;
199 __setup("traceoff_on_warning", stop_trace_on_warning);
201 static int __init boot_alloc_snapshot(char *str)
203 allocate_snapshot = true;
204 /* We also need the main ring buffer expanded */
205 ring_buffer_expanded = true;
208 __setup("alloc_snapshot", boot_alloc_snapshot);
211 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
213 static int __init set_trace_boot_options(char *str)
215 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
218 __setup("trace_options=", set_trace_boot_options);
220 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
221 static char *trace_boot_clock __initdata;
223 static int __init set_trace_boot_clock(char *str)
225 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
226 trace_boot_clock = trace_boot_clock_buf;
229 __setup("trace_clock=", set_trace_boot_clock);
231 static int __init set_tracepoint_printk(char *str)
233 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
234 tracepoint_printk = 1;
237 __setup("tp_printk", set_tracepoint_printk);
239 unsigned long long ns2usecs(u64 nsec)
246 /* trace_flags holds trace_options default values */
247 #define TRACE_DEFAULT_FLAGS \
248 (FUNCTION_DEFAULT_FLAGS | \
249 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
250 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
251 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
252 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS)
254 /* trace_options that are only supported by global_trace */
255 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
256 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
258 /* trace_flags that are default zero for instances */
259 #define ZEROED_TRACE_FLAGS \
260 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
263 * The global_trace is the descriptor that holds the top-level tracing
264 * buffers for the live tracing.
266 static struct trace_array global_trace = {
267 .trace_flags = TRACE_DEFAULT_FLAGS,
270 LIST_HEAD(ftrace_trace_arrays);
272 int trace_array_get(struct trace_array *this_tr)
274 struct trace_array *tr;
277 mutex_lock(&trace_types_lock);
278 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
285 mutex_unlock(&trace_types_lock);
290 static void __trace_array_put(struct trace_array *this_tr)
292 WARN_ON(!this_tr->ref);
296 void trace_array_put(struct trace_array *this_tr)
298 mutex_lock(&trace_types_lock);
299 __trace_array_put(this_tr);
300 mutex_unlock(&trace_types_lock);
303 int call_filter_check_discard(struct trace_event_call *call, void *rec,
304 struct ring_buffer *buffer,
305 struct ring_buffer_event *event)
307 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
308 !filter_match_preds(call->filter, rec)) {
309 __trace_event_discard_commit(buffer, event);
316 void trace_free_pid_list(struct trace_pid_list *pid_list)
318 vfree(pid_list->pids);
323 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
324 * @filtered_pids: The list of pids to check
325 * @search_pid: The PID to find in @filtered_pids
327 * Returns true if @search_pid is fonud in @filtered_pids, and false otherwis.
330 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
333 * If pid_max changed after filtered_pids was created, we
334 * by default ignore all pids greater than the previous pid_max.
336 if (search_pid >= filtered_pids->pid_max)
339 return test_bit(search_pid, filtered_pids->pids);
343 * trace_ignore_this_task - should a task be ignored for tracing
344 * @filtered_pids: The list of pids to check
345 * @task: The task that should be ignored if not filtered
347 * Checks if @task should be traced or not from @filtered_pids.
348 * Returns true if @task should *NOT* be traced.
349 * Returns false if @task should be traced.
352 trace_ignore_this_task(struct trace_pid_list *filtered_pids, struct task_struct *task)
355 * Return false, because if filtered_pids does not exist,
356 * all pids are good to trace.
361 return !trace_find_filtered_pid(filtered_pids, task->pid);
365 * trace_pid_filter_add_remove - Add or remove a task from a pid_list
366 * @pid_list: The list to modify
367 * @self: The current task for fork or NULL for exit
368 * @task: The task to add or remove
370 * If adding a task, if @self is defined, the task is only added if @self
371 * is also included in @pid_list. This happens on fork and tasks should
372 * only be added when the parent is listed. If @self is NULL, then the
373 * @task pid will be removed from the list, which would happen on exit
376 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
377 struct task_struct *self,
378 struct task_struct *task)
383 /* For forks, we only add if the forking task is listed */
385 if (!trace_find_filtered_pid(pid_list, self->pid))
389 /* Sorry, but we don't support pid_max changing after setting */
390 if (task->pid >= pid_list->pid_max)
393 /* "self" is set for forks, and NULL for exits */
395 set_bit(task->pid, pid_list->pids);
397 clear_bit(task->pid, pid_list->pids);
401 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
402 * @pid_list: The pid list to show
403 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
404 * @pos: The position of the file
406 * This is used by the seq_file "next" operation to iterate the pids
407 * listed in a trace_pid_list structure.
409 * Returns the pid+1 as we want to display pid of zero, but NULL would
410 * stop the iteration.
412 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
414 unsigned long pid = (unsigned long)v;
418 /* pid already is +1 of the actual prevous bit */
419 pid = find_next_bit(pid_list->pids, pid_list->pid_max, pid);
421 /* Return pid + 1 to allow zero to be represented */
422 if (pid < pid_list->pid_max)
423 return (void *)(pid + 1);
429 * trace_pid_start - Used for seq_file to start reading pid lists
430 * @pid_list: The pid list to show
431 * @pos: The position of the file
433 * This is used by seq_file "start" operation to start the iteration
436 * Returns the pid+1 as we want to display pid of zero, but NULL would
437 * stop the iteration.
439 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
444 pid = find_first_bit(pid_list->pids, pid_list->pid_max);
445 if (pid >= pid_list->pid_max)
448 /* Return pid + 1 so that zero can be the exit value */
449 for (pid++; pid && l < *pos;
450 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
456 * trace_pid_show - show the current pid in seq_file processing
457 * @m: The seq_file structure to write into
458 * @v: A void pointer of the pid (+1) value to display
460 * Can be directly used by seq_file operations to display the current
463 int trace_pid_show(struct seq_file *m, void *v)
465 unsigned long pid = (unsigned long)v - 1;
467 seq_printf(m, "%lu\n", pid);
471 /* 128 should be much more than enough */
472 #define PID_BUF_SIZE 127
474 int trace_pid_write(struct trace_pid_list *filtered_pids,
475 struct trace_pid_list **new_pid_list,
476 const char __user *ubuf, size_t cnt)
478 struct trace_pid_list *pid_list;
479 struct trace_parser parser;
487 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
491 * Always recreate a new array. The write is an all or nothing
492 * operation. Always create a new array when adding new pids by
493 * the user. If the operation fails, then the current list is
496 pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
500 pid_list->pid_max = READ_ONCE(pid_max);
502 /* Only truncating will shrink pid_max */
503 if (filtered_pids && filtered_pids->pid_max > pid_list->pid_max)
504 pid_list->pid_max = filtered_pids->pid_max;
506 pid_list->pids = vzalloc((pid_list->pid_max + 7) >> 3);
507 if (!pid_list->pids) {
513 /* copy the current bits to the new max */
514 for_each_set_bit(pid, filtered_pids->pids,
515 filtered_pids->pid_max) {
516 set_bit(pid, pid_list->pids);
525 ret = trace_get_user(&parser, ubuf, cnt, &pos);
526 if (ret < 0 || !trace_parser_loaded(&parser))
533 parser.buffer[parser.idx] = 0;
536 if (kstrtoul(parser.buffer, 0, &val))
538 if (val >= pid_list->pid_max)
543 set_bit(pid, pid_list->pids);
546 trace_parser_clear(&parser);
549 trace_parser_put(&parser);
552 trace_free_pid_list(pid_list);
557 /* Cleared the list of pids */
558 trace_free_pid_list(pid_list);
563 *new_pid_list = pid_list;
568 static u64 buffer_ftrace_now(struct trace_buffer *buf, int cpu)
572 /* Early boot up does not have a buffer yet */
574 return trace_clock_local();
576 ts = ring_buffer_time_stamp(buf->buffer, cpu);
577 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
582 u64 ftrace_now(int cpu)
584 return buffer_ftrace_now(&global_trace.trace_buffer, cpu);
588 * tracing_is_enabled - Show if global_trace has been disabled
590 * Shows if the global trace has been enabled or not. It uses the
591 * mirror flag "buffer_disabled" to be used in fast paths such as for
592 * the irqsoff tracer. But it may be inaccurate due to races. If you
593 * need to know the accurate state, use tracing_is_on() which is a little
594 * slower, but accurate.
596 int tracing_is_enabled(void)
599 * For quick access (irqsoff uses this in fast path), just
600 * return the mirror variable of the state of the ring buffer.
601 * It's a little racy, but we don't really care.
604 return !global_trace.buffer_disabled;
608 * trace_buf_size is the size in bytes that is allocated
609 * for a buffer. Note, the number of bytes is always rounded
612 * This number is purposely set to a low number of 16384.
613 * If the dump on oops happens, it will be much appreciated
614 * to not have to wait for all that output. Anyway this can be
615 * boot time and run time configurable.
617 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
619 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
621 /* trace_types holds a link list of available tracers. */
622 static struct tracer *trace_types __read_mostly;
625 * trace_types_lock is used to protect the trace_types list.
627 DEFINE_MUTEX(trace_types_lock);
630 * serialize the access of the ring buffer
632 * ring buffer serializes readers, but it is low level protection.
633 * The validity of the events (which returns by ring_buffer_peek() ..etc)
634 * are not protected by ring buffer.
636 * The content of events may become garbage if we allow other process consumes
637 * these events concurrently:
638 * A) the page of the consumed events may become a normal page
639 * (not reader page) in ring buffer, and this page will be rewrited
640 * by events producer.
641 * B) The page of the consumed events may become a page for splice_read,
642 * and this page will be returned to system.
644 * These primitives allow multi process access to different cpu ring buffer
647 * These primitives don't distinguish read-only and read-consume access.
648 * Multi read-only access are also serialized.
652 static DECLARE_RWSEM(all_cpu_access_lock);
653 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
655 static inline void trace_access_lock(int cpu)
657 if (cpu == RING_BUFFER_ALL_CPUS) {
658 /* gain it for accessing the whole ring buffer. */
659 down_write(&all_cpu_access_lock);
661 /* gain it for accessing a cpu ring buffer. */
663 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
664 down_read(&all_cpu_access_lock);
666 /* Secondly block other access to this @cpu ring buffer. */
667 mutex_lock(&per_cpu(cpu_access_lock, cpu));
671 static inline void trace_access_unlock(int cpu)
673 if (cpu == RING_BUFFER_ALL_CPUS) {
674 up_write(&all_cpu_access_lock);
676 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
677 up_read(&all_cpu_access_lock);
681 static inline void trace_access_lock_init(void)
685 for_each_possible_cpu(cpu)
686 mutex_init(&per_cpu(cpu_access_lock, cpu));
691 static DEFINE_MUTEX(access_lock);
693 static inline void trace_access_lock(int cpu)
696 mutex_lock(&access_lock);
699 static inline void trace_access_unlock(int cpu)
702 mutex_unlock(&access_lock);
705 static inline void trace_access_lock_init(void)
711 #ifdef CONFIG_STACKTRACE
712 static void __ftrace_trace_stack(struct ring_buffer *buffer,
714 int skip, int pc, struct pt_regs *regs);
715 static inline void ftrace_trace_stack(struct trace_array *tr,
716 struct ring_buffer *buffer,
718 int skip, int pc, struct pt_regs *regs);
721 static inline void __ftrace_trace_stack(struct ring_buffer *buffer,
723 int skip, int pc, struct pt_regs *regs)
726 static inline void ftrace_trace_stack(struct trace_array *tr,
727 struct ring_buffer *buffer,
729 int skip, int pc, struct pt_regs *regs)
735 static __always_inline void
736 trace_event_setup(struct ring_buffer_event *event,
737 int type, unsigned long flags, int pc)
739 struct trace_entry *ent = ring_buffer_event_data(event);
741 tracing_generic_entry_update(ent, flags, pc);
745 static __always_inline struct ring_buffer_event *
746 __trace_buffer_lock_reserve(struct ring_buffer *buffer,
749 unsigned long flags, int pc)
751 struct ring_buffer_event *event;
753 event = ring_buffer_lock_reserve(buffer, len);
755 trace_event_setup(event, type, flags, pc);
760 void tracer_tracing_on(struct trace_array *tr)
762 if (tr->trace_buffer.buffer)
763 ring_buffer_record_on(tr->trace_buffer.buffer);
765 * This flag is looked at when buffers haven't been allocated
766 * yet, or by some tracers (like irqsoff), that just want to
767 * know if the ring buffer has been disabled, but it can handle
768 * races of where it gets disabled but we still do a record.
769 * As the check is in the fast path of the tracers, it is more
770 * important to be fast than accurate.
772 tr->buffer_disabled = 0;
773 /* Make the flag seen by readers */
778 * tracing_on - enable tracing buffers
780 * This function enables tracing buffers that may have been
781 * disabled with tracing_off.
783 void tracing_on(void)
785 tracer_tracing_on(&global_trace);
787 EXPORT_SYMBOL_GPL(tracing_on);
790 static __always_inline void
791 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
793 __this_cpu_write(trace_taskinfo_save, true);
795 /* If this is the temp buffer, we need to commit fully */
796 if (this_cpu_read(trace_buffered_event) == event) {
797 /* Length is in event->array[0] */
798 ring_buffer_write(buffer, event->array[0], &event->array[1]);
799 /* Release the temp buffer */
800 this_cpu_dec(trace_buffered_event_cnt);
802 ring_buffer_unlock_commit(buffer, event);
806 * __trace_puts - write a constant string into the trace buffer.
807 * @ip: The address of the caller
808 * @str: The constant string to write
809 * @size: The size of the string.
811 int __trace_puts(unsigned long ip, const char *str, int size)
813 struct ring_buffer_event *event;
814 struct ring_buffer *buffer;
815 struct print_entry *entry;
816 unsigned long irq_flags;
820 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
823 pc = preempt_count();
825 if (unlikely(tracing_selftest_running || tracing_disabled))
828 alloc = sizeof(*entry) + size + 2; /* possible \n added */
830 local_save_flags(irq_flags);
831 buffer = global_trace.trace_buffer.buffer;
832 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
837 entry = ring_buffer_event_data(event);
840 memcpy(&entry->buf, str, size);
842 /* Add a newline if necessary */
843 if (entry->buf[size - 1] != '\n') {
844 entry->buf[size] = '\n';
845 entry->buf[size + 1] = '\0';
847 entry->buf[size] = '\0';
849 __buffer_unlock_commit(buffer, event);
850 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
854 EXPORT_SYMBOL_GPL(__trace_puts);
857 * __trace_bputs - write the pointer to a constant string into trace buffer
858 * @ip: The address of the caller
859 * @str: The constant string to write to the buffer to
861 int __trace_bputs(unsigned long ip, const char *str)
863 struct ring_buffer_event *event;
864 struct ring_buffer *buffer;
865 struct bputs_entry *entry;
866 unsigned long irq_flags;
867 int size = sizeof(struct bputs_entry);
870 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
873 pc = preempt_count();
875 if (unlikely(tracing_selftest_running || tracing_disabled))
878 local_save_flags(irq_flags);
879 buffer = global_trace.trace_buffer.buffer;
880 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
885 entry = ring_buffer_event_data(event);
889 __buffer_unlock_commit(buffer, event);
890 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
894 EXPORT_SYMBOL_GPL(__trace_bputs);
896 #ifdef CONFIG_TRACER_SNAPSHOT
897 static void tracing_snapshot_instance(struct trace_array *tr)
899 struct tracer *tracer = tr->current_trace;
903 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
904 internal_trace_puts("*** snapshot is being ignored ***\n");
908 if (!tr->allocated_snapshot) {
909 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
910 internal_trace_puts("*** stopping trace here! ***\n");
915 /* Note, snapshot can not be used when the tracer uses it */
916 if (tracer->use_max_tr) {
917 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
918 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
922 local_irq_save(flags);
923 update_max_tr(tr, current, smp_processor_id());
924 local_irq_restore(flags);
928 * trace_snapshot - take a snapshot of the current buffer.
930 * This causes a swap between the snapshot buffer and the current live
931 * tracing buffer. You can use this to take snapshots of the live
932 * trace when some condition is triggered, but continue to trace.
934 * Note, make sure to allocate the snapshot with either
935 * a tracing_snapshot_alloc(), or by doing it manually
936 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
938 * If the snapshot buffer is not allocated, it will stop tracing.
939 * Basically making a permanent snapshot.
941 void tracing_snapshot(void)
943 struct trace_array *tr = &global_trace;
945 tracing_snapshot_instance(tr);
947 EXPORT_SYMBOL_GPL(tracing_snapshot);
949 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
950 struct trace_buffer *size_buf, int cpu_id);
951 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val);
953 static int alloc_snapshot(struct trace_array *tr)
957 if (!tr->allocated_snapshot) {
959 /* allocate spare buffer */
960 ret = resize_buffer_duplicate_size(&tr->max_buffer,
961 &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
965 tr->allocated_snapshot = true;
971 static void free_snapshot(struct trace_array *tr)
974 * We don't free the ring buffer. instead, resize it because
975 * The max_tr ring buffer has some state (e.g. ring->clock) and
976 * we want preserve it.
978 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
979 set_buffer_entries(&tr->max_buffer, 1);
980 tracing_reset_online_cpus(&tr->max_buffer);
981 tr->allocated_snapshot = false;
985 * tracing_alloc_snapshot - allocate snapshot buffer.
987 * This only allocates the snapshot buffer if it isn't already
988 * allocated - it doesn't also take a snapshot.
990 * This is meant to be used in cases where the snapshot buffer needs
991 * to be set up for events that can't sleep but need to be able to
992 * trigger a snapshot.
994 int tracing_alloc_snapshot(void)
996 struct trace_array *tr = &global_trace;
999 ret = alloc_snapshot(tr);
1004 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1007 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
1009 * This is similar to trace_snapshot(), but it will allocate the
1010 * snapshot buffer if it isn't already allocated. Use this only
1011 * where it is safe to sleep, as the allocation may sleep.
1013 * This causes a swap between the snapshot buffer and the current live
1014 * tracing buffer. You can use this to take snapshots of the live
1015 * trace when some condition is triggered, but continue to trace.
1017 void tracing_snapshot_alloc(void)
1021 ret = tracing_alloc_snapshot();
1027 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1029 void tracing_snapshot(void)
1031 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1033 EXPORT_SYMBOL_GPL(tracing_snapshot);
1034 int tracing_alloc_snapshot(void)
1036 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1039 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1040 void tracing_snapshot_alloc(void)
1045 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1046 #endif /* CONFIG_TRACER_SNAPSHOT */
1048 void tracer_tracing_off(struct trace_array *tr)
1050 if (tr->trace_buffer.buffer)
1051 ring_buffer_record_off(tr->trace_buffer.buffer);
1053 * This flag is looked at when buffers haven't been allocated
1054 * yet, or by some tracers (like irqsoff), that just want to
1055 * know if the ring buffer has been disabled, but it can handle
1056 * races of where it gets disabled but we still do a record.
1057 * As the check is in the fast path of the tracers, it is more
1058 * important to be fast than accurate.
1060 tr->buffer_disabled = 1;
1061 /* Make the flag seen by readers */
1066 * tracing_off - turn off tracing buffers
1068 * This function stops the tracing buffers from recording data.
1069 * It does not disable any overhead the tracers themselves may
1070 * be causing. This function simply causes all recording to
1071 * the ring buffers to fail.
1073 void tracing_off(void)
1075 tracer_tracing_off(&global_trace);
1077 EXPORT_SYMBOL_GPL(tracing_off);
1079 void disable_trace_on_warning(void)
1081 if (__disable_trace_on_warning)
1086 * tracer_tracing_is_on - show real state of ring buffer enabled
1087 * @tr : the trace array to know if ring buffer is enabled
1089 * Shows real state of the ring buffer if it is enabled or not.
1091 int tracer_tracing_is_on(struct trace_array *tr)
1093 if (tr->trace_buffer.buffer)
1094 return ring_buffer_record_is_on(tr->trace_buffer.buffer);
1095 return !tr->buffer_disabled;
1099 * tracing_is_on - show state of ring buffers enabled
1101 int tracing_is_on(void)
1103 return tracer_tracing_is_on(&global_trace);
1105 EXPORT_SYMBOL_GPL(tracing_is_on);
1107 static int __init set_buf_size(char *str)
1109 unsigned long buf_size;
1113 buf_size = memparse(str, &str);
1114 /* nr_entries can not be zero */
1117 trace_buf_size = buf_size;
1120 __setup("trace_buf_size=", set_buf_size);
1122 static int __init set_tracing_thresh(char *str)
1124 unsigned long threshold;
1129 ret = kstrtoul(str, 0, &threshold);
1132 tracing_thresh = threshold * 1000;
1135 __setup("tracing_thresh=", set_tracing_thresh);
1137 unsigned long nsecs_to_usecs(unsigned long nsecs)
1139 return nsecs / 1000;
1143 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1144 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1145 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1146 * of strings in the order that the evals (enum) were defined.
1151 /* These must match the bit postions in trace_iterator_flags */
1152 static const char *trace_options[] = {
1160 int in_ns; /* is this clock in nanoseconds? */
1161 } trace_clocks[] = {
1162 { trace_clock_local, "local", 1 },
1163 { trace_clock_global, "global", 1 },
1164 { trace_clock_counter, "counter", 0 },
1165 { trace_clock_jiffies, "uptime", 0 },
1166 { trace_clock, "perf", 1 },
1167 { ktime_get_mono_fast_ns, "mono", 1 },
1168 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1169 { ktime_get_boot_fast_ns, "boot", 1 },
1174 * trace_parser_get_init - gets the buffer for trace parser
1176 int trace_parser_get_init(struct trace_parser *parser, int size)
1178 memset(parser, 0, sizeof(*parser));
1180 parser->buffer = kmalloc(size, GFP_KERNEL);
1181 if (!parser->buffer)
1184 parser->size = size;
1189 * trace_parser_put - frees the buffer for trace parser
1191 void trace_parser_put(struct trace_parser *parser)
1193 kfree(parser->buffer);
1194 parser->buffer = NULL;
1198 * trace_get_user - reads the user input string separated by space
1199 * (matched by isspace(ch))
1201 * For each string found the 'struct trace_parser' is updated,
1202 * and the function returns.
1204 * Returns number of bytes read.
1206 * See kernel/trace/trace.h for 'struct trace_parser' details.
1208 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1209 size_t cnt, loff_t *ppos)
1216 trace_parser_clear(parser);
1218 ret = get_user(ch, ubuf++);
1226 * The parser is not finished with the last write,
1227 * continue reading the user input without skipping spaces.
1229 if (!parser->cont) {
1230 /* skip white space */
1231 while (cnt && isspace(ch)) {
1232 ret = get_user(ch, ubuf++);
1239 /* only spaces were written */
1249 /* read the non-space input */
1250 while (cnt && !isspace(ch)) {
1251 if (parser->idx < parser->size - 1)
1252 parser->buffer[parser->idx++] = ch;
1257 ret = get_user(ch, ubuf++);
1264 /* We either got finished input or we have to wait for another call. */
1266 parser->buffer[parser->idx] = 0;
1267 parser->cont = false;
1268 } else if (parser->idx < parser->size - 1) {
1269 parser->cont = true;
1270 parser->buffer[parser->idx++] = ch;
1283 /* TODO add a seq_buf_to_buffer() */
1284 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1288 if (trace_seq_used(s) <= s->seq.readpos)
1291 len = trace_seq_used(s) - s->seq.readpos;
1294 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1296 s->seq.readpos += cnt;
1300 unsigned long __read_mostly tracing_thresh;
1302 #ifdef CONFIG_TRACER_MAX_TRACE
1304 * Copy the new maximum trace into the separate maximum-trace
1305 * structure. (this way the maximum trace is permanently saved,
1306 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
1309 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1311 struct trace_buffer *trace_buf = &tr->trace_buffer;
1312 struct trace_buffer *max_buf = &tr->max_buffer;
1313 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1314 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1317 max_buf->time_start = data->preempt_timestamp;
1319 max_data->saved_latency = tr->max_latency;
1320 max_data->critical_start = data->critical_start;
1321 max_data->critical_end = data->critical_end;
1323 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1324 max_data->pid = tsk->pid;
1326 * If tsk == current, then use current_uid(), as that does not use
1327 * RCU. The irq tracer can be called out of RCU scope.
1330 max_data->uid = current_uid();
1332 max_data->uid = task_uid(tsk);
1334 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1335 max_data->policy = tsk->policy;
1336 max_data->rt_priority = tsk->rt_priority;
1338 /* record this tasks comm */
1339 tracing_record_cmdline(tsk);
1343 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1345 * @tsk: the task with the latency
1346 * @cpu: The cpu that initiated the trace.
1348 * Flip the buffers between the @tr and the max_tr and record information
1349 * about which task was the cause of this latency.
1352 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1354 struct ring_buffer *buf;
1359 WARN_ON_ONCE(!irqs_disabled());
1361 if (!tr->allocated_snapshot) {
1362 /* Only the nop tracer should hit this when disabling */
1363 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1367 arch_spin_lock(&tr->max_lock);
1369 buf = tr->trace_buffer.buffer;
1370 tr->trace_buffer.buffer = tr->max_buffer.buffer;
1371 tr->max_buffer.buffer = buf;
1373 __update_max_tr(tr, tsk, cpu);
1374 arch_spin_unlock(&tr->max_lock);
1378 * update_max_tr_single - only copy one trace over, and reset the rest
1380 * @tsk - task with the latency
1381 * @cpu - the cpu of the buffer to copy.
1383 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1386 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1393 WARN_ON_ONCE(!irqs_disabled());
1394 if (!tr->allocated_snapshot) {
1395 /* Only the nop tracer should hit this when disabling */
1396 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1400 arch_spin_lock(&tr->max_lock);
1402 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1404 if (ret == -EBUSY) {
1406 * We failed to swap the buffer due to a commit taking
1407 * place on this CPU. We fail to record, but we reset
1408 * the max trace buffer (no one writes directly to it)
1409 * and flag that it failed.
1411 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1412 "Failed to swap buffers due to commit in progress\n");
1415 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1417 __update_max_tr(tr, tsk, cpu);
1418 arch_spin_unlock(&tr->max_lock);
1420 #endif /* CONFIG_TRACER_MAX_TRACE */
1422 static int wait_on_pipe(struct trace_iterator *iter, bool full)
1424 /* Iterators are static, they should be filled or empty */
1425 if (trace_buffer_iter(iter, iter->cpu_file))
1428 return ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file,
1432 #ifdef CONFIG_FTRACE_STARTUP_TEST
1433 static bool selftests_can_run;
1435 struct trace_selftests {
1436 struct list_head list;
1437 struct tracer *type;
1440 static LIST_HEAD(postponed_selftests);
1442 static int save_selftest(struct tracer *type)
1444 struct trace_selftests *selftest;
1446 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1450 selftest->type = type;
1451 list_add(&selftest->list, &postponed_selftests);
1455 static int run_tracer_selftest(struct tracer *type)
1457 struct trace_array *tr = &global_trace;
1458 struct tracer *saved_tracer = tr->current_trace;
1461 if (!type->selftest || tracing_selftest_disabled)
1465 * If a tracer registers early in boot up (before scheduling is
1466 * initialized and such), then do not run its selftests yet.
1467 * Instead, run it a little later in the boot process.
1469 if (!selftests_can_run)
1470 return save_selftest(type);
1473 * Run a selftest on this tracer.
1474 * Here we reset the trace buffer, and set the current
1475 * tracer to be this tracer. The tracer can then run some
1476 * internal tracing to verify that everything is in order.
1477 * If we fail, we do not register this tracer.
1479 tracing_reset_online_cpus(&tr->trace_buffer);
1481 tr->current_trace = type;
1483 #ifdef CONFIG_TRACER_MAX_TRACE
1484 if (type->use_max_tr) {
1485 /* If we expanded the buffers, make sure the max is expanded too */
1486 if (ring_buffer_expanded)
1487 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1488 RING_BUFFER_ALL_CPUS);
1489 tr->allocated_snapshot = true;
1493 /* the test is responsible for initializing and enabling */
1494 pr_info("Testing tracer %s: ", type->name);
1495 ret = type->selftest(type, tr);
1496 /* the test is responsible for resetting too */
1497 tr->current_trace = saved_tracer;
1499 printk(KERN_CONT "FAILED!\n");
1500 /* Add the warning after printing 'FAILED' */
1504 /* Only reset on passing, to avoid touching corrupted buffers */
1505 tracing_reset_online_cpus(&tr->trace_buffer);
1507 #ifdef CONFIG_TRACER_MAX_TRACE
1508 if (type->use_max_tr) {
1509 tr->allocated_snapshot = false;
1511 /* Shrink the max buffer again */
1512 if (ring_buffer_expanded)
1513 ring_buffer_resize(tr->max_buffer.buffer, 1,
1514 RING_BUFFER_ALL_CPUS);
1518 printk(KERN_CONT "PASSED\n");
1522 static __init int init_trace_selftests(void)
1524 struct trace_selftests *p, *n;
1525 struct tracer *t, **last;
1528 selftests_can_run = true;
1530 mutex_lock(&trace_types_lock);
1532 if (list_empty(&postponed_selftests))
1535 pr_info("Running postponed tracer tests:\n");
1537 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
1538 ret = run_tracer_selftest(p->type);
1539 /* If the test fails, then warn and remove from available_tracers */
1541 WARN(1, "tracer: %s failed selftest, disabling\n",
1543 last = &trace_types;
1544 for (t = trace_types; t; t = t->next) {
1557 mutex_unlock(&trace_types_lock);
1561 core_initcall(init_trace_selftests);
1563 static inline int run_tracer_selftest(struct tracer *type)
1567 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1569 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
1571 static void __init apply_trace_boot_options(void);
1574 * register_tracer - register a tracer with the ftrace system.
1575 * @type - the plugin for the tracer
1577 * Register a new plugin tracer.
1579 int __init register_tracer(struct tracer *type)
1585 pr_info("Tracer must have a name\n");
1589 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1590 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1594 mutex_lock(&trace_types_lock);
1596 tracing_selftest_running = true;
1598 for (t = trace_types; t; t = t->next) {
1599 if (strcmp(type->name, t->name) == 0) {
1601 pr_info("Tracer %s already registered\n",
1608 if (!type->set_flag)
1609 type->set_flag = &dummy_set_flag;
1611 /*allocate a dummy tracer_flags*/
1612 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
1617 type->flags->val = 0;
1618 type->flags->opts = dummy_tracer_opt;
1620 if (!type->flags->opts)
1621 type->flags->opts = dummy_tracer_opt;
1623 /* store the tracer for __set_tracer_option */
1624 type->flags->trace = type;
1626 ret = run_tracer_selftest(type);
1630 type->next = trace_types;
1632 add_tracer_options(&global_trace, type);
1635 tracing_selftest_running = false;
1636 mutex_unlock(&trace_types_lock);
1638 if (ret || !default_bootup_tracer)
1641 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1644 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1645 /* Do we want this tracer to start on bootup? */
1646 tracing_set_tracer(&global_trace, type->name);
1647 default_bootup_tracer = NULL;
1649 apply_trace_boot_options();
1651 /* disable other selftests, since this will break it. */
1652 tracing_selftest_disabled = true;
1653 #ifdef CONFIG_FTRACE_STARTUP_TEST
1654 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1662 void tracing_reset(struct trace_buffer *buf, int cpu)
1664 struct ring_buffer *buffer = buf->buffer;
1669 ring_buffer_record_disable(buffer);
1671 /* Make sure all commits have finished */
1672 synchronize_sched();
1673 ring_buffer_reset_cpu(buffer, cpu);
1675 ring_buffer_record_enable(buffer);
1678 void tracing_reset_online_cpus(struct trace_buffer *buf)
1680 struct ring_buffer *buffer = buf->buffer;
1686 ring_buffer_record_disable(buffer);
1688 /* Make sure all commits have finished */
1689 synchronize_sched();
1691 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
1693 for_each_online_cpu(cpu)
1694 ring_buffer_reset_cpu(buffer, cpu);
1696 ring_buffer_record_enable(buffer);
1699 /* Must have trace_types_lock held */
1700 void tracing_reset_all_online_cpus(void)
1702 struct trace_array *tr;
1704 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1705 if (!tr->clear_trace)
1707 tr->clear_trace = false;
1708 tracing_reset_online_cpus(&tr->trace_buffer);
1709 #ifdef CONFIG_TRACER_MAX_TRACE
1710 tracing_reset_online_cpus(&tr->max_buffer);
1715 static int *tgid_map;
1717 #define SAVED_CMDLINES_DEFAULT 128
1718 #define NO_CMDLINE_MAP UINT_MAX
1719 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1720 struct saved_cmdlines_buffer {
1721 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1722 unsigned *map_cmdline_to_pid;
1723 unsigned cmdline_num;
1725 char *saved_cmdlines;
1727 static struct saved_cmdlines_buffer *savedcmd;
1729 /* temporary disable recording */
1730 static atomic_t trace_record_taskinfo_disabled __read_mostly;
1732 static inline char *get_saved_cmdlines(int idx)
1734 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
1737 static inline void set_cmdline(int idx, const char *cmdline)
1739 memcpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
1742 static int allocate_cmdlines_buffer(unsigned int val,
1743 struct saved_cmdlines_buffer *s)
1745 s->map_cmdline_to_pid = kmalloc(val * sizeof(*s->map_cmdline_to_pid),
1747 if (!s->map_cmdline_to_pid)
1750 s->saved_cmdlines = kmalloc(val * TASK_COMM_LEN, GFP_KERNEL);
1751 if (!s->saved_cmdlines) {
1752 kfree(s->map_cmdline_to_pid);
1757 s->cmdline_num = val;
1758 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
1759 sizeof(s->map_pid_to_cmdline));
1760 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
1761 val * sizeof(*s->map_cmdline_to_pid));
1766 static int trace_create_savedcmd(void)
1770 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
1774 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
1784 int is_tracing_stopped(void)
1786 return global_trace.stop_count;
1790 * tracing_start - quick start of the tracer
1792 * If tracing is enabled but was stopped by tracing_stop,
1793 * this will start the tracer back up.
1795 void tracing_start(void)
1797 struct ring_buffer *buffer;
1798 unsigned long flags;
1800 if (tracing_disabled)
1803 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1804 if (--global_trace.stop_count) {
1805 if (global_trace.stop_count < 0) {
1806 /* Someone screwed up their debugging */
1808 global_trace.stop_count = 0;
1813 /* Prevent the buffers from switching */
1814 arch_spin_lock(&global_trace.max_lock);
1816 buffer = global_trace.trace_buffer.buffer;
1818 ring_buffer_record_enable(buffer);
1820 #ifdef CONFIG_TRACER_MAX_TRACE
1821 buffer = global_trace.max_buffer.buffer;
1823 ring_buffer_record_enable(buffer);
1826 arch_spin_unlock(&global_trace.max_lock);
1829 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1832 static void tracing_start_tr(struct trace_array *tr)
1834 struct ring_buffer *buffer;
1835 unsigned long flags;
1837 if (tracing_disabled)
1840 /* If global, we need to also start the max tracer */
1841 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1842 return tracing_start();
1844 raw_spin_lock_irqsave(&tr->start_lock, flags);
1846 if (--tr->stop_count) {
1847 if (tr->stop_count < 0) {
1848 /* Someone screwed up their debugging */
1855 buffer = tr->trace_buffer.buffer;
1857 ring_buffer_record_enable(buffer);
1860 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1864 * tracing_stop - quick stop of the tracer
1866 * Light weight way to stop tracing. Use in conjunction with
1869 void tracing_stop(void)
1871 struct ring_buffer *buffer;
1872 unsigned long flags;
1874 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1875 if (global_trace.stop_count++)
1878 /* Prevent the buffers from switching */
1879 arch_spin_lock(&global_trace.max_lock);
1881 buffer = global_trace.trace_buffer.buffer;
1883 ring_buffer_record_disable(buffer);
1885 #ifdef CONFIG_TRACER_MAX_TRACE
1886 buffer = global_trace.max_buffer.buffer;
1888 ring_buffer_record_disable(buffer);
1891 arch_spin_unlock(&global_trace.max_lock);
1894 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1897 static void tracing_stop_tr(struct trace_array *tr)
1899 struct ring_buffer *buffer;
1900 unsigned long flags;
1902 /* If global, we need to also stop the max tracer */
1903 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1904 return tracing_stop();
1906 raw_spin_lock_irqsave(&tr->start_lock, flags);
1907 if (tr->stop_count++)
1910 buffer = tr->trace_buffer.buffer;
1912 ring_buffer_record_disable(buffer);
1915 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1918 static int trace_save_cmdline(struct task_struct *tsk)
1922 /* treat recording of idle task as a success */
1926 if (unlikely(tsk->pid > PID_MAX_DEFAULT))
1930 * It's not the end of the world if we don't get
1931 * the lock, but we also don't want to spin
1932 * nor do we want to disable interrupts,
1933 * so if we miss here, then better luck next time.
1935 if (!arch_spin_trylock(&trace_cmdline_lock))
1938 idx = savedcmd->map_pid_to_cmdline[tsk->pid];
1939 if (idx == NO_CMDLINE_MAP) {
1940 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
1943 * Check whether the cmdline buffer at idx has a pid
1944 * mapped. We are going to overwrite that entry so we
1945 * need to clear the map_pid_to_cmdline. Otherwise we
1946 * would read the new comm for the old pid.
1948 pid = savedcmd->map_cmdline_to_pid[idx];
1949 if (pid != NO_CMDLINE_MAP)
1950 savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1952 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
1953 savedcmd->map_pid_to_cmdline[tsk->pid] = idx;
1955 savedcmd->cmdline_idx = idx;
1958 set_cmdline(idx, tsk->comm);
1960 arch_spin_unlock(&trace_cmdline_lock);
1965 static void __trace_find_cmdline(int pid, char comm[])
1970 strcpy(comm, "<idle>");
1974 if (WARN_ON_ONCE(pid < 0)) {
1975 strcpy(comm, "<XXX>");
1979 if (pid > PID_MAX_DEFAULT) {
1980 strcpy(comm, "<...>");
1984 map = savedcmd->map_pid_to_cmdline[pid];
1985 if (map != NO_CMDLINE_MAP)
1986 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
1988 strcpy(comm, "<...>");
1991 void trace_find_cmdline(int pid, char comm[])
1994 arch_spin_lock(&trace_cmdline_lock);
1996 __trace_find_cmdline(pid, comm);
1998 arch_spin_unlock(&trace_cmdline_lock);
2002 int trace_find_tgid(int pid)
2004 if (unlikely(!tgid_map || !pid || pid > PID_MAX_DEFAULT))
2007 return tgid_map[pid];
2010 static int trace_save_tgid(struct task_struct *tsk)
2012 /* treat recording of idle task as a success */
2016 if (unlikely(!tgid_map || tsk->pid > PID_MAX_DEFAULT))
2019 tgid_map[tsk->pid] = tsk->tgid;
2023 static bool tracing_record_taskinfo_skip(int flags)
2025 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2027 if (atomic_read(&trace_record_taskinfo_disabled) || !tracing_is_on())
2029 if (!__this_cpu_read(trace_taskinfo_save))
2035 * tracing_record_taskinfo - record the task info of a task
2037 * @task - task to record
2038 * @flags - TRACE_RECORD_CMDLINE for recording comm
2039 * - TRACE_RECORD_TGID for recording tgid
2041 void tracing_record_taskinfo(struct task_struct *task, int flags)
2045 if (tracing_record_taskinfo_skip(flags))
2049 * Record as much task information as possible. If some fail, continue
2050 * to try to record the others.
2052 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2053 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2055 /* If recording any information failed, retry again soon. */
2059 __this_cpu_write(trace_taskinfo_save, false);
2063 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2065 * @prev - previous task during sched_switch
2066 * @next - next task during sched_switch
2067 * @flags - TRACE_RECORD_CMDLINE for recording comm
2068 * TRACE_RECORD_TGID for recording tgid
2070 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2071 struct task_struct *next, int flags)
2075 if (tracing_record_taskinfo_skip(flags))
2079 * Record as much task information as possible. If some fail, continue
2080 * to try to record the others.
2082 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2083 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2084 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2085 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2087 /* If recording any information failed, retry again soon. */
2091 __this_cpu_write(trace_taskinfo_save, false);
2094 /* Helpers to record a specific task information */
2095 void tracing_record_cmdline(struct task_struct *task)
2097 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2100 void tracing_record_tgid(struct task_struct *task)
2102 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2106 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2107 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2108 * simplifies those functions and keeps them in sync.
2110 enum print_line_t trace_handle_return(struct trace_seq *s)
2112 return trace_seq_has_overflowed(s) ?
2113 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2115 EXPORT_SYMBOL_GPL(trace_handle_return);
2118 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
2121 struct task_struct *tsk = current;
2123 entry->preempt_count = pc & 0xff;
2124 entry->pid = (tsk) ? tsk->pid : 0;
2126 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
2127 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
2129 TRACE_FLAG_IRQS_NOSUPPORT |
2131 ((pc & NMI_MASK ) ? TRACE_FLAG_NMI : 0) |
2132 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
2133 ((pc & SOFTIRQ_OFFSET) ? TRACE_FLAG_SOFTIRQ : 0) |
2134 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
2135 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
2137 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
2139 struct ring_buffer_event *
2140 trace_buffer_lock_reserve(struct ring_buffer *buffer,
2143 unsigned long flags, int pc)
2145 return __trace_buffer_lock_reserve(buffer, type, len, flags, pc);
2148 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2149 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2150 static int trace_buffered_event_ref;
2153 * trace_buffered_event_enable - enable buffering events
2155 * When events are being filtered, it is quicker to use a temporary
2156 * buffer to write the event data into if there's a likely chance
2157 * that it will not be committed. The discard of the ring buffer
2158 * is not as fast as committing, and is much slower than copying
2161 * When an event is to be filtered, allocate per cpu buffers to
2162 * write the event data into, and if the event is filtered and discarded
2163 * it is simply dropped, otherwise, the entire data is to be committed
2166 void trace_buffered_event_enable(void)
2168 struct ring_buffer_event *event;
2172 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2174 if (trace_buffered_event_ref++)
2177 for_each_tracing_cpu(cpu) {
2178 page = alloc_pages_node(cpu_to_node(cpu),
2179 GFP_KERNEL | __GFP_NORETRY, 0);
2183 event = page_address(page);
2184 memset(event, 0, sizeof(*event));
2186 per_cpu(trace_buffered_event, cpu) = event;
2189 if (cpu == smp_processor_id() &&
2190 this_cpu_read(trace_buffered_event) !=
2191 per_cpu(trace_buffered_event, cpu))
2198 trace_buffered_event_disable();
2201 static void enable_trace_buffered_event(void *data)
2203 /* Probably not needed, but do it anyway */
2205 this_cpu_dec(trace_buffered_event_cnt);
2208 static void disable_trace_buffered_event(void *data)
2210 this_cpu_inc(trace_buffered_event_cnt);
2214 * trace_buffered_event_disable - disable buffering events
2216 * When a filter is removed, it is faster to not use the buffered
2217 * events, and to commit directly into the ring buffer. Free up
2218 * the temp buffers when there are no more users. This requires
2219 * special synchronization with current events.
2221 void trace_buffered_event_disable(void)
2225 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2227 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2230 if (--trace_buffered_event_ref)
2234 /* For each CPU, set the buffer as used. */
2235 smp_call_function_many(tracing_buffer_mask,
2236 disable_trace_buffered_event, NULL, 1);
2239 /* Wait for all current users to finish */
2240 synchronize_sched();
2242 for_each_tracing_cpu(cpu) {
2243 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2244 per_cpu(trace_buffered_event, cpu) = NULL;
2247 * Make sure trace_buffered_event is NULL before clearing
2248 * trace_buffered_event_cnt.
2253 /* Do the work on each cpu */
2254 smp_call_function_many(tracing_buffer_mask,
2255 enable_trace_buffered_event, NULL, 1);
2259 static struct ring_buffer *temp_buffer;
2261 struct ring_buffer_event *
2262 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
2263 struct trace_event_file *trace_file,
2264 int type, unsigned long len,
2265 unsigned long flags, int pc)
2267 struct ring_buffer_event *entry;
2270 *current_rb = trace_file->tr->trace_buffer.buffer;
2272 if ((trace_file->flags &
2273 (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED)) &&
2274 (entry = this_cpu_read(trace_buffered_event))) {
2275 /* Try to use the per cpu buffer first */
2276 val = this_cpu_inc_return(trace_buffered_event_cnt);
2278 trace_event_setup(entry, type, flags, pc);
2279 entry->array[0] = len;
2282 this_cpu_dec(trace_buffered_event_cnt);
2285 entry = __trace_buffer_lock_reserve(*current_rb,
2286 type, len, flags, pc);
2288 * If tracing is off, but we have triggers enabled
2289 * we still need to look at the event data. Use the temp_buffer
2290 * to store the trace event for the tigger to use. It's recusive
2291 * safe and will not be recorded anywhere.
2293 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2294 *current_rb = temp_buffer;
2295 entry = __trace_buffer_lock_reserve(*current_rb,
2296 type, len, flags, pc);
2300 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2302 static DEFINE_SPINLOCK(tracepoint_iter_lock);
2303 static DEFINE_MUTEX(tracepoint_printk_mutex);
2305 static void output_printk(struct trace_event_buffer *fbuffer)
2307 struct trace_event_call *event_call;
2308 struct trace_event *event;
2309 unsigned long flags;
2310 struct trace_iterator *iter = tracepoint_print_iter;
2312 /* We should never get here if iter is NULL */
2313 if (WARN_ON_ONCE(!iter))
2316 event_call = fbuffer->trace_file->event_call;
2317 if (!event_call || !event_call->event.funcs ||
2318 !event_call->event.funcs->trace)
2321 event = &fbuffer->trace_file->event_call->event;
2323 spin_lock_irqsave(&tracepoint_iter_lock, flags);
2324 trace_seq_init(&iter->seq);
2325 iter->ent = fbuffer->entry;
2326 event_call->event.funcs->trace(iter, 0, event);
2327 trace_seq_putc(&iter->seq, 0);
2328 printk("%s", iter->seq.buffer);
2330 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2333 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2334 void __user *buffer, size_t *lenp,
2337 int save_tracepoint_printk;
2340 mutex_lock(&tracepoint_printk_mutex);
2341 save_tracepoint_printk = tracepoint_printk;
2343 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2346 * This will force exiting early, as tracepoint_printk
2347 * is always zero when tracepoint_printk_iter is not allocated
2349 if (!tracepoint_print_iter)
2350 tracepoint_printk = 0;
2352 if (save_tracepoint_printk == tracepoint_printk)
2355 if (tracepoint_printk)
2356 static_key_enable(&tracepoint_printk_key.key);
2358 static_key_disable(&tracepoint_printk_key.key);
2361 mutex_unlock(&tracepoint_printk_mutex);
2366 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2368 if (static_key_false(&tracepoint_printk_key.key))
2369 output_printk(fbuffer);
2371 event_trigger_unlock_commit(fbuffer->trace_file, fbuffer->buffer,
2372 fbuffer->event, fbuffer->entry,
2373 fbuffer->flags, fbuffer->pc);
2375 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2377 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2378 struct ring_buffer *buffer,
2379 struct ring_buffer_event *event,
2380 unsigned long flags, int pc,
2381 struct pt_regs *regs)
2383 __buffer_unlock_commit(buffer, event);
2386 * If regs is not set, then skip the following callers:
2387 * trace_buffer_unlock_commit_regs
2388 * event_trigger_unlock_commit
2389 * trace_event_buffer_commit
2390 * trace_event_raw_event_sched_switch
2391 * Note, we can still get here via blktrace, wakeup tracer
2392 * and mmiotrace, but that's ok if they lose a function or
2393 * two. They are that meaningful.
2395 ftrace_trace_stack(tr, buffer, flags, regs ? 0 : 4, pc, regs);
2396 ftrace_trace_userstack(buffer, flags, pc);
2400 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2403 trace_buffer_unlock_commit_nostack(struct ring_buffer *buffer,
2404 struct ring_buffer_event *event)
2406 __buffer_unlock_commit(buffer, event);
2410 trace_process_export(struct trace_export *export,
2411 struct ring_buffer_event *event)
2413 struct trace_entry *entry;
2414 unsigned int size = 0;
2416 entry = ring_buffer_event_data(event);
2417 size = ring_buffer_event_length(event);
2418 export->write(entry, size);
2421 static DEFINE_MUTEX(ftrace_export_lock);
2423 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
2425 static DEFINE_STATIC_KEY_FALSE(ftrace_exports_enabled);
2427 static inline void ftrace_exports_enable(void)
2429 static_branch_enable(&ftrace_exports_enabled);
2432 static inline void ftrace_exports_disable(void)
2434 static_branch_disable(&ftrace_exports_enabled);
2437 void ftrace_exports(struct ring_buffer_event *event)
2439 struct trace_export *export;
2441 preempt_disable_notrace();
2443 export = rcu_dereference_raw_notrace(ftrace_exports_list);
2445 trace_process_export(export, event);
2446 export = rcu_dereference_raw_notrace(export->next);
2449 preempt_enable_notrace();
2453 add_trace_export(struct trace_export **list, struct trace_export *export)
2455 rcu_assign_pointer(export->next, *list);
2457 * We are entering export into the list but another
2458 * CPU might be walking that list. We need to make sure
2459 * the export->next pointer is valid before another CPU sees
2460 * the export pointer included into the list.
2462 rcu_assign_pointer(*list, export);
2466 rm_trace_export(struct trace_export **list, struct trace_export *export)
2468 struct trace_export **p;
2470 for (p = list; *p != NULL; p = &(*p)->next)
2477 rcu_assign_pointer(*p, (*p)->next);
2483 add_ftrace_export(struct trace_export **list, struct trace_export *export)
2486 ftrace_exports_enable();
2488 add_trace_export(list, export);
2492 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
2496 ret = rm_trace_export(list, export);
2498 ftrace_exports_disable();
2503 int register_ftrace_export(struct trace_export *export)
2505 if (WARN_ON_ONCE(!export->write))
2508 mutex_lock(&ftrace_export_lock);
2510 add_ftrace_export(&ftrace_exports_list, export);
2512 mutex_unlock(&ftrace_export_lock);
2516 EXPORT_SYMBOL_GPL(register_ftrace_export);
2518 int unregister_ftrace_export(struct trace_export *export)
2522 mutex_lock(&ftrace_export_lock);
2524 ret = rm_ftrace_export(&ftrace_exports_list, export);
2526 mutex_unlock(&ftrace_export_lock);
2530 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
2533 trace_function(struct trace_array *tr,
2534 unsigned long ip, unsigned long parent_ip, unsigned long flags,
2537 struct trace_event_call *call = &event_function;
2538 struct ring_buffer *buffer = tr->trace_buffer.buffer;
2539 struct ring_buffer_event *event;
2540 struct ftrace_entry *entry;
2542 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2546 entry = ring_buffer_event_data(event);
2548 entry->parent_ip = parent_ip;
2550 if (!call_filter_check_discard(call, entry, buffer, event)) {
2551 if (static_branch_unlikely(&ftrace_exports_enabled))
2552 ftrace_exports(event);
2553 __buffer_unlock_commit(buffer, event);
2557 #ifdef CONFIG_STACKTRACE
2559 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
2560 struct ftrace_stack {
2561 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
2564 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
2565 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2567 static void __ftrace_trace_stack(struct ring_buffer *buffer,
2568 unsigned long flags,
2569 int skip, int pc, struct pt_regs *regs)
2571 struct trace_event_call *call = &event_kernel_stack;
2572 struct ring_buffer_event *event;
2573 struct stack_entry *entry;
2574 struct stack_trace trace;
2576 int size = FTRACE_STACK_ENTRIES;
2578 trace.nr_entries = 0;
2582 * Add two, for this function and the call to save_stack_trace()
2583 * If regs is set, then these functions will not be in the way.
2589 * Since events can happen in NMIs there's no safe way to
2590 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
2591 * or NMI comes in, it will just have to use the default
2592 * FTRACE_STACK_SIZE.
2594 preempt_disable_notrace();
2596 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
2598 * We don't need any atomic variables, just a barrier.
2599 * If an interrupt comes in, we don't care, because it would
2600 * have exited and put the counter back to what we want.
2601 * We just need a barrier to keep gcc from moving things
2605 if (use_stack == 1) {
2606 trace.entries = this_cpu_ptr(ftrace_stack.calls);
2607 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
2610 save_stack_trace_regs(regs, &trace);
2612 save_stack_trace(&trace);
2614 if (trace.nr_entries > size)
2615 size = trace.nr_entries;
2617 /* From now on, use_stack is a boolean */
2620 size *= sizeof(unsigned long);
2622 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2623 sizeof(*entry) + size, flags, pc);
2626 entry = ring_buffer_event_data(event);
2628 memset(&entry->caller, 0, size);
2631 memcpy(&entry->caller, trace.entries,
2632 trace.nr_entries * sizeof(unsigned long));
2634 trace.max_entries = FTRACE_STACK_ENTRIES;
2635 trace.entries = entry->caller;
2637 save_stack_trace_regs(regs, &trace);
2639 save_stack_trace(&trace);
2642 entry->size = trace.nr_entries;
2644 if (!call_filter_check_discard(call, entry, buffer, event))
2645 __buffer_unlock_commit(buffer, event);
2648 /* Again, don't let gcc optimize things here */
2650 __this_cpu_dec(ftrace_stack_reserve);
2651 preempt_enable_notrace();
2655 static inline void ftrace_trace_stack(struct trace_array *tr,
2656 struct ring_buffer *buffer,
2657 unsigned long flags,
2658 int skip, int pc, struct pt_regs *regs)
2660 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
2663 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
2666 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
2669 struct ring_buffer *buffer = tr->trace_buffer.buffer;
2671 if (rcu_is_watching()) {
2672 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
2677 * When an NMI triggers, RCU is enabled via rcu_nmi_enter(),
2678 * but if the above rcu_is_watching() failed, then the NMI
2679 * triggered someplace critical, and rcu_irq_enter() should
2680 * not be called from NMI.
2682 if (unlikely(in_nmi()))
2686 * It is possible that a function is being traced in a
2687 * location that RCU is not watching. A call to
2688 * rcu_irq_enter() will make sure that it is, but there's
2689 * a few internal rcu functions that could be traced
2690 * where that wont work either. In those cases, we just
2693 if (unlikely(rcu_irq_enter_disabled()))
2696 rcu_irq_enter_irqson();
2697 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
2698 rcu_irq_exit_irqson();
2702 * trace_dump_stack - record a stack back trace in the trace buffer
2703 * @skip: Number of functions to skip (helper handlers)
2705 void trace_dump_stack(int skip)
2707 unsigned long flags;
2709 if (tracing_disabled || tracing_selftest_running)
2712 local_save_flags(flags);
2715 * Skip 3 more, seems to get us at the caller of
2719 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
2720 flags, skip, preempt_count(), NULL);
2723 static DEFINE_PER_CPU(int, user_stack_count);
2726 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
2728 struct trace_event_call *call = &event_user_stack;
2729 struct ring_buffer_event *event;
2730 struct userstack_entry *entry;
2731 struct stack_trace trace;
2733 if (!(global_trace.trace_flags & TRACE_ITER_USERSTACKTRACE))
2737 * NMIs can not handle page faults, even with fix ups.
2738 * The save user stack can (and often does) fault.
2740 if (unlikely(in_nmi()))
2744 * prevent recursion, since the user stack tracing may
2745 * trigger other kernel events.
2748 if (__this_cpu_read(user_stack_count))
2751 __this_cpu_inc(user_stack_count);
2753 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
2754 sizeof(*entry), flags, pc);
2756 goto out_drop_count;
2757 entry = ring_buffer_event_data(event);
2759 entry->tgid = current->tgid;
2760 memset(&entry->caller, 0, sizeof(entry->caller));
2762 trace.nr_entries = 0;
2763 trace.max_entries = FTRACE_STACK_ENTRIES;
2765 trace.entries = entry->caller;
2767 save_stack_trace_user(&trace);
2768 if (!call_filter_check_discard(call, entry, buffer, event))
2769 __buffer_unlock_commit(buffer, event);
2772 __this_cpu_dec(user_stack_count);
2778 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
2780 ftrace_trace_userstack(tr, flags, preempt_count());
2784 #endif /* CONFIG_STACKTRACE */
2786 /* created for use with alloc_percpu */
2787 struct trace_buffer_struct {
2789 char buffer[4][TRACE_BUF_SIZE];
2792 static struct trace_buffer_struct *trace_percpu_buffer;
2795 * Thise allows for lockless recording. If we're nested too deeply, then
2796 * this returns NULL.
2798 static char *get_trace_buf(void)
2800 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
2802 if (!buffer || buffer->nesting >= 4)
2807 /* Interrupts must see nesting incremented before we use the buffer */
2809 return &buffer->buffer[buffer->nesting][0];
2812 static void put_trace_buf(void)
2814 /* Don't let the decrement of nesting leak before this */
2816 this_cpu_dec(trace_percpu_buffer->nesting);
2819 static int alloc_percpu_trace_buffer(void)
2821 struct trace_buffer_struct *buffers;
2823 buffers = alloc_percpu(struct trace_buffer_struct);
2824 if (WARN(!buffers, "Could not allocate percpu trace_printk buffer"))
2827 trace_percpu_buffer = buffers;
2831 static int buffers_allocated;
2833 void trace_printk_init_buffers(void)
2835 if (buffers_allocated)
2838 if (alloc_percpu_trace_buffer())
2841 /* trace_printk() is for debug use only. Don't use it in production. */
2844 pr_warn("**********************************************************\n");
2845 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
2847 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
2849 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
2850 pr_warn("** unsafe for production use. **\n");
2852 pr_warn("** If you see this message and you are not debugging **\n");
2853 pr_warn("** the kernel, report this immediately to your vendor! **\n");
2855 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
2856 pr_warn("**********************************************************\n");
2858 /* Expand the buffers to set size */
2859 tracing_update_buffers();
2861 buffers_allocated = 1;
2864 * trace_printk_init_buffers() can be called by modules.
2865 * If that happens, then we need to start cmdline recording
2866 * directly here. If the global_trace.buffer is already
2867 * allocated here, then this was called by module code.
2869 if (global_trace.trace_buffer.buffer)
2870 tracing_start_cmdline_record();
2873 void trace_printk_start_comm(void)
2875 /* Start tracing comms if trace printk is set */
2876 if (!buffers_allocated)
2878 tracing_start_cmdline_record();
2881 static void trace_printk_start_stop_comm(int enabled)
2883 if (!buffers_allocated)
2887 tracing_start_cmdline_record();
2889 tracing_stop_cmdline_record();
2893 * trace_vbprintk - write binary msg to tracing buffer
2896 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
2898 struct trace_event_call *call = &event_bprint;
2899 struct ring_buffer_event *event;
2900 struct ring_buffer *buffer;
2901 struct trace_array *tr = &global_trace;
2902 struct bprint_entry *entry;
2903 unsigned long flags;
2905 int len = 0, size, pc;
2907 if (unlikely(tracing_selftest_running || tracing_disabled))
2910 /* Don't pollute graph traces with trace_vprintk internals */
2911 pause_graph_tracing();
2913 pc = preempt_count();
2914 preempt_disable_notrace();
2916 tbuffer = get_trace_buf();
2922 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
2924 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
2927 local_save_flags(flags);
2928 size = sizeof(*entry) + sizeof(u32) * len;
2929 buffer = tr->trace_buffer.buffer;
2930 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2934 entry = ring_buffer_event_data(event);
2938 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2939 if (!call_filter_check_discard(call, entry, buffer, event)) {
2940 __buffer_unlock_commit(buffer, event);
2941 ftrace_trace_stack(tr, buffer, flags, 6, pc, NULL);
2948 preempt_enable_notrace();
2949 unpause_graph_tracing();
2953 EXPORT_SYMBOL_GPL(trace_vbprintk);
2956 __trace_array_vprintk(struct ring_buffer *buffer,
2957 unsigned long ip, const char *fmt, va_list args)
2959 struct trace_event_call *call = &event_print;
2960 struct ring_buffer_event *event;
2961 int len = 0, size, pc;
2962 struct print_entry *entry;
2963 unsigned long flags;
2966 if (tracing_disabled || tracing_selftest_running)
2969 /* Don't pollute graph traces with trace_vprintk internals */
2970 pause_graph_tracing();
2972 pc = preempt_count();
2973 preempt_disable_notrace();
2976 tbuffer = get_trace_buf();
2982 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2984 local_save_flags(flags);
2985 size = sizeof(*entry) + len + 1;
2986 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2990 entry = ring_buffer_event_data(event);
2993 memcpy(&entry->buf, tbuffer, len + 1);
2994 if (!call_filter_check_discard(call, entry, buffer, event)) {
2995 __buffer_unlock_commit(buffer, event);
2996 ftrace_trace_stack(&global_trace, buffer, flags, 6, pc, NULL);
3003 preempt_enable_notrace();
3004 unpause_graph_tracing();
3009 int trace_array_vprintk(struct trace_array *tr,
3010 unsigned long ip, const char *fmt, va_list args)
3012 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
3015 int trace_array_printk(struct trace_array *tr,
3016 unsigned long ip, const char *fmt, ...)
3021 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3025 ret = trace_array_vprintk(tr, ip, fmt, ap);
3030 int trace_array_printk_buf(struct ring_buffer *buffer,
3031 unsigned long ip, const char *fmt, ...)
3036 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3040 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3045 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3047 return trace_array_vprintk(&global_trace, ip, fmt, args);
3049 EXPORT_SYMBOL_GPL(trace_vprintk);
3051 static void trace_iterator_increment(struct trace_iterator *iter)
3053 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3057 ring_buffer_read(buf_iter, NULL);
3060 static struct trace_entry *
3061 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3062 unsigned long *lost_events)
3064 struct ring_buffer_event *event;
3065 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3068 event = ring_buffer_iter_peek(buf_iter, ts);
3070 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
3074 iter->ent_size = ring_buffer_event_length(event);
3075 return ring_buffer_event_data(event);
3081 static struct trace_entry *
3082 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3083 unsigned long *missing_events, u64 *ent_ts)
3085 struct ring_buffer *buffer = iter->trace_buffer->buffer;
3086 struct trace_entry *ent, *next = NULL;
3087 unsigned long lost_events = 0, next_lost = 0;
3088 int cpu_file = iter->cpu_file;
3089 u64 next_ts = 0, ts;
3095 * If we are in a per_cpu trace file, don't bother by iterating over
3096 * all cpu and peek directly.
3098 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3099 if (ring_buffer_empty_cpu(buffer, cpu_file))
3101 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3103 *ent_cpu = cpu_file;
3108 for_each_tracing_cpu(cpu) {
3110 if (ring_buffer_empty_cpu(buffer, cpu))
3113 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3116 * Pick the entry with the smallest timestamp:
3118 if (ent && (!next || ts < next_ts)) {
3122 next_lost = lost_events;
3123 next_size = iter->ent_size;
3127 iter->ent_size = next_size;
3130 *ent_cpu = next_cpu;
3136 *missing_events = next_lost;
3141 /* Find the next real entry, without updating the iterator itself */
3142 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3143 int *ent_cpu, u64 *ent_ts)
3145 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3148 /* Find the next real entry, and increment the iterator to the next entry */
3149 void *trace_find_next_entry_inc(struct trace_iterator *iter)
3151 iter->ent = __find_next_entry(iter, &iter->cpu,
3152 &iter->lost_events, &iter->ts);
3155 trace_iterator_increment(iter);
3157 return iter->ent ? iter : NULL;
3160 static void trace_consume(struct trace_iterator *iter)
3162 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
3163 &iter->lost_events);
3166 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3168 struct trace_iterator *iter = m->private;
3172 WARN_ON_ONCE(iter->leftover);
3176 /* can't go backwards */
3181 ent = trace_find_next_entry_inc(iter);
3185 while (ent && iter->idx < i)
3186 ent = trace_find_next_entry_inc(iter);
3193 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3195 struct ring_buffer_event *event;
3196 struct ring_buffer_iter *buf_iter;
3197 unsigned long entries = 0;
3200 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
3202 buf_iter = trace_buffer_iter(iter, cpu);
3206 ring_buffer_iter_reset(buf_iter);
3209 * We could have the case with the max latency tracers
3210 * that a reset never took place on a cpu. This is evident
3211 * by the timestamp being before the start of the buffer.
3213 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
3214 if (ts >= iter->trace_buffer->time_start)
3217 ring_buffer_read(buf_iter, NULL);
3220 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
3224 * The current tracer is copied to avoid a global locking
3227 static void *s_start(struct seq_file *m, loff_t *pos)
3229 struct trace_iterator *iter = m->private;
3230 struct trace_array *tr = iter->tr;
3231 int cpu_file = iter->cpu_file;
3237 * copy the tracer to avoid using a global lock all around.
3238 * iter->trace is a copy of current_trace, the pointer to the
3239 * name may be used instead of a strcmp(), as iter->trace->name
3240 * will point to the same string as current_trace->name.
3242 mutex_lock(&trace_types_lock);
3243 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
3244 *iter->trace = *tr->current_trace;
3245 mutex_unlock(&trace_types_lock);
3247 #ifdef CONFIG_TRACER_MAX_TRACE
3248 if (iter->snapshot && iter->trace->use_max_tr)
3249 return ERR_PTR(-EBUSY);
3252 if (!iter->snapshot)
3253 atomic_inc(&trace_record_taskinfo_disabled);
3255 if (*pos != iter->pos) {
3260 if (cpu_file == RING_BUFFER_ALL_CPUS) {
3261 for_each_tracing_cpu(cpu)
3262 tracing_iter_reset(iter, cpu);
3264 tracing_iter_reset(iter, cpu_file);
3267 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
3272 * If we overflowed the seq_file before, then we want
3273 * to just reuse the trace_seq buffer again.
3279 p = s_next(m, p, &l);
3283 trace_event_read_lock();
3284 trace_access_lock(cpu_file);
3288 static void s_stop(struct seq_file *m, void *p)
3290 struct trace_iterator *iter = m->private;
3292 #ifdef CONFIG_TRACER_MAX_TRACE
3293 if (iter->snapshot && iter->trace->use_max_tr)
3297 if (!iter->snapshot)
3298 atomic_dec(&trace_record_taskinfo_disabled);
3300 trace_access_unlock(iter->cpu_file);
3301 trace_event_read_unlock();
3305 get_total_entries(struct trace_buffer *buf,
3306 unsigned long *total, unsigned long *entries)
3308 unsigned long count;
3314 for_each_tracing_cpu(cpu) {
3315 count = ring_buffer_entries_cpu(buf->buffer, cpu);
3317 * If this buffer has skipped entries, then we hold all
3318 * entries for the trace and we need to ignore the
3319 * ones before the time stamp.
3321 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
3322 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
3323 /* total is the same as the entries */
3327 ring_buffer_overrun_cpu(buf->buffer, cpu);
3332 static void print_lat_help_header(struct seq_file *m)
3334 seq_puts(m, "# _------=> CPU# \n"
3335 "# / _-----=> irqs-off \n"
3336 "# | / _----=> need-resched \n"
3337 "# || / _---=> hardirq/softirq \n"
3338 "# ||| / _--=> preempt-depth \n"
3340 "# cmd pid ||||| time | caller \n"
3341 "# \\ / ||||| \\ | / \n");
3344 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
3346 unsigned long total;
3347 unsigned long entries;
3349 get_total_entries(buf, &total, &entries);
3350 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
3351 entries, total, num_online_cpus());
3355 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m,
3358 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3360 print_event_info(buf, m);
3362 seq_printf(m, "# TASK-PID CPU# %s TIMESTAMP FUNCTION\n", tgid ? "TGID " : "");
3363 seq_printf(m, "# | | | %s | |\n", tgid ? " | " : "");
3366 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m,
3369 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3370 const char tgid_space[] = " ";
3371 const char space[] = " ";
3373 seq_printf(m, "# %s _-----=> irqs-off\n",
3374 tgid ? tgid_space : space);
3375 seq_printf(m, "# %s / _----=> need-resched\n",
3376 tgid ? tgid_space : space);
3377 seq_printf(m, "# %s| / _---=> hardirq/softirq\n",
3378 tgid ? tgid_space : space);
3379 seq_printf(m, "# %s|| / _--=> preempt-depth\n",
3380 tgid ? tgid_space : space);
3381 seq_printf(m, "# %s||| / delay\n",
3382 tgid ? tgid_space : space);
3383 seq_printf(m, "# TASK-PID CPU#%s|||| TIMESTAMP FUNCTION\n",
3384 tgid ? " TGID " : space);
3385 seq_printf(m, "# | | | %s|||| | |\n",
3386 tgid ? " | " : space);
3390 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
3392 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
3393 struct trace_buffer *buf = iter->trace_buffer;
3394 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
3395 struct tracer *type = iter->trace;
3396 unsigned long entries;
3397 unsigned long total;
3398 const char *name = "preemption";
3402 get_total_entries(buf, &total, &entries);
3404 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
3406 seq_puts(m, "# -----------------------------------"
3407 "---------------------------------\n");
3408 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
3409 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
3410 nsecs_to_usecs(data->saved_latency),
3414 #if defined(CONFIG_PREEMPT_NONE)
3416 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
3418 #elif defined(CONFIG_PREEMPT)
3423 /* These are reserved for later use */
3426 seq_printf(m, " #P:%d)\n", num_online_cpus());
3430 seq_puts(m, "# -----------------\n");
3431 seq_printf(m, "# | task: %.16s-%d "
3432 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
3433 data->comm, data->pid,
3434 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
3435 data->policy, data->rt_priority);
3436 seq_puts(m, "# -----------------\n");
3438 if (data->critical_start) {
3439 seq_puts(m, "# => started at: ");
3440 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
3441 trace_print_seq(m, &iter->seq);
3442 seq_puts(m, "\n# => ended at: ");
3443 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
3444 trace_print_seq(m, &iter->seq);
3445 seq_puts(m, "\n#\n");
3451 static void test_cpu_buff_start(struct trace_iterator *iter)
3453 struct trace_seq *s = &iter->seq;
3454 struct trace_array *tr = iter->tr;
3456 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
3459 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
3462 if (cpumask_available(iter->started) &&
3463 cpumask_test_cpu(iter->cpu, iter->started))
3466 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
3469 if (cpumask_available(iter->started))
3470 cpumask_set_cpu(iter->cpu, iter->started);
3472 /* Don't print started cpu buffer for the first entry of the trace */
3474 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
3478 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
3480 struct trace_array *tr = iter->tr;
3481 struct trace_seq *s = &iter->seq;
3482 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
3483 struct trace_entry *entry;
3484 struct trace_event *event;
3488 test_cpu_buff_start(iter);
3490 event = ftrace_find_event(entry->type);
3492 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3493 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
3494 trace_print_lat_context(iter);
3496 trace_print_context(iter);
3499 if (trace_seq_has_overflowed(s))
3500 return TRACE_TYPE_PARTIAL_LINE;
3503 return event->funcs->trace(iter, sym_flags, event);
3505 trace_seq_printf(s, "Unknown type %d\n", entry->type);
3507 return trace_handle_return(s);
3510 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
3512 struct trace_array *tr = iter->tr;
3513 struct trace_seq *s = &iter->seq;
3514 struct trace_entry *entry;
3515 struct trace_event *event;
3519 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
3520 trace_seq_printf(s, "%d %d %llu ",
3521 entry->pid, iter->cpu, iter->ts);
3523 if (trace_seq_has_overflowed(s))
3524 return TRACE_TYPE_PARTIAL_LINE;
3526 event = ftrace_find_event(entry->type);
3528 return event->funcs->raw(iter, 0, event);
3530 trace_seq_printf(s, "%d ?\n", entry->type);
3532 return trace_handle_return(s);
3535 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
3537 struct trace_array *tr = iter->tr;
3538 struct trace_seq *s = &iter->seq;
3539 unsigned char newline = '\n';
3540 struct trace_entry *entry;
3541 struct trace_event *event;
3545 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3546 SEQ_PUT_HEX_FIELD(s, entry->pid);
3547 SEQ_PUT_HEX_FIELD(s, iter->cpu);
3548 SEQ_PUT_HEX_FIELD(s, iter->ts);
3549 if (trace_seq_has_overflowed(s))
3550 return TRACE_TYPE_PARTIAL_LINE;
3553 event = ftrace_find_event(entry->type);
3555 enum print_line_t ret = event->funcs->hex(iter, 0, event);
3556 if (ret != TRACE_TYPE_HANDLED)
3560 SEQ_PUT_FIELD(s, newline);
3562 return trace_handle_return(s);
3565 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
3567 struct trace_array *tr = iter->tr;
3568 struct trace_seq *s = &iter->seq;
3569 struct trace_entry *entry;
3570 struct trace_event *event;
3574 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3575 SEQ_PUT_FIELD(s, entry->pid);
3576 SEQ_PUT_FIELD(s, iter->cpu);
3577 SEQ_PUT_FIELD(s, iter->ts);
3578 if (trace_seq_has_overflowed(s))
3579 return TRACE_TYPE_PARTIAL_LINE;
3582 event = ftrace_find_event(entry->type);
3583 return event ? event->funcs->binary(iter, 0, event) :
3587 int trace_empty(struct trace_iterator *iter)
3589 struct ring_buffer_iter *buf_iter;
3592 /* If we are looking at one CPU buffer, only check that one */
3593 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
3594 cpu = iter->cpu_file;
3595 buf_iter = trace_buffer_iter(iter, cpu);
3597 if (!ring_buffer_iter_empty(buf_iter))
3600 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
3606 for_each_tracing_cpu(cpu) {
3607 buf_iter = trace_buffer_iter(iter, cpu);
3609 if (!ring_buffer_iter_empty(buf_iter))
3612 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
3620 /* Called with trace_event_read_lock() held. */
3621 enum print_line_t print_trace_line(struct trace_iterator *iter)
3623 struct trace_array *tr = iter->tr;
3624 unsigned long trace_flags = tr->trace_flags;
3625 enum print_line_t ret;
3627 if (iter->lost_events) {
3628 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
3629 iter->cpu, iter->lost_events);
3630 if (trace_seq_has_overflowed(&iter->seq))
3631 return TRACE_TYPE_PARTIAL_LINE;
3634 if (iter->trace && iter->trace->print_line) {
3635 ret = iter->trace->print_line(iter);
3636 if (ret != TRACE_TYPE_UNHANDLED)
3640 if (iter->ent->type == TRACE_BPUTS &&
3641 trace_flags & TRACE_ITER_PRINTK &&
3642 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
3643 return trace_print_bputs_msg_only(iter);
3645 if (iter->ent->type == TRACE_BPRINT &&
3646 trace_flags & TRACE_ITER_PRINTK &&
3647 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
3648 return trace_print_bprintk_msg_only(iter);
3650 if (iter->ent->type == TRACE_PRINT &&
3651 trace_flags & TRACE_ITER_PRINTK &&
3652 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
3653 return trace_print_printk_msg_only(iter);
3655 if (trace_flags & TRACE_ITER_BIN)
3656 return print_bin_fmt(iter);
3658 if (trace_flags & TRACE_ITER_HEX)
3659 return print_hex_fmt(iter);
3661 if (trace_flags & TRACE_ITER_RAW)
3662 return print_raw_fmt(iter);
3664 return print_trace_fmt(iter);
3667 void trace_latency_header(struct seq_file *m)
3669 struct trace_iterator *iter = m->private;
3670 struct trace_array *tr = iter->tr;
3672 /* print nothing if the buffers are empty */
3673 if (trace_empty(iter))
3676 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
3677 print_trace_header(m, iter);
3679 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
3680 print_lat_help_header(m);
3683 void trace_default_header(struct seq_file *m)
3685 struct trace_iterator *iter = m->private;
3686 struct trace_array *tr = iter->tr;
3687 unsigned long trace_flags = tr->trace_flags;
3689 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
3692 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
3693 /* print nothing if the buffers are empty */
3694 if (trace_empty(iter))
3696 print_trace_header(m, iter);
3697 if (!(trace_flags & TRACE_ITER_VERBOSE))
3698 print_lat_help_header(m);
3700 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
3701 if (trace_flags & TRACE_ITER_IRQ_INFO)
3702 print_func_help_header_irq(iter->trace_buffer,
3705 print_func_help_header(iter->trace_buffer, m,
3711 static void test_ftrace_alive(struct seq_file *m)
3713 if (!ftrace_is_dead())
3715 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
3716 "# MAY BE MISSING FUNCTION EVENTS\n");
3719 #ifdef CONFIG_TRACER_MAX_TRACE
3720 static void show_snapshot_main_help(struct seq_file *m)
3722 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
3723 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
3724 "# Takes a snapshot of the main buffer.\n"
3725 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
3726 "# (Doesn't have to be '2' works with any number that\n"
3727 "# is not a '0' or '1')\n");
3730 static void show_snapshot_percpu_help(struct seq_file *m)
3732 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
3733 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
3734 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
3735 "# Takes a snapshot of the main buffer for this cpu.\n");
3737 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
3738 "# Must use main snapshot file to allocate.\n");
3740 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
3741 "# (Doesn't have to be '2' works with any number that\n"
3742 "# is not a '0' or '1')\n");
3745 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
3747 if (iter->tr->allocated_snapshot)
3748 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
3750 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
3752 seq_puts(m, "# Snapshot commands:\n");
3753 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
3754 show_snapshot_main_help(m);
3756 show_snapshot_percpu_help(m);
3759 /* Should never be called */
3760 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
3763 static int s_show(struct seq_file *m, void *v)
3765 struct trace_iterator *iter = v;
3768 if (iter->ent == NULL) {
3770 seq_printf(m, "# tracer: %s\n", iter->trace->name);
3772 test_ftrace_alive(m);
3774 if (iter->snapshot && trace_empty(iter))
3775 print_snapshot_help(m, iter);
3776 else if (iter->trace && iter->trace->print_header)
3777 iter->trace->print_header(m);
3779 trace_default_header(m);
3781 } else if (iter->leftover) {
3783 * If we filled the seq_file buffer earlier, we
3784 * want to just show it now.
3786 ret = trace_print_seq(m, &iter->seq);
3788 /* ret should this time be zero, but you never know */
3789 iter->leftover = ret;
3792 print_trace_line(iter);
3793 ret = trace_print_seq(m, &iter->seq);
3795 * If we overflow the seq_file buffer, then it will
3796 * ask us for this data again at start up.
3798 * ret is 0 if seq_file write succeeded.
3801 iter->leftover = ret;
3808 * Should be used after trace_array_get(), trace_types_lock
3809 * ensures that i_cdev was already initialized.
3811 static inline int tracing_get_cpu(struct inode *inode)
3813 if (inode->i_cdev) /* See trace_create_cpu_file() */
3814 return (long)inode->i_cdev - 1;
3815 return RING_BUFFER_ALL_CPUS;
3818 static const struct seq_operations tracer_seq_ops = {
3825 static struct trace_iterator *
3826 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
3828 struct trace_array *tr = inode->i_private;
3829 struct trace_iterator *iter;
3832 if (tracing_disabled)
3833 return ERR_PTR(-ENODEV);
3835 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
3837 return ERR_PTR(-ENOMEM);
3839 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
3841 if (!iter->buffer_iter)
3845 * We make a copy of the current tracer to avoid concurrent
3846 * changes on it while we are reading.
3848 mutex_lock(&trace_types_lock);
3849 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
3853 *iter->trace = *tr->current_trace;
3855 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
3860 #ifdef CONFIG_TRACER_MAX_TRACE
3861 /* Currently only the top directory has a snapshot */
3862 if (tr->current_trace->print_max || snapshot)
3863 iter->trace_buffer = &tr->max_buffer;
3866 iter->trace_buffer = &tr->trace_buffer;
3867 iter->snapshot = snapshot;
3869 iter->cpu_file = tracing_get_cpu(inode);
3870 mutex_init(&iter->mutex);
3872 /* Notify the tracer early; before we stop tracing. */
3873 if (iter->trace && iter->trace->open)
3874 iter->trace->open(iter);
3876 /* Annotate start of buffers if we had overruns */
3877 if (ring_buffer_overruns(iter->trace_buffer->buffer))
3878 iter->iter_flags |= TRACE_FILE_ANNOTATE;
3880 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3881 if (trace_clocks[tr->clock_id].in_ns)
3882 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3884 /* stop the trace while dumping if we are not opening "snapshot" */
3885 if (!iter->snapshot)
3886 tracing_stop_tr(tr);
3888 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
3889 for_each_tracing_cpu(cpu) {
3890 iter->buffer_iter[cpu] =
3891 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
3893 ring_buffer_read_prepare_sync();
3894 for_each_tracing_cpu(cpu) {
3895 ring_buffer_read_start(iter->buffer_iter[cpu]);
3896 tracing_iter_reset(iter, cpu);
3899 cpu = iter->cpu_file;
3900 iter->buffer_iter[cpu] =
3901 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
3902 ring_buffer_read_prepare_sync();
3903 ring_buffer_read_start(iter->buffer_iter[cpu]);
3904 tracing_iter_reset(iter, cpu);
3907 mutex_unlock(&trace_types_lock);
3912 mutex_unlock(&trace_types_lock);
3914 kfree(iter->buffer_iter);
3916 seq_release_private(inode, file);
3917 return ERR_PTR(-ENOMEM);
3920 int tracing_open_generic(struct inode *inode, struct file *filp)
3922 if (tracing_disabled)
3925 filp->private_data = inode->i_private;
3929 bool tracing_is_disabled(void)
3931 return (tracing_disabled) ? true: false;
3935 * Open and update trace_array ref count.
3936 * Must have the current trace_array passed to it.
3938 static int tracing_open_generic_tr(struct inode *inode, struct file *filp)
3940 struct trace_array *tr = inode->i_private;
3942 if (tracing_disabled)
3945 if (trace_array_get(tr) < 0)
3948 filp->private_data = inode->i_private;
3953 static int tracing_release(struct inode *inode, struct file *file)
3955 struct trace_array *tr = inode->i_private;
3956 struct seq_file *m = file->private_data;
3957 struct trace_iterator *iter;
3960 if (!(file->f_mode & FMODE_READ)) {
3961 trace_array_put(tr);
3965 /* Writes do not use seq_file */
3967 mutex_lock(&trace_types_lock);
3969 for_each_tracing_cpu(cpu) {
3970 if (iter->buffer_iter[cpu])
3971 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3974 if (iter->trace && iter->trace->close)
3975 iter->trace->close(iter);
3977 if (!iter->snapshot)
3978 /* reenable tracing if it was previously enabled */
3979 tracing_start_tr(tr);
3981 __trace_array_put(tr);
3983 mutex_unlock(&trace_types_lock);
3985 mutex_destroy(&iter->mutex);
3986 free_cpumask_var(iter->started);
3988 kfree(iter->buffer_iter);
3989 seq_release_private(inode, file);
3994 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3996 struct trace_array *tr = inode->i_private;
3998 trace_array_put(tr);
4002 static int tracing_single_release_tr(struct inode *inode, struct file *file)
4004 struct trace_array *tr = inode->i_private;
4006 trace_array_put(tr);
4008 return single_release(inode, file);
4011 static int tracing_open(struct inode *inode, struct file *file)
4013 struct trace_array *tr = inode->i_private;
4014 struct trace_iterator *iter;
4017 if (trace_array_get(tr) < 0)
4020 /* If this file was open for write, then erase contents */
4021 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4022 int cpu = tracing_get_cpu(inode);
4023 struct trace_buffer *trace_buf = &tr->trace_buffer;
4025 #ifdef CONFIG_TRACER_MAX_TRACE
4026 if (tr->current_trace->print_max)
4027 trace_buf = &tr->max_buffer;
4030 if (cpu == RING_BUFFER_ALL_CPUS)
4031 tracing_reset_online_cpus(trace_buf);
4033 tracing_reset(trace_buf, cpu);
4036 if (file->f_mode & FMODE_READ) {
4037 iter = __tracing_open(inode, file, false);
4039 ret = PTR_ERR(iter);
4040 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4041 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4045 trace_array_put(tr);
4051 * Some tracers are not suitable for instance buffers.
4052 * A tracer is always available for the global array (toplevel)
4053 * or if it explicitly states that it is.
4056 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4058 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4061 /* Find the next tracer that this trace array may use */
4062 static struct tracer *
4063 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4065 while (t && !trace_ok_for_array(t, tr))
4072 t_next(struct seq_file *m, void *v, loff_t *pos)
4074 struct trace_array *tr = m->private;
4075 struct tracer *t = v;
4080 t = get_tracer_for_array(tr, t->next);
4085 static void *t_start(struct seq_file *m, loff_t *pos)
4087 struct trace_array *tr = m->private;
4091 mutex_lock(&trace_types_lock);
4093 t = get_tracer_for_array(tr, trace_types);
4094 for (; t && l < *pos; t = t_next(m, t, &l))
4100 static void t_stop(struct seq_file *m, void *p)
4102 mutex_unlock(&trace_types_lock);
4105 static int t_show(struct seq_file *m, void *v)
4107 struct tracer *t = v;
4112 seq_puts(m, t->name);
4121 static const struct seq_operations show_traces_seq_ops = {
4128 static int show_traces_open(struct inode *inode, struct file *file)
4130 struct trace_array *tr = inode->i_private;
4134 if (tracing_disabled)
4137 ret = seq_open(file, &show_traces_seq_ops);
4141 m = file->private_data;
4148 tracing_write_stub(struct file *filp, const char __user *ubuf,
4149 size_t count, loff_t *ppos)
4154 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
4158 if (file->f_mode & FMODE_READ)
4159 ret = seq_lseek(file, offset, whence);
4161 file->f_pos = ret = 0;
4166 static const struct file_operations tracing_fops = {
4167 .open = tracing_open,
4169 .write = tracing_write_stub,
4170 .llseek = tracing_lseek,
4171 .release = tracing_release,
4174 static const struct file_operations show_traces_fops = {
4175 .open = show_traces_open,
4177 .release = seq_release,
4178 .llseek = seq_lseek,
4182 * The tracer itself will not take this lock, but still we want
4183 * to provide a consistent cpumask to user-space:
4185 static DEFINE_MUTEX(tracing_cpumask_update_lock);
4188 * Temporary storage for the character representation of the
4189 * CPU bitmask (and one more byte for the newline):
4191 static char mask_str[NR_CPUS + 1];
4194 tracing_cpumask_read(struct file *filp, char __user *ubuf,
4195 size_t count, loff_t *ppos)
4197 struct trace_array *tr = file_inode(filp)->i_private;
4200 mutex_lock(&tracing_cpumask_update_lock);
4202 len = snprintf(mask_str, count, "%*pb\n",
4203 cpumask_pr_args(tr->tracing_cpumask));
4208 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
4211 mutex_unlock(&tracing_cpumask_update_lock);
4217 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
4218 size_t count, loff_t *ppos)
4220 struct trace_array *tr = file_inode(filp)->i_private;
4221 cpumask_var_t tracing_cpumask_new;
4224 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
4227 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
4231 mutex_lock(&tracing_cpumask_update_lock);
4233 local_irq_disable();
4234 arch_spin_lock(&tr->max_lock);
4235 for_each_tracing_cpu(cpu) {
4237 * Increase/decrease the disabled counter if we are
4238 * about to flip a bit in the cpumask:
4240 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4241 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4242 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
4243 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
4245 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4246 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4247 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
4248 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
4251 arch_spin_unlock(&tr->max_lock);
4254 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
4256 mutex_unlock(&tracing_cpumask_update_lock);
4257 free_cpumask_var(tracing_cpumask_new);
4262 free_cpumask_var(tracing_cpumask_new);
4267 static const struct file_operations tracing_cpumask_fops = {
4268 .open = tracing_open_generic_tr,
4269 .read = tracing_cpumask_read,
4270 .write = tracing_cpumask_write,
4271 .release = tracing_release_generic_tr,
4272 .llseek = generic_file_llseek,
4275 static int tracing_trace_options_show(struct seq_file *m, void *v)
4277 struct tracer_opt *trace_opts;
4278 struct trace_array *tr = m->private;
4282 mutex_lock(&trace_types_lock);
4283 tracer_flags = tr->current_trace->flags->val;
4284 trace_opts = tr->current_trace->flags->opts;
4286 for (i = 0; trace_options[i]; i++) {
4287 if (tr->trace_flags & (1 << i))
4288 seq_printf(m, "%s\n", trace_options[i]);
4290 seq_printf(m, "no%s\n", trace_options[i]);
4293 for (i = 0; trace_opts[i].name; i++) {
4294 if (tracer_flags & trace_opts[i].bit)
4295 seq_printf(m, "%s\n", trace_opts[i].name);
4297 seq_printf(m, "no%s\n", trace_opts[i].name);
4299 mutex_unlock(&trace_types_lock);
4304 static int __set_tracer_option(struct trace_array *tr,
4305 struct tracer_flags *tracer_flags,
4306 struct tracer_opt *opts, int neg)
4308 struct tracer *trace = tracer_flags->trace;
4311 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
4316 tracer_flags->val &= ~opts->bit;
4318 tracer_flags->val |= opts->bit;
4322 /* Try to assign a tracer specific option */
4323 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
4325 struct tracer *trace = tr->current_trace;
4326 struct tracer_flags *tracer_flags = trace->flags;
4327 struct tracer_opt *opts = NULL;
4330 for (i = 0; tracer_flags->opts[i].name; i++) {
4331 opts = &tracer_flags->opts[i];
4333 if (strcmp(cmp, opts->name) == 0)
4334 return __set_tracer_option(tr, trace->flags, opts, neg);
4340 /* Some tracers require overwrite to stay enabled */
4341 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
4343 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
4349 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
4351 /* do nothing if flag is already set */
4352 if (!!(tr->trace_flags & mask) == !!enabled)
4355 /* Give the tracer a chance to approve the change */
4356 if (tr->current_trace->flag_changed)
4357 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
4361 tr->trace_flags |= mask;
4363 tr->trace_flags &= ~mask;
4365 if (mask == TRACE_ITER_RECORD_CMD)
4366 trace_event_enable_cmd_record(enabled);
4368 if (mask == TRACE_ITER_RECORD_TGID) {
4370 tgid_map = kzalloc((PID_MAX_DEFAULT + 1) * sizeof(*tgid_map),
4373 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
4377 trace_event_enable_tgid_record(enabled);
4380 if (mask == TRACE_ITER_EVENT_FORK)
4381 trace_event_follow_fork(tr, enabled);
4383 if (mask == TRACE_ITER_FUNC_FORK)
4384 ftrace_pid_follow_fork(tr, enabled);
4386 if (mask == TRACE_ITER_OVERWRITE) {
4387 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
4388 #ifdef CONFIG_TRACER_MAX_TRACE
4389 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
4393 if (mask == TRACE_ITER_PRINTK) {
4394 trace_printk_start_stop_comm(enabled);
4395 trace_printk_control(enabled);
4401 static int trace_set_options(struct trace_array *tr, char *option)
4407 size_t orig_len = strlen(option);
4409 cmp = strstrip(option);
4411 if (strncmp(cmp, "no", 2) == 0) {
4416 mutex_lock(&trace_types_lock);
4418 for (i = 0; trace_options[i]; i++) {
4419 if (strcmp(cmp, trace_options[i]) == 0) {
4420 ret = set_tracer_flag(tr, 1 << i, !neg);
4425 /* If no option could be set, test the specific tracer options */
4426 if (!trace_options[i])
4427 ret = set_tracer_option(tr, cmp, neg);
4429 mutex_unlock(&trace_types_lock);
4432 * If the first trailing whitespace is replaced with '\0' by strstrip,
4433 * turn it back into a space.
4435 if (orig_len > strlen(option))
4436 option[strlen(option)] = ' ';
4441 static void __init apply_trace_boot_options(void)
4443 char *buf = trace_boot_options_buf;
4447 option = strsep(&buf, ",");
4453 trace_set_options(&global_trace, option);
4455 /* Put back the comma to allow this to be called again */
4462 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
4463 size_t cnt, loff_t *ppos)
4465 struct seq_file *m = filp->private_data;
4466 struct trace_array *tr = m->private;
4470 if (cnt >= sizeof(buf))
4473 if (copy_from_user(buf, ubuf, cnt))
4478 ret = trace_set_options(tr, buf);
4487 static int tracing_trace_options_open(struct inode *inode, struct file *file)
4489 struct trace_array *tr = inode->i_private;
4492 if (tracing_disabled)
4495 if (trace_array_get(tr) < 0)
4498 ret = single_open(file, tracing_trace_options_show, inode->i_private);
4500 trace_array_put(tr);
4505 static const struct file_operations tracing_iter_fops = {
4506 .open = tracing_trace_options_open,
4508 .llseek = seq_lseek,
4509 .release = tracing_single_release_tr,
4510 .write = tracing_trace_options_write,
4513 static const char readme_msg[] =
4514 "tracing mini-HOWTO:\n\n"
4515 "# echo 0 > tracing_on : quick way to disable tracing\n"
4516 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
4517 " Important files:\n"
4518 " trace\t\t\t- The static contents of the buffer\n"
4519 "\t\t\t To clear the buffer write into this file: echo > trace\n"
4520 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
4521 " current_tracer\t- function and latency tracers\n"
4522 " available_tracers\t- list of configured tracers for current_tracer\n"
4523 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
4524 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
4525 " trace_clock\t\t-change the clock used to order events\n"
4526 " local: Per cpu clock but may not be synced across CPUs\n"
4527 " global: Synced across CPUs but slows tracing down.\n"
4528 " counter: Not a clock, but just an increment\n"
4529 " uptime: Jiffy counter from time of boot\n"
4530 " perf: Same clock that perf events use\n"
4531 #ifdef CONFIG_X86_64
4532 " x86-tsc: TSC cycle counter\n"
4534 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
4535 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
4536 " tracing_cpumask\t- Limit which CPUs to trace\n"
4537 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
4538 "\t\t\t Remove sub-buffer with rmdir\n"
4539 " trace_options\t\t- Set format or modify how tracing happens\n"
4540 "\t\t\t Disable an option by adding a suffix 'no' to the\n"
4541 "\t\t\t option name\n"
4542 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
4543 #ifdef CONFIG_DYNAMIC_FTRACE
4544 "\n available_filter_functions - list of functions that can be filtered on\n"
4545 " set_ftrace_filter\t- echo function name in here to only trace these\n"
4546 "\t\t\t functions\n"
4547 "\t accepts: func_full_name or glob-matching-pattern\n"
4548 "\t modules: Can select a group via module\n"
4549 "\t Format: :mod:<module-name>\n"
4550 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
4551 "\t triggers: a command to perform when function is hit\n"
4552 "\t Format: <function>:<trigger>[:count]\n"
4553 "\t trigger: traceon, traceoff\n"
4554 "\t\t enable_event:<system>:<event>\n"
4555 "\t\t disable_event:<system>:<event>\n"
4556 #ifdef CONFIG_STACKTRACE
4559 #ifdef CONFIG_TRACER_SNAPSHOT
4564 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
4565 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
4566 "\t The first one will disable tracing every time do_fault is hit\n"
4567 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
4568 "\t The first time do trap is hit and it disables tracing, the\n"
4569 "\t counter will decrement to 2. If tracing is already disabled,\n"
4570 "\t the counter will not decrement. It only decrements when the\n"
4571 "\t trigger did work\n"
4572 "\t To remove trigger without count:\n"
4573 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
4574 "\t To remove trigger with a count:\n"
4575 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
4576 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
4577 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
4578 "\t modules: Can select a group via module command :mod:\n"
4579 "\t Does not accept triggers\n"
4580 #endif /* CONFIG_DYNAMIC_FTRACE */
4581 #ifdef CONFIG_FUNCTION_TRACER
4582 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
4585 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4586 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
4587 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
4588 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
4590 #ifdef CONFIG_TRACER_SNAPSHOT
4591 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
4592 "\t\t\t snapshot buffer. Read the contents for more\n"
4593 "\t\t\t information\n"
4595 #ifdef CONFIG_STACK_TRACER
4596 " stack_trace\t\t- Shows the max stack trace when active\n"
4597 " stack_max_size\t- Shows current max stack size that was traced\n"
4598 "\t\t\t Write into this file to reset the max size (trigger a\n"
4599 "\t\t\t new trace)\n"
4600 #ifdef CONFIG_DYNAMIC_FTRACE
4601 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
4604 #endif /* CONFIG_STACK_TRACER */
4605 #ifdef CONFIG_KPROBE_EVENTS
4606 " kprobe_events\t\t- Add/remove/show the kernel dynamic events\n"
4607 "\t\t\t Write into this file to define/undefine new trace events.\n"
4609 #ifdef CONFIG_UPROBE_EVENTS
4610 " uprobe_events\t\t- Add/remove/show the userspace dynamic events\n"
4611 "\t\t\t Write into this file to define/undefine new trace events.\n"
4613 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
4614 "\t accepts: event-definitions (one definition per line)\n"
4615 "\t Format: p[:[<group>/]<event>] <place> [<args>]\n"
4616 "\t r[maxactive][:[<group>/]<event>] <place> [<args>]\n"
4617 "\t -:[<group>/]<event>\n"
4618 #ifdef CONFIG_KPROBE_EVENTS
4619 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
4620 "place (kretprobe): [<module>:]<symbol>[+<offset>]|<memaddr>\n"
4622 #ifdef CONFIG_UPROBE_EVENTS
4623 "\t place: <path>:<offset>\n"
4625 "\t args: <name>=fetcharg[:type]\n"
4626 "\t fetcharg: %<register>, @<address>, @<symbol>[+|-<offset>],\n"
4627 "\t $stack<index>, $stack, $retval, $comm\n"
4628 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string,\n"
4629 "\t b<bit-width>@<bit-offset>/<container-size>\n"
4631 " events/\t\t- Directory containing all trace event subsystems:\n"
4632 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
4633 " events/<system>/\t- Directory containing all trace events for <system>:\n"
4634 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
4636 " filter\t\t- If set, only events passing filter are traced\n"
4637 " events/<system>/<event>/\t- Directory containing control files for\n"
4639 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
4640 " filter\t\t- If set, only events passing filter are traced\n"
4641 " trigger\t\t- If set, a command to perform when event is hit\n"
4642 "\t Format: <trigger>[:count][if <filter>]\n"
4643 "\t trigger: traceon, traceoff\n"
4644 "\t enable_event:<system>:<event>\n"
4645 "\t disable_event:<system>:<event>\n"
4646 #ifdef CONFIG_HIST_TRIGGERS
4647 "\t enable_hist:<system>:<event>\n"
4648 "\t disable_hist:<system>:<event>\n"
4650 #ifdef CONFIG_STACKTRACE
4653 #ifdef CONFIG_TRACER_SNAPSHOT
4656 #ifdef CONFIG_HIST_TRIGGERS
4657 "\t\t hist (see below)\n"
4659 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
4660 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
4661 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
4662 "\t events/block/block_unplug/trigger\n"
4663 "\t The first disables tracing every time block_unplug is hit.\n"
4664 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
4665 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
4666 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
4667 "\t Like function triggers, the counter is only decremented if it\n"
4668 "\t enabled or disabled tracing.\n"
4669 "\t To remove a trigger without a count:\n"
4670 "\t echo '!<trigger> > <system>/<event>/trigger\n"
4671 "\t To remove a trigger with a count:\n"
4672 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
4673 "\t Filters can be ignored when removing a trigger.\n"
4674 #ifdef CONFIG_HIST_TRIGGERS
4675 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
4676 "\t Format: hist:keys=<field1[,field2,...]>\n"
4677 "\t [:values=<field1[,field2,...]>]\n"
4678 "\t [:sort=<field1[,field2,...]>]\n"
4679 "\t [:size=#entries]\n"
4680 "\t [:pause][:continue][:clear]\n"
4681 "\t [:name=histname1]\n"
4682 "\t [if <filter>]\n\n"
4683 "\t When a matching event is hit, an entry is added to a hash\n"
4684 "\t table using the key(s) and value(s) named, and the value of a\n"
4685 "\t sum called 'hitcount' is incremented. Keys and values\n"
4686 "\t correspond to fields in the event's format description. Keys\n"
4687 "\t can be any field, or the special string 'stacktrace'.\n"
4688 "\t Compound keys consisting of up to two fields can be specified\n"
4689 "\t by the 'keys' keyword. Values must correspond to numeric\n"
4690 "\t fields. Sort keys consisting of up to two fields can be\n"
4691 "\t specified using the 'sort' keyword. The sort direction can\n"
4692 "\t be modified by appending '.descending' or '.ascending' to a\n"
4693 "\t sort field. The 'size' parameter can be used to specify more\n"
4694 "\t or fewer than the default 2048 entries for the hashtable size.\n"
4695 "\t If a hist trigger is given a name using the 'name' parameter,\n"
4696 "\t its histogram data will be shared with other triggers of the\n"
4697 "\t same name, and trigger hits will update this common data.\n\n"
4698 "\t Reading the 'hist' file for the event will dump the hash\n"
4699 "\t table in its entirety to stdout. If there are multiple hist\n"
4700 "\t triggers attached to an event, there will be a table for each\n"
4701 "\t trigger in the output. The table displayed for a named\n"
4702 "\t trigger will be the same as any other instance having the\n"
4703 "\t same name. The default format used to display a given field\n"
4704 "\t can be modified by appending any of the following modifiers\n"
4705 "\t to the field name, as applicable:\n\n"
4706 "\t .hex display a number as a hex value\n"
4707 "\t .sym display an address as a symbol\n"
4708 "\t .sym-offset display an address as a symbol and offset\n"
4709 "\t .execname display a common_pid as a program name\n"
4710 "\t .syscall display a syscall id as a syscall name\n\n"
4711 "\t .log2 display log2 value rather than raw number\n\n"
4712 "\t The 'pause' parameter can be used to pause an existing hist\n"
4713 "\t trigger or to start a hist trigger but not log any events\n"
4714 "\t until told to do so. 'continue' can be used to start or\n"
4715 "\t restart a paused hist trigger.\n\n"
4716 "\t The 'clear' parameter will clear the contents of a running\n"
4717 "\t hist trigger and leave its current paused/active state\n"
4719 "\t The enable_hist and disable_hist triggers can be used to\n"
4720 "\t have one event conditionally start and stop another event's\n"
4721 "\t already-attached hist trigger. The syntax is analagous to\n"
4722 "\t the enable_event and disable_event triggers.\n"
4727 tracing_readme_read(struct file *filp, char __user *ubuf,
4728 size_t cnt, loff_t *ppos)
4730 return simple_read_from_buffer(ubuf, cnt, ppos,
4731 readme_msg, strlen(readme_msg));
4734 static const struct file_operations tracing_readme_fops = {
4735 .open = tracing_open_generic,
4736 .read = tracing_readme_read,
4737 .llseek = generic_file_llseek,
4740 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
4744 if (*pos || m->count)
4749 for (; ptr <= &tgid_map[PID_MAX_DEFAULT]; ptr++) {
4750 if (trace_find_tgid(*ptr))
4757 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
4767 v = saved_tgids_next(m, v, &l);
4775 static void saved_tgids_stop(struct seq_file *m, void *v)
4779 static int saved_tgids_show(struct seq_file *m, void *v)
4781 int pid = (int *)v - tgid_map;
4783 seq_printf(m, "%d %d\n", pid, trace_find_tgid(pid));
4787 static const struct seq_operations tracing_saved_tgids_seq_ops = {
4788 .start = saved_tgids_start,
4789 .stop = saved_tgids_stop,
4790 .next = saved_tgids_next,
4791 .show = saved_tgids_show,
4794 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
4796 if (tracing_disabled)
4799 return seq_open(filp, &tracing_saved_tgids_seq_ops);
4803 static const struct file_operations tracing_saved_tgids_fops = {
4804 .open = tracing_saved_tgids_open,
4806 .llseek = seq_lseek,
4807 .release = seq_release,
4810 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
4812 unsigned int *ptr = v;
4814 if (*pos || m->count)
4819 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
4821 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
4830 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
4836 arch_spin_lock(&trace_cmdline_lock);
4838 v = &savedcmd->map_cmdline_to_pid[0];
4840 v = saved_cmdlines_next(m, v, &l);
4848 static void saved_cmdlines_stop(struct seq_file *m, void *v)
4850 arch_spin_unlock(&trace_cmdline_lock);
4854 static int saved_cmdlines_show(struct seq_file *m, void *v)
4856 char buf[TASK_COMM_LEN];
4857 unsigned int *pid = v;
4859 __trace_find_cmdline(*pid, buf);
4860 seq_printf(m, "%d %s\n", *pid, buf);
4864 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
4865 .start = saved_cmdlines_start,
4866 .next = saved_cmdlines_next,
4867 .stop = saved_cmdlines_stop,
4868 .show = saved_cmdlines_show,
4871 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
4873 if (tracing_disabled)
4876 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
4879 static const struct file_operations tracing_saved_cmdlines_fops = {
4880 .open = tracing_saved_cmdlines_open,
4882 .llseek = seq_lseek,
4883 .release = seq_release,
4887 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
4888 size_t cnt, loff_t *ppos)
4893 arch_spin_lock(&trace_cmdline_lock);
4894 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
4895 arch_spin_unlock(&trace_cmdline_lock);
4897 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4900 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
4902 kfree(s->saved_cmdlines);
4903 kfree(s->map_cmdline_to_pid);
4907 static int tracing_resize_saved_cmdlines(unsigned int val)
4909 struct saved_cmdlines_buffer *s, *savedcmd_temp;
4911 s = kmalloc(sizeof(*s), GFP_KERNEL);
4915 if (allocate_cmdlines_buffer(val, s) < 0) {
4920 arch_spin_lock(&trace_cmdline_lock);
4921 savedcmd_temp = savedcmd;
4923 arch_spin_unlock(&trace_cmdline_lock);
4924 free_saved_cmdlines_buffer(savedcmd_temp);
4930 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
4931 size_t cnt, loff_t *ppos)
4936 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4940 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
4941 if (!val || val > PID_MAX_DEFAULT)
4944 ret = tracing_resize_saved_cmdlines((unsigned int)val);
4953 static const struct file_operations tracing_saved_cmdlines_size_fops = {
4954 .open = tracing_open_generic,
4955 .read = tracing_saved_cmdlines_size_read,
4956 .write = tracing_saved_cmdlines_size_write,
4959 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
4960 static union trace_eval_map_item *
4961 update_eval_map(union trace_eval_map_item *ptr)
4963 if (!ptr->map.eval_string) {
4964 if (ptr->tail.next) {
4965 ptr = ptr->tail.next;
4966 /* Set ptr to the next real item (skip head) */
4974 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
4976 union trace_eval_map_item *ptr = v;
4979 * Paranoid! If ptr points to end, we don't want to increment past it.
4980 * This really should never happen.
4982 ptr = update_eval_map(ptr);
4983 if (WARN_ON_ONCE(!ptr))
4990 ptr = update_eval_map(ptr);
4995 static void *eval_map_start(struct seq_file *m, loff_t *pos)
4997 union trace_eval_map_item *v;
5000 mutex_lock(&trace_eval_mutex);
5002 v = trace_eval_maps;
5006 while (v && l < *pos) {
5007 v = eval_map_next(m, v, &l);
5013 static void eval_map_stop(struct seq_file *m, void *v)
5015 mutex_unlock(&trace_eval_mutex);
5018 static int eval_map_show(struct seq_file *m, void *v)
5020 union trace_eval_map_item *ptr = v;
5022 seq_printf(m, "%s %ld (%s)\n",
5023 ptr->map.eval_string, ptr->map.eval_value,
5029 static const struct seq_operations tracing_eval_map_seq_ops = {
5030 .start = eval_map_start,
5031 .next = eval_map_next,
5032 .stop = eval_map_stop,
5033 .show = eval_map_show,
5036 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5038 if (tracing_disabled)
5041 return seq_open(filp, &tracing_eval_map_seq_ops);
5044 static const struct file_operations tracing_eval_map_fops = {
5045 .open = tracing_eval_map_open,
5047 .llseek = seq_lseek,
5048 .release = seq_release,
5051 static inline union trace_eval_map_item *
5052 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
5054 /* Return tail of array given the head */
5055 return ptr + ptr->head.length + 1;
5059 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
5062 struct trace_eval_map **stop;
5063 struct trace_eval_map **map;
5064 union trace_eval_map_item *map_array;
5065 union trace_eval_map_item *ptr;
5070 * The trace_eval_maps contains the map plus a head and tail item,
5071 * where the head holds the module and length of array, and the
5072 * tail holds a pointer to the next list.
5074 map_array = kmalloc(sizeof(*map_array) * (len + 2), GFP_KERNEL);
5076 pr_warn("Unable to allocate trace eval mapping\n");
5080 mutex_lock(&trace_eval_mutex);
5082 if (!trace_eval_maps)
5083 trace_eval_maps = map_array;
5085 ptr = trace_eval_maps;
5087 ptr = trace_eval_jmp_to_tail(ptr);
5088 if (!ptr->tail.next)
5090 ptr = ptr->tail.next;
5093 ptr->tail.next = map_array;
5095 map_array->head.mod = mod;
5096 map_array->head.length = len;
5099 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
5100 map_array->map = **map;
5103 memset(map_array, 0, sizeof(*map_array));
5105 mutex_unlock(&trace_eval_mutex);
5108 static void trace_create_eval_file(struct dentry *d_tracer)
5110 trace_create_file("eval_map", 0444, d_tracer,
5111 NULL, &tracing_eval_map_fops);
5114 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
5115 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
5116 static inline void trace_insert_eval_map_file(struct module *mod,
5117 struct trace_eval_map **start, int len) { }
5118 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
5120 static void trace_insert_eval_map(struct module *mod,
5121 struct trace_eval_map **start, int len)
5123 struct trace_eval_map **map;
5130 trace_event_eval_update(map, len);
5132 trace_insert_eval_map_file(mod, start, len);
5136 tracing_set_trace_read(struct file *filp, char __user *ubuf,
5137 size_t cnt, loff_t *ppos)
5139 struct trace_array *tr = filp->private_data;
5140 char buf[MAX_TRACER_SIZE+2];
5143 mutex_lock(&trace_types_lock);
5144 r = sprintf(buf, "%s\n", tr->current_trace->name);
5145 mutex_unlock(&trace_types_lock);
5147 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5150 int tracer_init(struct tracer *t, struct trace_array *tr)
5152 tracing_reset_online_cpus(&tr->trace_buffer);
5156 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
5160 for_each_tracing_cpu(cpu)
5161 per_cpu_ptr(buf->data, cpu)->entries = val;
5164 #ifdef CONFIG_TRACER_MAX_TRACE
5165 /* resize @tr's buffer to the size of @size_tr's entries */
5166 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
5167 struct trace_buffer *size_buf, int cpu_id)
5171 if (cpu_id == RING_BUFFER_ALL_CPUS) {
5172 for_each_tracing_cpu(cpu) {
5173 ret = ring_buffer_resize(trace_buf->buffer,
5174 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
5177 per_cpu_ptr(trace_buf->data, cpu)->entries =
5178 per_cpu_ptr(size_buf->data, cpu)->entries;
5181 ret = ring_buffer_resize(trace_buf->buffer,
5182 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
5184 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
5185 per_cpu_ptr(size_buf->data, cpu_id)->entries;
5190 #endif /* CONFIG_TRACER_MAX_TRACE */
5192 static int __tracing_resize_ring_buffer(struct trace_array *tr,
5193 unsigned long size, int cpu)
5198 * If kernel or user changes the size of the ring buffer
5199 * we use the size that was given, and we can forget about
5200 * expanding it later.
5202 ring_buffer_expanded = true;
5204 /* May be called before buffers are initialized */
5205 if (!tr->trace_buffer.buffer)
5208 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
5212 #ifdef CONFIG_TRACER_MAX_TRACE
5213 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
5214 !tr->current_trace->use_max_tr)
5217 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
5219 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
5220 &tr->trace_buffer, cpu);
5223 * AARGH! We are left with different
5224 * size max buffer!!!!
5225 * The max buffer is our "snapshot" buffer.
5226 * When a tracer needs a snapshot (one of the
5227 * latency tracers), it swaps the max buffer
5228 * with the saved snap shot. We succeeded to
5229 * update the size of the main buffer, but failed to
5230 * update the size of the max buffer. But when we tried
5231 * to reset the main buffer to the original size, we
5232 * failed there too. This is very unlikely to
5233 * happen, but if it does, warn and kill all
5237 tracing_disabled = 1;
5242 if (cpu == RING_BUFFER_ALL_CPUS)
5243 set_buffer_entries(&tr->max_buffer, size);
5245 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
5248 #endif /* CONFIG_TRACER_MAX_TRACE */
5250 if (cpu == RING_BUFFER_ALL_CPUS)
5251 set_buffer_entries(&tr->trace_buffer, size);
5253 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
5258 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
5259 unsigned long size, int cpu_id)
5263 mutex_lock(&trace_types_lock);
5265 if (cpu_id != RING_BUFFER_ALL_CPUS) {
5266 /* make sure, this cpu is enabled in the mask */
5267 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
5273 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
5278 mutex_unlock(&trace_types_lock);
5285 * tracing_update_buffers - used by tracing facility to expand ring buffers
5287 * To save on memory when the tracing is never used on a system with it
5288 * configured in. The ring buffers are set to a minimum size. But once
5289 * a user starts to use the tracing facility, then they need to grow
5290 * to their default size.
5292 * This function is to be called when a tracer is about to be used.
5294 int tracing_update_buffers(void)
5298 mutex_lock(&trace_types_lock);
5299 if (!ring_buffer_expanded)
5300 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
5301 RING_BUFFER_ALL_CPUS);
5302 mutex_unlock(&trace_types_lock);
5307 struct trace_option_dentry;
5310 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
5313 * Used to clear out the tracer before deletion of an instance.
5314 * Must have trace_types_lock held.
5316 static void tracing_set_nop(struct trace_array *tr)
5318 if (tr->current_trace == &nop_trace)
5321 tr->current_trace->enabled--;
5323 if (tr->current_trace->reset)
5324 tr->current_trace->reset(tr);
5326 tr->current_trace = &nop_trace;
5329 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
5331 /* Only enable if the directory has been created already. */
5335 create_trace_option_files(tr, t);
5338 static int tracing_set_tracer(struct trace_array *tr, const char *buf)
5341 #ifdef CONFIG_TRACER_MAX_TRACE
5346 mutex_lock(&trace_types_lock);
5348 if (!ring_buffer_expanded) {
5349 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
5350 RING_BUFFER_ALL_CPUS);
5356 for (t = trace_types; t; t = t->next) {
5357 if (strcmp(t->name, buf) == 0)
5364 if (t == tr->current_trace)
5367 /* Some tracers won't work on kernel command line */
5368 if (system_state < SYSTEM_RUNNING && t->noboot) {
5369 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
5374 /* Some tracers are only allowed for the top level buffer */
5375 if (!trace_ok_for_array(t, tr)) {
5380 /* If trace pipe files are being read, we can't change the tracer */
5381 if (tr->current_trace->ref) {
5386 trace_branch_disable();
5388 tr->current_trace->enabled--;
5390 if (tr->current_trace->reset)
5391 tr->current_trace->reset(tr);
5393 /* Current trace needs to be nop_trace before synchronize_sched */
5394 tr->current_trace = &nop_trace;
5396 #ifdef CONFIG_TRACER_MAX_TRACE
5397 had_max_tr = tr->allocated_snapshot;
5399 if (had_max_tr && !t->use_max_tr) {
5401 * We need to make sure that the update_max_tr sees that
5402 * current_trace changed to nop_trace to keep it from
5403 * swapping the buffers after we resize it.
5404 * The update_max_tr is called from interrupts disabled
5405 * so a synchronized_sched() is sufficient.
5407 synchronize_sched();
5412 #ifdef CONFIG_TRACER_MAX_TRACE
5413 if (t->use_max_tr && !had_max_tr) {
5414 ret = alloc_snapshot(tr);
5421 ret = tracer_init(t, tr);
5426 tr->current_trace = t;
5427 tr->current_trace->enabled++;
5428 trace_branch_enable(tr);
5430 mutex_unlock(&trace_types_lock);
5436 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
5437 size_t cnt, loff_t *ppos)
5439 struct trace_array *tr = filp->private_data;
5440 char buf[MAX_TRACER_SIZE+1];
5447 if (cnt > MAX_TRACER_SIZE)
5448 cnt = MAX_TRACER_SIZE;
5450 if (copy_from_user(buf, ubuf, cnt))
5455 /* strip ending whitespace. */
5456 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
5459 err = tracing_set_tracer(tr, buf);
5469 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
5470 size_t cnt, loff_t *ppos)
5475 r = snprintf(buf, sizeof(buf), "%ld\n",
5476 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
5477 if (r > sizeof(buf))
5479 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5483 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
5484 size_t cnt, loff_t *ppos)
5489 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5499 tracing_thresh_read(struct file *filp, char __user *ubuf,
5500 size_t cnt, loff_t *ppos)
5502 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
5506 tracing_thresh_write(struct file *filp, const char __user *ubuf,
5507 size_t cnt, loff_t *ppos)
5509 struct trace_array *tr = filp->private_data;
5512 mutex_lock(&trace_types_lock);
5513 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
5517 if (tr->current_trace->update_thresh) {
5518 ret = tr->current_trace->update_thresh(tr);
5525 mutex_unlock(&trace_types_lock);
5530 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
5533 tracing_max_lat_read(struct file *filp, char __user *ubuf,
5534 size_t cnt, loff_t *ppos)
5536 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
5540 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
5541 size_t cnt, loff_t *ppos)
5543 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
5548 static int tracing_open_pipe(struct inode *inode, struct file *filp)
5550 struct trace_array *tr = inode->i_private;
5551 struct trace_iterator *iter;
5554 if (tracing_disabled)
5557 if (trace_array_get(tr) < 0)
5560 mutex_lock(&trace_types_lock);
5562 /* create a buffer to store the information to pass to userspace */
5563 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
5566 __trace_array_put(tr);
5570 trace_seq_init(&iter->seq);
5571 iter->trace = tr->current_trace;
5573 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
5578 /* trace pipe does not show start of buffer */
5579 cpumask_setall(iter->started);
5581 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
5582 iter->iter_flags |= TRACE_FILE_LAT_FMT;
5584 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
5585 if (trace_clocks[tr->clock_id].in_ns)
5586 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
5589 iter->trace_buffer = &tr->trace_buffer;
5590 iter->cpu_file = tracing_get_cpu(inode);
5591 mutex_init(&iter->mutex);
5592 filp->private_data = iter;
5594 if (iter->trace->pipe_open)
5595 iter->trace->pipe_open(iter);
5597 nonseekable_open(inode, filp);
5599 tr->current_trace->ref++;
5601 mutex_unlock(&trace_types_lock);
5607 __trace_array_put(tr);
5608 mutex_unlock(&trace_types_lock);
5612 static int tracing_release_pipe(struct inode *inode, struct file *file)
5614 struct trace_iterator *iter = file->private_data;
5615 struct trace_array *tr = inode->i_private;
5617 mutex_lock(&trace_types_lock);
5619 tr->current_trace->ref--;
5621 if (iter->trace->pipe_close)
5622 iter->trace->pipe_close(iter);
5624 mutex_unlock(&trace_types_lock);
5626 free_cpumask_var(iter->started);
5627 mutex_destroy(&iter->mutex);
5630 trace_array_put(tr);
5636 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
5638 struct trace_array *tr = iter->tr;
5640 /* Iterators are static, they should be filled or empty */
5641 if (trace_buffer_iter(iter, iter->cpu_file))
5642 return POLLIN | POLLRDNORM;
5644 if (tr->trace_flags & TRACE_ITER_BLOCK)
5646 * Always select as readable when in blocking mode
5648 return POLLIN | POLLRDNORM;
5650 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
5655 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
5657 struct trace_iterator *iter = filp->private_data;
5659 return trace_poll(iter, filp, poll_table);
5662 /* Must be called with iter->mutex held. */
5663 static int tracing_wait_pipe(struct file *filp)
5665 struct trace_iterator *iter = filp->private_data;
5668 while (trace_empty(iter)) {
5670 if ((filp->f_flags & O_NONBLOCK)) {
5675 * We block until we read something and tracing is disabled.
5676 * We still block if tracing is disabled, but we have never
5677 * read anything. This allows a user to cat this file, and
5678 * then enable tracing. But after we have read something,
5679 * we give an EOF when tracing is again disabled.
5681 * iter->pos will be 0 if we haven't read anything.
5683 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
5686 mutex_unlock(&iter->mutex);
5688 ret = wait_on_pipe(iter, false);
5690 mutex_lock(&iter->mutex);
5703 tracing_read_pipe(struct file *filp, char __user *ubuf,
5704 size_t cnt, loff_t *ppos)
5706 struct trace_iterator *iter = filp->private_data;
5710 * Avoid more than one consumer on a single file descriptor
5711 * This is just a matter of traces coherency, the ring buffer itself
5714 mutex_lock(&iter->mutex);
5716 /* return any leftover data */
5717 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
5721 trace_seq_init(&iter->seq);
5723 if (iter->trace->read) {
5724 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
5730 sret = tracing_wait_pipe(filp);
5734 /* stop when tracing is finished */
5735 if (trace_empty(iter)) {
5740 if (cnt >= PAGE_SIZE)
5741 cnt = PAGE_SIZE - 1;
5743 /* reset all but tr, trace, and overruns */
5744 memset(&iter->seq, 0,
5745 sizeof(struct trace_iterator) -
5746 offsetof(struct trace_iterator, seq));
5747 cpumask_clear(iter->started);
5750 trace_event_read_lock();
5751 trace_access_lock(iter->cpu_file);
5752 while (trace_find_next_entry_inc(iter) != NULL) {
5753 enum print_line_t ret;
5754 int save_len = iter->seq.seq.len;
5756 ret = print_trace_line(iter);
5757 if (ret == TRACE_TYPE_PARTIAL_LINE) {
5758 /* don't print partial lines */
5759 iter->seq.seq.len = save_len;
5762 if (ret != TRACE_TYPE_NO_CONSUME)
5763 trace_consume(iter);
5765 if (trace_seq_used(&iter->seq) >= cnt)
5769 * Setting the full flag means we reached the trace_seq buffer
5770 * size and we should leave by partial output condition above.
5771 * One of the trace_seq_* functions is not used properly.
5773 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
5776 trace_access_unlock(iter->cpu_file);
5777 trace_event_read_unlock();
5779 /* Now copy what we have to the user */
5780 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
5781 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
5782 trace_seq_init(&iter->seq);
5785 * If there was nothing to send to user, in spite of consuming trace
5786 * entries, go back to wait for more entries.
5792 mutex_unlock(&iter->mutex);
5797 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
5800 __free_page(spd->pages[idx]);
5803 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
5805 .confirm = generic_pipe_buf_confirm,
5806 .release = generic_pipe_buf_release,
5807 .steal = generic_pipe_buf_steal,
5808 .get = generic_pipe_buf_get,
5812 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
5818 /* Seq buffer is page-sized, exactly what we need. */
5820 save_len = iter->seq.seq.len;
5821 ret = print_trace_line(iter);
5823 if (trace_seq_has_overflowed(&iter->seq)) {
5824 iter->seq.seq.len = save_len;
5829 * This should not be hit, because it should only
5830 * be set if the iter->seq overflowed. But check it
5831 * anyway to be safe.
5833 if (ret == TRACE_TYPE_PARTIAL_LINE) {
5834 iter->seq.seq.len = save_len;
5838 count = trace_seq_used(&iter->seq) - save_len;
5841 iter->seq.seq.len = save_len;
5845 if (ret != TRACE_TYPE_NO_CONSUME)
5846 trace_consume(iter);
5848 if (!trace_find_next_entry_inc(iter)) {
5858 static ssize_t tracing_splice_read_pipe(struct file *filp,
5860 struct pipe_inode_info *pipe,
5864 struct page *pages_def[PIPE_DEF_BUFFERS];
5865 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5866 struct trace_iterator *iter = filp->private_data;
5867 struct splice_pipe_desc spd = {
5869 .partial = partial_def,
5870 .nr_pages = 0, /* This gets updated below. */
5871 .nr_pages_max = PIPE_DEF_BUFFERS,
5872 .ops = &tracing_pipe_buf_ops,
5873 .spd_release = tracing_spd_release_pipe,
5879 if (splice_grow_spd(pipe, &spd))
5882 mutex_lock(&iter->mutex);
5884 if (iter->trace->splice_read) {
5885 ret = iter->trace->splice_read(iter, filp,
5886 ppos, pipe, len, flags);
5891 ret = tracing_wait_pipe(filp);
5895 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
5900 trace_event_read_lock();
5901 trace_access_lock(iter->cpu_file);
5903 /* Fill as many pages as possible. */
5904 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
5905 spd.pages[i] = alloc_page(GFP_KERNEL);
5909 rem = tracing_fill_pipe_page(rem, iter);
5911 /* Copy the data into the page, so we can start over. */
5912 ret = trace_seq_to_buffer(&iter->seq,
5913 page_address(spd.pages[i]),
5914 trace_seq_used(&iter->seq));
5916 __free_page(spd.pages[i]);
5919 spd.partial[i].offset = 0;
5920 spd.partial[i].len = trace_seq_used(&iter->seq);
5922 trace_seq_init(&iter->seq);
5925 trace_access_unlock(iter->cpu_file);
5926 trace_event_read_unlock();
5927 mutex_unlock(&iter->mutex);
5932 ret = splice_to_pipe(pipe, &spd);
5936 splice_shrink_spd(&spd);
5940 mutex_unlock(&iter->mutex);
5945 tracing_entries_read(struct file *filp, char __user *ubuf,
5946 size_t cnt, loff_t *ppos)
5948 struct inode *inode = file_inode(filp);
5949 struct trace_array *tr = inode->i_private;
5950 int cpu = tracing_get_cpu(inode);
5955 mutex_lock(&trace_types_lock);
5957 if (cpu == RING_BUFFER_ALL_CPUS) {
5958 int cpu, buf_size_same;
5963 /* check if all cpu sizes are same */
5964 for_each_tracing_cpu(cpu) {
5965 /* fill in the size from first enabled cpu */
5967 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
5968 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
5974 if (buf_size_same) {
5975 if (!ring_buffer_expanded)
5976 r = sprintf(buf, "%lu (expanded: %lu)\n",
5978 trace_buf_size >> 10);
5980 r = sprintf(buf, "%lu\n", size >> 10);
5982 r = sprintf(buf, "X\n");
5984 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
5986 mutex_unlock(&trace_types_lock);
5988 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5993 tracing_entries_write(struct file *filp, const char __user *ubuf,
5994 size_t cnt, loff_t *ppos)
5996 struct inode *inode = file_inode(filp);
5997 struct trace_array *tr = inode->i_private;
6001 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6005 /* must have at least 1 entry */
6009 /* value is in KB */
6011 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6021 tracing_total_entries_read(struct file *filp, char __user *ubuf,
6022 size_t cnt, loff_t *ppos)
6024 struct trace_array *tr = filp->private_data;
6027 unsigned long size = 0, expanded_size = 0;
6029 mutex_lock(&trace_types_lock);
6030 for_each_tracing_cpu(cpu) {
6031 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
6032 if (!ring_buffer_expanded)
6033 expanded_size += trace_buf_size >> 10;
6035 if (ring_buffer_expanded)
6036 r = sprintf(buf, "%lu\n", size);
6038 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6039 mutex_unlock(&trace_types_lock);
6041 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6045 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
6046 size_t cnt, loff_t *ppos)
6049 * There is no need to read what the user has written, this function
6050 * is just to make sure that there is no error when "echo" is used
6059 tracing_free_buffer_release(struct inode *inode, struct file *filp)
6061 struct trace_array *tr = inode->i_private;
6063 /* disable tracing ? */
6064 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
6065 tracer_tracing_off(tr);
6066 /* resize the ring buffer to 0 */
6067 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
6069 trace_array_put(tr);
6075 tracing_mark_write(struct file *filp, const char __user *ubuf,
6076 size_t cnt, loff_t *fpos)
6078 struct trace_array *tr = filp->private_data;
6079 struct ring_buffer_event *event;
6080 struct ring_buffer *buffer;
6081 struct print_entry *entry;
6082 unsigned long irq_flags;
6083 const char faulted[] = "<faulted>";
6088 /* Used in tracing_mark_raw_write() as well */
6089 #define FAULTED_SIZE (sizeof(faulted) - 1) /* '\0' is already accounted for */
6091 if (tracing_disabled)
6094 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6097 if (cnt > TRACE_BUF_SIZE)
6098 cnt = TRACE_BUF_SIZE;
6100 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6102 local_save_flags(irq_flags);
6103 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
6105 /* If less than "<faulted>", then make sure we can still add that */
6106 if (cnt < FAULTED_SIZE)
6107 size += FAULTED_SIZE - cnt;
6109 buffer = tr->trace_buffer.buffer;
6110 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
6111 irq_flags, preempt_count());
6112 if (unlikely(!event))
6113 /* Ring buffer disabled, return as if not open for write */
6116 entry = ring_buffer_event_data(event);
6117 entry->ip = _THIS_IP_;
6119 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
6121 memcpy(&entry->buf, faulted, FAULTED_SIZE);
6128 if (entry->buf[cnt - 1] != '\n') {
6129 entry->buf[cnt] = '\n';
6130 entry->buf[cnt + 1] = '\0';
6132 entry->buf[cnt] = '\0';
6134 __buffer_unlock_commit(buffer, event);
6142 /* Limit it for now to 3K (including tag) */
6143 #define RAW_DATA_MAX_SIZE (1024*3)
6146 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
6147 size_t cnt, loff_t *fpos)
6149 struct trace_array *tr = filp->private_data;
6150 struct ring_buffer_event *event;
6151 struct ring_buffer *buffer;
6152 struct raw_data_entry *entry;
6153 const char faulted[] = "<faulted>";
6154 unsigned long irq_flags;
6159 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
6161 if (tracing_disabled)
6164 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6167 /* The marker must at least have a tag id */
6168 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
6171 if (cnt > TRACE_BUF_SIZE)
6172 cnt = TRACE_BUF_SIZE;
6174 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6176 local_save_flags(irq_flags);
6177 size = sizeof(*entry) + cnt;
6178 if (cnt < FAULT_SIZE_ID)
6179 size += FAULT_SIZE_ID - cnt;
6181 buffer = tr->trace_buffer.buffer;
6182 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
6183 irq_flags, preempt_count());
6185 /* Ring buffer disabled, return as if not open for write */
6188 entry = ring_buffer_event_data(event);
6190 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
6193 memcpy(&entry->buf, faulted, FAULTED_SIZE);
6198 __buffer_unlock_commit(buffer, event);
6206 static int tracing_clock_show(struct seq_file *m, void *v)
6208 struct trace_array *tr = m->private;
6211 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
6213 "%s%s%s%s", i ? " " : "",
6214 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
6215 i == tr->clock_id ? "]" : "");
6221 static int tracing_set_clock(struct trace_array *tr, const char *clockstr)
6225 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
6226 if (strcmp(trace_clocks[i].name, clockstr) == 0)
6229 if (i == ARRAY_SIZE(trace_clocks))
6232 mutex_lock(&trace_types_lock);
6236 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
6239 * New clock may not be consistent with the previous clock.
6240 * Reset the buffer so that it doesn't have incomparable timestamps.
6242 tracing_reset_online_cpus(&tr->trace_buffer);
6244 #ifdef CONFIG_TRACER_MAX_TRACE
6245 if (tr->max_buffer.buffer)
6246 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
6247 tracing_reset_online_cpus(&tr->max_buffer);
6250 mutex_unlock(&trace_types_lock);
6255 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
6256 size_t cnt, loff_t *fpos)
6258 struct seq_file *m = filp->private_data;
6259 struct trace_array *tr = m->private;
6261 const char *clockstr;
6264 if (cnt >= sizeof(buf))
6267 if (copy_from_user(buf, ubuf, cnt))
6272 clockstr = strstrip(buf);
6274 ret = tracing_set_clock(tr, clockstr);
6283 static int tracing_clock_open(struct inode *inode, struct file *file)
6285 struct trace_array *tr = inode->i_private;
6288 if (tracing_disabled)
6291 if (trace_array_get(tr))
6294 ret = single_open(file, tracing_clock_show, inode->i_private);
6296 trace_array_put(tr);
6301 struct ftrace_buffer_info {
6302 struct trace_iterator iter;
6304 unsigned int spare_cpu;
6308 #ifdef CONFIG_TRACER_SNAPSHOT
6309 static int tracing_snapshot_open(struct inode *inode, struct file *file)
6311 struct trace_array *tr = inode->i_private;
6312 struct trace_iterator *iter;
6316 if (trace_array_get(tr) < 0)
6319 if (file->f_mode & FMODE_READ) {
6320 iter = __tracing_open(inode, file, true);
6322 ret = PTR_ERR(iter);
6324 /* Writes still need the seq_file to hold the private data */
6326 m = kzalloc(sizeof(*m), GFP_KERNEL);
6329 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6337 iter->trace_buffer = &tr->max_buffer;
6338 iter->cpu_file = tracing_get_cpu(inode);
6340 file->private_data = m;
6344 trace_array_put(tr);
6350 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
6353 struct seq_file *m = filp->private_data;
6354 struct trace_iterator *iter = m->private;
6355 struct trace_array *tr = iter->tr;
6359 ret = tracing_update_buffers();
6363 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6367 mutex_lock(&trace_types_lock);
6369 if (tr->current_trace->use_max_tr) {
6376 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
6380 if (tr->allocated_snapshot)
6384 /* Only allow per-cpu swap if the ring buffer supports it */
6385 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
6386 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
6391 if (!tr->allocated_snapshot) {
6392 ret = alloc_snapshot(tr);
6396 local_irq_disable();
6397 /* Now, we're going to swap */
6398 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
6399 update_max_tr(tr, current, smp_processor_id());
6401 update_max_tr_single(tr, current, iter->cpu_file);
6405 if (tr->allocated_snapshot) {
6406 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
6407 tracing_reset_online_cpus(&tr->max_buffer);
6409 tracing_reset(&tr->max_buffer, iter->cpu_file);
6419 mutex_unlock(&trace_types_lock);
6423 static int tracing_snapshot_release(struct inode *inode, struct file *file)
6425 struct seq_file *m = file->private_data;
6428 ret = tracing_release(inode, file);
6430 if (file->f_mode & FMODE_READ)
6433 /* If write only, the seq_file is just a stub */
6441 static int tracing_buffers_open(struct inode *inode, struct file *filp);
6442 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
6443 size_t count, loff_t *ppos);
6444 static int tracing_buffers_release(struct inode *inode, struct file *file);
6445 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
6446 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
6448 static int snapshot_raw_open(struct inode *inode, struct file *filp)
6450 struct ftrace_buffer_info *info;
6453 ret = tracing_buffers_open(inode, filp);
6457 info = filp->private_data;
6459 if (info->iter.trace->use_max_tr) {
6460 tracing_buffers_release(inode, filp);
6464 info->iter.snapshot = true;
6465 info->iter.trace_buffer = &info->iter.tr->max_buffer;
6470 #endif /* CONFIG_TRACER_SNAPSHOT */
6473 static const struct file_operations tracing_thresh_fops = {
6474 .open = tracing_open_generic,
6475 .read = tracing_thresh_read,
6476 .write = tracing_thresh_write,
6477 .llseek = generic_file_llseek,
6480 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
6481 static const struct file_operations tracing_max_lat_fops = {
6482 .open = tracing_open_generic,
6483 .read = tracing_max_lat_read,
6484 .write = tracing_max_lat_write,
6485 .llseek = generic_file_llseek,
6489 static const struct file_operations set_tracer_fops = {
6490 .open = tracing_open_generic,
6491 .read = tracing_set_trace_read,
6492 .write = tracing_set_trace_write,
6493 .llseek = generic_file_llseek,
6496 static const struct file_operations tracing_pipe_fops = {
6497 .open = tracing_open_pipe,
6498 .poll = tracing_poll_pipe,
6499 .read = tracing_read_pipe,
6500 .splice_read = tracing_splice_read_pipe,
6501 .release = tracing_release_pipe,
6502 .llseek = no_llseek,
6505 static const struct file_operations tracing_entries_fops = {
6506 .open = tracing_open_generic_tr,
6507 .read = tracing_entries_read,
6508 .write = tracing_entries_write,
6509 .llseek = generic_file_llseek,
6510 .release = tracing_release_generic_tr,
6513 static const struct file_operations tracing_total_entries_fops = {
6514 .open = tracing_open_generic_tr,
6515 .read = tracing_total_entries_read,
6516 .llseek = generic_file_llseek,
6517 .release = tracing_release_generic_tr,
6520 static const struct file_operations tracing_free_buffer_fops = {
6521 .open = tracing_open_generic_tr,
6522 .write = tracing_free_buffer_write,
6523 .release = tracing_free_buffer_release,
6526 static const struct file_operations tracing_mark_fops = {
6527 .open = tracing_open_generic_tr,
6528 .write = tracing_mark_write,
6529 .llseek = generic_file_llseek,
6530 .release = tracing_release_generic_tr,
6533 static const struct file_operations tracing_mark_raw_fops = {
6534 .open = tracing_open_generic_tr,
6535 .write = tracing_mark_raw_write,
6536 .llseek = generic_file_llseek,
6537 .release = tracing_release_generic_tr,
6540 static const struct file_operations trace_clock_fops = {
6541 .open = tracing_clock_open,
6543 .llseek = seq_lseek,
6544 .release = tracing_single_release_tr,
6545 .write = tracing_clock_write,
6548 #ifdef CONFIG_TRACER_SNAPSHOT
6549 static const struct file_operations snapshot_fops = {
6550 .open = tracing_snapshot_open,
6552 .write = tracing_snapshot_write,
6553 .llseek = tracing_lseek,
6554 .release = tracing_snapshot_release,
6557 static const struct file_operations snapshot_raw_fops = {
6558 .open = snapshot_raw_open,
6559 .read = tracing_buffers_read,
6560 .release = tracing_buffers_release,
6561 .splice_read = tracing_buffers_splice_read,
6562 .llseek = no_llseek,
6565 #endif /* CONFIG_TRACER_SNAPSHOT */
6567 static int tracing_buffers_open(struct inode *inode, struct file *filp)
6569 struct trace_array *tr = inode->i_private;
6570 struct ftrace_buffer_info *info;
6573 if (tracing_disabled)
6576 if (trace_array_get(tr) < 0)
6579 info = kzalloc(sizeof(*info), GFP_KERNEL);
6581 trace_array_put(tr);
6585 mutex_lock(&trace_types_lock);
6588 info->iter.cpu_file = tracing_get_cpu(inode);
6589 info->iter.trace = tr->current_trace;
6590 info->iter.trace_buffer = &tr->trace_buffer;
6592 /* Force reading ring buffer for first read */
6593 info->read = (unsigned int)-1;
6595 filp->private_data = info;
6597 tr->current_trace->ref++;
6599 mutex_unlock(&trace_types_lock);
6601 ret = nonseekable_open(inode, filp);
6603 trace_array_put(tr);
6609 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
6611 struct ftrace_buffer_info *info = filp->private_data;
6612 struct trace_iterator *iter = &info->iter;
6614 return trace_poll(iter, filp, poll_table);
6618 tracing_buffers_read(struct file *filp, char __user *ubuf,
6619 size_t count, loff_t *ppos)
6621 struct ftrace_buffer_info *info = filp->private_data;
6622 struct trace_iterator *iter = &info->iter;
6629 #ifdef CONFIG_TRACER_MAX_TRACE
6630 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
6635 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
6637 if (IS_ERR(info->spare)) {
6638 ret = PTR_ERR(info->spare);
6641 info->spare_cpu = iter->cpu_file;
6647 /* Do we have previous read data to read? */
6648 if (info->read < PAGE_SIZE)
6652 trace_access_lock(iter->cpu_file);
6653 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
6657 trace_access_unlock(iter->cpu_file);
6660 if (trace_empty(iter)) {
6661 if ((filp->f_flags & O_NONBLOCK))
6664 ret = wait_on_pipe(iter, false);
6675 size = PAGE_SIZE - info->read;
6679 ret = copy_to_user(ubuf, info->spare + info->read, size);
6691 static int tracing_buffers_release(struct inode *inode, struct file *file)
6693 struct ftrace_buffer_info *info = file->private_data;
6694 struct trace_iterator *iter = &info->iter;
6696 mutex_lock(&trace_types_lock);
6698 iter->tr->current_trace->ref--;
6700 __trace_array_put(iter->tr);
6703 ring_buffer_free_read_page(iter->trace_buffer->buffer,
6704 info->spare_cpu, info->spare);
6707 mutex_unlock(&trace_types_lock);
6713 struct ring_buffer *buffer;
6719 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
6720 struct pipe_buffer *buf)
6722 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
6727 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
6732 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
6733 struct pipe_buffer *buf)
6735 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
6740 /* Pipe buffer operations for a buffer. */
6741 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
6743 .confirm = generic_pipe_buf_confirm,
6744 .release = buffer_pipe_buf_release,
6745 .steal = generic_pipe_buf_steal,
6746 .get = buffer_pipe_buf_get,
6750 * Callback from splice_to_pipe(), if we need to release some pages
6751 * at the end of the spd in case we error'ed out in filling the pipe.
6753 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
6755 struct buffer_ref *ref =
6756 (struct buffer_ref *)spd->partial[i].private;
6761 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
6763 spd->partial[i].private = 0;
6767 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
6768 struct pipe_inode_info *pipe, size_t len,
6771 struct ftrace_buffer_info *info = file->private_data;
6772 struct trace_iterator *iter = &info->iter;
6773 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6774 struct page *pages_def[PIPE_DEF_BUFFERS];
6775 struct splice_pipe_desc spd = {
6777 .partial = partial_def,
6778 .nr_pages_max = PIPE_DEF_BUFFERS,
6779 .ops = &buffer_pipe_buf_ops,
6780 .spd_release = buffer_spd_release,
6782 struct buffer_ref *ref;
6783 int entries, size, i;
6786 #ifdef CONFIG_TRACER_MAX_TRACE
6787 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
6791 if (*ppos & (PAGE_SIZE - 1))
6794 if (len & (PAGE_SIZE - 1)) {
6795 if (len < PAGE_SIZE)
6800 if (splice_grow_spd(pipe, &spd))
6804 trace_access_lock(iter->cpu_file);
6805 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
6807 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
6811 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
6818 ref->buffer = iter->trace_buffer->buffer;
6819 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
6820 if (IS_ERR(ref->page)) {
6821 ret = PTR_ERR(ref->page);
6826 ref->cpu = iter->cpu_file;
6828 r = ring_buffer_read_page(ref->buffer, &ref->page,
6829 len, iter->cpu_file, 1);
6831 ring_buffer_free_read_page(ref->buffer, ref->cpu,
6838 * zero out any left over data, this is going to
6841 size = ring_buffer_page_len(ref->page);
6842 if (size < PAGE_SIZE)
6843 memset(ref->page + size, 0, PAGE_SIZE - size);
6845 page = virt_to_page(ref->page);
6847 spd.pages[i] = page;
6848 spd.partial[i].len = PAGE_SIZE;
6849 spd.partial[i].offset = 0;
6850 spd.partial[i].private = (unsigned long)ref;
6854 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
6857 trace_access_unlock(iter->cpu_file);
6860 /* did we read anything? */
6861 if (!spd.nr_pages) {
6866 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
6869 ret = wait_on_pipe(iter, true);
6876 ret = splice_to_pipe(pipe, &spd);
6878 splice_shrink_spd(&spd);
6883 static const struct file_operations tracing_buffers_fops = {
6884 .open = tracing_buffers_open,
6885 .read = tracing_buffers_read,
6886 .poll = tracing_buffers_poll,
6887 .release = tracing_buffers_release,
6888 .splice_read = tracing_buffers_splice_read,
6889 .llseek = no_llseek,
6893 tracing_stats_read(struct file *filp, char __user *ubuf,
6894 size_t count, loff_t *ppos)
6896 struct inode *inode = file_inode(filp);
6897 struct trace_array *tr = inode->i_private;
6898 struct trace_buffer *trace_buf = &tr->trace_buffer;
6899 int cpu = tracing_get_cpu(inode);
6900 struct trace_seq *s;
6902 unsigned long long t;
6903 unsigned long usec_rem;
6905 s = kmalloc(sizeof(*s), GFP_KERNEL);
6911 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
6912 trace_seq_printf(s, "entries: %ld\n", cnt);
6914 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
6915 trace_seq_printf(s, "overrun: %ld\n", cnt);
6917 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
6918 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
6920 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
6921 trace_seq_printf(s, "bytes: %ld\n", cnt);
6923 if (trace_clocks[tr->clock_id].in_ns) {
6924 /* local or global for trace_clock */
6925 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
6926 usec_rem = do_div(t, USEC_PER_SEC);
6927 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
6930 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
6931 usec_rem = do_div(t, USEC_PER_SEC);
6932 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
6934 /* counter or tsc mode for trace_clock */
6935 trace_seq_printf(s, "oldest event ts: %llu\n",
6936 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
6938 trace_seq_printf(s, "now ts: %llu\n",
6939 ring_buffer_time_stamp(trace_buf->buffer, cpu));
6942 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
6943 trace_seq_printf(s, "dropped events: %ld\n", cnt);
6945 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
6946 trace_seq_printf(s, "read events: %ld\n", cnt);
6948 count = simple_read_from_buffer(ubuf, count, ppos,
6949 s->buffer, trace_seq_used(s));
6956 static const struct file_operations tracing_stats_fops = {
6957 .open = tracing_open_generic_tr,
6958 .read = tracing_stats_read,
6959 .llseek = generic_file_llseek,
6960 .release = tracing_release_generic_tr,
6963 #ifdef CONFIG_DYNAMIC_FTRACE
6966 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
6967 size_t cnt, loff_t *ppos)
6969 unsigned long *p = filp->private_data;
6970 char buf[64]; /* Not too big for a shallow stack */
6973 r = scnprintf(buf, 63, "%ld", *p);
6976 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6979 static const struct file_operations tracing_dyn_info_fops = {
6980 .open = tracing_open_generic,
6981 .read = tracing_read_dyn_info,
6982 .llseek = generic_file_llseek,
6984 #endif /* CONFIG_DYNAMIC_FTRACE */
6986 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
6988 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
6989 struct trace_array *tr, struct ftrace_probe_ops *ops,
6992 tracing_snapshot_instance(tr);
6996 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
6997 struct trace_array *tr, struct ftrace_probe_ops *ops,
7000 struct ftrace_func_mapper *mapper = data;
7004 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7014 tracing_snapshot_instance(tr);
7018 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
7019 struct ftrace_probe_ops *ops, void *data)
7021 struct ftrace_func_mapper *mapper = data;
7024 seq_printf(m, "%ps:", (void *)ip);
7026 seq_puts(m, "snapshot");
7029 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7032 seq_printf(m, ":count=%ld\n", *count);
7034 seq_puts(m, ":unlimited\n");
7040 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
7041 unsigned long ip, void *init_data, void **data)
7043 struct ftrace_func_mapper *mapper = *data;
7046 mapper = allocate_ftrace_func_mapper();
7052 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
7056 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
7057 unsigned long ip, void *data)
7059 struct ftrace_func_mapper *mapper = data;
7064 free_ftrace_func_mapper(mapper, NULL);
7068 ftrace_func_mapper_remove_ip(mapper, ip);
7071 static struct ftrace_probe_ops snapshot_probe_ops = {
7072 .func = ftrace_snapshot,
7073 .print = ftrace_snapshot_print,
7076 static struct ftrace_probe_ops snapshot_count_probe_ops = {
7077 .func = ftrace_count_snapshot,
7078 .print = ftrace_snapshot_print,
7079 .init = ftrace_snapshot_init,
7080 .free = ftrace_snapshot_free,
7084 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
7085 char *glob, char *cmd, char *param, int enable)
7087 struct ftrace_probe_ops *ops;
7088 void *count = (void *)-1;
7095 /* hash funcs only work with set_ftrace_filter */
7099 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
7102 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
7107 number = strsep(¶m, ":");
7109 if (!strlen(number))
7113 * We use the callback data field (which is a pointer)
7116 ret = kstrtoul(number, 0, (unsigned long *)&count);
7121 ret = alloc_snapshot(tr);
7125 ret = register_ftrace_function_probe(glob, tr, ops, count);
7128 return ret < 0 ? ret : 0;
7131 static struct ftrace_func_command ftrace_snapshot_cmd = {
7133 .func = ftrace_trace_snapshot_callback,
7136 static __init int register_snapshot_cmd(void)
7138 return register_ftrace_command(&ftrace_snapshot_cmd);
7141 static inline __init int register_snapshot_cmd(void) { return 0; }
7142 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
7144 static struct dentry *tracing_get_dentry(struct trace_array *tr)
7146 if (WARN_ON(!tr->dir))
7147 return ERR_PTR(-ENODEV);
7149 /* Top directory uses NULL as the parent */
7150 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
7153 /* All sub buffers have a descriptor */
7157 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
7159 struct dentry *d_tracer;
7162 return tr->percpu_dir;
7164 d_tracer = tracing_get_dentry(tr);
7165 if (IS_ERR(d_tracer))
7168 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
7170 WARN_ONCE(!tr->percpu_dir,
7171 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
7173 return tr->percpu_dir;
7176 static struct dentry *
7177 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
7178 void *data, long cpu, const struct file_operations *fops)
7180 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
7182 if (ret) /* See tracing_get_cpu() */
7183 d_inode(ret)->i_cdev = (void *)(cpu + 1);
7188 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
7190 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
7191 struct dentry *d_cpu;
7192 char cpu_dir[30]; /* 30 characters should be more than enough */
7197 snprintf(cpu_dir, 30, "cpu%ld", cpu);
7198 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
7200 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
7204 /* per cpu trace_pipe */
7205 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
7206 tr, cpu, &tracing_pipe_fops);
7209 trace_create_cpu_file("trace", 0644, d_cpu,
7210 tr, cpu, &tracing_fops);
7212 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
7213 tr, cpu, &tracing_buffers_fops);
7215 trace_create_cpu_file("stats", 0444, d_cpu,
7216 tr, cpu, &tracing_stats_fops);
7218 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
7219 tr, cpu, &tracing_entries_fops);
7221 #ifdef CONFIG_TRACER_SNAPSHOT
7222 trace_create_cpu_file("snapshot", 0644, d_cpu,
7223 tr, cpu, &snapshot_fops);
7225 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
7226 tr, cpu, &snapshot_raw_fops);
7230 #ifdef CONFIG_FTRACE_SELFTEST
7231 /* Let selftest have access to static functions in this file */
7232 #include "trace_selftest.c"
7236 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
7239 struct trace_option_dentry *topt = filp->private_data;
7242 if (topt->flags->val & topt->opt->bit)
7247 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
7251 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
7254 struct trace_option_dentry *topt = filp->private_data;
7258 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7262 if (val != 0 && val != 1)
7265 if (!!(topt->flags->val & topt->opt->bit) != val) {
7266 mutex_lock(&trace_types_lock);
7267 ret = __set_tracer_option(topt->tr, topt->flags,
7269 mutex_unlock(&trace_types_lock);
7280 static const struct file_operations trace_options_fops = {
7281 .open = tracing_open_generic,
7282 .read = trace_options_read,
7283 .write = trace_options_write,
7284 .llseek = generic_file_llseek,
7288 * In order to pass in both the trace_array descriptor as well as the index
7289 * to the flag that the trace option file represents, the trace_array
7290 * has a character array of trace_flags_index[], which holds the index
7291 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
7292 * The address of this character array is passed to the flag option file
7293 * read/write callbacks.
7295 * In order to extract both the index and the trace_array descriptor,
7296 * get_tr_index() uses the following algorithm.
7300 * As the pointer itself contains the address of the index (remember
7303 * Then to get the trace_array descriptor, by subtracting that index
7304 * from the ptr, we get to the start of the index itself.
7306 * ptr - idx == &index[0]
7308 * Then a simple container_of() from that pointer gets us to the
7309 * trace_array descriptor.
7311 static void get_tr_index(void *data, struct trace_array **ptr,
7312 unsigned int *pindex)
7314 *pindex = *(unsigned char *)data;
7316 *ptr = container_of(data - *pindex, struct trace_array,
7321 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
7324 void *tr_index = filp->private_data;
7325 struct trace_array *tr;
7329 get_tr_index(tr_index, &tr, &index);
7331 if (tr->trace_flags & (1 << index))
7336 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
7340 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
7343 void *tr_index = filp->private_data;
7344 struct trace_array *tr;
7349 get_tr_index(tr_index, &tr, &index);
7351 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7355 if (val != 0 && val != 1)
7358 mutex_lock(&trace_types_lock);
7359 ret = set_tracer_flag(tr, 1 << index, val);
7360 mutex_unlock(&trace_types_lock);
7370 static const struct file_operations trace_options_core_fops = {
7371 .open = tracing_open_generic,
7372 .read = trace_options_core_read,
7373 .write = trace_options_core_write,
7374 .llseek = generic_file_llseek,
7377 struct dentry *trace_create_file(const char *name,
7379 struct dentry *parent,
7381 const struct file_operations *fops)
7385 ret = tracefs_create_file(name, mode, parent, data, fops);
7387 pr_warn("Could not create tracefs '%s' entry\n", name);
7393 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
7395 struct dentry *d_tracer;
7400 d_tracer = tracing_get_dentry(tr);
7401 if (IS_ERR(d_tracer))
7404 tr->options = tracefs_create_dir("options", d_tracer);
7406 pr_warn("Could not create tracefs directory 'options'\n");
7414 create_trace_option_file(struct trace_array *tr,
7415 struct trace_option_dentry *topt,
7416 struct tracer_flags *flags,
7417 struct tracer_opt *opt)
7419 struct dentry *t_options;
7421 t_options = trace_options_init_dentry(tr);
7425 topt->flags = flags;
7429 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
7430 &trace_options_fops);
7435 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
7437 struct trace_option_dentry *topts;
7438 struct trace_options *tr_topts;
7439 struct tracer_flags *flags;
7440 struct tracer_opt *opts;
7447 flags = tracer->flags;
7449 if (!flags || !flags->opts)
7453 * If this is an instance, only create flags for tracers
7454 * the instance may have.
7456 if (!trace_ok_for_array(tracer, tr))
7459 for (i = 0; i < tr->nr_topts; i++) {
7460 /* Make sure there's no duplicate flags. */
7461 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
7467 for (cnt = 0; opts[cnt].name; cnt++)
7470 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
7474 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
7481 tr->topts = tr_topts;
7482 tr->topts[tr->nr_topts].tracer = tracer;
7483 tr->topts[tr->nr_topts].topts = topts;
7486 for (cnt = 0; opts[cnt].name; cnt++) {
7487 create_trace_option_file(tr, &topts[cnt], flags,
7489 WARN_ONCE(topts[cnt].entry == NULL,
7490 "Failed to create trace option: %s",
7495 static struct dentry *
7496 create_trace_option_core_file(struct trace_array *tr,
7497 const char *option, long index)
7499 struct dentry *t_options;
7501 t_options = trace_options_init_dentry(tr);
7505 return trace_create_file(option, 0644, t_options,
7506 (void *)&tr->trace_flags_index[index],
7507 &trace_options_core_fops);
7510 static void create_trace_options_dir(struct trace_array *tr)
7512 struct dentry *t_options;
7513 bool top_level = tr == &global_trace;
7516 t_options = trace_options_init_dentry(tr);
7520 for (i = 0; trace_options[i]; i++) {
7522 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
7523 create_trace_option_core_file(tr, trace_options[i], i);
7528 rb_simple_read(struct file *filp, char __user *ubuf,
7529 size_t cnt, loff_t *ppos)
7531 struct trace_array *tr = filp->private_data;
7535 r = tracer_tracing_is_on(tr);
7536 r = sprintf(buf, "%d\n", r);
7538 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7542 rb_simple_write(struct file *filp, const char __user *ubuf,
7543 size_t cnt, loff_t *ppos)
7545 struct trace_array *tr = filp->private_data;
7546 struct ring_buffer *buffer = tr->trace_buffer.buffer;
7550 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7555 mutex_lock(&trace_types_lock);
7557 tracer_tracing_on(tr);
7558 if (tr->current_trace->start)
7559 tr->current_trace->start(tr);
7561 tracer_tracing_off(tr);
7562 if (tr->current_trace->stop)
7563 tr->current_trace->stop(tr);
7565 mutex_unlock(&trace_types_lock);
7573 static const struct file_operations rb_simple_fops = {
7574 .open = tracing_open_generic_tr,
7575 .read = rb_simple_read,
7576 .write = rb_simple_write,
7577 .release = tracing_release_generic_tr,
7578 .llseek = default_llseek,
7581 struct dentry *trace_instance_dir;
7584 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
7587 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
7589 enum ring_buffer_flags rb_flags;
7591 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
7595 buf->buffer = ring_buffer_alloc(size, rb_flags);
7599 buf->data = alloc_percpu(struct trace_array_cpu);
7601 ring_buffer_free(buf->buffer);
7605 /* Allocate the first page for all buffers */
7606 set_buffer_entries(&tr->trace_buffer,
7607 ring_buffer_size(tr->trace_buffer.buffer, 0));
7612 static int allocate_trace_buffers(struct trace_array *tr, int size)
7616 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
7620 #ifdef CONFIG_TRACER_MAX_TRACE
7621 ret = allocate_trace_buffer(tr, &tr->max_buffer,
7622 allocate_snapshot ? size : 1);
7624 ring_buffer_free(tr->trace_buffer.buffer);
7625 free_percpu(tr->trace_buffer.data);
7628 tr->allocated_snapshot = allocate_snapshot;
7631 * Only the top level trace array gets its snapshot allocated
7632 * from the kernel command line.
7634 allocate_snapshot = false;
7639 static void free_trace_buffer(struct trace_buffer *buf)
7642 ring_buffer_free(buf->buffer);
7644 free_percpu(buf->data);
7649 static void free_trace_buffers(struct trace_array *tr)
7654 free_trace_buffer(&tr->trace_buffer);
7656 #ifdef CONFIG_TRACER_MAX_TRACE
7657 free_trace_buffer(&tr->max_buffer);
7661 static void init_trace_flags_index(struct trace_array *tr)
7665 /* Used by the trace options files */
7666 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
7667 tr->trace_flags_index[i] = i;
7670 static void __update_tracer_options(struct trace_array *tr)
7674 for (t = trace_types; t; t = t->next)
7675 add_tracer_options(tr, t);
7678 static void update_tracer_options(struct trace_array *tr)
7680 mutex_lock(&trace_types_lock);
7681 __update_tracer_options(tr);
7682 mutex_unlock(&trace_types_lock);
7685 static int instance_mkdir(const char *name)
7687 struct trace_array *tr;
7690 mutex_lock(&event_mutex);
7691 mutex_lock(&trace_types_lock);
7694 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7695 if (tr->name && strcmp(tr->name, name) == 0)
7700 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
7704 tr->name = kstrdup(name, GFP_KERNEL);
7708 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
7711 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
7713 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
7715 raw_spin_lock_init(&tr->start_lock);
7717 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
7719 tr->current_trace = &nop_trace;
7721 INIT_LIST_HEAD(&tr->systems);
7722 INIT_LIST_HEAD(&tr->events);
7724 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
7727 tr->dir = tracefs_create_dir(name, trace_instance_dir);
7731 ret = event_trace_add_tracer(tr->dir, tr);
7733 tracefs_remove_recursive(tr->dir);
7737 ftrace_init_trace_array(tr);
7739 init_tracer_tracefs(tr, tr->dir);
7740 init_trace_flags_index(tr);
7741 __update_tracer_options(tr);
7743 list_add(&tr->list, &ftrace_trace_arrays);
7745 mutex_unlock(&trace_types_lock);
7746 mutex_unlock(&event_mutex);
7751 free_trace_buffers(tr);
7752 free_cpumask_var(tr->tracing_cpumask);
7757 mutex_unlock(&trace_types_lock);
7758 mutex_unlock(&event_mutex);
7764 static int instance_rmdir(const char *name)
7766 struct trace_array *tr;
7771 mutex_lock(&event_mutex);
7772 mutex_lock(&trace_types_lock);
7775 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
7776 if (tr->name && strcmp(tr->name, name) == 0) {
7785 if (tr->ref || (tr->current_trace && tr->current_trace->ref))
7788 list_del(&tr->list);
7790 /* Disable all the flags that were enabled coming in */
7791 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
7792 if ((1 << i) & ZEROED_TRACE_FLAGS)
7793 set_tracer_flag(tr, 1 << i, 0);
7796 tracing_set_nop(tr);
7797 clear_ftrace_function_probes(tr);
7798 event_trace_del_tracer(tr);
7799 ftrace_clear_pids(tr);
7800 ftrace_destroy_function_files(tr);
7801 tracefs_remove_recursive(tr->dir);
7802 free_trace_buffers(tr);
7804 for (i = 0; i < tr->nr_topts; i++) {
7805 kfree(tr->topts[i].topts);
7809 free_cpumask_var(tr->tracing_cpumask);
7816 mutex_unlock(&trace_types_lock);
7817 mutex_unlock(&event_mutex);
7822 static __init void create_trace_instances(struct dentry *d_tracer)
7824 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
7827 if (WARN_ON(!trace_instance_dir))
7832 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
7836 trace_create_file("available_tracers", 0444, d_tracer,
7837 tr, &show_traces_fops);
7839 trace_create_file("current_tracer", 0644, d_tracer,
7840 tr, &set_tracer_fops);
7842 trace_create_file("tracing_cpumask", 0644, d_tracer,
7843 tr, &tracing_cpumask_fops);
7845 trace_create_file("trace_options", 0644, d_tracer,
7846 tr, &tracing_iter_fops);
7848 trace_create_file("trace", 0644, d_tracer,
7851 trace_create_file("trace_pipe", 0444, d_tracer,
7852 tr, &tracing_pipe_fops);
7854 trace_create_file("buffer_size_kb", 0644, d_tracer,
7855 tr, &tracing_entries_fops);
7857 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
7858 tr, &tracing_total_entries_fops);
7860 trace_create_file("free_buffer", 0200, d_tracer,
7861 tr, &tracing_free_buffer_fops);
7863 trace_create_file("trace_marker", 0220, d_tracer,
7864 tr, &tracing_mark_fops);
7866 trace_create_file("trace_marker_raw", 0220, d_tracer,
7867 tr, &tracing_mark_raw_fops);
7869 trace_create_file("trace_clock", 0644, d_tracer, tr,
7872 trace_create_file("tracing_on", 0644, d_tracer,
7873 tr, &rb_simple_fops);
7875 create_trace_options_dir(tr);
7877 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
7878 trace_create_file("tracing_max_latency", 0644, d_tracer,
7879 &tr->max_latency, &tracing_max_lat_fops);
7882 if (ftrace_create_function_files(tr, d_tracer))
7883 WARN(1, "Could not allocate function filter files");
7885 #ifdef CONFIG_TRACER_SNAPSHOT
7886 trace_create_file("snapshot", 0644, d_tracer,
7887 tr, &snapshot_fops);
7890 for_each_tracing_cpu(cpu)
7891 tracing_init_tracefs_percpu(tr, cpu);
7893 ftrace_init_tracefs(tr, d_tracer);
7896 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
7898 struct vfsmount *mnt;
7899 struct file_system_type *type;
7902 * To maintain backward compatibility for tools that mount
7903 * debugfs to get to the tracing facility, tracefs is automatically
7904 * mounted to the debugfs/tracing directory.
7906 type = get_fs_type("tracefs");
7909 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
7910 put_filesystem(type);
7919 * tracing_init_dentry - initialize top level trace array
7921 * This is called when creating files or directories in the tracing
7922 * directory. It is called via fs_initcall() by any of the boot up code
7923 * and expects to return the dentry of the top level tracing directory.
7925 struct dentry *tracing_init_dentry(void)
7927 struct trace_array *tr = &global_trace;
7929 /* The top level trace array uses NULL as parent */
7933 if (WARN_ON(!tracefs_initialized()) ||
7934 (IS_ENABLED(CONFIG_DEBUG_FS) &&
7935 WARN_ON(!debugfs_initialized())))
7936 return ERR_PTR(-ENODEV);
7939 * As there may still be users that expect the tracing
7940 * files to exist in debugfs/tracing, we must automount
7941 * the tracefs file system there, so older tools still
7942 * work with the newer kerenl.
7944 tr->dir = debugfs_create_automount("tracing", NULL,
7945 trace_automount, NULL);
7947 pr_warn_once("Could not create debugfs directory 'tracing'\n");
7948 return ERR_PTR(-ENOMEM);
7954 extern struct trace_eval_map *__start_ftrace_eval_maps[];
7955 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
7957 static void __init trace_eval_init(void)
7961 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
7962 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
7965 #ifdef CONFIG_MODULES
7966 static void trace_module_add_evals(struct module *mod)
7968 if (!mod->num_trace_evals)
7972 * Modules with bad taint do not have events created, do
7973 * not bother with enums either.
7975 if (trace_module_has_bad_taint(mod))
7978 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
7981 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
7982 static void trace_module_remove_evals(struct module *mod)
7984 union trace_eval_map_item *map;
7985 union trace_eval_map_item **last = &trace_eval_maps;
7987 if (!mod->num_trace_evals)
7990 mutex_lock(&trace_eval_mutex);
7992 map = trace_eval_maps;
7995 if (map->head.mod == mod)
7997 map = trace_eval_jmp_to_tail(map);
7998 last = &map->tail.next;
7999 map = map->tail.next;
8004 *last = trace_eval_jmp_to_tail(map)->tail.next;
8007 mutex_unlock(&trace_eval_mutex);
8010 static inline void trace_module_remove_evals(struct module *mod) { }
8011 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
8013 static int trace_module_notify(struct notifier_block *self,
8014 unsigned long val, void *data)
8016 struct module *mod = data;
8019 case MODULE_STATE_COMING:
8020 trace_module_add_evals(mod);
8022 case MODULE_STATE_GOING:
8023 trace_module_remove_evals(mod);
8030 static struct notifier_block trace_module_nb = {
8031 .notifier_call = trace_module_notify,
8034 #endif /* CONFIG_MODULES */
8036 static __init int tracer_init_tracefs(void)
8038 struct dentry *d_tracer;
8040 trace_access_lock_init();
8042 d_tracer = tracing_init_dentry();
8043 if (IS_ERR(d_tracer))
8046 init_tracer_tracefs(&global_trace, d_tracer);
8047 ftrace_init_tracefs_toplevel(&global_trace, d_tracer);
8049 trace_create_file("tracing_thresh", 0644, d_tracer,
8050 &global_trace, &tracing_thresh_fops);
8052 trace_create_file("README", 0444, d_tracer,
8053 NULL, &tracing_readme_fops);
8055 trace_create_file("saved_cmdlines", 0444, d_tracer,
8056 NULL, &tracing_saved_cmdlines_fops);
8058 trace_create_file("saved_cmdlines_size", 0644, d_tracer,
8059 NULL, &tracing_saved_cmdlines_size_fops);
8061 trace_create_file("saved_tgids", 0444, d_tracer,
8062 NULL, &tracing_saved_tgids_fops);
8066 trace_create_eval_file(d_tracer);
8068 #ifdef CONFIG_MODULES
8069 register_module_notifier(&trace_module_nb);
8072 #ifdef CONFIG_DYNAMIC_FTRACE
8073 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
8074 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
8077 create_trace_instances(d_tracer);
8079 update_tracer_options(&global_trace);
8084 static int trace_panic_handler(struct notifier_block *this,
8085 unsigned long event, void *unused)
8087 if (ftrace_dump_on_oops)
8088 ftrace_dump(ftrace_dump_on_oops);
8092 static struct notifier_block trace_panic_notifier = {
8093 .notifier_call = trace_panic_handler,
8095 .priority = 150 /* priority: INT_MAX >= x >= 0 */
8098 static int trace_die_handler(struct notifier_block *self,
8104 if (ftrace_dump_on_oops)
8105 ftrace_dump(ftrace_dump_on_oops);
8113 static struct notifier_block trace_die_notifier = {
8114 .notifier_call = trace_die_handler,
8119 * printk is set to max of 1024, we really don't need it that big.
8120 * Nothing should be printing 1000 characters anyway.
8122 #define TRACE_MAX_PRINT 1000
8125 * Define here KERN_TRACE so that we have one place to modify
8126 * it if we decide to change what log level the ftrace dump
8129 #define KERN_TRACE KERN_EMERG
8132 trace_printk_seq(struct trace_seq *s)
8134 /* Probably should print a warning here. */
8135 if (s->seq.len >= TRACE_MAX_PRINT)
8136 s->seq.len = TRACE_MAX_PRINT;
8139 * More paranoid code. Although the buffer size is set to
8140 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
8141 * an extra layer of protection.
8143 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
8144 s->seq.len = s->seq.size - 1;
8146 /* should be zero ended, but we are paranoid. */
8147 s->buffer[s->seq.len] = 0;
8149 printk(KERN_TRACE "%s", s->buffer);
8154 void trace_init_global_iter(struct trace_iterator *iter)
8156 iter->tr = &global_trace;
8157 iter->trace = iter->tr->current_trace;
8158 iter->cpu_file = RING_BUFFER_ALL_CPUS;
8159 iter->trace_buffer = &global_trace.trace_buffer;
8161 if (iter->trace && iter->trace->open)
8162 iter->trace->open(iter);
8164 /* Annotate start of buffers if we had overruns */
8165 if (ring_buffer_overruns(iter->trace_buffer->buffer))
8166 iter->iter_flags |= TRACE_FILE_ANNOTATE;
8168 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
8169 if (trace_clocks[iter->tr->clock_id].in_ns)
8170 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
8173 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
8175 /* use static because iter can be a bit big for the stack */
8176 static struct trace_iterator iter;
8177 static atomic_t dump_running;
8178 struct trace_array *tr = &global_trace;
8179 unsigned int old_userobj;
8180 unsigned long flags;
8183 /* Only allow one dump user at a time. */
8184 if (atomic_inc_return(&dump_running) != 1) {
8185 atomic_dec(&dump_running);
8190 * Always turn off tracing when we dump.
8191 * We don't need to show trace output of what happens
8192 * between multiple crashes.
8194 * If the user does a sysrq-z, then they can re-enable
8195 * tracing with echo 1 > tracing_on.
8199 local_irq_save(flags);
8201 /* Simulate the iterator */
8202 trace_init_global_iter(&iter);
8204 for_each_tracing_cpu(cpu) {
8205 atomic_inc(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
8208 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
8210 /* don't look at user memory in panic mode */
8211 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
8213 switch (oops_dump_mode) {
8215 iter.cpu_file = RING_BUFFER_ALL_CPUS;
8218 iter.cpu_file = raw_smp_processor_id();
8223 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
8224 iter.cpu_file = RING_BUFFER_ALL_CPUS;
8227 printk(KERN_TRACE "Dumping ftrace buffer:\n");
8229 /* Did function tracer already get disabled? */
8230 if (ftrace_is_dead()) {
8231 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
8232 printk("# MAY BE MISSING FUNCTION EVENTS\n");
8236 * We need to stop all tracing on all CPUS to read the
8237 * the next buffer. This is a bit expensive, but is
8238 * not done often. We fill all what we can read,
8239 * and then release the locks again.
8242 while (!trace_empty(&iter)) {
8245 printk(KERN_TRACE "---------------------------------\n");
8249 /* reset all but tr, trace, and overruns */
8250 memset(&iter.seq, 0,
8251 sizeof(struct trace_iterator) -
8252 offsetof(struct trace_iterator, seq));
8253 iter.iter_flags |= TRACE_FILE_LAT_FMT;
8256 if (trace_find_next_entry_inc(&iter) != NULL) {
8259 ret = print_trace_line(&iter);
8260 if (ret != TRACE_TYPE_NO_CONSUME)
8261 trace_consume(&iter);
8263 touch_nmi_watchdog();
8265 trace_printk_seq(&iter.seq);
8269 printk(KERN_TRACE " (ftrace buffer empty)\n");
8271 printk(KERN_TRACE "---------------------------------\n");
8274 tr->trace_flags |= old_userobj;
8276 for_each_tracing_cpu(cpu) {
8277 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
8279 atomic_dec(&dump_running);
8280 local_irq_restore(flags);
8282 EXPORT_SYMBOL_GPL(ftrace_dump);
8284 int trace_run_command(const char *buf, int (*createfn)(int, char **))
8291 argv = argv_split(GFP_KERNEL, buf, &argc);
8296 ret = createfn(argc, argv);
8303 #define WRITE_BUFSIZE 4096
8305 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
8306 size_t count, loff_t *ppos,
8307 int (*createfn)(int, char **))
8309 char *kbuf, *buf, *tmp;
8314 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
8318 while (done < count) {
8319 size = count - done;
8321 if (size >= WRITE_BUFSIZE)
8322 size = WRITE_BUFSIZE - 1;
8324 if (copy_from_user(kbuf, buffer + done, size)) {
8331 tmp = strchr(buf, '\n');
8334 size = tmp - buf + 1;
8337 if (done + size < count) {
8340 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
8341 pr_warn("Line length is too long: Should be less than %d\n",
8349 /* Remove comments */
8350 tmp = strchr(buf, '#');
8355 ret = trace_run_command(buf, createfn);
8360 } while (done < count);
8370 __init static int tracer_alloc_buffers(void)
8376 * Make sure we don't accidently add more trace options
8377 * than we have bits for.
8379 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
8381 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
8384 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
8385 goto out_free_buffer_mask;
8387 /* Only allocate trace_printk buffers if a trace_printk exists */
8388 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
8389 /* Must be called before global_trace.buffer is allocated */
8390 trace_printk_init_buffers();
8392 /* To save memory, keep the ring buffer size to its minimum */
8393 if (ring_buffer_expanded)
8394 ring_buf_size = trace_buf_size;
8398 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
8399 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
8401 raw_spin_lock_init(&global_trace.start_lock);
8404 * The prepare callbacks allocates some memory for the ring buffer. We
8405 * don't free the buffer if the if the CPU goes down. If we were to free
8406 * the buffer, then the user would lose any trace that was in the
8407 * buffer. The memory will be removed once the "instance" is removed.
8409 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
8410 "trace/RB:preapre", trace_rb_cpu_prepare,
8413 goto out_free_cpumask;
8414 /* Used for event triggers */
8416 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
8418 goto out_rm_hp_state;
8420 if (trace_create_savedcmd() < 0)
8421 goto out_free_temp_buffer;
8423 /* TODO: make the number of buffers hot pluggable with CPUS */
8424 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
8425 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
8427 goto out_free_savedcmd;
8430 if (global_trace.buffer_disabled)
8433 if (trace_boot_clock) {
8434 ret = tracing_set_clock(&global_trace, trace_boot_clock);
8436 pr_warn("Trace clock %s not defined, going back to default\n",
8441 * register_tracer() might reference current_trace, so it
8442 * needs to be set before we register anything. This is
8443 * just a bootstrap of current_trace anyway.
8445 global_trace.current_trace = &nop_trace;
8447 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
8449 ftrace_init_global_array_ops(&global_trace);
8451 init_trace_flags_index(&global_trace);
8453 register_tracer(&nop_trace);
8455 /* Function tracing may start here (via kernel command line) */
8456 init_function_trace();
8458 /* All seems OK, enable tracing */
8459 tracing_disabled = 0;
8461 atomic_notifier_chain_register(&panic_notifier_list,
8462 &trace_panic_notifier);
8464 register_die_notifier(&trace_die_notifier);
8466 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
8468 INIT_LIST_HEAD(&global_trace.systems);
8469 INIT_LIST_HEAD(&global_trace.events);
8470 list_add(&global_trace.list, &ftrace_trace_arrays);
8472 apply_trace_boot_options();
8474 register_snapshot_cmd();
8479 free_saved_cmdlines_buffer(savedcmd);
8480 out_free_temp_buffer:
8481 ring_buffer_free(temp_buffer);
8483 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
8485 free_cpumask_var(global_trace.tracing_cpumask);
8486 out_free_buffer_mask:
8487 free_cpumask_var(tracing_buffer_mask);
8492 void __init early_trace_init(void)
8494 if (tracepoint_printk) {
8495 tracepoint_print_iter =
8496 kmalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
8497 if (WARN_ON(!tracepoint_print_iter))
8498 tracepoint_printk = 0;
8500 static_key_enable(&tracepoint_printk_key.key);
8502 tracer_alloc_buffers();
8505 void __init trace_init(void)
8510 __init static int clear_boot_tracer(void)
8513 * The default tracer at boot buffer is an init section.
8514 * This function is called in lateinit. If we did not
8515 * find the boot tracer, then clear it out, to prevent
8516 * later registration from accessing the buffer that is
8517 * about to be freed.
8519 if (!default_bootup_tracer)
8522 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
8523 default_bootup_tracer);
8524 default_bootup_tracer = NULL;
8529 fs_initcall(tracer_init_tracefs);
8530 late_initcall_sync(clear_boot_tracer);