1 // SPDX-License-Identifier: GPL-2.0
3 * ring buffer based function tracer
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
15 #include <linux/ring_buffer.h>
16 #include <generated/utsrelease.h>
17 #include <linux/stacktrace.h>
18 #include <linux/writeback.h>
19 #include <linux/kallsyms.h>
20 #include <linux/security.h>
21 #include <linux/seq_file.h>
22 #include <linux/notifier.h>
23 #include <linux/irqflags.h>
24 #include <linux/debugfs.h>
25 #include <linux/tracefs.h>
26 #include <linux/pagemap.h>
27 #include <linux/hardirq.h>
28 #include <linux/linkage.h>
29 #include <linux/uaccess.h>
30 #include <linux/vmalloc.h>
31 #include <linux/ftrace.h>
32 #include <linux/module.h>
33 #include <linux/percpu.h>
34 #include <linux/splice.h>
35 #include <linux/kdebug.h>
36 #include <linux/string.h>
37 #include <linux/mount.h>
38 #include <linux/rwsem.h>
39 #include <linux/slab.h>
40 #include <linux/ctype.h>
41 #include <linux/init.h>
42 #include <linux/poll.h>
43 #include <linux/nmi.h>
45 #include <linux/trace.h>
46 #include <linux/sched/clock.h>
47 #include <linux/sched/rt.h>
48 #include <linux/fsnotify.h>
49 #include <linux/irq_work.h>
50 #include <linux/workqueue.h>
53 #include "trace_output.h"
56 * On boot up, the ring buffer is set to the minimum size, so that
57 * we do not waste memory on systems that are not using tracing.
59 bool ring_buffer_expanded;
62 * We need to change this state when a selftest is running.
63 * A selftest will lurk into the ring-buffer to count the
64 * entries inserted during the selftest although some concurrent
65 * insertions into the ring-buffer such as trace_printk could occurred
66 * at the same time, giving false positive or negative results.
68 static bool __read_mostly tracing_selftest_running;
71 * If boot-time tracing including tracers/events via kernel cmdline
72 * is running, we do not want to run SELFTEST.
74 bool __read_mostly tracing_selftest_disabled;
76 #ifdef CONFIG_FTRACE_STARTUP_TEST
77 void __init disable_tracing_selftest(const char *reason)
79 if (!tracing_selftest_disabled) {
80 tracing_selftest_disabled = true;
81 pr_info("Ftrace startup test is disabled due to %s\n", reason);
86 /* Pipe tracepoints to printk */
87 struct trace_iterator *tracepoint_print_iter;
88 int tracepoint_printk;
89 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
91 /* For tracers that don't implement custom flags */
92 static struct tracer_opt dummy_tracer_opt[] = {
97 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
103 * To prevent the comm cache from being overwritten when no
104 * tracing is active, only save the comm when a trace event
107 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
110 * Kill all tracing for good (never come back).
111 * It is initialized to 1 but will turn to zero if the initialization
112 * of the tracer is successful. But that is the only place that sets
115 static int tracing_disabled = 1;
117 cpumask_var_t __read_mostly tracing_buffer_mask;
120 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
122 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
123 * is set, then ftrace_dump is called. This will output the contents
124 * of the ftrace buffers to the console. This is very useful for
125 * capturing traces that lead to crashes and outputing it to a
128 * It is default off, but you can enable it with either specifying
129 * "ftrace_dump_on_oops" in the kernel command line, or setting
130 * /proc/sys/kernel/ftrace_dump_on_oops
131 * Set 1 if you want to dump buffers of all CPUs
132 * Set 2 if you want to dump the buffer of the CPU that triggered oops
135 enum ftrace_dump_mode ftrace_dump_on_oops;
137 /* When set, tracing will stop when a WARN*() is hit */
138 int __disable_trace_on_warning;
140 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
141 /* Map of enums to their values, for "eval_map" file */
142 struct trace_eval_map_head {
144 unsigned long length;
147 union trace_eval_map_item;
149 struct trace_eval_map_tail {
151 * "end" is first and points to NULL as it must be different
152 * than "mod" or "eval_string"
154 union trace_eval_map_item *next;
155 const char *end; /* points to NULL */
158 static DEFINE_MUTEX(trace_eval_mutex);
161 * The trace_eval_maps are saved in an array with two extra elements,
162 * one at the beginning, and one at the end. The beginning item contains
163 * the count of the saved maps (head.length), and the module they
164 * belong to if not built in (head.mod). The ending item contains a
165 * pointer to the next array of saved eval_map items.
167 union trace_eval_map_item {
168 struct trace_eval_map map;
169 struct trace_eval_map_head head;
170 struct trace_eval_map_tail tail;
173 static union trace_eval_map_item *trace_eval_maps;
174 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
176 int tracing_set_tracer(struct trace_array *tr, const char *buf);
177 static void ftrace_trace_userstack(struct trace_array *tr,
178 struct trace_buffer *buffer,
179 unsigned int trace_ctx);
181 #define MAX_TRACER_SIZE 100
182 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
183 static char *default_bootup_tracer;
185 static bool allocate_snapshot;
187 static int __init set_cmdline_ftrace(char *str)
189 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
190 default_bootup_tracer = bootup_tracer_buf;
191 /* We are using ftrace early, expand it */
192 ring_buffer_expanded = true;
195 __setup("ftrace=", set_cmdline_ftrace);
197 static int __init set_ftrace_dump_on_oops(char *str)
199 if (*str++ != '=' || !*str) {
200 ftrace_dump_on_oops = DUMP_ALL;
204 if (!strcmp("orig_cpu", str)) {
205 ftrace_dump_on_oops = DUMP_ORIG;
211 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
213 static int __init stop_trace_on_warning(char *str)
215 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
216 __disable_trace_on_warning = 1;
219 __setup("traceoff_on_warning", stop_trace_on_warning);
221 static int __init boot_alloc_snapshot(char *str)
223 allocate_snapshot = true;
224 /* We also need the main ring buffer expanded */
225 ring_buffer_expanded = true;
228 __setup("alloc_snapshot", boot_alloc_snapshot);
231 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
233 static int __init set_trace_boot_options(char *str)
235 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
238 __setup("trace_options=", set_trace_boot_options);
240 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
241 static char *trace_boot_clock __initdata;
243 static int __init set_trace_boot_clock(char *str)
245 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
246 trace_boot_clock = trace_boot_clock_buf;
249 __setup("trace_clock=", set_trace_boot_clock);
251 static int __init set_tracepoint_printk(char *str)
253 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
254 tracepoint_printk = 1;
257 __setup("tp_printk", set_tracepoint_printk);
259 unsigned long long ns2usecs(u64 nsec)
267 trace_process_export(struct trace_export *export,
268 struct ring_buffer_event *event, int flag)
270 struct trace_entry *entry;
271 unsigned int size = 0;
273 if (export->flags & flag) {
274 entry = ring_buffer_event_data(event);
275 size = ring_buffer_event_length(event);
276 export->write(export, entry, size);
280 static DEFINE_MUTEX(ftrace_export_lock);
282 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
284 static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
285 static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
286 static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
288 static inline void ftrace_exports_enable(struct trace_export *export)
290 if (export->flags & TRACE_EXPORT_FUNCTION)
291 static_branch_inc(&trace_function_exports_enabled);
293 if (export->flags & TRACE_EXPORT_EVENT)
294 static_branch_inc(&trace_event_exports_enabled);
296 if (export->flags & TRACE_EXPORT_MARKER)
297 static_branch_inc(&trace_marker_exports_enabled);
300 static inline void ftrace_exports_disable(struct trace_export *export)
302 if (export->flags & TRACE_EXPORT_FUNCTION)
303 static_branch_dec(&trace_function_exports_enabled);
305 if (export->flags & TRACE_EXPORT_EVENT)
306 static_branch_dec(&trace_event_exports_enabled);
308 if (export->flags & TRACE_EXPORT_MARKER)
309 static_branch_dec(&trace_marker_exports_enabled);
312 static void ftrace_exports(struct ring_buffer_event *event, int flag)
314 struct trace_export *export;
316 preempt_disable_notrace();
318 export = rcu_dereference_raw_check(ftrace_exports_list);
320 trace_process_export(export, event, flag);
321 export = rcu_dereference_raw_check(export->next);
324 preempt_enable_notrace();
328 add_trace_export(struct trace_export **list, struct trace_export *export)
330 rcu_assign_pointer(export->next, *list);
332 * We are entering export into the list but another
333 * CPU might be walking that list. We need to make sure
334 * the export->next pointer is valid before another CPU sees
335 * the export pointer included into the list.
337 rcu_assign_pointer(*list, export);
341 rm_trace_export(struct trace_export **list, struct trace_export *export)
343 struct trace_export **p;
345 for (p = list; *p != NULL; p = &(*p)->next)
352 rcu_assign_pointer(*p, (*p)->next);
358 add_ftrace_export(struct trace_export **list, struct trace_export *export)
360 ftrace_exports_enable(export);
362 add_trace_export(list, export);
366 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
370 ret = rm_trace_export(list, export);
371 ftrace_exports_disable(export);
376 int register_ftrace_export(struct trace_export *export)
378 if (WARN_ON_ONCE(!export->write))
381 mutex_lock(&ftrace_export_lock);
383 add_ftrace_export(&ftrace_exports_list, export);
385 mutex_unlock(&ftrace_export_lock);
389 EXPORT_SYMBOL_GPL(register_ftrace_export);
391 int unregister_ftrace_export(struct trace_export *export)
395 mutex_lock(&ftrace_export_lock);
397 ret = rm_ftrace_export(&ftrace_exports_list, export);
399 mutex_unlock(&ftrace_export_lock);
403 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
405 /* trace_flags holds trace_options default values */
406 #define TRACE_DEFAULT_FLAGS \
407 (FUNCTION_DEFAULT_FLAGS | \
408 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
409 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
410 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
411 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
414 /* trace_options that are only supported by global_trace */
415 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
416 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
418 /* trace_flags that are default zero for instances */
419 #define ZEROED_TRACE_FLAGS \
420 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
423 * The global_trace is the descriptor that holds the top-level tracing
424 * buffers for the live tracing.
426 static struct trace_array global_trace = {
427 .trace_flags = TRACE_DEFAULT_FLAGS,
430 LIST_HEAD(ftrace_trace_arrays);
432 int trace_array_get(struct trace_array *this_tr)
434 struct trace_array *tr;
437 mutex_lock(&trace_types_lock);
438 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
445 mutex_unlock(&trace_types_lock);
450 static void __trace_array_put(struct trace_array *this_tr)
452 WARN_ON(!this_tr->ref);
457 * trace_array_put - Decrement the reference counter for this trace array.
458 * @this_tr : pointer to the trace array
460 * NOTE: Use this when we no longer need the trace array returned by
461 * trace_array_get_by_name(). This ensures the trace array can be later
465 void trace_array_put(struct trace_array *this_tr)
470 mutex_lock(&trace_types_lock);
471 __trace_array_put(this_tr);
472 mutex_unlock(&trace_types_lock);
474 EXPORT_SYMBOL_GPL(trace_array_put);
476 int tracing_check_open_get_tr(struct trace_array *tr)
480 ret = security_locked_down(LOCKDOWN_TRACEFS);
484 if (tracing_disabled)
487 if (tr && trace_array_get(tr) < 0)
493 int call_filter_check_discard(struct trace_event_call *call, void *rec,
494 struct trace_buffer *buffer,
495 struct ring_buffer_event *event)
497 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
498 !filter_match_preds(call->filter, rec)) {
499 __trace_event_discard_commit(buffer, event);
506 void trace_free_pid_list(struct trace_pid_list *pid_list)
508 vfree(pid_list->pids);
513 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
514 * @filtered_pids: The list of pids to check
515 * @search_pid: The PID to find in @filtered_pids
517 * Returns true if @search_pid is fonud in @filtered_pids, and false otherwis.
520 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
523 * If pid_max changed after filtered_pids was created, we
524 * by default ignore all pids greater than the previous pid_max.
526 if (search_pid >= filtered_pids->pid_max)
529 return test_bit(search_pid, filtered_pids->pids);
533 * trace_ignore_this_task - should a task be ignored for tracing
534 * @filtered_pids: The list of pids to check
535 * @filtered_no_pids: The list of pids not to be traced
536 * @task: The task that should be ignored if not filtered
538 * Checks if @task should be traced or not from @filtered_pids.
539 * Returns true if @task should *NOT* be traced.
540 * Returns false if @task should be traced.
543 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
544 struct trace_pid_list *filtered_no_pids,
545 struct task_struct *task)
548 * If filterd_no_pids is not empty, and the task's pid is listed
549 * in filtered_no_pids, then return true.
550 * Otherwise, if filtered_pids is empty, that means we can
551 * trace all tasks. If it has content, then only trace pids
552 * within filtered_pids.
555 return (filtered_pids &&
556 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
558 trace_find_filtered_pid(filtered_no_pids, task->pid));
562 * trace_filter_add_remove_task - Add or remove a task from a pid_list
563 * @pid_list: The list to modify
564 * @self: The current task for fork or NULL for exit
565 * @task: The task to add or remove
567 * If adding a task, if @self is defined, the task is only added if @self
568 * is also included in @pid_list. This happens on fork and tasks should
569 * only be added when the parent is listed. If @self is NULL, then the
570 * @task pid will be removed from the list, which would happen on exit
573 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
574 struct task_struct *self,
575 struct task_struct *task)
580 /* For forks, we only add if the forking task is listed */
582 if (!trace_find_filtered_pid(pid_list, self->pid))
586 /* Sorry, but we don't support pid_max changing after setting */
587 if (task->pid >= pid_list->pid_max)
590 /* "self" is set for forks, and NULL for exits */
592 set_bit(task->pid, pid_list->pids);
594 clear_bit(task->pid, pid_list->pids);
598 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
599 * @pid_list: The pid list to show
600 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
601 * @pos: The position of the file
603 * This is used by the seq_file "next" operation to iterate the pids
604 * listed in a trace_pid_list structure.
606 * Returns the pid+1 as we want to display pid of zero, but NULL would
607 * stop the iteration.
609 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
611 unsigned long pid = (unsigned long)v;
615 /* pid already is +1 of the actual prevous bit */
616 pid = find_next_bit(pid_list->pids, pid_list->pid_max, pid);
618 /* Return pid + 1 to allow zero to be represented */
619 if (pid < pid_list->pid_max)
620 return (void *)(pid + 1);
626 * trace_pid_start - Used for seq_file to start reading pid lists
627 * @pid_list: The pid list to show
628 * @pos: The position of the file
630 * This is used by seq_file "start" operation to start the iteration
633 * Returns the pid+1 as we want to display pid of zero, but NULL would
634 * stop the iteration.
636 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
641 pid = find_first_bit(pid_list->pids, pid_list->pid_max);
642 if (pid >= pid_list->pid_max)
645 /* Return pid + 1 so that zero can be the exit value */
646 for (pid++; pid && l < *pos;
647 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
653 * trace_pid_show - show the current pid in seq_file processing
654 * @m: The seq_file structure to write into
655 * @v: A void pointer of the pid (+1) value to display
657 * Can be directly used by seq_file operations to display the current
660 int trace_pid_show(struct seq_file *m, void *v)
662 unsigned long pid = (unsigned long)v - 1;
664 seq_printf(m, "%lu\n", pid);
668 /* 128 should be much more than enough */
669 #define PID_BUF_SIZE 127
671 int trace_pid_write(struct trace_pid_list *filtered_pids,
672 struct trace_pid_list **new_pid_list,
673 const char __user *ubuf, size_t cnt)
675 struct trace_pid_list *pid_list;
676 struct trace_parser parser;
684 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
688 * Always recreate a new array. The write is an all or nothing
689 * operation. Always create a new array when adding new pids by
690 * the user. If the operation fails, then the current list is
693 pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
695 trace_parser_put(&parser);
699 pid_list->pid_max = READ_ONCE(pid_max);
701 /* Only truncating will shrink pid_max */
702 if (filtered_pids && filtered_pids->pid_max > pid_list->pid_max)
703 pid_list->pid_max = filtered_pids->pid_max;
705 pid_list->pids = vzalloc((pid_list->pid_max + 7) >> 3);
706 if (!pid_list->pids) {
707 trace_parser_put(&parser);
713 /* copy the current bits to the new max */
714 for_each_set_bit(pid, filtered_pids->pids,
715 filtered_pids->pid_max) {
716 set_bit(pid, pid_list->pids);
725 ret = trace_get_user(&parser, ubuf, cnt, &pos);
726 if (ret < 0 || !trace_parser_loaded(&parser))
734 if (kstrtoul(parser.buffer, 0, &val))
736 if (val >= pid_list->pid_max)
741 set_bit(pid, pid_list->pids);
744 trace_parser_clear(&parser);
747 trace_parser_put(&parser);
750 trace_free_pid_list(pid_list);
755 /* Cleared the list of pids */
756 trace_free_pid_list(pid_list);
761 *new_pid_list = pid_list;
766 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
770 /* Early boot up does not have a buffer yet */
772 return trace_clock_local();
774 ts = ring_buffer_time_stamp(buf->buffer, cpu);
775 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
780 u64 ftrace_now(int cpu)
782 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
786 * tracing_is_enabled - Show if global_trace has been enabled
788 * Shows if the global trace has been enabled or not. It uses the
789 * mirror flag "buffer_disabled" to be used in fast paths such as for
790 * the irqsoff tracer. But it may be inaccurate due to races. If you
791 * need to know the accurate state, use tracing_is_on() which is a little
792 * slower, but accurate.
794 int tracing_is_enabled(void)
797 * For quick access (irqsoff uses this in fast path), just
798 * return the mirror variable of the state of the ring buffer.
799 * It's a little racy, but we don't really care.
802 return !global_trace.buffer_disabled;
806 * trace_buf_size is the size in bytes that is allocated
807 * for a buffer. Note, the number of bytes is always rounded
810 * This number is purposely set to a low number of 16384.
811 * If the dump on oops happens, it will be much appreciated
812 * to not have to wait for all that output. Anyway this can be
813 * boot time and run time configurable.
815 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
817 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
819 /* trace_types holds a link list of available tracers. */
820 static struct tracer *trace_types __read_mostly;
823 * trace_types_lock is used to protect the trace_types list.
825 DEFINE_MUTEX(trace_types_lock);
828 * serialize the access of the ring buffer
830 * ring buffer serializes readers, but it is low level protection.
831 * The validity of the events (which returns by ring_buffer_peek() ..etc)
832 * are not protected by ring buffer.
834 * The content of events may become garbage if we allow other process consumes
835 * these events concurrently:
836 * A) the page of the consumed events may become a normal page
837 * (not reader page) in ring buffer, and this page will be rewrited
838 * by events producer.
839 * B) The page of the consumed events may become a page for splice_read,
840 * and this page will be returned to system.
842 * These primitives allow multi process access to different cpu ring buffer
845 * These primitives don't distinguish read-only and read-consume access.
846 * Multi read-only access are also serialized.
850 static DECLARE_RWSEM(all_cpu_access_lock);
851 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
853 static inline void trace_access_lock(int cpu)
855 if (cpu == RING_BUFFER_ALL_CPUS) {
856 /* gain it for accessing the whole ring buffer. */
857 down_write(&all_cpu_access_lock);
859 /* gain it for accessing a cpu ring buffer. */
861 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
862 down_read(&all_cpu_access_lock);
864 /* Secondly block other access to this @cpu ring buffer. */
865 mutex_lock(&per_cpu(cpu_access_lock, cpu));
869 static inline void trace_access_unlock(int cpu)
871 if (cpu == RING_BUFFER_ALL_CPUS) {
872 up_write(&all_cpu_access_lock);
874 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
875 up_read(&all_cpu_access_lock);
879 static inline void trace_access_lock_init(void)
883 for_each_possible_cpu(cpu)
884 mutex_init(&per_cpu(cpu_access_lock, cpu));
889 static DEFINE_MUTEX(access_lock);
891 static inline void trace_access_lock(int cpu)
894 mutex_lock(&access_lock);
897 static inline void trace_access_unlock(int cpu)
900 mutex_unlock(&access_lock);
903 static inline void trace_access_lock_init(void)
909 #ifdef CONFIG_STACKTRACE
910 static void __ftrace_trace_stack(struct trace_buffer *buffer,
911 unsigned int trace_ctx,
912 int skip, struct pt_regs *regs);
913 static inline void ftrace_trace_stack(struct trace_array *tr,
914 struct trace_buffer *buffer,
915 unsigned int trace_ctx,
916 int skip, struct pt_regs *regs);
919 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
920 unsigned int trace_ctx,
921 int skip, struct pt_regs *regs)
924 static inline void ftrace_trace_stack(struct trace_array *tr,
925 struct trace_buffer *buffer,
926 unsigned long trace_ctx,
927 int skip, struct pt_regs *regs)
933 static __always_inline void
934 trace_event_setup(struct ring_buffer_event *event,
935 int type, unsigned int trace_ctx)
937 struct trace_entry *ent = ring_buffer_event_data(event);
939 tracing_generic_entry_update(ent, type, trace_ctx);
942 static __always_inline struct ring_buffer_event *
943 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
946 unsigned int trace_ctx)
948 struct ring_buffer_event *event;
950 event = ring_buffer_lock_reserve(buffer, len);
952 trace_event_setup(event, type, trace_ctx);
957 void tracer_tracing_on(struct trace_array *tr)
959 if (tr->array_buffer.buffer)
960 ring_buffer_record_on(tr->array_buffer.buffer);
962 * This flag is looked at when buffers haven't been allocated
963 * yet, or by some tracers (like irqsoff), that just want to
964 * know if the ring buffer has been disabled, but it can handle
965 * races of where it gets disabled but we still do a record.
966 * As the check is in the fast path of the tracers, it is more
967 * important to be fast than accurate.
969 tr->buffer_disabled = 0;
970 /* Make the flag seen by readers */
975 * tracing_on - enable tracing buffers
977 * This function enables tracing buffers that may have been
978 * disabled with tracing_off.
980 void tracing_on(void)
982 tracer_tracing_on(&global_trace);
984 EXPORT_SYMBOL_GPL(tracing_on);
987 static __always_inline void
988 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
990 __this_cpu_write(trace_taskinfo_save, true);
992 /* If this is the temp buffer, we need to commit fully */
993 if (this_cpu_read(trace_buffered_event) == event) {
994 /* Length is in event->array[0] */
995 ring_buffer_write(buffer, event->array[0], &event->array[1]);
996 /* Release the temp buffer */
997 this_cpu_dec(trace_buffered_event_cnt);
999 ring_buffer_unlock_commit(buffer, event);
1003 * __trace_puts - write a constant string into the trace buffer.
1004 * @ip: The address of the caller
1005 * @str: The constant string to write
1006 * @size: The size of the string.
1008 int __trace_puts(unsigned long ip, const char *str, int size)
1010 struct ring_buffer_event *event;
1011 struct trace_buffer *buffer;
1012 struct print_entry *entry;
1013 unsigned int trace_ctx;
1016 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1019 if (unlikely(tracing_selftest_running || tracing_disabled))
1022 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1024 trace_ctx = tracing_gen_ctx();
1025 buffer = global_trace.array_buffer.buffer;
1026 ring_buffer_nest_start(buffer);
1027 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1034 entry = ring_buffer_event_data(event);
1037 memcpy(&entry->buf, str, size);
1039 /* Add a newline if necessary */
1040 if (entry->buf[size - 1] != '\n') {
1041 entry->buf[size] = '\n';
1042 entry->buf[size + 1] = '\0';
1044 entry->buf[size] = '\0';
1046 __buffer_unlock_commit(buffer, event);
1047 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1049 ring_buffer_nest_end(buffer);
1052 EXPORT_SYMBOL_GPL(__trace_puts);
1055 * __trace_bputs - write the pointer to a constant string into trace buffer
1056 * @ip: The address of the caller
1057 * @str: The constant string to write to the buffer to
1059 int __trace_bputs(unsigned long ip, const char *str)
1061 struct ring_buffer_event *event;
1062 struct trace_buffer *buffer;
1063 struct bputs_entry *entry;
1064 unsigned int trace_ctx;
1065 int size = sizeof(struct bputs_entry);
1068 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1071 if (unlikely(tracing_selftest_running || tracing_disabled))
1074 trace_ctx = tracing_gen_ctx();
1075 buffer = global_trace.array_buffer.buffer;
1077 ring_buffer_nest_start(buffer);
1078 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1083 entry = ring_buffer_event_data(event);
1087 __buffer_unlock_commit(buffer, event);
1088 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1092 ring_buffer_nest_end(buffer);
1095 EXPORT_SYMBOL_GPL(__trace_bputs);
1097 #ifdef CONFIG_TRACER_SNAPSHOT
1098 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1101 struct tracer *tracer = tr->current_trace;
1102 unsigned long flags;
1105 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1106 internal_trace_puts("*** snapshot is being ignored ***\n");
1110 if (!tr->allocated_snapshot) {
1111 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
1112 internal_trace_puts("*** stopping trace here! ***\n");
1117 /* Note, snapshot can not be used when the tracer uses it */
1118 if (tracer->use_max_tr) {
1119 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
1120 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
1124 local_irq_save(flags);
1125 update_max_tr(tr, current, smp_processor_id(), cond_data);
1126 local_irq_restore(flags);
1129 void tracing_snapshot_instance(struct trace_array *tr)
1131 tracing_snapshot_instance_cond(tr, NULL);
1135 * tracing_snapshot - take a snapshot of the current buffer.
1137 * This causes a swap between the snapshot buffer and the current live
1138 * tracing buffer. You can use this to take snapshots of the live
1139 * trace when some condition is triggered, but continue to trace.
1141 * Note, make sure to allocate the snapshot with either
1142 * a tracing_snapshot_alloc(), or by doing it manually
1143 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
1145 * If the snapshot buffer is not allocated, it will stop tracing.
1146 * Basically making a permanent snapshot.
1148 void tracing_snapshot(void)
1150 struct trace_array *tr = &global_trace;
1152 tracing_snapshot_instance(tr);
1154 EXPORT_SYMBOL_GPL(tracing_snapshot);
1157 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1158 * @tr: The tracing instance to snapshot
1159 * @cond_data: The data to be tested conditionally, and possibly saved
1161 * This is the same as tracing_snapshot() except that the snapshot is
1162 * conditional - the snapshot will only happen if the
1163 * cond_snapshot.update() implementation receiving the cond_data
1164 * returns true, which means that the trace array's cond_snapshot
1165 * update() operation used the cond_data to determine whether the
1166 * snapshot should be taken, and if it was, presumably saved it along
1167 * with the snapshot.
1169 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1171 tracing_snapshot_instance_cond(tr, cond_data);
1173 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1176 * tracing_snapshot_cond_data - get the user data associated with a snapshot
1177 * @tr: The tracing instance
1179 * When the user enables a conditional snapshot using
1180 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1181 * with the snapshot. This accessor is used to retrieve it.
1183 * Should not be called from cond_snapshot.update(), since it takes
1184 * the tr->max_lock lock, which the code calling
1185 * cond_snapshot.update() has already done.
1187 * Returns the cond_data associated with the trace array's snapshot.
1189 void *tracing_cond_snapshot_data(struct trace_array *tr)
1191 void *cond_data = NULL;
1193 arch_spin_lock(&tr->max_lock);
1195 if (tr->cond_snapshot)
1196 cond_data = tr->cond_snapshot->cond_data;
1198 arch_spin_unlock(&tr->max_lock);
1202 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1204 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1205 struct array_buffer *size_buf, int cpu_id);
1206 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1208 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1212 if (!tr->allocated_snapshot) {
1214 /* allocate spare buffer */
1215 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1216 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1220 tr->allocated_snapshot = true;
1226 static void free_snapshot(struct trace_array *tr)
1229 * We don't free the ring buffer. instead, resize it because
1230 * The max_tr ring buffer has some state (e.g. ring->clock) and
1231 * we want preserve it.
1233 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1234 set_buffer_entries(&tr->max_buffer, 1);
1235 tracing_reset_online_cpus(&tr->max_buffer);
1236 tr->allocated_snapshot = false;
1240 * tracing_alloc_snapshot - allocate snapshot buffer.
1242 * This only allocates the snapshot buffer if it isn't already
1243 * allocated - it doesn't also take a snapshot.
1245 * This is meant to be used in cases where the snapshot buffer needs
1246 * to be set up for events that can't sleep but need to be able to
1247 * trigger a snapshot.
1249 int tracing_alloc_snapshot(void)
1251 struct trace_array *tr = &global_trace;
1254 ret = tracing_alloc_snapshot_instance(tr);
1259 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1262 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1264 * This is similar to tracing_snapshot(), but it will allocate the
1265 * snapshot buffer if it isn't already allocated. Use this only
1266 * where it is safe to sleep, as the allocation may sleep.
1268 * This causes a swap between the snapshot buffer and the current live
1269 * tracing buffer. You can use this to take snapshots of the live
1270 * trace when some condition is triggered, but continue to trace.
1272 void tracing_snapshot_alloc(void)
1276 ret = tracing_alloc_snapshot();
1282 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1285 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1286 * @tr: The tracing instance
1287 * @cond_data: User data to associate with the snapshot
1288 * @update: Implementation of the cond_snapshot update function
1290 * Check whether the conditional snapshot for the given instance has
1291 * already been enabled, or if the current tracer is already using a
1292 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1293 * save the cond_data and update function inside.
1295 * Returns 0 if successful, error otherwise.
1297 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1298 cond_update_fn_t update)
1300 struct cond_snapshot *cond_snapshot;
1303 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1307 cond_snapshot->cond_data = cond_data;
1308 cond_snapshot->update = update;
1310 mutex_lock(&trace_types_lock);
1312 ret = tracing_alloc_snapshot_instance(tr);
1316 if (tr->current_trace->use_max_tr) {
1322 * The cond_snapshot can only change to NULL without the
1323 * trace_types_lock. We don't care if we race with it going
1324 * to NULL, but we want to make sure that it's not set to
1325 * something other than NULL when we get here, which we can
1326 * do safely with only holding the trace_types_lock and not
1327 * having to take the max_lock.
1329 if (tr->cond_snapshot) {
1334 arch_spin_lock(&tr->max_lock);
1335 tr->cond_snapshot = cond_snapshot;
1336 arch_spin_unlock(&tr->max_lock);
1338 mutex_unlock(&trace_types_lock);
1343 mutex_unlock(&trace_types_lock);
1344 kfree(cond_snapshot);
1347 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1350 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1351 * @tr: The tracing instance
1353 * Check whether the conditional snapshot for the given instance is
1354 * enabled; if so, free the cond_snapshot associated with it,
1355 * otherwise return -EINVAL.
1357 * Returns 0 if successful, error otherwise.
1359 int tracing_snapshot_cond_disable(struct trace_array *tr)
1363 arch_spin_lock(&tr->max_lock);
1365 if (!tr->cond_snapshot)
1368 kfree(tr->cond_snapshot);
1369 tr->cond_snapshot = NULL;
1372 arch_spin_unlock(&tr->max_lock);
1376 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1378 void tracing_snapshot(void)
1380 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1382 EXPORT_SYMBOL_GPL(tracing_snapshot);
1383 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1385 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1387 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1388 int tracing_alloc_snapshot(void)
1390 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1393 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1394 void tracing_snapshot_alloc(void)
1399 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1400 void *tracing_cond_snapshot_data(struct trace_array *tr)
1404 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1405 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1409 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1410 int tracing_snapshot_cond_disable(struct trace_array *tr)
1414 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1415 #endif /* CONFIG_TRACER_SNAPSHOT */
1417 void tracer_tracing_off(struct trace_array *tr)
1419 if (tr->array_buffer.buffer)
1420 ring_buffer_record_off(tr->array_buffer.buffer);
1422 * This flag is looked at when buffers haven't been allocated
1423 * yet, or by some tracers (like irqsoff), that just want to
1424 * know if the ring buffer has been disabled, but it can handle
1425 * races of where it gets disabled but we still do a record.
1426 * As the check is in the fast path of the tracers, it is more
1427 * important to be fast than accurate.
1429 tr->buffer_disabled = 1;
1430 /* Make the flag seen by readers */
1435 * tracing_off - turn off tracing buffers
1437 * This function stops the tracing buffers from recording data.
1438 * It does not disable any overhead the tracers themselves may
1439 * be causing. This function simply causes all recording to
1440 * the ring buffers to fail.
1442 void tracing_off(void)
1444 tracer_tracing_off(&global_trace);
1446 EXPORT_SYMBOL_GPL(tracing_off);
1448 void disable_trace_on_warning(void)
1450 if (__disable_trace_on_warning) {
1451 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1452 "Disabling tracing due to warning\n");
1458 * tracer_tracing_is_on - show real state of ring buffer enabled
1459 * @tr : the trace array to know if ring buffer is enabled
1461 * Shows real state of the ring buffer if it is enabled or not.
1463 bool tracer_tracing_is_on(struct trace_array *tr)
1465 if (tr->array_buffer.buffer)
1466 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1467 return !tr->buffer_disabled;
1471 * tracing_is_on - show state of ring buffers enabled
1473 int tracing_is_on(void)
1475 return tracer_tracing_is_on(&global_trace);
1477 EXPORT_SYMBOL_GPL(tracing_is_on);
1479 static int __init set_buf_size(char *str)
1481 unsigned long buf_size;
1485 buf_size = memparse(str, &str);
1486 /* nr_entries can not be zero */
1489 trace_buf_size = buf_size;
1492 __setup("trace_buf_size=", set_buf_size);
1494 static int __init set_tracing_thresh(char *str)
1496 unsigned long threshold;
1501 ret = kstrtoul(str, 0, &threshold);
1504 tracing_thresh = threshold * 1000;
1507 __setup("tracing_thresh=", set_tracing_thresh);
1509 unsigned long nsecs_to_usecs(unsigned long nsecs)
1511 return nsecs / 1000;
1515 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1516 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1517 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1518 * of strings in the order that the evals (enum) were defined.
1523 /* These must match the bit postions in trace_iterator_flags */
1524 static const char *trace_options[] = {
1532 int in_ns; /* is this clock in nanoseconds? */
1533 } trace_clocks[] = {
1534 { trace_clock_local, "local", 1 },
1535 { trace_clock_global, "global", 1 },
1536 { trace_clock_counter, "counter", 0 },
1537 { trace_clock_jiffies, "uptime", 0 },
1538 { trace_clock, "perf", 1 },
1539 { ktime_get_mono_fast_ns, "mono", 1 },
1540 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1541 { ktime_get_boot_fast_ns, "boot", 1 },
1545 bool trace_clock_in_ns(struct trace_array *tr)
1547 if (trace_clocks[tr->clock_id].in_ns)
1554 * trace_parser_get_init - gets the buffer for trace parser
1556 int trace_parser_get_init(struct trace_parser *parser, int size)
1558 memset(parser, 0, sizeof(*parser));
1560 parser->buffer = kmalloc(size, GFP_KERNEL);
1561 if (!parser->buffer)
1564 parser->size = size;
1569 * trace_parser_put - frees the buffer for trace parser
1571 void trace_parser_put(struct trace_parser *parser)
1573 kfree(parser->buffer);
1574 parser->buffer = NULL;
1578 * trace_get_user - reads the user input string separated by space
1579 * (matched by isspace(ch))
1581 * For each string found the 'struct trace_parser' is updated,
1582 * and the function returns.
1584 * Returns number of bytes read.
1586 * See kernel/trace/trace.h for 'struct trace_parser' details.
1588 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1589 size_t cnt, loff_t *ppos)
1596 trace_parser_clear(parser);
1598 ret = get_user(ch, ubuf++);
1606 * The parser is not finished with the last write,
1607 * continue reading the user input without skipping spaces.
1609 if (!parser->cont) {
1610 /* skip white space */
1611 while (cnt && isspace(ch)) {
1612 ret = get_user(ch, ubuf++);
1621 /* only spaces were written */
1622 if (isspace(ch) || !ch) {
1629 /* read the non-space input */
1630 while (cnt && !isspace(ch) && ch) {
1631 if (parser->idx < parser->size - 1)
1632 parser->buffer[parser->idx++] = ch;
1637 ret = get_user(ch, ubuf++);
1644 /* We either got finished input or we have to wait for another call. */
1645 if (isspace(ch) || !ch) {
1646 parser->buffer[parser->idx] = 0;
1647 parser->cont = false;
1648 } else if (parser->idx < parser->size - 1) {
1649 parser->cont = true;
1650 parser->buffer[parser->idx++] = ch;
1651 /* Make sure the parsed string always terminates with '\0'. */
1652 parser->buffer[parser->idx] = 0;
1665 /* TODO add a seq_buf_to_buffer() */
1666 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1670 if (trace_seq_used(s) <= s->seq.readpos)
1673 len = trace_seq_used(s) - s->seq.readpos;
1676 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1678 s->seq.readpos += cnt;
1682 unsigned long __read_mostly tracing_thresh;
1683 static const struct file_operations tracing_max_lat_fops;
1685 #if (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1686 defined(CONFIG_FSNOTIFY)
1688 static struct workqueue_struct *fsnotify_wq;
1690 static void latency_fsnotify_workfn(struct work_struct *work)
1692 struct trace_array *tr = container_of(work, struct trace_array,
1694 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1697 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1699 struct trace_array *tr = container_of(iwork, struct trace_array,
1701 queue_work(fsnotify_wq, &tr->fsnotify_work);
1704 static void trace_create_maxlat_file(struct trace_array *tr,
1705 struct dentry *d_tracer)
1707 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1708 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1709 tr->d_max_latency = trace_create_file("tracing_max_latency", 0644,
1710 d_tracer, &tr->max_latency,
1711 &tracing_max_lat_fops);
1714 __init static int latency_fsnotify_init(void)
1716 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1717 WQ_UNBOUND | WQ_HIGHPRI, 0);
1719 pr_err("Unable to allocate tr_max_lat_wq\n");
1725 late_initcall_sync(latency_fsnotify_init);
1727 void latency_fsnotify(struct trace_array *tr)
1732 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1733 * possible that we are called from __schedule() or do_idle(), which
1734 * could cause a deadlock.
1736 irq_work_queue(&tr->fsnotify_irqwork);
1740 * (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1741 * defined(CONFIG_FSNOTIFY)
1745 #define trace_create_maxlat_file(tr, d_tracer) \
1746 trace_create_file("tracing_max_latency", 0644, d_tracer, \
1747 &tr->max_latency, &tracing_max_lat_fops)
1751 #ifdef CONFIG_TRACER_MAX_TRACE
1753 * Copy the new maximum trace into the separate maximum-trace
1754 * structure. (this way the maximum trace is permanently saved,
1755 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1758 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1760 struct array_buffer *trace_buf = &tr->array_buffer;
1761 struct array_buffer *max_buf = &tr->max_buffer;
1762 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1763 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1766 max_buf->time_start = data->preempt_timestamp;
1768 max_data->saved_latency = tr->max_latency;
1769 max_data->critical_start = data->critical_start;
1770 max_data->critical_end = data->critical_end;
1772 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1773 max_data->pid = tsk->pid;
1775 * If tsk == current, then use current_uid(), as that does not use
1776 * RCU. The irq tracer can be called out of RCU scope.
1779 max_data->uid = current_uid();
1781 max_data->uid = task_uid(tsk);
1783 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1784 max_data->policy = tsk->policy;
1785 max_data->rt_priority = tsk->rt_priority;
1787 /* record this tasks comm */
1788 tracing_record_cmdline(tsk);
1789 latency_fsnotify(tr);
1793 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1795 * @tsk: the task with the latency
1796 * @cpu: The cpu that initiated the trace.
1797 * @cond_data: User data associated with a conditional snapshot
1799 * Flip the buffers between the @tr and the max_tr and record information
1800 * about which task was the cause of this latency.
1803 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1809 WARN_ON_ONCE(!irqs_disabled());
1811 if (!tr->allocated_snapshot) {
1812 /* Only the nop tracer should hit this when disabling */
1813 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1817 arch_spin_lock(&tr->max_lock);
1819 /* Inherit the recordable setting from array_buffer */
1820 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1821 ring_buffer_record_on(tr->max_buffer.buffer);
1823 ring_buffer_record_off(tr->max_buffer.buffer);
1825 #ifdef CONFIG_TRACER_SNAPSHOT
1826 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data))
1829 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1831 __update_max_tr(tr, tsk, cpu);
1834 arch_spin_unlock(&tr->max_lock);
1838 * update_max_tr_single - only copy one trace over, and reset the rest
1840 * @tsk: task with the latency
1841 * @cpu: the cpu of the buffer to copy.
1843 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1846 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1853 WARN_ON_ONCE(!irqs_disabled());
1854 if (!tr->allocated_snapshot) {
1855 /* Only the nop tracer should hit this when disabling */
1856 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1860 arch_spin_lock(&tr->max_lock);
1862 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1864 if (ret == -EBUSY) {
1866 * We failed to swap the buffer due to a commit taking
1867 * place on this CPU. We fail to record, but we reset
1868 * the max trace buffer (no one writes directly to it)
1869 * and flag that it failed.
1871 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1872 "Failed to swap buffers due to commit in progress\n");
1875 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1877 __update_max_tr(tr, tsk, cpu);
1878 arch_spin_unlock(&tr->max_lock);
1880 #endif /* CONFIG_TRACER_MAX_TRACE */
1882 static int wait_on_pipe(struct trace_iterator *iter, int full)
1884 /* Iterators are static, they should be filled or empty */
1885 if (trace_buffer_iter(iter, iter->cpu_file))
1888 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1892 #ifdef CONFIG_FTRACE_STARTUP_TEST
1893 static bool selftests_can_run;
1895 struct trace_selftests {
1896 struct list_head list;
1897 struct tracer *type;
1900 static LIST_HEAD(postponed_selftests);
1902 static int save_selftest(struct tracer *type)
1904 struct trace_selftests *selftest;
1906 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1910 selftest->type = type;
1911 list_add(&selftest->list, &postponed_selftests);
1915 static int run_tracer_selftest(struct tracer *type)
1917 struct trace_array *tr = &global_trace;
1918 struct tracer *saved_tracer = tr->current_trace;
1921 if (!type->selftest || tracing_selftest_disabled)
1925 * If a tracer registers early in boot up (before scheduling is
1926 * initialized and such), then do not run its selftests yet.
1927 * Instead, run it a little later in the boot process.
1929 if (!selftests_can_run)
1930 return save_selftest(type);
1932 if (!tracing_is_on()) {
1933 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
1939 * Run a selftest on this tracer.
1940 * Here we reset the trace buffer, and set the current
1941 * tracer to be this tracer. The tracer can then run some
1942 * internal tracing to verify that everything is in order.
1943 * If we fail, we do not register this tracer.
1945 tracing_reset_online_cpus(&tr->array_buffer);
1947 tr->current_trace = type;
1949 #ifdef CONFIG_TRACER_MAX_TRACE
1950 if (type->use_max_tr) {
1951 /* If we expanded the buffers, make sure the max is expanded too */
1952 if (ring_buffer_expanded)
1953 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1954 RING_BUFFER_ALL_CPUS);
1955 tr->allocated_snapshot = true;
1959 /* the test is responsible for initializing and enabling */
1960 pr_info("Testing tracer %s: ", type->name);
1961 ret = type->selftest(type, tr);
1962 /* the test is responsible for resetting too */
1963 tr->current_trace = saved_tracer;
1965 printk(KERN_CONT "FAILED!\n");
1966 /* Add the warning after printing 'FAILED' */
1970 /* Only reset on passing, to avoid touching corrupted buffers */
1971 tracing_reset_online_cpus(&tr->array_buffer);
1973 #ifdef CONFIG_TRACER_MAX_TRACE
1974 if (type->use_max_tr) {
1975 tr->allocated_snapshot = false;
1977 /* Shrink the max buffer again */
1978 if (ring_buffer_expanded)
1979 ring_buffer_resize(tr->max_buffer.buffer, 1,
1980 RING_BUFFER_ALL_CPUS);
1984 printk(KERN_CONT "PASSED\n");
1988 static __init int init_trace_selftests(void)
1990 struct trace_selftests *p, *n;
1991 struct tracer *t, **last;
1994 selftests_can_run = true;
1996 mutex_lock(&trace_types_lock);
1998 if (list_empty(&postponed_selftests))
2001 pr_info("Running postponed tracer tests:\n");
2003 tracing_selftest_running = true;
2004 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2005 /* This loop can take minutes when sanitizers are enabled, so
2006 * lets make sure we allow RCU processing.
2009 ret = run_tracer_selftest(p->type);
2010 /* If the test fails, then warn and remove from available_tracers */
2012 WARN(1, "tracer: %s failed selftest, disabling\n",
2014 last = &trace_types;
2015 for (t = trace_types; t; t = t->next) {
2026 tracing_selftest_running = false;
2029 mutex_unlock(&trace_types_lock);
2033 core_initcall(init_trace_selftests);
2035 static inline int run_tracer_selftest(struct tracer *type)
2039 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2041 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2043 static void __init apply_trace_boot_options(void);
2046 * register_tracer - register a tracer with the ftrace system.
2047 * @type: the plugin for the tracer
2049 * Register a new plugin tracer.
2051 int __init register_tracer(struct tracer *type)
2057 pr_info("Tracer must have a name\n");
2061 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2062 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2066 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2067 pr_warn("Can not register tracer %s due to lockdown\n",
2072 mutex_lock(&trace_types_lock);
2074 tracing_selftest_running = true;
2076 for (t = trace_types; t; t = t->next) {
2077 if (strcmp(type->name, t->name) == 0) {
2079 pr_info("Tracer %s already registered\n",
2086 if (!type->set_flag)
2087 type->set_flag = &dummy_set_flag;
2089 /*allocate a dummy tracer_flags*/
2090 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2095 type->flags->val = 0;
2096 type->flags->opts = dummy_tracer_opt;
2098 if (!type->flags->opts)
2099 type->flags->opts = dummy_tracer_opt;
2101 /* store the tracer for __set_tracer_option */
2102 type->flags->trace = type;
2104 ret = run_tracer_selftest(type);
2108 type->next = trace_types;
2110 add_tracer_options(&global_trace, type);
2113 tracing_selftest_running = false;
2114 mutex_unlock(&trace_types_lock);
2116 if (ret || !default_bootup_tracer)
2119 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2122 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2123 /* Do we want this tracer to start on bootup? */
2124 tracing_set_tracer(&global_trace, type->name);
2125 default_bootup_tracer = NULL;
2127 apply_trace_boot_options();
2129 /* disable other selftests, since this will break it. */
2130 disable_tracing_selftest("running a tracer");
2136 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2138 struct trace_buffer *buffer = buf->buffer;
2143 ring_buffer_record_disable(buffer);
2145 /* Make sure all commits have finished */
2147 ring_buffer_reset_cpu(buffer, cpu);
2149 ring_buffer_record_enable(buffer);
2152 void tracing_reset_online_cpus(struct array_buffer *buf)
2154 struct trace_buffer *buffer = buf->buffer;
2159 ring_buffer_record_disable(buffer);
2161 /* Make sure all commits have finished */
2164 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2166 ring_buffer_reset_online_cpus(buffer);
2168 ring_buffer_record_enable(buffer);
2171 /* Must have trace_types_lock held */
2172 void tracing_reset_all_online_cpus(void)
2174 struct trace_array *tr;
2176 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2177 if (!tr->clear_trace)
2179 tr->clear_trace = false;
2180 tracing_reset_online_cpus(&tr->array_buffer);
2181 #ifdef CONFIG_TRACER_MAX_TRACE
2182 tracing_reset_online_cpus(&tr->max_buffer);
2187 static int *tgid_map;
2189 #define SAVED_CMDLINES_DEFAULT 128
2190 #define NO_CMDLINE_MAP UINT_MAX
2191 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2192 struct saved_cmdlines_buffer {
2193 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2194 unsigned *map_cmdline_to_pid;
2195 unsigned cmdline_num;
2197 char *saved_cmdlines;
2199 static struct saved_cmdlines_buffer *savedcmd;
2201 /* temporary disable recording */
2202 static atomic_t trace_record_taskinfo_disabled __read_mostly;
2204 static inline char *get_saved_cmdlines(int idx)
2206 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2209 static inline void set_cmdline(int idx, const char *cmdline)
2211 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2214 static int allocate_cmdlines_buffer(unsigned int val,
2215 struct saved_cmdlines_buffer *s)
2217 s->map_cmdline_to_pid = kmalloc_array(val,
2218 sizeof(*s->map_cmdline_to_pid),
2220 if (!s->map_cmdline_to_pid)
2223 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2224 if (!s->saved_cmdlines) {
2225 kfree(s->map_cmdline_to_pid);
2230 s->cmdline_num = val;
2231 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2232 sizeof(s->map_pid_to_cmdline));
2233 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2234 val * sizeof(*s->map_cmdline_to_pid));
2239 static int trace_create_savedcmd(void)
2243 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2247 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2257 int is_tracing_stopped(void)
2259 return global_trace.stop_count;
2263 * tracing_start - quick start of the tracer
2265 * If tracing is enabled but was stopped by tracing_stop,
2266 * this will start the tracer back up.
2268 void tracing_start(void)
2270 struct trace_buffer *buffer;
2271 unsigned long flags;
2273 if (tracing_disabled)
2276 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2277 if (--global_trace.stop_count) {
2278 if (global_trace.stop_count < 0) {
2279 /* Someone screwed up their debugging */
2281 global_trace.stop_count = 0;
2286 /* Prevent the buffers from switching */
2287 arch_spin_lock(&global_trace.max_lock);
2289 buffer = global_trace.array_buffer.buffer;
2291 ring_buffer_record_enable(buffer);
2293 #ifdef CONFIG_TRACER_MAX_TRACE
2294 buffer = global_trace.max_buffer.buffer;
2296 ring_buffer_record_enable(buffer);
2299 arch_spin_unlock(&global_trace.max_lock);
2302 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2305 static void tracing_start_tr(struct trace_array *tr)
2307 struct trace_buffer *buffer;
2308 unsigned long flags;
2310 if (tracing_disabled)
2313 /* If global, we need to also start the max tracer */
2314 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2315 return tracing_start();
2317 raw_spin_lock_irqsave(&tr->start_lock, flags);
2319 if (--tr->stop_count) {
2320 if (tr->stop_count < 0) {
2321 /* Someone screwed up their debugging */
2328 buffer = tr->array_buffer.buffer;
2330 ring_buffer_record_enable(buffer);
2333 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2337 * tracing_stop - quick stop of the tracer
2339 * Light weight way to stop tracing. Use in conjunction with
2342 void tracing_stop(void)
2344 struct trace_buffer *buffer;
2345 unsigned long flags;
2347 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2348 if (global_trace.stop_count++)
2351 /* Prevent the buffers from switching */
2352 arch_spin_lock(&global_trace.max_lock);
2354 buffer = global_trace.array_buffer.buffer;
2356 ring_buffer_record_disable(buffer);
2358 #ifdef CONFIG_TRACER_MAX_TRACE
2359 buffer = global_trace.max_buffer.buffer;
2361 ring_buffer_record_disable(buffer);
2364 arch_spin_unlock(&global_trace.max_lock);
2367 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2370 static void tracing_stop_tr(struct trace_array *tr)
2372 struct trace_buffer *buffer;
2373 unsigned long flags;
2375 /* If global, we need to also stop the max tracer */
2376 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2377 return tracing_stop();
2379 raw_spin_lock_irqsave(&tr->start_lock, flags);
2380 if (tr->stop_count++)
2383 buffer = tr->array_buffer.buffer;
2385 ring_buffer_record_disable(buffer);
2388 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2391 static int trace_save_cmdline(struct task_struct *tsk)
2395 /* treat recording of idle task as a success */
2399 if (unlikely(tsk->pid > PID_MAX_DEFAULT))
2403 * It's not the end of the world if we don't get
2404 * the lock, but we also don't want to spin
2405 * nor do we want to disable interrupts,
2406 * so if we miss here, then better luck next time.
2408 if (!arch_spin_trylock(&trace_cmdline_lock))
2411 idx = savedcmd->map_pid_to_cmdline[tsk->pid];
2412 if (idx == NO_CMDLINE_MAP) {
2413 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2416 * Check whether the cmdline buffer at idx has a pid
2417 * mapped. We are going to overwrite that entry so we
2418 * need to clear the map_pid_to_cmdline. Otherwise we
2419 * would read the new comm for the old pid.
2421 pid = savedcmd->map_cmdline_to_pid[idx];
2422 if (pid != NO_CMDLINE_MAP)
2423 savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
2425 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2426 savedcmd->map_pid_to_cmdline[tsk->pid] = idx;
2428 savedcmd->cmdline_idx = idx;
2431 set_cmdline(idx, tsk->comm);
2433 arch_spin_unlock(&trace_cmdline_lock);
2438 static void __trace_find_cmdline(int pid, char comm[])
2443 strcpy(comm, "<idle>");
2447 if (WARN_ON_ONCE(pid < 0)) {
2448 strcpy(comm, "<XXX>");
2452 if (pid > PID_MAX_DEFAULT) {
2453 strcpy(comm, "<...>");
2457 map = savedcmd->map_pid_to_cmdline[pid];
2458 if (map != NO_CMDLINE_MAP)
2459 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2461 strcpy(comm, "<...>");
2464 void trace_find_cmdline(int pid, char comm[])
2467 arch_spin_lock(&trace_cmdline_lock);
2469 __trace_find_cmdline(pid, comm);
2471 arch_spin_unlock(&trace_cmdline_lock);
2475 int trace_find_tgid(int pid)
2477 if (unlikely(!tgid_map || !pid || pid > PID_MAX_DEFAULT))
2480 return tgid_map[pid];
2483 static int trace_save_tgid(struct task_struct *tsk)
2485 /* treat recording of idle task as a success */
2489 if (unlikely(!tgid_map || tsk->pid > PID_MAX_DEFAULT))
2492 tgid_map[tsk->pid] = tsk->tgid;
2496 static bool tracing_record_taskinfo_skip(int flags)
2498 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2500 if (atomic_read(&trace_record_taskinfo_disabled) || !tracing_is_on())
2502 if (!__this_cpu_read(trace_taskinfo_save))
2508 * tracing_record_taskinfo - record the task info of a task
2510 * @task: task to record
2511 * @flags: TRACE_RECORD_CMDLINE for recording comm
2512 * TRACE_RECORD_TGID for recording tgid
2514 void tracing_record_taskinfo(struct task_struct *task, int flags)
2518 if (tracing_record_taskinfo_skip(flags))
2522 * Record as much task information as possible. If some fail, continue
2523 * to try to record the others.
2525 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2526 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2528 /* If recording any information failed, retry again soon. */
2532 __this_cpu_write(trace_taskinfo_save, false);
2536 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2538 * @prev: previous task during sched_switch
2539 * @next: next task during sched_switch
2540 * @flags: TRACE_RECORD_CMDLINE for recording comm
2541 * TRACE_RECORD_TGID for recording tgid
2543 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2544 struct task_struct *next, int flags)
2548 if (tracing_record_taskinfo_skip(flags))
2552 * Record as much task information as possible. If some fail, continue
2553 * to try to record the others.
2555 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2556 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2557 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2558 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2560 /* If recording any information failed, retry again soon. */
2564 __this_cpu_write(trace_taskinfo_save, false);
2567 /* Helpers to record a specific task information */
2568 void tracing_record_cmdline(struct task_struct *task)
2570 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2573 void tracing_record_tgid(struct task_struct *task)
2575 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2579 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2580 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2581 * simplifies those functions and keeps them in sync.
2583 enum print_line_t trace_handle_return(struct trace_seq *s)
2585 return trace_seq_has_overflowed(s) ?
2586 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2588 EXPORT_SYMBOL_GPL(trace_handle_return);
2590 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2592 unsigned int trace_flags = irqs_status;
2595 pc = preempt_count();
2598 trace_flags |= TRACE_FLAG_NMI;
2599 if (pc & HARDIRQ_MASK)
2600 trace_flags |= TRACE_FLAG_HARDIRQ;
2601 if (in_serving_softirq())
2602 trace_flags |= TRACE_FLAG_SOFTIRQ;
2604 if (tif_need_resched())
2605 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2606 if (test_preempt_need_resched())
2607 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2608 return (trace_flags << 16) | (pc & 0xff);
2611 struct ring_buffer_event *
2612 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2615 unsigned int trace_ctx)
2617 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2620 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2621 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2622 static int trace_buffered_event_ref;
2625 * trace_buffered_event_enable - enable buffering events
2627 * When events are being filtered, it is quicker to use a temporary
2628 * buffer to write the event data into if there's a likely chance
2629 * that it will not be committed. The discard of the ring buffer
2630 * is not as fast as committing, and is much slower than copying
2633 * When an event is to be filtered, allocate per cpu buffers to
2634 * write the event data into, and if the event is filtered and discarded
2635 * it is simply dropped, otherwise, the entire data is to be committed
2638 void trace_buffered_event_enable(void)
2640 struct ring_buffer_event *event;
2644 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2646 if (trace_buffered_event_ref++)
2649 for_each_tracing_cpu(cpu) {
2650 page = alloc_pages_node(cpu_to_node(cpu),
2651 GFP_KERNEL | __GFP_NORETRY, 0);
2655 event = page_address(page);
2656 memset(event, 0, sizeof(*event));
2658 per_cpu(trace_buffered_event, cpu) = event;
2661 if (cpu == smp_processor_id() &&
2662 __this_cpu_read(trace_buffered_event) !=
2663 per_cpu(trace_buffered_event, cpu))
2670 trace_buffered_event_disable();
2673 static void enable_trace_buffered_event(void *data)
2675 /* Probably not needed, but do it anyway */
2677 this_cpu_dec(trace_buffered_event_cnt);
2680 static void disable_trace_buffered_event(void *data)
2682 this_cpu_inc(trace_buffered_event_cnt);
2686 * trace_buffered_event_disable - disable buffering events
2688 * When a filter is removed, it is faster to not use the buffered
2689 * events, and to commit directly into the ring buffer. Free up
2690 * the temp buffers when there are no more users. This requires
2691 * special synchronization with current events.
2693 void trace_buffered_event_disable(void)
2697 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2699 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2702 if (--trace_buffered_event_ref)
2706 /* For each CPU, set the buffer as used. */
2707 smp_call_function_many(tracing_buffer_mask,
2708 disable_trace_buffered_event, NULL, 1);
2711 /* Wait for all current users to finish */
2714 for_each_tracing_cpu(cpu) {
2715 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2716 per_cpu(trace_buffered_event, cpu) = NULL;
2719 * Make sure trace_buffered_event is NULL before clearing
2720 * trace_buffered_event_cnt.
2725 /* Do the work on each cpu */
2726 smp_call_function_many(tracing_buffer_mask,
2727 enable_trace_buffered_event, NULL, 1);
2731 static struct trace_buffer *temp_buffer;
2733 struct ring_buffer_event *
2734 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2735 struct trace_event_file *trace_file,
2736 int type, unsigned long len,
2737 unsigned int trace_ctx)
2739 struct ring_buffer_event *entry;
2742 *current_rb = trace_file->tr->array_buffer.buffer;
2744 if (!ring_buffer_time_stamp_abs(*current_rb) && (trace_file->flags &
2745 (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED)) &&
2746 (entry = this_cpu_read(trace_buffered_event))) {
2747 /* Try to use the per cpu buffer first */
2748 val = this_cpu_inc_return(trace_buffered_event_cnt);
2749 if ((len < (PAGE_SIZE - sizeof(*entry))) && val == 1) {
2750 trace_event_setup(entry, type, trace_ctx);
2751 entry->array[0] = len;
2754 this_cpu_dec(trace_buffered_event_cnt);
2757 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2760 * If tracing is off, but we have triggers enabled
2761 * we still need to look at the event data. Use the temp_buffer
2762 * to store the trace event for the trigger to use. It's recursive
2763 * safe and will not be recorded anywhere.
2765 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2766 *current_rb = temp_buffer;
2767 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2772 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2774 static DEFINE_SPINLOCK(tracepoint_iter_lock);
2775 static DEFINE_MUTEX(tracepoint_printk_mutex);
2777 static void output_printk(struct trace_event_buffer *fbuffer)
2779 struct trace_event_call *event_call;
2780 struct trace_event_file *file;
2781 struct trace_event *event;
2782 unsigned long flags;
2783 struct trace_iterator *iter = tracepoint_print_iter;
2785 /* We should never get here if iter is NULL */
2786 if (WARN_ON_ONCE(!iter))
2789 event_call = fbuffer->trace_file->event_call;
2790 if (!event_call || !event_call->event.funcs ||
2791 !event_call->event.funcs->trace)
2794 file = fbuffer->trace_file;
2795 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2796 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2797 !filter_match_preds(file->filter, fbuffer->entry)))
2800 event = &fbuffer->trace_file->event_call->event;
2802 spin_lock_irqsave(&tracepoint_iter_lock, flags);
2803 trace_seq_init(&iter->seq);
2804 iter->ent = fbuffer->entry;
2805 event_call->event.funcs->trace(iter, 0, event);
2806 trace_seq_putc(&iter->seq, 0);
2807 printk("%s", iter->seq.buffer);
2809 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2812 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2813 void *buffer, size_t *lenp,
2816 int save_tracepoint_printk;
2819 mutex_lock(&tracepoint_printk_mutex);
2820 save_tracepoint_printk = tracepoint_printk;
2822 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2825 * This will force exiting early, as tracepoint_printk
2826 * is always zero when tracepoint_printk_iter is not allocated
2828 if (!tracepoint_print_iter)
2829 tracepoint_printk = 0;
2831 if (save_tracepoint_printk == tracepoint_printk)
2834 if (tracepoint_printk)
2835 static_key_enable(&tracepoint_printk_key.key);
2837 static_key_disable(&tracepoint_printk_key.key);
2840 mutex_unlock(&tracepoint_printk_mutex);
2845 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2847 if (static_key_false(&tracepoint_printk_key.key))
2848 output_printk(fbuffer);
2850 if (static_branch_unlikely(&trace_event_exports_enabled))
2851 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2852 event_trigger_unlock_commit_regs(fbuffer->trace_file, fbuffer->buffer,
2853 fbuffer->event, fbuffer->entry,
2854 fbuffer->trace_ctx, fbuffer->regs);
2856 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2861 * trace_buffer_unlock_commit_regs()
2862 * trace_event_buffer_commit()
2863 * trace_event_raw_event_xxx()
2865 # define STACK_SKIP 3
2867 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2868 struct trace_buffer *buffer,
2869 struct ring_buffer_event *event,
2870 unsigned int trace_ctx,
2871 struct pt_regs *regs)
2873 __buffer_unlock_commit(buffer, event);
2876 * If regs is not set, then skip the necessary functions.
2877 * Note, we can still get here via blktrace, wakeup tracer
2878 * and mmiotrace, but that's ok if they lose a function or
2879 * two. They are not that meaningful.
2881 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
2882 ftrace_trace_userstack(tr, buffer, trace_ctx);
2886 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2889 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2890 struct ring_buffer_event *event)
2892 __buffer_unlock_commit(buffer, event);
2896 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
2897 parent_ip, unsigned int trace_ctx)
2899 struct trace_event_call *call = &event_function;
2900 struct trace_buffer *buffer = tr->array_buffer.buffer;
2901 struct ring_buffer_event *event;
2902 struct ftrace_entry *entry;
2904 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2908 entry = ring_buffer_event_data(event);
2910 entry->parent_ip = parent_ip;
2912 if (!call_filter_check_discard(call, entry, buffer, event)) {
2913 if (static_branch_unlikely(&trace_function_exports_enabled))
2914 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
2915 __buffer_unlock_commit(buffer, event);
2919 #ifdef CONFIG_STACKTRACE
2921 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2922 #define FTRACE_KSTACK_NESTING 4
2924 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
2926 struct ftrace_stack {
2927 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2931 struct ftrace_stacks {
2932 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
2935 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
2936 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2938 static void __ftrace_trace_stack(struct trace_buffer *buffer,
2939 unsigned int trace_ctx,
2940 int skip, struct pt_regs *regs)
2942 struct trace_event_call *call = &event_kernel_stack;
2943 struct ring_buffer_event *event;
2944 unsigned int size, nr_entries;
2945 struct ftrace_stack *fstack;
2946 struct stack_entry *entry;
2950 * Add one, for this function and the call to save_stack_trace()
2951 * If regs is set, then these functions will not be in the way.
2953 #ifndef CONFIG_UNWINDER_ORC
2958 preempt_disable_notrace();
2960 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
2962 /* This should never happen. If it does, yell once and skip */
2963 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
2967 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
2968 * interrupt will either see the value pre increment or post
2969 * increment. If the interrupt happens pre increment it will have
2970 * restored the counter when it returns. We just need a barrier to
2971 * keep gcc from moving things around.
2975 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
2976 size = ARRAY_SIZE(fstack->calls);
2979 nr_entries = stack_trace_save_regs(regs, fstack->calls,
2982 nr_entries = stack_trace_save(fstack->calls, size, skip);
2985 size = nr_entries * sizeof(unsigned long);
2986 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2987 (sizeof(*entry) - sizeof(entry->caller)) + size,
2991 entry = ring_buffer_event_data(event);
2993 memcpy(&entry->caller, fstack->calls, size);
2994 entry->size = nr_entries;
2996 if (!call_filter_check_discard(call, entry, buffer, event))
2997 __buffer_unlock_commit(buffer, event);
3000 /* Again, don't let gcc optimize things here */
3002 __this_cpu_dec(ftrace_stack_reserve);
3003 preempt_enable_notrace();
3007 static inline void ftrace_trace_stack(struct trace_array *tr,
3008 struct trace_buffer *buffer,
3009 unsigned int trace_ctx,
3010 int skip, struct pt_regs *regs)
3012 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3015 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3018 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3021 struct trace_buffer *buffer = tr->array_buffer.buffer;
3023 if (rcu_is_watching()) {
3024 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3029 * When an NMI triggers, RCU is enabled via rcu_nmi_enter(),
3030 * but if the above rcu_is_watching() failed, then the NMI
3031 * triggered someplace critical, and rcu_irq_enter() should
3032 * not be called from NMI.
3034 if (unlikely(in_nmi()))
3037 rcu_irq_enter_irqson();
3038 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3039 rcu_irq_exit_irqson();
3043 * trace_dump_stack - record a stack back trace in the trace buffer
3044 * @skip: Number of functions to skip (helper handlers)
3046 void trace_dump_stack(int skip)
3048 if (tracing_disabled || tracing_selftest_running)
3051 #ifndef CONFIG_UNWINDER_ORC
3052 /* Skip 1 to skip this function. */
3055 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3056 tracing_gen_ctx(), skip, NULL);
3058 EXPORT_SYMBOL_GPL(trace_dump_stack);
3060 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3061 static DEFINE_PER_CPU(int, user_stack_count);
3064 ftrace_trace_userstack(struct trace_array *tr,
3065 struct trace_buffer *buffer, unsigned int trace_ctx)
3067 struct trace_event_call *call = &event_user_stack;
3068 struct ring_buffer_event *event;
3069 struct userstack_entry *entry;
3071 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3075 * NMIs can not handle page faults, even with fix ups.
3076 * The save user stack can (and often does) fault.
3078 if (unlikely(in_nmi()))
3082 * prevent recursion, since the user stack tracing may
3083 * trigger other kernel events.
3086 if (__this_cpu_read(user_stack_count))
3089 __this_cpu_inc(user_stack_count);
3091 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3092 sizeof(*entry), trace_ctx);
3094 goto out_drop_count;
3095 entry = ring_buffer_event_data(event);
3097 entry->tgid = current->tgid;
3098 memset(&entry->caller, 0, sizeof(entry->caller));
3100 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3101 if (!call_filter_check_discard(call, entry, buffer, event))
3102 __buffer_unlock_commit(buffer, event);
3105 __this_cpu_dec(user_stack_count);
3109 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3110 static void ftrace_trace_userstack(struct trace_array *tr,
3111 struct trace_buffer *buffer,
3112 unsigned int trace_ctx)
3115 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3117 #endif /* CONFIG_STACKTRACE */
3119 /* created for use with alloc_percpu */
3120 struct trace_buffer_struct {
3122 char buffer[4][TRACE_BUF_SIZE];
3125 static struct trace_buffer_struct *trace_percpu_buffer;
3128 * This allows for lockless recording. If we're nested too deeply, then
3129 * this returns NULL.
3131 static char *get_trace_buf(void)
3133 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3135 if (!buffer || buffer->nesting >= 4)
3140 /* Interrupts must see nesting incremented before we use the buffer */
3142 return &buffer->buffer[buffer->nesting - 1][0];
3145 static void put_trace_buf(void)
3147 /* Don't let the decrement of nesting leak before this */
3149 this_cpu_dec(trace_percpu_buffer->nesting);
3152 static int alloc_percpu_trace_buffer(void)
3154 struct trace_buffer_struct *buffers;
3156 if (trace_percpu_buffer)
3159 buffers = alloc_percpu(struct trace_buffer_struct);
3160 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3163 trace_percpu_buffer = buffers;
3167 static int buffers_allocated;
3169 void trace_printk_init_buffers(void)
3171 if (buffers_allocated)
3174 if (alloc_percpu_trace_buffer())
3177 /* trace_printk() is for debug use only. Don't use it in production. */
3180 pr_warn("**********************************************************\n");
3181 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3183 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3185 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3186 pr_warn("** unsafe for production use. **\n");
3188 pr_warn("** If you see this message and you are not debugging **\n");
3189 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3191 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3192 pr_warn("**********************************************************\n");
3194 /* Expand the buffers to set size */
3195 tracing_update_buffers();
3197 buffers_allocated = 1;
3200 * trace_printk_init_buffers() can be called by modules.
3201 * If that happens, then we need to start cmdline recording
3202 * directly here. If the global_trace.buffer is already
3203 * allocated here, then this was called by module code.
3205 if (global_trace.array_buffer.buffer)
3206 tracing_start_cmdline_record();
3208 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3210 void trace_printk_start_comm(void)
3212 /* Start tracing comms if trace printk is set */
3213 if (!buffers_allocated)
3215 tracing_start_cmdline_record();
3218 static void trace_printk_start_stop_comm(int enabled)
3220 if (!buffers_allocated)
3224 tracing_start_cmdline_record();
3226 tracing_stop_cmdline_record();
3230 * trace_vbprintk - write binary msg to tracing buffer
3231 * @ip: The address of the caller
3232 * @fmt: The string format to write to the buffer
3233 * @args: Arguments for @fmt
3235 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3237 struct trace_event_call *call = &event_bprint;
3238 struct ring_buffer_event *event;
3239 struct trace_buffer *buffer;
3240 struct trace_array *tr = &global_trace;
3241 struct bprint_entry *entry;
3242 unsigned int trace_ctx;
3246 if (unlikely(tracing_selftest_running || tracing_disabled))
3249 /* Don't pollute graph traces with trace_vprintk internals */
3250 pause_graph_tracing();
3252 trace_ctx = tracing_gen_ctx();
3253 preempt_disable_notrace();
3255 tbuffer = get_trace_buf();
3261 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3263 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3266 size = sizeof(*entry) + sizeof(u32) * len;
3267 buffer = tr->array_buffer.buffer;
3268 ring_buffer_nest_start(buffer);
3269 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3273 entry = ring_buffer_event_data(event);
3277 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3278 if (!call_filter_check_discard(call, entry, buffer, event)) {
3279 __buffer_unlock_commit(buffer, event);
3280 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3284 ring_buffer_nest_end(buffer);
3289 preempt_enable_notrace();
3290 unpause_graph_tracing();
3294 EXPORT_SYMBOL_GPL(trace_vbprintk);
3298 __trace_array_vprintk(struct trace_buffer *buffer,
3299 unsigned long ip, const char *fmt, va_list args)
3301 struct trace_event_call *call = &event_print;
3302 struct ring_buffer_event *event;
3304 struct print_entry *entry;
3305 unsigned int trace_ctx;
3308 if (tracing_disabled || tracing_selftest_running)
3311 /* Don't pollute graph traces with trace_vprintk internals */
3312 pause_graph_tracing();
3314 trace_ctx = tracing_gen_ctx();
3315 preempt_disable_notrace();
3318 tbuffer = get_trace_buf();
3324 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3326 size = sizeof(*entry) + len + 1;
3327 ring_buffer_nest_start(buffer);
3328 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3332 entry = ring_buffer_event_data(event);
3335 memcpy(&entry->buf, tbuffer, len + 1);
3336 if (!call_filter_check_discard(call, entry, buffer, event)) {
3337 __buffer_unlock_commit(buffer, event);
3338 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3342 ring_buffer_nest_end(buffer);
3346 preempt_enable_notrace();
3347 unpause_graph_tracing();
3353 int trace_array_vprintk(struct trace_array *tr,
3354 unsigned long ip, const char *fmt, va_list args)
3356 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3360 * trace_array_printk - Print a message to a specific instance
3361 * @tr: The instance trace_array descriptor
3362 * @ip: The instruction pointer that this is called from.
3363 * @fmt: The format to print (printf format)
3365 * If a subsystem sets up its own instance, they have the right to
3366 * printk strings into their tracing instance buffer using this
3367 * function. Note, this function will not write into the top level
3368 * buffer (use trace_printk() for that), as writing into the top level
3369 * buffer should only have events that can be individually disabled.
3370 * trace_printk() is only used for debugging a kernel, and should not
3371 * be ever encorporated in normal use.
3373 * trace_array_printk() can be used, as it will not add noise to the
3374 * top level tracing buffer.
3376 * Note, trace_array_init_printk() must be called on @tr before this
3380 int trace_array_printk(struct trace_array *tr,
3381 unsigned long ip, const char *fmt, ...)
3389 /* This is only allowed for created instances */
3390 if (tr == &global_trace)
3393 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3397 ret = trace_array_vprintk(tr, ip, fmt, ap);
3401 EXPORT_SYMBOL_GPL(trace_array_printk);
3404 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3405 * @tr: The trace array to initialize the buffers for
3407 * As trace_array_printk() only writes into instances, they are OK to
3408 * have in the kernel (unlike trace_printk()). This needs to be called
3409 * before trace_array_printk() can be used on a trace_array.
3411 int trace_array_init_printk(struct trace_array *tr)
3416 /* This is only allowed for created instances */
3417 if (tr == &global_trace)
3420 return alloc_percpu_trace_buffer();
3422 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3425 int trace_array_printk_buf(struct trace_buffer *buffer,
3426 unsigned long ip, const char *fmt, ...)
3431 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3435 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3441 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3443 return trace_array_vprintk(&global_trace, ip, fmt, args);
3445 EXPORT_SYMBOL_GPL(trace_vprintk);
3447 static void trace_iterator_increment(struct trace_iterator *iter)
3449 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3453 ring_buffer_iter_advance(buf_iter);
3456 static struct trace_entry *
3457 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3458 unsigned long *lost_events)
3460 struct ring_buffer_event *event;
3461 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3464 event = ring_buffer_iter_peek(buf_iter, ts);
3466 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3467 (unsigned long)-1 : 0;
3469 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3474 iter->ent_size = ring_buffer_event_length(event);
3475 return ring_buffer_event_data(event);
3481 static struct trace_entry *
3482 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3483 unsigned long *missing_events, u64 *ent_ts)
3485 struct trace_buffer *buffer = iter->array_buffer->buffer;
3486 struct trace_entry *ent, *next = NULL;
3487 unsigned long lost_events = 0, next_lost = 0;
3488 int cpu_file = iter->cpu_file;
3489 u64 next_ts = 0, ts;
3495 * If we are in a per_cpu trace file, don't bother by iterating over
3496 * all cpu and peek directly.
3498 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3499 if (ring_buffer_empty_cpu(buffer, cpu_file))
3501 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3503 *ent_cpu = cpu_file;
3508 for_each_tracing_cpu(cpu) {
3510 if (ring_buffer_empty_cpu(buffer, cpu))
3513 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3516 * Pick the entry with the smallest timestamp:
3518 if (ent && (!next || ts < next_ts)) {
3522 next_lost = lost_events;
3523 next_size = iter->ent_size;
3527 iter->ent_size = next_size;
3530 *ent_cpu = next_cpu;
3536 *missing_events = next_lost;
3541 #define STATIC_FMT_BUF_SIZE 128
3542 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3544 static char *trace_iter_expand_format(struct trace_iterator *iter)
3548 if (iter->fmt == static_fmt_buf)
3551 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3554 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3561 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3563 const char *p, *new_fmt;
3566 if (WARN_ON_ONCE(!fmt))
3569 if (iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
3573 new_fmt = q = iter->fmt;
3575 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
3576 if (!trace_iter_expand_format(iter))
3579 q += iter->fmt - new_fmt;
3580 new_fmt = iter->fmt;
3585 /* Replace %p with %px */
3589 } else if (p[0] == 'p' && !isalnum(p[1])) {
3600 #define STATIC_TEMP_BUF_SIZE 128
3601 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3603 /* Find the next real entry, without updating the iterator itself */
3604 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3605 int *ent_cpu, u64 *ent_ts)
3607 /* __find_next_entry will reset ent_size */
3608 int ent_size = iter->ent_size;
3609 struct trace_entry *entry;
3612 * If called from ftrace_dump(), then the iter->temp buffer
3613 * will be the static_temp_buf and not created from kmalloc.
3614 * If the entry size is greater than the buffer, we can
3615 * not save it. Just return NULL in that case. This is only
3616 * used to add markers when two consecutive events' time
3617 * stamps have a large delta. See trace_print_lat_context()
3619 if (iter->temp == static_temp_buf &&
3620 STATIC_TEMP_BUF_SIZE < ent_size)
3624 * The __find_next_entry() may call peek_next_entry(), which may
3625 * call ring_buffer_peek() that may make the contents of iter->ent
3626 * undefined. Need to copy iter->ent now.
3628 if (iter->ent && iter->ent != iter->temp) {
3629 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3630 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3632 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3637 iter->temp_size = iter->ent_size;
3639 memcpy(iter->temp, iter->ent, iter->ent_size);
3640 iter->ent = iter->temp;
3642 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3643 /* Put back the original ent_size */
3644 iter->ent_size = ent_size;
3649 /* Find the next real entry, and increment the iterator to the next entry */
3650 void *trace_find_next_entry_inc(struct trace_iterator *iter)
3652 iter->ent = __find_next_entry(iter, &iter->cpu,
3653 &iter->lost_events, &iter->ts);
3656 trace_iterator_increment(iter);
3658 return iter->ent ? iter : NULL;
3661 static void trace_consume(struct trace_iterator *iter)
3663 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3664 &iter->lost_events);
3667 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3669 struct trace_iterator *iter = m->private;
3673 WARN_ON_ONCE(iter->leftover);
3677 /* can't go backwards */
3682 ent = trace_find_next_entry_inc(iter);
3686 while (ent && iter->idx < i)
3687 ent = trace_find_next_entry_inc(iter);
3694 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3696 struct ring_buffer_iter *buf_iter;
3697 unsigned long entries = 0;
3700 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
3702 buf_iter = trace_buffer_iter(iter, cpu);
3706 ring_buffer_iter_reset(buf_iter);
3709 * We could have the case with the max latency tracers
3710 * that a reset never took place on a cpu. This is evident
3711 * by the timestamp being before the start of the buffer.
3713 while (ring_buffer_iter_peek(buf_iter, &ts)) {
3714 if (ts >= iter->array_buffer->time_start)
3717 ring_buffer_iter_advance(buf_iter);
3720 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
3724 * The current tracer is copied to avoid a global locking
3727 static void *s_start(struct seq_file *m, loff_t *pos)
3729 struct trace_iterator *iter = m->private;
3730 struct trace_array *tr = iter->tr;
3731 int cpu_file = iter->cpu_file;
3737 * copy the tracer to avoid using a global lock all around.
3738 * iter->trace is a copy of current_trace, the pointer to the
3739 * name may be used instead of a strcmp(), as iter->trace->name
3740 * will point to the same string as current_trace->name.
3742 mutex_lock(&trace_types_lock);
3743 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
3744 *iter->trace = *tr->current_trace;
3745 mutex_unlock(&trace_types_lock);
3747 #ifdef CONFIG_TRACER_MAX_TRACE
3748 if (iter->snapshot && iter->trace->use_max_tr)
3749 return ERR_PTR(-EBUSY);
3752 if (!iter->snapshot)
3753 atomic_inc(&trace_record_taskinfo_disabled);
3755 if (*pos != iter->pos) {
3760 if (cpu_file == RING_BUFFER_ALL_CPUS) {
3761 for_each_tracing_cpu(cpu)
3762 tracing_iter_reset(iter, cpu);
3764 tracing_iter_reset(iter, cpu_file);
3767 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
3772 * If we overflowed the seq_file before, then we want
3773 * to just reuse the trace_seq buffer again.
3779 p = s_next(m, p, &l);
3783 trace_event_read_lock();
3784 trace_access_lock(cpu_file);
3788 static void s_stop(struct seq_file *m, void *p)
3790 struct trace_iterator *iter = m->private;
3792 #ifdef CONFIG_TRACER_MAX_TRACE
3793 if (iter->snapshot && iter->trace->use_max_tr)
3797 if (!iter->snapshot)
3798 atomic_dec(&trace_record_taskinfo_disabled);
3800 trace_access_unlock(iter->cpu_file);
3801 trace_event_read_unlock();
3805 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
3806 unsigned long *entries, int cpu)
3808 unsigned long count;
3810 count = ring_buffer_entries_cpu(buf->buffer, cpu);
3812 * If this buffer has skipped entries, then we hold all
3813 * entries for the trace and we need to ignore the
3814 * ones before the time stamp.
3816 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
3817 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
3818 /* total is the same as the entries */
3822 ring_buffer_overrun_cpu(buf->buffer, cpu);
3827 get_total_entries(struct array_buffer *buf,
3828 unsigned long *total, unsigned long *entries)
3836 for_each_tracing_cpu(cpu) {
3837 get_total_entries_cpu(buf, &t, &e, cpu);
3843 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
3845 unsigned long total, entries;
3850 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
3855 unsigned long trace_total_entries(struct trace_array *tr)
3857 unsigned long total, entries;
3862 get_total_entries(&tr->array_buffer, &total, &entries);
3867 static void print_lat_help_header(struct seq_file *m)
3869 seq_puts(m, "# _------=> CPU# \n"
3870 "# / _-----=> irqs-off \n"
3871 "# | / _----=> need-resched \n"
3872 "# || / _---=> hardirq/softirq \n"
3873 "# ||| / _--=> preempt-depth \n"
3875 "# cmd pid ||||| time | caller \n"
3876 "# \\ / ||||| \\ | / \n");
3879 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
3881 unsigned long total;
3882 unsigned long entries;
3884 get_total_entries(buf, &total, &entries);
3885 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
3886 entries, total, num_online_cpus());
3890 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
3893 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3895 print_event_info(buf, m);
3897 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
3898 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
3901 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
3904 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3905 const char *space = " ";
3906 int prec = tgid ? 12 : 2;
3908 print_event_info(buf, m);
3910 seq_printf(m, "# %.*s _-----=> irqs-off\n", prec, space);
3911 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
3912 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
3913 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
3914 seq_printf(m, "# %.*s||| / delay\n", prec, space);
3915 seq_printf(m, "# TASK-PID %.*s CPU# |||| TIMESTAMP FUNCTION\n", prec, " TGID ");
3916 seq_printf(m, "# | | %.*s | |||| | |\n", prec, " | ");
3920 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
3922 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
3923 struct array_buffer *buf = iter->array_buffer;
3924 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
3925 struct tracer *type = iter->trace;
3926 unsigned long entries;
3927 unsigned long total;
3928 const char *name = "preemption";
3932 get_total_entries(buf, &total, &entries);
3934 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
3936 seq_puts(m, "# -----------------------------------"
3937 "---------------------------------\n");
3938 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
3939 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
3940 nsecs_to_usecs(data->saved_latency),
3944 #if defined(CONFIG_PREEMPT_NONE)
3946 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
3948 #elif defined(CONFIG_PREEMPT)
3950 #elif defined(CONFIG_PREEMPT_RT)
3955 /* These are reserved for later use */
3958 seq_printf(m, " #P:%d)\n", num_online_cpus());
3962 seq_puts(m, "# -----------------\n");
3963 seq_printf(m, "# | task: %.16s-%d "
3964 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
3965 data->comm, data->pid,
3966 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
3967 data->policy, data->rt_priority);
3968 seq_puts(m, "# -----------------\n");
3970 if (data->critical_start) {
3971 seq_puts(m, "# => started at: ");
3972 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
3973 trace_print_seq(m, &iter->seq);
3974 seq_puts(m, "\n# => ended at: ");
3975 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
3976 trace_print_seq(m, &iter->seq);
3977 seq_puts(m, "\n#\n");
3983 static void test_cpu_buff_start(struct trace_iterator *iter)
3985 struct trace_seq *s = &iter->seq;
3986 struct trace_array *tr = iter->tr;
3988 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
3991 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
3994 if (cpumask_available(iter->started) &&
3995 cpumask_test_cpu(iter->cpu, iter->started))
3998 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4001 if (cpumask_available(iter->started))
4002 cpumask_set_cpu(iter->cpu, iter->started);
4004 /* Don't print started cpu buffer for the first entry of the trace */
4006 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4010 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4012 struct trace_array *tr = iter->tr;
4013 struct trace_seq *s = &iter->seq;
4014 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4015 struct trace_entry *entry;
4016 struct trace_event *event;
4020 test_cpu_buff_start(iter);
4022 event = ftrace_find_event(entry->type);
4024 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4025 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4026 trace_print_lat_context(iter);
4028 trace_print_context(iter);
4031 if (trace_seq_has_overflowed(s))
4032 return TRACE_TYPE_PARTIAL_LINE;
4035 return event->funcs->trace(iter, sym_flags, event);
4037 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4039 return trace_handle_return(s);
4042 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4044 struct trace_array *tr = iter->tr;
4045 struct trace_seq *s = &iter->seq;
4046 struct trace_entry *entry;
4047 struct trace_event *event;
4051 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4052 trace_seq_printf(s, "%d %d %llu ",
4053 entry->pid, iter->cpu, iter->ts);
4055 if (trace_seq_has_overflowed(s))
4056 return TRACE_TYPE_PARTIAL_LINE;
4058 event = ftrace_find_event(entry->type);
4060 return event->funcs->raw(iter, 0, event);
4062 trace_seq_printf(s, "%d ?\n", entry->type);
4064 return trace_handle_return(s);
4067 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4069 struct trace_array *tr = iter->tr;
4070 struct trace_seq *s = &iter->seq;
4071 unsigned char newline = '\n';
4072 struct trace_entry *entry;
4073 struct trace_event *event;
4077 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4078 SEQ_PUT_HEX_FIELD(s, entry->pid);
4079 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4080 SEQ_PUT_HEX_FIELD(s, iter->ts);
4081 if (trace_seq_has_overflowed(s))
4082 return TRACE_TYPE_PARTIAL_LINE;
4085 event = ftrace_find_event(entry->type);
4087 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4088 if (ret != TRACE_TYPE_HANDLED)
4092 SEQ_PUT_FIELD(s, newline);
4094 return trace_handle_return(s);
4097 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4099 struct trace_array *tr = iter->tr;
4100 struct trace_seq *s = &iter->seq;
4101 struct trace_entry *entry;
4102 struct trace_event *event;
4106 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4107 SEQ_PUT_FIELD(s, entry->pid);
4108 SEQ_PUT_FIELD(s, iter->cpu);
4109 SEQ_PUT_FIELD(s, iter->ts);
4110 if (trace_seq_has_overflowed(s))
4111 return TRACE_TYPE_PARTIAL_LINE;
4114 event = ftrace_find_event(entry->type);
4115 return event ? event->funcs->binary(iter, 0, event) :
4119 int trace_empty(struct trace_iterator *iter)
4121 struct ring_buffer_iter *buf_iter;
4124 /* If we are looking at one CPU buffer, only check that one */
4125 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4126 cpu = iter->cpu_file;
4127 buf_iter = trace_buffer_iter(iter, cpu);
4129 if (!ring_buffer_iter_empty(buf_iter))
4132 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4138 for_each_tracing_cpu(cpu) {
4139 buf_iter = trace_buffer_iter(iter, cpu);
4141 if (!ring_buffer_iter_empty(buf_iter))
4144 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4152 /* Called with trace_event_read_lock() held. */
4153 enum print_line_t print_trace_line(struct trace_iterator *iter)
4155 struct trace_array *tr = iter->tr;
4156 unsigned long trace_flags = tr->trace_flags;
4157 enum print_line_t ret;
4159 if (iter->lost_events) {
4160 if (iter->lost_events == (unsigned long)-1)
4161 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4164 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4165 iter->cpu, iter->lost_events);
4166 if (trace_seq_has_overflowed(&iter->seq))
4167 return TRACE_TYPE_PARTIAL_LINE;
4170 if (iter->trace && iter->trace->print_line) {
4171 ret = iter->trace->print_line(iter);
4172 if (ret != TRACE_TYPE_UNHANDLED)
4176 if (iter->ent->type == TRACE_BPUTS &&
4177 trace_flags & TRACE_ITER_PRINTK &&
4178 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4179 return trace_print_bputs_msg_only(iter);
4181 if (iter->ent->type == TRACE_BPRINT &&
4182 trace_flags & TRACE_ITER_PRINTK &&
4183 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4184 return trace_print_bprintk_msg_only(iter);
4186 if (iter->ent->type == TRACE_PRINT &&
4187 trace_flags & TRACE_ITER_PRINTK &&
4188 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4189 return trace_print_printk_msg_only(iter);
4191 if (trace_flags & TRACE_ITER_BIN)
4192 return print_bin_fmt(iter);
4194 if (trace_flags & TRACE_ITER_HEX)
4195 return print_hex_fmt(iter);
4197 if (trace_flags & TRACE_ITER_RAW)
4198 return print_raw_fmt(iter);
4200 return print_trace_fmt(iter);
4203 void trace_latency_header(struct seq_file *m)
4205 struct trace_iterator *iter = m->private;
4206 struct trace_array *tr = iter->tr;
4208 /* print nothing if the buffers are empty */
4209 if (trace_empty(iter))
4212 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4213 print_trace_header(m, iter);
4215 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4216 print_lat_help_header(m);
4219 void trace_default_header(struct seq_file *m)
4221 struct trace_iterator *iter = m->private;
4222 struct trace_array *tr = iter->tr;
4223 unsigned long trace_flags = tr->trace_flags;
4225 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4228 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4229 /* print nothing if the buffers are empty */
4230 if (trace_empty(iter))
4232 print_trace_header(m, iter);
4233 if (!(trace_flags & TRACE_ITER_VERBOSE))
4234 print_lat_help_header(m);
4236 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4237 if (trace_flags & TRACE_ITER_IRQ_INFO)
4238 print_func_help_header_irq(iter->array_buffer,
4241 print_func_help_header(iter->array_buffer, m,
4247 static void test_ftrace_alive(struct seq_file *m)
4249 if (!ftrace_is_dead())
4251 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4252 "# MAY BE MISSING FUNCTION EVENTS\n");
4255 #ifdef CONFIG_TRACER_MAX_TRACE
4256 static void show_snapshot_main_help(struct seq_file *m)
4258 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4259 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4260 "# Takes a snapshot of the main buffer.\n"
4261 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4262 "# (Doesn't have to be '2' works with any number that\n"
4263 "# is not a '0' or '1')\n");
4266 static void show_snapshot_percpu_help(struct seq_file *m)
4268 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4269 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4270 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4271 "# Takes a snapshot of the main buffer for this cpu.\n");
4273 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4274 "# Must use main snapshot file to allocate.\n");
4276 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4277 "# (Doesn't have to be '2' works with any number that\n"
4278 "# is not a '0' or '1')\n");
4281 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4283 if (iter->tr->allocated_snapshot)
4284 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4286 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4288 seq_puts(m, "# Snapshot commands:\n");
4289 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4290 show_snapshot_main_help(m);
4292 show_snapshot_percpu_help(m);
4295 /* Should never be called */
4296 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4299 static int s_show(struct seq_file *m, void *v)
4301 struct trace_iterator *iter = v;
4304 if (iter->ent == NULL) {
4306 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4308 test_ftrace_alive(m);
4310 if (iter->snapshot && trace_empty(iter))
4311 print_snapshot_help(m, iter);
4312 else if (iter->trace && iter->trace->print_header)
4313 iter->trace->print_header(m);
4315 trace_default_header(m);
4317 } else if (iter->leftover) {
4319 * If we filled the seq_file buffer earlier, we
4320 * want to just show it now.
4322 ret = trace_print_seq(m, &iter->seq);
4324 /* ret should this time be zero, but you never know */
4325 iter->leftover = ret;
4328 print_trace_line(iter);
4329 ret = trace_print_seq(m, &iter->seq);
4331 * If we overflow the seq_file buffer, then it will
4332 * ask us for this data again at start up.
4334 * ret is 0 if seq_file write succeeded.
4337 iter->leftover = ret;
4344 * Should be used after trace_array_get(), trace_types_lock
4345 * ensures that i_cdev was already initialized.
4347 static inline int tracing_get_cpu(struct inode *inode)
4349 if (inode->i_cdev) /* See trace_create_cpu_file() */
4350 return (long)inode->i_cdev - 1;
4351 return RING_BUFFER_ALL_CPUS;
4354 static const struct seq_operations tracer_seq_ops = {
4361 static struct trace_iterator *
4362 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4364 struct trace_array *tr = inode->i_private;
4365 struct trace_iterator *iter;
4368 if (tracing_disabled)
4369 return ERR_PTR(-ENODEV);
4371 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4373 return ERR_PTR(-ENOMEM);
4375 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4377 if (!iter->buffer_iter)
4381 * trace_find_next_entry() may need to save off iter->ent.
4382 * It will place it into the iter->temp buffer. As most
4383 * events are less than 128, allocate a buffer of that size.
4384 * If one is greater, then trace_find_next_entry() will
4385 * allocate a new buffer to adjust for the bigger iter->ent.
4386 * It's not critical if it fails to get allocated here.
4388 iter->temp = kmalloc(128, GFP_KERNEL);
4390 iter->temp_size = 128;
4393 * trace_event_printf() may need to modify given format
4394 * string to replace %p with %px so that it shows real address
4395 * instead of hash value. However, that is only for the event
4396 * tracing, other tracer may not need. Defer the allocation
4397 * until it is needed.
4403 * We make a copy of the current tracer to avoid concurrent
4404 * changes on it while we are reading.
4406 mutex_lock(&trace_types_lock);
4407 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4411 *iter->trace = *tr->current_trace;
4413 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4418 #ifdef CONFIG_TRACER_MAX_TRACE
4419 /* Currently only the top directory has a snapshot */
4420 if (tr->current_trace->print_max || snapshot)
4421 iter->array_buffer = &tr->max_buffer;
4424 iter->array_buffer = &tr->array_buffer;
4425 iter->snapshot = snapshot;
4427 iter->cpu_file = tracing_get_cpu(inode);
4428 mutex_init(&iter->mutex);
4430 /* Notify the tracer early; before we stop tracing. */
4431 if (iter->trace->open)
4432 iter->trace->open(iter);
4434 /* Annotate start of buffers if we had overruns */
4435 if (ring_buffer_overruns(iter->array_buffer->buffer))
4436 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4438 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4439 if (trace_clocks[tr->clock_id].in_ns)
4440 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4443 * If pause-on-trace is enabled, then stop the trace while
4444 * dumping, unless this is the "snapshot" file
4446 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4447 tracing_stop_tr(tr);
4449 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4450 for_each_tracing_cpu(cpu) {
4451 iter->buffer_iter[cpu] =
4452 ring_buffer_read_prepare(iter->array_buffer->buffer,
4455 ring_buffer_read_prepare_sync();
4456 for_each_tracing_cpu(cpu) {
4457 ring_buffer_read_start(iter->buffer_iter[cpu]);
4458 tracing_iter_reset(iter, cpu);
4461 cpu = iter->cpu_file;
4462 iter->buffer_iter[cpu] =
4463 ring_buffer_read_prepare(iter->array_buffer->buffer,
4465 ring_buffer_read_prepare_sync();
4466 ring_buffer_read_start(iter->buffer_iter[cpu]);
4467 tracing_iter_reset(iter, cpu);
4470 mutex_unlock(&trace_types_lock);
4475 mutex_unlock(&trace_types_lock);
4478 kfree(iter->buffer_iter);
4480 seq_release_private(inode, file);
4481 return ERR_PTR(-ENOMEM);
4484 int tracing_open_generic(struct inode *inode, struct file *filp)
4488 ret = tracing_check_open_get_tr(NULL);
4492 filp->private_data = inode->i_private;
4496 bool tracing_is_disabled(void)
4498 return (tracing_disabled) ? true: false;
4502 * Open and update trace_array ref count.
4503 * Must have the current trace_array passed to it.
4505 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4507 struct trace_array *tr = inode->i_private;
4510 ret = tracing_check_open_get_tr(tr);
4514 filp->private_data = inode->i_private;
4519 static int tracing_release(struct inode *inode, struct file *file)
4521 struct trace_array *tr = inode->i_private;
4522 struct seq_file *m = file->private_data;
4523 struct trace_iterator *iter;
4526 if (!(file->f_mode & FMODE_READ)) {
4527 trace_array_put(tr);
4531 /* Writes do not use seq_file */
4533 mutex_lock(&trace_types_lock);
4535 for_each_tracing_cpu(cpu) {
4536 if (iter->buffer_iter[cpu])
4537 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4540 if (iter->trace && iter->trace->close)
4541 iter->trace->close(iter);
4543 if (!iter->snapshot && tr->stop_count)
4544 /* reenable tracing if it was previously enabled */
4545 tracing_start_tr(tr);
4547 __trace_array_put(tr);
4549 mutex_unlock(&trace_types_lock);
4551 mutex_destroy(&iter->mutex);
4552 free_cpumask_var(iter->started);
4556 kfree(iter->buffer_iter);
4557 seq_release_private(inode, file);
4562 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4564 struct trace_array *tr = inode->i_private;
4566 trace_array_put(tr);
4570 static int tracing_single_release_tr(struct inode *inode, struct file *file)
4572 struct trace_array *tr = inode->i_private;
4574 trace_array_put(tr);
4576 return single_release(inode, file);
4579 static int tracing_open(struct inode *inode, struct file *file)
4581 struct trace_array *tr = inode->i_private;
4582 struct trace_iterator *iter;
4585 ret = tracing_check_open_get_tr(tr);
4589 /* If this file was open for write, then erase contents */
4590 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4591 int cpu = tracing_get_cpu(inode);
4592 struct array_buffer *trace_buf = &tr->array_buffer;
4594 #ifdef CONFIG_TRACER_MAX_TRACE
4595 if (tr->current_trace->print_max)
4596 trace_buf = &tr->max_buffer;
4599 if (cpu == RING_BUFFER_ALL_CPUS)
4600 tracing_reset_online_cpus(trace_buf);
4602 tracing_reset_cpu(trace_buf, cpu);
4605 if (file->f_mode & FMODE_READ) {
4606 iter = __tracing_open(inode, file, false);
4608 ret = PTR_ERR(iter);
4609 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4610 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4614 trace_array_put(tr);
4620 * Some tracers are not suitable for instance buffers.
4621 * A tracer is always available for the global array (toplevel)
4622 * or if it explicitly states that it is.
4625 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4627 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4630 /* Find the next tracer that this trace array may use */
4631 static struct tracer *
4632 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4634 while (t && !trace_ok_for_array(t, tr))
4641 t_next(struct seq_file *m, void *v, loff_t *pos)
4643 struct trace_array *tr = m->private;
4644 struct tracer *t = v;
4649 t = get_tracer_for_array(tr, t->next);
4654 static void *t_start(struct seq_file *m, loff_t *pos)
4656 struct trace_array *tr = m->private;
4660 mutex_lock(&trace_types_lock);
4662 t = get_tracer_for_array(tr, trace_types);
4663 for (; t && l < *pos; t = t_next(m, t, &l))
4669 static void t_stop(struct seq_file *m, void *p)
4671 mutex_unlock(&trace_types_lock);
4674 static int t_show(struct seq_file *m, void *v)
4676 struct tracer *t = v;
4681 seq_puts(m, t->name);
4690 static const struct seq_operations show_traces_seq_ops = {
4697 static int show_traces_open(struct inode *inode, struct file *file)
4699 struct trace_array *tr = inode->i_private;
4703 ret = tracing_check_open_get_tr(tr);
4707 ret = seq_open(file, &show_traces_seq_ops);
4709 trace_array_put(tr);
4713 m = file->private_data;
4719 static int show_traces_release(struct inode *inode, struct file *file)
4721 struct trace_array *tr = inode->i_private;
4723 trace_array_put(tr);
4724 return seq_release(inode, file);
4728 tracing_write_stub(struct file *filp, const char __user *ubuf,
4729 size_t count, loff_t *ppos)
4734 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
4738 if (file->f_mode & FMODE_READ)
4739 ret = seq_lseek(file, offset, whence);
4741 file->f_pos = ret = 0;
4746 static const struct file_operations tracing_fops = {
4747 .open = tracing_open,
4749 .write = tracing_write_stub,
4750 .llseek = tracing_lseek,
4751 .release = tracing_release,
4754 static const struct file_operations show_traces_fops = {
4755 .open = show_traces_open,
4757 .llseek = seq_lseek,
4758 .release = show_traces_release,
4762 tracing_cpumask_read(struct file *filp, char __user *ubuf,
4763 size_t count, loff_t *ppos)
4765 struct trace_array *tr = file_inode(filp)->i_private;
4769 len = snprintf(NULL, 0, "%*pb\n",
4770 cpumask_pr_args(tr->tracing_cpumask)) + 1;
4771 mask_str = kmalloc(len, GFP_KERNEL);
4775 len = snprintf(mask_str, len, "%*pb\n",
4776 cpumask_pr_args(tr->tracing_cpumask));
4781 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
4789 int tracing_set_cpumask(struct trace_array *tr,
4790 cpumask_var_t tracing_cpumask_new)
4797 local_irq_disable();
4798 arch_spin_lock(&tr->max_lock);
4799 for_each_tracing_cpu(cpu) {
4801 * Increase/decrease the disabled counter if we are
4802 * about to flip a bit in the cpumask:
4804 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4805 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4806 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
4807 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
4809 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4810 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4811 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
4812 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
4815 arch_spin_unlock(&tr->max_lock);
4818 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
4824 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
4825 size_t count, loff_t *ppos)
4827 struct trace_array *tr = file_inode(filp)->i_private;
4828 cpumask_var_t tracing_cpumask_new;
4831 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
4834 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
4838 err = tracing_set_cpumask(tr, tracing_cpumask_new);
4842 free_cpumask_var(tracing_cpumask_new);
4847 free_cpumask_var(tracing_cpumask_new);
4852 static const struct file_operations tracing_cpumask_fops = {
4853 .open = tracing_open_generic_tr,
4854 .read = tracing_cpumask_read,
4855 .write = tracing_cpumask_write,
4856 .release = tracing_release_generic_tr,
4857 .llseek = generic_file_llseek,
4860 static int tracing_trace_options_show(struct seq_file *m, void *v)
4862 struct tracer_opt *trace_opts;
4863 struct trace_array *tr = m->private;
4867 mutex_lock(&trace_types_lock);
4868 tracer_flags = tr->current_trace->flags->val;
4869 trace_opts = tr->current_trace->flags->opts;
4871 for (i = 0; trace_options[i]; i++) {
4872 if (tr->trace_flags & (1 << i))
4873 seq_printf(m, "%s\n", trace_options[i]);
4875 seq_printf(m, "no%s\n", trace_options[i]);
4878 for (i = 0; trace_opts[i].name; i++) {
4879 if (tracer_flags & trace_opts[i].bit)
4880 seq_printf(m, "%s\n", trace_opts[i].name);
4882 seq_printf(m, "no%s\n", trace_opts[i].name);
4884 mutex_unlock(&trace_types_lock);
4889 static int __set_tracer_option(struct trace_array *tr,
4890 struct tracer_flags *tracer_flags,
4891 struct tracer_opt *opts, int neg)
4893 struct tracer *trace = tracer_flags->trace;
4896 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
4901 tracer_flags->val &= ~opts->bit;
4903 tracer_flags->val |= opts->bit;
4907 /* Try to assign a tracer specific option */
4908 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
4910 struct tracer *trace = tr->current_trace;
4911 struct tracer_flags *tracer_flags = trace->flags;
4912 struct tracer_opt *opts = NULL;
4915 for (i = 0; tracer_flags->opts[i].name; i++) {
4916 opts = &tracer_flags->opts[i];
4918 if (strcmp(cmp, opts->name) == 0)
4919 return __set_tracer_option(tr, trace->flags, opts, neg);
4925 /* Some tracers require overwrite to stay enabled */
4926 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
4928 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
4934 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
4936 if ((mask == TRACE_ITER_RECORD_TGID) ||
4937 (mask == TRACE_ITER_RECORD_CMD))
4938 lockdep_assert_held(&event_mutex);
4940 /* do nothing if flag is already set */
4941 if (!!(tr->trace_flags & mask) == !!enabled)
4944 /* Give the tracer a chance to approve the change */
4945 if (tr->current_trace->flag_changed)
4946 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
4950 tr->trace_flags |= mask;
4952 tr->trace_flags &= ~mask;
4954 if (mask == TRACE_ITER_RECORD_CMD)
4955 trace_event_enable_cmd_record(enabled);
4957 if (mask == TRACE_ITER_RECORD_TGID) {
4959 tgid_map = kvcalloc(PID_MAX_DEFAULT + 1,
4963 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
4967 trace_event_enable_tgid_record(enabled);
4970 if (mask == TRACE_ITER_EVENT_FORK)
4971 trace_event_follow_fork(tr, enabled);
4973 if (mask == TRACE_ITER_FUNC_FORK)
4974 ftrace_pid_follow_fork(tr, enabled);
4976 if (mask == TRACE_ITER_OVERWRITE) {
4977 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
4978 #ifdef CONFIG_TRACER_MAX_TRACE
4979 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
4983 if (mask == TRACE_ITER_PRINTK) {
4984 trace_printk_start_stop_comm(enabled);
4985 trace_printk_control(enabled);
4991 int trace_set_options(struct trace_array *tr, char *option)
4996 size_t orig_len = strlen(option);
4999 cmp = strstrip(option);
5001 len = str_has_prefix(cmp, "no");
5007 mutex_lock(&event_mutex);
5008 mutex_lock(&trace_types_lock);
5010 ret = match_string(trace_options, -1, cmp);
5011 /* If no option could be set, test the specific tracer options */
5013 ret = set_tracer_option(tr, cmp, neg);
5015 ret = set_tracer_flag(tr, 1 << ret, !neg);
5017 mutex_unlock(&trace_types_lock);
5018 mutex_unlock(&event_mutex);
5021 * If the first trailing whitespace is replaced with '\0' by strstrip,
5022 * turn it back into a space.
5024 if (orig_len > strlen(option))
5025 option[strlen(option)] = ' ';
5030 static void __init apply_trace_boot_options(void)
5032 char *buf = trace_boot_options_buf;
5036 option = strsep(&buf, ",");
5042 trace_set_options(&global_trace, option);
5044 /* Put back the comma to allow this to be called again */
5051 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5052 size_t cnt, loff_t *ppos)
5054 struct seq_file *m = filp->private_data;
5055 struct trace_array *tr = m->private;
5059 if (cnt >= sizeof(buf))
5062 if (copy_from_user(buf, ubuf, cnt))
5067 ret = trace_set_options(tr, buf);
5076 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5078 struct trace_array *tr = inode->i_private;
5081 ret = tracing_check_open_get_tr(tr);
5085 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5087 trace_array_put(tr);
5092 static const struct file_operations tracing_iter_fops = {
5093 .open = tracing_trace_options_open,
5095 .llseek = seq_lseek,
5096 .release = tracing_single_release_tr,
5097 .write = tracing_trace_options_write,
5100 static const char readme_msg[] =
5101 "tracing mini-HOWTO:\n\n"
5102 "# echo 0 > tracing_on : quick way to disable tracing\n"
5103 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5104 " Important files:\n"
5105 " trace\t\t\t- The static contents of the buffer\n"
5106 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5107 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5108 " current_tracer\t- function and latency tracers\n"
5109 " available_tracers\t- list of configured tracers for current_tracer\n"
5110 " error_log\t- error log for failed commands (that support it)\n"
5111 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5112 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5113 " trace_clock\t\t-change the clock used to order events\n"
5114 " local: Per cpu clock but may not be synced across CPUs\n"
5115 " global: Synced across CPUs but slows tracing down.\n"
5116 " counter: Not a clock, but just an increment\n"
5117 " uptime: Jiffy counter from time of boot\n"
5118 " perf: Same clock that perf events use\n"
5119 #ifdef CONFIG_X86_64
5120 " x86-tsc: TSC cycle counter\n"
5122 "\n timestamp_mode\t-view the mode used to timestamp events\n"
5123 " delta: Delta difference against a buffer-wide timestamp\n"
5124 " absolute: Absolute (standalone) timestamp\n"
5125 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5126 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5127 " tracing_cpumask\t- Limit which CPUs to trace\n"
5128 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5129 "\t\t\t Remove sub-buffer with rmdir\n"
5130 " trace_options\t\t- Set format or modify how tracing happens\n"
5131 "\t\t\t Disable an option by prefixing 'no' to the\n"
5132 "\t\t\t option name\n"
5133 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5134 #ifdef CONFIG_DYNAMIC_FTRACE
5135 "\n available_filter_functions - list of functions that can be filtered on\n"
5136 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5137 "\t\t\t functions\n"
5138 "\t accepts: func_full_name or glob-matching-pattern\n"
5139 "\t modules: Can select a group via module\n"
5140 "\t Format: :mod:<module-name>\n"
5141 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5142 "\t triggers: a command to perform when function is hit\n"
5143 "\t Format: <function>:<trigger>[:count]\n"
5144 "\t trigger: traceon, traceoff\n"
5145 "\t\t enable_event:<system>:<event>\n"
5146 "\t\t disable_event:<system>:<event>\n"
5147 #ifdef CONFIG_STACKTRACE
5150 #ifdef CONFIG_TRACER_SNAPSHOT
5155 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5156 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5157 "\t The first one will disable tracing every time do_fault is hit\n"
5158 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5159 "\t The first time do trap is hit and it disables tracing, the\n"
5160 "\t counter will decrement to 2. If tracing is already disabled,\n"
5161 "\t the counter will not decrement. It only decrements when the\n"
5162 "\t trigger did work\n"
5163 "\t To remove trigger without count:\n"
5164 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5165 "\t To remove trigger with a count:\n"
5166 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5167 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5168 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5169 "\t modules: Can select a group via module command :mod:\n"
5170 "\t Does not accept triggers\n"
5171 #endif /* CONFIG_DYNAMIC_FTRACE */
5172 #ifdef CONFIG_FUNCTION_TRACER
5173 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5175 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5178 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5179 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5180 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5181 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5183 #ifdef CONFIG_TRACER_SNAPSHOT
5184 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5185 "\t\t\t snapshot buffer. Read the contents for more\n"
5186 "\t\t\t information\n"
5188 #ifdef CONFIG_STACK_TRACER
5189 " stack_trace\t\t- Shows the max stack trace when active\n"
5190 " stack_max_size\t- Shows current max stack size that was traced\n"
5191 "\t\t\t Write into this file to reset the max size (trigger a\n"
5192 "\t\t\t new trace)\n"
5193 #ifdef CONFIG_DYNAMIC_FTRACE
5194 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5197 #endif /* CONFIG_STACK_TRACER */
5198 #ifdef CONFIG_DYNAMIC_EVENTS
5199 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5200 "\t\t\t Write into this file to define/undefine new trace events.\n"
5202 #ifdef CONFIG_KPROBE_EVENTS
5203 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5204 "\t\t\t Write into this file to define/undefine new trace events.\n"
5206 #ifdef CONFIG_UPROBE_EVENTS
5207 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5208 "\t\t\t Write into this file to define/undefine new trace events.\n"
5210 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5211 "\t accepts: event-definitions (one definition per line)\n"
5212 "\t Format: p[:[<group>/]<event>] <place> [<args>]\n"
5213 "\t r[maxactive][:[<group>/]<event>] <place> [<args>]\n"
5214 #ifdef CONFIG_HIST_TRIGGERS
5215 "\t s:[synthetic/]<event> <field> [<field>]\n"
5217 "\t -:[<group>/]<event>\n"
5218 #ifdef CONFIG_KPROBE_EVENTS
5219 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5220 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5222 #ifdef CONFIG_UPROBE_EVENTS
5223 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5225 "\t args: <name>=fetcharg[:type]\n"
5226 "\t fetcharg: %<register>, @<address>, @<symbol>[+|-<offset>],\n"
5227 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5228 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5230 "\t $stack<index>, $stack, $retval, $comm,\n"
5232 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5233 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string, symbol,\n"
5234 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5235 "\t <type>\\[<array-size>\\]\n"
5236 #ifdef CONFIG_HIST_TRIGGERS
5237 "\t field: <stype> <name>;\n"
5238 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5239 "\t [unsigned] char/int/long\n"
5242 " events/\t\t- Directory containing all trace event subsystems:\n"
5243 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5244 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5245 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5247 " filter\t\t- If set, only events passing filter are traced\n"
5248 " events/<system>/<event>/\t- Directory containing control files for\n"
5250 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5251 " filter\t\t- If set, only events passing filter are traced\n"
5252 " trigger\t\t- If set, a command to perform when event is hit\n"
5253 "\t Format: <trigger>[:count][if <filter>]\n"
5254 "\t trigger: traceon, traceoff\n"
5255 "\t enable_event:<system>:<event>\n"
5256 "\t disable_event:<system>:<event>\n"
5257 #ifdef CONFIG_HIST_TRIGGERS
5258 "\t enable_hist:<system>:<event>\n"
5259 "\t disable_hist:<system>:<event>\n"
5261 #ifdef CONFIG_STACKTRACE
5264 #ifdef CONFIG_TRACER_SNAPSHOT
5267 #ifdef CONFIG_HIST_TRIGGERS
5268 "\t\t hist (see below)\n"
5270 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5271 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5272 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5273 "\t events/block/block_unplug/trigger\n"
5274 "\t The first disables tracing every time block_unplug is hit.\n"
5275 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5276 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5277 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5278 "\t Like function triggers, the counter is only decremented if it\n"
5279 "\t enabled or disabled tracing.\n"
5280 "\t To remove a trigger without a count:\n"
5281 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5282 "\t To remove a trigger with a count:\n"
5283 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5284 "\t Filters can be ignored when removing a trigger.\n"
5285 #ifdef CONFIG_HIST_TRIGGERS
5286 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5287 "\t Format: hist:keys=<field1[,field2,...]>\n"
5288 "\t [:values=<field1[,field2,...]>]\n"
5289 "\t [:sort=<field1[,field2,...]>]\n"
5290 "\t [:size=#entries]\n"
5291 "\t [:pause][:continue][:clear]\n"
5292 "\t [:name=histname1]\n"
5293 "\t [:<handler>.<action>]\n"
5294 "\t [if <filter>]\n\n"
5295 "\t When a matching event is hit, an entry is added to a hash\n"
5296 "\t table using the key(s) and value(s) named, and the value of a\n"
5297 "\t sum called 'hitcount' is incremented. Keys and values\n"
5298 "\t correspond to fields in the event's format description. Keys\n"
5299 "\t can be any field, or the special string 'stacktrace'.\n"
5300 "\t Compound keys consisting of up to two fields can be specified\n"
5301 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5302 "\t fields. Sort keys consisting of up to two fields can be\n"
5303 "\t specified using the 'sort' keyword. The sort direction can\n"
5304 "\t be modified by appending '.descending' or '.ascending' to a\n"
5305 "\t sort field. The 'size' parameter can be used to specify more\n"
5306 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5307 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5308 "\t its histogram data will be shared with other triggers of the\n"
5309 "\t same name, and trigger hits will update this common data.\n\n"
5310 "\t Reading the 'hist' file for the event will dump the hash\n"
5311 "\t table in its entirety to stdout. If there are multiple hist\n"
5312 "\t triggers attached to an event, there will be a table for each\n"
5313 "\t trigger in the output. The table displayed for a named\n"
5314 "\t trigger will be the same as any other instance having the\n"
5315 "\t same name. The default format used to display a given field\n"
5316 "\t can be modified by appending any of the following modifiers\n"
5317 "\t to the field name, as applicable:\n\n"
5318 "\t .hex display a number as a hex value\n"
5319 "\t .sym display an address as a symbol\n"
5320 "\t .sym-offset display an address as a symbol and offset\n"
5321 "\t .execname display a common_pid as a program name\n"
5322 "\t .syscall display a syscall id as a syscall name\n"
5323 "\t .log2 display log2 value rather than raw number\n"
5324 "\t .usecs display a common_timestamp in microseconds\n\n"
5325 "\t The 'pause' parameter can be used to pause an existing hist\n"
5326 "\t trigger or to start a hist trigger but not log any events\n"
5327 "\t until told to do so. 'continue' can be used to start or\n"
5328 "\t restart a paused hist trigger.\n\n"
5329 "\t The 'clear' parameter will clear the contents of a running\n"
5330 "\t hist trigger and leave its current paused/active state\n"
5332 "\t The enable_hist and disable_hist triggers can be used to\n"
5333 "\t have one event conditionally start and stop another event's\n"
5334 "\t already-attached hist trigger. The syntax is analogous to\n"
5335 "\t the enable_event and disable_event triggers.\n\n"
5336 "\t Hist trigger handlers and actions are executed whenever a\n"
5337 "\t a histogram entry is added or updated. They take the form:\n\n"
5338 "\t <handler>.<action>\n\n"
5339 "\t The available handlers are:\n\n"
5340 "\t onmatch(matching.event) - invoke on addition or update\n"
5341 "\t onmax(var) - invoke if var exceeds current max\n"
5342 "\t onchange(var) - invoke action if var changes\n\n"
5343 "\t The available actions are:\n\n"
5344 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5345 "\t save(field,...) - save current event fields\n"
5346 #ifdef CONFIG_TRACER_SNAPSHOT
5347 "\t snapshot() - snapshot the trace buffer\n\n"
5349 #ifdef CONFIG_SYNTH_EVENTS
5350 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5351 "\t Write into this file to define/undefine new synthetic events.\n"
5352 "\t example: echo 'myevent u64 lat; char name[]' >> synthetic_events\n"
5358 tracing_readme_read(struct file *filp, char __user *ubuf,
5359 size_t cnt, loff_t *ppos)
5361 return simple_read_from_buffer(ubuf, cnt, ppos,
5362 readme_msg, strlen(readme_msg));
5365 static const struct file_operations tracing_readme_fops = {
5366 .open = tracing_open_generic,
5367 .read = tracing_readme_read,
5368 .llseek = generic_file_llseek,
5371 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5375 if (*pos || m->count)
5380 for (; ptr <= &tgid_map[PID_MAX_DEFAULT]; ptr++) {
5381 if (trace_find_tgid(*ptr))
5388 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5398 v = saved_tgids_next(m, v, &l);
5406 static void saved_tgids_stop(struct seq_file *m, void *v)
5410 static int saved_tgids_show(struct seq_file *m, void *v)
5412 int pid = (int *)v - tgid_map;
5414 seq_printf(m, "%d %d\n", pid, trace_find_tgid(pid));
5418 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5419 .start = saved_tgids_start,
5420 .stop = saved_tgids_stop,
5421 .next = saved_tgids_next,
5422 .show = saved_tgids_show,
5425 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5429 ret = tracing_check_open_get_tr(NULL);
5433 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5437 static const struct file_operations tracing_saved_tgids_fops = {
5438 .open = tracing_saved_tgids_open,
5440 .llseek = seq_lseek,
5441 .release = seq_release,
5444 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5446 unsigned int *ptr = v;
5448 if (*pos || m->count)
5453 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5455 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5464 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5470 arch_spin_lock(&trace_cmdline_lock);
5472 v = &savedcmd->map_cmdline_to_pid[0];
5474 v = saved_cmdlines_next(m, v, &l);
5482 static void saved_cmdlines_stop(struct seq_file *m, void *v)
5484 arch_spin_unlock(&trace_cmdline_lock);
5488 static int saved_cmdlines_show(struct seq_file *m, void *v)
5490 char buf[TASK_COMM_LEN];
5491 unsigned int *pid = v;
5493 __trace_find_cmdline(*pid, buf);
5494 seq_printf(m, "%d %s\n", *pid, buf);
5498 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5499 .start = saved_cmdlines_start,
5500 .next = saved_cmdlines_next,
5501 .stop = saved_cmdlines_stop,
5502 .show = saved_cmdlines_show,
5505 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5509 ret = tracing_check_open_get_tr(NULL);
5513 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5516 static const struct file_operations tracing_saved_cmdlines_fops = {
5517 .open = tracing_saved_cmdlines_open,
5519 .llseek = seq_lseek,
5520 .release = seq_release,
5524 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5525 size_t cnt, loff_t *ppos)
5530 arch_spin_lock(&trace_cmdline_lock);
5531 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5532 arch_spin_unlock(&trace_cmdline_lock);
5534 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5537 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5539 kfree(s->saved_cmdlines);
5540 kfree(s->map_cmdline_to_pid);
5544 static int tracing_resize_saved_cmdlines(unsigned int val)
5546 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5548 s = kmalloc(sizeof(*s), GFP_KERNEL);
5552 if (allocate_cmdlines_buffer(val, s) < 0) {
5557 arch_spin_lock(&trace_cmdline_lock);
5558 savedcmd_temp = savedcmd;
5560 arch_spin_unlock(&trace_cmdline_lock);
5561 free_saved_cmdlines_buffer(savedcmd_temp);
5567 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
5568 size_t cnt, loff_t *ppos)
5573 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5577 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
5578 if (!val || val > PID_MAX_DEFAULT)
5581 ret = tracing_resize_saved_cmdlines((unsigned int)val);
5590 static const struct file_operations tracing_saved_cmdlines_size_fops = {
5591 .open = tracing_open_generic,
5592 .read = tracing_saved_cmdlines_size_read,
5593 .write = tracing_saved_cmdlines_size_write,
5596 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
5597 static union trace_eval_map_item *
5598 update_eval_map(union trace_eval_map_item *ptr)
5600 if (!ptr->map.eval_string) {
5601 if (ptr->tail.next) {
5602 ptr = ptr->tail.next;
5603 /* Set ptr to the next real item (skip head) */
5611 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5613 union trace_eval_map_item *ptr = v;
5616 * Paranoid! If ptr points to end, we don't want to increment past it.
5617 * This really should never happen.
5620 ptr = update_eval_map(ptr);
5621 if (WARN_ON_ONCE(!ptr))
5625 ptr = update_eval_map(ptr);
5630 static void *eval_map_start(struct seq_file *m, loff_t *pos)
5632 union trace_eval_map_item *v;
5635 mutex_lock(&trace_eval_mutex);
5637 v = trace_eval_maps;
5641 while (v && l < *pos) {
5642 v = eval_map_next(m, v, &l);
5648 static void eval_map_stop(struct seq_file *m, void *v)
5650 mutex_unlock(&trace_eval_mutex);
5653 static int eval_map_show(struct seq_file *m, void *v)
5655 union trace_eval_map_item *ptr = v;
5657 seq_printf(m, "%s %ld (%s)\n",
5658 ptr->map.eval_string, ptr->map.eval_value,
5664 static const struct seq_operations tracing_eval_map_seq_ops = {
5665 .start = eval_map_start,
5666 .next = eval_map_next,
5667 .stop = eval_map_stop,
5668 .show = eval_map_show,
5671 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5675 ret = tracing_check_open_get_tr(NULL);
5679 return seq_open(filp, &tracing_eval_map_seq_ops);
5682 static const struct file_operations tracing_eval_map_fops = {
5683 .open = tracing_eval_map_open,
5685 .llseek = seq_lseek,
5686 .release = seq_release,
5689 static inline union trace_eval_map_item *
5690 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
5692 /* Return tail of array given the head */
5693 return ptr + ptr->head.length + 1;
5697 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
5700 struct trace_eval_map **stop;
5701 struct trace_eval_map **map;
5702 union trace_eval_map_item *map_array;
5703 union trace_eval_map_item *ptr;
5708 * The trace_eval_maps contains the map plus a head and tail item,
5709 * where the head holds the module and length of array, and the
5710 * tail holds a pointer to the next list.
5712 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
5714 pr_warn("Unable to allocate trace eval mapping\n");
5718 mutex_lock(&trace_eval_mutex);
5720 if (!trace_eval_maps)
5721 trace_eval_maps = map_array;
5723 ptr = trace_eval_maps;
5725 ptr = trace_eval_jmp_to_tail(ptr);
5726 if (!ptr->tail.next)
5728 ptr = ptr->tail.next;
5731 ptr->tail.next = map_array;
5733 map_array->head.mod = mod;
5734 map_array->head.length = len;
5737 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
5738 map_array->map = **map;
5741 memset(map_array, 0, sizeof(*map_array));
5743 mutex_unlock(&trace_eval_mutex);
5746 static void trace_create_eval_file(struct dentry *d_tracer)
5748 trace_create_file("eval_map", 0444, d_tracer,
5749 NULL, &tracing_eval_map_fops);
5752 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
5753 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
5754 static inline void trace_insert_eval_map_file(struct module *mod,
5755 struct trace_eval_map **start, int len) { }
5756 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
5758 static void trace_insert_eval_map(struct module *mod,
5759 struct trace_eval_map **start, int len)
5761 struct trace_eval_map **map;
5768 trace_event_eval_update(map, len);
5770 trace_insert_eval_map_file(mod, start, len);
5774 tracing_set_trace_read(struct file *filp, char __user *ubuf,
5775 size_t cnt, loff_t *ppos)
5777 struct trace_array *tr = filp->private_data;
5778 char buf[MAX_TRACER_SIZE+2];
5781 mutex_lock(&trace_types_lock);
5782 r = sprintf(buf, "%s\n", tr->current_trace->name);
5783 mutex_unlock(&trace_types_lock);
5785 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5788 int tracer_init(struct tracer *t, struct trace_array *tr)
5790 tracing_reset_online_cpus(&tr->array_buffer);
5794 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
5798 for_each_tracing_cpu(cpu)
5799 per_cpu_ptr(buf->data, cpu)->entries = val;
5802 #ifdef CONFIG_TRACER_MAX_TRACE
5803 /* resize @tr's buffer to the size of @size_tr's entries */
5804 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
5805 struct array_buffer *size_buf, int cpu_id)
5809 if (cpu_id == RING_BUFFER_ALL_CPUS) {
5810 for_each_tracing_cpu(cpu) {
5811 ret = ring_buffer_resize(trace_buf->buffer,
5812 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
5815 per_cpu_ptr(trace_buf->data, cpu)->entries =
5816 per_cpu_ptr(size_buf->data, cpu)->entries;
5819 ret = ring_buffer_resize(trace_buf->buffer,
5820 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
5822 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
5823 per_cpu_ptr(size_buf->data, cpu_id)->entries;
5828 #endif /* CONFIG_TRACER_MAX_TRACE */
5830 static int __tracing_resize_ring_buffer(struct trace_array *tr,
5831 unsigned long size, int cpu)
5836 * If kernel or user changes the size of the ring buffer
5837 * we use the size that was given, and we can forget about
5838 * expanding it later.
5840 ring_buffer_expanded = true;
5842 /* May be called before buffers are initialized */
5843 if (!tr->array_buffer.buffer)
5846 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
5850 #ifdef CONFIG_TRACER_MAX_TRACE
5851 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
5852 !tr->current_trace->use_max_tr)
5855 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
5857 int r = resize_buffer_duplicate_size(&tr->array_buffer,
5858 &tr->array_buffer, cpu);
5861 * AARGH! We are left with different
5862 * size max buffer!!!!
5863 * The max buffer is our "snapshot" buffer.
5864 * When a tracer needs a snapshot (one of the
5865 * latency tracers), it swaps the max buffer
5866 * with the saved snap shot. We succeeded to
5867 * update the size of the main buffer, but failed to
5868 * update the size of the max buffer. But when we tried
5869 * to reset the main buffer to the original size, we
5870 * failed there too. This is very unlikely to
5871 * happen, but if it does, warn and kill all
5875 tracing_disabled = 1;
5880 if (cpu == RING_BUFFER_ALL_CPUS)
5881 set_buffer_entries(&tr->max_buffer, size);
5883 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
5886 #endif /* CONFIG_TRACER_MAX_TRACE */
5888 if (cpu == RING_BUFFER_ALL_CPUS)
5889 set_buffer_entries(&tr->array_buffer, size);
5891 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
5896 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
5897 unsigned long size, int cpu_id)
5901 mutex_lock(&trace_types_lock);
5903 if (cpu_id != RING_BUFFER_ALL_CPUS) {
5904 /* make sure, this cpu is enabled in the mask */
5905 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
5911 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
5916 mutex_unlock(&trace_types_lock);
5923 * tracing_update_buffers - used by tracing facility to expand ring buffers
5925 * To save on memory when the tracing is never used on a system with it
5926 * configured in. The ring buffers are set to a minimum size. But once
5927 * a user starts to use the tracing facility, then they need to grow
5928 * to their default size.
5930 * This function is to be called when a tracer is about to be used.
5932 int tracing_update_buffers(void)
5936 mutex_lock(&trace_types_lock);
5937 if (!ring_buffer_expanded)
5938 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
5939 RING_BUFFER_ALL_CPUS);
5940 mutex_unlock(&trace_types_lock);
5945 struct trace_option_dentry;
5948 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
5951 * Used to clear out the tracer before deletion of an instance.
5952 * Must have trace_types_lock held.
5954 static void tracing_set_nop(struct trace_array *tr)
5956 if (tr->current_trace == &nop_trace)
5959 tr->current_trace->enabled--;
5961 if (tr->current_trace->reset)
5962 tr->current_trace->reset(tr);
5964 tr->current_trace = &nop_trace;
5967 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
5969 /* Only enable if the directory has been created already. */
5973 create_trace_option_files(tr, t);
5976 int tracing_set_tracer(struct trace_array *tr, const char *buf)
5979 #ifdef CONFIG_TRACER_MAX_TRACE
5984 mutex_lock(&trace_types_lock);
5986 if (!ring_buffer_expanded) {
5987 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
5988 RING_BUFFER_ALL_CPUS);
5994 for (t = trace_types; t; t = t->next) {
5995 if (strcmp(t->name, buf) == 0)
6002 if (t == tr->current_trace)
6005 #ifdef CONFIG_TRACER_SNAPSHOT
6006 if (t->use_max_tr) {
6007 arch_spin_lock(&tr->max_lock);
6008 if (tr->cond_snapshot)
6010 arch_spin_unlock(&tr->max_lock);
6015 /* Some tracers won't work on kernel command line */
6016 if (system_state < SYSTEM_RUNNING && t->noboot) {
6017 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6022 /* Some tracers are only allowed for the top level buffer */
6023 if (!trace_ok_for_array(t, tr)) {
6028 /* If trace pipe files are being read, we can't change the tracer */
6029 if (tr->trace_ref) {
6034 trace_branch_disable();
6036 tr->current_trace->enabled--;
6038 if (tr->current_trace->reset)
6039 tr->current_trace->reset(tr);
6041 /* Current trace needs to be nop_trace before synchronize_rcu */
6042 tr->current_trace = &nop_trace;
6044 #ifdef CONFIG_TRACER_MAX_TRACE
6045 had_max_tr = tr->allocated_snapshot;
6047 if (had_max_tr && !t->use_max_tr) {
6049 * We need to make sure that the update_max_tr sees that
6050 * current_trace changed to nop_trace to keep it from
6051 * swapping the buffers after we resize it.
6052 * The update_max_tr is called from interrupts disabled
6053 * so a synchronized_sched() is sufficient.
6060 #ifdef CONFIG_TRACER_MAX_TRACE
6061 if (t->use_max_tr && !had_max_tr) {
6062 ret = tracing_alloc_snapshot_instance(tr);
6069 ret = tracer_init(t, tr);
6074 tr->current_trace = t;
6075 tr->current_trace->enabled++;
6076 trace_branch_enable(tr);
6078 mutex_unlock(&trace_types_lock);
6084 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6085 size_t cnt, loff_t *ppos)
6087 struct trace_array *tr = filp->private_data;
6088 char buf[MAX_TRACER_SIZE+1];
6095 if (cnt > MAX_TRACER_SIZE)
6096 cnt = MAX_TRACER_SIZE;
6098 if (copy_from_user(buf, ubuf, cnt))
6103 /* strip ending whitespace. */
6104 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
6107 err = tracing_set_tracer(tr, buf);
6117 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6118 size_t cnt, loff_t *ppos)
6123 r = snprintf(buf, sizeof(buf), "%ld\n",
6124 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6125 if (r > sizeof(buf))
6127 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6131 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6132 size_t cnt, loff_t *ppos)
6137 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6147 tracing_thresh_read(struct file *filp, char __user *ubuf,
6148 size_t cnt, loff_t *ppos)
6150 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6154 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6155 size_t cnt, loff_t *ppos)
6157 struct trace_array *tr = filp->private_data;
6160 mutex_lock(&trace_types_lock);
6161 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6165 if (tr->current_trace->update_thresh) {
6166 ret = tr->current_trace->update_thresh(tr);
6173 mutex_unlock(&trace_types_lock);
6178 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
6181 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6182 size_t cnt, loff_t *ppos)
6184 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6188 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6189 size_t cnt, loff_t *ppos)
6191 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6196 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6198 struct trace_array *tr = inode->i_private;
6199 struct trace_iterator *iter;
6202 ret = tracing_check_open_get_tr(tr);
6206 mutex_lock(&trace_types_lock);
6208 /* create a buffer to store the information to pass to userspace */
6209 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6212 __trace_array_put(tr);
6216 trace_seq_init(&iter->seq);
6217 iter->trace = tr->current_trace;
6219 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6224 /* trace pipe does not show start of buffer */
6225 cpumask_setall(iter->started);
6227 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6228 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6230 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6231 if (trace_clocks[tr->clock_id].in_ns)
6232 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6235 iter->array_buffer = &tr->array_buffer;
6236 iter->cpu_file = tracing_get_cpu(inode);
6237 mutex_init(&iter->mutex);
6238 filp->private_data = iter;
6240 if (iter->trace->pipe_open)
6241 iter->trace->pipe_open(iter);
6243 nonseekable_open(inode, filp);
6247 mutex_unlock(&trace_types_lock);
6252 __trace_array_put(tr);
6253 mutex_unlock(&trace_types_lock);
6257 static int tracing_release_pipe(struct inode *inode, struct file *file)
6259 struct trace_iterator *iter = file->private_data;
6260 struct trace_array *tr = inode->i_private;
6262 mutex_lock(&trace_types_lock);
6266 if (iter->trace->pipe_close)
6267 iter->trace->pipe_close(iter);
6269 mutex_unlock(&trace_types_lock);
6271 free_cpumask_var(iter->started);
6272 mutex_destroy(&iter->mutex);
6275 trace_array_put(tr);
6281 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6283 struct trace_array *tr = iter->tr;
6285 /* Iterators are static, they should be filled or empty */
6286 if (trace_buffer_iter(iter, iter->cpu_file))
6287 return EPOLLIN | EPOLLRDNORM;
6289 if (tr->trace_flags & TRACE_ITER_BLOCK)
6291 * Always select as readable when in blocking mode
6293 return EPOLLIN | EPOLLRDNORM;
6295 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6300 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6302 struct trace_iterator *iter = filp->private_data;
6304 return trace_poll(iter, filp, poll_table);
6307 /* Must be called with iter->mutex held. */
6308 static int tracing_wait_pipe(struct file *filp)
6310 struct trace_iterator *iter = filp->private_data;
6313 while (trace_empty(iter)) {
6315 if ((filp->f_flags & O_NONBLOCK)) {
6320 * We block until we read something and tracing is disabled.
6321 * We still block if tracing is disabled, but we have never
6322 * read anything. This allows a user to cat this file, and
6323 * then enable tracing. But after we have read something,
6324 * we give an EOF when tracing is again disabled.
6326 * iter->pos will be 0 if we haven't read anything.
6328 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6331 mutex_unlock(&iter->mutex);
6333 ret = wait_on_pipe(iter, 0);
6335 mutex_lock(&iter->mutex);
6348 tracing_read_pipe(struct file *filp, char __user *ubuf,
6349 size_t cnt, loff_t *ppos)
6351 struct trace_iterator *iter = filp->private_data;
6355 * Avoid more than one consumer on a single file descriptor
6356 * This is just a matter of traces coherency, the ring buffer itself
6359 mutex_lock(&iter->mutex);
6361 /* return any leftover data */
6362 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6366 trace_seq_init(&iter->seq);
6368 if (iter->trace->read) {
6369 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6375 sret = tracing_wait_pipe(filp);
6379 /* stop when tracing is finished */
6380 if (trace_empty(iter)) {
6385 if (cnt >= PAGE_SIZE)
6386 cnt = PAGE_SIZE - 1;
6388 /* reset all but tr, trace, and overruns */
6389 memset(&iter->seq, 0,
6390 sizeof(struct trace_iterator) -
6391 offsetof(struct trace_iterator, seq));
6392 cpumask_clear(iter->started);
6393 trace_seq_init(&iter->seq);
6396 trace_event_read_lock();
6397 trace_access_lock(iter->cpu_file);
6398 while (trace_find_next_entry_inc(iter) != NULL) {
6399 enum print_line_t ret;
6400 int save_len = iter->seq.seq.len;
6402 ret = print_trace_line(iter);
6403 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6404 /* don't print partial lines */
6405 iter->seq.seq.len = save_len;
6408 if (ret != TRACE_TYPE_NO_CONSUME)
6409 trace_consume(iter);
6411 if (trace_seq_used(&iter->seq) >= cnt)
6415 * Setting the full flag means we reached the trace_seq buffer
6416 * size and we should leave by partial output condition above.
6417 * One of the trace_seq_* functions is not used properly.
6419 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6422 trace_access_unlock(iter->cpu_file);
6423 trace_event_read_unlock();
6425 /* Now copy what we have to the user */
6426 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6427 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6428 trace_seq_init(&iter->seq);
6431 * If there was nothing to send to user, in spite of consuming trace
6432 * entries, go back to wait for more entries.
6438 mutex_unlock(&iter->mutex);
6443 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6446 __free_page(spd->pages[idx]);
6450 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6456 /* Seq buffer is page-sized, exactly what we need. */
6458 save_len = iter->seq.seq.len;
6459 ret = print_trace_line(iter);
6461 if (trace_seq_has_overflowed(&iter->seq)) {
6462 iter->seq.seq.len = save_len;
6467 * This should not be hit, because it should only
6468 * be set if the iter->seq overflowed. But check it
6469 * anyway to be safe.
6471 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6472 iter->seq.seq.len = save_len;
6476 count = trace_seq_used(&iter->seq) - save_len;
6479 iter->seq.seq.len = save_len;
6483 if (ret != TRACE_TYPE_NO_CONSUME)
6484 trace_consume(iter);
6486 if (!trace_find_next_entry_inc(iter)) {
6496 static ssize_t tracing_splice_read_pipe(struct file *filp,
6498 struct pipe_inode_info *pipe,
6502 struct page *pages_def[PIPE_DEF_BUFFERS];
6503 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6504 struct trace_iterator *iter = filp->private_data;
6505 struct splice_pipe_desc spd = {
6507 .partial = partial_def,
6508 .nr_pages = 0, /* This gets updated below. */
6509 .nr_pages_max = PIPE_DEF_BUFFERS,
6510 .ops = &default_pipe_buf_ops,
6511 .spd_release = tracing_spd_release_pipe,
6517 if (splice_grow_spd(pipe, &spd))
6520 mutex_lock(&iter->mutex);
6522 if (iter->trace->splice_read) {
6523 ret = iter->trace->splice_read(iter, filp,
6524 ppos, pipe, len, flags);
6529 ret = tracing_wait_pipe(filp);
6533 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6538 trace_event_read_lock();
6539 trace_access_lock(iter->cpu_file);
6541 /* Fill as many pages as possible. */
6542 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6543 spd.pages[i] = alloc_page(GFP_KERNEL);
6547 rem = tracing_fill_pipe_page(rem, iter);
6549 /* Copy the data into the page, so we can start over. */
6550 ret = trace_seq_to_buffer(&iter->seq,
6551 page_address(spd.pages[i]),
6552 trace_seq_used(&iter->seq));
6554 __free_page(spd.pages[i]);
6557 spd.partial[i].offset = 0;
6558 spd.partial[i].len = trace_seq_used(&iter->seq);
6560 trace_seq_init(&iter->seq);
6563 trace_access_unlock(iter->cpu_file);
6564 trace_event_read_unlock();
6565 mutex_unlock(&iter->mutex);
6570 ret = splice_to_pipe(pipe, &spd);
6574 splice_shrink_spd(&spd);
6578 mutex_unlock(&iter->mutex);
6583 tracing_entries_read(struct file *filp, char __user *ubuf,
6584 size_t cnt, loff_t *ppos)
6586 struct inode *inode = file_inode(filp);
6587 struct trace_array *tr = inode->i_private;
6588 int cpu = tracing_get_cpu(inode);
6593 mutex_lock(&trace_types_lock);
6595 if (cpu == RING_BUFFER_ALL_CPUS) {
6596 int cpu, buf_size_same;
6601 /* check if all cpu sizes are same */
6602 for_each_tracing_cpu(cpu) {
6603 /* fill in the size from first enabled cpu */
6605 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6606 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6612 if (buf_size_same) {
6613 if (!ring_buffer_expanded)
6614 r = sprintf(buf, "%lu (expanded: %lu)\n",
6616 trace_buf_size >> 10);
6618 r = sprintf(buf, "%lu\n", size >> 10);
6620 r = sprintf(buf, "X\n");
6622 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6624 mutex_unlock(&trace_types_lock);
6626 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6631 tracing_entries_write(struct file *filp, const char __user *ubuf,
6632 size_t cnt, loff_t *ppos)
6634 struct inode *inode = file_inode(filp);
6635 struct trace_array *tr = inode->i_private;
6639 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6643 /* must have at least 1 entry */
6647 /* value is in KB */
6649 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6659 tracing_total_entries_read(struct file *filp, char __user *ubuf,
6660 size_t cnt, loff_t *ppos)
6662 struct trace_array *tr = filp->private_data;
6665 unsigned long size = 0, expanded_size = 0;
6667 mutex_lock(&trace_types_lock);
6668 for_each_tracing_cpu(cpu) {
6669 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6670 if (!ring_buffer_expanded)
6671 expanded_size += trace_buf_size >> 10;
6673 if (ring_buffer_expanded)
6674 r = sprintf(buf, "%lu\n", size);
6676 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6677 mutex_unlock(&trace_types_lock);
6679 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6683 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
6684 size_t cnt, loff_t *ppos)
6687 * There is no need to read what the user has written, this function
6688 * is just to make sure that there is no error when "echo" is used
6697 tracing_free_buffer_release(struct inode *inode, struct file *filp)
6699 struct trace_array *tr = inode->i_private;
6701 /* disable tracing ? */
6702 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
6703 tracer_tracing_off(tr);
6704 /* resize the ring buffer to 0 */
6705 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
6707 trace_array_put(tr);
6713 tracing_mark_write(struct file *filp, const char __user *ubuf,
6714 size_t cnt, loff_t *fpos)
6716 struct trace_array *tr = filp->private_data;
6717 struct ring_buffer_event *event;
6718 enum event_trigger_type tt = ETT_NONE;
6719 struct trace_buffer *buffer;
6720 struct print_entry *entry;
6725 /* Used in tracing_mark_raw_write() as well */
6726 #define FAULTED_STR "<faulted>"
6727 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
6729 if (tracing_disabled)
6732 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6735 if (cnt > TRACE_BUF_SIZE)
6736 cnt = TRACE_BUF_SIZE;
6738 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6740 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
6742 /* If less than "<faulted>", then make sure we can still add that */
6743 if (cnt < FAULTED_SIZE)
6744 size += FAULTED_SIZE - cnt;
6746 buffer = tr->array_buffer.buffer;
6747 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
6749 if (unlikely(!event))
6750 /* Ring buffer disabled, return as if not open for write */
6753 entry = ring_buffer_event_data(event);
6754 entry->ip = _THIS_IP_;
6756 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
6758 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6764 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
6765 /* do not add \n before testing triggers, but add \0 */
6766 entry->buf[cnt] = '\0';
6767 tt = event_triggers_call(tr->trace_marker_file, entry, event);
6770 if (entry->buf[cnt - 1] != '\n') {
6771 entry->buf[cnt] = '\n';
6772 entry->buf[cnt + 1] = '\0';
6774 entry->buf[cnt] = '\0';
6776 if (static_branch_unlikely(&trace_marker_exports_enabled))
6777 ftrace_exports(event, TRACE_EXPORT_MARKER);
6778 __buffer_unlock_commit(buffer, event);
6781 event_triggers_post_call(tr->trace_marker_file, tt);
6789 /* Limit it for now to 3K (including tag) */
6790 #define RAW_DATA_MAX_SIZE (1024*3)
6793 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
6794 size_t cnt, loff_t *fpos)
6796 struct trace_array *tr = filp->private_data;
6797 struct ring_buffer_event *event;
6798 struct trace_buffer *buffer;
6799 struct raw_data_entry *entry;
6804 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
6806 if (tracing_disabled)
6809 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6812 /* The marker must at least have a tag id */
6813 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
6816 if (cnt > TRACE_BUF_SIZE)
6817 cnt = TRACE_BUF_SIZE;
6819 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6821 size = sizeof(*entry) + cnt;
6822 if (cnt < FAULT_SIZE_ID)
6823 size += FAULT_SIZE_ID - cnt;
6825 buffer = tr->array_buffer.buffer;
6826 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
6829 /* Ring buffer disabled, return as if not open for write */
6832 entry = ring_buffer_event_data(event);
6834 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
6837 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6842 __buffer_unlock_commit(buffer, event);
6850 static int tracing_clock_show(struct seq_file *m, void *v)
6852 struct trace_array *tr = m->private;
6855 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
6857 "%s%s%s%s", i ? " " : "",
6858 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
6859 i == tr->clock_id ? "]" : "");
6865 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
6869 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
6870 if (strcmp(trace_clocks[i].name, clockstr) == 0)
6873 if (i == ARRAY_SIZE(trace_clocks))
6876 mutex_lock(&trace_types_lock);
6880 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
6883 * New clock may not be consistent with the previous clock.
6884 * Reset the buffer so that it doesn't have incomparable timestamps.
6886 tracing_reset_online_cpus(&tr->array_buffer);
6888 #ifdef CONFIG_TRACER_MAX_TRACE
6889 if (tr->max_buffer.buffer)
6890 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
6891 tracing_reset_online_cpus(&tr->max_buffer);
6894 mutex_unlock(&trace_types_lock);
6899 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
6900 size_t cnt, loff_t *fpos)
6902 struct seq_file *m = filp->private_data;
6903 struct trace_array *tr = m->private;
6905 const char *clockstr;
6908 if (cnt >= sizeof(buf))
6911 if (copy_from_user(buf, ubuf, cnt))
6916 clockstr = strstrip(buf);
6918 ret = tracing_set_clock(tr, clockstr);
6927 static int tracing_clock_open(struct inode *inode, struct file *file)
6929 struct trace_array *tr = inode->i_private;
6932 ret = tracing_check_open_get_tr(tr);
6936 ret = single_open(file, tracing_clock_show, inode->i_private);
6938 trace_array_put(tr);
6943 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
6945 struct trace_array *tr = m->private;
6947 mutex_lock(&trace_types_lock);
6949 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
6950 seq_puts(m, "delta [absolute]\n");
6952 seq_puts(m, "[delta] absolute\n");
6954 mutex_unlock(&trace_types_lock);
6959 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
6961 struct trace_array *tr = inode->i_private;
6964 ret = tracing_check_open_get_tr(tr);
6968 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
6970 trace_array_put(tr);
6975 int tracing_set_time_stamp_abs(struct trace_array *tr, bool abs)
6979 mutex_lock(&trace_types_lock);
6981 if (abs && tr->time_stamp_abs_ref++)
6985 if (WARN_ON_ONCE(!tr->time_stamp_abs_ref)) {
6990 if (--tr->time_stamp_abs_ref)
6994 ring_buffer_set_time_stamp_abs(tr->array_buffer.buffer, abs);
6996 #ifdef CONFIG_TRACER_MAX_TRACE
6997 if (tr->max_buffer.buffer)
6998 ring_buffer_set_time_stamp_abs(tr->max_buffer.buffer, abs);
7001 mutex_unlock(&trace_types_lock);
7006 struct ftrace_buffer_info {
7007 struct trace_iterator iter;
7009 unsigned int spare_cpu;
7013 #ifdef CONFIG_TRACER_SNAPSHOT
7014 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7016 struct trace_array *tr = inode->i_private;
7017 struct trace_iterator *iter;
7021 ret = tracing_check_open_get_tr(tr);
7025 if (file->f_mode & FMODE_READ) {
7026 iter = __tracing_open(inode, file, true);
7028 ret = PTR_ERR(iter);
7030 /* Writes still need the seq_file to hold the private data */
7032 m = kzalloc(sizeof(*m), GFP_KERNEL);
7035 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7043 iter->array_buffer = &tr->max_buffer;
7044 iter->cpu_file = tracing_get_cpu(inode);
7046 file->private_data = m;
7050 trace_array_put(tr);
7056 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7059 struct seq_file *m = filp->private_data;
7060 struct trace_iterator *iter = m->private;
7061 struct trace_array *tr = iter->tr;
7065 ret = tracing_update_buffers();
7069 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7073 mutex_lock(&trace_types_lock);
7075 if (tr->current_trace->use_max_tr) {
7080 arch_spin_lock(&tr->max_lock);
7081 if (tr->cond_snapshot)
7083 arch_spin_unlock(&tr->max_lock);
7089 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7093 if (tr->allocated_snapshot)
7097 /* Only allow per-cpu swap if the ring buffer supports it */
7098 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7099 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7104 if (tr->allocated_snapshot)
7105 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7106 &tr->array_buffer, iter->cpu_file);
7108 ret = tracing_alloc_snapshot_instance(tr);
7111 local_irq_disable();
7112 /* Now, we're going to swap */
7113 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7114 update_max_tr(tr, current, smp_processor_id(), NULL);
7116 update_max_tr_single(tr, current, iter->cpu_file);
7120 if (tr->allocated_snapshot) {
7121 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7122 tracing_reset_online_cpus(&tr->max_buffer);
7124 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7134 mutex_unlock(&trace_types_lock);
7138 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7140 struct seq_file *m = file->private_data;
7143 ret = tracing_release(inode, file);
7145 if (file->f_mode & FMODE_READ)
7148 /* If write only, the seq_file is just a stub */
7156 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7157 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7158 size_t count, loff_t *ppos);
7159 static int tracing_buffers_release(struct inode *inode, struct file *file);
7160 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7161 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7163 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7165 struct ftrace_buffer_info *info;
7168 /* The following checks for tracefs lockdown */
7169 ret = tracing_buffers_open(inode, filp);
7173 info = filp->private_data;
7175 if (info->iter.trace->use_max_tr) {
7176 tracing_buffers_release(inode, filp);
7180 info->iter.snapshot = true;
7181 info->iter.array_buffer = &info->iter.tr->max_buffer;
7186 #endif /* CONFIG_TRACER_SNAPSHOT */
7189 static const struct file_operations tracing_thresh_fops = {
7190 .open = tracing_open_generic,
7191 .read = tracing_thresh_read,
7192 .write = tracing_thresh_write,
7193 .llseek = generic_file_llseek,
7196 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
7197 static const struct file_operations tracing_max_lat_fops = {
7198 .open = tracing_open_generic,
7199 .read = tracing_max_lat_read,
7200 .write = tracing_max_lat_write,
7201 .llseek = generic_file_llseek,
7205 static const struct file_operations set_tracer_fops = {
7206 .open = tracing_open_generic,
7207 .read = tracing_set_trace_read,
7208 .write = tracing_set_trace_write,
7209 .llseek = generic_file_llseek,
7212 static const struct file_operations tracing_pipe_fops = {
7213 .open = tracing_open_pipe,
7214 .poll = tracing_poll_pipe,
7215 .read = tracing_read_pipe,
7216 .splice_read = tracing_splice_read_pipe,
7217 .release = tracing_release_pipe,
7218 .llseek = no_llseek,
7221 static const struct file_operations tracing_entries_fops = {
7222 .open = tracing_open_generic_tr,
7223 .read = tracing_entries_read,
7224 .write = tracing_entries_write,
7225 .llseek = generic_file_llseek,
7226 .release = tracing_release_generic_tr,
7229 static const struct file_operations tracing_total_entries_fops = {
7230 .open = tracing_open_generic_tr,
7231 .read = tracing_total_entries_read,
7232 .llseek = generic_file_llseek,
7233 .release = tracing_release_generic_tr,
7236 static const struct file_operations tracing_free_buffer_fops = {
7237 .open = tracing_open_generic_tr,
7238 .write = tracing_free_buffer_write,
7239 .release = tracing_free_buffer_release,
7242 static const struct file_operations tracing_mark_fops = {
7243 .open = tracing_open_generic_tr,
7244 .write = tracing_mark_write,
7245 .llseek = generic_file_llseek,
7246 .release = tracing_release_generic_tr,
7249 static const struct file_operations tracing_mark_raw_fops = {
7250 .open = tracing_open_generic_tr,
7251 .write = tracing_mark_raw_write,
7252 .llseek = generic_file_llseek,
7253 .release = tracing_release_generic_tr,
7256 static const struct file_operations trace_clock_fops = {
7257 .open = tracing_clock_open,
7259 .llseek = seq_lseek,
7260 .release = tracing_single_release_tr,
7261 .write = tracing_clock_write,
7264 static const struct file_operations trace_time_stamp_mode_fops = {
7265 .open = tracing_time_stamp_mode_open,
7267 .llseek = seq_lseek,
7268 .release = tracing_single_release_tr,
7271 #ifdef CONFIG_TRACER_SNAPSHOT
7272 static const struct file_operations snapshot_fops = {
7273 .open = tracing_snapshot_open,
7275 .write = tracing_snapshot_write,
7276 .llseek = tracing_lseek,
7277 .release = tracing_snapshot_release,
7280 static const struct file_operations snapshot_raw_fops = {
7281 .open = snapshot_raw_open,
7282 .read = tracing_buffers_read,
7283 .release = tracing_buffers_release,
7284 .splice_read = tracing_buffers_splice_read,
7285 .llseek = no_llseek,
7288 #endif /* CONFIG_TRACER_SNAPSHOT */
7290 #define TRACING_LOG_ERRS_MAX 8
7291 #define TRACING_LOG_LOC_MAX 128
7293 #define CMD_PREFIX " Command: "
7296 const char **errs; /* ptr to loc-specific array of err strings */
7297 u8 type; /* index into errs -> specific err string */
7298 u8 pos; /* MAX_FILTER_STR_VAL = 256 */
7302 struct tracing_log_err {
7303 struct list_head list;
7304 struct err_info info;
7305 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7306 char cmd[MAX_FILTER_STR_VAL]; /* what caused err */
7309 static DEFINE_MUTEX(tracing_err_log_lock);
7311 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr)
7313 struct tracing_log_err *err;
7315 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7316 err = kzalloc(sizeof(*err), GFP_KERNEL);
7318 err = ERR_PTR(-ENOMEM);
7319 tr->n_err_log_entries++;
7324 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7325 list_del(&err->list);
7331 * err_pos - find the position of a string within a command for error careting
7332 * @cmd: The tracing command that caused the error
7333 * @str: The string to position the caret at within @cmd
7335 * Finds the position of the first occurence of @str within @cmd. The
7336 * return value can be passed to tracing_log_err() for caret placement
7339 * Returns the index within @cmd of the first occurence of @str or 0
7340 * if @str was not found.
7342 unsigned int err_pos(char *cmd, const char *str)
7346 if (WARN_ON(!strlen(cmd)))
7349 found = strstr(cmd, str);
7357 * tracing_log_err - write an error to the tracing error log
7358 * @tr: The associated trace array for the error (NULL for top level array)
7359 * @loc: A string describing where the error occurred
7360 * @cmd: The tracing command that caused the error
7361 * @errs: The array of loc-specific static error strings
7362 * @type: The index into errs[], which produces the specific static err string
7363 * @pos: The position the caret should be placed in the cmd
7365 * Writes an error into tracing/error_log of the form:
7367 * <loc>: error: <text>
7371 * tracing/error_log is a small log file containing the last
7372 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7373 * unless there has been a tracing error, and the error log can be
7374 * cleared and have its memory freed by writing the empty string in
7375 * truncation mode to it i.e. echo > tracing/error_log.
7377 * NOTE: the @errs array along with the @type param are used to
7378 * produce a static error string - this string is not copied and saved
7379 * when the error is logged - only a pointer to it is saved. See
7380 * existing callers for examples of how static strings are typically
7381 * defined for use with tracing_log_err().
7383 void tracing_log_err(struct trace_array *tr,
7384 const char *loc, const char *cmd,
7385 const char **errs, u8 type, u8 pos)
7387 struct tracing_log_err *err;
7392 mutex_lock(&tracing_err_log_lock);
7393 err = get_tracing_log_err(tr);
7394 if (PTR_ERR(err) == -ENOMEM) {
7395 mutex_unlock(&tracing_err_log_lock);
7399 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7400 snprintf(err->cmd, MAX_FILTER_STR_VAL,"\n" CMD_PREFIX "%s\n", cmd);
7402 err->info.errs = errs;
7403 err->info.type = type;
7404 err->info.pos = pos;
7405 err->info.ts = local_clock();
7407 list_add_tail(&err->list, &tr->err_log);
7408 mutex_unlock(&tracing_err_log_lock);
7411 static void clear_tracing_err_log(struct trace_array *tr)
7413 struct tracing_log_err *err, *next;
7415 mutex_lock(&tracing_err_log_lock);
7416 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7417 list_del(&err->list);
7421 tr->n_err_log_entries = 0;
7422 mutex_unlock(&tracing_err_log_lock);
7425 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7427 struct trace_array *tr = m->private;
7429 mutex_lock(&tracing_err_log_lock);
7431 return seq_list_start(&tr->err_log, *pos);
7434 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7436 struct trace_array *tr = m->private;
7438 return seq_list_next(v, &tr->err_log, pos);
7441 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7443 mutex_unlock(&tracing_err_log_lock);
7446 static void tracing_err_log_show_pos(struct seq_file *m, u8 pos)
7450 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7452 for (i = 0; i < pos; i++)
7457 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7459 struct tracing_log_err *err = v;
7462 const char *err_text = err->info.errs[err->info.type];
7463 u64 sec = err->info.ts;
7466 nsec = do_div(sec, NSEC_PER_SEC);
7467 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7468 err->loc, err_text);
7469 seq_printf(m, "%s", err->cmd);
7470 tracing_err_log_show_pos(m, err->info.pos);
7476 static const struct seq_operations tracing_err_log_seq_ops = {
7477 .start = tracing_err_log_seq_start,
7478 .next = tracing_err_log_seq_next,
7479 .stop = tracing_err_log_seq_stop,
7480 .show = tracing_err_log_seq_show
7483 static int tracing_err_log_open(struct inode *inode, struct file *file)
7485 struct trace_array *tr = inode->i_private;
7488 ret = tracing_check_open_get_tr(tr);
7492 /* If this file was opened for write, then erase contents */
7493 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7494 clear_tracing_err_log(tr);
7496 if (file->f_mode & FMODE_READ) {
7497 ret = seq_open(file, &tracing_err_log_seq_ops);
7499 struct seq_file *m = file->private_data;
7502 trace_array_put(tr);
7508 static ssize_t tracing_err_log_write(struct file *file,
7509 const char __user *buffer,
7510 size_t count, loff_t *ppos)
7515 static int tracing_err_log_release(struct inode *inode, struct file *file)
7517 struct trace_array *tr = inode->i_private;
7519 trace_array_put(tr);
7521 if (file->f_mode & FMODE_READ)
7522 seq_release(inode, file);
7527 static const struct file_operations tracing_err_log_fops = {
7528 .open = tracing_err_log_open,
7529 .write = tracing_err_log_write,
7531 .llseek = seq_lseek,
7532 .release = tracing_err_log_release,
7535 static int tracing_buffers_open(struct inode *inode, struct file *filp)
7537 struct trace_array *tr = inode->i_private;
7538 struct ftrace_buffer_info *info;
7541 ret = tracing_check_open_get_tr(tr);
7545 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7547 trace_array_put(tr);
7551 mutex_lock(&trace_types_lock);
7554 info->iter.cpu_file = tracing_get_cpu(inode);
7555 info->iter.trace = tr->current_trace;
7556 info->iter.array_buffer = &tr->array_buffer;
7558 /* Force reading ring buffer for first read */
7559 info->read = (unsigned int)-1;
7561 filp->private_data = info;
7565 mutex_unlock(&trace_types_lock);
7567 ret = nonseekable_open(inode, filp);
7569 trace_array_put(tr);
7575 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7577 struct ftrace_buffer_info *info = filp->private_data;
7578 struct trace_iterator *iter = &info->iter;
7580 return trace_poll(iter, filp, poll_table);
7584 tracing_buffers_read(struct file *filp, char __user *ubuf,
7585 size_t count, loff_t *ppos)
7587 struct ftrace_buffer_info *info = filp->private_data;
7588 struct trace_iterator *iter = &info->iter;
7595 #ifdef CONFIG_TRACER_MAX_TRACE
7596 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7601 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
7603 if (IS_ERR(info->spare)) {
7604 ret = PTR_ERR(info->spare);
7607 info->spare_cpu = iter->cpu_file;
7613 /* Do we have previous read data to read? */
7614 if (info->read < PAGE_SIZE)
7618 trace_access_lock(iter->cpu_file);
7619 ret = ring_buffer_read_page(iter->array_buffer->buffer,
7623 trace_access_unlock(iter->cpu_file);
7626 if (trace_empty(iter)) {
7627 if ((filp->f_flags & O_NONBLOCK))
7630 ret = wait_on_pipe(iter, 0);
7641 size = PAGE_SIZE - info->read;
7645 ret = copy_to_user(ubuf, info->spare + info->read, size);
7657 static int tracing_buffers_release(struct inode *inode, struct file *file)
7659 struct ftrace_buffer_info *info = file->private_data;
7660 struct trace_iterator *iter = &info->iter;
7662 mutex_lock(&trace_types_lock);
7664 iter->tr->trace_ref--;
7666 __trace_array_put(iter->tr);
7669 ring_buffer_free_read_page(iter->array_buffer->buffer,
7670 info->spare_cpu, info->spare);
7673 mutex_unlock(&trace_types_lock);
7679 struct trace_buffer *buffer;
7682 refcount_t refcount;
7685 static void buffer_ref_release(struct buffer_ref *ref)
7687 if (!refcount_dec_and_test(&ref->refcount))
7689 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
7693 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
7694 struct pipe_buffer *buf)
7696 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
7698 buffer_ref_release(ref);
7702 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
7703 struct pipe_buffer *buf)
7705 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
7707 if (refcount_read(&ref->refcount) > INT_MAX/2)
7710 refcount_inc(&ref->refcount);
7714 /* Pipe buffer operations for a buffer. */
7715 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
7716 .release = buffer_pipe_buf_release,
7717 .get = buffer_pipe_buf_get,
7721 * Callback from splice_to_pipe(), if we need to release some pages
7722 * at the end of the spd in case we error'ed out in filling the pipe.
7724 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
7726 struct buffer_ref *ref =
7727 (struct buffer_ref *)spd->partial[i].private;
7729 buffer_ref_release(ref);
7730 spd->partial[i].private = 0;
7734 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7735 struct pipe_inode_info *pipe, size_t len,
7738 struct ftrace_buffer_info *info = file->private_data;
7739 struct trace_iterator *iter = &info->iter;
7740 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7741 struct page *pages_def[PIPE_DEF_BUFFERS];
7742 struct splice_pipe_desc spd = {
7744 .partial = partial_def,
7745 .nr_pages_max = PIPE_DEF_BUFFERS,
7746 .ops = &buffer_pipe_buf_ops,
7747 .spd_release = buffer_spd_release,
7749 struct buffer_ref *ref;
7753 #ifdef CONFIG_TRACER_MAX_TRACE
7754 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7758 if (*ppos & (PAGE_SIZE - 1))
7761 if (len & (PAGE_SIZE - 1)) {
7762 if (len < PAGE_SIZE)
7767 if (splice_grow_spd(pipe, &spd))
7771 trace_access_lock(iter->cpu_file);
7772 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
7774 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
7778 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
7784 refcount_set(&ref->refcount, 1);
7785 ref->buffer = iter->array_buffer->buffer;
7786 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
7787 if (IS_ERR(ref->page)) {
7788 ret = PTR_ERR(ref->page);
7793 ref->cpu = iter->cpu_file;
7795 r = ring_buffer_read_page(ref->buffer, &ref->page,
7796 len, iter->cpu_file, 1);
7798 ring_buffer_free_read_page(ref->buffer, ref->cpu,
7804 page = virt_to_page(ref->page);
7806 spd.pages[i] = page;
7807 spd.partial[i].len = PAGE_SIZE;
7808 spd.partial[i].offset = 0;
7809 spd.partial[i].private = (unsigned long)ref;
7813 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
7816 trace_access_unlock(iter->cpu_file);
7819 /* did we read anything? */
7820 if (!spd.nr_pages) {
7825 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
7828 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
7835 ret = splice_to_pipe(pipe, &spd);
7837 splice_shrink_spd(&spd);
7842 static const struct file_operations tracing_buffers_fops = {
7843 .open = tracing_buffers_open,
7844 .read = tracing_buffers_read,
7845 .poll = tracing_buffers_poll,
7846 .release = tracing_buffers_release,
7847 .splice_read = tracing_buffers_splice_read,
7848 .llseek = no_llseek,
7852 tracing_stats_read(struct file *filp, char __user *ubuf,
7853 size_t count, loff_t *ppos)
7855 struct inode *inode = file_inode(filp);
7856 struct trace_array *tr = inode->i_private;
7857 struct array_buffer *trace_buf = &tr->array_buffer;
7858 int cpu = tracing_get_cpu(inode);
7859 struct trace_seq *s;
7861 unsigned long long t;
7862 unsigned long usec_rem;
7864 s = kmalloc(sizeof(*s), GFP_KERNEL);
7870 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
7871 trace_seq_printf(s, "entries: %ld\n", cnt);
7873 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
7874 trace_seq_printf(s, "overrun: %ld\n", cnt);
7876 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
7877 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
7879 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
7880 trace_seq_printf(s, "bytes: %ld\n", cnt);
7882 if (trace_clocks[tr->clock_id].in_ns) {
7883 /* local or global for trace_clock */
7884 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
7885 usec_rem = do_div(t, USEC_PER_SEC);
7886 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
7889 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
7890 usec_rem = do_div(t, USEC_PER_SEC);
7891 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
7893 /* counter or tsc mode for trace_clock */
7894 trace_seq_printf(s, "oldest event ts: %llu\n",
7895 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
7897 trace_seq_printf(s, "now ts: %llu\n",
7898 ring_buffer_time_stamp(trace_buf->buffer, cpu));
7901 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
7902 trace_seq_printf(s, "dropped events: %ld\n", cnt);
7904 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
7905 trace_seq_printf(s, "read events: %ld\n", cnt);
7907 count = simple_read_from_buffer(ubuf, count, ppos,
7908 s->buffer, trace_seq_used(s));
7915 static const struct file_operations tracing_stats_fops = {
7916 .open = tracing_open_generic_tr,
7917 .read = tracing_stats_read,
7918 .llseek = generic_file_llseek,
7919 .release = tracing_release_generic_tr,
7922 #ifdef CONFIG_DYNAMIC_FTRACE
7925 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
7926 size_t cnt, loff_t *ppos)
7932 /* 256 should be plenty to hold the amount needed */
7933 buf = kmalloc(256, GFP_KERNEL);
7937 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
7938 ftrace_update_tot_cnt,
7939 ftrace_number_of_pages,
7940 ftrace_number_of_groups);
7942 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7947 static const struct file_operations tracing_dyn_info_fops = {
7948 .open = tracing_open_generic,
7949 .read = tracing_read_dyn_info,
7950 .llseek = generic_file_llseek,
7952 #endif /* CONFIG_DYNAMIC_FTRACE */
7954 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
7956 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
7957 struct trace_array *tr, struct ftrace_probe_ops *ops,
7960 tracing_snapshot_instance(tr);
7964 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
7965 struct trace_array *tr, struct ftrace_probe_ops *ops,
7968 struct ftrace_func_mapper *mapper = data;
7972 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7982 tracing_snapshot_instance(tr);
7986 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
7987 struct ftrace_probe_ops *ops, void *data)
7989 struct ftrace_func_mapper *mapper = data;
7992 seq_printf(m, "%ps:", (void *)ip);
7994 seq_puts(m, "snapshot");
7997 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8000 seq_printf(m, ":count=%ld\n", *count);
8002 seq_puts(m, ":unlimited\n");
8008 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8009 unsigned long ip, void *init_data, void **data)
8011 struct ftrace_func_mapper *mapper = *data;
8014 mapper = allocate_ftrace_func_mapper();
8020 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8024 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8025 unsigned long ip, void *data)
8027 struct ftrace_func_mapper *mapper = data;
8032 free_ftrace_func_mapper(mapper, NULL);
8036 ftrace_func_mapper_remove_ip(mapper, ip);
8039 static struct ftrace_probe_ops snapshot_probe_ops = {
8040 .func = ftrace_snapshot,
8041 .print = ftrace_snapshot_print,
8044 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8045 .func = ftrace_count_snapshot,
8046 .print = ftrace_snapshot_print,
8047 .init = ftrace_snapshot_init,
8048 .free = ftrace_snapshot_free,
8052 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8053 char *glob, char *cmd, char *param, int enable)
8055 struct ftrace_probe_ops *ops;
8056 void *count = (void *)-1;
8063 /* hash funcs only work with set_ftrace_filter */
8067 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8070 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8075 number = strsep(¶m, ":");
8077 if (!strlen(number))
8081 * We use the callback data field (which is a pointer)
8084 ret = kstrtoul(number, 0, (unsigned long *)&count);
8089 ret = tracing_alloc_snapshot_instance(tr);
8093 ret = register_ftrace_function_probe(glob, tr, ops, count);
8096 return ret < 0 ? ret : 0;
8099 static struct ftrace_func_command ftrace_snapshot_cmd = {
8101 .func = ftrace_trace_snapshot_callback,
8104 static __init int register_snapshot_cmd(void)
8106 return register_ftrace_command(&ftrace_snapshot_cmd);
8109 static inline __init int register_snapshot_cmd(void) { return 0; }
8110 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8112 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8114 if (WARN_ON(!tr->dir))
8115 return ERR_PTR(-ENODEV);
8117 /* Top directory uses NULL as the parent */
8118 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8121 /* All sub buffers have a descriptor */
8125 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8127 struct dentry *d_tracer;
8130 return tr->percpu_dir;
8132 d_tracer = tracing_get_dentry(tr);
8133 if (IS_ERR(d_tracer))
8136 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8138 MEM_FAIL(!tr->percpu_dir,
8139 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8141 return tr->percpu_dir;
8144 static struct dentry *
8145 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8146 void *data, long cpu, const struct file_operations *fops)
8148 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8150 if (ret) /* See tracing_get_cpu() */
8151 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8156 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8158 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8159 struct dentry *d_cpu;
8160 char cpu_dir[30]; /* 30 characters should be more than enough */
8165 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8166 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8168 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8172 /* per cpu trace_pipe */
8173 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
8174 tr, cpu, &tracing_pipe_fops);
8177 trace_create_cpu_file("trace", 0644, d_cpu,
8178 tr, cpu, &tracing_fops);
8180 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
8181 tr, cpu, &tracing_buffers_fops);
8183 trace_create_cpu_file("stats", 0444, d_cpu,
8184 tr, cpu, &tracing_stats_fops);
8186 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
8187 tr, cpu, &tracing_entries_fops);
8189 #ifdef CONFIG_TRACER_SNAPSHOT
8190 trace_create_cpu_file("snapshot", 0644, d_cpu,
8191 tr, cpu, &snapshot_fops);
8193 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
8194 tr, cpu, &snapshot_raw_fops);
8198 #ifdef CONFIG_FTRACE_SELFTEST
8199 /* Let selftest have access to static functions in this file */
8200 #include "trace_selftest.c"
8204 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8207 struct trace_option_dentry *topt = filp->private_data;
8210 if (topt->flags->val & topt->opt->bit)
8215 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8219 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8222 struct trace_option_dentry *topt = filp->private_data;
8226 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8230 if (val != 0 && val != 1)
8233 if (!!(topt->flags->val & topt->opt->bit) != val) {
8234 mutex_lock(&trace_types_lock);
8235 ret = __set_tracer_option(topt->tr, topt->flags,
8237 mutex_unlock(&trace_types_lock);
8248 static const struct file_operations trace_options_fops = {
8249 .open = tracing_open_generic,
8250 .read = trace_options_read,
8251 .write = trace_options_write,
8252 .llseek = generic_file_llseek,
8256 * In order to pass in both the trace_array descriptor as well as the index
8257 * to the flag that the trace option file represents, the trace_array
8258 * has a character array of trace_flags_index[], which holds the index
8259 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8260 * The address of this character array is passed to the flag option file
8261 * read/write callbacks.
8263 * In order to extract both the index and the trace_array descriptor,
8264 * get_tr_index() uses the following algorithm.
8268 * As the pointer itself contains the address of the index (remember
8271 * Then to get the trace_array descriptor, by subtracting that index
8272 * from the ptr, we get to the start of the index itself.
8274 * ptr - idx == &index[0]
8276 * Then a simple container_of() from that pointer gets us to the
8277 * trace_array descriptor.
8279 static void get_tr_index(void *data, struct trace_array **ptr,
8280 unsigned int *pindex)
8282 *pindex = *(unsigned char *)data;
8284 *ptr = container_of(data - *pindex, struct trace_array,
8289 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8292 void *tr_index = filp->private_data;
8293 struct trace_array *tr;
8297 get_tr_index(tr_index, &tr, &index);
8299 if (tr->trace_flags & (1 << index))
8304 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8308 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8311 void *tr_index = filp->private_data;
8312 struct trace_array *tr;
8317 get_tr_index(tr_index, &tr, &index);
8319 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8323 if (val != 0 && val != 1)
8326 mutex_lock(&event_mutex);
8327 mutex_lock(&trace_types_lock);
8328 ret = set_tracer_flag(tr, 1 << index, val);
8329 mutex_unlock(&trace_types_lock);
8330 mutex_unlock(&event_mutex);
8340 static const struct file_operations trace_options_core_fops = {
8341 .open = tracing_open_generic,
8342 .read = trace_options_core_read,
8343 .write = trace_options_core_write,
8344 .llseek = generic_file_llseek,
8347 struct dentry *trace_create_file(const char *name,
8349 struct dentry *parent,
8351 const struct file_operations *fops)
8355 ret = tracefs_create_file(name, mode, parent, data, fops);
8357 pr_warn("Could not create tracefs '%s' entry\n", name);
8363 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8365 struct dentry *d_tracer;
8370 d_tracer = tracing_get_dentry(tr);
8371 if (IS_ERR(d_tracer))
8374 tr->options = tracefs_create_dir("options", d_tracer);
8376 pr_warn("Could not create tracefs directory 'options'\n");
8384 create_trace_option_file(struct trace_array *tr,
8385 struct trace_option_dentry *topt,
8386 struct tracer_flags *flags,
8387 struct tracer_opt *opt)
8389 struct dentry *t_options;
8391 t_options = trace_options_init_dentry(tr);
8395 topt->flags = flags;
8399 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
8400 &trace_options_fops);
8405 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8407 struct trace_option_dentry *topts;
8408 struct trace_options *tr_topts;
8409 struct tracer_flags *flags;
8410 struct tracer_opt *opts;
8417 flags = tracer->flags;
8419 if (!flags || !flags->opts)
8423 * If this is an instance, only create flags for tracers
8424 * the instance may have.
8426 if (!trace_ok_for_array(tracer, tr))
8429 for (i = 0; i < tr->nr_topts; i++) {
8430 /* Make sure there's no duplicate flags. */
8431 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8437 for (cnt = 0; opts[cnt].name; cnt++)
8440 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8444 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8451 tr->topts = tr_topts;
8452 tr->topts[tr->nr_topts].tracer = tracer;
8453 tr->topts[tr->nr_topts].topts = topts;
8456 for (cnt = 0; opts[cnt].name; cnt++) {
8457 create_trace_option_file(tr, &topts[cnt], flags,
8459 MEM_FAIL(topts[cnt].entry == NULL,
8460 "Failed to create trace option: %s",
8465 static struct dentry *
8466 create_trace_option_core_file(struct trace_array *tr,
8467 const char *option, long index)
8469 struct dentry *t_options;
8471 t_options = trace_options_init_dentry(tr);
8475 return trace_create_file(option, 0644, t_options,
8476 (void *)&tr->trace_flags_index[index],
8477 &trace_options_core_fops);
8480 static void create_trace_options_dir(struct trace_array *tr)
8482 struct dentry *t_options;
8483 bool top_level = tr == &global_trace;
8486 t_options = trace_options_init_dentry(tr);
8490 for (i = 0; trace_options[i]; i++) {
8492 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8493 create_trace_option_core_file(tr, trace_options[i], i);
8498 rb_simple_read(struct file *filp, char __user *ubuf,
8499 size_t cnt, loff_t *ppos)
8501 struct trace_array *tr = filp->private_data;
8505 r = tracer_tracing_is_on(tr);
8506 r = sprintf(buf, "%d\n", r);
8508 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8512 rb_simple_write(struct file *filp, const char __user *ubuf,
8513 size_t cnt, loff_t *ppos)
8515 struct trace_array *tr = filp->private_data;
8516 struct trace_buffer *buffer = tr->array_buffer.buffer;
8520 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8525 mutex_lock(&trace_types_lock);
8526 if (!!val == tracer_tracing_is_on(tr)) {
8527 val = 0; /* do nothing */
8529 tracer_tracing_on(tr);
8530 if (tr->current_trace->start)
8531 tr->current_trace->start(tr);
8533 tracer_tracing_off(tr);
8534 if (tr->current_trace->stop)
8535 tr->current_trace->stop(tr);
8537 mutex_unlock(&trace_types_lock);
8545 static const struct file_operations rb_simple_fops = {
8546 .open = tracing_open_generic_tr,
8547 .read = rb_simple_read,
8548 .write = rb_simple_write,
8549 .release = tracing_release_generic_tr,
8550 .llseek = default_llseek,
8554 buffer_percent_read(struct file *filp, char __user *ubuf,
8555 size_t cnt, loff_t *ppos)
8557 struct trace_array *tr = filp->private_data;
8561 r = tr->buffer_percent;
8562 r = sprintf(buf, "%d\n", r);
8564 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8568 buffer_percent_write(struct file *filp, const char __user *ubuf,
8569 size_t cnt, loff_t *ppos)
8571 struct trace_array *tr = filp->private_data;
8575 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8585 tr->buffer_percent = val;
8592 static const struct file_operations buffer_percent_fops = {
8593 .open = tracing_open_generic_tr,
8594 .read = buffer_percent_read,
8595 .write = buffer_percent_write,
8596 .release = tracing_release_generic_tr,
8597 .llseek = default_llseek,
8600 static struct dentry *trace_instance_dir;
8603 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
8606 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
8608 enum ring_buffer_flags rb_flags;
8610 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
8614 buf->buffer = ring_buffer_alloc(size, rb_flags);
8618 buf->data = alloc_percpu(struct trace_array_cpu);
8620 ring_buffer_free(buf->buffer);
8625 /* Allocate the first page for all buffers */
8626 set_buffer_entries(&tr->array_buffer,
8627 ring_buffer_size(tr->array_buffer.buffer, 0));
8632 static int allocate_trace_buffers(struct trace_array *tr, int size)
8636 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
8640 #ifdef CONFIG_TRACER_MAX_TRACE
8641 ret = allocate_trace_buffer(tr, &tr->max_buffer,
8642 allocate_snapshot ? size : 1);
8643 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
8644 ring_buffer_free(tr->array_buffer.buffer);
8645 tr->array_buffer.buffer = NULL;
8646 free_percpu(tr->array_buffer.data);
8647 tr->array_buffer.data = NULL;
8650 tr->allocated_snapshot = allocate_snapshot;
8653 * Only the top level trace array gets its snapshot allocated
8654 * from the kernel command line.
8656 allocate_snapshot = false;
8662 static void free_trace_buffer(struct array_buffer *buf)
8665 ring_buffer_free(buf->buffer);
8667 free_percpu(buf->data);
8672 static void free_trace_buffers(struct trace_array *tr)
8677 free_trace_buffer(&tr->array_buffer);
8679 #ifdef CONFIG_TRACER_MAX_TRACE
8680 free_trace_buffer(&tr->max_buffer);
8684 static void init_trace_flags_index(struct trace_array *tr)
8688 /* Used by the trace options files */
8689 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
8690 tr->trace_flags_index[i] = i;
8693 static void __update_tracer_options(struct trace_array *tr)
8697 for (t = trace_types; t; t = t->next)
8698 add_tracer_options(tr, t);
8701 static void update_tracer_options(struct trace_array *tr)
8703 mutex_lock(&trace_types_lock);
8704 __update_tracer_options(tr);
8705 mutex_unlock(&trace_types_lock);
8708 /* Must have trace_types_lock held */
8709 struct trace_array *trace_array_find(const char *instance)
8711 struct trace_array *tr, *found = NULL;
8713 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8714 if (tr->name && strcmp(tr->name, instance) == 0) {
8723 struct trace_array *trace_array_find_get(const char *instance)
8725 struct trace_array *tr;
8727 mutex_lock(&trace_types_lock);
8728 tr = trace_array_find(instance);
8731 mutex_unlock(&trace_types_lock);
8736 static int trace_array_create_dir(struct trace_array *tr)
8740 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
8744 ret = event_trace_add_tracer(tr->dir, tr);
8746 tracefs_remove(tr->dir);
8748 init_tracer_tracefs(tr, tr->dir);
8749 __update_tracer_options(tr);
8754 static struct trace_array *trace_array_create(const char *name)
8756 struct trace_array *tr;
8760 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
8762 return ERR_PTR(ret);
8764 tr->name = kstrdup(name, GFP_KERNEL);
8768 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
8771 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
8773 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
8775 raw_spin_lock_init(&tr->start_lock);
8777 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
8779 tr->current_trace = &nop_trace;
8781 INIT_LIST_HEAD(&tr->systems);
8782 INIT_LIST_HEAD(&tr->events);
8783 INIT_LIST_HEAD(&tr->hist_vars);
8784 INIT_LIST_HEAD(&tr->err_log);
8786 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
8789 if (ftrace_allocate_ftrace_ops(tr) < 0)
8792 ftrace_init_trace_array(tr);
8794 init_trace_flags_index(tr);
8796 if (trace_instance_dir) {
8797 ret = trace_array_create_dir(tr);
8801 __trace_early_add_events(tr);
8803 list_add(&tr->list, &ftrace_trace_arrays);
8810 ftrace_free_ftrace_ops(tr);
8811 free_trace_buffers(tr);
8812 free_cpumask_var(tr->tracing_cpumask);
8816 return ERR_PTR(ret);
8819 static int instance_mkdir(const char *name)
8821 struct trace_array *tr;
8824 mutex_lock(&event_mutex);
8825 mutex_lock(&trace_types_lock);
8828 if (trace_array_find(name))
8831 tr = trace_array_create(name);
8833 ret = PTR_ERR_OR_ZERO(tr);
8836 mutex_unlock(&trace_types_lock);
8837 mutex_unlock(&event_mutex);
8842 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
8843 * @name: The name of the trace array to be looked up/created.
8845 * Returns pointer to trace array with given name.
8846 * NULL, if it cannot be created.
8848 * NOTE: This function increments the reference counter associated with the
8849 * trace array returned. This makes sure it cannot be freed while in use.
8850 * Use trace_array_put() once the trace array is no longer needed.
8851 * If the trace_array is to be freed, trace_array_destroy() needs to
8852 * be called after the trace_array_put(), or simply let user space delete
8853 * it from the tracefs instances directory. But until the
8854 * trace_array_put() is called, user space can not delete it.
8857 struct trace_array *trace_array_get_by_name(const char *name)
8859 struct trace_array *tr;
8861 mutex_lock(&event_mutex);
8862 mutex_lock(&trace_types_lock);
8864 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8865 if (tr->name && strcmp(tr->name, name) == 0)
8869 tr = trace_array_create(name);
8877 mutex_unlock(&trace_types_lock);
8878 mutex_unlock(&event_mutex);
8881 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
8883 static int __remove_instance(struct trace_array *tr)
8887 /* Reference counter for a newly created trace array = 1. */
8888 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
8891 list_del(&tr->list);
8893 /* Disable all the flags that were enabled coming in */
8894 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
8895 if ((1 << i) & ZEROED_TRACE_FLAGS)
8896 set_tracer_flag(tr, 1 << i, 0);
8899 tracing_set_nop(tr);
8900 clear_ftrace_function_probes(tr);
8901 event_trace_del_tracer(tr);
8902 ftrace_clear_pids(tr);
8903 ftrace_destroy_function_files(tr);
8904 tracefs_remove(tr->dir);
8905 free_trace_buffers(tr);
8907 for (i = 0; i < tr->nr_topts; i++) {
8908 kfree(tr->topts[i].topts);
8912 free_cpumask_var(tr->tracing_cpumask);
8919 int trace_array_destroy(struct trace_array *this_tr)
8921 struct trace_array *tr;
8927 mutex_lock(&event_mutex);
8928 mutex_lock(&trace_types_lock);
8932 /* Making sure trace array exists before destroying it. */
8933 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8934 if (tr == this_tr) {
8935 ret = __remove_instance(tr);
8940 mutex_unlock(&trace_types_lock);
8941 mutex_unlock(&event_mutex);
8945 EXPORT_SYMBOL_GPL(trace_array_destroy);
8947 static int instance_rmdir(const char *name)
8949 struct trace_array *tr;
8952 mutex_lock(&event_mutex);
8953 mutex_lock(&trace_types_lock);
8956 tr = trace_array_find(name);
8958 ret = __remove_instance(tr);
8960 mutex_unlock(&trace_types_lock);
8961 mutex_unlock(&event_mutex);
8966 static __init void create_trace_instances(struct dentry *d_tracer)
8968 struct trace_array *tr;
8970 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
8973 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
8976 mutex_lock(&event_mutex);
8977 mutex_lock(&trace_types_lock);
8979 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8982 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
8983 "Failed to create instance directory\n"))
8987 mutex_unlock(&trace_types_lock);
8988 mutex_unlock(&event_mutex);
8992 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
8994 struct trace_event_file *file;
8997 trace_create_file("available_tracers", 0444, d_tracer,
8998 tr, &show_traces_fops);
9000 trace_create_file("current_tracer", 0644, d_tracer,
9001 tr, &set_tracer_fops);
9003 trace_create_file("tracing_cpumask", 0644, d_tracer,
9004 tr, &tracing_cpumask_fops);
9006 trace_create_file("trace_options", 0644, d_tracer,
9007 tr, &tracing_iter_fops);
9009 trace_create_file("trace", 0644, d_tracer,
9012 trace_create_file("trace_pipe", 0444, d_tracer,
9013 tr, &tracing_pipe_fops);
9015 trace_create_file("buffer_size_kb", 0644, d_tracer,
9016 tr, &tracing_entries_fops);
9018 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
9019 tr, &tracing_total_entries_fops);
9021 trace_create_file("free_buffer", 0200, d_tracer,
9022 tr, &tracing_free_buffer_fops);
9024 trace_create_file("trace_marker", 0220, d_tracer,
9025 tr, &tracing_mark_fops);
9027 file = __find_event_file(tr, "ftrace", "print");
9028 if (file && file->dir)
9029 trace_create_file("trigger", 0644, file->dir, file,
9030 &event_trigger_fops);
9031 tr->trace_marker_file = file;
9033 trace_create_file("trace_marker_raw", 0220, d_tracer,
9034 tr, &tracing_mark_raw_fops);
9036 trace_create_file("trace_clock", 0644, d_tracer, tr,
9039 trace_create_file("tracing_on", 0644, d_tracer,
9040 tr, &rb_simple_fops);
9042 trace_create_file("timestamp_mode", 0444, d_tracer, tr,
9043 &trace_time_stamp_mode_fops);
9045 tr->buffer_percent = 50;
9047 trace_create_file("buffer_percent", 0444, d_tracer,
9048 tr, &buffer_percent_fops);
9050 create_trace_options_dir(tr);
9052 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
9053 trace_create_maxlat_file(tr, d_tracer);
9056 if (ftrace_create_function_files(tr, d_tracer))
9057 MEM_FAIL(1, "Could not allocate function filter files");
9059 #ifdef CONFIG_TRACER_SNAPSHOT
9060 trace_create_file("snapshot", 0644, d_tracer,
9061 tr, &snapshot_fops);
9064 trace_create_file("error_log", 0644, d_tracer,
9065 tr, &tracing_err_log_fops);
9067 for_each_tracing_cpu(cpu)
9068 tracing_init_tracefs_percpu(tr, cpu);
9070 ftrace_init_tracefs(tr, d_tracer);
9073 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9075 struct vfsmount *mnt;
9076 struct file_system_type *type;
9079 * To maintain backward compatibility for tools that mount
9080 * debugfs to get to the tracing facility, tracefs is automatically
9081 * mounted to the debugfs/tracing directory.
9083 type = get_fs_type("tracefs");
9086 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9087 put_filesystem(type);
9096 * tracing_init_dentry - initialize top level trace array
9098 * This is called when creating files or directories in the tracing
9099 * directory. It is called via fs_initcall() by any of the boot up code
9100 * and expects to return the dentry of the top level tracing directory.
9102 int tracing_init_dentry(void)
9104 struct trace_array *tr = &global_trace;
9106 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9107 pr_warn("Tracing disabled due to lockdown\n");
9111 /* The top level trace array uses NULL as parent */
9115 if (WARN_ON(!tracefs_initialized()))
9119 * As there may still be users that expect the tracing
9120 * files to exist in debugfs/tracing, we must automount
9121 * the tracefs file system there, so older tools still
9122 * work with the newer kerenl.
9124 tr->dir = debugfs_create_automount("tracing", NULL,
9125 trace_automount, NULL);
9130 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9131 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9133 static struct workqueue_struct *eval_map_wq __initdata;
9134 static struct work_struct eval_map_work __initdata;
9136 static void __init eval_map_work_func(struct work_struct *work)
9140 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9141 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9144 static int __init trace_eval_init(void)
9146 INIT_WORK(&eval_map_work, eval_map_work_func);
9148 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9150 pr_err("Unable to allocate eval_map_wq\n");
9152 eval_map_work_func(&eval_map_work);
9156 queue_work(eval_map_wq, &eval_map_work);
9160 static int __init trace_eval_sync(void)
9162 /* Make sure the eval map updates are finished */
9164 destroy_workqueue(eval_map_wq);
9168 late_initcall_sync(trace_eval_sync);
9171 #ifdef CONFIG_MODULES
9172 static void trace_module_add_evals(struct module *mod)
9174 if (!mod->num_trace_evals)
9178 * Modules with bad taint do not have events created, do
9179 * not bother with enums either.
9181 if (trace_module_has_bad_taint(mod))
9184 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9187 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
9188 static void trace_module_remove_evals(struct module *mod)
9190 union trace_eval_map_item *map;
9191 union trace_eval_map_item **last = &trace_eval_maps;
9193 if (!mod->num_trace_evals)
9196 mutex_lock(&trace_eval_mutex);
9198 map = trace_eval_maps;
9201 if (map->head.mod == mod)
9203 map = trace_eval_jmp_to_tail(map);
9204 last = &map->tail.next;
9205 map = map->tail.next;
9210 *last = trace_eval_jmp_to_tail(map)->tail.next;
9213 mutex_unlock(&trace_eval_mutex);
9216 static inline void trace_module_remove_evals(struct module *mod) { }
9217 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9219 static int trace_module_notify(struct notifier_block *self,
9220 unsigned long val, void *data)
9222 struct module *mod = data;
9225 case MODULE_STATE_COMING:
9226 trace_module_add_evals(mod);
9228 case MODULE_STATE_GOING:
9229 trace_module_remove_evals(mod);
9236 static struct notifier_block trace_module_nb = {
9237 .notifier_call = trace_module_notify,
9240 #endif /* CONFIG_MODULES */
9242 static __init int tracer_init_tracefs(void)
9246 trace_access_lock_init();
9248 ret = tracing_init_dentry();
9254 init_tracer_tracefs(&global_trace, NULL);
9255 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9257 trace_create_file("tracing_thresh", 0644, NULL,
9258 &global_trace, &tracing_thresh_fops);
9260 trace_create_file("README", 0444, NULL,
9261 NULL, &tracing_readme_fops);
9263 trace_create_file("saved_cmdlines", 0444, NULL,
9264 NULL, &tracing_saved_cmdlines_fops);
9266 trace_create_file("saved_cmdlines_size", 0644, NULL,
9267 NULL, &tracing_saved_cmdlines_size_fops);
9269 trace_create_file("saved_tgids", 0444, NULL,
9270 NULL, &tracing_saved_tgids_fops);
9274 trace_create_eval_file(NULL);
9276 #ifdef CONFIG_MODULES
9277 register_module_notifier(&trace_module_nb);
9280 #ifdef CONFIG_DYNAMIC_FTRACE
9281 trace_create_file("dyn_ftrace_total_info", 0444, NULL,
9282 NULL, &tracing_dyn_info_fops);
9285 create_trace_instances(NULL);
9287 update_tracer_options(&global_trace);
9292 static int trace_panic_handler(struct notifier_block *this,
9293 unsigned long event, void *unused)
9295 if (ftrace_dump_on_oops)
9296 ftrace_dump(ftrace_dump_on_oops);
9300 static struct notifier_block trace_panic_notifier = {
9301 .notifier_call = trace_panic_handler,
9303 .priority = 150 /* priority: INT_MAX >= x >= 0 */
9306 static int trace_die_handler(struct notifier_block *self,
9312 if (ftrace_dump_on_oops)
9313 ftrace_dump(ftrace_dump_on_oops);
9321 static struct notifier_block trace_die_notifier = {
9322 .notifier_call = trace_die_handler,
9327 * printk is set to max of 1024, we really don't need it that big.
9328 * Nothing should be printing 1000 characters anyway.
9330 #define TRACE_MAX_PRINT 1000
9333 * Define here KERN_TRACE so that we have one place to modify
9334 * it if we decide to change what log level the ftrace dump
9337 #define KERN_TRACE KERN_EMERG
9340 trace_printk_seq(struct trace_seq *s)
9342 /* Probably should print a warning here. */
9343 if (s->seq.len >= TRACE_MAX_PRINT)
9344 s->seq.len = TRACE_MAX_PRINT;
9347 * More paranoid code. Although the buffer size is set to
9348 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9349 * an extra layer of protection.
9351 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9352 s->seq.len = s->seq.size - 1;
9354 /* should be zero ended, but we are paranoid. */
9355 s->buffer[s->seq.len] = 0;
9357 printk(KERN_TRACE "%s", s->buffer);
9362 void trace_init_global_iter(struct trace_iterator *iter)
9364 iter->tr = &global_trace;
9365 iter->trace = iter->tr->current_trace;
9366 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9367 iter->array_buffer = &global_trace.array_buffer;
9369 if (iter->trace && iter->trace->open)
9370 iter->trace->open(iter);
9372 /* Annotate start of buffers if we had overruns */
9373 if (ring_buffer_overruns(iter->array_buffer->buffer))
9374 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9376 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9377 if (trace_clocks[iter->tr->clock_id].in_ns)
9378 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9381 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
9383 /* use static because iter can be a bit big for the stack */
9384 static struct trace_iterator iter;
9385 static atomic_t dump_running;
9386 struct trace_array *tr = &global_trace;
9387 unsigned int old_userobj;
9388 unsigned long flags;
9391 /* Only allow one dump user at a time. */
9392 if (atomic_inc_return(&dump_running) != 1) {
9393 atomic_dec(&dump_running);
9398 * Always turn off tracing when we dump.
9399 * We don't need to show trace output of what happens
9400 * between multiple crashes.
9402 * If the user does a sysrq-z, then they can re-enable
9403 * tracing with echo 1 > tracing_on.
9407 local_irq_save(flags);
9408 printk_nmi_direct_enter();
9410 /* Simulate the iterator */
9411 trace_init_global_iter(&iter);
9412 /* Can not use kmalloc for iter.temp and iter.fmt */
9413 iter.temp = static_temp_buf;
9414 iter.temp_size = STATIC_TEMP_BUF_SIZE;
9415 iter.fmt = static_fmt_buf;
9416 iter.fmt_size = STATIC_FMT_BUF_SIZE;
9418 for_each_tracing_cpu(cpu) {
9419 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9422 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
9424 /* don't look at user memory in panic mode */
9425 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
9427 switch (oops_dump_mode) {
9429 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9432 iter.cpu_file = raw_smp_processor_id();
9437 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
9438 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9441 printk(KERN_TRACE "Dumping ftrace buffer:\n");
9443 /* Did function tracer already get disabled? */
9444 if (ftrace_is_dead()) {
9445 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
9446 printk("# MAY BE MISSING FUNCTION EVENTS\n");
9450 * We need to stop all tracing on all CPUS to read
9451 * the next buffer. This is a bit expensive, but is
9452 * not done often. We fill all what we can read,
9453 * and then release the locks again.
9456 while (!trace_empty(&iter)) {
9459 printk(KERN_TRACE "---------------------------------\n");
9463 trace_iterator_reset(&iter);
9464 iter.iter_flags |= TRACE_FILE_LAT_FMT;
9466 if (trace_find_next_entry_inc(&iter) != NULL) {
9469 ret = print_trace_line(&iter);
9470 if (ret != TRACE_TYPE_NO_CONSUME)
9471 trace_consume(&iter);
9473 touch_nmi_watchdog();
9475 trace_printk_seq(&iter.seq);
9479 printk(KERN_TRACE " (ftrace buffer empty)\n");
9481 printk(KERN_TRACE "---------------------------------\n");
9484 tr->trace_flags |= old_userobj;
9486 for_each_tracing_cpu(cpu) {
9487 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9489 atomic_dec(&dump_running);
9490 printk_nmi_direct_exit();
9491 local_irq_restore(flags);
9493 EXPORT_SYMBOL_GPL(ftrace_dump);
9495 #define WRITE_BUFSIZE 4096
9497 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
9498 size_t count, loff_t *ppos,
9499 int (*createfn)(const char *))
9501 char *kbuf, *buf, *tmp;
9506 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
9510 while (done < count) {
9511 size = count - done;
9513 if (size >= WRITE_BUFSIZE)
9514 size = WRITE_BUFSIZE - 1;
9516 if (copy_from_user(kbuf, buffer + done, size)) {
9523 tmp = strchr(buf, '\n');
9526 size = tmp - buf + 1;
9529 if (done + size < count) {
9532 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
9533 pr_warn("Line length is too long: Should be less than %d\n",
9541 /* Remove comments */
9542 tmp = strchr(buf, '#');
9547 ret = createfn(buf);
9552 } while (done < count);
9562 __init static int tracer_alloc_buffers(void)
9568 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9569 pr_warn("Tracing disabled due to lockdown\n");
9574 * Make sure we don't accidentally add more trace options
9575 * than we have bits for.
9577 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
9579 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
9582 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
9583 goto out_free_buffer_mask;
9585 /* Only allocate trace_printk buffers if a trace_printk exists */
9586 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
9587 /* Must be called before global_trace.buffer is allocated */
9588 trace_printk_init_buffers();
9590 /* To save memory, keep the ring buffer size to its minimum */
9591 if (ring_buffer_expanded)
9592 ring_buf_size = trace_buf_size;
9596 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
9597 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
9599 raw_spin_lock_init(&global_trace.start_lock);
9602 * The prepare callbacks allocates some memory for the ring buffer. We
9603 * don't free the buffer if the CPU goes down. If we were to free
9604 * the buffer, then the user would lose any trace that was in the
9605 * buffer. The memory will be removed once the "instance" is removed.
9607 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
9608 "trace/RB:preapre", trace_rb_cpu_prepare,
9611 goto out_free_cpumask;
9612 /* Used for event triggers */
9614 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
9616 goto out_rm_hp_state;
9618 if (trace_create_savedcmd() < 0)
9619 goto out_free_temp_buffer;
9621 /* TODO: make the number of buffers hot pluggable with CPUS */
9622 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
9623 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
9624 goto out_free_savedcmd;
9627 if (global_trace.buffer_disabled)
9630 if (trace_boot_clock) {
9631 ret = tracing_set_clock(&global_trace, trace_boot_clock);
9633 pr_warn("Trace clock %s not defined, going back to default\n",
9638 * register_tracer() might reference current_trace, so it
9639 * needs to be set before we register anything. This is
9640 * just a bootstrap of current_trace anyway.
9642 global_trace.current_trace = &nop_trace;
9644 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9646 ftrace_init_global_array_ops(&global_trace);
9648 init_trace_flags_index(&global_trace);
9650 register_tracer(&nop_trace);
9652 /* Function tracing may start here (via kernel command line) */
9653 init_function_trace();
9655 /* All seems OK, enable tracing */
9656 tracing_disabled = 0;
9658 atomic_notifier_chain_register(&panic_notifier_list,
9659 &trace_panic_notifier);
9661 register_die_notifier(&trace_die_notifier);
9663 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
9665 INIT_LIST_HEAD(&global_trace.systems);
9666 INIT_LIST_HEAD(&global_trace.events);
9667 INIT_LIST_HEAD(&global_trace.hist_vars);
9668 INIT_LIST_HEAD(&global_trace.err_log);
9669 list_add(&global_trace.list, &ftrace_trace_arrays);
9671 apply_trace_boot_options();
9673 register_snapshot_cmd();
9678 free_saved_cmdlines_buffer(savedcmd);
9679 out_free_temp_buffer:
9680 ring_buffer_free(temp_buffer);
9682 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
9684 free_cpumask_var(global_trace.tracing_cpumask);
9685 out_free_buffer_mask:
9686 free_cpumask_var(tracing_buffer_mask);
9691 void __init early_trace_init(void)
9693 if (tracepoint_printk) {
9694 tracepoint_print_iter =
9695 kmalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
9696 if (MEM_FAIL(!tracepoint_print_iter,
9697 "Failed to allocate trace iterator\n"))
9698 tracepoint_printk = 0;
9700 static_key_enable(&tracepoint_printk_key.key);
9702 tracer_alloc_buffers();
9705 void __init trace_init(void)
9710 __init static int clear_boot_tracer(void)
9713 * The default tracer at boot buffer is an init section.
9714 * This function is called in lateinit. If we did not
9715 * find the boot tracer, then clear it out, to prevent
9716 * later registration from accessing the buffer that is
9717 * about to be freed.
9719 if (!default_bootup_tracer)
9722 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
9723 default_bootup_tracer);
9724 default_bootup_tracer = NULL;
9729 fs_initcall(tracer_init_tracefs);
9730 late_initcall_sync(clear_boot_tracer);
9732 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
9733 __init static int tracing_set_default_clock(void)
9735 /* sched_clock_stable() is determined in late_initcall */
9736 if (!trace_boot_clock && !sched_clock_stable()) {
9737 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9738 pr_warn("Can not set tracing clock due to lockdown\n");
9743 "Unstable clock detected, switching default tracing clock to \"global\"\n"
9744 "If you want to keep using the local clock, then add:\n"
9745 " \"trace_clock=local\"\n"
9746 "on the kernel command line\n");
9747 tracing_set_clock(&global_trace, "global");
9752 late_initcall_sync(tracing_set_default_clock);