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 long flags, int pc);
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)
413 /* trace_options that are only supported by global_trace */
414 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
415 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
417 /* trace_flags that are default zero for instances */
418 #define ZEROED_TRACE_FLAGS \
419 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
422 * The global_trace is the descriptor that holds the top-level tracing
423 * buffers for the live tracing.
425 static struct trace_array global_trace = {
426 .trace_flags = TRACE_DEFAULT_FLAGS,
429 LIST_HEAD(ftrace_trace_arrays);
431 int trace_array_get(struct trace_array *this_tr)
433 struct trace_array *tr;
436 mutex_lock(&trace_types_lock);
437 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
444 mutex_unlock(&trace_types_lock);
449 static void __trace_array_put(struct trace_array *this_tr)
451 WARN_ON(!this_tr->ref);
456 * trace_array_put - Decrement the reference counter for this trace array.
458 * NOTE: Use this when we no longer need the trace array returned by
459 * trace_array_get_by_name(). This ensures the trace array can be later
463 void trace_array_put(struct trace_array *this_tr)
468 mutex_lock(&trace_types_lock);
469 __trace_array_put(this_tr);
470 mutex_unlock(&trace_types_lock);
472 EXPORT_SYMBOL_GPL(trace_array_put);
474 int tracing_check_open_get_tr(struct trace_array *tr)
478 ret = security_locked_down(LOCKDOWN_TRACEFS);
482 if (tracing_disabled)
485 if (tr && trace_array_get(tr) < 0)
491 int call_filter_check_discard(struct trace_event_call *call, void *rec,
492 struct trace_buffer *buffer,
493 struct ring_buffer_event *event)
495 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
496 !filter_match_preds(call->filter, rec)) {
497 __trace_event_discard_commit(buffer, event);
504 void trace_free_pid_list(struct trace_pid_list *pid_list)
506 vfree(pid_list->pids);
511 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
512 * @filtered_pids: The list of pids to check
513 * @search_pid: The PID to find in @filtered_pids
515 * Returns true if @search_pid is fonud in @filtered_pids, and false otherwis.
518 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
521 * If pid_max changed after filtered_pids was created, we
522 * by default ignore all pids greater than the previous pid_max.
524 if (search_pid >= filtered_pids->pid_max)
527 return test_bit(search_pid, filtered_pids->pids);
531 * trace_ignore_this_task - should a task be ignored for tracing
532 * @filtered_pids: The list of pids to check
533 * @task: The task that should be ignored if not filtered
535 * Checks if @task should be traced or not from @filtered_pids.
536 * Returns true if @task should *NOT* be traced.
537 * Returns false if @task should be traced.
540 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
541 struct trace_pid_list *filtered_no_pids,
542 struct task_struct *task)
545 * If filterd_no_pids is not empty, and the task's pid is listed
546 * in filtered_no_pids, then return true.
547 * Otherwise, if filtered_pids is empty, that means we can
548 * trace all tasks. If it has content, then only trace pids
549 * within filtered_pids.
552 return (filtered_pids &&
553 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
555 trace_find_filtered_pid(filtered_no_pids, task->pid));
559 * trace_filter_add_remove_task - Add or remove a task from a pid_list
560 * @pid_list: The list to modify
561 * @self: The current task for fork or NULL for exit
562 * @task: The task to add or remove
564 * If adding a task, if @self is defined, the task is only added if @self
565 * is also included in @pid_list. This happens on fork and tasks should
566 * only be added when the parent is listed. If @self is NULL, then the
567 * @task pid will be removed from the list, which would happen on exit
570 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
571 struct task_struct *self,
572 struct task_struct *task)
577 /* For forks, we only add if the forking task is listed */
579 if (!trace_find_filtered_pid(pid_list, self->pid))
583 /* Sorry, but we don't support pid_max changing after setting */
584 if (task->pid >= pid_list->pid_max)
587 /* "self" is set for forks, and NULL for exits */
589 set_bit(task->pid, pid_list->pids);
591 clear_bit(task->pid, pid_list->pids);
595 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
596 * @pid_list: The pid list to show
597 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
598 * @pos: The position of the file
600 * This is used by the seq_file "next" operation to iterate the pids
601 * listed in a trace_pid_list structure.
603 * Returns the pid+1 as we want to display pid of zero, but NULL would
604 * stop the iteration.
606 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
608 unsigned long pid = (unsigned long)v;
612 /* pid already is +1 of the actual prevous bit */
613 pid = find_next_bit(pid_list->pids, pid_list->pid_max, pid);
615 /* Return pid + 1 to allow zero to be represented */
616 if (pid < pid_list->pid_max)
617 return (void *)(pid + 1);
623 * trace_pid_start - Used for seq_file to start reading pid lists
624 * @pid_list: The pid list to show
625 * @pos: The position of the file
627 * This is used by seq_file "start" operation to start the iteration
630 * Returns the pid+1 as we want to display pid of zero, but NULL would
631 * stop the iteration.
633 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
638 pid = find_first_bit(pid_list->pids, pid_list->pid_max);
639 if (pid >= pid_list->pid_max)
642 /* Return pid + 1 so that zero can be the exit value */
643 for (pid++; pid && l < *pos;
644 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
650 * trace_pid_show - show the current pid in seq_file processing
651 * @m: The seq_file structure to write into
652 * @v: A void pointer of the pid (+1) value to display
654 * Can be directly used by seq_file operations to display the current
657 int trace_pid_show(struct seq_file *m, void *v)
659 unsigned long pid = (unsigned long)v - 1;
661 seq_printf(m, "%lu\n", pid);
665 /* 128 should be much more than enough */
666 #define PID_BUF_SIZE 127
668 int trace_pid_write(struct trace_pid_list *filtered_pids,
669 struct trace_pid_list **new_pid_list,
670 const char __user *ubuf, size_t cnt)
672 struct trace_pid_list *pid_list;
673 struct trace_parser parser;
681 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
685 * Always recreate a new array. The write is an all or nothing
686 * operation. Always create a new array when adding new pids by
687 * the user. If the operation fails, then the current list is
690 pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
692 trace_parser_put(&parser);
696 pid_list->pid_max = READ_ONCE(pid_max);
698 /* Only truncating will shrink pid_max */
699 if (filtered_pids && filtered_pids->pid_max > pid_list->pid_max)
700 pid_list->pid_max = filtered_pids->pid_max;
702 pid_list->pids = vzalloc((pid_list->pid_max + 7) >> 3);
703 if (!pid_list->pids) {
704 trace_parser_put(&parser);
710 /* copy the current bits to the new max */
711 for_each_set_bit(pid, filtered_pids->pids,
712 filtered_pids->pid_max) {
713 set_bit(pid, pid_list->pids);
722 ret = trace_get_user(&parser, ubuf, cnt, &pos);
723 if (ret < 0 || !trace_parser_loaded(&parser))
731 if (kstrtoul(parser.buffer, 0, &val))
733 if (val >= pid_list->pid_max)
738 set_bit(pid, pid_list->pids);
741 trace_parser_clear(&parser);
744 trace_parser_put(&parser);
747 trace_free_pid_list(pid_list);
752 /* Cleared the list of pids */
753 trace_free_pid_list(pid_list);
758 *new_pid_list = pid_list;
763 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
767 /* Early boot up does not have a buffer yet */
769 return trace_clock_local();
771 ts = ring_buffer_time_stamp(buf->buffer, cpu);
772 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
777 u64 ftrace_now(int cpu)
779 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
783 * tracing_is_enabled - Show if global_trace has been disabled
785 * Shows if the global trace has been enabled or not. It uses the
786 * mirror flag "buffer_disabled" to be used in fast paths such as for
787 * the irqsoff tracer. But it may be inaccurate due to races. If you
788 * need to know the accurate state, use tracing_is_on() which is a little
789 * slower, but accurate.
791 int tracing_is_enabled(void)
794 * For quick access (irqsoff uses this in fast path), just
795 * return the mirror variable of the state of the ring buffer.
796 * It's a little racy, but we don't really care.
799 return !global_trace.buffer_disabled;
803 * trace_buf_size is the size in bytes that is allocated
804 * for a buffer. Note, the number of bytes is always rounded
807 * This number is purposely set to a low number of 16384.
808 * If the dump on oops happens, it will be much appreciated
809 * to not have to wait for all that output. Anyway this can be
810 * boot time and run time configurable.
812 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
814 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
816 /* trace_types holds a link list of available tracers. */
817 static struct tracer *trace_types __read_mostly;
820 * trace_types_lock is used to protect the trace_types list.
822 DEFINE_MUTEX(trace_types_lock);
825 * serialize the access of the ring buffer
827 * ring buffer serializes readers, but it is low level protection.
828 * The validity of the events (which returns by ring_buffer_peek() ..etc)
829 * are not protected by ring buffer.
831 * The content of events may become garbage if we allow other process consumes
832 * these events concurrently:
833 * A) the page of the consumed events may become a normal page
834 * (not reader page) in ring buffer, and this page will be rewrited
835 * by events producer.
836 * B) The page of the consumed events may become a page for splice_read,
837 * and this page will be returned to system.
839 * These primitives allow multi process access to different cpu ring buffer
842 * These primitives don't distinguish read-only and read-consume access.
843 * Multi read-only access are also serialized.
847 static DECLARE_RWSEM(all_cpu_access_lock);
848 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
850 static inline void trace_access_lock(int cpu)
852 if (cpu == RING_BUFFER_ALL_CPUS) {
853 /* gain it for accessing the whole ring buffer. */
854 down_write(&all_cpu_access_lock);
856 /* gain it for accessing a cpu ring buffer. */
858 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
859 down_read(&all_cpu_access_lock);
861 /* Secondly block other access to this @cpu ring buffer. */
862 mutex_lock(&per_cpu(cpu_access_lock, cpu));
866 static inline void trace_access_unlock(int cpu)
868 if (cpu == RING_BUFFER_ALL_CPUS) {
869 up_write(&all_cpu_access_lock);
871 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
872 up_read(&all_cpu_access_lock);
876 static inline void trace_access_lock_init(void)
880 for_each_possible_cpu(cpu)
881 mutex_init(&per_cpu(cpu_access_lock, cpu));
886 static DEFINE_MUTEX(access_lock);
888 static inline void trace_access_lock(int cpu)
891 mutex_lock(&access_lock);
894 static inline void trace_access_unlock(int cpu)
897 mutex_unlock(&access_lock);
900 static inline void trace_access_lock_init(void)
906 #ifdef CONFIG_STACKTRACE
907 static void __ftrace_trace_stack(struct trace_buffer *buffer,
909 int skip, int pc, struct pt_regs *regs);
910 static inline void ftrace_trace_stack(struct trace_array *tr,
911 struct trace_buffer *buffer,
913 int skip, int pc, struct pt_regs *regs);
916 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
918 int skip, int pc, struct pt_regs *regs)
921 static inline void ftrace_trace_stack(struct trace_array *tr,
922 struct trace_buffer *buffer,
924 int skip, int pc, struct pt_regs *regs)
930 static __always_inline void
931 trace_event_setup(struct ring_buffer_event *event,
932 int type, unsigned long flags, int pc)
934 struct trace_entry *ent = ring_buffer_event_data(event);
936 tracing_generic_entry_update(ent, type, flags, pc);
939 static __always_inline struct ring_buffer_event *
940 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
943 unsigned long flags, int pc)
945 struct ring_buffer_event *event;
947 event = ring_buffer_lock_reserve(buffer, len);
949 trace_event_setup(event, type, flags, pc);
954 void tracer_tracing_on(struct trace_array *tr)
956 if (tr->array_buffer.buffer)
957 ring_buffer_record_on(tr->array_buffer.buffer);
959 * This flag is looked at when buffers haven't been allocated
960 * yet, or by some tracers (like irqsoff), that just want to
961 * know if the ring buffer has been disabled, but it can handle
962 * races of where it gets disabled but we still do a record.
963 * As the check is in the fast path of the tracers, it is more
964 * important to be fast than accurate.
966 tr->buffer_disabled = 0;
967 /* Make the flag seen by readers */
972 * tracing_on - enable tracing buffers
974 * This function enables tracing buffers that may have been
975 * disabled with tracing_off.
977 void tracing_on(void)
979 tracer_tracing_on(&global_trace);
981 EXPORT_SYMBOL_GPL(tracing_on);
984 static __always_inline void
985 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
987 __this_cpu_write(trace_taskinfo_save, true);
989 /* If this is the temp buffer, we need to commit fully */
990 if (this_cpu_read(trace_buffered_event) == event) {
991 /* Length is in event->array[0] */
992 ring_buffer_write(buffer, event->array[0], &event->array[1]);
993 /* Release the temp buffer */
994 this_cpu_dec(trace_buffered_event_cnt);
996 ring_buffer_unlock_commit(buffer, event);
1000 * __trace_puts - write a constant string into the trace buffer.
1001 * @ip: The address of the caller
1002 * @str: The constant string to write
1003 * @size: The size of the string.
1005 int __trace_puts(unsigned long ip, const char *str, int size)
1007 struct ring_buffer_event *event;
1008 struct trace_buffer *buffer;
1009 struct print_entry *entry;
1010 unsigned long irq_flags;
1014 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1017 pc = preempt_count();
1019 if (unlikely(tracing_selftest_running || tracing_disabled))
1022 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1024 local_save_flags(irq_flags);
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, irq_flags, 4, pc, 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 long irq_flags;
1065 int size = sizeof(struct bputs_entry);
1069 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1072 pc = preempt_count();
1074 if (unlikely(tracing_selftest_running || tracing_disabled))
1077 local_save_flags(irq_flags);
1078 buffer = global_trace.array_buffer.buffer;
1080 ring_buffer_nest_start(buffer);
1081 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1086 entry = ring_buffer_event_data(event);
1090 __buffer_unlock_commit(buffer, event);
1091 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
1095 ring_buffer_nest_end(buffer);
1098 EXPORT_SYMBOL_GPL(__trace_bputs);
1100 #ifdef CONFIG_TRACER_SNAPSHOT
1101 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1104 struct tracer *tracer = tr->current_trace;
1105 unsigned long flags;
1108 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1109 internal_trace_puts("*** snapshot is being ignored ***\n");
1113 if (!tr->allocated_snapshot) {
1114 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
1115 internal_trace_puts("*** stopping trace here! ***\n");
1120 /* Note, snapshot can not be used when the tracer uses it */
1121 if (tracer->use_max_tr) {
1122 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
1123 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
1127 local_irq_save(flags);
1128 update_max_tr(tr, current, smp_processor_id(), cond_data);
1129 local_irq_restore(flags);
1132 void tracing_snapshot_instance(struct trace_array *tr)
1134 tracing_snapshot_instance_cond(tr, NULL);
1138 * tracing_snapshot - take a snapshot of the current buffer.
1140 * This causes a swap between the snapshot buffer and the current live
1141 * tracing buffer. You can use this to take snapshots of the live
1142 * trace when some condition is triggered, but continue to trace.
1144 * Note, make sure to allocate the snapshot with either
1145 * a tracing_snapshot_alloc(), or by doing it manually
1146 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
1148 * If the snapshot buffer is not allocated, it will stop tracing.
1149 * Basically making a permanent snapshot.
1151 void tracing_snapshot(void)
1153 struct trace_array *tr = &global_trace;
1155 tracing_snapshot_instance(tr);
1157 EXPORT_SYMBOL_GPL(tracing_snapshot);
1160 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1161 * @tr: The tracing instance to snapshot
1162 * @cond_data: The data to be tested conditionally, and possibly saved
1164 * This is the same as tracing_snapshot() except that the snapshot is
1165 * conditional - the snapshot will only happen if the
1166 * cond_snapshot.update() implementation receiving the cond_data
1167 * returns true, which means that the trace array's cond_snapshot
1168 * update() operation used the cond_data to determine whether the
1169 * snapshot should be taken, and if it was, presumably saved it along
1170 * with the snapshot.
1172 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1174 tracing_snapshot_instance_cond(tr, cond_data);
1176 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1179 * tracing_snapshot_cond_data - get the user data associated with a snapshot
1180 * @tr: The tracing instance
1182 * When the user enables a conditional snapshot using
1183 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1184 * with the snapshot. This accessor is used to retrieve it.
1186 * Should not be called from cond_snapshot.update(), since it takes
1187 * the tr->max_lock lock, which the code calling
1188 * cond_snapshot.update() has already done.
1190 * Returns the cond_data associated with the trace array's snapshot.
1192 void *tracing_cond_snapshot_data(struct trace_array *tr)
1194 void *cond_data = NULL;
1196 arch_spin_lock(&tr->max_lock);
1198 if (tr->cond_snapshot)
1199 cond_data = tr->cond_snapshot->cond_data;
1201 arch_spin_unlock(&tr->max_lock);
1205 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1207 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1208 struct array_buffer *size_buf, int cpu_id);
1209 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1211 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1215 if (!tr->allocated_snapshot) {
1217 /* allocate spare buffer */
1218 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1219 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1223 tr->allocated_snapshot = true;
1229 static void free_snapshot(struct trace_array *tr)
1232 * We don't free the ring buffer. instead, resize it because
1233 * The max_tr ring buffer has some state (e.g. ring->clock) and
1234 * we want preserve it.
1236 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1237 set_buffer_entries(&tr->max_buffer, 1);
1238 tracing_reset_online_cpus(&tr->max_buffer);
1239 tr->allocated_snapshot = false;
1243 * tracing_alloc_snapshot - allocate snapshot buffer.
1245 * This only allocates the snapshot buffer if it isn't already
1246 * allocated - it doesn't also take a snapshot.
1248 * This is meant to be used in cases where the snapshot buffer needs
1249 * to be set up for events that can't sleep but need to be able to
1250 * trigger a snapshot.
1252 int tracing_alloc_snapshot(void)
1254 struct trace_array *tr = &global_trace;
1257 ret = tracing_alloc_snapshot_instance(tr);
1262 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1265 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1267 * This is similar to tracing_snapshot(), but it will allocate the
1268 * snapshot buffer if it isn't already allocated. Use this only
1269 * where it is safe to sleep, as the allocation may sleep.
1271 * This causes a swap between the snapshot buffer and the current live
1272 * tracing buffer. You can use this to take snapshots of the live
1273 * trace when some condition is triggered, but continue to trace.
1275 void tracing_snapshot_alloc(void)
1279 ret = tracing_alloc_snapshot();
1285 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1288 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1289 * @tr: The tracing instance
1290 * @cond_data: User data to associate with the snapshot
1291 * @update: Implementation of the cond_snapshot update function
1293 * Check whether the conditional snapshot for the given instance has
1294 * already been enabled, or if the current tracer is already using a
1295 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1296 * save the cond_data and update function inside.
1298 * Returns 0 if successful, error otherwise.
1300 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1301 cond_update_fn_t update)
1303 struct cond_snapshot *cond_snapshot;
1306 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1310 cond_snapshot->cond_data = cond_data;
1311 cond_snapshot->update = update;
1313 mutex_lock(&trace_types_lock);
1315 ret = tracing_alloc_snapshot_instance(tr);
1319 if (tr->current_trace->use_max_tr) {
1325 * The cond_snapshot can only change to NULL without the
1326 * trace_types_lock. We don't care if we race with it going
1327 * to NULL, but we want to make sure that it's not set to
1328 * something other than NULL when we get here, which we can
1329 * do safely with only holding the trace_types_lock and not
1330 * having to take the max_lock.
1332 if (tr->cond_snapshot) {
1337 arch_spin_lock(&tr->max_lock);
1338 tr->cond_snapshot = cond_snapshot;
1339 arch_spin_unlock(&tr->max_lock);
1341 mutex_unlock(&trace_types_lock);
1346 mutex_unlock(&trace_types_lock);
1347 kfree(cond_snapshot);
1350 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1353 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1354 * @tr: The tracing instance
1356 * Check whether the conditional snapshot for the given instance is
1357 * enabled; if so, free the cond_snapshot associated with it,
1358 * otherwise return -EINVAL.
1360 * Returns 0 if successful, error otherwise.
1362 int tracing_snapshot_cond_disable(struct trace_array *tr)
1366 arch_spin_lock(&tr->max_lock);
1368 if (!tr->cond_snapshot)
1371 kfree(tr->cond_snapshot);
1372 tr->cond_snapshot = NULL;
1375 arch_spin_unlock(&tr->max_lock);
1379 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1381 void tracing_snapshot(void)
1383 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1385 EXPORT_SYMBOL_GPL(tracing_snapshot);
1386 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1388 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1390 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1391 int tracing_alloc_snapshot(void)
1393 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1396 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1397 void tracing_snapshot_alloc(void)
1402 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1403 void *tracing_cond_snapshot_data(struct trace_array *tr)
1407 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1408 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1412 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1413 int tracing_snapshot_cond_disable(struct trace_array *tr)
1417 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1418 #endif /* CONFIG_TRACER_SNAPSHOT */
1420 void tracer_tracing_off(struct trace_array *tr)
1422 if (tr->array_buffer.buffer)
1423 ring_buffer_record_off(tr->array_buffer.buffer);
1425 * This flag is looked at when buffers haven't been allocated
1426 * yet, or by some tracers (like irqsoff), that just want to
1427 * know if the ring buffer has been disabled, but it can handle
1428 * races of where it gets disabled but we still do a record.
1429 * As the check is in the fast path of the tracers, it is more
1430 * important to be fast than accurate.
1432 tr->buffer_disabled = 1;
1433 /* Make the flag seen by readers */
1438 * tracing_off - turn off tracing buffers
1440 * This function stops the tracing buffers from recording data.
1441 * It does not disable any overhead the tracers themselves may
1442 * be causing. This function simply causes all recording to
1443 * the ring buffers to fail.
1445 void tracing_off(void)
1447 tracer_tracing_off(&global_trace);
1449 EXPORT_SYMBOL_GPL(tracing_off);
1451 void disable_trace_on_warning(void)
1453 if (__disable_trace_on_warning) {
1454 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1455 "Disabling tracing due to warning\n");
1461 * tracer_tracing_is_on - show real state of ring buffer enabled
1462 * @tr : the trace array to know if ring buffer is enabled
1464 * Shows real state of the ring buffer if it is enabled or not.
1466 bool tracer_tracing_is_on(struct trace_array *tr)
1468 if (tr->array_buffer.buffer)
1469 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1470 return !tr->buffer_disabled;
1474 * tracing_is_on - show state of ring buffers enabled
1476 int tracing_is_on(void)
1478 return tracer_tracing_is_on(&global_trace);
1480 EXPORT_SYMBOL_GPL(tracing_is_on);
1482 static int __init set_buf_size(char *str)
1484 unsigned long buf_size;
1488 buf_size = memparse(str, &str);
1489 /* nr_entries can not be zero */
1492 trace_buf_size = buf_size;
1495 __setup("trace_buf_size=", set_buf_size);
1497 static int __init set_tracing_thresh(char *str)
1499 unsigned long threshold;
1504 ret = kstrtoul(str, 0, &threshold);
1507 tracing_thresh = threshold * 1000;
1510 __setup("tracing_thresh=", set_tracing_thresh);
1512 unsigned long nsecs_to_usecs(unsigned long nsecs)
1514 return nsecs / 1000;
1518 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1519 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1520 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1521 * of strings in the order that the evals (enum) were defined.
1526 /* These must match the bit postions in trace_iterator_flags */
1527 static const char *trace_options[] = {
1535 int in_ns; /* is this clock in nanoseconds? */
1536 } trace_clocks[] = {
1537 { trace_clock_local, "local", 1 },
1538 { trace_clock_global, "global", 1 },
1539 { trace_clock_counter, "counter", 0 },
1540 { trace_clock_jiffies, "uptime", 0 },
1541 { trace_clock, "perf", 1 },
1542 { ktime_get_mono_fast_ns, "mono", 1 },
1543 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1544 { ktime_get_boot_fast_ns, "boot", 1 },
1548 bool trace_clock_in_ns(struct trace_array *tr)
1550 if (trace_clocks[tr->clock_id].in_ns)
1557 * trace_parser_get_init - gets the buffer for trace parser
1559 int trace_parser_get_init(struct trace_parser *parser, int size)
1561 memset(parser, 0, sizeof(*parser));
1563 parser->buffer = kmalloc(size, GFP_KERNEL);
1564 if (!parser->buffer)
1567 parser->size = size;
1572 * trace_parser_put - frees the buffer for trace parser
1574 void trace_parser_put(struct trace_parser *parser)
1576 kfree(parser->buffer);
1577 parser->buffer = NULL;
1581 * trace_get_user - reads the user input string separated by space
1582 * (matched by isspace(ch))
1584 * For each string found the 'struct trace_parser' is updated,
1585 * and the function returns.
1587 * Returns number of bytes read.
1589 * See kernel/trace/trace.h for 'struct trace_parser' details.
1591 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1592 size_t cnt, loff_t *ppos)
1599 trace_parser_clear(parser);
1601 ret = get_user(ch, ubuf++);
1609 * The parser is not finished with the last write,
1610 * continue reading the user input without skipping spaces.
1612 if (!parser->cont) {
1613 /* skip white space */
1614 while (cnt && isspace(ch)) {
1615 ret = get_user(ch, ubuf++);
1624 /* only spaces were written */
1625 if (isspace(ch) || !ch) {
1632 /* read the non-space input */
1633 while (cnt && !isspace(ch) && ch) {
1634 if (parser->idx < parser->size - 1)
1635 parser->buffer[parser->idx++] = ch;
1640 ret = get_user(ch, ubuf++);
1647 /* We either got finished input or we have to wait for another call. */
1648 if (isspace(ch) || !ch) {
1649 parser->buffer[parser->idx] = 0;
1650 parser->cont = false;
1651 } else if (parser->idx < parser->size - 1) {
1652 parser->cont = true;
1653 parser->buffer[parser->idx++] = ch;
1654 /* Make sure the parsed string always terminates with '\0'. */
1655 parser->buffer[parser->idx] = 0;
1668 /* TODO add a seq_buf_to_buffer() */
1669 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1673 if (trace_seq_used(s) <= s->seq.readpos)
1676 len = trace_seq_used(s) - s->seq.readpos;
1679 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1681 s->seq.readpos += cnt;
1685 unsigned long __read_mostly tracing_thresh;
1686 static const struct file_operations tracing_max_lat_fops;
1688 #if (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1689 defined(CONFIG_FSNOTIFY)
1691 static struct workqueue_struct *fsnotify_wq;
1693 static void latency_fsnotify_workfn(struct work_struct *work)
1695 struct trace_array *tr = container_of(work, struct trace_array,
1697 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1700 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1702 struct trace_array *tr = container_of(iwork, struct trace_array,
1704 queue_work(fsnotify_wq, &tr->fsnotify_work);
1707 static void trace_create_maxlat_file(struct trace_array *tr,
1708 struct dentry *d_tracer)
1710 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1711 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1712 tr->d_max_latency = trace_create_file("tracing_max_latency", 0644,
1713 d_tracer, &tr->max_latency,
1714 &tracing_max_lat_fops);
1717 __init static int latency_fsnotify_init(void)
1719 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1720 WQ_UNBOUND | WQ_HIGHPRI, 0);
1722 pr_err("Unable to allocate tr_max_lat_wq\n");
1728 late_initcall_sync(latency_fsnotify_init);
1730 void latency_fsnotify(struct trace_array *tr)
1735 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1736 * possible that we are called from __schedule() or do_idle(), which
1737 * could cause a deadlock.
1739 irq_work_queue(&tr->fsnotify_irqwork);
1743 * (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1744 * defined(CONFIG_FSNOTIFY)
1748 #define trace_create_maxlat_file(tr, d_tracer) \
1749 trace_create_file("tracing_max_latency", 0644, d_tracer, \
1750 &tr->max_latency, &tracing_max_lat_fops)
1754 #ifdef CONFIG_TRACER_MAX_TRACE
1756 * Copy the new maximum trace into the separate maximum-trace
1757 * structure. (this way the maximum trace is permanently saved,
1758 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1761 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1763 struct array_buffer *trace_buf = &tr->array_buffer;
1764 struct array_buffer *max_buf = &tr->max_buffer;
1765 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1766 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1769 max_buf->time_start = data->preempt_timestamp;
1771 max_data->saved_latency = tr->max_latency;
1772 max_data->critical_start = data->critical_start;
1773 max_data->critical_end = data->critical_end;
1775 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1776 max_data->pid = tsk->pid;
1778 * If tsk == current, then use current_uid(), as that does not use
1779 * RCU. The irq tracer can be called out of RCU scope.
1782 max_data->uid = current_uid();
1784 max_data->uid = task_uid(tsk);
1786 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1787 max_data->policy = tsk->policy;
1788 max_data->rt_priority = tsk->rt_priority;
1790 /* record this tasks comm */
1791 tracing_record_cmdline(tsk);
1792 latency_fsnotify(tr);
1796 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1798 * @tsk: the task with the latency
1799 * @cpu: The cpu that initiated the trace.
1800 * @cond_data: User data associated with a conditional snapshot
1802 * Flip the buffers between the @tr and the max_tr and record information
1803 * about which task was the cause of this latency.
1806 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1812 WARN_ON_ONCE(!irqs_disabled());
1814 if (!tr->allocated_snapshot) {
1815 /* Only the nop tracer should hit this when disabling */
1816 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1820 arch_spin_lock(&tr->max_lock);
1822 /* Inherit the recordable setting from array_buffer */
1823 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1824 ring_buffer_record_on(tr->max_buffer.buffer);
1826 ring_buffer_record_off(tr->max_buffer.buffer);
1828 #ifdef CONFIG_TRACER_SNAPSHOT
1829 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data))
1832 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1834 __update_max_tr(tr, tsk, cpu);
1837 arch_spin_unlock(&tr->max_lock);
1841 * update_max_tr_single - only copy one trace over, and reset the rest
1843 * @tsk: task with the latency
1844 * @cpu: the cpu of the buffer to copy.
1846 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1849 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1856 WARN_ON_ONCE(!irqs_disabled());
1857 if (!tr->allocated_snapshot) {
1858 /* Only the nop tracer should hit this when disabling */
1859 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1863 arch_spin_lock(&tr->max_lock);
1865 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1867 if (ret == -EBUSY) {
1869 * We failed to swap the buffer due to a commit taking
1870 * place on this CPU. We fail to record, but we reset
1871 * the max trace buffer (no one writes directly to it)
1872 * and flag that it failed.
1874 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1875 "Failed to swap buffers due to commit in progress\n");
1878 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1880 __update_max_tr(tr, tsk, cpu);
1881 arch_spin_unlock(&tr->max_lock);
1883 #endif /* CONFIG_TRACER_MAX_TRACE */
1885 static int wait_on_pipe(struct trace_iterator *iter, int full)
1887 /* Iterators are static, they should be filled or empty */
1888 if (trace_buffer_iter(iter, iter->cpu_file))
1891 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1895 #ifdef CONFIG_FTRACE_STARTUP_TEST
1896 static bool selftests_can_run;
1898 struct trace_selftests {
1899 struct list_head list;
1900 struct tracer *type;
1903 static LIST_HEAD(postponed_selftests);
1905 static int save_selftest(struct tracer *type)
1907 struct trace_selftests *selftest;
1909 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1913 selftest->type = type;
1914 list_add(&selftest->list, &postponed_selftests);
1918 static int run_tracer_selftest(struct tracer *type)
1920 struct trace_array *tr = &global_trace;
1921 struct tracer *saved_tracer = tr->current_trace;
1924 if (!type->selftest || tracing_selftest_disabled)
1928 * If a tracer registers early in boot up (before scheduling is
1929 * initialized and such), then do not run its selftests yet.
1930 * Instead, run it a little later in the boot process.
1932 if (!selftests_can_run)
1933 return save_selftest(type);
1936 * Run a selftest on this tracer.
1937 * Here we reset the trace buffer, and set the current
1938 * tracer to be this tracer. The tracer can then run some
1939 * internal tracing to verify that everything is in order.
1940 * If we fail, we do not register this tracer.
1942 tracing_reset_online_cpus(&tr->array_buffer);
1944 tr->current_trace = type;
1946 #ifdef CONFIG_TRACER_MAX_TRACE
1947 if (type->use_max_tr) {
1948 /* If we expanded the buffers, make sure the max is expanded too */
1949 if (ring_buffer_expanded)
1950 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1951 RING_BUFFER_ALL_CPUS);
1952 tr->allocated_snapshot = true;
1956 /* the test is responsible for initializing and enabling */
1957 pr_info("Testing tracer %s: ", type->name);
1958 ret = type->selftest(type, tr);
1959 /* the test is responsible for resetting too */
1960 tr->current_trace = saved_tracer;
1962 printk(KERN_CONT "FAILED!\n");
1963 /* Add the warning after printing 'FAILED' */
1967 /* Only reset on passing, to avoid touching corrupted buffers */
1968 tracing_reset_online_cpus(&tr->array_buffer);
1970 #ifdef CONFIG_TRACER_MAX_TRACE
1971 if (type->use_max_tr) {
1972 tr->allocated_snapshot = false;
1974 /* Shrink the max buffer again */
1975 if (ring_buffer_expanded)
1976 ring_buffer_resize(tr->max_buffer.buffer, 1,
1977 RING_BUFFER_ALL_CPUS);
1981 printk(KERN_CONT "PASSED\n");
1985 static __init int init_trace_selftests(void)
1987 struct trace_selftests *p, *n;
1988 struct tracer *t, **last;
1991 selftests_can_run = true;
1993 mutex_lock(&trace_types_lock);
1995 if (list_empty(&postponed_selftests))
1998 pr_info("Running postponed tracer tests:\n");
2000 tracing_selftest_running = true;
2001 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2002 /* This loop can take minutes when sanitizers are enabled, so
2003 * lets make sure we allow RCU processing.
2006 ret = run_tracer_selftest(p->type);
2007 /* If the test fails, then warn and remove from available_tracers */
2009 WARN(1, "tracer: %s failed selftest, disabling\n",
2011 last = &trace_types;
2012 for (t = trace_types; t; t = t->next) {
2023 tracing_selftest_running = false;
2026 mutex_unlock(&trace_types_lock);
2030 core_initcall(init_trace_selftests);
2032 static inline int run_tracer_selftest(struct tracer *type)
2036 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2038 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2040 static void __init apply_trace_boot_options(void);
2043 * register_tracer - register a tracer with the ftrace system.
2044 * @type: the plugin for the tracer
2046 * Register a new plugin tracer.
2048 int __init register_tracer(struct tracer *type)
2054 pr_info("Tracer must have a name\n");
2058 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2059 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2063 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2064 pr_warn("Can not register tracer %s due to lockdown\n",
2069 mutex_lock(&trace_types_lock);
2071 tracing_selftest_running = true;
2073 for (t = trace_types; t; t = t->next) {
2074 if (strcmp(type->name, t->name) == 0) {
2076 pr_info("Tracer %s already registered\n",
2083 if (!type->set_flag)
2084 type->set_flag = &dummy_set_flag;
2086 /*allocate a dummy tracer_flags*/
2087 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2092 type->flags->val = 0;
2093 type->flags->opts = dummy_tracer_opt;
2095 if (!type->flags->opts)
2096 type->flags->opts = dummy_tracer_opt;
2098 /* store the tracer for __set_tracer_option */
2099 type->flags->trace = type;
2101 ret = run_tracer_selftest(type);
2105 type->next = trace_types;
2107 add_tracer_options(&global_trace, type);
2110 tracing_selftest_running = false;
2111 mutex_unlock(&trace_types_lock);
2113 if (ret || !default_bootup_tracer)
2116 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2119 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2120 /* Do we want this tracer to start on bootup? */
2121 tracing_set_tracer(&global_trace, type->name);
2122 default_bootup_tracer = NULL;
2124 apply_trace_boot_options();
2126 /* disable other selftests, since this will break it. */
2127 disable_tracing_selftest("running a tracer");
2133 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2135 struct trace_buffer *buffer = buf->buffer;
2140 ring_buffer_record_disable(buffer);
2142 /* Make sure all commits have finished */
2144 ring_buffer_reset_cpu(buffer, cpu);
2146 ring_buffer_record_enable(buffer);
2149 void tracing_reset_online_cpus(struct array_buffer *buf)
2151 struct trace_buffer *buffer = buf->buffer;
2156 ring_buffer_record_disable(buffer);
2158 /* Make sure all commits have finished */
2161 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2163 ring_buffer_reset_online_cpus(buffer);
2165 ring_buffer_record_enable(buffer);
2168 /* Must have trace_types_lock held */
2169 void tracing_reset_all_online_cpus(void)
2171 struct trace_array *tr;
2173 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2174 if (!tr->clear_trace)
2176 tr->clear_trace = false;
2177 tracing_reset_online_cpus(&tr->array_buffer);
2178 #ifdef CONFIG_TRACER_MAX_TRACE
2179 tracing_reset_online_cpus(&tr->max_buffer);
2185 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2186 * is the tgid last observed corresponding to pid=i.
2188 static int *tgid_map;
2190 /* The maximum valid index into tgid_map. */
2191 static size_t tgid_map_max;
2193 #define SAVED_CMDLINES_DEFAULT 128
2194 #define NO_CMDLINE_MAP UINT_MAX
2195 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2196 struct saved_cmdlines_buffer {
2197 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2198 unsigned *map_cmdline_to_pid;
2199 unsigned cmdline_num;
2201 char *saved_cmdlines;
2203 static struct saved_cmdlines_buffer *savedcmd;
2205 static inline char *get_saved_cmdlines(int idx)
2207 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2210 static inline void set_cmdline(int idx, const char *cmdline)
2212 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2215 static int allocate_cmdlines_buffer(unsigned int val,
2216 struct saved_cmdlines_buffer *s)
2218 s->map_cmdline_to_pid = kmalloc_array(val,
2219 sizeof(*s->map_cmdline_to_pid),
2221 if (!s->map_cmdline_to_pid)
2224 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2225 if (!s->saved_cmdlines) {
2226 kfree(s->map_cmdline_to_pid);
2231 s->cmdline_num = val;
2232 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2233 sizeof(s->map_pid_to_cmdline));
2234 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2235 val * sizeof(*s->map_cmdline_to_pid));
2240 static int trace_create_savedcmd(void)
2244 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2248 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2258 int is_tracing_stopped(void)
2260 return global_trace.stop_count;
2264 * tracing_start - quick start of the tracer
2266 * If tracing is enabled but was stopped by tracing_stop,
2267 * this will start the tracer back up.
2269 void tracing_start(void)
2271 struct trace_buffer *buffer;
2272 unsigned long flags;
2274 if (tracing_disabled)
2277 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2278 if (--global_trace.stop_count) {
2279 if (global_trace.stop_count < 0) {
2280 /* Someone screwed up their debugging */
2282 global_trace.stop_count = 0;
2287 /* Prevent the buffers from switching */
2288 arch_spin_lock(&global_trace.max_lock);
2290 buffer = global_trace.array_buffer.buffer;
2292 ring_buffer_record_enable(buffer);
2294 #ifdef CONFIG_TRACER_MAX_TRACE
2295 buffer = global_trace.max_buffer.buffer;
2297 ring_buffer_record_enable(buffer);
2300 arch_spin_unlock(&global_trace.max_lock);
2303 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2306 static void tracing_start_tr(struct trace_array *tr)
2308 struct trace_buffer *buffer;
2309 unsigned long flags;
2311 if (tracing_disabled)
2314 /* If global, we need to also start the max tracer */
2315 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2316 return tracing_start();
2318 raw_spin_lock_irqsave(&tr->start_lock, flags);
2320 if (--tr->stop_count) {
2321 if (tr->stop_count < 0) {
2322 /* Someone screwed up their debugging */
2329 buffer = tr->array_buffer.buffer;
2331 ring_buffer_record_enable(buffer);
2334 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2338 * tracing_stop - quick stop of the tracer
2340 * Light weight way to stop tracing. Use in conjunction with
2343 void tracing_stop(void)
2345 struct trace_buffer *buffer;
2346 unsigned long flags;
2348 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2349 if (global_trace.stop_count++)
2352 /* Prevent the buffers from switching */
2353 arch_spin_lock(&global_trace.max_lock);
2355 buffer = global_trace.array_buffer.buffer;
2357 ring_buffer_record_disable(buffer);
2359 #ifdef CONFIG_TRACER_MAX_TRACE
2360 buffer = global_trace.max_buffer.buffer;
2362 ring_buffer_record_disable(buffer);
2365 arch_spin_unlock(&global_trace.max_lock);
2368 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2371 static void tracing_stop_tr(struct trace_array *tr)
2373 struct trace_buffer *buffer;
2374 unsigned long flags;
2376 /* If global, we need to also stop the max tracer */
2377 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2378 return tracing_stop();
2380 raw_spin_lock_irqsave(&tr->start_lock, flags);
2381 if (tr->stop_count++)
2384 buffer = tr->array_buffer.buffer;
2386 ring_buffer_record_disable(buffer);
2389 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2392 static int trace_save_cmdline(struct task_struct *tsk)
2396 /* treat recording of idle task as a success */
2400 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
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[tpid];
2412 if (idx == NO_CMDLINE_MAP) {
2413 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2415 savedcmd->map_pid_to_cmdline[tpid] = idx;
2416 savedcmd->cmdline_idx = idx;
2419 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2420 set_cmdline(idx, tsk->comm);
2422 arch_spin_unlock(&trace_cmdline_lock);
2427 static void __trace_find_cmdline(int pid, char comm[])
2433 strcpy(comm, "<idle>");
2437 if (WARN_ON_ONCE(pid < 0)) {
2438 strcpy(comm, "<XXX>");
2442 tpid = pid & (PID_MAX_DEFAULT - 1);
2443 map = savedcmd->map_pid_to_cmdline[tpid];
2444 if (map != NO_CMDLINE_MAP) {
2445 tpid = savedcmd->map_cmdline_to_pid[map];
2447 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2451 strcpy(comm, "<...>");
2454 void trace_find_cmdline(int pid, char comm[])
2457 arch_spin_lock(&trace_cmdline_lock);
2459 __trace_find_cmdline(pid, comm);
2461 arch_spin_unlock(&trace_cmdline_lock);
2465 static int *trace_find_tgid_ptr(int pid)
2468 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2469 * if we observe a non-NULL tgid_map then we also observe the correct
2472 int *map = smp_load_acquire(&tgid_map);
2474 if (unlikely(!map || pid > tgid_map_max))
2480 int trace_find_tgid(int pid)
2482 int *ptr = trace_find_tgid_ptr(pid);
2484 return ptr ? *ptr : 0;
2487 static int trace_save_tgid(struct task_struct *tsk)
2491 /* treat recording of idle task as a success */
2495 ptr = trace_find_tgid_ptr(tsk->pid);
2503 static bool tracing_record_taskinfo_skip(int flags)
2505 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2507 if (!__this_cpu_read(trace_taskinfo_save))
2513 * tracing_record_taskinfo - record the task info of a task
2515 * @task: task to record
2516 * @flags: TRACE_RECORD_CMDLINE for recording comm
2517 * TRACE_RECORD_TGID for recording tgid
2519 void tracing_record_taskinfo(struct task_struct *task, int flags)
2523 if (tracing_record_taskinfo_skip(flags))
2527 * Record as much task information as possible. If some fail, continue
2528 * to try to record the others.
2530 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2531 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2533 /* If recording any information failed, retry again soon. */
2537 __this_cpu_write(trace_taskinfo_save, false);
2541 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2543 * @prev: previous task during sched_switch
2544 * @next: next task during sched_switch
2545 * @flags: TRACE_RECORD_CMDLINE for recording comm
2546 * TRACE_RECORD_TGID for recording tgid
2548 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2549 struct task_struct *next, int flags)
2553 if (tracing_record_taskinfo_skip(flags))
2557 * Record as much task information as possible. If some fail, continue
2558 * to try to record the others.
2560 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2561 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2562 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2563 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2565 /* If recording any information failed, retry again soon. */
2569 __this_cpu_write(trace_taskinfo_save, false);
2572 /* Helpers to record a specific task information */
2573 void tracing_record_cmdline(struct task_struct *task)
2575 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2578 void tracing_record_tgid(struct task_struct *task)
2580 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2584 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2585 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2586 * simplifies those functions and keeps them in sync.
2588 enum print_line_t trace_handle_return(struct trace_seq *s)
2590 return trace_seq_has_overflowed(s) ?
2591 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2593 EXPORT_SYMBOL_GPL(trace_handle_return);
2596 tracing_generic_entry_update(struct trace_entry *entry, unsigned short type,
2597 unsigned long flags, int pc)
2599 struct task_struct *tsk = current;
2601 entry->preempt_count = pc & 0xff;
2602 entry->pid = (tsk) ? tsk->pid : 0;
2605 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
2606 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
2608 TRACE_FLAG_IRQS_NOSUPPORT |
2610 ((pc & NMI_MASK ) ? TRACE_FLAG_NMI : 0) |
2611 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
2612 ((pc & SOFTIRQ_OFFSET) ? TRACE_FLAG_SOFTIRQ : 0) |
2613 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
2614 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
2616 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
2618 struct ring_buffer_event *
2619 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2622 unsigned long flags, int pc)
2624 return __trace_buffer_lock_reserve(buffer, type, len, flags, pc);
2627 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2628 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2629 static int trace_buffered_event_ref;
2632 * trace_buffered_event_enable - enable buffering events
2634 * When events are being filtered, it is quicker to use a temporary
2635 * buffer to write the event data into if there's a likely chance
2636 * that it will not be committed. The discard of the ring buffer
2637 * is not as fast as committing, and is much slower than copying
2640 * When an event is to be filtered, allocate per cpu buffers to
2641 * write the event data into, and if the event is filtered and discarded
2642 * it is simply dropped, otherwise, the entire data is to be committed
2645 void trace_buffered_event_enable(void)
2647 struct ring_buffer_event *event;
2651 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2653 if (trace_buffered_event_ref++)
2656 for_each_tracing_cpu(cpu) {
2657 page = alloc_pages_node(cpu_to_node(cpu),
2658 GFP_KERNEL | __GFP_NORETRY, 0);
2662 event = page_address(page);
2663 memset(event, 0, sizeof(*event));
2665 per_cpu(trace_buffered_event, cpu) = event;
2668 if (cpu == smp_processor_id() &&
2669 __this_cpu_read(trace_buffered_event) !=
2670 per_cpu(trace_buffered_event, cpu))
2677 trace_buffered_event_disable();
2680 static void enable_trace_buffered_event(void *data)
2682 /* Probably not needed, but do it anyway */
2684 this_cpu_dec(trace_buffered_event_cnt);
2687 static void disable_trace_buffered_event(void *data)
2689 this_cpu_inc(trace_buffered_event_cnt);
2693 * trace_buffered_event_disable - disable buffering events
2695 * When a filter is removed, it is faster to not use the buffered
2696 * events, and to commit directly into the ring buffer. Free up
2697 * the temp buffers when there are no more users. This requires
2698 * special synchronization with current events.
2700 void trace_buffered_event_disable(void)
2704 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2706 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2709 if (--trace_buffered_event_ref)
2713 /* For each CPU, set the buffer as used. */
2714 smp_call_function_many(tracing_buffer_mask,
2715 disable_trace_buffered_event, NULL, 1);
2718 /* Wait for all current users to finish */
2721 for_each_tracing_cpu(cpu) {
2722 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2723 per_cpu(trace_buffered_event, cpu) = NULL;
2726 * Make sure trace_buffered_event is NULL before clearing
2727 * trace_buffered_event_cnt.
2732 /* Do the work on each cpu */
2733 smp_call_function_many(tracing_buffer_mask,
2734 enable_trace_buffered_event, NULL, 1);
2738 static struct trace_buffer *temp_buffer;
2740 struct ring_buffer_event *
2741 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2742 struct trace_event_file *trace_file,
2743 int type, unsigned long len,
2744 unsigned long flags, int pc)
2746 struct ring_buffer_event *entry;
2749 *current_rb = trace_file->tr->array_buffer.buffer;
2751 if (!ring_buffer_time_stamp_abs(*current_rb) && (trace_file->flags &
2752 (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED)) &&
2753 (entry = this_cpu_read(trace_buffered_event))) {
2754 /* Try to use the per cpu buffer first */
2755 val = this_cpu_inc_return(trace_buffered_event_cnt);
2756 if ((len < (PAGE_SIZE - sizeof(*entry) - sizeof(entry->array[0]))) && val == 1) {
2757 trace_event_setup(entry, type, flags, pc);
2758 entry->array[0] = len;
2761 this_cpu_dec(trace_buffered_event_cnt);
2764 entry = __trace_buffer_lock_reserve(*current_rb,
2765 type, len, flags, pc);
2767 * If tracing is off, but we have triggers enabled
2768 * we still need to look at the event data. Use the temp_buffer
2769 * to store the trace event for the trigger to use. It's recursive
2770 * safe and will not be recorded anywhere.
2772 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2773 *current_rb = temp_buffer;
2774 entry = __trace_buffer_lock_reserve(*current_rb,
2775 type, len, flags, pc);
2779 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2781 static DEFINE_SPINLOCK(tracepoint_iter_lock);
2782 static DEFINE_MUTEX(tracepoint_printk_mutex);
2784 static void output_printk(struct trace_event_buffer *fbuffer)
2786 struct trace_event_call *event_call;
2787 struct trace_event_file *file;
2788 struct trace_event *event;
2789 unsigned long flags;
2790 struct trace_iterator *iter = tracepoint_print_iter;
2792 /* We should never get here if iter is NULL */
2793 if (WARN_ON_ONCE(!iter))
2796 event_call = fbuffer->trace_file->event_call;
2797 if (!event_call || !event_call->event.funcs ||
2798 !event_call->event.funcs->trace)
2801 file = fbuffer->trace_file;
2802 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2803 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2804 !filter_match_preds(file->filter, fbuffer->entry)))
2807 event = &fbuffer->trace_file->event_call->event;
2809 spin_lock_irqsave(&tracepoint_iter_lock, flags);
2810 trace_seq_init(&iter->seq);
2811 iter->ent = fbuffer->entry;
2812 event_call->event.funcs->trace(iter, 0, event);
2813 trace_seq_putc(&iter->seq, 0);
2814 printk("%s", iter->seq.buffer);
2816 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2819 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2820 void *buffer, size_t *lenp,
2823 int save_tracepoint_printk;
2826 mutex_lock(&tracepoint_printk_mutex);
2827 save_tracepoint_printk = tracepoint_printk;
2829 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2832 * This will force exiting early, as tracepoint_printk
2833 * is always zero when tracepoint_printk_iter is not allocated
2835 if (!tracepoint_print_iter)
2836 tracepoint_printk = 0;
2838 if (save_tracepoint_printk == tracepoint_printk)
2841 if (tracepoint_printk)
2842 static_key_enable(&tracepoint_printk_key.key);
2844 static_key_disable(&tracepoint_printk_key.key);
2847 mutex_unlock(&tracepoint_printk_mutex);
2852 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2854 if (static_key_false(&tracepoint_printk_key.key))
2855 output_printk(fbuffer);
2857 if (static_branch_unlikely(&trace_event_exports_enabled))
2858 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2859 event_trigger_unlock_commit_regs(fbuffer->trace_file, fbuffer->buffer,
2860 fbuffer->event, fbuffer->entry,
2861 fbuffer->flags, fbuffer->pc, fbuffer->regs);
2863 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2868 * trace_buffer_unlock_commit_regs()
2869 * trace_event_buffer_commit()
2870 * trace_event_raw_event_xxx()
2872 # define STACK_SKIP 3
2874 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2875 struct trace_buffer *buffer,
2876 struct ring_buffer_event *event,
2877 unsigned long flags, int pc,
2878 struct pt_regs *regs)
2880 __buffer_unlock_commit(buffer, event);
2883 * If regs is not set, then skip the necessary functions.
2884 * Note, we can still get here via blktrace, wakeup tracer
2885 * and mmiotrace, but that's ok if they lose a function or
2886 * two. They are not that meaningful.
2888 ftrace_trace_stack(tr, buffer, flags, regs ? 0 : STACK_SKIP, pc, regs);
2889 ftrace_trace_userstack(tr, buffer, flags, pc);
2893 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2896 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2897 struct ring_buffer_event *event)
2899 __buffer_unlock_commit(buffer, event);
2903 trace_function(struct trace_array *tr,
2904 unsigned long ip, unsigned long parent_ip, unsigned long flags,
2907 struct trace_event_call *call = &event_function;
2908 struct trace_buffer *buffer = tr->array_buffer.buffer;
2909 struct ring_buffer_event *event;
2910 struct ftrace_entry *entry;
2912 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2916 entry = ring_buffer_event_data(event);
2918 entry->parent_ip = parent_ip;
2920 if (!call_filter_check_discard(call, entry, buffer, event)) {
2921 if (static_branch_unlikely(&trace_function_exports_enabled))
2922 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
2923 __buffer_unlock_commit(buffer, event);
2927 #ifdef CONFIG_STACKTRACE
2929 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2930 #define FTRACE_KSTACK_NESTING 4
2932 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
2934 struct ftrace_stack {
2935 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2939 struct ftrace_stacks {
2940 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
2943 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
2944 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2946 static void __ftrace_trace_stack(struct trace_buffer *buffer,
2947 unsigned long flags,
2948 int skip, int pc, struct pt_regs *regs)
2950 struct trace_event_call *call = &event_kernel_stack;
2951 struct ring_buffer_event *event;
2952 unsigned int size, nr_entries;
2953 struct ftrace_stack *fstack;
2954 struct stack_entry *entry;
2958 * Add one, for this function and the call to save_stack_trace()
2959 * If regs is set, then these functions will not be in the way.
2961 #ifndef CONFIG_UNWINDER_ORC
2966 preempt_disable_notrace();
2968 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
2970 /* This should never happen. If it does, yell once and skip */
2971 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
2975 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
2976 * interrupt will either see the value pre increment or post
2977 * increment. If the interrupt happens pre increment it will have
2978 * restored the counter when it returns. We just need a barrier to
2979 * keep gcc from moving things around.
2983 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
2984 size = ARRAY_SIZE(fstack->calls);
2987 nr_entries = stack_trace_save_regs(regs, fstack->calls,
2990 nr_entries = stack_trace_save(fstack->calls, size, skip);
2993 size = nr_entries * sizeof(unsigned long);
2994 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2995 (sizeof(*entry) - sizeof(entry->caller)) + size,
2999 entry = ring_buffer_event_data(event);
3001 memcpy(&entry->caller, fstack->calls, size);
3002 entry->size = nr_entries;
3004 if (!call_filter_check_discard(call, entry, buffer, event))
3005 __buffer_unlock_commit(buffer, event);
3008 /* Again, don't let gcc optimize things here */
3010 __this_cpu_dec(ftrace_stack_reserve);
3011 preempt_enable_notrace();
3015 static inline void ftrace_trace_stack(struct trace_array *tr,
3016 struct trace_buffer *buffer,
3017 unsigned long flags,
3018 int skip, int pc, struct pt_regs *regs)
3020 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3023 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
3026 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
3029 struct trace_buffer *buffer = tr->array_buffer.buffer;
3031 if (rcu_is_watching()) {
3032 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
3037 * When an NMI triggers, RCU is enabled via rcu_nmi_enter(),
3038 * but if the above rcu_is_watching() failed, then the NMI
3039 * triggered someplace critical, and rcu_irq_enter() should
3040 * not be called from NMI.
3042 if (unlikely(in_nmi()))
3045 rcu_irq_enter_irqson();
3046 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
3047 rcu_irq_exit_irqson();
3051 * trace_dump_stack - record a stack back trace in the trace buffer
3052 * @skip: Number of functions to skip (helper handlers)
3054 void trace_dump_stack(int skip)
3056 unsigned long flags;
3058 if (tracing_disabled || tracing_selftest_running)
3061 local_save_flags(flags);
3063 #ifndef CONFIG_UNWINDER_ORC
3064 /* Skip 1 to skip this function. */
3067 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3068 flags, skip, preempt_count(), NULL);
3070 EXPORT_SYMBOL_GPL(trace_dump_stack);
3072 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3073 static DEFINE_PER_CPU(int, user_stack_count);
3076 ftrace_trace_userstack(struct trace_array *tr,
3077 struct trace_buffer *buffer, unsigned long flags, int pc)
3079 struct trace_event_call *call = &event_user_stack;
3080 struct ring_buffer_event *event;
3081 struct userstack_entry *entry;
3083 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3087 * NMIs can not handle page faults, even with fix ups.
3088 * The save user stack can (and often does) fault.
3090 if (unlikely(in_nmi()))
3094 * prevent recursion, since the user stack tracing may
3095 * trigger other kernel events.
3098 if (__this_cpu_read(user_stack_count))
3101 __this_cpu_inc(user_stack_count);
3103 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3104 sizeof(*entry), flags, pc);
3106 goto out_drop_count;
3107 entry = ring_buffer_event_data(event);
3109 entry->tgid = current->tgid;
3110 memset(&entry->caller, 0, sizeof(entry->caller));
3112 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3113 if (!call_filter_check_discard(call, entry, buffer, event))
3114 __buffer_unlock_commit(buffer, event);
3117 __this_cpu_dec(user_stack_count);
3121 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3122 static void ftrace_trace_userstack(struct trace_array *tr,
3123 struct trace_buffer *buffer,
3124 unsigned long flags, int pc)
3127 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3129 #endif /* CONFIG_STACKTRACE */
3131 /* created for use with alloc_percpu */
3132 struct trace_buffer_struct {
3134 char buffer[4][TRACE_BUF_SIZE];
3137 static struct trace_buffer_struct *trace_percpu_buffer;
3140 * Thise allows for lockless recording. If we're nested too deeply, then
3141 * this returns NULL.
3143 static char *get_trace_buf(void)
3145 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3147 if (!trace_percpu_buffer || buffer->nesting >= 4)
3152 /* Interrupts must see nesting incremented before we use the buffer */
3154 return &buffer->buffer[buffer->nesting - 1][0];
3157 static void put_trace_buf(void)
3159 /* Don't let the decrement of nesting leak before this */
3161 this_cpu_dec(trace_percpu_buffer->nesting);
3164 static int alloc_percpu_trace_buffer(void)
3166 struct trace_buffer_struct *buffers;
3168 if (trace_percpu_buffer)
3171 buffers = alloc_percpu(struct trace_buffer_struct);
3172 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3175 trace_percpu_buffer = buffers;
3179 static int buffers_allocated;
3181 void trace_printk_init_buffers(void)
3183 if (buffers_allocated)
3186 if (alloc_percpu_trace_buffer())
3189 /* trace_printk() is for debug use only. Don't use it in production. */
3192 pr_warn("**********************************************************\n");
3193 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3195 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3197 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3198 pr_warn("** unsafe for production use. **\n");
3200 pr_warn("** If you see this message and you are not debugging **\n");
3201 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3203 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3204 pr_warn("**********************************************************\n");
3206 /* Expand the buffers to set size */
3207 tracing_update_buffers();
3209 buffers_allocated = 1;
3212 * trace_printk_init_buffers() can be called by modules.
3213 * If that happens, then we need to start cmdline recording
3214 * directly here. If the global_trace.buffer is already
3215 * allocated here, then this was called by module code.
3217 if (global_trace.array_buffer.buffer)
3218 tracing_start_cmdline_record();
3220 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3222 void trace_printk_start_comm(void)
3224 /* Start tracing comms if trace printk is set */
3225 if (!buffers_allocated)
3227 tracing_start_cmdline_record();
3230 static void trace_printk_start_stop_comm(int enabled)
3232 if (!buffers_allocated)
3236 tracing_start_cmdline_record();
3238 tracing_stop_cmdline_record();
3242 * trace_vbprintk - write binary msg to tracing buffer
3243 * @ip: The address of the caller
3244 * @fmt: The string format to write to the buffer
3245 * @args: Arguments for @fmt
3247 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3249 struct trace_event_call *call = &event_bprint;
3250 struct ring_buffer_event *event;
3251 struct trace_buffer *buffer;
3252 struct trace_array *tr = &global_trace;
3253 struct bprint_entry *entry;
3254 unsigned long flags;
3256 int len = 0, size, pc;
3258 if (unlikely(tracing_selftest_running || tracing_disabled))
3261 /* Don't pollute graph traces with trace_vprintk internals */
3262 pause_graph_tracing();
3264 pc = preempt_count();
3265 preempt_disable_notrace();
3267 tbuffer = get_trace_buf();
3273 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3275 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3278 local_save_flags(flags);
3279 size = sizeof(*entry) + sizeof(u32) * len;
3280 buffer = tr->array_buffer.buffer;
3281 ring_buffer_nest_start(buffer);
3282 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3286 entry = ring_buffer_event_data(event);
3290 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3291 if (!call_filter_check_discard(call, entry, buffer, event)) {
3292 __buffer_unlock_commit(buffer, event);
3293 ftrace_trace_stack(tr, buffer, flags, 6, pc, NULL);
3297 ring_buffer_nest_end(buffer);
3302 preempt_enable_notrace();
3303 unpause_graph_tracing();
3307 EXPORT_SYMBOL_GPL(trace_vbprintk);
3311 __trace_array_vprintk(struct trace_buffer *buffer,
3312 unsigned long ip, const char *fmt, va_list args)
3314 struct trace_event_call *call = &event_print;
3315 struct ring_buffer_event *event;
3316 int len = 0, size, pc;
3317 struct print_entry *entry;
3318 unsigned long flags;
3321 if (tracing_disabled || tracing_selftest_running)
3324 /* Don't pollute graph traces with trace_vprintk internals */
3325 pause_graph_tracing();
3327 pc = preempt_count();
3328 preempt_disable_notrace();
3331 tbuffer = get_trace_buf();
3337 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3339 local_save_flags(flags);
3340 size = sizeof(*entry) + len + 1;
3341 ring_buffer_nest_start(buffer);
3342 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3346 entry = ring_buffer_event_data(event);
3349 memcpy(&entry->buf, tbuffer, len + 1);
3350 if (!call_filter_check_discard(call, entry, buffer, event)) {
3351 __buffer_unlock_commit(buffer, event);
3352 ftrace_trace_stack(&global_trace, buffer, flags, 6, pc, NULL);
3356 ring_buffer_nest_end(buffer);
3360 preempt_enable_notrace();
3361 unpause_graph_tracing();
3367 int trace_array_vprintk(struct trace_array *tr,
3368 unsigned long ip, const char *fmt, va_list args)
3370 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3374 * trace_array_printk - Print a message to a specific instance
3375 * @tr: The instance trace_array descriptor
3376 * @ip: The instruction pointer that this is called from.
3377 * @fmt: The format to print (printf format)
3379 * If a subsystem sets up its own instance, they have the right to
3380 * printk strings into their tracing instance buffer using this
3381 * function. Note, this function will not write into the top level
3382 * buffer (use trace_printk() for that), as writing into the top level
3383 * buffer should only have events that can be individually disabled.
3384 * trace_printk() is only used for debugging a kernel, and should not
3385 * be ever encorporated in normal use.
3387 * trace_array_printk() can be used, as it will not add noise to the
3388 * top level tracing buffer.
3390 * Note, trace_array_init_printk() must be called on @tr before this
3394 int trace_array_printk(struct trace_array *tr,
3395 unsigned long ip, const char *fmt, ...)
3403 /* This is only allowed for created instances */
3404 if (tr == &global_trace)
3407 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3411 ret = trace_array_vprintk(tr, ip, fmt, ap);
3415 EXPORT_SYMBOL_GPL(trace_array_printk);
3418 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3419 * @tr: The trace array to initialize the buffers for
3421 * As trace_array_printk() only writes into instances, they are OK to
3422 * have in the kernel (unlike trace_printk()). This needs to be called
3423 * before trace_array_printk() can be used on a trace_array.
3425 int trace_array_init_printk(struct trace_array *tr)
3430 /* This is only allowed for created instances */
3431 if (tr == &global_trace)
3434 return alloc_percpu_trace_buffer();
3436 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3439 int trace_array_printk_buf(struct trace_buffer *buffer,
3440 unsigned long ip, const char *fmt, ...)
3445 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3449 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3455 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3457 return trace_array_vprintk(&global_trace, ip, fmt, args);
3459 EXPORT_SYMBOL_GPL(trace_vprintk);
3461 static void trace_iterator_increment(struct trace_iterator *iter)
3463 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3467 ring_buffer_iter_advance(buf_iter);
3470 static struct trace_entry *
3471 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3472 unsigned long *lost_events)
3474 struct ring_buffer_event *event;
3475 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3478 event = ring_buffer_iter_peek(buf_iter, ts);
3480 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3481 (unsigned long)-1 : 0;
3483 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3488 iter->ent_size = ring_buffer_event_length(event);
3489 return ring_buffer_event_data(event);
3495 static struct trace_entry *
3496 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3497 unsigned long *missing_events, u64 *ent_ts)
3499 struct trace_buffer *buffer = iter->array_buffer->buffer;
3500 struct trace_entry *ent, *next = NULL;
3501 unsigned long lost_events = 0, next_lost = 0;
3502 int cpu_file = iter->cpu_file;
3503 u64 next_ts = 0, ts;
3509 * If we are in a per_cpu trace file, don't bother by iterating over
3510 * all cpu and peek directly.
3512 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3513 if (ring_buffer_empty_cpu(buffer, cpu_file))
3515 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3517 *ent_cpu = cpu_file;
3522 for_each_tracing_cpu(cpu) {
3524 if (ring_buffer_empty_cpu(buffer, cpu))
3527 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3530 * Pick the entry with the smallest timestamp:
3532 if (ent && (!next || ts < next_ts)) {
3536 next_lost = lost_events;
3537 next_size = iter->ent_size;
3541 iter->ent_size = next_size;
3544 *ent_cpu = next_cpu;
3550 *missing_events = next_lost;
3555 #define STATIC_TEMP_BUF_SIZE 128
3556 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3558 /* Find the next real entry, without updating the iterator itself */
3559 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3560 int *ent_cpu, u64 *ent_ts)
3562 /* __find_next_entry will reset ent_size */
3563 int ent_size = iter->ent_size;
3564 struct trace_entry *entry;
3567 * If called from ftrace_dump(), then the iter->temp buffer
3568 * will be the static_temp_buf and not created from kmalloc.
3569 * If the entry size is greater than the buffer, we can
3570 * not save it. Just return NULL in that case. This is only
3571 * used to add markers when two consecutive events' time
3572 * stamps have a large delta. See trace_print_lat_context()
3574 if (iter->temp == static_temp_buf &&
3575 STATIC_TEMP_BUF_SIZE < ent_size)
3579 * The __find_next_entry() may call peek_next_entry(), which may
3580 * call ring_buffer_peek() that may make the contents of iter->ent
3581 * undefined. Need to copy iter->ent now.
3583 if (iter->ent && iter->ent != iter->temp) {
3584 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3585 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3587 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3592 iter->temp_size = iter->ent_size;
3594 memcpy(iter->temp, iter->ent, iter->ent_size);
3595 iter->ent = iter->temp;
3597 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3598 /* Put back the original ent_size */
3599 iter->ent_size = ent_size;
3604 /* Find the next real entry, and increment the iterator to the next entry */
3605 void *trace_find_next_entry_inc(struct trace_iterator *iter)
3607 iter->ent = __find_next_entry(iter, &iter->cpu,
3608 &iter->lost_events, &iter->ts);
3611 trace_iterator_increment(iter);
3613 return iter->ent ? iter : NULL;
3616 static void trace_consume(struct trace_iterator *iter)
3618 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3619 &iter->lost_events);
3622 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3624 struct trace_iterator *iter = m->private;
3628 WARN_ON_ONCE(iter->leftover);
3632 /* can't go backwards */
3637 ent = trace_find_next_entry_inc(iter);
3641 while (ent && iter->idx < i)
3642 ent = trace_find_next_entry_inc(iter);
3649 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3651 struct ring_buffer_iter *buf_iter;
3652 unsigned long entries = 0;
3655 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
3657 buf_iter = trace_buffer_iter(iter, cpu);
3661 ring_buffer_iter_reset(buf_iter);
3664 * We could have the case with the max latency tracers
3665 * that a reset never took place on a cpu. This is evident
3666 * by the timestamp being before the start of the buffer.
3668 while (ring_buffer_iter_peek(buf_iter, &ts)) {
3669 if (ts >= iter->array_buffer->time_start)
3672 ring_buffer_iter_advance(buf_iter);
3675 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
3679 * The current tracer is copied to avoid a global locking
3682 static void *s_start(struct seq_file *m, loff_t *pos)
3684 struct trace_iterator *iter = m->private;
3685 struct trace_array *tr = iter->tr;
3686 int cpu_file = iter->cpu_file;
3692 * copy the tracer to avoid using a global lock all around.
3693 * iter->trace is a copy of current_trace, the pointer to the
3694 * name may be used instead of a strcmp(), as iter->trace->name
3695 * will point to the same string as current_trace->name.
3697 mutex_lock(&trace_types_lock);
3698 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
3699 *iter->trace = *tr->current_trace;
3700 mutex_unlock(&trace_types_lock);
3702 #ifdef CONFIG_TRACER_MAX_TRACE
3703 if (iter->snapshot && iter->trace->use_max_tr)
3704 return ERR_PTR(-EBUSY);
3707 if (*pos != iter->pos) {
3712 if (cpu_file == RING_BUFFER_ALL_CPUS) {
3713 for_each_tracing_cpu(cpu)
3714 tracing_iter_reset(iter, cpu);
3716 tracing_iter_reset(iter, cpu_file);
3719 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
3724 * If we overflowed the seq_file before, then we want
3725 * to just reuse the trace_seq buffer again.
3731 p = s_next(m, p, &l);
3735 trace_event_read_lock();
3736 trace_access_lock(cpu_file);
3740 static void s_stop(struct seq_file *m, void *p)
3742 struct trace_iterator *iter = m->private;
3744 #ifdef CONFIG_TRACER_MAX_TRACE
3745 if (iter->snapshot && iter->trace->use_max_tr)
3749 trace_access_unlock(iter->cpu_file);
3750 trace_event_read_unlock();
3754 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
3755 unsigned long *entries, int cpu)
3757 unsigned long count;
3759 count = ring_buffer_entries_cpu(buf->buffer, cpu);
3761 * If this buffer has skipped entries, then we hold all
3762 * entries for the trace and we need to ignore the
3763 * ones before the time stamp.
3765 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
3766 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
3767 /* total is the same as the entries */
3771 ring_buffer_overrun_cpu(buf->buffer, cpu);
3776 get_total_entries(struct array_buffer *buf,
3777 unsigned long *total, unsigned long *entries)
3785 for_each_tracing_cpu(cpu) {
3786 get_total_entries_cpu(buf, &t, &e, cpu);
3792 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
3794 unsigned long total, entries;
3799 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
3804 unsigned long trace_total_entries(struct trace_array *tr)
3806 unsigned long total, entries;
3811 get_total_entries(&tr->array_buffer, &total, &entries);
3816 static void print_lat_help_header(struct seq_file *m)
3818 seq_puts(m, "# _------=> CPU# \n"
3819 "# / _-----=> irqs-off \n"
3820 "# | / _----=> need-resched \n"
3821 "# || / _---=> hardirq/softirq \n"
3822 "# ||| / _--=> preempt-depth \n"
3824 "# cmd pid ||||| time | caller \n"
3825 "# \\ / ||||| \\ | / \n");
3828 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
3830 unsigned long total;
3831 unsigned long entries;
3833 get_total_entries(buf, &total, &entries);
3834 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
3835 entries, total, num_online_cpus());
3839 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
3842 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3844 print_event_info(buf, m);
3846 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
3847 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
3850 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
3853 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3854 const char *space = " ";
3855 int prec = tgid ? 12 : 2;
3857 print_event_info(buf, m);
3859 seq_printf(m, "# %.*s _-----=> irqs-off\n", prec, space);
3860 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
3861 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
3862 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
3863 seq_printf(m, "# %.*s||| / delay\n", prec, space);
3864 seq_printf(m, "# TASK-PID %.*s CPU# |||| TIMESTAMP FUNCTION\n", prec, " TGID ");
3865 seq_printf(m, "# | | %.*s | |||| | |\n", prec, " | ");
3869 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
3871 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
3872 struct array_buffer *buf = iter->array_buffer;
3873 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
3874 struct tracer *type = iter->trace;
3875 unsigned long entries;
3876 unsigned long total;
3877 const char *name = "preemption";
3881 get_total_entries(buf, &total, &entries);
3883 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
3885 seq_puts(m, "# -----------------------------------"
3886 "---------------------------------\n");
3887 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
3888 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
3889 nsecs_to_usecs(data->saved_latency),
3893 #if defined(CONFIG_PREEMPT_NONE)
3895 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
3897 #elif defined(CONFIG_PREEMPT)
3899 #elif defined(CONFIG_PREEMPT_RT)
3904 /* These are reserved for later use */
3907 seq_printf(m, " #P:%d)\n", num_online_cpus());
3911 seq_puts(m, "# -----------------\n");
3912 seq_printf(m, "# | task: %.16s-%d "
3913 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
3914 data->comm, data->pid,
3915 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
3916 data->policy, data->rt_priority);
3917 seq_puts(m, "# -----------------\n");
3919 if (data->critical_start) {
3920 seq_puts(m, "# => started at: ");
3921 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
3922 trace_print_seq(m, &iter->seq);
3923 seq_puts(m, "\n# => ended at: ");
3924 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
3925 trace_print_seq(m, &iter->seq);
3926 seq_puts(m, "\n#\n");
3932 static void test_cpu_buff_start(struct trace_iterator *iter)
3934 struct trace_seq *s = &iter->seq;
3935 struct trace_array *tr = iter->tr;
3937 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
3940 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
3943 if (cpumask_available(iter->started) &&
3944 cpumask_test_cpu(iter->cpu, iter->started))
3947 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
3950 if (cpumask_available(iter->started))
3951 cpumask_set_cpu(iter->cpu, iter->started);
3953 /* Don't print started cpu buffer for the first entry of the trace */
3955 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
3959 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
3961 struct trace_array *tr = iter->tr;
3962 struct trace_seq *s = &iter->seq;
3963 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
3964 struct trace_entry *entry;
3965 struct trace_event *event;
3969 test_cpu_buff_start(iter);
3971 event = ftrace_find_event(entry->type);
3973 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3974 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
3975 trace_print_lat_context(iter);
3977 trace_print_context(iter);
3980 if (trace_seq_has_overflowed(s))
3981 return TRACE_TYPE_PARTIAL_LINE;
3984 return event->funcs->trace(iter, sym_flags, event);
3986 trace_seq_printf(s, "Unknown type %d\n", entry->type);
3988 return trace_handle_return(s);
3991 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
3993 struct trace_array *tr = iter->tr;
3994 struct trace_seq *s = &iter->seq;
3995 struct trace_entry *entry;
3996 struct trace_event *event;
4000 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4001 trace_seq_printf(s, "%d %d %llu ",
4002 entry->pid, iter->cpu, iter->ts);
4004 if (trace_seq_has_overflowed(s))
4005 return TRACE_TYPE_PARTIAL_LINE;
4007 event = ftrace_find_event(entry->type);
4009 return event->funcs->raw(iter, 0, event);
4011 trace_seq_printf(s, "%d ?\n", entry->type);
4013 return trace_handle_return(s);
4016 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4018 struct trace_array *tr = iter->tr;
4019 struct trace_seq *s = &iter->seq;
4020 unsigned char newline = '\n';
4021 struct trace_entry *entry;
4022 struct trace_event *event;
4026 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4027 SEQ_PUT_HEX_FIELD(s, entry->pid);
4028 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4029 SEQ_PUT_HEX_FIELD(s, iter->ts);
4030 if (trace_seq_has_overflowed(s))
4031 return TRACE_TYPE_PARTIAL_LINE;
4034 event = ftrace_find_event(entry->type);
4036 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4037 if (ret != TRACE_TYPE_HANDLED)
4041 SEQ_PUT_FIELD(s, newline);
4043 return trace_handle_return(s);
4046 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4048 struct trace_array *tr = iter->tr;
4049 struct trace_seq *s = &iter->seq;
4050 struct trace_entry *entry;
4051 struct trace_event *event;
4055 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4056 SEQ_PUT_FIELD(s, entry->pid);
4057 SEQ_PUT_FIELD(s, iter->cpu);
4058 SEQ_PUT_FIELD(s, iter->ts);
4059 if (trace_seq_has_overflowed(s))
4060 return TRACE_TYPE_PARTIAL_LINE;
4063 event = ftrace_find_event(entry->type);
4064 return event ? event->funcs->binary(iter, 0, event) :
4068 int trace_empty(struct trace_iterator *iter)
4070 struct ring_buffer_iter *buf_iter;
4073 /* If we are looking at one CPU buffer, only check that one */
4074 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4075 cpu = iter->cpu_file;
4076 buf_iter = trace_buffer_iter(iter, cpu);
4078 if (!ring_buffer_iter_empty(buf_iter))
4081 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4087 for_each_tracing_cpu(cpu) {
4088 buf_iter = trace_buffer_iter(iter, cpu);
4090 if (!ring_buffer_iter_empty(buf_iter))
4093 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4101 /* Called with trace_event_read_lock() held. */
4102 enum print_line_t print_trace_line(struct trace_iterator *iter)
4104 struct trace_array *tr = iter->tr;
4105 unsigned long trace_flags = tr->trace_flags;
4106 enum print_line_t ret;
4108 if (iter->lost_events) {
4109 if (iter->lost_events == (unsigned long)-1)
4110 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4113 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4114 iter->cpu, iter->lost_events);
4115 if (trace_seq_has_overflowed(&iter->seq))
4116 return TRACE_TYPE_PARTIAL_LINE;
4119 if (iter->trace && iter->trace->print_line) {
4120 ret = iter->trace->print_line(iter);
4121 if (ret != TRACE_TYPE_UNHANDLED)
4125 if (iter->ent->type == TRACE_BPUTS &&
4126 trace_flags & TRACE_ITER_PRINTK &&
4127 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4128 return trace_print_bputs_msg_only(iter);
4130 if (iter->ent->type == TRACE_BPRINT &&
4131 trace_flags & TRACE_ITER_PRINTK &&
4132 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4133 return trace_print_bprintk_msg_only(iter);
4135 if (iter->ent->type == TRACE_PRINT &&
4136 trace_flags & TRACE_ITER_PRINTK &&
4137 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4138 return trace_print_printk_msg_only(iter);
4140 if (trace_flags & TRACE_ITER_BIN)
4141 return print_bin_fmt(iter);
4143 if (trace_flags & TRACE_ITER_HEX)
4144 return print_hex_fmt(iter);
4146 if (trace_flags & TRACE_ITER_RAW)
4147 return print_raw_fmt(iter);
4149 return print_trace_fmt(iter);
4152 void trace_latency_header(struct seq_file *m)
4154 struct trace_iterator *iter = m->private;
4155 struct trace_array *tr = iter->tr;
4157 /* print nothing if the buffers are empty */
4158 if (trace_empty(iter))
4161 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4162 print_trace_header(m, iter);
4164 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4165 print_lat_help_header(m);
4168 void trace_default_header(struct seq_file *m)
4170 struct trace_iterator *iter = m->private;
4171 struct trace_array *tr = iter->tr;
4172 unsigned long trace_flags = tr->trace_flags;
4174 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4177 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4178 /* print nothing if the buffers are empty */
4179 if (trace_empty(iter))
4181 print_trace_header(m, iter);
4182 if (!(trace_flags & TRACE_ITER_VERBOSE))
4183 print_lat_help_header(m);
4185 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4186 if (trace_flags & TRACE_ITER_IRQ_INFO)
4187 print_func_help_header_irq(iter->array_buffer,
4190 print_func_help_header(iter->array_buffer, m,
4196 static void test_ftrace_alive(struct seq_file *m)
4198 if (!ftrace_is_dead())
4200 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4201 "# MAY BE MISSING FUNCTION EVENTS\n");
4204 #ifdef CONFIG_TRACER_MAX_TRACE
4205 static void show_snapshot_main_help(struct seq_file *m)
4207 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4208 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4209 "# Takes a snapshot of the main buffer.\n"
4210 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4211 "# (Doesn't have to be '2' works with any number that\n"
4212 "# is not a '0' or '1')\n");
4215 static void show_snapshot_percpu_help(struct seq_file *m)
4217 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4218 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4219 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4220 "# Takes a snapshot of the main buffer for this cpu.\n");
4222 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4223 "# Must use main snapshot file to allocate.\n");
4225 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4226 "# (Doesn't have to be '2' works with any number that\n"
4227 "# is not a '0' or '1')\n");
4230 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4232 if (iter->tr->allocated_snapshot)
4233 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4235 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4237 seq_puts(m, "# Snapshot commands:\n");
4238 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4239 show_snapshot_main_help(m);
4241 show_snapshot_percpu_help(m);
4244 /* Should never be called */
4245 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4248 static int s_show(struct seq_file *m, void *v)
4250 struct trace_iterator *iter = v;
4253 if (iter->ent == NULL) {
4255 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4257 test_ftrace_alive(m);
4259 if (iter->snapshot && trace_empty(iter))
4260 print_snapshot_help(m, iter);
4261 else if (iter->trace && iter->trace->print_header)
4262 iter->trace->print_header(m);
4264 trace_default_header(m);
4266 } else if (iter->leftover) {
4268 * If we filled the seq_file buffer earlier, we
4269 * want to just show it now.
4271 ret = trace_print_seq(m, &iter->seq);
4273 /* ret should this time be zero, but you never know */
4274 iter->leftover = ret;
4277 print_trace_line(iter);
4278 ret = trace_print_seq(m, &iter->seq);
4280 * If we overflow the seq_file buffer, then it will
4281 * ask us for this data again at start up.
4283 * ret is 0 if seq_file write succeeded.
4286 iter->leftover = ret;
4293 * Should be used after trace_array_get(), trace_types_lock
4294 * ensures that i_cdev was already initialized.
4296 static inline int tracing_get_cpu(struct inode *inode)
4298 if (inode->i_cdev) /* See trace_create_cpu_file() */
4299 return (long)inode->i_cdev - 1;
4300 return RING_BUFFER_ALL_CPUS;
4303 static const struct seq_operations tracer_seq_ops = {
4310 static struct trace_iterator *
4311 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4313 struct trace_array *tr = inode->i_private;
4314 struct trace_iterator *iter;
4317 if (tracing_disabled)
4318 return ERR_PTR(-ENODEV);
4320 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4322 return ERR_PTR(-ENOMEM);
4324 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4326 if (!iter->buffer_iter)
4330 * trace_find_next_entry() may need to save off iter->ent.
4331 * It will place it into the iter->temp buffer. As most
4332 * events are less than 128, allocate a buffer of that size.
4333 * If one is greater, then trace_find_next_entry() will
4334 * allocate a new buffer to adjust for the bigger iter->ent.
4335 * It's not critical if it fails to get allocated here.
4337 iter->temp = kmalloc(128, GFP_KERNEL);
4339 iter->temp_size = 128;
4342 * We make a copy of the current tracer to avoid concurrent
4343 * changes on it while we are reading.
4345 mutex_lock(&trace_types_lock);
4346 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4350 *iter->trace = *tr->current_trace;
4352 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4357 #ifdef CONFIG_TRACER_MAX_TRACE
4358 /* Currently only the top directory has a snapshot */
4359 if (tr->current_trace->print_max || snapshot)
4360 iter->array_buffer = &tr->max_buffer;
4363 iter->array_buffer = &tr->array_buffer;
4364 iter->snapshot = snapshot;
4366 iter->cpu_file = tracing_get_cpu(inode);
4367 mutex_init(&iter->mutex);
4369 /* Notify the tracer early; before we stop tracing. */
4370 if (iter->trace->open)
4371 iter->trace->open(iter);
4373 /* Annotate start of buffers if we had overruns */
4374 if (ring_buffer_overruns(iter->array_buffer->buffer))
4375 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4377 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4378 if (trace_clocks[tr->clock_id].in_ns)
4379 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4382 * If pause-on-trace is enabled, then stop the trace while
4383 * dumping, unless this is the "snapshot" file
4385 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4386 tracing_stop_tr(tr);
4388 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4389 for_each_tracing_cpu(cpu) {
4390 iter->buffer_iter[cpu] =
4391 ring_buffer_read_prepare(iter->array_buffer->buffer,
4394 ring_buffer_read_prepare_sync();
4395 for_each_tracing_cpu(cpu) {
4396 ring_buffer_read_start(iter->buffer_iter[cpu]);
4397 tracing_iter_reset(iter, cpu);
4400 cpu = iter->cpu_file;
4401 iter->buffer_iter[cpu] =
4402 ring_buffer_read_prepare(iter->array_buffer->buffer,
4404 ring_buffer_read_prepare_sync();
4405 ring_buffer_read_start(iter->buffer_iter[cpu]);
4406 tracing_iter_reset(iter, cpu);
4409 mutex_unlock(&trace_types_lock);
4414 mutex_unlock(&trace_types_lock);
4417 kfree(iter->buffer_iter);
4419 seq_release_private(inode, file);
4420 return ERR_PTR(-ENOMEM);
4423 int tracing_open_generic(struct inode *inode, struct file *filp)
4427 ret = tracing_check_open_get_tr(NULL);
4431 filp->private_data = inode->i_private;
4435 bool tracing_is_disabled(void)
4437 return (tracing_disabled) ? true: false;
4441 * Open and update trace_array ref count.
4442 * Must have the current trace_array passed to it.
4444 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4446 struct trace_array *tr = inode->i_private;
4449 ret = tracing_check_open_get_tr(tr);
4453 filp->private_data = inode->i_private;
4458 static int tracing_release(struct inode *inode, struct file *file)
4460 struct trace_array *tr = inode->i_private;
4461 struct seq_file *m = file->private_data;
4462 struct trace_iterator *iter;
4465 if (!(file->f_mode & FMODE_READ)) {
4466 trace_array_put(tr);
4470 /* Writes do not use seq_file */
4472 mutex_lock(&trace_types_lock);
4474 for_each_tracing_cpu(cpu) {
4475 if (iter->buffer_iter[cpu])
4476 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4479 if (iter->trace && iter->trace->close)
4480 iter->trace->close(iter);
4482 if (!iter->snapshot && tr->stop_count)
4483 /* reenable tracing if it was previously enabled */
4484 tracing_start_tr(tr);
4486 __trace_array_put(tr);
4488 mutex_unlock(&trace_types_lock);
4490 mutex_destroy(&iter->mutex);
4491 free_cpumask_var(iter->started);
4494 kfree(iter->buffer_iter);
4495 seq_release_private(inode, file);
4500 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4502 struct trace_array *tr = inode->i_private;
4504 trace_array_put(tr);
4508 static int tracing_single_release_tr(struct inode *inode, struct file *file)
4510 struct trace_array *tr = inode->i_private;
4512 trace_array_put(tr);
4514 return single_release(inode, file);
4517 static int tracing_open(struct inode *inode, struct file *file)
4519 struct trace_array *tr = inode->i_private;
4520 struct trace_iterator *iter;
4523 ret = tracing_check_open_get_tr(tr);
4527 /* If this file was open for write, then erase contents */
4528 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4529 int cpu = tracing_get_cpu(inode);
4530 struct array_buffer *trace_buf = &tr->array_buffer;
4532 #ifdef CONFIG_TRACER_MAX_TRACE
4533 if (tr->current_trace->print_max)
4534 trace_buf = &tr->max_buffer;
4537 if (cpu == RING_BUFFER_ALL_CPUS)
4538 tracing_reset_online_cpus(trace_buf);
4540 tracing_reset_cpu(trace_buf, cpu);
4543 if (file->f_mode & FMODE_READ) {
4544 iter = __tracing_open(inode, file, false);
4546 ret = PTR_ERR(iter);
4547 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4548 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4552 trace_array_put(tr);
4558 * Some tracers are not suitable for instance buffers.
4559 * A tracer is always available for the global array (toplevel)
4560 * or if it explicitly states that it is.
4563 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4565 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4568 /* Find the next tracer that this trace array may use */
4569 static struct tracer *
4570 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4572 while (t && !trace_ok_for_array(t, tr))
4579 t_next(struct seq_file *m, void *v, loff_t *pos)
4581 struct trace_array *tr = m->private;
4582 struct tracer *t = v;
4587 t = get_tracer_for_array(tr, t->next);
4592 static void *t_start(struct seq_file *m, loff_t *pos)
4594 struct trace_array *tr = m->private;
4598 mutex_lock(&trace_types_lock);
4600 t = get_tracer_for_array(tr, trace_types);
4601 for (; t && l < *pos; t = t_next(m, t, &l))
4607 static void t_stop(struct seq_file *m, void *p)
4609 mutex_unlock(&trace_types_lock);
4612 static int t_show(struct seq_file *m, void *v)
4614 struct tracer *t = v;
4619 seq_puts(m, t->name);
4628 static const struct seq_operations show_traces_seq_ops = {
4635 static int show_traces_open(struct inode *inode, struct file *file)
4637 struct trace_array *tr = inode->i_private;
4641 ret = tracing_check_open_get_tr(tr);
4645 ret = seq_open(file, &show_traces_seq_ops);
4647 trace_array_put(tr);
4651 m = file->private_data;
4657 static int show_traces_release(struct inode *inode, struct file *file)
4659 struct trace_array *tr = inode->i_private;
4661 trace_array_put(tr);
4662 return seq_release(inode, file);
4666 tracing_write_stub(struct file *filp, const char __user *ubuf,
4667 size_t count, loff_t *ppos)
4672 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
4676 if (file->f_mode & FMODE_READ)
4677 ret = seq_lseek(file, offset, whence);
4679 file->f_pos = ret = 0;
4684 static const struct file_operations tracing_fops = {
4685 .open = tracing_open,
4687 .write = tracing_write_stub,
4688 .llseek = tracing_lseek,
4689 .release = tracing_release,
4692 static const struct file_operations show_traces_fops = {
4693 .open = show_traces_open,
4695 .llseek = seq_lseek,
4696 .release = show_traces_release,
4700 tracing_cpumask_read(struct file *filp, char __user *ubuf,
4701 size_t count, loff_t *ppos)
4703 struct trace_array *tr = file_inode(filp)->i_private;
4707 len = snprintf(NULL, 0, "%*pb\n",
4708 cpumask_pr_args(tr->tracing_cpumask)) + 1;
4709 mask_str = kmalloc(len, GFP_KERNEL);
4713 len = snprintf(mask_str, len, "%*pb\n",
4714 cpumask_pr_args(tr->tracing_cpumask));
4719 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
4727 int tracing_set_cpumask(struct trace_array *tr,
4728 cpumask_var_t tracing_cpumask_new)
4735 local_irq_disable();
4736 arch_spin_lock(&tr->max_lock);
4737 for_each_tracing_cpu(cpu) {
4739 * Increase/decrease the disabled counter if we are
4740 * about to flip a bit in the cpumask:
4742 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4743 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4744 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
4745 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
4747 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4748 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4749 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
4750 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
4753 arch_spin_unlock(&tr->max_lock);
4756 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
4762 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
4763 size_t count, loff_t *ppos)
4765 struct trace_array *tr = file_inode(filp)->i_private;
4766 cpumask_var_t tracing_cpumask_new;
4769 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
4772 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
4776 err = tracing_set_cpumask(tr, tracing_cpumask_new);
4780 free_cpumask_var(tracing_cpumask_new);
4785 free_cpumask_var(tracing_cpumask_new);
4790 static const struct file_operations tracing_cpumask_fops = {
4791 .open = tracing_open_generic_tr,
4792 .read = tracing_cpumask_read,
4793 .write = tracing_cpumask_write,
4794 .release = tracing_release_generic_tr,
4795 .llseek = generic_file_llseek,
4798 static int tracing_trace_options_show(struct seq_file *m, void *v)
4800 struct tracer_opt *trace_opts;
4801 struct trace_array *tr = m->private;
4805 mutex_lock(&trace_types_lock);
4806 tracer_flags = tr->current_trace->flags->val;
4807 trace_opts = tr->current_trace->flags->opts;
4809 for (i = 0; trace_options[i]; i++) {
4810 if (tr->trace_flags & (1 << i))
4811 seq_printf(m, "%s\n", trace_options[i]);
4813 seq_printf(m, "no%s\n", trace_options[i]);
4816 for (i = 0; trace_opts[i].name; i++) {
4817 if (tracer_flags & trace_opts[i].bit)
4818 seq_printf(m, "%s\n", trace_opts[i].name);
4820 seq_printf(m, "no%s\n", trace_opts[i].name);
4822 mutex_unlock(&trace_types_lock);
4827 static int __set_tracer_option(struct trace_array *tr,
4828 struct tracer_flags *tracer_flags,
4829 struct tracer_opt *opts, int neg)
4831 struct tracer *trace = tracer_flags->trace;
4834 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
4839 tracer_flags->val &= ~opts->bit;
4841 tracer_flags->val |= opts->bit;
4845 /* Try to assign a tracer specific option */
4846 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
4848 struct tracer *trace = tr->current_trace;
4849 struct tracer_flags *tracer_flags = trace->flags;
4850 struct tracer_opt *opts = NULL;
4853 for (i = 0; tracer_flags->opts[i].name; i++) {
4854 opts = &tracer_flags->opts[i];
4856 if (strcmp(cmp, opts->name) == 0)
4857 return __set_tracer_option(tr, trace->flags, opts, neg);
4863 /* Some tracers require overwrite to stay enabled */
4864 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
4866 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
4872 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
4876 if ((mask == TRACE_ITER_RECORD_TGID) ||
4877 (mask == TRACE_ITER_RECORD_CMD))
4878 lockdep_assert_held(&event_mutex);
4880 /* do nothing if flag is already set */
4881 if (!!(tr->trace_flags & mask) == !!enabled)
4884 /* Give the tracer a chance to approve the change */
4885 if (tr->current_trace->flag_changed)
4886 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
4890 tr->trace_flags |= mask;
4892 tr->trace_flags &= ~mask;
4894 if (mask == TRACE_ITER_RECORD_CMD)
4895 trace_event_enable_cmd_record(enabled);
4897 if (mask == TRACE_ITER_RECORD_TGID) {
4899 tgid_map_max = pid_max;
4900 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
4904 * Pairs with smp_load_acquire() in
4905 * trace_find_tgid_ptr() to ensure that if it observes
4906 * the tgid_map we just allocated then it also observes
4907 * the corresponding tgid_map_max value.
4909 smp_store_release(&tgid_map, map);
4912 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
4916 trace_event_enable_tgid_record(enabled);
4919 if (mask == TRACE_ITER_EVENT_FORK)
4920 trace_event_follow_fork(tr, enabled);
4922 if (mask == TRACE_ITER_FUNC_FORK)
4923 ftrace_pid_follow_fork(tr, enabled);
4925 if (mask == TRACE_ITER_OVERWRITE) {
4926 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
4927 #ifdef CONFIG_TRACER_MAX_TRACE
4928 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
4932 if (mask == TRACE_ITER_PRINTK) {
4933 trace_printk_start_stop_comm(enabled);
4934 trace_printk_control(enabled);
4940 int trace_set_options(struct trace_array *tr, char *option)
4945 size_t orig_len = strlen(option);
4948 cmp = strstrip(option);
4950 len = str_has_prefix(cmp, "no");
4956 mutex_lock(&event_mutex);
4957 mutex_lock(&trace_types_lock);
4959 ret = match_string(trace_options, -1, cmp);
4960 /* If no option could be set, test the specific tracer options */
4962 ret = set_tracer_option(tr, cmp, neg);
4964 ret = set_tracer_flag(tr, 1 << ret, !neg);
4966 mutex_unlock(&trace_types_lock);
4967 mutex_unlock(&event_mutex);
4970 * If the first trailing whitespace is replaced with '\0' by strstrip,
4971 * turn it back into a space.
4973 if (orig_len > strlen(option))
4974 option[strlen(option)] = ' ';
4979 static void __init apply_trace_boot_options(void)
4981 char *buf = trace_boot_options_buf;
4985 option = strsep(&buf, ",");
4991 trace_set_options(&global_trace, option);
4993 /* Put back the comma to allow this to be called again */
5000 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5001 size_t cnt, loff_t *ppos)
5003 struct seq_file *m = filp->private_data;
5004 struct trace_array *tr = m->private;
5008 if (cnt >= sizeof(buf))
5011 if (copy_from_user(buf, ubuf, cnt))
5016 ret = trace_set_options(tr, buf);
5025 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5027 struct trace_array *tr = inode->i_private;
5030 ret = tracing_check_open_get_tr(tr);
5034 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5036 trace_array_put(tr);
5041 static const struct file_operations tracing_iter_fops = {
5042 .open = tracing_trace_options_open,
5044 .llseek = seq_lseek,
5045 .release = tracing_single_release_tr,
5046 .write = tracing_trace_options_write,
5049 static const char readme_msg[] =
5050 "tracing mini-HOWTO:\n\n"
5051 "# echo 0 > tracing_on : quick way to disable tracing\n"
5052 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5053 " Important files:\n"
5054 " trace\t\t\t- The static contents of the buffer\n"
5055 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5056 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5057 " current_tracer\t- function and latency tracers\n"
5058 " available_tracers\t- list of configured tracers for current_tracer\n"
5059 " error_log\t- error log for failed commands (that support it)\n"
5060 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5061 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5062 " trace_clock\t\t-change the clock used to order events\n"
5063 " local: Per cpu clock but may not be synced across CPUs\n"
5064 " global: Synced across CPUs but slows tracing down.\n"
5065 " counter: Not a clock, but just an increment\n"
5066 " uptime: Jiffy counter from time of boot\n"
5067 " perf: Same clock that perf events use\n"
5068 #ifdef CONFIG_X86_64
5069 " x86-tsc: TSC cycle counter\n"
5071 "\n timestamp_mode\t-view the mode used to timestamp events\n"
5072 " delta: Delta difference against a buffer-wide timestamp\n"
5073 " absolute: Absolute (standalone) timestamp\n"
5074 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5075 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5076 " tracing_cpumask\t- Limit which CPUs to trace\n"
5077 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5078 "\t\t\t Remove sub-buffer with rmdir\n"
5079 " trace_options\t\t- Set format or modify how tracing happens\n"
5080 "\t\t\t Disable an option by prefixing 'no' to the\n"
5081 "\t\t\t option name\n"
5082 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5083 #ifdef CONFIG_DYNAMIC_FTRACE
5084 "\n available_filter_functions - list of functions that can be filtered on\n"
5085 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5086 "\t\t\t functions\n"
5087 "\t accepts: func_full_name or glob-matching-pattern\n"
5088 "\t modules: Can select a group via module\n"
5089 "\t Format: :mod:<module-name>\n"
5090 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5091 "\t triggers: a command to perform when function is hit\n"
5092 "\t Format: <function>:<trigger>[:count]\n"
5093 "\t trigger: traceon, traceoff\n"
5094 "\t\t enable_event:<system>:<event>\n"
5095 "\t\t disable_event:<system>:<event>\n"
5096 #ifdef CONFIG_STACKTRACE
5099 #ifdef CONFIG_TRACER_SNAPSHOT
5104 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5105 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5106 "\t The first one will disable tracing every time do_fault is hit\n"
5107 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5108 "\t The first time do trap is hit and it disables tracing, the\n"
5109 "\t counter will decrement to 2. If tracing is already disabled,\n"
5110 "\t the counter will not decrement. It only decrements when the\n"
5111 "\t trigger did work\n"
5112 "\t To remove trigger without count:\n"
5113 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5114 "\t To remove trigger with a count:\n"
5115 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5116 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5117 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5118 "\t modules: Can select a group via module command :mod:\n"
5119 "\t Does not accept triggers\n"
5120 #endif /* CONFIG_DYNAMIC_FTRACE */
5121 #ifdef CONFIG_FUNCTION_TRACER
5122 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5124 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5127 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5128 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5129 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5130 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5132 #ifdef CONFIG_TRACER_SNAPSHOT
5133 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5134 "\t\t\t snapshot buffer. Read the contents for more\n"
5135 "\t\t\t information\n"
5137 #ifdef CONFIG_STACK_TRACER
5138 " stack_trace\t\t- Shows the max stack trace when active\n"
5139 " stack_max_size\t- Shows current max stack size that was traced\n"
5140 "\t\t\t Write into this file to reset the max size (trigger a\n"
5141 "\t\t\t new trace)\n"
5142 #ifdef CONFIG_DYNAMIC_FTRACE
5143 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5146 #endif /* CONFIG_STACK_TRACER */
5147 #ifdef CONFIG_DYNAMIC_EVENTS
5148 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5149 "\t\t\t Write into this file to define/undefine new trace events.\n"
5151 #ifdef CONFIG_KPROBE_EVENTS
5152 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5153 "\t\t\t Write into this file to define/undefine new trace events.\n"
5155 #ifdef CONFIG_UPROBE_EVENTS
5156 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5157 "\t\t\t Write into this file to define/undefine new trace events.\n"
5159 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5160 "\t accepts: event-definitions (one definition per line)\n"
5161 "\t Format: p[:[<group>/]<event>] <place> [<args>]\n"
5162 "\t r[maxactive][:[<group>/]<event>] <place> [<args>]\n"
5163 #ifdef CONFIG_HIST_TRIGGERS
5164 "\t s:[synthetic/]<event> <field> [<field>]\n"
5166 "\t -:[<group>/]<event>\n"
5167 #ifdef CONFIG_KPROBE_EVENTS
5168 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5169 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5171 #ifdef CONFIG_UPROBE_EVENTS
5172 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5174 "\t args: <name>=fetcharg[:type]\n"
5175 "\t fetcharg: %<register>, @<address>, @<symbol>[+|-<offset>],\n"
5176 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5177 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5179 "\t $stack<index>, $stack, $retval, $comm,\n"
5181 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5182 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string, symbol,\n"
5183 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5184 "\t <type>\\[<array-size>\\]\n"
5185 #ifdef CONFIG_HIST_TRIGGERS
5186 "\t field: <stype> <name>;\n"
5187 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5188 "\t [unsigned] char/int/long\n"
5191 " events/\t\t- Directory containing all trace event subsystems:\n"
5192 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5193 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5194 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5196 " filter\t\t- If set, only events passing filter are traced\n"
5197 " events/<system>/<event>/\t- Directory containing control files for\n"
5199 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5200 " filter\t\t- If set, only events passing filter are traced\n"
5201 " trigger\t\t- If set, a command to perform when event is hit\n"
5202 "\t Format: <trigger>[:count][if <filter>]\n"
5203 "\t trigger: traceon, traceoff\n"
5204 "\t enable_event:<system>:<event>\n"
5205 "\t disable_event:<system>:<event>\n"
5206 #ifdef CONFIG_HIST_TRIGGERS
5207 "\t enable_hist:<system>:<event>\n"
5208 "\t disable_hist:<system>:<event>\n"
5210 #ifdef CONFIG_STACKTRACE
5213 #ifdef CONFIG_TRACER_SNAPSHOT
5216 #ifdef CONFIG_HIST_TRIGGERS
5217 "\t\t hist (see below)\n"
5219 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5220 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5221 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5222 "\t events/block/block_unplug/trigger\n"
5223 "\t The first disables tracing every time block_unplug is hit.\n"
5224 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5225 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5226 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5227 "\t Like function triggers, the counter is only decremented if it\n"
5228 "\t enabled or disabled tracing.\n"
5229 "\t To remove a trigger without a count:\n"
5230 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5231 "\t To remove a trigger with a count:\n"
5232 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5233 "\t Filters can be ignored when removing a trigger.\n"
5234 #ifdef CONFIG_HIST_TRIGGERS
5235 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5236 "\t Format: hist:keys=<field1[,field2,...]>\n"
5237 "\t [:values=<field1[,field2,...]>]\n"
5238 "\t [:sort=<field1[,field2,...]>]\n"
5239 "\t [:size=#entries]\n"
5240 "\t [:pause][:continue][:clear]\n"
5241 "\t [:name=histname1]\n"
5242 "\t [:<handler>.<action>]\n"
5243 "\t [if <filter>]\n\n"
5244 "\t Note, special fields can be used as well:\n"
5245 "\t common_timestamp - to record current timestamp\n"
5246 "\t common_cpu - to record the CPU the event happened on\n"
5248 "\t When a matching event is hit, an entry is added to a hash\n"
5249 "\t table using the key(s) and value(s) named, and the value of a\n"
5250 "\t sum called 'hitcount' is incremented. Keys and values\n"
5251 "\t correspond to fields in the event's format description. Keys\n"
5252 "\t can be any field, or the special string 'stacktrace'.\n"
5253 "\t Compound keys consisting of up to two fields can be specified\n"
5254 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5255 "\t fields. Sort keys consisting of up to two fields can be\n"
5256 "\t specified using the 'sort' keyword. The sort direction can\n"
5257 "\t be modified by appending '.descending' or '.ascending' to a\n"
5258 "\t sort field. The 'size' parameter can be used to specify more\n"
5259 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5260 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5261 "\t its histogram data will be shared with other triggers of the\n"
5262 "\t same name, and trigger hits will update this common data.\n\n"
5263 "\t Reading the 'hist' file for the event will dump the hash\n"
5264 "\t table in its entirety to stdout. If there are multiple hist\n"
5265 "\t triggers attached to an event, there will be a table for each\n"
5266 "\t trigger in the output. The table displayed for a named\n"
5267 "\t trigger will be the same as any other instance having the\n"
5268 "\t same name. The default format used to display a given field\n"
5269 "\t can be modified by appending any of the following modifiers\n"
5270 "\t to the field name, as applicable:\n\n"
5271 "\t .hex display a number as a hex value\n"
5272 "\t .sym display an address as a symbol\n"
5273 "\t .sym-offset display an address as a symbol and offset\n"
5274 "\t .execname display a common_pid as a program name\n"
5275 "\t .syscall display a syscall id as a syscall name\n"
5276 "\t .log2 display log2 value rather than raw number\n"
5277 "\t .usecs display a common_timestamp in microseconds\n\n"
5278 "\t The 'pause' parameter can be used to pause an existing hist\n"
5279 "\t trigger or to start a hist trigger but not log any events\n"
5280 "\t until told to do so. 'continue' can be used to start or\n"
5281 "\t restart a paused hist trigger.\n\n"
5282 "\t The 'clear' parameter will clear the contents of a running\n"
5283 "\t hist trigger and leave its current paused/active state\n"
5285 "\t The enable_hist and disable_hist triggers can be used to\n"
5286 "\t have one event conditionally start and stop another event's\n"
5287 "\t already-attached hist trigger. The syntax is analogous to\n"
5288 "\t the enable_event and disable_event triggers.\n\n"
5289 "\t Hist trigger handlers and actions are executed whenever a\n"
5290 "\t a histogram entry is added or updated. They take the form:\n\n"
5291 "\t <handler>.<action>\n\n"
5292 "\t The available handlers are:\n\n"
5293 "\t onmatch(matching.event) - invoke on addition or update\n"
5294 "\t onmax(var) - invoke if var exceeds current max\n"
5295 "\t onchange(var) - invoke action if var changes\n\n"
5296 "\t The available actions are:\n\n"
5297 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5298 "\t save(field,...) - save current event fields\n"
5299 #ifdef CONFIG_TRACER_SNAPSHOT
5300 "\t snapshot() - snapshot the trace buffer\n\n"
5302 #ifdef CONFIG_SYNTH_EVENTS
5303 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5304 "\t Write into this file to define/undefine new synthetic events.\n"
5305 "\t example: echo 'myevent u64 lat; char name[]' >> synthetic_events\n"
5311 tracing_readme_read(struct file *filp, char __user *ubuf,
5312 size_t cnt, loff_t *ppos)
5314 return simple_read_from_buffer(ubuf, cnt, ppos,
5315 readme_msg, strlen(readme_msg));
5318 static const struct file_operations tracing_readme_fops = {
5319 .open = tracing_open_generic,
5320 .read = tracing_readme_read,
5321 .llseek = generic_file_llseek,
5324 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5328 return trace_find_tgid_ptr(pid);
5331 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5335 return trace_find_tgid_ptr(pid);
5338 static void saved_tgids_stop(struct seq_file *m, void *v)
5342 static int saved_tgids_show(struct seq_file *m, void *v)
5344 int *entry = (int *)v;
5345 int pid = entry - tgid_map;
5351 seq_printf(m, "%d %d\n", pid, tgid);
5355 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5356 .start = saved_tgids_start,
5357 .stop = saved_tgids_stop,
5358 .next = saved_tgids_next,
5359 .show = saved_tgids_show,
5362 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5366 ret = tracing_check_open_get_tr(NULL);
5370 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5374 static const struct file_operations tracing_saved_tgids_fops = {
5375 .open = tracing_saved_tgids_open,
5377 .llseek = seq_lseek,
5378 .release = seq_release,
5381 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5383 unsigned int *ptr = v;
5385 if (*pos || m->count)
5390 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5392 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5401 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5407 arch_spin_lock(&trace_cmdline_lock);
5409 v = &savedcmd->map_cmdline_to_pid[0];
5411 v = saved_cmdlines_next(m, v, &l);
5419 static void saved_cmdlines_stop(struct seq_file *m, void *v)
5421 arch_spin_unlock(&trace_cmdline_lock);
5425 static int saved_cmdlines_show(struct seq_file *m, void *v)
5427 char buf[TASK_COMM_LEN];
5428 unsigned int *pid = v;
5430 __trace_find_cmdline(*pid, buf);
5431 seq_printf(m, "%d %s\n", *pid, buf);
5435 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5436 .start = saved_cmdlines_start,
5437 .next = saved_cmdlines_next,
5438 .stop = saved_cmdlines_stop,
5439 .show = saved_cmdlines_show,
5442 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5446 ret = tracing_check_open_get_tr(NULL);
5450 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5453 static const struct file_operations tracing_saved_cmdlines_fops = {
5454 .open = tracing_saved_cmdlines_open,
5456 .llseek = seq_lseek,
5457 .release = seq_release,
5461 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5462 size_t cnt, loff_t *ppos)
5467 arch_spin_lock(&trace_cmdline_lock);
5468 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5469 arch_spin_unlock(&trace_cmdline_lock);
5471 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5474 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5476 kfree(s->saved_cmdlines);
5477 kfree(s->map_cmdline_to_pid);
5481 static int tracing_resize_saved_cmdlines(unsigned int val)
5483 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5485 s = kmalloc(sizeof(*s), GFP_KERNEL);
5489 if (allocate_cmdlines_buffer(val, s) < 0) {
5494 arch_spin_lock(&trace_cmdline_lock);
5495 savedcmd_temp = savedcmd;
5497 arch_spin_unlock(&trace_cmdline_lock);
5498 free_saved_cmdlines_buffer(savedcmd_temp);
5504 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
5505 size_t cnt, loff_t *ppos)
5510 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5514 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
5515 if (!val || val > PID_MAX_DEFAULT)
5518 ret = tracing_resize_saved_cmdlines((unsigned int)val);
5527 static const struct file_operations tracing_saved_cmdlines_size_fops = {
5528 .open = tracing_open_generic,
5529 .read = tracing_saved_cmdlines_size_read,
5530 .write = tracing_saved_cmdlines_size_write,
5533 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
5534 static union trace_eval_map_item *
5535 update_eval_map(union trace_eval_map_item *ptr)
5537 if (!ptr->map.eval_string) {
5538 if (ptr->tail.next) {
5539 ptr = ptr->tail.next;
5540 /* Set ptr to the next real item (skip head) */
5548 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5550 union trace_eval_map_item *ptr = v;
5553 * Paranoid! If ptr points to end, we don't want to increment past it.
5554 * This really should never happen.
5557 ptr = update_eval_map(ptr);
5558 if (WARN_ON_ONCE(!ptr))
5562 ptr = update_eval_map(ptr);
5567 static void *eval_map_start(struct seq_file *m, loff_t *pos)
5569 union trace_eval_map_item *v;
5572 mutex_lock(&trace_eval_mutex);
5574 v = trace_eval_maps;
5578 while (v && l < *pos) {
5579 v = eval_map_next(m, v, &l);
5585 static void eval_map_stop(struct seq_file *m, void *v)
5587 mutex_unlock(&trace_eval_mutex);
5590 static int eval_map_show(struct seq_file *m, void *v)
5592 union trace_eval_map_item *ptr = v;
5594 seq_printf(m, "%s %ld (%s)\n",
5595 ptr->map.eval_string, ptr->map.eval_value,
5601 static const struct seq_operations tracing_eval_map_seq_ops = {
5602 .start = eval_map_start,
5603 .next = eval_map_next,
5604 .stop = eval_map_stop,
5605 .show = eval_map_show,
5608 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5612 ret = tracing_check_open_get_tr(NULL);
5616 return seq_open(filp, &tracing_eval_map_seq_ops);
5619 static const struct file_operations tracing_eval_map_fops = {
5620 .open = tracing_eval_map_open,
5622 .llseek = seq_lseek,
5623 .release = seq_release,
5626 static inline union trace_eval_map_item *
5627 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
5629 /* Return tail of array given the head */
5630 return ptr + ptr->head.length + 1;
5634 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
5637 struct trace_eval_map **stop;
5638 struct trace_eval_map **map;
5639 union trace_eval_map_item *map_array;
5640 union trace_eval_map_item *ptr;
5645 * The trace_eval_maps contains the map plus a head and tail item,
5646 * where the head holds the module and length of array, and the
5647 * tail holds a pointer to the next list.
5649 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
5651 pr_warn("Unable to allocate trace eval mapping\n");
5655 mutex_lock(&trace_eval_mutex);
5657 if (!trace_eval_maps)
5658 trace_eval_maps = map_array;
5660 ptr = trace_eval_maps;
5662 ptr = trace_eval_jmp_to_tail(ptr);
5663 if (!ptr->tail.next)
5665 ptr = ptr->tail.next;
5668 ptr->tail.next = map_array;
5670 map_array->head.mod = mod;
5671 map_array->head.length = len;
5674 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
5675 map_array->map = **map;
5678 memset(map_array, 0, sizeof(*map_array));
5680 mutex_unlock(&trace_eval_mutex);
5683 static void trace_create_eval_file(struct dentry *d_tracer)
5685 trace_create_file("eval_map", 0444, d_tracer,
5686 NULL, &tracing_eval_map_fops);
5689 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
5690 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
5691 static inline void trace_insert_eval_map_file(struct module *mod,
5692 struct trace_eval_map **start, int len) { }
5693 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
5695 static void trace_insert_eval_map(struct module *mod,
5696 struct trace_eval_map **start, int len)
5698 struct trace_eval_map **map;
5705 trace_event_eval_update(map, len);
5707 trace_insert_eval_map_file(mod, start, len);
5711 tracing_set_trace_read(struct file *filp, char __user *ubuf,
5712 size_t cnt, loff_t *ppos)
5714 struct trace_array *tr = filp->private_data;
5715 char buf[MAX_TRACER_SIZE+2];
5718 mutex_lock(&trace_types_lock);
5719 r = sprintf(buf, "%s\n", tr->current_trace->name);
5720 mutex_unlock(&trace_types_lock);
5722 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5725 int tracer_init(struct tracer *t, struct trace_array *tr)
5727 tracing_reset_online_cpus(&tr->array_buffer);
5731 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
5735 for_each_tracing_cpu(cpu)
5736 per_cpu_ptr(buf->data, cpu)->entries = val;
5739 #ifdef CONFIG_TRACER_MAX_TRACE
5740 /* resize @tr's buffer to the size of @size_tr's entries */
5741 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
5742 struct array_buffer *size_buf, int cpu_id)
5746 if (cpu_id == RING_BUFFER_ALL_CPUS) {
5747 for_each_tracing_cpu(cpu) {
5748 ret = ring_buffer_resize(trace_buf->buffer,
5749 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
5752 per_cpu_ptr(trace_buf->data, cpu)->entries =
5753 per_cpu_ptr(size_buf->data, cpu)->entries;
5756 ret = ring_buffer_resize(trace_buf->buffer,
5757 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
5759 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
5760 per_cpu_ptr(size_buf->data, cpu_id)->entries;
5765 #endif /* CONFIG_TRACER_MAX_TRACE */
5767 static int __tracing_resize_ring_buffer(struct trace_array *tr,
5768 unsigned long size, int cpu)
5773 * If kernel or user changes the size of the ring buffer
5774 * we use the size that was given, and we can forget about
5775 * expanding it later.
5777 ring_buffer_expanded = true;
5779 /* May be called before buffers are initialized */
5780 if (!tr->array_buffer.buffer)
5783 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
5787 #ifdef CONFIG_TRACER_MAX_TRACE
5788 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
5789 !tr->current_trace->use_max_tr)
5792 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
5794 int r = resize_buffer_duplicate_size(&tr->array_buffer,
5795 &tr->array_buffer, cpu);
5798 * AARGH! We are left with different
5799 * size max buffer!!!!
5800 * The max buffer is our "snapshot" buffer.
5801 * When a tracer needs a snapshot (one of the
5802 * latency tracers), it swaps the max buffer
5803 * with the saved snap shot. We succeeded to
5804 * update the size of the main buffer, but failed to
5805 * update the size of the max buffer. But when we tried
5806 * to reset the main buffer to the original size, we
5807 * failed there too. This is very unlikely to
5808 * happen, but if it does, warn and kill all
5812 tracing_disabled = 1;
5817 if (cpu == RING_BUFFER_ALL_CPUS)
5818 set_buffer_entries(&tr->max_buffer, size);
5820 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
5823 #endif /* CONFIG_TRACER_MAX_TRACE */
5825 if (cpu == RING_BUFFER_ALL_CPUS)
5826 set_buffer_entries(&tr->array_buffer, size);
5828 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
5833 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
5834 unsigned long size, int cpu_id)
5838 mutex_lock(&trace_types_lock);
5840 if (cpu_id != RING_BUFFER_ALL_CPUS) {
5841 /* make sure, this cpu is enabled in the mask */
5842 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
5848 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
5853 mutex_unlock(&trace_types_lock);
5860 * tracing_update_buffers - used by tracing facility to expand ring buffers
5862 * To save on memory when the tracing is never used on a system with it
5863 * configured in. The ring buffers are set to a minimum size. But once
5864 * a user starts to use the tracing facility, then they need to grow
5865 * to their default size.
5867 * This function is to be called when a tracer is about to be used.
5869 int tracing_update_buffers(void)
5873 mutex_lock(&trace_types_lock);
5874 if (!ring_buffer_expanded)
5875 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
5876 RING_BUFFER_ALL_CPUS);
5877 mutex_unlock(&trace_types_lock);
5882 struct trace_option_dentry;
5885 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
5888 * Used to clear out the tracer before deletion of an instance.
5889 * Must have trace_types_lock held.
5891 static void tracing_set_nop(struct trace_array *tr)
5893 if (tr->current_trace == &nop_trace)
5896 tr->current_trace->enabled--;
5898 if (tr->current_trace->reset)
5899 tr->current_trace->reset(tr);
5901 tr->current_trace = &nop_trace;
5904 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
5906 /* Only enable if the directory has been created already. */
5910 create_trace_option_files(tr, t);
5913 int tracing_set_tracer(struct trace_array *tr, const char *buf)
5916 #ifdef CONFIG_TRACER_MAX_TRACE
5921 mutex_lock(&trace_types_lock);
5923 if (!ring_buffer_expanded) {
5924 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
5925 RING_BUFFER_ALL_CPUS);
5931 for (t = trace_types; t; t = t->next) {
5932 if (strcmp(t->name, buf) == 0)
5939 if (t == tr->current_trace)
5942 #ifdef CONFIG_TRACER_SNAPSHOT
5943 if (t->use_max_tr) {
5944 arch_spin_lock(&tr->max_lock);
5945 if (tr->cond_snapshot)
5947 arch_spin_unlock(&tr->max_lock);
5952 /* Some tracers won't work on kernel command line */
5953 if (system_state < SYSTEM_RUNNING && t->noboot) {
5954 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
5959 /* Some tracers are only allowed for the top level buffer */
5960 if (!trace_ok_for_array(t, tr)) {
5965 /* If trace pipe files are being read, we can't change the tracer */
5966 if (tr->trace_ref) {
5971 trace_branch_disable();
5973 tr->current_trace->enabled--;
5975 if (tr->current_trace->reset)
5976 tr->current_trace->reset(tr);
5978 /* Current trace needs to be nop_trace before synchronize_rcu */
5979 tr->current_trace = &nop_trace;
5981 #ifdef CONFIG_TRACER_MAX_TRACE
5982 had_max_tr = tr->allocated_snapshot;
5984 if (had_max_tr && !t->use_max_tr) {
5986 * We need to make sure that the update_max_tr sees that
5987 * current_trace changed to nop_trace to keep it from
5988 * swapping the buffers after we resize it.
5989 * The update_max_tr is called from interrupts disabled
5990 * so a synchronized_sched() is sufficient.
5997 #ifdef CONFIG_TRACER_MAX_TRACE
5998 if (t->use_max_tr && !had_max_tr) {
5999 ret = tracing_alloc_snapshot_instance(tr);
6006 ret = tracer_init(t, tr);
6011 tr->current_trace = t;
6012 tr->current_trace->enabled++;
6013 trace_branch_enable(tr);
6015 mutex_unlock(&trace_types_lock);
6021 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6022 size_t cnt, loff_t *ppos)
6024 struct trace_array *tr = filp->private_data;
6025 char buf[MAX_TRACER_SIZE+1];
6032 if (cnt > MAX_TRACER_SIZE)
6033 cnt = MAX_TRACER_SIZE;
6035 if (copy_from_user(buf, ubuf, cnt))
6040 /* strip ending whitespace. */
6041 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
6044 err = tracing_set_tracer(tr, buf);
6054 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6055 size_t cnt, loff_t *ppos)
6060 r = snprintf(buf, sizeof(buf), "%ld\n",
6061 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6062 if (r > sizeof(buf))
6064 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6068 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6069 size_t cnt, loff_t *ppos)
6074 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6084 tracing_thresh_read(struct file *filp, char __user *ubuf,
6085 size_t cnt, loff_t *ppos)
6087 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6091 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6092 size_t cnt, loff_t *ppos)
6094 struct trace_array *tr = filp->private_data;
6097 mutex_lock(&trace_types_lock);
6098 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6102 if (tr->current_trace->update_thresh) {
6103 ret = tr->current_trace->update_thresh(tr);
6110 mutex_unlock(&trace_types_lock);
6115 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
6118 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6119 size_t cnt, loff_t *ppos)
6121 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6125 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6126 size_t cnt, loff_t *ppos)
6128 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6133 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6135 struct trace_array *tr = inode->i_private;
6136 struct trace_iterator *iter;
6139 ret = tracing_check_open_get_tr(tr);
6143 mutex_lock(&trace_types_lock);
6145 /* create a buffer to store the information to pass to userspace */
6146 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6149 __trace_array_put(tr);
6153 trace_seq_init(&iter->seq);
6154 iter->trace = tr->current_trace;
6156 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6161 /* trace pipe does not show start of buffer */
6162 cpumask_setall(iter->started);
6164 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6165 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6167 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6168 if (trace_clocks[tr->clock_id].in_ns)
6169 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6172 iter->array_buffer = &tr->array_buffer;
6173 iter->cpu_file = tracing_get_cpu(inode);
6174 mutex_init(&iter->mutex);
6175 filp->private_data = iter;
6177 if (iter->trace->pipe_open)
6178 iter->trace->pipe_open(iter);
6180 nonseekable_open(inode, filp);
6184 mutex_unlock(&trace_types_lock);
6189 __trace_array_put(tr);
6190 mutex_unlock(&trace_types_lock);
6194 static int tracing_release_pipe(struct inode *inode, struct file *file)
6196 struct trace_iterator *iter = file->private_data;
6197 struct trace_array *tr = inode->i_private;
6199 mutex_lock(&trace_types_lock);
6203 if (iter->trace->pipe_close)
6204 iter->trace->pipe_close(iter);
6206 mutex_unlock(&trace_types_lock);
6208 free_cpumask_var(iter->started);
6209 mutex_destroy(&iter->mutex);
6212 trace_array_put(tr);
6218 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6220 struct trace_array *tr = iter->tr;
6222 /* Iterators are static, they should be filled or empty */
6223 if (trace_buffer_iter(iter, iter->cpu_file))
6224 return EPOLLIN | EPOLLRDNORM;
6226 if (tr->trace_flags & TRACE_ITER_BLOCK)
6228 * Always select as readable when in blocking mode
6230 return EPOLLIN | EPOLLRDNORM;
6232 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6237 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6239 struct trace_iterator *iter = filp->private_data;
6241 return trace_poll(iter, filp, poll_table);
6244 /* Must be called with iter->mutex held. */
6245 static int tracing_wait_pipe(struct file *filp)
6247 struct trace_iterator *iter = filp->private_data;
6250 while (trace_empty(iter)) {
6252 if ((filp->f_flags & O_NONBLOCK)) {
6257 * We block until we read something and tracing is disabled.
6258 * We still block if tracing is disabled, but we have never
6259 * read anything. This allows a user to cat this file, and
6260 * then enable tracing. But after we have read something,
6261 * we give an EOF when tracing is again disabled.
6263 * iter->pos will be 0 if we haven't read anything.
6265 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6268 mutex_unlock(&iter->mutex);
6270 ret = wait_on_pipe(iter, 0);
6272 mutex_lock(&iter->mutex);
6285 tracing_read_pipe(struct file *filp, char __user *ubuf,
6286 size_t cnt, loff_t *ppos)
6288 struct trace_iterator *iter = filp->private_data;
6292 * Avoid more than one consumer on a single file descriptor
6293 * This is just a matter of traces coherency, the ring buffer itself
6296 mutex_lock(&iter->mutex);
6298 /* return any leftover data */
6299 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6303 trace_seq_init(&iter->seq);
6305 if (iter->trace->read) {
6306 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6312 sret = tracing_wait_pipe(filp);
6316 /* stop when tracing is finished */
6317 if (trace_empty(iter)) {
6322 if (cnt >= PAGE_SIZE)
6323 cnt = PAGE_SIZE - 1;
6325 /* reset all but tr, trace, and overruns */
6326 memset(&iter->seq, 0,
6327 sizeof(struct trace_iterator) -
6328 offsetof(struct trace_iterator, seq));
6329 cpumask_clear(iter->started);
6330 trace_seq_init(&iter->seq);
6333 trace_event_read_lock();
6334 trace_access_lock(iter->cpu_file);
6335 while (trace_find_next_entry_inc(iter) != NULL) {
6336 enum print_line_t ret;
6337 int save_len = iter->seq.seq.len;
6339 ret = print_trace_line(iter);
6340 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6341 /* don't print partial lines */
6342 iter->seq.seq.len = save_len;
6345 if (ret != TRACE_TYPE_NO_CONSUME)
6346 trace_consume(iter);
6348 if (trace_seq_used(&iter->seq) >= cnt)
6352 * Setting the full flag means we reached the trace_seq buffer
6353 * size and we should leave by partial output condition above.
6354 * One of the trace_seq_* functions is not used properly.
6356 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6359 trace_access_unlock(iter->cpu_file);
6360 trace_event_read_unlock();
6362 /* Now copy what we have to the user */
6363 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6364 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6365 trace_seq_init(&iter->seq);
6368 * If there was nothing to send to user, in spite of consuming trace
6369 * entries, go back to wait for more entries.
6375 mutex_unlock(&iter->mutex);
6380 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6383 __free_page(spd->pages[idx]);
6387 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6393 /* Seq buffer is page-sized, exactly what we need. */
6395 save_len = iter->seq.seq.len;
6396 ret = print_trace_line(iter);
6398 if (trace_seq_has_overflowed(&iter->seq)) {
6399 iter->seq.seq.len = save_len;
6404 * This should not be hit, because it should only
6405 * be set if the iter->seq overflowed. But check it
6406 * anyway to be safe.
6408 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6409 iter->seq.seq.len = save_len;
6413 count = trace_seq_used(&iter->seq) - save_len;
6416 iter->seq.seq.len = save_len;
6420 if (ret != TRACE_TYPE_NO_CONSUME)
6421 trace_consume(iter);
6423 if (!trace_find_next_entry_inc(iter)) {
6433 static ssize_t tracing_splice_read_pipe(struct file *filp,
6435 struct pipe_inode_info *pipe,
6439 struct page *pages_def[PIPE_DEF_BUFFERS];
6440 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6441 struct trace_iterator *iter = filp->private_data;
6442 struct splice_pipe_desc spd = {
6444 .partial = partial_def,
6445 .nr_pages = 0, /* This gets updated below. */
6446 .nr_pages_max = PIPE_DEF_BUFFERS,
6447 .ops = &default_pipe_buf_ops,
6448 .spd_release = tracing_spd_release_pipe,
6454 if (splice_grow_spd(pipe, &spd))
6457 mutex_lock(&iter->mutex);
6459 if (iter->trace->splice_read) {
6460 ret = iter->trace->splice_read(iter, filp,
6461 ppos, pipe, len, flags);
6466 ret = tracing_wait_pipe(filp);
6470 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6475 trace_event_read_lock();
6476 trace_access_lock(iter->cpu_file);
6478 /* Fill as many pages as possible. */
6479 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6480 spd.pages[i] = alloc_page(GFP_KERNEL);
6484 rem = tracing_fill_pipe_page(rem, iter);
6486 /* Copy the data into the page, so we can start over. */
6487 ret = trace_seq_to_buffer(&iter->seq,
6488 page_address(spd.pages[i]),
6489 trace_seq_used(&iter->seq));
6491 __free_page(spd.pages[i]);
6494 spd.partial[i].offset = 0;
6495 spd.partial[i].len = trace_seq_used(&iter->seq);
6497 trace_seq_init(&iter->seq);
6500 trace_access_unlock(iter->cpu_file);
6501 trace_event_read_unlock();
6502 mutex_unlock(&iter->mutex);
6507 ret = splice_to_pipe(pipe, &spd);
6511 splice_shrink_spd(&spd);
6515 mutex_unlock(&iter->mutex);
6520 tracing_entries_read(struct file *filp, char __user *ubuf,
6521 size_t cnt, loff_t *ppos)
6523 struct inode *inode = file_inode(filp);
6524 struct trace_array *tr = inode->i_private;
6525 int cpu = tracing_get_cpu(inode);
6530 mutex_lock(&trace_types_lock);
6532 if (cpu == RING_BUFFER_ALL_CPUS) {
6533 int cpu, buf_size_same;
6538 /* check if all cpu sizes are same */
6539 for_each_tracing_cpu(cpu) {
6540 /* fill in the size from first enabled cpu */
6542 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6543 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6549 if (buf_size_same) {
6550 if (!ring_buffer_expanded)
6551 r = sprintf(buf, "%lu (expanded: %lu)\n",
6553 trace_buf_size >> 10);
6555 r = sprintf(buf, "%lu\n", size >> 10);
6557 r = sprintf(buf, "X\n");
6559 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6561 mutex_unlock(&trace_types_lock);
6563 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6568 tracing_entries_write(struct file *filp, const char __user *ubuf,
6569 size_t cnt, loff_t *ppos)
6571 struct inode *inode = file_inode(filp);
6572 struct trace_array *tr = inode->i_private;
6576 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6580 /* must have at least 1 entry */
6584 /* value is in KB */
6586 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6596 tracing_total_entries_read(struct file *filp, char __user *ubuf,
6597 size_t cnt, loff_t *ppos)
6599 struct trace_array *tr = filp->private_data;
6602 unsigned long size = 0, expanded_size = 0;
6604 mutex_lock(&trace_types_lock);
6605 for_each_tracing_cpu(cpu) {
6606 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6607 if (!ring_buffer_expanded)
6608 expanded_size += trace_buf_size >> 10;
6610 if (ring_buffer_expanded)
6611 r = sprintf(buf, "%lu\n", size);
6613 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6614 mutex_unlock(&trace_types_lock);
6616 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6620 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
6621 size_t cnt, loff_t *ppos)
6624 * There is no need to read what the user has written, this function
6625 * is just to make sure that there is no error when "echo" is used
6634 tracing_free_buffer_release(struct inode *inode, struct file *filp)
6636 struct trace_array *tr = inode->i_private;
6638 /* disable tracing ? */
6639 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
6640 tracer_tracing_off(tr);
6641 /* resize the ring buffer to 0 */
6642 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
6644 trace_array_put(tr);
6650 tracing_mark_write(struct file *filp, const char __user *ubuf,
6651 size_t cnt, loff_t *fpos)
6653 struct trace_array *tr = filp->private_data;
6654 struct ring_buffer_event *event;
6655 enum event_trigger_type tt = ETT_NONE;
6656 struct trace_buffer *buffer;
6657 struct print_entry *entry;
6658 unsigned long irq_flags;
6663 /* Used in tracing_mark_raw_write() as well */
6664 #define FAULTED_STR "<faulted>"
6665 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
6667 if (tracing_disabled)
6670 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6673 if (cnt > TRACE_BUF_SIZE)
6674 cnt = TRACE_BUF_SIZE;
6676 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6678 local_save_flags(irq_flags);
6679 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
6681 /* If less than "<faulted>", then make sure we can still add that */
6682 if (cnt < FAULTED_SIZE)
6683 size += FAULTED_SIZE - cnt;
6685 buffer = tr->array_buffer.buffer;
6686 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
6687 irq_flags, preempt_count());
6688 if (unlikely(!event))
6689 /* Ring buffer disabled, return as if not open for write */
6692 entry = ring_buffer_event_data(event);
6693 entry->ip = _THIS_IP_;
6695 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
6697 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6703 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
6704 /* do not add \n before testing triggers, but add \0 */
6705 entry->buf[cnt] = '\0';
6706 tt = event_triggers_call(tr->trace_marker_file, entry, event);
6709 if (entry->buf[cnt - 1] != '\n') {
6710 entry->buf[cnt] = '\n';
6711 entry->buf[cnt + 1] = '\0';
6713 entry->buf[cnt] = '\0';
6715 if (static_branch_unlikely(&trace_marker_exports_enabled))
6716 ftrace_exports(event, TRACE_EXPORT_MARKER);
6717 __buffer_unlock_commit(buffer, event);
6720 event_triggers_post_call(tr->trace_marker_file, tt);
6728 /* Limit it for now to 3K (including tag) */
6729 #define RAW_DATA_MAX_SIZE (1024*3)
6732 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
6733 size_t cnt, loff_t *fpos)
6735 struct trace_array *tr = filp->private_data;
6736 struct ring_buffer_event *event;
6737 struct trace_buffer *buffer;
6738 struct raw_data_entry *entry;
6739 unsigned long irq_flags;
6744 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
6746 if (tracing_disabled)
6749 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6752 /* The marker must at least have a tag id */
6753 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
6756 if (cnt > TRACE_BUF_SIZE)
6757 cnt = TRACE_BUF_SIZE;
6759 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6761 local_save_flags(irq_flags);
6762 size = sizeof(*entry) + cnt;
6763 if (cnt < FAULT_SIZE_ID)
6764 size += FAULT_SIZE_ID - cnt;
6766 buffer = tr->array_buffer.buffer;
6767 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
6768 irq_flags, preempt_count());
6770 /* Ring buffer disabled, return as if not open for write */
6773 entry = ring_buffer_event_data(event);
6775 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
6778 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6783 __buffer_unlock_commit(buffer, event);
6791 static int tracing_clock_show(struct seq_file *m, void *v)
6793 struct trace_array *tr = m->private;
6796 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
6798 "%s%s%s%s", i ? " " : "",
6799 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
6800 i == tr->clock_id ? "]" : "");
6806 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
6810 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
6811 if (strcmp(trace_clocks[i].name, clockstr) == 0)
6814 if (i == ARRAY_SIZE(trace_clocks))
6817 mutex_lock(&trace_types_lock);
6821 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
6824 * New clock may not be consistent with the previous clock.
6825 * Reset the buffer so that it doesn't have incomparable timestamps.
6827 tracing_reset_online_cpus(&tr->array_buffer);
6829 #ifdef CONFIG_TRACER_MAX_TRACE
6830 if (tr->max_buffer.buffer)
6831 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
6832 tracing_reset_online_cpus(&tr->max_buffer);
6835 mutex_unlock(&trace_types_lock);
6840 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
6841 size_t cnt, loff_t *fpos)
6843 struct seq_file *m = filp->private_data;
6844 struct trace_array *tr = m->private;
6846 const char *clockstr;
6849 if (cnt >= sizeof(buf))
6852 if (copy_from_user(buf, ubuf, cnt))
6857 clockstr = strstrip(buf);
6859 ret = tracing_set_clock(tr, clockstr);
6868 static int tracing_clock_open(struct inode *inode, struct file *file)
6870 struct trace_array *tr = inode->i_private;
6873 ret = tracing_check_open_get_tr(tr);
6877 ret = single_open(file, tracing_clock_show, inode->i_private);
6879 trace_array_put(tr);
6884 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
6886 struct trace_array *tr = m->private;
6888 mutex_lock(&trace_types_lock);
6890 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
6891 seq_puts(m, "delta [absolute]\n");
6893 seq_puts(m, "[delta] absolute\n");
6895 mutex_unlock(&trace_types_lock);
6900 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
6902 struct trace_array *tr = inode->i_private;
6905 ret = tracing_check_open_get_tr(tr);
6909 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
6911 trace_array_put(tr);
6916 int tracing_set_time_stamp_abs(struct trace_array *tr, bool abs)
6920 mutex_lock(&trace_types_lock);
6922 if (abs && tr->time_stamp_abs_ref++)
6926 if (WARN_ON_ONCE(!tr->time_stamp_abs_ref)) {
6931 if (--tr->time_stamp_abs_ref)
6935 ring_buffer_set_time_stamp_abs(tr->array_buffer.buffer, abs);
6937 #ifdef CONFIG_TRACER_MAX_TRACE
6938 if (tr->max_buffer.buffer)
6939 ring_buffer_set_time_stamp_abs(tr->max_buffer.buffer, abs);
6942 mutex_unlock(&trace_types_lock);
6947 struct ftrace_buffer_info {
6948 struct trace_iterator iter;
6950 unsigned int spare_cpu;
6954 #ifdef CONFIG_TRACER_SNAPSHOT
6955 static int tracing_snapshot_open(struct inode *inode, struct file *file)
6957 struct trace_array *tr = inode->i_private;
6958 struct trace_iterator *iter;
6962 ret = tracing_check_open_get_tr(tr);
6966 if (file->f_mode & FMODE_READ) {
6967 iter = __tracing_open(inode, file, true);
6969 ret = PTR_ERR(iter);
6971 /* Writes still need the seq_file to hold the private data */
6973 m = kzalloc(sizeof(*m), GFP_KERNEL);
6976 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6984 iter->array_buffer = &tr->max_buffer;
6985 iter->cpu_file = tracing_get_cpu(inode);
6987 file->private_data = m;
6991 trace_array_put(tr);
6997 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7000 struct seq_file *m = filp->private_data;
7001 struct trace_iterator *iter = m->private;
7002 struct trace_array *tr = iter->tr;
7006 ret = tracing_update_buffers();
7010 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7014 mutex_lock(&trace_types_lock);
7016 if (tr->current_trace->use_max_tr) {
7021 arch_spin_lock(&tr->max_lock);
7022 if (tr->cond_snapshot)
7024 arch_spin_unlock(&tr->max_lock);
7030 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7034 if (tr->allocated_snapshot)
7038 /* Only allow per-cpu swap if the ring buffer supports it */
7039 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7040 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7045 if (tr->allocated_snapshot)
7046 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7047 &tr->array_buffer, iter->cpu_file);
7049 ret = tracing_alloc_snapshot_instance(tr);
7052 local_irq_disable();
7053 /* Now, we're going to swap */
7054 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7055 update_max_tr(tr, current, smp_processor_id(), NULL);
7057 update_max_tr_single(tr, current, iter->cpu_file);
7061 if (tr->allocated_snapshot) {
7062 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7063 tracing_reset_online_cpus(&tr->max_buffer);
7065 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7075 mutex_unlock(&trace_types_lock);
7079 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7081 struct seq_file *m = file->private_data;
7084 ret = tracing_release(inode, file);
7086 if (file->f_mode & FMODE_READ)
7089 /* If write only, the seq_file is just a stub */
7097 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7098 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7099 size_t count, loff_t *ppos);
7100 static int tracing_buffers_release(struct inode *inode, struct file *file);
7101 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7102 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7104 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7106 struct ftrace_buffer_info *info;
7109 /* The following checks for tracefs lockdown */
7110 ret = tracing_buffers_open(inode, filp);
7114 info = filp->private_data;
7116 if (info->iter.trace->use_max_tr) {
7117 tracing_buffers_release(inode, filp);
7121 info->iter.snapshot = true;
7122 info->iter.array_buffer = &info->iter.tr->max_buffer;
7127 #endif /* CONFIG_TRACER_SNAPSHOT */
7130 static const struct file_operations tracing_thresh_fops = {
7131 .open = tracing_open_generic,
7132 .read = tracing_thresh_read,
7133 .write = tracing_thresh_write,
7134 .llseek = generic_file_llseek,
7137 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
7138 static const struct file_operations tracing_max_lat_fops = {
7139 .open = tracing_open_generic,
7140 .read = tracing_max_lat_read,
7141 .write = tracing_max_lat_write,
7142 .llseek = generic_file_llseek,
7146 static const struct file_operations set_tracer_fops = {
7147 .open = tracing_open_generic,
7148 .read = tracing_set_trace_read,
7149 .write = tracing_set_trace_write,
7150 .llseek = generic_file_llseek,
7153 static const struct file_operations tracing_pipe_fops = {
7154 .open = tracing_open_pipe,
7155 .poll = tracing_poll_pipe,
7156 .read = tracing_read_pipe,
7157 .splice_read = tracing_splice_read_pipe,
7158 .release = tracing_release_pipe,
7159 .llseek = no_llseek,
7162 static const struct file_operations tracing_entries_fops = {
7163 .open = tracing_open_generic_tr,
7164 .read = tracing_entries_read,
7165 .write = tracing_entries_write,
7166 .llseek = generic_file_llseek,
7167 .release = tracing_release_generic_tr,
7170 static const struct file_operations tracing_total_entries_fops = {
7171 .open = tracing_open_generic_tr,
7172 .read = tracing_total_entries_read,
7173 .llseek = generic_file_llseek,
7174 .release = tracing_release_generic_tr,
7177 static const struct file_operations tracing_free_buffer_fops = {
7178 .open = tracing_open_generic_tr,
7179 .write = tracing_free_buffer_write,
7180 .release = tracing_free_buffer_release,
7183 static const struct file_operations tracing_mark_fops = {
7184 .open = tracing_open_generic_tr,
7185 .write = tracing_mark_write,
7186 .llseek = generic_file_llseek,
7187 .release = tracing_release_generic_tr,
7190 static const struct file_operations tracing_mark_raw_fops = {
7191 .open = tracing_open_generic_tr,
7192 .write = tracing_mark_raw_write,
7193 .llseek = generic_file_llseek,
7194 .release = tracing_release_generic_tr,
7197 static const struct file_operations trace_clock_fops = {
7198 .open = tracing_clock_open,
7200 .llseek = seq_lseek,
7201 .release = tracing_single_release_tr,
7202 .write = tracing_clock_write,
7205 static const struct file_operations trace_time_stamp_mode_fops = {
7206 .open = tracing_time_stamp_mode_open,
7208 .llseek = seq_lseek,
7209 .release = tracing_single_release_tr,
7212 #ifdef CONFIG_TRACER_SNAPSHOT
7213 static const struct file_operations snapshot_fops = {
7214 .open = tracing_snapshot_open,
7216 .write = tracing_snapshot_write,
7217 .llseek = tracing_lseek,
7218 .release = tracing_snapshot_release,
7221 static const struct file_operations snapshot_raw_fops = {
7222 .open = snapshot_raw_open,
7223 .read = tracing_buffers_read,
7224 .release = tracing_buffers_release,
7225 .splice_read = tracing_buffers_splice_read,
7226 .llseek = no_llseek,
7229 #endif /* CONFIG_TRACER_SNAPSHOT */
7231 #define TRACING_LOG_ERRS_MAX 8
7232 #define TRACING_LOG_LOC_MAX 128
7234 #define CMD_PREFIX " Command: "
7237 const char **errs; /* ptr to loc-specific array of err strings */
7238 u8 type; /* index into errs -> specific err string */
7239 u8 pos; /* MAX_FILTER_STR_VAL = 256 */
7243 struct tracing_log_err {
7244 struct list_head list;
7245 struct err_info info;
7246 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7247 char cmd[MAX_FILTER_STR_VAL]; /* what caused err */
7250 static DEFINE_MUTEX(tracing_err_log_lock);
7252 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr)
7254 struct tracing_log_err *err;
7256 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7257 err = kzalloc(sizeof(*err), GFP_KERNEL);
7259 err = ERR_PTR(-ENOMEM);
7260 tr->n_err_log_entries++;
7265 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7266 list_del(&err->list);
7272 * err_pos - find the position of a string within a command for error careting
7273 * @cmd: The tracing command that caused the error
7274 * @str: The string to position the caret at within @cmd
7276 * Finds the position of the first occurence of @str within @cmd. The
7277 * return value can be passed to tracing_log_err() for caret placement
7280 * Returns the index within @cmd of the first occurence of @str or 0
7281 * if @str was not found.
7283 unsigned int err_pos(char *cmd, const char *str)
7287 if (WARN_ON(!strlen(cmd)))
7290 found = strstr(cmd, str);
7298 * tracing_log_err - write an error to the tracing error log
7299 * @tr: The associated trace array for the error (NULL for top level array)
7300 * @loc: A string describing where the error occurred
7301 * @cmd: The tracing command that caused the error
7302 * @errs: The array of loc-specific static error strings
7303 * @type: The index into errs[], which produces the specific static err string
7304 * @pos: The position the caret should be placed in the cmd
7306 * Writes an error into tracing/error_log of the form:
7308 * <loc>: error: <text>
7312 * tracing/error_log is a small log file containing the last
7313 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7314 * unless there has been a tracing error, and the error log can be
7315 * cleared and have its memory freed by writing the empty string in
7316 * truncation mode to it i.e. echo > tracing/error_log.
7318 * NOTE: the @errs array along with the @type param are used to
7319 * produce a static error string - this string is not copied and saved
7320 * when the error is logged - only a pointer to it is saved. See
7321 * existing callers for examples of how static strings are typically
7322 * defined for use with tracing_log_err().
7324 void tracing_log_err(struct trace_array *tr,
7325 const char *loc, const char *cmd,
7326 const char **errs, u8 type, u8 pos)
7328 struct tracing_log_err *err;
7333 mutex_lock(&tracing_err_log_lock);
7334 err = get_tracing_log_err(tr);
7335 if (PTR_ERR(err) == -ENOMEM) {
7336 mutex_unlock(&tracing_err_log_lock);
7340 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7341 snprintf(err->cmd, MAX_FILTER_STR_VAL,"\n" CMD_PREFIX "%s\n", cmd);
7343 err->info.errs = errs;
7344 err->info.type = type;
7345 err->info.pos = pos;
7346 err->info.ts = local_clock();
7348 list_add_tail(&err->list, &tr->err_log);
7349 mutex_unlock(&tracing_err_log_lock);
7352 static void clear_tracing_err_log(struct trace_array *tr)
7354 struct tracing_log_err *err, *next;
7356 mutex_lock(&tracing_err_log_lock);
7357 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7358 list_del(&err->list);
7362 tr->n_err_log_entries = 0;
7363 mutex_unlock(&tracing_err_log_lock);
7366 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7368 struct trace_array *tr = m->private;
7370 mutex_lock(&tracing_err_log_lock);
7372 return seq_list_start(&tr->err_log, *pos);
7375 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7377 struct trace_array *tr = m->private;
7379 return seq_list_next(v, &tr->err_log, pos);
7382 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7384 mutex_unlock(&tracing_err_log_lock);
7387 static void tracing_err_log_show_pos(struct seq_file *m, u8 pos)
7391 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7393 for (i = 0; i < pos; i++)
7398 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7400 struct tracing_log_err *err = v;
7403 const char *err_text = err->info.errs[err->info.type];
7404 u64 sec = err->info.ts;
7407 nsec = do_div(sec, NSEC_PER_SEC);
7408 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7409 err->loc, err_text);
7410 seq_printf(m, "%s", err->cmd);
7411 tracing_err_log_show_pos(m, err->info.pos);
7417 static const struct seq_operations tracing_err_log_seq_ops = {
7418 .start = tracing_err_log_seq_start,
7419 .next = tracing_err_log_seq_next,
7420 .stop = tracing_err_log_seq_stop,
7421 .show = tracing_err_log_seq_show
7424 static int tracing_err_log_open(struct inode *inode, struct file *file)
7426 struct trace_array *tr = inode->i_private;
7429 ret = tracing_check_open_get_tr(tr);
7433 /* If this file was opened for write, then erase contents */
7434 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7435 clear_tracing_err_log(tr);
7437 if (file->f_mode & FMODE_READ) {
7438 ret = seq_open(file, &tracing_err_log_seq_ops);
7440 struct seq_file *m = file->private_data;
7443 trace_array_put(tr);
7449 static ssize_t tracing_err_log_write(struct file *file,
7450 const char __user *buffer,
7451 size_t count, loff_t *ppos)
7456 static int tracing_err_log_release(struct inode *inode, struct file *file)
7458 struct trace_array *tr = inode->i_private;
7460 trace_array_put(tr);
7462 if (file->f_mode & FMODE_READ)
7463 seq_release(inode, file);
7468 static const struct file_operations tracing_err_log_fops = {
7469 .open = tracing_err_log_open,
7470 .write = tracing_err_log_write,
7472 .llseek = seq_lseek,
7473 .release = tracing_err_log_release,
7476 static int tracing_buffers_open(struct inode *inode, struct file *filp)
7478 struct trace_array *tr = inode->i_private;
7479 struct ftrace_buffer_info *info;
7482 ret = tracing_check_open_get_tr(tr);
7486 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7488 trace_array_put(tr);
7492 mutex_lock(&trace_types_lock);
7495 info->iter.cpu_file = tracing_get_cpu(inode);
7496 info->iter.trace = tr->current_trace;
7497 info->iter.array_buffer = &tr->array_buffer;
7499 /* Force reading ring buffer for first read */
7500 info->read = (unsigned int)-1;
7502 filp->private_data = info;
7506 mutex_unlock(&trace_types_lock);
7508 ret = nonseekable_open(inode, filp);
7510 trace_array_put(tr);
7516 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7518 struct ftrace_buffer_info *info = filp->private_data;
7519 struct trace_iterator *iter = &info->iter;
7521 return trace_poll(iter, filp, poll_table);
7525 tracing_buffers_read(struct file *filp, char __user *ubuf,
7526 size_t count, loff_t *ppos)
7528 struct ftrace_buffer_info *info = filp->private_data;
7529 struct trace_iterator *iter = &info->iter;
7536 #ifdef CONFIG_TRACER_MAX_TRACE
7537 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7542 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
7544 if (IS_ERR(info->spare)) {
7545 ret = PTR_ERR(info->spare);
7548 info->spare_cpu = iter->cpu_file;
7554 /* Do we have previous read data to read? */
7555 if (info->read < PAGE_SIZE)
7559 trace_access_lock(iter->cpu_file);
7560 ret = ring_buffer_read_page(iter->array_buffer->buffer,
7564 trace_access_unlock(iter->cpu_file);
7567 if (trace_empty(iter)) {
7568 if ((filp->f_flags & O_NONBLOCK))
7571 ret = wait_on_pipe(iter, 0);
7582 size = PAGE_SIZE - info->read;
7586 ret = copy_to_user(ubuf, info->spare + info->read, size);
7598 static int tracing_buffers_release(struct inode *inode, struct file *file)
7600 struct ftrace_buffer_info *info = file->private_data;
7601 struct trace_iterator *iter = &info->iter;
7603 mutex_lock(&trace_types_lock);
7605 iter->tr->trace_ref--;
7607 __trace_array_put(iter->tr);
7610 ring_buffer_free_read_page(iter->array_buffer->buffer,
7611 info->spare_cpu, info->spare);
7614 mutex_unlock(&trace_types_lock);
7620 struct trace_buffer *buffer;
7623 refcount_t refcount;
7626 static void buffer_ref_release(struct buffer_ref *ref)
7628 if (!refcount_dec_and_test(&ref->refcount))
7630 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
7634 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
7635 struct pipe_buffer *buf)
7637 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
7639 buffer_ref_release(ref);
7643 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
7644 struct pipe_buffer *buf)
7646 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
7648 if (refcount_read(&ref->refcount) > INT_MAX/2)
7651 refcount_inc(&ref->refcount);
7655 /* Pipe buffer operations for a buffer. */
7656 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
7657 .release = buffer_pipe_buf_release,
7658 .get = buffer_pipe_buf_get,
7662 * Callback from splice_to_pipe(), if we need to release some pages
7663 * at the end of the spd in case we error'ed out in filling the pipe.
7665 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
7667 struct buffer_ref *ref =
7668 (struct buffer_ref *)spd->partial[i].private;
7670 buffer_ref_release(ref);
7671 spd->partial[i].private = 0;
7675 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7676 struct pipe_inode_info *pipe, size_t len,
7679 struct ftrace_buffer_info *info = file->private_data;
7680 struct trace_iterator *iter = &info->iter;
7681 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7682 struct page *pages_def[PIPE_DEF_BUFFERS];
7683 struct splice_pipe_desc spd = {
7685 .partial = partial_def,
7686 .nr_pages_max = PIPE_DEF_BUFFERS,
7687 .ops = &buffer_pipe_buf_ops,
7688 .spd_release = buffer_spd_release,
7690 struct buffer_ref *ref;
7694 #ifdef CONFIG_TRACER_MAX_TRACE
7695 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7699 if (*ppos & (PAGE_SIZE - 1))
7702 if (len & (PAGE_SIZE - 1)) {
7703 if (len < PAGE_SIZE)
7708 if (splice_grow_spd(pipe, &spd))
7712 trace_access_lock(iter->cpu_file);
7713 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
7715 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
7719 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
7725 refcount_set(&ref->refcount, 1);
7726 ref->buffer = iter->array_buffer->buffer;
7727 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
7728 if (IS_ERR(ref->page)) {
7729 ret = PTR_ERR(ref->page);
7734 ref->cpu = iter->cpu_file;
7736 r = ring_buffer_read_page(ref->buffer, &ref->page,
7737 len, iter->cpu_file, 1);
7739 ring_buffer_free_read_page(ref->buffer, ref->cpu,
7745 page = virt_to_page(ref->page);
7747 spd.pages[i] = page;
7748 spd.partial[i].len = PAGE_SIZE;
7749 spd.partial[i].offset = 0;
7750 spd.partial[i].private = (unsigned long)ref;
7754 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
7757 trace_access_unlock(iter->cpu_file);
7760 /* did we read anything? */
7761 if (!spd.nr_pages) {
7766 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
7769 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
7776 ret = splice_to_pipe(pipe, &spd);
7778 splice_shrink_spd(&spd);
7783 static const struct file_operations tracing_buffers_fops = {
7784 .open = tracing_buffers_open,
7785 .read = tracing_buffers_read,
7786 .poll = tracing_buffers_poll,
7787 .release = tracing_buffers_release,
7788 .splice_read = tracing_buffers_splice_read,
7789 .llseek = no_llseek,
7793 tracing_stats_read(struct file *filp, char __user *ubuf,
7794 size_t count, loff_t *ppos)
7796 struct inode *inode = file_inode(filp);
7797 struct trace_array *tr = inode->i_private;
7798 struct array_buffer *trace_buf = &tr->array_buffer;
7799 int cpu = tracing_get_cpu(inode);
7800 struct trace_seq *s;
7802 unsigned long long t;
7803 unsigned long usec_rem;
7805 s = kmalloc(sizeof(*s), GFP_KERNEL);
7811 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
7812 trace_seq_printf(s, "entries: %ld\n", cnt);
7814 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
7815 trace_seq_printf(s, "overrun: %ld\n", cnt);
7817 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
7818 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
7820 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
7821 trace_seq_printf(s, "bytes: %ld\n", cnt);
7823 if (trace_clocks[tr->clock_id].in_ns) {
7824 /* local or global for trace_clock */
7825 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
7826 usec_rem = do_div(t, USEC_PER_SEC);
7827 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
7830 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
7831 usec_rem = do_div(t, USEC_PER_SEC);
7832 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
7834 /* counter or tsc mode for trace_clock */
7835 trace_seq_printf(s, "oldest event ts: %llu\n",
7836 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
7838 trace_seq_printf(s, "now ts: %llu\n",
7839 ring_buffer_time_stamp(trace_buf->buffer, cpu));
7842 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
7843 trace_seq_printf(s, "dropped events: %ld\n", cnt);
7845 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
7846 trace_seq_printf(s, "read events: %ld\n", cnt);
7848 count = simple_read_from_buffer(ubuf, count, ppos,
7849 s->buffer, trace_seq_used(s));
7856 static const struct file_operations tracing_stats_fops = {
7857 .open = tracing_open_generic_tr,
7858 .read = tracing_stats_read,
7859 .llseek = generic_file_llseek,
7860 .release = tracing_release_generic_tr,
7863 #ifdef CONFIG_DYNAMIC_FTRACE
7866 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
7867 size_t cnt, loff_t *ppos)
7873 /* 256 should be plenty to hold the amount needed */
7874 buf = kmalloc(256, GFP_KERNEL);
7878 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
7879 ftrace_update_tot_cnt,
7880 ftrace_number_of_pages,
7881 ftrace_number_of_groups);
7883 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7888 static const struct file_operations tracing_dyn_info_fops = {
7889 .open = tracing_open_generic,
7890 .read = tracing_read_dyn_info,
7891 .llseek = generic_file_llseek,
7893 #endif /* CONFIG_DYNAMIC_FTRACE */
7895 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
7897 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
7898 struct trace_array *tr, struct ftrace_probe_ops *ops,
7901 tracing_snapshot_instance(tr);
7905 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
7906 struct trace_array *tr, struct ftrace_probe_ops *ops,
7909 struct ftrace_func_mapper *mapper = data;
7913 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7923 tracing_snapshot_instance(tr);
7927 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
7928 struct ftrace_probe_ops *ops, void *data)
7930 struct ftrace_func_mapper *mapper = data;
7933 seq_printf(m, "%ps:", (void *)ip);
7935 seq_puts(m, "snapshot");
7938 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7941 seq_printf(m, ":count=%ld\n", *count);
7943 seq_puts(m, ":unlimited\n");
7949 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
7950 unsigned long ip, void *init_data, void **data)
7952 struct ftrace_func_mapper *mapper = *data;
7955 mapper = allocate_ftrace_func_mapper();
7961 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
7965 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
7966 unsigned long ip, void *data)
7968 struct ftrace_func_mapper *mapper = data;
7973 free_ftrace_func_mapper(mapper, NULL);
7977 ftrace_func_mapper_remove_ip(mapper, ip);
7980 static struct ftrace_probe_ops snapshot_probe_ops = {
7981 .func = ftrace_snapshot,
7982 .print = ftrace_snapshot_print,
7985 static struct ftrace_probe_ops snapshot_count_probe_ops = {
7986 .func = ftrace_count_snapshot,
7987 .print = ftrace_snapshot_print,
7988 .init = ftrace_snapshot_init,
7989 .free = ftrace_snapshot_free,
7993 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
7994 char *glob, char *cmd, char *param, int enable)
7996 struct ftrace_probe_ops *ops;
7997 void *count = (void *)-1;
8004 /* hash funcs only work with set_ftrace_filter */
8008 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8011 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8016 number = strsep(¶m, ":");
8018 if (!strlen(number))
8022 * We use the callback data field (which is a pointer)
8025 ret = kstrtoul(number, 0, (unsigned long *)&count);
8030 ret = tracing_alloc_snapshot_instance(tr);
8034 ret = register_ftrace_function_probe(glob, tr, ops, count);
8037 return ret < 0 ? ret : 0;
8040 static struct ftrace_func_command ftrace_snapshot_cmd = {
8042 .func = ftrace_trace_snapshot_callback,
8045 static __init int register_snapshot_cmd(void)
8047 return register_ftrace_command(&ftrace_snapshot_cmd);
8050 static inline __init int register_snapshot_cmd(void) { return 0; }
8051 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8053 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8055 if (WARN_ON(!tr->dir))
8056 return ERR_PTR(-ENODEV);
8058 /* Top directory uses NULL as the parent */
8059 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8062 /* All sub buffers have a descriptor */
8066 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8068 struct dentry *d_tracer;
8071 return tr->percpu_dir;
8073 d_tracer = tracing_get_dentry(tr);
8074 if (IS_ERR(d_tracer))
8077 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8079 MEM_FAIL(!tr->percpu_dir,
8080 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8082 return tr->percpu_dir;
8085 static struct dentry *
8086 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8087 void *data, long cpu, const struct file_operations *fops)
8089 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8091 if (ret) /* See tracing_get_cpu() */
8092 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8097 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8099 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8100 struct dentry *d_cpu;
8101 char cpu_dir[30]; /* 30 characters should be more than enough */
8106 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8107 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8109 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8113 /* per cpu trace_pipe */
8114 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
8115 tr, cpu, &tracing_pipe_fops);
8118 trace_create_cpu_file("trace", 0644, d_cpu,
8119 tr, cpu, &tracing_fops);
8121 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
8122 tr, cpu, &tracing_buffers_fops);
8124 trace_create_cpu_file("stats", 0444, d_cpu,
8125 tr, cpu, &tracing_stats_fops);
8127 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
8128 tr, cpu, &tracing_entries_fops);
8130 #ifdef CONFIG_TRACER_SNAPSHOT
8131 trace_create_cpu_file("snapshot", 0644, d_cpu,
8132 tr, cpu, &snapshot_fops);
8134 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
8135 tr, cpu, &snapshot_raw_fops);
8139 #ifdef CONFIG_FTRACE_SELFTEST
8140 /* Let selftest have access to static functions in this file */
8141 #include "trace_selftest.c"
8145 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8148 struct trace_option_dentry *topt = filp->private_data;
8151 if (topt->flags->val & topt->opt->bit)
8156 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8160 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8163 struct trace_option_dentry *topt = filp->private_data;
8167 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8171 if (val != 0 && val != 1)
8174 if (!!(topt->flags->val & topt->opt->bit) != val) {
8175 mutex_lock(&trace_types_lock);
8176 ret = __set_tracer_option(topt->tr, topt->flags,
8178 mutex_unlock(&trace_types_lock);
8189 static const struct file_operations trace_options_fops = {
8190 .open = tracing_open_generic,
8191 .read = trace_options_read,
8192 .write = trace_options_write,
8193 .llseek = generic_file_llseek,
8197 * In order to pass in both the trace_array descriptor as well as the index
8198 * to the flag that the trace option file represents, the trace_array
8199 * has a character array of trace_flags_index[], which holds the index
8200 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8201 * The address of this character array is passed to the flag option file
8202 * read/write callbacks.
8204 * In order to extract both the index and the trace_array descriptor,
8205 * get_tr_index() uses the following algorithm.
8209 * As the pointer itself contains the address of the index (remember
8212 * Then to get the trace_array descriptor, by subtracting that index
8213 * from the ptr, we get to the start of the index itself.
8215 * ptr - idx == &index[0]
8217 * Then a simple container_of() from that pointer gets us to the
8218 * trace_array descriptor.
8220 static void get_tr_index(void *data, struct trace_array **ptr,
8221 unsigned int *pindex)
8223 *pindex = *(unsigned char *)data;
8225 *ptr = container_of(data - *pindex, struct trace_array,
8230 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8233 void *tr_index = filp->private_data;
8234 struct trace_array *tr;
8238 get_tr_index(tr_index, &tr, &index);
8240 if (tr->trace_flags & (1 << index))
8245 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8249 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8252 void *tr_index = filp->private_data;
8253 struct trace_array *tr;
8258 get_tr_index(tr_index, &tr, &index);
8260 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8264 if (val != 0 && val != 1)
8267 mutex_lock(&event_mutex);
8268 mutex_lock(&trace_types_lock);
8269 ret = set_tracer_flag(tr, 1 << index, val);
8270 mutex_unlock(&trace_types_lock);
8271 mutex_unlock(&event_mutex);
8281 static const struct file_operations trace_options_core_fops = {
8282 .open = tracing_open_generic,
8283 .read = trace_options_core_read,
8284 .write = trace_options_core_write,
8285 .llseek = generic_file_llseek,
8288 struct dentry *trace_create_file(const char *name,
8290 struct dentry *parent,
8292 const struct file_operations *fops)
8296 ret = tracefs_create_file(name, mode, parent, data, fops);
8298 pr_warn("Could not create tracefs '%s' entry\n", name);
8304 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8306 struct dentry *d_tracer;
8311 d_tracer = tracing_get_dentry(tr);
8312 if (IS_ERR(d_tracer))
8315 tr->options = tracefs_create_dir("options", d_tracer);
8317 pr_warn("Could not create tracefs directory 'options'\n");
8325 create_trace_option_file(struct trace_array *tr,
8326 struct trace_option_dentry *topt,
8327 struct tracer_flags *flags,
8328 struct tracer_opt *opt)
8330 struct dentry *t_options;
8332 t_options = trace_options_init_dentry(tr);
8336 topt->flags = flags;
8340 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
8341 &trace_options_fops);
8346 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8348 struct trace_option_dentry *topts;
8349 struct trace_options *tr_topts;
8350 struct tracer_flags *flags;
8351 struct tracer_opt *opts;
8358 flags = tracer->flags;
8360 if (!flags || !flags->opts)
8364 * If this is an instance, only create flags for tracers
8365 * the instance may have.
8367 if (!trace_ok_for_array(tracer, tr))
8370 for (i = 0; i < tr->nr_topts; i++) {
8371 /* Make sure there's no duplicate flags. */
8372 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8378 for (cnt = 0; opts[cnt].name; cnt++)
8381 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8385 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8392 tr->topts = tr_topts;
8393 tr->topts[tr->nr_topts].tracer = tracer;
8394 tr->topts[tr->nr_topts].topts = topts;
8397 for (cnt = 0; opts[cnt].name; cnt++) {
8398 create_trace_option_file(tr, &topts[cnt], flags,
8400 MEM_FAIL(topts[cnt].entry == NULL,
8401 "Failed to create trace option: %s",
8406 static struct dentry *
8407 create_trace_option_core_file(struct trace_array *tr,
8408 const char *option, long index)
8410 struct dentry *t_options;
8412 t_options = trace_options_init_dentry(tr);
8416 return trace_create_file(option, 0644, t_options,
8417 (void *)&tr->trace_flags_index[index],
8418 &trace_options_core_fops);
8421 static void create_trace_options_dir(struct trace_array *tr)
8423 struct dentry *t_options;
8424 bool top_level = tr == &global_trace;
8427 t_options = trace_options_init_dentry(tr);
8431 for (i = 0; trace_options[i]; i++) {
8433 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8434 create_trace_option_core_file(tr, trace_options[i], i);
8439 rb_simple_read(struct file *filp, char __user *ubuf,
8440 size_t cnt, loff_t *ppos)
8442 struct trace_array *tr = filp->private_data;
8446 r = tracer_tracing_is_on(tr);
8447 r = sprintf(buf, "%d\n", r);
8449 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8453 rb_simple_write(struct file *filp, const char __user *ubuf,
8454 size_t cnt, loff_t *ppos)
8456 struct trace_array *tr = filp->private_data;
8457 struct trace_buffer *buffer = tr->array_buffer.buffer;
8461 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8466 mutex_lock(&trace_types_lock);
8467 if (!!val == tracer_tracing_is_on(tr)) {
8468 val = 0; /* do nothing */
8470 tracer_tracing_on(tr);
8471 if (tr->current_trace->start)
8472 tr->current_trace->start(tr);
8474 tracer_tracing_off(tr);
8475 if (tr->current_trace->stop)
8476 tr->current_trace->stop(tr);
8478 mutex_unlock(&trace_types_lock);
8486 static const struct file_operations rb_simple_fops = {
8487 .open = tracing_open_generic_tr,
8488 .read = rb_simple_read,
8489 .write = rb_simple_write,
8490 .release = tracing_release_generic_tr,
8491 .llseek = default_llseek,
8495 buffer_percent_read(struct file *filp, char __user *ubuf,
8496 size_t cnt, loff_t *ppos)
8498 struct trace_array *tr = filp->private_data;
8502 r = tr->buffer_percent;
8503 r = sprintf(buf, "%d\n", r);
8505 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8509 buffer_percent_write(struct file *filp, const char __user *ubuf,
8510 size_t cnt, loff_t *ppos)
8512 struct trace_array *tr = filp->private_data;
8516 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8526 tr->buffer_percent = val;
8533 static const struct file_operations buffer_percent_fops = {
8534 .open = tracing_open_generic_tr,
8535 .read = buffer_percent_read,
8536 .write = buffer_percent_write,
8537 .release = tracing_release_generic_tr,
8538 .llseek = default_llseek,
8541 static struct dentry *trace_instance_dir;
8544 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
8547 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
8549 enum ring_buffer_flags rb_flags;
8551 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
8555 buf->buffer = ring_buffer_alloc(size, rb_flags);
8559 buf->data = alloc_percpu(struct trace_array_cpu);
8561 ring_buffer_free(buf->buffer);
8566 /* Allocate the first page for all buffers */
8567 set_buffer_entries(&tr->array_buffer,
8568 ring_buffer_size(tr->array_buffer.buffer, 0));
8573 static int allocate_trace_buffers(struct trace_array *tr, int size)
8577 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
8581 #ifdef CONFIG_TRACER_MAX_TRACE
8582 ret = allocate_trace_buffer(tr, &tr->max_buffer,
8583 allocate_snapshot ? size : 1);
8584 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
8585 ring_buffer_free(tr->array_buffer.buffer);
8586 tr->array_buffer.buffer = NULL;
8587 free_percpu(tr->array_buffer.data);
8588 tr->array_buffer.data = NULL;
8591 tr->allocated_snapshot = allocate_snapshot;
8594 * Only the top level trace array gets its snapshot allocated
8595 * from the kernel command line.
8597 allocate_snapshot = false;
8603 static void free_trace_buffer(struct array_buffer *buf)
8606 ring_buffer_free(buf->buffer);
8608 free_percpu(buf->data);
8613 static void free_trace_buffers(struct trace_array *tr)
8618 free_trace_buffer(&tr->array_buffer);
8620 #ifdef CONFIG_TRACER_MAX_TRACE
8621 free_trace_buffer(&tr->max_buffer);
8625 static void init_trace_flags_index(struct trace_array *tr)
8629 /* Used by the trace options files */
8630 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
8631 tr->trace_flags_index[i] = i;
8634 static void __update_tracer_options(struct trace_array *tr)
8638 for (t = trace_types; t; t = t->next)
8639 add_tracer_options(tr, t);
8642 static void update_tracer_options(struct trace_array *tr)
8644 mutex_lock(&trace_types_lock);
8645 __update_tracer_options(tr);
8646 mutex_unlock(&trace_types_lock);
8649 /* Must have trace_types_lock held */
8650 struct trace_array *trace_array_find(const char *instance)
8652 struct trace_array *tr, *found = NULL;
8654 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8655 if (tr->name && strcmp(tr->name, instance) == 0) {
8664 struct trace_array *trace_array_find_get(const char *instance)
8666 struct trace_array *tr;
8668 mutex_lock(&trace_types_lock);
8669 tr = trace_array_find(instance);
8672 mutex_unlock(&trace_types_lock);
8677 static int trace_array_create_dir(struct trace_array *tr)
8681 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
8685 ret = event_trace_add_tracer(tr->dir, tr);
8687 tracefs_remove(tr->dir);
8691 init_tracer_tracefs(tr, tr->dir);
8692 __update_tracer_options(tr);
8697 static struct trace_array *trace_array_create(const char *name)
8699 struct trace_array *tr;
8703 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
8705 return ERR_PTR(ret);
8707 tr->name = kstrdup(name, GFP_KERNEL);
8711 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
8714 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
8716 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
8718 raw_spin_lock_init(&tr->start_lock);
8720 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
8722 tr->current_trace = &nop_trace;
8724 INIT_LIST_HEAD(&tr->systems);
8725 INIT_LIST_HEAD(&tr->events);
8726 INIT_LIST_HEAD(&tr->hist_vars);
8727 INIT_LIST_HEAD(&tr->err_log);
8729 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
8732 if (ftrace_allocate_ftrace_ops(tr) < 0)
8735 ftrace_init_trace_array(tr);
8737 init_trace_flags_index(tr);
8739 if (trace_instance_dir) {
8740 ret = trace_array_create_dir(tr);
8744 __trace_early_add_events(tr);
8746 list_add(&tr->list, &ftrace_trace_arrays);
8753 ftrace_free_ftrace_ops(tr);
8754 free_trace_buffers(tr);
8755 free_cpumask_var(tr->tracing_cpumask);
8759 return ERR_PTR(ret);
8762 static int instance_mkdir(const char *name)
8764 struct trace_array *tr;
8767 mutex_lock(&event_mutex);
8768 mutex_lock(&trace_types_lock);
8771 if (trace_array_find(name))
8774 tr = trace_array_create(name);
8776 ret = PTR_ERR_OR_ZERO(tr);
8779 mutex_unlock(&trace_types_lock);
8780 mutex_unlock(&event_mutex);
8785 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
8786 * @name: The name of the trace array to be looked up/created.
8788 * Returns pointer to trace array with given name.
8789 * NULL, if it cannot be created.
8791 * NOTE: This function increments the reference counter associated with the
8792 * trace array returned. This makes sure it cannot be freed while in use.
8793 * Use trace_array_put() once the trace array is no longer needed.
8794 * If the trace_array is to be freed, trace_array_destroy() needs to
8795 * be called after the trace_array_put(), or simply let user space delete
8796 * it from the tracefs instances directory. But until the
8797 * trace_array_put() is called, user space can not delete it.
8800 struct trace_array *trace_array_get_by_name(const char *name)
8802 struct trace_array *tr;
8804 mutex_lock(&event_mutex);
8805 mutex_lock(&trace_types_lock);
8807 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8808 if (tr->name && strcmp(tr->name, name) == 0)
8812 tr = trace_array_create(name);
8820 mutex_unlock(&trace_types_lock);
8821 mutex_unlock(&event_mutex);
8824 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
8826 static int __remove_instance(struct trace_array *tr)
8830 /* Reference counter for a newly created trace array = 1. */
8831 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
8834 list_del(&tr->list);
8836 /* Disable all the flags that were enabled coming in */
8837 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
8838 if ((1 << i) & ZEROED_TRACE_FLAGS)
8839 set_tracer_flag(tr, 1 << i, 0);
8842 tracing_set_nop(tr);
8843 clear_ftrace_function_probes(tr);
8844 event_trace_del_tracer(tr);
8845 ftrace_clear_pids(tr);
8846 ftrace_destroy_function_files(tr);
8847 tracefs_remove(tr->dir);
8848 free_trace_buffers(tr);
8850 for (i = 0; i < tr->nr_topts; i++) {
8851 kfree(tr->topts[i].topts);
8855 free_cpumask_var(tr->tracing_cpumask);
8862 int trace_array_destroy(struct trace_array *this_tr)
8864 struct trace_array *tr;
8870 mutex_lock(&event_mutex);
8871 mutex_lock(&trace_types_lock);
8875 /* Making sure trace array exists before destroying it. */
8876 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8877 if (tr == this_tr) {
8878 ret = __remove_instance(tr);
8883 mutex_unlock(&trace_types_lock);
8884 mutex_unlock(&event_mutex);
8888 EXPORT_SYMBOL_GPL(trace_array_destroy);
8890 static int instance_rmdir(const char *name)
8892 struct trace_array *tr;
8895 mutex_lock(&event_mutex);
8896 mutex_lock(&trace_types_lock);
8899 tr = trace_array_find(name);
8901 ret = __remove_instance(tr);
8903 mutex_unlock(&trace_types_lock);
8904 mutex_unlock(&event_mutex);
8909 static __init void create_trace_instances(struct dentry *d_tracer)
8911 struct trace_array *tr;
8913 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
8916 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
8919 mutex_lock(&event_mutex);
8920 mutex_lock(&trace_types_lock);
8922 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8925 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
8926 "Failed to create instance directory\n"))
8930 mutex_unlock(&trace_types_lock);
8931 mutex_unlock(&event_mutex);
8935 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
8937 struct trace_event_file *file;
8940 trace_create_file("available_tracers", 0444, d_tracer,
8941 tr, &show_traces_fops);
8943 trace_create_file("current_tracer", 0644, d_tracer,
8944 tr, &set_tracer_fops);
8946 trace_create_file("tracing_cpumask", 0644, d_tracer,
8947 tr, &tracing_cpumask_fops);
8949 trace_create_file("trace_options", 0644, d_tracer,
8950 tr, &tracing_iter_fops);
8952 trace_create_file("trace", 0644, d_tracer,
8955 trace_create_file("trace_pipe", 0444, d_tracer,
8956 tr, &tracing_pipe_fops);
8958 trace_create_file("buffer_size_kb", 0644, d_tracer,
8959 tr, &tracing_entries_fops);
8961 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
8962 tr, &tracing_total_entries_fops);
8964 trace_create_file("free_buffer", 0200, d_tracer,
8965 tr, &tracing_free_buffer_fops);
8967 trace_create_file("trace_marker", 0220, d_tracer,
8968 tr, &tracing_mark_fops);
8970 file = __find_event_file(tr, "ftrace", "print");
8971 if (file && file->dir)
8972 trace_create_file("trigger", 0644, file->dir, file,
8973 &event_trigger_fops);
8974 tr->trace_marker_file = file;
8976 trace_create_file("trace_marker_raw", 0220, d_tracer,
8977 tr, &tracing_mark_raw_fops);
8979 trace_create_file("trace_clock", 0644, d_tracer, tr,
8982 trace_create_file("tracing_on", 0644, d_tracer,
8983 tr, &rb_simple_fops);
8985 trace_create_file("timestamp_mode", 0444, d_tracer, tr,
8986 &trace_time_stamp_mode_fops);
8988 tr->buffer_percent = 50;
8990 trace_create_file("buffer_percent", 0444, d_tracer,
8991 tr, &buffer_percent_fops);
8993 create_trace_options_dir(tr);
8995 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
8996 trace_create_maxlat_file(tr, d_tracer);
8999 if (ftrace_create_function_files(tr, d_tracer))
9000 MEM_FAIL(1, "Could not allocate function filter files");
9002 #ifdef CONFIG_TRACER_SNAPSHOT
9003 trace_create_file("snapshot", 0644, d_tracer,
9004 tr, &snapshot_fops);
9007 trace_create_file("error_log", 0644, d_tracer,
9008 tr, &tracing_err_log_fops);
9010 for_each_tracing_cpu(cpu)
9011 tracing_init_tracefs_percpu(tr, cpu);
9013 ftrace_init_tracefs(tr, d_tracer);
9016 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9018 struct vfsmount *mnt;
9019 struct file_system_type *type;
9022 * To maintain backward compatibility for tools that mount
9023 * debugfs to get to the tracing facility, tracefs is automatically
9024 * mounted to the debugfs/tracing directory.
9026 type = get_fs_type("tracefs");
9029 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9030 put_filesystem(type);
9039 * tracing_init_dentry - initialize top level trace array
9041 * This is called when creating files or directories in the tracing
9042 * directory. It is called via fs_initcall() by any of the boot up code
9043 * and expects to return the dentry of the top level tracing directory.
9045 int tracing_init_dentry(void)
9047 struct trace_array *tr = &global_trace;
9049 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9050 pr_warn("Tracing disabled due to lockdown\n");
9054 /* The top level trace array uses NULL as parent */
9058 if (WARN_ON(!tracefs_initialized()))
9062 * As there may still be users that expect the tracing
9063 * files to exist in debugfs/tracing, we must automount
9064 * the tracefs file system there, so older tools still
9065 * work with the newer kerenl.
9067 tr->dir = debugfs_create_automount("tracing", NULL,
9068 trace_automount, NULL);
9073 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9074 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9076 static void __init trace_eval_init(void)
9080 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9081 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9084 #ifdef CONFIG_MODULES
9085 static void trace_module_add_evals(struct module *mod)
9087 if (!mod->num_trace_evals)
9091 * Modules with bad taint do not have events created, do
9092 * not bother with enums either.
9094 if (trace_module_has_bad_taint(mod))
9097 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9100 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
9101 static void trace_module_remove_evals(struct module *mod)
9103 union trace_eval_map_item *map;
9104 union trace_eval_map_item **last = &trace_eval_maps;
9106 if (!mod->num_trace_evals)
9109 mutex_lock(&trace_eval_mutex);
9111 map = trace_eval_maps;
9114 if (map->head.mod == mod)
9116 map = trace_eval_jmp_to_tail(map);
9117 last = &map->tail.next;
9118 map = map->tail.next;
9123 *last = trace_eval_jmp_to_tail(map)->tail.next;
9126 mutex_unlock(&trace_eval_mutex);
9129 static inline void trace_module_remove_evals(struct module *mod) { }
9130 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9132 static int trace_module_notify(struct notifier_block *self,
9133 unsigned long val, void *data)
9135 struct module *mod = data;
9138 case MODULE_STATE_COMING:
9139 trace_module_add_evals(mod);
9141 case MODULE_STATE_GOING:
9142 trace_module_remove_evals(mod);
9149 static struct notifier_block trace_module_nb = {
9150 .notifier_call = trace_module_notify,
9153 #endif /* CONFIG_MODULES */
9155 static __init int tracer_init_tracefs(void)
9159 trace_access_lock_init();
9161 ret = tracing_init_dentry();
9167 init_tracer_tracefs(&global_trace, NULL);
9168 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9170 trace_create_file("tracing_thresh", 0644, NULL,
9171 &global_trace, &tracing_thresh_fops);
9173 trace_create_file("README", 0444, NULL,
9174 NULL, &tracing_readme_fops);
9176 trace_create_file("saved_cmdlines", 0444, NULL,
9177 NULL, &tracing_saved_cmdlines_fops);
9179 trace_create_file("saved_cmdlines_size", 0644, NULL,
9180 NULL, &tracing_saved_cmdlines_size_fops);
9182 trace_create_file("saved_tgids", 0444, NULL,
9183 NULL, &tracing_saved_tgids_fops);
9187 trace_create_eval_file(NULL);
9189 #ifdef CONFIG_MODULES
9190 register_module_notifier(&trace_module_nb);
9193 #ifdef CONFIG_DYNAMIC_FTRACE
9194 trace_create_file("dyn_ftrace_total_info", 0444, NULL,
9195 NULL, &tracing_dyn_info_fops);
9198 create_trace_instances(NULL);
9200 update_tracer_options(&global_trace);
9205 static int trace_panic_handler(struct notifier_block *this,
9206 unsigned long event, void *unused)
9208 if (ftrace_dump_on_oops)
9209 ftrace_dump(ftrace_dump_on_oops);
9213 static struct notifier_block trace_panic_notifier = {
9214 .notifier_call = trace_panic_handler,
9216 .priority = 150 /* priority: INT_MAX >= x >= 0 */
9219 static int trace_die_handler(struct notifier_block *self,
9225 if (ftrace_dump_on_oops)
9226 ftrace_dump(ftrace_dump_on_oops);
9234 static struct notifier_block trace_die_notifier = {
9235 .notifier_call = trace_die_handler,
9240 * printk is set to max of 1024, we really don't need it that big.
9241 * Nothing should be printing 1000 characters anyway.
9243 #define TRACE_MAX_PRINT 1000
9246 * Define here KERN_TRACE so that we have one place to modify
9247 * it if we decide to change what log level the ftrace dump
9250 #define KERN_TRACE KERN_EMERG
9253 trace_printk_seq(struct trace_seq *s)
9255 /* Probably should print a warning here. */
9256 if (s->seq.len >= TRACE_MAX_PRINT)
9257 s->seq.len = TRACE_MAX_PRINT;
9260 * More paranoid code. Although the buffer size is set to
9261 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9262 * an extra layer of protection.
9264 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9265 s->seq.len = s->seq.size - 1;
9267 /* should be zero ended, but we are paranoid. */
9268 s->buffer[s->seq.len] = 0;
9270 printk(KERN_TRACE "%s", s->buffer);
9275 void trace_init_global_iter(struct trace_iterator *iter)
9277 iter->tr = &global_trace;
9278 iter->trace = iter->tr->current_trace;
9279 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9280 iter->array_buffer = &global_trace.array_buffer;
9282 if (iter->trace && iter->trace->open)
9283 iter->trace->open(iter);
9285 /* Annotate start of buffers if we had overruns */
9286 if (ring_buffer_overruns(iter->array_buffer->buffer))
9287 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9289 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9290 if (trace_clocks[iter->tr->clock_id].in_ns)
9291 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9294 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
9296 /* use static because iter can be a bit big for the stack */
9297 static struct trace_iterator iter;
9298 static atomic_t dump_running;
9299 struct trace_array *tr = &global_trace;
9300 unsigned int old_userobj;
9301 unsigned long flags;
9304 /* Only allow one dump user at a time. */
9305 if (atomic_inc_return(&dump_running) != 1) {
9306 atomic_dec(&dump_running);
9311 * Always turn off tracing when we dump.
9312 * We don't need to show trace output of what happens
9313 * between multiple crashes.
9315 * If the user does a sysrq-z, then they can re-enable
9316 * tracing with echo 1 > tracing_on.
9320 local_irq_save(flags);
9321 printk_nmi_direct_enter();
9323 /* Simulate the iterator */
9324 trace_init_global_iter(&iter);
9325 /* Can not use kmalloc for iter.temp */
9326 iter.temp = static_temp_buf;
9327 iter.temp_size = STATIC_TEMP_BUF_SIZE;
9329 for_each_tracing_cpu(cpu) {
9330 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9333 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
9335 /* don't look at user memory in panic mode */
9336 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
9338 switch (oops_dump_mode) {
9340 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9343 iter.cpu_file = raw_smp_processor_id();
9348 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
9349 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9352 printk(KERN_TRACE "Dumping ftrace buffer:\n");
9354 /* Did function tracer already get disabled? */
9355 if (ftrace_is_dead()) {
9356 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
9357 printk("# MAY BE MISSING FUNCTION EVENTS\n");
9361 * We need to stop all tracing on all CPUS to read
9362 * the next buffer. This is a bit expensive, but is
9363 * not done often. We fill all what we can read,
9364 * and then release the locks again.
9367 while (!trace_empty(&iter)) {
9370 printk(KERN_TRACE "---------------------------------\n");
9374 trace_iterator_reset(&iter);
9375 iter.iter_flags |= TRACE_FILE_LAT_FMT;
9377 if (trace_find_next_entry_inc(&iter) != NULL) {
9380 ret = print_trace_line(&iter);
9381 if (ret != TRACE_TYPE_NO_CONSUME)
9382 trace_consume(&iter);
9384 touch_nmi_watchdog();
9386 trace_printk_seq(&iter.seq);
9390 printk(KERN_TRACE " (ftrace buffer empty)\n");
9392 printk(KERN_TRACE "---------------------------------\n");
9395 tr->trace_flags |= old_userobj;
9397 for_each_tracing_cpu(cpu) {
9398 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9400 atomic_dec(&dump_running);
9401 printk_nmi_direct_exit();
9402 local_irq_restore(flags);
9404 EXPORT_SYMBOL_GPL(ftrace_dump);
9406 int trace_run_command(const char *buf, int (*createfn)(int, char **))
9413 argv = argv_split(GFP_KERNEL, buf, &argc);
9418 ret = createfn(argc, argv);
9425 #define WRITE_BUFSIZE 4096
9427 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
9428 size_t count, loff_t *ppos,
9429 int (*createfn)(int, char **))
9431 char *kbuf, *buf, *tmp;
9436 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
9440 while (done < count) {
9441 size = count - done;
9443 if (size >= WRITE_BUFSIZE)
9444 size = WRITE_BUFSIZE - 1;
9446 if (copy_from_user(kbuf, buffer + done, size)) {
9453 tmp = strchr(buf, '\n');
9456 size = tmp - buf + 1;
9459 if (done + size < count) {
9462 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
9463 pr_warn("Line length is too long: Should be less than %d\n",
9471 /* Remove comments */
9472 tmp = strchr(buf, '#');
9477 ret = trace_run_command(buf, createfn);
9482 } while (done < count);
9492 __init static int tracer_alloc_buffers(void)
9498 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9499 pr_warn("Tracing disabled due to lockdown\n");
9504 * Make sure we don't accidentally add more trace options
9505 * than we have bits for.
9507 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
9509 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
9512 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
9513 goto out_free_buffer_mask;
9515 /* Only allocate trace_printk buffers if a trace_printk exists */
9516 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
9517 /* Must be called before global_trace.buffer is allocated */
9518 trace_printk_init_buffers();
9520 /* To save memory, keep the ring buffer size to its minimum */
9521 if (ring_buffer_expanded)
9522 ring_buf_size = trace_buf_size;
9526 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
9527 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
9529 raw_spin_lock_init(&global_trace.start_lock);
9532 * The prepare callbacks allocates some memory for the ring buffer. We
9533 * don't free the buffer if the CPU goes down. If we were to free
9534 * the buffer, then the user would lose any trace that was in the
9535 * buffer. The memory will be removed once the "instance" is removed.
9537 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
9538 "trace/RB:preapre", trace_rb_cpu_prepare,
9541 goto out_free_cpumask;
9542 /* Used for event triggers */
9544 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
9546 goto out_rm_hp_state;
9548 if (trace_create_savedcmd() < 0)
9549 goto out_free_temp_buffer;
9551 /* TODO: make the number of buffers hot pluggable with CPUS */
9552 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
9553 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
9554 goto out_free_savedcmd;
9557 if (global_trace.buffer_disabled)
9560 if (trace_boot_clock) {
9561 ret = tracing_set_clock(&global_trace, trace_boot_clock);
9563 pr_warn("Trace clock %s not defined, going back to default\n",
9568 * register_tracer() might reference current_trace, so it
9569 * needs to be set before we register anything. This is
9570 * just a bootstrap of current_trace anyway.
9572 global_trace.current_trace = &nop_trace;
9574 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9576 ftrace_init_global_array_ops(&global_trace);
9578 init_trace_flags_index(&global_trace);
9580 register_tracer(&nop_trace);
9582 /* Function tracing may start here (via kernel command line) */
9583 init_function_trace();
9585 /* All seems OK, enable tracing */
9586 tracing_disabled = 0;
9588 atomic_notifier_chain_register(&panic_notifier_list,
9589 &trace_panic_notifier);
9591 register_die_notifier(&trace_die_notifier);
9593 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
9595 INIT_LIST_HEAD(&global_trace.systems);
9596 INIT_LIST_HEAD(&global_trace.events);
9597 INIT_LIST_HEAD(&global_trace.hist_vars);
9598 INIT_LIST_HEAD(&global_trace.err_log);
9599 list_add(&global_trace.list, &ftrace_trace_arrays);
9601 apply_trace_boot_options();
9603 register_snapshot_cmd();
9608 free_saved_cmdlines_buffer(savedcmd);
9609 out_free_temp_buffer:
9610 ring_buffer_free(temp_buffer);
9612 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
9614 free_cpumask_var(global_trace.tracing_cpumask);
9615 out_free_buffer_mask:
9616 free_cpumask_var(tracing_buffer_mask);
9621 void __init early_trace_init(void)
9623 if (tracepoint_printk) {
9624 tracepoint_print_iter =
9625 kmalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
9626 if (MEM_FAIL(!tracepoint_print_iter,
9627 "Failed to allocate trace iterator\n"))
9628 tracepoint_printk = 0;
9630 static_key_enable(&tracepoint_printk_key.key);
9632 tracer_alloc_buffers();
9635 void __init trace_init(void)
9640 __init static int clear_boot_tracer(void)
9643 * The default tracer at boot buffer is an init section.
9644 * This function is called in lateinit. If we did not
9645 * find the boot tracer, then clear it out, to prevent
9646 * later registration from accessing the buffer that is
9647 * about to be freed.
9649 if (!default_bootup_tracer)
9652 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
9653 default_bootup_tracer);
9654 default_bootup_tracer = NULL;
9659 fs_initcall(tracer_init_tracefs);
9660 late_initcall_sync(clear_boot_tracer);
9662 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
9663 __init static int tracing_set_default_clock(void)
9665 /* sched_clock_stable() is determined in late_initcall */
9666 if (!trace_boot_clock && !sched_clock_stable()) {
9667 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9668 pr_warn("Can not set tracing clock due to lockdown\n");
9673 "Unstable clock detected, switching default tracing clock to \"global\"\n"
9674 "If you want to keep using the local clock, then add:\n"
9675 " \"trace_clock=local\"\n"
9676 "on the kernel command line\n");
9677 tracing_set_clock(&global_trace, "global");
9682 late_initcall_sync(tracing_set_default_clock);