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 a tracer is running, we do not want to run SELFTEST.
73 bool __read_mostly tracing_selftest_disabled;
75 /* Pipe tracepoints to printk */
76 struct trace_iterator *tracepoint_print_iter;
77 int tracepoint_printk;
78 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
80 /* For tracers that don't implement custom flags */
81 static struct tracer_opt dummy_tracer_opt[] = {
86 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
92 * To prevent the comm cache from being overwritten when no
93 * tracing is active, only save the comm when a trace event
96 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
99 * Kill all tracing for good (never come back).
100 * It is initialized to 1 but will turn to zero if the initialization
101 * of the tracer is successful. But that is the only place that sets
104 static int tracing_disabled = 1;
106 cpumask_var_t __read_mostly tracing_buffer_mask;
109 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
111 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
112 * is set, then ftrace_dump is called. This will output the contents
113 * of the ftrace buffers to the console. This is very useful for
114 * capturing traces that lead to crashes and outputing it to a
117 * It is default off, but you can enable it with either specifying
118 * "ftrace_dump_on_oops" in the kernel command line, or setting
119 * /proc/sys/kernel/ftrace_dump_on_oops
120 * Set 1 if you want to dump buffers of all CPUs
121 * Set 2 if you want to dump the buffer of the CPU that triggered oops
124 enum ftrace_dump_mode ftrace_dump_on_oops;
126 /* When set, tracing will stop when a WARN*() is hit */
127 int __disable_trace_on_warning;
129 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
130 /* Map of enums to their values, for "eval_map" file */
131 struct trace_eval_map_head {
133 unsigned long length;
136 union trace_eval_map_item;
138 struct trace_eval_map_tail {
140 * "end" is first and points to NULL as it must be different
141 * than "mod" or "eval_string"
143 union trace_eval_map_item *next;
144 const char *end; /* points to NULL */
147 static DEFINE_MUTEX(trace_eval_mutex);
150 * The trace_eval_maps are saved in an array with two extra elements,
151 * one at the beginning, and one at the end. The beginning item contains
152 * the count of the saved maps (head.length), and the module they
153 * belong to if not built in (head.mod). The ending item contains a
154 * pointer to the next array of saved eval_map items.
156 union trace_eval_map_item {
157 struct trace_eval_map map;
158 struct trace_eval_map_head head;
159 struct trace_eval_map_tail tail;
162 static union trace_eval_map_item *trace_eval_maps;
163 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
165 int tracing_set_tracer(struct trace_array *tr, const char *buf);
166 static void ftrace_trace_userstack(struct trace_buffer *buffer,
167 unsigned long flags, int pc);
169 #define MAX_TRACER_SIZE 100
170 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
171 static char *default_bootup_tracer;
173 static bool allocate_snapshot;
175 static int __init set_cmdline_ftrace(char *str)
177 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
178 default_bootup_tracer = bootup_tracer_buf;
179 /* We are using ftrace early, expand it */
180 ring_buffer_expanded = true;
183 __setup("ftrace=", set_cmdline_ftrace);
185 static int __init set_ftrace_dump_on_oops(char *str)
187 if (*str++ != '=' || !*str) {
188 ftrace_dump_on_oops = DUMP_ALL;
192 if (!strcmp("orig_cpu", str)) {
193 ftrace_dump_on_oops = DUMP_ORIG;
199 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
201 static int __init stop_trace_on_warning(char *str)
203 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
204 __disable_trace_on_warning = 1;
207 __setup("traceoff_on_warning", stop_trace_on_warning);
209 static int __init boot_alloc_snapshot(char *str)
211 allocate_snapshot = true;
212 /* We also need the main ring buffer expanded */
213 ring_buffer_expanded = true;
216 __setup("alloc_snapshot", boot_alloc_snapshot);
219 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
221 static int __init set_trace_boot_options(char *str)
223 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
226 __setup("trace_options=", set_trace_boot_options);
228 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
229 static char *trace_boot_clock __initdata;
231 static int __init set_trace_boot_clock(char *str)
233 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
234 trace_boot_clock = trace_boot_clock_buf;
237 __setup("trace_clock=", set_trace_boot_clock);
239 static int __init set_tracepoint_printk(char *str)
241 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
242 tracepoint_printk = 1;
245 __setup("tp_printk", set_tracepoint_printk);
247 unsigned long long ns2usecs(u64 nsec)
255 trace_process_export(struct trace_export *export,
256 struct ring_buffer_event *event, int flag)
258 struct trace_entry *entry;
259 unsigned int size = 0;
261 if (export->flags & flag) {
262 entry = ring_buffer_event_data(event);
263 size = ring_buffer_event_length(event);
264 export->write(export, entry, size);
268 static DEFINE_MUTEX(ftrace_export_lock);
270 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
272 static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
273 static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
274 static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
276 static inline void ftrace_exports_enable(struct trace_export *export)
278 if (export->flags & TRACE_EXPORT_FUNCTION)
279 static_branch_inc(&trace_function_exports_enabled);
281 if (export->flags & TRACE_EXPORT_EVENT)
282 static_branch_inc(&trace_event_exports_enabled);
284 if (export->flags & TRACE_EXPORT_MARKER)
285 static_branch_inc(&trace_marker_exports_enabled);
288 static inline void ftrace_exports_disable(struct trace_export *export)
290 if (export->flags & TRACE_EXPORT_FUNCTION)
291 static_branch_dec(&trace_function_exports_enabled);
293 if (export->flags & TRACE_EXPORT_EVENT)
294 static_branch_dec(&trace_event_exports_enabled);
296 if (export->flags & TRACE_EXPORT_MARKER)
297 static_branch_dec(&trace_marker_exports_enabled);
300 static void ftrace_exports(struct ring_buffer_event *event, int flag)
302 struct trace_export *export;
304 preempt_disable_notrace();
306 export = rcu_dereference_raw_check(ftrace_exports_list);
308 trace_process_export(export, event, flag);
309 export = rcu_dereference_raw_check(export->next);
312 preempt_enable_notrace();
316 add_trace_export(struct trace_export **list, struct trace_export *export)
318 rcu_assign_pointer(export->next, *list);
320 * We are entering export into the list but another
321 * CPU might be walking that list. We need to make sure
322 * the export->next pointer is valid before another CPU sees
323 * the export pointer included into the list.
325 rcu_assign_pointer(*list, export);
329 rm_trace_export(struct trace_export **list, struct trace_export *export)
331 struct trace_export **p;
333 for (p = list; *p != NULL; p = &(*p)->next)
340 rcu_assign_pointer(*p, (*p)->next);
346 add_ftrace_export(struct trace_export **list, struct trace_export *export)
348 ftrace_exports_enable(export);
350 add_trace_export(list, export);
354 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
358 ret = rm_trace_export(list, export);
359 ftrace_exports_disable(export);
364 int register_ftrace_export(struct trace_export *export)
366 if (WARN_ON_ONCE(!export->write))
369 mutex_lock(&ftrace_export_lock);
371 add_ftrace_export(&ftrace_exports_list, export);
373 mutex_unlock(&ftrace_export_lock);
377 EXPORT_SYMBOL_GPL(register_ftrace_export);
379 int unregister_ftrace_export(struct trace_export *export)
383 mutex_lock(&ftrace_export_lock);
385 ret = rm_ftrace_export(&ftrace_exports_list, export);
387 mutex_unlock(&ftrace_export_lock);
391 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
393 /* trace_flags holds trace_options default values */
394 #define TRACE_DEFAULT_FLAGS \
395 (FUNCTION_DEFAULT_FLAGS | \
396 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
397 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
398 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
399 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS)
401 /* trace_options that are only supported by global_trace */
402 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
403 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
405 /* trace_flags that are default zero for instances */
406 #define ZEROED_TRACE_FLAGS \
407 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
410 * The global_trace is the descriptor that holds the top-level tracing
411 * buffers for the live tracing.
413 static struct trace_array global_trace = {
414 .trace_flags = TRACE_DEFAULT_FLAGS,
417 LIST_HEAD(ftrace_trace_arrays);
419 int trace_array_get(struct trace_array *this_tr)
421 struct trace_array *tr;
424 mutex_lock(&trace_types_lock);
425 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
432 mutex_unlock(&trace_types_lock);
437 static void __trace_array_put(struct trace_array *this_tr)
439 WARN_ON(!this_tr->ref);
444 * trace_array_put - Decrement the reference counter for this trace array.
446 * NOTE: Use this when we no longer need the trace array returned by
447 * trace_array_get_by_name(). This ensures the trace array can be later
451 void trace_array_put(struct trace_array *this_tr)
456 mutex_lock(&trace_types_lock);
457 __trace_array_put(this_tr);
458 mutex_unlock(&trace_types_lock);
460 EXPORT_SYMBOL_GPL(trace_array_put);
462 int tracing_check_open_get_tr(struct trace_array *tr)
466 ret = security_locked_down(LOCKDOWN_TRACEFS);
470 if (tracing_disabled)
473 if (tr && trace_array_get(tr) < 0)
479 int call_filter_check_discard(struct trace_event_call *call, void *rec,
480 struct trace_buffer *buffer,
481 struct ring_buffer_event *event)
483 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
484 !filter_match_preds(call->filter, rec)) {
485 __trace_event_discard_commit(buffer, event);
492 void trace_free_pid_list(struct trace_pid_list *pid_list)
494 vfree(pid_list->pids);
499 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
500 * @filtered_pids: The list of pids to check
501 * @search_pid: The PID to find in @filtered_pids
503 * Returns true if @search_pid is fonud in @filtered_pids, and false otherwis.
506 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
509 * If pid_max changed after filtered_pids was created, we
510 * by default ignore all pids greater than the previous pid_max.
512 if (search_pid >= filtered_pids->pid_max)
515 return test_bit(search_pid, filtered_pids->pids);
519 * trace_ignore_this_task - should a task be ignored for tracing
520 * @filtered_pids: The list of pids to check
521 * @task: The task that should be ignored if not filtered
523 * Checks if @task should be traced or not from @filtered_pids.
524 * Returns true if @task should *NOT* be traced.
525 * Returns false if @task should be traced.
528 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
529 struct trace_pid_list *filtered_no_pids,
530 struct task_struct *task)
533 * If filterd_no_pids is not empty, and the task's pid is listed
534 * in filtered_no_pids, then return true.
535 * Otherwise, if filtered_pids is empty, that means we can
536 * trace all tasks. If it has content, then only trace pids
537 * within filtered_pids.
540 return (filtered_pids &&
541 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
543 trace_find_filtered_pid(filtered_no_pids, task->pid));
547 * trace_filter_add_remove_task - Add or remove a task from a pid_list
548 * @pid_list: The list to modify
549 * @self: The current task for fork or NULL for exit
550 * @task: The task to add or remove
552 * If adding a task, if @self is defined, the task is only added if @self
553 * is also included in @pid_list. This happens on fork and tasks should
554 * only be added when the parent is listed. If @self is NULL, then the
555 * @task pid will be removed from the list, which would happen on exit
558 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
559 struct task_struct *self,
560 struct task_struct *task)
565 /* For forks, we only add if the forking task is listed */
567 if (!trace_find_filtered_pid(pid_list, self->pid))
571 /* Sorry, but we don't support pid_max changing after setting */
572 if (task->pid >= pid_list->pid_max)
575 /* "self" is set for forks, and NULL for exits */
577 set_bit(task->pid, pid_list->pids);
579 clear_bit(task->pid, pid_list->pids);
583 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
584 * @pid_list: The pid list to show
585 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
586 * @pos: The position of the file
588 * This is used by the seq_file "next" operation to iterate the pids
589 * listed in a trace_pid_list structure.
591 * Returns the pid+1 as we want to display pid of zero, but NULL would
592 * stop the iteration.
594 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
596 unsigned long pid = (unsigned long)v;
600 /* pid already is +1 of the actual prevous bit */
601 pid = find_next_bit(pid_list->pids, pid_list->pid_max, pid);
603 /* Return pid + 1 to allow zero to be represented */
604 if (pid < pid_list->pid_max)
605 return (void *)(pid + 1);
611 * trace_pid_start - Used for seq_file to start reading pid lists
612 * @pid_list: The pid list to show
613 * @pos: The position of the file
615 * This is used by seq_file "start" operation to start the iteration
618 * Returns the pid+1 as we want to display pid of zero, but NULL would
619 * stop the iteration.
621 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
626 pid = find_first_bit(pid_list->pids, pid_list->pid_max);
627 if (pid >= pid_list->pid_max)
630 /* Return pid + 1 so that zero can be the exit value */
631 for (pid++; pid && l < *pos;
632 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
638 * trace_pid_show - show the current pid in seq_file processing
639 * @m: The seq_file structure to write into
640 * @v: A void pointer of the pid (+1) value to display
642 * Can be directly used by seq_file operations to display the current
645 int trace_pid_show(struct seq_file *m, void *v)
647 unsigned long pid = (unsigned long)v - 1;
649 seq_printf(m, "%lu\n", pid);
653 /* 128 should be much more than enough */
654 #define PID_BUF_SIZE 127
656 int trace_pid_write(struct trace_pid_list *filtered_pids,
657 struct trace_pid_list **new_pid_list,
658 const char __user *ubuf, size_t cnt)
660 struct trace_pid_list *pid_list;
661 struct trace_parser parser;
669 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
673 * Always recreate a new array. The write is an all or nothing
674 * operation. Always create a new array when adding new pids by
675 * the user. If the operation fails, then the current list is
678 pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
680 trace_parser_put(&parser);
684 pid_list->pid_max = READ_ONCE(pid_max);
686 /* Only truncating will shrink pid_max */
687 if (filtered_pids && filtered_pids->pid_max > pid_list->pid_max)
688 pid_list->pid_max = filtered_pids->pid_max;
690 pid_list->pids = vzalloc((pid_list->pid_max + 7) >> 3);
691 if (!pid_list->pids) {
692 trace_parser_put(&parser);
698 /* copy the current bits to the new max */
699 for_each_set_bit(pid, filtered_pids->pids,
700 filtered_pids->pid_max) {
701 set_bit(pid, pid_list->pids);
710 ret = trace_get_user(&parser, ubuf, cnt, &pos);
711 if (ret < 0 || !trace_parser_loaded(&parser))
719 if (kstrtoul(parser.buffer, 0, &val))
721 if (val >= pid_list->pid_max)
726 set_bit(pid, pid_list->pids);
729 trace_parser_clear(&parser);
732 trace_parser_put(&parser);
735 trace_free_pid_list(pid_list);
740 /* Cleared the list of pids */
741 trace_free_pid_list(pid_list);
746 *new_pid_list = pid_list;
751 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
755 /* Early boot up does not have a buffer yet */
757 return trace_clock_local();
759 ts = ring_buffer_time_stamp(buf->buffer, cpu);
760 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
765 u64 ftrace_now(int cpu)
767 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
771 * tracing_is_enabled - Show if global_trace has been disabled
773 * Shows if the global trace has been enabled or not. It uses the
774 * mirror flag "buffer_disabled" to be used in fast paths such as for
775 * the irqsoff tracer. But it may be inaccurate due to races. If you
776 * need to know the accurate state, use tracing_is_on() which is a little
777 * slower, but accurate.
779 int tracing_is_enabled(void)
782 * For quick access (irqsoff uses this in fast path), just
783 * return the mirror variable of the state of the ring buffer.
784 * It's a little racy, but we don't really care.
787 return !global_trace.buffer_disabled;
791 * trace_buf_size is the size in bytes that is allocated
792 * for a buffer. Note, the number of bytes is always rounded
795 * This number is purposely set to a low number of 16384.
796 * If the dump on oops happens, it will be much appreciated
797 * to not have to wait for all that output. Anyway this can be
798 * boot time and run time configurable.
800 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
802 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
804 /* trace_types holds a link list of available tracers. */
805 static struct tracer *trace_types __read_mostly;
808 * trace_types_lock is used to protect the trace_types list.
810 DEFINE_MUTEX(trace_types_lock);
813 * serialize the access of the ring buffer
815 * ring buffer serializes readers, but it is low level protection.
816 * The validity of the events (which returns by ring_buffer_peek() ..etc)
817 * are not protected by ring buffer.
819 * The content of events may become garbage if we allow other process consumes
820 * these events concurrently:
821 * A) the page of the consumed events may become a normal page
822 * (not reader page) in ring buffer, and this page will be rewrited
823 * by events producer.
824 * B) The page of the consumed events may become a page for splice_read,
825 * and this page will be returned to system.
827 * These primitives allow multi process access to different cpu ring buffer
830 * These primitives don't distinguish read-only and read-consume access.
831 * Multi read-only access are also serialized.
835 static DECLARE_RWSEM(all_cpu_access_lock);
836 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
838 static inline void trace_access_lock(int cpu)
840 if (cpu == RING_BUFFER_ALL_CPUS) {
841 /* gain it for accessing the whole ring buffer. */
842 down_write(&all_cpu_access_lock);
844 /* gain it for accessing a cpu ring buffer. */
846 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
847 down_read(&all_cpu_access_lock);
849 /* Secondly block other access to this @cpu ring buffer. */
850 mutex_lock(&per_cpu(cpu_access_lock, cpu));
854 static inline void trace_access_unlock(int cpu)
856 if (cpu == RING_BUFFER_ALL_CPUS) {
857 up_write(&all_cpu_access_lock);
859 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
860 up_read(&all_cpu_access_lock);
864 static inline void trace_access_lock_init(void)
868 for_each_possible_cpu(cpu)
869 mutex_init(&per_cpu(cpu_access_lock, cpu));
874 static DEFINE_MUTEX(access_lock);
876 static inline void trace_access_lock(int cpu)
879 mutex_lock(&access_lock);
882 static inline void trace_access_unlock(int cpu)
885 mutex_unlock(&access_lock);
888 static inline void trace_access_lock_init(void)
894 #ifdef CONFIG_STACKTRACE
895 static void __ftrace_trace_stack(struct trace_buffer *buffer,
897 int skip, int pc, struct pt_regs *regs);
898 static inline void ftrace_trace_stack(struct trace_array *tr,
899 struct trace_buffer *buffer,
901 int skip, int pc, struct pt_regs *regs);
904 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
906 int skip, int pc, struct pt_regs *regs)
909 static inline void ftrace_trace_stack(struct trace_array *tr,
910 struct trace_buffer *buffer,
912 int skip, int pc, struct pt_regs *regs)
918 static __always_inline void
919 trace_event_setup(struct ring_buffer_event *event,
920 int type, unsigned long flags, int pc)
922 struct trace_entry *ent = ring_buffer_event_data(event);
924 tracing_generic_entry_update(ent, type, flags, pc);
927 static __always_inline struct ring_buffer_event *
928 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
931 unsigned long flags, int pc)
933 struct ring_buffer_event *event;
935 event = ring_buffer_lock_reserve(buffer, len);
937 trace_event_setup(event, type, flags, pc);
942 void tracer_tracing_on(struct trace_array *tr)
944 if (tr->array_buffer.buffer)
945 ring_buffer_record_on(tr->array_buffer.buffer);
947 * This flag is looked at when buffers haven't been allocated
948 * yet, or by some tracers (like irqsoff), that just want to
949 * know if the ring buffer has been disabled, but it can handle
950 * races of where it gets disabled but we still do a record.
951 * As the check is in the fast path of the tracers, it is more
952 * important to be fast than accurate.
954 tr->buffer_disabled = 0;
955 /* Make the flag seen by readers */
960 * tracing_on - enable tracing buffers
962 * This function enables tracing buffers that may have been
963 * disabled with tracing_off.
965 void tracing_on(void)
967 tracer_tracing_on(&global_trace);
969 EXPORT_SYMBOL_GPL(tracing_on);
972 static __always_inline void
973 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
975 __this_cpu_write(trace_taskinfo_save, true);
977 /* If this is the temp buffer, we need to commit fully */
978 if (this_cpu_read(trace_buffered_event) == event) {
979 /* Length is in event->array[0] */
980 ring_buffer_write(buffer, event->array[0], &event->array[1]);
981 /* Release the temp buffer */
982 this_cpu_dec(trace_buffered_event_cnt);
984 ring_buffer_unlock_commit(buffer, event);
988 * __trace_puts - write a constant string into the trace buffer.
989 * @ip: The address of the caller
990 * @str: The constant string to write
991 * @size: The size of the string.
993 int __trace_puts(unsigned long ip, const char *str, int size)
995 struct ring_buffer_event *event;
996 struct trace_buffer *buffer;
997 struct print_entry *entry;
998 unsigned long irq_flags;
1002 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1005 pc = preempt_count();
1007 if (unlikely(tracing_selftest_running || tracing_disabled))
1010 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1012 local_save_flags(irq_flags);
1013 buffer = global_trace.array_buffer.buffer;
1014 ring_buffer_nest_start(buffer);
1015 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1022 entry = ring_buffer_event_data(event);
1025 memcpy(&entry->buf, str, size);
1027 /* Add a newline if necessary */
1028 if (entry->buf[size - 1] != '\n') {
1029 entry->buf[size] = '\n';
1030 entry->buf[size + 1] = '\0';
1032 entry->buf[size] = '\0';
1034 __buffer_unlock_commit(buffer, event);
1035 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
1037 ring_buffer_nest_end(buffer);
1040 EXPORT_SYMBOL_GPL(__trace_puts);
1043 * __trace_bputs - write the pointer to a constant string into trace buffer
1044 * @ip: The address of the caller
1045 * @str: The constant string to write to the buffer to
1047 int __trace_bputs(unsigned long ip, const char *str)
1049 struct ring_buffer_event *event;
1050 struct trace_buffer *buffer;
1051 struct bputs_entry *entry;
1052 unsigned long irq_flags;
1053 int size = sizeof(struct bputs_entry);
1057 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1060 pc = preempt_count();
1062 if (unlikely(tracing_selftest_running || tracing_disabled))
1065 local_save_flags(irq_flags);
1066 buffer = global_trace.array_buffer.buffer;
1068 ring_buffer_nest_start(buffer);
1069 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1074 entry = ring_buffer_event_data(event);
1078 __buffer_unlock_commit(buffer, event);
1079 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
1083 ring_buffer_nest_end(buffer);
1086 EXPORT_SYMBOL_GPL(__trace_bputs);
1088 #ifdef CONFIG_TRACER_SNAPSHOT
1089 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1092 struct tracer *tracer = tr->current_trace;
1093 unsigned long flags;
1096 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1097 internal_trace_puts("*** snapshot is being ignored ***\n");
1101 if (!tr->allocated_snapshot) {
1102 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
1103 internal_trace_puts("*** stopping trace here! ***\n");
1108 /* Note, snapshot can not be used when the tracer uses it */
1109 if (tracer->use_max_tr) {
1110 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
1111 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
1115 local_irq_save(flags);
1116 update_max_tr(tr, current, smp_processor_id(), cond_data);
1117 local_irq_restore(flags);
1120 void tracing_snapshot_instance(struct trace_array *tr)
1122 tracing_snapshot_instance_cond(tr, NULL);
1126 * tracing_snapshot - take a snapshot of the current buffer.
1128 * This causes a swap between the snapshot buffer and the current live
1129 * tracing buffer. You can use this to take snapshots of the live
1130 * trace when some condition is triggered, but continue to trace.
1132 * Note, make sure to allocate the snapshot with either
1133 * a tracing_snapshot_alloc(), or by doing it manually
1134 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
1136 * If the snapshot buffer is not allocated, it will stop tracing.
1137 * Basically making a permanent snapshot.
1139 void tracing_snapshot(void)
1141 struct trace_array *tr = &global_trace;
1143 tracing_snapshot_instance(tr);
1145 EXPORT_SYMBOL_GPL(tracing_snapshot);
1148 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1149 * @tr: The tracing instance to snapshot
1150 * @cond_data: The data to be tested conditionally, and possibly saved
1152 * This is the same as tracing_snapshot() except that the snapshot is
1153 * conditional - the snapshot will only happen if the
1154 * cond_snapshot.update() implementation receiving the cond_data
1155 * returns true, which means that the trace array's cond_snapshot
1156 * update() operation used the cond_data to determine whether the
1157 * snapshot should be taken, and if it was, presumably saved it along
1158 * with the snapshot.
1160 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1162 tracing_snapshot_instance_cond(tr, cond_data);
1164 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1167 * tracing_snapshot_cond_data - get the user data associated with a snapshot
1168 * @tr: The tracing instance
1170 * When the user enables a conditional snapshot using
1171 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1172 * with the snapshot. This accessor is used to retrieve it.
1174 * Should not be called from cond_snapshot.update(), since it takes
1175 * the tr->max_lock lock, which the code calling
1176 * cond_snapshot.update() has already done.
1178 * Returns the cond_data associated with the trace array's snapshot.
1180 void *tracing_cond_snapshot_data(struct trace_array *tr)
1182 void *cond_data = NULL;
1184 arch_spin_lock(&tr->max_lock);
1186 if (tr->cond_snapshot)
1187 cond_data = tr->cond_snapshot->cond_data;
1189 arch_spin_unlock(&tr->max_lock);
1193 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1195 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1196 struct array_buffer *size_buf, int cpu_id);
1197 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1199 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1203 if (!tr->allocated_snapshot) {
1205 /* allocate spare buffer */
1206 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1207 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1211 tr->allocated_snapshot = true;
1217 static void free_snapshot(struct trace_array *tr)
1220 * We don't free the ring buffer. instead, resize it because
1221 * The max_tr ring buffer has some state (e.g. ring->clock) and
1222 * we want preserve it.
1224 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1225 set_buffer_entries(&tr->max_buffer, 1);
1226 tracing_reset_online_cpus(&tr->max_buffer);
1227 tr->allocated_snapshot = false;
1231 * tracing_alloc_snapshot - allocate snapshot buffer.
1233 * This only allocates the snapshot buffer if it isn't already
1234 * allocated - it doesn't also take a snapshot.
1236 * This is meant to be used in cases where the snapshot buffer needs
1237 * to be set up for events that can't sleep but need to be able to
1238 * trigger a snapshot.
1240 int tracing_alloc_snapshot(void)
1242 struct trace_array *tr = &global_trace;
1245 ret = tracing_alloc_snapshot_instance(tr);
1250 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1253 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1255 * This is similar to tracing_snapshot(), but it will allocate the
1256 * snapshot buffer if it isn't already allocated. Use this only
1257 * where it is safe to sleep, as the allocation may sleep.
1259 * This causes a swap between the snapshot buffer and the current live
1260 * tracing buffer. You can use this to take snapshots of the live
1261 * trace when some condition is triggered, but continue to trace.
1263 void tracing_snapshot_alloc(void)
1267 ret = tracing_alloc_snapshot();
1273 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1276 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1277 * @tr: The tracing instance
1278 * @cond_data: User data to associate with the snapshot
1279 * @update: Implementation of the cond_snapshot update function
1281 * Check whether the conditional snapshot for the given instance has
1282 * already been enabled, or if the current tracer is already using a
1283 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1284 * save the cond_data and update function inside.
1286 * Returns 0 if successful, error otherwise.
1288 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1289 cond_update_fn_t update)
1291 struct cond_snapshot *cond_snapshot;
1294 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1298 cond_snapshot->cond_data = cond_data;
1299 cond_snapshot->update = update;
1301 mutex_lock(&trace_types_lock);
1303 ret = tracing_alloc_snapshot_instance(tr);
1307 if (tr->current_trace->use_max_tr) {
1313 * The cond_snapshot can only change to NULL without the
1314 * trace_types_lock. We don't care if we race with it going
1315 * to NULL, but we want to make sure that it's not set to
1316 * something other than NULL when we get here, which we can
1317 * do safely with only holding the trace_types_lock and not
1318 * having to take the max_lock.
1320 if (tr->cond_snapshot) {
1325 arch_spin_lock(&tr->max_lock);
1326 tr->cond_snapshot = cond_snapshot;
1327 arch_spin_unlock(&tr->max_lock);
1329 mutex_unlock(&trace_types_lock);
1334 mutex_unlock(&trace_types_lock);
1335 kfree(cond_snapshot);
1338 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1341 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1342 * @tr: The tracing instance
1344 * Check whether the conditional snapshot for the given instance is
1345 * enabled; if so, free the cond_snapshot associated with it,
1346 * otherwise return -EINVAL.
1348 * Returns 0 if successful, error otherwise.
1350 int tracing_snapshot_cond_disable(struct trace_array *tr)
1354 arch_spin_lock(&tr->max_lock);
1356 if (!tr->cond_snapshot)
1359 kfree(tr->cond_snapshot);
1360 tr->cond_snapshot = NULL;
1363 arch_spin_unlock(&tr->max_lock);
1367 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1369 void tracing_snapshot(void)
1371 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1373 EXPORT_SYMBOL_GPL(tracing_snapshot);
1374 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1376 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1378 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1379 int tracing_alloc_snapshot(void)
1381 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1384 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1385 void tracing_snapshot_alloc(void)
1390 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1391 void *tracing_cond_snapshot_data(struct trace_array *tr)
1395 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1396 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1400 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1401 int tracing_snapshot_cond_disable(struct trace_array *tr)
1405 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1406 #endif /* CONFIG_TRACER_SNAPSHOT */
1408 void tracer_tracing_off(struct trace_array *tr)
1410 if (tr->array_buffer.buffer)
1411 ring_buffer_record_off(tr->array_buffer.buffer);
1413 * This flag is looked at when buffers haven't been allocated
1414 * yet, or by some tracers (like irqsoff), that just want to
1415 * know if the ring buffer has been disabled, but it can handle
1416 * races of where it gets disabled but we still do a record.
1417 * As the check is in the fast path of the tracers, it is more
1418 * important to be fast than accurate.
1420 tr->buffer_disabled = 1;
1421 /* Make the flag seen by readers */
1426 * tracing_off - turn off tracing buffers
1428 * This function stops the tracing buffers from recording data.
1429 * It does not disable any overhead the tracers themselves may
1430 * be causing. This function simply causes all recording to
1431 * the ring buffers to fail.
1433 void tracing_off(void)
1435 tracer_tracing_off(&global_trace);
1437 EXPORT_SYMBOL_GPL(tracing_off);
1439 void disable_trace_on_warning(void)
1441 if (__disable_trace_on_warning) {
1442 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1443 "Disabling tracing due to warning\n");
1449 * tracer_tracing_is_on - show real state of ring buffer enabled
1450 * @tr : the trace array to know if ring buffer is enabled
1452 * Shows real state of the ring buffer if it is enabled or not.
1454 bool tracer_tracing_is_on(struct trace_array *tr)
1456 if (tr->array_buffer.buffer)
1457 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1458 return !tr->buffer_disabled;
1462 * tracing_is_on - show state of ring buffers enabled
1464 int tracing_is_on(void)
1466 return tracer_tracing_is_on(&global_trace);
1468 EXPORT_SYMBOL_GPL(tracing_is_on);
1470 static int __init set_buf_size(char *str)
1472 unsigned long buf_size;
1476 buf_size = memparse(str, &str);
1477 /* nr_entries can not be zero */
1480 trace_buf_size = buf_size;
1483 __setup("trace_buf_size=", set_buf_size);
1485 static int __init set_tracing_thresh(char *str)
1487 unsigned long threshold;
1492 ret = kstrtoul(str, 0, &threshold);
1495 tracing_thresh = threshold * 1000;
1498 __setup("tracing_thresh=", set_tracing_thresh);
1500 unsigned long nsecs_to_usecs(unsigned long nsecs)
1502 return nsecs / 1000;
1506 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1507 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1508 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1509 * of strings in the order that the evals (enum) were defined.
1514 /* These must match the bit postions in trace_iterator_flags */
1515 static const char *trace_options[] = {
1523 int in_ns; /* is this clock in nanoseconds? */
1524 } trace_clocks[] = {
1525 { trace_clock_local, "local", 1 },
1526 { trace_clock_global, "global", 1 },
1527 { trace_clock_counter, "counter", 0 },
1528 { trace_clock_jiffies, "uptime", 0 },
1529 { trace_clock, "perf", 1 },
1530 { ktime_get_mono_fast_ns, "mono", 1 },
1531 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1532 { ktime_get_boot_fast_ns, "boot", 1 },
1536 bool trace_clock_in_ns(struct trace_array *tr)
1538 if (trace_clocks[tr->clock_id].in_ns)
1545 * trace_parser_get_init - gets the buffer for trace parser
1547 int trace_parser_get_init(struct trace_parser *parser, int size)
1549 memset(parser, 0, sizeof(*parser));
1551 parser->buffer = kmalloc(size, GFP_KERNEL);
1552 if (!parser->buffer)
1555 parser->size = size;
1560 * trace_parser_put - frees the buffer for trace parser
1562 void trace_parser_put(struct trace_parser *parser)
1564 kfree(parser->buffer);
1565 parser->buffer = NULL;
1569 * trace_get_user - reads the user input string separated by space
1570 * (matched by isspace(ch))
1572 * For each string found the 'struct trace_parser' is updated,
1573 * and the function returns.
1575 * Returns number of bytes read.
1577 * See kernel/trace/trace.h for 'struct trace_parser' details.
1579 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1580 size_t cnt, loff_t *ppos)
1587 trace_parser_clear(parser);
1589 ret = get_user(ch, ubuf++);
1597 * The parser is not finished with the last write,
1598 * continue reading the user input without skipping spaces.
1600 if (!parser->cont) {
1601 /* skip white space */
1602 while (cnt && isspace(ch)) {
1603 ret = get_user(ch, ubuf++);
1612 /* only spaces were written */
1613 if (isspace(ch) || !ch) {
1620 /* read the non-space input */
1621 while (cnt && !isspace(ch) && ch) {
1622 if (parser->idx < parser->size - 1)
1623 parser->buffer[parser->idx++] = ch;
1628 ret = get_user(ch, ubuf++);
1635 /* We either got finished input or we have to wait for another call. */
1636 if (isspace(ch) || !ch) {
1637 parser->buffer[parser->idx] = 0;
1638 parser->cont = false;
1639 } else if (parser->idx < parser->size - 1) {
1640 parser->cont = true;
1641 parser->buffer[parser->idx++] = ch;
1642 /* Make sure the parsed string always terminates with '\0'. */
1643 parser->buffer[parser->idx] = 0;
1656 /* TODO add a seq_buf_to_buffer() */
1657 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1661 if (trace_seq_used(s) <= s->seq.readpos)
1664 len = trace_seq_used(s) - s->seq.readpos;
1667 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1669 s->seq.readpos += cnt;
1673 unsigned long __read_mostly tracing_thresh;
1674 static const struct file_operations tracing_max_lat_fops;
1676 #if (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1677 defined(CONFIG_FSNOTIFY)
1679 static struct workqueue_struct *fsnotify_wq;
1681 static void latency_fsnotify_workfn(struct work_struct *work)
1683 struct trace_array *tr = container_of(work, struct trace_array,
1685 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1688 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1690 struct trace_array *tr = container_of(iwork, struct trace_array,
1692 queue_work(fsnotify_wq, &tr->fsnotify_work);
1695 static void trace_create_maxlat_file(struct trace_array *tr,
1696 struct dentry *d_tracer)
1698 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1699 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1700 tr->d_max_latency = trace_create_file("tracing_max_latency", 0644,
1701 d_tracer, &tr->max_latency,
1702 &tracing_max_lat_fops);
1705 __init static int latency_fsnotify_init(void)
1707 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1708 WQ_UNBOUND | WQ_HIGHPRI, 0);
1710 pr_err("Unable to allocate tr_max_lat_wq\n");
1716 late_initcall_sync(latency_fsnotify_init);
1718 void latency_fsnotify(struct trace_array *tr)
1723 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1724 * possible that we are called from __schedule() or do_idle(), which
1725 * could cause a deadlock.
1727 irq_work_queue(&tr->fsnotify_irqwork);
1731 * (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1732 * defined(CONFIG_FSNOTIFY)
1736 #define trace_create_maxlat_file(tr, d_tracer) \
1737 trace_create_file("tracing_max_latency", 0644, d_tracer, \
1738 &tr->max_latency, &tracing_max_lat_fops)
1742 #ifdef CONFIG_TRACER_MAX_TRACE
1744 * Copy the new maximum trace into the separate maximum-trace
1745 * structure. (this way the maximum trace is permanently saved,
1746 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1749 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1751 struct array_buffer *trace_buf = &tr->array_buffer;
1752 struct array_buffer *max_buf = &tr->max_buffer;
1753 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1754 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1757 max_buf->time_start = data->preempt_timestamp;
1759 max_data->saved_latency = tr->max_latency;
1760 max_data->critical_start = data->critical_start;
1761 max_data->critical_end = data->critical_end;
1763 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1764 max_data->pid = tsk->pid;
1766 * If tsk == current, then use current_uid(), as that does not use
1767 * RCU. The irq tracer can be called out of RCU scope.
1770 max_data->uid = current_uid();
1772 max_data->uid = task_uid(tsk);
1774 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1775 max_data->policy = tsk->policy;
1776 max_data->rt_priority = tsk->rt_priority;
1778 /* record this tasks comm */
1779 tracing_record_cmdline(tsk);
1780 latency_fsnotify(tr);
1784 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1786 * @tsk: the task with the latency
1787 * @cpu: The cpu that initiated the trace.
1788 * @cond_data: User data associated with a conditional snapshot
1790 * Flip the buffers between the @tr and the max_tr and record information
1791 * about which task was the cause of this latency.
1794 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1800 WARN_ON_ONCE(!irqs_disabled());
1802 if (!tr->allocated_snapshot) {
1803 /* Only the nop tracer should hit this when disabling */
1804 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1808 arch_spin_lock(&tr->max_lock);
1810 /* Inherit the recordable setting from array_buffer */
1811 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1812 ring_buffer_record_on(tr->max_buffer.buffer);
1814 ring_buffer_record_off(tr->max_buffer.buffer);
1816 #ifdef CONFIG_TRACER_SNAPSHOT
1817 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data))
1820 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1822 __update_max_tr(tr, tsk, cpu);
1825 arch_spin_unlock(&tr->max_lock);
1829 * update_max_tr_single - only copy one trace over, and reset the rest
1831 * @tsk: task with the latency
1832 * @cpu: the cpu of the buffer to copy.
1834 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1837 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1844 WARN_ON_ONCE(!irqs_disabled());
1845 if (!tr->allocated_snapshot) {
1846 /* Only the nop tracer should hit this when disabling */
1847 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1851 arch_spin_lock(&tr->max_lock);
1853 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1855 if (ret == -EBUSY) {
1857 * We failed to swap the buffer due to a commit taking
1858 * place on this CPU. We fail to record, but we reset
1859 * the max trace buffer (no one writes directly to it)
1860 * and flag that it failed.
1862 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1863 "Failed to swap buffers due to commit in progress\n");
1866 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1868 __update_max_tr(tr, tsk, cpu);
1869 arch_spin_unlock(&tr->max_lock);
1871 #endif /* CONFIG_TRACER_MAX_TRACE */
1873 static int wait_on_pipe(struct trace_iterator *iter, int full)
1875 /* Iterators are static, they should be filled or empty */
1876 if (trace_buffer_iter(iter, iter->cpu_file))
1879 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1883 #ifdef CONFIG_FTRACE_STARTUP_TEST
1884 static bool selftests_can_run;
1886 struct trace_selftests {
1887 struct list_head list;
1888 struct tracer *type;
1891 static LIST_HEAD(postponed_selftests);
1893 static int save_selftest(struct tracer *type)
1895 struct trace_selftests *selftest;
1897 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1901 selftest->type = type;
1902 list_add(&selftest->list, &postponed_selftests);
1906 static int run_tracer_selftest(struct tracer *type)
1908 struct trace_array *tr = &global_trace;
1909 struct tracer *saved_tracer = tr->current_trace;
1912 if (!type->selftest || tracing_selftest_disabled)
1916 * If a tracer registers early in boot up (before scheduling is
1917 * initialized and such), then do not run its selftests yet.
1918 * Instead, run it a little later in the boot process.
1920 if (!selftests_can_run)
1921 return save_selftest(type);
1924 * Run a selftest on this tracer.
1925 * Here we reset the trace buffer, and set the current
1926 * tracer to be this tracer. The tracer can then run some
1927 * internal tracing to verify that everything is in order.
1928 * If we fail, we do not register this tracer.
1930 tracing_reset_online_cpus(&tr->array_buffer);
1932 tr->current_trace = type;
1934 #ifdef CONFIG_TRACER_MAX_TRACE
1935 if (type->use_max_tr) {
1936 /* If we expanded the buffers, make sure the max is expanded too */
1937 if (ring_buffer_expanded)
1938 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1939 RING_BUFFER_ALL_CPUS);
1940 tr->allocated_snapshot = true;
1944 /* the test is responsible for initializing and enabling */
1945 pr_info("Testing tracer %s: ", type->name);
1946 ret = type->selftest(type, tr);
1947 /* the test is responsible for resetting too */
1948 tr->current_trace = saved_tracer;
1950 printk(KERN_CONT "FAILED!\n");
1951 /* Add the warning after printing 'FAILED' */
1955 /* Only reset on passing, to avoid touching corrupted buffers */
1956 tracing_reset_online_cpus(&tr->array_buffer);
1958 #ifdef CONFIG_TRACER_MAX_TRACE
1959 if (type->use_max_tr) {
1960 tr->allocated_snapshot = false;
1962 /* Shrink the max buffer again */
1963 if (ring_buffer_expanded)
1964 ring_buffer_resize(tr->max_buffer.buffer, 1,
1965 RING_BUFFER_ALL_CPUS);
1969 printk(KERN_CONT "PASSED\n");
1973 static __init int init_trace_selftests(void)
1975 struct trace_selftests *p, *n;
1976 struct tracer *t, **last;
1979 selftests_can_run = true;
1981 mutex_lock(&trace_types_lock);
1983 if (list_empty(&postponed_selftests))
1986 pr_info("Running postponed tracer tests:\n");
1988 tracing_selftest_running = true;
1989 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
1990 /* This loop can take minutes when sanitizers are enabled, so
1991 * lets make sure we allow RCU processing.
1994 ret = run_tracer_selftest(p->type);
1995 /* If the test fails, then warn and remove from available_tracers */
1997 WARN(1, "tracer: %s failed selftest, disabling\n",
1999 last = &trace_types;
2000 for (t = trace_types; t; t = t->next) {
2011 tracing_selftest_running = false;
2014 mutex_unlock(&trace_types_lock);
2018 core_initcall(init_trace_selftests);
2020 static inline int run_tracer_selftest(struct tracer *type)
2024 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2026 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2028 static void __init apply_trace_boot_options(void);
2031 * register_tracer - register a tracer with the ftrace system.
2032 * @type: the plugin for the tracer
2034 * Register a new plugin tracer.
2036 int __init register_tracer(struct tracer *type)
2042 pr_info("Tracer must have a name\n");
2046 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2047 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2051 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2052 pr_warn("Can not register tracer %s due to lockdown\n",
2057 mutex_lock(&trace_types_lock);
2059 tracing_selftest_running = true;
2061 for (t = trace_types; t; t = t->next) {
2062 if (strcmp(type->name, t->name) == 0) {
2064 pr_info("Tracer %s already registered\n",
2071 if (!type->set_flag)
2072 type->set_flag = &dummy_set_flag;
2074 /*allocate a dummy tracer_flags*/
2075 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2080 type->flags->val = 0;
2081 type->flags->opts = dummy_tracer_opt;
2083 if (!type->flags->opts)
2084 type->flags->opts = dummy_tracer_opt;
2086 /* store the tracer for __set_tracer_option */
2087 type->flags->trace = type;
2089 ret = run_tracer_selftest(type);
2093 type->next = trace_types;
2095 add_tracer_options(&global_trace, type);
2098 tracing_selftest_running = false;
2099 mutex_unlock(&trace_types_lock);
2101 if (ret || !default_bootup_tracer)
2104 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2107 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2108 /* Do we want this tracer to start on bootup? */
2109 tracing_set_tracer(&global_trace, type->name);
2110 default_bootup_tracer = NULL;
2112 apply_trace_boot_options();
2114 /* disable other selftests, since this will break it. */
2115 tracing_selftest_disabled = true;
2116 #ifdef CONFIG_FTRACE_STARTUP_TEST
2117 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
2125 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2127 struct trace_buffer *buffer = buf->buffer;
2132 ring_buffer_record_disable(buffer);
2134 /* Make sure all commits have finished */
2136 ring_buffer_reset_cpu(buffer, cpu);
2138 ring_buffer_record_enable(buffer);
2141 void tracing_reset_online_cpus(struct array_buffer *buf)
2143 struct trace_buffer *buffer = buf->buffer;
2148 ring_buffer_record_disable(buffer);
2150 /* Make sure all commits have finished */
2153 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2155 ring_buffer_reset_online_cpus(buffer);
2157 ring_buffer_record_enable(buffer);
2160 /* Must have trace_types_lock held */
2161 void tracing_reset_all_online_cpus(void)
2163 struct trace_array *tr;
2165 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2166 if (!tr->clear_trace)
2168 tr->clear_trace = false;
2169 tracing_reset_online_cpus(&tr->array_buffer);
2170 #ifdef CONFIG_TRACER_MAX_TRACE
2171 tracing_reset_online_cpus(&tr->max_buffer);
2176 static int *tgid_map;
2178 #define SAVED_CMDLINES_DEFAULT 128
2179 #define NO_CMDLINE_MAP UINT_MAX
2180 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2181 struct saved_cmdlines_buffer {
2182 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2183 unsigned *map_cmdline_to_pid;
2184 unsigned cmdline_num;
2186 char *saved_cmdlines;
2188 static struct saved_cmdlines_buffer *savedcmd;
2190 /* temporary disable recording */
2191 static atomic_t trace_record_taskinfo_disabled __read_mostly;
2193 static inline char *get_saved_cmdlines(int idx)
2195 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2198 static inline void set_cmdline(int idx, const char *cmdline)
2200 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2203 static int allocate_cmdlines_buffer(unsigned int val,
2204 struct saved_cmdlines_buffer *s)
2206 s->map_cmdline_to_pid = kmalloc_array(val,
2207 sizeof(*s->map_cmdline_to_pid),
2209 if (!s->map_cmdline_to_pid)
2212 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2213 if (!s->saved_cmdlines) {
2214 kfree(s->map_cmdline_to_pid);
2219 s->cmdline_num = val;
2220 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2221 sizeof(s->map_pid_to_cmdline));
2222 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2223 val * sizeof(*s->map_cmdline_to_pid));
2228 static int trace_create_savedcmd(void)
2232 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2236 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2246 int is_tracing_stopped(void)
2248 return global_trace.stop_count;
2252 * tracing_start - quick start of the tracer
2254 * If tracing is enabled but was stopped by tracing_stop,
2255 * this will start the tracer back up.
2257 void tracing_start(void)
2259 struct trace_buffer *buffer;
2260 unsigned long flags;
2262 if (tracing_disabled)
2265 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2266 if (--global_trace.stop_count) {
2267 if (global_trace.stop_count < 0) {
2268 /* Someone screwed up their debugging */
2270 global_trace.stop_count = 0;
2275 /* Prevent the buffers from switching */
2276 arch_spin_lock(&global_trace.max_lock);
2278 buffer = global_trace.array_buffer.buffer;
2280 ring_buffer_record_enable(buffer);
2282 #ifdef CONFIG_TRACER_MAX_TRACE
2283 buffer = global_trace.max_buffer.buffer;
2285 ring_buffer_record_enable(buffer);
2288 arch_spin_unlock(&global_trace.max_lock);
2291 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2294 static void tracing_start_tr(struct trace_array *tr)
2296 struct trace_buffer *buffer;
2297 unsigned long flags;
2299 if (tracing_disabled)
2302 /* If global, we need to also start the max tracer */
2303 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2304 return tracing_start();
2306 raw_spin_lock_irqsave(&tr->start_lock, flags);
2308 if (--tr->stop_count) {
2309 if (tr->stop_count < 0) {
2310 /* Someone screwed up their debugging */
2317 buffer = tr->array_buffer.buffer;
2319 ring_buffer_record_enable(buffer);
2322 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2326 * tracing_stop - quick stop of the tracer
2328 * Light weight way to stop tracing. Use in conjunction with
2331 void tracing_stop(void)
2333 struct trace_buffer *buffer;
2334 unsigned long flags;
2336 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2337 if (global_trace.stop_count++)
2340 /* Prevent the buffers from switching */
2341 arch_spin_lock(&global_trace.max_lock);
2343 buffer = global_trace.array_buffer.buffer;
2345 ring_buffer_record_disable(buffer);
2347 #ifdef CONFIG_TRACER_MAX_TRACE
2348 buffer = global_trace.max_buffer.buffer;
2350 ring_buffer_record_disable(buffer);
2353 arch_spin_unlock(&global_trace.max_lock);
2356 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2359 static void tracing_stop_tr(struct trace_array *tr)
2361 struct trace_buffer *buffer;
2362 unsigned long flags;
2364 /* If global, we need to also stop the max tracer */
2365 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2366 return tracing_stop();
2368 raw_spin_lock_irqsave(&tr->start_lock, flags);
2369 if (tr->stop_count++)
2372 buffer = tr->array_buffer.buffer;
2374 ring_buffer_record_disable(buffer);
2377 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2380 static int trace_save_cmdline(struct task_struct *tsk)
2384 /* treat recording of idle task as a success */
2388 if (unlikely(tsk->pid > PID_MAX_DEFAULT))
2392 * It's not the end of the world if we don't get
2393 * the lock, but we also don't want to spin
2394 * nor do we want to disable interrupts,
2395 * so if we miss here, then better luck next time.
2397 if (!arch_spin_trylock(&trace_cmdline_lock))
2400 idx = savedcmd->map_pid_to_cmdline[tsk->pid];
2401 if (idx == NO_CMDLINE_MAP) {
2402 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2405 * Check whether the cmdline buffer at idx has a pid
2406 * mapped. We are going to overwrite that entry so we
2407 * need to clear the map_pid_to_cmdline. Otherwise we
2408 * would read the new comm for the old pid.
2410 pid = savedcmd->map_cmdline_to_pid[idx];
2411 if (pid != NO_CMDLINE_MAP)
2412 savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
2414 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2415 savedcmd->map_pid_to_cmdline[tsk->pid] = idx;
2417 savedcmd->cmdline_idx = idx;
2420 set_cmdline(idx, tsk->comm);
2422 arch_spin_unlock(&trace_cmdline_lock);
2427 static void __trace_find_cmdline(int pid, char comm[])
2432 strcpy(comm, "<idle>");
2436 if (WARN_ON_ONCE(pid < 0)) {
2437 strcpy(comm, "<XXX>");
2441 if (pid > PID_MAX_DEFAULT) {
2442 strcpy(comm, "<...>");
2446 map = savedcmd->map_pid_to_cmdline[pid];
2447 if (map != NO_CMDLINE_MAP)
2448 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2450 strcpy(comm, "<...>");
2453 void trace_find_cmdline(int pid, char comm[])
2456 arch_spin_lock(&trace_cmdline_lock);
2458 __trace_find_cmdline(pid, comm);
2460 arch_spin_unlock(&trace_cmdline_lock);
2464 int trace_find_tgid(int pid)
2466 if (unlikely(!tgid_map || !pid || pid > PID_MAX_DEFAULT))
2469 return tgid_map[pid];
2472 static int trace_save_tgid(struct task_struct *tsk)
2474 /* treat recording of idle task as a success */
2478 if (unlikely(!tgid_map || tsk->pid > PID_MAX_DEFAULT))
2481 tgid_map[tsk->pid] = tsk->tgid;
2485 static bool tracing_record_taskinfo_skip(int flags)
2487 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2489 if (atomic_read(&trace_record_taskinfo_disabled) || !tracing_is_on())
2491 if (!__this_cpu_read(trace_taskinfo_save))
2497 * tracing_record_taskinfo - record the task info of a task
2499 * @task: task to record
2500 * @flags: TRACE_RECORD_CMDLINE for recording comm
2501 * TRACE_RECORD_TGID for recording tgid
2503 void tracing_record_taskinfo(struct task_struct *task, int flags)
2507 if (tracing_record_taskinfo_skip(flags))
2511 * Record as much task information as possible. If some fail, continue
2512 * to try to record the others.
2514 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2515 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2517 /* If recording any information failed, retry again soon. */
2521 __this_cpu_write(trace_taskinfo_save, false);
2525 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2527 * @prev: previous task during sched_switch
2528 * @next: next task during sched_switch
2529 * @flags: TRACE_RECORD_CMDLINE for recording comm
2530 * TRACE_RECORD_TGID for recording tgid
2532 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2533 struct task_struct *next, int flags)
2537 if (tracing_record_taskinfo_skip(flags))
2541 * Record as much task information as possible. If some fail, continue
2542 * to try to record the others.
2544 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2545 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2546 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2547 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2549 /* If recording any information failed, retry again soon. */
2553 __this_cpu_write(trace_taskinfo_save, false);
2556 /* Helpers to record a specific task information */
2557 void tracing_record_cmdline(struct task_struct *task)
2559 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2562 void tracing_record_tgid(struct task_struct *task)
2564 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2568 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2569 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2570 * simplifies those functions and keeps them in sync.
2572 enum print_line_t trace_handle_return(struct trace_seq *s)
2574 return trace_seq_has_overflowed(s) ?
2575 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2577 EXPORT_SYMBOL_GPL(trace_handle_return);
2580 tracing_generic_entry_update(struct trace_entry *entry, unsigned short type,
2581 unsigned long flags, int pc)
2583 struct task_struct *tsk = current;
2585 entry->preempt_count = pc & 0xff;
2586 entry->pid = (tsk) ? tsk->pid : 0;
2589 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
2590 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
2592 TRACE_FLAG_IRQS_NOSUPPORT |
2594 ((pc & NMI_MASK ) ? TRACE_FLAG_NMI : 0) |
2595 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
2596 ((pc & SOFTIRQ_OFFSET) ? TRACE_FLAG_SOFTIRQ : 0) |
2597 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
2598 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
2600 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
2602 struct ring_buffer_event *
2603 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2606 unsigned long flags, int pc)
2608 return __trace_buffer_lock_reserve(buffer, type, len, flags, pc);
2611 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2612 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2613 static int trace_buffered_event_ref;
2616 * trace_buffered_event_enable - enable buffering events
2618 * When events are being filtered, it is quicker to use a temporary
2619 * buffer to write the event data into if there's a likely chance
2620 * that it will not be committed. The discard of the ring buffer
2621 * is not as fast as committing, and is much slower than copying
2624 * When an event is to be filtered, allocate per cpu buffers to
2625 * write the event data into, and if the event is filtered and discarded
2626 * it is simply dropped, otherwise, the entire data is to be committed
2629 void trace_buffered_event_enable(void)
2631 struct ring_buffer_event *event;
2635 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2637 if (trace_buffered_event_ref++)
2640 for_each_tracing_cpu(cpu) {
2641 page = alloc_pages_node(cpu_to_node(cpu),
2642 GFP_KERNEL | __GFP_NORETRY, 0);
2646 event = page_address(page);
2647 memset(event, 0, sizeof(*event));
2649 per_cpu(trace_buffered_event, cpu) = event;
2652 if (cpu == smp_processor_id() &&
2653 this_cpu_read(trace_buffered_event) !=
2654 per_cpu(trace_buffered_event, cpu))
2661 trace_buffered_event_disable();
2664 static void enable_trace_buffered_event(void *data)
2666 /* Probably not needed, but do it anyway */
2668 this_cpu_dec(trace_buffered_event_cnt);
2671 static void disable_trace_buffered_event(void *data)
2673 this_cpu_inc(trace_buffered_event_cnt);
2677 * trace_buffered_event_disable - disable buffering events
2679 * When a filter is removed, it is faster to not use the buffered
2680 * events, and to commit directly into the ring buffer. Free up
2681 * the temp buffers when there are no more users. This requires
2682 * special synchronization with current events.
2684 void trace_buffered_event_disable(void)
2688 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2690 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2693 if (--trace_buffered_event_ref)
2697 /* For each CPU, set the buffer as used. */
2698 smp_call_function_many(tracing_buffer_mask,
2699 disable_trace_buffered_event, NULL, 1);
2702 /* Wait for all current users to finish */
2705 for_each_tracing_cpu(cpu) {
2706 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2707 per_cpu(trace_buffered_event, cpu) = NULL;
2710 * Make sure trace_buffered_event is NULL before clearing
2711 * trace_buffered_event_cnt.
2716 /* Do the work on each cpu */
2717 smp_call_function_many(tracing_buffer_mask,
2718 enable_trace_buffered_event, NULL, 1);
2722 static struct trace_buffer *temp_buffer;
2724 struct ring_buffer_event *
2725 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2726 struct trace_event_file *trace_file,
2727 int type, unsigned long len,
2728 unsigned long flags, int pc)
2730 struct ring_buffer_event *entry;
2733 *current_rb = trace_file->tr->array_buffer.buffer;
2735 if (!ring_buffer_time_stamp_abs(*current_rb) && (trace_file->flags &
2736 (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED)) &&
2737 (entry = this_cpu_read(trace_buffered_event))) {
2738 /* Try to use the per cpu buffer first */
2739 val = this_cpu_inc_return(trace_buffered_event_cnt);
2741 trace_event_setup(entry, type, flags, pc);
2742 entry->array[0] = len;
2745 this_cpu_dec(trace_buffered_event_cnt);
2748 entry = __trace_buffer_lock_reserve(*current_rb,
2749 type, len, flags, pc);
2751 * If tracing is off, but we have triggers enabled
2752 * we still need to look at the event data. Use the temp_buffer
2753 * to store the trace event for the tigger to use. It's recusive
2754 * safe and will not be recorded anywhere.
2756 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2757 *current_rb = temp_buffer;
2758 entry = __trace_buffer_lock_reserve(*current_rb,
2759 type, len, flags, pc);
2763 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2765 static DEFINE_SPINLOCK(tracepoint_iter_lock);
2766 static DEFINE_MUTEX(tracepoint_printk_mutex);
2768 static void output_printk(struct trace_event_buffer *fbuffer)
2770 struct trace_event_call *event_call;
2771 struct trace_event_file *file;
2772 struct trace_event *event;
2773 unsigned long flags;
2774 struct trace_iterator *iter = tracepoint_print_iter;
2776 /* We should never get here if iter is NULL */
2777 if (WARN_ON_ONCE(!iter))
2780 event_call = fbuffer->trace_file->event_call;
2781 if (!event_call || !event_call->event.funcs ||
2782 !event_call->event.funcs->trace)
2785 file = fbuffer->trace_file;
2786 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2787 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2788 !filter_match_preds(file->filter, fbuffer->entry)))
2791 event = &fbuffer->trace_file->event_call->event;
2793 spin_lock_irqsave(&tracepoint_iter_lock, flags);
2794 trace_seq_init(&iter->seq);
2795 iter->ent = fbuffer->entry;
2796 event_call->event.funcs->trace(iter, 0, event);
2797 trace_seq_putc(&iter->seq, 0);
2798 printk("%s", iter->seq.buffer);
2800 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2803 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2804 void *buffer, size_t *lenp,
2807 int save_tracepoint_printk;
2810 mutex_lock(&tracepoint_printk_mutex);
2811 save_tracepoint_printk = tracepoint_printk;
2813 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2816 * This will force exiting early, as tracepoint_printk
2817 * is always zero when tracepoint_printk_iter is not allocated
2819 if (!tracepoint_print_iter)
2820 tracepoint_printk = 0;
2822 if (save_tracepoint_printk == tracepoint_printk)
2825 if (tracepoint_printk)
2826 static_key_enable(&tracepoint_printk_key.key);
2828 static_key_disable(&tracepoint_printk_key.key);
2831 mutex_unlock(&tracepoint_printk_mutex);
2836 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2838 if (static_key_false(&tracepoint_printk_key.key))
2839 output_printk(fbuffer);
2841 if (static_branch_unlikely(&trace_event_exports_enabled))
2842 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2843 event_trigger_unlock_commit_regs(fbuffer->trace_file, fbuffer->buffer,
2844 fbuffer->event, fbuffer->entry,
2845 fbuffer->flags, fbuffer->pc, fbuffer->regs);
2847 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2852 * trace_buffer_unlock_commit_regs()
2853 * trace_event_buffer_commit()
2854 * trace_event_raw_event_xxx()
2856 # define STACK_SKIP 3
2858 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2859 struct trace_buffer *buffer,
2860 struct ring_buffer_event *event,
2861 unsigned long flags, int pc,
2862 struct pt_regs *regs)
2864 __buffer_unlock_commit(buffer, event);
2867 * If regs is not set, then skip the necessary functions.
2868 * Note, we can still get here via blktrace, wakeup tracer
2869 * and mmiotrace, but that's ok if they lose a function or
2870 * two. They are not that meaningful.
2872 ftrace_trace_stack(tr, buffer, flags, regs ? 0 : STACK_SKIP, pc, regs);
2873 ftrace_trace_userstack(buffer, flags, pc);
2877 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2880 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2881 struct ring_buffer_event *event)
2883 __buffer_unlock_commit(buffer, event);
2887 trace_function(struct trace_array *tr,
2888 unsigned long ip, unsigned long parent_ip, unsigned long flags,
2891 struct trace_event_call *call = &event_function;
2892 struct trace_buffer *buffer = tr->array_buffer.buffer;
2893 struct ring_buffer_event *event;
2894 struct ftrace_entry *entry;
2896 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2900 entry = ring_buffer_event_data(event);
2902 entry->parent_ip = parent_ip;
2904 if (!call_filter_check_discard(call, entry, buffer, event)) {
2905 if (static_branch_unlikely(&trace_function_exports_enabled))
2906 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
2907 __buffer_unlock_commit(buffer, event);
2911 #ifdef CONFIG_STACKTRACE
2913 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2914 #define FTRACE_KSTACK_NESTING 4
2916 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
2918 struct ftrace_stack {
2919 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2923 struct ftrace_stacks {
2924 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
2927 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
2928 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
2930 static void __ftrace_trace_stack(struct trace_buffer *buffer,
2931 unsigned long flags,
2932 int skip, int pc, struct pt_regs *regs)
2934 struct trace_event_call *call = &event_kernel_stack;
2935 struct ring_buffer_event *event;
2936 unsigned int size, nr_entries;
2937 struct ftrace_stack *fstack;
2938 struct stack_entry *entry;
2942 * Add one, for this function and the call to save_stack_trace()
2943 * If regs is set, then these functions will not be in the way.
2945 #ifndef CONFIG_UNWINDER_ORC
2950 preempt_disable_notrace();
2952 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
2954 /* This should never happen. If it does, yell once and skip */
2955 if (WARN_ON_ONCE(stackidx > FTRACE_KSTACK_NESTING))
2959 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
2960 * interrupt will either see the value pre increment or post
2961 * increment. If the interrupt happens pre increment it will have
2962 * restored the counter when it returns. We just need a barrier to
2963 * keep gcc from moving things around.
2967 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
2968 size = ARRAY_SIZE(fstack->calls);
2971 nr_entries = stack_trace_save_regs(regs, fstack->calls,
2974 nr_entries = stack_trace_save(fstack->calls, size, skip);
2977 size = nr_entries * sizeof(unsigned long);
2978 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
2979 sizeof(*entry) + size, flags, pc);
2982 entry = ring_buffer_event_data(event);
2984 memcpy(&entry->caller, fstack->calls, size);
2985 entry->size = nr_entries;
2987 if (!call_filter_check_discard(call, entry, buffer, event))
2988 __buffer_unlock_commit(buffer, event);
2991 /* Again, don't let gcc optimize things here */
2993 __this_cpu_dec(ftrace_stack_reserve);
2994 preempt_enable_notrace();
2998 static inline void ftrace_trace_stack(struct trace_array *tr,
2999 struct trace_buffer *buffer,
3000 unsigned long flags,
3001 int skip, int pc, struct pt_regs *regs)
3003 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3006 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
3009 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
3012 struct trace_buffer *buffer = tr->array_buffer.buffer;
3014 if (rcu_is_watching()) {
3015 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
3020 * When an NMI triggers, RCU is enabled via rcu_nmi_enter(),
3021 * but if the above rcu_is_watching() failed, then the NMI
3022 * triggered someplace critical, and rcu_irq_enter() should
3023 * not be called from NMI.
3025 if (unlikely(in_nmi()))
3028 rcu_irq_enter_irqson();
3029 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
3030 rcu_irq_exit_irqson();
3034 * trace_dump_stack - record a stack back trace in the trace buffer
3035 * @skip: Number of functions to skip (helper handlers)
3037 void trace_dump_stack(int skip)
3039 unsigned long flags;
3041 if (tracing_disabled || tracing_selftest_running)
3044 local_save_flags(flags);
3046 #ifndef CONFIG_UNWINDER_ORC
3047 /* Skip 1 to skip this function. */
3050 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3051 flags, skip, preempt_count(), NULL);
3053 EXPORT_SYMBOL_GPL(trace_dump_stack);
3055 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3056 static DEFINE_PER_CPU(int, user_stack_count);
3059 ftrace_trace_userstack(struct trace_buffer *buffer, unsigned long flags, int pc)
3061 struct trace_event_call *call = &event_user_stack;
3062 struct ring_buffer_event *event;
3063 struct userstack_entry *entry;
3065 if (!(global_trace.trace_flags & TRACE_ITER_USERSTACKTRACE))
3069 * NMIs can not handle page faults, even with fix ups.
3070 * The save user stack can (and often does) fault.
3072 if (unlikely(in_nmi()))
3076 * prevent recursion, since the user stack tracing may
3077 * trigger other kernel events.
3080 if (__this_cpu_read(user_stack_count))
3083 __this_cpu_inc(user_stack_count);
3085 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3086 sizeof(*entry), flags, pc);
3088 goto out_drop_count;
3089 entry = ring_buffer_event_data(event);
3091 entry->tgid = current->tgid;
3092 memset(&entry->caller, 0, sizeof(entry->caller));
3094 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3095 if (!call_filter_check_discard(call, entry, buffer, event))
3096 __buffer_unlock_commit(buffer, event);
3099 __this_cpu_dec(user_stack_count);
3103 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3104 static void ftrace_trace_userstack(struct trace_buffer *buffer,
3105 unsigned long flags, int pc)
3108 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3110 #endif /* CONFIG_STACKTRACE */
3112 /* created for use with alloc_percpu */
3113 struct trace_buffer_struct {
3115 char buffer[4][TRACE_BUF_SIZE];
3118 static struct trace_buffer_struct *trace_percpu_buffer;
3121 * Thise allows for lockless recording. If we're nested too deeply, then
3122 * this returns NULL.
3124 static char *get_trace_buf(void)
3126 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3128 if (!buffer || buffer->nesting >= 4)
3133 /* Interrupts must see nesting incremented before we use the buffer */
3135 return &buffer->buffer[buffer->nesting][0];
3138 static void put_trace_buf(void)
3140 /* Don't let the decrement of nesting leak before this */
3142 this_cpu_dec(trace_percpu_buffer->nesting);
3145 static int alloc_percpu_trace_buffer(void)
3147 struct trace_buffer_struct *buffers;
3149 if (trace_percpu_buffer)
3152 buffers = alloc_percpu(struct trace_buffer_struct);
3153 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3156 trace_percpu_buffer = buffers;
3160 static int buffers_allocated;
3162 void trace_printk_init_buffers(void)
3164 if (buffers_allocated)
3167 if (alloc_percpu_trace_buffer())
3170 /* trace_printk() is for debug use only. Don't use it in production. */
3173 pr_warn("**********************************************************\n");
3174 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3176 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3178 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3179 pr_warn("** unsafe for production use. **\n");
3181 pr_warn("** If you see this message and you are not debugging **\n");
3182 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3184 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3185 pr_warn("**********************************************************\n");
3187 /* Expand the buffers to set size */
3188 tracing_update_buffers();
3190 buffers_allocated = 1;
3193 * trace_printk_init_buffers() can be called by modules.
3194 * If that happens, then we need to start cmdline recording
3195 * directly here. If the global_trace.buffer is already
3196 * allocated here, then this was called by module code.
3198 if (global_trace.array_buffer.buffer)
3199 tracing_start_cmdline_record();
3201 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3203 void trace_printk_start_comm(void)
3205 /* Start tracing comms if trace printk is set */
3206 if (!buffers_allocated)
3208 tracing_start_cmdline_record();
3211 static void trace_printk_start_stop_comm(int enabled)
3213 if (!buffers_allocated)
3217 tracing_start_cmdline_record();
3219 tracing_stop_cmdline_record();
3223 * trace_vbprintk - write binary msg to tracing buffer
3224 * @ip: The address of the caller
3225 * @fmt: The string format to write to the buffer
3226 * @args: Arguments for @fmt
3228 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3230 struct trace_event_call *call = &event_bprint;
3231 struct ring_buffer_event *event;
3232 struct trace_buffer *buffer;
3233 struct trace_array *tr = &global_trace;
3234 struct bprint_entry *entry;
3235 unsigned long flags;
3237 int len = 0, size, pc;
3239 if (unlikely(tracing_selftest_running || tracing_disabled))
3242 /* Don't pollute graph traces with trace_vprintk internals */
3243 pause_graph_tracing();
3245 pc = preempt_count();
3246 preempt_disable_notrace();
3248 tbuffer = get_trace_buf();
3254 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3256 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3259 local_save_flags(flags);
3260 size = sizeof(*entry) + sizeof(u32) * len;
3261 buffer = tr->array_buffer.buffer;
3262 ring_buffer_nest_start(buffer);
3263 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3267 entry = ring_buffer_event_data(event);
3271 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3272 if (!call_filter_check_discard(call, entry, buffer, event)) {
3273 __buffer_unlock_commit(buffer, event);
3274 ftrace_trace_stack(tr, buffer, flags, 6, pc, NULL);
3278 ring_buffer_nest_end(buffer);
3283 preempt_enable_notrace();
3284 unpause_graph_tracing();
3288 EXPORT_SYMBOL_GPL(trace_vbprintk);
3292 __trace_array_vprintk(struct trace_buffer *buffer,
3293 unsigned long ip, const char *fmt, va_list args)
3295 struct trace_event_call *call = &event_print;
3296 struct ring_buffer_event *event;
3297 int len = 0, size, pc;
3298 struct print_entry *entry;
3299 unsigned long flags;
3302 if (tracing_disabled || tracing_selftest_running)
3305 /* Don't pollute graph traces with trace_vprintk internals */
3306 pause_graph_tracing();
3308 pc = preempt_count();
3309 preempt_disable_notrace();
3312 tbuffer = get_trace_buf();
3318 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3320 local_save_flags(flags);
3321 size = sizeof(*entry) + len + 1;
3322 ring_buffer_nest_start(buffer);
3323 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3327 entry = ring_buffer_event_data(event);
3330 memcpy(&entry->buf, tbuffer, len + 1);
3331 if (!call_filter_check_discard(call, entry, buffer, event)) {
3332 __buffer_unlock_commit(buffer, event);
3333 ftrace_trace_stack(&global_trace, buffer, flags, 6, pc, NULL);
3337 ring_buffer_nest_end(buffer);
3341 preempt_enable_notrace();
3342 unpause_graph_tracing();
3348 int trace_array_vprintk(struct trace_array *tr,
3349 unsigned long ip, const char *fmt, va_list args)
3351 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3355 * trace_array_printk - Print a message to a specific instance
3356 * @tr: The instance trace_array descriptor
3357 * @ip: The instruction pointer that this is called from.
3358 * @fmt: The format to print (printf format)
3360 * If a subsystem sets up its own instance, they have the right to
3361 * printk strings into their tracing instance buffer using this
3362 * function. Note, this function will not write into the top level
3363 * buffer (use trace_printk() for that), as writing into the top level
3364 * buffer should only have events that can be individually disabled.
3365 * trace_printk() is only used for debugging a kernel, and should not
3366 * be ever encorporated in normal use.
3368 * trace_array_printk() can be used, as it will not add noise to the
3369 * top level tracing buffer.
3371 * Note, trace_array_init_printk() must be called on @tr before this
3375 int trace_array_printk(struct trace_array *tr,
3376 unsigned long ip, const char *fmt, ...)
3384 /* This is only allowed for created instances */
3385 if (tr == &global_trace)
3388 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3392 ret = trace_array_vprintk(tr, ip, fmt, ap);
3396 EXPORT_SYMBOL_GPL(trace_array_printk);
3399 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3400 * @tr: The trace array to initialize the buffers for
3402 * As trace_array_printk() only writes into instances, they are OK to
3403 * have in the kernel (unlike trace_printk()). This needs to be called
3404 * before trace_array_printk() can be used on a trace_array.
3406 int trace_array_init_printk(struct trace_array *tr)
3411 /* This is only allowed for created instances */
3412 if (tr == &global_trace)
3415 return alloc_percpu_trace_buffer();
3417 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3420 int trace_array_printk_buf(struct trace_buffer *buffer,
3421 unsigned long ip, const char *fmt, ...)
3426 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3430 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3436 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3438 return trace_array_vprintk(&global_trace, ip, fmt, args);
3440 EXPORT_SYMBOL_GPL(trace_vprintk);
3442 static void trace_iterator_increment(struct trace_iterator *iter)
3444 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3448 ring_buffer_iter_advance(buf_iter);
3451 static struct trace_entry *
3452 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3453 unsigned long *lost_events)
3455 struct ring_buffer_event *event;
3456 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3459 event = ring_buffer_iter_peek(buf_iter, ts);
3461 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3462 (unsigned long)-1 : 0;
3464 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3469 iter->ent_size = ring_buffer_event_length(event);
3470 return ring_buffer_event_data(event);
3476 static struct trace_entry *
3477 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3478 unsigned long *missing_events, u64 *ent_ts)
3480 struct trace_buffer *buffer = iter->array_buffer->buffer;
3481 struct trace_entry *ent, *next = NULL;
3482 unsigned long lost_events = 0, next_lost = 0;
3483 int cpu_file = iter->cpu_file;
3484 u64 next_ts = 0, ts;
3490 * If we are in a per_cpu trace file, don't bother by iterating over
3491 * all cpu and peek directly.
3493 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3494 if (ring_buffer_empty_cpu(buffer, cpu_file))
3496 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3498 *ent_cpu = cpu_file;
3503 for_each_tracing_cpu(cpu) {
3505 if (ring_buffer_empty_cpu(buffer, cpu))
3508 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3511 * Pick the entry with the smallest timestamp:
3513 if (ent && (!next || ts < next_ts)) {
3517 next_lost = lost_events;
3518 next_size = iter->ent_size;
3522 iter->ent_size = next_size;
3525 *ent_cpu = next_cpu;
3531 *missing_events = next_lost;
3536 #define STATIC_TEMP_BUF_SIZE 128
3537 static char static_temp_buf[STATIC_TEMP_BUF_SIZE];
3539 /* Find the next real entry, without updating the iterator itself */
3540 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3541 int *ent_cpu, u64 *ent_ts)
3543 /* __find_next_entry will reset ent_size */
3544 int ent_size = iter->ent_size;
3545 struct trace_entry *entry;
3548 * If called from ftrace_dump(), then the iter->temp buffer
3549 * will be the static_temp_buf and not created from kmalloc.
3550 * If the entry size is greater than the buffer, we can
3551 * not save it. Just return NULL in that case. This is only
3552 * used to add markers when two consecutive events' time
3553 * stamps have a large delta. See trace_print_lat_context()
3555 if (iter->temp == static_temp_buf &&
3556 STATIC_TEMP_BUF_SIZE < ent_size)
3560 * The __find_next_entry() may call peek_next_entry(), which may
3561 * call ring_buffer_peek() that may make the contents of iter->ent
3562 * undefined. Need to copy iter->ent now.
3564 if (iter->ent && iter->ent != iter->temp) {
3565 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3566 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3568 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3573 iter->temp_size = iter->ent_size;
3575 memcpy(iter->temp, iter->ent, iter->ent_size);
3576 iter->ent = iter->temp;
3578 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3579 /* Put back the original ent_size */
3580 iter->ent_size = ent_size;
3585 /* Find the next real entry, and increment the iterator to the next entry */
3586 void *trace_find_next_entry_inc(struct trace_iterator *iter)
3588 iter->ent = __find_next_entry(iter, &iter->cpu,
3589 &iter->lost_events, &iter->ts);
3592 trace_iterator_increment(iter);
3594 return iter->ent ? iter : NULL;
3597 static void trace_consume(struct trace_iterator *iter)
3599 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3600 &iter->lost_events);
3603 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3605 struct trace_iterator *iter = m->private;
3609 WARN_ON_ONCE(iter->leftover);
3613 /* can't go backwards */
3618 ent = trace_find_next_entry_inc(iter);
3622 while (ent && iter->idx < i)
3623 ent = trace_find_next_entry_inc(iter);
3630 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
3632 struct ring_buffer_iter *buf_iter;
3633 unsigned long entries = 0;
3636 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
3638 buf_iter = trace_buffer_iter(iter, cpu);
3642 ring_buffer_iter_reset(buf_iter);
3645 * We could have the case with the max latency tracers
3646 * that a reset never took place on a cpu. This is evident
3647 * by the timestamp being before the start of the buffer.
3649 while (ring_buffer_iter_peek(buf_iter, &ts)) {
3650 if (ts >= iter->array_buffer->time_start)
3653 ring_buffer_iter_advance(buf_iter);
3656 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
3660 * The current tracer is copied to avoid a global locking
3663 static void *s_start(struct seq_file *m, loff_t *pos)
3665 struct trace_iterator *iter = m->private;
3666 struct trace_array *tr = iter->tr;
3667 int cpu_file = iter->cpu_file;
3673 * copy the tracer to avoid using a global lock all around.
3674 * iter->trace is a copy of current_trace, the pointer to the
3675 * name may be used instead of a strcmp(), as iter->trace->name
3676 * will point to the same string as current_trace->name.
3678 mutex_lock(&trace_types_lock);
3679 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
3680 *iter->trace = *tr->current_trace;
3681 mutex_unlock(&trace_types_lock);
3683 #ifdef CONFIG_TRACER_MAX_TRACE
3684 if (iter->snapshot && iter->trace->use_max_tr)
3685 return ERR_PTR(-EBUSY);
3688 if (!iter->snapshot)
3689 atomic_inc(&trace_record_taskinfo_disabled);
3691 if (*pos != iter->pos) {
3696 if (cpu_file == RING_BUFFER_ALL_CPUS) {
3697 for_each_tracing_cpu(cpu)
3698 tracing_iter_reset(iter, cpu);
3700 tracing_iter_reset(iter, cpu_file);
3703 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
3708 * If we overflowed the seq_file before, then we want
3709 * to just reuse the trace_seq buffer again.
3715 p = s_next(m, p, &l);
3719 trace_event_read_lock();
3720 trace_access_lock(cpu_file);
3724 static void s_stop(struct seq_file *m, void *p)
3726 struct trace_iterator *iter = m->private;
3728 #ifdef CONFIG_TRACER_MAX_TRACE
3729 if (iter->snapshot && iter->trace->use_max_tr)
3733 if (!iter->snapshot)
3734 atomic_dec(&trace_record_taskinfo_disabled);
3736 trace_access_unlock(iter->cpu_file);
3737 trace_event_read_unlock();
3741 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
3742 unsigned long *entries, int cpu)
3744 unsigned long count;
3746 count = ring_buffer_entries_cpu(buf->buffer, cpu);
3748 * If this buffer has skipped entries, then we hold all
3749 * entries for the trace and we need to ignore the
3750 * ones before the time stamp.
3752 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
3753 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
3754 /* total is the same as the entries */
3758 ring_buffer_overrun_cpu(buf->buffer, cpu);
3763 get_total_entries(struct array_buffer *buf,
3764 unsigned long *total, unsigned long *entries)
3772 for_each_tracing_cpu(cpu) {
3773 get_total_entries_cpu(buf, &t, &e, cpu);
3779 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
3781 unsigned long total, entries;
3786 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
3791 unsigned long trace_total_entries(struct trace_array *tr)
3793 unsigned long total, entries;
3798 get_total_entries(&tr->array_buffer, &total, &entries);
3803 static void print_lat_help_header(struct seq_file *m)
3805 seq_puts(m, "# _------=> CPU# \n"
3806 "# / _-----=> irqs-off \n"
3807 "# | / _----=> need-resched \n"
3808 "# || / _---=> hardirq/softirq \n"
3809 "# ||| / _--=> preempt-depth \n"
3811 "# cmd pid ||||| time | caller \n"
3812 "# \\ / ||||| \\ | / \n");
3815 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
3817 unsigned long total;
3818 unsigned long entries;
3820 get_total_entries(buf, &total, &entries);
3821 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
3822 entries, total, num_online_cpus());
3826 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
3829 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3831 print_event_info(buf, m);
3833 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
3834 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
3837 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
3840 bool tgid = flags & TRACE_ITER_RECORD_TGID;
3841 const char *space = " ";
3842 int prec = tgid ? 12 : 2;
3844 print_event_info(buf, m);
3846 seq_printf(m, "# %.*s _-----=> irqs-off\n", prec, space);
3847 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
3848 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
3849 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
3850 seq_printf(m, "# %.*s||| / delay\n", prec, space);
3851 seq_printf(m, "# TASK-PID %.*s CPU# |||| TIMESTAMP FUNCTION\n", prec, " TGID ");
3852 seq_printf(m, "# | | %.*s | |||| | |\n", prec, " | ");
3856 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
3858 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
3859 struct array_buffer *buf = iter->array_buffer;
3860 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
3861 struct tracer *type = iter->trace;
3862 unsigned long entries;
3863 unsigned long total;
3864 const char *name = "preemption";
3868 get_total_entries(buf, &total, &entries);
3870 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
3872 seq_puts(m, "# -----------------------------------"
3873 "---------------------------------\n");
3874 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
3875 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
3876 nsecs_to_usecs(data->saved_latency),
3880 #if defined(CONFIG_PREEMPT_NONE)
3882 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
3884 #elif defined(CONFIG_PREEMPT)
3886 #elif defined(CONFIG_PREEMPT_RT)
3891 /* These are reserved for later use */
3894 seq_printf(m, " #P:%d)\n", num_online_cpus());
3898 seq_puts(m, "# -----------------\n");
3899 seq_printf(m, "# | task: %.16s-%d "
3900 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
3901 data->comm, data->pid,
3902 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
3903 data->policy, data->rt_priority);
3904 seq_puts(m, "# -----------------\n");
3906 if (data->critical_start) {
3907 seq_puts(m, "# => started at: ");
3908 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
3909 trace_print_seq(m, &iter->seq);
3910 seq_puts(m, "\n# => ended at: ");
3911 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
3912 trace_print_seq(m, &iter->seq);
3913 seq_puts(m, "\n#\n");
3919 static void test_cpu_buff_start(struct trace_iterator *iter)
3921 struct trace_seq *s = &iter->seq;
3922 struct trace_array *tr = iter->tr;
3924 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
3927 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
3930 if (cpumask_available(iter->started) &&
3931 cpumask_test_cpu(iter->cpu, iter->started))
3934 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
3937 if (cpumask_available(iter->started))
3938 cpumask_set_cpu(iter->cpu, iter->started);
3940 /* Don't print started cpu buffer for the first entry of the trace */
3942 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
3946 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
3948 struct trace_array *tr = iter->tr;
3949 struct trace_seq *s = &iter->seq;
3950 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
3951 struct trace_entry *entry;
3952 struct trace_event *event;
3956 test_cpu_buff_start(iter);
3958 event = ftrace_find_event(entry->type);
3960 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
3961 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
3962 trace_print_lat_context(iter);
3964 trace_print_context(iter);
3967 if (trace_seq_has_overflowed(s))
3968 return TRACE_TYPE_PARTIAL_LINE;
3971 return event->funcs->trace(iter, sym_flags, event);
3973 trace_seq_printf(s, "Unknown type %d\n", entry->type);
3975 return trace_handle_return(s);
3978 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
3980 struct trace_array *tr = iter->tr;
3981 struct trace_seq *s = &iter->seq;
3982 struct trace_entry *entry;
3983 struct trace_event *event;
3987 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
3988 trace_seq_printf(s, "%d %d %llu ",
3989 entry->pid, iter->cpu, iter->ts);
3991 if (trace_seq_has_overflowed(s))
3992 return TRACE_TYPE_PARTIAL_LINE;
3994 event = ftrace_find_event(entry->type);
3996 return event->funcs->raw(iter, 0, event);
3998 trace_seq_printf(s, "%d ?\n", entry->type);
4000 return trace_handle_return(s);
4003 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4005 struct trace_array *tr = iter->tr;
4006 struct trace_seq *s = &iter->seq;
4007 unsigned char newline = '\n';
4008 struct trace_entry *entry;
4009 struct trace_event *event;
4013 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4014 SEQ_PUT_HEX_FIELD(s, entry->pid);
4015 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4016 SEQ_PUT_HEX_FIELD(s, iter->ts);
4017 if (trace_seq_has_overflowed(s))
4018 return TRACE_TYPE_PARTIAL_LINE;
4021 event = ftrace_find_event(entry->type);
4023 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4024 if (ret != TRACE_TYPE_HANDLED)
4028 SEQ_PUT_FIELD(s, newline);
4030 return trace_handle_return(s);
4033 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4035 struct trace_array *tr = iter->tr;
4036 struct trace_seq *s = &iter->seq;
4037 struct trace_entry *entry;
4038 struct trace_event *event;
4042 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4043 SEQ_PUT_FIELD(s, entry->pid);
4044 SEQ_PUT_FIELD(s, iter->cpu);
4045 SEQ_PUT_FIELD(s, iter->ts);
4046 if (trace_seq_has_overflowed(s))
4047 return TRACE_TYPE_PARTIAL_LINE;
4050 event = ftrace_find_event(entry->type);
4051 return event ? event->funcs->binary(iter, 0, event) :
4055 int trace_empty(struct trace_iterator *iter)
4057 struct ring_buffer_iter *buf_iter;
4060 /* If we are looking at one CPU buffer, only check that one */
4061 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4062 cpu = iter->cpu_file;
4063 buf_iter = trace_buffer_iter(iter, cpu);
4065 if (!ring_buffer_iter_empty(buf_iter))
4068 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4074 for_each_tracing_cpu(cpu) {
4075 buf_iter = trace_buffer_iter(iter, cpu);
4077 if (!ring_buffer_iter_empty(buf_iter))
4080 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4088 /* Called with trace_event_read_lock() held. */
4089 enum print_line_t print_trace_line(struct trace_iterator *iter)
4091 struct trace_array *tr = iter->tr;
4092 unsigned long trace_flags = tr->trace_flags;
4093 enum print_line_t ret;
4095 if (iter->lost_events) {
4096 if (iter->lost_events == (unsigned long)-1)
4097 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4100 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4101 iter->cpu, iter->lost_events);
4102 if (trace_seq_has_overflowed(&iter->seq))
4103 return TRACE_TYPE_PARTIAL_LINE;
4106 if (iter->trace && iter->trace->print_line) {
4107 ret = iter->trace->print_line(iter);
4108 if (ret != TRACE_TYPE_UNHANDLED)
4112 if (iter->ent->type == TRACE_BPUTS &&
4113 trace_flags & TRACE_ITER_PRINTK &&
4114 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4115 return trace_print_bputs_msg_only(iter);
4117 if (iter->ent->type == TRACE_BPRINT &&
4118 trace_flags & TRACE_ITER_PRINTK &&
4119 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4120 return trace_print_bprintk_msg_only(iter);
4122 if (iter->ent->type == TRACE_PRINT &&
4123 trace_flags & TRACE_ITER_PRINTK &&
4124 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4125 return trace_print_printk_msg_only(iter);
4127 if (trace_flags & TRACE_ITER_BIN)
4128 return print_bin_fmt(iter);
4130 if (trace_flags & TRACE_ITER_HEX)
4131 return print_hex_fmt(iter);
4133 if (trace_flags & TRACE_ITER_RAW)
4134 return print_raw_fmt(iter);
4136 return print_trace_fmt(iter);
4139 void trace_latency_header(struct seq_file *m)
4141 struct trace_iterator *iter = m->private;
4142 struct trace_array *tr = iter->tr;
4144 /* print nothing if the buffers are empty */
4145 if (trace_empty(iter))
4148 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4149 print_trace_header(m, iter);
4151 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4152 print_lat_help_header(m);
4155 void trace_default_header(struct seq_file *m)
4157 struct trace_iterator *iter = m->private;
4158 struct trace_array *tr = iter->tr;
4159 unsigned long trace_flags = tr->trace_flags;
4161 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4164 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4165 /* print nothing if the buffers are empty */
4166 if (trace_empty(iter))
4168 print_trace_header(m, iter);
4169 if (!(trace_flags & TRACE_ITER_VERBOSE))
4170 print_lat_help_header(m);
4172 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4173 if (trace_flags & TRACE_ITER_IRQ_INFO)
4174 print_func_help_header_irq(iter->array_buffer,
4177 print_func_help_header(iter->array_buffer, m,
4183 static void test_ftrace_alive(struct seq_file *m)
4185 if (!ftrace_is_dead())
4187 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4188 "# MAY BE MISSING FUNCTION EVENTS\n");
4191 #ifdef CONFIG_TRACER_MAX_TRACE
4192 static void show_snapshot_main_help(struct seq_file *m)
4194 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4195 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4196 "# Takes a snapshot of the main buffer.\n"
4197 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4198 "# (Doesn't have to be '2' works with any number that\n"
4199 "# is not a '0' or '1')\n");
4202 static void show_snapshot_percpu_help(struct seq_file *m)
4204 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4205 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4206 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4207 "# Takes a snapshot of the main buffer for this cpu.\n");
4209 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4210 "# Must use main snapshot file to allocate.\n");
4212 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4213 "# (Doesn't have to be '2' works with any number that\n"
4214 "# is not a '0' or '1')\n");
4217 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4219 if (iter->tr->allocated_snapshot)
4220 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4222 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4224 seq_puts(m, "# Snapshot commands:\n");
4225 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4226 show_snapshot_main_help(m);
4228 show_snapshot_percpu_help(m);
4231 /* Should never be called */
4232 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4235 static int s_show(struct seq_file *m, void *v)
4237 struct trace_iterator *iter = v;
4240 if (iter->ent == NULL) {
4242 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4244 test_ftrace_alive(m);
4246 if (iter->snapshot && trace_empty(iter))
4247 print_snapshot_help(m, iter);
4248 else if (iter->trace && iter->trace->print_header)
4249 iter->trace->print_header(m);
4251 trace_default_header(m);
4253 } else if (iter->leftover) {
4255 * If we filled the seq_file buffer earlier, we
4256 * want to just show it now.
4258 ret = trace_print_seq(m, &iter->seq);
4260 /* ret should this time be zero, but you never know */
4261 iter->leftover = ret;
4264 print_trace_line(iter);
4265 ret = trace_print_seq(m, &iter->seq);
4267 * If we overflow the seq_file buffer, then it will
4268 * ask us for this data again at start up.
4270 * ret is 0 if seq_file write succeeded.
4273 iter->leftover = ret;
4280 * Should be used after trace_array_get(), trace_types_lock
4281 * ensures that i_cdev was already initialized.
4283 static inline int tracing_get_cpu(struct inode *inode)
4285 if (inode->i_cdev) /* See trace_create_cpu_file() */
4286 return (long)inode->i_cdev - 1;
4287 return RING_BUFFER_ALL_CPUS;
4290 static const struct seq_operations tracer_seq_ops = {
4297 static struct trace_iterator *
4298 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4300 struct trace_array *tr = inode->i_private;
4301 struct trace_iterator *iter;
4304 if (tracing_disabled)
4305 return ERR_PTR(-ENODEV);
4307 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4309 return ERR_PTR(-ENOMEM);
4311 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4313 if (!iter->buffer_iter)
4317 * trace_find_next_entry() may need to save off iter->ent.
4318 * It will place it into the iter->temp buffer. As most
4319 * events are less than 128, allocate a buffer of that size.
4320 * If one is greater, then trace_find_next_entry() will
4321 * allocate a new buffer to adjust for the bigger iter->ent.
4322 * It's not critical if it fails to get allocated here.
4324 iter->temp = kmalloc(128, GFP_KERNEL);
4326 iter->temp_size = 128;
4329 * We make a copy of the current tracer to avoid concurrent
4330 * changes on it while we are reading.
4332 mutex_lock(&trace_types_lock);
4333 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4337 *iter->trace = *tr->current_trace;
4339 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4344 #ifdef CONFIG_TRACER_MAX_TRACE
4345 /* Currently only the top directory has a snapshot */
4346 if (tr->current_trace->print_max || snapshot)
4347 iter->array_buffer = &tr->max_buffer;
4350 iter->array_buffer = &tr->array_buffer;
4351 iter->snapshot = snapshot;
4353 iter->cpu_file = tracing_get_cpu(inode);
4354 mutex_init(&iter->mutex);
4356 /* Notify the tracer early; before we stop tracing. */
4357 if (iter->trace->open)
4358 iter->trace->open(iter);
4360 /* Annotate start of buffers if we had overruns */
4361 if (ring_buffer_overruns(iter->array_buffer->buffer))
4362 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4364 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4365 if (trace_clocks[tr->clock_id].in_ns)
4366 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4369 * If pause-on-trace is enabled, then stop the trace while
4370 * dumping, unless this is the "snapshot" file
4372 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4373 tracing_stop_tr(tr);
4375 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4376 for_each_tracing_cpu(cpu) {
4377 iter->buffer_iter[cpu] =
4378 ring_buffer_read_prepare(iter->array_buffer->buffer,
4381 ring_buffer_read_prepare_sync();
4382 for_each_tracing_cpu(cpu) {
4383 ring_buffer_read_start(iter->buffer_iter[cpu]);
4384 tracing_iter_reset(iter, cpu);
4387 cpu = iter->cpu_file;
4388 iter->buffer_iter[cpu] =
4389 ring_buffer_read_prepare(iter->array_buffer->buffer,
4391 ring_buffer_read_prepare_sync();
4392 ring_buffer_read_start(iter->buffer_iter[cpu]);
4393 tracing_iter_reset(iter, cpu);
4396 mutex_unlock(&trace_types_lock);
4401 mutex_unlock(&trace_types_lock);
4404 kfree(iter->buffer_iter);
4406 seq_release_private(inode, file);
4407 return ERR_PTR(-ENOMEM);
4410 int tracing_open_generic(struct inode *inode, struct file *filp)
4414 ret = tracing_check_open_get_tr(NULL);
4418 filp->private_data = inode->i_private;
4422 bool tracing_is_disabled(void)
4424 return (tracing_disabled) ? true: false;
4428 * Open and update trace_array ref count.
4429 * Must have the current trace_array passed to it.
4431 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4433 struct trace_array *tr = inode->i_private;
4436 ret = tracing_check_open_get_tr(tr);
4440 filp->private_data = inode->i_private;
4445 static int tracing_release(struct inode *inode, struct file *file)
4447 struct trace_array *tr = inode->i_private;
4448 struct seq_file *m = file->private_data;
4449 struct trace_iterator *iter;
4452 if (!(file->f_mode & FMODE_READ)) {
4453 trace_array_put(tr);
4457 /* Writes do not use seq_file */
4459 mutex_lock(&trace_types_lock);
4461 for_each_tracing_cpu(cpu) {
4462 if (iter->buffer_iter[cpu])
4463 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4466 if (iter->trace && iter->trace->close)
4467 iter->trace->close(iter);
4469 if (!iter->snapshot && tr->stop_count)
4470 /* reenable tracing if it was previously enabled */
4471 tracing_start_tr(tr);
4473 __trace_array_put(tr);
4475 mutex_unlock(&trace_types_lock);
4477 mutex_destroy(&iter->mutex);
4478 free_cpumask_var(iter->started);
4481 kfree(iter->buffer_iter);
4482 seq_release_private(inode, file);
4487 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4489 struct trace_array *tr = inode->i_private;
4491 trace_array_put(tr);
4495 static int tracing_single_release_tr(struct inode *inode, struct file *file)
4497 struct trace_array *tr = inode->i_private;
4499 trace_array_put(tr);
4501 return single_release(inode, file);
4504 static int tracing_open(struct inode *inode, struct file *file)
4506 struct trace_array *tr = inode->i_private;
4507 struct trace_iterator *iter;
4510 ret = tracing_check_open_get_tr(tr);
4514 /* If this file was open for write, then erase contents */
4515 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4516 int cpu = tracing_get_cpu(inode);
4517 struct array_buffer *trace_buf = &tr->array_buffer;
4519 #ifdef CONFIG_TRACER_MAX_TRACE
4520 if (tr->current_trace->print_max)
4521 trace_buf = &tr->max_buffer;
4524 if (cpu == RING_BUFFER_ALL_CPUS)
4525 tracing_reset_online_cpus(trace_buf);
4527 tracing_reset_cpu(trace_buf, cpu);
4530 if (file->f_mode & FMODE_READ) {
4531 iter = __tracing_open(inode, file, false);
4533 ret = PTR_ERR(iter);
4534 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4535 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4539 trace_array_put(tr);
4545 * Some tracers are not suitable for instance buffers.
4546 * A tracer is always available for the global array (toplevel)
4547 * or if it explicitly states that it is.
4550 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4552 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4555 /* Find the next tracer that this trace array may use */
4556 static struct tracer *
4557 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4559 while (t && !trace_ok_for_array(t, tr))
4566 t_next(struct seq_file *m, void *v, loff_t *pos)
4568 struct trace_array *tr = m->private;
4569 struct tracer *t = v;
4574 t = get_tracer_for_array(tr, t->next);
4579 static void *t_start(struct seq_file *m, loff_t *pos)
4581 struct trace_array *tr = m->private;
4585 mutex_lock(&trace_types_lock);
4587 t = get_tracer_for_array(tr, trace_types);
4588 for (; t && l < *pos; t = t_next(m, t, &l))
4594 static void t_stop(struct seq_file *m, void *p)
4596 mutex_unlock(&trace_types_lock);
4599 static int t_show(struct seq_file *m, void *v)
4601 struct tracer *t = v;
4606 seq_puts(m, t->name);
4615 static const struct seq_operations show_traces_seq_ops = {
4622 static int show_traces_open(struct inode *inode, struct file *file)
4624 struct trace_array *tr = inode->i_private;
4628 ret = tracing_check_open_get_tr(tr);
4632 ret = seq_open(file, &show_traces_seq_ops);
4634 trace_array_put(tr);
4638 m = file->private_data;
4644 static int show_traces_release(struct inode *inode, struct file *file)
4646 struct trace_array *tr = inode->i_private;
4648 trace_array_put(tr);
4649 return seq_release(inode, file);
4653 tracing_write_stub(struct file *filp, const char __user *ubuf,
4654 size_t count, loff_t *ppos)
4659 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
4663 if (file->f_mode & FMODE_READ)
4664 ret = seq_lseek(file, offset, whence);
4666 file->f_pos = ret = 0;
4671 static const struct file_operations tracing_fops = {
4672 .open = tracing_open,
4674 .write = tracing_write_stub,
4675 .llseek = tracing_lseek,
4676 .release = tracing_release,
4679 static const struct file_operations show_traces_fops = {
4680 .open = show_traces_open,
4682 .llseek = seq_lseek,
4683 .release = show_traces_release,
4687 tracing_cpumask_read(struct file *filp, char __user *ubuf,
4688 size_t count, loff_t *ppos)
4690 struct trace_array *tr = file_inode(filp)->i_private;
4694 len = snprintf(NULL, 0, "%*pb\n",
4695 cpumask_pr_args(tr->tracing_cpumask)) + 1;
4696 mask_str = kmalloc(len, GFP_KERNEL);
4700 len = snprintf(mask_str, len, "%*pb\n",
4701 cpumask_pr_args(tr->tracing_cpumask));
4706 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
4714 int tracing_set_cpumask(struct trace_array *tr,
4715 cpumask_var_t tracing_cpumask_new)
4722 local_irq_disable();
4723 arch_spin_lock(&tr->max_lock);
4724 for_each_tracing_cpu(cpu) {
4726 * Increase/decrease the disabled counter if we are
4727 * about to flip a bit in the cpumask:
4729 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4730 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4731 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
4732 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
4734 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
4735 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
4736 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
4737 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
4740 arch_spin_unlock(&tr->max_lock);
4743 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
4749 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
4750 size_t count, loff_t *ppos)
4752 struct trace_array *tr = file_inode(filp)->i_private;
4753 cpumask_var_t tracing_cpumask_new;
4756 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
4759 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
4763 err = tracing_set_cpumask(tr, tracing_cpumask_new);
4767 free_cpumask_var(tracing_cpumask_new);
4772 free_cpumask_var(tracing_cpumask_new);
4777 static const struct file_operations tracing_cpumask_fops = {
4778 .open = tracing_open_generic_tr,
4779 .read = tracing_cpumask_read,
4780 .write = tracing_cpumask_write,
4781 .release = tracing_release_generic_tr,
4782 .llseek = generic_file_llseek,
4785 static int tracing_trace_options_show(struct seq_file *m, void *v)
4787 struct tracer_opt *trace_opts;
4788 struct trace_array *tr = m->private;
4792 mutex_lock(&trace_types_lock);
4793 tracer_flags = tr->current_trace->flags->val;
4794 trace_opts = tr->current_trace->flags->opts;
4796 for (i = 0; trace_options[i]; i++) {
4797 if (tr->trace_flags & (1 << i))
4798 seq_printf(m, "%s\n", trace_options[i]);
4800 seq_printf(m, "no%s\n", trace_options[i]);
4803 for (i = 0; trace_opts[i].name; i++) {
4804 if (tracer_flags & trace_opts[i].bit)
4805 seq_printf(m, "%s\n", trace_opts[i].name);
4807 seq_printf(m, "no%s\n", trace_opts[i].name);
4809 mutex_unlock(&trace_types_lock);
4814 static int __set_tracer_option(struct trace_array *tr,
4815 struct tracer_flags *tracer_flags,
4816 struct tracer_opt *opts, int neg)
4818 struct tracer *trace = tracer_flags->trace;
4821 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
4826 tracer_flags->val &= ~opts->bit;
4828 tracer_flags->val |= opts->bit;
4832 /* Try to assign a tracer specific option */
4833 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
4835 struct tracer *trace = tr->current_trace;
4836 struct tracer_flags *tracer_flags = trace->flags;
4837 struct tracer_opt *opts = NULL;
4840 for (i = 0; tracer_flags->opts[i].name; i++) {
4841 opts = &tracer_flags->opts[i];
4843 if (strcmp(cmp, opts->name) == 0)
4844 return __set_tracer_option(tr, trace->flags, opts, neg);
4850 /* Some tracers require overwrite to stay enabled */
4851 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
4853 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
4859 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
4861 if ((mask == TRACE_ITER_RECORD_TGID) ||
4862 (mask == TRACE_ITER_RECORD_CMD))
4863 lockdep_assert_held(&event_mutex);
4865 /* do nothing if flag is already set */
4866 if (!!(tr->trace_flags & mask) == !!enabled)
4869 /* Give the tracer a chance to approve the change */
4870 if (tr->current_trace->flag_changed)
4871 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
4875 tr->trace_flags |= mask;
4877 tr->trace_flags &= ~mask;
4879 if (mask == TRACE_ITER_RECORD_CMD)
4880 trace_event_enable_cmd_record(enabled);
4882 if (mask == TRACE_ITER_RECORD_TGID) {
4884 tgid_map = kvcalloc(PID_MAX_DEFAULT + 1,
4888 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
4892 trace_event_enable_tgid_record(enabled);
4895 if (mask == TRACE_ITER_EVENT_FORK)
4896 trace_event_follow_fork(tr, enabled);
4898 if (mask == TRACE_ITER_FUNC_FORK)
4899 ftrace_pid_follow_fork(tr, enabled);
4901 if (mask == TRACE_ITER_OVERWRITE) {
4902 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
4903 #ifdef CONFIG_TRACER_MAX_TRACE
4904 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
4908 if (mask == TRACE_ITER_PRINTK) {
4909 trace_printk_start_stop_comm(enabled);
4910 trace_printk_control(enabled);
4916 int trace_set_options(struct trace_array *tr, char *option)
4921 size_t orig_len = strlen(option);
4924 cmp = strstrip(option);
4926 len = str_has_prefix(cmp, "no");
4932 mutex_lock(&event_mutex);
4933 mutex_lock(&trace_types_lock);
4935 ret = match_string(trace_options, -1, cmp);
4936 /* If no option could be set, test the specific tracer options */
4938 ret = set_tracer_option(tr, cmp, neg);
4940 ret = set_tracer_flag(tr, 1 << ret, !neg);
4942 mutex_unlock(&trace_types_lock);
4943 mutex_unlock(&event_mutex);
4946 * If the first trailing whitespace is replaced with '\0' by strstrip,
4947 * turn it back into a space.
4949 if (orig_len > strlen(option))
4950 option[strlen(option)] = ' ';
4955 static void __init apply_trace_boot_options(void)
4957 char *buf = trace_boot_options_buf;
4961 option = strsep(&buf, ",");
4967 trace_set_options(&global_trace, option);
4969 /* Put back the comma to allow this to be called again */
4976 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
4977 size_t cnt, loff_t *ppos)
4979 struct seq_file *m = filp->private_data;
4980 struct trace_array *tr = m->private;
4984 if (cnt >= sizeof(buf))
4987 if (copy_from_user(buf, ubuf, cnt))
4992 ret = trace_set_options(tr, buf);
5001 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5003 struct trace_array *tr = inode->i_private;
5006 ret = tracing_check_open_get_tr(tr);
5010 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5012 trace_array_put(tr);
5017 static const struct file_operations tracing_iter_fops = {
5018 .open = tracing_trace_options_open,
5020 .llseek = seq_lseek,
5021 .release = tracing_single_release_tr,
5022 .write = tracing_trace_options_write,
5025 static const char readme_msg[] =
5026 "tracing mini-HOWTO:\n\n"
5027 "# echo 0 > tracing_on : quick way to disable tracing\n"
5028 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5029 " Important files:\n"
5030 " trace\t\t\t- The static contents of the buffer\n"
5031 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5032 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5033 " current_tracer\t- function and latency tracers\n"
5034 " available_tracers\t- list of configured tracers for current_tracer\n"
5035 " error_log\t- error log for failed commands (that support it)\n"
5036 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5037 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5038 " trace_clock\t\t-change the clock used to order events\n"
5039 " local: Per cpu clock but may not be synced across CPUs\n"
5040 " global: Synced across CPUs but slows tracing down.\n"
5041 " counter: Not a clock, but just an increment\n"
5042 " uptime: Jiffy counter from time of boot\n"
5043 " perf: Same clock that perf events use\n"
5044 #ifdef CONFIG_X86_64
5045 " x86-tsc: TSC cycle counter\n"
5047 "\n timestamp_mode\t-view the mode used to timestamp events\n"
5048 " delta: Delta difference against a buffer-wide timestamp\n"
5049 " absolute: Absolute (standalone) timestamp\n"
5050 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5051 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5052 " tracing_cpumask\t- Limit which CPUs to trace\n"
5053 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5054 "\t\t\t Remove sub-buffer with rmdir\n"
5055 " trace_options\t\t- Set format or modify how tracing happens\n"
5056 "\t\t\t Disable an option by prefixing 'no' to the\n"
5057 "\t\t\t option name\n"
5058 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5059 #ifdef CONFIG_DYNAMIC_FTRACE
5060 "\n available_filter_functions - list of functions that can be filtered on\n"
5061 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5062 "\t\t\t functions\n"
5063 "\t accepts: func_full_name or glob-matching-pattern\n"
5064 "\t modules: Can select a group via module\n"
5065 "\t Format: :mod:<module-name>\n"
5066 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5067 "\t triggers: a command to perform when function is hit\n"
5068 "\t Format: <function>:<trigger>[:count]\n"
5069 "\t trigger: traceon, traceoff\n"
5070 "\t\t enable_event:<system>:<event>\n"
5071 "\t\t disable_event:<system>:<event>\n"
5072 #ifdef CONFIG_STACKTRACE
5075 #ifdef CONFIG_TRACER_SNAPSHOT
5080 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5081 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5082 "\t The first one will disable tracing every time do_fault is hit\n"
5083 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5084 "\t The first time do trap is hit and it disables tracing, the\n"
5085 "\t counter will decrement to 2. If tracing is already disabled,\n"
5086 "\t the counter will not decrement. It only decrements when the\n"
5087 "\t trigger did work\n"
5088 "\t To remove trigger without count:\n"
5089 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5090 "\t To remove trigger with a count:\n"
5091 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5092 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5093 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5094 "\t modules: Can select a group via module command :mod:\n"
5095 "\t Does not accept triggers\n"
5096 #endif /* CONFIG_DYNAMIC_FTRACE */
5097 #ifdef CONFIG_FUNCTION_TRACER
5098 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5100 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5103 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5104 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5105 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5106 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5108 #ifdef CONFIG_TRACER_SNAPSHOT
5109 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5110 "\t\t\t snapshot buffer. Read the contents for more\n"
5111 "\t\t\t information\n"
5113 #ifdef CONFIG_STACK_TRACER
5114 " stack_trace\t\t- Shows the max stack trace when active\n"
5115 " stack_max_size\t- Shows current max stack size that was traced\n"
5116 "\t\t\t Write into this file to reset the max size (trigger a\n"
5117 "\t\t\t new trace)\n"
5118 #ifdef CONFIG_DYNAMIC_FTRACE
5119 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5122 #endif /* CONFIG_STACK_TRACER */
5123 #ifdef CONFIG_DYNAMIC_EVENTS
5124 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5125 "\t\t\t Write into this file to define/undefine new trace events.\n"
5127 #ifdef CONFIG_KPROBE_EVENTS
5128 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5129 "\t\t\t Write into this file to define/undefine new trace events.\n"
5131 #ifdef CONFIG_UPROBE_EVENTS
5132 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5133 "\t\t\t Write into this file to define/undefine new trace events.\n"
5135 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5136 "\t accepts: event-definitions (one definition per line)\n"
5137 "\t Format: p[:[<group>/]<event>] <place> [<args>]\n"
5138 "\t r[maxactive][:[<group>/]<event>] <place> [<args>]\n"
5139 #ifdef CONFIG_HIST_TRIGGERS
5140 "\t s:[synthetic/]<event> <field> [<field>]\n"
5142 "\t -:[<group>/]<event>\n"
5143 #ifdef CONFIG_KPROBE_EVENTS
5144 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5145 "place (kretprobe): [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5147 #ifdef CONFIG_UPROBE_EVENTS
5148 " place (uprobe): <path>:<offset>[(ref_ctr_offset)]\n"
5150 "\t args: <name>=fetcharg[:type]\n"
5151 "\t fetcharg: %<register>, @<address>, @<symbol>[+|-<offset>],\n"
5152 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5153 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5155 "\t $stack<index>, $stack, $retval, $comm,\n"
5157 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5158 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string, symbol,\n"
5159 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5160 "\t <type>\\[<array-size>\\]\n"
5161 #ifdef CONFIG_HIST_TRIGGERS
5162 "\t field: <stype> <name>;\n"
5163 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5164 "\t [unsigned] char/int/long\n"
5167 " events/\t\t- Directory containing all trace event subsystems:\n"
5168 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5169 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5170 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5172 " filter\t\t- If set, only events passing filter are traced\n"
5173 " events/<system>/<event>/\t- Directory containing control files for\n"
5175 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5176 " filter\t\t- If set, only events passing filter are traced\n"
5177 " trigger\t\t- If set, a command to perform when event is hit\n"
5178 "\t Format: <trigger>[:count][if <filter>]\n"
5179 "\t trigger: traceon, traceoff\n"
5180 "\t enable_event:<system>:<event>\n"
5181 "\t disable_event:<system>:<event>\n"
5182 #ifdef CONFIG_HIST_TRIGGERS
5183 "\t enable_hist:<system>:<event>\n"
5184 "\t disable_hist:<system>:<event>\n"
5186 #ifdef CONFIG_STACKTRACE
5189 #ifdef CONFIG_TRACER_SNAPSHOT
5192 #ifdef CONFIG_HIST_TRIGGERS
5193 "\t\t hist (see below)\n"
5195 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5196 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5197 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5198 "\t events/block/block_unplug/trigger\n"
5199 "\t The first disables tracing every time block_unplug is hit.\n"
5200 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5201 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5202 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5203 "\t Like function triggers, the counter is only decremented if it\n"
5204 "\t enabled or disabled tracing.\n"
5205 "\t To remove a trigger without a count:\n"
5206 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5207 "\t To remove a trigger with a count:\n"
5208 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5209 "\t Filters can be ignored when removing a trigger.\n"
5210 #ifdef CONFIG_HIST_TRIGGERS
5211 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5212 "\t Format: hist:keys=<field1[,field2,...]>\n"
5213 "\t [:values=<field1[,field2,...]>]\n"
5214 "\t [:sort=<field1[,field2,...]>]\n"
5215 "\t [:size=#entries]\n"
5216 "\t [:pause][:continue][:clear]\n"
5217 "\t [:name=histname1]\n"
5218 "\t [:<handler>.<action>]\n"
5219 "\t [if <filter>]\n\n"
5220 "\t When a matching event is hit, an entry is added to a hash\n"
5221 "\t table using the key(s) and value(s) named, and the value of a\n"
5222 "\t sum called 'hitcount' is incremented. Keys and values\n"
5223 "\t correspond to fields in the event's format description. Keys\n"
5224 "\t can be any field, or the special string 'stacktrace'.\n"
5225 "\t Compound keys consisting of up to two fields can be specified\n"
5226 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5227 "\t fields. Sort keys consisting of up to two fields can be\n"
5228 "\t specified using the 'sort' keyword. The sort direction can\n"
5229 "\t be modified by appending '.descending' or '.ascending' to a\n"
5230 "\t sort field. The 'size' parameter can be used to specify more\n"
5231 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5232 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5233 "\t its histogram data will be shared with other triggers of the\n"
5234 "\t same name, and trigger hits will update this common data.\n\n"
5235 "\t Reading the 'hist' file for the event will dump the hash\n"
5236 "\t table in its entirety to stdout. If there are multiple hist\n"
5237 "\t triggers attached to an event, there will be a table for each\n"
5238 "\t trigger in the output. The table displayed for a named\n"
5239 "\t trigger will be the same as any other instance having the\n"
5240 "\t same name. The default format used to display a given field\n"
5241 "\t can be modified by appending any of the following modifiers\n"
5242 "\t to the field name, as applicable:\n\n"
5243 "\t .hex display a number as a hex value\n"
5244 "\t .sym display an address as a symbol\n"
5245 "\t .sym-offset display an address as a symbol and offset\n"
5246 "\t .execname display a common_pid as a program name\n"
5247 "\t .syscall display a syscall id as a syscall name\n"
5248 "\t .log2 display log2 value rather than raw number\n"
5249 "\t .usecs display a common_timestamp in microseconds\n\n"
5250 "\t The 'pause' parameter can be used to pause an existing hist\n"
5251 "\t trigger or to start a hist trigger but not log any events\n"
5252 "\t until told to do so. 'continue' can be used to start or\n"
5253 "\t restart a paused hist trigger.\n\n"
5254 "\t The 'clear' parameter will clear the contents of a running\n"
5255 "\t hist trigger and leave its current paused/active state\n"
5257 "\t The enable_hist and disable_hist triggers can be used to\n"
5258 "\t have one event conditionally start and stop another event's\n"
5259 "\t already-attached hist trigger. The syntax is analogous to\n"
5260 "\t the enable_event and disable_event triggers.\n\n"
5261 "\t Hist trigger handlers and actions are executed whenever a\n"
5262 "\t a histogram entry is added or updated. They take the form:\n\n"
5263 "\t <handler>.<action>\n\n"
5264 "\t The available handlers are:\n\n"
5265 "\t onmatch(matching.event) - invoke on addition or update\n"
5266 "\t onmax(var) - invoke if var exceeds current max\n"
5267 "\t onchange(var) - invoke action if var changes\n\n"
5268 "\t The available actions are:\n\n"
5269 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5270 "\t save(field,...) - save current event fields\n"
5271 #ifdef CONFIG_TRACER_SNAPSHOT
5272 "\t snapshot() - snapshot the trace buffer\n"
5278 tracing_readme_read(struct file *filp, char __user *ubuf,
5279 size_t cnt, loff_t *ppos)
5281 return simple_read_from_buffer(ubuf, cnt, ppos,
5282 readme_msg, strlen(readme_msg));
5285 static const struct file_operations tracing_readme_fops = {
5286 .open = tracing_open_generic,
5287 .read = tracing_readme_read,
5288 .llseek = generic_file_llseek,
5291 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5295 if (*pos || m->count)
5300 for (; ptr <= &tgid_map[PID_MAX_DEFAULT]; ptr++) {
5301 if (trace_find_tgid(*ptr))
5308 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5318 v = saved_tgids_next(m, v, &l);
5326 static void saved_tgids_stop(struct seq_file *m, void *v)
5330 static int saved_tgids_show(struct seq_file *m, void *v)
5332 int pid = (int *)v - tgid_map;
5334 seq_printf(m, "%d %d\n", pid, trace_find_tgid(pid));
5338 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5339 .start = saved_tgids_start,
5340 .stop = saved_tgids_stop,
5341 .next = saved_tgids_next,
5342 .show = saved_tgids_show,
5345 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5349 ret = tracing_check_open_get_tr(NULL);
5353 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5357 static const struct file_operations tracing_saved_tgids_fops = {
5358 .open = tracing_saved_tgids_open,
5360 .llseek = seq_lseek,
5361 .release = seq_release,
5364 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5366 unsigned int *ptr = v;
5368 if (*pos || m->count)
5373 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5375 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5384 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5390 arch_spin_lock(&trace_cmdline_lock);
5392 v = &savedcmd->map_cmdline_to_pid[0];
5394 v = saved_cmdlines_next(m, v, &l);
5402 static void saved_cmdlines_stop(struct seq_file *m, void *v)
5404 arch_spin_unlock(&trace_cmdline_lock);
5408 static int saved_cmdlines_show(struct seq_file *m, void *v)
5410 char buf[TASK_COMM_LEN];
5411 unsigned int *pid = v;
5413 __trace_find_cmdline(*pid, buf);
5414 seq_printf(m, "%d %s\n", *pid, buf);
5418 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5419 .start = saved_cmdlines_start,
5420 .next = saved_cmdlines_next,
5421 .stop = saved_cmdlines_stop,
5422 .show = saved_cmdlines_show,
5425 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5429 ret = tracing_check_open_get_tr(NULL);
5433 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5436 static const struct file_operations tracing_saved_cmdlines_fops = {
5437 .open = tracing_saved_cmdlines_open,
5439 .llseek = seq_lseek,
5440 .release = seq_release,
5444 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5445 size_t cnt, loff_t *ppos)
5450 arch_spin_lock(&trace_cmdline_lock);
5451 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5452 arch_spin_unlock(&trace_cmdline_lock);
5454 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5457 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5459 kfree(s->saved_cmdlines);
5460 kfree(s->map_cmdline_to_pid);
5464 static int tracing_resize_saved_cmdlines(unsigned int val)
5466 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5468 s = kmalloc(sizeof(*s), GFP_KERNEL);
5472 if (allocate_cmdlines_buffer(val, s) < 0) {
5477 arch_spin_lock(&trace_cmdline_lock);
5478 savedcmd_temp = savedcmd;
5480 arch_spin_unlock(&trace_cmdline_lock);
5481 free_saved_cmdlines_buffer(savedcmd_temp);
5487 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
5488 size_t cnt, loff_t *ppos)
5493 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5497 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
5498 if (!val || val > PID_MAX_DEFAULT)
5501 ret = tracing_resize_saved_cmdlines((unsigned int)val);
5510 static const struct file_operations tracing_saved_cmdlines_size_fops = {
5511 .open = tracing_open_generic,
5512 .read = tracing_saved_cmdlines_size_read,
5513 .write = tracing_saved_cmdlines_size_write,
5516 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
5517 static union trace_eval_map_item *
5518 update_eval_map(union trace_eval_map_item *ptr)
5520 if (!ptr->map.eval_string) {
5521 if (ptr->tail.next) {
5522 ptr = ptr->tail.next;
5523 /* Set ptr to the next real item (skip head) */
5531 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5533 union trace_eval_map_item *ptr = v;
5536 * Paranoid! If ptr points to end, we don't want to increment past it.
5537 * This really should never happen.
5540 ptr = update_eval_map(ptr);
5541 if (WARN_ON_ONCE(!ptr))
5545 ptr = update_eval_map(ptr);
5550 static void *eval_map_start(struct seq_file *m, loff_t *pos)
5552 union trace_eval_map_item *v;
5555 mutex_lock(&trace_eval_mutex);
5557 v = trace_eval_maps;
5561 while (v && l < *pos) {
5562 v = eval_map_next(m, v, &l);
5568 static void eval_map_stop(struct seq_file *m, void *v)
5570 mutex_unlock(&trace_eval_mutex);
5573 static int eval_map_show(struct seq_file *m, void *v)
5575 union trace_eval_map_item *ptr = v;
5577 seq_printf(m, "%s %ld (%s)\n",
5578 ptr->map.eval_string, ptr->map.eval_value,
5584 static const struct seq_operations tracing_eval_map_seq_ops = {
5585 .start = eval_map_start,
5586 .next = eval_map_next,
5587 .stop = eval_map_stop,
5588 .show = eval_map_show,
5591 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5595 ret = tracing_check_open_get_tr(NULL);
5599 return seq_open(filp, &tracing_eval_map_seq_ops);
5602 static const struct file_operations tracing_eval_map_fops = {
5603 .open = tracing_eval_map_open,
5605 .llseek = seq_lseek,
5606 .release = seq_release,
5609 static inline union trace_eval_map_item *
5610 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
5612 /* Return tail of array given the head */
5613 return ptr + ptr->head.length + 1;
5617 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
5620 struct trace_eval_map **stop;
5621 struct trace_eval_map **map;
5622 union trace_eval_map_item *map_array;
5623 union trace_eval_map_item *ptr;
5628 * The trace_eval_maps contains the map plus a head and tail item,
5629 * where the head holds the module and length of array, and the
5630 * tail holds a pointer to the next list.
5632 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
5634 pr_warn("Unable to allocate trace eval mapping\n");
5638 mutex_lock(&trace_eval_mutex);
5640 if (!trace_eval_maps)
5641 trace_eval_maps = map_array;
5643 ptr = trace_eval_maps;
5645 ptr = trace_eval_jmp_to_tail(ptr);
5646 if (!ptr->tail.next)
5648 ptr = ptr->tail.next;
5651 ptr->tail.next = map_array;
5653 map_array->head.mod = mod;
5654 map_array->head.length = len;
5657 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
5658 map_array->map = **map;
5661 memset(map_array, 0, sizeof(*map_array));
5663 mutex_unlock(&trace_eval_mutex);
5666 static void trace_create_eval_file(struct dentry *d_tracer)
5668 trace_create_file("eval_map", 0444, d_tracer,
5669 NULL, &tracing_eval_map_fops);
5672 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
5673 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
5674 static inline void trace_insert_eval_map_file(struct module *mod,
5675 struct trace_eval_map **start, int len) { }
5676 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
5678 static void trace_insert_eval_map(struct module *mod,
5679 struct trace_eval_map **start, int len)
5681 struct trace_eval_map **map;
5688 trace_event_eval_update(map, len);
5690 trace_insert_eval_map_file(mod, start, len);
5694 tracing_set_trace_read(struct file *filp, char __user *ubuf,
5695 size_t cnt, loff_t *ppos)
5697 struct trace_array *tr = filp->private_data;
5698 char buf[MAX_TRACER_SIZE+2];
5701 mutex_lock(&trace_types_lock);
5702 r = sprintf(buf, "%s\n", tr->current_trace->name);
5703 mutex_unlock(&trace_types_lock);
5705 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5708 int tracer_init(struct tracer *t, struct trace_array *tr)
5710 tracing_reset_online_cpus(&tr->array_buffer);
5714 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
5718 for_each_tracing_cpu(cpu)
5719 per_cpu_ptr(buf->data, cpu)->entries = val;
5722 #ifdef CONFIG_TRACER_MAX_TRACE
5723 /* resize @tr's buffer to the size of @size_tr's entries */
5724 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
5725 struct array_buffer *size_buf, int cpu_id)
5729 if (cpu_id == RING_BUFFER_ALL_CPUS) {
5730 for_each_tracing_cpu(cpu) {
5731 ret = ring_buffer_resize(trace_buf->buffer,
5732 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
5735 per_cpu_ptr(trace_buf->data, cpu)->entries =
5736 per_cpu_ptr(size_buf->data, cpu)->entries;
5739 ret = ring_buffer_resize(trace_buf->buffer,
5740 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
5742 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
5743 per_cpu_ptr(size_buf->data, cpu_id)->entries;
5748 #endif /* CONFIG_TRACER_MAX_TRACE */
5750 static int __tracing_resize_ring_buffer(struct trace_array *tr,
5751 unsigned long size, int cpu)
5756 * If kernel or user changes the size of the ring buffer
5757 * we use the size that was given, and we can forget about
5758 * expanding it later.
5760 ring_buffer_expanded = true;
5762 /* May be called before buffers are initialized */
5763 if (!tr->array_buffer.buffer)
5766 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
5770 #ifdef CONFIG_TRACER_MAX_TRACE
5771 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
5772 !tr->current_trace->use_max_tr)
5775 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
5777 int r = resize_buffer_duplicate_size(&tr->array_buffer,
5778 &tr->array_buffer, cpu);
5781 * AARGH! We are left with different
5782 * size max buffer!!!!
5783 * The max buffer is our "snapshot" buffer.
5784 * When a tracer needs a snapshot (one of the
5785 * latency tracers), it swaps the max buffer
5786 * with the saved snap shot. We succeeded to
5787 * update the size of the main buffer, but failed to
5788 * update the size of the max buffer. But when we tried
5789 * to reset the main buffer to the original size, we
5790 * failed there too. This is very unlikely to
5791 * happen, but if it does, warn and kill all
5795 tracing_disabled = 1;
5800 if (cpu == RING_BUFFER_ALL_CPUS)
5801 set_buffer_entries(&tr->max_buffer, size);
5803 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
5806 #endif /* CONFIG_TRACER_MAX_TRACE */
5808 if (cpu == RING_BUFFER_ALL_CPUS)
5809 set_buffer_entries(&tr->array_buffer, size);
5811 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
5816 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
5817 unsigned long size, int cpu_id)
5821 mutex_lock(&trace_types_lock);
5823 if (cpu_id != RING_BUFFER_ALL_CPUS) {
5824 /* make sure, this cpu is enabled in the mask */
5825 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
5831 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
5836 mutex_unlock(&trace_types_lock);
5843 * tracing_update_buffers - used by tracing facility to expand ring buffers
5845 * To save on memory when the tracing is never used on a system with it
5846 * configured in. The ring buffers are set to a minimum size. But once
5847 * a user starts to use the tracing facility, then they need to grow
5848 * to their default size.
5850 * This function is to be called when a tracer is about to be used.
5852 int tracing_update_buffers(void)
5856 mutex_lock(&trace_types_lock);
5857 if (!ring_buffer_expanded)
5858 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
5859 RING_BUFFER_ALL_CPUS);
5860 mutex_unlock(&trace_types_lock);
5865 struct trace_option_dentry;
5868 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
5871 * Used to clear out the tracer before deletion of an instance.
5872 * Must have trace_types_lock held.
5874 static void tracing_set_nop(struct trace_array *tr)
5876 if (tr->current_trace == &nop_trace)
5879 tr->current_trace->enabled--;
5881 if (tr->current_trace->reset)
5882 tr->current_trace->reset(tr);
5884 tr->current_trace = &nop_trace;
5887 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
5889 /* Only enable if the directory has been created already. */
5893 create_trace_option_files(tr, t);
5896 int tracing_set_tracer(struct trace_array *tr, const char *buf)
5899 #ifdef CONFIG_TRACER_MAX_TRACE
5904 mutex_lock(&trace_types_lock);
5906 if (!ring_buffer_expanded) {
5907 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
5908 RING_BUFFER_ALL_CPUS);
5914 for (t = trace_types; t; t = t->next) {
5915 if (strcmp(t->name, buf) == 0)
5922 if (t == tr->current_trace)
5925 #ifdef CONFIG_TRACER_SNAPSHOT
5926 if (t->use_max_tr) {
5927 arch_spin_lock(&tr->max_lock);
5928 if (tr->cond_snapshot)
5930 arch_spin_unlock(&tr->max_lock);
5935 /* Some tracers won't work on kernel command line */
5936 if (system_state < SYSTEM_RUNNING && t->noboot) {
5937 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
5942 /* Some tracers are only allowed for the top level buffer */
5943 if (!trace_ok_for_array(t, tr)) {
5948 /* If trace pipe files are being read, we can't change the tracer */
5949 if (tr->trace_ref) {
5954 trace_branch_disable();
5956 tr->current_trace->enabled--;
5958 if (tr->current_trace->reset)
5959 tr->current_trace->reset(tr);
5961 /* Current trace needs to be nop_trace before synchronize_rcu */
5962 tr->current_trace = &nop_trace;
5964 #ifdef CONFIG_TRACER_MAX_TRACE
5965 had_max_tr = tr->allocated_snapshot;
5967 if (had_max_tr && !t->use_max_tr) {
5969 * We need to make sure that the update_max_tr sees that
5970 * current_trace changed to nop_trace to keep it from
5971 * swapping the buffers after we resize it.
5972 * The update_max_tr is called from interrupts disabled
5973 * so a synchronized_sched() is sufficient.
5980 #ifdef CONFIG_TRACER_MAX_TRACE
5981 if (t->use_max_tr && !had_max_tr) {
5982 ret = tracing_alloc_snapshot_instance(tr);
5989 ret = tracer_init(t, tr);
5994 tr->current_trace = t;
5995 tr->current_trace->enabled++;
5996 trace_branch_enable(tr);
5998 mutex_unlock(&trace_types_lock);
6004 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6005 size_t cnt, loff_t *ppos)
6007 struct trace_array *tr = filp->private_data;
6008 char buf[MAX_TRACER_SIZE+1];
6015 if (cnt > MAX_TRACER_SIZE)
6016 cnt = MAX_TRACER_SIZE;
6018 if (copy_from_user(buf, ubuf, cnt))
6023 /* strip ending whitespace. */
6024 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
6027 err = tracing_set_tracer(tr, buf);
6037 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6038 size_t cnt, loff_t *ppos)
6043 r = snprintf(buf, sizeof(buf), "%ld\n",
6044 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6045 if (r > sizeof(buf))
6047 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6051 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6052 size_t cnt, loff_t *ppos)
6057 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6067 tracing_thresh_read(struct file *filp, char __user *ubuf,
6068 size_t cnt, loff_t *ppos)
6070 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6074 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6075 size_t cnt, loff_t *ppos)
6077 struct trace_array *tr = filp->private_data;
6080 mutex_lock(&trace_types_lock);
6081 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6085 if (tr->current_trace->update_thresh) {
6086 ret = tr->current_trace->update_thresh(tr);
6093 mutex_unlock(&trace_types_lock);
6098 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
6101 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6102 size_t cnt, loff_t *ppos)
6104 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6108 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6109 size_t cnt, loff_t *ppos)
6111 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6116 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6118 struct trace_array *tr = inode->i_private;
6119 struct trace_iterator *iter;
6122 ret = tracing_check_open_get_tr(tr);
6126 mutex_lock(&trace_types_lock);
6128 /* create a buffer to store the information to pass to userspace */
6129 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6132 __trace_array_put(tr);
6136 trace_seq_init(&iter->seq);
6137 iter->trace = tr->current_trace;
6139 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6144 /* trace pipe does not show start of buffer */
6145 cpumask_setall(iter->started);
6147 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6148 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6150 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6151 if (trace_clocks[tr->clock_id].in_ns)
6152 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6155 iter->array_buffer = &tr->array_buffer;
6156 iter->cpu_file = tracing_get_cpu(inode);
6157 mutex_init(&iter->mutex);
6158 filp->private_data = iter;
6160 if (iter->trace->pipe_open)
6161 iter->trace->pipe_open(iter);
6163 nonseekable_open(inode, filp);
6167 mutex_unlock(&trace_types_lock);
6172 __trace_array_put(tr);
6173 mutex_unlock(&trace_types_lock);
6177 static int tracing_release_pipe(struct inode *inode, struct file *file)
6179 struct trace_iterator *iter = file->private_data;
6180 struct trace_array *tr = inode->i_private;
6182 mutex_lock(&trace_types_lock);
6186 if (iter->trace->pipe_close)
6187 iter->trace->pipe_close(iter);
6189 mutex_unlock(&trace_types_lock);
6191 free_cpumask_var(iter->started);
6192 mutex_destroy(&iter->mutex);
6195 trace_array_put(tr);
6201 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6203 struct trace_array *tr = iter->tr;
6205 /* Iterators are static, they should be filled or empty */
6206 if (trace_buffer_iter(iter, iter->cpu_file))
6207 return EPOLLIN | EPOLLRDNORM;
6209 if (tr->trace_flags & TRACE_ITER_BLOCK)
6211 * Always select as readable when in blocking mode
6213 return EPOLLIN | EPOLLRDNORM;
6215 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6220 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6222 struct trace_iterator *iter = filp->private_data;
6224 return trace_poll(iter, filp, poll_table);
6227 /* Must be called with iter->mutex held. */
6228 static int tracing_wait_pipe(struct file *filp)
6230 struct trace_iterator *iter = filp->private_data;
6233 while (trace_empty(iter)) {
6235 if ((filp->f_flags & O_NONBLOCK)) {
6240 * We block until we read something and tracing is disabled.
6241 * We still block if tracing is disabled, but we have never
6242 * read anything. This allows a user to cat this file, and
6243 * then enable tracing. But after we have read something,
6244 * we give an EOF when tracing is again disabled.
6246 * iter->pos will be 0 if we haven't read anything.
6248 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6251 mutex_unlock(&iter->mutex);
6253 ret = wait_on_pipe(iter, 0);
6255 mutex_lock(&iter->mutex);
6268 tracing_read_pipe(struct file *filp, char __user *ubuf,
6269 size_t cnt, loff_t *ppos)
6271 struct trace_iterator *iter = filp->private_data;
6275 * Avoid more than one consumer on a single file descriptor
6276 * This is just a matter of traces coherency, the ring buffer itself
6279 mutex_lock(&iter->mutex);
6281 /* return any leftover data */
6282 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6286 trace_seq_init(&iter->seq);
6288 if (iter->trace->read) {
6289 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6295 sret = tracing_wait_pipe(filp);
6299 /* stop when tracing is finished */
6300 if (trace_empty(iter)) {
6305 if (cnt >= PAGE_SIZE)
6306 cnt = PAGE_SIZE - 1;
6308 /* reset all but tr, trace, and overruns */
6309 memset(&iter->seq, 0,
6310 sizeof(struct trace_iterator) -
6311 offsetof(struct trace_iterator, seq));
6312 cpumask_clear(iter->started);
6313 trace_seq_init(&iter->seq);
6316 trace_event_read_lock();
6317 trace_access_lock(iter->cpu_file);
6318 while (trace_find_next_entry_inc(iter) != NULL) {
6319 enum print_line_t ret;
6320 int save_len = iter->seq.seq.len;
6322 ret = print_trace_line(iter);
6323 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6324 /* don't print partial lines */
6325 iter->seq.seq.len = save_len;
6328 if (ret != TRACE_TYPE_NO_CONSUME)
6329 trace_consume(iter);
6331 if (trace_seq_used(&iter->seq) >= cnt)
6335 * Setting the full flag means we reached the trace_seq buffer
6336 * size and we should leave by partial output condition above.
6337 * One of the trace_seq_* functions is not used properly.
6339 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6342 trace_access_unlock(iter->cpu_file);
6343 trace_event_read_unlock();
6345 /* Now copy what we have to the user */
6346 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6347 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6348 trace_seq_init(&iter->seq);
6351 * If there was nothing to send to user, in spite of consuming trace
6352 * entries, go back to wait for more entries.
6358 mutex_unlock(&iter->mutex);
6363 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6366 __free_page(spd->pages[idx]);
6370 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6376 /* Seq buffer is page-sized, exactly what we need. */
6378 save_len = iter->seq.seq.len;
6379 ret = print_trace_line(iter);
6381 if (trace_seq_has_overflowed(&iter->seq)) {
6382 iter->seq.seq.len = save_len;
6387 * This should not be hit, because it should only
6388 * be set if the iter->seq overflowed. But check it
6389 * anyway to be safe.
6391 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6392 iter->seq.seq.len = save_len;
6396 count = trace_seq_used(&iter->seq) - save_len;
6399 iter->seq.seq.len = save_len;
6403 if (ret != TRACE_TYPE_NO_CONSUME)
6404 trace_consume(iter);
6406 if (!trace_find_next_entry_inc(iter)) {
6416 static ssize_t tracing_splice_read_pipe(struct file *filp,
6418 struct pipe_inode_info *pipe,
6422 struct page *pages_def[PIPE_DEF_BUFFERS];
6423 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6424 struct trace_iterator *iter = filp->private_data;
6425 struct splice_pipe_desc spd = {
6427 .partial = partial_def,
6428 .nr_pages = 0, /* This gets updated below. */
6429 .nr_pages_max = PIPE_DEF_BUFFERS,
6430 .ops = &default_pipe_buf_ops,
6431 .spd_release = tracing_spd_release_pipe,
6437 if (splice_grow_spd(pipe, &spd))
6440 mutex_lock(&iter->mutex);
6442 if (iter->trace->splice_read) {
6443 ret = iter->trace->splice_read(iter, filp,
6444 ppos, pipe, len, flags);
6449 ret = tracing_wait_pipe(filp);
6453 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6458 trace_event_read_lock();
6459 trace_access_lock(iter->cpu_file);
6461 /* Fill as many pages as possible. */
6462 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6463 spd.pages[i] = alloc_page(GFP_KERNEL);
6467 rem = tracing_fill_pipe_page(rem, iter);
6469 /* Copy the data into the page, so we can start over. */
6470 ret = trace_seq_to_buffer(&iter->seq,
6471 page_address(spd.pages[i]),
6472 trace_seq_used(&iter->seq));
6474 __free_page(spd.pages[i]);
6477 spd.partial[i].offset = 0;
6478 spd.partial[i].len = trace_seq_used(&iter->seq);
6480 trace_seq_init(&iter->seq);
6483 trace_access_unlock(iter->cpu_file);
6484 trace_event_read_unlock();
6485 mutex_unlock(&iter->mutex);
6490 ret = splice_to_pipe(pipe, &spd);
6494 splice_shrink_spd(&spd);
6498 mutex_unlock(&iter->mutex);
6503 tracing_entries_read(struct file *filp, char __user *ubuf,
6504 size_t cnt, loff_t *ppos)
6506 struct inode *inode = file_inode(filp);
6507 struct trace_array *tr = inode->i_private;
6508 int cpu = tracing_get_cpu(inode);
6513 mutex_lock(&trace_types_lock);
6515 if (cpu == RING_BUFFER_ALL_CPUS) {
6516 int cpu, buf_size_same;
6521 /* check if all cpu sizes are same */
6522 for_each_tracing_cpu(cpu) {
6523 /* fill in the size from first enabled cpu */
6525 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6526 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6532 if (buf_size_same) {
6533 if (!ring_buffer_expanded)
6534 r = sprintf(buf, "%lu (expanded: %lu)\n",
6536 trace_buf_size >> 10);
6538 r = sprintf(buf, "%lu\n", size >> 10);
6540 r = sprintf(buf, "X\n");
6542 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6544 mutex_unlock(&trace_types_lock);
6546 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6551 tracing_entries_write(struct file *filp, const char __user *ubuf,
6552 size_t cnt, loff_t *ppos)
6554 struct inode *inode = file_inode(filp);
6555 struct trace_array *tr = inode->i_private;
6559 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6563 /* must have at least 1 entry */
6567 /* value is in KB */
6569 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6579 tracing_total_entries_read(struct file *filp, char __user *ubuf,
6580 size_t cnt, loff_t *ppos)
6582 struct trace_array *tr = filp->private_data;
6585 unsigned long size = 0, expanded_size = 0;
6587 mutex_lock(&trace_types_lock);
6588 for_each_tracing_cpu(cpu) {
6589 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6590 if (!ring_buffer_expanded)
6591 expanded_size += trace_buf_size >> 10;
6593 if (ring_buffer_expanded)
6594 r = sprintf(buf, "%lu\n", size);
6596 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6597 mutex_unlock(&trace_types_lock);
6599 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6603 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
6604 size_t cnt, loff_t *ppos)
6607 * There is no need to read what the user has written, this function
6608 * is just to make sure that there is no error when "echo" is used
6617 tracing_free_buffer_release(struct inode *inode, struct file *filp)
6619 struct trace_array *tr = inode->i_private;
6621 /* disable tracing ? */
6622 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
6623 tracer_tracing_off(tr);
6624 /* resize the ring buffer to 0 */
6625 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
6627 trace_array_put(tr);
6633 tracing_mark_write(struct file *filp, const char __user *ubuf,
6634 size_t cnt, loff_t *fpos)
6636 struct trace_array *tr = filp->private_data;
6637 struct ring_buffer_event *event;
6638 enum event_trigger_type tt = ETT_NONE;
6639 struct trace_buffer *buffer;
6640 struct print_entry *entry;
6641 unsigned long irq_flags;
6646 /* Used in tracing_mark_raw_write() as well */
6647 #define FAULTED_STR "<faulted>"
6648 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
6650 if (tracing_disabled)
6653 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6656 if (cnt > TRACE_BUF_SIZE)
6657 cnt = TRACE_BUF_SIZE;
6659 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6661 local_save_flags(irq_flags);
6662 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
6664 /* If less than "<faulted>", then make sure we can still add that */
6665 if (cnt < FAULTED_SIZE)
6666 size += FAULTED_SIZE - cnt;
6668 buffer = tr->array_buffer.buffer;
6669 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
6670 irq_flags, preempt_count());
6671 if (unlikely(!event))
6672 /* Ring buffer disabled, return as if not open for write */
6675 entry = ring_buffer_event_data(event);
6676 entry->ip = _THIS_IP_;
6678 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
6680 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6687 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
6688 /* do not add \n before testing triggers, but add \0 */
6689 entry->buf[cnt] = '\0';
6690 tt = event_triggers_call(tr->trace_marker_file, entry, event);
6693 if (entry->buf[cnt - 1] != '\n') {
6694 entry->buf[cnt] = '\n';
6695 entry->buf[cnt + 1] = '\0';
6697 entry->buf[cnt] = '\0';
6699 if (static_branch_unlikely(&trace_marker_exports_enabled))
6700 ftrace_exports(event, TRACE_EXPORT_MARKER);
6701 __buffer_unlock_commit(buffer, event);
6704 event_triggers_post_call(tr->trace_marker_file, tt);
6712 /* Limit it for now to 3K (including tag) */
6713 #define RAW_DATA_MAX_SIZE (1024*3)
6716 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
6717 size_t cnt, loff_t *fpos)
6719 struct trace_array *tr = filp->private_data;
6720 struct ring_buffer_event *event;
6721 struct trace_buffer *buffer;
6722 struct raw_data_entry *entry;
6723 unsigned long irq_flags;
6728 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
6730 if (tracing_disabled)
6733 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
6736 /* The marker must at least have a tag id */
6737 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
6740 if (cnt > TRACE_BUF_SIZE)
6741 cnt = TRACE_BUF_SIZE;
6743 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
6745 local_save_flags(irq_flags);
6746 size = sizeof(*entry) + cnt;
6747 if (cnt < FAULT_SIZE_ID)
6748 size += FAULT_SIZE_ID - cnt;
6750 buffer = tr->array_buffer.buffer;
6751 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
6752 irq_flags, preempt_count());
6754 /* Ring buffer disabled, return as if not open for write */
6757 entry = ring_buffer_event_data(event);
6759 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
6762 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
6767 __buffer_unlock_commit(buffer, event);
6775 static int tracing_clock_show(struct seq_file *m, void *v)
6777 struct trace_array *tr = m->private;
6780 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
6782 "%s%s%s%s", i ? " " : "",
6783 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
6784 i == tr->clock_id ? "]" : "");
6790 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
6794 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
6795 if (strcmp(trace_clocks[i].name, clockstr) == 0)
6798 if (i == ARRAY_SIZE(trace_clocks))
6801 mutex_lock(&trace_types_lock);
6805 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
6808 * New clock may not be consistent with the previous clock.
6809 * Reset the buffer so that it doesn't have incomparable timestamps.
6811 tracing_reset_online_cpus(&tr->array_buffer);
6813 #ifdef CONFIG_TRACER_MAX_TRACE
6814 if (tr->max_buffer.buffer)
6815 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
6816 tracing_reset_online_cpus(&tr->max_buffer);
6819 mutex_unlock(&trace_types_lock);
6824 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
6825 size_t cnt, loff_t *fpos)
6827 struct seq_file *m = filp->private_data;
6828 struct trace_array *tr = m->private;
6830 const char *clockstr;
6833 if (cnt >= sizeof(buf))
6836 if (copy_from_user(buf, ubuf, cnt))
6841 clockstr = strstrip(buf);
6843 ret = tracing_set_clock(tr, clockstr);
6852 static int tracing_clock_open(struct inode *inode, struct file *file)
6854 struct trace_array *tr = inode->i_private;
6857 ret = tracing_check_open_get_tr(tr);
6861 ret = single_open(file, tracing_clock_show, inode->i_private);
6863 trace_array_put(tr);
6868 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
6870 struct trace_array *tr = m->private;
6872 mutex_lock(&trace_types_lock);
6874 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
6875 seq_puts(m, "delta [absolute]\n");
6877 seq_puts(m, "[delta] absolute\n");
6879 mutex_unlock(&trace_types_lock);
6884 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
6886 struct trace_array *tr = inode->i_private;
6889 ret = tracing_check_open_get_tr(tr);
6893 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
6895 trace_array_put(tr);
6900 int tracing_set_time_stamp_abs(struct trace_array *tr, bool abs)
6904 mutex_lock(&trace_types_lock);
6906 if (abs && tr->time_stamp_abs_ref++)
6910 if (WARN_ON_ONCE(!tr->time_stamp_abs_ref)) {
6915 if (--tr->time_stamp_abs_ref)
6919 ring_buffer_set_time_stamp_abs(tr->array_buffer.buffer, abs);
6921 #ifdef CONFIG_TRACER_MAX_TRACE
6922 if (tr->max_buffer.buffer)
6923 ring_buffer_set_time_stamp_abs(tr->max_buffer.buffer, abs);
6926 mutex_unlock(&trace_types_lock);
6931 struct ftrace_buffer_info {
6932 struct trace_iterator iter;
6934 unsigned int spare_cpu;
6938 #ifdef CONFIG_TRACER_SNAPSHOT
6939 static int tracing_snapshot_open(struct inode *inode, struct file *file)
6941 struct trace_array *tr = inode->i_private;
6942 struct trace_iterator *iter;
6946 ret = tracing_check_open_get_tr(tr);
6950 if (file->f_mode & FMODE_READ) {
6951 iter = __tracing_open(inode, file, true);
6953 ret = PTR_ERR(iter);
6955 /* Writes still need the seq_file to hold the private data */
6957 m = kzalloc(sizeof(*m), GFP_KERNEL);
6960 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6968 iter->array_buffer = &tr->max_buffer;
6969 iter->cpu_file = tracing_get_cpu(inode);
6971 file->private_data = m;
6975 trace_array_put(tr);
6981 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
6984 struct seq_file *m = filp->private_data;
6985 struct trace_iterator *iter = m->private;
6986 struct trace_array *tr = iter->tr;
6990 ret = tracing_update_buffers();
6994 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6998 mutex_lock(&trace_types_lock);
7000 if (tr->current_trace->use_max_tr) {
7005 arch_spin_lock(&tr->max_lock);
7006 if (tr->cond_snapshot)
7008 arch_spin_unlock(&tr->max_lock);
7014 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7018 if (tr->allocated_snapshot)
7022 /* Only allow per-cpu swap if the ring buffer supports it */
7023 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7024 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7029 if (tr->allocated_snapshot)
7030 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7031 &tr->array_buffer, iter->cpu_file);
7033 ret = tracing_alloc_snapshot_instance(tr);
7036 local_irq_disable();
7037 /* Now, we're going to swap */
7038 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7039 update_max_tr(tr, current, smp_processor_id(), NULL);
7041 update_max_tr_single(tr, current, iter->cpu_file);
7045 if (tr->allocated_snapshot) {
7046 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7047 tracing_reset_online_cpus(&tr->max_buffer);
7049 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7059 mutex_unlock(&trace_types_lock);
7063 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7065 struct seq_file *m = file->private_data;
7068 ret = tracing_release(inode, file);
7070 if (file->f_mode & FMODE_READ)
7073 /* If write only, the seq_file is just a stub */
7081 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7082 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7083 size_t count, loff_t *ppos);
7084 static int tracing_buffers_release(struct inode *inode, struct file *file);
7085 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7086 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7088 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7090 struct ftrace_buffer_info *info;
7093 /* The following checks for tracefs lockdown */
7094 ret = tracing_buffers_open(inode, filp);
7098 info = filp->private_data;
7100 if (info->iter.trace->use_max_tr) {
7101 tracing_buffers_release(inode, filp);
7105 info->iter.snapshot = true;
7106 info->iter.array_buffer = &info->iter.tr->max_buffer;
7111 #endif /* CONFIG_TRACER_SNAPSHOT */
7114 static const struct file_operations tracing_thresh_fops = {
7115 .open = tracing_open_generic,
7116 .read = tracing_thresh_read,
7117 .write = tracing_thresh_write,
7118 .llseek = generic_file_llseek,
7121 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
7122 static const struct file_operations tracing_max_lat_fops = {
7123 .open = tracing_open_generic,
7124 .read = tracing_max_lat_read,
7125 .write = tracing_max_lat_write,
7126 .llseek = generic_file_llseek,
7130 static const struct file_operations set_tracer_fops = {
7131 .open = tracing_open_generic,
7132 .read = tracing_set_trace_read,
7133 .write = tracing_set_trace_write,
7134 .llseek = generic_file_llseek,
7137 static const struct file_operations tracing_pipe_fops = {
7138 .open = tracing_open_pipe,
7139 .poll = tracing_poll_pipe,
7140 .read = tracing_read_pipe,
7141 .splice_read = tracing_splice_read_pipe,
7142 .release = tracing_release_pipe,
7143 .llseek = no_llseek,
7146 static const struct file_operations tracing_entries_fops = {
7147 .open = tracing_open_generic_tr,
7148 .read = tracing_entries_read,
7149 .write = tracing_entries_write,
7150 .llseek = generic_file_llseek,
7151 .release = tracing_release_generic_tr,
7154 static const struct file_operations tracing_total_entries_fops = {
7155 .open = tracing_open_generic_tr,
7156 .read = tracing_total_entries_read,
7157 .llseek = generic_file_llseek,
7158 .release = tracing_release_generic_tr,
7161 static const struct file_operations tracing_free_buffer_fops = {
7162 .open = tracing_open_generic_tr,
7163 .write = tracing_free_buffer_write,
7164 .release = tracing_free_buffer_release,
7167 static const struct file_operations tracing_mark_fops = {
7168 .open = tracing_open_generic_tr,
7169 .write = tracing_mark_write,
7170 .llseek = generic_file_llseek,
7171 .release = tracing_release_generic_tr,
7174 static const struct file_operations tracing_mark_raw_fops = {
7175 .open = tracing_open_generic_tr,
7176 .write = tracing_mark_raw_write,
7177 .llseek = generic_file_llseek,
7178 .release = tracing_release_generic_tr,
7181 static const struct file_operations trace_clock_fops = {
7182 .open = tracing_clock_open,
7184 .llseek = seq_lseek,
7185 .release = tracing_single_release_tr,
7186 .write = tracing_clock_write,
7189 static const struct file_operations trace_time_stamp_mode_fops = {
7190 .open = tracing_time_stamp_mode_open,
7192 .llseek = seq_lseek,
7193 .release = tracing_single_release_tr,
7196 #ifdef CONFIG_TRACER_SNAPSHOT
7197 static const struct file_operations snapshot_fops = {
7198 .open = tracing_snapshot_open,
7200 .write = tracing_snapshot_write,
7201 .llseek = tracing_lseek,
7202 .release = tracing_snapshot_release,
7205 static const struct file_operations snapshot_raw_fops = {
7206 .open = snapshot_raw_open,
7207 .read = tracing_buffers_read,
7208 .release = tracing_buffers_release,
7209 .splice_read = tracing_buffers_splice_read,
7210 .llseek = no_llseek,
7213 #endif /* CONFIG_TRACER_SNAPSHOT */
7215 #define TRACING_LOG_ERRS_MAX 8
7216 #define TRACING_LOG_LOC_MAX 128
7218 #define CMD_PREFIX " Command: "
7221 const char **errs; /* ptr to loc-specific array of err strings */
7222 u8 type; /* index into errs -> specific err string */
7223 u8 pos; /* MAX_FILTER_STR_VAL = 256 */
7227 struct tracing_log_err {
7228 struct list_head list;
7229 struct err_info info;
7230 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7231 char cmd[MAX_FILTER_STR_VAL]; /* what caused err */
7234 static DEFINE_MUTEX(tracing_err_log_lock);
7236 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr)
7238 struct tracing_log_err *err;
7240 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7241 err = kzalloc(sizeof(*err), GFP_KERNEL);
7243 err = ERR_PTR(-ENOMEM);
7244 tr->n_err_log_entries++;
7249 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7250 list_del(&err->list);
7256 * err_pos - find the position of a string within a command for error careting
7257 * @cmd: The tracing command that caused the error
7258 * @str: The string to position the caret at within @cmd
7260 * Finds the position of the first occurence of @str within @cmd. The
7261 * return value can be passed to tracing_log_err() for caret placement
7264 * Returns the index within @cmd of the first occurence of @str or 0
7265 * if @str was not found.
7267 unsigned int err_pos(char *cmd, const char *str)
7271 if (WARN_ON(!strlen(cmd)))
7274 found = strstr(cmd, str);
7282 * tracing_log_err - write an error to the tracing error log
7283 * @tr: The associated trace array for the error (NULL for top level array)
7284 * @loc: A string describing where the error occurred
7285 * @cmd: The tracing command that caused the error
7286 * @errs: The array of loc-specific static error strings
7287 * @type: The index into errs[], which produces the specific static err string
7288 * @pos: The position the caret should be placed in the cmd
7290 * Writes an error into tracing/error_log of the form:
7292 * <loc>: error: <text>
7296 * tracing/error_log is a small log file containing the last
7297 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7298 * unless there has been a tracing error, and the error log can be
7299 * cleared and have its memory freed by writing the empty string in
7300 * truncation mode to it i.e. echo > tracing/error_log.
7302 * NOTE: the @errs array along with the @type param are used to
7303 * produce a static error string - this string is not copied and saved
7304 * when the error is logged - only a pointer to it is saved. See
7305 * existing callers for examples of how static strings are typically
7306 * defined for use with tracing_log_err().
7308 void tracing_log_err(struct trace_array *tr,
7309 const char *loc, const char *cmd,
7310 const char **errs, u8 type, u8 pos)
7312 struct tracing_log_err *err;
7317 mutex_lock(&tracing_err_log_lock);
7318 err = get_tracing_log_err(tr);
7319 if (PTR_ERR(err) == -ENOMEM) {
7320 mutex_unlock(&tracing_err_log_lock);
7324 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7325 snprintf(err->cmd, MAX_FILTER_STR_VAL,"\n" CMD_PREFIX "%s\n", cmd);
7327 err->info.errs = errs;
7328 err->info.type = type;
7329 err->info.pos = pos;
7330 err->info.ts = local_clock();
7332 list_add_tail(&err->list, &tr->err_log);
7333 mutex_unlock(&tracing_err_log_lock);
7336 static void clear_tracing_err_log(struct trace_array *tr)
7338 struct tracing_log_err *err, *next;
7340 mutex_lock(&tracing_err_log_lock);
7341 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7342 list_del(&err->list);
7346 tr->n_err_log_entries = 0;
7347 mutex_unlock(&tracing_err_log_lock);
7350 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7352 struct trace_array *tr = m->private;
7354 mutex_lock(&tracing_err_log_lock);
7356 return seq_list_start(&tr->err_log, *pos);
7359 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7361 struct trace_array *tr = m->private;
7363 return seq_list_next(v, &tr->err_log, pos);
7366 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7368 mutex_unlock(&tracing_err_log_lock);
7371 static void tracing_err_log_show_pos(struct seq_file *m, u8 pos)
7375 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7377 for (i = 0; i < pos; i++)
7382 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7384 struct tracing_log_err *err = v;
7387 const char *err_text = err->info.errs[err->info.type];
7388 u64 sec = err->info.ts;
7391 nsec = do_div(sec, NSEC_PER_SEC);
7392 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7393 err->loc, err_text);
7394 seq_printf(m, "%s", err->cmd);
7395 tracing_err_log_show_pos(m, err->info.pos);
7401 static const struct seq_operations tracing_err_log_seq_ops = {
7402 .start = tracing_err_log_seq_start,
7403 .next = tracing_err_log_seq_next,
7404 .stop = tracing_err_log_seq_stop,
7405 .show = tracing_err_log_seq_show
7408 static int tracing_err_log_open(struct inode *inode, struct file *file)
7410 struct trace_array *tr = inode->i_private;
7413 ret = tracing_check_open_get_tr(tr);
7417 /* If this file was opened for write, then erase contents */
7418 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7419 clear_tracing_err_log(tr);
7421 if (file->f_mode & FMODE_READ) {
7422 ret = seq_open(file, &tracing_err_log_seq_ops);
7424 struct seq_file *m = file->private_data;
7427 trace_array_put(tr);
7433 static ssize_t tracing_err_log_write(struct file *file,
7434 const char __user *buffer,
7435 size_t count, loff_t *ppos)
7440 static int tracing_err_log_release(struct inode *inode, struct file *file)
7442 struct trace_array *tr = inode->i_private;
7444 trace_array_put(tr);
7446 if (file->f_mode & FMODE_READ)
7447 seq_release(inode, file);
7452 static const struct file_operations tracing_err_log_fops = {
7453 .open = tracing_err_log_open,
7454 .write = tracing_err_log_write,
7456 .llseek = seq_lseek,
7457 .release = tracing_err_log_release,
7460 static int tracing_buffers_open(struct inode *inode, struct file *filp)
7462 struct trace_array *tr = inode->i_private;
7463 struct ftrace_buffer_info *info;
7466 ret = tracing_check_open_get_tr(tr);
7470 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7472 trace_array_put(tr);
7476 mutex_lock(&trace_types_lock);
7479 info->iter.cpu_file = tracing_get_cpu(inode);
7480 info->iter.trace = tr->current_trace;
7481 info->iter.array_buffer = &tr->array_buffer;
7483 /* Force reading ring buffer for first read */
7484 info->read = (unsigned int)-1;
7486 filp->private_data = info;
7490 mutex_unlock(&trace_types_lock);
7492 ret = nonseekable_open(inode, filp);
7494 trace_array_put(tr);
7500 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7502 struct ftrace_buffer_info *info = filp->private_data;
7503 struct trace_iterator *iter = &info->iter;
7505 return trace_poll(iter, filp, poll_table);
7509 tracing_buffers_read(struct file *filp, char __user *ubuf,
7510 size_t count, loff_t *ppos)
7512 struct ftrace_buffer_info *info = filp->private_data;
7513 struct trace_iterator *iter = &info->iter;
7520 #ifdef CONFIG_TRACER_MAX_TRACE
7521 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7526 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
7528 if (IS_ERR(info->spare)) {
7529 ret = PTR_ERR(info->spare);
7532 info->spare_cpu = iter->cpu_file;
7538 /* Do we have previous read data to read? */
7539 if (info->read < PAGE_SIZE)
7543 trace_access_lock(iter->cpu_file);
7544 ret = ring_buffer_read_page(iter->array_buffer->buffer,
7548 trace_access_unlock(iter->cpu_file);
7551 if (trace_empty(iter)) {
7552 if ((filp->f_flags & O_NONBLOCK))
7555 ret = wait_on_pipe(iter, 0);
7566 size = PAGE_SIZE - info->read;
7570 ret = copy_to_user(ubuf, info->spare + info->read, size);
7582 static int tracing_buffers_release(struct inode *inode, struct file *file)
7584 struct ftrace_buffer_info *info = file->private_data;
7585 struct trace_iterator *iter = &info->iter;
7587 mutex_lock(&trace_types_lock);
7589 iter->tr->trace_ref--;
7591 __trace_array_put(iter->tr);
7594 ring_buffer_free_read_page(iter->array_buffer->buffer,
7595 info->spare_cpu, info->spare);
7598 mutex_unlock(&trace_types_lock);
7604 struct trace_buffer *buffer;
7607 refcount_t refcount;
7610 static void buffer_ref_release(struct buffer_ref *ref)
7612 if (!refcount_dec_and_test(&ref->refcount))
7614 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
7618 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
7619 struct pipe_buffer *buf)
7621 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
7623 buffer_ref_release(ref);
7627 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
7628 struct pipe_buffer *buf)
7630 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
7632 if (refcount_read(&ref->refcount) > INT_MAX/2)
7635 refcount_inc(&ref->refcount);
7639 /* Pipe buffer operations for a buffer. */
7640 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
7641 .release = buffer_pipe_buf_release,
7642 .get = buffer_pipe_buf_get,
7646 * Callback from splice_to_pipe(), if we need to release some pages
7647 * at the end of the spd in case we error'ed out in filling the pipe.
7649 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
7651 struct buffer_ref *ref =
7652 (struct buffer_ref *)spd->partial[i].private;
7654 buffer_ref_release(ref);
7655 spd->partial[i].private = 0;
7659 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7660 struct pipe_inode_info *pipe, size_t len,
7663 struct ftrace_buffer_info *info = file->private_data;
7664 struct trace_iterator *iter = &info->iter;
7665 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7666 struct page *pages_def[PIPE_DEF_BUFFERS];
7667 struct splice_pipe_desc spd = {
7669 .partial = partial_def,
7670 .nr_pages_max = PIPE_DEF_BUFFERS,
7671 .ops = &buffer_pipe_buf_ops,
7672 .spd_release = buffer_spd_release,
7674 struct buffer_ref *ref;
7678 #ifdef CONFIG_TRACER_MAX_TRACE
7679 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
7683 if (*ppos & (PAGE_SIZE - 1))
7686 if (len & (PAGE_SIZE - 1)) {
7687 if (len < PAGE_SIZE)
7692 if (splice_grow_spd(pipe, &spd))
7696 trace_access_lock(iter->cpu_file);
7697 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
7699 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
7703 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
7709 refcount_set(&ref->refcount, 1);
7710 ref->buffer = iter->array_buffer->buffer;
7711 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
7712 if (IS_ERR(ref->page)) {
7713 ret = PTR_ERR(ref->page);
7718 ref->cpu = iter->cpu_file;
7720 r = ring_buffer_read_page(ref->buffer, &ref->page,
7721 len, iter->cpu_file, 1);
7723 ring_buffer_free_read_page(ref->buffer, ref->cpu,
7729 page = virt_to_page(ref->page);
7731 spd.pages[i] = page;
7732 spd.partial[i].len = PAGE_SIZE;
7733 spd.partial[i].offset = 0;
7734 spd.partial[i].private = (unsigned long)ref;
7738 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
7741 trace_access_unlock(iter->cpu_file);
7744 /* did we read anything? */
7745 if (!spd.nr_pages) {
7750 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
7753 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
7760 ret = splice_to_pipe(pipe, &spd);
7762 splice_shrink_spd(&spd);
7767 static const struct file_operations tracing_buffers_fops = {
7768 .open = tracing_buffers_open,
7769 .read = tracing_buffers_read,
7770 .poll = tracing_buffers_poll,
7771 .release = tracing_buffers_release,
7772 .splice_read = tracing_buffers_splice_read,
7773 .llseek = no_llseek,
7777 tracing_stats_read(struct file *filp, char __user *ubuf,
7778 size_t count, loff_t *ppos)
7780 struct inode *inode = file_inode(filp);
7781 struct trace_array *tr = inode->i_private;
7782 struct array_buffer *trace_buf = &tr->array_buffer;
7783 int cpu = tracing_get_cpu(inode);
7784 struct trace_seq *s;
7786 unsigned long long t;
7787 unsigned long usec_rem;
7789 s = kmalloc(sizeof(*s), GFP_KERNEL);
7795 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
7796 trace_seq_printf(s, "entries: %ld\n", cnt);
7798 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
7799 trace_seq_printf(s, "overrun: %ld\n", cnt);
7801 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
7802 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
7804 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
7805 trace_seq_printf(s, "bytes: %ld\n", cnt);
7807 if (trace_clocks[tr->clock_id].in_ns) {
7808 /* local or global for trace_clock */
7809 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
7810 usec_rem = do_div(t, USEC_PER_SEC);
7811 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
7814 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
7815 usec_rem = do_div(t, USEC_PER_SEC);
7816 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
7818 /* counter or tsc mode for trace_clock */
7819 trace_seq_printf(s, "oldest event ts: %llu\n",
7820 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
7822 trace_seq_printf(s, "now ts: %llu\n",
7823 ring_buffer_time_stamp(trace_buf->buffer, cpu));
7826 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
7827 trace_seq_printf(s, "dropped events: %ld\n", cnt);
7829 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
7830 trace_seq_printf(s, "read events: %ld\n", cnt);
7832 count = simple_read_from_buffer(ubuf, count, ppos,
7833 s->buffer, trace_seq_used(s));
7840 static const struct file_operations tracing_stats_fops = {
7841 .open = tracing_open_generic_tr,
7842 .read = tracing_stats_read,
7843 .llseek = generic_file_llseek,
7844 .release = tracing_release_generic_tr,
7847 #ifdef CONFIG_DYNAMIC_FTRACE
7850 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
7851 size_t cnt, loff_t *ppos)
7857 /* 256 should be plenty to hold the amount needed */
7858 buf = kmalloc(256, GFP_KERNEL);
7862 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
7863 ftrace_update_tot_cnt,
7864 ftrace_number_of_pages,
7865 ftrace_number_of_groups);
7867 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7872 static const struct file_operations tracing_dyn_info_fops = {
7873 .open = tracing_open_generic,
7874 .read = tracing_read_dyn_info,
7875 .llseek = generic_file_llseek,
7877 #endif /* CONFIG_DYNAMIC_FTRACE */
7879 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
7881 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
7882 struct trace_array *tr, struct ftrace_probe_ops *ops,
7885 tracing_snapshot_instance(tr);
7889 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
7890 struct trace_array *tr, struct ftrace_probe_ops *ops,
7893 struct ftrace_func_mapper *mapper = data;
7897 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7907 tracing_snapshot_instance(tr);
7911 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
7912 struct ftrace_probe_ops *ops, void *data)
7914 struct ftrace_func_mapper *mapper = data;
7917 seq_printf(m, "%ps:", (void *)ip);
7919 seq_puts(m, "snapshot");
7922 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
7925 seq_printf(m, ":count=%ld\n", *count);
7927 seq_puts(m, ":unlimited\n");
7933 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
7934 unsigned long ip, void *init_data, void **data)
7936 struct ftrace_func_mapper *mapper = *data;
7939 mapper = allocate_ftrace_func_mapper();
7945 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
7949 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
7950 unsigned long ip, void *data)
7952 struct ftrace_func_mapper *mapper = data;
7957 free_ftrace_func_mapper(mapper, NULL);
7961 ftrace_func_mapper_remove_ip(mapper, ip);
7964 static struct ftrace_probe_ops snapshot_probe_ops = {
7965 .func = ftrace_snapshot,
7966 .print = ftrace_snapshot_print,
7969 static struct ftrace_probe_ops snapshot_count_probe_ops = {
7970 .func = ftrace_count_snapshot,
7971 .print = ftrace_snapshot_print,
7972 .init = ftrace_snapshot_init,
7973 .free = ftrace_snapshot_free,
7977 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
7978 char *glob, char *cmd, char *param, int enable)
7980 struct ftrace_probe_ops *ops;
7981 void *count = (void *)-1;
7988 /* hash funcs only work with set_ftrace_filter */
7992 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
7995 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8000 number = strsep(¶m, ":");
8002 if (!strlen(number))
8006 * We use the callback data field (which is a pointer)
8009 ret = kstrtoul(number, 0, (unsigned long *)&count);
8014 ret = tracing_alloc_snapshot_instance(tr);
8018 ret = register_ftrace_function_probe(glob, tr, ops, count);
8021 return ret < 0 ? ret : 0;
8024 static struct ftrace_func_command ftrace_snapshot_cmd = {
8026 .func = ftrace_trace_snapshot_callback,
8029 static __init int register_snapshot_cmd(void)
8031 return register_ftrace_command(&ftrace_snapshot_cmd);
8034 static inline __init int register_snapshot_cmd(void) { return 0; }
8035 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8037 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8039 if (WARN_ON(!tr->dir))
8040 return ERR_PTR(-ENODEV);
8042 /* Top directory uses NULL as the parent */
8043 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8046 /* All sub buffers have a descriptor */
8050 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8052 struct dentry *d_tracer;
8055 return tr->percpu_dir;
8057 d_tracer = tracing_get_dentry(tr);
8058 if (IS_ERR(d_tracer))
8061 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8063 MEM_FAIL(!tr->percpu_dir,
8064 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8066 return tr->percpu_dir;
8069 static struct dentry *
8070 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8071 void *data, long cpu, const struct file_operations *fops)
8073 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8075 if (ret) /* See tracing_get_cpu() */
8076 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8081 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8083 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8084 struct dentry *d_cpu;
8085 char cpu_dir[30]; /* 30 characters should be more than enough */
8090 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8091 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8093 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8097 /* per cpu trace_pipe */
8098 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
8099 tr, cpu, &tracing_pipe_fops);
8102 trace_create_cpu_file("trace", 0644, d_cpu,
8103 tr, cpu, &tracing_fops);
8105 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
8106 tr, cpu, &tracing_buffers_fops);
8108 trace_create_cpu_file("stats", 0444, d_cpu,
8109 tr, cpu, &tracing_stats_fops);
8111 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
8112 tr, cpu, &tracing_entries_fops);
8114 #ifdef CONFIG_TRACER_SNAPSHOT
8115 trace_create_cpu_file("snapshot", 0644, d_cpu,
8116 tr, cpu, &snapshot_fops);
8118 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
8119 tr, cpu, &snapshot_raw_fops);
8123 #ifdef CONFIG_FTRACE_SELFTEST
8124 /* Let selftest have access to static functions in this file */
8125 #include "trace_selftest.c"
8129 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8132 struct trace_option_dentry *topt = filp->private_data;
8135 if (topt->flags->val & topt->opt->bit)
8140 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8144 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8147 struct trace_option_dentry *topt = filp->private_data;
8151 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8155 if (val != 0 && val != 1)
8158 if (!!(topt->flags->val & topt->opt->bit) != val) {
8159 mutex_lock(&trace_types_lock);
8160 ret = __set_tracer_option(topt->tr, topt->flags,
8162 mutex_unlock(&trace_types_lock);
8173 static const struct file_operations trace_options_fops = {
8174 .open = tracing_open_generic,
8175 .read = trace_options_read,
8176 .write = trace_options_write,
8177 .llseek = generic_file_llseek,
8181 * In order to pass in both the trace_array descriptor as well as the index
8182 * to the flag that the trace option file represents, the trace_array
8183 * has a character array of trace_flags_index[], which holds the index
8184 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8185 * The address of this character array is passed to the flag option file
8186 * read/write callbacks.
8188 * In order to extract both the index and the trace_array descriptor,
8189 * get_tr_index() uses the following algorithm.
8193 * As the pointer itself contains the address of the index (remember
8196 * Then to get the trace_array descriptor, by subtracting that index
8197 * from the ptr, we get to the start of the index itself.
8199 * ptr - idx == &index[0]
8201 * Then a simple container_of() from that pointer gets us to the
8202 * trace_array descriptor.
8204 static void get_tr_index(void *data, struct trace_array **ptr,
8205 unsigned int *pindex)
8207 *pindex = *(unsigned char *)data;
8209 *ptr = container_of(data - *pindex, struct trace_array,
8214 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8217 void *tr_index = filp->private_data;
8218 struct trace_array *tr;
8222 get_tr_index(tr_index, &tr, &index);
8224 if (tr->trace_flags & (1 << index))
8229 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8233 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8236 void *tr_index = filp->private_data;
8237 struct trace_array *tr;
8242 get_tr_index(tr_index, &tr, &index);
8244 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8248 if (val != 0 && val != 1)
8251 mutex_lock(&event_mutex);
8252 mutex_lock(&trace_types_lock);
8253 ret = set_tracer_flag(tr, 1 << index, val);
8254 mutex_unlock(&trace_types_lock);
8255 mutex_unlock(&event_mutex);
8265 static const struct file_operations trace_options_core_fops = {
8266 .open = tracing_open_generic,
8267 .read = trace_options_core_read,
8268 .write = trace_options_core_write,
8269 .llseek = generic_file_llseek,
8272 struct dentry *trace_create_file(const char *name,
8274 struct dentry *parent,
8276 const struct file_operations *fops)
8280 ret = tracefs_create_file(name, mode, parent, data, fops);
8282 pr_warn("Could not create tracefs '%s' entry\n", name);
8288 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8290 struct dentry *d_tracer;
8295 d_tracer = tracing_get_dentry(tr);
8296 if (IS_ERR(d_tracer))
8299 tr->options = tracefs_create_dir("options", d_tracer);
8301 pr_warn("Could not create tracefs directory 'options'\n");
8309 create_trace_option_file(struct trace_array *tr,
8310 struct trace_option_dentry *topt,
8311 struct tracer_flags *flags,
8312 struct tracer_opt *opt)
8314 struct dentry *t_options;
8316 t_options = trace_options_init_dentry(tr);
8320 topt->flags = flags;
8324 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
8325 &trace_options_fops);
8330 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8332 struct trace_option_dentry *topts;
8333 struct trace_options *tr_topts;
8334 struct tracer_flags *flags;
8335 struct tracer_opt *opts;
8342 flags = tracer->flags;
8344 if (!flags || !flags->opts)
8348 * If this is an instance, only create flags for tracers
8349 * the instance may have.
8351 if (!trace_ok_for_array(tracer, tr))
8354 for (i = 0; i < tr->nr_topts; i++) {
8355 /* Make sure there's no duplicate flags. */
8356 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8362 for (cnt = 0; opts[cnt].name; cnt++)
8365 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8369 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8376 tr->topts = tr_topts;
8377 tr->topts[tr->nr_topts].tracer = tracer;
8378 tr->topts[tr->nr_topts].topts = topts;
8381 for (cnt = 0; opts[cnt].name; cnt++) {
8382 create_trace_option_file(tr, &topts[cnt], flags,
8384 MEM_FAIL(topts[cnt].entry == NULL,
8385 "Failed to create trace option: %s",
8390 static struct dentry *
8391 create_trace_option_core_file(struct trace_array *tr,
8392 const char *option, long index)
8394 struct dentry *t_options;
8396 t_options = trace_options_init_dentry(tr);
8400 return trace_create_file(option, 0644, t_options,
8401 (void *)&tr->trace_flags_index[index],
8402 &trace_options_core_fops);
8405 static void create_trace_options_dir(struct trace_array *tr)
8407 struct dentry *t_options;
8408 bool top_level = tr == &global_trace;
8411 t_options = trace_options_init_dentry(tr);
8415 for (i = 0; trace_options[i]; i++) {
8417 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8418 create_trace_option_core_file(tr, trace_options[i], i);
8423 rb_simple_read(struct file *filp, char __user *ubuf,
8424 size_t cnt, loff_t *ppos)
8426 struct trace_array *tr = filp->private_data;
8430 r = tracer_tracing_is_on(tr);
8431 r = sprintf(buf, "%d\n", r);
8433 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8437 rb_simple_write(struct file *filp, const char __user *ubuf,
8438 size_t cnt, loff_t *ppos)
8440 struct trace_array *tr = filp->private_data;
8441 struct trace_buffer *buffer = tr->array_buffer.buffer;
8445 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8450 mutex_lock(&trace_types_lock);
8451 if (!!val == tracer_tracing_is_on(tr)) {
8452 val = 0; /* do nothing */
8454 tracer_tracing_on(tr);
8455 if (tr->current_trace->start)
8456 tr->current_trace->start(tr);
8458 tracer_tracing_off(tr);
8459 if (tr->current_trace->stop)
8460 tr->current_trace->stop(tr);
8462 mutex_unlock(&trace_types_lock);
8470 static const struct file_operations rb_simple_fops = {
8471 .open = tracing_open_generic_tr,
8472 .read = rb_simple_read,
8473 .write = rb_simple_write,
8474 .release = tracing_release_generic_tr,
8475 .llseek = default_llseek,
8479 buffer_percent_read(struct file *filp, char __user *ubuf,
8480 size_t cnt, loff_t *ppos)
8482 struct trace_array *tr = filp->private_data;
8486 r = tr->buffer_percent;
8487 r = sprintf(buf, "%d\n", r);
8489 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8493 buffer_percent_write(struct file *filp, const char __user *ubuf,
8494 size_t cnt, loff_t *ppos)
8496 struct trace_array *tr = filp->private_data;
8500 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8510 tr->buffer_percent = val;
8517 static const struct file_operations buffer_percent_fops = {
8518 .open = tracing_open_generic_tr,
8519 .read = buffer_percent_read,
8520 .write = buffer_percent_write,
8521 .release = tracing_release_generic_tr,
8522 .llseek = default_llseek,
8525 static struct dentry *trace_instance_dir;
8528 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
8531 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
8533 enum ring_buffer_flags rb_flags;
8535 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
8539 buf->buffer = ring_buffer_alloc(size, rb_flags);
8543 buf->data = alloc_percpu(struct trace_array_cpu);
8545 ring_buffer_free(buf->buffer);
8550 /* Allocate the first page for all buffers */
8551 set_buffer_entries(&tr->array_buffer,
8552 ring_buffer_size(tr->array_buffer.buffer, 0));
8557 static int allocate_trace_buffers(struct trace_array *tr, int size)
8561 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
8565 #ifdef CONFIG_TRACER_MAX_TRACE
8566 ret = allocate_trace_buffer(tr, &tr->max_buffer,
8567 allocate_snapshot ? size : 1);
8568 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
8569 ring_buffer_free(tr->array_buffer.buffer);
8570 tr->array_buffer.buffer = NULL;
8571 free_percpu(tr->array_buffer.data);
8572 tr->array_buffer.data = NULL;
8575 tr->allocated_snapshot = allocate_snapshot;
8578 * Only the top level trace array gets its snapshot allocated
8579 * from the kernel command line.
8581 allocate_snapshot = false;
8587 static void free_trace_buffer(struct array_buffer *buf)
8590 ring_buffer_free(buf->buffer);
8592 free_percpu(buf->data);
8597 static void free_trace_buffers(struct trace_array *tr)
8602 free_trace_buffer(&tr->array_buffer);
8604 #ifdef CONFIG_TRACER_MAX_TRACE
8605 free_trace_buffer(&tr->max_buffer);
8609 static void init_trace_flags_index(struct trace_array *tr)
8613 /* Used by the trace options files */
8614 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
8615 tr->trace_flags_index[i] = i;
8618 static void __update_tracer_options(struct trace_array *tr)
8622 for (t = trace_types; t; t = t->next)
8623 add_tracer_options(tr, t);
8626 static void update_tracer_options(struct trace_array *tr)
8628 mutex_lock(&trace_types_lock);
8629 __update_tracer_options(tr);
8630 mutex_unlock(&trace_types_lock);
8633 /* Must have trace_types_lock held */
8634 struct trace_array *trace_array_find(const char *instance)
8636 struct trace_array *tr, *found = NULL;
8638 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8639 if (tr->name && strcmp(tr->name, instance) == 0) {
8648 struct trace_array *trace_array_find_get(const char *instance)
8650 struct trace_array *tr;
8652 mutex_lock(&trace_types_lock);
8653 tr = trace_array_find(instance);
8656 mutex_unlock(&trace_types_lock);
8661 static struct trace_array *trace_array_create(const char *name)
8663 struct trace_array *tr;
8667 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
8669 return ERR_PTR(ret);
8671 tr->name = kstrdup(name, GFP_KERNEL);
8675 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
8678 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
8680 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
8682 raw_spin_lock_init(&tr->start_lock);
8684 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
8686 tr->current_trace = &nop_trace;
8688 INIT_LIST_HEAD(&tr->systems);
8689 INIT_LIST_HEAD(&tr->events);
8690 INIT_LIST_HEAD(&tr->hist_vars);
8691 INIT_LIST_HEAD(&tr->err_log);
8693 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
8696 tr->dir = tracefs_create_dir(name, trace_instance_dir);
8700 ret = event_trace_add_tracer(tr->dir, tr);
8702 tracefs_remove(tr->dir);
8706 ftrace_init_trace_array(tr);
8708 init_tracer_tracefs(tr, tr->dir);
8709 init_trace_flags_index(tr);
8710 __update_tracer_options(tr);
8712 list_add(&tr->list, &ftrace_trace_arrays);
8720 free_trace_buffers(tr);
8721 free_cpumask_var(tr->tracing_cpumask);
8725 return ERR_PTR(ret);
8728 static int instance_mkdir(const char *name)
8730 struct trace_array *tr;
8733 mutex_lock(&event_mutex);
8734 mutex_lock(&trace_types_lock);
8737 if (trace_array_find(name))
8740 tr = trace_array_create(name);
8742 ret = PTR_ERR_OR_ZERO(tr);
8745 mutex_unlock(&trace_types_lock);
8746 mutex_unlock(&event_mutex);
8751 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
8752 * @name: The name of the trace array to be looked up/created.
8754 * Returns pointer to trace array with given name.
8755 * NULL, if it cannot be created.
8757 * NOTE: This function increments the reference counter associated with the
8758 * trace array returned. This makes sure it cannot be freed while in use.
8759 * Use trace_array_put() once the trace array is no longer needed.
8760 * If the trace_array is to be freed, trace_array_destroy() needs to
8761 * be called after the trace_array_put(), or simply let user space delete
8762 * it from the tracefs instances directory. But until the
8763 * trace_array_put() is called, user space can not delete it.
8766 struct trace_array *trace_array_get_by_name(const char *name)
8768 struct trace_array *tr;
8770 mutex_lock(&event_mutex);
8771 mutex_lock(&trace_types_lock);
8773 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8774 if (tr->name && strcmp(tr->name, name) == 0)
8778 tr = trace_array_create(name);
8786 mutex_unlock(&trace_types_lock);
8787 mutex_unlock(&event_mutex);
8790 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
8792 static int __remove_instance(struct trace_array *tr)
8796 /* Reference counter for a newly created trace array = 1. */
8797 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
8800 list_del(&tr->list);
8802 /* Disable all the flags that were enabled coming in */
8803 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
8804 if ((1 << i) & ZEROED_TRACE_FLAGS)
8805 set_tracer_flag(tr, 1 << i, 0);
8808 tracing_set_nop(tr);
8809 clear_ftrace_function_probes(tr);
8810 event_trace_del_tracer(tr);
8811 ftrace_clear_pids(tr);
8812 ftrace_destroy_function_files(tr);
8813 tracefs_remove(tr->dir);
8814 free_trace_buffers(tr);
8816 for (i = 0; i < tr->nr_topts; i++) {
8817 kfree(tr->topts[i].topts);
8821 free_cpumask_var(tr->tracing_cpumask);
8829 int trace_array_destroy(struct trace_array *this_tr)
8831 struct trace_array *tr;
8837 mutex_lock(&event_mutex);
8838 mutex_lock(&trace_types_lock);
8842 /* Making sure trace array exists before destroying it. */
8843 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
8844 if (tr == this_tr) {
8845 ret = __remove_instance(tr);
8850 mutex_unlock(&trace_types_lock);
8851 mutex_unlock(&event_mutex);
8855 EXPORT_SYMBOL_GPL(trace_array_destroy);
8857 static int instance_rmdir(const char *name)
8859 struct trace_array *tr;
8862 mutex_lock(&event_mutex);
8863 mutex_lock(&trace_types_lock);
8866 tr = trace_array_find(name);
8868 ret = __remove_instance(tr);
8870 mutex_unlock(&trace_types_lock);
8871 mutex_unlock(&event_mutex);
8876 static __init void create_trace_instances(struct dentry *d_tracer)
8878 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
8881 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
8886 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
8888 struct trace_event_file *file;
8891 trace_create_file("available_tracers", 0444, d_tracer,
8892 tr, &show_traces_fops);
8894 trace_create_file("current_tracer", 0644, d_tracer,
8895 tr, &set_tracer_fops);
8897 trace_create_file("tracing_cpumask", 0644, d_tracer,
8898 tr, &tracing_cpumask_fops);
8900 trace_create_file("trace_options", 0644, d_tracer,
8901 tr, &tracing_iter_fops);
8903 trace_create_file("trace", 0644, d_tracer,
8906 trace_create_file("trace_pipe", 0444, d_tracer,
8907 tr, &tracing_pipe_fops);
8909 trace_create_file("buffer_size_kb", 0644, d_tracer,
8910 tr, &tracing_entries_fops);
8912 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
8913 tr, &tracing_total_entries_fops);
8915 trace_create_file("free_buffer", 0200, d_tracer,
8916 tr, &tracing_free_buffer_fops);
8918 trace_create_file("trace_marker", 0220, d_tracer,
8919 tr, &tracing_mark_fops);
8921 file = __find_event_file(tr, "ftrace", "print");
8922 if (file && file->dir)
8923 trace_create_file("trigger", 0644, file->dir, file,
8924 &event_trigger_fops);
8925 tr->trace_marker_file = file;
8927 trace_create_file("trace_marker_raw", 0220, d_tracer,
8928 tr, &tracing_mark_raw_fops);
8930 trace_create_file("trace_clock", 0644, d_tracer, tr,
8933 trace_create_file("tracing_on", 0644, d_tracer,
8934 tr, &rb_simple_fops);
8936 trace_create_file("timestamp_mode", 0444, d_tracer, tr,
8937 &trace_time_stamp_mode_fops);
8939 tr->buffer_percent = 50;
8941 trace_create_file("buffer_percent", 0444, d_tracer,
8942 tr, &buffer_percent_fops);
8944 create_trace_options_dir(tr);
8946 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
8947 trace_create_maxlat_file(tr, d_tracer);
8950 if (ftrace_create_function_files(tr, d_tracer))
8951 MEM_FAIL(1, "Could not allocate function filter files");
8953 #ifdef CONFIG_TRACER_SNAPSHOT
8954 trace_create_file("snapshot", 0644, d_tracer,
8955 tr, &snapshot_fops);
8958 trace_create_file("error_log", 0644, d_tracer,
8959 tr, &tracing_err_log_fops);
8961 for_each_tracing_cpu(cpu)
8962 tracing_init_tracefs_percpu(tr, cpu);
8964 ftrace_init_tracefs(tr, d_tracer);
8967 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
8969 struct vfsmount *mnt;
8970 struct file_system_type *type;
8973 * To maintain backward compatibility for tools that mount
8974 * debugfs to get to the tracing facility, tracefs is automatically
8975 * mounted to the debugfs/tracing directory.
8977 type = get_fs_type("tracefs");
8980 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
8981 put_filesystem(type);
8990 * tracing_init_dentry - initialize top level trace array
8992 * This is called when creating files or directories in the tracing
8993 * directory. It is called via fs_initcall() by any of the boot up code
8994 * and expects to return the dentry of the top level tracing directory.
8996 struct dentry *tracing_init_dentry(void)
8998 struct trace_array *tr = &global_trace;
9000 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9001 pr_warn("Tracing disabled due to lockdown\n");
9002 return ERR_PTR(-EPERM);
9005 /* The top level trace array uses NULL as parent */
9009 if (WARN_ON(!tracefs_initialized()))
9010 return ERR_PTR(-ENODEV);
9013 * As there may still be users that expect the tracing
9014 * files to exist in debugfs/tracing, we must automount
9015 * the tracefs file system there, so older tools still
9016 * work with the newer kerenl.
9018 tr->dir = debugfs_create_automount("tracing", NULL,
9019 trace_automount, NULL);
9024 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9025 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9027 static void __init trace_eval_init(void)
9031 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9032 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9035 #ifdef CONFIG_MODULES
9036 static void trace_module_add_evals(struct module *mod)
9038 if (!mod->num_trace_evals)
9042 * Modules with bad taint do not have events created, do
9043 * not bother with enums either.
9045 if (trace_module_has_bad_taint(mod))
9048 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9051 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
9052 static void trace_module_remove_evals(struct module *mod)
9054 union trace_eval_map_item *map;
9055 union trace_eval_map_item **last = &trace_eval_maps;
9057 if (!mod->num_trace_evals)
9060 mutex_lock(&trace_eval_mutex);
9062 map = trace_eval_maps;
9065 if (map->head.mod == mod)
9067 map = trace_eval_jmp_to_tail(map);
9068 last = &map->tail.next;
9069 map = map->tail.next;
9074 *last = trace_eval_jmp_to_tail(map)->tail.next;
9077 mutex_unlock(&trace_eval_mutex);
9080 static inline void trace_module_remove_evals(struct module *mod) { }
9081 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9083 static int trace_module_notify(struct notifier_block *self,
9084 unsigned long val, void *data)
9086 struct module *mod = data;
9089 case MODULE_STATE_COMING:
9090 trace_module_add_evals(mod);
9092 case MODULE_STATE_GOING:
9093 trace_module_remove_evals(mod);
9100 static struct notifier_block trace_module_nb = {
9101 .notifier_call = trace_module_notify,
9104 #endif /* CONFIG_MODULES */
9106 static __init int tracer_init_tracefs(void)
9108 struct dentry *d_tracer;
9110 trace_access_lock_init();
9112 d_tracer = tracing_init_dentry();
9113 if (IS_ERR(d_tracer))
9118 init_tracer_tracefs(&global_trace, d_tracer);
9119 ftrace_init_tracefs_toplevel(&global_trace, d_tracer);
9121 trace_create_file("tracing_thresh", 0644, d_tracer,
9122 &global_trace, &tracing_thresh_fops);
9124 trace_create_file("README", 0444, d_tracer,
9125 NULL, &tracing_readme_fops);
9127 trace_create_file("saved_cmdlines", 0444, d_tracer,
9128 NULL, &tracing_saved_cmdlines_fops);
9130 trace_create_file("saved_cmdlines_size", 0644, d_tracer,
9131 NULL, &tracing_saved_cmdlines_size_fops);
9133 trace_create_file("saved_tgids", 0444, d_tracer,
9134 NULL, &tracing_saved_tgids_fops);
9138 trace_create_eval_file(d_tracer);
9140 #ifdef CONFIG_MODULES
9141 register_module_notifier(&trace_module_nb);
9144 #ifdef CONFIG_DYNAMIC_FTRACE
9145 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
9146 NULL, &tracing_dyn_info_fops);
9149 create_trace_instances(d_tracer);
9151 update_tracer_options(&global_trace);
9156 static int trace_panic_handler(struct notifier_block *this,
9157 unsigned long event, void *unused)
9159 if (ftrace_dump_on_oops)
9160 ftrace_dump(ftrace_dump_on_oops);
9164 static struct notifier_block trace_panic_notifier = {
9165 .notifier_call = trace_panic_handler,
9167 .priority = 150 /* priority: INT_MAX >= x >= 0 */
9170 static int trace_die_handler(struct notifier_block *self,
9176 if (ftrace_dump_on_oops)
9177 ftrace_dump(ftrace_dump_on_oops);
9185 static struct notifier_block trace_die_notifier = {
9186 .notifier_call = trace_die_handler,
9191 * printk is set to max of 1024, we really don't need it that big.
9192 * Nothing should be printing 1000 characters anyway.
9194 #define TRACE_MAX_PRINT 1000
9197 * Define here KERN_TRACE so that we have one place to modify
9198 * it if we decide to change what log level the ftrace dump
9201 #define KERN_TRACE KERN_EMERG
9204 trace_printk_seq(struct trace_seq *s)
9206 /* Probably should print a warning here. */
9207 if (s->seq.len >= TRACE_MAX_PRINT)
9208 s->seq.len = TRACE_MAX_PRINT;
9211 * More paranoid code. Although the buffer size is set to
9212 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9213 * an extra layer of protection.
9215 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9216 s->seq.len = s->seq.size - 1;
9218 /* should be zero ended, but we are paranoid. */
9219 s->buffer[s->seq.len] = 0;
9221 printk(KERN_TRACE "%s", s->buffer);
9226 void trace_init_global_iter(struct trace_iterator *iter)
9228 iter->tr = &global_trace;
9229 iter->trace = iter->tr->current_trace;
9230 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9231 iter->array_buffer = &global_trace.array_buffer;
9233 if (iter->trace && iter->trace->open)
9234 iter->trace->open(iter);
9236 /* Annotate start of buffers if we had overruns */
9237 if (ring_buffer_overruns(iter->array_buffer->buffer))
9238 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9240 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9241 if (trace_clocks[iter->tr->clock_id].in_ns)
9242 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9245 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
9247 /* use static because iter can be a bit big for the stack */
9248 static struct trace_iterator iter;
9249 static atomic_t dump_running;
9250 struct trace_array *tr = &global_trace;
9251 unsigned int old_userobj;
9252 unsigned long flags;
9255 /* Only allow one dump user at a time. */
9256 if (atomic_inc_return(&dump_running) != 1) {
9257 atomic_dec(&dump_running);
9262 * Always turn off tracing when we dump.
9263 * We don't need to show trace output of what happens
9264 * between multiple crashes.
9266 * If the user does a sysrq-z, then they can re-enable
9267 * tracing with echo 1 > tracing_on.
9271 local_irq_save(flags);
9272 printk_nmi_direct_enter();
9274 /* Simulate the iterator */
9275 trace_init_global_iter(&iter);
9276 /* Can not use kmalloc for iter.temp */
9277 iter.temp = static_temp_buf;
9278 iter.temp_size = STATIC_TEMP_BUF_SIZE;
9280 for_each_tracing_cpu(cpu) {
9281 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9284 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
9286 /* don't look at user memory in panic mode */
9287 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
9289 switch (oops_dump_mode) {
9291 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9294 iter.cpu_file = raw_smp_processor_id();
9299 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
9300 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9303 printk(KERN_TRACE "Dumping ftrace buffer:\n");
9305 /* Did function tracer already get disabled? */
9306 if (ftrace_is_dead()) {
9307 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
9308 printk("# MAY BE MISSING FUNCTION EVENTS\n");
9312 * We need to stop all tracing on all CPUS to read the
9313 * the next buffer. This is a bit expensive, but is
9314 * not done often. We fill all what we can read,
9315 * and then release the locks again.
9318 while (!trace_empty(&iter)) {
9321 printk(KERN_TRACE "---------------------------------\n");
9325 trace_iterator_reset(&iter);
9326 iter.iter_flags |= TRACE_FILE_LAT_FMT;
9328 if (trace_find_next_entry_inc(&iter) != NULL) {
9331 ret = print_trace_line(&iter);
9332 if (ret != TRACE_TYPE_NO_CONSUME)
9333 trace_consume(&iter);
9335 touch_nmi_watchdog();
9337 trace_printk_seq(&iter.seq);
9341 printk(KERN_TRACE " (ftrace buffer empty)\n");
9343 printk(KERN_TRACE "---------------------------------\n");
9346 tr->trace_flags |= old_userobj;
9348 for_each_tracing_cpu(cpu) {
9349 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9351 atomic_dec(&dump_running);
9352 printk_nmi_direct_exit();
9353 local_irq_restore(flags);
9355 EXPORT_SYMBOL_GPL(ftrace_dump);
9357 int trace_run_command(const char *buf, int (*createfn)(int, char **))
9364 argv = argv_split(GFP_KERNEL, buf, &argc);
9369 ret = createfn(argc, argv);
9376 #define WRITE_BUFSIZE 4096
9378 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
9379 size_t count, loff_t *ppos,
9380 int (*createfn)(int, char **))
9382 char *kbuf, *buf, *tmp;
9387 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
9391 while (done < count) {
9392 size = count - done;
9394 if (size >= WRITE_BUFSIZE)
9395 size = WRITE_BUFSIZE - 1;
9397 if (copy_from_user(kbuf, buffer + done, size)) {
9404 tmp = strchr(buf, '\n');
9407 size = tmp - buf + 1;
9410 if (done + size < count) {
9413 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
9414 pr_warn("Line length is too long: Should be less than %d\n",
9422 /* Remove comments */
9423 tmp = strchr(buf, '#');
9428 ret = trace_run_command(buf, createfn);
9433 } while (done < count);
9443 __init static int tracer_alloc_buffers(void)
9449 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9450 pr_warn("Tracing disabled due to lockdown\n");
9455 * Make sure we don't accidently add more trace options
9456 * than we have bits for.
9458 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
9460 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
9463 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
9464 goto out_free_buffer_mask;
9466 /* Only allocate trace_printk buffers if a trace_printk exists */
9467 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
9468 /* Must be called before global_trace.buffer is allocated */
9469 trace_printk_init_buffers();
9471 /* To save memory, keep the ring buffer size to its minimum */
9472 if (ring_buffer_expanded)
9473 ring_buf_size = trace_buf_size;
9477 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
9478 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
9480 raw_spin_lock_init(&global_trace.start_lock);
9483 * The prepare callbacks allocates some memory for the ring buffer. We
9484 * don't free the buffer if the if the CPU goes down. If we were to free
9485 * the buffer, then the user would lose any trace that was in the
9486 * buffer. The memory will be removed once the "instance" is removed.
9488 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
9489 "trace/RB:preapre", trace_rb_cpu_prepare,
9492 goto out_free_cpumask;
9493 /* Used for event triggers */
9495 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
9497 goto out_rm_hp_state;
9499 if (trace_create_savedcmd() < 0)
9500 goto out_free_temp_buffer;
9502 /* TODO: make the number of buffers hot pluggable with CPUS */
9503 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
9504 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
9505 goto out_free_savedcmd;
9508 if (global_trace.buffer_disabled)
9511 if (trace_boot_clock) {
9512 ret = tracing_set_clock(&global_trace, trace_boot_clock);
9514 pr_warn("Trace clock %s not defined, going back to default\n",
9519 * register_tracer() might reference current_trace, so it
9520 * needs to be set before we register anything. This is
9521 * just a bootstrap of current_trace anyway.
9523 global_trace.current_trace = &nop_trace;
9525 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9527 ftrace_init_global_array_ops(&global_trace);
9529 init_trace_flags_index(&global_trace);
9531 register_tracer(&nop_trace);
9533 /* Function tracing may start here (via kernel command line) */
9534 init_function_trace();
9536 /* All seems OK, enable tracing */
9537 tracing_disabled = 0;
9539 atomic_notifier_chain_register(&panic_notifier_list,
9540 &trace_panic_notifier);
9542 register_die_notifier(&trace_die_notifier);
9544 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
9546 INIT_LIST_HEAD(&global_trace.systems);
9547 INIT_LIST_HEAD(&global_trace.events);
9548 INIT_LIST_HEAD(&global_trace.hist_vars);
9549 INIT_LIST_HEAD(&global_trace.err_log);
9550 list_add(&global_trace.list, &ftrace_trace_arrays);
9552 apply_trace_boot_options();
9554 register_snapshot_cmd();
9559 free_saved_cmdlines_buffer(savedcmd);
9560 out_free_temp_buffer:
9561 ring_buffer_free(temp_buffer);
9563 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
9565 free_cpumask_var(global_trace.tracing_cpumask);
9566 out_free_buffer_mask:
9567 free_cpumask_var(tracing_buffer_mask);
9572 void __init early_trace_init(void)
9574 if (tracepoint_printk) {
9575 tracepoint_print_iter =
9576 kmalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
9577 if (MEM_FAIL(!tracepoint_print_iter,
9578 "Failed to allocate trace iterator\n"))
9579 tracepoint_printk = 0;
9581 static_key_enable(&tracepoint_printk_key.key);
9583 tracer_alloc_buffers();
9586 void __init trace_init(void)
9591 __init static int clear_boot_tracer(void)
9594 * The default tracer at boot buffer is an init section.
9595 * This function is called in lateinit. If we did not
9596 * find the boot tracer, then clear it out, to prevent
9597 * later registration from accessing the buffer that is
9598 * about to be freed.
9600 if (!default_bootup_tracer)
9603 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
9604 default_bootup_tracer);
9605 default_bootup_tracer = NULL;
9610 fs_initcall(tracer_init_tracefs);
9611 late_initcall_sync(clear_boot_tracer);
9613 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
9614 __init static int tracing_set_default_clock(void)
9616 /* sched_clock_stable() is determined in late_initcall */
9617 if (!trace_boot_clock && !sched_clock_stable()) {
9618 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9619 pr_warn("Can not set tracing clock due to lockdown\n");
9624 "Unstable clock detected, switching default tracing clock to \"global\"\n"
9625 "If you want to keep using the local clock, then add:\n"
9626 " \"trace_clock=local\"\n"
9627 "on the kernel command line\n");
9628 tracing_set_clock(&global_trace, "global");
9633 late_initcall_sync(tracing_set_default_clock);