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/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/tracefs.h>
25 #include <linux/pagemap.h>
26 #include <linux/hardirq.h>
27 #include <linux/linkage.h>
28 #include <linux/uaccess.h>
29 #include <linux/vmalloc.h>
30 #include <linux/ftrace.h>
31 #include <linux/module.h>
32 #include <linux/percpu.h>
33 #include <linux/splice.h>
34 #include <linux/kdebug.h>
35 #include <linux/string.h>
36 #include <linux/mount.h>
37 #include <linux/rwsem.h>
38 #include <linux/slab.h>
39 #include <linux/ctype.h>
40 #include <linux/init.h>
41 #include <linux/panic_notifier.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>
52 #include <asm/setup.h> /* COMMAND_LINE_SIZE */
55 #include "trace_output.h"
58 * On boot up, the ring buffer is set to the minimum size, so that
59 * we do not waste memory on systems that are not using tracing.
61 bool ring_buffer_expanded;
64 * We need to change this state when a selftest is running.
65 * A selftest will lurk into the ring-buffer to count the
66 * entries inserted during the selftest although some concurrent
67 * insertions into the ring-buffer such as trace_printk could occurred
68 * at the same time, giving false positive or negative results.
70 static bool __read_mostly tracing_selftest_running;
73 * If boot-time tracing including tracers/events via kernel cmdline
74 * is running, we do not want to run SELFTEST.
76 bool __read_mostly tracing_selftest_disabled;
78 #ifdef CONFIG_FTRACE_STARTUP_TEST
79 void __init disable_tracing_selftest(const char *reason)
81 if (!tracing_selftest_disabled) {
82 tracing_selftest_disabled = true;
83 pr_info("Ftrace startup test is disabled due to %s\n", reason);
88 /* Pipe tracepoints to printk */
89 static struct trace_iterator *tracepoint_print_iter;
90 int tracepoint_printk;
91 static bool tracepoint_printk_stop_on_boot __initdata;
92 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
94 /* For tracers that don't implement custom flags */
95 static struct tracer_opt dummy_tracer_opt[] = {
100 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
106 * To prevent the comm cache from being overwritten when no
107 * tracing is active, only save the comm when a trace event
110 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
113 * Kill all tracing for good (never come back).
114 * It is initialized to 1 but will turn to zero if the initialization
115 * of the tracer is successful. But that is the only place that sets
118 static int tracing_disabled = 1;
120 cpumask_var_t __read_mostly tracing_buffer_mask;
123 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
125 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
126 * is set, then ftrace_dump is called. This will output the contents
127 * of the ftrace buffers to the console. This is very useful for
128 * capturing traces that lead to crashes and outputing it to a
131 * It is default off, but you can enable it with either specifying
132 * "ftrace_dump_on_oops" in the kernel command line, or setting
133 * /proc/sys/kernel/ftrace_dump_on_oops
134 * Set 1 if you want to dump buffers of all CPUs
135 * Set 2 if you want to dump the buffer of the CPU that triggered oops
138 enum ftrace_dump_mode ftrace_dump_on_oops;
140 /* When set, tracing will stop when a WARN*() is hit */
141 int __disable_trace_on_warning;
143 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
144 /* Map of enums to their values, for "eval_map" file */
145 struct trace_eval_map_head {
147 unsigned long length;
150 union trace_eval_map_item;
152 struct trace_eval_map_tail {
154 * "end" is first and points to NULL as it must be different
155 * than "mod" or "eval_string"
157 union trace_eval_map_item *next;
158 const char *end; /* points to NULL */
161 static DEFINE_MUTEX(trace_eval_mutex);
164 * The trace_eval_maps are saved in an array with two extra elements,
165 * one at the beginning, and one at the end. The beginning item contains
166 * the count of the saved maps (head.length), and the module they
167 * belong to if not built in (head.mod). The ending item contains a
168 * pointer to the next array of saved eval_map items.
170 union trace_eval_map_item {
171 struct trace_eval_map map;
172 struct trace_eval_map_head head;
173 struct trace_eval_map_tail tail;
176 static union trace_eval_map_item *trace_eval_maps;
177 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
179 int tracing_set_tracer(struct trace_array *tr, const char *buf);
180 static void ftrace_trace_userstack(struct trace_array *tr,
181 struct trace_buffer *buffer,
182 unsigned int trace_ctx);
184 #define MAX_TRACER_SIZE 100
185 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
186 static char *default_bootup_tracer;
188 static bool allocate_snapshot;
189 static bool snapshot_at_boot;
191 static char boot_instance_info[COMMAND_LINE_SIZE] __initdata;
192 static int boot_instance_index;
194 static char boot_snapshot_info[COMMAND_LINE_SIZE] __initdata;
195 static int boot_snapshot_index;
197 static int __init set_cmdline_ftrace(char *str)
199 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
200 default_bootup_tracer = bootup_tracer_buf;
201 /* We are using ftrace early, expand it */
202 ring_buffer_expanded = true;
205 __setup("ftrace=", set_cmdline_ftrace);
207 static int __init set_ftrace_dump_on_oops(char *str)
209 if (*str++ != '=' || !*str || !strcmp("1", str)) {
210 ftrace_dump_on_oops = DUMP_ALL;
214 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
215 ftrace_dump_on_oops = DUMP_ORIG;
221 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
223 static int __init stop_trace_on_warning(char *str)
225 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
226 __disable_trace_on_warning = 1;
229 __setup("traceoff_on_warning", stop_trace_on_warning);
231 static int __init boot_alloc_snapshot(char *str)
233 char *slot = boot_snapshot_info + boot_snapshot_index;
234 int left = sizeof(boot_snapshot_info) - boot_snapshot_index;
239 if (strlen(str) >= left)
242 ret = snprintf(slot, left, "%s\t", str);
243 boot_snapshot_index += ret;
245 allocate_snapshot = true;
246 /* We also need the main ring buffer expanded */
247 ring_buffer_expanded = true;
251 __setup("alloc_snapshot", boot_alloc_snapshot);
254 static int __init boot_snapshot(char *str)
256 snapshot_at_boot = true;
257 boot_alloc_snapshot(str);
260 __setup("ftrace_boot_snapshot", boot_snapshot);
263 static int __init boot_instance(char *str)
265 char *slot = boot_instance_info + boot_instance_index;
266 int left = sizeof(boot_instance_info) - boot_instance_index;
269 if (strlen(str) >= left)
272 ret = snprintf(slot, left, "%s\t", str);
273 boot_instance_index += ret;
277 __setup("trace_instance=", boot_instance);
280 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
282 static int __init set_trace_boot_options(char *str)
284 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
287 __setup("trace_options=", set_trace_boot_options);
289 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
290 static char *trace_boot_clock __initdata;
292 static int __init set_trace_boot_clock(char *str)
294 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
295 trace_boot_clock = trace_boot_clock_buf;
298 __setup("trace_clock=", set_trace_boot_clock);
300 static int __init set_tracepoint_printk(char *str)
302 /* Ignore the "tp_printk_stop_on_boot" param */
306 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
307 tracepoint_printk = 1;
310 __setup("tp_printk", set_tracepoint_printk);
312 static int __init set_tracepoint_printk_stop(char *str)
314 tracepoint_printk_stop_on_boot = true;
317 __setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
319 unsigned long long ns2usecs(u64 nsec)
327 trace_process_export(struct trace_export *export,
328 struct ring_buffer_event *event, int flag)
330 struct trace_entry *entry;
331 unsigned int size = 0;
333 if (export->flags & flag) {
334 entry = ring_buffer_event_data(event);
335 size = ring_buffer_event_length(event);
336 export->write(export, entry, size);
340 static DEFINE_MUTEX(ftrace_export_lock);
342 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
344 static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
345 static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
346 static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
348 static inline void ftrace_exports_enable(struct trace_export *export)
350 if (export->flags & TRACE_EXPORT_FUNCTION)
351 static_branch_inc(&trace_function_exports_enabled);
353 if (export->flags & TRACE_EXPORT_EVENT)
354 static_branch_inc(&trace_event_exports_enabled);
356 if (export->flags & TRACE_EXPORT_MARKER)
357 static_branch_inc(&trace_marker_exports_enabled);
360 static inline void ftrace_exports_disable(struct trace_export *export)
362 if (export->flags & TRACE_EXPORT_FUNCTION)
363 static_branch_dec(&trace_function_exports_enabled);
365 if (export->flags & TRACE_EXPORT_EVENT)
366 static_branch_dec(&trace_event_exports_enabled);
368 if (export->flags & TRACE_EXPORT_MARKER)
369 static_branch_dec(&trace_marker_exports_enabled);
372 static void ftrace_exports(struct ring_buffer_event *event, int flag)
374 struct trace_export *export;
376 preempt_disable_notrace();
378 export = rcu_dereference_raw_check(ftrace_exports_list);
380 trace_process_export(export, event, flag);
381 export = rcu_dereference_raw_check(export->next);
384 preempt_enable_notrace();
388 add_trace_export(struct trace_export **list, struct trace_export *export)
390 rcu_assign_pointer(export->next, *list);
392 * We are entering export into the list but another
393 * CPU might be walking that list. We need to make sure
394 * the export->next pointer is valid before another CPU sees
395 * the export pointer included into the list.
397 rcu_assign_pointer(*list, export);
401 rm_trace_export(struct trace_export **list, struct trace_export *export)
403 struct trace_export **p;
405 for (p = list; *p != NULL; p = &(*p)->next)
412 rcu_assign_pointer(*p, (*p)->next);
418 add_ftrace_export(struct trace_export **list, struct trace_export *export)
420 ftrace_exports_enable(export);
422 add_trace_export(list, export);
426 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
430 ret = rm_trace_export(list, export);
431 ftrace_exports_disable(export);
436 int register_ftrace_export(struct trace_export *export)
438 if (WARN_ON_ONCE(!export->write))
441 mutex_lock(&ftrace_export_lock);
443 add_ftrace_export(&ftrace_exports_list, export);
445 mutex_unlock(&ftrace_export_lock);
449 EXPORT_SYMBOL_GPL(register_ftrace_export);
451 int unregister_ftrace_export(struct trace_export *export)
455 mutex_lock(&ftrace_export_lock);
457 ret = rm_ftrace_export(&ftrace_exports_list, export);
459 mutex_unlock(&ftrace_export_lock);
463 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
465 /* trace_flags holds trace_options default values */
466 #define TRACE_DEFAULT_FLAGS \
467 (FUNCTION_DEFAULT_FLAGS | \
468 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
469 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
470 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
471 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
474 /* trace_options that are only supported by global_trace */
475 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
476 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
478 /* trace_flags that are default zero for instances */
479 #define ZEROED_TRACE_FLAGS \
480 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
483 * The global_trace is the descriptor that holds the top-level tracing
484 * buffers for the live tracing.
486 static struct trace_array global_trace = {
487 .trace_flags = TRACE_DEFAULT_FLAGS,
490 LIST_HEAD(ftrace_trace_arrays);
492 int trace_array_get(struct trace_array *this_tr)
494 struct trace_array *tr;
497 mutex_lock(&trace_types_lock);
498 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
505 mutex_unlock(&trace_types_lock);
510 static void __trace_array_put(struct trace_array *this_tr)
512 WARN_ON(!this_tr->ref);
517 * trace_array_put - Decrement the reference counter for this trace array.
518 * @this_tr : pointer to the trace array
520 * NOTE: Use this when we no longer need the trace array returned by
521 * trace_array_get_by_name(). This ensures the trace array can be later
525 void trace_array_put(struct trace_array *this_tr)
530 mutex_lock(&trace_types_lock);
531 __trace_array_put(this_tr);
532 mutex_unlock(&trace_types_lock);
534 EXPORT_SYMBOL_GPL(trace_array_put);
536 int tracing_check_open_get_tr(struct trace_array *tr)
540 ret = security_locked_down(LOCKDOWN_TRACEFS);
544 if (tracing_disabled)
547 if (tr && trace_array_get(tr) < 0)
553 int call_filter_check_discard(struct trace_event_call *call, void *rec,
554 struct trace_buffer *buffer,
555 struct ring_buffer_event *event)
557 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
558 !filter_match_preds(call->filter, rec)) {
559 __trace_event_discard_commit(buffer, event);
567 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
568 * @filtered_pids: The list of pids to check
569 * @search_pid: The PID to find in @filtered_pids
571 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
574 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
576 return trace_pid_list_is_set(filtered_pids, search_pid);
580 * trace_ignore_this_task - should a task be ignored for tracing
581 * @filtered_pids: The list of pids to check
582 * @filtered_no_pids: The list of pids not to be traced
583 * @task: The task that should be ignored if not filtered
585 * Checks if @task should be traced or not from @filtered_pids.
586 * Returns true if @task should *NOT* be traced.
587 * Returns false if @task should be traced.
590 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
591 struct trace_pid_list *filtered_no_pids,
592 struct task_struct *task)
595 * If filtered_no_pids is not empty, and the task's pid is listed
596 * in filtered_no_pids, then return true.
597 * Otherwise, if filtered_pids is empty, that means we can
598 * trace all tasks. If it has content, then only trace pids
599 * within filtered_pids.
602 return (filtered_pids &&
603 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
605 trace_find_filtered_pid(filtered_no_pids, task->pid));
609 * trace_filter_add_remove_task - Add or remove a task from a pid_list
610 * @pid_list: The list to modify
611 * @self: The current task for fork or NULL for exit
612 * @task: The task to add or remove
614 * If adding a task, if @self is defined, the task is only added if @self
615 * is also included in @pid_list. This happens on fork and tasks should
616 * only be added when the parent is listed. If @self is NULL, then the
617 * @task pid will be removed from the list, which would happen on exit
620 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
621 struct task_struct *self,
622 struct task_struct *task)
627 /* For forks, we only add if the forking task is listed */
629 if (!trace_find_filtered_pid(pid_list, self->pid))
633 /* "self" is set for forks, and NULL for exits */
635 trace_pid_list_set(pid_list, task->pid);
637 trace_pid_list_clear(pid_list, task->pid);
641 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
642 * @pid_list: The pid list to show
643 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
644 * @pos: The position of the file
646 * This is used by the seq_file "next" operation to iterate the pids
647 * listed in a trace_pid_list structure.
649 * Returns the pid+1 as we want to display pid of zero, but NULL would
650 * stop the iteration.
652 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
654 long pid = (unsigned long)v;
659 /* pid already is +1 of the actual previous bit */
660 if (trace_pid_list_next(pid_list, pid, &next) < 0)
665 /* Return pid + 1 to allow zero to be represented */
666 return (void *)(pid + 1);
670 * trace_pid_start - Used for seq_file to start reading pid lists
671 * @pid_list: The pid list to show
672 * @pos: The position of the file
674 * This is used by seq_file "start" operation to start the iteration
677 * Returns the pid+1 as we want to display pid of zero, but NULL would
678 * stop the iteration.
680 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
686 if (trace_pid_list_first(pid_list, &first) < 0)
691 /* Return pid + 1 so that zero can be the exit value */
692 for (pid++; pid && l < *pos;
693 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
699 * trace_pid_show - show the current pid in seq_file processing
700 * @m: The seq_file structure to write into
701 * @v: A void pointer of the pid (+1) value to display
703 * Can be directly used by seq_file operations to display the current
706 int trace_pid_show(struct seq_file *m, void *v)
708 unsigned long pid = (unsigned long)v - 1;
710 seq_printf(m, "%lu\n", pid);
714 /* 128 should be much more than enough */
715 #define PID_BUF_SIZE 127
717 int trace_pid_write(struct trace_pid_list *filtered_pids,
718 struct trace_pid_list **new_pid_list,
719 const char __user *ubuf, size_t cnt)
721 struct trace_pid_list *pid_list;
722 struct trace_parser parser;
730 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
734 * Always recreate a new array. The write is an all or nothing
735 * operation. Always create a new array when adding new pids by
736 * the user. If the operation fails, then the current list is
739 pid_list = trace_pid_list_alloc();
741 trace_parser_put(&parser);
746 /* copy the current bits to the new max */
747 ret = trace_pid_list_first(filtered_pids, &pid);
749 trace_pid_list_set(pid_list, pid);
750 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
760 ret = trace_get_user(&parser, ubuf, cnt, &pos);
768 if (!trace_parser_loaded(&parser))
772 if (kstrtoul(parser.buffer, 0, &val))
777 if (trace_pid_list_set(pid_list, pid) < 0) {
783 trace_parser_clear(&parser);
786 trace_parser_put(&parser);
789 trace_pid_list_free(pid_list);
794 /* Cleared the list of pids */
795 trace_pid_list_free(pid_list);
799 *new_pid_list = pid_list;
804 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
808 /* Early boot up does not have a buffer yet */
810 return trace_clock_local();
812 ts = ring_buffer_time_stamp(buf->buffer);
813 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
818 u64 ftrace_now(int cpu)
820 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
824 * tracing_is_enabled - Show if global_trace has been enabled
826 * Shows if the global trace has been enabled or not. It uses the
827 * mirror flag "buffer_disabled" to be used in fast paths such as for
828 * the irqsoff tracer. But it may be inaccurate due to races. If you
829 * need to know the accurate state, use tracing_is_on() which is a little
830 * slower, but accurate.
832 int tracing_is_enabled(void)
835 * For quick access (irqsoff uses this in fast path), just
836 * return the mirror variable of the state of the ring buffer.
837 * It's a little racy, but we don't really care.
840 return !global_trace.buffer_disabled;
844 * trace_buf_size is the size in bytes that is allocated
845 * for a buffer. Note, the number of bytes is always rounded
848 * This number is purposely set to a low number of 16384.
849 * If the dump on oops happens, it will be much appreciated
850 * to not have to wait for all that output. Anyway this can be
851 * boot time and run time configurable.
853 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
855 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
857 /* trace_types holds a link list of available tracers. */
858 static struct tracer *trace_types __read_mostly;
861 * trace_types_lock is used to protect the trace_types list.
863 DEFINE_MUTEX(trace_types_lock);
866 * serialize the access of the ring buffer
868 * ring buffer serializes readers, but it is low level protection.
869 * The validity of the events (which returns by ring_buffer_peek() ..etc)
870 * are not protected by ring buffer.
872 * The content of events may become garbage if we allow other process consumes
873 * these events concurrently:
874 * A) the page of the consumed events may become a normal page
875 * (not reader page) in ring buffer, and this page will be rewritten
876 * by events producer.
877 * B) The page of the consumed events may become a page for splice_read,
878 * and this page will be returned to system.
880 * These primitives allow multi process access to different cpu ring buffer
883 * These primitives don't distinguish read-only and read-consume access.
884 * Multi read-only access are also serialized.
888 static DECLARE_RWSEM(all_cpu_access_lock);
889 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
891 static inline void trace_access_lock(int cpu)
893 if (cpu == RING_BUFFER_ALL_CPUS) {
894 /* gain it for accessing the whole ring buffer. */
895 down_write(&all_cpu_access_lock);
897 /* gain it for accessing a cpu ring buffer. */
899 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
900 down_read(&all_cpu_access_lock);
902 /* Secondly block other access to this @cpu ring buffer. */
903 mutex_lock(&per_cpu(cpu_access_lock, cpu));
907 static inline void trace_access_unlock(int cpu)
909 if (cpu == RING_BUFFER_ALL_CPUS) {
910 up_write(&all_cpu_access_lock);
912 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
913 up_read(&all_cpu_access_lock);
917 static inline void trace_access_lock_init(void)
921 for_each_possible_cpu(cpu)
922 mutex_init(&per_cpu(cpu_access_lock, cpu));
927 static DEFINE_MUTEX(access_lock);
929 static inline void trace_access_lock(int cpu)
932 mutex_lock(&access_lock);
935 static inline void trace_access_unlock(int cpu)
938 mutex_unlock(&access_lock);
941 static inline void trace_access_lock_init(void)
947 #ifdef CONFIG_STACKTRACE
948 static void __ftrace_trace_stack(struct trace_buffer *buffer,
949 unsigned int trace_ctx,
950 int skip, struct pt_regs *regs);
951 static inline void ftrace_trace_stack(struct trace_array *tr,
952 struct trace_buffer *buffer,
953 unsigned int trace_ctx,
954 int skip, struct pt_regs *regs);
957 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
958 unsigned int trace_ctx,
959 int skip, struct pt_regs *regs)
962 static inline void ftrace_trace_stack(struct trace_array *tr,
963 struct trace_buffer *buffer,
964 unsigned long trace_ctx,
965 int skip, struct pt_regs *regs)
971 static __always_inline void
972 trace_event_setup(struct ring_buffer_event *event,
973 int type, unsigned int trace_ctx)
975 struct trace_entry *ent = ring_buffer_event_data(event);
977 tracing_generic_entry_update(ent, type, trace_ctx);
980 static __always_inline struct ring_buffer_event *
981 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
984 unsigned int trace_ctx)
986 struct ring_buffer_event *event;
988 event = ring_buffer_lock_reserve(buffer, len);
990 trace_event_setup(event, type, trace_ctx);
995 void tracer_tracing_on(struct trace_array *tr)
997 if (tr->array_buffer.buffer)
998 ring_buffer_record_on(tr->array_buffer.buffer);
1000 * This flag is looked at when buffers haven't been allocated
1001 * yet, or by some tracers (like irqsoff), that just want to
1002 * know if the ring buffer has been disabled, but it can handle
1003 * races of where it gets disabled but we still do a record.
1004 * As the check is in the fast path of the tracers, it is more
1005 * important to be fast than accurate.
1007 tr->buffer_disabled = 0;
1008 /* Make the flag seen by readers */
1013 * tracing_on - enable tracing buffers
1015 * This function enables tracing buffers that may have been
1016 * disabled with tracing_off.
1018 void tracing_on(void)
1020 tracer_tracing_on(&global_trace);
1022 EXPORT_SYMBOL_GPL(tracing_on);
1025 static __always_inline void
1026 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
1028 __this_cpu_write(trace_taskinfo_save, true);
1030 /* If this is the temp buffer, we need to commit fully */
1031 if (this_cpu_read(trace_buffered_event) == event) {
1032 /* Length is in event->array[0] */
1033 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1034 /* Release the temp buffer */
1035 this_cpu_dec(trace_buffered_event_cnt);
1036 /* ring_buffer_unlock_commit() enables preemption */
1037 preempt_enable_notrace();
1039 ring_buffer_unlock_commit(buffer);
1042 int __trace_array_puts(struct trace_array *tr, unsigned long ip,
1043 const char *str, int size)
1045 struct ring_buffer_event *event;
1046 struct trace_buffer *buffer;
1047 struct print_entry *entry;
1048 unsigned int trace_ctx;
1051 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1054 if (unlikely(tracing_selftest_running || tracing_disabled))
1057 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1059 trace_ctx = tracing_gen_ctx();
1060 buffer = tr->array_buffer.buffer;
1061 ring_buffer_nest_start(buffer);
1062 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1069 entry = ring_buffer_event_data(event);
1072 memcpy(&entry->buf, str, size);
1074 /* Add a newline if necessary */
1075 if (entry->buf[size - 1] != '\n') {
1076 entry->buf[size] = '\n';
1077 entry->buf[size + 1] = '\0';
1079 entry->buf[size] = '\0';
1081 __buffer_unlock_commit(buffer, event);
1082 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1084 ring_buffer_nest_end(buffer);
1087 EXPORT_SYMBOL_GPL(__trace_array_puts);
1090 * __trace_puts - write a constant string into the trace buffer.
1091 * @ip: The address of the caller
1092 * @str: The constant string to write
1093 * @size: The size of the string.
1095 int __trace_puts(unsigned long ip, const char *str, int size)
1097 return __trace_array_puts(&global_trace, ip, str, size);
1099 EXPORT_SYMBOL_GPL(__trace_puts);
1102 * __trace_bputs - write the pointer to a constant string into trace buffer
1103 * @ip: The address of the caller
1104 * @str: The constant string to write to the buffer to
1106 int __trace_bputs(unsigned long ip, const char *str)
1108 struct ring_buffer_event *event;
1109 struct trace_buffer *buffer;
1110 struct bputs_entry *entry;
1111 unsigned int trace_ctx;
1112 int size = sizeof(struct bputs_entry);
1115 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1118 if (unlikely(tracing_selftest_running || tracing_disabled))
1121 trace_ctx = tracing_gen_ctx();
1122 buffer = global_trace.array_buffer.buffer;
1124 ring_buffer_nest_start(buffer);
1125 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1130 entry = ring_buffer_event_data(event);
1134 __buffer_unlock_commit(buffer, event);
1135 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1139 ring_buffer_nest_end(buffer);
1142 EXPORT_SYMBOL_GPL(__trace_bputs);
1144 #ifdef CONFIG_TRACER_SNAPSHOT
1145 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1148 struct tracer *tracer = tr->current_trace;
1149 unsigned long flags;
1152 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1153 internal_trace_puts("*** snapshot is being ignored ***\n");
1157 if (!tr->allocated_snapshot) {
1158 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
1159 internal_trace_puts("*** stopping trace here! ***\n");
1164 /* Note, snapshot can not be used when the tracer uses it */
1165 if (tracer->use_max_tr) {
1166 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
1167 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
1171 local_irq_save(flags);
1172 update_max_tr(tr, current, smp_processor_id(), cond_data);
1173 local_irq_restore(flags);
1176 void tracing_snapshot_instance(struct trace_array *tr)
1178 tracing_snapshot_instance_cond(tr, NULL);
1182 * tracing_snapshot - take a snapshot of the current buffer.
1184 * This causes a swap between the snapshot buffer and the current live
1185 * tracing buffer. You can use this to take snapshots of the live
1186 * trace when some condition is triggered, but continue to trace.
1188 * Note, make sure to allocate the snapshot with either
1189 * a tracing_snapshot_alloc(), or by doing it manually
1190 * with: echo 1 > /sys/kernel/tracing/snapshot
1192 * If the snapshot buffer is not allocated, it will stop tracing.
1193 * Basically making a permanent snapshot.
1195 void tracing_snapshot(void)
1197 struct trace_array *tr = &global_trace;
1199 tracing_snapshot_instance(tr);
1201 EXPORT_SYMBOL_GPL(tracing_snapshot);
1204 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1205 * @tr: The tracing instance to snapshot
1206 * @cond_data: The data to be tested conditionally, and possibly saved
1208 * This is the same as tracing_snapshot() except that the snapshot is
1209 * conditional - the snapshot will only happen if the
1210 * cond_snapshot.update() implementation receiving the cond_data
1211 * returns true, which means that the trace array's cond_snapshot
1212 * update() operation used the cond_data to determine whether the
1213 * snapshot should be taken, and if it was, presumably saved it along
1214 * with the snapshot.
1216 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1218 tracing_snapshot_instance_cond(tr, cond_data);
1220 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1223 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1224 * @tr: The tracing instance
1226 * When the user enables a conditional snapshot using
1227 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1228 * with the snapshot. This accessor is used to retrieve it.
1230 * Should not be called from cond_snapshot.update(), since it takes
1231 * the tr->max_lock lock, which the code calling
1232 * cond_snapshot.update() has already done.
1234 * Returns the cond_data associated with the trace array's snapshot.
1236 void *tracing_cond_snapshot_data(struct trace_array *tr)
1238 void *cond_data = NULL;
1240 local_irq_disable();
1241 arch_spin_lock(&tr->max_lock);
1243 if (tr->cond_snapshot)
1244 cond_data = tr->cond_snapshot->cond_data;
1246 arch_spin_unlock(&tr->max_lock);
1251 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1253 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1254 struct array_buffer *size_buf, int cpu_id);
1255 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1257 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1261 if (!tr->allocated_snapshot) {
1263 /* allocate spare buffer */
1264 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1265 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1269 tr->allocated_snapshot = true;
1275 static void free_snapshot(struct trace_array *tr)
1278 * We don't free the ring buffer. instead, resize it because
1279 * The max_tr ring buffer has some state (e.g. ring->clock) and
1280 * we want preserve it.
1282 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1283 set_buffer_entries(&tr->max_buffer, 1);
1284 tracing_reset_online_cpus(&tr->max_buffer);
1285 tr->allocated_snapshot = false;
1289 * tracing_alloc_snapshot - allocate snapshot buffer.
1291 * This only allocates the snapshot buffer if it isn't already
1292 * allocated - it doesn't also take a snapshot.
1294 * This is meant to be used in cases where the snapshot buffer needs
1295 * to be set up for events that can't sleep but need to be able to
1296 * trigger a snapshot.
1298 int tracing_alloc_snapshot(void)
1300 struct trace_array *tr = &global_trace;
1303 ret = tracing_alloc_snapshot_instance(tr);
1308 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1311 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1313 * This is similar to tracing_snapshot(), but it will allocate the
1314 * snapshot buffer if it isn't already allocated. Use this only
1315 * where it is safe to sleep, as the allocation may sleep.
1317 * This causes a swap between the snapshot buffer and the current live
1318 * tracing buffer. You can use this to take snapshots of the live
1319 * trace when some condition is triggered, but continue to trace.
1321 void tracing_snapshot_alloc(void)
1325 ret = tracing_alloc_snapshot();
1331 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1334 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1335 * @tr: The tracing instance
1336 * @cond_data: User data to associate with the snapshot
1337 * @update: Implementation of the cond_snapshot update function
1339 * Check whether the conditional snapshot for the given instance has
1340 * already been enabled, or if the current tracer is already using a
1341 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1342 * save the cond_data and update function inside.
1344 * Returns 0 if successful, error otherwise.
1346 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1347 cond_update_fn_t update)
1349 struct cond_snapshot *cond_snapshot;
1352 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1356 cond_snapshot->cond_data = cond_data;
1357 cond_snapshot->update = update;
1359 mutex_lock(&trace_types_lock);
1361 ret = tracing_alloc_snapshot_instance(tr);
1365 if (tr->current_trace->use_max_tr) {
1371 * The cond_snapshot can only change to NULL without the
1372 * trace_types_lock. We don't care if we race with it going
1373 * to NULL, but we want to make sure that it's not set to
1374 * something other than NULL when we get here, which we can
1375 * do safely with only holding the trace_types_lock and not
1376 * having to take the max_lock.
1378 if (tr->cond_snapshot) {
1383 local_irq_disable();
1384 arch_spin_lock(&tr->max_lock);
1385 tr->cond_snapshot = cond_snapshot;
1386 arch_spin_unlock(&tr->max_lock);
1389 mutex_unlock(&trace_types_lock);
1394 mutex_unlock(&trace_types_lock);
1395 kfree(cond_snapshot);
1398 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1401 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1402 * @tr: The tracing instance
1404 * Check whether the conditional snapshot for the given instance is
1405 * enabled; if so, free the cond_snapshot associated with it,
1406 * otherwise return -EINVAL.
1408 * Returns 0 if successful, error otherwise.
1410 int tracing_snapshot_cond_disable(struct trace_array *tr)
1414 local_irq_disable();
1415 arch_spin_lock(&tr->max_lock);
1417 if (!tr->cond_snapshot)
1420 kfree(tr->cond_snapshot);
1421 tr->cond_snapshot = NULL;
1424 arch_spin_unlock(&tr->max_lock);
1429 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1431 void tracing_snapshot(void)
1433 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1435 EXPORT_SYMBOL_GPL(tracing_snapshot);
1436 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1438 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1440 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1441 int tracing_alloc_snapshot(void)
1443 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1446 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1447 void tracing_snapshot_alloc(void)
1452 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1453 void *tracing_cond_snapshot_data(struct trace_array *tr)
1457 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1458 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1462 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1463 int tracing_snapshot_cond_disable(struct trace_array *tr)
1467 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1468 #define free_snapshot(tr) do { } while (0)
1469 #endif /* CONFIG_TRACER_SNAPSHOT */
1471 void tracer_tracing_off(struct trace_array *tr)
1473 if (tr->array_buffer.buffer)
1474 ring_buffer_record_off(tr->array_buffer.buffer);
1476 * This flag is looked at when buffers haven't been allocated
1477 * yet, or by some tracers (like irqsoff), that just want to
1478 * know if the ring buffer has been disabled, but it can handle
1479 * races of where it gets disabled but we still do a record.
1480 * As the check is in the fast path of the tracers, it is more
1481 * important to be fast than accurate.
1483 tr->buffer_disabled = 1;
1484 /* Make the flag seen by readers */
1489 * tracing_off - turn off tracing buffers
1491 * This function stops the tracing buffers from recording data.
1492 * It does not disable any overhead the tracers themselves may
1493 * be causing. This function simply causes all recording to
1494 * the ring buffers to fail.
1496 void tracing_off(void)
1498 tracer_tracing_off(&global_trace);
1500 EXPORT_SYMBOL_GPL(tracing_off);
1502 void disable_trace_on_warning(void)
1504 if (__disable_trace_on_warning) {
1505 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1506 "Disabling tracing due to warning\n");
1512 * tracer_tracing_is_on - show real state of ring buffer enabled
1513 * @tr : the trace array to know if ring buffer is enabled
1515 * Shows real state of the ring buffer if it is enabled or not.
1517 bool tracer_tracing_is_on(struct trace_array *tr)
1519 if (tr->array_buffer.buffer)
1520 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1521 return !tr->buffer_disabled;
1525 * tracing_is_on - show state of ring buffers enabled
1527 int tracing_is_on(void)
1529 return tracer_tracing_is_on(&global_trace);
1531 EXPORT_SYMBOL_GPL(tracing_is_on);
1533 static int __init set_buf_size(char *str)
1535 unsigned long buf_size;
1539 buf_size = memparse(str, &str);
1541 * nr_entries can not be zero and the startup
1542 * tests require some buffer space. Therefore
1543 * ensure we have at least 4096 bytes of buffer.
1545 trace_buf_size = max(4096UL, buf_size);
1548 __setup("trace_buf_size=", set_buf_size);
1550 static int __init set_tracing_thresh(char *str)
1552 unsigned long threshold;
1557 ret = kstrtoul(str, 0, &threshold);
1560 tracing_thresh = threshold * 1000;
1563 __setup("tracing_thresh=", set_tracing_thresh);
1565 unsigned long nsecs_to_usecs(unsigned long nsecs)
1567 return nsecs / 1000;
1571 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1572 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1573 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1574 * of strings in the order that the evals (enum) were defined.
1579 /* These must match the bit positions in trace_iterator_flags */
1580 static const char *trace_options[] = {
1588 int in_ns; /* is this clock in nanoseconds? */
1589 } trace_clocks[] = {
1590 { trace_clock_local, "local", 1 },
1591 { trace_clock_global, "global", 1 },
1592 { trace_clock_counter, "counter", 0 },
1593 { trace_clock_jiffies, "uptime", 0 },
1594 { trace_clock, "perf", 1 },
1595 { ktime_get_mono_fast_ns, "mono", 1 },
1596 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1597 { ktime_get_boot_fast_ns, "boot", 1 },
1598 { ktime_get_tai_fast_ns, "tai", 1 },
1602 bool trace_clock_in_ns(struct trace_array *tr)
1604 if (trace_clocks[tr->clock_id].in_ns)
1611 * trace_parser_get_init - gets the buffer for trace parser
1613 int trace_parser_get_init(struct trace_parser *parser, int size)
1615 memset(parser, 0, sizeof(*parser));
1617 parser->buffer = kmalloc(size, GFP_KERNEL);
1618 if (!parser->buffer)
1621 parser->size = size;
1626 * trace_parser_put - frees the buffer for trace parser
1628 void trace_parser_put(struct trace_parser *parser)
1630 kfree(parser->buffer);
1631 parser->buffer = NULL;
1635 * trace_get_user - reads the user input string separated by space
1636 * (matched by isspace(ch))
1638 * For each string found the 'struct trace_parser' is updated,
1639 * and the function returns.
1641 * Returns number of bytes read.
1643 * See kernel/trace/trace.h for 'struct trace_parser' details.
1645 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1646 size_t cnt, loff_t *ppos)
1653 trace_parser_clear(parser);
1655 ret = get_user(ch, ubuf++);
1663 * The parser is not finished with the last write,
1664 * continue reading the user input without skipping spaces.
1666 if (!parser->cont) {
1667 /* skip white space */
1668 while (cnt && isspace(ch)) {
1669 ret = get_user(ch, ubuf++);
1678 /* only spaces were written */
1679 if (isspace(ch) || !ch) {
1686 /* read the non-space input */
1687 while (cnt && !isspace(ch) && ch) {
1688 if (parser->idx < parser->size - 1)
1689 parser->buffer[parser->idx++] = ch;
1694 ret = get_user(ch, ubuf++);
1701 /* We either got finished input or we have to wait for another call. */
1702 if (isspace(ch) || !ch) {
1703 parser->buffer[parser->idx] = 0;
1704 parser->cont = false;
1705 } else if (parser->idx < parser->size - 1) {
1706 parser->cont = true;
1707 parser->buffer[parser->idx++] = ch;
1708 /* Make sure the parsed string always terminates with '\0'. */
1709 parser->buffer[parser->idx] = 0;
1722 /* TODO add a seq_buf_to_buffer() */
1723 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1727 if (trace_seq_used(s) <= s->seq.readpos)
1730 len = trace_seq_used(s) - s->seq.readpos;
1733 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1735 s->seq.readpos += cnt;
1739 unsigned long __read_mostly tracing_thresh;
1741 #ifdef CONFIG_TRACER_MAX_TRACE
1742 static const struct file_operations tracing_max_lat_fops;
1744 #ifdef LATENCY_FS_NOTIFY
1746 static struct workqueue_struct *fsnotify_wq;
1748 static void latency_fsnotify_workfn(struct work_struct *work)
1750 struct trace_array *tr = container_of(work, struct trace_array,
1752 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1755 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1757 struct trace_array *tr = container_of(iwork, struct trace_array,
1759 queue_work(fsnotify_wq, &tr->fsnotify_work);
1762 static void trace_create_maxlat_file(struct trace_array *tr,
1763 struct dentry *d_tracer)
1765 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1766 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1767 tr->d_max_latency = trace_create_file("tracing_max_latency",
1769 d_tracer, &tr->max_latency,
1770 &tracing_max_lat_fops);
1773 __init static int latency_fsnotify_init(void)
1775 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1776 WQ_UNBOUND | WQ_HIGHPRI, 0);
1778 pr_err("Unable to allocate tr_max_lat_wq\n");
1784 late_initcall_sync(latency_fsnotify_init);
1786 void latency_fsnotify(struct trace_array *tr)
1791 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1792 * possible that we are called from __schedule() or do_idle(), which
1793 * could cause a deadlock.
1795 irq_work_queue(&tr->fsnotify_irqwork);
1798 #else /* !LATENCY_FS_NOTIFY */
1800 #define trace_create_maxlat_file(tr, d_tracer) \
1801 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1802 d_tracer, &tr->max_latency, &tracing_max_lat_fops)
1807 * Copy the new maximum trace into the separate maximum-trace
1808 * structure. (this way the maximum trace is permanently saved,
1809 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1812 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1814 struct array_buffer *trace_buf = &tr->array_buffer;
1815 struct array_buffer *max_buf = &tr->max_buffer;
1816 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1817 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1820 max_buf->time_start = data->preempt_timestamp;
1822 max_data->saved_latency = tr->max_latency;
1823 max_data->critical_start = data->critical_start;
1824 max_data->critical_end = data->critical_end;
1826 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1827 max_data->pid = tsk->pid;
1829 * If tsk == current, then use current_uid(), as that does not use
1830 * RCU. The irq tracer can be called out of RCU scope.
1833 max_data->uid = current_uid();
1835 max_data->uid = task_uid(tsk);
1837 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1838 max_data->policy = tsk->policy;
1839 max_data->rt_priority = tsk->rt_priority;
1841 /* record this tasks comm */
1842 tracing_record_cmdline(tsk);
1843 latency_fsnotify(tr);
1847 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1849 * @tsk: the task with the latency
1850 * @cpu: The cpu that initiated the trace.
1851 * @cond_data: User data associated with a conditional snapshot
1853 * Flip the buffers between the @tr and the max_tr and record information
1854 * about which task was the cause of this latency.
1857 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1863 WARN_ON_ONCE(!irqs_disabled());
1865 if (!tr->allocated_snapshot) {
1866 /* Only the nop tracer should hit this when disabling */
1867 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1871 arch_spin_lock(&tr->max_lock);
1873 /* Inherit the recordable setting from array_buffer */
1874 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1875 ring_buffer_record_on(tr->max_buffer.buffer);
1877 ring_buffer_record_off(tr->max_buffer.buffer);
1879 #ifdef CONFIG_TRACER_SNAPSHOT
1880 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1881 arch_spin_unlock(&tr->max_lock);
1885 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1887 __update_max_tr(tr, tsk, cpu);
1889 arch_spin_unlock(&tr->max_lock);
1893 * update_max_tr_single - only copy one trace over, and reset the rest
1895 * @tsk: task with the latency
1896 * @cpu: the cpu of the buffer to copy.
1898 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1901 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1908 WARN_ON_ONCE(!irqs_disabled());
1909 if (!tr->allocated_snapshot) {
1910 /* Only the nop tracer should hit this when disabling */
1911 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1915 arch_spin_lock(&tr->max_lock);
1917 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1919 if (ret == -EBUSY) {
1921 * We failed to swap the buffer due to a commit taking
1922 * place on this CPU. We fail to record, but we reset
1923 * the max trace buffer (no one writes directly to it)
1924 * and flag that it failed.
1926 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1927 "Failed to swap buffers due to commit in progress\n");
1930 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1932 __update_max_tr(tr, tsk, cpu);
1933 arch_spin_unlock(&tr->max_lock);
1936 #endif /* CONFIG_TRACER_MAX_TRACE */
1938 static int wait_on_pipe(struct trace_iterator *iter, int full)
1940 /* Iterators are static, they should be filled or empty */
1941 if (trace_buffer_iter(iter, iter->cpu_file))
1944 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1948 #ifdef CONFIG_FTRACE_STARTUP_TEST
1949 static bool selftests_can_run;
1951 struct trace_selftests {
1952 struct list_head list;
1953 struct tracer *type;
1956 static LIST_HEAD(postponed_selftests);
1958 static int save_selftest(struct tracer *type)
1960 struct trace_selftests *selftest;
1962 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1966 selftest->type = type;
1967 list_add(&selftest->list, &postponed_selftests);
1971 static int run_tracer_selftest(struct tracer *type)
1973 struct trace_array *tr = &global_trace;
1974 struct tracer *saved_tracer = tr->current_trace;
1977 if (!type->selftest || tracing_selftest_disabled)
1981 * If a tracer registers early in boot up (before scheduling is
1982 * initialized and such), then do not run its selftests yet.
1983 * Instead, run it a little later in the boot process.
1985 if (!selftests_can_run)
1986 return save_selftest(type);
1988 if (!tracing_is_on()) {
1989 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
1995 * Run a selftest on this tracer.
1996 * Here we reset the trace buffer, and set the current
1997 * tracer to be this tracer. The tracer can then run some
1998 * internal tracing to verify that everything is in order.
1999 * If we fail, we do not register this tracer.
2001 tracing_reset_online_cpus(&tr->array_buffer);
2003 tr->current_trace = type;
2005 #ifdef CONFIG_TRACER_MAX_TRACE
2006 if (type->use_max_tr) {
2007 /* If we expanded the buffers, make sure the max is expanded too */
2008 if (ring_buffer_expanded)
2009 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2010 RING_BUFFER_ALL_CPUS);
2011 tr->allocated_snapshot = true;
2015 /* the test is responsible for initializing and enabling */
2016 pr_info("Testing tracer %s: ", type->name);
2017 ret = type->selftest(type, tr);
2018 /* the test is responsible for resetting too */
2019 tr->current_trace = saved_tracer;
2021 printk(KERN_CONT "FAILED!\n");
2022 /* Add the warning after printing 'FAILED' */
2026 /* Only reset on passing, to avoid touching corrupted buffers */
2027 tracing_reset_online_cpus(&tr->array_buffer);
2029 #ifdef CONFIG_TRACER_MAX_TRACE
2030 if (type->use_max_tr) {
2031 tr->allocated_snapshot = false;
2033 /* Shrink the max buffer again */
2034 if (ring_buffer_expanded)
2035 ring_buffer_resize(tr->max_buffer.buffer, 1,
2036 RING_BUFFER_ALL_CPUS);
2040 printk(KERN_CONT "PASSED\n");
2044 static __init int init_trace_selftests(void)
2046 struct trace_selftests *p, *n;
2047 struct tracer *t, **last;
2050 selftests_can_run = true;
2052 mutex_lock(&trace_types_lock);
2054 if (list_empty(&postponed_selftests))
2057 pr_info("Running postponed tracer tests:\n");
2059 tracing_selftest_running = true;
2060 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2061 /* This loop can take minutes when sanitizers are enabled, so
2062 * lets make sure we allow RCU processing.
2065 ret = run_tracer_selftest(p->type);
2066 /* If the test fails, then warn and remove from available_tracers */
2068 WARN(1, "tracer: %s failed selftest, disabling\n",
2070 last = &trace_types;
2071 for (t = trace_types; t; t = t->next) {
2082 tracing_selftest_running = false;
2085 mutex_unlock(&trace_types_lock);
2089 core_initcall(init_trace_selftests);
2091 static inline int run_tracer_selftest(struct tracer *type)
2095 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2097 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2099 static void __init apply_trace_boot_options(void);
2102 * register_tracer - register a tracer with the ftrace system.
2103 * @type: the plugin for the tracer
2105 * Register a new plugin tracer.
2107 int __init register_tracer(struct tracer *type)
2113 pr_info("Tracer must have a name\n");
2117 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2118 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2122 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2123 pr_warn("Can not register tracer %s due to lockdown\n",
2128 mutex_lock(&trace_types_lock);
2130 tracing_selftest_running = true;
2132 for (t = trace_types; t; t = t->next) {
2133 if (strcmp(type->name, t->name) == 0) {
2135 pr_info("Tracer %s already registered\n",
2142 if (!type->set_flag)
2143 type->set_flag = &dummy_set_flag;
2145 /*allocate a dummy tracer_flags*/
2146 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2151 type->flags->val = 0;
2152 type->flags->opts = dummy_tracer_opt;
2154 if (!type->flags->opts)
2155 type->flags->opts = dummy_tracer_opt;
2157 /* store the tracer for __set_tracer_option */
2158 type->flags->trace = type;
2160 ret = run_tracer_selftest(type);
2164 type->next = trace_types;
2166 add_tracer_options(&global_trace, type);
2169 tracing_selftest_running = false;
2170 mutex_unlock(&trace_types_lock);
2172 if (ret || !default_bootup_tracer)
2175 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2178 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2179 /* Do we want this tracer to start on bootup? */
2180 tracing_set_tracer(&global_trace, type->name);
2181 default_bootup_tracer = NULL;
2183 apply_trace_boot_options();
2185 /* disable other selftests, since this will break it. */
2186 disable_tracing_selftest("running a tracer");
2192 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2194 struct trace_buffer *buffer = buf->buffer;
2199 ring_buffer_record_disable(buffer);
2201 /* Make sure all commits have finished */
2203 ring_buffer_reset_cpu(buffer, cpu);
2205 ring_buffer_record_enable(buffer);
2208 void tracing_reset_online_cpus(struct array_buffer *buf)
2210 struct trace_buffer *buffer = buf->buffer;
2215 ring_buffer_record_disable(buffer);
2217 /* Make sure all commits have finished */
2220 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2222 ring_buffer_reset_online_cpus(buffer);
2224 ring_buffer_record_enable(buffer);
2227 /* Must have trace_types_lock held */
2228 void tracing_reset_all_online_cpus_unlocked(void)
2230 struct trace_array *tr;
2232 lockdep_assert_held(&trace_types_lock);
2234 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2235 if (!tr->clear_trace)
2237 tr->clear_trace = false;
2238 tracing_reset_online_cpus(&tr->array_buffer);
2239 #ifdef CONFIG_TRACER_MAX_TRACE
2240 tracing_reset_online_cpus(&tr->max_buffer);
2245 void tracing_reset_all_online_cpus(void)
2247 mutex_lock(&trace_types_lock);
2248 tracing_reset_all_online_cpus_unlocked();
2249 mutex_unlock(&trace_types_lock);
2253 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2254 * is the tgid last observed corresponding to pid=i.
2256 static int *tgid_map;
2258 /* The maximum valid index into tgid_map. */
2259 static size_t tgid_map_max;
2261 #define SAVED_CMDLINES_DEFAULT 128
2262 #define NO_CMDLINE_MAP UINT_MAX
2264 * Preemption must be disabled before acquiring trace_cmdline_lock.
2265 * The various trace_arrays' max_lock must be acquired in a context
2266 * where interrupt is disabled.
2268 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2269 struct saved_cmdlines_buffer {
2270 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2271 unsigned *map_cmdline_to_pid;
2272 unsigned cmdline_num;
2274 char *saved_cmdlines;
2276 static struct saved_cmdlines_buffer *savedcmd;
2278 static inline char *get_saved_cmdlines(int idx)
2280 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2283 static inline void set_cmdline(int idx, const char *cmdline)
2285 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2288 static int allocate_cmdlines_buffer(unsigned int val,
2289 struct saved_cmdlines_buffer *s)
2291 s->map_cmdline_to_pid = kmalloc_array(val,
2292 sizeof(*s->map_cmdline_to_pid),
2294 if (!s->map_cmdline_to_pid)
2297 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2298 if (!s->saved_cmdlines) {
2299 kfree(s->map_cmdline_to_pid);
2304 s->cmdline_num = val;
2305 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2306 sizeof(s->map_pid_to_cmdline));
2307 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2308 val * sizeof(*s->map_cmdline_to_pid));
2313 static int trace_create_savedcmd(void)
2317 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2321 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2331 int is_tracing_stopped(void)
2333 return global_trace.stop_count;
2337 * tracing_start - quick start of the tracer
2339 * If tracing is enabled but was stopped by tracing_stop,
2340 * this will start the tracer back up.
2342 void tracing_start(void)
2344 struct trace_buffer *buffer;
2345 unsigned long flags;
2347 if (tracing_disabled)
2350 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2351 if (--global_trace.stop_count) {
2352 if (global_trace.stop_count < 0) {
2353 /* Someone screwed up their debugging */
2355 global_trace.stop_count = 0;
2360 /* Prevent the buffers from switching */
2361 arch_spin_lock(&global_trace.max_lock);
2363 buffer = global_trace.array_buffer.buffer;
2365 ring_buffer_record_enable(buffer);
2367 #ifdef CONFIG_TRACER_MAX_TRACE
2368 buffer = global_trace.max_buffer.buffer;
2370 ring_buffer_record_enable(buffer);
2373 arch_spin_unlock(&global_trace.max_lock);
2376 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2379 static void tracing_start_tr(struct trace_array *tr)
2381 struct trace_buffer *buffer;
2382 unsigned long flags;
2384 if (tracing_disabled)
2387 /* If global, we need to also start the max tracer */
2388 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2389 return tracing_start();
2391 raw_spin_lock_irqsave(&tr->start_lock, flags);
2393 if (--tr->stop_count) {
2394 if (tr->stop_count < 0) {
2395 /* Someone screwed up their debugging */
2402 buffer = tr->array_buffer.buffer;
2404 ring_buffer_record_enable(buffer);
2407 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2411 * tracing_stop - quick stop of the tracer
2413 * Light weight way to stop tracing. Use in conjunction with
2416 void tracing_stop(void)
2418 struct trace_buffer *buffer;
2419 unsigned long flags;
2421 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2422 if (global_trace.stop_count++)
2425 /* Prevent the buffers from switching */
2426 arch_spin_lock(&global_trace.max_lock);
2428 buffer = global_trace.array_buffer.buffer;
2430 ring_buffer_record_disable(buffer);
2432 #ifdef CONFIG_TRACER_MAX_TRACE
2433 buffer = global_trace.max_buffer.buffer;
2435 ring_buffer_record_disable(buffer);
2438 arch_spin_unlock(&global_trace.max_lock);
2441 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2444 static void tracing_stop_tr(struct trace_array *tr)
2446 struct trace_buffer *buffer;
2447 unsigned long flags;
2449 /* If global, we need to also stop the max tracer */
2450 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2451 return tracing_stop();
2453 raw_spin_lock_irqsave(&tr->start_lock, flags);
2454 if (tr->stop_count++)
2457 buffer = tr->array_buffer.buffer;
2459 ring_buffer_record_disable(buffer);
2462 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2465 static int trace_save_cmdline(struct task_struct *tsk)
2469 /* treat recording of idle task as a success */
2473 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2476 * It's not the end of the world if we don't get
2477 * the lock, but we also don't want to spin
2478 * nor do we want to disable interrupts,
2479 * so if we miss here, then better luck next time.
2481 * This is called within the scheduler and wake up, so interrupts
2482 * had better been disabled and run queue lock been held.
2484 lockdep_assert_preemption_disabled();
2485 if (!arch_spin_trylock(&trace_cmdline_lock))
2488 idx = savedcmd->map_pid_to_cmdline[tpid];
2489 if (idx == NO_CMDLINE_MAP) {
2490 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2492 savedcmd->map_pid_to_cmdline[tpid] = idx;
2493 savedcmd->cmdline_idx = idx;
2496 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2497 set_cmdline(idx, tsk->comm);
2499 arch_spin_unlock(&trace_cmdline_lock);
2504 static void __trace_find_cmdline(int pid, char comm[])
2510 strcpy(comm, "<idle>");
2514 if (WARN_ON_ONCE(pid < 0)) {
2515 strcpy(comm, "<XXX>");
2519 tpid = pid & (PID_MAX_DEFAULT - 1);
2520 map = savedcmd->map_pid_to_cmdline[tpid];
2521 if (map != NO_CMDLINE_MAP) {
2522 tpid = savedcmd->map_cmdline_to_pid[map];
2524 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2528 strcpy(comm, "<...>");
2531 void trace_find_cmdline(int pid, char comm[])
2534 arch_spin_lock(&trace_cmdline_lock);
2536 __trace_find_cmdline(pid, comm);
2538 arch_spin_unlock(&trace_cmdline_lock);
2542 static int *trace_find_tgid_ptr(int pid)
2545 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2546 * if we observe a non-NULL tgid_map then we also observe the correct
2549 int *map = smp_load_acquire(&tgid_map);
2551 if (unlikely(!map || pid > tgid_map_max))
2557 int trace_find_tgid(int pid)
2559 int *ptr = trace_find_tgid_ptr(pid);
2561 return ptr ? *ptr : 0;
2564 static int trace_save_tgid(struct task_struct *tsk)
2568 /* treat recording of idle task as a success */
2572 ptr = trace_find_tgid_ptr(tsk->pid);
2580 static bool tracing_record_taskinfo_skip(int flags)
2582 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2584 if (!__this_cpu_read(trace_taskinfo_save))
2590 * tracing_record_taskinfo - record the task info of a task
2592 * @task: task to record
2593 * @flags: TRACE_RECORD_CMDLINE for recording comm
2594 * TRACE_RECORD_TGID for recording tgid
2596 void tracing_record_taskinfo(struct task_struct *task, int flags)
2600 if (tracing_record_taskinfo_skip(flags))
2604 * Record as much task information as possible. If some fail, continue
2605 * to try to record the others.
2607 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2608 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2610 /* If recording any information failed, retry again soon. */
2614 __this_cpu_write(trace_taskinfo_save, false);
2618 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2620 * @prev: previous task during sched_switch
2621 * @next: next task during sched_switch
2622 * @flags: TRACE_RECORD_CMDLINE for recording comm
2623 * TRACE_RECORD_TGID for recording tgid
2625 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2626 struct task_struct *next, int flags)
2630 if (tracing_record_taskinfo_skip(flags))
2634 * Record as much task information as possible. If some fail, continue
2635 * to try to record the others.
2637 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2638 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2639 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2640 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2642 /* If recording any information failed, retry again soon. */
2646 __this_cpu_write(trace_taskinfo_save, false);
2649 /* Helpers to record a specific task information */
2650 void tracing_record_cmdline(struct task_struct *task)
2652 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2655 void tracing_record_tgid(struct task_struct *task)
2657 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2661 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2662 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2663 * simplifies those functions and keeps them in sync.
2665 enum print_line_t trace_handle_return(struct trace_seq *s)
2667 return trace_seq_has_overflowed(s) ?
2668 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2670 EXPORT_SYMBOL_GPL(trace_handle_return);
2672 static unsigned short migration_disable_value(void)
2674 #if defined(CONFIG_SMP)
2675 return current->migration_disabled;
2681 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2683 unsigned int trace_flags = irqs_status;
2686 pc = preempt_count();
2689 trace_flags |= TRACE_FLAG_NMI;
2690 if (pc & HARDIRQ_MASK)
2691 trace_flags |= TRACE_FLAG_HARDIRQ;
2692 if (in_serving_softirq())
2693 trace_flags |= TRACE_FLAG_SOFTIRQ;
2694 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2695 trace_flags |= TRACE_FLAG_BH_OFF;
2697 if (tif_need_resched())
2698 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2699 if (test_preempt_need_resched())
2700 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2701 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2702 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2705 struct ring_buffer_event *
2706 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2709 unsigned int trace_ctx)
2711 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2714 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2715 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2716 static int trace_buffered_event_ref;
2719 * trace_buffered_event_enable - enable buffering events
2721 * When events are being filtered, it is quicker to use a temporary
2722 * buffer to write the event data into if there's a likely chance
2723 * that it will not be committed. The discard of the ring buffer
2724 * is not as fast as committing, and is much slower than copying
2727 * When an event is to be filtered, allocate per cpu buffers to
2728 * write the event data into, and if the event is filtered and discarded
2729 * it is simply dropped, otherwise, the entire data is to be committed
2732 void trace_buffered_event_enable(void)
2734 struct ring_buffer_event *event;
2738 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2740 if (trace_buffered_event_ref++)
2743 for_each_tracing_cpu(cpu) {
2744 page = alloc_pages_node(cpu_to_node(cpu),
2745 GFP_KERNEL | __GFP_NORETRY, 0);
2749 event = page_address(page);
2750 memset(event, 0, sizeof(*event));
2752 per_cpu(trace_buffered_event, cpu) = event;
2755 if (cpu == smp_processor_id() &&
2756 __this_cpu_read(trace_buffered_event) !=
2757 per_cpu(trace_buffered_event, cpu))
2764 trace_buffered_event_disable();
2767 static void enable_trace_buffered_event(void *data)
2769 /* Probably not needed, but do it anyway */
2771 this_cpu_dec(trace_buffered_event_cnt);
2774 static void disable_trace_buffered_event(void *data)
2776 this_cpu_inc(trace_buffered_event_cnt);
2780 * trace_buffered_event_disable - disable buffering events
2782 * When a filter is removed, it is faster to not use the buffered
2783 * events, and to commit directly into the ring buffer. Free up
2784 * the temp buffers when there are no more users. This requires
2785 * special synchronization with current events.
2787 void trace_buffered_event_disable(void)
2791 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2793 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2796 if (--trace_buffered_event_ref)
2800 /* For each CPU, set the buffer as used. */
2801 smp_call_function_many(tracing_buffer_mask,
2802 disable_trace_buffered_event, NULL, 1);
2805 /* Wait for all current users to finish */
2808 for_each_tracing_cpu(cpu) {
2809 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2810 per_cpu(trace_buffered_event, cpu) = NULL;
2813 * Make sure trace_buffered_event is NULL before clearing
2814 * trace_buffered_event_cnt.
2819 /* Do the work on each cpu */
2820 smp_call_function_many(tracing_buffer_mask,
2821 enable_trace_buffered_event, NULL, 1);
2825 static struct trace_buffer *temp_buffer;
2827 struct ring_buffer_event *
2828 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2829 struct trace_event_file *trace_file,
2830 int type, unsigned long len,
2831 unsigned int trace_ctx)
2833 struct ring_buffer_event *entry;
2834 struct trace_array *tr = trace_file->tr;
2837 *current_rb = tr->array_buffer.buffer;
2839 if (!tr->no_filter_buffering_ref &&
2840 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2841 preempt_disable_notrace();
2843 * Filtering is on, so try to use the per cpu buffer first.
2844 * This buffer will simulate a ring_buffer_event,
2845 * where the type_len is zero and the array[0] will
2846 * hold the full length.
2847 * (see include/linux/ring-buffer.h for details on
2848 * how the ring_buffer_event is structured).
2850 * Using a temp buffer during filtering and copying it
2851 * on a matched filter is quicker than writing directly
2852 * into the ring buffer and then discarding it when
2853 * it doesn't match. That is because the discard
2854 * requires several atomic operations to get right.
2855 * Copying on match and doing nothing on a failed match
2856 * is still quicker than no copy on match, but having
2857 * to discard out of the ring buffer on a failed match.
2859 if ((entry = __this_cpu_read(trace_buffered_event))) {
2860 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2862 val = this_cpu_inc_return(trace_buffered_event_cnt);
2865 * Preemption is disabled, but interrupts and NMIs
2866 * can still come in now. If that happens after
2867 * the above increment, then it will have to go
2868 * back to the old method of allocating the event
2869 * on the ring buffer, and if the filter fails, it
2870 * will have to call ring_buffer_discard_commit()
2873 * Need to also check the unlikely case that the
2874 * length is bigger than the temp buffer size.
2875 * If that happens, then the reserve is pretty much
2876 * guaranteed to fail, as the ring buffer currently
2877 * only allows events less than a page. But that may
2878 * change in the future, so let the ring buffer reserve
2879 * handle the failure in that case.
2881 if (val == 1 && likely(len <= max_len)) {
2882 trace_event_setup(entry, type, trace_ctx);
2883 entry->array[0] = len;
2884 /* Return with preemption disabled */
2887 this_cpu_dec(trace_buffered_event_cnt);
2889 /* __trace_buffer_lock_reserve() disables preemption */
2890 preempt_enable_notrace();
2893 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2896 * If tracing is off, but we have triggers enabled
2897 * we still need to look at the event data. Use the temp_buffer
2898 * to store the trace event for the trigger to use. It's recursive
2899 * safe and will not be recorded anywhere.
2901 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2902 *current_rb = temp_buffer;
2903 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2908 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2910 static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2911 static DEFINE_MUTEX(tracepoint_printk_mutex);
2913 static void output_printk(struct trace_event_buffer *fbuffer)
2915 struct trace_event_call *event_call;
2916 struct trace_event_file *file;
2917 struct trace_event *event;
2918 unsigned long flags;
2919 struct trace_iterator *iter = tracepoint_print_iter;
2921 /* We should never get here if iter is NULL */
2922 if (WARN_ON_ONCE(!iter))
2925 event_call = fbuffer->trace_file->event_call;
2926 if (!event_call || !event_call->event.funcs ||
2927 !event_call->event.funcs->trace)
2930 file = fbuffer->trace_file;
2931 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2932 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2933 !filter_match_preds(file->filter, fbuffer->entry)))
2936 event = &fbuffer->trace_file->event_call->event;
2938 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2939 trace_seq_init(&iter->seq);
2940 iter->ent = fbuffer->entry;
2941 event_call->event.funcs->trace(iter, 0, event);
2942 trace_seq_putc(&iter->seq, 0);
2943 printk("%s", iter->seq.buffer);
2945 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2948 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2949 void *buffer, size_t *lenp,
2952 int save_tracepoint_printk;
2955 mutex_lock(&tracepoint_printk_mutex);
2956 save_tracepoint_printk = tracepoint_printk;
2958 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2961 * This will force exiting early, as tracepoint_printk
2962 * is always zero when tracepoint_printk_iter is not allocated
2964 if (!tracepoint_print_iter)
2965 tracepoint_printk = 0;
2967 if (save_tracepoint_printk == tracepoint_printk)
2970 if (tracepoint_printk)
2971 static_key_enable(&tracepoint_printk_key.key);
2973 static_key_disable(&tracepoint_printk_key.key);
2976 mutex_unlock(&tracepoint_printk_mutex);
2981 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2983 enum event_trigger_type tt = ETT_NONE;
2984 struct trace_event_file *file = fbuffer->trace_file;
2986 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2987 fbuffer->entry, &tt))
2990 if (static_key_false(&tracepoint_printk_key.key))
2991 output_printk(fbuffer);
2993 if (static_branch_unlikely(&trace_event_exports_enabled))
2994 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2996 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2997 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
3001 event_triggers_post_call(file, tt);
3004 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
3009 * trace_buffer_unlock_commit_regs()
3010 * trace_event_buffer_commit()
3011 * trace_event_raw_event_xxx()
3013 # define STACK_SKIP 3
3015 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
3016 struct trace_buffer *buffer,
3017 struct ring_buffer_event *event,
3018 unsigned int trace_ctx,
3019 struct pt_regs *regs)
3021 __buffer_unlock_commit(buffer, event);
3024 * If regs is not set, then skip the necessary functions.
3025 * Note, we can still get here via blktrace, wakeup tracer
3026 * and mmiotrace, but that's ok if they lose a function or
3027 * two. They are not that meaningful.
3029 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
3030 ftrace_trace_userstack(tr, buffer, trace_ctx);
3034 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
3037 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
3038 struct ring_buffer_event *event)
3040 __buffer_unlock_commit(buffer, event);
3044 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
3045 parent_ip, unsigned int trace_ctx)
3047 struct trace_event_call *call = &event_function;
3048 struct trace_buffer *buffer = tr->array_buffer.buffer;
3049 struct ring_buffer_event *event;
3050 struct ftrace_entry *entry;
3052 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
3056 entry = ring_buffer_event_data(event);
3058 entry->parent_ip = parent_ip;
3060 if (!call_filter_check_discard(call, entry, buffer, event)) {
3061 if (static_branch_unlikely(&trace_function_exports_enabled))
3062 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
3063 __buffer_unlock_commit(buffer, event);
3067 #ifdef CONFIG_STACKTRACE
3069 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
3070 #define FTRACE_KSTACK_NESTING 4
3072 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3074 struct ftrace_stack {
3075 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3079 struct ftrace_stacks {
3080 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3083 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3084 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3086 static void __ftrace_trace_stack(struct trace_buffer *buffer,
3087 unsigned int trace_ctx,
3088 int skip, struct pt_regs *regs)
3090 struct trace_event_call *call = &event_kernel_stack;
3091 struct ring_buffer_event *event;
3092 unsigned int size, nr_entries;
3093 struct ftrace_stack *fstack;
3094 struct stack_entry *entry;
3098 * Add one, for this function and the call to save_stack_trace()
3099 * If regs is set, then these functions will not be in the way.
3101 #ifndef CONFIG_UNWINDER_ORC
3106 preempt_disable_notrace();
3108 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3110 /* This should never happen. If it does, yell once and skip */
3111 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3115 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3116 * interrupt will either see the value pre increment or post
3117 * increment. If the interrupt happens pre increment it will have
3118 * restored the counter when it returns. We just need a barrier to
3119 * keep gcc from moving things around.
3123 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3124 size = ARRAY_SIZE(fstack->calls);
3127 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3130 nr_entries = stack_trace_save(fstack->calls, size, skip);
3133 size = nr_entries * sizeof(unsigned long);
3134 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3135 (sizeof(*entry) - sizeof(entry->caller)) + size,
3139 entry = ring_buffer_event_data(event);
3141 memcpy(&entry->caller, fstack->calls, size);
3142 entry->size = nr_entries;
3144 if (!call_filter_check_discard(call, entry, buffer, event))
3145 __buffer_unlock_commit(buffer, event);
3148 /* Again, don't let gcc optimize things here */
3150 __this_cpu_dec(ftrace_stack_reserve);
3151 preempt_enable_notrace();
3155 static inline void ftrace_trace_stack(struct trace_array *tr,
3156 struct trace_buffer *buffer,
3157 unsigned int trace_ctx,
3158 int skip, struct pt_regs *regs)
3160 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3163 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3166 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3169 struct trace_buffer *buffer = tr->array_buffer.buffer;
3171 if (rcu_is_watching()) {
3172 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3176 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3180 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3181 * but if the above rcu_is_watching() failed, then the NMI
3182 * triggered someplace critical, and ct_irq_enter() should
3183 * not be called from NMI.
3185 if (unlikely(in_nmi()))
3188 ct_irq_enter_irqson();
3189 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3190 ct_irq_exit_irqson();
3194 * trace_dump_stack - record a stack back trace in the trace buffer
3195 * @skip: Number of functions to skip (helper handlers)
3197 void trace_dump_stack(int skip)
3199 if (tracing_disabled || tracing_selftest_running)
3202 #ifndef CONFIG_UNWINDER_ORC
3203 /* Skip 1 to skip this function. */
3206 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3207 tracing_gen_ctx(), skip, NULL);
3209 EXPORT_SYMBOL_GPL(trace_dump_stack);
3211 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3212 static DEFINE_PER_CPU(int, user_stack_count);
3215 ftrace_trace_userstack(struct trace_array *tr,
3216 struct trace_buffer *buffer, unsigned int trace_ctx)
3218 struct trace_event_call *call = &event_user_stack;
3219 struct ring_buffer_event *event;
3220 struct userstack_entry *entry;
3222 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3226 * NMIs can not handle page faults, even with fix ups.
3227 * The save user stack can (and often does) fault.
3229 if (unlikely(in_nmi()))
3233 * prevent recursion, since the user stack tracing may
3234 * trigger other kernel events.
3237 if (__this_cpu_read(user_stack_count))
3240 __this_cpu_inc(user_stack_count);
3242 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3243 sizeof(*entry), trace_ctx);
3245 goto out_drop_count;
3246 entry = ring_buffer_event_data(event);
3248 entry->tgid = current->tgid;
3249 memset(&entry->caller, 0, sizeof(entry->caller));
3251 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3252 if (!call_filter_check_discard(call, entry, buffer, event))
3253 __buffer_unlock_commit(buffer, event);
3256 __this_cpu_dec(user_stack_count);
3260 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3261 static void ftrace_trace_userstack(struct trace_array *tr,
3262 struct trace_buffer *buffer,
3263 unsigned int trace_ctx)
3266 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3268 #endif /* CONFIG_STACKTRACE */
3271 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3272 unsigned long long delta)
3274 entry->bottom_delta_ts = delta & U32_MAX;
3275 entry->top_delta_ts = (delta >> 32);
3278 void trace_last_func_repeats(struct trace_array *tr,
3279 struct trace_func_repeats *last_info,
3280 unsigned int trace_ctx)
3282 struct trace_buffer *buffer = tr->array_buffer.buffer;
3283 struct func_repeats_entry *entry;
3284 struct ring_buffer_event *event;
3287 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3288 sizeof(*entry), trace_ctx);
3292 delta = ring_buffer_event_time_stamp(buffer, event) -
3293 last_info->ts_last_call;
3295 entry = ring_buffer_event_data(event);
3296 entry->ip = last_info->ip;
3297 entry->parent_ip = last_info->parent_ip;
3298 entry->count = last_info->count;
3299 func_repeats_set_delta_ts(entry, delta);
3301 __buffer_unlock_commit(buffer, event);
3304 /* created for use with alloc_percpu */
3305 struct trace_buffer_struct {
3307 char buffer[4][TRACE_BUF_SIZE];
3310 static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3313 * This allows for lockless recording. If we're nested too deeply, then
3314 * this returns NULL.
3316 static char *get_trace_buf(void)
3318 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3320 if (!trace_percpu_buffer || buffer->nesting >= 4)
3325 /* Interrupts must see nesting incremented before we use the buffer */
3327 return &buffer->buffer[buffer->nesting - 1][0];
3330 static void put_trace_buf(void)
3332 /* Don't let the decrement of nesting leak before this */
3334 this_cpu_dec(trace_percpu_buffer->nesting);
3337 static int alloc_percpu_trace_buffer(void)
3339 struct trace_buffer_struct __percpu *buffers;
3341 if (trace_percpu_buffer)
3344 buffers = alloc_percpu(struct trace_buffer_struct);
3345 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3348 trace_percpu_buffer = buffers;
3352 static int buffers_allocated;
3354 void trace_printk_init_buffers(void)
3356 if (buffers_allocated)
3359 if (alloc_percpu_trace_buffer())
3362 /* trace_printk() is for debug use only. Don't use it in production. */
3365 pr_warn("**********************************************************\n");
3366 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3368 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3370 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3371 pr_warn("** unsafe for production use. **\n");
3373 pr_warn("** If you see this message and you are not debugging **\n");
3374 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3376 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3377 pr_warn("**********************************************************\n");
3379 /* Expand the buffers to set size */
3380 tracing_update_buffers();
3382 buffers_allocated = 1;
3385 * trace_printk_init_buffers() can be called by modules.
3386 * If that happens, then we need to start cmdline recording
3387 * directly here. If the global_trace.buffer is already
3388 * allocated here, then this was called by module code.
3390 if (global_trace.array_buffer.buffer)
3391 tracing_start_cmdline_record();
3393 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3395 void trace_printk_start_comm(void)
3397 /* Start tracing comms if trace printk is set */
3398 if (!buffers_allocated)
3400 tracing_start_cmdline_record();
3403 static void trace_printk_start_stop_comm(int enabled)
3405 if (!buffers_allocated)
3409 tracing_start_cmdline_record();
3411 tracing_stop_cmdline_record();
3415 * trace_vbprintk - write binary msg to tracing buffer
3416 * @ip: The address of the caller
3417 * @fmt: The string format to write to the buffer
3418 * @args: Arguments for @fmt
3420 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3422 struct trace_event_call *call = &event_bprint;
3423 struct ring_buffer_event *event;
3424 struct trace_buffer *buffer;
3425 struct trace_array *tr = &global_trace;
3426 struct bprint_entry *entry;
3427 unsigned int trace_ctx;
3431 if (unlikely(tracing_selftest_running || tracing_disabled))
3434 /* Don't pollute graph traces with trace_vprintk internals */
3435 pause_graph_tracing();
3437 trace_ctx = tracing_gen_ctx();
3438 preempt_disable_notrace();
3440 tbuffer = get_trace_buf();
3446 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3448 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3451 size = sizeof(*entry) + sizeof(u32) * len;
3452 buffer = tr->array_buffer.buffer;
3453 ring_buffer_nest_start(buffer);
3454 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3458 entry = ring_buffer_event_data(event);
3462 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3463 if (!call_filter_check_discard(call, entry, buffer, event)) {
3464 __buffer_unlock_commit(buffer, event);
3465 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3469 ring_buffer_nest_end(buffer);
3474 preempt_enable_notrace();
3475 unpause_graph_tracing();
3479 EXPORT_SYMBOL_GPL(trace_vbprintk);
3483 __trace_array_vprintk(struct trace_buffer *buffer,
3484 unsigned long ip, const char *fmt, va_list args)
3486 struct trace_event_call *call = &event_print;
3487 struct ring_buffer_event *event;
3489 struct print_entry *entry;
3490 unsigned int trace_ctx;
3493 if (tracing_disabled || tracing_selftest_running)
3496 /* Don't pollute graph traces with trace_vprintk internals */
3497 pause_graph_tracing();
3499 trace_ctx = tracing_gen_ctx();
3500 preempt_disable_notrace();
3503 tbuffer = get_trace_buf();
3509 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3511 size = sizeof(*entry) + len + 1;
3512 ring_buffer_nest_start(buffer);
3513 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3517 entry = ring_buffer_event_data(event);
3520 memcpy(&entry->buf, tbuffer, len + 1);
3521 if (!call_filter_check_discard(call, entry, buffer, event)) {
3522 __buffer_unlock_commit(buffer, event);
3523 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3527 ring_buffer_nest_end(buffer);
3531 preempt_enable_notrace();
3532 unpause_graph_tracing();
3538 int trace_array_vprintk(struct trace_array *tr,
3539 unsigned long ip, const char *fmt, va_list args)
3541 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3545 * trace_array_printk - Print a message to a specific instance
3546 * @tr: The instance trace_array descriptor
3547 * @ip: The instruction pointer that this is called from.
3548 * @fmt: The format to print (printf format)
3550 * If a subsystem sets up its own instance, they have the right to
3551 * printk strings into their tracing instance buffer using this
3552 * function. Note, this function will not write into the top level
3553 * buffer (use trace_printk() for that), as writing into the top level
3554 * buffer should only have events that can be individually disabled.
3555 * trace_printk() is only used for debugging a kernel, and should not
3556 * be ever incorporated in normal use.
3558 * trace_array_printk() can be used, as it will not add noise to the
3559 * top level tracing buffer.
3561 * Note, trace_array_init_printk() must be called on @tr before this
3565 int trace_array_printk(struct trace_array *tr,
3566 unsigned long ip, const char *fmt, ...)
3574 /* This is only allowed for created instances */
3575 if (tr == &global_trace)
3578 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3582 ret = trace_array_vprintk(tr, ip, fmt, ap);
3586 EXPORT_SYMBOL_GPL(trace_array_printk);
3589 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3590 * @tr: The trace array to initialize the buffers for
3592 * As trace_array_printk() only writes into instances, they are OK to
3593 * have in the kernel (unlike trace_printk()). This needs to be called
3594 * before trace_array_printk() can be used on a trace_array.
3596 int trace_array_init_printk(struct trace_array *tr)
3601 /* This is only allowed for created instances */
3602 if (tr == &global_trace)
3605 return alloc_percpu_trace_buffer();
3607 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3610 int trace_array_printk_buf(struct trace_buffer *buffer,
3611 unsigned long ip, const char *fmt, ...)
3616 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3620 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3626 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3628 return trace_array_vprintk(&global_trace, ip, fmt, args);
3630 EXPORT_SYMBOL_GPL(trace_vprintk);
3632 static void trace_iterator_increment(struct trace_iterator *iter)
3634 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3638 ring_buffer_iter_advance(buf_iter);
3641 static struct trace_entry *
3642 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3643 unsigned long *lost_events)
3645 struct ring_buffer_event *event;
3646 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3649 event = ring_buffer_iter_peek(buf_iter, ts);
3651 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3652 (unsigned long)-1 : 0;
3654 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3659 iter->ent_size = ring_buffer_event_length(event);
3660 return ring_buffer_event_data(event);
3666 static struct trace_entry *
3667 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3668 unsigned long *missing_events, u64 *ent_ts)
3670 struct trace_buffer *buffer = iter->array_buffer->buffer;
3671 struct trace_entry *ent, *next = NULL;
3672 unsigned long lost_events = 0, next_lost = 0;
3673 int cpu_file = iter->cpu_file;
3674 u64 next_ts = 0, ts;
3680 * If we are in a per_cpu trace file, don't bother by iterating over
3681 * all cpu and peek directly.
3683 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3684 if (ring_buffer_empty_cpu(buffer, cpu_file))
3686 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3688 *ent_cpu = cpu_file;
3693 for_each_tracing_cpu(cpu) {
3695 if (ring_buffer_empty_cpu(buffer, cpu))
3698 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3701 * Pick the entry with the smallest timestamp:
3703 if (ent && (!next || ts < next_ts)) {
3707 next_lost = lost_events;
3708 next_size = iter->ent_size;
3712 iter->ent_size = next_size;
3715 *ent_cpu = next_cpu;
3721 *missing_events = next_lost;
3726 #define STATIC_FMT_BUF_SIZE 128
3727 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3729 static char *trace_iter_expand_format(struct trace_iterator *iter)
3734 * iter->tr is NULL when used with tp_printk, which makes
3735 * this get called where it is not safe to call krealloc().
3737 if (!iter->tr || iter->fmt == static_fmt_buf)
3740 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3743 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3750 /* Returns true if the string is safe to dereference from an event */
3751 static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3754 unsigned long addr = (unsigned long)str;
3755 struct trace_event *trace_event;
3756 struct trace_event_call *event;
3758 /* Ignore strings with no length */
3762 /* OK if part of the event data */
3763 if ((addr >= (unsigned long)iter->ent) &&
3764 (addr < (unsigned long)iter->ent + iter->ent_size))
3767 /* OK if part of the temp seq buffer */
3768 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3769 (addr < (unsigned long)iter->tmp_seq.buffer + PAGE_SIZE))
3772 /* Core rodata can not be freed */
3773 if (is_kernel_rodata(addr))
3776 if (trace_is_tracepoint_string(str))
3780 * Now this could be a module event, referencing core module
3781 * data, which is OK.
3786 trace_event = ftrace_find_event(iter->ent->type);
3790 event = container_of(trace_event, struct trace_event_call, event);
3791 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3794 /* Would rather have rodata, but this will suffice */
3795 if (within_module_core(addr, event->module))
3801 static const char *show_buffer(struct trace_seq *s)
3803 struct seq_buf *seq = &s->seq;
3805 seq_buf_terminate(seq);
3810 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3812 static int test_can_verify_check(const char *fmt, ...)
3819 * The verifier is dependent on vsnprintf() modifies the va_list
3820 * passed to it, where it is sent as a reference. Some architectures
3821 * (like x86_32) passes it by value, which means that vsnprintf()
3822 * does not modify the va_list passed to it, and the verifier
3823 * would then need to be able to understand all the values that
3824 * vsnprintf can use. If it is passed by value, then the verifier
3828 vsnprintf(buf, 16, "%d", ap);
3829 ret = va_arg(ap, int);
3835 static void test_can_verify(void)
3837 if (!test_can_verify_check("%d %d", 0, 1)) {
3838 pr_info("trace event string verifier disabled\n");
3839 static_branch_inc(&trace_no_verify);
3844 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3845 * @iter: The iterator that holds the seq buffer and the event being printed
3846 * @fmt: The format used to print the event
3847 * @ap: The va_list holding the data to print from @fmt.
3849 * This writes the data into the @iter->seq buffer using the data from
3850 * @fmt and @ap. If the format has a %s, then the source of the string
3851 * is examined to make sure it is safe to print, otherwise it will
3852 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3855 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3858 const char *p = fmt;
3862 if (WARN_ON_ONCE(!fmt))
3865 if (static_branch_unlikely(&trace_no_verify))
3868 /* Don't bother checking when doing a ftrace_dump() */
3869 if (iter->fmt == static_fmt_buf)
3878 /* We only care about %s and variants */
3879 for (i = 0; p[i]; i++) {
3880 if (i + 1 >= iter->fmt_size) {
3882 * If we can't expand the copy buffer,
3885 if (!trace_iter_expand_format(iter))
3889 if (p[i] == '\\' && p[i+1]) {
3894 /* Need to test cases like %08.*s */
3895 for (j = 1; p[i+j]; j++) {
3896 if (isdigit(p[i+j]) ||
3899 if (p[i+j] == '*') {
3911 /* If no %s found then just print normally */
3915 /* Copy up to the %s, and print that */
3916 strncpy(iter->fmt, p, i);
3917 iter->fmt[i] = '\0';
3918 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3921 * If iter->seq is full, the above call no longer guarantees
3922 * that ap is in sync with fmt processing, and further calls
3923 * to va_arg() can return wrong positional arguments.
3925 * Ensure that ap is no longer used in this case.
3927 if (iter->seq.full) {
3933 len = va_arg(ap, int);
3935 /* The ap now points to the string data of the %s */
3936 str = va_arg(ap, const char *);
3939 * If you hit this warning, it is likely that the
3940 * trace event in question used %s on a string that
3941 * was saved at the time of the event, but may not be
3942 * around when the trace is read. Use __string(),
3943 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3944 * instead. See samples/trace_events/trace-events-sample.h
3947 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3948 "fmt: '%s' current_buffer: '%s'",
3949 fmt, show_buffer(&iter->seq))) {
3952 /* Try to safely read the string */
3954 if (len + 1 > iter->fmt_size)
3955 len = iter->fmt_size - 1;
3958 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3962 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3966 trace_seq_printf(&iter->seq, "(0x%px)", str);
3968 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3970 str = "[UNSAFE-MEMORY]";
3971 strcpy(iter->fmt, "%s");
3973 strncpy(iter->fmt, p + i, j + 1);
3974 iter->fmt[j+1] = '\0';
3977 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3979 trace_seq_printf(&iter->seq, iter->fmt, str);
3985 trace_seq_vprintf(&iter->seq, p, ap);
3988 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3990 const char *p, *new_fmt;
3993 if (WARN_ON_ONCE(!fmt))
3996 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
4000 new_fmt = q = iter->fmt;
4002 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
4003 if (!trace_iter_expand_format(iter))
4006 q += iter->fmt - new_fmt;
4007 new_fmt = iter->fmt;
4012 /* Replace %p with %px */
4016 } else if (p[0] == 'p' && !isalnum(p[1])) {
4027 #define STATIC_TEMP_BUF_SIZE 128
4028 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
4030 /* Find the next real entry, without updating the iterator itself */
4031 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
4032 int *ent_cpu, u64 *ent_ts)
4034 /* __find_next_entry will reset ent_size */
4035 int ent_size = iter->ent_size;
4036 struct trace_entry *entry;
4039 * If called from ftrace_dump(), then the iter->temp buffer
4040 * will be the static_temp_buf and not created from kmalloc.
4041 * If the entry size is greater than the buffer, we can
4042 * not save it. Just return NULL in that case. This is only
4043 * used to add markers when two consecutive events' time
4044 * stamps have a large delta. See trace_print_lat_context()
4046 if (iter->temp == static_temp_buf &&
4047 STATIC_TEMP_BUF_SIZE < ent_size)
4051 * The __find_next_entry() may call peek_next_entry(), which may
4052 * call ring_buffer_peek() that may make the contents of iter->ent
4053 * undefined. Need to copy iter->ent now.
4055 if (iter->ent && iter->ent != iter->temp) {
4056 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4057 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4059 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4064 iter->temp_size = iter->ent_size;
4066 memcpy(iter->temp, iter->ent, iter->ent_size);
4067 iter->ent = iter->temp;
4069 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4070 /* Put back the original ent_size */
4071 iter->ent_size = ent_size;
4076 /* Find the next real entry, and increment the iterator to the next entry */
4077 void *trace_find_next_entry_inc(struct trace_iterator *iter)
4079 iter->ent = __find_next_entry(iter, &iter->cpu,
4080 &iter->lost_events, &iter->ts);
4083 trace_iterator_increment(iter);
4085 return iter->ent ? iter : NULL;
4088 static void trace_consume(struct trace_iterator *iter)
4090 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4091 &iter->lost_events);
4094 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4096 struct trace_iterator *iter = m->private;
4100 WARN_ON_ONCE(iter->leftover);
4104 /* can't go backwards */
4109 ent = trace_find_next_entry_inc(iter);
4113 while (ent && iter->idx < i)
4114 ent = trace_find_next_entry_inc(iter);
4121 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4123 struct ring_buffer_iter *buf_iter;
4124 unsigned long entries = 0;
4127 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4129 buf_iter = trace_buffer_iter(iter, cpu);
4133 ring_buffer_iter_reset(buf_iter);
4136 * We could have the case with the max latency tracers
4137 * that a reset never took place on a cpu. This is evident
4138 * by the timestamp being before the start of the buffer.
4140 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4141 if (ts >= iter->array_buffer->time_start)
4144 ring_buffer_iter_advance(buf_iter);
4147 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4151 * The current tracer is copied to avoid a global locking
4154 static void *s_start(struct seq_file *m, loff_t *pos)
4156 struct trace_iterator *iter = m->private;
4157 struct trace_array *tr = iter->tr;
4158 int cpu_file = iter->cpu_file;
4164 * copy the tracer to avoid using a global lock all around.
4165 * iter->trace is a copy of current_trace, the pointer to the
4166 * name may be used instead of a strcmp(), as iter->trace->name
4167 * will point to the same string as current_trace->name.
4169 mutex_lock(&trace_types_lock);
4170 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
4171 *iter->trace = *tr->current_trace;
4172 mutex_unlock(&trace_types_lock);
4174 #ifdef CONFIG_TRACER_MAX_TRACE
4175 if (iter->snapshot && iter->trace->use_max_tr)
4176 return ERR_PTR(-EBUSY);
4179 if (*pos != iter->pos) {
4184 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4185 for_each_tracing_cpu(cpu)
4186 tracing_iter_reset(iter, cpu);
4188 tracing_iter_reset(iter, cpu_file);
4191 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4196 * If we overflowed the seq_file before, then we want
4197 * to just reuse the trace_seq buffer again.
4203 p = s_next(m, p, &l);
4207 trace_event_read_lock();
4208 trace_access_lock(cpu_file);
4212 static void s_stop(struct seq_file *m, void *p)
4214 struct trace_iterator *iter = m->private;
4216 #ifdef CONFIG_TRACER_MAX_TRACE
4217 if (iter->snapshot && iter->trace->use_max_tr)
4221 trace_access_unlock(iter->cpu_file);
4222 trace_event_read_unlock();
4226 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4227 unsigned long *entries, int cpu)
4229 unsigned long count;
4231 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4233 * If this buffer has skipped entries, then we hold all
4234 * entries for the trace and we need to ignore the
4235 * ones before the time stamp.
4237 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4238 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4239 /* total is the same as the entries */
4243 ring_buffer_overrun_cpu(buf->buffer, cpu);
4248 get_total_entries(struct array_buffer *buf,
4249 unsigned long *total, unsigned long *entries)
4257 for_each_tracing_cpu(cpu) {
4258 get_total_entries_cpu(buf, &t, &e, cpu);
4264 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4266 unsigned long total, entries;
4271 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4276 unsigned long trace_total_entries(struct trace_array *tr)
4278 unsigned long total, entries;
4283 get_total_entries(&tr->array_buffer, &total, &entries);
4288 static void print_lat_help_header(struct seq_file *m)
4290 seq_puts(m, "# _------=> CPU# \n"
4291 "# / _-----=> irqs-off/BH-disabled\n"
4292 "# | / _----=> need-resched \n"
4293 "# || / _---=> hardirq/softirq \n"
4294 "# ||| / _--=> preempt-depth \n"
4295 "# |||| / _-=> migrate-disable \n"
4296 "# ||||| / delay \n"
4297 "# cmd pid |||||| time | caller \n"
4298 "# \\ / |||||| \\ | / \n");
4301 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4303 unsigned long total;
4304 unsigned long entries;
4306 get_total_entries(buf, &total, &entries);
4307 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4308 entries, total, num_online_cpus());
4312 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4315 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4317 print_event_info(buf, m);
4319 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4320 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4323 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4326 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4327 static const char space[] = " ";
4328 int prec = tgid ? 12 : 2;
4330 print_event_info(buf, m);
4332 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4333 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4334 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4335 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4336 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4337 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4338 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4339 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4343 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4345 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4346 struct array_buffer *buf = iter->array_buffer;
4347 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4348 struct tracer *type = iter->trace;
4349 unsigned long entries;
4350 unsigned long total;
4351 const char *name = type->name;
4353 get_total_entries(buf, &total, &entries);
4355 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4357 seq_puts(m, "# -----------------------------------"
4358 "---------------------------------\n");
4359 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4360 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4361 nsecs_to_usecs(data->saved_latency),
4365 preempt_model_none() ? "server" :
4366 preempt_model_voluntary() ? "desktop" :
4367 preempt_model_full() ? "preempt" :
4368 preempt_model_rt() ? "preempt_rt" :
4370 /* These are reserved for later use */
4373 seq_printf(m, " #P:%d)\n", num_online_cpus());
4377 seq_puts(m, "# -----------------\n");
4378 seq_printf(m, "# | task: %.16s-%d "
4379 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4380 data->comm, data->pid,
4381 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4382 data->policy, data->rt_priority);
4383 seq_puts(m, "# -----------------\n");
4385 if (data->critical_start) {
4386 seq_puts(m, "# => started at: ");
4387 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4388 trace_print_seq(m, &iter->seq);
4389 seq_puts(m, "\n# => ended at: ");
4390 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4391 trace_print_seq(m, &iter->seq);
4392 seq_puts(m, "\n#\n");
4398 static void test_cpu_buff_start(struct trace_iterator *iter)
4400 struct trace_seq *s = &iter->seq;
4401 struct trace_array *tr = iter->tr;
4403 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4406 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4409 if (cpumask_available(iter->started) &&
4410 cpumask_test_cpu(iter->cpu, iter->started))
4413 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4416 if (cpumask_available(iter->started))
4417 cpumask_set_cpu(iter->cpu, iter->started);
4419 /* Don't print started cpu buffer for the first entry of the trace */
4421 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4425 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4427 struct trace_array *tr = iter->tr;
4428 struct trace_seq *s = &iter->seq;
4429 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4430 struct trace_entry *entry;
4431 struct trace_event *event;
4435 test_cpu_buff_start(iter);
4437 event = ftrace_find_event(entry->type);
4439 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4440 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4441 trace_print_lat_context(iter);
4443 trace_print_context(iter);
4446 if (trace_seq_has_overflowed(s))
4447 return TRACE_TYPE_PARTIAL_LINE;
4450 return event->funcs->trace(iter, sym_flags, event);
4452 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4454 return trace_handle_return(s);
4457 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4459 struct trace_array *tr = iter->tr;
4460 struct trace_seq *s = &iter->seq;
4461 struct trace_entry *entry;
4462 struct trace_event *event;
4466 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4467 trace_seq_printf(s, "%d %d %llu ",
4468 entry->pid, iter->cpu, iter->ts);
4470 if (trace_seq_has_overflowed(s))
4471 return TRACE_TYPE_PARTIAL_LINE;
4473 event = ftrace_find_event(entry->type);
4475 return event->funcs->raw(iter, 0, event);
4477 trace_seq_printf(s, "%d ?\n", entry->type);
4479 return trace_handle_return(s);
4482 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4484 struct trace_array *tr = iter->tr;
4485 struct trace_seq *s = &iter->seq;
4486 unsigned char newline = '\n';
4487 struct trace_entry *entry;
4488 struct trace_event *event;
4492 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4493 SEQ_PUT_HEX_FIELD(s, entry->pid);
4494 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4495 SEQ_PUT_HEX_FIELD(s, iter->ts);
4496 if (trace_seq_has_overflowed(s))
4497 return TRACE_TYPE_PARTIAL_LINE;
4500 event = ftrace_find_event(entry->type);
4502 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4503 if (ret != TRACE_TYPE_HANDLED)
4507 SEQ_PUT_FIELD(s, newline);
4509 return trace_handle_return(s);
4512 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4514 struct trace_array *tr = iter->tr;
4515 struct trace_seq *s = &iter->seq;
4516 struct trace_entry *entry;
4517 struct trace_event *event;
4521 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4522 SEQ_PUT_FIELD(s, entry->pid);
4523 SEQ_PUT_FIELD(s, iter->cpu);
4524 SEQ_PUT_FIELD(s, iter->ts);
4525 if (trace_seq_has_overflowed(s))
4526 return TRACE_TYPE_PARTIAL_LINE;
4529 event = ftrace_find_event(entry->type);
4530 return event ? event->funcs->binary(iter, 0, event) :
4534 int trace_empty(struct trace_iterator *iter)
4536 struct ring_buffer_iter *buf_iter;
4539 /* If we are looking at one CPU buffer, only check that one */
4540 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4541 cpu = iter->cpu_file;
4542 buf_iter = trace_buffer_iter(iter, cpu);
4544 if (!ring_buffer_iter_empty(buf_iter))
4547 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4553 for_each_tracing_cpu(cpu) {
4554 buf_iter = trace_buffer_iter(iter, cpu);
4556 if (!ring_buffer_iter_empty(buf_iter))
4559 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4567 /* Called with trace_event_read_lock() held. */
4568 enum print_line_t print_trace_line(struct trace_iterator *iter)
4570 struct trace_array *tr = iter->tr;
4571 unsigned long trace_flags = tr->trace_flags;
4572 enum print_line_t ret;
4574 if (iter->lost_events) {
4575 if (iter->lost_events == (unsigned long)-1)
4576 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4579 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4580 iter->cpu, iter->lost_events);
4581 if (trace_seq_has_overflowed(&iter->seq))
4582 return TRACE_TYPE_PARTIAL_LINE;
4585 if (iter->trace && iter->trace->print_line) {
4586 ret = iter->trace->print_line(iter);
4587 if (ret != TRACE_TYPE_UNHANDLED)
4591 if (iter->ent->type == TRACE_BPUTS &&
4592 trace_flags & TRACE_ITER_PRINTK &&
4593 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4594 return trace_print_bputs_msg_only(iter);
4596 if (iter->ent->type == TRACE_BPRINT &&
4597 trace_flags & TRACE_ITER_PRINTK &&
4598 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4599 return trace_print_bprintk_msg_only(iter);
4601 if (iter->ent->type == TRACE_PRINT &&
4602 trace_flags & TRACE_ITER_PRINTK &&
4603 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4604 return trace_print_printk_msg_only(iter);
4606 if (trace_flags & TRACE_ITER_BIN)
4607 return print_bin_fmt(iter);
4609 if (trace_flags & TRACE_ITER_HEX)
4610 return print_hex_fmt(iter);
4612 if (trace_flags & TRACE_ITER_RAW)
4613 return print_raw_fmt(iter);
4615 return print_trace_fmt(iter);
4618 void trace_latency_header(struct seq_file *m)
4620 struct trace_iterator *iter = m->private;
4621 struct trace_array *tr = iter->tr;
4623 /* print nothing if the buffers are empty */
4624 if (trace_empty(iter))
4627 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4628 print_trace_header(m, iter);
4630 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4631 print_lat_help_header(m);
4634 void trace_default_header(struct seq_file *m)
4636 struct trace_iterator *iter = m->private;
4637 struct trace_array *tr = iter->tr;
4638 unsigned long trace_flags = tr->trace_flags;
4640 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4643 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4644 /* print nothing if the buffers are empty */
4645 if (trace_empty(iter))
4647 print_trace_header(m, iter);
4648 if (!(trace_flags & TRACE_ITER_VERBOSE))
4649 print_lat_help_header(m);
4651 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4652 if (trace_flags & TRACE_ITER_IRQ_INFO)
4653 print_func_help_header_irq(iter->array_buffer,
4656 print_func_help_header(iter->array_buffer, m,
4662 static void test_ftrace_alive(struct seq_file *m)
4664 if (!ftrace_is_dead())
4666 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4667 "# MAY BE MISSING FUNCTION EVENTS\n");
4670 #ifdef CONFIG_TRACER_MAX_TRACE
4671 static void show_snapshot_main_help(struct seq_file *m)
4673 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4674 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4675 "# Takes a snapshot of the main buffer.\n"
4676 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4677 "# (Doesn't have to be '2' works with any number that\n"
4678 "# is not a '0' or '1')\n");
4681 static void show_snapshot_percpu_help(struct seq_file *m)
4683 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4684 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4685 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4686 "# Takes a snapshot of the main buffer for this cpu.\n");
4688 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4689 "# Must use main snapshot file to allocate.\n");
4691 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4692 "# (Doesn't have to be '2' works with any number that\n"
4693 "# is not a '0' or '1')\n");
4696 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4698 if (iter->tr->allocated_snapshot)
4699 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4701 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4703 seq_puts(m, "# Snapshot commands:\n");
4704 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4705 show_snapshot_main_help(m);
4707 show_snapshot_percpu_help(m);
4710 /* Should never be called */
4711 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4714 static int s_show(struct seq_file *m, void *v)
4716 struct trace_iterator *iter = v;
4719 if (iter->ent == NULL) {
4721 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4723 test_ftrace_alive(m);
4725 if (iter->snapshot && trace_empty(iter))
4726 print_snapshot_help(m, iter);
4727 else if (iter->trace && iter->trace->print_header)
4728 iter->trace->print_header(m);
4730 trace_default_header(m);
4732 } else if (iter->leftover) {
4734 * If we filled the seq_file buffer earlier, we
4735 * want to just show it now.
4737 ret = trace_print_seq(m, &iter->seq);
4739 /* ret should this time be zero, but you never know */
4740 iter->leftover = ret;
4743 print_trace_line(iter);
4744 ret = trace_print_seq(m, &iter->seq);
4746 * If we overflow the seq_file buffer, then it will
4747 * ask us for this data again at start up.
4749 * ret is 0 if seq_file write succeeded.
4752 iter->leftover = ret;
4759 * Should be used after trace_array_get(), trace_types_lock
4760 * ensures that i_cdev was already initialized.
4762 static inline int tracing_get_cpu(struct inode *inode)
4764 if (inode->i_cdev) /* See trace_create_cpu_file() */
4765 return (long)inode->i_cdev - 1;
4766 return RING_BUFFER_ALL_CPUS;
4769 static const struct seq_operations tracer_seq_ops = {
4776 static struct trace_iterator *
4777 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4779 struct trace_array *tr = inode->i_private;
4780 struct trace_iterator *iter;
4783 if (tracing_disabled)
4784 return ERR_PTR(-ENODEV);
4786 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4788 return ERR_PTR(-ENOMEM);
4790 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4792 if (!iter->buffer_iter)
4796 * trace_find_next_entry() may need to save off iter->ent.
4797 * It will place it into the iter->temp buffer. As most
4798 * events are less than 128, allocate a buffer of that size.
4799 * If one is greater, then trace_find_next_entry() will
4800 * allocate a new buffer to adjust for the bigger iter->ent.
4801 * It's not critical if it fails to get allocated here.
4803 iter->temp = kmalloc(128, GFP_KERNEL);
4805 iter->temp_size = 128;
4808 * trace_event_printf() may need to modify given format
4809 * string to replace %p with %px so that it shows real address
4810 * instead of hash value. However, that is only for the event
4811 * tracing, other tracer may not need. Defer the allocation
4812 * until it is needed.
4818 * We make a copy of the current tracer to avoid concurrent
4819 * changes on it while we are reading.
4821 mutex_lock(&trace_types_lock);
4822 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4826 *iter->trace = *tr->current_trace;
4828 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4833 #ifdef CONFIG_TRACER_MAX_TRACE
4834 /* Currently only the top directory has a snapshot */
4835 if (tr->current_trace->print_max || snapshot)
4836 iter->array_buffer = &tr->max_buffer;
4839 iter->array_buffer = &tr->array_buffer;
4840 iter->snapshot = snapshot;
4842 iter->cpu_file = tracing_get_cpu(inode);
4843 mutex_init(&iter->mutex);
4845 /* Notify the tracer early; before we stop tracing. */
4846 if (iter->trace->open)
4847 iter->trace->open(iter);
4849 /* Annotate start of buffers if we had overruns */
4850 if (ring_buffer_overruns(iter->array_buffer->buffer))
4851 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4853 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4854 if (trace_clocks[tr->clock_id].in_ns)
4855 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4858 * If pause-on-trace is enabled, then stop the trace while
4859 * dumping, unless this is the "snapshot" file
4861 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4862 tracing_stop_tr(tr);
4864 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4865 for_each_tracing_cpu(cpu) {
4866 iter->buffer_iter[cpu] =
4867 ring_buffer_read_prepare(iter->array_buffer->buffer,
4870 ring_buffer_read_prepare_sync();
4871 for_each_tracing_cpu(cpu) {
4872 ring_buffer_read_start(iter->buffer_iter[cpu]);
4873 tracing_iter_reset(iter, cpu);
4876 cpu = iter->cpu_file;
4877 iter->buffer_iter[cpu] =
4878 ring_buffer_read_prepare(iter->array_buffer->buffer,
4880 ring_buffer_read_prepare_sync();
4881 ring_buffer_read_start(iter->buffer_iter[cpu]);
4882 tracing_iter_reset(iter, cpu);
4885 mutex_unlock(&trace_types_lock);
4890 mutex_unlock(&trace_types_lock);
4893 kfree(iter->buffer_iter);
4895 seq_release_private(inode, file);
4896 return ERR_PTR(-ENOMEM);
4899 int tracing_open_generic(struct inode *inode, struct file *filp)
4903 ret = tracing_check_open_get_tr(NULL);
4907 filp->private_data = inode->i_private;
4911 bool tracing_is_disabled(void)
4913 return (tracing_disabled) ? true: false;
4917 * Open and update trace_array ref count.
4918 * Must have the current trace_array passed to it.
4920 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4922 struct trace_array *tr = inode->i_private;
4925 ret = tracing_check_open_get_tr(tr);
4929 filp->private_data = inode->i_private;
4934 static int tracing_mark_open(struct inode *inode, struct file *filp)
4936 stream_open(inode, filp);
4937 return tracing_open_generic_tr(inode, filp);
4940 static int tracing_release(struct inode *inode, struct file *file)
4942 struct trace_array *tr = inode->i_private;
4943 struct seq_file *m = file->private_data;
4944 struct trace_iterator *iter;
4947 if (!(file->f_mode & FMODE_READ)) {
4948 trace_array_put(tr);
4952 /* Writes do not use seq_file */
4954 mutex_lock(&trace_types_lock);
4956 for_each_tracing_cpu(cpu) {
4957 if (iter->buffer_iter[cpu])
4958 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4961 if (iter->trace && iter->trace->close)
4962 iter->trace->close(iter);
4964 if (!iter->snapshot && tr->stop_count)
4965 /* reenable tracing if it was previously enabled */
4966 tracing_start_tr(tr);
4968 __trace_array_put(tr);
4970 mutex_unlock(&trace_types_lock);
4972 mutex_destroy(&iter->mutex);
4973 free_cpumask_var(iter->started);
4977 kfree(iter->buffer_iter);
4978 seq_release_private(inode, file);
4983 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4985 struct trace_array *tr = inode->i_private;
4987 trace_array_put(tr);
4991 static int tracing_single_release_tr(struct inode *inode, struct file *file)
4993 struct trace_array *tr = inode->i_private;
4995 trace_array_put(tr);
4997 return single_release(inode, file);
5000 static int tracing_open(struct inode *inode, struct file *file)
5002 struct trace_array *tr = inode->i_private;
5003 struct trace_iterator *iter;
5006 ret = tracing_check_open_get_tr(tr);
5010 /* If this file was open for write, then erase contents */
5011 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
5012 int cpu = tracing_get_cpu(inode);
5013 struct array_buffer *trace_buf = &tr->array_buffer;
5015 #ifdef CONFIG_TRACER_MAX_TRACE
5016 if (tr->current_trace->print_max)
5017 trace_buf = &tr->max_buffer;
5020 if (cpu == RING_BUFFER_ALL_CPUS)
5021 tracing_reset_online_cpus(trace_buf);
5023 tracing_reset_cpu(trace_buf, cpu);
5026 if (file->f_mode & FMODE_READ) {
5027 iter = __tracing_open(inode, file, false);
5029 ret = PTR_ERR(iter);
5030 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
5031 iter->iter_flags |= TRACE_FILE_LAT_FMT;
5035 trace_array_put(tr);
5041 * Some tracers are not suitable for instance buffers.
5042 * A tracer is always available for the global array (toplevel)
5043 * or if it explicitly states that it is.
5046 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
5048 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5051 /* Find the next tracer that this trace array may use */
5052 static struct tracer *
5053 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5055 while (t && !trace_ok_for_array(t, tr))
5062 t_next(struct seq_file *m, void *v, loff_t *pos)
5064 struct trace_array *tr = m->private;
5065 struct tracer *t = v;
5070 t = get_tracer_for_array(tr, t->next);
5075 static void *t_start(struct seq_file *m, loff_t *pos)
5077 struct trace_array *tr = m->private;
5081 mutex_lock(&trace_types_lock);
5083 t = get_tracer_for_array(tr, trace_types);
5084 for (; t && l < *pos; t = t_next(m, t, &l))
5090 static void t_stop(struct seq_file *m, void *p)
5092 mutex_unlock(&trace_types_lock);
5095 static int t_show(struct seq_file *m, void *v)
5097 struct tracer *t = v;
5102 seq_puts(m, t->name);
5111 static const struct seq_operations show_traces_seq_ops = {
5118 static int show_traces_open(struct inode *inode, struct file *file)
5120 struct trace_array *tr = inode->i_private;
5124 ret = tracing_check_open_get_tr(tr);
5128 ret = seq_open(file, &show_traces_seq_ops);
5130 trace_array_put(tr);
5134 m = file->private_data;
5140 static int show_traces_release(struct inode *inode, struct file *file)
5142 struct trace_array *tr = inode->i_private;
5144 trace_array_put(tr);
5145 return seq_release(inode, file);
5149 tracing_write_stub(struct file *filp, const char __user *ubuf,
5150 size_t count, loff_t *ppos)
5155 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5159 if (file->f_mode & FMODE_READ)
5160 ret = seq_lseek(file, offset, whence);
5162 file->f_pos = ret = 0;
5167 static const struct file_operations tracing_fops = {
5168 .open = tracing_open,
5170 .write = tracing_write_stub,
5171 .llseek = tracing_lseek,
5172 .release = tracing_release,
5175 static const struct file_operations show_traces_fops = {
5176 .open = show_traces_open,
5178 .llseek = seq_lseek,
5179 .release = show_traces_release,
5183 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5184 size_t count, loff_t *ppos)
5186 struct trace_array *tr = file_inode(filp)->i_private;
5190 len = snprintf(NULL, 0, "%*pb\n",
5191 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5192 mask_str = kmalloc(len, GFP_KERNEL);
5196 len = snprintf(mask_str, len, "%*pb\n",
5197 cpumask_pr_args(tr->tracing_cpumask));
5202 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5210 int tracing_set_cpumask(struct trace_array *tr,
5211 cpumask_var_t tracing_cpumask_new)
5218 local_irq_disable();
5219 arch_spin_lock(&tr->max_lock);
5220 for_each_tracing_cpu(cpu) {
5222 * Increase/decrease the disabled counter if we are
5223 * about to flip a bit in the cpumask:
5225 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5226 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5227 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5228 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5230 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5231 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5232 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5233 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5236 arch_spin_unlock(&tr->max_lock);
5239 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5245 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5246 size_t count, loff_t *ppos)
5248 struct trace_array *tr = file_inode(filp)->i_private;
5249 cpumask_var_t tracing_cpumask_new;
5252 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5255 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5259 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5263 free_cpumask_var(tracing_cpumask_new);
5268 free_cpumask_var(tracing_cpumask_new);
5273 static const struct file_operations tracing_cpumask_fops = {
5274 .open = tracing_open_generic_tr,
5275 .read = tracing_cpumask_read,
5276 .write = tracing_cpumask_write,
5277 .release = tracing_release_generic_tr,
5278 .llseek = generic_file_llseek,
5281 static int tracing_trace_options_show(struct seq_file *m, void *v)
5283 struct tracer_opt *trace_opts;
5284 struct trace_array *tr = m->private;
5288 mutex_lock(&trace_types_lock);
5289 tracer_flags = tr->current_trace->flags->val;
5290 trace_opts = tr->current_trace->flags->opts;
5292 for (i = 0; trace_options[i]; i++) {
5293 if (tr->trace_flags & (1 << i))
5294 seq_printf(m, "%s\n", trace_options[i]);
5296 seq_printf(m, "no%s\n", trace_options[i]);
5299 for (i = 0; trace_opts[i].name; i++) {
5300 if (tracer_flags & trace_opts[i].bit)
5301 seq_printf(m, "%s\n", trace_opts[i].name);
5303 seq_printf(m, "no%s\n", trace_opts[i].name);
5305 mutex_unlock(&trace_types_lock);
5310 static int __set_tracer_option(struct trace_array *tr,
5311 struct tracer_flags *tracer_flags,
5312 struct tracer_opt *opts, int neg)
5314 struct tracer *trace = tracer_flags->trace;
5317 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5322 tracer_flags->val &= ~opts->bit;
5324 tracer_flags->val |= opts->bit;
5328 /* Try to assign a tracer specific option */
5329 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5331 struct tracer *trace = tr->current_trace;
5332 struct tracer_flags *tracer_flags = trace->flags;
5333 struct tracer_opt *opts = NULL;
5336 for (i = 0; tracer_flags->opts[i].name; i++) {
5337 opts = &tracer_flags->opts[i];
5339 if (strcmp(cmp, opts->name) == 0)
5340 return __set_tracer_option(tr, trace->flags, opts, neg);
5346 /* Some tracers require overwrite to stay enabled */
5347 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5349 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5355 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5359 if ((mask == TRACE_ITER_RECORD_TGID) ||
5360 (mask == TRACE_ITER_RECORD_CMD))
5361 lockdep_assert_held(&event_mutex);
5363 /* do nothing if flag is already set */
5364 if (!!(tr->trace_flags & mask) == !!enabled)
5367 /* Give the tracer a chance to approve the change */
5368 if (tr->current_trace->flag_changed)
5369 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5373 tr->trace_flags |= mask;
5375 tr->trace_flags &= ~mask;
5377 if (mask == TRACE_ITER_RECORD_CMD)
5378 trace_event_enable_cmd_record(enabled);
5380 if (mask == TRACE_ITER_RECORD_TGID) {
5382 tgid_map_max = pid_max;
5383 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5387 * Pairs with smp_load_acquire() in
5388 * trace_find_tgid_ptr() to ensure that if it observes
5389 * the tgid_map we just allocated then it also observes
5390 * the corresponding tgid_map_max value.
5392 smp_store_release(&tgid_map, map);
5395 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5399 trace_event_enable_tgid_record(enabled);
5402 if (mask == TRACE_ITER_EVENT_FORK)
5403 trace_event_follow_fork(tr, enabled);
5405 if (mask == TRACE_ITER_FUNC_FORK)
5406 ftrace_pid_follow_fork(tr, enabled);
5408 if (mask == TRACE_ITER_OVERWRITE) {
5409 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5410 #ifdef CONFIG_TRACER_MAX_TRACE
5411 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5415 if (mask == TRACE_ITER_PRINTK) {
5416 trace_printk_start_stop_comm(enabled);
5417 trace_printk_control(enabled);
5423 int trace_set_options(struct trace_array *tr, char *option)
5428 size_t orig_len = strlen(option);
5431 cmp = strstrip(option);
5433 len = str_has_prefix(cmp, "no");
5439 mutex_lock(&event_mutex);
5440 mutex_lock(&trace_types_lock);
5442 ret = match_string(trace_options, -1, cmp);
5443 /* If no option could be set, test the specific tracer options */
5445 ret = set_tracer_option(tr, cmp, neg);
5447 ret = set_tracer_flag(tr, 1 << ret, !neg);
5449 mutex_unlock(&trace_types_lock);
5450 mutex_unlock(&event_mutex);
5453 * If the first trailing whitespace is replaced with '\0' by strstrip,
5454 * turn it back into a space.
5456 if (orig_len > strlen(option))
5457 option[strlen(option)] = ' ';
5462 static void __init apply_trace_boot_options(void)
5464 char *buf = trace_boot_options_buf;
5468 option = strsep(&buf, ",");
5474 trace_set_options(&global_trace, option);
5476 /* Put back the comma to allow this to be called again */
5483 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5484 size_t cnt, loff_t *ppos)
5486 struct seq_file *m = filp->private_data;
5487 struct trace_array *tr = m->private;
5491 if (cnt >= sizeof(buf))
5494 if (copy_from_user(buf, ubuf, cnt))
5499 ret = trace_set_options(tr, buf);
5508 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5510 struct trace_array *tr = inode->i_private;
5513 ret = tracing_check_open_get_tr(tr);
5517 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5519 trace_array_put(tr);
5524 static const struct file_operations tracing_iter_fops = {
5525 .open = tracing_trace_options_open,
5527 .llseek = seq_lseek,
5528 .release = tracing_single_release_tr,
5529 .write = tracing_trace_options_write,
5532 static const char readme_msg[] =
5533 "tracing mini-HOWTO:\n\n"
5534 "# echo 0 > tracing_on : quick way to disable tracing\n"
5535 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5536 " Important files:\n"
5537 " trace\t\t\t- The static contents of the buffer\n"
5538 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5539 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5540 " current_tracer\t- function and latency tracers\n"
5541 " available_tracers\t- list of configured tracers for current_tracer\n"
5542 " error_log\t- error log for failed commands (that support it)\n"
5543 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5544 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5545 " trace_clock\t\t- change the clock used to order events\n"
5546 " local: Per cpu clock but may not be synced across CPUs\n"
5547 " global: Synced across CPUs but slows tracing down.\n"
5548 " counter: Not a clock, but just an increment\n"
5549 " uptime: Jiffy counter from time of boot\n"
5550 " perf: Same clock that perf events use\n"
5551 #ifdef CONFIG_X86_64
5552 " x86-tsc: TSC cycle counter\n"
5554 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5555 " delta: Delta difference against a buffer-wide timestamp\n"
5556 " absolute: Absolute (standalone) timestamp\n"
5557 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5558 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5559 " tracing_cpumask\t- Limit which CPUs to trace\n"
5560 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5561 "\t\t\t Remove sub-buffer with rmdir\n"
5562 " trace_options\t\t- Set format or modify how tracing happens\n"
5563 "\t\t\t Disable an option by prefixing 'no' to the\n"
5564 "\t\t\t option name\n"
5565 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5566 #ifdef CONFIG_DYNAMIC_FTRACE
5567 "\n available_filter_functions - list of functions that can be filtered on\n"
5568 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5569 "\t\t\t functions\n"
5570 "\t accepts: func_full_name or glob-matching-pattern\n"
5571 "\t modules: Can select a group via module\n"
5572 "\t Format: :mod:<module-name>\n"
5573 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5574 "\t triggers: a command to perform when function is hit\n"
5575 "\t Format: <function>:<trigger>[:count]\n"
5576 "\t trigger: traceon, traceoff\n"
5577 "\t\t enable_event:<system>:<event>\n"
5578 "\t\t disable_event:<system>:<event>\n"
5579 #ifdef CONFIG_STACKTRACE
5582 #ifdef CONFIG_TRACER_SNAPSHOT
5587 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5588 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5589 "\t The first one will disable tracing every time do_fault is hit\n"
5590 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5591 "\t The first time do trap is hit and it disables tracing, the\n"
5592 "\t counter will decrement to 2. If tracing is already disabled,\n"
5593 "\t the counter will not decrement. It only decrements when the\n"
5594 "\t trigger did work\n"
5595 "\t To remove trigger without count:\n"
5596 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5597 "\t To remove trigger with a count:\n"
5598 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5599 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5600 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5601 "\t modules: Can select a group via module command :mod:\n"
5602 "\t Does not accept triggers\n"
5603 #endif /* CONFIG_DYNAMIC_FTRACE */
5604 #ifdef CONFIG_FUNCTION_TRACER
5605 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5607 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5610 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5611 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5612 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5613 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5615 #ifdef CONFIG_TRACER_SNAPSHOT
5616 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5617 "\t\t\t snapshot buffer. Read the contents for more\n"
5618 "\t\t\t information\n"
5620 #ifdef CONFIG_STACK_TRACER
5621 " stack_trace\t\t- Shows the max stack trace when active\n"
5622 " stack_max_size\t- Shows current max stack size that was traced\n"
5623 "\t\t\t Write into this file to reset the max size (trigger a\n"
5624 "\t\t\t new trace)\n"
5625 #ifdef CONFIG_DYNAMIC_FTRACE
5626 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5629 #endif /* CONFIG_STACK_TRACER */
5630 #ifdef CONFIG_DYNAMIC_EVENTS
5631 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5632 "\t\t\t Write into this file to define/undefine new trace events.\n"
5634 #ifdef CONFIG_KPROBE_EVENTS
5635 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5636 "\t\t\t Write into this file to define/undefine new trace events.\n"
5638 #ifdef CONFIG_UPROBE_EVENTS
5639 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5640 "\t\t\t Write into this file to define/undefine new trace events.\n"
5642 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5643 "\t accepts: event-definitions (one definition per line)\n"
5644 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5645 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5646 #ifdef CONFIG_HIST_TRIGGERS
5647 "\t s:[synthetic/]<event> <field> [<field>]\n"
5649 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5650 "\t -:[<group>/][<event>]\n"
5651 #ifdef CONFIG_KPROBE_EVENTS
5652 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5653 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5655 #ifdef CONFIG_UPROBE_EVENTS
5656 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5658 "\t args: <name>=fetcharg[:type]\n"
5659 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5660 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5661 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5663 "\t $stack<index>, $stack, $retval, $comm,\n"
5665 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5666 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5667 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5668 "\t symstr, <type>\\[<array-size>\\]\n"
5669 #ifdef CONFIG_HIST_TRIGGERS
5670 "\t field: <stype> <name>;\n"
5671 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5672 "\t [unsigned] char/int/long\n"
5674 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5675 "\t of the <attached-group>/<attached-event>.\n"
5677 " events/\t\t- Directory containing all trace event subsystems:\n"
5678 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5679 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5680 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5682 " filter\t\t- If set, only events passing filter are traced\n"
5683 " events/<system>/<event>/\t- Directory containing control files for\n"
5685 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5686 " filter\t\t- If set, only events passing filter are traced\n"
5687 " trigger\t\t- If set, a command to perform when event is hit\n"
5688 "\t Format: <trigger>[:count][if <filter>]\n"
5689 "\t trigger: traceon, traceoff\n"
5690 "\t enable_event:<system>:<event>\n"
5691 "\t disable_event:<system>:<event>\n"
5692 #ifdef CONFIG_HIST_TRIGGERS
5693 "\t enable_hist:<system>:<event>\n"
5694 "\t disable_hist:<system>:<event>\n"
5696 #ifdef CONFIG_STACKTRACE
5699 #ifdef CONFIG_TRACER_SNAPSHOT
5702 #ifdef CONFIG_HIST_TRIGGERS
5703 "\t\t hist (see below)\n"
5705 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5706 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5707 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5708 "\t events/block/block_unplug/trigger\n"
5709 "\t The first disables tracing every time block_unplug is hit.\n"
5710 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5711 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5712 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5713 "\t Like function triggers, the counter is only decremented if it\n"
5714 "\t enabled or disabled tracing.\n"
5715 "\t To remove a trigger without a count:\n"
5716 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5717 "\t To remove a trigger with a count:\n"
5718 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5719 "\t Filters can be ignored when removing a trigger.\n"
5720 #ifdef CONFIG_HIST_TRIGGERS
5721 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5722 "\t Format: hist:keys=<field1[,field2,...]>\n"
5723 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5724 "\t [:values=<field1[,field2,...]>]\n"
5725 "\t [:sort=<field1[,field2,...]>]\n"
5726 "\t [:size=#entries]\n"
5727 "\t [:pause][:continue][:clear]\n"
5728 "\t [:name=histname1]\n"
5729 "\t [:nohitcount]\n"
5730 "\t [:<handler>.<action>]\n"
5731 "\t [if <filter>]\n\n"
5732 "\t Note, special fields can be used as well:\n"
5733 "\t common_timestamp - to record current timestamp\n"
5734 "\t common_cpu - to record the CPU the event happened on\n"
5736 "\t A hist trigger variable can be:\n"
5737 "\t - a reference to a field e.g. x=current_timestamp,\n"
5738 "\t - a reference to another variable e.g. y=$x,\n"
5739 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5740 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5742 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5743 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5744 "\t variable reference, field or numeric literal.\n"
5746 "\t When a matching event is hit, an entry is added to a hash\n"
5747 "\t table using the key(s) and value(s) named, and the value of a\n"
5748 "\t sum called 'hitcount' is incremented. Keys and values\n"
5749 "\t correspond to fields in the event's format description. Keys\n"
5750 "\t can be any field, or the special string 'stacktrace'.\n"
5751 "\t Compound keys consisting of up to two fields can be specified\n"
5752 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5753 "\t fields. Sort keys consisting of up to two fields can be\n"
5754 "\t specified using the 'sort' keyword. The sort direction can\n"
5755 "\t be modified by appending '.descending' or '.ascending' to a\n"
5756 "\t sort field. The 'size' parameter can be used to specify more\n"
5757 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5758 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5759 "\t its histogram data will be shared with other triggers of the\n"
5760 "\t same name, and trigger hits will update this common data.\n\n"
5761 "\t Reading the 'hist' file for the event will dump the hash\n"
5762 "\t table in its entirety to stdout. If there are multiple hist\n"
5763 "\t triggers attached to an event, there will be a table for each\n"
5764 "\t trigger in the output. The table displayed for a named\n"
5765 "\t trigger will be the same as any other instance having the\n"
5766 "\t same name. The default format used to display a given field\n"
5767 "\t can be modified by appending any of the following modifiers\n"
5768 "\t to the field name, as applicable:\n\n"
5769 "\t .hex display a number as a hex value\n"
5770 "\t .sym display an address as a symbol\n"
5771 "\t .sym-offset display an address as a symbol and offset\n"
5772 "\t .execname display a common_pid as a program name\n"
5773 "\t .syscall display a syscall id as a syscall name\n"
5774 "\t .log2 display log2 value rather than raw number\n"
5775 "\t .buckets=size display values in groups of size rather than raw number\n"
5776 "\t .usecs display a common_timestamp in microseconds\n"
5777 "\t .percent display a number of percentage value\n"
5778 "\t .graph display a bar-graph of a value\n\n"
5779 "\t The 'pause' parameter can be used to pause an existing hist\n"
5780 "\t trigger or to start a hist trigger but not log any events\n"
5781 "\t until told to do so. 'continue' can be used to start or\n"
5782 "\t restart a paused hist trigger.\n\n"
5783 "\t The 'clear' parameter will clear the contents of a running\n"
5784 "\t hist trigger and leave its current paused/active state\n"
5786 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5787 "\t raw hitcount in the histogram.\n\n"
5788 "\t The enable_hist and disable_hist triggers can be used to\n"
5789 "\t have one event conditionally start and stop another event's\n"
5790 "\t already-attached hist trigger. The syntax is analogous to\n"
5791 "\t the enable_event and disable_event triggers.\n\n"
5792 "\t Hist trigger handlers and actions are executed whenever a\n"
5793 "\t a histogram entry is added or updated. They take the form:\n\n"
5794 "\t <handler>.<action>\n\n"
5795 "\t The available handlers are:\n\n"
5796 "\t onmatch(matching.event) - invoke on addition or update\n"
5797 "\t onmax(var) - invoke if var exceeds current max\n"
5798 "\t onchange(var) - invoke action if var changes\n\n"
5799 "\t The available actions are:\n\n"
5800 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5801 "\t save(field,...) - save current event fields\n"
5802 #ifdef CONFIG_TRACER_SNAPSHOT
5803 "\t snapshot() - snapshot the trace buffer\n\n"
5805 #ifdef CONFIG_SYNTH_EVENTS
5806 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5807 "\t Write into this file to define/undefine new synthetic events.\n"
5808 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5814 tracing_readme_read(struct file *filp, char __user *ubuf,
5815 size_t cnt, loff_t *ppos)
5817 return simple_read_from_buffer(ubuf, cnt, ppos,
5818 readme_msg, strlen(readme_msg));
5821 static const struct file_operations tracing_readme_fops = {
5822 .open = tracing_open_generic,
5823 .read = tracing_readme_read,
5824 .llseek = generic_file_llseek,
5827 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5831 return trace_find_tgid_ptr(pid);
5834 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5838 return trace_find_tgid_ptr(pid);
5841 static void saved_tgids_stop(struct seq_file *m, void *v)
5845 static int saved_tgids_show(struct seq_file *m, void *v)
5847 int *entry = (int *)v;
5848 int pid = entry - tgid_map;
5854 seq_printf(m, "%d %d\n", pid, tgid);
5858 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5859 .start = saved_tgids_start,
5860 .stop = saved_tgids_stop,
5861 .next = saved_tgids_next,
5862 .show = saved_tgids_show,
5865 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5869 ret = tracing_check_open_get_tr(NULL);
5873 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5877 static const struct file_operations tracing_saved_tgids_fops = {
5878 .open = tracing_saved_tgids_open,
5880 .llseek = seq_lseek,
5881 .release = seq_release,
5884 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5886 unsigned int *ptr = v;
5888 if (*pos || m->count)
5893 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5895 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5904 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5910 arch_spin_lock(&trace_cmdline_lock);
5912 v = &savedcmd->map_cmdline_to_pid[0];
5914 v = saved_cmdlines_next(m, v, &l);
5922 static void saved_cmdlines_stop(struct seq_file *m, void *v)
5924 arch_spin_unlock(&trace_cmdline_lock);
5928 static int saved_cmdlines_show(struct seq_file *m, void *v)
5930 char buf[TASK_COMM_LEN];
5931 unsigned int *pid = v;
5933 __trace_find_cmdline(*pid, buf);
5934 seq_printf(m, "%d %s\n", *pid, buf);
5938 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5939 .start = saved_cmdlines_start,
5940 .next = saved_cmdlines_next,
5941 .stop = saved_cmdlines_stop,
5942 .show = saved_cmdlines_show,
5945 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5949 ret = tracing_check_open_get_tr(NULL);
5953 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5956 static const struct file_operations tracing_saved_cmdlines_fops = {
5957 .open = tracing_saved_cmdlines_open,
5959 .llseek = seq_lseek,
5960 .release = seq_release,
5964 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5965 size_t cnt, loff_t *ppos)
5971 arch_spin_lock(&trace_cmdline_lock);
5972 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5973 arch_spin_unlock(&trace_cmdline_lock);
5976 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5979 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5981 kfree(s->saved_cmdlines);
5982 kfree(s->map_cmdline_to_pid);
5986 static int tracing_resize_saved_cmdlines(unsigned int val)
5988 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5990 s = kmalloc(sizeof(*s), GFP_KERNEL);
5994 if (allocate_cmdlines_buffer(val, s) < 0) {
6000 arch_spin_lock(&trace_cmdline_lock);
6001 savedcmd_temp = savedcmd;
6003 arch_spin_unlock(&trace_cmdline_lock);
6005 free_saved_cmdlines_buffer(savedcmd_temp);
6011 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
6012 size_t cnt, loff_t *ppos)
6017 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6021 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
6022 if (!val || val > PID_MAX_DEFAULT)
6025 ret = tracing_resize_saved_cmdlines((unsigned int)val);
6034 static const struct file_operations tracing_saved_cmdlines_size_fops = {
6035 .open = tracing_open_generic,
6036 .read = tracing_saved_cmdlines_size_read,
6037 .write = tracing_saved_cmdlines_size_write,
6040 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
6041 static union trace_eval_map_item *
6042 update_eval_map(union trace_eval_map_item *ptr)
6044 if (!ptr->map.eval_string) {
6045 if (ptr->tail.next) {
6046 ptr = ptr->tail.next;
6047 /* Set ptr to the next real item (skip head) */
6055 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6057 union trace_eval_map_item *ptr = v;
6060 * Paranoid! If ptr points to end, we don't want to increment past it.
6061 * This really should never happen.
6064 ptr = update_eval_map(ptr);
6065 if (WARN_ON_ONCE(!ptr))
6069 ptr = update_eval_map(ptr);
6074 static void *eval_map_start(struct seq_file *m, loff_t *pos)
6076 union trace_eval_map_item *v;
6079 mutex_lock(&trace_eval_mutex);
6081 v = trace_eval_maps;
6085 while (v && l < *pos) {
6086 v = eval_map_next(m, v, &l);
6092 static void eval_map_stop(struct seq_file *m, void *v)
6094 mutex_unlock(&trace_eval_mutex);
6097 static int eval_map_show(struct seq_file *m, void *v)
6099 union trace_eval_map_item *ptr = v;
6101 seq_printf(m, "%s %ld (%s)\n",
6102 ptr->map.eval_string, ptr->map.eval_value,
6108 static const struct seq_operations tracing_eval_map_seq_ops = {
6109 .start = eval_map_start,
6110 .next = eval_map_next,
6111 .stop = eval_map_stop,
6112 .show = eval_map_show,
6115 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6119 ret = tracing_check_open_get_tr(NULL);
6123 return seq_open(filp, &tracing_eval_map_seq_ops);
6126 static const struct file_operations tracing_eval_map_fops = {
6127 .open = tracing_eval_map_open,
6129 .llseek = seq_lseek,
6130 .release = seq_release,
6133 static inline union trace_eval_map_item *
6134 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6136 /* Return tail of array given the head */
6137 return ptr + ptr->head.length + 1;
6141 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6144 struct trace_eval_map **stop;
6145 struct trace_eval_map **map;
6146 union trace_eval_map_item *map_array;
6147 union trace_eval_map_item *ptr;
6152 * The trace_eval_maps contains the map plus a head and tail item,
6153 * where the head holds the module and length of array, and the
6154 * tail holds a pointer to the next list.
6156 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6158 pr_warn("Unable to allocate trace eval mapping\n");
6162 mutex_lock(&trace_eval_mutex);
6164 if (!trace_eval_maps)
6165 trace_eval_maps = map_array;
6167 ptr = trace_eval_maps;
6169 ptr = trace_eval_jmp_to_tail(ptr);
6170 if (!ptr->tail.next)
6172 ptr = ptr->tail.next;
6175 ptr->tail.next = map_array;
6177 map_array->head.mod = mod;
6178 map_array->head.length = len;
6181 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6182 map_array->map = **map;
6185 memset(map_array, 0, sizeof(*map_array));
6187 mutex_unlock(&trace_eval_mutex);
6190 static void trace_create_eval_file(struct dentry *d_tracer)
6192 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6193 NULL, &tracing_eval_map_fops);
6196 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
6197 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6198 static inline void trace_insert_eval_map_file(struct module *mod,
6199 struct trace_eval_map **start, int len) { }
6200 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6202 static void trace_insert_eval_map(struct module *mod,
6203 struct trace_eval_map **start, int len)
6205 struct trace_eval_map **map;
6212 trace_event_eval_update(map, len);
6214 trace_insert_eval_map_file(mod, start, len);
6218 tracing_set_trace_read(struct file *filp, char __user *ubuf,
6219 size_t cnt, loff_t *ppos)
6221 struct trace_array *tr = filp->private_data;
6222 char buf[MAX_TRACER_SIZE+2];
6225 mutex_lock(&trace_types_lock);
6226 r = sprintf(buf, "%s\n", tr->current_trace->name);
6227 mutex_unlock(&trace_types_lock);
6229 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6232 int tracer_init(struct tracer *t, struct trace_array *tr)
6234 tracing_reset_online_cpus(&tr->array_buffer);
6238 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6242 for_each_tracing_cpu(cpu)
6243 per_cpu_ptr(buf->data, cpu)->entries = val;
6246 #ifdef CONFIG_TRACER_MAX_TRACE
6247 /* resize @tr's buffer to the size of @size_tr's entries */
6248 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6249 struct array_buffer *size_buf, int cpu_id)
6253 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6254 for_each_tracing_cpu(cpu) {
6255 ret = ring_buffer_resize(trace_buf->buffer,
6256 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6259 per_cpu_ptr(trace_buf->data, cpu)->entries =
6260 per_cpu_ptr(size_buf->data, cpu)->entries;
6263 ret = ring_buffer_resize(trace_buf->buffer,
6264 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6266 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6267 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6272 #endif /* CONFIG_TRACER_MAX_TRACE */
6274 static int __tracing_resize_ring_buffer(struct trace_array *tr,
6275 unsigned long size, int cpu)
6280 * If kernel or user changes the size of the ring buffer
6281 * we use the size that was given, and we can forget about
6282 * expanding it later.
6284 ring_buffer_expanded = true;
6286 /* May be called before buffers are initialized */
6287 if (!tr->array_buffer.buffer)
6290 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6294 #ifdef CONFIG_TRACER_MAX_TRACE
6295 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
6296 !tr->current_trace->use_max_tr)
6299 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6301 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6302 &tr->array_buffer, cpu);
6305 * AARGH! We are left with different
6306 * size max buffer!!!!
6307 * The max buffer is our "snapshot" buffer.
6308 * When a tracer needs a snapshot (one of the
6309 * latency tracers), it swaps the max buffer
6310 * with the saved snap shot. We succeeded to
6311 * update the size of the main buffer, but failed to
6312 * update the size of the max buffer. But when we tried
6313 * to reset the main buffer to the original size, we
6314 * failed there too. This is very unlikely to
6315 * happen, but if it does, warn and kill all
6319 tracing_disabled = 1;
6324 if (cpu == RING_BUFFER_ALL_CPUS)
6325 set_buffer_entries(&tr->max_buffer, size);
6327 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
6330 #endif /* CONFIG_TRACER_MAX_TRACE */
6332 if (cpu == RING_BUFFER_ALL_CPUS)
6333 set_buffer_entries(&tr->array_buffer, size);
6335 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
6340 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6341 unsigned long size, int cpu_id)
6345 mutex_lock(&trace_types_lock);
6347 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6348 /* make sure, this cpu is enabled in the mask */
6349 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6355 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6360 mutex_unlock(&trace_types_lock);
6367 * tracing_update_buffers - used by tracing facility to expand ring buffers
6369 * To save on memory when the tracing is never used on a system with it
6370 * configured in. The ring buffers are set to a minimum size. But once
6371 * a user starts to use the tracing facility, then they need to grow
6372 * to their default size.
6374 * This function is to be called when a tracer is about to be used.
6376 int tracing_update_buffers(void)
6380 mutex_lock(&trace_types_lock);
6381 if (!ring_buffer_expanded)
6382 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
6383 RING_BUFFER_ALL_CPUS);
6384 mutex_unlock(&trace_types_lock);
6389 struct trace_option_dentry;
6392 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6395 * Used to clear out the tracer before deletion of an instance.
6396 * Must have trace_types_lock held.
6398 static void tracing_set_nop(struct trace_array *tr)
6400 if (tr->current_trace == &nop_trace)
6403 tr->current_trace->enabled--;
6405 if (tr->current_trace->reset)
6406 tr->current_trace->reset(tr);
6408 tr->current_trace = &nop_trace;
6411 static bool tracer_options_updated;
6413 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6415 /* Only enable if the directory has been created already. */
6419 /* Only create trace option files after update_tracer_options finish */
6420 if (!tracer_options_updated)
6423 create_trace_option_files(tr, t);
6426 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6429 #ifdef CONFIG_TRACER_MAX_TRACE
6434 mutex_lock(&trace_types_lock);
6436 if (!ring_buffer_expanded) {
6437 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6438 RING_BUFFER_ALL_CPUS);
6444 for (t = trace_types; t; t = t->next) {
6445 if (strcmp(t->name, buf) == 0)
6452 if (t == tr->current_trace)
6455 #ifdef CONFIG_TRACER_SNAPSHOT
6456 if (t->use_max_tr) {
6457 local_irq_disable();
6458 arch_spin_lock(&tr->max_lock);
6459 if (tr->cond_snapshot)
6461 arch_spin_unlock(&tr->max_lock);
6467 /* Some tracers won't work on kernel command line */
6468 if (system_state < SYSTEM_RUNNING && t->noboot) {
6469 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6474 /* Some tracers are only allowed for the top level buffer */
6475 if (!trace_ok_for_array(t, tr)) {
6480 /* If trace pipe files are being read, we can't change the tracer */
6481 if (tr->trace_ref) {
6486 trace_branch_disable();
6488 tr->current_trace->enabled--;
6490 if (tr->current_trace->reset)
6491 tr->current_trace->reset(tr);
6493 #ifdef CONFIG_TRACER_MAX_TRACE
6494 had_max_tr = tr->current_trace->use_max_tr;
6496 /* Current trace needs to be nop_trace before synchronize_rcu */
6497 tr->current_trace = &nop_trace;
6499 if (had_max_tr && !t->use_max_tr) {
6501 * We need to make sure that the update_max_tr sees that
6502 * current_trace changed to nop_trace to keep it from
6503 * swapping the buffers after we resize it.
6504 * The update_max_tr is called from interrupts disabled
6505 * so a synchronized_sched() is sufficient.
6511 if (t->use_max_tr && !tr->allocated_snapshot) {
6512 ret = tracing_alloc_snapshot_instance(tr);
6517 tr->current_trace = &nop_trace;
6521 ret = tracer_init(t, tr);
6526 tr->current_trace = t;
6527 tr->current_trace->enabled++;
6528 trace_branch_enable(tr);
6530 mutex_unlock(&trace_types_lock);
6536 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6537 size_t cnt, loff_t *ppos)
6539 struct trace_array *tr = filp->private_data;
6540 char buf[MAX_TRACER_SIZE+1];
6547 if (cnt > MAX_TRACER_SIZE)
6548 cnt = MAX_TRACER_SIZE;
6550 if (copy_from_user(buf, ubuf, cnt))
6557 err = tracing_set_tracer(tr, name);
6567 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6568 size_t cnt, loff_t *ppos)
6573 r = snprintf(buf, sizeof(buf), "%ld\n",
6574 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6575 if (r > sizeof(buf))
6577 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6581 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6582 size_t cnt, loff_t *ppos)
6587 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6597 tracing_thresh_read(struct file *filp, char __user *ubuf,
6598 size_t cnt, loff_t *ppos)
6600 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6604 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6605 size_t cnt, loff_t *ppos)
6607 struct trace_array *tr = filp->private_data;
6610 mutex_lock(&trace_types_lock);
6611 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6615 if (tr->current_trace->update_thresh) {
6616 ret = tr->current_trace->update_thresh(tr);
6623 mutex_unlock(&trace_types_lock);
6628 #ifdef CONFIG_TRACER_MAX_TRACE
6631 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6632 size_t cnt, loff_t *ppos)
6634 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6638 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6639 size_t cnt, loff_t *ppos)
6641 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6646 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6648 struct trace_array *tr = inode->i_private;
6649 struct trace_iterator *iter;
6652 ret = tracing_check_open_get_tr(tr);
6656 mutex_lock(&trace_types_lock);
6658 /* create a buffer to store the information to pass to userspace */
6659 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6662 __trace_array_put(tr);
6666 trace_seq_init(&iter->seq);
6667 iter->trace = tr->current_trace;
6669 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6674 /* trace pipe does not show start of buffer */
6675 cpumask_setall(iter->started);
6677 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6678 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6680 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6681 if (trace_clocks[tr->clock_id].in_ns)
6682 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6685 iter->array_buffer = &tr->array_buffer;
6686 iter->cpu_file = tracing_get_cpu(inode);
6687 mutex_init(&iter->mutex);
6688 filp->private_data = iter;
6690 if (iter->trace->pipe_open)
6691 iter->trace->pipe_open(iter);
6693 nonseekable_open(inode, filp);
6697 mutex_unlock(&trace_types_lock);
6702 __trace_array_put(tr);
6703 mutex_unlock(&trace_types_lock);
6707 static int tracing_release_pipe(struct inode *inode, struct file *file)
6709 struct trace_iterator *iter = file->private_data;
6710 struct trace_array *tr = inode->i_private;
6712 mutex_lock(&trace_types_lock);
6716 if (iter->trace->pipe_close)
6717 iter->trace->pipe_close(iter);
6719 mutex_unlock(&trace_types_lock);
6721 free_cpumask_var(iter->started);
6723 mutex_destroy(&iter->mutex);
6726 trace_array_put(tr);
6732 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6734 struct trace_array *tr = iter->tr;
6736 /* Iterators are static, they should be filled or empty */
6737 if (trace_buffer_iter(iter, iter->cpu_file))
6738 return EPOLLIN | EPOLLRDNORM;
6740 if (tr->trace_flags & TRACE_ITER_BLOCK)
6742 * Always select as readable when in blocking mode
6744 return EPOLLIN | EPOLLRDNORM;
6746 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6747 filp, poll_table, iter->tr->buffer_percent);
6751 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6753 struct trace_iterator *iter = filp->private_data;
6755 return trace_poll(iter, filp, poll_table);
6758 /* Must be called with iter->mutex held. */
6759 static int tracing_wait_pipe(struct file *filp)
6761 struct trace_iterator *iter = filp->private_data;
6764 while (trace_empty(iter)) {
6766 if ((filp->f_flags & O_NONBLOCK)) {
6771 * We block until we read something and tracing is disabled.
6772 * We still block if tracing is disabled, but we have never
6773 * read anything. This allows a user to cat this file, and
6774 * then enable tracing. But after we have read something,
6775 * we give an EOF when tracing is again disabled.
6777 * iter->pos will be 0 if we haven't read anything.
6779 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6782 mutex_unlock(&iter->mutex);
6784 ret = wait_on_pipe(iter, 0);
6786 mutex_lock(&iter->mutex);
6799 tracing_read_pipe(struct file *filp, char __user *ubuf,
6800 size_t cnt, loff_t *ppos)
6802 struct trace_iterator *iter = filp->private_data;
6806 * Avoid more than one consumer on a single file descriptor
6807 * This is just a matter of traces coherency, the ring buffer itself
6810 mutex_lock(&iter->mutex);
6812 /* return any leftover data */
6813 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6817 trace_seq_init(&iter->seq);
6819 if (iter->trace->read) {
6820 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6826 sret = tracing_wait_pipe(filp);
6830 /* stop when tracing is finished */
6831 if (trace_empty(iter)) {
6836 if (cnt >= PAGE_SIZE)
6837 cnt = PAGE_SIZE - 1;
6839 /* reset all but tr, trace, and overruns */
6840 trace_iterator_reset(iter);
6841 cpumask_clear(iter->started);
6842 trace_seq_init(&iter->seq);
6844 trace_event_read_lock();
6845 trace_access_lock(iter->cpu_file);
6846 while (trace_find_next_entry_inc(iter) != NULL) {
6847 enum print_line_t ret;
6848 int save_len = iter->seq.seq.len;
6850 ret = print_trace_line(iter);
6851 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6853 * If one print_trace_line() fills entire trace_seq in one shot,
6854 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6855 * In this case, we need to consume it, otherwise, loop will peek
6856 * this event next time, resulting in an infinite loop.
6858 if (save_len == 0) {
6860 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6861 trace_consume(iter);
6865 /* In other cases, don't print partial lines */
6866 iter->seq.seq.len = save_len;
6869 if (ret != TRACE_TYPE_NO_CONSUME)
6870 trace_consume(iter);
6872 if (trace_seq_used(&iter->seq) >= cnt)
6876 * Setting the full flag means we reached the trace_seq buffer
6877 * size and we should leave by partial output condition above.
6878 * One of the trace_seq_* functions is not used properly.
6880 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6883 trace_access_unlock(iter->cpu_file);
6884 trace_event_read_unlock();
6886 /* Now copy what we have to the user */
6887 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6888 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6889 trace_seq_init(&iter->seq);
6892 * If there was nothing to send to user, in spite of consuming trace
6893 * entries, go back to wait for more entries.
6899 mutex_unlock(&iter->mutex);
6904 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6907 __free_page(spd->pages[idx]);
6911 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6917 /* Seq buffer is page-sized, exactly what we need. */
6919 save_len = iter->seq.seq.len;
6920 ret = print_trace_line(iter);
6922 if (trace_seq_has_overflowed(&iter->seq)) {
6923 iter->seq.seq.len = save_len;
6928 * This should not be hit, because it should only
6929 * be set if the iter->seq overflowed. But check it
6930 * anyway to be safe.
6932 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6933 iter->seq.seq.len = save_len;
6937 count = trace_seq_used(&iter->seq) - save_len;
6940 iter->seq.seq.len = save_len;
6944 if (ret != TRACE_TYPE_NO_CONSUME)
6945 trace_consume(iter);
6947 if (!trace_find_next_entry_inc(iter)) {
6957 static ssize_t tracing_splice_read_pipe(struct file *filp,
6959 struct pipe_inode_info *pipe,
6963 struct page *pages_def[PIPE_DEF_BUFFERS];
6964 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6965 struct trace_iterator *iter = filp->private_data;
6966 struct splice_pipe_desc spd = {
6968 .partial = partial_def,
6969 .nr_pages = 0, /* This gets updated below. */
6970 .nr_pages_max = PIPE_DEF_BUFFERS,
6971 .ops = &default_pipe_buf_ops,
6972 .spd_release = tracing_spd_release_pipe,
6978 if (splice_grow_spd(pipe, &spd))
6981 mutex_lock(&iter->mutex);
6983 if (iter->trace->splice_read) {
6984 ret = iter->trace->splice_read(iter, filp,
6985 ppos, pipe, len, flags);
6990 ret = tracing_wait_pipe(filp);
6994 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6999 trace_event_read_lock();
7000 trace_access_lock(iter->cpu_file);
7002 /* Fill as many pages as possible. */
7003 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
7004 spd.pages[i] = alloc_page(GFP_KERNEL);
7008 rem = tracing_fill_pipe_page(rem, iter);
7010 /* Copy the data into the page, so we can start over. */
7011 ret = trace_seq_to_buffer(&iter->seq,
7012 page_address(spd.pages[i]),
7013 trace_seq_used(&iter->seq));
7015 __free_page(spd.pages[i]);
7018 spd.partial[i].offset = 0;
7019 spd.partial[i].len = trace_seq_used(&iter->seq);
7021 trace_seq_init(&iter->seq);
7024 trace_access_unlock(iter->cpu_file);
7025 trace_event_read_unlock();
7026 mutex_unlock(&iter->mutex);
7031 ret = splice_to_pipe(pipe, &spd);
7035 splice_shrink_spd(&spd);
7039 mutex_unlock(&iter->mutex);
7044 tracing_entries_read(struct file *filp, char __user *ubuf,
7045 size_t cnt, loff_t *ppos)
7047 struct inode *inode = file_inode(filp);
7048 struct trace_array *tr = inode->i_private;
7049 int cpu = tracing_get_cpu(inode);
7054 mutex_lock(&trace_types_lock);
7056 if (cpu == RING_BUFFER_ALL_CPUS) {
7057 int cpu, buf_size_same;
7062 /* check if all cpu sizes are same */
7063 for_each_tracing_cpu(cpu) {
7064 /* fill in the size from first enabled cpu */
7066 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7067 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7073 if (buf_size_same) {
7074 if (!ring_buffer_expanded)
7075 r = sprintf(buf, "%lu (expanded: %lu)\n",
7077 trace_buf_size >> 10);
7079 r = sprintf(buf, "%lu\n", size >> 10);
7081 r = sprintf(buf, "X\n");
7083 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7085 mutex_unlock(&trace_types_lock);
7087 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7092 tracing_entries_write(struct file *filp, const char __user *ubuf,
7093 size_t cnt, loff_t *ppos)
7095 struct inode *inode = file_inode(filp);
7096 struct trace_array *tr = inode->i_private;
7100 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7104 /* must have at least 1 entry */
7108 /* value is in KB */
7110 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7120 tracing_total_entries_read(struct file *filp, char __user *ubuf,
7121 size_t cnt, loff_t *ppos)
7123 struct trace_array *tr = filp->private_data;
7126 unsigned long size = 0, expanded_size = 0;
7128 mutex_lock(&trace_types_lock);
7129 for_each_tracing_cpu(cpu) {
7130 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7131 if (!ring_buffer_expanded)
7132 expanded_size += trace_buf_size >> 10;
7134 if (ring_buffer_expanded)
7135 r = sprintf(buf, "%lu\n", size);
7137 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7138 mutex_unlock(&trace_types_lock);
7140 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7144 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7145 size_t cnt, loff_t *ppos)
7148 * There is no need to read what the user has written, this function
7149 * is just to make sure that there is no error when "echo" is used
7158 tracing_free_buffer_release(struct inode *inode, struct file *filp)
7160 struct trace_array *tr = inode->i_private;
7162 /* disable tracing ? */
7163 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7164 tracer_tracing_off(tr);
7165 /* resize the ring buffer to 0 */
7166 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7168 trace_array_put(tr);
7174 tracing_mark_write(struct file *filp, const char __user *ubuf,
7175 size_t cnt, loff_t *fpos)
7177 struct trace_array *tr = filp->private_data;
7178 struct ring_buffer_event *event;
7179 enum event_trigger_type tt = ETT_NONE;
7180 struct trace_buffer *buffer;
7181 struct print_entry *entry;
7186 /* Used in tracing_mark_raw_write() as well */
7187 #define FAULTED_STR "<faulted>"
7188 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7190 if (tracing_disabled)
7193 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7196 if (cnt > TRACE_BUF_SIZE)
7197 cnt = TRACE_BUF_SIZE;
7199 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7201 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
7203 /* If less than "<faulted>", then make sure we can still add that */
7204 if (cnt < FAULTED_SIZE)
7205 size += FAULTED_SIZE - cnt;
7207 buffer = tr->array_buffer.buffer;
7208 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7210 if (unlikely(!event))
7211 /* Ring buffer disabled, return as if not open for write */
7214 entry = ring_buffer_event_data(event);
7215 entry->ip = _THIS_IP_;
7217 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7219 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7225 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7226 /* do not add \n before testing triggers, but add \0 */
7227 entry->buf[cnt] = '\0';
7228 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7231 if (entry->buf[cnt - 1] != '\n') {
7232 entry->buf[cnt] = '\n';
7233 entry->buf[cnt + 1] = '\0';
7235 entry->buf[cnt] = '\0';
7237 if (static_branch_unlikely(&trace_marker_exports_enabled))
7238 ftrace_exports(event, TRACE_EXPORT_MARKER);
7239 __buffer_unlock_commit(buffer, event);
7242 event_triggers_post_call(tr->trace_marker_file, tt);
7247 /* Limit it for now to 3K (including tag) */
7248 #define RAW_DATA_MAX_SIZE (1024*3)
7251 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7252 size_t cnt, loff_t *fpos)
7254 struct trace_array *tr = filp->private_data;
7255 struct ring_buffer_event *event;
7256 struct trace_buffer *buffer;
7257 struct raw_data_entry *entry;
7262 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7264 if (tracing_disabled)
7267 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7270 /* The marker must at least have a tag id */
7271 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
7274 if (cnt > TRACE_BUF_SIZE)
7275 cnt = TRACE_BUF_SIZE;
7277 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7279 size = sizeof(*entry) + cnt;
7280 if (cnt < FAULT_SIZE_ID)
7281 size += FAULT_SIZE_ID - cnt;
7283 buffer = tr->array_buffer.buffer;
7284 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7287 /* Ring buffer disabled, return as if not open for write */
7290 entry = ring_buffer_event_data(event);
7292 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7295 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7300 __buffer_unlock_commit(buffer, event);
7305 static int tracing_clock_show(struct seq_file *m, void *v)
7307 struct trace_array *tr = m->private;
7310 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7312 "%s%s%s%s", i ? " " : "",
7313 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7314 i == tr->clock_id ? "]" : "");
7320 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7324 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7325 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7328 if (i == ARRAY_SIZE(trace_clocks))
7331 mutex_lock(&trace_types_lock);
7335 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7338 * New clock may not be consistent with the previous clock.
7339 * Reset the buffer so that it doesn't have incomparable timestamps.
7341 tracing_reset_online_cpus(&tr->array_buffer);
7343 #ifdef CONFIG_TRACER_MAX_TRACE
7344 if (tr->max_buffer.buffer)
7345 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7346 tracing_reset_online_cpus(&tr->max_buffer);
7349 mutex_unlock(&trace_types_lock);
7354 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7355 size_t cnt, loff_t *fpos)
7357 struct seq_file *m = filp->private_data;
7358 struct trace_array *tr = m->private;
7360 const char *clockstr;
7363 if (cnt >= sizeof(buf))
7366 if (copy_from_user(buf, ubuf, cnt))
7371 clockstr = strstrip(buf);
7373 ret = tracing_set_clock(tr, clockstr);
7382 static int tracing_clock_open(struct inode *inode, struct file *file)
7384 struct trace_array *tr = inode->i_private;
7387 ret = tracing_check_open_get_tr(tr);
7391 ret = single_open(file, tracing_clock_show, inode->i_private);
7393 trace_array_put(tr);
7398 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7400 struct trace_array *tr = m->private;
7402 mutex_lock(&trace_types_lock);
7404 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7405 seq_puts(m, "delta [absolute]\n");
7407 seq_puts(m, "[delta] absolute\n");
7409 mutex_unlock(&trace_types_lock);
7414 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7416 struct trace_array *tr = inode->i_private;
7419 ret = tracing_check_open_get_tr(tr);
7423 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7425 trace_array_put(tr);
7430 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7432 if (rbe == this_cpu_read(trace_buffered_event))
7433 return ring_buffer_time_stamp(buffer);
7435 return ring_buffer_event_time_stamp(buffer, rbe);
7439 * Set or disable using the per CPU trace_buffer_event when possible.
7441 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7445 mutex_lock(&trace_types_lock);
7447 if (set && tr->no_filter_buffering_ref++)
7451 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7456 --tr->no_filter_buffering_ref;
7459 mutex_unlock(&trace_types_lock);
7464 struct ftrace_buffer_info {
7465 struct trace_iterator iter;
7467 unsigned int spare_cpu;
7471 #ifdef CONFIG_TRACER_SNAPSHOT
7472 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7474 struct trace_array *tr = inode->i_private;
7475 struct trace_iterator *iter;
7479 ret = tracing_check_open_get_tr(tr);
7483 if (file->f_mode & FMODE_READ) {
7484 iter = __tracing_open(inode, file, true);
7486 ret = PTR_ERR(iter);
7488 /* Writes still need the seq_file to hold the private data */
7490 m = kzalloc(sizeof(*m), GFP_KERNEL);
7493 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7501 iter->array_buffer = &tr->max_buffer;
7502 iter->cpu_file = tracing_get_cpu(inode);
7504 file->private_data = m;
7508 trace_array_put(tr);
7514 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7517 struct seq_file *m = filp->private_data;
7518 struct trace_iterator *iter = m->private;
7519 struct trace_array *tr = iter->tr;
7523 ret = tracing_update_buffers();
7527 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7531 mutex_lock(&trace_types_lock);
7533 if (tr->current_trace->use_max_tr) {
7538 local_irq_disable();
7539 arch_spin_lock(&tr->max_lock);
7540 if (tr->cond_snapshot)
7542 arch_spin_unlock(&tr->max_lock);
7549 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7553 if (tr->allocated_snapshot)
7557 /* Only allow per-cpu swap if the ring buffer supports it */
7558 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7559 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7564 if (tr->allocated_snapshot)
7565 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7566 &tr->array_buffer, iter->cpu_file);
7568 ret = tracing_alloc_snapshot_instance(tr);
7571 local_irq_disable();
7572 /* Now, we're going to swap */
7573 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7574 update_max_tr(tr, current, smp_processor_id(), NULL);
7576 update_max_tr_single(tr, current, iter->cpu_file);
7580 if (tr->allocated_snapshot) {
7581 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7582 tracing_reset_online_cpus(&tr->max_buffer);
7584 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7594 mutex_unlock(&trace_types_lock);
7598 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7600 struct seq_file *m = file->private_data;
7603 ret = tracing_release(inode, file);
7605 if (file->f_mode & FMODE_READ)
7608 /* If write only, the seq_file is just a stub */
7616 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7617 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7618 size_t count, loff_t *ppos);
7619 static int tracing_buffers_release(struct inode *inode, struct file *file);
7620 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7621 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7623 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7625 struct ftrace_buffer_info *info;
7628 /* The following checks for tracefs lockdown */
7629 ret = tracing_buffers_open(inode, filp);
7633 info = filp->private_data;
7635 if (info->iter.trace->use_max_tr) {
7636 tracing_buffers_release(inode, filp);
7640 info->iter.snapshot = true;
7641 info->iter.array_buffer = &info->iter.tr->max_buffer;
7646 #endif /* CONFIG_TRACER_SNAPSHOT */
7649 static const struct file_operations tracing_thresh_fops = {
7650 .open = tracing_open_generic,
7651 .read = tracing_thresh_read,
7652 .write = tracing_thresh_write,
7653 .llseek = generic_file_llseek,
7656 #ifdef CONFIG_TRACER_MAX_TRACE
7657 static const struct file_operations tracing_max_lat_fops = {
7658 .open = tracing_open_generic,
7659 .read = tracing_max_lat_read,
7660 .write = tracing_max_lat_write,
7661 .llseek = generic_file_llseek,
7665 static const struct file_operations set_tracer_fops = {
7666 .open = tracing_open_generic,
7667 .read = tracing_set_trace_read,
7668 .write = tracing_set_trace_write,
7669 .llseek = generic_file_llseek,
7672 static const struct file_operations tracing_pipe_fops = {
7673 .open = tracing_open_pipe,
7674 .poll = tracing_poll_pipe,
7675 .read = tracing_read_pipe,
7676 .splice_read = tracing_splice_read_pipe,
7677 .release = tracing_release_pipe,
7678 .llseek = no_llseek,
7681 static const struct file_operations tracing_entries_fops = {
7682 .open = tracing_open_generic_tr,
7683 .read = tracing_entries_read,
7684 .write = tracing_entries_write,
7685 .llseek = generic_file_llseek,
7686 .release = tracing_release_generic_tr,
7689 static const struct file_operations tracing_total_entries_fops = {
7690 .open = tracing_open_generic_tr,
7691 .read = tracing_total_entries_read,
7692 .llseek = generic_file_llseek,
7693 .release = tracing_release_generic_tr,
7696 static const struct file_operations tracing_free_buffer_fops = {
7697 .open = tracing_open_generic_tr,
7698 .write = tracing_free_buffer_write,
7699 .release = tracing_free_buffer_release,
7702 static const struct file_operations tracing_mark_fops = {
7703 .open = tracing_mark_open,
7704 .write = tracing_mark_write,
7705 .release = tracing_release_generic_tr,
7708 static const struct file_operations tracing_mark_raw_fops = {
7709 .open = tracing_mark_open,
7710 .write = tracing_mark_raw_write,
7711 .release = tracing_release_generic_tr,
7714 static const struct file_operations trace_clock_fops = {
7715 .open = tracing_clock_open,
7717 .llseek = seq_lseek,
7718 .release = tracing_single_release_tr,
7719 .write = tracing_clock_write,
7722 static const struct file_operations trace_time_stamp_mode_fops = {
7723 .open = tracing_time_stamp_mode_open,
7725 .llseek = seq_lseek,
7726 .release = tracing_single_release_tr,
7729 #ifdef CONFIG_TRACER_SNAPSHOT
7730 static const struct file_operations snapshot_fops = {
7731 .open = tracing_snapshot_open,
7733 .write = tracing_snapshot_write,
7734 .llseek = tracing_lseek,
7735 .release = tracing_snapshot_release,
7738 static const struct file_operations snapshot_raw_fops = {
7739 .open = snapshot_raw_open,
7740 .read = tracing_buffers_read,
7741 .release = tracing_buffers_release,
7742 .splice_read = tracing_buffers_splice_read,
7743 .llseek = no_llseek,
7746 #endif /* CONFIG_TRACER_SNAPSHOT */
7749 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7750 * @filp: The active open file structure
7751 * @ubuf: The userspace provided buffer to read value into
7752 * @cnt: The maximum number of bytes to read
7753 * @ppos: The current "file" position
7755 * This function implements the write interface for a struct trace_min_max_param.
7756 * The filp->private_data must point to a trace_min_max_param structure that
7757 * defines where to write the value, the min and the max acceptable values,
7758 * and a lock to protect the write.
7761 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7763 struct trace_min_max_param *param = filp->private_data;
7770 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7775 mutex_lock(param->lock);
7777 if (param->min && val < *param->min)
7780 if (param->max && val > *param->max)
7787 mutex_unlock(param->lock);
7796 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7797 * @filp: The active open file structure
7798 * @ubuf: The userspace provided buffer to read value into
7799 * @cnt: The maximum number of bytes to read
7800 * @ppos: The current "file" position
7802 * This function implements the read interface for a struct trace_min_max_param.
7803 * The filp->private_data must point to a trace_min_max_param struct with valid
7807 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7809 struct trace_min_max_param *param = filp->private_data;
7810 char buf[U64_STR_SIZE];
7819 if (cnt > sizeof(buf))
7822 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7824 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7827 const struct file_operations trace_min_max_fops = {
7828 .open = tracing_open_generic,
7829 .read = trace_min_max_read,
7830 .write = trace_min_max_write,
7833 #define TRACING_LOG_ERRS_MAX 8
7834 #define TRACING_LOG_LOC_MAX 128
7836 #define CMD_PREFIX " Command: "
7839 const char **errs; /* ptr to loc-specific array of err strings */
7840 u8 type; /* index into errs -> specific err string */
7841 u16 pos; /* caret position */
7845 struct tracing_log_err {
7846 struct list_head list;
7847 struct err_info info;
7848 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7849 char *cmd; /* what caused err */
7852 static DEFINE_MUTEX(tracing_err_log_lock);
7854 static struct tracing_log_err *alloc_tracing_log_err(int len)
7856 struct tracing_log_err *err;
7858 err = kzalloc(sizeof(*err), GFP_KERNEL);
7860 return ERR_PTR(-ENOMEM);
7862 err->cmd = kzalloc(len, GFP_KERNEL);
7865 return ERR_PTR(-ENOMEM);
7871 static void free_tracing_log_err(struct tracing_log_err *err)
7877 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7880 struct tracing_log_err *err;
7883 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7884 err = alloc_tracing_log_err(len);
7885 if (PTR_ERR(err) != -ENOMEM)
7886 tr->n_err_log_entries++;
7890 cmd = kzalloc(len, GFP_KERNEL);
7892 return ERR_PTR(-ENOMEM);
7893 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7896 list_del(&err->list);
7902 * err_pos - find the position of a string within a command for error careting
7903 * @cmd: The tracing command that caused the error
7904 * @str: The string to position the caret at within @cmd
7906 * Finds the position of the first occurrence of @str within @cmd. The
7907 * return value can be passed to tracing_log_err() for caret placement
7910 * Returns the index within @cmd of the first occurrence of @str or 0
7911 * if @str was not found.
7913 unsigned int err_pos(char *cmd, const char *str)
7917 if (WARN_ON(!strlen(cmd)))
7920 found = strstr(cmd, str);
7928 * tracing_log_err - write an error to the tracing error log
7929 * @tr: The associated trace array for the error (NULL for top level array)
7930 * @loc: A string describing where the error occurred
7931 * @cmd: The tracing command that caused the error
7932 * @errs: The array of loc-specific static error strings
7933 * @type: The index into errs[], which produces the specific static err string
7934 * @pos: The position the caret should be placed in the cmd
7936 * Writes an error into tracing/error_log of the form:
7938 * <loc>: error: <text>
7942 * tracing/error_log is a small log file containing the last
7943 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7944 * unless there has been a tracing error, and the error log can be
7945 * cleared and have its memory freed by writing the empty string in
7946 * truncation mode to it i.e. echo > tracing/error_log.
7948 * NOTE: the @errs array along with the @type param are used to
7949 * produce a static error string - this string is not copied and saved
7950 * when the error is logged - only a pointer to it is saved. See
7951 * existing callers for examples of how static strings are typically
7952 * defined for use with tracing_log_err().
7954 void tracing_log_err(struct trace_array *tr,
7955 const char *loc, const char *cmd,
7956 const char **errs, u8 type, u16 pos)
7958 struct tracing_log_err *err;
7964 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
7966 mutex_lock(&tracing_err_log_lock);
7967 err = get_tracing_log_err(tr, len);
7968 if (PTR_ERR(err) == -ENOMEM) {
7969 mutex_unlock(&tracing_err_log_lock);
7973 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7974 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
7976 err->info.errs = errs;
7977 err->info.type = type;
7978 err->info.pos = pos;
7979 err->info.ts = local_clock();
7981 list_add_tail(&err->list, &tr->err_log);
7982 mutex_unlock(&tracing_err_log_lock);
7985 static void clear_tracing_err_log(struct trace_array *tr)
7987 struct tracing_log_err *err, *next;
7989 mutex_lock(&tracing_err_log_lock);
7990 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7991 list_del(&err->list);
7992 free_tracing_log_err(err);
7995 tr->n_err_log_entries = 0;
7996 mutex_unlock(&tracing_err_log_lock);
7999 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
8001 struct trace_array *tr = m->private;
8003 mutex_lock(&tracing_err_log_lock);
8005 return seq_list_start(&tr->err_log, *pos);
8008 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
8010 struct trace_array *tr = m->private;
8012 return seq_list_next(v, &tr->err_log, pos);
8015 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
8017 mutex_unlock(&tracing_err_log_lock);
8020 static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
8024 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
8026 for (i = 0; i < pos; i++)
8031 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
8033 struct tracing_log_err *err = v;
8036 const char *err_text = err->info.errs[err->info.type];
8037 u64 sec = err->info.ts;
8040 nsec = do_div(sec, NSEC_PER_SEC);
8041 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
8042 err->loc, err_text);
8043 seq_printf(m, "%s", err->cmd);
8044 tracing_err_log_show_pos(m, err->info.pos);
8050 static const struct seq_operations tracing_err_log_seq_ops = {
8051 .start = tracing_err_log_seq_start,
8052 .next = tracing_err_log_seq_next,
8053 .stop = tracing_err_log_seq_stop,
8054 .show = tracing_err_log_seq_show
8057 static int tracing_err_log_open(struct inode *inode, struct file *file)
8059 struct trace_array *tr = inode->i_private;
8062 ret = tracing_check_open_get_tr(tr);
8066 /* If this file was opened for write, then erase contents */
8067 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8068 clear_tracing_err_log(tr);
8070 if (file->f_mode & FMODE_READ) {
8071 ret = seq_open(file, &tracing_err_log_seq_ops);
8073 struct seq_file *m = file->private_data;
8076 trace_array_put(tr);
8082 static ssize_t tracing_err_log_write(struct file *file,
8083 const char __user *buffer,
8084 size_t count, loff_t *ppos)
8089 static int tracing_err_log_release(struct inode *inode, struct file *file)
8091 struct trace_array *tr = inode->i_private;
8093 trace_array_put(tr);
8095 if (file->f_mode & FMODE_READ)
8096 seq_release(inode, file);
8101 static const struct file_operations tracing_err_log_fops = {
8102 .open = tracing_err_log_open,
8103 .write = tracing_err_log_write,
8105 .llseek = seq_lseek,
8106 .release = tracing_err_log_release,
8109 static int tracing_buffers_open(struct inode *inode, struct file *filp)
8111 struct trace_array *tr = inode->i_private;
8112 struct ftrace_buffer_info *info;
8115 ret = tracing_check_open_get_tr(tr);
8119 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8121 trace_array_put(tr);
8125 mutex_lock(&trace_types_lock);
8128 info->iter.cpu_file = tracing_get_cpu(inode);
8129 info->iter.trace = tr->current_trace;
8130 info->iter.array_buffer = &tr->array_buffer;
8132 /* Force reading ring buffer for first read */
8133 info->read = (unsigned int)-1;
8135 filp->private_data = info;
8139 mutex_unlock(&trace_types_lock);
8141 ret = nonseekable_open(inode, filp);
8143 trace_array_put(tr);
8149 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8151 struct ftrace_buffer_info *info = filp->private_data;
8152 struct trace_iterator *iter = &info->iter;
8154 return trace_poll(iter, filp, poll_table);
8158 tracing_buffers_read(struct file *filp, char __user *ubuf,
8159 size_t count, loff_t *ppos)
8161 struct ftrace_buffer_info *info = filp->private_data;
8162 struct trace_iterator *iter = &info->iter;
8169 #ifdef CONFIG_TRACER_MAX_TRACE
8170 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8175 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8177 if (IS_ERR(info->spare)) {
8178 ret = PTR_ERR(info->spare);
8181 info->spare_cpu = iter->cpu_file;
8187 /* Do we have previous read data to read? */
8188 if (info->read < PAGE_SIZE)
8192 trace_access_lock(iter->cpu_file);
8193 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8197 trace_access_unlock(iter->cpu_file);
8200 if (trace_empty(iter)) {
8201 if ((filp->f_flags & O_NONBLOCK))
8204 ret = wait_on_pipe(iter, 0);
8215 size = PAGE_SIZE - info->read;
8219 ret = copy_to_user(ubuf, info->spare + info->read, size);
8231 static int tracing_buffers_release(struct inode *inode, struct file *file)
8233 struct ftrace_buffer_info *info = file->private_data;
8234 struct trace_iterator *iter = &info->iter;
8236 mutex_lock(&trace_types_lock);
8238 iter->tr->trace_ref--;
8240 __trace_array_put(iter->tr);
8243 /* Make sure the waiters see the new wait_index */
8246 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8249 ring_buffer_free_read_page(iter->array_buffer->buffer,
8250 info->spare_cpu, info->spare);
8253 mutex_unlock(&trace_types_lock);
8259 struct trace_buffer *buffer;
8262 refcount_t refcount;
8265 static void buffer_ref_release(struct buffer_ref *ref)
8267 if (!refcount_dec_and_test(&ref->refcount))
8269 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8273 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8274 struct pipe_buffer *buf)
8276 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8278 buffer_ref_release(ref);
8282 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8283 struct pipe_buffer *buf)
8285 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8287 if (refcount_read(&ref->refcount) > INT_MAX/2)
8290 refcount_inc(&ref->refcount);
8294 /* Pipe buffer operations for a buffer. */
8295 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8296 .release = buffer_pipe_buf_release,
8297 .get = buffer_pipe_buf_get,
8301 * Callback from splice_to_pipe(), if we need to release some pages
8302 * at the end of the spd in case we error'ed out in filling the pipe.
8304 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8306 struct buffer_ref *ref =
8307 (struct buffer_ref *)spd->partial[i].private;
8309 buffer_ref_release(ref);
8310 spd->partial[i].private = 0;
8314 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8315 struct pipe_inode_info *pipe, size_t len,
8318 struct ftrace_buffer_info *info = file->private_data;
8319 struct trace_iterator *iter = &info->iter;
8320 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8321 struct page *pages_def[PIPE_DEF_BUFFERS];
8322 struct splice_pipe_desc spd = {
8324 .partial = partial_def,
8325 .nr_pages_max = PIPE_DEF_BUFFERS,
8326 .ops = &buffer_pipe_buf_ops,
8327 .spd_release = buffer_spd_release,
8329 struct buffer_ref *ref;
8333 #ifdef CONFIG_TRACER_MAX_TRACE
8334 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8338 if (*ppos & (PAGE_SIZE - 1))
8341 if (len & (PAGE_SIZE - 1)) {
8342 if (len < PAGE_SIZE)
8347 if (splice_grow_spd(pipe, &spd))
8351 trace_access_lock(iter->cpu_file);
8352 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8354 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
8358 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8364 refcount_set(&ref->refcount, 1);
8365 ref->buffer = iter->array_buffer->buffer;
8366 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8367 if (IS_ERR(ref->page)) {
8368 ret = PTR_ERR(ref->page);
8373 ref->cpu = iter->cpu_file;
8375 r = ring_buffer_read_page(ref->buffer, &ref->page,
8376 len, iter->cpu_file, 1);
8378 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8384 page = virt_to_page(ref->page);
8386 spd.pages[i] = page;
8387 spd.partial[i].len = PAGE_SIZE;
8388 spd.partial[i].offset = 0;
8389 spd.partial[i].private = (unsigned long)ref;
8393 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8396 trace_access_unlock(iter->cpu_file);
8399 /* did we read anything? */
8400 if (!spd.nr_pages) {
8407 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8410 wait_index = READ_ONCE(iter->wait_index);
8412 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
8416 /* No need to wait after waking up when tracing is off */
8417 if (!tracer_tracing_is_on(iter->tr))
8420 /* Make sure we see the new wait_index */
8422 if (wait_index != iter->wait_index)
8428 ret = splice_to_pipe(pipe, &spd);
8430 splice_shrink_spd(&spd);
8435 /* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8436 static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8438 struct ftrace_buffer_info *info = file->private_data;
8439 struct trace_iterator *iter = &info->iter;
8442 return -ENOIOCTLCMD;
8444 mutex_lock(&trace_types_lock);
8447 /* Make sure the waiters see the new wait_index */
8450 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8452 mutex_unlock(&trace_types_lock);
8456 static const struct file_operations tracing_buffers_fops = {
8457 .open = tracing_buffers_open,
8458 .read = tracing_buffers_read,
8459 .poll = tracing_buffers_poll,
8460 .release = tracing_buffers_release,
8461 .splice_read = tracing_buffers_splice_read,
8462 .unlocked_ioctl = tracing_buffers_ioctl,
8463 .llseek = no_llseek,
8467 tracing_stats_read(struct file *filp, char __user *ubuf,
8468 size_t count, loff_t *ppos)
8470 struct inode *inode = file_inode(filp);
8471 struct trace_array *tr = inode->i_private;
8472 struct array_buffer *trace_buf = &tr->array_buffer;
8473 int cpu = tracing_get_cpu(inode);
8474 struct trace_seq *s;
8476 unsigned long long t;
8477 unsigned long usec_rem;
8479 s = kmalloc(sizeof(*s), GFP_KERNEL);
8485 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8486 trace_seq_printf(s, "entries: %ld\n", cnt);
8488 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8489 trace_seq_printf(s, "overrun: %ld\n", cnt);
8491 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8492 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8494 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8495 trace_seq_printf(s, "bytes: %ld\n", cnt);
8497 if (trace_clocks[tr->clock_id].in_ns) {
8498 /* local or global for trace_clock */
8499 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8500 usec_rem = do_div(t, USEC_PER_SEC);
8501 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8504 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8505 usec_rem = do_div(t, USEC_PER_SEC);
8506 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8508 /* counter or tsc mode for trace_clock */
8509 trace_seq_printf(s, "oldest event ts: %llu\n",
8510 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8512 trace_seq_printf(s, "now ts: %llu\n",
8513 ring_buffer_time_stamp(trace_buf->buffer));
8516 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8517 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8519 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8520 trace_seq_printf(s, "read events: %ld\n", cnt);
8522 count = simple_read_from_buffer(ubuf, count, ppos,
8523 s->buffer, trace_seq_used(s));
8530 static const struct file_operations tracing_stats_fops = {
8531 .open = tracing_open_generic_tr,
8532 .read = tracing_stats_read,
8533 .llseek = generic_file_llseek,
8534 .release = tracing_release_generic_tr,
8537 #ifdef CONFIG_DYNAMIC_FTRACE
8540 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8541 size_t cnt, loff_t *ppos)
8547 /* 256 should be plenty to hold the amount needed */
8548 buf = kmalloc(256, GFP_KERNEL);
8552 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8553 ftrace_update_tot_cnt,
8554 ftrace_number_of_pages,
8555 ftrace_number_of_groups);
8557 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8562 static const struct file_operations tracing_dyn_info_fops = {
8563 .open = tracing_open_generic,
8564 .read = tracing_read_dyn_info,
8565 .llseek = generic_file_llseek,
8567 #endif /* CONFIG_DYNAMIC_FTRACE */
8569 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8571 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8572 struct trace_array *tr, struct ftrace_probe_ops *ops,
8575 tracing_snapshot_instance(tr);
8579 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8580 struct trace_array *tr, struct ftrace_probe_ops *ops,
8583 struct ftrace_func_mapper *mapper = data;
8587 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8597 tracing_snapshot_instance(tr);
8601 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8602 struct ftrace_probe_ops *ops, void *data)
8604 struct ftrace_func_mapper *mapper = data;
8607 seq_printf(m, "%ps:", (void *)ip);
8609 seq_puts(m, "snapshot");
8612 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8615 seq_printf(m, ":count=%ld\n", *count);
8617 seq_puts(m, ":unlimited\n");
8623 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8624 unsigned long ip, void *init_data, void **data)
8626 struct ftrace_func_mapper *mapper = *data;
8629 mapper = allocate_ftrace_func_mapper();
8635 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8639 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8640 unsigned long ip, void *data)
8642 struct ftrace_func_mapper *mapper = data;
8647 free_ftrace_func_mapper(mapper, NULL);
8651 ftrace_func_mapper_remove_ip(mapper, ip);
8654 static struct ftrace_probe_ops snapshot_probe_ops = {
8655 .func = ftrace_snapshot,
8656 .print = ftrace_snapshot_print,
8659 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8660 .func = ftrace_count_snapshot,
8661 .print = ftrace_snapshot_print,
8662 .init = ftrace_snapshot_init,
8663 .free = ftrace_snapshot_free,
8667 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8668 char *glob, char *cmd, char *param, int enable)
8670 struct ftrace_probe_ops *ops;
8671 void *count = (void *)-1;
8678 /* hash funcs only work with set_ftrace_filter */
8682 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8685 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8690 number = strsep(¶m, ":");
8692 if (!strlen(number))
8696 * We use the callback data field (which is a pointer)
8699 ret = kstrtoul(number, 0, (unsigned long *)&count);
8704 ret = tracing_alloc_snapshot_instance(tr);
8708 ret = register_ftrace_function_probe(glob, tr, ops, count);
8711 return ret < 0 ? ret : 0;
8714 static struct ftrace_func_command ftrace_snapshot_cmd = {
8716 .func = ftrace_trace_snapshot_callback,
8719 static __init int register_snapshot_cmd(void)
8721 return register_ftrace_command(&ftrace_snapshot_cmd);
8724 static inline __init int register_snapshot_cmd(void) { return 0; }
8725 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8727 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8729 if (WARN_ON(!tr->dir))
8730 return ERR_PTR(-ENODEV);
8732 /* Top directory uses NULL as the parent */
8733 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8736 /* All sub buffers have a descriptor */
8740 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8742 struct dentry *d_tracer;
8745 return tr->percpu_dir;
8747 d_tracer = tracing_get_dentry(tr);
8748 if (IS_ERR(d_tracer))
8751 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8753 MEM_FAIL(!tr->percpu_dir,
8754 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8756 return tr->percpu_dir;
8759 static struct dentry *
8760 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8761 void *data, long cpu, const struct file_operations *fops)
8763 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8765 if (ret) /* See tracing_get_cpu() */
8766 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8771 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8773 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8774 struct dentry *d_cpu;
8775 char cpu_dir[30]; /* 30 characters should be more than enough */
8780 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8781 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8783 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8787 /* per cpu trace_pipe */
8788 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8789 tr, cpu, &tracing_pipe_fops);
8792 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8793 tr, cpu, &tracing_fops);
8795 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8796 tr, cpu, &tracing_buffers_fops);
8798 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8799 tr, cpu, &tracing_stats_fops);
8801 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8802 tr, cpu, &tracing_entries_fops);
8804 #ifdef CONFIG_TRACER_SNAPSHOT
8805 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8806 tr, cpu, &snapshot_fops);
8808 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8809 tr, cpu, &snapshot_raw_fops);
8813 #ifdef CONFIG_FTRACE_SELFTEST
8814 /* Let selftest have access to static functions in this file */
8815 #include "trace_selftest.c"
8819 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8822 struct trace_option_dentry *topt = filp->private_data;
8825 if (topt->flags->val & topt->opt->bit)
8830 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8834 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8837 struct trace_option_dentry *topt = filp->private_data;
8841 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8845 if (val != 0 && val != 1)
8848 if (!!(topt->flags->val & topt->opt->bit) != val) {
8849 mutex_lock(&trace_types_lock);
8850 ret = __set_tracer_option(topt->tr, topt->flags,
8852 mutex_unlock(&trace_types_lock);
8863 static const struct file_operations trace_options_fops = {
8864 .open = tracing_open_generic,
8865 .read = trace_options_read,
8866 .write = trace_options_write,
8867 .llseek = generic_file_llseek,
8871 * In order to pass in both the trace_array descriptor as well as the index
8872 * to the flag that the trace option file represents, the trace_array
8873 * has a character array of trace_flags_index[], which holds the index
8874 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8875 * The address of this character array is passed to the flag option file
8876 * read/write callbacks.
8878 * In order to extract both the index and the trace_array descriptor,
8879 * get_tr_index() uses the following algorithm.
8883 * As the pointer itself contains the address of the index (remember
8886 * Then to get the trace_array descriptor, by subtracting that index
8887 * from the ptr, we get to the start of the index itself.
8889 * ptr - idx == &index[0]
8891 * Then a simple container_of() from that pointer gets us to the
8892 * trace_array descriptor.
8894 static void get_tr_index(void *data, struct trace_array **ptr,
8895 unsigned int *pindex)
8897 *pindex = *(unsigned char *)data;
8899 *ptr = container_of(data - *pindex, struct trace_array,
8904 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8907 void *tr_index = filp->private_data;
8908 struct trace_array *tr;
8912 get_tr_index(tr_index, &tr, &index);
8914 if (tr->trace_flags & (1 << index))
8919 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8923 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8926 void *tr_index = filp->private_data;
8927 struct trace_array *tr;
8932 get_tr_index(tr_index, &tr, &index);
8934 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8938 if (val != 0 && val != 1)
8941 mutex_lock(&event_mutex);
8942 mutex_lock(&trace_types_lock);
8943 ret = set_tracer_flag(tr, 1 << index, val);
8944 mutex_unlock(&trace_types_lock);
8945 mutex_unlock(&event_mutex);
8955 static const struct file_operations trace_options_core_fops = {
8956 .open = tracing_open_generic,
8957 .read = trace_options_core_read,
8958 .write = trace_options_core_write,
8959 .llseek = generic_file_llseek,
8962 struct dentry *trace_create_file(const char *name,
8964 struct dentry *parent,
8966 const struct file_operations *fops)
8970 ret = tracefs_create_file(name, mode, parent, data, fops);
8972 pr_warn("Could not create tracefs '%s' entry\n", name);
8978 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8980 struct dentry *d_tracer;
8985 d_tracer = tracing_get_dentry(tr);
8986 if (IS_ERR(d_tracer))
8989 tr->options = tracefs_create_dir("options", d_tracer);
8991 pr_warn("Could not create tracefs directory 'options'\n");
8999 create_trace_option_file(struct trace_array *tr,
9000 struct trace_option_dentry *topt,
9001 struct tracer_flags *flags,
9002 struct tracer_opt *opt)
9004 struct dentry *t_options;
9006 t_options = trace_options_init_dentry(tr);
9010 topt->flags = flags;
9014 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
9015 t_options, topt, &trace_options_fops);
9020 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
9022 struct trace_option_dentry *topts;
9023 struct trace_options *tr_topts;
9024 struct tracer_flags *flags;
9025 struct tracer_opt *opts;
9032 flags = tracer->flags;
9034 if (!flags || !flags->opts)
9038 * If this is an instance, only create flags for tracers
9039 * the instance may have.
9041 if (!trace_ok_for_array(tracer, tr))
9044 for (i = 0; i < tr->nr_topts; i++) {
9045 /* Make sure there's no duplicate flags. */
9046 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
9052 for (cnt = 0; opts[cnt].name; cnt++)
9055 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
9059 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9066 tr->topts = tr_topts;
9067 tr->topts[tr->nr_topts].tracer = tracer;
9068 tr->topts[tr->nr_topts].topts = topts;
9071 for (cnt = 0; opts[cnt].name; cnt++) {
9072 create_trace_option_file(tr, &topts[cnt], flags,
9074 MEM_FAIL(topts[cnt].entry == NULL,
9075 "Failed to create trace option: %s",
9080 static struct dentry *
9081 create_trace_option_core_file(struct trace_array *tr,
9082 const char *option, long index)
9084 struct dentry *t_options;
9086 t_options = trace_options_init_dentry(tr);
9090 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9091 (void *)&tr->trace_flags_index[index],
9092 &trace_options_core_fops);
9095 static void create_trace_options_dir(struct trace_array *tr)
9097 struct dentry *t_options;
9098 bool top_level = tr == &global_trace;
9101 t_options = trace_options_init_dentry(tr);
9105 for (i = 0; trace_options[i]; i++) {
9107 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9108 create_trace_option_core_file(tr, trace_options[i], i);
9113 rb_simple_read(struct file *filp, char __user *ubuf,
9114 size_t cnt, loff_t *ppos)
9116 struct trace_array *tr = filp->private_data;
9120 r = tracer_tracing_is_on(tr);
9121 r = sprintf(buf, "%d\n", r);
9123 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9127 rb_simple_write(struct file *filp, const char __user *ubuf,
9128 size_t cnt, loff_t *ppos)
9130 struct trace_array *tr = filp->private_data;
9131 struct trace_buffer *buffer = tr->array_buffer.buffer;
9135 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9140 mutex_lock(&trace_types_lock);
9141 if (!!val == tracer_tracing_is_on(tr)) {
9142 val = 0; /* do nothing */
9144 tracer_tracing_on(tr);
9145 if (tr->current_trace->start)
9146 tr->current_trace->start(tr);
9148 tracer_tracing_off(tr);
9149 if (tr->current_trace->stop)
9150 tr->current_trace->stop(tr);
9151 /* Wake up any waiters */
9152 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9154 mutex_unlock(&trace_types_lock);
9162 static const struct file_operations rb_simple_fops = {
9163 .open = tracing_open_generic_tr,
9164 .read = rb_simple_read,
9165 .write = rb_simple_write,
9166 .release = tracing_release_generic_tr,
9167 .llseek = default_llseek,
9171 buffer_percent_read(struct file *filp, char __user *ubuf,
9172 size_t cnt, loff_t *ppos)
9174 struct trace_array *tr = filp->private_data;
9178 r = tr->buffer_percent;
9179 r = sprintf(buf, "%d\n", r);
9181 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9185 buffer_percent_write(struct file *filp, const char __user *ubuf,
9186 size_t cnt, loff_t *ppos)
9188 struct trace_array *tr = filp->private_data;
9192 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9199 tr->buffer_percent = val;
9206 static const struct file_operations buffer_percent_fops = {
9207 .open = tracing_open_generic_tr,
9208 .read = buffer_percent_read,
9209 .write = buffer_percent_write,
9210 .release = tracing_release_generic_tr,
9211 .llseek = default_llseek,
9214 static struct dentry *trace_instance_dir;
9217 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9220 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9222 enum ring_buffer_flags rb_flags;
9224 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9228 buf->buffer = ring_buffer_alloc(size, rb_flags);
9232 buf->data = alloc_percpu(struct trace_array_cpu);
9234 ring_buffer_free(buf->buffer);
9239 /* Allocate the first page for all buffers */
9240 set_buffer_entries(&tr->array_buffer,
9241 ring_buffer_size(tr->array_buffer.buffer, 0));
9246 static void free_trace_buffer(struct array_buffer *buf)
9249 ring_buffer_free(buf->buffer);
9251 free_percpu(buf->data);
9256 static int allocate_trace_buffers(struct trace_array *tr, int size)
9260 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9264 #ifdef CONFIG_TRACER_MAX_TRACE
9265 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9266 allocate_snapshot ? size : 1);
9267 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9268 free_trace_buffer(&tr->array_buffer);
9271 tr->allocated_snapshot = allocate_snapshot;
9273 allocate_snapshot = false;
9279 static void free_trace_buffers(struct trace_array *tr)
9284 free_trace_buffer(&tr->array_buffer);
9286 #ifdef CONFIG_TRACER_MAX_TRACE
9287 free_trace_buffer(&tr->max_buffer);
9291 static void init_trace_flags_index(struct trace_array *tr)
9295 /* Used by the trace options files */
9296 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9297 tr->trace_flags_index[i] = i;
9300 static void __update_tracer_options(struct trace_array *tr)
9304 for (t = trace_types; t; t = t->next)
9305 add_tracer_options(tr, t);
9308 static void update_tracer_options(struct trace_array *tr)
9310 mutex_lock(&trace_types_lock);
9311 tracer_options_updated = true;
9312 __update_tracer_options(tr);
9313 mutex_unlock(&trace_types_lock);
9316 /* Must have trace_types_lock held */
9317 struct trace_array *trace_array_find(const char *instance)
9319 struct trace_array *tr, *found = NULL;
9321 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9322 if (tr->name && strcmp(tr->name, instance) == 0) {
9331 struct trace_array *trace_array_find_get(const char *instance)
9333 struct trace_array *tr;
9335 mutex_lock(&trace_types_lock);
9336 tr = trace_array_find(instance);
9339 mutex_unlock(&trace_types_lock);
9344 static int trace_array_create_dir(struct trace_array *tr)
9348 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9352 ret = event_trace_add_tracer(tr->dir, tr);
9354 tracefs_remove(tr->dir);
9358 init_tracer_tracefs(tr, tr->dir);
9359 __update_tracer_options(tr);
9364 static struct trace_array *trace_array_create(const char *name)
9366 struct trace_array *tr;
9370 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9372 return ERR_PTR(ret);
9374 tr->name = kstrdup(name, GFP_KERNEL);
9378 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9381 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9383 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9385 raw_spin_lock_init(&tr->start_lock);
9387 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9389 tr->current_trace = &nop_trace;
9391 INIT_LIST_HEAD(&tr->systems);
9392 INIT_LIST_HEAD(&tr->events);
9393 INIT_LIST_HEAD(&tr->hist_vars);
9394 INIT_LIST_HEAD(&tr->err_log);
9396 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9399 if (ftrace_allocate_ftrace_ops(tr) < 0)
9402 ftrace_init_trace_array(tr);
9404 init_trace_flags_index(tr);
9406 if (trace_instance_dir) {
9407 ret = trace_array_create_dir(tr);
9411 __trace_early_add_events(tr);
9413 list_add(&tr->list, &ftrace_trace_arrays);
9420 ftrace_free_ftrace_ops(tr);
9421 free_trace_buffers(tr);
9422 free_cpumask_var(tr->tracing_cpumask);
9426 return ERR_PTR(ret);
9429 static int instance_mkdir(const char *name)
9431 struct trace_array *tr;
9434 mutex_lock(&event_mutex);
9435 mutex_lock(&trace_types_lock);
9438 if (trace_array_find(name))
9441 tr = trace_array_create(name);
9443 ret = PTR_ERR_OR_ZERO(tr);
9446 mutex_unlock(&trace_types_lock);
9447 mutex_unlock(&event_mutex);
9452 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9453 * @name: The name of the trace array to be looked up/created.
9455 * Returns pointer to trace array with given name.
9456 * NULL, if it cannot be created.
9458 * NOTE: This function increments the reference counter associated with the
9459 * trace array returned. This makes sure it cannot be freed while in use.
9460 * Use trace_array_put() once the trace array is no longer needed.
9461 * If the trace_array is to be freed, trace_array_destroy() needs to
9462 * be called after the trace_array_put(), or simply let user space delete
9463 * it from the tracefs instances directory. But until the
9464 * trace_array_put() is called, user space can not delete it.
9467 struct trace_array *trace_array_get_by_name(const char *name)
9469 struct trace_array *tr;
9471 mutex_lock(&event_mutex);
9472 mutex_lock(&trace_types_lock);
9474 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9475 if (tr->name && strcmp(tr->name, name) == 0)
9479 tr = trace_array_create(name);
9487 mutex_unlock(&trace_types_lock);
9488 mutex_unlock(&event_mutex);
9491 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9493 static int __remove_instance(struct trace_array *tr)
9497 /* Reference counter for a newly created trace array = 1. */
9498 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9501 list_del(&tr->list);
9503 /* Disable all the flags that were enabled coming in */
9504 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9505 if ((1 << i) & ZEROED_TRACE_FLAGS)
9506 set_tracer_flag(tr, 1 << i, 0);
9509 tracing_set_nop(tr);
9510 clear_ftrace_function_probes(tr);
9511 event_trace_del_tracer(tr);
9512 ftrace_clear_pids(tr);
9513 ftrace_destroy_function_files(tr);
9514 tracefs_remove(tr->dir);
9515 free_percpu(tr->last_func_repeats);
9516 free_trace_buffers(tr);
9518 for (i = 0; i < tr->nr_topts; i++) {
9519 kfree(tr->topts[i].topts);
9523 free_cpumask_var(tr->tracing_cpumask);
9530 int trace_array_destroy(struct trace_array *this_tr)
9532 struct trace_array *tr;
9538 mutex_lock(&event_mutex);
9539 mutex_lock(&trace_types_lock);
9543 /* Making sure trace array exists before destroying it. */
9544 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9545 if (tr == this_tr) {
9546 ret = __remove_instance(tr);
9551 mutex_unlock(&trace_types_lock);
9552 mutex_unlock(&event_mutex);
9556 EXPORT_SYMBOL_GPL(trace_array_destroy);
9558 static int instance_rmdir(const char *name)
9560 struct trace_array *tr;
9563 mutex_lock(&event_mutex);
9564 mutex_lock(&trace_types_lock);
9567 tr = trace_array_find(name);
9569 ret = __remove_instance(tr);
9571 mutex_unlock(&trace_types_lock);
9572 mutex_unlock(&event_mutex);
9577 static __init void create_trace_instances(struct dentry *d_tracer)
9579 struct trace_array *tr;
9581 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9584 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9587 mutex_lock(&event_mutex);
9588 mutex_lock(&trace_types_lock);
9590 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9593 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9594 "Failed to create instance directory\n"))
9598 mutex_unlock(&trace_types_lock);
9599 mutex_unlock(&event_mutex);
9603 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9605 struct trace_event_file *file;
9608 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9609 tr, &show_traces_fops);
9611 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9612 tr, &set_tracer_fops);
9614 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9615 tr, &tracing_cpumask_fops);
9617 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9618 tr, &tracing_iter_fops);
9620 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9623 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9624 tr, &tracing_pipe_fops);
9626 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9627 tr, &tracing_entries_fops);
9629 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9630 tr, &tracing_total_entries_fops);
9632 trace_create_file("free_buffer", 0200, d_tracer,
9633 tr, &tracing_free_buffer_fops);
9635 trace_create_file("trace_marker", 0220, d_tracer,
9636 tr, &tracing_mark_fops);
9638 file = __find_event_file(tr, "ftrace", "print");
9639 if (file && file->dir)
9640 trace_create_file("trigger", TRACE_MODE_WRITE, file->dir,
9641 file, &event_trigger_fops);
9642 tr->trace_marker_file = file;
9644 trace_create_file("trace_marker_raw", 0220, d_tracer,
9645 tr, &tracing_mark_raw_fops);
9647 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9650 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9651 tr, &rb_simple_fops);
9653 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9654 &trace_time_stamp_mode_fops);
9656 tr->buffer_percent = 50;
9658 trace_create_file("buffer_percent", TRACE_MODE_READ, d_tracer,
9659 tr, &buffer_percent_fops);
9661 create_trace_options_dir(tr);
9663 #ifdef CONFIG_TRACER_MAX_TRACE
9664 trace_create_maxlat_file(tr, d_tracer);
9667 if (ftrace_create_function_files(tr, d_tracer))
9668 MEM_FAIL(1, "Could not allocate function filter files");
9670 #ifdef CONFIG_TRACER_SNAPSHOT
9671 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9672 tr, &snapshot_fops);
9675 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9676 tr, &tracing_err_log_fops);
9678 for_each_tracing_cpu(cpu)
9679 tracing_init_tracefs_percpu(tr, cpu);
9681 ftrace_init_tracefs(tr, d_tracer);
9684 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9686 struct vfsmount *mnt;
9687 struct file_system_type *type;
9690 * To maintain backward compatibility for tools that mount
9691 * debugfs to get to the tracing facility, tracefs is automatically
9692 * mounted to the debugfs/tracing directory.
9694 type = get_fs_type("tracefs");
9697 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9698 put_filesystem(type);
9707 * tracing_init_dentry - initialize top level trace array
9709 * This is called when creating files or directories in the tracing
9710 * directory. It is called via fs_initcall() by any of the boot up code
9711 * and expects to return the dentry of the top level tracing directory.
9713 int tracing_init_dentry(void)
9715 struct trace_array *tr = &global_trace;
9717 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9718 pr_warn("Tracing disabled due to lockdown\n");
9722 /* The top level trace array uses NULL as parent */
9726 if (WARN_ON(!tracefs_initialized()))
9730 * As there may still be users that expect the tracing
9731 * files to exist in debugfs/tracing, we must automount
9732 * the tracefs file system there, so older tools still
9733 * work with the newer kernel.
9735 tr->dir = debugfs_create_automount("tracing", NULL,
9736 trace_automount, NULL);
9741 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9742 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9744 static struct workqueue_struct *eval_map_wq __initdata;
9745 static struct work_struct eval_map_work __initdata;
9746 static struct work_struct tracerfs_init_work __initdata;
9748 static void __init eval_map_work_func(struct work_struct *work)
9752 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9753 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9756 static int __init trace_eval_init(void)
9758 INIT_WORK(&eval_map_work, eval_map_work_func);
9760 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9762 pr_err("Unable to allocate eval_map_wq\n");
9764 eval_map_work_func(&eval_map_work);
9768 queue_work(eval_map_wq, &eval_map_work);
9772 subsys_initcall(trace_eval_init);
9774 static int __init trace_eval_sync(void)
9776 /* Make sure the eval map updates are finished */
9778 destroy_workqueue(eval_map_wq);
9782 late_initcall_sync(trace_eval_sync);
9785 #ifdef CONFIG_MODULES
9786 static void trace_module_add_evals(struct module *mod)
9788 if (!mod->num_trace_evals)
9792 * Modules with bad taint do not have events created, do
9793 * not bother with enums either.
9795 if (trace_module_has_bad_taint(mod))
9798 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9801 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
9802 static void trace_module_remove_evals(struct module *mod)
9804 union trace_eval_map_item *map;
9805 union trace_eval_map_item **last = &trace_eval_maps;
9807 if (!mod->num_trace_evals)
9810 mutex_lock(&trace_eval_mutex);
9812 map = trace_eval_maps;
9815 if (map->head.mod == mod)
9817 map = trace_eval_jmp_to_tail(map);
9818 last = &map->tail.next;
9819 map = map->tail.next;
9824 *last = trace_eval_jmp_to_tail(map)->tail.next;
9827 mutex_unlock(&trace_eval_mutex);
9830 static inline void trace_module_remove_evals(struct module *mod) { }
9831 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9833 static int trace_module_notify(struct notifier_block *self,
9834 unsigned long val, void *data)
9836 struct module *mod = data;
9839 case MODULE_STATE_COMING:
9840 trace_module_add_evals(mod);
9842 case MODULE_STATE_GOING:
9843 trace_module_remove_evals(mod);
9850 static struct notifier_block trace_module_nb = {
9851 .notifier_call = trace_module_notify,
9854 #endif /* CONFIG_MODULES */
9856 static __init void tracer_init_tracefs_work_func(struct work_struct *work)
9861 init_tracer_tracefs(&global_trace, NULL);
9862 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9864 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9865 &global_trace, &tracing_thresh_fops);
9867 trace_create_file("README", TRACE_MODE_READ, NULL,
9868 NULL, &tracing_readme_fops);
9870 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9871 NULL, &tracing_saved_cmdlines_fops);
9873 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9874 NULL, &tracing_saved_cmdlines_size_fops);
9876 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9877 NULL, &tracing_saved_tgids_fops);
9879 trace_create_eval_file(NULL);
9881 #ifdef CONFIG_MODULES
9882 register_module_notifier(&trace_module_nb);
9885 #ifdef CONFIG_DYNAMIC_FTRACE
9886 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
9887 NULL, &tracing_dyn_info_fops);
9890 create_trace_instances(NULL);
9892 update_tracer_options(&global_trace);
9895 static __init int tracer_init_tracefs(void)
9899 trace_access_lock_init();
9901 ret = tracing_init_dentry();
9906 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
9907 queue_work(eval_map_wq, &tracerfs_init_work);
9909 tracer_init_tracefs_work_func(NULL);
9912 rv_init_interface();
9917 fs_initcall(tracer_init_tracefs);
9919 static int trace_die_panic_handler(struct notifier_block *self,
9920 unsigned long ev, void *unused);
9922 static struct notifier_block trace_panic_notifier = {
9923 .notifier_call = trace_die_panic_handler,
9924 .priority = INT_MAX - 1,
9927 static struct notifier_block trace_die_notifier = {
9928 .notifier_call = trace_die_panic_handler,
9929 .priority = INT_MAX - 1,
9933 * The idea is to execute the following die/panic callback early, in order
9934 * to avoid showing irrelevant information in the trace (like other panic
9935 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
9936 * warnings get disabled (to prevent potential log flooding).
9938 static int trace_die_panic_handler(struct notifier_block *self,
9939 unsigned long ev, void *unused)
9941 if (!ftrace_dump_on_oops)
9944 /* The die notifier requires DIE_OOPS to trigger */
9945 if (self == &trace_die_notifier && ev != DIE_OOPS)
9948 ftrace_dump(ftrace_dump_on_oops);
9954 * printk is set to max of 1024, we really don't need it that big.
9955 * Nothing should be printing 1000 characters anyway.
9957 #define TRACE_MAX_PRINT 1000
9960 * Define here KERN_TRACE so that we have one place to modify
9961 * it if we decide to change what log level the ftrace dump
9964 #define KERN_TRACE KERN_EMERG
9967 trace_printk_seq(struct trace_seq *s)
9969 /* Probably should print a warning here. */
9970 if (s->seq.len >= TRACE_MAX_PRINT)
9971 s->seq.len = TRACE_MAX_PRINT;
9974 * More paranoid code. Although the buffer size is set to
9975 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9976 * an extra layer of protection.
9978 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9979 s->seq.len = s->seq.size - 1;
9981 /* should be zero ended, but we are paranoid. */
9982 s->buffer[s->seq.len] = 0;
9984 printk(KERN_TRACE "%s", s->buffer);
9989 void trace_init_global_iter(struct trace_iterator *iter)
9991 iter->tr = &global_trace;
9992 iter->trace = iter->tr->current_trace;
9993 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9994 iter->array_buffer = &global_trace.array_buffer;
9996 if (iter->trace && iter->trace->open)
9997 iter->trace->open(iter);
9999 /* Annotate start of buffers if we had overruns */
10000 if (ring_buffer_overruns(iter->array_buffer->buffer))
10001 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10003 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10004 if (trace_clocks[iter->tr->clock_id].in_ns)
10005 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10007 /* Can not use kmalloc for iter.temp and iter.fmt */
10008 iter->temp = static_temp_buf;
10009 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10010 iter->fmt = static_fmt_buf;
10011 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10014 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10016 /* use static because iter can be a bit big for the stack */
10017 static struct trace_iterator iter;
10018 static atomic_t dump_running;
10019 struct trace_array *tr = &global_trace;
10020 unsigned int old_userobj;
10021 unsigned long flags;
10024 /* Only allow one dump user at a time. */
10025 if (atomic_inc_return(&dump_running) != 1) {
10026 atomic_dec(&dump_running);
10031 * Always turn off tracing when we dump.
10032 * We don't need to show trace output of what happens
10033 * between multiple crashes.
10035 * If the user does a sysrq-z, then they can re-enable
10036 * tracing with echo 1 > tracing_on.
10040 local_irq_save(flags);
10042 /* Simulate the iterator */
10043 trace_init_global_iter(&iter);
10045 for_each_tracing_cpu(cpu) {
10046 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10049 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10051 /* don't look at user memory in panic mode */
10052 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10054 switch (oops_dump_mode) {
10056 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10059 iter.cpu_file = raw_smp_processor_id();
10064 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10065 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10068 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10070 /* Did function tracer already get disabled? */
10071 if (ftrace_is_dead()) {
10072 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10073 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10077 * We need to stop all tracing on all CPUS to read
10078 * the next buffer. This is a bit expensive, but is
10079 * not done often. We fill all what we can read,
10080 * and then release the locks again.
10083 while (!trace_empty(&iter)) {
10086 printk(KERN_TRACE "---------------------------------\n");
10090 trace_iterator_reset(&iter);
10091 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10093 if (trace_find_next_entry_inc(&iter) != NULL) {
10096 ret = print_trace_line(&iter);
10097 if (ret != TRACE_TYPE_NO_CONSUME)
10098 trace_consume(&iter);
10100 touch_nmi_watchdog();
10102 trace_printk_seq(&iter.seq);
10106 printk(KERN_TRACE " (ftrace buffer empty)\n");
10108 printk(KERN_TRACE "---------------------------------\n");
10111 tr->trace_flags |= old_userobj;
10113 for_each_tracing_cpu(cpu) {
10114 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10116 atomic_dec(&dump_running);
10117 local_irq_restore(flags);
10119 EXPORT_SYMBOL_GPL(ftrace_dump);
10121 #define WRITE_BUFSIZE 4096
10123 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10124 size_t count, loff_t *ppos,
10125 int (*createfn)(const char *))
10127 char *kbuf, *buf, *tmp;
10132 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10136 while (done < count) {
10137 size = count - done;
10139 if (size >= WRITE_BUFSIZE)
10140 size = WRITE_BUFSIZE - 1;
10142 if (copy_from_user(kbuf, buffer + done, size)) {
10149 tmp = strchr(buf, '\n');
10152 size = tmp - buf + 1;
10154 size = strlen(buf);
10155 if (done + size < count) {
10158 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10159 pr_warn("Line length is too long: Should be less than %d\n",
10160 WRITE_BUFSIZE - 2);
10167 /* Remove comments */
10168 tmp = strchr(buf, '#');
10173 ret = createfn(buf);
10178 } while (done < count);
10188 #ifdef CONFIG_TRACER_MAX_TRACE
10189 __init static bool tr_needs_alloc_snapshot(const char *name)
10192 int len = strlen(name);
10195 if (!boot_snapshot_index)
10198 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10199 boot_snapshot_info[len] == '\t')
10202 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10206 sprintf(test, "\t%s\t", name);
10207 ret = strstr(boot_snapshot_info, test) == NULL;
10212 __init static void do_allocate_snapshot(const char *name)
10214 if (!tr_needs_alloc_snapshot(name))
10218 * When allocate_snapshot is set, the next call to
10219 * allocate_trace_buffers() (called by trace_array_get_by_name())
10220 * will allocate the snapshot buffer. That will alse clear
10223 allocate_snapshot = true;
10226 static inline void do_allocate_snapshot(const char *name) { }
10229 __init static void enable_instances(void)
10231 struct trace_array *tr;
10236 /* A tab is always appended */
10237 boot_instance_info[boot_instance_index - 1] = '\0';
10238 str = boot_instance_info;
10240 while ((curr_str = strsep(&str, "\t"))) {
10242 tok = strsep(&curr_str, ",");
10244 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10245 do_allocate_snapshot(tok);
10247 tr = trace_array_get_by_name(tok);
10249 pr_warn("Failed to create instance buffer %s\n", curr_str);
10252 /* Allow user space to delete it */
10253 trace_array_put(tr);
10255 while ((tok = strsep(&curr_str, ","))) {
10256 early_enable_events(tr, tok, true);
10261 __init static int tracer_alloc_buffers(void)
10267 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10268 pr_warn("Tracing disabled due to lockdown\n");
10273 * Make sure we don't accidentally add more trace options
10274 * than we have bits for.
10276 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10278 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10281 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10282 goto out_free_buffer_mask;
10284 /* Only allocate trace_printk buffers if a trace_printk exists */
10285 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10286 /* Must be called before global_trace.buffer is allocated */
10287 trace_printk_init_buffers();
10289 /* To save memory, keep the ring buffer size to its minimum */
10290 if (ring_buffer_expanded)
10291 ring_buf_size = trace_buf_size;
10295 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10296 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10298 raw_spin_lock_init(&global_trace.start_lock);
10301 * The prepare callbacks allocates some memory for the ring buffer. We
10302 * don't free the buffer if the CPU goes down. If we were to free
10303 * the buffer, then the user would lose any trace that was in the
10304 * buffer. The memory will be removed once the "instance" is removed.
10306 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10307 "trace/RB:prepare", trace_rb_cpu_prepare,
10310 goto out_free_cpumask;
10311 /* Used for event triggers */
10313 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10315 goto out_rm_hp_state;
10317 if (trace_create_savedcmd() < 0)
10318 goto out_free_temp_buffer;
10320 /* TODO: make the number of buffers hot pluggable with CPUS */
10321 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10322 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10323 goto out_free_savedcmd;
10326 if (global_trace.buffer_disabled)
10329 if (trace_boot_clock) {
10330 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10332 pr_warn("Trace clock %s not defined, going back to default\n",
10337 * register_tracer() might reference current_trace, so it
10338 * needs to be set before we register anything. This is
10339 * just a bootstrap of current_trace anyway.
10341 global_trace.current_trace = &nop_trace;
10343 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10345 ftrace_init_global_array_ops(&global_trace);
10347 init_trace_flags_index(&global_trace);
10349 register_tracer(&nop_trace);
10351 /* Function tracing may start here (via kernel command line) */
10352 init_function_trace();
10354 /* All seems OK, enable tracing */
10355 tracing_disabled = 0;
10357 atomic_notifier_chain_register(&panic_notifier_list,
10358 &trace_panic_notifier);
10360 register_die_notifier(&trace_die_notifier);
10362 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10364 INIT_LIST_HEAD(&global_trace.systems);
10365 INIT_LIST_HEAD(&global_trace.events);
10366 INIT_LIST_HEAD(&global_trace.hist_vars);
10367 INIT_LIST_HEAD(&global_trace.err_log);
10368 list_add(&global_trace.list, &ftrace_trace_arrays);
10370 apply_trace_boot_options();
10372 register_snapshot_cmd();
10379 free_saved_cmdlines_buffer(savedcmd);
10380 out_free_temp_buffer:
10381 ring_buffer_free(temp_buffer);
10383 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10385 free_cpumask_var(global_trace.tracing_cpumask);
10386 out_free_buffer_mask:
10387 free_cpumask_var(tracing_buffer_mask);
10392 void __init ftrace_boot_snapshot(void)
10394 struct trace_array *tr;
10396 if (snapshot_at_boot) {
10397 tracing_snapshot();
10398 internal_trace_puts("** Boot snapshot taken **\n");
10401 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10402 if (tr == &global_trace)
10404 trace_array_puts(tr, "** Boot snapshot taken **\n");
10405 tracing_snapshot_instance(tr);
10409 void __init early_trace_init(void)
10411 if (tracepoint_printk) {
10412 tracepoint_print_iter =
10413 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10414 if (MEM_FAIL(!tracepoint_print_iter,
10415 "Failed to allocate trace iterator\n"))
10416 tracepoint_printk = 0;
10418 static_key_enable(&tracepoint_printk_key.key);
10420 tracer_alloc_buffers();
10425 void __init trace_init(void)
10427 trace_event_init();
10429 if (boot_instance_index)
10430 enable_instances();
10433 __init static void clear_boot_tracer(void)
10436 * The default tracer at boot buffer is an init section.
10437 * This function is called in lateinit. If we did not
10438 * find the boot tracer, then clear it out, to prevent
10439 * later registration from accessing the buffer that is
10440 * about to be freed.
10442 if (!default_bootup_tracer)
10445 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10446 default_bootup_tracer);
10447 default_bootup_tracer = NULL;
10450 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10451 __init static void tracing_set_default_clock(void)
10453 /* sched_clock_stable() is determined in late_initcall */
10454 if (!trace_boot_clock && !sched_clock_stable()) {
10455 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10456 pr_warn("Can not set tracing clock due to lockdown\n");
10460 printk(KERN_WARNING
10461 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10462 "If you want to keep using the local clock, then add:\n"
10463 " \"trace_clock=local\"\n"
10464 "on the kernel command line\n");
10465 tracing_set_clock(&global_trace, "global");
10469 static inline void tracing_set_default_clock(void) { }
10472 __init static int late_trace_init(void)
10474 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10475 static_key_disable(&tracepoint_printk_key.key);
10476 tracepoint_printk = 0;
10479 tracing_set_default_clock();
10480 clear_boot_tracer();
10484 late_initcall_sync(late_trace_init);