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;
63 #ifdef CONFIG_FTRACE_STARTUP_TEST
65 * We need to change this state when a selftest is running.
66 * A selftest will lurk into the ring-buffer to count the
67 * entries inserted during the selftest although some concurrent
68 * insertions into the ring-buffer such as trace_printk could occurred
69 * at the same time, giving false positive or negative results.
71 static bool __read_mostly tracing_selftest_running;
74 * If boot-time tracing including tracers/events via kernel cmdline
75 * is running, we do not want to run SELFTEST.
77 bool __read_mostly tracing_selftest_disabled;
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);
87 #define tracing_selftest_running 0
88 #define tracing_selftest_disabled 0
91 /* Pipe tracepoints to printk */
92 static struct trace_iterator *tracepoint_print_iter;
93 int tracepoint_printk;
94 static bool tracepoint_printk_stop_on_boot __initdata;
95 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
97 /* For tracers that don't implement custom flags */
98 static struct tracer_opt dummy_tracer_opt[] = {
103 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
109 * To prevent the comm cache from being overwritten when no
110 * tracing is active, only save the comm when a trace event
113 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
116 * Kill all tracing for good (never come back).
117 * It is initialized to 1 but will turn to zero if the initialization
118 * of the tracer is successful. But that is the only place that sets
121 static int tracing_disabled = 1;
123 cpumask_var_t __read_mostly tracing_buffer_mask;
126 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
128 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
129 * is set, then ftrace_dump is called. This will output the contents
130 * of the ftrace buffers to the console. This is very useful for
131 * capturing traces that lead to crashes and outputing it to a
134 * It is default off, but you can enable it with either specifying
135 * "ftrace_dump_on_oops" in the kernel command line, or setting
136 * /proc/sys/kernel/ftrace_dump_on_oops
137 * Set 1 if you want to dump buffers of all CPUs
138 * Set 2 if you want to dump the buffer of the CPU that triggered oops
141 enum ftrace_dump_mode ftrace_dump_on_oops;
143 /* When set, tracing will stop when a WARN*() is hit */
144 int __disable_trace_on_warning;
146 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
147 /* Map of enums to their values, for "eval_map" file */
148 struct trace_eval_map_head {
150 unsigned long length;
153 union trace_eval_map_item;
155 struct trace_eval_map_tail {
157 * "end" is first and points to NULL as it must be different
158 * than "mod" or "eval_string"
160 union trace_eval_map_item *next;
161 const char *end; /* points to NULL */
164 static DEFINE_MUTEX(trace_eval_mutex);
167 * The trace_eval_maps are saved in an array with two extra elements,
168 * one at the beginning, and one at the end. The beginning item contains
169 * the count of the saved maps (head.length), and the module they
170 * belong to if not built in (head.mod). The ending item contains a
171 * pointer to the next array of saved eval_map items.
173 union trace_eval_map_item {
174 struct trace_eval_map map;
175 struct trace_eval_map_head head;
176 struct trace_eval_map_tail tail;
179 static union trace_eval_map_item *trace_eval_maps;
180 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
182 int tracing_set_tracer(struct trace_array *tr, const char *buf);
183 static void ftrace_trace_userstack(struct trace_array *tr,
184 struct trace_buffer *buffer,
185 unsigned int trace_ctx);
187 #define MAX_TRACER_SIZE 100
188 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
189 static char *default_bootup_tracer;
191 static bool allocate_snapshot;
192 static bool snapshot_at_boot;
194 static char boot_instance_info[COMMAND_LINE_SIZE] __initdata;
195 static int boot_instance_index;
197 static char boot_snapshot_info[COMMAND_LINE_SIZE] __initdata;
198 static int boot_snapshot_index;
200 static int __init set_cmdline_ftrace(char *str)
202 strscpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
203 default_bootup_tracer = bootup_tracer_buf;
204 /* We are using ftrace early, expand it */
205 ring_buffer_expanded = true;
208 __setup("ftrace=", set_cmdline_ftrace);
210 static int __init set_ftrace_dump_on_oops(char *str)
212 if (*str++ != '=' || !*str || !strcmp("1", str)) {
213 ftrace_dump_on_oops = DUMP_ALL;
217 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
218 ftrace_dump_on_oops = DUMP_ORIG;
224 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
226 static int __init stop_trace_on_warning(char *str)
228 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
229 __disable_trace_on_warning = 1;
232 __setup("traceoff_on_warning", stop_trace_on_warning);
234 static int __init boot_alloc_snapshot(char *str)
236 char *slot = boot_snapshot_info + boot_snapshot_index;
237 int left = sizeof(boot_snapshot_info) - boot_snapshot_index;
242 if (strlen(str) >= left)
245 ret = snprintf(slot, left, "%s\t", str);
246 boot_snapshot_index += ret;
248 allocate_snapshot = true;
249 /* We also need the main ring buffer expanded */
250 ring_buffer_expanded = true;
254 __setup("alloc_snapshot", boot_alloc_snapshot);
257 static int __init boot_snapshot(char *str)
259 snapshot_at_boot = true;
260 boot_alloc_snapshot(str);
263 __setup("ftrace_boot_snapshot", boot_snapshot);
266 static int __init boot_instance(char *str)
268 char *slot = boot_instance_info + boot_instance_index;
269 int left = sizeof(boot_instance_info) - boot_instance_index;
272 if (strlen(str) >= left)
275 ret = snprintf(slot, left, "%s\t", str);
276 boot_instance_index += ret;
280 __setup("trace_instance=", boot_instance);
283 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
285 static int __init set_trace_boot_options(char *str)
287 strscpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
290 __setup("trace_options=", set_trace_boot_options);
292 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
293 static char *trace_boot_clock __initdata;
295 static int __init set_trace_boot_clock(char *str)
297 strscpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
298 trace_boot_clock = trace_boot_clock_buf;
301 __setup("trace_clock=", set_trace_boot_clock);
303 static int __init set_tracepoint_printk(char *str)
305 /* Ignore the "tp_printk_stop_on_boot" param */
309 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
310 tracepoint_printk = 1;
313 __setup("tp_printk", set_tracepoint_printk);
315 static int __init set_tracepoint_printk_stop(char *str)
317 tracepoint_printk_stop_on_boot = true;
320 __setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
322 unsigned long long ns2usecs(u64 nsec)
330 trace_process_export(struct trace_export *export,
331 struct ring_buffer_event *event, int flag)
333 struct trace_entry *entry;
334 unsigned int size = 0;
336 if (export->flags & flag) {
337 entry = ring_buffer_event_data(event);
338 size = ring_buffer_event_length(event);
339 export->write(export, entry, size);
343 static DEFINE_MUTEX(ftrace_export_lock);
345 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
347 static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
348 static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
349 static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
351 static inline void ftrace_exports_enable(struct trace_export *export)
353 if (export->flags & TRACE_EXPORT_FUNCTION)
354 static_branch_inc(&trace_function_exports_enabled);
356 if (export->flags & TRACE_EXPORT_EVENT)
357 static_branch_inc(&trace_event_exports_enabled);
359 if (export->flags & TRACE_EXPORT_MARKER)
360 static_branch_inc(&trace_marker_exports_enabled);
363 static inline void ftrace_exports_disable(struct trace_export *export)
365 if (export->flags & TRACE_EXPORT_FUNCTION)
366 static_branch_dec(&trace_function_exports_enabled);
368 if (export->flags & TRACE_EXPORT_EVENT)
369 static_branch_dec(&trace_event_exports_enabled);
371 if (export->flags & TRACE_EXPORT_MARKER)
372 static_branch_dec(&trace_marker_exports_enabled);
375 static void ftrace_exports(struct ring_buffer_event *event, int flag)
377 struct trace_export *export;
379 preempt_disable_notrace();
381 export = rcu_dereference_raw_check(ftrace_exports_list);
383 trace_process_export(export, event, flag);
384 export = rcu_dereference_raw_check(export->next);
387 preempt_enable_notrace();
391 add_trace_export(struct trace_export **list, struct trace_export *export)
393 rcu_assign_pointer(export->next, *list);
395 * We are entering export into the list but another
396 * CPU might be walking that list. We need to make sure
397 * the export->next pointer is valid before another CPU sees
398 * the export pointer included into the list.
400 rcu_assign_pointer(*list, export);
404 rm_trace_export(struct trace_export **list, struct trace_export *export)
406 struct trace_export **p;
408 for (p = list; *p != NULL; p = &(*p)->next)
415 rcu_assign_pointer(*p, (*p)->next);
421 add_ftrace_export(struct trace_export **list, struct trace_export *export)
423 ftrace_exports_enable(export);
425 add_trace_export(list, export);
429 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
433 ret = rm_trace_export(list, export);
434 ftrace_exports_disable(export);
439 int register_ftrace_export(struct trace_export *export)
441 if (WARN_ON_ONCE(!export->write))
444 mutex_lock(&ftrace_export_lock);
446 add_ftrace_export(&ftrace_exports_list, export);
448 mutex_unlock(&ftrace_export_lock);
452 EXPORT_SYMBOL_GPL(register_ftrace_export);
454 int unregister_ftrace_export(struct trace_export *export)
458 mutex_lock(&ftrace_export_lock);
460 ret = rm_ftrace_export(&ftrace_exports_list, export);
462 mutex_unlock(&ftrace_export_lock);
466 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
468 /* trace_flags holds trace_options default values */
469 #define TRACE_DEFAULT_FLAGS \
470 (FUNCTION_DEFAULT_FLAGS | \
471 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
472 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
473 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
474 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
477 /* trace_options that are only supported by global_trace */
478 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
479 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
481 /* trace_flags that are default zero for instances */
482 #define ZEROED_TRACE_FLAGS \
483 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
486 * The global_trace is the descriptor that holds the top-level tracing
487 * buffers for the live tracing.
489 static struct trace_array global_trace = {
490 .trace_flags = TRACE_DEFAULT_FLAGS,
493 LIST_HEAD(ftrace_trace_arrays);
495 int trace_array_get(struct trace_array *this_tr)
497 struct trace_array *tr;
500 mutex_lock(&trace_types_lock);
501 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
508 mutex_unlock(&trace_types_lock);
513 static void __trace_array_put(struct trace_array *this_tr)
515 WARN_ON(!this_tr->ref);
520 * trace_array_put - Decrement the reference counter for this trace array.
521 * @this_tr : pointer to the trace array
523 * NOTE: Use this when we no longer need the trace array returned by
524 * trace_array_get_by_name(). This ensures the trace array can be later
528 void trace_array_put(struct trace_array *this_tr)
533 mutex_lock(&trace_types_lock);
534 __trace_array_put(this_tr);
535 mutex_unlock(&trace_types_lock);
537 EXPORT_SYMBOL_GPL(trace_array_put);
539 int tracing_check_open_get_tr(struct trace_array *tr)
543 ret = security_locked_down(LOCKDOWN_TRACEFS);
547 if (tracing_disabled)
550 if (tr && trace_array_get(tr) < 0)
556 int call_filter_check_discard(struct trace_event_call *call, void *rec,
557 struct trace_buffer *buffer,
558 struct ring_buffer_event *event)
560 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
561 !filter_match_preds(call->filter, rec)) {
562 __trace_event_discard_commit(buffer, event);
570 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
571 * @filtered_pids: The list of pids to check
572 * @search_pid: The PID to find in @filtered_pids
574 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
577 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
579 return trace_pid_list_is_set(filtered_pids, search_pid);
583 * trace_ignore_this_task - should a task be ignored for tracing
584 * @filtered_pids: The list of pids to check
585 * @filtered_no_pids: The list of pids not to be traced
586 * @task: The task that should be ignored if not filtered
588 * Checks if @task should be traced or not from @filtered_pids.
589 * Returns true if @task should *NOT* be traced.
590 * Returns false if @task should be traced.
593 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
594 struct trace_pid_list *filtered_no_pids,
595 struct task_struct *task)
598 * If filtered_no_pids is not empty, and the task's pid is listed
599 * in filtered_no_pids, then return true.
600 * Otherwise, if filtered_pids is empty, that means we can
601 * trace all tasks. If it has content, then only trace pids
602 * within filtered_pids.
605 return (filtered_pids &&
606 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
608 trace_find_filtered_pid(filtered_no_pids, task->pid));
612 * trace_filter_add_remove_task - Add or remove a task from a pid_list
613 * @pid_list: The list to modify
614 * @self: The current task for fork or NULL for exit
615 * @task: The task to add or remove
617 * If adding a task, if @self is defined, the task is only added if @self
618 * is also included in @pid_list. This happens on fork and tasks should
619 * only be added when the parent is listed. If @self is NULL, then the
620 * @task pid will be removed from the list, which would happen on exit
623 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
624 struct task_struct *self,
625 struct task_struct *task)
630 /* For forks, we only add if the forking task is listed */
632 if (!trace_find_filtered_pid(pid_list, self->pid))
636 /* "self" is set for forks, and NULL for exits */
638 trace_pid_list_set(pid_list, task->pid);
640 trace_pid_list_clear(pid_list, task->pid);
644 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
645 * @pid_list: The pid list to show
646 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
647 * @pos: The position of the file
649 * This is used by the seq_file "next" operation to iterate the pids
650 * listed in a trace_pid_list structure.
652 * Returns the pid+1 as we want to display pid of zero, but NULL would
653 * stop the iteration.
655 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
657 long pid = (unsigned long)v;
662 /* pid already is +1 of the actual previous bit */
663 if (trace_pid_list_next(pid_list, pid, &next) < 0)
668 /* Return pid + 1 to allow zero to be represented */
669 return (void *)(pid + 1);
673 * trace_pid_start - Used for seq_file to start reading pid lists
674 * @pid_list: The pid list to show
675 * @pos: The position of the file
677 * This is used by seq_file "start" operation to start the iteration
680 * Returns the pid+1 as we want to display pid of zero, but NULL would
681 * stop the iteration.
683 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
689 if (trace_pid_list_first(pid_list, &first) < 0)
694 /* Return pid + 1 so that zero can be the exit value */
695 for (pid++; pid && l < *pos;
696 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
702 * trace_pid_show - show the current pid in seq_file processing
703 * @m: The seq_file structure to write into
704 * @v: A void pointer of the pid (+1) value to display
706 * Can be directly used by seq_file operations to display the current
709 int trace_pid_show(struct seq_file *m, void *v)
711 unsigned long pid = (unsigned long)v - 1;
713 seq_printf(m, "%lu\n", pid);
717 /* 128 should be much more than enough */
718 #define PID_BUF_SIZE 127
720 int trace_pid_write(struct trace_pid_list *filtered_pids,
721 struct trace_pid_list **new_pid_list,
722 const char __user *ubuf, size_t cnt)
724 struct trace_pid_list *pid_list;
725 struct trace_parser parser;
733 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
737 * Always recreate a new array. The write is an all or nothing
738 * operation. Always create a new array when adding new pids by
739 * the user. If the operation fails, then the current list is
742 pid_list = trace_pid_list_alloc();
744 trace_parser_put(&parser);
749 /* copy the current bits to the new max */
750 ret = trace_pid_list_first(filtered_pids, &pid);
752 trace_pid_list_set(pid_list, pid);
753 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
763 ret = trace_get_user(&parser, ubuf, cnt, &pos);
771 if (!trace_parser_loaded(&parser))
775 if (kstrtoul(parser.buffer, 0, &val))
780 if (trace_pid_list_set(pid_list, pid) < 0) {
786 trace_parser_clear(&parser);
789 trace_parser_put(&parser);
792 trace_pid_list_free(pid_list);
797 /* Cleared the list of pids */
798 trace_pid_list_free(pid_list);
802 *new_pid_list = pid_list;
807 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
811 /* Early boot up does not have a buffer yet */
813 return trace_clock_local();
815 ts = ring_buffer_time_stamp(buf->buffer);
816 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
821 u64 ftrace_now(int cpu)
823 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
827 * tracing_is_enabled - Show if global_trace has been enabled
829 * Shows if the global trace has been enabled or not. It uses the
830 * mirror flag "buffer_disabled" to be used in fast paths such as for
831 * the irqsoff tracer. But it may be inaccurate due to races. If you
832 * need to know the accurate state, use tracing_is_on() which is a little
833 * slower, but accurate.
835 int tracing_is_enabled(void)
838 * For quick access (irqsoff uses this in fast path), just
839 * return the mirror variable of the state of the ring buffer.
840 * It's a little racy, but we don't really care.
843 return !global_trace.buffer_disabled;
847 * trace_buf_size is the size in bytes that is allocated
848 * for a buffer. Note, the number of bytes is always rounded
851 * This number is purposely set to a low number of 16384.
852 * If the dump on oops happens, it will be much appreciated
853 * to not have to wait for all that output. Anyway this can be
854 * boot time and run time configurable.
856 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
858 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
860 /* trace_types holds a link list of available tracers. */
861 static struct tracer *trace_types __read_mostly;
864 * trace_types_lock is used to protect the trace_types list.
866 DEFINE_MUTEX(trace_types_lock);
869 * serialize the access of the ring buffer
871 * ring buffer serializes readers, but it is low level protection.
872 * The validity of the events (which returns by ring_buffer_peek() ..etc)
873 * are not protected by ring buffer.
875 * The content of events may become garbage if we allow other process consumes
876 * these events concurrently:
877 * A) the page of the consumed events may become a normal page
878 * (not reader page) in ring buffer, and this page will be rewritten
879 * by events producer.
880 * B) The page of the consumed events may become a page for splice_read,
881 * and this page will be returned to system.
883 * These primitives allow multi process access to different cpu ring buffer
886 * These primitives don't distinguish read-only and read-consume access.
887 * Multi read-only access are also serialized.
891 static DECLARE_RWSEM(all_cpu_access_lock);
892 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
894 static inline void trace_access_lock(int cpu)
896 if (cpu == RING_BUFFER_ALL_CPUS) {
897 /* gain it for accessing the whole ring buffer. */
898 down_write(&all_cpu_access_lock);
900 /* gain it for accessing a cpu ring buffer. */
902 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
903 down_read(&all_cpu_access_lock);
905 /* Secondly block other access to this @cpu ring buffer. */
906 mutex_lock(&per_cpu(cpu_access_lock, cpu));
910 static inline void trace_access_unlock(int cpu)
912 if (cpu == RING_BUFFER_ALL_CPUS) {
913 up_write(&all_cpu_access_lock);
915 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
916 up_read(&all_cpu_access_lock);
920 static inline void trace_access_lock_init(void)
924 for_each_possible_cpu(cpu)
925 mutex_init(&per_cpu(cpu_access_lock, cpu));
930 static DEFINE_MUTEX(access_lock);
932 static inline void trace_access_lock(int cpu)
935 mutex_lock(&access_lock);
938 static inline void trace_access_unlock(int cpu)
941 mutex_unlock(&access_lock);
944 static inline void trace_access_lock_init(void)
950 #ifdef CONFIG_STACKTRACE
951 static void __ftrace_trace_stack(struct trace_buffer *buffer,
952 unsigned int trace_ctx,
953 int skip, struct pt_regs *regs);
954 static inline void ftrace_trace_stack(struct trace_array *tr,
955 struct trace_buffer *buffer,
956 unsigned int trace_ctx,
957 int skip, struct pt_regs *regs);
960 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
961 unsigned int trace_ctx,
962 int skip, struct pt_regs *regs)
965 static inline void ftrace_trace_stack(struct trace_array *tr,
966 struct trace_buffer *buffer,
967 unsigned long trace_ctx,
968 int skip, struct pt_regs *regs)
974 static __always_inline void
975 trace_event_setup(struct ring_buffer_event *event,
976 int type, unsigned int trace_ctx)
978 struct trace_entry *ent = ring_buffer_event_data(event);
980 tracing_generic_entry_update(ent, type, trace_ctx);
983 static __always_inline struct ring_buffer_event *
984 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
987 unsigned int trace_ctx)
989 struct ring_buffer_event *event;
991 event = ring_buffer_lock_reserve(buffer, len);
993 trace_event_setup(event, type, trace_ctx);
998 void tracer_tracing_on(struct trace_array *tr)
1000 if (tr->array_buffer.buffer)
1001 ring_buffer_record_on(tr->array_buffer.buffer);
1003 * This flag is looked at when buffers haven't been allocated
1004 * yet, or by some tracers (like irqsoff), that just want to
1005 * know if the ring buffer has been disabled, but it can handle
1006 * races of where it gets disabled but we still do a record.
1007 * As the check is in the fast path of the tracers, it is more
1008 * important to be fast than accurate.
1010 tr->buffer_disabled = 0;
1011 /* Make the flag seen by readers */
1016 * tracing_on - enable tracing buffers
1018 * This function enables tracing buffers that may have been
1019 * disabled with tracing_off.
1021 void tracing_on(void)
1023 tracer_tracing_on(&global_trace);
1025 EXPORT_SYMBOL_GPL(tracing_on);
1028 static __always_inline void
1029 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
1031 __this_cpu_write(trace_taskinfo_save, true);
1033 /* If this is the temp buffer, we need to commit fully */
1034 if (this_cpu_read(trace_buffered_event) == event) {
1035 /* Length is in event->array[0] */
1036 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1037 /* Release the temp buffer */
1038 this_cpu_dec(trace_buffered_event_cnt);
1039 /* ring_buffer_unlock_commit() enables preemption */
1040 preempt_enable_notrace();
1042 ring_buffer_unlock_commit(buffer);
1045 int __trace_array_puts(struct trace_array *tr, unsigned long ip,
1046 const char *str, int size)
1048 struct ring_buffer_event *event;
1049 struct trace_buffer *buffer;
1050 struct print_entry *entry;
1051 unsigned int trace_ctx;
1054 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1057 if (unlikely(tracing_selftest_running && tr == &global_trace))
1060 if (unlikely(tracing_disabled))
1063 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1065 trace_ctx = tracing_gen_ctx();
1066 buffer = tr->array_buffer.buffer;
1067 ring_buffer_nest_start(buffer);
1068 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1075 entry = ring_buffer_event_data(event);
1078 memcpy(&entry->buf, str, size);
1080 /* Add a newline if necessary */
1081 if (entry->buf[size - 1] != '\n') {
1082 entry->buf[size] = '\n';
1083 entry->buf[size + 1] = '\0';
1085 entry->buf[size] = '\0';
1087 __buffer_unlock_commit(buffer, event);
1088 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1090 ring_buffer_nest_end(buffer);
1093 EXPORT_SYMBOL_GPL(__trace_array_puts);
1096 * __trace_puts - write a constant string into the trace buffer.
1097 * @ip: The address of the caller
1098 * @str: The constant string to write
1099 * @size: The size of the string.
1101 int __trace_puts(unsigned long ip, const char *str, int size)
1103 return __trace_array_puts(&global_trace, ip, str, size);
1105 EXPORT_SYMBOL_GPL(__trace_puts);
1108 * __trace_bputs - write the pointer to a constant string into trace buffer
1109 * @ip: The address of the caller
1110 * @str: The constant string to write to the buffer to
1112 int __trace_bputs(unsigned long ip, const char *str)
1114 struct ring_buffer_event *event;
1115 struct trace_buffer *buffer;
1116 struct bputs_entry *entry;
1117 unsigned int trace_ctx;
1118 int size = sizeof(struct bputs_entry);
1121 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1124 if (unlikely(tracing_selftest_running || tracing_disabled))
1127 trace_ctx = tracing_gen_ctx();
1128 buffer = global_trace.array_buffer.buffer;
1130 ring_buffer_nest_start(buffer);
1131 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1136 entry = ring_buffer_event_data(event);
1140 __buffer_unlock_commit(buffer, event);
1141 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1145 ring_buffer_nest_end(buffer);
1148 EXPORT_SYMBOL_GPL(__trace_bputs);
1150 #ifdef CONFIG_TRACER_SNAPSHOT
1151 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1154 struct tracer *tracer = tr->current_trace;
1155 unsigned long flags;
1158 trace_array_puts(tr, "*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1159 trace_array_puts(tr, "*** snapshot is being ignored ***\n");
1163 if (!tr->allocated_snapshot) {
1164 trace_array_puts(tr, "*** SNAPSHOT NOT ALLOCATED ***\n");
1165 trace_array_puts(tr, "*** stopping trace here! ***\n");
1166 tracer_tracing_off(tr);
1170 /* Note, snapshot can not be used when the tracer uses it */
1171 if (tracer->use_max_tr) {
1172 trace_array_puts(tr, "*** LATENCY TRACER ACTIVE ***\n");
1173 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1177 local_irq_save(flags);
1178 update_max_tr(tr, current, smp_processor_id(), cond_data);
1179 local_irq_restore(flags);
1182 void tracing_snapshot_instance(struct trace_array *tr)
1184 tracing_snapshot_instance_cond(tr, NULL);
1188 * tracing_snapshot - take a snapshot of the current buffer.
1190 * This causes a swap between the snapshot buffer and the current live
1191 * tracing buffer. You can use this to take snapshots of the live
1192 * trace when some condition is triggered, but continue to trace.
1194 * Note, make sure to allocate the snapshot with either
1195 * a tracing_snapshot_alloc(), or by doing it manually
1196 * with: echo 1 > /sys/kernel/tracing/snapshot
1198 * If the snapshot buffer is not allocated, it will stop tracing.
1199 * Basically making a permanent snapshot.
1201 void tracing_snapshot(void)
1203 struct trace_array *tr = &global_trace;
1205 tracing_snapshot_instance(tr);
1207 EXPORT_SYMBOL_GPL(tracing_snapshot);
1210 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1211 * @tr: The tracing instance to snapshot
1212 * @cond_data: The data to be tested conditionally, and possibly saved
1214 * This is the same as tracing_snapshot() except that the snapshot is
1215 * conditional - the snapshot will only happen if the
1216 * cond_snapshot.update() implementation receiving the cond_data
1217 * returns true, which means that the trace array's cond_snapshot
1218 * update() operation used the cond_data to determine whether the
1219 * snapshot should be taken, and if it was, presumably saved it along
1220 * with the snapshot.
1222 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1224 tracing_snapshot_instance_cond(tr, cond_data);
1226 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1229 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1230 * @tr: The tracing instance
1232 * When the user enables a conditional snapshot using
1233 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1234 * with the snapshot. This accessor is used to retrieve it.
1236 * Should not be called from cond_snapshot.update(), since it takes
1237 * the tr->max_lock lock, which the code calling
1238 * cond_snapshot.update() has already done.
1240 * Returns the cond_data associated with the trace array's snapshot.
1242 void *tracing_cond_snapshot_data(struct trace_array *tr)
1244 void *cond_data = NULL;
1246 local_irq_disable();
1247 arch_spin_lock(&tr->max_lock);
1249 if (tr->cond_snapshot)
1250 cond_data = tr->cond_snapshot->cond_data;
1252 arch_spin_unlock(&tr->max_lock);
1257 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1259 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1260 struct array_buffer *size_buf, int cpu_id);
1261 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1263 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1267 if (!tr->allocated_snapshot) {
1269 /* allocate spare buffer */
1270 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1271 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1275 tr->allocated_snapshot = true;
1281 static void free_snapshot(struct trace_array *tr)
1284 * We don't free the ring buffer. instead, resize it because
1285 * The max_tr ring buffer has some state (e.g. ring->clock) and
1286 * we want preserve it.
1288 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1289 set_buffer_entries(&tr->max_buffer, 1);
1290 tracing_reset_online_cpus(&tr->max_buffer);
1291 tr->allocated_snapshot = false;
1295 * tracing_alloc_snapshot - allocate snapshot buffer.
1297 * This only allocates the snapshot buffer if it isn't already
1298 * allocated - it doesn't also take a snapshot.
1300 * This is meant to be used in cases where the snapshot buffer needs
1301 * to be set up for events that can't sleep but need to be able to
1302 * trigger a snapshot.
1304 int tracing_alloc_snapshot(void)
1306 struct trace_array *tr = &global_trace;
1309 ret = tracing_alloc_snapshot_instance(tr);
1314 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1317 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1319 * This is similar to tracing_snapshot(), but it will allocate the
1320 * snapshot buffer if it isn't already allocated. Use this only
1321 * where it is safe to sleep, as the allocation may sleep.
1323 * This causes a swap between the snapshot buffer and the current live
1324 * tracing buffer. You can use this to take snapshots of the live
1325 * trace when some condition is triggered, but continue to trace.
1327 void tracing_snapshot_alloc(void)
1331 ret = tracing_alloc_snapshot();
1337 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1340 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1341 * @tr: The tracing instance
1342 * @cond_data: User data to associate with the snapshot
1343 * @update: Implementation of the cond_snapshot update function
1345 * Check whether the conditional snapshot for the given instance has
1346 * already been enabled, or if the current tracer is already using a
1347 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1348 * save the cond_data and update function inside.
1350 * Returns 0 if successful, error otherwise.
1352 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1353 cond_update_fn_t update)
1355 struct cond_snapshot *cond_snapshot;
1358 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1362 cond_snapshot->cond_data = cond_data;
1363 cond_snapshot->update = update;
1365 mutex_lock(&trace_types_lock);
1367 ret = tracing_alloc_snapshot_instance(tr);
1371 if (tr->current_trace->use_max_tr) {
1377 * The cond_snapshot can only change to NULL without the
1378 * trace_types_lock. We don't care if we race with it going
1379 * to NULL, but we want to make sure that it's not set to
1380 * something other than NULL when we get here, which we can
1381 * do safely with only holding the trace_types_lock and not
1382 * having to take the max_lock.
1384 if (tr->cond_snapshot) {
1389 local_irq_disable();
1390 arch_spin_lock(&tr->max_lock);
1391 tr->cond_snapshot = cond_snapshot;
1392 arch_spin_unlock(&tr->max_lock);
1395 mutex_unlock(&trace_types_lock);
1400 mutex_unlock(&trace_types_lock);
1401 kfree(cond_snapshot);
1404 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1407 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1408 * @tr: The tracing instance
1410 * Check whether the conditional snapshot for the given instance is
1411 * enabled; if so, free the cond_snapshot associated with it,
1412 * otherwise return -EINVAL.
1414 * Returns 0 if successful, error otherwise.
1416 int tracing_snapshot_cond_disable(struct trace_array *tr)
1420 local_irq_disable();
1421 arch_spin_lock(&tr->max_lock);
1423 if (!tr->cond_snapshot)
1426 kfree(tr->cond_snapshot);
1427 tr->cond_snapshot = NULL;
1430 arch_spin_unlock(&tr->max_lock);
1435 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1437 void tracing_snapshot(void)
1439 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1441 EXPORT_SYMBOL_GPL(tracing_snapshot);
1442 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1444 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1446 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1447 int tracing_alloc_snapshot(void)
1449 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1452 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1453 void tracing_snapshot_alloc(void)
1458 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1459 void *tracing_cond_snapshot_data(struct trace_array *tr)
1463 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1464 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1468 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1469 int tracing_snapshot_cond_disable(struct trace_array *tr)
1473 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1474 #define free_snapshot(tr) do { } while (0)
1475 #endif /* CONFIG_TRACER_SNAPSHOT */
1477 void tracer_tracing_off(struct trace_array *tr)
1479 if (tr->array_buffer.buffer)
1480 ring_buffer_record_off(tr->array_buffer.buffer);
1482 * This flag is looked at when buffers haven't been allocated
1483 * yet, or by some tracers (like irqsoff), that just want to
1484 * know if the ring buffer has been disabled, but it can handle
1485 * races of where it gets disabled but we still do a record.
1486 * As the check is in the fast path of the tracers, it is more
1487 * important to be fast than accurate.
1489 tr->buffer_disabled = 1;
1490 /* Make the flag seen by readers */
1495 * tracing_off - turn off tracing buffers
1497 * This function stops the tracing buffers from recording data.
1498 * It does not disable any overhead the tracers themselves may
1499 * be causing. This function simply causes all recording to
1500 * the ring buffers to fail.
1502 void tracing_off(void)
1504 tracer_tracing_off(&global_trace);
1506 EXPORT_SYMBOL_GPL(tracing_off);
1508 void disable_trace_on_warning(void)
1510 if (__disable_trace_on_warning) {
1511 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1512 "Disabling tracing due to warning\n");
1518 * tracer_tracing_is_on - show real state of ring buffer enabled
1519 * @tr : the trace array to know if ring buffer is enabled
1521 * Shows real state of the ring buffer if it is enabled or not.
1523 bool tracer_tracing_is_on(struct trace_array *tr)
1525 if (tr->array_buffer.buffer)
1526 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1527 return !tr->buffer_disabled;
1531 * tracing_is_on - show state of ring buffers enabled
1533 int tracing_is_on(void)
1535 return tracer_tracing_is_on(&global_trace);
1537 EXPORT_SYMBOL_GPL(tracing_is_on);
1539 static int __init set_buf_size(char *str)
1541 unsigned long buf_size;
1545 buf_size = memparse(str, &str);
1547 * nr_entries can not be zero and the startup
1548 * tests require some buffer space. Therefore
1549 * ensure we have at least 4096 bytes of buffer.
1551 trace_buf_size = max(4096UL, buf_size);
1554 __setup("trace_buf_size=", set_buf_size);
1556 static int __init set_tracing_thresh(char *str)
1558 unsigned long threshold;
1563 ret = kstrtoul(str, 0, &threshold);
1566 tracing_thresh = threshold * 1000;
1569 __setup("tracing_thresh=", set_tracing_thresh);
1571 unsigned long nsecs_to_usecs(unsigned long nsecs)
1573 return nsecs / 1000;
1577 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1578 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1579 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1580 * of strings in the order that the evals (enum) were defined.
1585 /* These must match the bit positions in trace_iterator_flags */
1586 static const char *trace_options[] = {
1594 int in_ns; /* is this clock in nanoseconds? */
1595 } trace_clocks[] = {
1596 { trace_clock_local, "local", 1 },
1597 { trace_clock_global, "global", 1 },
1598 { trace_clock_counter, "counter", 0 },
1599 { trace_clock_jiffies, "uptime", 0 },
1600 { trace_clock, "perf", 1 },
1601 { ktime_get_mono_fast_ns, "mono", 1 },
1602 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1603 { ktime_get_boot_fast_ns, "boot", 1 },
1604 { ktime_get_tai_fast_ns, "tai", 1 },
1608 bool trace_clock_in_ns(struct trace_array *tr)
1610 if (trace_clocks[tr->clock_id].in_ns)
1617 * trace_parser_get_init - gets the buffer for trace parser
1619 int trace_parser_get_init(struct trace_parser *parser, int size)
1621 memset(parser, 0, sizeof(*parser));
1623 parser->buffer = kmalloc(size, GFP_KERNEL);
1624 if (!parser->buffer)
1627 parser->size = size;
1632 * trace_parser_put - frees the buffer for trace parser
1634 void trace_parser_put(struct trace_parser *parser)
1636 kfree(parser->buffer);
1637 parser->buffer = NULL;
1641 * trace_get_user - reads the user input string separated by space
1642 * (matched by isspace(ch))
1644 * For each string found the 'struct trace_parser' is updated,
1645 * and the function returns.
1647 * Returns number of bytes read.
1649 * See kernel/trace/trace.h for 'struct trace_parser' details.
1651 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1652 size_t cnt, loff_t *ppos)
1659 trace_parser_clear(parser);
1661 ret = get_user(ch, ubuf++);
1669 * The parser is not finished with the last write,
1670 * continue reading the user input without skipping spaces.
1672 if (!parser->cont) {
1673 /* skip white space */
1674 while (cnt && isspace(ch)) {
1675 ret = get_user(ch, ubuf++);
1684 /* only spaces were written */
1685 if (isspace(ch) || !ch) {
1692 /* read the non-space input */
1693 while (cnt && !isspace(ch) && ch) {
1694 if (parser->idx < parser->size - 1)
1695 parser->buffer[parser->idx++] = ch;
1700 ret = get_user(ch, ubuf++);
1707 /* We either got finished input or we have to wait for another call. */
1708 if (isspace(ch) || !ch) {
1709 parser->buffer[parser->idx] = 0;
1710 parser->cont = false;
1711 } else if (parser->idx < parser->size - 1) {
1712 parser->cont = true;
1713 parser->buffer[parser->idx++] = ch;
1714 /* Make sure the parsed string always terminates with '\0'. */
1715 parser->buffer[parser->idx] = 0;
1728 /* TODO add a seq_buf_to_buffer() */
1729 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1733 if (trace_seq_used(s) <= s->seq.readpos)
1736 len = trace_seq_used(s) - s->seq.readpos;
1739 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1741 s->seq.readpos += cnt;
1745 unsigned long __read_mostly tracing_thresh;
1747 #ifdef CONFIG_TRACER_MAX_TRACE
1748 static const struct file_operations tracing_max_lat_fops;
1750 #ifdef LATENCY_FS_NOTIFY
1752 static struct workqueue_struct *fsnotify_wq;
1754 static void latency_fsnotify_workfn(struct work_struct *work)
1756 struct trace_array *tr = container_of(work, struct trace_array,
1758 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1761 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1763 struct trace_array *tr = container_of(iwork, struct trace_array,
1765 queue_work(fsnotify_wq, &tr->fsnotify_work);
1768 static void trace_create_maxlat_file(struct trace_array *tr,
1769 struct dentry *d_tracer)
1771 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1772 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1773 tr->d_max_latency = trace_create_file("tracing_max_latency",
1775 d_tracer, &tr->max_latency,
1776 &tracing_max_lat_fops);
1779 __init static int latency_fsnotify_init(void)
1781 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1782 WQ_UNBOUND | WQ_HIGHPRI, 0);
1784 pr_err("Unable to allocate tr_max_lat_wq\n");
1790 late_initcall_sync(latency_fsnotify_init);
1792 void latency_fsnotify(struct trace_array *tr)
1797 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1798 * possible that we are called from __schedule() or do_idle(), which
1799 * could cause a deadlock.
1801 irq_work_queue(&tr->fsnotify_irqwork);
1804 #else /* !LATENCY_FS_NOTIFY */
1806 #define trace_create_maxlat_file(tr, d_tracer) \
1807 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1808 d_tracer, &tr->max_latency, &tracing_max_lat_fops)
1813 * Copy the new maximum trace into the separate maximum-trace
1814 * structure. (this way the maximum trace is permanently saved,
1815 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1818 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1820 struct array_buffer *trace_buf = &tr->array_buffer;
1821 struct array_buffer *max_buf = &tr->max_buffer;
1822 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1823 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1826 max_buf->time_start = data->preempt_timestamp;
1828 max_data->saved_latency = tr->max_latency;
1829 max_data->critical_start = data->critical_start;
1830 max_data->critical_end = data->critical_end;
1832 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1833 max_data->pid = tsk->pid;
1835 * If tsk == current, then use current_uid(), as that does not use
1836 * RCU. The irq tracer can be called out of RCU scope.
1839 max_data->uid = current_uid();
1841 max_data->uid = task_uid(tsk);
1843 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1844 max_data->policy = tsk->policy;
1845 max_data->rt_priority = tsk->rt_priority;
1847 /* record this tasks comm */
1848 tracing_record_cmdline(tsk);
1849 latency_fsnotify(tr);
1853 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1855 * @tsk: the task with the latency
1856 * @cpu: The cpu that initiated the trace.
1857 * @cond_data: User data associated with a conditional snapshot
1859 * Flip the buffers between the @tr and the max_tr and record information
1860 * about which task was the cause of this latency.
1863 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1869 WARN_ON_ONCE(!irqs_disabled());
1871 if (!tr->allocated_snapshot) {
1872 /* Only the nop tracer should hit this when disabling */
1873 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1877 arch_spin_lock(&tr->max_lock);
1879 /* Inherit the recordable setting from array_buffer */
1880 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1881 ring_buffer_record_on(tr->max_buffer.buffer);
1883 ring_buffer_record_off(tr->max_buffer.buffer);
1885 #ifdef CONFIG_TRACER_SNAPSHOT
1886 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1887 arch_spin_unlock(&tr->max_lock);
1891 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1893 __update_max_tr(tr, tsk, cpu);
1895 arch_spin_unlock(&tr->max_lock);
1899 * update_max_tr_single - only copy one trace over, and reset the rest
1901 * @tsk: task with the latency
1902 * @cpu: the cpu of the buffer to copy.
1904 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1907 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1914 WARN_ON_ONCE(!irqs_disabled());
1915 if (!tr->allocated_snapshot) {
1916 /* Only the nop tracer should hit this when disabling */
1917 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1921 arch_spin_lock(&tr->max_lock);
1923 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1925 if (ret == -EBUSY) {
1927 * We failed to swap the buffer due to a commit taking
1928 * place on this CPU. We fail to record, but we reset
1929 * the max trace buffer (no one writes directly to it)
1930 * and flag that it failed.
1931 * Another reason is resize is in progress.
1933 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1934 "Failed to swap buffers due to commit or resize in progress\n");
1937 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1939 __update_max_tr(tr, tsk, cpu);
1940 arch_spin_unlock(&tr->max_lock);
1943 #endif /* CONFIG_TRACER_MAX_TRACE */
1945 static int wait_on_pipe(struct trace_iterator *iter, int full)
1947 /* Iterators are static, they should be filled or empty */
1948 if (trace_buffer_iter(iter, iter->cpu_file))
1951 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1955 #ifdef CONFIG_FTRACE_STARTUP_TEST
1956 static bool selftests_can_run;
1958 struct trace_selftests {
1959 struct list_head list;
1960 struct tracer *type;
1963 static LIST_HEAD(postponed_selftests);
1965 static int save_selftest(struct tracer *type)
1967 struct trace_selftests *selftest;
1969 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1973 selftest->type = type;
1974 list_add(&selftest->list, &postponed_selftests);
1978 static int run_tracer_selftest(struct tracer *type)
1980 struct trace_array *tr = &global_trace;
1981 struct tracer *saved_tracer = tr->current_trace;
1984 if (!type->selftest || tracing_selftest_disabled)
1988 * If a tracer registers early in boot up (before scheduling is
1989 * initialized and such), then do not run its selftests yet.
1990 * Instead, run it a little later in the boot process.
1992 if (!selftests_can_run)
1993 return save_selftest(type);
1995 if (!tracing_is_on()) {
1996 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2002 * Run a selftest on this tracer.
2003 * Here we reset the trace buffer, and set the current
2004 * tracer to be this tracer. The tracer can then run some
2005 * internal tracing to verify that everything is in order.
2006 * If we fail, we do not register this tracer.
2008 tracing_reset_online_cpus(&tr->array_buffer);
2010 tr->current_trace = type;
2012 #ifdef CONFIG_TRACER_MAX_TRACE
2013 if (type->use_max_tr) {
2014 /* If we expanded the buffers, make sure the max is expanded too */
2015 if (ring_buffer_expanded)
2016 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2017 RING_BUFFER_ALL_CPUS);
2018 tr->allocated_snapshot = true;
2022 /* the test is responsible for initializing and enabling */
2023 pr_info("Testing tracer %s: ", type->name);
2024 ret = type->selftest(type, tr);
2025 /* the test is responsible for resetting too */
2026 tr->current_trace = saved_tracer;
2028 printk(KERN_CONT "FAILED!\n");
2029 /* Add the warning after printing 'FAILED' */
2033 /* Only reset on passing, to avoid touching corrupted buffers */
2034 tracing_reset_online_cpus(&tr->array_buffer);
2036 #ifdef CONFIG_TRACER_MAX_TRACE
2037 if (type->use_max_tr) {
2038 tr->allocated_snapshot = false;
2040 /* Shrink the max buffer again */
2041 if (ring_buffer_expanded)
2042 ring_buffer_resize(tr->max_buffer.buffer, 1,
2043 RING_BUFFER_ALL_CPUS);
2047 printk(KERN_CONT "PASSED\n");
2051 static int do_run_tracer_selftest(struct tracer *type)
2056 * Tests can take a long time, especially if they are run one after the
2057 * other, as does happen during bootup when all the tracers are
2058 * registered. This could cause the soft lockup watchdog to trigger.
2062 tracing_selftest_running = true;
2063 ret = run_tracer_selftest(type);
2064 tracing_selftest_running = false;
2069 static __init int init_trace_selftests(void)
2071 struct trace_selftests *p, *n;
2072 struct tracer *t, **last;
2075 selftests_can_run = true;
2077 mutex_lock(&trace_types_lock);
2079 if (list_empty(&postponed_selftests))
2082 pr_info("Running postponed tracer tests:\n");
2084 tracing_selftest_running = true;
2085 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2086 /* This loop can take minutes when sanitizers are enabled, so
2087 * lets make sure we allow RCU processing.
2090 ret = run_tracer_selftest(p->type);
2091 /* If the test fails, then warn and remove from available_tracers */
2093 WARN(1, "tracer: %s failed selftest, disabling\n",
2095 last = &trace_types;
2096 for (t = trace_types; t; t = t->next) {
2107 tracing_selftest_running = false;
2110 mutex_unlock(&trace_types_lock);
2114 core_initcall(init_trace_selftests);
2116 static inline int run_tracer_selftest(struct tracer *type)
2120 static inline int do_run_tracer_selftest(struct tracer *type)
2124 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2126 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2128 static void __init apply_trace_boot_options(void);
2131 * register_tracer - register a tracer with the ftrace system.
2132 * @type: the plugin for the tracer
2134 * Register a new plugin tracer.
2136 int __init register_tracer(struct tracer *type)
2142 pr_info("Tracer must have a name\n");
2146 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2147 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2151 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2152 pr_warn("Can not register tracer %s due to lockdown\n",
2157 mutex_lock(&trace_types_lock);
2159 for (t = trace_types; t; t = t->next) {
2160 if (strcmp(type->name, t->name) == 0) {
2162 pr_info("Tracer %s already registered\n",
2169 if (!type->set_flag)
2170 type->set_flag = &dummy_set_flag;
2172 /*allocate a dummy tracer_flags*/
2173 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2178 type->flags->val = 0;
2179 type->flags->opts = dummy_tracer_opt;
2181 if (!type->flags->opts)
2182 type->flags->opts = dummy_tracer_opt;
2184 /* store the tracer for __set_tracer_option */
2185 type->flags->trace = type;
2187 ret = do_run_tracer_selftest(type);
2191 type->next = trace_types;
2193 add_tracer_options(&global_trace, type);
2196 mutex_unlock(&trace_types_lock);
2198 if (ret || !default_bootup_tracer)
2201 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2204 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2205 /* Do we want this tracer to start on bootup? */
2206 tracing_set_tracer(&global_trace, type->name);
2207 default_bootup_tracer = NULL;
2209 apply_trace_boot_options();
2211 /* disable other selftests, since this will break it. */
2212 disable_tracing_selftest("running a tracer");
2218 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2220 struct trace_buffer *buffer = buf->buffer;
2225 ring_buffer_record_disable(buffer);
2227 /* Make sure all commits have finished */
2229 ring_buffer_reset_cpu(buffer, cpu);
2231 ring_buffer_record_enable(buffer);
2234 void tracing_reset_online_cpus(struct array_buffer *buf)
2236 struct trace_buffer *buffer = buf->buffer;
2241 ring_buffer_record_disable(buffer);
2243 /* Make sure all commits have finished */
2246 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2248 ring_buffer_reset_online_cpus(buffer);
2250 ring_buffer_record_enable(buffer);
2253 /* Must have trace_types_lock held */
2254 void tracing_reset_all_online_cpus_unlocked(void)
2256 struct trace_array *tr;
2258 lockdep_assert_held(&trace_types_lock);
2260 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2261 if (!tr->clear_trace)
2263 tr->clear_trace = false;
2264 tracing_reset_online_cpus(&tr->array_buffer);
2265 #ifdef CONFIG_TRACER_MAX_TRACE
2266 tracing_reset_online_cpus(&tr->max_buffer);
2271 void tracing_reset_all_online_cpus(void)
2273 mutex_lock(&trace_types_lock);
2274 tracing_reset_all_online_cpus_unlocked();
2275 mutex_unlock(&trace_types_lock);
2279 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2280 * is the tgid last observed corresponding to pid=i.
2282 static int *tgid_map;
2284 /* The maximum valid index into tgid_map. */
2285 static size_t tgid_map_max;
2287 #define SAVED_CMDLINES_DEFAULT 128
2288 #define NO_CMDLINE_MAP UINT_MAX
2290 * Preemption must be disabled before acquiring trace_cmdline_lock.
2291 * The various trace_arrays' max_lock must be acquired in a context
2292 * where interrupt is disabled.
2294 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2295 struct saved_cmdlines_buffer {
2296 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2297 unsigned *map_cmdline_to_pid;
2298 unsigned cmdline_num;
2300 char *saved_cmdlines;
2302 static struct saved_cmdlines_buffer *savedcmd;
2304 static inline char *get_saved_cmdlines(int idx)
2306 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2309 static inline void set_cmdline(int idx, const char *cmdline)
2311 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2314 static int allocate_cmdlines_buffer(unsigned int val,
2315 struct saved_cmdlines_buffer *s)
2317 s->map_cmdline_to_pid = kmalloc_array(val,
2318 sizeof(*s->map_cmdline_to_pid),
2320 if (!s->map_cmdline_to_pid)
2323 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2324 if (!s->saved_cmdlines) {
2325 kfree(s->map_cmdline_to_pid);
2330 s->cmdline_num = val;
2331 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2332 sizeof(s->map_pid_to_cmdline));
2333 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2334 val * sizeof(*s->map_cmdline_to_pid));
2339 static int trace_create_savedcmd(void)
2343 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2347 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2357 int is_tracing_stopped(void)
2359 return global_trace.stop_count;
2363 * tracing_start - quick start of the tracer
2365 * If tracing is enabled but was stopped by tracing_stop,
2366 * this will start the tracer back up.
2368 void tracing_start(void)
2370 struct trace_buffer *buffer;
2371 unsigned long flags;
2373 if (tracing_disabled)
2376 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2377 if (--global_trace.stop_count) {
2378 if (global_trace.stop_count < 0) {
2379 /* Someone screwed up their debugging */
2381 global_trace.stop_count = 0;
2386 /* Prevent the buffers from switching */
2387 arch_spin_lock(&global_trace.max_lock);
2389 buffer = global_trace.array_buffer.buffer;
2391 ring_buffer_record_enable(buffer);
2393 #ifdef CONFIG_TRACER_MAX_TRACE
2394 buffer = global_trace.max_buffer.buffer;
2396 ring_buffer_record_enable(buffer);
2399 arch_spin_unlock(&global_trace.max_lock);
2402 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2405 static void tracing_start_tr(struct trace_array *tr)
2407 struct trace_buffer *buffer;
2408 unsigned long flags;
2410 if (tracing_disabled)
2413 /* If global, we need to also start the max tracer */
2414 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2415 return tracing_start();
2417 raw_spin_lock_irqsave(&tr->start_lock, flags);
2419 if (--tr->stop_count) {
2420 if (tr->stop_count < 0) {
2421 /* Someone screwed up their debugging */
2428 buffer = tr->array_buffer.buffer;
2430 ring_buffer_record_enable(buffer);
2433 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2437 * tracing_stop - quick stop of the tracer
2439 * Light weight way to stop tracing. Use in conjunction with
2442 void tracing_stop(void)
2444 struct trace_buffer *buffer;
2445 unsigned long flags;
2447 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2448 if (global_trace.stop_count++)
2451 /* Prevent the buffers from switching */
2452 arch_spin_lock(&global_trace.max_lock);
2454 buffer = global_trace.array_buffer.buffer;
2456 ring_buffer_record_disable(buffer);
2458 #ifdef CONFIG_TRACER_MAX_TRACE
2459 buffer = global_trace.max_buffer.buffer;
2461 ring_buffer_record_disable(buffer);
2464 arch_spin_unlock(&global_trace.max_lock);
2467 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2470 static void tracing_stop_tr(struct trace_array *tr)
2472 struct trace_buffer *buffer;
2473 unsigned long flags;
2475 /* If global, we need to also stop the max tracer */
2476 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2477 return tracing_stop();
2479 raw_spin_lock_irqsave(&tr->start_lock, flags);
2480 if (tr->stop_count++)
2483 buffer = tr->array_buffer.buffer;
2485 ring_buffer_record_disable(buffer);
2488 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2491 static int trace_save_cmdline(struct task_struct *tsk)
2495 /* treat recording of idle task as a success */
2499 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2502 * It's not the end of the world if we don't get
2503 * the lock, but we also don't want to spin
2504 * nor do we want to disable interrupts,
2505 * so if we miss here, then better luck next time.
2507 * This is called within the scheduler and wake up, so interrupts
2508 * had better been disabled and run queue lock been held.
2510 lockdep_assert_preemption_disabled();
2511 if (!arch_spin_trylock(&trace_cmdline_lock))
2514 idx = savedcmd->map_pid_to_cmdline[tpid];
2515 if (idx == NO_CMDLINE_MAP) {
2516 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2518 savedcmd->map_pid_to_cmdline[tpid] = idx;
2519 savedcmd->cmdline_idx = idx;
2522 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2523 set_cmdline(idx, tsk->comm);
2525 arch_spin_unlock(&trace_cmdline_lock);
2530 static void __trace_find_cmdline(int pid, char comm[])
2536 strcpy(comm, "<idle>");
2540 if (WARN_ON_ONCE(pid < 0)) {
2541 strcpy(comm, "<XXX>");
2545 tpid = pid & (PID_MAX_DEFAULT - 1);
2546 map = savedcmd->map_pid_to_cmdline[tpid];
2547 if (map != NO_CMDLINE_MAP) {
2548 tpid = savedcmd->map_cmdline_to_pid[map];
2550 strscpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2554 strcpy(comm, "<...>");
2557 void trace_find_cmdline(int pid, char comm[])
2560 arch_spin_lock(&trace_cmdline_lock);
2562 __trace_find_cmdline(pid, comm);
2564 arch_spin_unlock(&trace_cmdline_lock);
2568 static int *trace_find_tgid_ptr(int pid)
2571 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2572 * if we observe a non-NULL tgid_map then we also observe the correct
2575 int *map = smp_load_acquire(&tgid_map);
2577 if (unlikely(!map || pid > tgid_map_max))
2583 int trace_find_tgid(int pid)
2585 int *ptr = trace_find_tgid_ptr(pid);
2587 return ptr ? *ptr : 0;
2590 static int trace_save_tgid(struct task_struct *tsk)
2594 /* treat recording of idle task as a success */
2598 ptr = trace_find_tgid_ptr(tsk->pid);
2606 static bool tracing_record_taskinfo_skip(int flags)
2608 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2610 if (!__this_cpu_read(trace_taskinfo_save))
2616 * tracing_record_taskinfo - record the task info of a task
2618 * @task: task to record
2619 * @flags: TRACE_RECORD_CMDLINE for recording comm
2620 * TRACE_RECORD_TGID for recording tgid
2622 void tracing_record_taskinfo(struct task_struct *task, int flags)
2626 if (tracing_record_taskinfo_skip(flags))
2630 * Record as much task information as possible. If some fail, continue
2631 * to try to record the others.
2633 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2634 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2636 /* If recording any information failed, retry again soon. */
2640 __this_cpu_write(trace_taskinfo_save, false);
2644 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2646 * @prev: previous task during sched_switch
2647 * @next: next task during sched_switch
2648 * @flags: TRACE_RECORD_CMDLINE for recording comm
2649 * TRACE_RECORD_TGID for recording tgid
2651 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2652 struct task_struct *next, int flags)
2656 if (tracing_record_taskinfo_skip(flags))
2660 * Record as much task information as possible. If some fail, continue
2661 * to try to record the others.
2663 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2664 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2665 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2666 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2668 /* If recording any information failed, retry again soon. */
2672 __this_cpu_write(trace_taskinfo_save, false);
2675 /* Helpers to record a specific task information */
2676 void tracing_record_cmdline(struct task_struct *task)
2678 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2681 void tracing_record_tgid(struct task_struct *task)
2683 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2687 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2688 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2689 * simplifies those functions and keeps them in sync.
2691 enum print_line_t trace_handle_return(struct trace_seq *s)
2693 return trace_seq_has_overflowed(s) ?
2694 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2696 EXPORT_SYMBOL_GPL(trace_handle_return);
2698 static unsigned short migration_disable_value(void)
2700 #if defined(CONFIG_SMP)
2701 return current->migration_disabled;
2707 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2709 unsigned int trace_flags = irqs_status;
2712 pc = preempt_count();
2715 trace_flags |= TRACE_FLAG_NMI;
2716 if (pc & HARDIRQ_MASK)
2717 trace_flags |= TRACE_FLAG_HARDIRQ;
2718 if (in_serving_softirq())
2719 trace_flags |= TRACE_FLAG_SOFTIRQ;
2720 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2721 trace_flags |= TRACE_FLAG_BH_OFF;
2723 if (tif_need_resched())
2724 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2725 if (test_preempt_need_resched())
2726 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2727 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2728 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2731 struct ring_buffer_event *
2732 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2735 unsigned int trace_ctx)
2737 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2740 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2741 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2742 static int trace_buffered_event_ref;
2745 * trace_buffered_event_enable - enable buffering events
2747 * When events are being filtered, it is quicker to use a temporary
2748 * buffer to write the event data into if there's a likely chance
2749 * that it will not be committed. The discard of the ring buffer
2750 * is not as fast as committing, and is much slower than copying
2753 * When an event is to be filtered, allocate per cpu buffers to
2754 * write the event data into, and if the event is filtered and discarded
2755 * it is simply dropped, otherwise, the entire data is to be committed
2758 void trace_buffered_event_enable(void)
2760 struct ring_buffer_event *event;
2764 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2766 if (trace_buffered_event_ref++)
2769 for_each_tracing_cpu(cpu) {
2770 page = alloc_pages_node(cpu_to_node(cpu),
2771 GFP_KERNEL | __GFP_NORETRY, 0);
2775 event = page_address(page);
2776 memset(event, 0, sizeof(*event));
2778 per_cpu(trace_buffered_event, cpu) = event;
2781 if (cpu == smp_processor_id() &&
2782 __this_cpu_read(trace_buffered_event) !=
2783 per_cpu(trace_buffered_event, cpu))
2790 trace_buffered_event_disable();
2793 static void enable_trace_buffered_event(void *data)
2795 /* Probably not needed, but do it anyway */
2797 this_cpu_dec(trace_buffered_event_cnt);
2800 static void disable_trace_buffered_event(void *data)
2802 this_cpu_inc(trace_buffered_event_cnt);
2806 * trace_buffered_event_disable - disable buffering events
2808 * When a filter is removed, it is faster to not use the buffered
2809 * events, and to commit directly into the ring buffer. Free up
2810 * the temp buffers when there are no more users. This requires
2811 * special synchronization with current events.
2813 void trace_buffered_event_disable(void)
2817 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2819 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2822 if (--trace_buffered_event_ref)
2826 /* For each CPU, set the buffer as used. */
2827 smp_call_function_many(tracing_buffer_mask,
2828 disable_trace_buffered_event, NULL, 1);
2831 /* Wait for all current users to finish */
2834 for_each_tracing_cpu(cpu) {
2835 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2836 per_cpu(trace_buffered_event, cpu) = NULL;
2839 * Make sure trace_buffered_event is NULL before clearing
2840 * trace_buffered_event_cnt.
2845 /* Do the work on each cpu */
2846 smp_call_function_many(tracing_buffer_mask,
2847 enable_trace_buffered_event, NULL, 1);
2851 static struct trace_buffer *temp_buffer;
2853 struct ring_buffer_event *
2854 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2855 struct trace_event_file *trace_file,
2856 int type, unsigned long len,
2857 unsigned int trace_ctx)
2859 struct ring_buffer_event *entry;
2860 struct trace_array *tr = trace_file->tr;
2863 *current_rb = tr->array_buffer.buffer;
2865 if (!tr->no_filter_buffering_ref &&
2866 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2867 preempt_disable_notrace();
2869 * Filtering is on, so try to use the per cpu buffer first.
2870 * This buffer will simulate a ring_buffer_event,
2871 * where the type_len is zero and the array[0] will
2872 * hold the full length.
2873 * (see include/linux/ring-buffer.h for details on
2874 * how the ring_buffer_event is structured).
2876 * Using a temp buffer during filtering and copying it
2877 * on a matched filter is quicker than writing directly
2878 * into the ring buffer and then discarding it when
2879 * it doesn't match. That is because the discard
2880 * requires several atomic operations to get right.
2881 * Copying on match and doing nothing on a failed match
2882 * is still quicker than no copy on match, but having
2883 * to discard out of the ring buffer on a failed match.
2885 if ((entry = __this_cpu_read(trace_buffered_event))) {
2886 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2888 val = this_cpu_inc_return(trace_buffered_event_cnt);
2891 * Preemption is disabled, but interrupts and NMIs
2892 * can still come in now. If that happens after
2893 * the above increment, then it will have to go
2894 * back to the old method of allocating the event
2895 * on the ring buffer, and if the filter fails, it
2896 * will have to call ring_buffer_discard_commit()
2899 * Need to also check the unlikely case that the
2900 * length is bigger than the temp buffer size.
2901 * If that happens, then the reserve is pretty much
2902 * guaranteed to fail, as the ring buffer currently
2903 * only allows events less than a page. But that may
2904 * change in the future, so let the ring buffer reserve
2905 * handle the failure in that case.
2907 if (val == 1 && likely(len <= max_len)) {
2908 trace_event_setup(entry, type, trace_ctx);
2909 entry->array[0] = len;
2910 /* Return with preemption disabled */
2913 this_cpu_dec(trace_buffered_event_cnt);
2915 /* __trace_buffer_lock_reserve() disables preemption */
2916 preempt_enable_notrace();
2919 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2922 * If tracing is off, but we have triggers enabled
2923 * we still need to look at the event data. Use the temp_buffer
2924 * to store the trace event for the trigger to use. It's recursive
2925 * safe and will not be recorded anywhere.
2927 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2928 *current_rb = temp_buffer;
2929 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2934 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2936 static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2937 static DEFINE_MUTEX(tracepoint_printk_mutex);
2939 static void output_printk(struct trace_event_buffer *fbuffer)
2941 struct trace_event_call *event_call;
2942 struct trace_event_file *file;
2943 struct trace_event *event;
2944 unsigned long flags;
2945 struct trace_iterator *iter = tracepoint_print_iter;
2947 /* We should never get here if iter is NULL */
2948 if (WARN_ON_ONCE(!iter))
2951 event_call = fbuffer->trace_file->event_call;
2952 if (!event_call || !event_call->event.funcs ||
2953 !event_call->event.funcs->trace)
2956 file = fbuffer->trace_file;
2957 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2958 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2959 !filter_match_preds(file->filter, fbuffer->entry)))
2962 event = &fbuffer->trace_file->event_call->event;
2964 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2965 trace_seq_init(&iter->seq);
2966 iter->ent = fbuffer->entry;
2967 event_call->event.funcs->trace(iter, 0, event);
2968 trace_seq_putc(&iter->seq, 0);
2969 printk("%s", iter->seq.buffer);
2971 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2974 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2975 void *buffer, size_t *lenp,
2978 int save_tracepoint_printk;
2981 mutex_lock(&tracepoint_printk_mutex);
2982 save_tracepoint_printk = tracepoint_printk;
2984 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2987 * This will force exiting early, as tracepoint_printk
2988 * is always zero when tracepoint_printk_iter is not allocated
2990 if (!tracepoint_print_iter)
2991 tracepoint_printk = 0;
2993 if (save_tracepoint_printk == tracepoint_printk)
2996 if (tracepoint_printk)
2997 static_key_enable(&tracepoint_printk_key.key);
2999 static_key_disable(&tracepoint_printk_key.key);
3002 mutex_unlock(&tracepoint_printk_mutex);
3007 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
3009 enum event_trigger_type tt = ETT_NONE;
3010 struct trace_event_file *file = fbuffer->trace_file;
3012 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
3013 fbuffer->entry, &tt))
3016 if (static_key_false(&tracepoint_printk_key.key))
3017 output_printk(fbuffer);
3019 if (static_branch_unlikely(&trace_event_exports_enabled))
3020 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
3022 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
3023 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
3027 event_triggers_post_call(file, tt);
3030 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
3035 * trace_buffer_unlock_commit_regs()
3036 * trace_event_buffer_commit()
3037 * trace_event_raw_event_xxx()
3039 # define STACK_SKIP 3
3041 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
3042 struct trace_buffer *buffer,
3043 struct ring_buffer_event *event,
3044 unsigned int trace_ctx,
3045 struct pt_regs *regs)
3047 __buffer_unlock_commit(buffer, event);
3050 * If regs is not set, then skip the necessary functions.
3051 * Note, we can still get here via blktrace, wakeup tracer
3052 * and mmiotrace, but that's ok if they lose a function or
3053 * two. They are not that meaningful.
3055 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
3056 ftrace_trace_userstack(tr, buffer, trace_ctx);
3060 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
3063 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
3064 struct ring_buffer_event *event)
3066 __buffer_unlock_commit(buffer, event);
3070 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
3071 parent_ip, unsigned int trace_ctx)
3073 struct trace_event_call *call = &event_function;
3074 struct trace_buffer *buffer = tr->array_buffer.buffer;
3075 struct ring_buffer_event *event;
3076 struct ftrace_entry *entry;
3078 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
3082 entry = ring_buffer_event_data(event);
3084 entry->parent_ip = parent_ip;
3086 if (!call_filter_check_discard(call, entry, buffer, event)) {
3087 if (static_branch_unlikely(&trace_function_exports_enabled))
3088 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
3089 __buffer_unlock_commit(buffer, event);
3093 #ifdef CONFIG_STACKTRACE
3095 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
3096 #define FTRACE_KSTACK_NESTING 4
3098 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3100 struct ftrace_stack {
3101 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3105 struct ftrace_stacks {
3106 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3109 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3110 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3112 static void __ftrace_trace_stack(struct trace_buffer *buffer,
3113 unsigned int trace_ctx,
3114 int skip, struct pt_regs *regs)
3116 struct trace_event_call *call = &event_kernel_stack;
3117 struct ring_buffer_event *event;
3118 unsigned int size, nr_entries;
3119 struct ftrace_stack *fstack;
3120 struct stack_entry *entry;
3124 * Add one, for this function and the call to save_stack_trace()
3125 * If regs is set, then these functions will not be in the way.
3127 #ifndef CONFIG_UNWINDER_ORC
3132 preempt_disable_notrace();
3134 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3136 /* This should never happen. If it does, yell once and skip */
3137 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3141 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3142 * interrupt will either see the value pre increment or post
3143 * increment. If the interrupt happens pre increment it will have
3144 * restored the counter when it returns. We just need a barrier to
3145 * keep gcc from moving things around.
3149 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3150 size = ARRAY_SIZE(fstack->calls);
3153 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3156 nr_entries = stack_trace_save(fstack->calls, size, skip);
3159 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3160 struct_size(entry, caller, nr_entries),
3164 entry = ring_buffer_event_data(event);
3166 entry->size = nr_entries;
3167 memcpy(&entry->caller, fstack->calls,
3168 flex_array_size(entry, caller, nr_entries));
3170 if (!call_filter_check_discard(call, entry, buffer, event))
3171 __buffer_unlock_commit(buffer, event);
3174 /* Again, don't let gcc optimize things here */
3176 __this_cpu_dec(ftrace_stack_reserve);
3177 preempt_enable_notrace();
3181 static inline void ftrace_trace_stack(struct trace_array *tr,
3182 struct trace_buffer *buffer,
3183 unsigned int trace_ctx,
3184 int skip, struct pt_regs *regs)
3186 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3189 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3192 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3195 struct trace_buffer *buffer = tr->array_buffer.buffer;
3197 if (rcu_is_watching()) {
3198 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3202 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3206 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3207 * but if the above rcu_is_watching() failed, then the NMI
3208 * triggered someplace critical, and ct_irq_enter() should
3209 * not be called from NMI.
3211 if (unlikely(in_nmi()))
3214 ct_irq_enter_irqson();
3215 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3216 ct_irq_exit_irqson();
3220 * trace_dump_stack - record a stack back trace in the trace buffer
3221 * @skip: Number of functions to skip (helper handlers)
3223 void trace_dump_stack(int skip)
3225 if (tracing_disabled || tracing_selftest_running)
3228 #ifndef CONFIG_UNWINDER_ORC
3229 /* Skip 1 to skip this function. */
3232 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3233 tracing_gen_ctx(), skip, NULL);
3235 EXPORT_SYMBOL_GPL(trace_dump_stack);
3237 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3238 static DEFINE_PER_CPU(int, user_stack_count);
3241 ftrace_trace_userstack(struct trace_array *tr,
3242 struct trace_buffer *buffer, unsigned int trace_ctx)
3244 struct trace_event_call *call = &event_user_stack;
3245 struct ring_buffer_event *event;
3246 struct userstack_entry *entry;
3248 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3252 * NMIs can not handle page faults, even with fix ups.
3253 * The save user stack can (and often does) fault.
3255 if (unlikely(in_nmi()))
3259 * prevent recursion, since the user stack tracing may
3260 * trigger other kernel events.
3263 if (__this_cpu_read(user_stack_count))
3266 __this_cpu_inc(user_stack_count);
3268 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3269 sizeof(*entry), trace_ctx);
3271 goto out_drop_count;
3272 entry = ring_buffer_event_data(event);
3274 entry->tgid = current->tgid;
3275 memset(&entry->caller, 0, sizeof(entry->caller));
3277 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3278 if (!call_filter_check_discard(call, entry, buffer, event))
3279 __buffer_unlock_commit(buffer, event);
3282 __this_cpu_dec(user_stack_count);
3286 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3287 static void ftrace_trace_userstack(struct trace_array *tr,
3288 struct trace_buffer *buffer,
3289 unsigned int trace_ctx)
3292 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3294 #endif /* CONFIG_STACKTRACE */
3297 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3298 unsigned long long delta)
3300 entry->bottom_delta_ts = delta & U32_MAX;
3301 entry->top_delta_ts = (delta >> 32);
3304 void trace_last_func_repeats(struct trace_array *tr,
3305 struct trace_func_repeats *last_info,
3306 unsigned int trace_ctx)
3308 struct trace_buffer *buffer = tr->array_buffer.buffer;
3309 struct func_repeats_entry *entry;
3310 struct ring_buffer_event *event;
3313 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3314 sizeof(*entry), trace_ctx);
3318 delta = ring_buffer_event_time_stamp(buffer, event) -
3319 last_info->ts_last_call;
3321 entry = ring_buffer_event_data(event);
3322 entry->ip = last_info->ip;
3323 entry->parent_ip = last_info->parent_ip;
3324 entry->count = last_info->count;
3325 func_repeats_set_delta_ts(entry, delta);
3327 __buffer_unlock_commit(buffer, event);
3330 /* created for use with alloc_percpu */
3331 struct trace_buffer_struct {
3333 char buffer[4][TRACE_BUF_SIZE];
3336 static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3339 * This allows for lockless recording. If we're nested too deeply, then
3340 * this returns NULL.
3342 static char *get_trace_buf(void)
3344 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3346 if (!trace_percpu_buffer || buffer->nesting >= 4)
3351 /* Interrupts must see nesting incremented before we use the buffer */
3353 return &buffer->buffer[buffer->nesting - 1][0];
3356 static void put_trace_buf(void)
3358 /* Don't let the decrement of nesting leak before this */
3360 this_cpu_dec(trace_percpu_buffer->nesting);
3363 static int alloc_percpu_trace_buffer(void)
3365 struct trace_buffer_struct __percpu *buffers;
3367 if (trace_percpu_buffer)
3370 buffers = alloc_percpu(struct trace_buffer_struct);
3371 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3374 trace_percpu_buffer = buffers;
3378 static int buffers_allocated;
3380 void trace_printk_init_buffers(void)
3382 if (buffers_allocated)
3385 if (alloc_percpu_trace_buffer())
3388 /* trace_printk() is for debug use only. Don't use it in production. */
3391 pr_warn("**********************************************************\n");
3392 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3394 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3396 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3397 pr_warn("** unsafe for production use. **\n");
3399 pr_warn("** If you see this message and you are not debugging **\n");
3400 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3402 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3403 pr_warn("**********************************************************\n");
3405 /* Expand the buffers to set size */
3406 tracing_update_buffers();
3408 buffers_allocated = 1;
3411 * trace_printk_init_buffers() can be called by modules.
3412 * If that happens, then we need to start cmdline recording
3413 * directly here. If the global_trace.buffer is already
3414 * allocated here, then this was called by module code.
3416 if (global_trace.array_buffer.buffer)
3417 tracing_start_cmdline_record();
3419 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3421 void trace_printk_start_comm(void)
3423 /* Start tracing comms if trace printk is set */
3424 if (!buffers_allocated)
3426 tracing_start_cmdline_record();
3429 static void trace_printk_start_stop_comm(int enabled)
3431 if (!buffers_allocated)
3435 tracing_start_cmdline_record();
3437 tracing_stop_cmdline_record();
3441 * trace_vbprintk - write binary msg to tracing buffer
3442 * @ip: The address of the caller
3443 * @fmt: The string format to write to the buffer
3444 * @args: Arguments for @fmt
3446 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3448 struct trace_event_call *call = &event_bprint;
3449 struct ring_buffer_event *event;
3450 struct trace_buffer *buffer;
3451 struct trace_array *tr = &global_trace;
3452 struct bprint_entry *entry;
3453 unsigned int trace_ctx;
3457 if (unlikely(tracing_selftest_running || tracing_disabled))
3460 /* Don't pollute graph traces with trace_vprintk internals */
3461 pause_graph_tracing();
3463 trace_ctx = tracing_gen_ctx();
3464 preempt_disable_notrace();
3466 tbuffer = get_trace_buf();
3472 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3474 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3477 size = sizeof(*entry) + sizeof(u32) * len;
3478 buffer = tr->array_buffer.buffer;
3479 ring_buffer_nest_start(buffer);
3480 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3484 entry = ring_buffer_event_data(event);
3488 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3489 if (!call_filter_check_discard(call, entry, buffer, event)) {
3490 __buffer_unlock_commit(buffer, event);
3491 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3495 ring_buffer_nest_end(buffer);
3500 preempt_enable_notrace();
3501 unpause_graph_tracing();
3505 EXPORT_SYMBOL_GPL(trace_vbprintk);
3509 __trace_array_vprintk(struct trace_buffer *buffer,
3510 unsigned long ip, const char *fmt, va_list args)
3512 struct trace_event_call *call = &event_print;
3513 struct ring_buffer_event *event;
3515 struct print_entry *entry;
3516 unsigned int trace_ctx;
3519 if (tracing_disabled)
3522 /* Don't pollute graph traces with trace_vprintk internals */
3523 pause_graph_tracing();
3525 trace_ctx = tracing_gen_ctx();
3526 preempt_disable_notrace();
3529 tbuffer = get_trace_buf();
3535 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3537 size = sizeof(*entry) + len + 1;
3538 ring_buffer_nest_start(buffer);
3539 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3543 entry = ring_buffer_event_data(event);
3546 memcpy(&entry->buf, tbuffer, len + 1);
3547 if (!call_filter_check_discard(call, entry, buffer, event)) {
3548 __buffer_unlock_commit(buffer, event);
3549 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3553 ring_buffer_nest_end(buffer);
3557 preempt_enable_notrace();
3558 unpause_graph_tracing();
3564 int trace_array_vprintk(struct trace_array *tr,
3565 unsigned long ip, const char *fmt, va_list args)
3567 if (tracing_selftest_running && tr == &global_trace)
3570 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3574 * trace_array_printk - Print a message to a specific instance
3575 * @tr: The instance trace_array descriptor
3576 * @ip: The instruction pointer that this is called from.
3577 * @fmt: The format to print (printf format)
3579 * If a subsystem sets up its own instance, they have the right to
3580 * printk strings into their tracing instance buffer using this
3581 * function. Note, this function will not write into the top level
3582 * buffer (use trace_printk() for that), as writing into the top level
3583 * buffer should only have events that can be individually disabled.
3584 * trace_printk() is only used for debugging a kernel, and should not
3585 * be ever incorporated in normal use.
3587 * trace_array_printk() can be used, as it will not add noise to the
3588 * top level tracing buffer.
3590 * Note, trace_array_init_printk() must be called on @tr before this
3594 int trace_array_printk(struct trace_array *tr,
3595 unsigned long ip, const char *fmt, ...)
3603 /* This is only allowed for created instances */
3604 if (tr == &global_trace)
3607 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3611 ret = trace_array_vprintk(tr, ip, fmt, ap);
3615 EXPORT_SYMBOL_GPL(trace_array_printk);
3618 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3619 * @tr: The trace array to initialize the buffers for
3621 * As trace_array_printk() only writes into instances, they are OK to
3622 * have in the kernel (unlike trace_printk()). This needs to be called
3623 * before trace_array_printk() can be used on a trace_array.
3625 int trace_array_init_printk(struct trace_array *tr)
3630 /* This is only allowed for created instances */
3631 if (tr == &global_trace)
3634 return alloc_percpu_trace_buffer();
3636 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3639 int trace_array_printk_buf(struct trace_buffer *buffer,
3640 unsigned long ip, const char *fmt, ...)
3645 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3649 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3655 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3657 return trace_array_vprintk(&global_trace, ip, fmt, args);
3659 EXPORT_SYMBOL_GPL(trace_vprintk);
3661 static void trace_iterator_increment(struct trace_iterator *iter)
3663 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3667 ring_buffer_iter_advance(buf_iter);
3670 static struct trace_entry *
3671 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3672 unsigned long *lost_events)
3674 struct ring_buffer_event *event;
3675 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3678 event = ring_buffer_iter_peek(buf_iter, ts);
3680 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3681 (unsigned long)-1 : 0;
3683 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3688 iter->ent_size = ring_buffer_event_length(event);
3689 return ring_buffer_event_data(event);
3695 static struct trace_entry *
3696 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3697 unsigned long *missing_events, u64 *ent_ts)
3699 struct trace_buffer *buffer = iter->array_buffer->buffer;
3700 struct trace_entry *ent, *next = NULL;
3701 unsigned long lost_events = 0, next_lost = 0;
3702 int cpu_file = iter->cpu_file;
3703 u64 next_ts = 0, ts;
3709 * If we are in a per_cpu trace file, don't bother by iterating over
3710 * all cpu and peek directly.
3712 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3713 if (ring_buffer_empty_cpu(buffer, cpu_file))
3715 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3717 *ent_cpu = cpu_file;
3722 for_each_tracing_cpu(cpu) {
3724 if (ring_buffer_empty_cpu(buffer, cpu))
3727 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3730 * Pick the entry with the smallest timestamp:
3732 if (ent && (!next || ts < next_ts)) {
3736 next_lost = lost_events;
3737 next_size = iter->ent_size;
3741 iter->ent_size = next_size;
3744 *ent_cpu = next_cpu;
3750 *missing_events = next_lost;
3755 #define STATIC_FMT_BUF_SIZE 128
3756 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3758 char *trace_iter_expand_format(struct trace_iterator *iter)
3763 * iter->tr is NULL when used with tp_printk, which makes
3764 * this get called where it is not safe to call krealloc().
3766 if (!iter->tr || iter->fmt == static_fmt_buf)
3769 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3772 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3779 /* Returns true if the string is safe to dereference from an event */
3780 static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3783 unsigned long addr = (unsigned long)str;
3784 struct trace_event *trace_event;
3785 struct trace_event_call *event;
3787 /* Ignore strings with no length */
3791 /* OK if part of the event data */
3792 if ((addr >= (unsigned long)iter->ent) &&
3793 (addr < (unsigned long)iter->ent + iter->ent_size))
3796 /* OK if part of the temp seq buffer */
3797 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3798 (addr < (unsigned long)iter->tmp_seq.buffer + PAGE_SIZE))
3801 /* Core rodata can not be freed */
3802 if (is_kernel_rodata(addr))
3805 if (trace_is_tracepoint_string(str))
3809 * Now this could be a module event, referencing core module
3810 * data, which is OK.
3815 trace_event = ftrace_find_event(iter->ent->type);
3819 event = container_of(trace_event, struct trace_event_call, event);
3820 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3823 /* Would rather have rodata, but this will suffice */
3824 if (within_module_core(addr, event->module))
3830 static const char *show_buffer(struct trace_seq *s)
3832 struct seq_buf *seq = &s->seq;
3834 seq_buf_terminate(seq);
3839 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3841 static int test_can_verify_check(const char *fmt, ...)
3848 * The verifier is dependent on vsnprintf() modifies the va_list
3849 * passed to it, where it is sent as a reference. Some architectures
3850 * (like x86_32) passes it by value, which means that vsnprintf()
3851 * does not modify the va_list passed to it, and the verifier
3852 * would then need to be able to understand all the values that
3853 * vsnprintf can use. If it is passed by value, then the verifier
3857 vsnprintf(buf, 16, "%d", ap);
3858 ret = va_arg(ap, int);
3864 static void test_can_verify(void)
3866 if (!test_can_verify_check("%d %d", 0, 1)) {
3867 pr_info("trace event string verifier disabled\n");
3868 static_branch_inc(&trace_no_verify);
3873 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3874 * @iter: The iterator that holds the seq buffer and the event being printed
3875 * @fmt: The format used to print the event
3876 * @ap: The va_list holding the data to print from @fmt.
3878 * This writes the data into the @iter->seq buffer using the data from
3879 * @fmt and @ap. If the format has a %s, then the source of the string
3880 * is examined to make sure it is safe to print, otherwise it will
3881 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3884 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3887 const char *p = fmt;
3891 if (WARN_ON_ONCE(!fmt))
3894 if (static_branch_unlikely(&trace_no_verify))
3897 /* Don't bother checking when doing a ftrace_dump() */
3898 if (iter->fmt == static_fmt_buf)
3907 /* We only care about %s and variants */
3908 for (i = 0; p[i]; i++) {
3909 if (i + 1 >= iter->fmt_size) {
3911 * If we can't expand the copy buffer,
3914 if (!trace_iter_expand_format(iter))
3918 if (p[i] == '\\' && p[i+1]) {
3923 /* Need to test cases like %08.*s */
3924 for (j = 1; p[i+j]; j++) {
3925 if (isdigit(p[i+j]) ||
3928 if (p[i+j] == '*') {
3940 /* If no %s found then just print normally */
3944 /* Copy up to the %s, and print that */
3945 strncpy(iter->fmt, p, i);
3946 iter->fmt[i] = '\0';
3947 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3950 * If iter->seq is full, the above call no longer guarantees
3951 * that ap is in sync with fmt processing, and further calls
3952 * to va_arg() can return wrong positional arguments.
3954 * Ensure that ap is no longer used in this case.
3956 if (iter->seq.full) {
3962 len = va_arg(ap, int);
3964 /* The ap now points to the string data of the %s */
3965 str = va_arg(ap, const char *);
3968 * If you hit this warning, it is likely that the
3969 * trace event in question used %s on a string that
3970 * was saved at the time of the event, but may not be
3971 * around when the trace is read. Use __string(),
3972 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3973 * instead. See samples/trace_events/trace-events-sample.h
3976 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3977 "fmt: '%s' current_buffer: '%s'",
3978 fmt, show_buffer(&iter->seq))) {
3981 /* Try to safely read the string */
3983 if (len + 1 > iter->fmt_size)
3984 len = iter->fmt_size - 1;
3987 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3991 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3995 trace_seq_printf(&iter->seq, "(0x%px)", str);
3997 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3999 str = "[UNSAFE-MEMORY]";
4000 strcpy(iter->fmt, "%s");
4002 strncpy(iter->fmt, p + i, j + 1);
4003 iter->fmt[j+1] = '\0';
4006 trace_seq_printf(&iter->seq, iter->fmt, len, str);
4008 trace_seq_printf(&iter->seq, iter->fmt, str);
4014 trace_seq_vprintf(&iter->seq, p, ap);
4017 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
4019 const char *p, *new_fmt;
4022 if (WARN_ON_ONCE(!fmt))
4025 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
4029 new_fmt = q = iter->fmt;
4031 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
4032 if (!trace_iter_expand_format(iter))
4035 q += iter->fmt - new_fmt;
4036 new_fmt = iter->fmt;
4041 /* Replace %p with %px */
4045 } else if (p[0] == 'p' && !isalnum(p[1])) {
4056 #define STATIC_TEMP_BUF_SIZE 128
4057 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
4059 /* Find the next real entry, without updating the iterator itself */
4060 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
4061 int *ent_cpu, u64 *ent_ts)
4063 /* __find_next_entry will reset ent_size */
4064 int ent_size = iter->ent_size;
4065 struct trace_entry *entry;
4068 * If called from ftrace_dump(), then the iter->temp buffer
4069 * will be the static_temp_buf and not created from kmalloc.
4070 * If the entry size is greater than the buffer, we can
4071 * not save it. Just return NULL in that case. This is only
4072 * used to add markers when two consecutive events' time
4073 * stamps have a large delta. See trace_print_lat_context()
4075 if (iter->temp == static_temp_buf &&
4076 STATIC_TEMP_BUF_SIZE < ent_size)
4080 * The __find_next_entry() may call peek_next_entry(), which may
4081 * call ring_buffer_peek() that may make the contents of iter->ent
4082 * undefined. Need to copy iter->ent now.
4084 if (iter->ent && iter->ent != iter->temp) {
4085 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4086 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4088 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4093 iter->temp_size = iter->ent_size;
4095 memcpy(iter->temp, iter->ent, iter->ent_size);
4096 iter->ent = iter->temp;
4098 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4099 /* Put back the original ent_size */
4100 iter->ent_size = ent_size;
4105 /* Find the next real entry, and increment the iterator to the next entry */
4106 void *trace_find_next_entry_inc(struct trace_iterator *iter)
4108 iter->ent = __find_next_entry(iter, &iter->cpu,
4109 &iter->lost_events, &iter->ts);
4112 trace_iterator_increment(iter);
4114 return iter->ent ? iter : NULL;
4117 static void trace_consume(struct trace_iterator *iter)
4119 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4120 &iter->lost_events);
4123 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4125 struct trace_iterator *iter = m->private;
4129 WARN_ON_ONCE(iter->leftover);
4133 /* can't go backwards */
4138 ent = trace_find_next_entry_inc(iter);
4142 while (ent && iter->idx < i)
4143 ent = trace_find_next_entry_inc(iter);
4150 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4152 struct ring_buffer_iter *buf_iter;
4153 unsigned long entries = 0;
4156 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4158 buf_iter = trace_buffer_iter(iter, cpu);
4162 ring_buffer_iter_reset(buf_iter);
4165 * We could have the case with the max latency tracers
4166 * that a reset never took place on a cpu. This is evident
4167 * by the timestamp being before the start of the buffer.
4169 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4170 if (ts >= iter->array_buffer->time_start)
4173 ring_buffer_iter_advance(buf_iter);
4176 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4180 * The current tracer is copied to avoid a global locking
4183 static void *s_start(struct seq_file *m, loff_t *pos)
4185 struct trace_iterator *iter = m->private;
4186 struct trace_array *tr = iter->tr;
4187 int cpu_file = iter->cpu_file;
4192 mutex_lock(&trace_types_lock);
4193 if (unlikely(tr->current_trace != iter->trace)) {
4194 /* Close iter->trace before switching to the new current tracer */
4195 if (iter->trace->close)
4196 iter->trace->close(iter);
4197 iter->trace = tr->current_trace;
4198 /* Reopen the new current tracer */
4199 if (iter->trace->open)
4200 iter->trace->open(iter);
4202 mutex_unlock(&trace_types_lock);
4204 #ifdef CONFIG_TRACER_MAX_TRACE
4205 if (iter->snapshot && iter->trace->use_max_tr)
4206 return ERR_PTR(-EBUSY);
4209 if (*pos != iter->pos) {
4214 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4215 for_each_tracing_cpu(cpu)
4216 tracing_iter_reset(iter, cpu);
4218 tracing_iter_reset(iter, cpu_file);
4221 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4226 * If we overflowed the seq_file before, then we want
4227 * to just reuse the trace_seq buffer again.
4233 p = s_next(m, p, &l);
4237 trace_event_read_lock();
4238 trace_access_lock(cpu_file);
4242 static void s_stop(struct seq_file *m, void *p)
4244 struct trace_iterator *iter = m->private;
4246 #ifdef CONFIG_TRACER_MAX_TRACE
4247 if (iter->snapshot && iter->trace->use_max_tr)
4251 trace_access_unlock(iter->cpu_file);
4252 trace_event_read_unlock();
4256 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4257 unsigned long *entries, int cpu)
4259 unsigned long count;
4261 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4263 * If this buffer has skipped entries, then we hold all
4264 * entries for the trace and we need to ignore the
4265 * ones before the time stamp.
4267 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4268 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4269 /* total is the same as the entries */
4273 ring_buffer_overrun_cpu(buf->buffer, cpu);
4278 get_total_entries(struct array_buffer *buf,
4279 unsigned long *total, unsigned long *entries)
4287 for_each_tracing_cpu(cpu) {
4288 get_total_entries_cpu(buf, &t, &e, cpu);
4294 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4296 unsigned long total, entries;
4301 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4306 unsigned long trace_total_entries(struct trace_array *tr)
4308 unsigned long total, entries;
4313 get_total_entries(&tr->array_buffer, &total, &entries);
4318 static void print_lat_help_header(struct seq_file *m)
4320 seq_puts(m, "# _------=> CPU# \n"
4321 "# / _-----=> irqs-off/BH-disabled\n"
4322 "# | / _----=> need-resched \n"
4323 "# || / _---=> hardirq/softirq \n"
4324 "# ||| / _--=> preempt-depth \n"
4325 "# |||| / _-=> migrate-disable \n"
4326 "# ||||| / delay \n"
4327 "# cmd pid |||||| time | caller \n"
4328 "# \\ / |||||| \\ | / \n");
4331 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4333 unsigned long total;
4334 unsigned long entries;
4336 get_total_entries(buf, &total, &entries);
4337 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4338 entries, total, num_online_cpus());
4342 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4345 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4347 print_event_info(buf, m);
4349 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4350 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4353 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4356 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4357 static const char space[] = " ";
4358 int prec = tgid ? 12 : 2;
4360 print_event_info(buf, m);
4362 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4363 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4364 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4365 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4366 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4367 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4368 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4369 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4373 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4375 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4376 struct array_buffer *buf = iter->array_buffer;
4377 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4378 struct tracer *type = iter->trace;
4379 unsigned long entries;
4380 unsigned long total;
4381 const char *name = type->name;
4383 get_total_entries(buf, &total, &entries);
4385 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4387 seq_puts(m, "# -----------------------------------"
4388 "---------------------------------\n");
4389 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4390 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4391 nsecs_to_usecs(data->saved_latency),
4395 preempt_model_none() ? "server" :
4396 preempt_model_voluntary() ? "desktop" :
4397 preempt_model_full() ? "preempt" :
4398 preempt_model_rt() ? "preempt_rt" :
4400 /* These are reserved for later use */
4403 seq_printf(m, " #P:%d)\n", num_online_cpus());
4407 seq_puts(m, "# -----------------\n");
4408 seq_printf(m, "# | task: %.16s-%d "
4409 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4410 data->comm, data->pid,
4411 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4412 data->policy, data->rt_priority);
4413 seq_puts(m, "# -----------------\n");
4415 if (data->critical_start) {
4416 seq_puts(m, "# => started at: ");
4417 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4418 trace_print_seq(m, &iter->seq);
4419 seq_puts(m, "\n# => ended at: ");
4420 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4421 trace_print_seq(m, &iter->seq);
4422 seq_puts(m, "\n#\n");
4428 static void test_cpu_buff_start(struct trace_iterator *iter)
4430 struct trace_seq *s = &iter->seq;
4431 struct trace_array *tr = iter->tr;
4433 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4436 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4439 if (cpumask_available(iter->started) &&
4440 cpumask_test_cpu(iter->cpu, iter->started))
4443 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4446 if (cpumask_available(iter->started))
4447 cpumask_set_cpu(iter->cpu, iter->started);
4449 /* Don't print started cpu buffer for the first entry of the trace */
4451 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4455 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4457 struct trace_array *tr = iter->tr;
4458 struct trace_seq *s = &iter->seq;
4459 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4460 struct trace_entry *entry;
4461 struct trace_event *event;
4465 test_cpu_buff_start(iter);
4467 event = ftrace_find_event(entry->type);
4469 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4470 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4471 trace_print_lat_context(iter);
4473 trace_print_context(iter);
4476 if (trace_seq_has_overflowed(s))
4477 return TRACE_TYPE_PARTIAL_LINE;
4480 if (tr->trace_flags & TRACE_ITER_FIELDS)
4481 return print_event_fields(iter, event);
4482 return event->funcs->trace(iter, sym_flags, event);
4485 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4487 return trace_handle_return(s);
4490 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4492 struct trace_array *tr = iter->tr;
4493 struct trace_seq *s = &iter->seq;
4494 struct trace_entry *entry;
4495 struct trace_event *event;
4499 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4500 trace_seq_printf(s, "%d %d %llu ",
4501 entry->pid, iter->cpu, iter->ts);
4503 if (trace_seq_has_overflowed(s))
4504 return TRACE_TYPE_PARTIAL_LINE;
4506 event = ftrace_find_event(entry->type);
4508 return event->funcs->raw(iter, 0, event);
4510 trace_seq_printf(s, "%d ?\n", entry->type);
4512 return trace_handle_return(s);
4515 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4517 struct trace_array *tr = iter->tr;
4518 struct trace_seq *s = &iter->seq;
4519 unsigned char newline = '\n';
4520 struct trace_entry *entry;
4521 struct trace_event *event;
4525 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4526 SEQ_PUT_HEX_FIELD(s, entry->pid);
4527 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4528 SEQ_PUT_HEX_FIELD(s, iter->ts);
4529 if (trace_seq_has_overflowed(s))
4530 return TRACE_TYPE_PARTIAL_LINE;
4533 event = ftrace_find_event(entry->type);
4535 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4536 if (ret != TRACE_TYPE_HANDLED)
4540 SEQ_PUT_FIELD(s, newline);
4542 return trace_handle_return(s);
4545 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4547 struct trace_array *tr = iter->tr;
4548 struct trace_seq *s = &iter->seq;
4549 struct trace_entry *entry;
4550 struct trace_event *event;
4554 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4555 SEQ_PUT_FIELD(s, entry->pid);
4556 SEQ_PUT_FIELD(s, iter->cpu);
4557 SEQ_PUT_FIELD(s, iter->ts);
4558 if (trace_seq_has_overflowed(s))
4559 return TRACE_TYPE_PARTIAL_LINE;
4562 event = ftrace_find_event(entry->type);
4563 return event ? event->funcs->binary(iter, 0, event) :
4567 int trace_empty(struct trace_iterator *iter)
4569 struct ring_buffer_iter *buf_iter;
4572 /* If we are looking at one CPU buffer, only check that one */
4573 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4574 cpu = iter->cpu_file;
4575 buf_iter = trace_buffer_iter(iter, cpu);
4577 if (!ring_buffer_iter_empty(buf_iter))
4580 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4586 for_each_tracing_cpu(cpu) {
4587 buf_iter = trace_buffer_iter(iter, cpu);
4589 if (!ring_buffer_iter_empty(buf_iter))
4592 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4600 /* Called with trace_event_read_lock() held. */
4601 enum print_line_t print_trace_line(struct trace_iterator *iter)
4603 struct trace_array *tr = iter->tr;
4604 unsigned long trace_flags = tr->trace_flags;
4605 enum print_line_t ret;
4607 if (iter->lost_events) {
4608 if (iter->lost_events == (unsigned long)-1)
4609 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4612 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4613 iter->cpu, iter->lost_events);
4614 if (trace_seq_has_overflowed(&iter->seq))
4615 return TRACE_TYPE_PARTIAL_LINE;
4618 if (iter->trace && iter->trace->print_line) {
4619 ret = iter->trace->print_line(iter);
4620 if (ret != TRACE_TYPE_UNHANDLED)
4624 if (iter->ent->type == TRACE_BPUTS &&
4625 trace_flags & TRACE_ITER_PRINTK &&
4626 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4627 return trace_print_bputs_msg_only(iter);
4629 if (iter->ent->type == TRACE_BPRINT &&
4630 trace_flags & TRACE_ITER_PRINTK &&
4631 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4632 return trace_print_bprintk_msg_only(iter);
4634 if (iter->ent->type == TRACE_PRINT &&
4635 trace_flags & TRACE_ITER_PRINTK &&
4636 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4637 return trace_print_printk_msg_only(iter);
4639 if (trace_flags & TRACE_ITER_BIN)
4640 return print_bin_fmt(iter);
4642 if (trace_flags & TRACE_ITER_HEX)
4643 return print_hex_fmt(iter);
4645 if (trace_flags & TRACE_ITER_RAW)
4646 return print_raw_fmt(iter);
4648 return print_trace_fmt(iter);
4651 void trace_latency_header(struct seq_file *m)
4653 struct trace_iterator *iter = m->private;
4654 struct trace_array *tr = iter->tr;
4656 /* print nothing if the buffers are empty */
4657 if (trace_empty(iter))
4660 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4661 print_trace_header(m, iter);
4663 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4664 print_lat_help_header(m);
4667 void trace_default_header(struct seq_file *m)
4669 struct trace_iterator *iter = m->private;
4670 struct trace_array *tr = iter->tr;
4671 unsigned long trace_flags = tr->trace_flags;
4673 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4676 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4677 /* print nothing if the buffers are empty */
4678 if (trace_empty(iter))
4680 print_trace_header(m, iter);
4681 if (!(trace_flags & TRACE_ITER_VERBOSE))
4682 print_lat_help_header(m);
4684 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4685 if (trace_flags & TRACE_ITER_IRQ_INFO)
4686 print_func_help_header_irq(iter->array_buffer,
4689 print_func_help_header(iter->array_buffer, m,
4695 static void test_ftrace_alive(struct seq_file *m)
4697 if (!ftrace_is_dead())
4699 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4700 "# MAY BE MISSING FUNCTION EVENTS\n");
4703 #ifdef CONFIG_TRACER_MAX_TRACE
4704 static void show_snapshot_main_help(struct seq_file *m)
4706 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4707 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4708 "# Takes a snapshot of the main buffer.\n"
4709 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4710 "# (Doesn't have to be '2' works with any number that\n"
4711 "# is not a '0' or '1')\n");
4714 static void show_snapshot_percpu_help(struct seq_file *m)
4716 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4717 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4718 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4719 "# Takes a snapshot of the main buffer for this cpu.\n");
4721 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4722 "# Must use main snapshot file to allocate.\n");
4724 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4725 "# (Doesn't have to be '2' works with any number that\n"
4726 "# is not a '0' or '1')\n");
4729 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4731 if (iter->tr->allocated_snapshot)
4732 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4734 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4736 seq_puts(m, "# Snapshot commands:\n");
4737 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4738 show_snapshot_main_help(m);
4740 show_snapshot_percpu_help(m);
4743 /* Should never be called */
4744 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4747 static int s_show(struct seq_file *m, void *v)
4749 struct trace_iterator *iter = v;
4752 if (iter->ent == NULL) {
4754 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4756 test_ftrace_alive(m);
4758 if (iter->snapshot && trace_empty(iter))
4759 print_snapshot_help(m, iter);
4760 else if (iter->trace && iter->trace->print_header)
4761 iter->trace->print_header(m);
4763 trace_default_header(m);
4765 } else if (iter->leftover) {
4767 * If we filled the seq_file buffer earlier, we
4768 * want to just show it now.
4770 ret = trace_print_seq(m, &iter->seq);
4772 /* ret should this time be zero, but you never know */
4773 iter->leftover = ret;
4776 print_trace_line(iter);
4777 ret = trace_print_seq(m, &iter->seq);
4779 * If we overflow the seq_file buffer, then it will
4780 * ask us for this data again at start up.
4782 * ret is 0 if seq_file write succeeded.
4785 iter->leftover = ret;
4792 * Should be used after trace_array_get(), trace_types_lock
4793 * ensures that i_cdev was already initialized.
4795 static inline int tracing_get_cpu(struct inode *inode)
4797 if (inode->i_cdev) /* See trace_create_cpu_file() */
4798 return (long)inode->i_cdev - 1;
4799 return RING_BUFFER_ALL_CPUS;
4802 static const struct seq_operations tracer_seq_ops = {
4810 * Note, as iter itself can be allocated and freed in different
4811 * ways, this function is only used to free its content, and not
4812 * the iterator itself. The only requirement to all the allocations
4813 * is that it must zero all fields (kzalloc), as freeing works with
4814 * ethier allocated content or NULL.
4816 static void free_trace_iter_content(struct trace_iterator *iter)
4818 /* The fmt is either NULL, allocated or points to static_fmt_buf */
4819 if (iter->fmt != static_fmt_buf)
4823 kfree(iter->buffer_iter);
4824 mutex_destroy(&iter->mutex);
4825 free_cpumask_var(iter->started);
4828 static struct trace_iterator *
4829 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4831 struct trace_array *tr = inode->i_private;
4832 struct trace_iterator *iter;
4835 if (tracing_disabled)
4836 return ERR_PTR(-ENODEV);
4838 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4840 return ERR_PTR(-ENOMEM);
4842 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4844 if (!iter->buffer_iter)
4848 * trace_find_next_entry() may need to save off iter->ent.
4849 * It will place it into the iter->temp buffer. As most
4850 * events are less than 128, allocate a buffer of that size.
4851 * If one is greater, then trace_find_next_entry() will
4852 * allocate a new buffer to adjust for the bigger iter->ent.
4853 * It's not critical if it fails to get allocated here.
4855 iter->temp = kmalloc(128, GFP_KERNEL);
4857 iter->temp_size = 128;
4860 * trace_event_printf() may need to modify given format
4861 * string to replace %p with %px so that it shows real address
4862 * instead of hash value. However, that is only for the event
4863 * tracing, other tracer may not need. Defer the allocation
4864 * until it is needed.
4869 mutex_lock(&trace_types_lock);
4870 iter->trace = tr->current_trace;
4872 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4877 #ifdef CONFIG_TRACER_MAX_TRACE
4878 /* Currently only the top directory has a snapshot */
4879 if (tr->current_trace->print_max || snapshot)
4880 iter->array_buffer = &tr->max_buffer;
4883 iter->array_buffer = &tr->array_buffer;
4884 iter->snapshot = snapshot;
4886 iter->cpu_file = tracing_get_cpu(inode);
4887 mutex_init(&iter->mutex);
4889 /* Notify the tracer early; before we stop tracing. */
4890 if (iter->trace->open)
4891 iter->trace->open(iter);
4893 /* Annotate start of buffers if we had overruns */
4894 if (ring_buffer_overruns(iter->array_buffer->buffer))
4895 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4897 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4898 if (trace_clocks[tr->clock_id].in_ns)
4899 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4902 * If pause-on-trace is enabled, then stop the trace while
4903 * dumping, unless this is the "snapshot" file
4905 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4906 tracing_stop_tr(tr);
4908 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4909 for_each_tracing_cpu(cpu) {
4910 iter->buffer_iter[cpu] =
4911 ring_buffer_read_prepare(iter->array_buffer->buffer,
4914 ring_buffer_read_prepare_sync();
4915 for_each_tracing_cpu(cpu) {
4916 ring_buffer_read_start(iter->buffer_iter[cpu]);
4917 tracing_iter_reset(iter, cpu);
4920 cpu = iter->cpu_file;
4921 iter->buffer_iter[cpu] =
4922 ring_buffer_read_prepare(iter->array_buffer->buffer,
4924 ring_buffer_read_prepare_sync();
4925 ring_buffer_read_start(iter->buffer_iter[cpu]);
4926 tracing_iter_reset(iter, cpu);
4929 mutex_unlock(&trace_types_lock);
4934 mutex_unlock(&trace_types_lock);
4935 free_trace_iter_content(iter);
4937 seq_release_private(inode, file);
4938 return ERR_PTR(-ENOMEM);
4941 int tracing_open_generic(struct inode *inode, struct file *filp)
4945 ret = tracing_check_open_get_tr(NULL);
4949 filp->private_data = inode->i_private;
4953 bool tracing_is_disabled(void)
4955 return (tracing_disabled) ? true: false;
4959 * Open and update trace_array ref count.
4960 * Must have the current trace_array passed to it.
4962 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4964 struct trace_array *tr = inode->i_private;
4967 ret = tracing_check_open_get_tr(tr);
4971 filp->private_data = inode->i_private;
4976 static int tracing_mark_open(struct inode *inode, struct file *filp)
4978 stream_open(inode, filp);
4979 return tracing_open_generic_tr(inode, filp);
4982 static int tracing_release(struct inode *inode, struct file *file)
4984 struct trace_array *tr = inode->i_private;
4985 struct seq_file *m = file->private_data;
4986 struct trace_iterator *iter;
4989 if (!(file->f_mode & FMODE_READ)) {
4990 trace_array_put(tr);
4994 /* Writes do not use seq_file */
4996 mutex_lock(&trace_types_lock);
4998 for_each_tracing_cpu(cpu) {
4999 if (iter->buffer_iter[cpu])
5000 ring_buffer_read_finish(iter->buffer_iter[cpu]);
5003 if (iter->trace && iter->trace->close)
5004 iter->trace->close(iter);
5006 if (!iter->snapshot && tr->stop_count)
5007 /* reenable tracing if it was previously enabled */
5008 tracing_start_tr(tr);
5010 __trace_array_put(tr);
5012 mutex_unlock(&trace_types_lock);
5014 free_trace_iter_content(iter);
5015 seq_release_private(inode, file);
5020 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
5022 struct trace_array *tr = inode->i_private;
5024 trace_array_put(tr);
5028 static int tracing_single_release_tr(struct inode *inode, struct file *file)
5030 struct trace_array *tr = inode->i_private;
5032 trace_array_put(tr);
5034 return single_release(inode, file);
5037 static int tracing_open(struct inode *inode, struct file *file)
5039 struct trace_array *tr = inode->i_private;
5040 struct trace_iterator *iter;
5043 ret = tracing_check_open_get_tr(tr);
5047 /* If this file was open for write, then erase contents */
5048 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
5049 int cpu = tracing_get_cpu(inode);
5050 struct array_buffer *trace_buf = &tr->array_buffer;
5052 #ifdef CONFIG_TRACER_MAX_TRACE
5053 if (tr->current_trace->print_max)
5054 trace_buf = &tr->max_buffer;
5057 if (cpu == RING_BUFFER_ALL_CPUS)
5058 tracing_reset_online_cpus(trace_buf);
5060 tracing_reset_cpu(trace_buf, cpu);
5063 if (file->f_mode & FMODE_READ) {
5064 iter = __tracing_open(inode, file, false);
5066 ret = PTR_ERR(iter);
5067 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
5068 iter->iter_flags |= TRACE_FILE_LAT_FMT;
5072 trace_array_put(tr);
5078 * Some tracers are not suitable for instance buffers.
5079 * A tracer is always available for the global array (toplevel)
5080 * or if it explicitly states that it is.
5083 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
5085 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5088 /* Find the next tracer that this trace array may use */
5089 static struct tracer *
5090 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5092 while (t && !trace_ok_for_array(t, tr))
5099 t_next(struct seq_file *m, void *v, loff_t *pos)
5101 struct trace_array *tr = m->private;
5102 struct tracer *t = v;
5107 t = get_tracer_for_array(tr, t->next);
5112 static void *t_start(struct seq_file *m, loff_t *pos)
5114 struct trace_array *tr = m->private;
5118 mutex_lock(&trace_types_lock);
5120 t = get_tracer_for_array(tr, trace_types);
5121 for (; t && l < *pos; t = t_next(m, t, &l))
5127 static void t_stop(struct seq_file *m, void *p)
5129 mutex_unlock(&trace_types_lock);
5132 static int t_show(struct seq_file *m, void *v)
5134 struct tracer *t = v;
5139 seq_puts(m, t->name);
5148 static const struct seq_operations show_traces_seq_ops = {
5155 static int show_traces_open(struct inode *inode, struct file *file)
5157 struct trace_array *tr = inode->i_private;
5161 ret = tracing_check_open_get_tr(tr);
5165 ret = seq_open(file, &show_traces_seq_ops);
5167 trace_array_put(tr);
5171 m = file->private_data;
5177 static int show_traces_release(struct inode *inode, struct file *file)
5179 struct trace_array *tr = inode->i_private;
5181 trace_array_put(tr);
5182 return seq_release(inode, file);
5186 tracing_write_stub(struct file *filp, const char __user *ubuf,
5187 size_t count, loff_t *ppos)
5192 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5196 if (file->f_mode & FMODE_READ)
5197 ret = seq_lseek(file, offset, whence);
5199 file->f_pos = ret = 0;
5204 static const struct file_operations tracing_fops = {
5205 .open = tracing_open,
5207 .read_iter = seq_read_iter,
5208 .splice_read = copy_splice_read,
5209 .write = tracing_write_stub,
5210 .llseek = tracing_lseek,
5211 .release = tracing_release,
5214 static const struct file_operations show_traces_fops = {
5215 .open = show_traces_open,
5217 .llseek = seq_lseek,
5218 .release = show_traces_release,
5222 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5223 size_t count, loff_t *ppos)
5225 struct trace_array *tr = file_inode(filp)->i_private;
5229 len = snprintf(NULL, 0, "%*pb\n",
5230 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5231 mask_str = kmalloc(len, GFP_KERNEL);
5235 len = snprintf(mask_str, len, "%*pb\n",
5236 cpumask_pr_args(tr->tracing_cpumask));
5241 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5249 int tracing_set_cpumask(struct trace_array *tr,
5250 cpumask_var_t tracing_cpumask_new)
5257 local_irq_disable();
5258 arch_spin_lock(&tr->max_lock);
5259 for_each_tracing_cpu(cpu) {
5261 * Increase/decrease the disabled counter if we are
5262 * about to flip a bit in the cpumask:
5264 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5265 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5266 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5267 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5268 #ifdef CONFIG_TRACER_MAX_TRACE
5269 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5272 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5273 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5274 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5275 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5276 #ifdef CONFIG_TRACER_MAX_TRACE
5277 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5281 arch_spin_unlock(&tr->max_lock);
5284 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5290 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5291 size_t count, loff_t *ppos)
5293 struct trace_array *tr = file_inode(filp)->i_private;
5294 cpumask_var_t tracing_cpumask_new;
5297 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5300 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5304 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5308 free_cpumask_var(tracing_cpumask_new);
5313 free_cpumask_var(tracing_cpumask_new);
5318 static const struct file_operations tracing_cpumask_fops = {
5319 .open = tracing_open_generic_tr,
5320 .read = tracing_cpumask_read,
5321 .write = tracing_cpumask_write,
5322 .release = tracing_release_generic_tr,
5323 .llseek = generic_file_llseek,
5326 static int tracing_trace_options_show(struct seq_file *m, void *v)
5328 struct tracer_opt *trace_opts;
5329 struct trace_array *tr = m->private;
5333 mutex_lock(&trace_types_lock);
5334 tracer_flags = tr->current_trace->flags->val;
5335 trace_opts = tr->current_trace->flags->opts;
5337 for (i = 0; trace_options[i]; i++) {
5338 if (tr->trace_flags & (1 << i))
5339 seq_printf(m, "%s\n", trace_options[i]);
5341 seq_printf(m, "no%s\n", trace_options[i]);
5344 for (i = 0; trace_opts[i].name; i++) {
5345 if (tracer_flags & trace_opts[i].bit)
5346 seq_printf(m, "%s\n", trace_opts[i].name);
5348 seq_printf(m, "no%s\n", trace_opts[i].name);
5350 mutex_unlock(&trace_types_lock);
5355 static int __set_tracer_option(struct trace_array *tr,
5356 struct tracer_flags *tracer_flags,
5357 struct tracer_opt *opts, int neg)
5359 struct tracer *trace = tracer_flags->trace;
5362 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5367 tracer_flags->val &= ~opts->bit;
5369 tracer_flags->val |= opts->bit;
5373 /* Try to assign a tracer specific option */
5374 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5376 struct tracer *trace = tr->current_trace;
5377 struct tracer_flags *tracer_flags = trace->flags;
5378 struct tracer_opt *opts = NULL;
5381 for (i = 0; tracer_flags->opts[i].name; i++) {
5382 opts = &tracer_flags->opts[i];
5384 if (strcmp(cmp, opts->name) == 0)
5385 return __set_tracer_option(tr, trace->flags, opts, neg);
5391 /* Some tracers require overwrite to stay enabled */
5392 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5394 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5400 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5404 if ((mask == TRACE_ITER_RECORD_TGID) ||
5405 (mask == TRACE_ITER_RECORD_CMD))
5406 lockdep_assert_held(&event_mutex);
5408 /* do nothing if flag is already set */
5409 if (!!(tr->trace_flags & mask) == !!enabled)
5412 /* Give the tracer a chance to approve the change */
5413 if (tr->current_trace->flag_changed)
5414 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5418 tr->trace_flags |= mask;
5420 tr->trace_flags &= ~mask;
5422 if (mask == TRACE_ITER_RECORD_CMD)
5423 trace_event_enable_cmd_record(enabled);
5425 if (mask == TRACE_ITER_RECORD_TGID) {
5427 tgid_map_max = pid_max;
5428 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5432 * Pairs with smp_load_acquire() in
5433 * trace_find_tgid_ptr() to ensure that if it observes
5434 * the tgid_map we just allocated then it also observes
5435 * the corresponding tgid_map_max value.
5437 smp_store_release(&tgid_map, map);
5440 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5444 trace_event_enable_tgid_record(enabled);
5447 if (mask == TRACE_ITER_EVENT_FORK)
5448 trace_event_follow_fork(tr, enabled);
5450 if (mask == TRACE_ITER_FUNC_FORK)
5451 ftrace_pid_follow_fork(tr, enabled);
5453 if (mask == TRACE_ITER_OVERWRITE) {
5454 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5455 #ifdef CONFIG_TRACER_MAX_TRACE
5456 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5460 if (mask == TRACE_ITER_PRINTK) {
5461 trace_printk_start_stop_comm(enabled);
5462 trace_printk_control(enabled);
5468 int trace_set_options(struct trace_array *tr, char *option)
5473 size_t orig_len = strlen(option);
5476 cmp = strstrip(option);
5478 len = str_has_prefix(cmp, "no");
5484 mutex_lock(&event_mutex);
5485 mutex_lock(&trace_types_lock);
5487 ret = match_string(trace_options, -1, cmp);
5488 /* If no option could be set, test the specific tracer options */
5490 ret = set_tracer_option(tr, cmp, neg);
5492 ret = set_tracer_flag(tr, 1 << ret, !neg);
5494 mutex_unlock(&trace_types_lock);
5495 mutex_unlock(&event_mutex);
5498 * If the first trailing whitespace is replaced with '\0' by strstrip,
5499 * turn it back into a space.
5501 if (orig_len > strlen(option))
5502 option[strlen(option)] = ' ';
5507 static void __init apply_trace_boot_options(void)
5509 char *buf = trace_boot_options_buf;
5513 option = strsep(&buf, ",");
5519 trace_set_options(&global_trace, option);
5521 /* Put back the comma to allow this to be called again */
5528 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5529 size_t cnt, loff_t *ppos)
5531 struct seq_file *m = filp->private_data;
5532 struct trace_array *tr = m->private;
5536 if (cnt >= sizeof(buf))
5539 if (copy_from_user(buf, ubuf, cnt))
5544 ret = trace_set_options(tr, buf);
5553 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5555 struct trace_array *tr = inode->i_private;
5558 ret = tracing_check_open_get_tr(tr);
5562 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5564 trace_array_put(tr);
5569 static const struct file_operations tracing_iter_fops = {
5570 .open = tracing_trace_options_open,
5572 .llseek = seq_lseek,
5573 .release = tracing_single_release_tr,
5574 .write = tracing_trace_options_write,
5577 static const char readme_msg[] =
5578 "tracing mini-HOWTO:\n\n"
5579 "# echo 0 > tracing_on : quick way to disable tracing\n"
5580 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5581 " Important files:\n"
5582 " trace\t\t\t- The static contents of the buffer\n"
5583 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5584 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5585 " current_tracer\t- function and latency tracers\n"
5586 " available_tracers\t- list of configured tracers for current_tracer\n"
5587 " error_log\t- error log for failed commands (that support it)\n"
5588 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5589 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5590 " trace_clock\t\t- change the clock used to order events\n"
5591 " local: Per cpu clock but may not be synced across CPUs\n"
5592 " global: Synced across CPUs but slows tracing down.\n"
5593 " counter: Not a clock, but just an increment\n"
5594 " uptime: Jiffy counter from time of boot\n"
5595 " perf: Same clock that perf events use\n"
5596 #ifdef CONFIG_X86_64
5597 " x86-tsc: TSC cycle counter\n"
5599 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5600 " delta: Delta difference against a buffer-wide timestamp\n"
5601 " absolute: Absolute (standalone) timestamp\n"
5602 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5603 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5604 " tracing_cpumask\t- Limit which CPUs to trace\n"
5605 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5606 "\t\t\t Remove sub-buffer with rmdir\n"
5607 " trace_options\t\t- Set format or modify how tracing happens\n"
5608 "\t\t\t Disable an option by prefixing 'no' to the\n"
5609 "\t\t\t option name\n"
5610 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5611 #ifdef CONFIG_DYNAMIC_FTRACE
5612 "\n available_filter_functions - list of functions that can be filtered on\n"
5613 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5614 "\t\t\t functions\n"
5615 "\t accepts: func_full_name or glob-matching-pattern\n"
5616 "\t modules: Can select a group via module\n"
5617 "\t Format: :mod:<module-name>\n"
5618 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5619 "\t triggers: a command to perform when function is hit\n"
5620 "\t Format: <function>:<trigger>[:count]\n"
5621 "\t trigger: traceon, traceoff\n"
5622 "\t\t enable_event:<system>:<event>\n"
5623 "\t\t disable_event:<system>:<event>\n"
5624 #ifdef CONFIG_STACKTRACE
5627 #ifdef CONFIG_TRACER_SNAPSHOT
5632 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5633 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5634 "\t The first one will disable tracing every time do_fault is hit\n"
5635 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5636 "\t The first time do trap is hit and it disables tracing, the\n"
5637 "\t counter will decrement to 2. If tracing is already disabled,\n"
5638 "\t the counter will not decrement. It only decrements when the\n"
5639 "\t trigger did work\n"
5640 "\t To remove trigger without count:\n"
5641 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5642 "\t To remove trigger with a count:\n"
5643 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5644 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5645 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5646 "\t modules: Can select a group via module command :mod:\n"
5647 "\t Does not accept triggers\n"
5648 #endif /* CONFIG_DYNAMIC_FTRACE */
5649 #ifdef CONFIG_FUNCTION_TRACER
5650 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5652 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5655 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5656 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5657 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5658 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5660 #ifdef CONFIG_TRACER_SNAPSHOT
5661 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5662 "\t\t\t snapshot buffer. Read the contents for more\n"
5663 "\t\t\t information\n"
5665 #ifdef CONFIG_STACK_TRACER
5666 " stack_trace\t\t- Shows the max stack trace when active\n"
5667 " stack_max_size\t- Shows current max stack size that was traced\n"
5668 "\t\t\t Write into this file to reset the max size (trigger a\n"
5669 "\t\t\t new trace)\n"
5670 #ifdef CONFIG_DYNAMIC_FTRACE
5671 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5674 #endif /* CONFIG_STACK_TRACER */
5675 #ifdef CONFIG_DYNAMIC_EVENTS
5676 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5677 "\t\t\t Write into this file to define/undefine new trace events.\n"
5679 #ifdef CONFIG_KPROBE_EVENTS
5680 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5681 "\t\t\t Write into this file to define/undefine new trace events.\n"
5683 #ifdef CONFIG_UPROBE_EVENTS
5684 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5685 "\t\t\t Write into this file to define/undefine new trace events.\n"
5687 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5688 defined(CONFIG_FPROBE_EVENTS)
5689 "\t accepts: event-definitions (one definition per line)\n"
5690 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5691 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5692 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5694 #ifdef CONFIG_FPROBE_EVENTS
5695 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5696 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5698 #ifdef CONFIG_HIST_TRIGGERS
5699 "\t s:[synthetic/]<event> <field> [<field>]\n"
5701 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5702 "\t -:[<group>/][<event>]\n"
5703 #ifdef CONFIG_KPROBE_EVENTS
5704 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5705 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5707 #ifdef CONFIG_UPROBE_EVENTS
5708 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5710 "\t args: <name>=fetcharg[:type]\n"
5711 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5712 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5713 #ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5714 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5715 "\t <argname>[->field[->field|.field...]],\n"
5717 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5720 "\t $stack<index>, $stack, $retval, $comm,\n"
5722 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5723 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5724 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5725 "\t symstr, <type>\\[<array-size>\\]\n"
5726 #ifdef CONFIG_HIST_TRIGGERS
5727 "\t field: <stype> <name>;\n"
5728 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5729 "\t [unsigned] char/int/long\n"
5731 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5732 "\t of the <attached-group>/<attached-event>.\n"
5734 " events/\t\t- Directory containing all trace event subsystems:\n"
5735 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5736 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5737 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5739 " filter\t\t- If set, only events passing filter are traced\n"
5740 " events/<system>/<event>/\t- Directory containing control files for\n"
5742 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5743 " filter\t\t- If set, only events passing filter are traced\n"
5744 " trigger\t\t- If set, a command to perform when event is hit\n"
5745 "\t Format: <trigger>[:count][if <filter>]\n"
5746 "\t trigger: traceon, traceoff\n"
5747 "\t enable_event:<system>:<event>\n"
5748 "\t disable_event:<system>:<event>\n"
5749 #ifdef CONFIG_HIST_TRIGGERS
5750 "\t enable_hist:<system>:<event>\n"
5751 "\t disable_hist:<system>:<event>\n"
5753 #ifdef CONFIG_STACKTRACE
5756 #ifdef CONFIG_TRACER_SNAPSHOT
5759 #ifdef CONFIG_HIST_TRIGGERS
5760 "\t\t hist (see below)\n"
5762 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5763 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5764 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5765 "\t events/block/block_unplug/trigger\n"
5766 "\t The first disables tracing every time block_unplug is hit.\n"
5767 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5768 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5769 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5770 "\t Like function triggers, the counter is only decremented if it\n"
5771 "\t enabled or disabled tracing.\n"
5772 "\t To remove a trigger without a count:\n"
5773 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5774 "\t To remove a trigger with a count:\n"
5775 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5776 "\t Filters can be ignored when removing a trigger.\n"
5777 #ifdef CONFIG_HIST_TRIGGERS
5778 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5779 "\t Format: hist:keys=<field1[,field2,...]>\n"
5780 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5781 "\t [:values=<field1[,field2,...]>]\n"
5782 "\t [:sort=<field1[,field2,...]>]\n"
5783 "\t [:size=#entries]\n"
5784 "\t [:pause][:continue][:clear]\n"
5785 "\t [:name=histname1]\n"
5786 "\t [:nohitcount]\n"
5787 "\t [:<handler>.<action>]\n"
5788 "\t [if <filter>]\n\n"
5789 "\t Note, special fields can be used as well:\n"
5790 "\t common_timestamp - to record current timestamp\n"
5791 "\t common_cpu - to record the CPU the event happened on\n"
5793 "\t A hist trigger variable can be:\n"
5794 "\t - a reference to a field e.g. x=current_timestamp,\n"
5795 "\t - a reference to another variable e.g. y=$x,\n"
5796 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5797 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5799 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5800 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5801 "\t variable reference, field or numeric literal.\n"
5803 "\t When a matching event is hit, an entry is added to a hash\n"
5804 "\t table using the key(s) and value(s) named, and the value of a\n"
5805 "\t sum called 'hitcount' is incremented. Keys and values\n"
5806 "\t correspond to fields in the event's format description. Keys\n"
5807 "\t can be any field, or the special string 'common_stacktrace'.\n"
5808 "\t Compound keys consisting of up to two fields can be specified\n"
5809 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5810 "\t fields. Sort keys consisting of up to two fields can be\n"
5811 "\t specified using the 'sort' keyword. The sort direction can\n"
5812 "\t be modified by appending '.descending' or '.ascending' to a\n"
5813 "\t sort field. The 'size' parameter can be used to specify more\n"
5814 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5815 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5816 "\t its histogram data will be shared with other triggers of the\n"
5817 "\t same name, and trigger hits will update this common data.\n\n"
5818 "\t Reading the 'hist' file for the event will dump the hash\n"
5819 "\t table in its entirety to stdout. If there are multiple hist\n"
5820 "\t triggers attached to an event, there will be a table for each\n"
5821 "\t trigger in the output. The table displayed for a named\n"
5822 "\t trigger will be the same as any other instance having the\n"
5823 "\t same name. The default format used to display a given field\n"
5824 "\t can be modified by appending any of the following modifiers\n"
5825 "\t to the field name, as applicable:\n\n"
5826 "\t .hex display a number as a hex value\n"
5827 "\t .sym display an address as a symbol\n"
5828 "\t .sym-offset display an address as a symbol and offset\n"
5829 "\t .execname display a common_pid as a program name\n"
5830 "\t .syscall display a syscall id as a syscall name\n"
5831 "\t .log2 display log2 value rather than raw number\n"
5832 "\t .buckets=size display values in groups of size rather than raw number\n"
5833 "\t .usecs display a common_timestamp in microseconds\n"
5834 "\t .percent display a number of percentage value\n"
5835 "\t .graph display a bar-graph of a value\n\n"
5836 "\t The 'pause' parameter can be used to pause an existing hist\n"
5837 "\t trigger or to start a hist trigger but not log any events\n"
5838 "\t until told to do so. 'continue' can be used to start or\n"
5839 "\t restart a paused hist trigger.\n\n"
5840 "\t The 'clear' parameter will clear the contents of a running\n"
5841 "\t hist trigger and leave its current paused/active state\n"
5843 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5844 "\t raw hitcount in the histogram.\n\n"
5845 "\t The enable_hist and disable_hist triggers can be used to\n"
5846 "\t have one event conditionally start and stop another event's\n"
5847 "\t already-attached hist trigger. The syntax is analogous to\n"
5848 "\t the enable_event and disable_event triggers.\n\n"
5849 "\t Hist trigger handlers and actions are executed whenever a\n"
5850 "\t a histogram entry is added or updated. They take the form:\n\n"
5851 "\t <handler>.<action>\n\n"
5852 "\t The available handlers are:\n\n"
5853 "\t onmatch(matching.event) - invoke on addition or update\n"
5854 "\t onmax(var) - invoke if var exceeds current max\n"
5855 "\t onchange(var) - invoke action if var changes\n\n"
5856 "\t The available actions are:\n\n"
5857 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5858 "\t save(field,...) - save current event fields\n"
5859 #ifdef CONFIG_TRACER_SNAPSHOT
5860 "\t snapshot() - snapshot the trace buffer\n\n"
5862 #ifdef CONFIG_SYNTH_EVENTS
5863 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5864 "\t Write into this file to define/undefine new synthetic events.\n"
5865 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5871 tracing_readme_read(struct file *filp, char __user *ubuf,
5872 size_t cnt, loff_t *ppos)
5874 return simple_read_from_buffer(ubuf, cnt, ppos,
5875 readme_msg, strlen(readme_msg));
5878 static const struct file_operations tracing_readme_fops = {
5879 .open = tracing_open_generic,
5880 .read = tracing_readme_read,
5881 .llseek = generic_file_llseek,
5884 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5888 return trace_find_tgid_ptr(pid);
5891 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5895 return trace_find_tgid_ptr(pid);
5898 static void saved_tgids_stop(struct seq_file *m, void *v)
5902 static int saved_tgids_show(struct seq_file *m, void *v)
5904 int *entry = (int *)v;
5905 int pid = entry - tgid_map;
5911 seq_printf(m, "%d %d\n", pid, tgid);
5915 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5916 .start = saved_tgids_start,
5917 .stop = saved_tgids_stop,
5918 .next = saved_tgids_next,
5919 .show = saved_tgids_show,
5922 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5926 ret = tracing_check_open_get_tr(NULL);
5930 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5934 static const struct file_operations tracing_saved_tgids_fops = {
5935 .open = tracing_saved_tgids_open,
5937 .llseek = seq_lseek,
5938 .release = seq_release,
5941 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5943 unsigned int *ptr = v;
5945 if (*pos || m->count)
5950 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5952 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5961 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5967 arch_spin_lock(&trace_cmdline_lock);
5969 v = &savedcmd->map_cmdline_to_pid[0];
5971 v = saved_cmdlines_next(m, v, &l);
5979 static void saved_cmdlines_stop(struct seq_file *m, void *v)
5981 arch_spin_unlock(&trace_cmdline_lock);
5985 static int saved_cmdlines_show(struct seq_file *m, void *v)
5987 char buf[TASK_COMM_LEN];
5988 unsigned int *pid = v;
5990 __trace_find_cmdline(*pid, buf);
5991 seq_printf(m, "%d %s\n", *pid, buf);
5995 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5996 .start = saved_cmdlines_start,
5997 .next = saved_cmdlines_next,
5998 .stop = saved_cmdlines_stop,
5999 .show = saved_cmdlines_show,
6002 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
6006 ret = tracing_check_open_get_tr(NULL);
6010 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
6013 static const struct file_operations tracing_saved_cmdlines_fops = {
6014 .open = tracing_saved_cmdlines_open,
6016 .llseek = seq_lseek,
6017 .release = seq_release,
6021 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
6022 size_t cnt, loff_t *ppos)
6028 arch_spin_lock(&trace_cmdline_lock);
6029 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
6030 arch_spin_unlock(&trace_cmdline_lock);
6033 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6036 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
6038 kfree(s->saved_cmdlines);
6039 kfree(s->map_cmdline_to_pid);
6043 static int tracing_resize_saved_cmdlines(unsigned int val)
6045 struct saved_cmdlines_buffer *s, *savedcmd_temp;
6047 s = kmalloc(sizeof(*s), GFP_KERNEL);
6051 if (allocate_cmdlines_buffer(val, s) < 0) {
6057 arch_spin_lock(&trace_cmdline_lock);
6058 savedcmd_temp = savedcmd;
6060 arch_spin_unlock(&trace_cmdline_lock);
6062 free_saved_cmdlines_buffer(savedcmd_temp);
6068 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
6069 size_t cnt, loff_t *ppos)
6074 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6078 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
6079 if (!val || val > PID_MAX_DEFAULT)
6082 ret = tracing_resize_saved_cmdlines((unsigned int)val);
6091 static const struct file_operations tracing_saved_cmdlines_size_fops = {
6092 .open = tracing_open_generic,
6093 .read = tracing_saved_cmdlines_size_read,
6094 .write = tracing_saved_cmdlines_size_write,
6097 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
6098 static union trace_eval_map_item *
6099 update_eval_map(union trace_eval_map_item *ptr)
6101 if (!ptr->map.eval_string) {
6102 if (ptr->tail.next) {
6103 ptr = ptr->tail.next;
6104 /* Set ptr to the next real item (skip head) */
6112 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6114 union trace_eval_map_item *ptr = v;
6117 * Paranoid! If ptr points to end, we don't want to increment past it.
6118 * This really should never happen.
6121 ptr = update_eval_map(ptr);
6122 if (WARN_ON_ONCE(!ptr))
6126 ptr = update_eval_map(ptr);
6131 static void *eval_map_start(struct seq_file *m, loff_t *pos)
6133 union trace_eval_map_item *v;
6136 mutex_lock(&trace_eval_mutex);
6138 v = trace_eval_maps;
6142 while (v && l < *pos) {
6143 v = eval_map_next(m, v, &l);
6149 static void eval_map_stop(struct seq_file *m, void *v)
6151 mutex_unlock(&trace_eval_mutex);
6154 static int eval_map_show(struct seq_file *m, void *v)
6156 union trace_eval_map_item *ptr = v;
6158 seq_printf(m, "%s %ld (%s)\n",
6159 ptr->map.eval_string, ptr->map.eval_value,
6165 static const struct seq_operations tracing_eval_map_seq_ops = {
6166 .start = eval_map_start,
6167 .next = eval_map_next,
6168 .stop = eval_map_stop,
6169 .show = eval_map_show,
6172 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6176 ret = tracing_check_open_get_tr(NULL);
6180 return seq_open(filp, &tracing_eval_map_seq_ops);
6183 static const struct file_operations tracing_eval_map_fops = {
6184 .open = tracing_eval_map_open,
6186 .llseek = seq_lseek,
6187 .release = seq_release,
6190 static inline union trace_eval_map_item *
6191 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6193 /* Return tail of array given the head */
6194 return ptr + ptr->head.length + 1;
6198 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6201 struct trace_eval_map **stop;
6202 struct trace_eval_map **map;
6203 union trace_eval_map_item *map_array;
6204 union trace_eval_map_item *ptr;
6209 * The trace_eval_maps contains the map plus a head and tail item,
6210 * where the head holds the module and length of array, and the
6211 * tail holds a pointer to the next list.
6213 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6215 pr_warn("Unable to allocate trace eval mapping\n");
6219 mutex_lock(&trace_eval_mutex);
6221 if (!trace_eval_maps)
6222 trace_eval_maps = map_array;
6224 ptr = trace_eval_maps;
6226 ptr = trace_eval_jmp_to_tail(ptr);
6227 if (!ptr->tail.next)
6229 ptr = ptr->tail.next;
6232 ptr->tail.next = map_array;
6234 map_array->head.mod = mod;
6235 map_array->head.length = len;
6238 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6239 map_array->map = **map;
6242 memset(map_array, 0, sizeof(*map_array));
6244 mutex_unlock(&trace_eval_mutex);
6247 static void trace_create_eval_file(struct dentry *d_tracer)
6249 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6250 NULL, &tracing_eval_map_fops);
6253 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
6254 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6255 static inline void trace_insert_eval_map_file(struct module *mod,
6256 struct trace_eval_map **start, int len) { }
6257 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6259 static void trace_insert_eval_map(struct module *mod,
6260 struct trace_eval_map **start, int len)
6262 struct trace_eval_map **map;
6269 trace_event_eval_update(map, len);
6271 trace_insert_eval_map_file(mod, start, len);
6275 tracing_set_trace_read(struct file *filp, char __user *ubuf,
6276 size_t cnt, loff_t *ppos)
6278 struct trace_array *tr = filp->private_data;
6279 char buf[MAX_TRACER_SIZE+2];
6282 mutex_lock(&trace_types_lock);
6283 r = sprintf(buf, "%s\n", tr->current_trace->name);
6284 mutex_unlock(&trace_types_lock);
6286 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6289 int tracer_init(struct tracer *t, struct trace_array *tr)
6291 tracing_reset_online_cpus(&tr->array_buffer);
6295 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6299 for_each_tracing_cpu(cpu)
6300 per_cpu_ptr(buf->data, cpu)->entries = val;
6303 static void update_buffer_entries(struct array_buffer *buf, int cpu)
6305 if (cpu == RING_BUFFER_ALL_CPUS) {
6306 set_buffer_entries(buf, ring_buffer_size(buf->buffer, 0));
6308 per_cpu_ptr(buf->data, cpu)->entries = ring_buffer_size(buf->buffer, cpu);
6312 #ifdef CONFIG_TRACER_MAX_TRACE
6313 /* resize @tr's buffer to the size of @size_tr's entries */
6314 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6315 struct array_buffer *size_buf, int cpu_id)
6319 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6320 for_each_tracing_cpu(cpu) {
6321 ret = ring_buffer_resize(trace_buf->buffer,
6322 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6325 per_cpu_ptr(trace_buf->data, cpu)->entries =
6326 per_cpu_ptr(size_buf->data, cpu)->entries;
6329 ret = ring_buffer_resize(trace_buf->buffer,
6330 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6332 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6333 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6338 #endif /* CONFIG_TRACER_MAX_TRACE */
6340 static int __tracing_resize_ring_buffer(struct trace_array *tr,
6341 unsigned long size, int cpu)
6346 * If kernel or user changes the size of the ring buffer
6347 * we use the size that was given, and we can forget about
6348 * expanding it later.
6350 ring_buffer_expanded = true;
6352 /* May be called before buffers are initialized */
6353 if (!tr->array_buffer.buffer)
6356 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6360 #ifdef CONFIG_TRACER_MAX_TRACE
6361 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
6362 !tr->current_trace->use_max_tr)
6365 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6367 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6368 &tr->array_buffer, cpu);
6371 * AARGH! We are left with different
6372 * size max buffer!!!!
6373 * The max buffer is our "snapshot" buffer.
6374 * When a tracer needs a snapshot (one of the
6375 * latency tracers), it swaps the max buffer
6376 * with the saved snap shot. We succeeded to
6377 * update the size of the main buffer, but failed to
6378 * update the size of the max buffer. But when we tried
6379 * to reset the main buffer to the original size, we
6380 * failed there too. This is very unlikely to
6381 * happen, but if it does, warn and kill all
6385 tracing_disabled = 1;
6390 update_buffer_entries(&tr->max_buffer, cpu);
6393 #endif /* CONFIG_TRACER_MAX_TRACE */
6395 update_buffer_entries(&tr->array_buffer, cpu);
6400 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6401 unsigned long size, int cpu_id)
6405 mutex_lock(&trace_types_lock);
6407 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6408 /* make sure, this cpu is enabled in the mask */
6409 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6415 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6420 mutex_unlock(&trace_types_lock);
6427 * tracing_update_buffers - used by tracing facility to expand ring buffers
6429 * To save on memory when the tracing is never used on a system with it
6430 * configured in. The ring buffers are set to a minimum size. But once
6431 * a user starts to use the tracing facility, then they need to grow
6432 * to their default size.
6434 * This function is to be called when a tracer is about to be used.
6436 int tracing_update_buffers(void)
6440 mutex_lock(&trace_types_lock);
6441 if (!ring_buffer_expanded)
6442 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
6443 RING_BUFFER_ALL_CPUS);
6444 mutex_unlock(&trace_types_lock);
6449 struct trace_option_dentry;
6452 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6455 * Used to clear out the tracer before deletion of an instance.
6456 * Must have trace_types_lock held.
6458 static void tracing_set_nop(struct trace_array *tr)
6460 if (tr->current_trace == &nop_trace)
6463 tr->current_trace->enabled--;
6465 if (tr->current_trace->reset)
6466 tr->current_trace->reset(tr);
6468 tr->current_trace = &nop_trace;
6471 static bool tracer_options_updated;
6473 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6475 /* Only enable if the directory has been created already. */
6479 /* Only create trace option files after update_tracer_options finish */
6480 if (!tracer_options_updated)
6483 create_trace_option_files(tr, t);
6486 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6489 #ifdef CONFIG_TRACER_MAX_TRACE
6494 mutex_lock(&trace_types_lock);
6496 if (!ring_buffer_expanded) {
6497 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6498 RING_BUFFER_ALL_CPUS);
6504 for (t = trace_types; t; t = t->next) {
6505 if (strcmp(t->name, buf) == 0)
6512 if (t == tr->current_trace)
6515 #ifdef CONFIG_TRACER_SNAPSHOT
6516 if (t->use_max_tr) {
6517 local_irq_disable();
6518 arch_spin_lock(&tr->max_lock);
6519 if (tr->cond_snapshot)
6521 arch_spin_unlock(&tr->max_lock);
6527 /* Some tracers won't work on kernel command line */
6528 if (system_state < SYSTEM_RUNNING && t->noboot) {
6529 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6534 /* Some tracers are only allowed for the top level buffer */
6535 if (!trace_ok_for_array(t, tr)) {
6540 /* If trace pipe files are being read, we can't change the tracer */
6541 if (tr->trace_ref) {
6546 trace_branch_disable();
6548 tr->current_trace->enabled--;
6550 if (tr->current_trace->reset)
6551 tr->current_trace->reset(tr);
6553 #ifdef CONFIG_TRACER_MAX_TRACE
6554 had_max_tr = tr->current_trace->use_max_tr;
6556 /* Current trace needs to be nop_trace before synchronize_rcu */
6557 tr->current_trace = &nop_trace;
6559 if (had_max_tr && !t->use_max_tr) {
6561 * We need to make sure that the update_max_tr sees that
6562 * current_trace changed to nop_trace to keep it from
6563 * swapping the buffers after we resize it.
6564 * The update_max_tr is called from interrupts disabled
6565 * so a synchronized_sched() is sufficient.
6571 if (t->use_max_tr && !tr->allocated_snapshot) {
6572 ret = tracing_alloc_snapshot_instance(tr);
6577 tr->current_trace = &nop_trace;
6581 ret = tracer_init(t, tr);
6586 tr->current_trace = t;
6587 tr->current_trace->enabled++;
6588 trace_branch_enable(tr);
6590 mutex_unlock(&trace_types_lock);
6596 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6597 size_t cnt, loff_t *ppos)
6599 struct trace_array *tr = filp->private_data;
6600 char buf[MAX_TRACER_SIZE+1];
6607 if (cnt > MAX_TRACER_SIZE)
6608 cnt = MAX_TRACER_SIZE;
6610 if (copy_from_user(buf, ubuf, cnt))
6617 err = tracing_set_tracer(tr, name);
6627 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6628 size_t cnt, loff_t *ppos)
6633 r = snprintf(buf, sizeof(buf), "%ld\n",
6634 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6635 if (r > sizeof(buf))
6637 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6641 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6642 size_t cnt, loff_t *ppos)
6647 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6657 tracing_thresh_read(struct file *filp, char __user *ubuf,
6658 size_t cnt, loff_t *ppos)
6660 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6664 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6665 size_t cnt, loff_t *ppos)
6667 struct trace_array *tr = filp->private_data;
6670 mutex_lock(&trace_types_lock);
6671 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6675 if (tr->current_trace->update_thresh) {
6676 ret = tr->current_trace->update_thresh(tr);
6683 mutex_unlock(&trace_types_lock);
6688 #ifdef CONFIG_TRACER_MAX_TRACE
6691 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6692 size_t cnt, loff_t *ppos)
6694 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6698 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6699 size_t cnt, loff_t *ppos)
6701 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6706 static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6708 if (cpu == RING_BUFFER_ALL_CPUS) {
6709 if (cpumask_empty(tr->pipe_cpumask)) {
6710 cpumask_setall(tr->pipe_cpumask);
6713 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6714 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6720 static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6722 if (cpu == RING_BUFFER_ALL_CPUS) {
6723 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6724 cpumask_clear(tr->pipe_cpumask);
6726 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6727 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6731 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6733 struct trace_array *tr = inode->i_private;
6734 struct trace_iterator *iter;
6738 ret = tracing_check_open_get_tr(tr);
6742 mutex_lock(&trace_types_lock);
6743 cpu = tracing_get_cpu(inode);
6744 ret = open_pipe_on_cpu(tr, cpu);
6746 goto fail_pipe_on_cpu;
6748 /* create a buffer to store the information to pass to userspace */
6749 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6752 goto fail_alloc_iter;
6755 trace_seq_init(&iter->seq);
6756 iter->trace = tr->current_trace;
6758 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6763 /* trace pipe does not show start of buffer */
6764 cpumask_setall(iter->started);
6766 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6767 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6769 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6770 if (trace_clocks[tr->clock_id].in_ns)
6771 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6774 iter->array_buffer = &tr->array_buffer;
6775 iter->cpu_file = cpu;
6776 mutex_init(&iter->mutex);
6777 filp->private_data = iter;
6779 if (iter->trace->pipe_open)
6780 iter->trace->pipe_open(iter);
6782 nonseekable_open(inode, filp);
6786 mutex_unlock(&trace_types_lock);
6792 close_pipe_on_cpu(tr, cpu);
6794 __trace_array_put(tr);
6795 mutex_unlock(&trace_types_lock);
6799 static int tracing_release_pipe(struct inode *inode, struct file *file)
6801 struct trace_iterator *iter = file->private_data;
6802 struct trace_array *tr = inode->i_private;
6804 mutex_lock(&trace_types_lock);
6808 if (iter->trace->pipe_close)
6809 iter->trace->pipe_close(iter);
6810 close_pipe_on_cpu(tr, iter->cpu_file);
6811 mutex_unlock(&trace_types_lock);
6813 free_trace_iter_content(iter);
6816 trace_array_put(tr);
6822 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6824 struct trace_array *tr = iter->tr;
6826 /* Iterators are static, they should be filled or empty */
6827 if (trace_buffer_iter(iter, iter->cpu_file))
6828 return EPOLLIN | EPOLLRDNORM;
6830 if (tr->trace_flags & TRACE_ITER_BLOCK)
6832 * Always select as readable when in blocking mode
6834 return EPOLLIN | EPOLLRDNORM;
6836 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6837 filp, poll_table, iter->tr->buffer_percent);
6841 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6843 struct trace_iterator *iter = filp->private_data;
6845 return trace_poll(iter, filp, poll_table);
6848 /* Must be called with iter->mutex held. */
6849 static int tracing_wait_pipe(struct file *filp)
6851 struct trace_iterator *iter = filp->private_data;
6854 while (trace_empty(iter)) {
6856 if ((filp->f_flags & O_NONBLOCK)) {
6861 * We block until we read something and tracing is disabled.
6862 * We still block if tracing is disabled, but we have never
6863 * read anything. This allows a user to cat this file, and
6864 * then enable tracing. But after we have read something,
6865 * we give an EOF when tracing is again disabled.
6867 * iter->pos will be 0 if we haven't read anything.
6869 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6872 mutex_unlock(&iter->mutex);
6874 ret = wait_on_pipe(iter, 0);
6876 mutex_lock(&iter->mutex);
6889 tracing_read_pipe(struct file *filp, char __user *ubuf,
6890 size_t cnt, loff_t *ppos)
6892 struct trace_iterator *iter = filp->private_data;
6896 * Avoid more than one consumer on a single file descriptor
6897 * This is just a matter of traces coherency, the ring buffer itself
6900 mutex_lock(&iter->mutex);
6902 /* return any leftover data */
6903 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6907 trace_seq_init(&iter->seq);
6909 if (iter->trace->read) {
6910 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6916 sret = tracing_wait_pipe(filp);
6920 /* stop when tracing is finished */
6921 if (trace_empty(iter)) {
6926 if (cnt >= PAGE_SIZE)
6927 cnt = PAGE_SIZE - 1;
6929 /* reset all but tr, trace, and overruns */
6930 trace_iterator_reset(iter);
6931 cpumask_clear(iter->started);
6932 trace_seq_init(&iter->seq);
6934 trace_event_read_lock();
6935 trace_access_lock(iter->cpu_file);
6936 while (trace_find_next_entry_inc(iter) != NULL) {
6937 enum print_line_t ret;
6938 int save_len = iter->seq.seq.len;
6940 ret = print_trace_line(iter);
6941 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6943 * If one print_trace_line() fills entire trace_seq in one shot,
6944 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6945 * In this case, we need to consume it, otherwise, loop will peek
6946 * this event next time, resulting in an infinite loop.
6948 if (save_len == 0) {
6950 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6951 trace_consume(iter);
6955 /* In other cases, don't print partial lines */
6956 iter->seq.seq.len = save_len;
6959 if (ret != TRACE_TYPE_NO_CONSUME)
6960 trace_consume(iter);
6962 if (trace_seq_used(&iter->seq) >= cnt)
6966 * Setting the full flag means we reached the trace_seq buffer
6967 * size and we should leave by partial output condition above.
6968 * One of the trace_seq_* functions is not used properly.
6970 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6973 trace_access_unlock(iter->cpu_file);
6974 trace_event_read_unlock();
6976 /* Now copy what we have to the user */
6977 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6978 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6979 trace_seq_init(&iter->seq);
6982 * If there was nothing to send to user, in spite of consuming trace
6983 * entries, go back to wait for more entries.
6989 mutex_unlock(&iter->mutex);
6994 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6997 __free_page(spd->pages[idx]);
7001 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
7007 /* Seq buffer is page-sized, exactly what we need. */
7009 save_len = iter->seq.seq.len;
7010 ret = print_trace_line(iter);
7012 if (trace_seq_has_overflowed(&iter->seq)) {
7013 iter->seq.seq.len = save_len;
7018 * This should not be hit, because it should only
7019 * be set if the iter->seq overflowed. But check it
7020 * anyway to be safe.
7022 if (ret == TRACE_TYPE_PARTIAL_LINE) {
7023 iter->seq.seq.len = save_len;
7027 count = trace_seq_used(&iter->seq) - save_len;
7030 iter->seq.seq.len = save_len;
7034 if (ret != TRACE_TYPE_NO_CONSUME)
7035 trace_consume(iter);
7037 if (!trace_find_next_entry_inc(iter)) {
7047 static ssize_t tracing_splice_read_pipe(struct file *filp,
7049 struct pipe_inode_info *pipe,
7053 struct page *pages_def[PIPE_DEF_BUFFERS];
7054 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7055 struct trace_iterator *iter = filp->private_data;
7056 struct splice_pipe_desc spd = {
7058 .partial = partial_def,
7059 .nr_pages = 0, /* This gets updated below. */
7060 .nr_pages_max = PIPE_DEF_BUFFERS,
7061 .ops = &default_pipe_buf_ops,
7062 .spd_release = tracing_spd_release_pipe,
7068 if (splice_grow_spd(pipe, &spd))
7071 mutex_lock(&iter->mutex);
7073 if (iter->trace->splice_read) {
7074 ret = iter->trace->splice_read(iter, filp,
7075 ppos, pipe, len, flags);
7080 ret = tracing_wait_pipe(filp);
7084 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
7089 trace_event_read_lock();
7090 trace_access_lock(iter->cpu_file);
7092 /* Fill as many pages as possible. */
7093 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
7094 spd.pages[i] = alloc_page(GFP_KERNEL);
7098 rem = tracing_fill_pipe_page(rem, iter);
7100 /* Copy the data into the page, so we can start over. */
7101 ret = trace_seq_to_buffer(&iter->seq,
7102 page_address(spd.pages[i]),
7103 trace_seq_used(&iter->seq));
7105 __free_page(spd.pages[i]);
7108 spd.partial[i].offset = 0;
7109 spd.partial[i].len = trace_seq_used(&iter->seq);
7111 trace_seq_init(&iter->seq);
7114 trace_access_unlock(iter->cpu_file);
7115 trace_event_read_unlock();
7116 mutex_unlock(&iter->mutex);
7121 ret = splice_to_pipe(pipe, &spd);
7125 splice_shrink_spd(&spd);
7129 mutex_unlock(&iter->mutex);
7134 tracing_entries_read(struct file *filp, char __user *ubuf,
7135 size_t cnt, loff_t *ppos)
7137 struct inode *inode = file_inode(filp);
7138 struct trace_array *tr = inode->i_private;
7139 int cpu = tracing_get_cpu(inode);
7144 mutex_lock(&trace_types_lock);
7146 if (cpu == RING_BUFFER_ALL_CPUS) {
7147 int cpu, buf_size_same;
7152 /* check if all cpu sizes are same */
7153 for_each_tracing_cpu(cpu) {
7154 /* fill in the size from first enabled cpu */
7156 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7157 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7163 if (buf_size_same) {
7164 if (!ring_buffer_expanded)
7165 r = sprintf(buf, "%lu (expanded: %lu)\n",
7167 trace_buf_size >> 10);
7169 r = sprintf(buf, "%lu\n", size >> 10);
7171 r = sprintf(buf, "X\n");
7173 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7175 mutex_unlock(&trace_types_lock);
7177 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7182 tracing_entries_write(struct file *filp, const char __user *ubuf,
7183 size_t cnt, loff_t *ppos)
7185 struct inode *inode = file_inode(filp);
7186 struct trace_array *tr = inode->i_private;
7190 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7194 /* must have at least 1 entry */
7198 /* value is in KB */
7200 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7210 tracing_total_entries_read(struct file *filp, char __user *ubuf,
7211 size_t cnt, loff_t *ppos)
7213 struct trace_array *tr = filp->private_data;
7216 unsigned long size = 0, expanded_size = 0;
7218 mutex_lock(&trace_types_lock);
7219 for_each_tracing_cpu(cpu) {
7220 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7221 if (!ring_buffer_expanded)
7222 expanded_size += trace_buf_size >> 10;
7224 if (ring_buffer_expanded)
7225 r = sprintf(buf, "%lu\n", size);
7227 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7228 mutex_unlock(&trace_types_lock);
7230 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7234 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7235 size_t cnt, loff_t *ppos)
7238 * There is no need to read what the user has written, this function
7239 * is just to make sure that there is no error when "echo" is used
7248 tracing_free_buffer_release(struct inode *inode, struct file *filp)
7250 struct trace_array *tr = inode->i_private;
7252 /* disable tracing ? */
7253 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7254 tracer_tracing_off(tr);
7255 /* resize the ring buffer to 0 */
7256 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7258 trace_array_put(tr);
7264 tracing_mark_write(struct file *filp, const char __user *ubuf,
7265 size_t cnt, loff_t *fpos)
7267 struct trace_array *tr = filp->private_data;
7268 struct ring_buffer_event *event;
7269 enum event_trigger_type tt = ETT_NONE;
7270 struct trace_buffer *buffer;
7271 struct print_entry *entry;
7276 /* Used in tracing_mark_raw_write() as well */
7277 #define FAULTED_STR "<faulted>"
7278 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7280 if (tracing_disabled)
7283 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7286 if (cnt > TRACE_BUF_SIZE)
7287 cnt = TRACE_BUF_SIZE;
7289 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7291 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
7293 /* If less than "<faulted>", then make sure we can still add that */
7294 if (cnt < FAULTED_SIZE)
7295 size += FAULTED_SIZE - cnt;
7297 buffer = tr->array_buffer.buffer;
7298 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7300 if (unlikely(!event))
7301 /* Ring buffer disabled, return as if not open for write */
7304 entry = ring_buffer_event_data(event);
7305 entry->ip = _THIS_IP_;
7307 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7309 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7315 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7316 /* do not add \n before testing triggers, but add \0 */
7317 entry->buf[cnt] = '\0';
7318 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7321 if (entry->buf[cnt - 1] != '\n') {
7322 entry->buf[cnt] = '\n';
7323 entry->buf[cnt + 1] = '\0';
7325 entry->buf[cnt] = '\0';
7327 if (static_branch_unlikely(&trace_marker_exports_enabled))
7328 ftrace_exports(event, TRACE_EXPORT_MARKER);
7329 __buffer_unlock_commit(buffer, event);
7332 event_triggers_post_call(tr->trace_marker_file, tt);
7337 /* Limit it for now to 3K (including tag) */
7338 #define RAW_DATA_MAX_SIZE (1024*3)
7341 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7342 size_t cnt, loff_t *fpos)
7344 struct trace_array *tr = filp->private_data;
7345 struct ring_buffer_event *event;
7346 struct trace_buffer *buffer;
7347 struct raw_data_entry *entry;
7352 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7354 if (tracing_disabled)
7357 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7360 /* The marker must at least have a tag id */
7361 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
7364 if (cnt > TRACE_BUF_SIZE)
7365 cnt = TRACE_BUF_SIZE;
7367 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7369 size = sizeof(*entry) + cnt;
7370 if (cnt < FAULT_SIZE_ID)
7371 size += FAULT_SIZE_ID - cnt;
7373 buffer = tr->array_buffer.buffer;
7374 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7377 /* Ring buffer disabled, return as if not open for write */
7380 entry = ring_buffer_event_data(event);
7382 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7385 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7390 __buffer_unlock_commit(buffer, event);
7395 static int tracing_clock_show(struct seq_file *m, void *v)
7397 struct trace_array *tr = m->private;
7400 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7402 "%s%s%s%s", i ? " " : "",
7403 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7404 i == tr->clock_id ? "]" : "");
7410 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7414 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7415 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7418 if (i == ARRAY_SIZE(trace_clocks))
7421 mutex_lock(&trace_types_lock);
7425 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7428 * New clock may not be consistent with the previous clock.
7429 * Reset the buffer so that it doesn't have incomparable timestamps.
7431 tracing_reset_online_cpus(&tr->array_buffer);
7433 #ifdef CONFIG_TRACER_MAX_TRACE
7434 if (tr->max_buffer.buffer)
7435 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7436 tracing_reset_online_cpus(&tr->max_buffer);
7439 mutex_unlock(&trace_types_lock);
7444 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7445 size_t cnt, loff_t *fpos)
7447 struct seq_file *m = filp->private_data;
7448 struct trace_array *tr = m->private;
7450 const char *clockstr;
7453 if (cnt >= sizeof(buf))
7456 if (copy_from_user(buf, ubuf, cnt))
7461 clockstr = strstrip(buf);
7463 ret = tracing_set_clock(tr, clockstr);
7472 static int tracing_clock_open(struct inode *inode, struct file *file)
7474 struct trace_array *tr = inode->i_private;
7477 ret = tracing_check_open_get_tr(tr);
7481 ret = single_open(file, tracing_clock_show, inode->i_private);
7483 trace_array_put(tr);
7488 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7490 struct trace_array *tr = m->private;
7492 mutex_lock(&trace_types_lock);
7494 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7495 seq_puts(m, "delta [absolute]\n");
7497 seq_puts(m, "[delta] absolute\n");
7499 mutex_unlock(&trace_types_lock);
7504 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7506 struct trace_array *tr = inode->i_private;
7509 ret = tracing_check_open_get_tr(tr);
7513 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7515 trace_array_put(tr);
7520 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7522 if (rbe == this_cpu_read(trace_buffered_event))
7523 return ring_buffer_time_stamp(buffer);
7525 return ring_buffer_event_time_stamp(buffer, rbe);
7529 * Set or disable using the per CPU trace_buffer_event when possible.
7531 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7535 mutex_lock(&trace_types_lock);
7537 if (set && tr->no_filter_buffering_ref++)
7541 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7546 --tr->no_filter_buffering_ref;
7549 mutex_unlock(&trace_types_lock);
7554 struct ftrace_buffer_info {
7555 struct trace_iterator iter;
7557 unsigned int spare_cpu;
7561 #ifdef CONFIG_TRACER_SNAPSHOT
7562 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7564 struct trace_array *tr = inode->i_private;
7565 struct trace_iterator *iter;
7569 ret = tracing_check_open_get_tr(tr);
7573 if (file->f_mode & FMODE_READ) {
7574 iter = __tracing_open(inode, file, true);
7576 ret = PTR_ERR(iter);
7578 /* Writes still need the seq_file to hold the private data */
7580 m = kzalloc(sizeof(*m), GFP_KERNEL);
7583 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7591 iter->array_buffer = &tr->max_buffer;
7592 iter->cpu_file = tracing_get_cpu(inode);
7594 file->private_data = m;
7598 trace_array_put(tr);
7603 static void tracing_swap_cpu_buffer(void *tr)
7605 update_max_tr_single((struct trace_array *)tr, current, smp_processor_id());
7609 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7612 struct seq_file *m = filp->private_data;
7613 struct trace_iterator *iter = m->private;
7614 struct trace_array *tr = iter->tr;
7618 ret = tracing_update_buffers();
7622 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7626 mutex_lock(&trace_types_lock);
7628 if (tr->current_trace->use_max_tr) {
7633 local_irq_disable();
7634 arch_spin_lock(&tr->max_lock);
7635 if (tr->cond_snapshot)
7637 arch_spin_unlock(&tr->max_lock);
7644 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7648 if (tr->allocated_snapshot)
7652 /* Only allow per-cpu swap if the ring buffer supports it */
7653 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7654 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7659 if (tr->allocated_snapshot)
7660 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7661 &tr->array_buffer, iter->cpu_file);
7663 ret = tracing_alloc_snapshot_instance(tr);
7666 /* Now, we're going to swap */
7667 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
7668 local_irq_disable();
7669 update_max_tr(tr, current, smp_processor_id(), NULL);
7672 smp_call_function_single(iter->cpu_file, tracing_swap_cpu_buffer,
7677 if (tr->allocated_snapshot) {
7678 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7679 tracing_reset_online_cpus(&tr->max_buffer);
7681 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7691 mutex_unlock(&trace_types_lock);
7695 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7697 struct seq_file *m = file->private_data;
7700 ret = tracing_release(inode, file);
7702 if (file->f_mode & FMODE_READ)
7705 /* If write only, the seq_file is just a stub */
7713 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7714 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7715 size_t count, loff_t *ppos);
7716 static int tracing_buffers_release(struct inode *inode, struct file *file);
7717 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7718 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7720 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7722 struct ftrace_buffer_info *info;
7725 /* The following checks for tracefs lockdown */
7726 ret = tracing_buffers_open(inode, filp);
7730 info = filp->private_data;
7732 if (info->iter.trace->use_max_tr) {
7733 tracing_buffers_release(inode, filp);
7737 info->iter.snapshot = true;
7738 info->iter.array_buffer = &info->iter.tr->max_buffer;
7743 #endif /* CONFIG_TRACER_SNAPSHOT */
7746 static const struct file_operations tracing_thresh_fops = {
7747 .open = tracing_open_generic,
7748 .read = tracing_thresh_read,
7749 .write = tracing_thresh_write,
7750 .llseek = generic_file_llseek,
7753 #ifdef CONFIG_TRACER_MAX_TRACE
7754 static const struct file_operations tracing_max_lat_fops = {
7755 .open = tracing_open_generic,
7756 .read = tracing_max_lat_read,
7757 .write = tracing_max_lat_write,
7758 .llseek = generic_file_llseek,
7762 static const struct file_operations set_tracer_fops = {
7763 .open = tracing_open_generic,
7764 .read = tracing_set_trace_read,
7765 .write = tracing_set_trace_write,
7766 .llseek = generic_file_llseek,
7769 static const struct file_operations tracing_pipe_fops = {
7770 .open = tracing_open_pipe,
7771 .poll = tracing_poll_pipe,
7772 .read = tracing_read_pipe,
7773 .splice_read = tracing_splice_read_pipe,
7774 .release = tracing_release_pipe,
7775 .llseek = no_llseek,
7778 static const struct file_operations tracing_entries_fops = {
7779 .open = tracing_open_generic_tr,
7780 .read = tracing_entries_read,
7781 .write = tracing_entries_write,
7782 .llseek = generic_file_llseek,
7783 .release = tracing_release_generic_tr,
7786 static const struct file_operations tracing_total_entries_fops = {
7787 .open = tracing_open_generic_tr,
7788 .read = tracing_total_entries_read,
7789 .llseek = generic_file_llseek,
7790 .release = tracing_release_generic_tr,
7793 static const struct file_operations tracing_free_buffer_fops = {
7794 .open = tracing_open_generic_tr,
7795 .write = tracing_free_buffer_write,
7796 .release = tracing_free_buffer_release,
7799 static const struct file_operations tracing_mark_fops = {
7800 .open = tracing_mark_open,
7801 .write = tracing_mark_write,
7802 .release = tracing_release_generic_tr,
7805 static const struct file_operations tracing_mark_raw_fops = {
7806 .open = tracing_mark_open,
7807 .write = tracing_mark_raw_write,
7808 .release = tracing_release_generic_tr,
7811 static const struct file_operations trace_clock_fops = {
7812 .open = tracing_clock_open,
7814 .llseek = seq_lseek,
7815 .release = tracing_single_release_tr,
7816 .write = tracing_clock_write,
7819 static const struct file_operations trace_time_stamp_mode_fops = {
7820 .open = tracing_time_stamp_mode_open,
7822 .llseek = seq_lseek,
7823 .release = tracing_single_release_tr,
7826 #ifdef CONFIG_TRACER_SNAPSHOT
7827 static const struct file_operations snapshot_fops = {
7828 .open = tracing_snapshot_open,
7830 .write = tracing_snapshot_write,
7831 .llseek = tracing_lseek,
7832 .release = tracing_snapshot_release,
7835 static const struct file_operations snapshot_raw_fops = {
7836 .open = snapshot_raw_open,
7837 .read = tracing_buffers_read,
7838 .release = tracing_buffers_release,
7839 .splice_read = tracing_buffers_splice_read,
7840 .llseek = no_llseek,
7843 #endif /* CONFIG_TRACER_SNAPSHOT */
7846 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7847 * @filp: The active open file structure
7848 * @ubuf: The userspace provided buffer to read value into
7849 * @cnt: The maximum number of bytes to read
7850 * @ppos: The current "file" position
7852 * This function implements the write interface for a struct trace_min_max_param.
7853 * The filp->private_data must point to a trace_min_max_param structure that
7854 * defines where to write the value, the min and the max acceptable values,
7855 * and a lock to protect the write.
7858 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7860 struct trace_min_max_param *param = filp->private_data;
7867 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7872 mutex_lock(param->lock);
7874 if (param->min && val < *param->min)
7877 if (param->max && val > *param->max)
7884 mutex_unlock(param->lock);
7893 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7894 * @filp: The active open file structure
7895 * @ubuf: The userspace provided buffer to read value into
7896 * @cnt: The maximum number of bytes to read
7897 * @ppos: The current "file" position
7899 * This function implements the read interface for a struct trace_min_max_param.
7900 * The filp->private_data must point to a trace_min_max_param struct with valid
7904 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7906 struct trace_min_max_param *param = filp->private_data;
7907 char buf[U64_STR_SIZE];
7916 if (cnt > sizeof(buf))
7919 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7921 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7924 const struct file_operations trace_min_max_fops = {
7925 .open = tracing_open_generic,
7926 .read = trace_min_max_read,
7927 .write = trace_min_max_write,
7930 #define TRACING_LOG_ERRS_MAX 8
7931 #define TRACING_LOG_LOC_MAX 128
7933 #define CMD_PREFIX " Command: "
7936 const char **errs; /* ptr to loc-specific array of err strings */
7937 u8 type; /* index into errs -> specific err string */
7938 u16 pos; /* caret position */
7942 struct tracing_log_err {
7943 struct list_head list;
7944 struct err_info info;
7945 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7946 char *cmd; /* what caused err */
7949 static DEFINE_MUTEX(tracing_err_log_lock);
7951 static struct tracing_log_err *alloc_tracing_log_err(int len)
7953 struct tracing_log_err *err;
7955 err = kzalloc(sizeof(*err), GFP_KERNEL);
7957 return ERR_PTR(-ENOMEM);
7959 err->cmd = kzalloc(len, GFP_KERNEL);
7962 return ERR_PTR(-ENOMEM);
7968 static void free_tracing_log_err(struct tracing_log_err *err)
7974 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7977 struct tracing_log_err *err;
7980 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7981 err = alloc_tracing_log_err(len);
7982 if (PTR_ERR(err) != -ENOMEM)
7983 tr->n_err_log_entries++;
7987 cmd = kzalloc(len, GFP_KERNEL);
7989 return ERR_PTR(-ENOMEM);
7990 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7993 list_del(&err->list);
7999 * err_pos - find the position of a string within a command for error careting
8000 * @cmd: The tracing command that caused the error
8001 * @str: The string to position the caret at within @cmd
8003 * Finds the position of the first occurrence of @str within @cmd. The
8004 * return value can be passed to tracing_log_err() for caret placement
8007 * Returns the index within @cmd of the first occurrence of @str or 0
8008 * if @str was not found.
8010 unsigned int err_pos(char *cmd, const char *str)
8014 if (WARN_ON(!strlen(cmd)))
8017 found = strstr(cmd, str);
8025 * tracing_log_err - write an error to the tracing error log
8026 * @tr: The associated trace array for the error (NULL for top level array)
8027 * @loc: A string describing where the error occurred
8028 * @cmd: The tracing command that caused the error
8029 * @errs: The array of loc-specific static error strings
8030 * @type: The index into errs[], which produces the specific static err string
8031 * @pos: The position the caret should be placed in the cmd
8033 * Writes an error into tracing/error_log of the form:
8035 * <loc>: error: <text>
8039 * tracing/error_log is a small log file containing the last
8040 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
8041 * unless there has been a tracing error, and the error log can be
8042 * cleared and have its memory freed by writing the empty string in
8043 * truncation mode to it i.e. echo > tracing/error_log.
8045 * NOTE: the @errs array along with the @type param are used to
8046 * produce a static error string - this string is not copied and saved
8047 * when the error is logged - only a pointer to it is saved. See
8048 * existing callers for examples of how static strings are typically
8049 * defined for use with tracing_log_err().
8051 void tracing_log_err(struct trace_array *tr,
8052 const char *loc, const char *cmd,
8053 const char **errs, u8 type, u16 pos)
8055 struct tracing_log_err *err;
8061 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
8063 mutex_lock(&tracing_err_log_lock);
8064 err = get_tracing_log_err(tr, len);
8065 if (PTR_ERR(err) == -ENOMEM) {
8066 mutex_unlock(&tracing_err_log_lock);
8070 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
8071 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
8073 err->info.errs = errs;
8074 err->info.type = type;
8075 err->info.pos = pos;
8076 err->info.ts = local_clock();
8078 list_add_tail(&err->list, &tr->err_log);
8079 mutex_unlock(&tracing_err_log_lock);
8082 static void clear_tracing_err_log(struct trace_array *tr)
8084 struct tracing_log_err *err, *next;
8086 mutex_lock(&tracing_err_log_lock);
8087 list_for_each_entry_safe(err, next, &tr->err_log, list) {
8088 list_del(&err->list);
8089 free_tracing_log_err(err);
8092 tr->n_err_log_entries = 0;
8093 mutex_unlock(&tracing_err_log_lock);
8096 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
8098 struct trace_array *tr = m->private;
8100 mutex_lock(&tracing_err_log_lock);
8102 return seq_list_start(&tr->err_log, *pos);
8105 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
8107 struct trace_array *tr = m->private;
8109 return seq_list_next(v, &tr->err_log, pos);
8112 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
8114 mutex_unlock(&tracing_err_log_lock);
8117 static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
8121 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
8123 for (i = 0; i < pos; i++)
8128 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
8130 struct tracing_log_err *err = v;
8133 const char *err_text = err->info.errs[err->info.type];
8134 u64 sec = err->info.ts;
8137 nsec = do_div(sec, NSEC_PER_SEC);
8138 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
8139 err->loc, err_text);
8140 seq_printf(m, "%s", err->cmd);
8141 tracing_err_log_show_pos(m, err->info.pos);
8147 static const struct seq_operations tracing_err_log_seq_ops = {
8148 .start = tracing_err_log_seq_start,
8149 .next = tracing_err_log_seq_next,
8150 .stop = tracing_err_log_seq_stop,
8151 .show = tracing_err_log_seq_show
8154 static int tracing_err_log_open(struct inode *inode, struct file *file)
8156 struct trace_array *tr = inode->i_private;
8159 ret = tracing_check_open_get_tr(tr);
8163 /* If this file was opened for write, then erase contents */
8164 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8165 clear_tracing_err_log(tr);
8167 if (file->f_mode & FMODE_READ) {
8168 ret = seq_open(file, &tracing_err_log_seq_ops);
8170 struct seq_file *m = file->private_data;
8173 trace_array_put(tr);
8179 static ssize_t tracing_err_log_write(struct file *file,
8180 const char __user *buffer,
8181 size_t count, loff_t *ppos)
8186 static int tracing_err_log_release(struct inode *inode, struct file *file)
8188 struct trace_array *tr = inode->i_private;
8190 trace_array_put(tr);
8192 if (file->f_mode & FMODE_READ)
8193 seq_release(inode, file);
8198 static const struct file_operations tracing_err_log_fops = {
8199 .open = tracing_err_log_open,
8200 .write = tracing_err_log_write,
8202 .llseek = tracing_lseek,
8203 .release = tracing_err_log_release,
8206 static int tracing_buffers_open(struct inode *inode, struct file *filp)
8208 struct trace_array *tr = inode->i_private;
8209 struct ftrace_buffer_info *info;
8212 ret = tracing_check_open_get_tr(tr);
8216 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8218 trace_array_put(tr);
8222 mutex_lock(&trace_types_lock);
8225 info->iter.cpu_file = tracing_get_cpu(inode);
8226 info->iter.trace = tr->current_trace;
8227 info->iter.array_buffer = &tr->array_buffer;
8229 /* Force reading ring buffer for first read */
8230 info->read = (unsigned int)-1;
8232 filp->private_data = info;
8236 mutex_unlock(&trace_types_lock);
8238 ret = nonseekable_open(inode, filp);
8240 trace_array_put(tr);
8246 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8248 struct ftrace_buffer_info *info = filp->private_data;
8249 struct trace_iterator *iter = &info->iter;
8251 return trace_poll(iter, filp, poll_table);
8255 tracing_buffers_read(struct file *filp, char __user *ubuf,
8256 size_t count, loff_t *ppos)
8258 struct ftrace_buffer_info *info = filp->private_data;
8259 struct trace_iterator *iter = &info->iter;
8266 #ifdef CONFIG_TRACER_MAX_TRACE
8267 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8272 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8274 if (IS_ERR(info->spare)) {
8275 ret = PTR_ERR(info->spare);
8278 info->spare_cpu = iter->cpu_file;
8284 /* Do we have previous read data to read? */
8285 if (info->read < PAGE_SIZE)
8289 trace_access_lock(iter->cpu_file);
8290 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8294 trace_access_unlock(iter->cpu_file);
8297 if (trace_empty(iter)) {
8298 if ((filp->f_flags & O_NONBLOCK))
8301 ret = wait_on_pipe(iter, 0);
8312 size = PAGE_SIZE - info->read;
8316 ret = copy_to_user(ubuf, info->spare + info->read, size);
8328 static int tracing_buffers_release(struct inode *inode, struct file *file)
8330 struct ftrace_buffer_info *info = file->private_data;
8331 struct trace_iterator *iter = &info->iter;
8333 mutex_lock(&trace_types_lock);
8335 iter->tr->trace_ref--;
8337 __trace_array_put(iter->tr);
8340 /* Make sure the waiters see the new wait_index */
8343 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8346 ring_buffer_free_read_page(iter->array_buffer->buffer,
8347 info->spare_cpu, info->spare);
8350 mutex_unlock(&trace_types_lock);
8356 struct trace_buffer *buffer;
8359 refcount_t refcount;
8362 static void buffer_ref_release(struct buffer_ref *ref)
8364 if (!refcount_dec_and_test(&ref->refcount))
8366 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8370 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8371 struct pipe_buffer *buf)
8373 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8375 buffer_ref_release(ref);
8379 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8380 struct pipe_buffer *buf)
8382 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8384 if (refcount_read(&ref->refcount) > INT_MAX/2)
8387 refcount_inc(&ref->refcount);
8391 /* Pipe buffer operations for a buffer. */
8392 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8393 .release = buffer_pipe_buf_release,
8394 .get = buffer_pipe_buf_get,
8398 * Callback from splice_to_pipe(), if we need to release some pages
8399 * at the end of the spd in case we error'ed out in filling the pipe.
8401 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8403 struct buffer_ref *ref =
8404 (struct buffer_ref *)spd->partial[i].private;
8406 buffer_ref_release(ref);
8407 spd->partial[i].private = 0;
8411 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8412 struct pipe_inode_info *pipe, size_t len,
8415 struct ftrace_buffer_info *info = file->private_data;
8416 struct trace_iterator *iter = &info->iter;
8417 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8418 struct page *pages_def[PIPE_DEF_BUFFERS];
8419 struct splice_pipe_desc spd = {
8421 .partial = partial_def,
8422 .nr_pages_max = PIPE_DEF_BUFFERS,
8423 .ops = &buffer_pipe_buf_ops,
8424 .spd_release = buffer_spd_release,
8426 struct buffer_ref *ref;
8430 #ifdef CONFIG_TRACER_MAX_TRACE
8431 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8435 if (*ppos & (PAGE_SIZE - 1))
8438 if (len & (PAGE_SIZE - 1)) {
8439 if (len < PAGE_SIZE)
8444 if (splice_grow_spd(pipe, &spd))
8448 trace_access_lock(iter->cpu_file);
8449 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8451 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
8455 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8461 refcount_set(&ref->refcount, 1);
8462 ref->buffer = iter->array_buffer->buffer;
8463 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8464 if (IS_ERR(ref->page)) {
8465 ret = PTR_ERR(ref->page);
8470 ref->cpu = iter->cpu_file;
8472 r = ring_buffer_read_page(ref->buffer, &ref->page,
8473 len, iter->cpu_file, 1);
8475 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8481 page = virt_to_page(ref->page);
8483 spd.pages[i] = page;
8484 spd.partial[i].len = PAGE_SIZE;
8485 spd.partial[i].offset = 0;
8486 spd.partial[i].private = (unsigned long)ref;
8490 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8493 trace_access_unlock(iter->cpu_file);
8496 /* did we read anything? */
8497 if (!spd.nr_pages) {
8504 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8507 wait_index = READ_ONCE(iter->wait_index);
8509 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
8513 /* No need to wait after waking up when tracing is off */
8514 if (!tracer_tracing_is_on(iter->tr))
8517 /* Make sure we see the new wait_index */
8519 if (wait_index != iter->wait_index)
8525 ret = splice_to_pipe(pipe, &spd);
8527 splice_shrink_spd(&spd);
8532 /* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8533 static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8535 struct ftrace_buffer_info *info = file->private_data;
8536 struct trace_iterator *iter = &info->iter;
8539 return -ENOIOCTLCMD;
8541 mutex_lock(&trace_types_lock);
8544 /* Make sure the waiters see the new wait_index */
8547 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8549 mutex_unlock(&trace_types_lock);
8553 static const struct file_operations tracing_buffers_fops = {
8554 .open = tracing_buffers_open,
8555 .read = tracing_buffers_read,
8556 .poll = tracing_buffers_poll,
8557 .release = tracing_buffers_release,
8558 .splice_read = tracing_buffers_splice_read,
8559 .unlocked_ioctl = tracing_buffers_ioctl,
8560 .llseek = no_llseek,
8564 tracing_stats_read(struct file *filp, char __user *ubuf,
8565 size_t count, loff_t *ppos)
8567 struct inode *inode = file_inode(filp);
8568 struct trace_array *tr = inode->i_private;
8569 struct array_buffer *trace_buf = &tr->array_buffer;
8570 int cpu = tracing_get_cpu(inode);
8571 struct trace_seq *s;
8573 unsigned long long t;
8574 unsigned long usec_rem;
8576 s = kmalloc(sizeof(*s), GFP_KERNEL);
8582 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8583 trace_seq_printf(s, "entries: %ld\n", cnt);
8585 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8586 trace_seq_printf(s, "overrun: %ld\n", cnt);
8588 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8589 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8591 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8592 trace_seq_printf(s, "bytes: %ld\n", cnt);
8594 if (trace_clocks[tr->clock_id].in_ns) {
8595 /* local or global for trace_clock */
8596 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8597 usec_rem = do_div(t, USEC_PER_SEC);
8598 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8601 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8602 usec_rem = do_div(t, USEC_PER_SEC);
8603 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8605 /* counter or tsc mode for trace_clock */
8606 trace_seq_printf(s, "oldest event ts: %llu\n",
8607 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8609 trace_seq_printf(s, "now ts: %llu\n",
8610 ring_buffer_time_stamp(trace_buf->buffer));
8613 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8614 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8616 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8617 trace_seq_printf(s, "read events: %ld\n", cnt);
8619 count = simple_read_from_buffer(ubuf, count, ppos,
8620 s->buffer, trace_seq_used(s));
8627 static const struct file_operations tracing_stats_fops = {
8628 .open = tracing_open_generic_tr,
8629 .read = tracing_stats_read,
8630 .llseek = generic_file_llseek,
8631 .release = tracing_release_generic_tr,
8634 #ifdef CONFIG_DYNAMIC_FTRACE
8637 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8638 size_t cnt, loff_t *ppos)
8644 /* 256 should be plenty to hold the amount needed */
8645 buf = kmalloc(256, GFP_KERNEL);
8649 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8650 ftrace_update_tot_cnt,
8651 ftrace_number_of_pages,
8652 ftrace_number_of_groups);
8654 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8659 static const struct file_operations tracing_dyn_info_fops = {
8660 .open = tracing_open_generic,
8661 .read = tracing_read_dyn_info,
8662 .llseek = generic_file_llseek,
8664 #endif /* CONFIG_DYNAMIC_FTRACE */
8666 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8668 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8669 struct trace_array *tr, struct ftrace_probe_ops *ops,
8672 tracing_snapshot_instance(tr);
8676 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8677 struct trace_array *tr, struct ftrace_probe_ops *ops,
8680 struct ftrace_func_mapper *mapper = data;
8684 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8694 tracing_snapshot_instance(tr);
8698 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8699 struct ftrace_probe_ops *ops, void *data)
8701 struct ftrace_func_mapper *mapper = data;
8704 seq_printf(m, "%ps:", (void *)ip);
8706 seq_puts(m, "snapshot");
8709 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8712 seq_printf(m, ":count=%ld\n", *count);
8714 seq_puts(m, ":unlimited\n");
8720 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8721 unsigned long ip, void *init_data, void **data)
8723 struct ftrace_func_mapper *mapper = *data;
8726 mapper = allocate_ftrace_func_mapper();
8732 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8736 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8737 unsigned long ip, void *data)
8739 struct ftrace_func_mapper *mapper = data;
8744 free_ftrace_func_mapper(mapper, NULL);
8748 ftrace_func_mapper_remove_ip(mapper, ip);
8751 static struct ftrace_probe_ops snapshot_probe_ops = {
8752 .func = ftrace_snapshot,
8753 .print = ftrace_snapshot_print,
8756 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8757 .func = ftrace_count_snapshot,
8758 .print = ftrace_snapshot_print,
8759 .init = ftrace_snapshot_init,
8760 .free = ftrace_snapshot_free,
8764 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8765 char *glob, char *cmd, char *param, int enable)
8767 struct ftrace_probe_ops *ops;
8768 void *count = (void *)-1;
8775 /* hash funcs only work with set_ftrace_filter */
8779 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8782 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8787 number = strsep(¶m, ":");
8789 if (!strlen(number))
8793 * We use the callback data field (which is a pointer)
8796 ret = kstrtoul(number, 0, (unsigned long *)&count);
8801 ret = tracing_alloc_snapshot_instance(tr);
8805 ret = register_ftrace_function_probe(glob, tr, ops, count);
8808 return ret < 0 ? ret : 0;
8811 static struct ftrace_func_command ftrace_snapshot_cmd = {
8813 .func = ftrace_trace_snapshot_callback,
8816 static __init int register_snapshot_cmd(void)
8818 return register_ftrace_command(&ftrace_snapshot_cmd);
8821 static inline __init int register_snapshot_cmd(void) { return 0; }
8822 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8824 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8826 if (WARN_ON(!tr->dir))
8827 return ERR_PTR(-ENODEV);
8829 /* Top directory uses NULL as the parent */
8830 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8833 /* All sub buffers have a descriptor */
8837 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8839 struct dentry *d_tracer;
8842 return tr->percpu_dir;
8844 d_tracer = tracing_get_dentry(tr);
8845 if (IS_ERR(d_tracer))
8848 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8850 MEM_FAIL(!tr->percpu_dir,
8851 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8853 return tr->percpu_dir;
8856 static struct dentry *
8857 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8858 void *data, long cpu, const struct file_operations *fops)
8860 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8862 if (ret) /* See tracing_get_cpu() */
8863 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8868 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8870 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8871 struct dentry *d_cpu;
8872 char cpu_dir[30]; /* 30 characters should be more than enough */
8877 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8878 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8880 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8884 /* per cpu trace_pipe */
8885 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8886 tr, cpu, &tracing_pipe_fops);
8889 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8890 tr, cpu, &tracing_fops);
8892 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8893 tr, cpu, &tracing_buffers_fops);
8895 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8896 tr, cpu, &tracing_stats_fops);
8898 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8899 tr, cpu, &tracing_entries_fops);
8901 #ifdef CONFIG_TRACER_SNAPSHOT
8902 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8903 tr, cpu, &snapshot_fops);
8905 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8906 tr, cpu, &snapshot_raw_fops);
8910 #ifdef CONFIG_FTRACE_SELFTEST
8911 /* Let selftest have access to static functions in this file */
8912 #include "trace_selftest.c"
8916 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8919 struct trace_option_dentry *topt = filp->private_data;
8922 if (topt->flags->val & topt->opt->bit)
8927 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8931 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8934 struct trace_option_dentry *topt = filp->private_data;
8938 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8942 if (val != 0 && val != 1)
8945 if (!!(topt->flags->val & topt->opt->bit) != val) {
8946 mutex_lock(&trace_types_lock);
8947 ret = __set_tracer_option(topt->tr, topt->flags,
8949 mutex_unlock(&trace_types_lock);
8960 static const struct file_operations trace_options_fops = {
8961 .open = tracing_open_generic,
8962 .read = trace_options_read,
8963 .write = trace_options_write,
8964 .llseek = generic_file_llseek,
8968 * In order to pass in both the trace_array descriptor as well as the index
8969 * to the flag that the trace option file represents, the trace_array
8970 * has a character array of trace_flags_index[], which holds the index
8971 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8972 * The address of this character array is passed to the flag option file
8973 * read/write callbacks.
8975 * In order to extract both the index and the trace_array descriptor,
8976 * get_tr_index() uses the following algorithm.
8980 * As the pointer itself contains the address of the index (remember
8983 * Then to get the trace_array descriptor, by subtracting that index
8984 * from the ptr, we get to the start of the index itself.
8986 * ptr - idx == &index[0]
8988 * Then a simple container_of() from that pointer gets us to the
8989 * trace_array descriptor.
8991 static void get_tr_index(void *data, struct trace_array **ptr,
8992 unsigned int *pindex)
8994 *pindex = *(unsigned char *)data;
8996 *ptr = container_of(data - *pindex, struct trace_array,
9001 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
9004 void *tr_index = filp->private_data;
9005 struct trace_array *tr;
9009 get_tr_index(tr_index, &tr, &index);
9011 if (tr->trace_flags & (1 << index))
9016 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9020 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
9023 void *tr_index = filp->private_data;
9024 struct trace_array *tr;
9029 get_tr_index(tr_index, &tr, &index);
9031 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9035 if (val != 0 && val != 1)
9038 mutex_lock(&event_mutex);
9039 mutex_lock(&trace_types_lock);
9040 ret = set_tracer_flag(tr, 1 << index, val);
9041 mutex_unlock(&trace_types_lock);
9042 mutex_unlock(&event_mutex);
9052 static const struct file_operations trace_options_core_fops = {
9053 .open = tracing_open_generic,
9054 .read = trace_options_core_read,
9055 .write = trace_options_core_write,
9056 .llseek = generic_file_llseek,
9059 struct dentry *trace_create_file(const char *name,
9061 struct dentry *parent,
9063 const struct file_operations *fops)
9067 ret = tracefs_create_file(name, mode, parent, data, fops);
9069 pr_warn("Could not create tracefs '%s' entry\n", name);
9075 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
9077 struct dentry *d_tracer;
9082 d_tracer = tracing_get_dentry(tr);
9083 if (IS_ERR(d_tracer))
9086 tr->options = tracefs_create_dir("options", d_tracer);
9088 pr_warn("Could not create tracefs directory 'options'\n");
9096 create_trace_option_file(struct trace_array *tr,
9097 struct trace_option_dentry *topt,
9098 struct tracer_flags *flags,
9099 struct tracer_opt *opt)
9101 struct dentry *t_options;
9103 t_options = trace_options_init_dentry(tr);
9107 topt->flags = flags;
9111 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
9112 t_options, topt, &trace_options_fops);
9117 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
9119 struct trace_option_dentry *topts;
9120 struct trace_options *tr_topts;
9121 struct tracer_flags *flags;
9122 struct tracer_opt *opts;
9129 flags = tracer->flags;
9131 if (!flags || !flags->opts)
9135 * If this is an instance, only create flags for tracers
9136 * the instance may have.
9138 if (!trace_ok_for_array(tracer, tr))
9141 for (i = 0; i < tr->nr_topts; i++) {
9142 /* Make sure there's no duplicate flags. */
9143 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
9149 for (cnt = 0; opts[cnt].name; cnt++)
9152 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
9156 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9163 tr->topts = tr_topts;
9164 tr->topts[tr->nr_topts].tracer = tracer;
9165 tr->topts[tr->nr_topts].topts = topts;
9168 for (cnt = 0; opts[cnt].name; cnt++) {
9169 create_trace_option_file(tr, &topts[cnt], flags,
9171 MEM_FAIL(topts[cnt].entry == NULL,
9172 "Failed to create trace option: %s",
9177 static struct dentry *
9178 create_trace_option_core_file(struct trace_array *tr,
9179 const char *option, long index)
9181 struct dentry *t_options;
9183 t_options = trace_options_init_dentry(tr);
9187 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9188 (void *)&tr->trace_flags_index[index],
9189 &trace_options_core_fops);
9192 static void create_trace_options_dir(struct trace_array *tr)
9194 struct dentry *t_options;
9195 bool top_level = tr == &global_trace;
9198 t_options = trace_options_init_dentry(tr);
9202 for (i = 0; trace_options[i]; i++) {
9204 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9205 create_trace_option_core_file(tr, trace_options[i], i);
9210 rb_simple_read(struct file *filp, char __user *ubuf,
9211 size_t cnt, loff_t *ppos)
9213 struct trace_array *tr = filp->private_data;
9217 r = tracer_tracing_is_on(tr);
9218 r = sprintf(buf, "%d\n", r);
9220 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9224 rb_simple_write(struct file *filp, const char __user *ubuf,
9225 size_t cnt, loff_t *ppos)
9227 struct trace_array *tr = filp->private_data;
9228 struct trace_buffer *buffer = tr->array_buffer.buffer;
9232 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9237 mutex_lock(&trace_types_lock);
9238 if (!!val == tracer_tracing_is_on(tr)) {
9239 val = 0; /* do nothing */
9241 tracer_tracing_on(tr);
9242 if (tr->current_trace->start)
9243 tr->current_trace->start(tr);
9245 tracer_tracing_off(tr);
9246 if (tr->current_trace->stop)
9247 tr->current_trace->stop(tr);
9248 /* Wake up any waiters */
9249 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9251 mutex_unlock(&trace_types_lock);
9259 static const struct file_operations rb_simple_fops = {
9260 .open = tracing_open_generic_tr,
9261 .read = rb_simple_read,
9262 .write = rb_simple_write,
9263 .release = tracing_release_generic_tr,
9264 .llseek = default_llseek,
9268 buffer_percent_read(struct file *filp, char __user *ubuf,
9269 size_t cnt, loff_t *ppos)
9271 struct trace_array *tr = filp->private_data;
9275 r = tr->buffer_percent;
9276 r = sprintf(buf, "%d\n", r);
9278 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9282 buffer_percent_write(struct file *filp, const char __user *ubuf,
9283 size_t cnt, loff_t *ppos)
9285 struct trace_array *tr = filp->private_data;
9289 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9296 tr->buffer_percent = val;
9303 static const struct file_operations buffer_percent_fops = {
9304 .open = tracing_open_generic_tr,
9305 .read = buffer_percent_read,
9306 .write = buffer_percent_write,
9307 .release = tracing_release_generic_tr,
9308 .llseek = default_llseek,
9311 static struct dentry *trace_instance_dir;
9314 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9317 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9319 enum ring_buffer_flags rb_flags;
9321 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9325 buf->buffer = ring_buffer_alloc(size, rb_flags);
9329 buf->data = alloc_percpu(struct trace_array_cpu);
9331 ring_buffer_free(buf->buffer);
9336 /* Allocate the first page for all buffers */
9337 set_buffer_entries(&tr->array_buffer,
9338 ring_buffer_size(tr->array_buffer.buffer, 0));
9343 static void free_trace_buffer(struct array_buffer *buf)
9346 ring_buffer_free(buf->buffer);
9348 free_percpu(buf->data);
9353 static int allocate_trace_buffers(struct trace_array *tr, int size)
9357 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9361 #ifdef CONFIG_TRACER_MAX_TRACE
9362 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9363 allocate_snapshot ? size : 1);
9364 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9365 free_trace_buffer(&tr->array_buffer);
9368 tr->allocated_snapshot = allocate_snapshot;
9370 allocate_snapshot = false;
9376 static void free_trace_buffers(struct trace_array *tr)
9381 free_trace_buffer(&tr->array_buffer);
9383 #ifdef CONFIG_TRACER_MAX_TRACE
9384 free_trace_buffer(&tr->max_buffer);
9388 static void init_trace_flags_index(struct trace_array *tr)
9392 /* Used by the trace options files */
9393 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9394 tr->trace_flags_index[i] = i;
9397 static void __update_tracer_options(struct trace_array *tr)
9401 for (t = trace_types; t; t = t->next)
9402 add_tracer_options(tr, t);
9405 static void update_tracer_options(struct trace_array *tr)
9407 mutex_lock(&trace_types_lock);
9408 tracer_options_updated = true;
9409 __update_tracer_options(tr);
9410 mutex_unlock(&trace_types_lock);
9413 /* Must have trace_types_lock held */
9414 struct trace_array *trace_array_find(const char *instance)
9416 struct trace_array *tr, *found = NULL;
9418 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9419 if (tr->name && strcmp(tr->name, instance) == 0) {
9428 struct trace_array *trace_array_find_get(const char *instance)
9430 struct trace_array *tr;
9432 mutex_lock(&trace_types_lock);
9433 tr = trace_array_find(instance);
9436 mutex_unlock(&trace_types_lock);
9441 static int trace_array_create_dir(struct trace_array *tr)
9445 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9449 ret = event_trace_add_tracer(tr->dir, tr);
9451 tracefs_remove(tr->dir);
9455 init_tracer_tracefs(tr, tr->dir);
9456 __update_tracer_options(tr);
9461 static struct trace_array *trace_array_create(const char *name)
9463 struct trace_array *tr;
9467 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9469 return ERR_PTR(ret);
9471 tr->name = kstrdup(name, GFP_KERNEL);
9475 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9478 if (!zalloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9481 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9483 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9485 raw_spin_lock_init(&tr->start_lock);
9487 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9489 tr->current_trace = &nop_trace;
9491 INIT_LIST_HEAD(&tr->systems);
9492 INIT_LIST_HEAD(&tr->events);
9493 INIT_LIST_HEAD(&tr->hist_vars);
9494 INIT_LIST_HEAD(&tr->err_log);
9496 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9499 if (ftrace_allocate_ftrace_ops(tr) < 0)
9502 ftrace_init_trace_array(tr);
9504 init_trace_flags_index(tr);
9506 if (trace_instance_dir) {
9507 ret = trace_array_create_dir(tr);
9511 __trace_early_add_events(tr);
9513 list_add(&tr->list, &ftrace_trace_arrays);
9520 ftrace_free_ftrace_ops(tr);
9521 free_trace_buffers(tr);
9522 free_cpumask_var(tr->pipe_cpumask);
9523 free_cpumask_var(tr->tracing_cpumask);
9527 return ERR_PTR(ret);
9530 static int instance_mkdir(const char *name)
9532 struct trace_array *tr;
9535 mutex_lock(&event_mutex);
9536 mutex_lock(&trace_types_lock);
9539 if (trace_array_find(name))
9542 tr = trace_array_create(name);
9544 ret = PTR_ERR_OR_ZERO(tr);
9547 mutex_unlock(&trace_types_lock);
9548 mutex_unlock(&event_mutex);
9553 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9554 * @name: The name of the trace array to be looked up/created.
9556 * Returns pointer to trace array with given name.
9557 * NULL, if it cannot be created.
9559 * NOTE: This function increments the reference counter associated with the
9560 * trace array returned. This makes sure it cannot be freed while in use.
9561 * Use trace_array_put() once the trace array is no longer needed.
9562 * If the trace_array is to be freed, trace_array_destroy() needs to
9563 * be called after the trace_array_put(), or simply let user space delete
9564 * it from the tracefs instances directory. But until the
9565 * trace_array_put() is called, user space can not delete it.
9568 struct trace_array *trace_array_get_by_name(const char *name)
9570 struct trace_array *tr;
9572 mutex_lock(&event_mutex);
9573 mutex_lock(&trace_types_lock);
9575 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9576 if (tr->name && strcmp(tr->name, name) == 0)
9580 tr = trace_array_create(name);
9588 mutex_unlock(&trace_types_lock);
9589 mutex_unlock(&event_mutex);
9592 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9594 static int __remove_instance(struct trace_array *tr)
9598 /* Reference counter for a newly created trace array = 1. */
9599 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9602 list_del(&tr->list);
9604 /* Disable all the flags that were enabled coming in */
9605 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9606 if ((1 << i) & ZEROED_TRACE_FLAGS)
9607 set_tracer_flag(tr, 1 << i, 0);
9610 tracing_set_nop(tr);
9611 clear_ftrace_function_probes(tr);
9612 event_trace_del_tracer(tr);
9613 ftrace_clear_pids(tr);
9614 ftrace_destroy_function_files(tr);
9615 tracefs_remove(tr->dir);
9616 free_percpu(tr->last_func_repeats);
9617 free_trace_buffers(tr);
9618 clear_tracing_err_log(tr);
9620 for (i = 0; i < tr->nr_topts; i++) {
9621 kfree(tr->topts[i].topts);
9625 free_cpumask_var(tr->pipe_cpumask);
9626 free_cpumask_var(tr->tracing_cpumask);
9633 int trace_array_destroy(struct trace_array *this_tr)
9635 struct trace_array *tr;
9641 mutex_lock(&event_mutex);
9642 mutex_lock(&trace_types_lock);
9646 /* Making sure trace array exists before destroying it. */
9647 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9648 if (tr == this_tr) {
9649 ret = __remove_instance(tr);
9654 mutex_unlock(&trace_types_lock);
9655 mutex_unlock(&event_mutex);
9659 EXPORT_SYMBOL_GPL(trace_array_destroy);
9661 static int instance_rmdir(const char *name)
9663 struct trace_array *tr;
9666 mutex_lock(&event_mutex);
9667 mutex_lock(&trace_types_lock);
9670 tr = trace_array_find(name);
9672 ret = __remove_instance(tr);
9674 mutex_unlock(&trace_types_lock);
9675 mutex_unlock(&event_mutex);
9680 static __init void create_trace_instances(struct dentry *d_tracer)
9682 struct trace_array *tr;
9684 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9687 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9690 mutex_lock(&event_mutex);
9691 mutex_lock(&trace_types_lock);
9693 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9696 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9697 "Failed to create instance directory\n"))
9701 mutex_unlock(&trace_types_lock);
9702 mutex_unlock(&event_mutex);
9706 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9708 struct trace_event_file *file;
9711 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9712 tr, &show_traces_fops);
9714 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9715 tr, &set_tracer_fops);
9717 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9718 tr, &tracing_cpumask_fops);
9720 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9721 tr, &tracing_iter_fops);
9723 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9726 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9727 tr, &tracing_pipe_fops);
9729 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9730 tr, &tracing_entries_fops);
9732 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9733 tr, &tracing_total_entries_fops);
9735 trace_create_file("free_buffer", 0200, d_tracer,
9736 tr, &tracing_free_buffer_fops);
9738 trace_create_file("trace_marker", 0220, d_tracer,
9739 tr, &tracing_mark_fops);
9741 file = __find_event_file(tr, "ftrace", "print");
9742 if (file && file->dir)
9743 trace_create_file("trigger", TRACE_MODE_WRITE, file->dir,
9744 file, &event_trigger_fops);
9745 tr->trace_marker_file = file;
9747 trace_create_file("trace_marker_raw", 0220, d_tracer,
9748 tr, &tracing_mark_raw_fops);
9750 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9753 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9754 tr, &rb_simple_fops);
9756 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9757 &trace_time_stamp_mode_fops);
9759 tr->buffer_percent = 50;
9761 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9762 tr, &buffer_percent_fops);
9764 create_trace_options_dir(tr);
9766 #ifdef CONFIG_TRACER_MAX_TRACE
9767 trace_create_maxlat_file(tr, d_tracer);
9770 if (ftrace_create_function_files(tr, d_tracer))
9771 MEM_FAIL(1, "Could not allocate function filter files");
9773 #ifdef CONFIG_TRACER_SNAPSHOT
9774 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9775 tr, &snapshot_fops);
9778 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9779 tr, &tracing_err_log_fops);
9781 for_each_tracing_cpu(cpu)
9782 tracing_init_tracefs_percpu(tr, cpu);
9784 ftrace_init_tracefs(tr, d_tracer);
9787 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9789 struct vfsmount *mnt;
9790 struct file_system_type *type;
9793 * To maintain backward compatibility for tools that mount
9794 * debugfs to get to the tracing facility, tracefs is automatically
9795 * mounted to the debugfs/tracing directory.
9797 type = get_fs_type("tracefs");
9800 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9801 put_filesystem(type);
9810 * tracing_init_dentry - initialize top level trace array
9812 * This is called when creating files or directories in the tracing
9813 * directory. It is called via fs_initcall() by any of the boot up code
9814 * and expects to return the dentry of the top level tracing directory.
9816 int tracing_init_dentry(void)
9818 struct trace_array *tr = &global_trace;
9820 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9821 pr_warn("Tracing disabled due to lockdown\n");
9825 /* The top level trace array uses NULL as parent */
9829 if (WARN_ON(!tracefs_initialized()))
9833 * As there may still be users that expect the tracing
9834 * files to exist in debugfs/tracing, we must automount
9835 * the tracefs file system there, so older tools still
9836 * work with the newer kernel.
9838 tr->dir = debugfs_create_automount("tracing", NULL,
9839 trace_automount, NULL);
9844 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9845 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9847 static struct workqueue_struct *eval_map_wq __initdata;
9848 static struct work_struct eval_map_work __initdata;
9849 static struct work_struct tracerfs_init_work __initdata;
9851 static void __init eval_map_work_func(struct work_struct *work)
9855 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9856 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9859 static int __init trace_eval_init(void)
9861 INIT_WORK(&eval_map_work, eval_map_work_func);
9863 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9865 pr_err("Unable to allocate eval_map_wq\n");
9867 eval_map_work_func(&eval_map_work);
9871 queue_work(eval_map_wq, &eval_map_work);
9875 subsys_initcall(trace_eval_init);
9877 static int __init trace_eval_sync(void)
9879 /* Make sure the eval map updates are finished */
9881 destroy_workqueue(eval_map_wq);
9885 late_initcall_sync(trace_eval_sync);
9888 #ifdef CONFIG_MODULES
9889 static void trace_module_add_evals(struct module *mod)
9891 if (!mod->num_trace_evals)
9895 * Modules with bad taint do not have events created, do
9896 * not bother with enums either.
9898 if (trace_module_has_bad_taint(mod))
9901 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9904 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
9905 static void trace_module_remove_evals(struct module *mod)
9907 union trace_eval_map_item *map;
9908 union trace_eval_map_item **last = &trace_eval_maps;
9910 if (!mod->num_trace_evals)
9913 mutex_lock(&trace_eval_mutex);
9915 map = trace_eval_maps;
9918 if (map->head.mod == mod)
9920 map = trace_eval_jmp_to_tail(map);
9921 last = &map->tail.next;
9922 map = map->tail.next;
9927 *last = trace_eval_jmp_to_tail(map)->tail.next;
9930 mutex_unlock(&trace_eval_mutex);
9933 static inline void trace_module_remove_evals(struct module *mod) { }
9934 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9936 static int trace_module_notify(struct notifier_block *self,
9937 unsigned long val, void *data)
9939 struct module *mod = data;
9942 case MODULE_STATE_COMING:
9943 trace_module_add_evals(mod);
9945 case MODULE_STATE_GOING:
9946 trace_module_remove_evals(mod);
9953 static struct notifier_block trace_module_nb = {
9954 .notifier_call = trace_module_notify,
9957 #endif /* CONFIG_MODULES */
9959 static __init void tracer_init_tracefs_work_func(struct work_struct *work)
9964 init_tracer_tracefs(&global_trace, NULL);
9965 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9967 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9968 &global_trace, &tracing_thresh_fops);
9970 trace_create_file("README", TRACE_MODE_READ, NULL,
9971 NULL, &tracing_readme_fops);
9973 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9974 NULL, &tracing_saved_cmdlines_fops);
9976 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9977 NULL, &tracing_saved_cmdlines_size_fops);
9979 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9980 NULL, &tracing_saved_tgids_fops);
9982 trace_create_eval_file(NULL);
9984 #ifdef CONFIG_MODULES
9985 register_module_notifier(&trace_module_nb);
9988 #ifdef CONFIG_DYNAMIC_FTRACE
9989 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
9990 NULL, &tracing_dyn_info_fops);
9993 create_trace_instances(NULL);
9995 update_tracer_options(&global_trace);
9998 static __init int tracer_init_tracefs(void)
10002 trace_access_lock_init();
10004 ret = tracing_init_dentry();
10009 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
10010 queue_work(eval_map_wq, &tracerfs_init_work);
10012 tracer_init_tracefs_work_func(NULL);
10015 rv_init_interface();
10020 fs_initcall(tracer_init_tracefs);
10022 static int trace_die_panic_handler(struct notifier_block *self,
10023 unsigned long ev, void *unused);
10025 static struct notifier_block trace_panic_notifier = {
10026 .notifier_call = trace_die_panic_handler,
10027 .priority = INT_MAX - 1,
10030 static struct notifier_block trace_die_notifier = {
10031 .notifier_call = trace_die_panic_handler,
10032 .priority = INT_MAX - 1,
10036 * The idea is to execute the following die/panic callback early, in order
10037 * to avoid showing irrelevant information in the trace (like other panic
10038 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
10039 * warnings get disabled (to prevent potential log flooding).
10041 static int trace_die_panic_handler(struct notifier_block *self,
10042 unsigned long ev, void *unused)
10044 if (!ftrace_dump_on_oops)
10045 return NOTIFY_DONE;
10047 /* The die notifier requires DIE_OOPS to trigger */
10048 if (self == &trace_die_notifier && ev != DIE_OOPS)
10049 return NOTIFY_DONE;
10051 ftrace_dump(ftrace_dump_on_oops);
10053 return NOTIFY_DONE;
10057 * printk is set to max of 1024, we really don't need it that big.
10058 * Nothing should be printing 1000 characters anyway.
10060 #define TRACE_MAX_PRINT 1000
10063 * Define here KERN_TRACE so that we have one place to modify
10064 * it if we decide to change what log level the ftrace dump
10067 #define KERN_TRACE KERN_EMERG
10070 trace_printk_seq(struct trace_seq *s)
10072 /* Probably should print a warning here. */
10073 if (s->seq.len >= TRACE_MAX_PRINT)
10074 s->seq.len = TRACE_MAX_PRINT;
10077 * More paranoid code. Although the buffer size is set to
10078 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
10079 * an extra layer of protection.
10081 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
10082 s->seq.len = s->seq.size - 1;
10084 /* should be zero ended, but we are paranoid. */
10085 s->buffer[s->seq.len] = 0;
10087 printk(KERN_TRACE "%s", s->buffer);
10092 void trace_init_global_iter(struct trace_iterator *iter)
10094 iter->tr = &global_trace;
10095 iter->trace = iter->tr->current_trace;
10096 iter->cpu_file = RING_BUFFER_ALL_CPUS;
10097 iter->array_buffer = &global_trace.array_buffer;
10099 if (iter->trace && iter->trace->open)
10100 iter->trace->open(iter);
10102 /* Annotate start of buffers if we had overruns */
10103 if (ring_buffer_overruns(iter->array_buffer->buffer))
10104 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10106 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10107 if (trace_clocks[iter->tr->clock_id].in_ns)
10108 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10110 /* Can not use kmalloc for iter.temp and iter.fmt */
10111 iter->temp = static_temp_buf;
10112 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10113 iter->fmt = static_fmt_buf;
10114 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10117 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10119 /* use static because iter can be a bit big for the stack */
10120 static struct trace_iterator iter;
10121 static atomic_t dump_running;
10122 struct trace_array *tr = &global_trace;
10123 unsigned int old_userobj;
10124 unsigned long flags;
10127 /* Only allow one dump user at a time. */
10128 if (atomic_inc_return(&dump_running) != 1) {
10129 atomic_dec(&dump_running);
10134 * Always turn off tracing when we dump.
10135 * We don't need to show trace output of what happens
10136 * between multiple crashes.
10138 * If the user does a sysrq-z, then they can re-enable
10139 * tracing with echo 1 > tracing_on.
10143 local_irq_save(flags);
10145 /* Simulate the iterator */
10146 trace_init_global_iter(&iter);
10148 for_each_tracing_cpu(cpu) {
10149 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10152 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10154 /* don't look at user memory in panic mode */
10155 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10157 switch (oops_dump_mode) {
10159 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10162 iter.cpu_file = raw_smp_processor_id();
10167 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10168 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10171 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10173 /* Did function tracer already get disabled? */
10174 if (ftrace_is_dead()) {
10175 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10176 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10180 * We need to stop all tracing on all CPUS to read
10181 * the next buffer. This is a bit expensive, but is
10182 * not done often. We fill all what we can read,
10183 * and then release the locks again.
10186 while (!trace_empty(&iter)) {
10189 printk(KERN_TRACE "---------------------------------\n");
10193 trace_iterator_reset(&iter);
10194 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10196 if (trace_find_next_entry_inc(&iter) != NULL) {
10199 ret = print_trace_line(&iter);
10200 if (ret != TRACE_TYPE_NO_CONSUME)
10201 trace_consume(&iter);
10203 touch_nmi_watchdog();
10205 trace_printk_seq(&iter.seq);
10209 printk(KERN_TRACE " (ftrace buffer empty)\n");
10211 printk(KERN_TRACE "---------------------------------\n");
10214 tr->trace_flags |= old_userobj;
10216 for_each_tracing_cpu(cpu) {
10217 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10219 atomic_dec(&dump_running);
10220 local_irq_restore(flags);
10222 EXPORT_SYMBOL_GPL(ftrace_dump);
10224 #define WRITE_BUFSIZE 4096
10226 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10227 size_t count, loff_t *ppos,
10228 int (*createfn)(const char *))
10230 char *kbuf, *buf, *tmp;
10235 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10239 while (done < count) {
10240 size = count - done;
10242 if (size >= WRITE_BUFSIZE)
10243 size = WRITE_BUFSIZE - 1;
10245 if (copy_from_user(kbuf, buffer + done, size)) {
10252 tmp = strchr(buf, '\n');
10255 size = tmp - buf + 1;
10257 size = strlen(buf);
10258 if (done + size < count) {
10261 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10262 pr_warn("Line length is too long: Should be less than %d\n",
10263 WRITE_BUFSIZE - 2);
10270 /* Remove comments */
10271 tmp = strchr(buf, '#');
10276 ret = createfn(buf);
10281 } while (done < count);
10291 #ifdef CONFIG_TRACER_MAX_TRACE
10292 __init static bool tr_needs_alloc_snapshot(const char *name)
10295 int len = strlen(name);
10298 if (!boot_snapshot_index)
10301 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10302 boot_snapshot_info[len] == '\t')
10305 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10309 sprintf(test, "\t%s\t", name);
10310 ret = strstr(boot_snapshot_info, test) == NULL;
10315 __init static void do_allocate_snapshot(const char *name)
10317 if (!tr_needs_alloc_snapshot(name))
10321 * When allocate_snapshot is set, the next call to
10322 * allocate_trace_buffers() (called by trace_array_get_by_name())
10323 * will allocate the snapshot buffer. That will alse clear
10326 allocate_snapshot = true;
10329 static inline void do_allocate_snapshot(const char *name) { }
10332 __init static void enable_instances(void)
10334 struct trace_array *tr;
10339 /* A tab is always appended */
10340 boot_instance_info[boot_instance_index - 1] = '\0';
10341 str = boot_instance_info;
10343 while ((curr_str = strsep(&str, "\t"))) {
10345 tok = strsep(&curr_str, ",");
10347 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10348 do_allocate_snapshot(tok);
10350 tr = trace_array_get_by_name(tok);
10352 pr_warn("Failed to create instance buffer %s\n", curr_str);
10355 /* Allow user space to delete it */
10356 trace_array_put(tr);
10358 while ((tok = strsep(&curr_str, ","))) {
10359 early_enable_events(tr, tok, true);
10364 __init static int tracer_alloc_buffers(void)
10370 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10371 pr_warn("Tracing disabled due to lockdown\n");
10376 * Make sure we don't accidentally add more trace options
10377 * than we have bits for.
10379 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10381 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10384 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10385 goto out_free_buffer_mask;
10387 /* Only allocate trace_printk buffers if a trace_printk exists */
10388 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10389 /* Must be called before global_trace.buffer is allocated */
10390 trace_printk_init_buffers();
10392 /* To save memory, keep the ring buffer size to its minimum */
10393 if (ring_buffer_expanded)
10394 ring_buf_size = trace_buf_size;
10398 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10399 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10401 raw_spin_lock_init(&global_trace.start_lock);
10404 * The prepare callbacks allocates some memory for the ring buffer. We
10405 * don't free the buffer if the CPU goes down. If we were to free
10406 * the buffer, then the user would lose any trace that was in the
10407 * buffer. The memory will be removed once the "instance" is removed.
10409 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10410 "trace/RB:prepare", trace_rb_cpu_prepare,
10413 goto out_free_cpumask;
10414 /* Used for event triggers */
10416 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10418 goto out_rm_hp_state;
10420 if (trace_create_savedcmd() < 0)
10421 goto out_free_temp_buffer;
10423 if (!zalloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10424 goto out_free_savedcmd;
10426 /* TODO: make the number of buffers hot pluggable with CPUS */
10427 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10428 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10429 goto out_free_pipe_cpumask;
10431 if (global_trace.buffer_disabled)
10434 if (trace_boot_clock) {
10435 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10437 pr_warn("Trace clock %s not defined, going back to default\n",
10442 * register_tracer() might reference current_trace, so it
10443 * needs to be set before we register anything. This is
10444 * just a bootstrap of current_trace anyway.
10446 global_trace.current_trace = &nop_trace;
10448 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10450 ftrace_init_global_array_ops(&global_trace);
10452 init_trace_flags_index(&global_trace);
10454 register_tracer(&nop_trace);
10456 /* Function tracing may start here (via kernel command line) */
10457 init_function_trace();
10459 /* All seems OK, enable tracing */
10460 tracing_disabled = 0;
10462 atomic_notifier_chain_register(&panic_notifier_list,
10463 &trace_panic_notifier);
10465 register_die_notifier(&trace_die_notifier);
10467 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10469 INIT_LIST_HEAD(&global_trace.systems);
10470 INIT_LIST_HEAD(&global_trace.events);
10471 INIT_LIST_HEAD(&global_trace.hist_vars);
10472 INIT_LIST_HEAD(&global_trace.err_log);
10473 list_add(&global_trace.list, &ftrace_trace_arrays);
10475 apply_trace_boot_options();
10477 register_snapshot_cmd();
10483 out_free_pipe_cpumask:
10484 free_cpumask_var(global_trace.pipe_cpumask);
10486 free_saved_cmdlines_buffer(savedcmd);
10487 out_free_temp_buffer:
10488 ring_buffer_free(temp_buffer);
10490 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10492 free_cpumask_var(global_trace.tracing_cpumask);
10493 out_free_buffer_mask:
10494 free_cpumask_var(tracing_buffer_mask);
10499 void __init ftrace_boot_snapshot(void)
10501 #ifdef CONFIG_TRACER_MAX_TRACE
10502 struct trace_array *tr;
10504 if (!snapshot_at_boot)
10507 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10508 if (!tr->allocated_snapshot)
10511 tracing_snapshot_instance(tr);
10512 trace_array_puts(tr, "** Boot snapshot taken **\n");
10517 void __init early_trace_init(void)
10519 if (tracepoint_printk) {
10520 tracepoint_print_iter =
10521 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10522 if (MEM_FAIL(!tracepoint_print_iter,
10523 "Failed to allocate trace iterator\n"))
10524 tracepoint_printk = 0;
10526 static_key_enable(&tracepoint_printk_key.key);
10528 tracer_alloc_buffers();
10533 void __init trace_init(void)
10535 trace_event_init();
10537 if (boot_instance_index)
10538 enable_instances();
10541 __init static void clear_boot_tracer(void)
10544 * The default tracer at boot buffer is an init section.
10545 * This function is called in lateinit. If we did not
10546 * find the boot tracer, then clear it out, to prevent
10547 * later registration from accessing the buffer that is
10548 * about to be freed.
10550 if (!default_bootup_tracer)
10553 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10554 default_bootup_tracer);
10555 default_bootup_tracer = NULL;
10558 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10559 __init static void tracing_set_default_clock(void)
10561 /* sched_clock_stable() is determined in late_initcall */
10562 if (!trace_boot_clock && !sched_clock_stable()) {
10563 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10564 pr_warn("Can not set tracing clock due to lockdown\n");
10568 printk(KERN_WARNING
10569 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10570 "If you want to keep using the local clock, then add:\n"
10571 " \"trace_clock=local\"\n"
10572 "on the kernel command line\n");
10573 tracing_set_clock(&global_trace, "global");
10577 static inline void tracing_set_default_clock(void) { }
10580 __init static int late_trace_init(void)
10582 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10583 static_key_disable(&tracepoint_printk_key.key);
10584 tracepoint_printk = 0;
10587 tracing_set_default_clock();
10588 clear_boot_tracer();
10592 late_initcall_sync(late_trace_init);