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;
3125 * Add one, for this function and the call to save_stack_trace()
3126 * If regs is set, then these functions will not be in the way.
3128 #ifndef CONFIG_UNWINDER_ORC
3133 preempt_disable_notrace();
3135 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3137 /* This should never happen. If it does, yell once and skip */
3138 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3142 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3143 * interrupt will either see the value pre increment or post
3144 * increment. If the interrupt happens pre increment it will have
3145 * restored the counter when it returns. We just need a barrier to
3146 * keep gcc from moving things around.
3150 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3151 size = ARRAY_SIZE(fstack->calls);
3154 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3157 nr_entries = stack_trace_save(fstack->calls, size, skip);
3160 size = nr_entries * sizeof(unsigned long);
3161 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3162 (sizeof(*entry) - sizeof(entry->caller)) + size,
3166 ptr = ring_buffer_event_data(event);
3170 * For backward compatibility reasons, the entry->caller is an
3171 * array of 8 slots to store the stack. This is also exported
3172 * to user space. The amount allocated on the ring buffer actually
3173 * holds enough for the stack specified by nr_entries. This will
3174 * go into the location of entry->caller. Due to string fortifiers
3175 * checking the size of the destination of memcpy() it triggers
3176 * when it detects that size is greater than 8. To hide this from
3177 * the fortifiers, we use "ptr" and pointer arithmetic to assign caller.
3179 * The below is really just:
3180 * memcpy(&entry->caller, fstack->calls, size);
3182 ptr += offsetof(typeof(*entry), caller);
3183 memcpy(ptr, fstack->calls, size);
3185 entry->size = nr_entries;
3187 if (!call_filter_check_discard(call, entry, buffer, event))
3188 __buffer_unlock_commit(buffer, event);
3191 /* Again, don't let gcc optimize things here */
3193 __this_cpu_dec(ftrace_stack_reserve);
3194 preempt_enable_notrace();
3198 static inline void ftrace_trace_stack(struct trace_array *tr,
3199 struct trace_buffer *buffer,
3200 unsigned int trace_ctx,
3201 int skip, struct pt_regs *regs)
3203 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3206 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3209 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3212 struct trace_buffer *buffer = tr->array_buffer.buffer;
3214 if (rcu_is_watching()) {
3215 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3219 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3223 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3224 * but if the above rcu_is_watching() failed, then the NMI
3225 * triggered someplace critical, and ct_irq_enter() should
3226 * not be called from NMI.
3228 if (unlikely(in_nmi()))
3231 ct_irq_enter_irqson();
3232 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3233 ct_irq_exit_irqson();
3237 * trace_dump_stack - record a stack back trace in the trace buffer
3238 * @skip: Number of functions to skip (helper handlers)
3240 void trace_dump_stack(int skip)
3242 if (tracing_disabled || tracing_selftest_running)
3245 #ifndef CONFIG_UNWINDER_ORC
3246 /* Skip 1 to skip this function. */
3249 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3250 tracing_gen_ctx(), skip, NULL);
3252 EXPORT_SYMBOL_GPL(trace_dump_stack);
3254 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3255 static DEFINE_PER_CPU(int, user_stack_count);
3258 ftrace_trace_userstack(struct trace_array *tr,
3259 struct trace_buffer *buffer, unsigned int trace_ctx)
3261 struct trace_event_call *call = &event_user_stack;
3262 struct ring_buffer_event *event;
3263 struct userstack_entry *entry;
3265 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3269 * NMIs can not handle page faults, even with fix ups.
3270 * The save user stack can (and often does) fault.
3272 if (unlikely(in_nmi()))
3276 * prevent recursion, since the user stack tracing may
3277 * trigger other kernel events.
3280 if (__this_cpu_read(user_stack_count))
3283 __this_cpu_inc(user_stack_count);
3285 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3286 sizeof(*entry), trace_ctx);
3288 goto out_drop_count;
3289 entry = ring_buffer_event_data(event);
3291 entry->tgid = current->tgid;
3292 memset(&entry->caller, 0, sizeof(entry->caller));
3294 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3295 if (!call_filter_check_discard(call, entry, buffer, event))
3296 __buffer_unlock_commit(buffer, event);
3299 __this_cpu_dec(user_stack_count);
3303 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3304 static void ftrace_trace_userstack(struct trace_array *tr,
3305 struct trace_buffer *buffer,
3306 unsigned int trace_ctx)
3309 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3311 #endif /* CONFIG_STACKTRACE */
3314 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3315 unsigned long long delta)
3317 entry->bottom_delta_ts = delta & U32_MAX;
3318 entry->top_delta_ts = (delta >> 32);
3321 void trace_last_func_repeats(struct trace_array *tr,
3322 struct trace_func_repeats *last_info,
3323 unsigned int trace_ctx)
3325 struct trace_buffer *buffer = tr->array_buffer.buffer;
3326 struct func_repeats_entry *entry;
3327 struct ring_buffer_event *event;
3330 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3331 sizeof(*entry), trace_ctx);
3335 delta = ring_buffer_event_time_stamp(buffer, event) -
3336 last_info->ts_last_call;
3338 entry = ring_buffer_event_data(event);
3339 entry->ip = last_info->ip;
3340 entry->parent_ip = last_info->parent_ip;
3341 entry->count = last_info->count;
3342 func_repeats_set_delta_ts(entry, delta);
3344 __buffer_unlock_commit(buffer, event);
3347 /* created for use with alloc_percpu */
3348 struct trace_buffer_struct {
3350 char buffer[4][TRACE_BUF_SIZE];
3353 static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3356 * This allows for lockless recording. If we're nested too deeply, then
3357 * this returns NULL.
3359 static char *get_trace_buf(void)
3361 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3363 if (!trace_percpu_buffer || buffer->nesting >= 4)
3368 /* Interrupts must see nesting incremented before we use the buffer */
3370 return &buffer->buffer[buffer->nesting - 1][0];
3373 static void put_trace_buf(void)
3375 /* Don't let the decrement of nesting leak before this */
3377 this_cpu_dec(trace_percpu_buffer->nesting);
3380 static int alloc_percpu_trace_buffer(void)
3382 struct trace_buffer_struct __percpu *buffers;
3384 if (trace_percpu_buffer)
3387 buffers = alloc_percpu(struct trace_buffer_struct);
3388 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3391 trace_percpu_buffer = buffers;
3395 static int buffers_allocated;
3397 void trace_printk_init_buffers(void)
3399 if (buffers_allocated)
3402 if (alloc_percpu_trace_buffer())
3405 /* trace_printk() is for debug use only. Don't use it in production. */
3408 pr_warn("**********************************************************\n");
3409 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3411 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3413 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3414 pr_warn("** unsafe for production use. **\n");
3416 pr_warn("** If you see this message and you are not debugging **\n");
3417 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3419 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3420 pr_warn("**********************************************************\n");
3422 /* Expand the buffers to set size */
3423 tracing_update_buffers();
3425 buffers_allocated = 1;
3428 * trace_printk_init_buffers() can be called by modules.
3429 * If that happens, then we need to start cmdline recording
3430 * directly here. If the global_trace.buffer is already
3431 * allocated here, then this was called by module code.
3433 if (global_trace.array_buffer.buffer)
3434 tracing_start_cmdline_record();
3436 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3438 void trace_printk_start_comm(void)
3440 /* Start tracing comms if trace printk is set */
3441 if (!buffers_allocated)
3443 tracing_start_cmdline_record();
3446 static void trace_printk_start_stop_comm(int enabled)
3448 if (!buffers_allocated)
3452 tracing_start_cmdline_record();
3454 tracing_stop_cmdline_record();
3458 * trace_vbprintk - write binary msg to tracing buffer
3459 * @ip: The address of the caller
3460 * @fmt: The string format to write to the buffer
3461 * @args: Arguments for @fmt
3463 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3465 struct trace_event_call *call = &event_bprint;
3466 struct ring_buffer_event *event;
3467 struct trace_buffer *buffer;
3468 struct trace_array *tr = &global_trace;
3469 struct bprint_entry *entry;
3470 unsigned int trace_ctx;
3474 if (unlikely(tracing_selftest_running || tracing_disabled))
3477 /* Don't pollute graph traces with trace_vprintk internals */
3478 pause_graph_tracing();
3480 trace_ctx = tracing_gen_ctx();
3481 preempt_disable_notrace();
3483 tbuffer = get_trace_buf();
3489 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3491 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3494 size = sizeof(*entry) + sizeof(u32) * len;
3495 buffer = tr->array_buffer.buffer;
3496 ring_buffer_nest_start(buffer);
3497 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3501 entry = ring_buffer_event_data(event);
3505 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3506 if (!call_filter_check_discard(call, entry, buffer, event)) {
3507 __buffer_unlock_commit(buffer, event);
3508 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3512 ring_buffer_nest_end(buffer);
3517 preempt_enable_notrace();
3518 unpause_graph_tracing();
3522 EXPORT_SYMBOL_GPL(trace_vbprintk);
3526 __trace_array_vprintk(struct trace_buffer *buffer,
3527 unsigned long ip, const char *fmt, va_list args)
3529 struct trace_event_call *call = &event_print;
3530 struct ring_buffer_event *event;
3532 struct print_entry *entry;
3533 unsigned int trace_ctx;
3536 if (tracing_disabled)
3539 /* Don't pollute graph traces with trace_vprintk internals */
3540 pause_graph_tracing();
3542 trace_ctx = tracing_gen_ctx();
3543 preempt_disable_notrace();
3546 tbuffer = get_trace_buf();
3552 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3554 size = sizeof(*entry) + len + 1;
3555 ring_buffer_nest_start(buffer);
3556 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3560 entry = ring_buffer_event_data(event);
3563 memcpy(&entry->buf, tbuffer, len + 1);
3564 if (!call_filter_check_discard(call, entry, buffer, event)) {
3565 __buffer_unlock_commit(buffer, event);
3566 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3570 ring_buffer_nest_end(buffer);
3574 preempt_enable_notrace();
3575 unpause_graph_tracing();
3581 int trace_array_vprintk(struct trace_array *tr,
3582 unsigned long ip, const char *fmt, va_list args)
3584 if (tracing_selftest_running && tr == &global_trace)
3587 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3591 * trace_array_printk - Print a message to a specific instance
3592 * @tr: The instance trace_array descriptor
3593 * @ip: The instruction pointer that this is called from.
3594 * @fmt: The format to print (printf format)
3596 * If a subsystem sets up its own instance, they have the right to
3597 * printk strings into their tracing instance buffer using this
3598 * function. Note, this function will not write into the top level
3599 * buffer (use trace_printk() for that), as writing into the top level
3600 * buffer should only have events that can be individually disabled.
3601 * trace_printk() is only used for debugging a kernel, and should not
3602 * be ever incorporated in normal use.
3604 * trace_array_printk() can be used, as it will not add noise to the
3605 * top level tracing buffer.
3607 * Note, trace_array_init_printk() must be called on @tr before this
3611 int trace_array_printk(struct trace_array *tr,
3612 unsigned long ip, const char *fmt, ...)
3620 /* This is only allowed for created instances */
3621 if (tr == &global_trace)
3624 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3628 ret = trace_array_vprintk(tr, ip, fmt, ap);
3632 EXPORT_SYMBOL_GPL(trace_array_printk);
3635 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3636 * @tr: The trace array to initialize the buffers for
3638 * As trace_array_printk() only writes into instances, they are OK to
3639 * have in the kernel (unlike trace_printk()). This needs to be called
3640 * before trace_array_printk() can be used on a trace_array.
3642 int trace_array_init_printk(struct trace_array *tr)
3647 /* This is only allowed for created instances */
3648 if (tr == &global_trace)
3651 return alloc_percpu_trace_buffer();
3653 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3656 int trace_array_printk_buf(struct trace_buffer *buffer,
3657 unsigned long ip, const char *fmt, ...)
3662 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3666 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3672 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3674 return trace_array_vprintk(&global_trace, ip, fmt, args);
3676 EXPORT_SYMBOL_GPL(trace_vprintk);
3678 static void trace_iterator_increment(struct trace_iterator *iter)
3680 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3684 ring_buffer_iter_advance(buf_iter);
3687 static struct trace_entry *
3688 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3689 unsigned long *lost_events)
3691 struct ring_buffer_event *event;
3692 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3695 event = ring_buffer_iter_peek(buf_iter, ts);
3697 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3698 (unsigned long)-1 : 0;
3700 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3705 iter->ent_size = ring_buffer_event_length(event);
3706 return ring_buffer_event_data(event);
3712 static struct trace_entry *
3713 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3714 unsigned long *missing_events, u64 *ent_ts)
3716 struct trace_buffer *buffer = iter->array_buffer->buffer;
3717 struct trace_entry *ent, *next = NULL;
3718 unsigned long lost_events = 0, next_lost = 0;
3719 int cpu_file = iter->cpu_file;
3720 u64 next_ts = 0, ts;
3726 * If we are in a per_cpu trace file, don't bother by iterating over
3727 * all cpu and peek directly.
3729 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3730 if (ring_buffer_empty_cpu(buffer, cpu_file))
3732 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3734 *ent_cpu = cpu_file;
3739 for_each_tracing_cpu(cpu) {
3741 if (ring_buffer_empty_cpu(buffer, cpu))
3744 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3747 * Pick the entry with the smallest timestamp:
3749 if (ent && (!next || ts < next_ts)) {
3753 next_lost = lost_events;
3754 next_size = iter->ent_size;
3758 iter->ent_size = next_size;
3761 *ent_cpu = next_cpu;
3767 *missing_events = next_lost;
3772 #define STATIC_FMT_BUF_SIZE 128
3773 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3775 char *trace_iter_expand_format(struct trace_iterator *iter)
3780 * iter->tr is NULL when used with tp_printk, which makes
3781 * this get called where it is not safe to call krealloc().
3783 if (!iter->tr || iter->fmt == static_fmt_buf)
3786 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3789 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3796 /* Returns true if the string is safe to dereference from an event */
3797 static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3800 unsigned long addr = (unsigned long)str;
3801 struct trace_event *trace_event;
3802 struct trace_event_call *event;
3804 /* Ignore strings with no length */
3808 /* OK if part of the event data */
3809 if ((addr >= (unsigned long)iter->ent) &&
3810 (addr < (unsigned long)iter->ent + iter->ent_size))
3813 /* OK if part of the temp seq buffer */
3814 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3815 (addr < (unsigned long)iter->tmp_seq.buffer + PAGE_SIZE))
3818 /* Core rodata can not be freed */
3819 if (is_kernel_rodata(addr))
3822 if (trace_is_tracepoint_string(str))
3826 * Now this could be a module event, referencing core module
3827 * data, which is OK.
3832 trace_event = ftrace_find_event(iter->ent->type);
3836 event = container_of(trace_event, struct trace_event_call, event);
3837 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3840 /* Would rather have rodata, but this will suffice */
3841 if (within_module_core(addr, event->module))
3847 static const char *show_buffer(struct trace_seq *s)
3849 struct seq_buf *seq = &s->seq;
3851 seq_buf_terminate(seq);
3856 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3858 static int test_can_verify_check(const char *fmt, ...)
3865 * The verifier is dependent on vsnprintf() modifies the va_list
3866 * passed to it, where it is sent as a reference. Some architectures
3867 * (like x86_32) passes it by value, which means that vsnprintf()
3868 * does not modify the va_list passed to it, and the verifier
3869 * would then need to be able to understand all the values that
3870 * vsnprintf can use. If it is passed by value, then the verifier
3874 vsnprintf(buf, 16, "%d", ap);
3875 ret = va_arg(ap, int);
3881 static void test_can_verify(void)
3883 if (!test_can_verify_check("%d %d", 0, 1)) {
3884 pr_info("trace event string verifier disabled\n");
3885 static_branch_inc(&trace_no_verify);
3890 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3891 * @iter: The iterator that holds the seq buffer and the event being printed
3892 * @fmt: The format used to print the event
3893 * @ap: The va_list holding the data to print from @fmt.
3895 * This writes the data into the @iter->seq buffer using the data from
3896 * @fmt and @ap. If the format has a %s, then the source of the string
3897 * is examined to make sure it is safe to print, otherwise it will
3898 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3901 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3904 const char *p = fmt;
3908 if (WARN_ON_ONCE(!fmt))
3911 if (static_branch_unlikely(&trace_no_verify))
3914 /* Don't bother checking when doing a ftrace_dump() */
3915 if (iter->fmt == static_fmt_buf)
3924 /* We only care about %s and variants */
3925 for (i = 0; p[i]; i++) {
3926 if (i + 1 >= iter->fmt_size) {
3928 * If we can't expand the copy buffer,
3931 if (!trace_iter_expand_format(iter))
3935 if (p[i] == '\\' && p[i+1]) {
3940 /* Need to test cases like %08.*s */
3941 for (j = 1; p[i+j]; j++) {
3942 if (isdigit(p[i+j]) ||
3945 if (p[i+j] == '*') {
3957 /* If no %s found then just print normally */
3961 /* Copy up to the %s, and print that */
3962 strncpy(iter->fmt, p, i);
3963 iter->fmt[i] = '\0';
3964 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3967 * If iter->seq is full, the above call no longer guarantees
3968 * that ap is in sync with fmt processing, and further calls
3969 * to va_arg() can return wrong positional arguments.
3971 * Ensure that ap is no longer used in this case.
3973 if (iter->seq.full) {
3979 len = va_arg(ap, int);
3981 /* The ap now points to the string data of the %s */
3982 str = va_arg(ap, const char *);
3985 * If you hit this warning, it is likely that the
3986 * trace event in question used %s on a string that
3987 * was saved at the time of the event, but may not be
3988 * around when the trace is read. Use __string(),
3989 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3990 * instead. See samples/trace_events/trace-events-sample.h
3993 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3994 "fmt: '%s' current_buffer: '%s'",
3995 fmt, show_buffer(&iter->seq))) {
3998 /* Try to safely read the string */
4000 if (len + 1 > iter->fmt_size)
4001 len = iter->fmt_size - 1;
4004 ret = copy_from_kernel_nofault(iter->fmt, str, len);
4008 ret = strncpy_from_kernel_nofault(iter->fmt, str,
4012 trace_seq_printf(&iter->seq, "(0x%px)", str);
4014 trace_seq_printf(&iter->seq, "(0x%px:%s)",
4016 str = "[UNSAFE-MEMORY]";
4017 strcpy(iter->fmt, "%s");
4019 strncpy(iter->fmt, p + i, j + 1);
4020 iter->fmt[j+1] = '\0';
4023 trace_seq_printf(&iter->seq, iter->fmt, len, str);
4025 trace_seq_printf(&iter->seq, iter->fmt, str);
4031 trace_seq_vprintf(&iter->seq, p, ap);
4034 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
4036 const char *p, *new_fmt;
4039 if (WARN_ON_ONCE(!fmt))
4042 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
4046 new_fmt = q = iter->fmt;
4048 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
4049 if (!trace_iter_expand_format(iter))
4052 q += iter->fmt - new_fmt;
4053 new_fmt = iter->fmt;
4058 /* Replace %p with %px */
4062 } else if (p[0] == 'p' && !isalnum(p[1])) {
4073 #define STATIC_TEMP_BUF_SIZE 128
4074 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
4076 /* Find the next real entry, without updating the iterator itself */
4077 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
4078 int *ent_cpu, u64 *ent_ts)
4080 /* __find_next_entry will reset ent_size */
4081 int ent_size = iter->ent_size;
4082 struct trace_entry *entry;
4085 * If called from ftrace_dump(), then the iter->temp buffer
4086 * will be the static_temp_buf and not created from kmalloc.
4087 * If the entry size is greater than the buffer, we can
4088 * not save it. Just return NULL in that case. This is only
4089 * used to add markers when two consecutive events' time
4090 * stamps have a large delta. See trace_print_lat_context()
4092 if (iter->temp == static_temp_buf &&
4093 STATIC_TEMP_BUF_SIZE < ent_size)
4097 * The __find_next_entry() may call peek_next_entry(), which may
4098 * call ring_buffer_peek() that may make the contents of iter->ent
4099 * undefined. Need to copy iter->ent now.
4101 if (iter->ent && iter->ent != iter->temp) {
4102 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4103 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4105 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4110 iter->temp_size = iter->ent_size;
4112 memcpy(iter->temp, iter->ent, iter->ent_size);
4113 iter->ent = iter->temp;
4115 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4116 /* Put back the original ent_size */
4117 iter->ent_size = ent_size;
4122 /* Find the next real entry, and increment the iterator to the next entry */
4123 void *trace_find_next_entry_inc(struct trace_iterator *iter)
4125 iter->ent = __find_next_entry(iter, &iter->cpu,
4126 &iter->lost_events, &iter->ts);
4129 trace_iterator_increment(iter);
4131 return iter->ent ? iter : NULL;
4134 static void trace_consume(struct trace_iterator *iter)
4136 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4137 &iter->lost_events);
4140 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4142 struct trace_iterator *iter = m->private;
4146 WARN_ON_ONCE(iter->leftover);
4150 /* can't go backwards */
4155 ent = trace_find_next_entry_inc(iter);
4159 while (ent && iter->idx < i)
4160 ent = trace_find_next_entry_inc(iter);
4167 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4169 struct ring_buffer_iter *buf_iter;
4170 unsigned long entries = 0;
4173 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4175 buf_iter = trace_buffer_iter(iter, cpu);
4179 ring_buffer_iter_reset(buf_iter);
4182 * We could have the case with the max latency tracers
4183 * that a reset never took place on a cpu. This is evident
4184 * by the timestamp being before the start of the buffer.
4186 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4187 if (ts >= iter->array_buffer->time_start)
4190 ring_buffer_iter_advance(buf_iter);
4193 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4197 * The current tracer is copied to avoid a global locking
4200 static void *s_start(struct seq_file *m, loff_t *pos)
4202 struct trace_iterator *iter = m->private;
4203 struct trace_array *tr = iter->tr;
4204 int cpu_file = iter->cpu_file;
4210 * copy the tracer to avoid using a global lock all around.
4211 * iter->trace is a copy of current_trace, the pointer to the
4212 * name may be used instead of a strcmp(), as iter->trace->name
4213 * will point to the same string as current_trace->name.
4215 mutex_lock(&trace_types_lock);
4216 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name)) {
4217 /* Close iter->trace before switching to the new current tracer */
4218 if (iter->trace->close)
4219 iter->trace->close(iter);
4220 *iter->trace = *tr->current_trace;
4221 /* Reopen the new current tracer */
4222 if (iter->trace->open)
4223 iter->trace->open(iter);
4225 mutex_unlock(&trace_types_lock);
4227 #ifdef CONFIG_TRACER_MAX_TRACE
4228 if (iter->snapshot && iter->trace->use_max_tr)
4229 return ERR_PTR(-EBUSY);
4232 if (*pos != iter->pos) {
4237 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4238 for_each_tracing_cpu(cpu)
4239 tracing_iter_reset(iter, cpu);
4241 tracing_iter_reset(iter, cpu_file);
4244 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4249 * If we overflowed the seq_file before, then we want
4250 * to just reuse the trace_seq buffer again.
4256 p = s_next(m, p, &l);
4260 trace_event_read_lock();
4261 trace_access_lock(cpu_file);
4265 static void s_stop(struct seq_file *m, void *p)
4267 struct trace_iterator *iter = m->private;
4269 #ifdef CONFIG_TRACER_MAX_TRACE
4270 if (iter->snapshot && iter->trace->use_max_tr)
4274 trace_access_unlock(iter->cpu_file);
4275 trace_event_read_unlock();
4279 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4280 unsigned long *entries, int cpu)
4282 unsigned long count;
4284 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4286 * If this buffer has skipped entries, then we hold all
4287 * entries for the trace and we need to ignore the
4288 * ones before the time stamp.
4290 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4291 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4292 /* total is the same as the entries */
4296 ring_buffer_overrun_cpu(buf->buffer, cpu);
4301 get_total_entries(struct array_buffer *buf,
4302 unsigned long *total, unsigned long *entries)
4310 for_each_tracing_cpu(cpu) {
4311 get_total_entries_cpu(buf, &t, &e, cpu);
4317 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4319 unsigned long total, entries;
4324 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4329 unsigned long trace_total_entries(struct trace_array *tr)
4331 unsigned long total, entries;
4336 get_total_entries(&tr->array_buffer, &total, &entries);
4341 static void print_lat_help_header(struct seq_file *m)
4343 seq_puts(m, "# _------=> CPU# \n"
4344 "# / _-----=> irqs-off/BH-disabled\n"
4345 "# | / _----=> need-resched \n"
4346 "# || / _---=> hardirq/softirq \n"
4347 "# ||| / _--=> preempt-depth \n"
4348 "# |||| / _-=> migrate-disable \n"
4349 "# ||||| / delay \n"
4350 "# cmd pid |||||| time | caller \n"
4351 "# \\ / |||||| \\ | / \n");
4354 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4356 unsigned long total;
4357 unsigned long entries;
4359 get_total_entries(buf, &total, &entries);
4360 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4361 entries, total, num_online_cpus());
4365 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4368 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4370 print_event_info(buf, m);
4372 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4373 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4376 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4379 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4380 static const char space[] = " ";
4381 int prec = tgid ? 12 : 2;
4383 print_event_info(buf, m);
4385 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4386 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4387 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4388 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4389 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4390 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4391 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4392 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4396 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4398 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4399 struct array_buffer *buf = iter->array_buffer;
4400 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4401 struct tracer *type = iter->trace;
4402 unsigned long entries;
4403 unsigned long total;
4404 const char *name = type->name;
4406 get_total_entries(buf, &total, &entries);
4408 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4410 seq_puts(m, "# -----------------------------------"
4411 "---------------------------------\n");
4412 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4413 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4414 nsecs_to_usecs(data->saved_latency),
4418 preempt_model_none() ? "server" :
4419 preempt_model_voluntary() ? "desktop" :
4420 preempt_model_full() ? "preempt" :
4421 preempt_model_rt() ? "preempt_rt" :
4423 /* These are reserved for later use */
4426 seq_printf(m, " #P:%d)\n", num_online_cpus());
4430 seq_puts(m, "# -----------------\n");
4431 seq_printf(m, "# | task: %.16s-%d "
4432 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4433 data->comm, data->pid,
4434 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4435 data->policy, data->rt_priority);
4436 seq_puts(m, "# -----------------\n");
4438 if (data->critical_start) {
4439 seq_puts(m, "# => started at: ");
4440 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4441 trace_print_seq(m, &iter->seq);
4442 seq_puts(m, "\n# => ended at: ");
4443 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4444 trace_print_seq(m, &iter->seq);
4445 seq_puts(m, "\n#\n");
4451 static void test_cpu_buff_start(struct trace_iterator *iter)
4453 struct trace_seq *s = &iter->seq;
4454 struct trace_array *tr = iter->tr;
4456 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4459 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4462 if (cpumask_available(iter->started) &&
4463 cpumask_test_cpu(iter->cpu, iter->started))
4466 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4469 if (cpumask_available(iter->started))
4470 cpumask_set_cpu(iter->cpu, iter->started);
4472 /* Don't print started cpu buffer for the first entry of the trace */
4474 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4478 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4480 struct trace_array *tr = iter->tr;
4481 struct trace_seq *s = &iter->seq;
4482 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4483 struct trace_entry *entry;
4484 struct trace_event *event;
4488 test_cpu_buff_start(iter);
4490 event = ftrace_find_event(entry->type);
4492 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4493 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4494 trace_print_lat_context(iter);
4496 trace_print_context(iter);
4499 if (trace_seq_has_overflowed(s))
4500 return TRACE_TYPE_PARTIAL_LINE;
4503 if (tr->trace_flags & TRACE_ITER_FIELDS)
4504 return print_event_fields(iter, event);
4505 return event->funcs->trace(iter, sym_flags, event);
4508 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4510 return trace_handle_return(s);
4513 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4515 struct trace_array *tr = iter->tr;
4516 struct trace_seq *s = &iter->seq;
4517 struct trace_entry *entry;
4518 struct trace_event *event;
4522 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4523 trace_seq_printf(s, "%d %d %llu ",
4524 entry->pid, iter->cpu, iter->ts);
4526 if (trace_seq_has_overflowed(s))
4527 return TRACE_TYPE_PARTIAL_LINE;
4529 event = ftrace_find_event(entry->type);
4531 return event->funcs->raw(iter, 0, event);
4533 trace_seq_printf(s, "%d ?\n", entry->type);
4535 return trace_handle_return(s);
4538 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4540 struct trace_array *tr = iter->tr;
4541 struct trace_seq *s = &iter->seq;
4542 unsigned char newline = '\n';
4543 struct trace_entry *entry;
4544 struct trace_event *event;
4548 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4549 SEQ_PUT_HEX_FIELD(s, entry->pid);
4550 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4551 SEQ_PUT_HEX_FIELD(s, iter->ts);
4552 if (trace_seq_has_overflowed(s))
4553 return TRACE_TYPE_PARTIAL_LINE;
4556 event = ftrace_find_event(entry->type);
4558 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4559 if (ret != TRACE_TYPE_HANDLED)
4563 SEQ_PUT_FIELD(s, newline);
4565 return trace_handle_return(s);
4568 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4570 struct trace_array *tr = iter->tr;
4571 struct trace_seq *s = &iter->seq;
4572 struct trace_entry *entry;
4573 struct trace_event *event;
4577 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4578 SEQ_PUT_FIELD(s, entry->pid);
4579 SEQ_PUT_FIELD(s, iter->cpu);
4580 SEQ_PUT_FIELD(s, iter->ts);
4581 if (trace_seq_has_overflowed(s))
4582 return TRACE_TYPE_PARTIAL_LINE;
4585 event = ftrace_find_event(entry->type);
4586 return event ? event->funcs->binary(iter, 0, event) :
4590 int trace_empty(struct trace_iterator *iter)
4592 struct ring_buffer_iter *buf_iter;
4595 /* If we are looking at one CPU buffer, only check that one */
4596 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4597 cpu = iter->cpu_file;
4598 buf_iter = trace_buffer_iter(iter, cpu);
4600 if (!ring_buffer_iter_empty(buf_iter))
4603 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4609 for_each_tracing_cpu(cpu) {
4610 buf_iter = trace_buffer_iter(iter, cpu);
4612 if (!ring_buffer_iter_empty(buf_iter))
4615 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4623 /* Called with trace_event_read_lock() held. */
4624 enum print_line_t print_trace_line(struct trace_iterator *iter)
4626 struct trace_array *tr = iter->tr;
4627 unsigned long trace_flags = tr->trace_flags;
4628 enum print_line_t ret;
4630 if (iter->lost_events) {
4631 if (iter->lost_events == (unsigned long)-1)
4632 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4635 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4636 iter->cpu, iter->lost_events);
4637 if (trace_seq_has_overflowed(&iter->seq))
4638 return TRACE_TYPE_PARTIAL_LINE;
4641 if (iter->trace && iter->trace->print_line) {
4642 ret = iter->trace->print_line(iter);
4643 if (ret != TRACE_TYPE_UNHANDLED)
4647 if (iter->ent->type == TRACE_BPUTS &&
4648 trace_flags & TRACE_ITER_PRINTK &&
4649 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4650 return trace_print_bputs_msg_only(iter);
4652 if (iter->ent->type == TRACE_BPRINT &&
4653 trace_flags & TRACE_ITER_PRINTK &&
4654 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4655 return trace_print_bprintk_msg_only(iter);
4657 if (iter->ent->type == TRACE_PRINT &&
4658 trace_flags & TRACE_ITER_PRINTK &&
4659 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4660 return trace_print_printk_msg_only(iter);
4662 if (trace_flags & TRACE_ITER_BIN)
4663 return print_bin_fmt(iter);
4665 if (trace_flags & TRACE_ITER_HEX)
4666 return print_hex_fmt(iter);
4668 if (trace_flags & TRACE_ITER_RAW)
4669 return print_raw_fmt(iter);
4671 return print_trace_fmt(iter);
4674 void trace_latency_header(struct seq_file *m)
4676 struct trace_iterator *iter = m->private;
4677 struct trace_array *tr = iter->tr;
4679 /* print nothing if the buffers are empty */
4680 if (trace_empty(iter))
4683 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4684 print_trace_header(m, iter);
4686 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4687 print_lat_help_header(m);
4690 void trace_default_header(struct seq_file *m)
4692 struct trace_iterator *iter = m->private;
4693 struct trace_array *tr = iter->tr;
4694 unsigned long trace_flags = tr->trace_flags;
4696 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4699 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4700 /* print nothing if the buffers are empty */
4701 if (trace_empty(iter))
4703 print_trace_header(m, iter);
4704 if (!(trace_flags & TRACE_ITER_VERBOSE))
4705 print_lat_help_header(m);
4707 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4708 if (trace_flags & TRACE_ITER_IRQ_INFO)
4709 print_func_help_header_irq(iter->array_buffer,
4712 print_func_help_header(iter->array_buffer, m,
4718 static void test_ftrace_alive(struct seq_file *m)
4720 if (!ftrace_is_dead())
4722 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4723 "# MAY BE MISSING FUNCTION EVENTS\n");
4726 #ifdef CONFIG_TRACER_MAX_TRACE
4727 static void show_snapshot_main_help(struct seq_file *m)
4729 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4730 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4731 "# Takes a snapshot of the main buffer.\n"
4732 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4733 "# (Doesn't have to be '2' works with any number that\n"
4734 "# is not a '0' or '1')\n");
4737 static void show_snapshot_percpu_help(struct seq_file *m)
4739 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4740 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4741 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4742 "# Takes a snapshot of the main buffer for this cpu.\n");
4744 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4745 "# Must use main snapshot file to allocate.\n");
4747 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4748 "# (Doesn't have to be '2' works with any number that\n"
4749 "# is not a '0' or '1')\n");
4752 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4754 if (iter->tr->allocated_snapshot)
4755 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4757 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4759 seq_puts(m, "# Snapshot commands:\n");
4760 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4761 show_snapshot_main_help(m);
4763 show_snapshot_percpu_help(m);
4766 /* Should never be called */
4767 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4770 static int s_show(struct seq_file *m, void *v)
4772 struct trace_iterator *iter = v;
4775 if (iter->ent == NULL) {
4777 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4779 test_ftrace_alive(m);
4781 if (iter->snapshot && trace_empty(iter))
4782 print_snapshot_help(m, iter);
4783 else if (iter->trace && iter->trace->print_header)
4784 iter->trace->print_header(m);
4786 trace_default_header(m);
4788 } else if (iter->leftover) {
4790 * If we filled the seq_file buffer earlier, we
4791 * want to just show it now.
4793 ret = trace_print_seq(m, &iter->seq);
4795 /* ret should this time be zero, but you never know */
4796 iter->leftover = ret;
4799 print_trace_line(iter);
4800 ret = trace_print_seq(m, &iter->seq);
4802 * If we overflow the seq_file buffer, then it will
4803 * ask us for this data again at start up.
4805 * ret is 0 if seq_file write succeeded.
4808 iter->leftover = ret;
4815 * Should be used after trace_array_get(), trace_types_lock
4816 * ensures that i_cdev was already initialized.
4818 static inline int tracing_get_cpu(struct inode *inode)
4820 if (inode->i_cdev) /* See trace_create_cpu_file() */
4821 return (long)inode->i_cdev - 1;
4822 return RING_BUFFER_ALL_CPUS;
4825 static const struct seq_operations tracer_seq_ops = {
4832 static struct trace_iterator *
4833 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4835 struct trace_array *tr = inode->i_private;
4836 struct trace_iterator *iter;
4839 if (tracing_disabled)
4840 return ERR_PTR(-ENODEV);
4842 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4844 return ERR_PTR(-ENOMEM);
4846 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4848 if (!iter->buffer_iter)
4852 * trace_find_next_entry() may need to save off iter->ent.
4853 * It will place it into the iter->temp buffer. As most
4854 * events are less than 128, allocate a buffer of that size.
4855 * If one is greater, then trace_find_next_entry() will
4856 * allocate a new buffer to adjust for the bigger iter->ent.
4857 * It's not critical if it fails to get allocated here.
4859 iter->temp = kmalloc(128, GFP_KERNEL);
4861 iter->temp_size = 128;
4864 * trace_event_printf() may need to modify given format
4865 * string to replace %p with %px so that it shows real address
4866 * instead of hash value. However, that is only for the event
4867 * tracing, other tracer may not need. Defer the allocation
4868 * until it is needed.
4874 * We make a copy of the current tracer to avoid concurrent
4875 * changes on it while we are reading.
4877 mutex_lock(&trace_types_lock);
4878 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4882 *iter->trace = *tr->current_trace;
4884 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4889 #ifdef CONFIG_TRACER_MAX_TRACE
4890 /* Currently only the top directory has a snapshot */
4891 if (tr->current_trace->print_max || snapshot)
4892 iter->array_buffer = &tr->max_buffer;
4895 iter->array_buffer = &tr->array_buffer;
4896 iter->snapshot = snapshot;
4898 iter->cpu_file = tracing_get_cpu(inode);
4899 mutex_init(&iter->mutex);
4901 /* Notify the tracer early; before we stop tracing. */
4902 if (iter->trace->open)
4903 iter->trace->open(iter);
4905 /* Annotate start of buffers if we had overruns */
4906 if (ring_buffer_overruns(iter->array_buffer->buffer))
4907 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4909 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4910 if (trace_clocks[tr->clock_id].in_ns)
4911 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4914 * If pause-on-trace is enabled, then stop the trace while
4915 * dumping, unless this is the "snapshot" file
4917 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4918 tracing_stop_tr(tr);
4920 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4921 for_each_tracing_cpu(cpu) {
4922 iter->buffer_iter[cpu] =
4923 ring_buffer_read_prepare(iter->array_buffer->buffer,
4926 ring_buffer_read_prepare_sync();
4927 for_each_tracing_cpu(cpu) {
4928 ring_buffer_read_start(iter->buffer_iter[cpu]);
4929 tracing_iter_reset(iter, cpu);
4932 cpu = iter->cpu_file;
4933 iter->buffer_iter[cpu] =
4934 ring_buffer_read_prepare(iter->array_buffer->buffer,
4936 ring_buffer_read_prepare_sync();
4937 ring_buffer_read_start(iter->buffer_iter[cpu]);
4938 tracing_iter_reset(iter, cpu);
4941 mutex_unlock(&trace_types_lock);
4946 mutex_unlock(&trace_types_lock);
4949 kfree(iter->buffer_iter);
4951 seq_release_private(inode, file);
4952 return ERR_PTR(-ENOMEM);
4955 int tracing_open_generic(struct inode *inode, struct file *filp)
4959 ret = tracing_check_open_get_tr(NULL);
4963 filp->private_data = inode->i_private;
4967 bool tracing_is_disabled(void)
4969 return (tracing_disabled) ? true: false;
4973 * Open and update trace_array ref count.
4974 * Must have the current trace_array passed to it.
4976 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4978 struct trace_array *tr = inode->i_private;
4981 ret = tracing_check_open_get_tr(tr);
4985 filp->private_data = inode->i_private;
4990 static int tracing_mark_open(struct inode *inode, struct file *filp)
4992 stream_open(inode, filp);
4993 return tracing_open_generic_tr(inode, filp);
4996 static int tracing_release(struct inode *inode, struct file *file)
4998 struct trace_array *tr = inode->i_private;
4999 struct seq_file *m = file->private_data;
5000 struct trace_iterator *iter;
5003 if (!(file->f_mode & FMODE_READ)) {
5004 trace_array_put(tr);
5008 /* Writes do not use seq_file */
5010 mutex_lock(&trace_types_lock);
5012 for_each_tracing_cpu(cpu) {
5013 if (iter->buffer_iter[cpu])
5014 ring_buffer_read_finish(iter->buffer_iter[cpu]);
5017 if (iter->trace && iter->trace->close)
5018 iter->trace->close(iter);
5020 if (!iter->snapshot && tr->stop_count)
5021 /* reenable tracing if it was previously enabled */
5022 tracing_start_tr(tr);
5024 __trace_array_put(tr);
5026 mutex_unlock(&trace_types_lock);
5028 mutex_destroy(&iter->mutex);
5029 free_cpumask_var(iter->started);
5033 kfree(iter->buffer_iter);
5034 seq_release_private(inode, file);
5039 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
5041 struct trace_array *tr = inode->i_private;
5043 trace_array_put(tr);
5047 static int tracing_single_release_tr(struct inode *inode, struct file *file)
5049 struct trace_array *tr = inode->i_private;
5051 trace_array_put(tr);
5053 return single_release(inode, file);
5056 static int tracing_open(struct inode *inode, struct file *file)
5058 struct trace_array *tr = inode->i_private;
5059 struct trace_iterator *iter;
5062 ret = tracing_check_open_get_tr(tr);
5066 /* If this file was open for write, then erase contents */
5067 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
5068 int cpu = tracing_get_cpu(inode);
5069 struct array_buffer *trace_buf = &tr->array_buffer;
5071 #ifdef CONFIG_TRACER_MAX_TRACE
5072 if (tr->current_trace->print_max)
5073 trace_buf = &tr->max_buffer;
5076 if (cpu == RING_BUFFER_ALL_CPUS)
5077 tracing_reset_online_cpus(trace_buf);
5079 tracing_reset_cpu(trace_buf, cpu);
5082 if (file->f_mode & FMODE_READ) {
5083 iter = __tracing_open(inode, file, false);
5085 ret = PTR_ERR(iter);
5086 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
5087 iter->iter_flags |= TRACE_FILE_LAT_FMT;
5091 trace_array_put(tr);
5097 * Some tracers are not suitable for instance buffers.
5098 * A tracer is always available for the global array (toplevel)
5099 * or if it explicitly states that it is.
5102 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
5104 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5107 /* Find the next tracer that this trace array may use */
5108 static struct tracer *
5109 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5111 while (t && !trace_ok_for_array(t, tr))
5118 t_next(struct seq_file *m, void *v, loff_t *pos)
5120 struct trace_array *tr = m->private;
5121 struct tracer *t = v;
5126 t = get_tracer_for_array(tr, t->next);
5131 static void *t_start(struct seq_file *m, loff_t *pos)
5133 struct trace_array *tr = m->private;
5137 mutex_lock(&trace_types_lock);
5139 t = get_tracer_for_array(tr, trace_types);
5140 for (; t && l < *pos; t = t_next(m, t, &l))
5146 static void t_stop(struct seq_file *m, void *p)
5148 mutex_unlock(&trace_types_lock);
5151 static int t_show(struct seq_file *m, void *v)
5153 struct tracer *t = v;
5158 seq_puts(m, t->name);
5167 static const struct seq_operations show_traces_seq_ops = {
5174 static int show_traces_open(struct inode *inode, struct file *file)
5176 struct trace_array *tr = inode->i_private;
5180 ret = tracing_check_open_get_tr(tr);
5184 ret = seq_open(file, &show_traces_seq_ops);
5186 trace_array_put(tr);
5190 m = file->private_data;
5196 static int show_traces_release(struct inode *inode, struct file *file)
5198 struct trace_array *tr = inode->i_private;
5200 trace_array_put(tr);
5201 return seq_release(inode, file);
5205 tracing_write_stub(struct file *filp, const char __user *ubuf,
5206 size_t count, loff_t *ppos)
5211 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5215 if (file->f_mode & FMODE_READ)
5216 ret = seq_lseek(file, offset, whence);
5218 file->f_pos = ret = 0;
5223 static const struct file_operations tracing_fops = {
5224 .open = tracing_open,
5226 .read_iter = seq_read_iter,
5227 .splice_read = copy_splice_read,
5228 .write = tracing_write_stub,
5229 .llseek = tracing_lseek,
5230 .release = tracing_release,
5233 static const struct file_operations show_traces_fops = {
5234 .open = show_traces_open,
5236 .llseek = seq_lseek,
5237 .release = show_traces_release,
5241 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5242 size_t count, loff_t *ppos)
5244 struct trace_array *tr = file_inode(filp)->i_private;
5248 len = snprintf(NULL, 0, "%*pb\n",
5249 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5250 mask_str = kmalloc(len, GFP_KERNEL);
5254 len = snprintf(mask_str, len, "%*pb\n",
5255 cpumask_pr_args(tr->tracing_cpumask));
5260 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5268 int tracing_set_cpumask(struct trace_array *tr,
5269 cpumask_var_t tracing_cpumask_new)
5276 local_irq_disable();
5277 arch_spin_lock(&tr->max_lock);
5278 for_each_tracing_cpu(cpu) {
5280 * Increase/decrease the disabled counter if we are
5281 * about to flip a bit in the cpumask:
5283 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5284 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5285 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5286 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5287 #ifdef CONFIG_TRACER_MAX_TRACE
5288 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5291 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5292 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5293 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5294 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5295 #ifdef CONFIG_TRACER_MAX_TRACE
5296 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5300 arch_spin_unlock(&tr->max_lock);
5303 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5309 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5310 size_t count, loff_t *ppos)
5312 struct trace_array *tr = file_inode(filp)->i_private;
5313 cpumask_var_t tracing_cpumask_new;
5316 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5319 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5323 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5327 free_cpumask_var(tracing_cpumask_new);
5332 free_cpumask_var(tracing_cpumask_new);
5337 static const struct file_operations tracing_cpumask_fops = {
5338 .open = tracing_open_generic_tr,
5339 .read = tracing_cpumask_read,
5340 .write = tracing_cpumask_write,
5341 .release = tracing_release_generic_tr,
5342 .llseek = generic_file_llseek,
5345 static int tracing_trace_options_show(struct seq_file *m, void *v)
5347 struct tracer_opt *trace_opts;
5348 struct trace_array *tr = m->private;
5352 mutex_lock(&trace_types_lock);
5353 tracer_flags = tr->current_trace->flags->val;
5354 trace_opts = tr->current_trace->flags->opts;
5356 for (i = 0; trace_options[i]; i++) {
5357 if (tr->trace_flags & (1 << i))
5358 seq_printf(m, "%s\n", trace_options[i]);
5360 seq_printf(m, "no%s\n", trace_options[i]);
5363 for (i = 0; trace_opts[i].name; i++) {
5364 if (tracer_flags & trace_opts[i].bit)
5365 seq_printf(m, "%s\n", trace_opts[i].name);
5367 seq_printf(m, "no%s\n", trace_opts[i].name);
5369 mutex_unlock(&trace_types_lock);
5374 static int __set_tracer_option(struct trace_array *tr,
5375 struct tracer_flags *tracer_flags,
5376 struct tracer_opt *opts, int neg)
5378 struct tracer *trace = tracer_flags->trace;
5381 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5386 tracer_flags->val &= ~opts->bit;
5388 tracer_flags->val |= opts->bit;
5392 /* Try to assign a tracer specific option */
5393 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5395 struct tracer *trace = tr->current_trace;
5396 struct tracer_flags *tracer_flags = trace->flags;
5397 struct tracer_opt *opts = NULL;
5400 for (i = 0; tracer_flags->opts[i].name; i++) {
5401 opts = &tracer_flags->opts[i];
5403 if (strcmp(cmp, opts->name) == 0)
5404 return __set_tracer_option(tr, trace->flags, opts, neg);
5410 /* Some tracers require overwrite to stay enabled */
5411 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5413 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5419 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5423 if ((mask == TRACE_ITER_RECORD_TGID) ||
5424 (mask == TRACE_ITER_RECORD_CMD))
5425 lockdep_assert_held(&event_mutex);
5427 /* do nothing if flag is already set */
5428 if (!!(tr->trace_flags & mask) == !!enabled)
5431 /* Give the tracer a chance to approve the change */
5432 if (tr->current_trace->flag_changed)
5433 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5437 tr->trace_flags |= mask;
5439 tr->trace_flags &= ~mask;
5441 if (mask == TRACE_ITER_RECORD_CMD)
5442 trace_event_enable_cmd_record(enabled);
5444 if (mask == TRACE_ITER_RECORD_TGID) {
5446 tgid_map_max = pid_max;
5447 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5451 * Pairs with smp_load_acquire() in
5452 * trace_find_tgid_ptr() to ensure that if it observes
5453 * the tgid_map we just allocated then it also observes
5454 * the corresponding tgid_map_max value.
5456 smp_store_release(&tgid_map, map);
5459 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5463 trace_event_enable_tgid_record(enabled);
5466 if (mask == TRACE_ITER_EVENT_FORK)
5467 trace_event_follow_fork(tr, enabled);
5469 if (mask == TRACE_ITER_FUNC_FORK)
5470 ftrace_pid_follow_fork(tr, enabled);
5472 if (mask == TRACE_ITER_OVERWRITE) {
5473 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5474 #ifdef CONFIG_TRACER_MAX_TRACE
5475 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5479 if (mask == TRACE_ITER_PRINTK) {
5480 trace_printk_start_stop_comm(enabled);
5481 trace_printk_control(enabled);
5487 int trace_set_options(struct trace_array *tr, char *option)
5492 size_t orig_len = strlen(option);
5495 cmp = strstrip(option);
5497 len = str_has_prefix(cmp, "no");
5503 mutex_lock(&event_mutex);
5504 mutex_lock(&trace_types_lock);
5506 ret = match_string(trace_options, -1, cmp);
5507 /* If no option could be set, test the specific tracer options */
5509 ret = set_tracer_option(tr, cmp, neg);
5511 ret = set_tracer_flag(tr, 1 << ret, !neg);
5513 mutex_unlock(&trace_types_lock);
5514 mutex_unlock(&event_mutex);
5517 * If the first trailing whitespace is replaced with '\0' by strstrip,
5518 * turn it back into a space.
5520 if (orig_len > strlen(option))
5521 option[strlen(option)] = ' ';
5526 static void __init apply_trace_boot_options(void)
5528 char *buf = trace_boot_options_buf;
5532 option = strsep(&buf, ",");
5538 trace_set_options(&global_trace, option);
5540 /* Put back the comma to allow this to be called again */
5547 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5548 size_t cnt, loff_t *ppos)
5550 struct seq_file *m = filp->private_data;
5551 struct trace_array *tr = m->private;
5555 if (cnt >= sizeof(buf))
5558 if (copy_from_user(buf, ubuf, cnt))
5563 ret = trace_set_options(tr, buf);
5572 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5574 struct trace_array *tr = inode->i_private;
5577 ret = tracing_check_open_get_tr(tr);
5581 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5583 trace_array_put(tr);
5588 static const struct file_operations tracing_iter_fops = {
5589 .open = tracing_trace_options_open,
5591 .llseek = seq_lseek,
5592 .release = tracing_single_release_tr,
5593 .write = tracing_trace_options_write,
5596 static const char readme_msg[] =
5597 "tracing mini-HOWTO:\n\n"
5598 "# echo 0 > tracing_on : quick way to disable tracing\n"
5599 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5600 " Important files:\n"
5601 " trace\t\t\t- The static contents of the buffer\n"
5602 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5603 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5604 " current_tracer\t- function and latency tracers\n"
5605 " available_tracers\t- list of configured tracers for current_tracer\n"
5606 " error_log\t- error log for failed commands (that support it)\n"
5607 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5608 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5609 " trace_clock\t\t- change the clock used to order events\n"
5610 " local: Per cpu clock but may not be synced across CPUs\n"
5611 " global: Synced across CPUs but slows tracing down.\n"
5612 " counter: Not a clock, but just an increment\n"
5613 " uptime: Jiffy counter from time of boot\n"
5614 " perf: Same clock that perf events use\n"
5615 #ifdef CONFIG_X86_64
5616 " x86-tsc: TSC cycle counter\n"
5618 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5619 " delta: Delta difference against a buffer-wide timestamp\n"
5620 " absolute: Absolute (standalone) timestamp\n"
5621 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5622 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5623 " tracing_cpumask\t- Limit which CPUs to trace\n"
5624 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5625 "\t\t\t Remove sub-buffer with rmdir\n"
5626 " trace_options\t\t- Set format or modify how tracing happens\n"
5627 "\t\t\t Disable an option by prefixing 'no' to the\n"
5628 "\t\t\t option name\n"
5629 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5630 #ifdef CONFIG_DYNAMIC_FTRACE
5631 "\n available_filter_functions - list of functions that can be filtered on\n"
5632 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5633 "\t\t\t functions\n"
5634 "\t accepts: func_full_name or glob-matching-pattern\n"
5635 "\t modules: Can select a group via module\n"
5636 "\t Format: :mod:<module-name>\n"
5637 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5638 "\t triggers: a command to perform when function is hit\n"
5639 "\t Format: <function>:<trigger>[:count]\n"
5640 "\t trigger: traceon, traceoff\n"
5641 "\t\t enable_event:<system>:<event>\n"
5642 "\t\t disable_event:<system>:<event>\n"
5643 #ifdef CONFIG_STACKTRACE
5646 #ifdef CONFIG_TRACER_SNAPSHOT
5651 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5652 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5653 "\t The first one will disable tracing every time do_fault is hit\n"
5654 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5655 "\t The first time do trap is hit and it disables tracing, the\n"
5656 "\t counter will decrement to 2. If tracing is already disabled,\n"
5657 "\t the counter will not decrement. It only decrements when the\n"
5658 "\t trigger did work\n"
5659 "\t To remove trigger without count:\n"
5660 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5661 "\t To remove trigger with a count:\n"
5662 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5663 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5664 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5665 "\t modules: Can select a group via module command :mod:\n"
5666 "\t Does not accept triggers\n"
5667 #endif /* CONFIG_DYNAMIC_FTRACE */
5668 #ifdef CONFIG_FUNCTION_TRACER
5669 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5671 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5674 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5675 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5676 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5677 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5679 #ifdef CONFIG_TRACER_SNAPSHOT
5680 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5681 "\t\t\t snapshot buffer. Read the contents for more\n"
5682 "\t\t\t information\n"
5684 #ifdef CONFIG_STACK_TRACER
5685 " stack_trace\t\t- Shows the max stack trace when active\n"
5686 " stack_max_size\t- Shows current max stack size that was traced\n"
5687 "\t\t\t Write into this file to reset the max size (trigger a\n"
5688 "\t\t\t new trace)\n"
5689 #ifdef CONFIG_DYNAMIC_FTRACE
5690 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5693 #endif /* CONFIG_STACK_TRACER */
5694 #ifdef CONFIG_DYNAMIC_EVENTS
5695 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5696 "\t\t\t Write into this file to define/undefine new trace events.\n"
5698 #ifdef CONFIG_KPROBE_EVENTS
5699 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5700 "\t\t\t Write into this file to define/undefine new trace events.\n"
5702 #ifdef CONFIG_UPROBE_EVENTS
5703 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5704 "\t\t\t Write into this file to define/undefine new trace events.\n"
5706 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5707 defined(CONFIG_FPROBE_EVENTS)
5708 "\t accepts: event-definitions (one definition per line)\n"
5709 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5710 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5711 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5713 #ifdef CONFIG_FPROBE_EVENTS
5714 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5715 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5717 #ifdef CONFIG_HIST_TRIGGERS
5718 "\t s:[synthetic/]<event> <field> [<field>]\n"
5720 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5721 "\t -:[<group>/][<event>]\n"
5722 #ifdef CONFIG_KPROBE_EVENTS
5723 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5724 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5726 #ifdef CONFIG_UPROBE_EVENTS
5727 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5729 "\t args: <name>=fetcharg[:type]\n"
5730 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5731 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5732 #ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5733 "\t $stack<index>, $stack, $retval, $comm, $arg<N>, <argname>\n"
5735 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5738 "\t $stack<index>, $stack, $retval, $comm,\n"
5740 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5741 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5742 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5743 "\t symstr, <type>\\[<array-size>\\]\n"
5744 #ifdef CONFIG_HIST_TRIGGERS
5745 "\t field: <stype> <name>;\n"
5746 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5747 "\t [unsigned] char/int/long\n"
5749 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5750 "\t of the <attached-group>/<attached-event>.\n"
5752 " events/\t\t- Directory containing all trace event subsystems:\n"
5753 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5754 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5755 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5757 " filter\t\t- If set, only events passing filter are traced\n"
5758 " events/<system>/<event>/\t- Directory containing control files for\n"
5760 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5761 " filter\t\t- If set, only events passing filter are traced\n"
5762 " trigger\t\t- If set, a command to perform when event is hit\n"
5763 "\t Format: <trigger>[:count][if <filter>]\n"
5764 "\t trigger: traceon, traceoff\n"
5765 "\t enable_event:<system>:<event>\n"
5766 "\t disable_event:<system>:<event>\n"
5767 #ifdef CONFIG_HIST_TRIGGERS
5768 "\t enable_hist:<system>:<event>\n"
5769 "\t disable_hist:<system>:<event>\n"
5771 #ifdef CONFIG_STACKTRACE
5774 #ifdef CONFIG_TRACER_SNAPSHOT
5777 #ifdef CONFIG_HIST_TRIGGERS
5778 "\t\t hist (see below)\n"
5780 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5781 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5782 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5783 "\t events/block/block_unplug/trigger\n"
5784 "\t The first disables tracing every time block_unplug is hit.\n"
5785 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5786 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5787 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5788 "\t Like function triggers, the counter is only decremented if it\n"
5789 "\t enabled or disabled tracing.\n"
5790 "\t To remove a trigger without a count:\n"
5791 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5792 "\t To remove a trigger with a count:\n"
5793 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5794 "\t Filters can be ignored when removing a trigger.\n"
5795 #ifdef CONFIG_HIST_TRIGGERS
5796 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5797 "\t Format: hist:keys=<field1[,field2,...]>\n"
5798 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5799 "\t [:values=<field1[,field2,...]>]\n"
5800 "\t [:sort=<field1[,field2,...]>]\n"
5801 "\t [:size=#entries]\n"
5802 "\t [:pause][:continue][:clear]\n"
5803 "\t [:name=histname1]\n"
5804 "\t [:nohitcount]\n"
5805 "\t [:<handler>.<action>]\n"
5806 "\t [if <filter>]\n\n"
5807 "\t Note, special fields can be used as well:\n"
5808 "\t common_timestamp - to record current timestamp\n"
5809 "\t common_cpu - to record the CPU the event happened on\n"
5811 "\t A hist trigger variable can be:\n"
5812 "\t - a reference to a field e.g. x=current_timestamp,\n"
5813 "\t - a reference to another variable e.g. y=$x,\n"
5814 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5815 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5817 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5818 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5819 "\t variable reference, field or numeric literal.\n"
5821 "\t When a matching event is hit, an entry is added to a hash\n"
5822 "\t table using the key(s) and value(s) named, and the value of a\n"
5823 "\t sum called 'hitcount' is incremented. Keys and values\n"
5824 "\t correspond to fields in the event's format description. Keys\n"
5825 "\t can be any field, or the special string 'common_stacktrace'.\n"
5826 "\t Compound keys consisting of up to two fields can be specified\n"
5827 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5828 "\t fields. Sort keys consisting of up to two fields can be\n"
5829 "\t specified using the 'sort' keyword. The sort direction can\n"
5830 "\t be modified by appending '.descending' or '.ascending' to a\n"
5831 "\t sort field. The 'size' parameter can be used to specify more\n"
5832 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5833 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5834 "\t its histogram data will be shared with other triggers of the\n"
5835 "\t same name, and trigger hits will update this common data.\n\n"
5836 "\t Reading the 'hist' file for the event will dump the hash\n"
5837 "\t table in its entirety to stdout. If there are multiple hist\n"
5838 "\t triggers attached to an event, there will be a table for each\n"
5839 "\t trigger in the output. The table displayed for a named\n"
5840 "\t trigger will be the same as any other instance having the\n"
5841 "\t same name. The default format used to display a given field\n"
5842 "\t can be modified by appending any of the following modifiers\n"
5843 "\t to the field name, as applicable:\n\n"
5844 "\t .hex display a number as a hex value\n"
5845 "\t .sym display an address as a symbol\n"
5846 "\t .sym-offset display an address as a symbol and offset\n"
5847 "\t .execname display a common_pid as a program name\n"
5848 "\t .syscall display a syscall id as a syscall name\n"
5849 "\t .log2 display log2 value rather than raw number\n"
5850 "\t .buckets=size display values in groups of size rather than raw number\n"
5851 "\t .usecs display a common_timestamp in microseconds\n"
5852 "\t .percent display a number of percentage value\n"
5853 "\t .graph display a bar-graph of a value\n\n"
5854 "\t The 'pause' parameter can be used to pause an existing hist\n"
5855 "\t trigger or to start a hist trigger but not log any events\n"
5856 "\t until told to do so. 'continue' can be used to start or\n"
5857 "\t restart a paused hist trigger.\n\n"
5858 "\t The 'clear' parameter will clear the contents of a running\n"
5859 "\t hist trigger and leave its current paused/active state\n"
5861 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5862 "\t raw hitcount in the histogram.\n\n"
5863 "\t The enable_hist and disable_hist triggers can be used to\n"
5864 "\t have one event conditionally start and stop another event's\n"
5865 "\t already-attached hist trigger. The syntax is analogous to\n"
5866 "\t the enable_event and disable_event triggers.\n\n"
5867 "\t Hist trigger handlers and actions are executed whenever a\n"
5868 "\t a histogram entry is added or updated. They take the form:\n\n"
5869 "\t <handler>.<action>\n\n"
5870 "\t The available handlers are:\n\n"
5871 "\t onmatch(matching.event) - invoke on addition or update\n"
5872 "\t onmax(var) - invoke if var exceeds current max\n"
5873 "\t onchange(var) - invoke action if var changes\n\n"
5874 "\t The available actions are:\n\n"
5875 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5876 "\t save(field,...) - save current event fields\n"
5877 #ifdef CONFIG_TRACER_SNAPSHOT
5878 "\t snapshot() - snapshot the trace buffer\n\n"
5880 #ifdef CONFIG_SYNTH_EVENTS
5881 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5882 "\t Write into this file to define/undefine new synthetic events.\n"
5883 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5889 tracing_readme_read(struct file *filp, char __user *ubuf,
5890 size_t cnt, loff_t *ppos)
5892 return simple_read_from_buffer(ubuf, cnt, ppos,
5893 readme_msg, strlen(readme_msg));
5896 static const struct file_operations tracing_readme_fops = {
5897 .open = tracing_open_generic,
5898 .read = tracing_readme_read,
5899 .llseek = generic_file_llseek,
5902 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5906 return trace_find_tgid_ptr(pid);
5909 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5913 return trace_find_tgid_ptr(pid);
5916 static void saved_tgids_stop(struct seq_file *m, void *v)
5920 static int saved_tgids_show(struct seq_file *m, void *v)
5922 int *entry = (int *)v;
5923 int pid = entry - tgid_map;
5929 seq_printf(m, "%d %d\n", pid, tgid);
5933 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5934 .start = saved_tgids_start,
5935 .stop = saved_tgids_stop,
5936 .next = saved_tgids_next,
5937 .show = saved_tgids_show,
5940 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5944 ret = tracing_check_open_get_tr(NULL);
5948 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5952 static const struct file_operations tracing_saved_tgids_fops = {
5953 .open = tracing_saved_tgids_open,
5955 .llseek = seq_lseek,
5956 .release = seq_release,
5959 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5961 unsigned int *ptr = v;
5963 if (*pos || m->count)
5968 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5970 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5979 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5985 arch_spin_lock(&trace_cmdline_lock);
5987 v = &savedcmd->map_cmdline_to_pid[0];
5989 v = saved_cmdlines_next(m, v, &l);
5997 static void saved_cmdlines_stop(struct seq_file *m, void *v)
5999 arch_spin_unlock(&trace_cmdline_lock);
6003 static int saved_cmdlines_show(struct seq_file *m, void *v)
6005 char buf[TASK_COMM_LEN];
6006 unsigned int *pid = v;
6008 __trace_find_cmdline(*pid, buf);
6009 seq_printf(m, "%d %s\n", *pid, buf);
6013 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
6014 .start = saved_cmdlines_start,
6015 .next = saved_cmdlines_next,
6016 .stop = saved_cmdlines_stop,
6017 .show = saved_cmdlines_show,
6020 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
6024 ret = tracing_check_open_get_tr(NULL);
6028 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
6031 static const struct file_operations tracing_saved_cmdlines_fops = {
6032 .open = tracing_saved_cmdlines_open,
6034 .llseek = seq_lseek,
6035 .release = seq_release,
6039 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
6040 size_t cnt, loff_t *ppos)
6046 arch_spin_lock(&trace_cmdline_lock);
6047 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
6048 arch_spin_unlock(&trace_cmdline_lock);
6051 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6054 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
6056 kfree(s->saved_cmdlines);
6057 kfree(s->map_cmdline_to_pid);
6061 static int tracing_resize_saved_cmdlines(unsigned int val)
6063 struct saved_cmdlines_buffer *s, *savedcmd_temp;
6065 s = kmalloc(sizeof(*s), GFP_KERNEL);
6069 if (allocate_cmdlines_buffer(val, s) < 0) {
6075 arch_spin_lock(&trace_cmdline_lock);
6076 savedcmd_temp = savedcmd;
6078 arch_spin_unlock(&trace_cmdline_lock);
6080 free_saved_cmdlines_buffer(savedcmd_temp);
6086 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
6087 size_t cnt, loff_t *ppos)
6092 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6096 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
6097 if (!val || val > PID_MAX_DEFAULT)
6100 ret = tracing_resize_saved_cmdlines((unsigned int)val);
6109 static const struct file_operations tracing_saved_cmdlines_size_fops = {
6110 .open = tracing_open_generic,
6111 .read = tracing_saved_cmdlines_size_read,
6112 .write = tracing_saved_cmdlines_size_write,
6115 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
6116 static union trace_eval_map_item *
6117 update_eval_map(union trace_eval_map_item *ptr)
6119 if (!ptr->map.eval_string) {
6120 if (ptr->tail.next) {
6121 ptr = ptr->tail.next;
6122 /* Set ptr to the next real item (skip head) */
6130 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6132 union trace_eval_map_item *ptr = v;
6135 * Paranoid! If ptr points to end, we don't want to increment past it.
6136 * This really should never happen.
6139 ptr = update_eval_map(ptr);
6140 if (WARN_ON_ONCE(!ptr))
6144 ptr = update_eval_map(ptr);
6149 static void *eval_map_start(struct seq_file *m, loff_t *pos)
6151 union trace_eval_map_item *v;
6154 mutex_lock(&trace_eval_mutex);
6156 v = trace_eval_maps;
6160 while (v && l < *pos) {
6161 v = eval_map_next(m, v, &l);
6167 static void eval_map_stop(struct seq_file *m, void *v)
6169 mutex_unlock(&trace_eval_mutex);
6172 static int eval_map_show(struct seq_file *m, void *v)
6174 union trace_eval_map_item *ptr = v;
6176 seq_printf(m, "%s %ld (%s)\n",
6177 ptr->map.eval_string, ptr->map.eval_value,
6183 static const struct seq_operations tracing_eval_map_seq_ops = {
6184 .start = eval_map_start,
6185 .next = eval_map_next,
6186 .stop = eval_map_stop,
6187 .show = eval_map_show,
6190 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6194 ret = tracing_check_open_get_tr(NULL);
6198 return seq_open(filp, &tracing_eval_map_seq_ops);
6201 static const struct file_operations tracing_eval_map_fops = {
6202 .open = tracing_eval_map_open,
6204 .llseek = seq_lseek,
6205 .release = seq_release,
6208 static inline union trace_eval_map_item *
6209 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6211 /* Return tail of array given the head */
6212 return ptr + ptr->head.length + 1;
6216 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6219 struct trace_eval_map **stop;
6220 struct trace_eval_map **map;
6221 union trace_eval_map_item *map_array;
6222 union trace_eval_map_item *ptr;
6227 * The trace_eval_maps contains the map plus a head and tail item,
6228 * where the head holds the module and length of array, and the
6229 * tail holds a pointer to the next list.
6231 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6233 pr_warn("Unable to allocate trace eval mapping\n");
6237 mutex_lock(&trace_eval_mutex);
6239 if (!trace_eval_maps)
6240 trace_eval_maps = map_array;
6242 ptr = trace_eval_maps;
6244 ptr = trace_eval_jmp_to_tail(ptr);
6245 if (!ptr->tail.next)
6247 ptr = ptr->tail.next;
6250 ptr->tail.next = map_array;
6252 map_array->head.mod = mod;
6253 map_array->head.length = len;
6256 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6257 map_array->map = **map;
6260 memset(map_array, 0, sizeof(*map_array));
6262 mutex_unlock(&trace_eval_mutex);
6265 static void trace_create_eval_file(struct dentry *d_tracer)
6267 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6268 NULL, &tracing_eval_map_fops);
6271 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
6272 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6273 static inline void trace_insert_eval_map_file(struct module *mod,
6274 struct trace_eval_map **start, int len) { }
6275 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6277 static void trace_insert_eval_map(struct module *mod,
6278 struct trace_eval_map **start, int len)
6280 struct trace_eval_map **map;
6287 trace_event_eval_update(map, len);
6289 trace_insert_eval_map_file(mod, start, len);
6293 tracing_set_trace_read(struct file *filp, char __user *ubuf,
6294 size_t cnt, loff_t *ppos)
6296 struct trace_array *tr = filp->private_data;
6297 char buf[MAX_TRACER_SIZE+2];
6300 mutex_lock(&trace_types_lock);
6301 r = sprintf(buf, "%s\n", tr->current_trace->name);
6302 mutex_unlock(&trace_types_lock);
6304 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6307 int tracer_init(struct tracer *t, struct trace_array *tr)
6309 tracing_reset_online_cpus(&tr->array_buffer);
6313 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6317 for_each_tracing_cpu(cpu)
6318 per_cpu_ptr(buf->data, cpu)->entries = val;
6321 #ifdef CONFIG_TRACER_MAX_TRACE
6322 /* resize @tr's buffer to the size of @size_tr's entries */
6323 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6324 struct array_buffer *size_buf, int cpu_id)
6328 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6329 for_each_tracing_cpu(cpu) {
6330 ret = ring_buffer_resize(trace_buf->buffer,
6331 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6334 per_cpu_ptr(trace_buf->data, cpu)->entries =
6335 per_cpu_ptr(size_buf->data, cpu)->entries;
6338 ret = ring_buffer_resize(trace_buf->buffer,
6339 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6341 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6342 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6347 #endif /* CONFIG_TRACER_MAX_TRACE */
6349 static int __tracing_resize_ring_buffer(struct trace_array *tr,
6350 unsigned long size, int cpu)
6355 * If kernel or user changes the size of the ring buffer
6356 * we use the size that was given, and we can forget about
6357 * expanding it later.
6359 ring_buffer_expanded = true;
6361 /* May be called before buffers are initialized */
6362 if (!tr->array_buffer.buffer)
6365 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6369 #ifdef CONFIG_TRACER_MAX_TRACE
6370 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
6371 !tr->current_trace->use_max_tr)
6374 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6376 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6377 &tr->array_buffer, cpu);
6380 * AARGH! We are left with different
6381 * size max buffer!!!!
6382 * The max buffer is our "snapshot" buffer.
6383 * When a tracer needs a snapshot (one of the
6384 * latency tracers), it swaps the max buffer
6385 * with the saved snap shot. We succeeded to
6386 * update the size of the main buffer, but failed to
6387 * update the size of the max buffer. But when we tried
6388 * to reset the main buffer to the original size, we
6389 * failed there too. This is very unlikely to
6390 * happen, but if it does, warn and kill all
6394 tracing_disabled = 1;
6399 if (cpu == RING_BUFFER_ALL_CPUS)
6400 set_buffer_entries(&tr->max_buffer, size);
6402 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
6405 #endif /* CONFIG_TRACER_MAX_TRACE */
6407 if (cpu == RING_BUFFER_ALL_CPUS)
6408 set_buffer_entries(&tr->array_buffer, size);
6410 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
6415 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6416 unsigned long size, int cpu_id)
6420 mutex_lock(&trace_types_lock);
6422 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6423 /* make sure, this cpu is enabled in the mask */
6424 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6430 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6435 mutex_unlock(&trace_types_lock);
6442 * tracing_update_buffers - used by tracing facility to expand ring buffers
6444 * To save on memory when the tracing is never used on a system with it
6445 * configured in. The ring buffers are set to a minimum size. But once
6446 * a user starts to use the tracing facility, then they need to grow
6447 * to their default size.
6449 * This function is to be called when a tracer is about to be used.
6451 int tracing_update_buffers(void)
6455 mutex_lock(&trace_types_lock);
6456 if (!ring_buffer_expanded)
6457 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
6458 RING_BUFFER_ALL_CPUS);
6459 mutex_unlock(&trace_types_lock);
6464 struct trace_option_dentry;
6467 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6470 * Used to clear out the tracer before deletion of an instance.
6471 * Must have trace_types_lock held.
6473 static void tracing_set_nop(struct trace_array *tr)
6475 if (tr->current_trace == &nop_trace)
6478 tr->current_trace->enabled--;
6480 if (tr->current_trace->reset)
6481 tr->current_trace->reset(tr);
6483 tr->current_trace = &nop_trace;
6486 static bool tracer_options_updated;
6488 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6490 /* Only enable if the directory has been created already. */
6494 /* Only create trace option files after update_tracer_options finish */
6495 if (!tracer_options_updated)
6498 create_trace_option_files(tr, t);
6501 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6504 #ifdef CONFIG_TRACER_MAX_TRACE
6509 mutex_lock(&trace_types_lock);
6511 if (!ring_buffer_expanded) {
6512 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6513 RING_BUFFER_ALL_CPUS);
6519 for (t = trace_types; t; t = t->next) {
6520 if (strcmp(t->name, buf) == 0)
6527 if (t == tr->current_trace)
6530 #ifdef CONFIG_TRACER_SNAPSHOT
6531 if (t->use_max_tr) {
6532 local_irq_disable();
6533 arch_spin_lock(&tr->max_lock);
6534 if (tr->cond_snapshot)
6536 arch_spin_unlock(&tr->max_lock);
6542 /* Some tracers won't work on kernel command line */
6543 if (system_state < SYSTEM_RUNNING && t->noboot) {
6544 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6549 /* Some tracers are only allowed for the top level buffer */
6550 if (!trace_ok_for_array(t, tr)) {
6555 /* If trace pipe files are being read, we can't change the tracer */
6556 if (tr->trace_ref) {
6561 trace_branch_disable();
6563 tr->current_trace->enabled--;
6565 if (tr->current_trace->reset)
6566 tr->current_trace->reset(tr);
6568 #ifdef CONFIG_TRACER_MAX_TRACE
6569 had_max_tr = tr->current_trace->use_max_tr;
6571 /* Current trace needs to be nop_trace before synchronize_rcu */
6572 tr->current_trace = &nop_trace;
6574 if (had_max_tr && !t->use_max_tr) {
6576 * We need to make sure that the update_max_tr sees that
6577 * current_trace changed to nop_trace to keep it from
6578 * swapping the buffers after we resize it.
6579 * The update_max_tr is called from interrupts disabled
6580 * so a synchronized_sched() is sufficient.
6586 if (t->use_max_tr && !tr->allocated_snapshot) {
6587 ret = tracing_alloc_snapshot_instance(tr);
6592 tr->current_trace = &nop_trace;
6596 ret = tracer_init(t, tr);
6601 tr->current_trace = t;
6602 tr->current_trace->enabled++;
6603 trace_branch_enable(tr);
6605 mutex_unlock(&trace_types_lock);
6611 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6612 size_t cnt, loff_t *ppos)
6614 struct trace_array *tr = filp->private_data;
6615 char buf[MAX_TRACER_SIZE+1];
6622 if (cnt > MAX_TRACER_SIZE)
6623 cnt = MAX_TRACER_SIZE;
6625 if (copy_from_user(buf, ubuf, cnt))
6632 err = tracing_set_tracer(tr, name);
6642 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6643 size_t cnt, loff_t *ppos)
6648 r = snprintf(buf, sizeof(buf), "%ld\n",
6649 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6650 if (r > sizeof(buf))
6652 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6656 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6657 size_t cnt, loff_t *ppos)
6662 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6672 tracing_thresh_read(struct file *filp, char __user *ubuf,
6673 size_t cnt, loff_t *ppos)
6675 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6679 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6680 size_t cnt, loff_t *ppos)
6682 struct trace_array *tr = filp->private_data;
6685 mutex_lock(&trace_types_lock);
6686 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6690 if (tr->current_trace->update_thresh) {
6691 ret = tr->current_trace->update_thresh(tr);
6698 mutex_unlock(&trace_types_lock);
6703 #ifdef CONFIG_TRACER_MAX_TRACE
6706 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6707 size_t cnt, loff_t *ppos)
6709 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6713 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6714 size_t cnt, loff_t *ppos)
6716 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6721 static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6723 if (cpu == RING_BUFFER_ALL_CPUS) {
6724 if (cpumask_empty(tr->pipe_cpumask)) {
6725 cpumask_setall(tr->pipe_cpumask);
6728 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6729 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6735 static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6737 if (cpu == RING_BUFFER_ALL_CPUS) {
6738 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6739 cpumask_clear(tr->pipe_cpumask);
6741 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6742 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6746 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6748 struct trace_array *tr = inode->i_private;
6749 struct trace_iterator *iter;
6753 ret = tracing_check_open_get_tr(tr);
6757 mutex_lock(&trace_types_lock);
6758 cpu = tracing_get_cpu(inode);
6759 ret = open_pipe_on_cpu(tr, cpu);
6761 goto fail_pipe_on_cpu;
6763 /* create a buffer to store the information to pass to userspace */
6764 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6767 goto fail_alloc_iter;
6770 trace_seq_init(&iter->seq);
6771 iter->trace = tr->current_trace;
6773 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6778 /* trace pipe does not show start of buffer */
6779 cpumask_setall(iter->started);
6781 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6782 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6784 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6785 if (trace_clocks[tr->clock_id].in_ns)
6786 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6789 iter->array_buffer = &tr->array_buffer;
6790 iter->cpu_file = cpu;
6791 mutex_init(&iter->mutex);
6792 filp->private_data = iter;
6794 if (iter->trace->pipe_open)
6795 iter->trace->pipe_open(iter);
6797 nonseekable_open(inode, filp);
6801 mutex_unlock(&trace_types_lock);
6807 close_pipe_on_cpu(tr, cpu);
6809 __trace_array_put(tr);
6810 mutex_unlock(&trace_types_lock);
6814 static int tracing_release_pipe(struct inode *inode, struct file *file)
6816 struct trace_iterator *iter = file->private_data;
6817 struct trace_array *tr = inode->i_private;
6819 mutex_lock(&trace_types_lock);
6823 if (iter->trace->pipe_close)
6824 iter->trace->pipe_close(iter);
6825 close_pipe_on_cpu(tr, iter->cpu_file);
6826 mutex_unlock(&trace_types_lock);
6828 free_cpumask_var(iter->started);
6831 mutex_destroy(&iter->mutex);
6834 trace_array_put(tr);
6840 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6842 struct trace_array *tr = iter->tr;
6844 /* Iterators are static, they should be filled or empty */
6845 if (trace_buffer_iter(iter, iter->cpu_file))
6846 return EPOLLIN | EPOLLRDNORM;
6848 if (tr->trace_flags & TRACE_ITER_BLOCK)
6850 * Always select as readable when in blocking mode
6852 return EPOLLIN | EPOLLRDNORM;
6854 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6855 filp, poll_table, iter->tr->buffer_percent);
6859 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6861 struct trace_iterator *iter = filp->private_data;
6863 return trace_poll(iter, filp, poll_table);
6866 /* Must be called with iter->mutex held. */
6867 static int tracing_wait_pipe(struct file *filp)
6869 struct trace_iterator *iter = filp->private_data;
6872 while (trace_empty(iter)) {
6874 if ((filp->f_flags & O_NONBLOCK)) {
6879 * We block until we read something and tracing is disabled.
6880 * We still block if tracing is disabled, but we have never
6881 * read anything. This allows a user to cat this file, and
6882 * then enable tracing. But after we have read something,
6883 * we give an EOF when tracing is again disabled.
6885 * iter->pos will be 0 if we haven't read anything.
6887 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6890 mutex_unlock(&iter->mutex);
6892 ret = wait_on_pipe(iter, 0);
6894 mutex_lock(&iter->mutex);
6907 tracing_read_pipe(struct file *filp, char __user *ubuf,
6908 size_t cnt, loff_t *ppos)
6910 struct trace_iterator *iter = filp->private_data;
6914 * Avoid more than one consumer on a single file descriptor
6915 * This is just a matter of traces coherency, the ring buffer itself
6918 mutex_lock(&iter->mutex);
6920 /* return any leftover data */
6921 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6925 trace_seq_init(&iter->seq);
6927 if (iter->trace->read) {
6928 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6934 sret = tracing_wait_pipe(filp);
6938 /* stop when tracing is finished */
6939 if (trace_empty(iter)) {
6944 if (cnt >= PAGE_SIZE)
6945 cnt = PAGE_SIZE - 1;
6947 /* reset all but tr, trace, and overruns */
6948 trace_iterator_reset(iter);
6949 cpumask_clear(iter->started);
6950 trace_seq_init(&iter->seq);
6952 trace_event_read_lock();
6953 trace_access_lock(iter->cpu_file);
6954 while (trace_find_next_entry_inc(iter) != NULL) {
6955 enum print_line_t ret;
6956 int save_len = iter->seq.seq.len;
6958 ret = print_trace_line(iter);
6959 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6961 * If one print_trace_line() fills entire trace_seq in one shot,
6962 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6963 * In this case, we need to consume it, otherwise, loop will peek
6964 * this event next time, resulting in an infinite loop.
6966 if (save_len == 0) {
6968 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6969 trace_consume(iter);
6973 /* In other cases, don't print partial lines */
6974 iter->seq.seq.len = save_len;
6977 if (ret != TRACE_TYPE_NO_CONSUME)
6978 trace_consume(iter);
6980 if (trace_seq_used(&iter->seq) >= cnt)
6984 * Setting the full flag means we reached the trace_seq buffer
6985 * size and we should leave by partial output condition above.
6986 * One of the trace_seq_* functions is not used properly.
6988 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6991 trace_access_unlock(iter->cpu_file);
6992 trace_event_read_unlock();
6994 /* Now copy what we have to the user */
6995 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6996 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6997 trace_seq_init(&iter->seq);
7000 * If there was nothing to send to user, in spite of consuming trace
7001 * entries, go back to wait for more entries.
7007 mutex_unlock(&iter->mutex);
7012 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
7015 __free_page(spd->pages[idx]);
7019 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
7025 /* Seq buffer is page-sized, exactly what we need. */
7027 save_len = iter->seq.seq.len;
7028 ret = print_trace_line(iter);
7030 if (trace_seq_has_overflowed(&iter->seq)) {
7031 iter->seq.seq.len = save_len;
7036 * This should not be hit, because it should only
7037 * be set if the iter->seq overflowed. But check it
7038 * anyway to be safe.
7040 if (ret == TRACE_TYPE_PARTIAL_LINE) {
7041 iter->seq.seq.len = save_len;
7045 count = trace_seq_used(&iter->seq) - save_len;
7048 iter->seq.seq.len = save_len;
7052 if (ret != TRACE_TYPE_NO_CONSUME)
7053 trace_consume(iter);
7055 if (!trace_find_next_entry_inc(iter)) {
7065 static ssize_t tracing_splice_read_pipe(struct file *filp,
7067 struct pipe_inode_info *pipe,
7071 struct page *pages_def[PIPE_DEF_BUFFERS];
7072 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7073 struct trace_iterator *iter = filp->private_data;
7074 struct splice_pipe_desc spd = {
7076 .partial = partial_def,
7077 .nr_pages = 0, /* This gets updated below. */
7078 .nr_pages_max = PIPE_DEF_BUFFERS,
7079 .ops = &default_pipe_buf_ops,
7080 .spd_release = tracing_spd_release_pipe,
7086 if (splice_grow_spd(pipe, &spd))
7089 mutex_lock(&iter->mutex);
7091 if (iter->trace->splice_read) {
7092 ret = iter->trace->splice_read(iter, filp,
7093 ppos, pipe, len, flags);
7098 ret = tracing_wait_pipe(filp);
7102 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
7107 trace_event_read_lock();
7108 trace_access_lock(iter->cpu_file);
7110 /* Fill as many pages as possible. */
7111 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
7112 spd.pages[i] = alloc_page(GFP_KERNEL);
7116 rem = tracing_fill_pipe_page(rem, iter);
7118 /* Copy the data into the page, so we can start over. */
7119 ret = trace_seq_to_buffer(&iter->seq,
7120 page_address(spd.pages[i]),
7121 trace_seq_used(&iter->seq));
7123 __free_page(spd.pages[i]);
7126 spd.partial[i].offset = 0;
7127 spd.partial[i].len = trace_seq_used(&iter->seq);
7129 trace_seq_init(&iter->seq);
7132 trace_access_unlock(iter->cpu_file);
7133 trace_event_read_unlock();
7134 mutex_unlock(&iter->mutex);
7139 ret = splice_to_pipe(pipe, &spd);
7143 splice_shrink_spd(&spd);
7147 mutex_unlock(&iter->mutex);
7152 tracing_entries_read(struct file *filp, char __user *ubuf,
7153 size_t cnt, loff_t *ppos)
7155 struct inode *inode = file_inode(filp);
7156 struct trace_array *tr = inode->i_private;
7157 int cpu = tracing_get_cpu(inode);
7162 mutex_lock(&trace_types_lock);
7164 if (cpu == RING_BUFFER_ALL_CPUS) {
7165 int cpu, buf_size_same;
7170 /* check if all cpu sizes are same */
7171 for_each_tracing_cpu(cpu) {
7172 /* fill in the size from first enabled cpu */
7174 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7175 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7181 if (buf_size_same) {
7182 if (!ring_buffer_expanded)
7183 r = sprintf(buf, "%lu (expanded: %lu)\n",
7185 trace_buf_size >> 10);
7187 r = sprintf(buf, "%lu\n", size >> 10);
7189 r = sprintf(buf, "X\n");
7191 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7193 mutex_unlock(&trace_types_lock);
7195 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7200 tracing_entries_write(struct file *filp, const char __user *ubuf,
7201 size_t cnt, loff_t *ppos)
7203 struct inode *inode = file_inode(filp);
7204 struct trace_array *tr = inode->i_private;
7208 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7212 /* must have at least 1 entry */
7216 /* value is in KB */
7218 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7228 tracing_total_entries_read(struct file *filp, char __user *ubuf,
7229 size_t cnt, loff_t *ppos)
7231 struct trace_array *tr = filp->private_data;
7234 unsigned long size = 0, expanded_size = 0;
7236 mutex_lock(&trace_types_lock);
7237 for_each_tracing_cpu(cpu) {
7238 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7239 if (!ring_buffer_expanded)
7240 expanded_size += trace_buf_size >> 10;
7242 if (ring_buffer_expanded)
7243 r = sprintf(buf, "%lu\n", size);
7245 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7246 mutex_unlock(&trace_types_lock);
7248 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7252 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7253 size_t cnt, loff_t *ppos)
7256 * There is no need to read what the user has written, this function
7257 * is just to make sure that there is no error when "echo" is used
7266 tracing_free_buffer_release(struct inode *inode, struct file *filp)
7268 struct trace_array *tr = inode->i_private;
7270 /* disable tracing ? */
7271 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7272 tracer_tracing_off(tr);
7273 /* resize the ring buffer to 0 */
7274 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7276 trace_array_put(tr);
7282 tracing_mark_write(struct file *filp, const char __user *ubuf,
7283 size_t cnt, loff_t *fpos)
7285 struct trace_array *tr = filp->private_data;
7286 struct ring_buffer_event *event;
7287 enum event_trigger_type tt = ETT_NONE;
7288 struct trace_buffer *buffer;
7289 struct print_entry *entry;
7294 /* Used in tracing_mark_raw_write() as well */
7295 #define FAULTED_STR "<faulted>"
7296 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7298 if (tracing_disabled)
7301 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7304 if (cnt > TRACE_BUF_SIZE)
7305 cnt = TRACE_BUF_SIZE;
7307 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7309 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
7311 /* If less than "<faulted>", then make sure we can still add that */
7312 if (cnt < FAULTED_SIZE)
7313 size += FAULTED_SIZE - cnt;
7315 buffer = tr->array_buffer.buffer;
7316 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7318 if (unlikely(!event))
7319 /* Ring buffer disabled, return as if not open for write */
7322 entry = ring_buffer_event_data(event);
7323 entry->ip = _THIS_IP_;
7325 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7327 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7333 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7334 /* do not add \n before testing triggers, but add \0 */
7335 entry->buf[cnt] = '\0';
7336 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7339 if (entry->buf[cnt - 1] != '\n') {
7340 entry->buf[cnt] = '\n';
7341 entry->buf[cnt + 1] = '\0';
7343 entry->buf[cnt] = '\0';
7345 if (static_branch_unlikely(&trace_marker_exports_enabled))
7346 ftrace_exports(event, TRACE_EXPORT_MARKER);
7347 __buffer_unlock_commit(buffer, event);
7350 event_triggers_post_call(tr->trace_marker_file, tt);
7355 /* Limit it for now to 3K (including tag) */
7356 #define RAW_DATA_MAX_SIZE (1024*3)
7359 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7360 size_t cnt, loff_t *fpos)
7362 struct trace_array *tr = filp->private_data;
7363 struct ring_buffer_event *event;
7364 struct trace_buffer *buffer;
7365 struct raw_data_entry *entry;
7370 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7372 if (tracing_disabled)
7375 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7378 /* The marker must at least have a tag id */
7379 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
7382 if (cnt > TRACE_BUF_SIZE)
7383 cnt = TRACE_BUF_SIZE;
7385 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7387 size = sizeof(*entry) + cnt;
7388 if (cnt < FAULT_SIZE_ID)
7389 size += FAULT_SIZE_ID - cnt;
7391 buffer = tr->array_buffer.buffer;
7392 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7395 /* Ring buffer disabled, return as if not open for write */
7398 entry = ring_buffer_event_data(event);
7400 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7403 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7408 __buffer_unlock_commit(buffer, event);
7413 static int tracing_clock_show(struct seq_file *m, void *v)
7415 struct trace_array *tr = m->private;
7418 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7420 "%s%s%s%s", i ? " " : "",
7421 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7422 i == tr->clock_id ? "]" : "");
7428 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7432 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7433 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7436 if (i == ARRAY_SIZE(trace_clocks))
7439 mutex_lock(&trace_types_lock);
7443 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7446 * New clock may not be consistent with the previous clock.
7447 * Reset the buffer so that it doesn't have incomparable timestamps.
7449 tracing_reset_online_cpus(&tr->array_buffer);
7451 #ifdef CONFIG_TRACER_MAX_TRACE
7452 if (tr->max_buffer.buffer)
7453 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7454 tracing_reset_online_cpus(&tr->max_buffer);
7457 mutex_unlock(&trace_types_lock);
7462 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7463 size_t cnt, loff_t *fpos)
7465 struct seq_file *m = filp->private_data;
7466 struct trace_array *tr = m->private;
7468 const char *clockstr;
7471 if (cnt >= sizeof(buf))
7474 if (copy_from_user(buf, ubuf, cnt))
7479 clockstr = strstrip(buf);
7481 ret = tracing_set_clock(tr, clockstr);
7490 static int tracing_clock_open(struct inode *inode, struct file *file)
7492 struct trace_array *tr = inode->i_private;
7495 ret = tracing_check_open_get_tr(tr);
7499 ret = single_open(file, tracing_clock_show, inode->i_private);
7501 trace_array_put(tr);
7506 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7508 struct trace_array *tr = m->private;
7510 mutex_lock(&trace_types_lock);
7512 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7513 seq_puts(m, "delta [absolute]\n");
7515 seq_puts(m, "[delta] absolute\n");
7517 mutex_unlock(&trace_types_lock);
7522 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7524 struct trace_array *tr = inode->i_private;
7527 ret = tracing_check_open_get_tr(tr);
7531 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7533 trace_array_put(tr);
7538 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7540 if (rbe == this_cpu_read(trace_buffered_event))
7541 return ring_buffer_time_stamp(buffer);
7543 return ring_buffer_event_time_stamp(buffer, rbe);
7547 * Set or disable using the per CPU trace_buffer_event when possible.
7549 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7553 mutex_lock(&trace_types_lock);
7555 if (set && tr->no_filter_buffering_ref++)
7559 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7564 --tr->no_filter_buffering_ref;
7567 mutex_unlock(&trace_types_lock);
7572 struct ftrace_buffer_info {
7573 struct trace_iterator iter;
7575 unsigned int spare_cpu;
7579 #ifdef CONFIG_TRACER_SNAPSHOT
7580 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7582 struct trace_array *tr = inode->i_private;
7583 struct trace_iterator *iter;
7587 ret = tracing_check_open_get_tr(tr);
7591 if (file->f_mode & FMODE_READ) {
7592 iter = __tracing_open(inode, file, true);
7594 ret = PTR_ERR(iter);
7596 /* Writes still need the seq_file to hold the private data */
7598 m = kzalloc(sizeof(*m), GFP_KERNEL);
7601 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7609 iter->array_buffer = &tr->max_buffer;
7610 iter->cpu_file = tracing_get_cpu(inode);
7612 file->private_data = m;
7616 trace_array_put(tr);
7622 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7625 struct seq_file *m = filp->private_data;
7626 struct trace_iterator *iter = m->private;
7627 struct trace_array *tr = iter->tr;
7631 ret = tracing_update_buffers();
7635 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7639 mutex_lock(&trace_types_lock);
7641 if (tr->current_trace->use_max_tr) {
7646 local_irq_disable();
7647 arch_spin_lock(&tr->max_lock);
7648 if (tr->cond_snapshot)
7650 arch_spin_unlock(&tr->max_lock);
7657 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7661 if (tr->allocated_snapshot)
7665 /* Only allow per-cpu swap if the ring buffer supports it */
7666 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7667 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7672 if (tr->allocated_snapshot)
7673 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7674 &tr->array_buffer, iter->cpu_file);
7676 ret = tracing_alloc_snapshot_instance(tr);
7679 local_irq_disable();
7680 /* Now, we're going to swap */
7681 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7682 update_max_tr(tr, current, smp_processor_id(), NULL);
7684 update_max_tr_single(tr, current, iter->cpu_file);
7688 if (tr->allocated_snapshot) {
7689 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7690 tracing_reset_online_cpus(&tr->max_buffer);
7692 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7702 mutex_unlock(&trace_types_lock);
7706 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7708 struct seq_file *m = file->private_data;
7711 ret = tracing_release(inode, file);
7713 if (file->f_mode & FMODE_READ)
7716 /* If write only, the seq_file is just a stub */
7724 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7725 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7726 size_t count, loff_t *ppos);
7727 static int tracing_buffers_release(struct inode *inode, struct file *file);
7728 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7729 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7731 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7733 struct ftrace_buffer_info *info;
7736 /* The following checks for tracefs lockdown */
7737 ret = tracing_buffers_open(inode, filp);
7741 info = filp->private_data;
7743 if (info->iter.trace->use_max_tr) {
7744 tracing_buffers_release(inode, filp);
7748 info->iter.snapshot = true;
7749 info->iter.array_buffer = &info->iter.tr->max_buffer;
7754 #endif /* CONFIG_TRACER_SNAPSHOT */
7757 static const struct file_operations tracing_thresh_fops = {
7758 .open = tracing_open_generic,
7759 .read = tracing_thresh_read,
7760 .write = tracing_thresh_write,
7761 .llseek = generic_file_llseek,
7764 #ifdef CONFIG_TRACER_MAX_TRACE
7765 static const struct file_operations tracing_max_lat_fops = {
7766 .open = tracing_open_generic,
7767 .read = tracing_max_lat_read,
7768 .write = tracing_max_lat_write,
7769 .llseek = generic_file_llseek,
7773 static const struct file_operations set_tracer_fops = {
7774 .open = tracing_open_generic,
7775 .read = tracing_set_trace_read,
7776 .write = tracing_set_trace_write,
7777 .llseek = generic_file_llseek,
7780 static const struct file_operations tracing_pipe_fops = {
7781 .open = tracing_open_pipe,
7782 .poll = tracing_poll_pipe,
7783 .read = tracing_read_pipe,
7784 .splice_read = tracing_splice_read_pipe,
7785 .release = tracing_release_pipe,
7786 .llseek = no_llseek,
7789 static const struct file_operations tracing_entries_fops = {
7790 .open = tracing_open_generic_tr,
7791 .read = tracing_entries_read,
7792 .write = tracing_entries_write,
7793 .llseek = generic_file_llseek,
7794 .release = tracing_release_generic_tr,
7797 static const struct file_operations tracing_total_entries_fops = {
7798 .open = tracing_open_generic_tr,
7799 .read = tracing_total_entries_read,
7800 .llseek = generic_file_llseek,
7801 .release = tracing_release_generic_tr,
7804 static const struct file_operations tracing_free_buffer_fops = {
7805 .open = tracing_open_generic_tr,
7806 .write = tracing_free_buffer_write,
7807 .release = tracing_free_buffer_release,
7810 static const struct file_operations tracing_mark_fops = {
7811 .open = tracing_mark_open,
7812 .write = tracing_mark_write,
7813 .release = tracing_release_generic_tr,
7816 static const struct file_operations tracing_mark_raw_fops = {
7817 .open = tracing_mark_open,
7818 .write = tracing_mark_raw_write,
7819 .release = tracing_release_generic_tr,
7822 static const struct file_operations trace_clock_fops = {
7823 .open = tracing_clock_open,
7825 .llseek = seq_lseek,
7826 .release = tracing_single_release_tr,
7827 .write = tracing_clock_write,
7830 static const struct file_operations trace_time_stamp_mode_fops = {
7831 .open = tracing_time_stamp_mode_open,
7833 .llseek = seq_lseek,
7834 .release = tracing_single_release_tr,
7837 #ifdef CONFIG_TRACER_SNAPSHOT
7838 static const struct file_operations snapshot_fops = {
7839 .open = tracing_snapshot_open,
7841 .write = tracing_snapshot_write,
7842 .llseek = tracing_lseek,
7843 .release = tracing_snapshot_release,
7846 static const struct file_operations snapshot_raw_fops = {
7847 .open = snapshot_raw_open,
7848 .read = tracing_buffers_read,
7849 .release = tracing_buffers_release,
7850 .splice_read = tracing_buffers_splice_read,
7851 .llseek = no_llseek,
7854 #endif /* CONFIG_TRACER_SNAPSHOT */
7857 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7858 * @filp: The active open file structure
7859 * @ubuf: The userspace provided buffer to read value into
7860 * @cnt: The maximum number of bytes to read
7861 * @ppos: The current "file" position
7863 * This function implements the write interface for a struct trace_min_max_param.
7864 * The filp->private_data must point to a trace_min_max_param structure that
7865 * defines where to write the value, the min and the max acceptable values,
7866 * and a lock to protect the write.
7869 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7871 struct trace_min_max_param *param = filp->private_data;
7878 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7883 mutex_lock(param->lock);
7885 if (param->min && val < *param->min)
7888 if (param->max && val > *param->max)
7895 mutex_unlock(param->lock);
7904 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7905 * @filp: The active open file structure
7906 * @ubuf: The userspace provided buffer to read value into
7907 * @cnt: The maximum number of bytes to read
7908 * @ppos: The current "file" position
7910 * This function implements the read interface for a struct trace_min_max_param.
7911 * The filp->private_data must point to a trace_min_max_param struct with valid
7915 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7917 struct trace_min_max_param *param = filp->private_data;
7918 char buf[U64_STR_SIZE];
7927 if (cnt > sizeof(buf))
7930 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7932 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7935 const struct file_operations trace_min_max_fops = {
7936 .open = tracing_open_generic,
7937 .read = trace_min_max_read,
7938 .write = trace_min_max_write,
7941 #define TRACING_LOG_ERRS_MAX 8
7942 #define TRACING_LOG_LOC_MAX 128
7944 #define CMD_PREFIX " Command: "
7947 const char **errs; /* ptr to loc-specific array of err strings */
7948 u8 type; /* index into errs -> specific err string */
7949 u16 pos; /* caret position */
7953 struct tracing_log_err {
7954 struct list_head list;
7955 struct err_info info;
7956 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7957 char *cmd; /* what caused err */
7960 static DEFINE_MUTEX(tracing_err_log_lock);
7962 static struct tracing_log_err *alloc_tracing_log_err(int len)
7964 struct tracing_log_err *err;
7966 err = kzalloc(sizeof(*err), GFP_KERNEL);
7968 return ERR_PTR(-ENOMEM);
7970 err->cmd = kzalloc(len, GFP_KERNEL);
7973 return ERR_PTR(-ENOMEM);
7979 static void free_tracing_log_err(struct tracing_log_err *err)
7985 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7988 struct tracing_log_err *err;
7991 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7992 err = alloc_tracing_log_err(len);
7993 if (PTR_ERR(err) != -ENOMEM)
7994 tr->n_err_log_entries++;
7998 cmd = kzalloc(len, GFP_KERNEL);
8000 return ERR_PTR(-ENOMEM);
8001 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
8004 list_del(&err->list);
8010 * err_pos - find the position of a string within a command for error careting
8011 * @cmd: The tracing command that caused the error
8012 * @str: The string to position the caret at within @cmd
8014 * Finds the position of the first occurrence of @str within @cmd. The
8015 * return value can be passed to tracing_log_err() for caret placement
8018 * Returns the index within @cmd of the first occurrence of @str or 0
8019 * if @str was not found.
8021 unsigned int err_pos(char *cmd, const char *str)
8025 if (WARN_ON(!strlen(cmd)))
8028 found = strstr(cmd, str);
8036 * tracing_log_err - write an error to the tracing error log
8037 * @tr: The associated trace array for the error (NULL for top level array)
8038 * @loc: A string describing where the error occurred
8039 * @cmd: The tracing command that caused the error
8040 * @errs: The array of loc-specific static error strings
8041 * @type: The index into errs[], which produces the specific static err string
8042 * @pos: The position the caret should be placed in the cmd
8044 * Writes an error into tracing/error_log of the form:
8046 * <loc>: error: <text>
8050 * tracing/error_log is a small log file containing the last
8051 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
8052 * unless there has been a tracing error, and the error log can be
8053 * cleared and have its memory freed by writing the empty string in
8054 * truncation mode to it i.e. echo > tracing/error_log.
8056 * NOTE: the @errs array along with the @type param are used to
8057 * produce a static error string - this string is not copied and saved
8058 * when the error is logged - only a pointer to it is saved. See
8059 * existing callers for examples of how static strings are typically
8060 * defined for use with tracing_log_err().
8062 void tracing_log_err(struct trace_array *tr,
8063 const char *loc, const char *cmd,
8064 const char **errs, u8 type, u16 pos)
8066 struct tracing_log_err *err;
8072 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
8074 mutex_lock(&tracing_err_log_lock);
8075 err = get_tracing_log_err(tr, len);
8076 if (PTR_ERR(err) == -ENOMEM) {
8077 mutex_unlock(&tracing_err_log_lock);
8081 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
8082 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
8084 err->info.errs = errs;
8085 err->info.type = type;
8086 err->info.pos = pos;
8087 err->info.ts = local_clock();
8089 list_add_tail(&err->list, &tr->err_log);
8090 mutex_unlock(&tracing_err_log_lock);
8093 static void clear_tracing_err_log(struct trace_array *tr)
8095 struct tracing_log_err *err, *next;
8097 mutex_lock(&tracing_err_log_lock);
8098 list_for_each_entry_safe(err, next, &tr->err_log, list) {
8099 list_del(&err->list);
8100 free_tracing_log_err(err);
8103 tr->n_err_log_entries = 0;
8104 mutex_unlock(&tracing_err_log_lock);
8107 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
8109 struct trace_array *tr = m->private;
8111 mutex_lock(&tracing_err_log_lock);
8113 return seq_list_start(&tr->err_log, *pos);
8116 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
8118 struct trace_array *tr = m->private;
8120 return seq_list_next(v, &tr->err_log, pos);
8123 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
8125 mutex_unlock(&tracing_err_log_lock);
8128 static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
8132 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
8134 for (i = 0; i < pos; i++)
8139 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
8141 struct tracing_log_err *err = v;
8144 const char *err_text = err->info.errs[err->info.type];
8145 u64 sec = err->info.ts;
8148 nsec = do_div(sec, NSEC_PER_SEC);
8149 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
8150 err->loc, err_text);
8151 seq_printf(m, "%s", err->cmd);
8152 tracing_err_log_show_pos(m, err->info.pos);
8158 static const struct seq_operations tracing_err_log_seq_ops = {
8159 .start = tracing_err_log_seq_start,
8160 .next = tracing_err_log_seq_next,
8161 .stop = tracing_err_log_seq_stop,
8162 .show = tracing_err_log_seq_show
8165 static int tracing_err_log_open(struct inode *inode, struct file *file)
8167 struct trace_array *tr = inode->i_private;
8170 ret = tracing_check_open_get_tr(tr);
8174 /* If this file was opened for write, then erase contents */
8175 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8176 clear_tracing_err_log(tr);
8178 if (file->f_mode & FMODE_READ) {
8179 ret = seq_open(file, &tracing_err_log_seq_ops);
8181 struct seq_file *m = file->private_data;
8184 trace_array_put(tr);
8190 static ssize_t tracing_err_log_write(struct file *file,
8191 const char __user *buffer,
8192 size_t count, loff_t *ppos)
8197 static int tracing_err_log_release(struct inode *inode, struct file *file)
8199 struct trace_array *tr = inode->i_private;
8201 trace_array_put(tr);
8203 if (file->f_mode & FMODE_READ)
8204 seq_release(inode, file);
8209 static const struct file_operations tracing_err_log_fops = {
8210 .open = tracing_err_log_open,
8211 .write = tracing_err_log_write,
8213 .llseek = tracing_lseek,
8214 .release = tracing_err_log_release,
8217 static int tracing_buffers_open(struct inode *inode, struct file *filp)
8219 struct trace_array *tr = inode->i_private;
8220 struct ftrace_buffer_info *info;
8223 ret = tracing_check_open_get_tr(tr);
8227 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8229 trace_array_put(tr);
8233 mutex_lock(&trace_types_lock);
8236 info->iter.cpu_file = tracing_get_cpu(inode);
8237 info->iter.trace = tr->current_trace;
8238 info->iter.array_buffer = &tr->array_buffer;
8240 /* Force reading ring buffer for first read */
8241 info->read = (unsigned int)-1;
8243 filp->private_data = info;
8247 mutex_unlock(&trace_types_lock);
8249 ret = nonseekable_open(inode, filp);
8251 trace_array_put(tr);
8257 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8259 struct ftrace_buffer_info *info = filp->private_data;
8260 struct trace_iterator *iter = &info->iter;
8262 return trace_poll(iter, filp, poll_table);
8266 tracing_buffers_read(struct file *filp, char __user *ubuf,
8267 size_t count, loff_t *ppos)
8269 struct ftrace_buffer_info *info = filp->private_data;
8270 struct trace_iterator *iter = &info->iter;
8277 #ifdef CONFIG_TRACER_MAX_TRACE
8278 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8283 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8285 if (IS_ERR(info->spare)) {
8286 ret = PTR_ERR(info->spare);
8289 info->spare_cpu = iter->cpu_file;
8295 /* Do we have previous read data to read? */
8296 if (info->read < PAGE_SIZE)
8300 trace_access_lock(iter->cpu_file);
8301 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8305 trace_access_unlock(iter->cpu_file);
8308 if (trace_empty(iter)) {
8309 if ((filp->f_flags & O_NONBLOCK))
8312 ret = wait_on_pipe(iter, 0);
8323 size = PAGE_SIZE - info->read;
8327 ret = copy_to_user(ubuf, info->spare + info->read, size);
8339 static int tracing_buffers_release(struct inode *inode, struct file *file)
8341 struct ftrace_buffer_info *info = file->private_data;
8342 struct trace_iterator *iter = &info->iter;
8344 mutex_lock(&trace_types_lock);
8346 iter->tr->trace_ref--;
8348 __trace_array_put(iter->tr);
8351 /* Make sure the waiters see the new wait_index */
8354 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8357 ring_buffer_free_read_page(iter->array_buffer->buffer,
8358 info->spare_cpu, info->spare);
8361 mutex_unlock(&trace_types_lock);
8367 struct trace_buffer *buffer;
8370 refcount_t refcount;
8373 static void buffer_ref_release(struct buffer_ref *ref)
8375 if (!refcount_dec_and_test(&ref->refcount))
8377 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8381 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8382 struct pipe_buffer *buf)
8384 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8386 buffer_ref_release(ref);
8390 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8391 struct pipe_buffer *buf)
8393 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8395 if (refcount_read(&ref->refcount) > INT_MAX/2)
8398 refcount_inc(&ref->refcount);
8402 /* Pipe buffer operations for a buffer. */
8403 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8404 .release = buffer_pipe_buf_release,
8405 .get = buffer_pipe_buf_get,
8409 * Callback from splice_to_pipe(), if we need to release some pages
8410 * at the end of the spd in case we error'ed out in filling the pipe.
8412 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8414 struct buffer_ref *ref =
8415 (struct buffer_ref *)spd->partial[i].private;
8417 buffer_ref_release(ref);
8418 spd->partial[i].private = 0;
8422 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8423 struct pipe_inode_info *pipe, size_t len,
8426 struct ftrace_buffer_info *info = file->private_data;
8427 struct trace_iterator *iter = &info->iter;
8428 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8429 struct page *pages_def[PIPE_DEF_BUFFERS];
8430 struct splice_pipe_desc spd = {
8432 .partial = partial_def,
8433 .nr_pages_max = PIPE_DEF_BUFFERS,
8434 .ops = &buffer_pipe_buf_ops,
8435 .spd_release = buffer_spd_release,
8437 struct buffer_ref *ref;
8441 #ifdef CONFIG_TRACER_MAX_TRACE
8442 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8446 if (*ppos & (PAGE_SIZE - 1))
8449 if (len & (PAGE_SIZE - 1)) {
8450 if (len < PAGE_SIZE)
8455 if (splice_grow_spd(pipe, &spd))
8459 trace_access_lock(iter->cpu_file);
8460 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8462 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
8466 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8472 refcount_set(&ref->refcount, 1);
8473 ref->buffer = iter->array_buffer->buffer;
8474 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8475 if (IS_ERR(ref->page)) {
8476 ret = PTR_ERR(ref->page);
8481 ref->cpu = iter->cpu_file;
8483 r = ring_buffer_read_page(ref->buffer, &ref->page,
8484 len, iter->cpu_file, 1);
8486 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8492 page = virt_to_page(ref->page);
8494 spd.pages[i] = page;
8495 spd.partial[i].len = PAGE_SIZE;
8496 spd.partial[i].offset = 0;
8497 spd.partial[i].private = (unsigned long)ref;
8501 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8504 trace_access_unlock(iter->cpu_file);
8507 /* did we read anything? */
8508 if (!spd.nr_pages) {
8515 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8518 wait_index = READ_ONCE(iter->wait_index);
8520 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
8524 /* No need to wait after waking up when tracing is off */
8525 if (!tracer_tracing_is_on(iter->tr))
8528 /* Make sure we see the new wait_index */
8530 if (wait_index != iter->wait_index)
8536 ret = splice_to_pipe(pipe, &spd);
8538 splice_shrink_spd(&spd);
8543 /* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8544 static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8546 struct ftrace_buffer_info *info = file->private_data;
8547 struct trace_iterator *iter = &info->iter;
8550 return -ENOIOCTLCMD;
8552 mutex_lock(&trace_types_lock);
8555 /* Make sure the waiters see the new wait_index */
8558 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8560 mutex_unlock(&trace_types_lock);
8564 static const struct file_operations tracing_buffers_fops = {
8565 .open = tracing_buffers_open,
8566 .read = tracing_buffers_read,
8567 .poll = tracing_buffers_poll,
8568 .release = tracing_buffers_release,
8569 .splice_read = tracing_buffers_splice_read,
8570 .unlocked_ioctl = tracing_buffers_ioctl,
8571 .llseek = no_llseek,
8575 tracing_stats_read(struct file *filp, char __user *ubuf,
8576 size_t count, loff_t *ppos)
8578 struct inode *inode = file_inode(filp);
8579 struct trace_array *tr = inode->i_private;
8580 struct array_buffer *trace_buf = &tr->array_buffer;
8581 int cpu = tracing_get_cpu(inode);
8582 struct trace_seq *s;
8584 unsigned long long t;
8585 unsigned long usec_rem;
8587 s = kmalloc(sizeof(*s), GFP_KERNEL);
8593 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8594 trace_seq_printf(s, "entries: %ld\n", cnt);
8596 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8597 trace_seq_printf(s, "overrun: %ld\n", cnt);
8599 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8600 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8602 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8603 trace_seq_printf(s, "bytes: %ld\n", cnt);
8605 if (trace_clocks[tr->clock_id].in_ns) {
8606 /* local or global for trace_clock */
8607 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8608 usec_rem = do_div(t, USEC_PER_SEC);
8609 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8612 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8613 usec_rem = do_div(t, USEC_PER_SEC);
8614 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8616 /* counter or tsc mode for trace_clock */
8617 trace_seq_printf(s, "oldest event ts: %llu\n",
8618 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8620 trace_seq_printf(s, "now ts: %llu\n",
8621 ring_buffer_time_stamp(trace_buf->buffer));
8624 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8625 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8627 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8628 trace_seq_printf(s, "read events: %ld\n", cnt);
8630 count = simple_read_from_buffer(ubuf, count, ppos,
8631 s->buffer, trace_seq_used(s));
8638 static const struct file_operations tracing_stats_fops = {
8639 .open = tracing_open_generic_tr,
8640 .read = tracing_stats_read,
8641 .llseek = generic_file_llseek,
8642 .release = tracing_release_generic_tr,
8645 #ifdef CONFIG_DYNAMIC_FTRACE
8648 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8649 size_t cnt, loff_t *ppos)
8655 /* 256 should be plenty to hold the amount needed */
8656 buf = kmalloc(256, GFP_KERNEL);
8660 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8661 ftrace_update_tot_cnt,
8662 ftrace_number_of_pages,
8663 ftrace_number_of_groups);
8665 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8670 static const struct file_operations tracing_dyn_info_fops = {
8671 .open = tracing_open_generic,
8672 .read = tracing_read_dyn_info,
8673 .llseek = generic_file_llseek,
8675 #endif /* CONFIG_DYNAMIC_FTRACE */
8677 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8679 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8680 struct trace_array *tr, struct ftrace_probe_ops *ops,
8683 tracing_snapshot_instance(tr);
8687 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8688 struct trace_array *tr, struct ftrace_probe_ops *ops,
8691 struct ftrace_func_mapper *mapper = data;
8695 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8705 tracing_snapshot_instance(tr);
8709 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8710 struct ftrace_probe_ops *ops, void *data)
8712 struct ftrace_func_mapper *mapper = data;
8715 seq_printf(m, "%ps:", (void *)ip);
8717 seq_puts(m, "snapshot");
8720 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8723 seq_printf(m, ":count=%ld\n", *count);
8725 seq_puts(m, ":unlimited\n");
8731 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8732 unsigned long ip, void *init_data, void **data)
8734 struct ftrace_func_mapper *mapper = *data;
8737 mapper = allocate_ftrace_func_mapper();
8743 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8747 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8748 unsigned long ip, void *data)
8750 struct ftrace_func_mapper *mapper = data;
8755 free_ftrace_func_mapper(mapper, NULL);
8759 ftrace_func_mapper_remove_ip(mapper, ip);
8762 static struct ftrace_probe_ops snapshot_probe_ops = {
8763 .func = ftrace_snapshot,
8764 .print = ftrace_snapshot_print,
8767 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8768 .func = ftrace_count_snapshot,
8769 .print = ftrace_snapshot_print,
8770 .init = ftrace_snapshot_init,
8771 .free = ftrace_snapshot_free,
8775 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8776 char *glob, char *cmd, char *param, int enable)
8778 struct ftrace_probe_ops *ops;
8779 void *count = (void *)-1;
8786 /* hash funcs only work with set_ftrace_filter */
8790 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8793 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8798 number = strsep(¶m, ":");
8800 if (!strlen(number))
8804 * We use the callback data field (which is a pointer)
8807 ret = kstrtoul(number, 0, (unsigned long *)&count);
8812 ret = tracing_alloc_snapshot_instance(tr);
8816 ret = register_ftrace_function_probe(glob, tr, ops, count);
8819 return ret < 0 ? ret : 0;
8822 static struct ftrace_func_command ftrace_snapshot_cmd = {
8824 .func = ftrace_trace_snapshot_callback,
8827 static __init int register_snapshot_cmd(void)
8829 return register_ftrace_command(&ftrace_snapshot_cmd);
8832 static inline __init int register_snapshot_cmd(void) { return 0; }
8833 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8835 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8837 if (WARN_ON(!tr->dir))
8838 return ERR_PTR(-ENODEV);
8840 /* Top directory uses NULL as the parent */
8841 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8844 /* All sub buffers have a descriptor */
8848 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8850 struct dentry *d_tracer;
8853 return tr->percpu_dir;
8855 d_tracer = tracing_get_dentry(tr);
8856 if (IS_ERR(d_tracer))
8859 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8861 MEM_FAIL(!tr->percpu_dir,
8862 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8864 return tr->percpu_dir;
8867 static struct dentry *
8868 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8869 void *data, long cpu, const struct file_operations *fops)
8871 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8873 if (ret) /* See tracing_get_cpu() */
8874 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8879 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8881 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8882 struct dentry *d_cpu;
8883 char cpu_dir[30]; /* 30 characters should be more than enough */
8888 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8889 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8891 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8895 /* per cpu trace_pipe */
8896 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8897 tr, cpu, &tracing_pipe_fops);
8900 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8901 tr, cpu, &tracing_fops);
8903 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8904 tr, cpu, &tracing_buffers_fops);
8906 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8907 tr, cpu, &tracing_stats_fops);
8909 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8910 tr, cpu, &tracing_entries_fops);
8912 #ifdef CONFIG_TRACER_SNAPSHOT
8913 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8914 tr, cpu, &snapshot_fops);
8916 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8917 tr, cpu, &snapshot_raw_fops);
8921 #ifdef CONFIG_FTRACE_SELFTEST
8922 /* Let selftest have access to static functions in this file */
8923 #include "trace_selftest.c"
8927 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8930 struct trace_option_dentry *topt = filp->private_data;
8933 if (topt->flags->val & topt->opt->bit)
8938 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8942 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8945 struct trace_option_dentry *topt = filp->private_data;
8949 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8953 if (val != 0 && val != 1)
8956 if (!!(topt->flags->val & topt->opt->bit) != val) {
8957 mutex_lock(&trace_types_lock);
8958 ret = __set_tracer_option(topt->tr, topt->flags,
8960 mutex_unlock(&trace_types_lock);
8971 static const struct file_operations trace_options_fops = {
8972 .open = tracing_open_generic,
8973 .read = trace_options_read,
8974 .write = trace_options_write,
8975 .llseek = generic_file_llseek,
8979 * In order to pass in both the trace_array descriptor as well as the index
8980 * to the flag that the trace option file represents, the trace_array
8981 * has a character array of trace_flags_index[], which holds the index
8982 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8983 * The address of this character array is passed to the flag option file
8984 * read/write callbacks.
8986 * In order to extract both the index and the trace_array descriptor,
8987 * get_tr_index() uses the following algorithm.
8991 * As the pointer itself contains the address of the index (remember
8994 * Then to get the trace_array descriptor, by subtracting that index
8995 * from the ptr, we get to the start of the index itself.
8997 * ptr - idx == &index[0]
8999 * Then a simple container_of() from that pointer gets us to the
9000 * trace_array descriptor.
9002 static void get_tr_index(void *data, struct trace_array **ptr,
9003 unsigned int *pindex)
9005 *pindex = *(unsigned char *)data;
9007 *ptr = container_of(data - *pindex, struct trace_array,
9012 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
9015 void *tr_index = filp->private_data;
9016 struct trace_array *tr;
9020 get_tr_index(tr_index, &tr, &index);
9022 if (tr->trace_flags & (1 << index))
9027 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9031 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
9034 void *tr_index = filp->private_data;
9035 struct trace_array *tr;
9040 get_tr_index(tr_index, &tr, &index);
9042 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9046 if (val != 0 && val != 1)
9049 mutex_lock(&event_mutex);
9050 mutex_lock(&trace_types_lock);
9051 ret = set_tracer_flag(tr, 1 << index, val);
9052 mutex_unlock(&trace_types_lock);
9053 mutex_unlock(&event_mutex);
9063 static const struct file_operations trace_options_core_fops = {
9064 .open = tracing_open_generic,
9065 .read = trace_options_core_read,
9066 .write = trace_options_core_write,
9067 .llseek = generic_file_llseek,
9070 struct dentry *trace_create_file(const char *name,
9072 struct dentry *parent,
9074 const struct file_operations *fops)
9078 ret = tracefs_create_file(name, mode, parent, data, fops);
9080 pr_warn("Could not create tracefs '%s' entry\n", name);
9086 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
9088 struct dentry *d_tracer;
9093 d_tracer = tracing_get_dentry(tr);
9094 if (IS_ERR(d_tracer))
9097 tr->options = tracefs_create_dir("options", d_tracer);
9099 pr_warn("Could not create tracefs directory 'options'\n");
9107 create_trace_option_file(struct trace_array *tr,
9108 struct trace_option_dentry *topt,
9109 struct tracer_flags *flags,
9110 struct tracer_opt *opt)
9112 struct dentry *t_options;
9114 t_options = trace_options_init_dentry(tr);
9118 topt->flags = flags;
9122 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
9123 t_options, topt, &trace_options_fops);
9128 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
9130 struct trace_option_dentry *topts;
9131 struct trace_options *tr_topts;
9132 struct tracer_flags *flags;
9133 struct tracer_opt *opts;
9140 flags = tracer->flags;
9142 if (!flags || !flags->opts)
9146 * If this is an instance, only create flags for tracers
9147 * the instance may have.
9149 if (!trace_ok_for_array(tracer, tr))
9152 for (i = 0; i < tr->nr_topts; i++) {
9153 /* Make sure there's no duplicate flags. */
9154 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
9160 for (cnt = 0; opts[cnt].name; cnt++)
9163 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
9167 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9174 tr->topts = tr_topts;
9175 tr->topts[tr->nr_topts].tracer = tracer;
9176 tr->topts[tr->nr_topts].topts = topts;
9179 for (cnt = 0; opts[cnt].name; cnt++) {
9180 create_trace_option_file(tr, &topts[cnt], flags,
9182 MEM_FAIL(topts[cnt].entry == NULL,
9183 "Failed to create trace option: %s",
9188 static struct dentry *
9189 create_trace_option_core_file(struct trace_array *tr,
9190 const char *option, long index)
9192 struct dentry *t_options;
9194 t_options = trace_options_init_dentry(tr);
9198 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9199 (void *)&tr->trace_flags_index[index],
9200 &trace_options_core_fops);
9203 static void create_trace_options_dir(struct trace_array *tr)
9205 struct dentry *t_options;
9206 bool top_level = tr == &global_trace;
9209 t_options = trace_options_init_dentry(tr);
9213 for (i = 0; trace_options[i]; i++) {
9215 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9216 create_trace_option_core_file(tr, trace_options[i], i);
9221 rb_simple_read(struct file *filp, char __user *ubuf,
9222 size_t cnt, loff_t *ppos)
9224 struct trace_array *tr = filp->private_data;
9228 r = tracer_tracing_is_on(tr);
9229 r = sprintf(buf, "%d\n", r);
9231 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9235 rb_simple_write(struct file *filp, const char __user *ubuf,
9236 size_t cnt, loff_t *ppos)
9238 struct trace_array *tr = filp->private_data;
9239 struct trace_buffer *buffer = tr->array_buffer.buffer;
9243 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9248 mutex_lock(&trace_types_lock);
9249 if (!!val == tracer_tracing_is_on(tr)) {
9250 val = 0; /* do nothing */
9252 tracer_tracing_on(tr);
9253 if (tr->current_trace->start)
9254 tr->current_trace->start(tr);
9256 tracer_tracing_off(tr);
9257 if (tr->current_trace->stop)
9258 tr->current_trace->stop(tr);
9259 /* Wake up any waiters */
9260 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9262 mutex_unlock(&trace_types_lock);
9270 static const struct file_operations rb_simple_fops = {
9271 .open = tracing_open_generic_tr,
9272 .read = rb_simple_read,
9273 .write = rb_simple_write,
9274 .release = tracing_release_generic_tr,
9275 .llseek = default_llseek,
9279 buffer_percent_read(struct file *filp, char __user *ubuf,
9280 size_t cnt, loff_t *ppos)
9282 struct trace_array *tr = filp->private_data;
9286 r = tr->buffer_percent;
9287 r = sprintf(buf, "%d\n", r);
9289 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9293 buffer_percent_write(struct file *filp, const char __user *ubuf,
9294 size_t cnt, loff_t *ppos)
9296 struct trace_array *tr = filp->private_data;
9300 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9307 tr->buffer_percent = val;
9314 static const struct file_operations buffer_percent_fops = {
9315 .open = tracing_open_generic_tr,
9316 .read = buffer_percent_read,
9317 .write = buffer_percent_write,
9318 .release = tracing_release_generic_tr,
9319 .llseek = default_llseek,
9322 static struct dentry *trace_instance_dir;
9325 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9328 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9330 enum ring_buffer_flags rb_flags;
9332 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9336 buf->buffer = ring_buffer_alloc(size, rb_flags);
9340 buf->data = alloc_percpu(struct trace_array_cpu);
9342 ring_buffer_free(buf->buffer);
9347 /* Allocate the first page for all buffers */
9348 set_buffer_entries(&tr->array_buffer,
9349 ring_buffer_size(tr->array_buffer.buffer, 0));
9354 static void free_trace_buffer(struct array_buffer *buf)
9357 ring_buffer_free(buf->buffer);
9359 free_percpu(buf->data);
9364 static int allocate_trace_buffers(struct trace_array *tr, int size)
9368 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9372 #ifdef CONFIG_TRACER_MAX_TRACE
9373 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9374 allocate_snapshot ? size : 1);
9375 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9376 free_trace_buffer(&tr->array_buffer);
9379 tr->allocated_snapshot = allocate_snapshot;
9381 allocate_snapshot = false;
9387 static void free_trace_buffers(struct trace_array *tr)
9392 free_trace_buffer(&tr->array_buffer);
9394 #ifdef CONFIG_TRACER_MAX_TRACE
9395 free_trace_buffer(&tr->max_buffer);
9399 static void init_trace_flags_index(struct trace_array *tr)
9403 /* Used by the trace options files */
9404 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9405 tr->trace_flags_index[i] = i;
9408 static void __update_tracer_options(struct trace_array *tr)
9412 for (t = trace_types; t; t = t->next)
9413 add_tracer_options(tr, t);
9416 static void update_tracer_options(struct trace_array *tr)
9418 mutex_lock(&trace_types_lock);
9419 tracer_options_updated = true;
9420 __update_tracer_options(tr);
9421 mutex_unlock(&trace_types_lock);
9424 /* Must have trace_types_lock held */
9425 struct trace_array *trace_array_find(const char *instance)
9427 struct trace_array *tr, *found = NULL;
9429 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9430 if (tr->name && strcmp(tr->name, instance) == 0) {
9439 struct trace_array *trace_array_find_get(const char *instance)
9441 struct trace_array *tr;
9443 mutex_lock(&trace_types_lock);
9444 tr = trace_array_find(instance);
9447 mutex_unlock(&trace_types_lock);
9452 static int trace_array_create_dir(struct trace_array *tr)
9456 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9460 ret = event_trace_add_tracer(tr->dir, tr);
9462 tracefs_remove(tr->dir);
9466 init_tracer_tracefs(tr, tr->dir);
9467 __update_tracer_options(tr);
9472 static struct trace_array *trace_array_create(const char *name)
9474 struct trace_array *tr;
9478 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9480 return ERR_PTR(ret);
9482 tr->name = kstrdup(name, GFP_KERNEL);
9486 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9489 if (!alloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9492 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9494 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9496 raw_spin_lock_init(&tr->start_lock);
9498 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9500 tr->current_trace = &nop_trace;
9502 INIT_LIST_HEAD(&tr->systems);
9503 INIT_LIST_HEAD(&tr->events);
9504 INIT_LIST_HEAD(&tr->hist_vars);
9505 INIT_LIST_HEAD(&tr->err_log);
9507 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9510 if (ftrace_allocate_ftrace_ops(tr) < 0)
9513 ftrace_init_trace_array(tr);
9515 init_trace_flags_index(tr);
9517 if (trace_instance_dir) {
9518 ret = trace_array_create_dir(tr);
9522 __trace_early_add_events(tr);
9524 list_add(&tr->list, &ftrace_trace_arrays);
9531 ftrace_free_ftrace_ops(tr);
9532 free_trace_buffers(tr);
9533 free_cpumask_var(tr->pipe_cpumask);
9534 free_cpumask_var(tr->tracing_cpumask);
9538 return ERR_PTR(ret);
9541 static int instance_mkdir(const char *name)
9543 struct trace_array *tr;
9546 mutex_lock(&event_mutex);
9547 mutex_lock(&trace_types_lock);
9550 if (trace_array_find(name))
9553 tr = trace_array_create(name);
9555 ret = PTR_ERR_OR_ZERO(tr);
9558 mutex_unlock(&trace_types_lock);
9559 mutex_unlock(&event_mutex);
9564 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9565 * @name: The name of the trace array to be looked up/created.
9567 * Returns pointer to trace array with given name.
9568 * NULL, if it cannot be created.
9570 * NOTE: This function increments the reference counter associated with the
9571 * trace array returned. This makes sure it cannot be freed while in use.
9572 * Use trace_array_put() once the trace array is no longer needed.
9573 * If the trace_array is to be freed, trace_array_destroy() needs to
9574 * be called after the trace_array_put(), or simply let user space delete
9575 * it from the tracefs instances directory. But until the
9576 * trace_array_put() is called, user space can not delete it.
9579 struct trace_array *trace_array_get_by_name(const char *name)
9581 struct trace_array *tr;
9583 mutex_lock(&event_mutex);
9584 mutex_lock(&trace_types_lock);
9586 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9587 if (tr->name && strcmp(tr->name, name) == 0)
9591 tr = trace_array_create(name);
9599 mutex_unlock(&trace_types_lock);
9600 mutex_unlock(&event_mutex);
9603 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9605 static int __remove_instance(struct trace_array *tr)
9609 /* Reference counter for a newly created trace array = 1. */
9610 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9613 list_del(&tr->list);
9615 /* Disable all the flags that were enabled coming in */
9616 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9617 if ((1 << i) & ZEROED_TRACE_FLAGS)
9618 set_tracer_flag(tr, 1 << i, 0);
9621 tracing_set_nop(tr);
9622 clear_ftrace_function_probes(tr);
9623 event_trace_del_tracer(tr);
9624 ftrace_clear_pids(tr);
9625 ftrace_destroy_function_files(tr);
9626 tracefs_remove(tr->dir);
9627 free_percpu(tr->last_func_repeats);
9628 free_trace_buffers(tr);
9629 clear_tracing_err_log(tr);
9631 for (i = 0; i < tr->nr_topts; i++) {
9632 kfree(tr->topts[i].topts);
9636 free_cpumask_var(tr->pipe_cpumask);
9637 free_cpumask_var(tr->tracing_cpumask);
9644 int trace_array_destroy(struct trace_array *this_tr)
9646 struct trace_array *tr;
9652 mutex_lock(&event_mutex);
9653 mutex_lock(&trace_types_lock);
9657 /* Making sure trace array exists before destroying it. */
9658 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9659 if (tr == this_tr) {
9660 ret = __remove_instance(tr);
9665 mutex_unlock(&trace_types_lock);
9666 mutex_unlock(&event_mutex);
9670 EXPORT_SYMBOL_GPL(trace_array_destroy);
9672 static int instance_rmdir(const char *name)
9674 struct trace_array *tr;
9677 mutex_lock(&event_mutex);
9678 mutex_lock(&trace_types_lock);
9681 tr = trace_array_find(name);
9683 ret = __remove_instance(tr);
9685 mutex_unlock(&trace_types_lock);
9686 mutex_unlock(&event_mutex);
9691 static __init void create_trace_instances(struct dentry *d_tracer)
9693 struct trace_array *tr;
9695 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9698 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9701 mutex_lock(&event_mutex);
9702 mutex_lock(&trace_types_lock);
9704 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9707 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9708 "Failed to create instance directory\n"))
9712 mutex_unlock(&trace_types_lock);
9713 mutex_unlock(&event_mutex);
9717 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9719 struct trace_event_file *file;
9722 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9723 tr, &show_traces_fops);
9725 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9726 tr, &set_tracer_fops);
9728 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9729 tr, &tracing_cpumask_fops);
9731 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9732 tr, &tracing_iter_fops);
9734 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9737 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9738 tr, &tracing_pipe_fops);
9740 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9741 tr, &tracing_entries_fops);
9743 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9744 tr, &tracing_total_entries_fops);
9746 trace_create_file("free_buffer", 0200, d_tracer,
9747 tr, &tracing_free_buffer_fops);
9749 trace_create_file("trace_marker", 0220, d_tracer,
9750 tr, &tracing_mark_fops);
9752 file = __find_event_file(tr, "ftrace", "print");
9753 if (file && file->dir)
9754 trace_create_file("trigger", TRACE_MODE_WRITE, file->dir,
9755 file, &event_trigger_fops);
9756 tr->trace_marker_file = file;
9758 trace_create_file("trace_marker_raw", 0220, d_tracer,
9759 tr, &tracing_mark_raw_fops);
9761 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9764 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9765 tr, &rb_simple_fops);
9767 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9768 &trace_time_stamp_mode_fops);
9770 tr->buffer_percent = 50;
9772 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9773 tr, &buffer_percent_fops);
9775 create_trace_options_dir(tr);
9777 #ifdef CONFIG_TRACER_MAX_TRACE
9778 trace_create_maxlat_file(tr, d_tracer);
9781 if (ftrace_create_function_files(tr, d_tracer))
9782 MEM_FAIL(1, "Could not allocate function filter files");
9784 #ifdef CONFIG_TRACER_SNAPSHOT
9785 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9786 tr, &snapshot_fops);
9789 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9790 tr, &tracing_err_log_fops);
9792 for_each_tracing_cpu(cpu)
9793 tracing_init_tracefs_percpu(tr, cpu);
9795 ftrace_init_tracefs(tr, d_tracer);
9798 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9800 struct vfsmount *mnt;
9801 struct file_system_type *type;
9804 * To maintain backward compatibility for tools that mount
9805 * debugfs to get to the tracing facility, tracefs is automatically
9806 * mounted to the debugfs/tracing directory.
9808 type = get_fs_type("tracefs");
9811 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9812 put_filesystem(type);
9821 * tracing_init_dentry - initialize top level trace array
9823 * This is called when creating files or directories in the tracing
9824 * directory. It is called via fs_initcall() by any of the boot up code
9825 * and expects to return the dentry of the top level tracing directory.
9827 int tracing_init_dentry(void)
9829 struct trace_array *tr = &global_trace;
9831 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9832 pr_warn("Tracing disabled due to lockdown\n");
9836 /* The top level trace array uses NULL as parent */
9840 if (WARN_ON(!tracefs_initialized()))
9844 * As there may still be users that expect the tracing
9845 * files to exist in debugfs/tracing, we must automount
9846 * the tracefs file system there, so older tools still
9847 * work with the newer kernel.
9849 tr->dir = debugfs_create_automount("tracing", NULL,
9850 trace_automount, NULL);
9855 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9856 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9858 static struct workqueue_struct *eval_map_wq __initdata;
9859 static struct work_struct eval_map_work __initdata;
9860 static struct work_struct tracerfs_init_work __initdata;
9862 static void __init eval_map_work_func(struct work_struct *work)
9866 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9867 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9870 static int __init trace_eval_init(void)
9872 INIT_WORK(&eval_map_work, eval_map_work_func);
9874 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9876 pr_err("Unable to allocate eval_map_wq\n");
9878 eval_map_work_func(&eval_map_work);
9882 queue_work(eval_map_wq, &eval_map_work);
9886 subsys_initcall(trace_eval_init);
9888 static int __init trace_eval_sync(void)
9890 /* Make sure the eval map updates are finished */
9892 destroy_workqueue(eval_map_wq);
9896 late_initcall_sync(trace_eval_sync);
9899 #ifdef CONFIG_MODULES
9900 static void trace_module_add_evals(struct module *mod)
9902 if (!mod->num_trace_evals)
9906 * Modules with bad taint do not have events created, do
9907 * not bother with enums either.
9909 if (trace_module_has_bad_taint(mod))
9912 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9915 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
9916 static void trace_module_remove_evals(struct module *mod)
9918 union trace_eval_map_item *map;
9919 union trace_eval_map_item **last = &trace_eval_maps;
9921 if (!mod->num_trace_evals)
9924 mutex_lock(&trace_eval_mutex);
9926 map = trace_eval_maps;
9929 if (map->head.mod == mod)
9931 map = trace_eval_jmp_to_tail(map);
9932 last = &map->tail.next;
9933 map = map->tail.next;
9938 *last = trace_eval_jmp_to_tail(map)->tail.next;
9941 mutex_unlock(&trace_eval_mutex);
9944 static inline void trace_module_remove_evals(struct module *mod) { }
9945 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9947 static int trace_module_notify(struct notifier_block *self,
9948 unsigned long val, void *data)
9950 struct module *mod = data;
9953 case MODULE_STATE_COMING:
9954 trace_module_add_evals(mod);
9956 case MODULE_STATE_GOING:
9957 trace_module_remove_evals(mod);
9964 static struct notifier_block trace_module_nb = {
9965 .notifier_call = trace_module_notify,
9968 #endif /* CONFIG_MODULES */
9970 static __init void tracer_init_tracefs_work_func(struct work_struct *work)
9975 init_tracer_tracefs(&global_trace, NULL);
9976 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9978 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9979 &global_trace, &tracing_thresh_fops);
9981 trace_create_file("README", TRACE_MODE_READ, NULL,
9982 NULL, &tracing_readme_fops);
9984 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9985 NULL, &tracing_saved_cmdlines_fops);
9987 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9988 NULL, &tracing_saved_cmdlines_size_fops);
9990 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9991 NULL, &tracing_saved_tgids_fops);
9993 trace_create_eval_file(NULL);
9995 #ifdef CONFIG_MODULES
9996 register_module_notifier(&trace_module_nb);
9999 #ifdef CONFIG_DYNAMIC_FTRACE
10000 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
10001 NULL, &tracing_dyn_info_fops);
10004 create_trace_instances(NULL);
10006 update_tracer_options(&global_trace);
10009 static __init int tracer_init_tracefs(void)
10013 trace_access_lock_init();
10015 ret = tracing_init_dentry();
10020 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
10021 queue_work(eval_map_wq, &tracerfs_init_work);
10023 tracer_init_tracefs_work_func(NULL);
10026 rv_init_interface();
10031 fs_initcall(tracer_init_tracefs);
10033 static int trace_die_panic_handler(struct notifier_block *self,
10034 unsigned long ev, void *unused);
10036 static struct notifier_block trace_panic_notifier = {
10037 .notifier_call = trace_die_panic_handler,
10038 .priority = INT_MAX - 1,
10041 static struct notifier_block trace_die_notifier = {
10042 .notifier_call = trace_die_panic_handler,
10043 .priority = INT_MAX - 1,
10047 * The idea is to execute the following die/panic callback early, in order
10048 * to avoid showing irrelevant information in the trace (like other panic
10049 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
10050 * warnings get disabled (to prevent potential log flooding).
10052 static int trace_die_panic_handler(struct notifier_block *self,
10053 unsigned long ev, void *unused)
10055 if (!ftrace_dump_on_oops)
10056 return NOTIFY_DONE;
10058 /* The die notifier requires DIE_OOPS to trigger */
10059 if (self == &trace_die_notifier && ev != DIE_OOPS)
10060 return NOTIFY_DONE;
10062 ftrace_dump(ftrace_dump_on_oops);
10064 return NOTIFY_DONE;
10068 * printk is set to max of 1024, we really don't need it that big.
10069 * Nothing should be printing 1000 characters anyway.
10071 #define TRACE_MAX_PRINT 1000
10074 * Define here KERN_TRACE so that we have one place to modify
10075 * it if we decide to change what log level the ftrace dump
10078 #define KERN_TRACE KERN_EMERG
10081 trace_printk_seq(struct trace_seq *s)
10083 /* Probably should print a warning here. */
10084 if (s->seq.len >= TRACE_MAX_PRINT)
10085 s->seq.len = TRACE_MAX_PRINT;
10088 * More paranoid code. Although the buffer size is set to
10089 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
10090 * an extra layer of protection.
10092 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
10093 s->seq.len = s->seq.size - 1;
10095 /* should be zero ended, but we are paranoid. */
10096 s->buffer[s->seq.len] = 0;
10098 printk(KERN_TRACE "%s", s->buffer);
10103 void trace_init_global_iter(struct trace_iterator *iter)
10105 iter->tr = &global_trace;
10106 iter->trace = iter->tr->current_trace;
10107 iter->cpu_file = RING_BUFFER_ALL_CPUS;
10108 iter->array_buffer = &global_trace.array_buffer;
10110 if (iter->trace && iter->trace->open)
10111 iter->trace->open(iter);
10113 /* Annotate start of buffers if we had overruns */
10114 if (ring_buffer_overruns(iter->array_buffer->buffer))
10115 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10117 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10118 if (trace_clocks[iter->tr->clock_id].in_ns)
10119 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10121 /* Can not use kmalloc for iter.temp and iter.fmt */
10122 iter->temp = static_temp_buf;
10123 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10124 iter->fmt = static_fmt_buf;
10125 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10128 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10130 /* use static because iter can be a bit big for the stack */
10131 static struct trace_iterator iter;
10132 static atomic_t dump_running;
10133 struct trace_array *tr = &global_trace;
10134 unsigned int old_userobj;
10135 unsigned long flags;
10138 /* Only allow one dump user at a time. */
10139 if (atomic_inc_return(&dump_running) != 1) {
10140 atomic_dec(&dump_running);
10145 * Always turn off tracing when we dump.
10146 * We don't need to show trace output of what happens
10147 * between multiple crashes.
10149 * If the user does a sysrq-z, then they can re-enable
10150 * tracing with echo 1 > tracing_on.
10154 local_irq_save(flags);
10156 /* Simulate the iterator */
10157 trace_init_global_iter(&iter);
10159 for_each_tracing_cpu(cpu) {
10160 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10163 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10165 /* don't look at user memory in panic mode */
10166 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10168 switch (oops_dump_mode) {
10170 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10173 iter.cpu_file = raw_smp_processor_id();
10178 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10179 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10182 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10184 /* Did function tracer already get disabled? */
10185 if (ftrace_is_dead()) {
10186 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10187 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10191 * We need to stop all tracing on all CPUS to read
10192 * the next buffer. This is a bit expensive, but is
10193 * not done often. We fill all what we can read,
10194 * and then release the locks again.
10197 while (!trace_empty(&iter)) {
10200 printk(KERN_TRACE "---------------------------------\n");
10204 trace_iterator_reset(&iter);
10205 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10207 if (trace_find_next_entry_inc(&iter) != NULL) {
10210 ret = print_trace_line(&iter);
10211 if (ret != TRACE_TYPE_NO_CONSUME)
10212 trace_consume(&iter);
10214 touch_nmi_watchdog();
10216 trace_printk_seq(&iter.seq);
10220 printk(KERN_TRACE " (ftrace buffer empty)\n");
10222 printk(KERN_TRACE "---------------------------------\n");
10225 tr->trace_flags |= old_userobj;
10227 for_each_tracing_cpu(cpu) {
10228 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10230 atomic_dec(&dump_running);
10231 local_irq_restore(flags);
10233 EXPORT_SYMBOL_GPL(ftrace_dump);
10235 #define WRITE_BUFSIZE 4096
10237 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10238 size_t count, loff_t *ppos,
10239 int (*createfn)(const char *))
10241 char *kbuf, *buf, *tmp;
10246 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10250 while (done < count) {
10251 size = count - done;
10253 if (size >= WRITE_BUFSIZE)
10254 size = WRITE_BUFSIZE - 1;
10256 if (copy_from_user(kbuf, buffer + done, size)) {
10263 tmp = strchr(buf, '\n');
10266 size = tmp - buf + 1;
10268 size = strlen(buf);
10269 if (done + size < count) {
10272 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10273 pr_warn("Line length is too long: Should be less than %d\n",
10274 WRITE_BUFSIZE - 2);
10281 /* Remove comments */
10282 tmp = strchr(buf, '#');
10287 ret = createfn(buf);
10292 } while (done < count);
10302 #ifdef CONFIG_TRACER_MAX_TRACE
10303 __init static bool tr_needs_alloc_snapshot(const char *name)
10306 int len = strlen(name);
10309 if (!boot_snapshot_index)
10312 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10313 boot_snapshot_info[len] == '\t')
10316 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10320 sprintf(test, "\t%s\t", name);
10321 ret = strstr(boot_snapshot_info, test) == NULL;
10326 __init static void do_allocate_snapshot(const char *name)
10328 if (!tr_needs_alloc_snapshot(name))
10332 * When allocate_snapshot is set, the next call to
10333 * allocate_trace_buffers() (called by trace_array_get_by_name())
10334 * will allocate the snapshot buffer. That will alse clear
10337 allocate_snapshot = true;
10340 static inline void do_allocate_snapshot(const char *name) { }
10343 __init static void enable_instances(void)
10345 struct trace_array *tr;
10350 /* A tab is always appended */
10351 boot_instance_info[boot_instance_index - 1] = '\0';
10352 str = boot_instance_info;
10354 while ((curr_str = strsep(&str, "\t"))) {
10356 tok = strsep(&curr_str, ",");
10358 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10359 do_allocate_snapshot(tok);
10361 tr = trace_array_get_by_name(tok);
10363 pr_warn("Failed to create instance buffer %s\n", curr_str);
10366 /* Allow user space to delete it */
10367 trace_array_put(tr);
10369 while ((tok = strsep(&curr_str, ","))) {
10370 early_enable_events(tr, tok, true);
10375 __init static int tracer_alloc_buffers(void)
10381 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10382 pr_warn("Tracing disabled due to lockdown\n");
10387 * Make sure we don't accidentally add more trace options
10388 * than we have bits for.
10390 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10392 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10395 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10396 goto out_free_buffer_mask;
10398 /* Only allocate trace_printk buffers if a trace_printk exists */
10399 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10400 /* Must be called before global_trace.buffer is allocated */
10401 trace_printk_init_buffers();
10403 /* To save memory, keep the ring buffer size to its minimum */
10404 if (ring_buffer_expanded)
10405 ring_buf_size = trace_buf_size;
10409 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10410 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10412 raw_spin_lock_init(&global_trace.start_lock);
10415 * The prepare callbacks allocates some memory for the ring buffer. We
10416 * don't free the buffer if the CPU goes down. If we were to free
10417 * the buffer, then the user would lose any trace that was in the
10418 * buffer. The memory will be removed once the "instance" is removed.
10420 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10421 "trace/RB:prepare", trace_rb_cpu_prepare,
10424 goto out_free_cpumask;
10425 /* Used for event triggers */
10427 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10429 goto out_rm_hp_state;
10431 if (trace_create_savedcmd() < 0)
10432 goto out_free_temp_buffer;
10434 if (!alloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10435 goto out_free_savedcmd;
10437 /* TODO: make the number of buffers hot pluggable with CPUS */
10438 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10439 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10440 goto out_free_pipe_cpumask;
10442 if (global_trace.buffer_disabled)
10445 if (trace_boot_clock) {
10446 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10448 pr_warn("Trace clock %s not defined, going back to default\n",
10453 * register_tracer() might reference current_trace, so it
10454 * needs to be set before we register anything. This is
10455 * just a bootstrap of current_trace anyway.
10457 global_trace.current_trace = &nop_trace;
10459 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10461 ftrace_init_global_array_ops(&global_trace);
10463 init_trace_flags_index(&global_trace);
10465 register_tracer(&nop_trace);
10467 /* Function tracing may start here (via kernel command line) */
10468 init_function_trace();
10470 /* All seems OK, enable tracing */
10471 tracing_disabled = 0;
10473 atomic_notifier_chain_register(&panic_notifier_list,
10474 &trace_panic_notifier);
10476 register_die_notifier(&trace_die_notifier);
10478 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10480 INIT_LIST_HEAD(&global_trace.systems);
10481 INIT_LIST_HEAD(&global_trace.events);
10482 INIT_LIST_HEAD(&global_trace.hist_vars);
10483 INIT_LIST_HEAD(&global_trace.err_log);
10484 list_add(&global_trace.list, &ftrace_trace_arrays);
10486 apply_trace_boot_options();
10488 register_snapshot_cmd();
10494 out_free_pipe_cpumask:
10495 free_cpumask_var(global_trace.pipe_cpumask);
10497 free_saved_cmdlines_buffer(savedcmd);
10498 out_free_temp_buffer:
10499 ring_buffer_free(temp_buffer);
10501 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10503 free_cpumask_var(global_trace.tracing_cpumask);
10504 out_free_buffer_mask:
10505 free_cpumask_var(tracing_buffer_mask);
10510 void __init ftrace_boot_snapshot(void)
10512 #ifdef CONFIG_TRACER_MAX_TRACE
10513 struct trace_array *tr;
10515 if (!snapshot_at_boot)
10518 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10519 if (!tr->allocated_snapshot)
10522 tracing_snapshot_instance(tr);
10523 trace_array_puts(tr, "** Boot snapshot taken **\n");
10528 void __init early_trace_init(void)
10530 if (tracepoint_printk) {
10531 tracepoint_print_iter =
10532 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10533 if (MEM_FAIL(!tracepoint_print_iter,
10534 "Failed to allocate trace iterator\n"))
10535 tracepoint_printk = 0;
10537 static_key_enable(&tracepoint_printk_key.key);
10539 tracer_alloc_buffers();
10544 void __init trace_init(void)
10546 trace_event_init();
10548 if (boot_instance_index)
10549 enable_instances();
10552 __init static void clear_boot_tracer(void)
10555 * The default tracer at boot buffer is an init section.
10556 * This function is called in lateinit. If we did not
10557 * find the boot tracer, then clear it out, to prevent
10558 * later registration from accessing the buffer that is
10559 * about to be freed.
10561 if (!default_bootup_tracer)
10564 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10565 default_bootup_tracer);
10566 default_bootup_tracer = NULL;
10569 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10570 __init static void tracing_set_default_clock(void)
10572 /* sched_clock_stable() is determined in late_initcall */
10573 if (!trace_boot_clock && !sched_clock_stable()) {
10574 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10575 pr_warn("Can not set tracing clock due to lockdown\n");
10579 printk(KERN_WARNING
10580 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10581 "If you want to keep using the local clock, then add:\n"
10582 " \"trace_clock=local\"\n"
10583 "on the kernel command line\n");
10584 tracing_set_clock(&global_trace, "global");
10588 static inline void tracing_set_default_clock(void) { }
10591 __init static int late_trace_init(void)
10593 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10594 static_key_disable(&tracepoint_printk_key.key);
10595 tracepoint_printk = 0;
10598 tracing_set_default_clock();
10599 clear_boot_tracer();
10603 late_initcall_sync(late_trace_init);