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",
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, &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);
1897 /* Any waiters on the old snapshot buffer need to wake up */
1898 ring_buffer_wake_waiters(tr->array_buffer.buffer, RING_BUFFER_ALL_CPUS);
1902 * update_max_tr_single - only copy one trace over, and reset the rest
1904 * @tsk: task with the latency
1905 * @cpu: the cpu of the buffer to copy.
1907 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1910 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1917 WARN_ON_ONCE(!irqs_disabled());
1918 if (!tr->allocated_snapshot) {
1919 /* Only the nop tracer should hit this when disabling */
1920 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1924 arch_spin_lock(&tr->max_lock);
1926 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1928 if (ret == -EBUSY) {
1930 * We failed to swap the buffer due to a commit taking
1931 * place on this CPU. We fail to record, but we reset
1932 * the max trace buffer (no one writes directly to it)
1933 * and flag that it failed.
1934 * Another reason is resize is in progress.
1936 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1937 "Failed to swap buffers due to commit or resize in progress\n");
1940 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1942 __update_max_tr(tr, tsk, cpu);
1943 arch_spin_unlock(&tr->max_lock);
1946 #endif /* CONFIG_TRACER_MAX_TRACE */
1948 static int wait_on_pipe(struct trace_iterator *iter, int full)
1952 /* Iterators are static, they should be filled or empty */
1953 if (trace_buffer_iter(iter, iter->cpu_file))
1956 ret = ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file, full);
1958 #ifdef CONFIG_TRACER_MAX_TRACE
1960 * Make sure this is still the snapshot buffer, as if a snapshot were
1961 * to happen, this would now be the main buffer.
1964 iter->array_buffer = &iter->tr->max_buffer;
1969 #ifdef CONFIG_FTRACE_STARTUP_TEST
1970 static bool selftests_can_run;
1972 struct trace_selftests {
1973 struct list_head list;
1974 struct tracer *type;
1977 static LIST_HEAD(postponed_selftests);
1979 static int save_selftest(struct tracer *type)
1981 struct trace_selftests *selftest;
1983 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1987 selftest->type = type;
1988 list_add(&selftest->list, &postponed_selftests);
1992 static int run_tracer_selftest(struct tracer *type)
1994 struct trace_array *tr = &global_trace;
1995 struct tracer *saved_tracer = tr->current_trace;
1998 if (!type->selftest || tracing_selftest_disabled)
2002 * If a tracer registers early in boot up (before scheduling is
2003 * initialized and such), then do not run its selftests yet.
2004 * Instead, run it a little later in the boot process.
2006 if (!selftests_can_run)
2007 return save_selftest(type);
2009 if (!tracing_is_on()) {
2010 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2016 * Run a selftest on this tracer.
2017 * Here we reset the trace buffer, and set the current
2018 * tracer to be this tracer. The tracer can then run some
2019 * internal tracing to verify that everything is in order.
2020 * If we fail, we do not register this tracer.
2022 tracing_reset_online_cpus(&tr->array_buffer);
2024 tr->current_trace = type;
2026 #ifdef CONFIG_TRACER_MAX_TRACE
2027 if (type->use_max_tr) {
2028 /* If we expanded the buffers, make sure the max is expanded too */
2029 if (ring_buffer_expanded)
2030 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2031 RING_BUFFER_ALL_CPUS);
2032 tr->allocated_snapshot = true;
2036 /* the test is responsible for initializing and enabling */
2037 pr_info("Testing tracer %s: ", type->name);
2038 ret = type->selftest(type, tr);
2039 /* the test is responsible for resetting too */
2040 tr->current_trace = saved_tracer;
2042 printk(KERN_CONT "FAILED!\n");
2043 /* Add the warning after printing 'FAILED' */
2047 /* Only reset on passing, to avoid touching corrupted buffers */
2048 tracing_reset_online_cpus(&tr->array_buffer);
2050 #ifdef CONFIG_TRACER_MAX_TRACE
2051 if (type->use_max_tr) {
2052 tr->allocated_snapshot = false;
2054 /* Shrink the max buffer again */
2055 if (ring_buffer_expanded)
2056 ring_buffer_resize(tr->max_buffer.buffer, 1,
2057 RING_BUFFER_ALL_CPUS);
2061 printk(KERN_CONT "PASSED\n");
2065 static int do_run_tracer_selftest(struct tracer *type)
2070 * Tests can take a long time, especially if they are run one after the
2071 * other, as does happen during bootup when all the tracers are
2072 * registered. This could cause the soft lockup watchdog to trigger.
2076 tracing_selftest_running = true;
2077 ret = run_tracer_selftest(type);
2078 tracing_selftest_running = false;
2083 static __init int init_trace_selftests(void)
2085 struct trace_selftests *p, *n;
2086 struct tracer *t, **last;
2089 selftests_can_run = true;
2091 mutex_lock(&trace_types_lock);
2093 if (list_empty(&postponed_selftests))
2096 pr_info("Running postponed tracer tests:\n");
2098 tracing_selftest_running = true;
2099 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2100 /* This loop can take minutes when sanitizers are enabled, so
2101 * lets make sure we allow RCU processing.
2104 ret = run_tracer_selftest(p->type);
2105 /* If the test fails, then warn and remove from available_tracers */
2107 WARN(1, "tracer: %s failed selftest, disabling\n",
2109 last = &trace_types;
2110 for (t = trace_types; t; t = t->next) {
2121 tracing_selftest_running = false;
2124 mutex_unlock(&trace_types_lock);
2128 core_initcall(init_trace_selftests);
2130 static inline int run_tracer_selftest(struct tracer *type)
2134 static inline int do_run_tracer_selftest(struct tracer *type)
2138 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2140 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2142 static void __init apply_trace_boot_options(void);
2145 * register_tracer - register a tracer with the ftrace system.
2146 * @type: the plugin for the tracer
2148 * Register a new plugin tracer.
2150 int __init register_tracer(struct tracer *type)
2156 pr_info("Tracer must have a name\n");
2160 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2161 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2165 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2166 pr_warn("Can not register tracer %s due to lockdown\n",
2171 mutex_lock(&trace_types_lock);
2173 for (t = trace_types; t; t = t->next) {
2174 if (strcmp(type->name, t->name) == 0) {
2176 pr_info("Tracer %s already registered\n",
2183 if (!type->set_flag)
2184 type->set_flag = &dummy_set_flag;
2186 /*allocate a dummy tracer_flags*/
2187 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2192 type->flags->val = 0;
2193 type->flags->opts = dummy_tracer_opt;
2195 if (!type->flags->opts)
2196 type->flags->opts = dummy_tracer_opt;
2198 /* store the tracer for __set_tracer_option */
2199 type->flags->trace = type;
2201 ret = do_run_tracer_selftest(type);
2205 type->next = trace_types;
2207 add_tracer_options(&global_trace, type);
2210 mutex_unlock(&trace_types_lock);
2212 if (ret || !default_bootup_tracer)
2215 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2218 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2219 /* Do we want this tracer to start on bootup? */
2220 tracing_set_tracer(&global_trace, type->name);
2221 default_bootup_tracer = NULL;
2223 apply_trace_boot_options();
2225 /* disable other selftests, since this will break it. */
2226 disable_tracing_selftest("running a tracer");
2232 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2234 struct trace_buffer *buffer = buf->buffer;
2239 ring_buffer_record_disable(buffer);
2241 /* Make sure all commits have finished */
2243 ring_buffer_reset_cpu(buffer, cpu);
2245 ring_buffer_record_enable(buffer);
2248 void tracing_reset_online_cpus(struct array_buffer *buf)
2250 struct trace_buffer *buffer = buf->buffer;
2255 ring_buffer_record_disable(buffer);
2257 /* Make sure all commits have finished */
2260 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2262 ring_buffer_reset_online_cpus(buffer);
2264 ring_buffer_record_enable(buffer);
2267 /* Must have trace_types_lock held */
2268 void tracing_reset_all_online_cpus_unlocked(void)
2270 struct trace_array *tr;
2272 lockdep_assert_held(&trace_types_lock);
2274 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2275 if (!tr->clear_trace)
2277 tr->clear_trace = false;
2278 tracing_reset_online_cpus(&tr->array_buffer);
2279 #ifdef CONFIG_TRACER_MAX_TRACE
2280 tracing_reset_online_cpus(&tr->max_buffer);
2285 void tracing_reset_all_online_cpus(void)
2287 mutex_lock(&trace_types_lock);
2288 tracing_reset_all_online_cpus_unlocked();
2289 mutex_unlock(&trace_types_lock);
2293 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2294 * is the tgid last observed corresponding to pid=i.
2296 static int *tgid_map;
2298 /* The maximum valid index into tgid_map. */
2299 static size_t tgid_map_max;
2301 #define SAVED_CMDLINES_DEFAULT 128
2302 #define NO_CMDLINE_MAP UINT_MAX
2304 * Preemption must be disabled before acquiring trace_cmdline_lock.
2305 * The various trace_arrays' max_lock must be acquired in a context
2306 * where interrupt is disabled.
2308 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2309 struct saved_cmdlines_buffer {
2310 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2311 unsigned *map_cmdline_to_pid;
2312 unsigned cmdline_num;
2314 char *saved_cmdlines;
2316 static struct saved_cmdlines_buffer *savedcmd;
2318 static inline char *get_saved_cmdlines(int idx)
2320 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2323 static inline void set_cmdline(int idx, const char *cmdline)
2325 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2328 static int allocate_cmdlines_buffer(unsigned int val,
2329 struct saved_cmdlines_buffer *s)
2331 s->map_cmdline_to_pid = kmalloc_array(val,
2332 sizeof(*s->map_cmdline_to_pid),
2334 if (!s->map_cmdline_to_pid)
2337 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2338 if (!s->saved_cmdlines) {
2339 kfree(s->map_cmdline_to_pid);
2344 s->cmdline_num = val;
2345 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2346 sizeof(s->map_pid_to_cmdline));
2347 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2348 val * sizeof(*s->map_cmdline_to_pid));
2353 static int trace_create_savedcmd(void)
2357 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2361 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2371 int is_tracing_stopped(void)
2373 return global_trace.stop_count;
2376 static void tracing_start_tr(struct trace_array *tr)
2378 struct trace_buffer *buffer;
2379 unsigned long flags;
2381 if (tracing_disabled)
2384 raw_spin_lock_irqsave(&tr->start_lock, flags);
2385 if (--tr->stop_count) {
2386 if (WARN_ON_ONCE(tr->stop_count < 0)) {
2387 /* Someone screwed up their debugging */
2393 /* Prevent the buffers from switching */
2394 arch_spin_lock(&tr->max_lock);
2396 buffer = tr->array_buffer.buffer;
2398 ring_buffer_record_enable(buffer);
2400 #ifdef CONFIG_TRACER_MAX_TRACE
2401 buffer = tr->max_buffer.buffer;
2403 ring_buffer_record_enable(buffer);
2406 arch_spin_unlock(&tr->max_lock);
2409 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2413 * tracing_start - quick start of the tracer
2415 * If tracing is enabled but was stopped by tracing_stop,
2416 * this will start the tracer back up.
2418 void tracing_start(void)
2421 return tracing_start_tr(&global_trace);
2424 static void tracing_stop_tr(struct trace_array *tr)
2426 struct trace_buffer *buffer;
2427 unsigned long flags;
2429 raw_spin_lock_irqsave(&tr->start_lock, flags);
2430 if (tr->stop_count++)
2433 /* Prevent the buffers from switching */
2434 arch_spin_lock(&tr->max_lock);
2436 buffer = tr->array_buffer.buffer;
2438 ring_buffer_record_disable(buffer);
2440 #ifdef CONFIG_TRACER_MAX_TRACE
2441 buffer = tr->max_buffer.buffer;
2443 ring_buffer_record_disable(buffer);
2446 arch_spin_unlock(&tr->max_lock);
2449 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2453 * tracing_stop - quick stop of the tracer
2455 * Light weight way to stop tracing. Use in conjunction with
2458 void tracing_stop(void)
2460 return tracing_stop_tr(&global_trace);
2463 static int trace_save_cmdline(struct task_struct *tsk)
2467 /* treat recording of idle task as a success */
2471 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2474 * It's not the end of the world if we don't get
2475 * the lock, but we also don't want to spin
2476 * nor do we want to disable interrupts,
2477 * so if we miss here, then better luck next time.
2479 * This is called within the scheduler and wake up, so interrupts
2480 * had better been disabled and run queue lock been held.
2482 lockdep_assert_preemption_disabled();
2483 if (!arch_spin_trylock(&trace_cmdline_lock))
2486 idx = savedcmd->map_pid_to_cmdline[tpid];
2487 if (idx == NO_CMDLINE_MAP) {
2488 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2490 savedcmd->map_pid_to_cmdline[tpid] = idx;
2491 savedcmd->cmdline_idx = idx;
2494 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2495 set_cmdline(idx, tsk->comm);
2497 arch_spin_unlock(&trace_cmdline_lock);
2502 static void __trace_find_cmdline(int pid, char comm[])
2508 strcpy(comm, "<idle>");
2512 if (WARN_ON_ONCE(pid < 0)) {
2513 strcpy(comm, "<XXX>");
2517 tpid = pid & (PID_MAX_DEFAULT - 1);
2518 map = savedcmd->map_pid_to_cmdline[tpid];
2519 if (map != NO_CMDLINE_MAP) {
2520 tpid = savedcmd->map_cmdline_to_pid[map];
2522 strscpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2526 strcpy(comm, "<...>");
2529 void trace_find_cmdline(int pid, char comm[])
2532 arch_spin_lock(&trace_cmdline_lock);
2534 __trace_find_cmdline(pid, comm);
2536 arch_spin_unlock(&trace_cmdline_lock);
2540 static int *trace_find_tgid_ptr(int pid)
2543 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2544 * if we observe a non-NULL tgid_map then we also observe the correct
2547 int *map = smp_load_acquire(&tgid_map);
2549 if (unlikely(!map || pid > tgid_map_max))
2555 int trace_find_tgid(int pid)
2557 int *ptr = trace_find_tgid_ptr(pid);
2559 return ptr ? *ptr : 0;
2562 static int trace_save_tgid(struct task_struct *tsk)
2566 /* treat recording of idle task as a success */
2570 ptr = trace_find_tgid_ptr(tsk->pid);
2578 static bool tracing_record_taskinfo_skip(int flags)
2580 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2582 if (!__this_cpu_read(trace_taskinfo_save))
2588 * tracing_record_taskinfo - record the task info of a task
2590 * @task: task to record
2591 * @flags: TRACE_RECORD_CMDLINE for recording comm
2592 * TRACE_RECORD_TGID for recording tgid
2594 void tracing_record_taskinfo(struct task_struct *task, int flags)
2598 if (tracing_record_taskinfo_skip(flags))
2602 * Record as much task information as possible. If some fail, continue
2603 * to try to record the others.
2605 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2606 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2608 /* If recording any information failed, retry again soon. */
2612 __this_cpu_write(trace_taskinfo_save, false);
2616 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2618 * @prev: previous task during sched_switch
2619 * @next: next task during sched_switch
2620 * @flags: TRACE_RECORD_CMDLINE for recording comm
2621 * TRACE_RECORD_TGID for recording tgid
2623 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2624 struct task_struct *next, int flags)
2628 if (tracing_record_taskinfo_skip(flags))
2632 * Record as much task information as possible. If some fail, continue
2633 * to try to record the others.
2635 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2636 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2637 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2638 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2640 /* If recording any information failed, retry again soon. */
2644 __this_cpu_write(trace_taskinfo_save, false);
2647 /* Helpers to record a specific task information */
2648 void tracing_record_cmdline(struct task_struct *task)
2650 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2653 void tracing_record_tgid(struct task_struct *task)
2655 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2659 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2660 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2661 * simplifies those functions and keeps them in sync.
2663 enum print_line_t trace_handle_return(struct trace_seq *s)
2665 return trace_seq_has_overflowed(s) ?
2666 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2668 EXPORT_SYMBOL_GPL(trace_handle_return);
2670 static unsigned short migration_disable_value(void)
2672 #if defined(CONFIG_SMP)
2673 return current->migration_disabled;
2679 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2681 unsigned int trace_flags = irqs_status;
2684 pc = preempt_count();
2687 trace_flags |= TRACE_FLAG_NMI;
2688 if (pc & HARDIRQ_MASK)
2689 trace_flags |= TRACE_FLAG_HARDIRQ;
2690 if (in_serving_softirq())
2691 trace_flags |= TRACE_FLAG_SOFTIRQ;
2692 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2693 trace_flags |= TRACE_FLAG_BH_OFF;
2695 if (tif_need_resched())
2696 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2697 if (test_preempt_need_resched())
2698 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2699 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2700 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2703 struct ring_buffer_event *
2704 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2707 unsigned int trace_ctx)
2709 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2712 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2713 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2714 static int trace_buffered_event_ref;
2717 * trace_buffered_event_enable - enable buffering events
2719 * When events are being filtered, it is quicker to use a temporary
2720 * buffer to write the event data into if there's a likely chance
2721 * that it will not be committed. The discard of the ring buffer
2722 * is not as fast as committing, and is much slower than copying
2725 * When an event is to be filtered, allocate per cpu buffers to
2726 * write the event data into, and if the event is filtered and discarded
2727 * it is simply dropped, otherwise, the entire data is to be committed
2730 void trace_buffered_event_enable(void)
2732 struct ring_buffer_event *event;
2736 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2738 if (trace_buffered_event_ref++)
2741 for_each_tracing_cpu(cpu) {
2742 page = alloc_pages_node(cpu_to_node(cpu),
2743 GFP_KERNEL | __GFP_NORETRY, 0);
2744 /* This is just an optimization and can handle failures */
2746 pr_err("Failed to allocate event buffer\n");
2750 event = page_address(page);
2751 memset(event, 0, sizeof(*event));
2753 per_cpu(trace_buffered_event, cpu) = event;
2756 if (cpu == smp_processor_id() &&
2757 __this_cpu_read(trace_buffered_event) !=
2758 per_cpu(trace_buffered_event, cpu))
2764 static void enable_trace_buffered_event(void *data)
2766 /* Probably not needed, but do it anyway */
2768 this_cpu_dec(trace_buffered_event_cnt);
2771 static void disable_trace_buffered_event(void *data)
2773 this_cpu_inc(trace_buffered_event_cnt);
2777 * trace_buffered_event_disable - disable buffering events
2779 * When a filter is removed, it is faster to not use the buffered
2780 * events, and to commit directly into the ring buffer. Free up
2781 * the temp buffers when there are no more users. This requires
2782 * special synchronization with current events.
2784 void trace_buffered_event_disable(void)
2788 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2790 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2793 if (--trace_buffered_event_ref)
2796 /* For each CPU, set the buffer as used. */
2797 on_each_cpu_mask(tracing_buffer_mask, disable_trace_buffered_event,
2800 /* Wait for all current users to finish */
2803 for_each_tracing_cpu(cpu) {
2804 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2805 per_cpu(trace_buffered_event, cpu) = NULL;
2809 * Wait for all CPUs that potentially started checking if they can use
2810 * their event buffer only after the previous synchronize_rcu() call and
2811 * they still read a valid pointer from trace_buffered_event. It must be
2812 * ensured they don't see cleared trace_buffered_event_cnt else they
2813 * could wrongly decide to use the pointed-to buffer which is now freed.
2817 /* For each CPU, relinquish the buffer */
2818 on_each_cpu_mask(tracing_buffer_mask, enable_trace_buffered_event, NULL,
2822 static struct trace_buffer *temp_buffer;
2824 struct ring_buffer_event *
2825 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2826 struct trace_event_file *trace_file,
2827 int type, unsigned long len,
2828 unsigned int trace_ctx)
2830 struct ring_buffer_event *entry;
2831 struct trace_array *tr = trace_file->tr;
2834 *current_rb = tr->array_buffer.buffer;
2836 if (!tr->no_filter_buffering_ref &&
2837 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2838 preempt_disable_notrace();
2840 * Filtering is on, so try to use the per cpu buffer first.
2841 * This buffer will simulate a ring_buffer_event,
2842 * where the type_len is zero and the array[0] will
2843 * hold the full length.
2844 * (see include/linux/ring-buffer.h for details on
2845 * how the ring_buffer_event is structured).
2847 * Using a temp buffer during filtering and copying it
2848 * on a matched filter is quicker than writing directly
2849 * into the ring buffer and then discarding it when
2850 * it doesn't match. That is because the discard
2851 * requires several atomic operations to get right.
2852 * Copying on match and doing nothing on a failed match
2853 * is still quicker than no copy on match, but having
2854 * to discard out of the ring buffer on a failed match.
2856 if ((entry = __this_cpu_read(trace_buffered_event))) {
2857 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2859 val = this_cpu_inc_return(trace_buffered_event_cnt);
2862 * Preemption is disabled, but interrupts and NMIs
2863 * can still come in now. If that happens after
2864 * the above increment, then it will have to go
2865 * back to the old method of allocating the event
2866 * on the ring buffer, and if the filter fails, it
2867 * will have to call ring_buffer_discard_commit()
2870 * Need to also check the unlikely case that the
2871 * length is bigger than the temp buffer size.
2872 * If that happens, then the reserve is pretty much
2873 * guaranteed to fail, as the ring buffer currently
2874 * only allows events less than a page. But that may
2875 * change in the future, so let the ring buffer reserve
2876 * handle the failure in that case.
2878 if (val == 1 && likely(len <= max_len)) {
2879 trace_event_setup(entry, type, trace_ctx);
2880 entry->array[0] = len;
2881 /* Return with preemption disabled */
2884 this_cpu_dec(trace_buffered_event_cnt);
2886 /* __trace_buffer_lock_reserve() disables preemption */
2887 preempt_enable_notrace();
2890 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2893 * If tracing is off, but we have triggers enabled
2894 * we still need to look at the event data. Use the temp_buffer
2895 * to store the trace event for the trigger to use. It's recursive
2896 * safe and will not be recorded anywhere.
2898 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2899 *current_rb = temp_buffer;
2900 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2905 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2907 static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2908 static DEFINE_MUTEX(tracepoint_printk_mutex);
2910 static void output_printk(struct trace_event_buffer *fbuffer)
2912 struct trace_event_call *event_call;
2913 struct trace_event_file *file;
2914 struct trace_event *event;
2915 unsigned long flags;
2916 struct trace_iterator *iter = tracepoint_print_iter;
2918 /* We should never get here if iter is NULL */
2919 if (WARN_ON_ONCE(!iter))
2922 event_call = fbuffer->trace_file->event_call;
2923 if (!event_call || !event_call->event.funcs ||
2924 !event_call->event.funcs->trace)
2927 file = fbuffer->trace_file;
2928 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2929 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2930 !filter_match_preds(file->filter, fbuffer->entry)))
2933 event = &fbuffer->trace_file->event_call->event;
2935 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2936 trace_seq_init(&iter->seq);
2937 iter->ent = fbuffer->entry;
2938 event_call->event.funcs->trace(iter, 0, event);
2939 trace_seq_putc(&iter->seq, 0);
2940 printk("%s", iter->seq.buffer);
2942 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2945 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2946 void *buffer, size_t *lenp,
2949 int save_tracepoint_printk;
2952 mutex_lock(&tracepoint_printk_mutex);
2953 save_tracepoint_printk = tracepoint_printk;
2955 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2958 * This will force exiting early, as tracepoint_printk
2959 * is always zero when tracepoint_printk_iter is not allocated
2961 if (!tracepoint_print_iter)
2962 tracepoint_printk = 0;
2964 if (save_tracepoint_printk == tracepoint_printk)
2967 if (tracepoint_printk)
2968 static_key_enable(&tracepoint_printk_key.key);
2970 static_key_disable(&tracepoint_printk_key.key);
2973 mutex_unlock(&tracepoint_printk_mutex);
2978 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2980 enum event_trigger_type tt = ETT_NONE;
2981 struct trace_event_file *file = fbuffer->trace_file;
2983 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2984 fbuffer->entry, &tt))
2987 if (static_key_false(&tracepoint_printk_key.key))
2988 output_printk(fbuffer);
2990 if (static_branch_unlikely(&trace_event_exports_enabled))
2991 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2993 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2994 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
2998 event_triggers_post_call(file, tt);
3001 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
3006 * trace_buffer_unlock_commit_regs()
3007 * trace_event_buffer_commit()
3008 * trace_event_raw_event_xxx()
3010 # define STACK_SKIP 3
3012 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
3013 struct trace_buffer *buffer,
3014 struct ring_buffer_event *event,
3015 unsigned int trace_ctx,
3016 struct pt_regs *regs)
3018 __buffer_unlock_commit(buffer, event);
3021 * If regs is not set, then skip the necessary functions.
3022 * Note, we can still get here via blktrace, wakeup tracer
3023 * and mmiotrace, but that's ok if they lose a function or
3024 * two. They are not that meaningful.
3026 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
3027 ftrace_trace_userstack(tr, buffer, trace_ctx);
3031 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
3034 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
3035 struct ring_buffer_event *event)
3037 __buffer_unlock_commit(buffer, event);
3041 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
3042 parent_ip, unsigned int trace_ctx)
3044 struct trace_event_call *call = &event_function;
3045 struct trace_buffer *buffer = tr->array_buffer.buffer;
3046 struct ring_buffer_event *event;
3047 struct ftrace_entry *entry;
3049 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
3053 entry = ring_buffer_event_data(event);
3055 entry->parent_ip = parent_ip;
3057 if (!call_filter_check_discard(call, entry, buffer, event)) {
3058 if (static_branch_unlikely(&trace_function_exports_enabled))
3059 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
3060 __buffer_unlock_commit(buffer, event);
3064 #ifdef CONFIG_STACKTRACE
3066 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
3067 #define FTRACE_KSTACK_NESTING 4
3069 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3071 struct ftrace_stack {
3072 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3076 struct ftrace_stacks {
3077 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3080 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3081 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3083 static void __ftrace_trace_stack(struct trace_buffer *buffer,
3084 unsigned int trace_ctx,
3085 int skip, struct pt_regs *regs)
3087 struct trace_event_call *call = &event_kernel_stack;
3088 struct ring_buffer_event *event;
3089 unsigned int size, nr_entries;
3090 struct ftrace_stack *fstack;
3091 struct stack_entry *entry;
3095 * Add one, for this function and the call to save_stack_trace()
3096 * If regs is set, then these functions will not be in the way.
3098 #ifndef CONFIG_UNWINDER_ORC
3103 preempt_disable_notrace();
3105 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3107 /* This should never happen. If it does, yell once and skip */
3108 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3112 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3113 * interrupt will either see the value pre increment or post
3114 * increment. If the interrupt happens pre increment it will have
3115 * restored the counter when it returns. We just need a barrier to
3116 * keep gcc from moving things around.
3120 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3121 size = ARRAY_SIZE(fstack->calls);
3124 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3127 nr_entries = stack_trace_save(fstack->calls, size, skip);
3130 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3131 struct_size(entry, caller, nr_entries),
3135 entry = ring_buffer_event_data(event);
3137 entry->size = nr_entries;
3138 memcpy(&entry->caller, fstack->calls,
3139 flex_array_size(entry, caller, nr_entries));
3141 if (!call_filter_check_discard(call, entry, buffer, event))
3142 __buffer_unlock_commit(buffer, event);
3145 /* Again, don't let gcc optimize things here */
3147 __this_cpu_dec(ftrace_stack_reserve);
3148 preempt_enable_notrace();
3152 static inline void ftrace_trace_stack(struct trace_array *tr,
3153 struct trace_buffer *buffer,
3154 unsigned int trace_ctx,
3155 int skip, struct pt_regs *regs)
3157 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3160 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3163 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3166 struct trace_buffer *buffer = tr->array_buffer.buffer;
3168 if (rcu_is_watching()) {
3169 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3173 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3177 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3178 * but if the above rcu_is_watching() failed, then the NMI
3179 * triggered someplace critical, and ct_irq_enter() should
3180 * not be called from NMI.
3182 if (unlikely(in_nmi()))
3185 ct_irq_enter_irqson();
3186 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3187 ct_irq_exit_irqson();
3191 * trace_dump_stack - record a stack back trace in the trace buffer
3192 * @skip: Number of functions to skip (helper handlers)
3194 void trace_dump_stack(int skip)
3196 if (tracing_disabled || tracing_selftest_running)
3199 #ifndef CONFIG_UNWINDER_ORC
3200 /* Skip 1 to skip this function. */
3203 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3204 tracing_gen_ctx(), skip, NULL);
3206 EXPORT_SYMBOL_GPL(trace_dump_stack);
3208 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3209 static DEFINE_PER_CPU(int, user_stack_count);
3212 ftrace_trace_userstack(struct trace_array *tr,
3213 struct trace_buffer *buffer, unsigned int trace_ctx)
3215 struct trace_event_call *call = &event_user_stack;
3216 struct ring_buffer_event *event;
3217 struct userstack_entry *entry;
3219 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3223 * NMIs can not handle page faults, even with fix ups.
3224 * The save user stack can (and often does) fault.
3226 if (unlikely(in_nmi()))
3230 * prevent recursion, since the user stack tracing may
3231 * trigger other kernel events.
3234 if (__this_cpu_read(user_stack_count))
3237 __this_cpu_inc(user_stack_count);
3239 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3240 sizeof(*entry), trace_ctx);
3242 goto out_drop_count;
3243 entry = ring_buffer_event_data(event);
3245 entry->tgid = current->tgid;
3246 memset(&entry->caller, 0, sizeof(entry->caller));
3248 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3249 if (!call_filter_check_discard(call, entry, buffer, event))
3250 __buffer_unlock_commit(buffer, event);
3253 __this_cpu_dec(user_stack_count);
3257 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3258 static void ftrace_trace_userstack(struct trace_array *tr,
3259 struct trace_buffer *buffer,
3260 unsigned int trace_ctx)
3263 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3265 #endif /* CONFIG_STACKTRACE */
3268 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3269 unsigned long long delta)
3271 entry->bottom_delta_ts = delta & U32_MAX;
3272 entry->top_delta_ts = (delta >> 32);
3275 void trace_last_func_repeats(struct trace_array *tr,
3276 struct trace_func_repeats *last_info,
3277 unsigned int trace_ctx)
3279 struct trace_buffer *buffer = tr->array_buffer.buffer;
3280 struct func_repeats_entry *entry;
3281 struct ring_buffer_event *event;
3284 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3285 sizeof(*entry), trace_ctx);
3289 delta = ring_buffer_event_time_stamp(buffer, event) -
3290 last_info->ts_last_call;
3292 entry = ring_buffer_event_data(event);
3293 entry->ip = last_info->ip;
3294 entry->parent_ip = last_info->parent_ip;
3295 entry->count = last_info->count;
3296 func_repeats_set_delta_ts(entry, delta);
3298 __buffer_unlock_commit(buffer, event);
3301 /* created for use with alloc_percpu */
3302 struct trace_buffer_struct {
3304 char buffer[4][TRACE_BUF_SIZE];
3307 static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3310 * This allows for lockless recording. If we're nested too deeply, then
3311 * this returns NULL.
3313 static char *get_trace_buf(void)
3315 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3317 if (!trace_percpu_buffer || buffer->nesting >= 4)
3322 /* Interrupts must see nesting incremented before we use the buffer */
3324 return &buffer->buffer[buffer->nesting - 1][0];
3327 static void put_trace_buf(void)
3329 /* Don't let the decrement of nesting leak before this */
3331 this_cpu_dec(trace_percpu_buffer->nesting);
3334 static int alloc_percpu_trace_buffer(void)
3336 struct trace_buffer_struct __percpu *buffers;
3338 if (trace_percpu_buffer)
3341 buffers = alloc_percpu(struct trace_buffer_struct);
3342 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3345 trace_percpu_buffer = buffers;
3349 static int buffers_allocated;
3351 void trace_printk_init_buffers(void)
3353 if (buffers_allocated)
3356 if (alloc_percpu_trace_buffer())
3359 /* trace_printk() is for debug use only. Don't use it in production. */
3362 pr_warn("**********************************************************\n");
3363 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3365 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3367 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3368 pr_warn("** unsafe for production use. **\n");
3370 pr_warn("** If you see this message and you are not debugging **\n");
3371 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3373 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3374 pr_warn("**********************************************************\n");
3376 /* Expand the buffers to set size */
3377 tracing_update_buffers();
3379 buffers_allocated = 1;
3382 * trace_printk_init_buffers() can be called by modules.
3383 * If that happens, then we need to start cmdline recording
3384 * directly here. If the global_trace.buffer is already
3385 * allocated here, then this was called by module code.
3387 if (global_trace.array_buffer.buffer)
3388 tracing_start_cmdline_record();
3390 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3392 void trace_printk_start_comm(void)
3394 /* Start tracing comms if trace printk is set */
3395 if (!buffers_allocated)
3397 tracing_start_cmdline_record();
3400 static void trace_printk_start_stop_comm(int enabled)
3402 if (!buffers_allocated)
3406 tracing_start_cmdline_record();
3408 tracing_stop_cmdline_record();
3412 * trace_vbprintk - write binary msg to tracing buffer
3413 * @ip: The address of the caller
3414 * @fmt: The string format to write to the buffer
3415 * @args: Arguments for @fmt
3417 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3419 struct trace_event_call *call = &event_bprint;
3420 struct ring_buffer_event *event;
3421 struct trace_buffer *buffer;
3422 struct trace_array *tr = &global_trace;
3423 struct bprint_entry *entry;
3424 unsigned int trace_ctx;
3428 if (unlikely(tracing_selftest_running || tracing_disabled))
3431 /* Don't pollute graph traces with trace_vprintk internals */
3432 pause_graph_tracing();
3434 trace_ctx = tracing_gen_ctx();
3435 preempt_disable_notrace();
3437 tbuffer = get_trace_buf();
3443 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3445 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3448 size = sizeof(*entry) + sizeof(u32) * len;
3449 buffer = tr->array_buffer.buffer;
3450 ring_buffer_nest_start(buffer);
3451 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3455 entry = ring_buffer_event_data(event);
3459 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3460 if (!call_filter_check_discard(call, entry, buffer, event)) {
3461 __buffer_unlock_commit(buffer, event);
3462 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3466 ring_buffer_nest_end(buffer);
3471 preempt_enable_notrace();
3472 unpause_graph_tracing();
3476 EXPORT_SYMBOL_GPL(trace_vbprintk);
3480 __trace_array_vprintk(struct trace_buffer *buffer,
3481 unsigned long ip, const char *fmt, va_list args)
3483 struct trace_event_call *call = &event_print;
3484 struct ring_buffer_event *event;
3486 struct print_entry *entry;
3487 unsigned int trace_ctx;
3490 if (tracing_disabled)
3493 /* Don't pollute graph traces with trace_vprintk internals */
3494 pause_graph_tracing();
3496 trace_ctx = tracing_gen_ctx();
3497 preempt_disable_notrace();
3500 tbuffer = get_trace_buf();
3506 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3508 size = sizeof(*entry) + len + 1;
3509 ring_buffer_nest_start(buffer);
3510 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3514 entry = ring_buffer_event_data(event);
3517 memcpy(&entry->buf, tbuffer, len + 1);
3518 if (!call_filter_check_discard(call, entry, buffer, event)) {
3519 __buffer_unlock_commit(buffer, event);
3520 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3524 ring_buffer_nest_end(buffer);
3528 preempt_enable_notrace();
3529 unpause_graph_tracing();
3535 int trace_array_vprintk(struct trace_array *tr,
3536 unsigned long ip, const char *fmt, va_list args)
3538 if (tracing_selftest_running && tr == &global_trace)
3541 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3545 * trace_array_printk - Print a message to a specific instance
3546 * @tr: The instance trace_array descriptor
3547 * @ip: The instruction pointer that this is called from.
3548 * @fmt: The format to print (printf format)
3550 * If a subsystem sets up its own instance, they have the right to
3551 * printk strings into their tracing instance buffer using this
3552 * function. Note, this function will not write into the top level
3553 * buffer (use trace_printk() for that), as writing into the top level
3554 * buffer should only have events that can be individually disabled.
3555 * trace_printk() is only used for debugging a kernel, and should not
3556 * be ever incorporated in normal use.
3558 * trace_array_printk() can be used, as it will not add noise to the
3559 * top level tracing buffer.
3561 * Note, trace_array_init_printk() must be called on @tr before this
3565 int trace_array_printk(struct trace_array *tr,
3566 unsigned long ip, const char *fmt, ...)
3574 /* This is only allowed for created instances */
3575 if (tr == &global_trace)
3578 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3582 ret = trace_array_vprintk(tr, ip, fmt, ap);
3586 EXPORT_SYMBOL_GPL(trace_array_printk);
3589 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3590 * @tr: The trace array to initialize the buffers for
3592 * As trace_array_printk() only writes into instances, they are OK to
3593 * have in the kernel (unlike trace_printk()). This needs to be called
3594 * before trace_array_printk() can be used on a trace_array.
3596 int trace_array_init_printk(struct trace_array *tr)
3601 /* This is only allowed for created instances */
3602 if (tr == &global_trace)
3605 return alloc_percpu_trace_buffer();
3607 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3610 int trace_array_printk_buf(struct trace_buffer *buffer,
3611 unsigned long ip, const char *fmt, ...)
3616 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3620 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3626 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3628 return trace_array_vprintk(&global_trace, ip, fmt, args);
3630 EXPORT_SYMBOL_GPL(trace_vprintk);
3632 static void trace_iterator_increment(struct trace_iterator *iter)
3634 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3638 ring_buffer_iter_advance(buf_iter);
3641 static struct trace_entry *
3642 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3643 unsigned long *lost_events)
3645 struct ring_buffer_event *event;
3646 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3649 event = ring_buffer_iter_peek(buf_iter, ts);
3651 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3652 (unsigned long)-1 : 0;
3654 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3659 iter->ent_size = ring_buffer_event_length(event);
3660 return ring_buffer_event_data(event);
3666 static struct trace_entry *
3667 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3668 unsigned long *missing_events, u64 *ent_ts)
3670 struct trace_buffer *buffer = iter->array_buffer->buffer;
3671 struct trace_entry *ent, *next = NULL;
3672 unsigned long lost_events = 0, next_lost = 0;
3673 int cpu_file = iter->cpu_file;
3674 u64 next_ts = 0, ts;
3680 * If we are in a per_cpu trace file, don't bother by iterating over
3681 * all cpu and peek directly.
3683 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3684 if (ring_buffer_empty_cpu(buffer, cpu_file))
3686 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3688 *ent_cpu = cpu_file;
3693 for_each_tracing_cpu(cpu) {
3695 if (ring_buffer_empty_cpu(buffer, cpu))
3698 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3701 * Pick the entry with the smallest timestamp:
3703 if (ent && (!next || ts < next_ts)) {
3707 next_lost = lost_events;
3708 next_size = iter->ent_size;
3712 iter->ent_size = next_size;
3715 *ent_cpu = next_cpu;
3721 *missing_events = next_lost;
3726 #define STATIC_FMT_BUF_SIZE 128
3727 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3729 char *trace_iter_expand_format(struct trace_iterator *iter)
3734 * iter->tr is NULL when used with tp_printk, which makes
3735 * this get called where it is not safe to call krealloc().
3737 if (!iter->tr || iter->fmt == static_fmt_buf)
3740 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3743 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3750 /* Returns true if the string is safe to dereference from an event */
3751 static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3754 unsigned long addr = (unsigned long)str;
3755 struct trace_event *trace_event;
3756 struct trace_event_call *event;
3758 /* Ignore strings with no length */
3762 /* OK if part of the event data */
3763 if ((addr >= (unsigned long)iter->ent) &&
3764 (addr < (unsigned long)iter->ent + iter->ent_size))
3767 /* OK if part of the temp seq buffer */
3768 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3769 (addr < (unsigned long)iter->tmp_seq.buffer + PAGE_SIZE))
3772 /* Core rodata can not be freed */
3773 if (is_kernel_rodata(addr))
3776 if (trace_is_tracepoint_string(str))
3780 * Now this could be a module event, referencing core module
3781 * data, which is OK.
3786 trace_event = ftrace_find_event(iter->ent->type);
3790 event = container_of(trace_event, struct trace_event_call, event);
3791 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3794 /* Would rather have rodata, but this will suffice */
3795 if (within_module_core(addr, event->module))
3801 static const char *show_buffer(struct trace_seq *s)
3803 struct seq_buf *seq = &s->seq;
3805 seq_buf_terminate(seq);
3810 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3812 static int test_can_verify_check(const char *fmt, ...)
3819 * The verifier is dependent on vsnprintf() modifies the va_list
3820 * passed to it, where it is sent as a reference. Some architectures
3821 * (like x86_32) passes it by value, which means that vsnprintf()
3822 * does not modify the va_list passed to it, and the verifier
3823 * would then need to be able to understand all the values that
3824 * vsnprintf can use. If it is passed by value, then the verifier
3828 vsnprintf(buf, 16, "%d", ap);
3829 ret = va_arg(ap, int);
3835 static void test_can_verify(void)
3837 if (!test_can_verify_check("%d %d", 0, 1)) {
3838 pr_info("trace event string verifier disabled\n");
3839 static_branch_inc(&trace_no_verify);
3844 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3845 * @iter: The iterator that holds the seq buffer and the event being printed
3846 * @fmt: The format used to print the event
3847 * @ap: The va_list holding the data to print from @fmt.
3849 * This writes the data into the @iter->seq buffer using the data from
3850 * @fmt and @ap. If the format has a %s, then the source of the string
3851 * is examined to make sure it is safe to print, otherwise it will
3852 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3855 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3858 const char *p = fmt;
3862 if (WARN_ON_ONCE(!fmt))
3865 if (static_branch_unlikely(&trace_no_verify))
3868 /* Don't bother checking when doing a ftrace_dump() */
3869 if (iter->fmt == static_fmt_buf)
3878 /* We only care about %s and variants */
3879 for (i = 0; p[i]; i++) {
3880 if (i + 1 >= iter->fmt_size) {
3882 * If we can't expand the copy buffer,
3885 if (!trace_iter_expand_format(iter))
3889 if (p[i] == '\\' && p[i+1]) {
3894 /* Need to test cases like %08.*s */
3895 for (j = 1; p[i+j]; j++) {
3896 if (isdigit(p[i+j]) ||
3899 if (p[i+j] == '*') {
3911 /* If no %s found then just print normally */
3915 /* Copy up to the %s, and print that */
3916 strncpy(iter->fmt, p, i);
3917 iter->fmt[i] = '\0';
3918 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3921 * If iter->seq is full, the above call no longer guarantees
3922 * that ap is in sync with fmt processing, and further calls
3923 * to va_arg() can return wrong positional arguments.
3925 * Ensure that ap is no longer used in this case.
3927 if (iter->seq.full) {
3933 len = va_arg(ap, int);
3935 /* The ap now points to the string data of the %s */
3936 str = va_arg(ap, const char *);
3939 * If you hit this warning, it is likely that the
3940 * trace event in question used %s on a string that
3941 * was saved at the time of the event, but may not be
3942 * around when the trace is read. Use __string(),
3943 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3944 * instead. See samples/trace_events/trace-events-sample.h
3947 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3948 "fmt: '%s' current_buffer: '%s'",
3949 fmt, show_buffer(&iter->seq))) {
3952 /* Try to safely read the string */
3954 if (len + 1 > iter->fmt_size)
3955 len = iter->fmt_size - 1;
3958 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3962 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3966 trace_seq_printf(&iter->seq, "(0x%px)", str);
3968 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3970 str = "[UNSAFE-MEMORY]";
3971 strcpy(iter->fmt, "%s");
3973 strncpy(iter->fmt, p + i, j + 1);
3974 iter->fmt[j+1] = '\0';
3977 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3979 trace_seq_printf(&iter->seq, iter->fmt, str);
3985 trace_seq_vprintf(&iter->seq, p, ap);
3988 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3990 const char *p, *new_fmt;
3993 if (WARN_ON_ONCE(!fmt))
3996 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
4000 new_fmt = q = iter->fmt;
4002 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
4003 if (!trace_iter_expand_format(iter))
4006 q += iter->fmt - new_fmt;
4007 new_fmt = iter->fmt;
4012 /* Replace %p with %px */
4016 } else if (p[0] == 'p' && !isalnum(p[1])) {
4027 #define STATIC_TEMP_BUF_SIZE 128
4028 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
4030 /* Find the next real entry, without updating the iterator itself */
4031 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
4032 int *ent_cpu, u64 *ent_ts)
4034 /* __find_next_entry will reset ent_size */
4035 int ent_size = iter->ent_size;
4036 struct trace_entry *entry;
4039 * If called from ftrace_dump(), then the iter->temp buffer
4040 * will be the static_temp_buf and not created from kmalloc.
4041 * If the entry size is greater than the buffer, we can
4042 * not save it. Just return NULL in that case. This is only
4043 * used to add markers when two consecutive events' time
4044 * stamps have a large delta. See trace_print_lat_context()
4046 if (iter->temp == static_temp_buf &&
4047 STATIC_TEMP_BUF_SIZE < ent_size)
4051 * The __find_next_entry() may call peek_next_entry(), which may
4052 * call ring_buffer_peek() that may make the contents of iter->ent
4053 * undefined. Need to copy iter->ent now.
4055 if (iter->ent && iter->ent != iter->temp) {
4056 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4057 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4059 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4064 iter->temp_size = iter->ent_size;
4066 memcpy(iter->temp, iter->ent, iter->ent_size);
4067 iter->ent = iter->temp;
4069 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4070 /* Put back the original ent_size */
4071 iter->ent_size = ent_size;
4076 /* Find the next real entry, and increment the iterator to the next entry */
4077 void *trace_find_next_entry_inc(struct trace_iterator *iter)
4079 iter->ent = __find_next_entry(iter, &iter->cpu,
4080 &iter->lost_events, &iter->ts);
4083 trace_iterator_increment(iter);
4085 return iter->ent ? iter : NULL;
4088 static void trace_consume(struct trace_iterator *iter)
4090 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4091 &iter->lost_events);
4094 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4096 struct trace_iterator *iter = m->private;
4100 WARN_ON_ONCE(iter->leftover);
4104 /* can't go backwards */
4109 ent = trace_find_next_entry_inc(iter);
4113 while (ent && iter->idx < i)
4114 ent = trace_find_next_entry_inc(iter);
4121 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4123 struct ring_buffer_iter *buf_iter;
4124 unsigned long entries = 0;
4127 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4129 buf_iter = trace_buffer_iter(iter, cpu);
4133 ring_buffer_iter_reset(buf_iter);
4136 * We could have the case with the max latency tracers
4137 * that a reset never took place on a cpu. This is evident
4138 * by the timestamp being before the start of the buffer.
4140 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4141 if (ts >= iter->array_buffer->time_start)
4144 ring_buffer_iter_advance(buf_iter);
4147 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4151 * The current tracer is copied to avoid a global locking
4154 static void *s_start(struct seq_file *m, loff_t *pos)
4156 struct trace_iterator *iter = m->private;
4157 struct trace_array *tr = iter->tr;
4158 int cpu_file = iter->cpu_file;
4163 mutex_lock(&trace_types_lock);
4164 if (unlikely(tr->current_trace != iter->trace)) {
4165 /* Close iter->trace before switching to the new current tracer */
4166 if (iter->trace->close)
4167 iter->trace->close(iter);
4168 iter->trace = tr->current_trace;
4169 /* Reopen the new current tracer */
4170 if (iter->trace->open)
4171 iter->trace->open(iter);
4173 mutex_unlock(&trace_types_lock);
4175 #ifdef CONFIG_TRACER_MAX_TRACE
4176 if (iter->snapshot && iter->trace->use_max_tr)
4177 return ERR_PTR(-EBUSY);
4180 if (*pos != iter->pos) {
4185 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4186 for_each_tracing_cpu(cpu)
4187 tracing_iter_reset(iter, cpu);
4189 tracing_iter_reset(iter, cpu_file);
4192 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4197 * If we overflowed the seq_file before, then we want
4198 * to just reuse the trace_seq buffer again.
4204 p = s_next(m, p, &l);
4208 trace_event_read_lock();
4209 trace_access_lock(cpu_file);
4213 static void s_stop(struct seq_file *m, void *p)
4215 struct trace_iterator *iter = m->private;
4217 #ifdef CONFIG_TRACER_MAX_TRACE
4218 if (iter->snapshot && iter->trace->use_max_tr)
4222 trace_access_unlock(iter->cpu_file);
4223 trace_event_read_unlock();
4227 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4228 unsigned long *entries, int cpu)
4230 unsigned long count;
4232 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4234 * If this buffer has skipped entries, then we hold all
4235 * entries for the trace and we need to ignore the
4236 * ones before the time stamp.
4238 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4239 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4240 /* total is the same as the entries */
4244 ring_buffer_overrun_cpu(buf->buffer, cpu);
4249 get_total_entries(struct array_buffer *buf,
4250 unsigned long *total, unsigned long *entries)
4258 for_each_tracing_cpu(cpu) {
4259 get_total_entries_cpu(buf, &t, &e, cpu);
4265 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4267 unsigned long total, entries;
4272 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4277 unsigned long trace_total_entries(struct trace_array *tr)
4279 unsigned long total, entries;
4284 get_total_entries(&tr->array_buffer, &total, &entries);
4289 static void print_lat_help_header(struct seq_file *m)
4291 seq_puts(m, "# _------=> CPU# \n"
4292 "# / _-----=> irqs-off/BH-disabled\n"
4293 "# | / _----=> need-resched \n"
4294 "# || / _---=> hardirq/softirq \n"
4295 "# ||| / _--=> preempt-depth \n"
4296 "# |||| / _-=> migrate-disable \n"
4297 "# ||||| / delay \n"
4298 "# cmd pid |||||| time | caller \n"
4299 "# \\ / |||||| \\ | / \n");
4302 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4304 unsigned long total;
4305 unsigned long entries;
4307 get_total_entries(buf, &total, &entries);
4308 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4309 entries, total, num_online_cpus());
4313 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4316 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4318 print_event_info(buf, m);
4320 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4321 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4324 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4327 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4328 static const char space[] = " ";
4329 int prec = tgid ? 12 : 2;
4331 print_event_info(buf, m);
4333 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4334 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4335 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4336 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4337 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4338 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4339 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4340 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4344 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4346 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4347 struct array_buffer *buf = iter->array_buffer;
4348 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4349 struct tracer *type = iter->trace;
4350 unsigned long entries;
4351 unsigned long total;
4352 const char *name = type->name;
4354 get_total_entries(buf, &total, &entries);
4356 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4358 seq_puts(m, "# -----------------------------------"
4359 "---------------------------------\n");
4360 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4361 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4362 nsecs_to_usecs(data->saved_latency),
4366 preempt_model_none() ? "server" :
4367 preempt_model_voluntary() ? "desktop" :
4368 preempt_model_full() ? "preempt" :
4369 preempt_model_rt() ? "preempt_rt" :
4371 /* These are reserved for later use */
4374 seq_printf(m, " #P:%d)\n", num_online_cpus());
4378 seq_puts(m, "# -----------------\n");
4379 seq_printf(m, "# | task: %.16s-%d "
4380 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4381 data->comm, data->pid,
4382 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4383 data->policy, data->rt_priority);
4384 seq_puts(m, "# -----------------\n");
4386 if (data->critical_start) {
4387 seq_puts(m, "# => started at: ");
4388 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4389 trace_print_seq(m, &iter->seq);
4390 seq_puts(m, "\n# => ended at: ");
4391 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4392 trace_print_seq(m, &iter->seq);
4393 seq_puts(m, "\n#\n");
4399 static void test_cpu_buff_start(struct trace_iterator *iter)
4401 struct trace_seq *s = &iter->seq;
4402 struct trace_array *tr = iter->tr;
4404 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4407 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4410 if (cpumask_available(iter->started) &&
4411 cpumask_test_cpu(iter->cpu, iter->started))
4414 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4417 if (cpumask_available(iter->started))
4418 cpumask_set_cpu(iter->cpu, iter->started);
4420 /* Don't print started cpu buffer for the first entry of the trace */
4422 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4426 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4428 struct trace_array *tr = iter->tr;
4429 struct trace_seq *s = &iter->seq;
4430 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4431 struct trace_entry *entry;
4432 struct trace_event *event;
4436 test_cpu_buff_start(iter);
4438 event = ftrace_find_event(entry->type);
4440 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4441 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4442 trace_print_lat_context(iter);
4444 trace_print_context(iter);
4447 if (trace_seq_has_overflowed(s))
4448 return TRACE_TYPE_PARTIAL_LINE;
4451 if (tr->trace_flags & TRACE_ITER_FIELDS)
4452 return print_event_fields(iter, event);
4453 return event->funcs->trace(iter, sym_flags, event);
4456 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4458 return trace_handle_return(s);
4461 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4463 struct trace_array *tr = iter->tr;
4464 struct trace_seq *s = &iter->seq;
4465 struct trace_entry *entry;
4466 struct trace_event *event;
4470 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4471 trace_seq_printf(s, "%d %d %llu ",
4472 entry->pid, iter->cpu, iter->ts);
4474 if (trace_seq_has_overflowed(s))
4475 return TRACE_TYPE_PARTIAL_LINE;
4477 event = ftrace_find_event(entry->type);
4479 return event->funcs->raw(iter, 0, event);
4481 trace_seq_printf(s, "%d ?\n", entry->type);
4483 return trace_handle_return(s);
4486 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4488 struct trace_array *tr = iter->tr;
4489 struct trace_seq *s = &iter->seq;
4490 unsigned char newline = '\n';
4491 struct trace_entry *entry;
4492 struct trace_event *event;
4496 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4497 SEQ_PUT_HEX_FIELD(s, entry->pid);
4498 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4499 SEQ_PUT_HEX_FIELD(s, iter->ts);
4500 if (trace_seq_has_overflowed(s))
4501 return TRACE_TYPE_PARTIAL_LINE;
4504 event = ftrace_find_event(entry->type);
4506 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4507 if (ret != TRACE_TYPE_HANDLED)
4511 SEQ_PUT_FIELD(s, newline);
4513 return trace_handle_return(s);
4516 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4518 struct trace_array *tr = iter->tr;
4519 struct trace_seq *s = &iter->seq;
4520 struct trace_entry *entry;
4521 struct trace_event *event;
4525 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4526 SEQ_PUT_FIELD(s, entry->pid);
4527 SEQ_PUT_FIELD(s, iter->cpu);
4528 SEQ_PUT_FIELD(s, iter->ts);
4529 if (trace_seq_has_overflowed(s))
4530 return TRACE_TYPE_PARTIAL_LINE;
4533 event = ftrace_find_event(entry->type);
4534 return event ? event->funcs->binary(iter, 0, event) :
4538 int trace_empty(struct trace_iterator *iter)
4540 struct ring_buffer_iter *buf_iter;
4543 /* If we are looking at one CPU buffer, only check that one */
4544 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4545 cpu = iter->cpu_file;
4546 buf_iter = trace_buffer_iter(iter, cpu);
4548 if (!ring_buffer_iter_empty(buf_iter))
4551 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4557 for_each_tracing_cpu(cpu) {
4558 buf_iter = trace_buffer_iter(iter, cpu);
4560 if (!ring_buffer_iter_empty(buf_iter))
4563 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4571 /* Called with trace_event_read_lock() held. */
4572 enum print_line_t print_trace_line(struct trace_iterator *iter)
4574 struct trace_array *tr = iter->tr;
4575 unsigned long trace_flags = tr->trace_flags;
4576 enum print_line_t ret;
4578 if (iter->lost_events) {
4579 if (iter->lost_events == (unsigned long)-1)
4580 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4583 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4584 iter->cpu, iter->lost_events);
4585 if (trace_seq_has_overflowed(&iter->seq))
4586 return TRACE_TYPE_PARTIAL_LINE;
4589 if (iter->trace && iter->trace->print_line) {
4590 ret = iter->trace->print_line(iter);
4591 if (ret != TRACE_TYPE_UNHANDLED)
4595 if (iter->ent->type == TRACE_BPUTS &&
4596 trace_flags & TRACE_ITER_PRINTK &&
4597 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4598 return trace_print_bputs_msg_only(iter);
4600 if (iter->ent->type == TRACE_BPRINT &&
4601 trace_flags & TRACE_ITER_PRINTK &&
4602 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4603 return trace_print_bprintk_msg_only(iter);
4605 if (iter->ent->type == TRACE_PRINT &&
4606 trace_flags & TRACE_ITER_PRINTK &&
4607 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4608 return trace_print_printk_msg_only(iter);
4610 if (trace_flags & TRACE_ITER_BIN)
4611 return print_bin_fmt(iter);
4613 if (trace_flags & TRACE_ITER_HEX)
4614 return print_hex_fmt(iter);
4616 if (trace_flags & TRACE_ITER_RAW)
4617 return print_raw_fmt(iter);
4619 return print_trace_fmt(iter);
4622 void trace_latency_header(struct seq_file *m)
4624 struct trace_iterator *iter = m->private;
4625 struct trace_array *tr = iter->tr;
4627 /* print nothing if the buffers are empty */
4628 if (trace_empty(iter))
4631 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4632 print_trace_header(m, iter);
4634 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4635 print_lat_help_header(m);
4638 void trace_default_header(struct seq_file *m)
4640 struct trace_iterator *iter = m->private;
4641 struct trace_array *tr = iter->tr;
4642 unsigned long trace_flags = tr->trace_flags;
4644 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4647 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4648 /* print nothing if the buffers are empty */
4649 if (trace_empty(iter))
4651 print_trace_header(m, iter);
4652 if (!(trace_flags & TRACE_ITER_VERBOSE))
4653 print_lat_help_header(m);
4655 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4656 if (trace_flags & TRACE_ITER_IRQ_INFO)
4657 print_func_help_header_irq(iter->array_buffer,
4660 print_func_help_header(iter->array_buffer, m,
4666 static void test_ftrace_alive(struct seq_file *m)
4668 if (!ftrace_is_dead())
4670 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4671 "# MAY BE MISSING FUNCTION EVENTS\n");
4674 #ifdef CONFIG_TRACER_MAX_TRACE
4675 static void show_snapshot_main_help(struct seq_file *m)
4677 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4678 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4679 "# Takes a snapshot of the main buffer.\n"
4680 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4681 "# (Doesn't have to be '2' works with any number that\n"
4682 "# is not a '0' or '1')\n");
4685 static void show_snapshot_percpu_help(struct seq_file *m)
4687 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4688 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4689 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4690 "# Takes a snapshot of the main buffer for this cpu.\n");
4692 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4693 "# Must use main snapshot file to allocate.\n");
4695 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4696 "# (Doesn't have to be '2' works with any number that\n"
4697 "# is not a '0' or '1')\n");
4700 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4702 if (iter->tr->allocated_snapshot)
4703 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4705 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4707 seq_puts(m, "# Snapshot commands:\n");
4708 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4709 show_snapshot_main_help(m);
4711 show_snapshot_percpu_help(m);
4714 /* Should never be called */
4715 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4718 static int s_show(struct seq_file *m, void *v)
4720 struct trace_iterator *iter = v;
4723 if (iter->ent == NULL) {
4725 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4727 test_ftrace_alive(m);
4729 if (iter->snapshot && trace_empty(iter))
4730 print_snapshot_help(m, iter);
4731 else if (iter->trace && iter->trace->print_header)
4732 iter->trace->print_header(m);
4734 trace_default_header(m);
4736 } else if (iter->leftover) {
4738 * If we filled the seq_file buffer earlier, we
4739 * want to just show it now.
4741 ret = trace_print_seq(m, &iter->seq);
4743 /* ret should this time be zero, but you never know */
4744 iter->leftover = ret;
4747 print_trace_line(iter);
4748 ret = trace_print_seq(m, &iter->seq);
4750 * If we overflow the seq_file buffer, then it will
4751 * ask us for this data again at start up.
4753 * ret is 0 if seq_file write succeeded.
4756 iter->leftover = ret;
4763 * Should be used after trace_array_get(), trace_types_lock
4764 * ensures that i_cdev was already initialized.
4766 static inline int tracing_get_cpu(struct inode *inode)
4768 if (inode->i_cdev) /* See trace_create_cpu_file() */
4769 return (long)inode->i_cdev - 1;
4770 return RING_BUFFER_ALL_CPUS;
4773 static const struct seq_operations tracer_seq_ops = {
4781 * Note, as iter itself can be allocated and freed in different
4782 * ways, this function is only used to free its content, and not
4783 * the iterator itself. The only requirement to all the allocations
4784 * is that it must zero all fields (kzalloc), as freeing works with
4785 * ethier allocated content or NULL.
4787 static void free_trace_iter_content(struct trace_iterator *iter)
4789 /* The fmt is either NULL, allocated or points to static_fmt_buf */
4790 if (iter->fmt != static_fmt_buf)
4794 kfree(iter->buffer_iter);
4795 mutex_destroy(&iter->mutex);
4796 free_cpumask_var(iter->started);
4799 static struct trace_iterator *
4800 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4802 struct trace_array *tr = inode->i_private;
4803 struct trace_iterator *iter;
4806 if (tracing_disabled)
4807 return ERR_PTR(-ENODEV);
4809 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4811 return ERR_PTR(-ENOMEM);
4813 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4815 if (!iter->buffer_iter)
4819 * trace_find_next_entry() may need to save off iter->ent.
4820 * It will place it into the iter->temp buffer. As most
4821 * events are less than 128, allocate a buffer of that size.
4822 * If one is greater, then trace_find_next_entry() will
4823 * allocate a new buffer to adjust for the bigger iter->ent.
4824 * It's not critical if it fails to get allocated here.
4826 iter->temp = kmalloc(128, GFP_KERNEL);
4828 iter->temp_size = 128;
4831 * trace_event_printf() may need to modify given format
4832 * string to replace %p with %px so that it shows real address
4833 * instead of hash value. However, that is only for the event
4834 * tracing, other tracer may not need. Defer the allocation
4835 * until it is needed.
4840 mutex_lock(&trace_types_lock);
4841 iter->trace = tr->current_trace;
4843 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4848 #ifdef CONFIG_TRACER_MAX_TRACE
4849 /* Currently only the top directory has a snapshot */
4850 if (tr->current_trace->print_max || snapshot)
4851 iter->array_buffer = &tr->max_buffer;
4854 iter->array_buffer = &tr->array_buffer;
4855 iter->snapshot = snapshot;
4857 iter->cpu_file = tracing_get_cpu(inode);
4858 mutex_init(&iter->mutex);
4860 /* Notify the tracer early; before we stop tracing. */
4861 if (iter->trace->open)
4862 iter->trace->open(iter);
4864 /* Annotate start of buffers if we had overruns */
4865 if (ring_buffer_overruns(iter->array_buffer->buffer))
4866 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4868 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4869 if (trace_clocks[tr->clock_id].in_ns)
4870 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4873 * If pause-on-trace is enabled, then stop the trace while
4874 * dumping, unless this is the "snapshot" file
4876 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4877 tracing_stop_tr(tr);
4879 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4880 for_each_tracing_cpu(cpu) {
4881 iter->buffer_iter[cpu] =
4882 ring_buffer_read_prepare(iter->array_buffer->buffer,
4885 ring_buffer_read_prepare_sync();
4886 for_each_tracing_cpu(cpu) {
4887 ring_buffer_read_start(iter->buffer_iter[cpu]);
4888 tracing_iter_reset(iter, cpu);
4891 cpu = iter->cpu_file;
4892 iter->buffer_iter[cpu] =
4893 ring_buffer_read_prepare(iter->array_buffer->buffer,
4895 ring_buffer_read_prepare_sync();
4896 ring_buffer_read_start(iter->buffer_iter[cpu]);
4897 tracing_iter_reset(iter, cpu);
4900 mutex_unlock(&trace_types_lock);
4905 mutex_unlock(&trace_types_lock);
4906 free_trace_iter_content(iter);
4908 seq_release_private(inode, file);
4909 return ERR_PTR(-ENOMEM);
4912 int tracing_open_generic(struct inode *inode, struct file *filp)
4916 ret = tracing_check_open_get_tr(NULL);
4920 filp->private_data = inode->i_private;
4924 bool tracing_is_disabled(void)
4926 return (tracing_disabled) ? true: false;
4930 * Open and update trace_array ref count.
4931 * Must have the current trace_array passed to it.
4933 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4935 struct trace_array *tr = inode->i_private;
4938 ret = tracing_check_open_get_tr(tr);
4942 filp->private_data = inode->i_private;
4948 * The private pointer of the inode is the trace_event_file.
4949 * Update the tr ref count associated to it.
4951 int tracing_open_file_tr(struct inode *inode, struct file *filp)
4953 struct trace_event_file *file = inode->i_private;
4956 ret = tracing_check_open_get_tr(file->tr);
4960 mutex_lock(&event_mutex);
4962 /* Fail if the file is marked for removal */
4963 if (file->flags & EVENT_FILE_FL_FREED) {
4964 trace_array_put(file->tr);
4967 event_file_get(file);
4970 mutex_unlock(&event_mutex);
4974 filp->private_data = inode->i_private;
4979 int tracing_release_file_tr(struct inode *inode, struct file *filp)
4981 struct trace_event_file *file = inode->i_private;
4983 trace_array_put(file->tr);
4984 event_file_put(file);
4989 static int tracing_mark_open(struct inode *inode, struct file *filp)
4991 stream_open(inode, filp);
4992 return tracing_open_generic_tr(inode, filp);
4995 static int tracing_release(struct inode *inode, struct file *file)
4997 struct trace_array *tr = inode->i_private;
4998 struct seq_file *m = file->private_data;
4999 struct trace_iterator *iter;
5002 if (!(file->f_mode & FMODE_READ)) {
5003 trace_array_put(tr);
5007 /* Writes do not use seq_file */
5009 mutex_lock(&trace_types_lock);
5011 for_each_tracing_cpu(cpu) {
5012 if (iter->buffer_iter[cpu])
5013 ring_buffer_read_finish(iter->buffer_iter[cpu]);
5016 if (iter->trace && iter->trace->close)
5017 iter->trace->close(iter);
5019 if (!iter->snapshot && tr->stop_count)
5020 /* reenable tracing if it was previously enabled */
5021 tracing_start_tr(tr);
5023 __trace_array_put(tr);
5025 mutex_unlock(&trace_types_lock);
5027 free_trace_iter_content(iter);
5028 seq_release_private(inode, file);
5033 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
5035 struct trace_array *tr = inode->i_private;
5037 trace_array_put(tr);
5041 static int tracing_single_release_tr(struct inode *inode, struct file *file)
5043 struct trace_array *tr = inode->i_private;
5045 trace_array_put(tr);
5047 return single_release(inode, file);
5050 static int tracing_open(struct inode *inode, struct file *file)
5052 struct trace_array *tr = inode->i_private;
5053 struct trace_iterator *iter;
5056 ret = tracing_check_open_get_tr(tr);
5060 /* If this file was open for write, then erase contents */
5061 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
5062 int cpu = tracing_get_cpu(inode);
5063 struct array_buffer *trace_buf = &tr->array_buffer;
5065 #ifdef CONFIG_TRACER_MAX_TRACE
5066 if (tr->current_trace->print_max)
5067 trace_buf = &tr->max_buffer;
5070 if (cpu == RING_BUFFER_ALL_CPUS)
5071 tracing_reset_online_cpus(trace_buf);
5073 tracing_reset_cpu(trace_buf, cpu);
5076 if (file->f_mode & FMODE_READ) {
5077 iter = __tracing_open(inode, file, false);
5079 ret = PTR_ERR(iter);
5080 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
5081 iter->iter_flags |= TRACE_FILE_LAT_FMT;
5085 trace_array_put(tr);
5091 * Some tracers are not suitable for instance buffers.
5092 * A tracer is always available for the global array (toplevel)
5093 * or if it explicitly states that it is.
5096 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
5098 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5101 /* Find the next tracer that this trace array may use */
5102 static struct tracer *
5103 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5105 while (t && !trace_ok_for_array(t, tr))
5112 t_next(struct seq_file *m, void *v, loff_t *pos)
5114 struct trace_array *tr = m->private;
5115 struct tracer *t = v;
5120 t = get_tracer_for_array(tr, t->next);
5125 static void *t_start(struct seq_file *m, loff_t *pos)
5127 struct trace_array *tr = m->private;
5131 mutex_lock(&trace_types_lock);
5133 t = get_tracer_for_array(tr, trace_types);
5134 for (; t && l < *pos; t = t_next(m, t, &l))
5140 static void t_stop(struct seq_file *m, void *p)
5142 mutex_unlock(&trace_types_lock);
5145 static int t_show(struct seq_file *m, void *v)
5147 struct tracer *t = v;
5152 seq_puts(m, t->name);
5161 static const struct seq_operations show_traces_seq_ops = {
5168 static int show_traces_open(struct inode *inode, struct file *file)
5170 struct trace_array *tr = inode->i_private;
5174 ret = tracing_check_open_get_tr(tr);
5178 ret = seq_open(file, &show_traces_seq_ops);
5180 trace_array_put(tr);
5184 m = file->private_data;
5190 static int show_traces_release(struct inode *inode, struct file *file)
5192 struct trace_array *tr = inode->i_private;
5194 trace_array_put(tr);
5195 return seq_release(inode, file);
5199 tracing_write_stub(struct file *filp, const char __user *ubuf,
5200 size_t count, loff_t *ppos)
5205 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5209 if (file->f_mode & FMODE_READ)
5210 ret = seq_lseek(file, offset, whence);
5212 file->f_pos = ret = 0;
5217 static const struct file_operations tracing_fops = {
5218 .open = tracing_open,
5220 .read_iter = seq_read_iter,
5221 .splice_read = copy_splice_read,
5222 .write = tracing_write_stub,
5223 .llseek = tracing_lseek,
5224 .release = tracing_release,
5227 static const struct file_operations show_traces_fops = {
5228 .open = show_traces_open,
5230 .llseek = seq_lseek,
5231 .release = show_traces_release,
5235 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5236 size_t count, loff_t *ppos)
5238 struct trace_array *tr = file_inode(filp)->i_private;
5242 len = snprintf(NULL, 0, "%*pb\n",
5243 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5244 mask_str = kmalloc(len, GFP_KERNEL);
5248 len = snprintf(mask_str, len, "%*pb\n",
5249 cpumask_pr_args(tr->tracing_cpumask));
5254 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5262 int tracing_set_cpumask(struct trace_array *tr,
5263 cpumask_var_t tracing_cpumask_new)
5270 local_irq_disable();
5271 arch_spin_lock(&tr->max_lock);
5272 for_each_tracing_cpu(cpu) {
5274 * Increase/decrease the disabled counter if we are
5275 * about to flip a bit in the cpumask:
5277 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5278 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5279 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5280 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5281 #ifdef CONFIG_TRACER_MAX_TRACE
5282 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5285 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5286 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5287 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5288 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5289 #ifdef CONFIG_TRACER_MAX_TRACE
5290 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5294 arch_spin_unlock(&tr->max_lock);
5297 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5303 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5304 size_t count, loff_t *ppos)
5306 struct trace_array *tr = file_inode(filp)->i_private;
5307 cpumask_var_t tracing_cpumask_new;
5310 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5313 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5317 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5321 free_cpumask_var(tracing_cpumask_new);
5326 free_cpumask_var(tracing_cpumask_new);
5331 static const struct file_operations tracing_cpumask_fops = {
5332 .open = tracing_open_generic_tr,
5333 .read = tracing_cpumask_read,
5334 .write = tracing_cpumask_write,
5335 .release = tracing_release_generic_tr,
5336 .llseek = generic_file_llseek,
5339 static int tracing_trace_options_show(struct seq_file *m, void *v)
5341 struct tracer_opt *trace_opts;
5342 struct trace_array *tr = m->private;
5346 mutex_lock(&trace_types_lock);
5347 tracer_flags = tr->current_trace->flags->val;
5348 trace_opts = tr->current_trace->flags->opts;
5350 for (i = 0; trace_options[i]; i++) {
5351 if (tr->trace_flags & (1 << i))
5352 seq_printf(m, "%s\n", trace_options[i]);
5354 seq_printf(m, "no%s\n", trace_options[i]);
5357 for (i = 0; trace_opts[i].name; i++) {
5358 if (tracer_flags & trace_opts[i].bit)
5359 seq_printf(m, "%s\n", trace_opts[i].name);
5361 seq_printf(m, "no%s\n", trace_opts[i].name);
5363 mutex_unlock(&trace_types_lock);
5368 static int __set_tracer_option(struct trace_array *tr,
5369 struct tracer_flags *tracer_flags,
5370 struct tracer_opt *opts, int neg)
5372 struct tracer *trace = tracer_flags->trace;
5375 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5380 tracer_flags->val &= ~opts->bit;
5382 tracer_flags->val |= opts->bit;
5386 /* Try to assign a tracer specific option */
5387 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5389 struct tracer *trace = tr->current_trace;
5390 struct tracer_flags *tracer_flags = trace->flags;
5391 struct tracer_opt *opts = NULL;
5394 for (i = 0; tracer_flags->opts[i].name; i++) {
5395 opts = &tracer_flags->opts[i];
5397 if (strcmp(cmp, opts->name) == 0)
5398 return __set_tracer_option(tr, trace->flags, opts, neg);
5404 /* Some tracers require overwrite to stay enabled */
5405 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5407 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5413 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5417 if ((mask == TRACE_ITER_RECORD_TGID) ||
5418 (mask == TRACE_ITER_RECORD_CMD))
5419 lockdep_assert_held(&event_mutex);
5421 /* do nothing if flag is already set */
5422 if (!!(tr->trace_flags & mask) == !!enabled)
5425 /* Give the tracer a chance to approve the change */
5426 if (tr->current_trace->flag_changed)
5427 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5431 tr->trace_flags |= mask;
5433 tr->trace_flags &= ~mask;
5435 if (mask == TRACE_ITER_RECORD_CMD)
5436 trace_event_enable_cmd_record(enabled);
5438 if (mask == TRACE_ITER_RECORD_TGID) {
5440 tgid_map_max = pid_max;
5441 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5445 * Pairs with smp_load_acquire() in
5446 * trace_find_tgid_ptr() to ensure that if it observes
5447 * the tgid_map we just allocated then it also observes
5448 * the corresponding tgid_map_max value.
5450 smp_store_release(&tgid_map, map);
5453 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5457 trace_event_enable_tgid_record(enabled);
5460 if (mask == TRACE_ITER_EVENT_FORK)
5461 trace_event_follow_fork(tr, enabled);
5463 if (mask == TRACE_ITER_FUNC_FORK)
5464 ftrace_pid_follow_fork(tr, enabled);
5466 if (mask == TRACE_ITER_OVERWRITE) {
5467 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5468 #ifdef CONFIG_TRACER_MAX_TRACE
5469 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5473 if (mask == TRACE_ITER_PRINTK) {
5474 trace_printk_start_stop_comm(enabled);
5475 trace_printk_control(enabled);
5481 int trace_set_options(struct trace_array *tr, char *option)
5486 size_t orig_len = strlen(option);
5489 cmp = strstrip(option);
5491 len = str_has_prefix(cmp, "no");
5497 mutex_lock(&event_mutex);
5498 mutex_lock(&trace_types_lock);
5500 ret = match_string(trace_options, -1, cmp);
5501 /* If no option could be set, test the specific tracer options */
5503 ret = set_tracer_option(tr, cmp, neg);
5505 ret = set_tracer_flag(tr, 1 << ret, !neg);
5507 mutex_unlock(&trace_types_lock);
5508 mutex_unlock(&event_mutex);
5511 * If the first trailing whitespace is replaced with '\0' by strstrip,
5512 * turn it back into a space.
5514 if (orig_len > strlen(option))
5515 option[strlen(option)] = ' ';
5520 static void __init apply_trace_boot_options(void)
5522 char *buf = trace_boot_options_buf;
5526 option = strsep(&buf, ",");
5532 trace_set_options(&global_trace, option);
5534 /* Put back the comma to allow this to be called again */
5541 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5542 size_t cnt, loff_t *ppos)
5544 struct seq_file *m = filp->private_data;
5545 struct trace_array *tr = m->private;
5549 if (cnt >= sizeof(buf))
5552 if (copy_from_user(buf, ubuf, cnt))
5557 ret = trace_set_options(tr, buf);
5566 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5568 struct trace_array *tr = inode->i_private;
5571 ret = tracing_check_open_get_tr(tr);
5575 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5577 trace_array_put(tr);
5582 static const struct file_operations tracing_iter_fops = {
5583 .open = tracing_trace_options_open,
5585 .llseek = seq_lseek,
5586 .release = tracing_single_release_tr,
5587 .write = tracing_trace_options_write,
5590 static const char readme_msg[] =
5591 "tracing mini-HOWTO:\n\n"
5592 "# echo 0 > tracing_on : quick way to disable tracing\n"
5593 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5594 " Important files:\n"
5595 " trace\t\t\t- The static contents of the buffer\n"
5596 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5597 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5598 " current_tracer\t- function and latency tracers\n"
5599 " available_tracers\t- list of configured tracers for current_tracer\n"
5600 " error_log\t- error log for failed commands (that support it)\n"
5601 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5602 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5603 " trace_clock\t\t- change the clock used to order events\n"
5604 " local: Per cpu clock but may not be synced across CPUs\n"
5605 " global: Synced across CPUs but slows tracing down.\n"
5606 " counter: Not a clock, but just an increment\n"
5607 " uptime: Jiffy counter from time of boot\n"
5608 " perf: Same clock that perf events use\n"
5609 #ifdef CONFIG_X86_64
5610 " x86-tsc: TSC cycle counter\n"
5612 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5613 " delta: Delta difference against a buffer-wide timestamp\n"
5614 " absolute: Absolute (standalone) timestamp\n"
5615 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5616 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5617 " tracing_cpumask\t- Limit which CPUs to trace\n"
5618 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5619 "\t\t\t Remove sub-buffer with rmdir\n"
5620 " trace_options\t\t- Set format or modify how tracing happens\n"
5621 "\t\t\t Disable an option by prefixing 'no' to the\n"
5622 "\t\t\t option name\n"
5623 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5624 #ifdef CONFIG_DYNAMIC_FTRACE
5625 "\n available_filter_functions - list of functions that can be filtered on\n"
5626 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5627 "\t\t\t functions\n"
5628 "\t accepts: func_full_name or glob-matching-pattern\n"
5629 "\t modules: Can select a group via module\n"
5630 "\t Format: :mod:<module-name>\n"
5631 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5632 "\t triggers: a command to perform when function is hit\n"
5633 "\t Format: <function>:<trigger>[:count]\n"
5634 "\t trigger: traceon, traceoff\n"
5635 "\t\t enable_event:<system>:<event>\n"
5636 "\t\t disable_event:<system>:<event>\n"
5637 #ifdef CONFIG_STACKTRACE
5640 #ifdef CONFIG_TRACER_SNAPSHOT
5645 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5646 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5647 "\t The first one will disable tracing every time do_fault is hit\n"
5648 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5649 "\t The first time do trap is hit and it disables tracing, the\n"
5650 "\t counter will decrement to 2. If tracing is already disabled,\n"
5651 "\t the counter will not decrement. It only decrements when the\n"
5652 "\t trigger did work\n"
5653 "\t To remove trigger without count:\n"
5654 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5655 "\t To remove trigger with a count:\n"
5656 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5657 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5658 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5659 "\t modules: Can select a group via module command :mod:\n"
5660 "\t Does not accept triggers\n"
5661 #endif /* CONFIG_DYNAMIC_FTRACE */
5662 #ifdef CONFIG_FUNCTION_TRACER
5663 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5665 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5668 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5669 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5670 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5671 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5673 #ifdef CONFIG_TRACER_SNAPSHOT
5674 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5675 "\t\t\t snapshot buffer. Read the contents for more\n"
5676 "\t\t\t information\n"
5678 #ifdef CONFIG_STACK_TRACER
5679 " stack_trace\t\t- Shows the max stack trace when active\n"
5680 " stack_max_size\t- Shows current max stack size that was traced\n"
5681 "\t\t\t Write into this file to reset the max size (trigger a\n"
5682 "\t\t\t new trace)\n"
5683 #ifdef CONFIG_DYNAMIC_FTRACE
5684 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5687 #endif /* CONFIG_STACK_TRACER */
5688 #ifdef CONFIG_DYNAMIC_EVENTS
5689 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5690 "\t\t\t Write into this file to define/undefine new trace events.\n"
5692 #ifdef CONFIG_KPROBE_EVENTS
5693 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5694 "\t\t\t Write into this file to define/undefine new trace events.\n"
5696 #ifdef CONFIG_UPROBE_EVENTS
5697 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5698 "\t\t\t Write into this file to define/undefine new trace events.\n"
5700 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5701 defined(CONFIG_FPROBE_EVENTS)
5702 "\t accepts: event-definitions (one definition per line)\n"
5703 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5704 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5705 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5707 #ifdef CONFIG_FPROBE_EVENTS
5708 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5709 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5711 #ifdef CONFIG_HIST_TRIGGERS
5712 "\t s:[synthetic/]<event> <field> [<field>]\n"
5714 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5715 "\t -:[<group>/][<event>]\n"
5716 #ifdef CONFIG_KPROBE_EVENTS
5717 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5718 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5720 #ifdef CONFIG_UPROBE_EVENTS
5721 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5723 "\t args: <name>=fetcharg[:type]\n"
5724 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5725 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5726 #ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5727 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5728 "\t <argname>[->field[->field|.field...]],\n"
5730 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5733 "\t $stack<index>, $stack, $retval, $comm,\n"
5735 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5736 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5737 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5738 "\t symstr, <type>\\[<array-size>\\]\n"
5739 #ifdef CONFIG_HIST_TRIGGERS
5740 "\t field: <stype> <name>;\n"
5741 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5742 "\t [unsigned] char/int/long\n"
5744 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5745 "\t of the <attached-group>/<attached-event>.\n"
5747 " events/\t\t- Directory containing all trace event subsystems:\n"
5748 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5749 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5750 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5752 " filter\t\t- If set, only events passing filter are traced\n"
5753 " events/<system>/<event>/\t- Directory containing control files for\n"
5755 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5756 " filter\t\t- If set, only events passing filter are traced\n"
5757 " trigger\t\t- If set, a command to perform when event is hit\n"
5758 "\t Format: <trigger>[:count][if <filter>]\n"
5759 "\t trigger: traceon, traceoff\n"
5760 "\t enable_event:<system>:<event>\n"
5761 "\t disable_event:<system>:<event>\n"
5762 #ifdef CONFIG_HIST_TRIGGERS
5763 "\t enable_hist:<system>:<event>\n"
5764 "\t disable_hist:<system>:<event>\n"
5766 #ifdef CONFIG_STACKTRACE
5769 #ifdef CONFIG_TRACER_SNAPSHOT
5772 #ifdef CONFIG_HIST_TRIGGERS
5773 "\t\t hist (see below)\n"
5775 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5776 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5777 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5778 "\t events/block/block_unplug/trigger\n"
5779 "\t The first disables tracing every time block_unplug is hit.\n"
5780 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5781 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5782 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5783 "\t Like function triggers, the counter is only decremented if it\n"
5784 "\t enabled or disabled tracing.\n"
5785 "\t To remove a trigger without a count:\n"
5786 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5787 "\t To remove a trigger with a count:\n"
5788 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5789 "\t Filters can be ignored when removing a trigger.\n"
5790 #ifdef CONFIG_HIST_TRIGGERS
5791 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5792 "\t Format: hist:keys=<field1[,field2,...]>\n"
5793 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5794 "\t [:values=<field1[,field2,...]>]\n"
5795 "\t [:sort=<field1[,field2,...]>]\n"
5796 "\t [:size=#entries]\n"
5797 "\t [:pause][:continue][:clear]\n"
5798 "\t [:name=histname1]\n"
5799 "\t [:nohitcount]\n"
5800 "\t [:<handler>.<action>]\n"
5801 "\t [if <filter>]\n\n"
5802 "\t Note, special fields can be used as well:\n"
5803 "\t common_timestamp - to record current timestamp\n"
5804 "\t common_cpu - to record the CPU the event happened on\n"
5806 "\t A hist trigger variable can be:\n"
5807 "\t - a reference to a field e.g. x=current_timestamp,\n"
5808 "\t - a reference to another variable e.g. y=$x,\n"
5809 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5810 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5812 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5813 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5814 "\t variable reference, field or numeric literal.\n"
5816 "\t When a matching event is hit, an entry is added to a hash\n"
5817 "\t table using the key(s) and value(s) named, and the value of a\n"
5818 "\t sum called 'hitcount' is incremented. Keys and values\n"
5819 "\t correspond to fields in the event's format description. Keys\n"
5820 "\t can be any field, or the special string 'common_stacktrace'.\n"
5821 "\t Compound keys consisting of up to two fields can be specified\n"
5822 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5823 "\t fields. Sort keys consisting of up to two fields can be\n"
5824 "\t specified using the 'sort' keyword. The sort direction can\n"
5825 "\t be modified by appending '.descending' or '.ascending' to a\n"
5826 "\t sort field. The 'size' parameter can be used to specify more\n"
5827 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5828 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5829 "\t its histogram data will be shared with other triggers of the\n"
5830 "\t same name, and trigger hits will update this common data.\n\n"
5831 "\t Reading the 'hist' file for the event will dump the hash\n"
5832 "\t table in its entirety to stdout. If there are multiple hist\n"
5833 "\t triggers attached to an event, there will be a table for each\n"
5834 "\t trigger in the output. The table displayed for a named\n"
5835 "\t trigger will be the same as any other instance having the\n"
5836 "\t same name. The default format used to display a given field\n"
5837 "\t can be modified by appending any of the following modifiers\n"
5838 "\t to the field name, as applicable:\n\n"
5839 "\t .hex display a number as a hex value\n"
5840 "\t .sym display an address as a symbol\n"
5841 "\t .sym-offset display an address as a symbol and offset\n"
5842 "\t .execname display a common_pid as a program name\n"
5843 "\t .syscall display a syscall id as a syscall name\n"
5844 "\t .log2 display log2 value rather than raw number\n"
5845 "\t .buckets=size display values in groups of size rather than raw number\n"
5846 "\t .usecs display a common_timestamp in microseconds\n"
5847 "\t .percent display a number of percentage value\n"
5848 "\t .graph display a bar-graph of a value\n\n"
5849 "\t The 'pause' parameter can be used to pause an existing hist\n"
5850 "\t trigger or to start a hist trigger but not log any events\n"
5851 "\t until told to do so. 'continue' can be used to start or\n"
5852 "\t restart a paused hist trigger.\n\n"
5853 "\t The 'clear' parameter will clear the contents of a running\n"
5854 "\t hist trigger and leave its current paused/active state\n"
5856 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5857 "\t raw hitcount in the histogram.\n\n"
5858 "\t The enable_hist and disable_hist triggers can be used to\n"
5859 "\t have one event conditionally start and stop another event's\n"
5860 "\t already-attached hist trigger. The syntax is analogous to\n"
5861 "\t the enable_event and disable_event triggers.\n\n"
5862 "\t Hist trigger handlers and actions are executed whenever a\n"
5863 "\t a histogram entry is added or updated. They take the form:\n\n"
5864 "\t <handler>.<action>\n\n"
5865 "\t The available handlers are:\n\n"
5866 "\t onmatch(matching.event) - invoke on addition or update\n"
5867 "\t onmax(var) - invoke if var exceeds current max\n"
5868 "\t onchange(var) - invoke action if var changes\n\n"
5869 "\t The available actions are:\n\n"
5870 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5871 "\t save(field,...) - save current event fields\n"
5872 #ifdef CONFIG_TRACER_SNAPSHOT
5873 "\t snapshot() - snapshot the trace buffer\n\n"
5875 #ifdef CONFIG_SYNTH_EVENTS
5876 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5877 "\t Write into this file to define/undefine new synthetic events.\n"
5878 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5884 tracing_readme_read(struct file *filp, char __user *ubuf,
5885 size_t cnt, loff_t *ppos)
5887 return simple_read_from_buffer(ubuf, cnt, ppos,
5888 readme_msg, strlen(readme_msg));
5891 static const struct file_operations tracing_readme_fops = {
5892 .open = tracing_open_generic,
5893 .read = tracing_readme_read,
5894 .llseek = generic_file_llseek,
5897 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5901 return trace_find_tgid_ptr(pid);
5904 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5908 return trace_find_tgid_ptr(pid);
5911 static void saved_tgids_stop(struct seq_file *m, void *v)
5915 static int saved_tgids_show(struct seq_file *m, void *v)
5917 int *entry = (int *)v;
5918 int pid = entry - tgid_map;
5924 seq_printf(m, "%d %d\n", pid, tgid);
5928 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5929 .start = saved_tgids_start,
5930 .stop = saved_tgids_stop,
5931 .next = saved_tgids_next,
5932 .show = saved_tgids_show,
5935 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5939 ret = tracing_check_open_get_tr(NULL);
5943 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5947 static const struct file_operations tracing_saved_tgids_fops = {
5948 .open = tracing_saved_tgids_open,
5950 .llseek = seq_lseek,
5951 .release = seq_release,
5954 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5956 unsigned int *ptr = v;
5958 if (*pos || m->count)
5963 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5965 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5974 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5980 arch_spin_lock(&trace_cmdline_lock);
5982 v = &savedcmd->map_cmdline_to_pid[0];
5984 v = saved_cmdlines_next(m, v, &l);
5992 static void saved_cmdlines_stop(struct seq_file *m, void *v)
5994 arch_spin_unlock(&trace_cmdline_lock);
5998 static int saved_cmdlines_show(struct seq_file *m, void *v)
6000 char buf[TASK_COMM_LEN];
6001 unsigned int *pid = v;
6003 __trace_find_cmdline(*pid, buf);
6004 seq_printf(m, "%d %s\n", *pid, buf);
6008 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
6009 .start = saved_cmdlines_start,
6010 .next = saved_cmdlines_next,
6011 .stop = saved_cmdlines_stop,
6012 .show = saved_cmdlines_show,
6015 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
6019 ret = tracing_check_open_get_tr(NULL);
6023 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
6026 static const struct file_operations tracing_saved_cmdlines_fops = {
6027 .open = tracing_saved_cmdlines_open,
6029 .llseek = seq_lseek,
6030 .release = seq_release,
6034 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
6035 size_t cnt, loff_t *ppos)
6041 arch_spin_lock(&trace_cmdline_lock);
6042 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
6043 arch_spin_unlock(&trace_cmdline_lock);
6046 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6049 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
6051 kfree(s->saved_cmdlines);
6052 kfree(s->map_cmdline_to_pid);
6056 static int tracing_resize_saved_cmdlines(unsigned int val)
6058 struct saved_cmdlines_buffer *s, *savedcmd_temp;
6060 s = kmalloc(sizeof(*s), GFP_KERNEL);
6064 if (allocate_cmdlines_buffer(val, s) < 0) {
6070 arch_spin_lock(&trace_cmdline_lock);
6071 savedcmd_temp = savedcmd;
6073 arch_spin_unlock(&trace_cmdline_lock);
6075 free_saved_cmdlines_buffer(savedcmd_temp);
6081 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
6082 size_t cnt, loff_t *ppos)
6087 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6091 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
6092 if (!val || val > PID_MAX_DEFAULT)
6095 ret = tracing_resize_saved_cmdlines((unsigned int)val);
6104 static const struct file_operations tracing_saved_cmdlines_size_fops = {
6105 .open = tracing_open_generic,
6106 .read = tracing_saved_cmdlines_size_read,
6107 .write = tracing_saved_cmdlines_size_write,
6110 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
6111 static union trace_eval_map_item *
6112 update_eval_map(union trace_eval_map_item *ptr)
6114 if (!ptr->map.eval_string) {
6115 if (ptr->tail.next) {
6116 ptr = ptr->tail.next;
6117 /* Set ptr to the next real item (skip head) */
6125 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6127 union trace_eval_map_item *ptr = v;
6130 * Paranoid! If ptr points to end, we don't want to increment past it.
6131 * This really should never happen.
6134 ptr = update_eval_map(ptr);
6135 if (WARN_ON_ONCE(!ptr))
6139 ptr = update_eval_map(ptr);
6144 static void *eval_map_start(struct seq_file *m, loff_t *pos)
6146 union trace_eval_map_item *v;
6149 mutex_lock(&trace_eval_mutex);
6151 v = trace_eval_maps;
6155 while (v && l < *pos) {
6156 v = eval_map_next(m, v, &l);
6162 static void eval_map_stop(struct seq_file *m, void *v)
6164 mutex_unlock(&trace_eval_mutex);
6167 static int eval_map_show(struct seq_file *m, void *v)
6169 union trace_eval_map_item *ptr = v;
6171 seq_printf(m, "%s %ld (%s)\n",
6172 ptr->map.eval_string, ptr->map.eval_value,
6178 static const struct seq_operations tracing_eval_map_seq_ops = {
6179 .start = eval_map_start,
6180 .next = eval_map_next,
6181 .stop = eval_map_stop,
6182 .show = eval_map_show,
6185 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6189 ret = tracing_check_open_get_tr(NULL);
6193 return seq_open(filp, &tracing_eval_map_seq_ops);
6196 static const struct file_operations tracing_eval_map_fops = {
6197 .open = tracing_eval_map_open,
6199 .llseek = seq_lseek,
6200 .release = seq_release,
6203 static inline union trace_eval_map_item *
6204 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6206 /* Return tail of array given the head */
6207 return ptr + ptr->head.length + 1;
6211 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6214 struct trace_eval_map **stop;
6215 struct trace_eval_map **map;
6216 union trace_eval_map_item *map_array;
6217 union trace_eval_map_item *ptr;
6222 * The trace_eval_maps contains the map plus a head and tail item,
6223 * where the head holds the module and length of array, and the
6224 * tail holds a pointer to the next list.
6226 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6228 pr_warn("Unable to allocate trace eval mapping\n");
6232 mutex_lock(&trace_eval_mutex);
6234 if (!trace_eval_maps)
6235 trace_eval_maps = map_array;
6237 ptr = trace_eval_maps;
6239 ptr = trace_eval_jmp_to_tail(ptr);
6240 if (!ptr->tail.next)
6242 ptr = ptr->tail.next;
6245 ptr->tail.next = map_array;
6247 map_array->head.mod = mod;
6248 map_array->head.length = len;
6251 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6252 map_array->map = **map;
6255 memset(map_array, 0, sizeof(*map_array));
6257 mutex_unlock(&trace_eval_mutex);
6260 static void trace_create_eval_file(struct dentry *d_tracer)
6262 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6263 NULL, &tracing_eval_map_fops);
6266 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
6267 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6268 static inline void trace_insert_eval_map_file(struct module *mod,
6269 struct trace_eval_map **start, int len) { }
6270 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6272 static void trace_insert_eval_map(struct module *mod,
6273 struct trace_eval_map **start, int len)
6275 struct trace_eval_map **map;
6282 trace_event_eval_update(map, len);
6284 trace_insert_eval_map_file(mod, start, len);
6288 tracing_set_trace_read(struct file *filp, char __user *ubuf,
6289 size_t cnt, loff_t *ppos)
6291 struct trace_array *tr = filp->private_data;
6292 char buf[MAX_TRACER_SIZE+2];
6295 mutex_lock(&trace_types_lock);
6296 r = sprintf(buf, "%s\n", tr->current_trace->name);
6297 mutex_unlock(&trace_types_lock);
6299 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6302 int tracer_init(struct tracer *t, struct trace_array *tr)
6304 tracing_reset_online_cpus(&tr->array_buffer);
6308 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6312 for_each_tracing_cpu(cpu)
6313 per_cpu_ptr(buf->data, cpu)->entries = val;
6316 static void update_buffer_entries(struct array_buffer *buf, int cpu)
6318 if (cpu == RING_BUFFER_ALL_CPUS) {
6319 set_buffer_entries(buf, ring_buffer_size(buf->buffer, 0));
6321 per_cpu_ptr(buf->data, cpu)->entries = ring_buffer_size(buf->buffer, cpu);
6325 #ifdef CONFIG_TRACER_MAX_TRACE
6326 /* resize @tr's buffer to the size of @size_tr's entries */
6327 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6328 struct array_buffer *size_buf, int cpu_id)
6332 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6333 for_each_tracing_cpu(cpu) {
6334 ret = ring_buffer_resize(trace_buf->buffer,
6335 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6338 per_cpu_ptr(trace_buf->data, cpu)->entries =
6339 per_cpu_ptr(size_buf->data, cpu)->entries;
6342 ret = ring_buffer_resize(trace_buf->buffer,
6343 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6345 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6346 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6351 #endif /* CONFIG_TRACER_MAX_TRACE */
6353 static int __tracing_resize_ring_buffer(struct trace_array *tr,
6354 unsigned long size, int cpu)
6359 * If kernel or user changes the size of the ring buffer
6360 * we use the size that was given, and we can forget about
6361 * expanding it later.
6363 ring_buffer_expanded = true;
6365 /* May be called before buffers are initialized */
6366 if (!tr->array_buffer.buffer)
6369 /* Do not allow tracing while resizing ring buffer */
6370 tracing_stop_tr(tr);
6372 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6376 #ifdef CONFIG_TRACER_MAX_TRACE
6377 if (!tr->allocated_snapshot)
6380 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6382 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6383 &tr->array_buffer, cpu);
6386 * AARGH! We are left with different
6387 * size max buffer!!!!
6388 * The max buffer is our "snapshot" buffer.
6389 * When a tracer needs a snapshot (one of the
6390 * latency tracers), it swaps the max buffer
6391 * with the saved snap shot. We succeeded to
6392 * update the size of the main buffer, but failed to
6393 * update the size of the max buffer. But when we tried
6394 * to reset the main buffer to the original size, we
6395 * failed there too. This is very unlikely to
6396 * happen, but if it does, warn and kill all
6400 tracing_disabled = 1;
6405 update_buffer_entries(&tr->max_buffer, cpu);
6408 #endif /* CONFIG_TRACER_MAX_TRACE */
6410 update_buffer_entries(&tr->array_buffer, cpu);
6412 tracing_start_tr(tr);
6416 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6417 unsigned long size, int cpu_id)
6421 mutex_lock(&trace_types_lock);
6423 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6424 /* make sure, this cpu is enabled in the mask */
6425 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6431 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6436 mutex_unlock(&trace_types_lock);
6443 * tracing_update_buffers - used by tracing facility to expand ring buffers
6445 * To save on memory when the tracing is never used on a system with it
6446 * configured in. The ring buffers are set to a minimum size. But once
6447 * a user starts to use the tracing facility, then they need to grow
6448 * to their default size.
6450 * This function is to be called when a tracer is about to be used.
6452 int tracing_update_buffers(void)
6456 mutex_lock(&trace_types_lock);
6457 if (!ring_buffer_expanded)
6458 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
6459 RING_BUFFER_ALL_CPUS);
6460 mutex_unlock(&trace_types_lock);
6465 struct trace_option_dentry;
6468 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6471 * Used to clear out the tracer before deletion of an instance.
6472 * Must have trace_types_lock held.
6474 static void tracing_set_nop(struct trace_array *tr)
6476 if (tr->current_trace == &nop_trace)
6479 tr->current_trace->enabled--;
6481 if (tr->current_trace->reset)
6482 tr->current_trace->reset(tr);
6484 tr->current_trace = &nop_trace;
6487 static bool tracer_options_updated;
6489 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6491 /* Only enable if the directory has been created already. */
6495 /* Only create trace option files after update_tracer_options finish */
6496 if (!tracer_options_updated)
6499 create_trace_option_files(tr, t);
6502 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6505 #ifdef CONFIG_TRACER_MAX_TRACE
6510 mutex_lock(&trace_types_lock);
6512 if (!ring_buffer_expanded) {
6513 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6514 RING_BUFFER_ALL_CPUS);
6520 for (t = trace_types; t; t = t->next) {
6521 if (strcmp(t->name, buf) == 0)
6528 if (t == tr->current_trace)
6531 #ifdef CONFIG_TRACER_SNAPSHOT
6532 if (t->use_max_tr) {
6533 local_irq_disable();
6534 arch_spin_lock(&tr->max_lock);
6535 if (tr->cond_snapshot)
6537 arch_spin_unlock(&tr->max_lock);
6543 /* Some tracers won't work on kernel command line */
6544 if (system_state < SYSTEM_RUNNING && t->noboot) {
6545 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6550 /* Some tracers are only allowed for the top level buffer */
6551 if (!trace_ok_for_array(t, tr)) {
6556 /* If trace pipe files are being read, we can't change the tracer */
6557 if (tr->trace_ref) {
6562 trace_branch_disable();
6564 tr->current_trace->enabled--;
6566 if (tr->current_trace->reset)
6567 tr->current_trace->reset(tr);
6569 #ifdef CONFIG_TRACER_MAX_TRACE
6570 had_max_tr = tr->current_trace->use_max_tr;
6572 /* Current trace needs to be nop_trace before synchronize_rcu */
6573 tr->current_trace = &nop_trace;
6575 if (had_max_tr && !t->use_max_tr) {
6577 * We need to make sure that the update_max_tr sees that
6578 * current_trace changed to nop_trace to keep it from
6579 * swapping the buffers after we resize it.
6580 * The update_max_tr is called from interrupts disabled
6581 * so a synchronized_sched() is sufficient.
6587 if (t->use_max_tr && !tr->allocated_snapshot) {
6588 ret = tracing_alloc_snapshot_instance(tr);
6593 tr->current_trace = &nop_trace;
6597 ret = tracer_init(t, tr);
6602 tr->current_trace = t;
6603 tr->current_trace->enabled++;
6604 trace_branch_enable(tr);
6606 mutex_unlock(&trace_types_lock);
6612 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6613 size_t cnt, loff_t *ppos)
6615 struct trace_array *tr = filp->private_data;
6616 char buf[MAX_TRACER_SIZE+1];
6623 if (cnt > MAX_TRACER_SIZE)
6624 cnt = MAX_TRACER_SIZE;
6626 if (copy_from_user(buf, ubuf, cnt))
6633 err = tracing_set_tracer(tr, name);
6643 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6644 size_t cnt, loff_t *ppos)
6649 r = snprintf(buf, sizeof(buf), "%ld\n",
6650 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6651 if (r > sizeof(buf))
6653 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6657 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6658 size_t cnt, loff_t *ppos)
6663 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6673 tracing_thresh_read(struct file *filp, char __user *ubuf,
6674 size_t cnt, loff_t *ppos)
6676 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6680 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6681 size_t cnt, loff_t *ppos)
6683 struct trace_array *tr = filp->private_data;
6686 mutex_lock(&trace_types_lock);
6687 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6691 if (tr->current_trace->update_thresh) {
6692 ret = tr->current_trace->update_thresh(tr);
6699 mutex_unlock(&trace_types_lock);
6704 #ifdef CONFIG_TRACER_MAX_TRACE
6707 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6708 size_t cnt, loff_t *ppos)
6710 struct trace_array *tr = filp->private_data;
6712 return tracing_nsecs_read(&tr->max_latency, ubuf, cnt, ppos);
6716 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6717 size_t cnt, loff_t *ppos)
6719 struct trace_array *tr = filp->private_data;
6721 return tracing_nsecs_write(&tr->max_latency, ubuf, cnt, ppos);
6726 static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6728 if (cpu == RING_BUFFER_ALL_CPUS) {
6729 if (cpumask_empty(tr->pipe_cpumask)) {
6730 cpumask_setall(tr->pipe_cpumask);
6733 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6734 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6740 static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6742 if (cpu == RING_BUFFER_ALL_CPUS) {
6743 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6744 cpumask_clear(tr->pipe_cpumask);
6746 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6747 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6751 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6753 struct trace_array *tr = inode->i_private;
6754 struct trace_iterator *iter;
6758 ret = tracing_check_open_get_tr(tr);
6762 mutex_lock(&trace_types_lock);
6763 cpu = tracing_get_cpu(inode);
6764 ret = open_pipe_on_cpu(tr, cpu);
6766 goto fail_pipe_on_cpu;
6768 /* create a buffer to store the information to pass to userspace */
6769 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6772 goto fail_alloc_iter;
6775 trace_seq_init(&iter->seq);
6776 iter->trace = tr->current_trace;
6778 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6783 /* trace pipe does not show start of buffer */
6784 cpumask_setall(iter->started);
6786 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6787 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6789 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6790 if (trace_clocks[tr->clock_id].in_ns)
6791 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6794 iter->array_buffer = &tr->array_buffer;
6795 iter->cpu_file = cpu;
6796 mutex_init(&iter->mutex);
6797 filp->private_data = iter;
6799 if (iter->trace->pipe_open)
6800 iter->trace->pipe_open(iter);
6802 nonseekable_open(inode, filp);
6806 mutex_unlock(&trace_types_lock);
6812 close_pipe_on_cpu(tr, cpu);
6814 __trace_array_put(tr);
6815 mutex_unlock(&trace_types_lock);
6819 static int tracing_release_pipe(struct inode *inode, struct file *file)
6821 struct trace_iterator *iter = file->private_data;
6822 struct trace_array *tr = inode->i_private;
6824 mutex_lock(&trace_types_lock);
6828 if (iter->trace->pipe_close)
6829 iter->trace->pipe_close(iter);
6830 close_pipe_on_cpu(tr, iter->cpu_file);
6831 mutex_unlock(&trace_types_lock);
6833 free_trace_iter_content(iter);
6836 trace_array_put(tr);
6842 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6844 struct trace_array *tr = iter->tr;
6846 /* Iterators are static, they should be filled or empty */
6847 if (trace_buffer_iter(iter, iter->cpu_file))
6848 return EPOLLIN | EPOLLRDNORM;
6850 if (tr->trace_flags & TRACE_ITER_BLOCK)
6852 * Always select as readable when in blocking mode
6854 return EPOLLIN | EPOLLRDNORM;
6856 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6857 filp, poll_table, iter->tr->buffer_percent);
6861 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6863 struct trace_iterator *iter = filp->private_data;
6865 return trace_poll(iter, filp, poll_table);
6868 /* Must be called with iter->mutex held. */
6869 static int tracing_wait_pipe(struct file *filp)
6871 struct trace_iterator *iter = filp->private_data;
6874 while (trace_empty(iter)) {
6876 if ((filp->f_flags & O_NONBLOCK)) {
6881 * We block until we read something and tracing is disabled.
6882 * We still block if tracing is disabled, but we have never
6883 * read anything. This allows a user to cat this file, and
6884 * then enable tracing. But after we have read something,
6885 * we give an EOF when tracing is again disabled.
6887 * iter->pos will be 0 if we haven't read anything.
6889 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6892 mutex_unlock(&iter->mutex);
6894 ret = wait_on_pipe(iter, 0);
6896 mutex_lock(&iter->mutex);
6909 tracing_read_pipe(struct file *filp, char __user *ubuf,
6910 size_t cnt, loff_t *ppos)
6912 struct trace_iterator *iter = filp->private_data;
6916 * Avoid more than one consumer on a single file descriptor
6917 * This is just a matter of traces coherency, the ring buffer itself
6920 mutex_lock(&iter->mutex);
6922 /* return any leftover data */
6923 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6927 trace_seq_init(&iter->seq);
6929 if (iter->trace->read) {
6930 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6936 sret = tracing_wait_pipe(filp);
6940 /* stop when tracing is finished */
6941 if (trace_empty(iter)) {
6946 if (cnt >= PAGE_SIZE)
6947 cnt = PAGE_SIZE - 1;
6949 /* reset all but tr, trace, and overruns */
6950 trace_iterator_reset(iter);
6951 cpumask_clear(iter->started);
6952 trace_seq_init(&iter->seq);
6954 trace_event_read_lock();
6955 trace_access_lock(iter->cpu_file);
6956 while (trace_find_next_entry_inc(iter) != NULL) {
6957 enum print_line_t ret;
6958 int save_len = iter->seq.seq.len;
6960 ret = print_trace_line(iter);
6961 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6963 * If one print_trace_line() fills entire trace_seq in one shot,
6964 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6965 * In this case, we need to consume it, otherwise, loop will peek
6966 * this event next time, resulting in an infinite loop.
6968 if (save_len == 0) {
6970 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6971 trace_consume(iter);
6975 /* In other cases, don't print partial lines */
6976 iter->seq.seq.len = save_len;
6979 if (ret != TRACE_TYPE_NO_CONSUME)
6980 trace_consume(iter);
6982 if (trace_seq_used(&iter->seq) >= cnt)
6986 * Setting the full flag means we reached the trace_seq buffer
6987 * size and we should leave by partial output condition above.
6988 * One of the trace_seq_* functions is not used properly.
6990 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6993 trace_access_unlock(iter->cpu_file);
6994 trace_event_read_unlock();
6996 /* Now copy what we have to the user */
6997 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6998 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6999 trace_seq_init(&iter->seq);
7002 * If there was nothing to send to user, in spite of consuming trace
7003 * entries, go back to wait for more entries.
7009 mutex_unlock(&iter->mutex);
7014 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
7017 __free_page(spd->pages[idx]);
7021 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
7027 /* Seq buffer is page-sized, exactly what we need. */
7029 save_len = iter->seq.seq.len;
7030 ret = print_trace_line(iter);
7032 if (trace_seq_has_overflowed(&iter->seq)) {
7033 iter->seq.seq.len = save_len;
7038 * This should not be hit, because it should only
7039 * be set if the iter->seq overflowed. But check it
7040 * anyway to be safe.
7042 if (ret == TRACE_TYPE_PARTIAL_LINE) {
7043 iter->seq.seq.len = save_len;
7047 count = trace_seq_used(&iter->seq) - save_len;
7050 iter->seq.seq.len = save_len;
7054 if (ret != TRACE_TYPE_NO_CONSUME)
7055 trace_consume(iter);
7057 if (!trace_find_next_entry_inc(iter)) {
7067 static ssize_t tracing_splice_read_pipe(struct file *filp,
7069 struct pipe_inode_info *pipe,
7073 struct page *pages_def[PIPE_DEF_BUFFERS];
7074 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7075 struct trace_iterator *iter = filp->private_data;
7076 struct splice_pipe_desc spd = {
7078 .partial = partial_def,
7079 .nr_pages = 0, /* This gets updated below. */
7080 .nr_pages_max = PIPE_DEF_BUFFERS,
7081 .ops = &default_pipe_buf_ops,
7082 .spd_release = tracing_spd_release_pipe,
7088 if (splice_grow_spd(pipe, &spd))
7091 mutex_lock(&iter->mutex);
7093 if (iter->trace->splice_read) {
7094 ret = iter->trace->splice_read(iter, filp,
7095 ppos, pipe, len, flags);
7100 ret = tracing_wait_pipe(filp);
7104 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
7109 trace_event_read_lock();
7110 trace_access_lock(iter->cpu_file);
7112 /* Fill as many pages as possible. */
7113 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
7114 spd.pages[i] = alloc_page(GFP_KERNEL);
7118 rem = tracing_fill_pipe_page(rem, iter);
7120 /* Copy the data into the page, so we can start over. */
7121 ret = trace_seq_to_buffer(&iter->seq,
7122 page_address(spd.pages[i]),
7123 trace_seq_used(&iter->seq));
7125 __free_page(spd.pages[i]);
7128 spd.partial[i].offset = 0;
7129 spd.partial[i].len = trace_seq_used(&iter->seq);
7131 trace_seq_init(&iter->seq);
7134 trace_access_unlock(iter->cpu_file);
7135 trace_event_read_unlock();
7136 mutex_unlock(&iter->mutex);
7141 ret = splice_to_pipe(pipe, &spd);
7145 splice_shrink_spd(&spd);
7149 mutex_unlock(&iter->mutex);
7154 tracing_entries_read(struct file *filp, char __user *ubuf,
7155 size_t cnt, loff_t *ppos)
7157 struct inode *inode = file_inode(filp);
7158 struct trace_array *tr = inode->i_private;
7159 int cpu = tracing_get_cpu(inode);
7164 mutex_lock(&trace_types_lock);
7166 if (cpu == RING_BUFFER_ALL_CPUS) {
7167 int cpu, buf_size_same;
7172 /* check if all cpu sizes are same */
7173 for_each_tracing_cpu(cpu) {
7174 /* fill in the size from first enabled cpu */
7176 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7177 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7183 if (buf_size_same) {
7184 if (!ring_buffer_expanded)
7185 r = sprintf(buf, "%lu (expanded: %lu)\n",
7187 trace_buf_size >> 10);
7189 r = sprintf(buf, "%lu\n", size >> 10);
7191 r = sprintf(buf, "X\n");
7193 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7195 mutex_unlock(&trace_types_lock);
7197 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7202 tracing_entries_write(struct file *filp, const char __user *ubuf,
7203 size_t cnt, loff_t *ppos)
7205 struct inode *inode = file_inode(filp);
7206 struct trace_array *tr = inode->i_private;
7210 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7214 /* must have at least 1 entry */
7218 /* value is in KB */
7220 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7230 tracing_total_entries_read(struct file *filp, char __user *ubuf,
7231 size_t cnt, loff_t *ppos)
7233 struct trace_array *tr = filp->private_data;
7236 unsigned long size = 0, expanded_size = 0;
7238 mutex_lock(&trace_types_lock);
7239 for_each_tracing_cpu(cpu) {
7240 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7241 if (!ring_buffer_expanded)
7242 expanded_size += trace_buf_size >> 10;
7244 if (ring_buffer_expanded)
7245 r = sprintf(buf, "%lu\n", size);
7247 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7248 mutex_unlock(&trace_types_lock);
7250 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7254 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7255 size_t cnt, loff_t *ppos)
7258 * There is no need to read what the user has written, this function
7259 * is just to make sure that there is no error when "echo" is used
7268 tracing_free_buffer_release(struct inode *inode, struct file *filp)
7270 struct trace_array *tr = inode->i_private;
7272 /* disable tracing ? */
7273 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7274 tracer_tracing_off(tr);
7275 /* resize the ring buffer to 0 */
7276 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7278 trace_array_put(tr);
7284 tracing_mark_write(struct file *filp, const char __user *ubuf,
7285 size_t cnt, loff_t *fpos)
7287 struct trace_array *tr = filp->private_data;
7288 struct ring_buffer_event *event;
7289 enum event_trigger_type tt = ETT_NONE;
7290 struct trace_buffer *buffer;
7291 struct print_entry *entry;
7296 /* Used in tracing_mark_raw_write() as well */
7297 #define FAULTED_STR "<faulted>"
7298 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7300 if (tracing_disabled)
7303 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7306 if (cnt > TRACE_BUF_SIZE)
7307 cnt = TRACE_BUF_SIZE;
7309 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7311 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
7313 /* If less than "<faulted>", then make sure we can still add that */
7314 if (cnt < FAULTED_SIZE)
7315 size += FAULTED_SIZE - cnt;
7317 buffer = tr->array_buffer.buffer;
7318 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7320 if (unlikely(!event))
7321 /* Ring buffer disabled, return as if not open for write */
7324 entry = ring_buffer_event_data(event);
7325 entry->ip = _THIS_IP_;
7327 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7329 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7335 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7336 /* do not add \n before testing triggers, but add \0 */
7337 entry->buf[cnt] = '\0';
7338 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7341 if (entry->buf[cnt - 1] != '\n') {
7342 entry->buf[cnt] = '\n';
7343 entry->buf[cnt + 1] = '\0';
7345 entry->buf[cnt] = '\0';
7347 if (static_branch_unlikely(&trace_marker_exports_enabled))
7348 ftrace_exports(event, TRACE_EXPORT_MARKER);
7349 __buffer_unlock_commit(buffer, event);
7352 event_triggers_post_call(tr->trace_marker_file, tt);
7357 /* Limit it for now to 3K (including tag) */
7358 #define RAW_DATA_MAX_SIZE (1024*3)
7361 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7362 size_t cnt, loff_t *fpos)
7364 struct trace_array *tr = filp->private_data;
7365 struct ring_buffer_event *event;
7366 struct trace_buffer *buffer;
7367 struct raw_data_entry *entry;
7372 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7374 if (tracing_disabled)
7377 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7380 /* The marker must at least have a tag id */
7381 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
7384 if (cnt > TRACE_BUF_SIZE)
7385 cnt = TRACE_BUF_SIZE;
7387 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7389 size = sizeof(*entry) + cnt;
7390 if (cnt < FAULT_SIZE_ID)
7391 size += FAULT_SIZE_ID - cnt;
7393 buffer = tr->array_buffer.buffer;
7394 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7397 /* Ring buffer disabled, return as if not open for write */
7400 entry = ring_buffer_event_data(event);
7402 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7405 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7410 __buffer_unlock_commit(buffer, event);
7415 static int tracing_clock_show(struct seq_file *m, void *v)
7417 struct trace_array *tr = m->private;
7420 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7422 "%s%s%s%s", i ? " " : "",
7423 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7424 i == tr->clock_id ? "]" : "");
7430 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7434 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7435 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7438 if (i == ARRAY_SIZE(trace_clocks))
7441 mutex_lock(&trace_types_lock);
7445 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7448 * New clock may not be consistent with the previous clock.
7449 * Reset the buffer so that it doesn't have incomparable timestamps.
7451 tracing_reset_online_cpus(&tr->array_buffer);
7453 #ifdef CONFIG_TRACER_MAX_TRACE
7454 if (tr->max_buffer.buffer)
7455 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7456 tracing_reset_online_cpus(&tr->max_buffer);
7459 mutex_unlock(&trace_types_lock);
7464 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7465 size_t cnt, loff_t *fpos)
7467 struct seq_file *m = filp->private_data;
7468 struct trace_array *tr = m->private;
7470 const char *clockstr;
7473 if (cnt >= sizeof(buf))
7476 if (copy_from_user(buf, ubuf, cnt))
7481 clockstr = strstrip(buf);
7483 ret = tracing_set_clock(tr, clockstr);
7492 static int tracing_clock_open(struct inode *inode, struct file *file)
7494 struct trace_array *tr = inode->i_private;
7497 ret = tracing_check_open_get_tr(tr);
7501 ret = single_open(file, tracing_clock_show, inode->i_private);
7503 trace_array_put(tr);
7508 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7510 struct trace_array *tr = m->private;
7512 mutex_lock(&trace_types_lock);
7514 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7515 seq_puts(m, "delta [absolute]\n");
7517 seq_puts(m, "[delta] absolute\n");
7519 mutex_unlock(&trace_types_lock);
7524 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7526 struct trace_array *tr = inode->i_private;
7529 ret = tracing_check_open_get_tr(tr);
7533 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7535 trace_array_put(tr);
7540 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7542 if (rbe == this_cpu_read(trace_buffered_event))
7543 return ring_buffer_time_stamp(buffer);
7545 return ring_buffer_event_time_stamp(buffer, rbe);
7549 * Set or disable using the per CPU trace_buffer_event when possible.
7551 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7555 mutex_lock(&trace_types_lock);
7557 if (set && tr->no_filter_buffering_ref++)
7561 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7566 --tr->no_filter_buffering_ref;
7569 mutex_unlock(&trace_types_lock);
7574 struct ftrace_buffer_info {
7575 struct trace_iterator iter;
7577 unsigned int spare_cpu;
7581 #ifdef CONFIG_TRACER_SNAPSHOT
7582 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7584 struct trace_array *tr = inode->i_private;
7585 struct trace_iterator *iter;
7589 ret = tracing_check_open_get_tr(tr);
7593 if (file->f_mode & FMODE_READ) {
7594 iter = __tracing_open(inode, file, true);
7596 ret = PTR_ERR(iter);
7598 /* Writes still need the seq_file to hold the private data */
7600 m = kzalloc(sizeof(*m), GFP_KERNEL);
7603 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7611 iter->array_buffer = &tr->max_buffer;
7612 iter->cpu_file = tracing_get_cpu(inode);
7614 file->private_data = m;
7618 trace_array_put(tr);
7623 static void tracing_swap_cpu_buffer(void *tr)
7625 update_max_tr_single((struct trace_array *)tr, current, smp_processor_id());
7629 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7632 struct seq_file *m = filp->private_data;
7633 struct trace_iterator *iter = m->private;
7634 struct trace_array *tr = iter->tr;
7638 ret = tracing_update_buffers();
7642 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7646 mutex_lock(&trace_types_lock);
7648 if (tr->current_trace->use_max_tr) {
7653 local_irq_disable();
7654 arch_spin_lock(&tr->max_lock);
7655 if (tr->cond_snapshot)
7657 arch_spin_unlock(&tr->max_lock);
7664 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7668 if (tr->allocated_snapshot)
7672 /* Only allow per-cpu swap if the ring buffer supports it */
7673 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7674 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7679 if (tr->allocated_snapshot)
7680 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7681 &tr->array_buffer, iter->cpu_file);
7683 ret = tracing_alloc_snapshot_instance(tr);
7686 /* Now, we're going to swap */
7687 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
7688 local_irq_disable();
7689 update_max_tr(tr, current, smp_processor_id(), NULL);
7692 smp_call_function_single(iter->cpu_file, tracing_swap_cpu_buffer,
7697 if (tr->allocated_snapshot) {
7698 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7699 tracing_reset_online_cpus(&tr->max_buffer);
7701 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7711 mutex_unlock(&trace_types_lock);
7715 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7717 struct seq_file *m = file->private_data;
7720 ret = tracing_release(inode, file);
7722 if (file->f_mode & FMODE_READ)
7725 /* If write only, the seq_file is just a stub */
7733 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7734 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7735 size_t count, loff_t *ppos);
7736 static int tracing_buffers_release(struct inode *inode, struct file *file);
7737 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7738 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7740 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7742 struct ftrace_buffer_info *info;
7745 /* The following checks for tracefs lockdown */
7746 ret = tracing_buffers_open(inode, filp);
7750 info = filp->private_data;
7752 if (info->iter.trace->use_max_tr) {
7753 tracing_buffers_release(inode, filp);
7757 info->iter.snapshot = true;
7758 info->iter.array_buffer = &info->iter.tr->max_buffer;
7763 #endif /* CONFIG_TRACER_SNAPSHOT */
7766 static const struct file_operations tracing_thresh_fops = {
7767 .open = tracing_open_generic,
7768 .read = tracing_thresh_read,
7769 .write = tracing_thresh_write,
7770 .llseek = generic_file_llseek,
7773 #ifdef CONFIG_TRACER_MAX_TRACE
7774 static const struct file_operations tracing_max_lat_fops = {
7775 .open = tracing_open_generic_tr,
7776 .read = tracing_max_lat_read,
7777 .write = tracing_max_lat_write,
7778 .llseek = generic_file_llseek,
7779 .release = tracing_release_generic_tr,
7783 static const struct file_operations set_tracer_fops = {
7784 .open = tracing_open_generic_tr,
7785 .read = tracing_set_trace_read,
7786 .write = tracing_set_trace_write,
7787 .llseek = generic_file_llseek,
7788 .release = tracing_release_generic_tr,
7791 static const struct file_operations tracing_pipe_fops = {
7792 .open = tracing_open_pipe,
7793 .poll = tracing_poll_pipe,
7794 .read = tracing_read_pipe,
7795 .splice_read = tracing_splice_read_pipe,
7796 .release = tracing_release_pipe,
7797 .llseek = no_llseek,
7800 static const struct file_operations tracing_entries_fops = {
7801 .open = tracing_open_generic_tr,
7802 .read = tracing_entries_read,
7803 .write = tracing_entries_write,
7804 .llseek = generic_file_llseek,
7805 .release = tracing_release_generic_tr,
7808 static const struct file_operations tracing_total_entries_fops = {
7809 .open = tracing_open_generic_tr,
7810 .read = tracing_total_entries_read,
7811 .llseek = generic_file_llseek,
7812 .release = tracing_release_generic_tr,
7815 static const struct file_operations tracing_free_buffer_fops = {
7816 .open = tracing_open_generic_tr,
7817 .write = tracing_free_buffer_write,
7818 .release = tracing_free_buffer_release,
7821 static const struct file_operations tracing_mark_fops = {
7822 .open = tracing_mark_open,
7823 .write = tracing_mark_write,
7824 .release = tracing_release_generic_tr,
7827 static const struct file_operations tracing_mark_raw_fops = {
7828 .open = tracing_mark_open,
7829 .write = tracing_mark_raw_write,
7830 .release = tracing_release_generic_tr,
7833 static const struct file_operations trace_clock_fops = {
7834 .open = tracing_clock_open,
7836 .llseek = seq_lseek,
7837 .release = tracing_single_release_tr,
7838 .write = tracing_clock_write,
7841 static const struct file_operations trace_time_stamp_mode_fops = {
7842 .open = tracing_time_stamp_mode_open,
7844 .llseek = seq_lseek,
7845 .release = tracing_single_release_tr,
7848 #ifdef CONFIG_TRACER_SNAPSHOT
7849 static const struct file_operations snapshot_fops = {
7850 .open = tracing_snapshot_open,
7852 .write = tracing_snapshot_write,
7853 .llseek = tracing_lseek,
7854 .release = tracing_snapshot_release,
7857 static const struct file_operations snapshot_raw_fops = {
7858 .open = snapshot_raw_open,
7859 .read = tracing_buffers_read,
7860 .release = tracing_buffers_release,
7861 .splice_read = tracing_buffers_splice_read,
7862 .llseek = no_llseek,
7865 #endif /* CONFIG_TRACER_SNAPSHOT */
7868 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7869 * @filp: The active open file structure
7870 * @ubuf: The userspace provided buffer to read value into
7871 * @cnt: The maximum number of bytes to read
7872 * @ppos: The current "file" position
7874 * This function implements the write interface for a struct trace_min_max_param.
7875 * The filp->private_data must point to a trace_min_max_param structure that
7876 * defines where to write the value, the min and the max acceptable values,
7877 * and a lock to protect the write.
7880 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7882 struct trace_min_max_param *param = filp->private_data;
7889 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7894 mutex_lock(param->lock);
7896 if (param->min && val < *param->min)
7899 if (param->max && val > *param->max)
7906 mutex_unlock(param->lock);
7915 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7916 * @filp: The active open file structure
7917 * @ubuf: The userspace provided buffer to read value into
7918 * @cnt: The maximum number of bytes to read
7919 * @ppos: The current "file" position
7921 * This function implements the read interface for a struct trace_min_max_param.
7922 * The filp->private_data must point to a trace_min_max_param struct with valid
7926 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7928 struct trace_min_max_param *param = filp->private_data;
7929 char buf[U64_STR_SIZE];
7938 if (cnt > sizeof(buf))
7941 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7943 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7946 const struct file_operations trace_min_max_fops = {
7947 .open = tracing_open_generic,
7948 .read = trace_min_max_read,
7949 .write = trace_min_max_write,
7952 #define TRACING_LOG_ERRS_MAX 8
7953 #define TRACING_LOG_LOC_MAX 128
7955 #define CMD_PREFIX " Command: "
7958 const char **errs; /* ptr to loc-specific array of err strings */
7959 u8 type; /* index into errs -> specific err string */
7960 u16 pos; /* caret position */
7964 struct tracing_log_err {
7965 struct list_head list;
7966 struct err_info info;
7967 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7968 char *cmd; /* what caused err */
7971 static DEFINE_MUTEX(tracing_err_log_lock);
7973 static struct tracing_log_err *alloc_tracing_log_err(int len)
7975 struct tracing_log_err *err;
7977 err = kzalloc(sizeof(*err), GFP_KERNEL);
7979 return ERR_PTR(-ENOMEM);
7981 err->cmd = kzalloc(len, GFP_KERNEL);
7984 return ERR_PTR(-ENOMEM);
7990 static void free_tracing_log_err(struct tracing_log_err *err)
7996 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7999 struct tracing_log_err *err;
8002 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
8003 err = alloc_tracing_log_err(len);
8004 if (PTR_ERR(err) != -ENOMEM)
8005 tr->n_err_log_entries++;
8009 cmd = kzalloc(len, GFP_KERNEL);
8011 return ERR_PTR(-ENOMEM);
8012 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
8015 list_del(&err->list);
8021 * err_pos - find the position of a string within a command for error careting
8022 * @cmd: The tracing command that caused the error
8023 * @str: The string to position the caret at within @cmd
8025 * Finds the position of the first occurrence of @str within @cmd. The
8026 * return value can be passed to tracing_log_err() for caret placement
8029 * Returns the index within @cmd of the first occurrence of @str or 0
8030 * if @str was not found.
8032 unsigned int err_pos(char *cmd, const char *str)
8036 if (WARN_ON(!strlen(cmd)))
8039 found = strstr(cmd, str);
8047 * tracing_log_err - write an error to the tracing error log
8048 * @tr: The associated trace array for the error (NULL for top level array)
8049 * @loc: A string describing where the error occurred
8050 * @cmd: The tracing command that caused the error
8051 * @errs: The array of loc-specific static error strings
8052 * @type: The index into errs[], which produces the specific static err string
8053 * @pos: The position the caret should be placed in the cmd
8055 * Writes an error into tracing/error_log of the form:
8057 * <loc>: error: <text>
8061 * tracing/error_log is a small log file containing the last
8062 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
8063 * unless there has been a tracing error, and the error log can be
8064 * cleared and have its memory freed by writing the empty string in
8065 * truncation mode to it i.e. echo > tracing/error_log.
8067 * NOTE: the @errs array along with the @type param are used to
8068 * produce a static error string - this string is not copied and saved
8069 * when the error is logged - only a pointer to it is saved. See
8070 * existing callers for examples of how static strings are typically
8071 * defined for use with tracing_log_err().
8073 void tracing_log_err(struct trace_array *tr,
8074 const char *loc, const char *cmd,
8075 const char **errs, u8 type, u16 pos)
8077 struct tracing_log_err *err;
8083 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
8085 mutex_lock(&tracing_err_log_lock);
8086 err = get_tracing_log_err(tr, len);
8087 if (PTR_ERR(err) == -ENOMEM) {
8088 mutex_unlock(&tracing_err_log_lock);
8092 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
8093 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
8095 err->info.errs = errs;
8096 err->info.type = type;
8097 err->info.pos = pos;
8098 err->info.ts = local_clock();
8100 list_add_tail(&err->list, &tr->err_log);
8101 mutex_unlock(&tracing_err_log_lock);
8104 static void clear_tracing_err_log(struct trace_array *tr)
8106 struct tracing_log_err *err, *next;
8108 mutex_lock(&tracing_err_log_lock);
8109 list_for_each_entry_safe(err, next, &tr->err_log, list) {
8110 list_del(&err->list);
8111 free_tracing_log_err(err);
8114 tr->n_err_log_entries = 0;
8115 mutex_unlock(&tracing_err_log_lock);
8118 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
8120 struct trace_array *tr = m->private;
8122 mutex_lock(&tracing_err_log_lock);
8124 return seq_list_start(&tr->err_log, *pos);
8127 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
8129 struct trace_array *tr = m->private;
8131 return seq_list_next(v, &tr->err_log, pos);
8134 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
8136 mutex_unlock(&tracing_err_log_lock);
8139 static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
8143 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
8145 for (i = 0; i < pos; i++)
8150 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
8152 struct tracing_log_err *err = v;
8155 const char *err_text = err->info.errs[err->info.type];
8156 u64 sec = err->info.ts;
8159 nsec = do_div(sec, NSEC_PER_SEC);
8160 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
8161 err->loc, err_text);
8162 seq_printf(m, "%s", err->cmd);
8163 tracing_err_log_show_pos(m, err->info.pos);
8169 static const struct seq_operations tracing_err_log_seq_ops = {
8170 .start = tracing_err_log_seq_start,
8171 .next = tracing_err_log_seq_next,
8172 .stop = tracing_err_log_seq_stop,
8173 .show = tracing_err_log_seq_show
8176 static int tracing_err_log_open(struct inode *inode, struct file *file)
8178 struct trace_array *tr = inode->i_private;
8181 ret = tracing_check_open_get_tr(tr);
8185 /* If this file was opened for write, then erase contents */
8186 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8187 clear_tracing_err_log(tr);
8189 if (file->f_mode & FMODE_READ) {
8190 ret = seq_open(file, &tracing_err_log_seq_ops);
8192 struct seq_file *m = file->private_data;
8195 trace_array_put(tr);
8201 static ssize_t tracing_err_log_write(struct file *file,
8202 const char __user *buffer,
8203 size_t count, loff_t *ppos)
8208 static int tracing_err_log_release(struct inode *inode, struct file *file)
8210 struct trace_array *tr = inode->i_private;
8212 trace_array_put(tr);
8214 if (file->f_mode & FMODE_READ)
8215 seq_release(inode, file);
8220 static const struct file_operations tracing_err_log_fops = {
8221 .open = tracing_err_log_open,
8222 .write = tracing_err_log_write,
8224 .llseek = tracing_lseek,
8225 .release = tracing_err_log_release,
8228 static int tracing_buffers_open(struct inode *inode, struct file *filp)
8230 struct trace_array *tr = inode->i_private;
8231 struct ftrace_buffer_info *info;
8234 ret = tracing_check_open_get_tr(tr);
8238 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8240 trace_array_put(tr);
8244 mutex_lock(&trace_types_lock);
8247 info->iter.cpu_file = tracing_get_cpu(inode);
8248 info->iter.trace = tr->current_trace;
8249 info->iter.array_buffer = &tr->array_buffer;
8251 /* Force reading ring buffer for first read */
8252 info->read = (unsigned int)-1;
8254 filp->private_data = info;
8258 mutex_unlock(&trace_types_lock);
8260 ret = nonseekable_open(inode, filp);
8262 trace_array_put(tr);
8268 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8270 struct ftrace_buffer_info *info = filp->private_data;
8271 struct trace_iterator *iter = &info->iter;
8273 return trace_poll(iter, filp, poll_table);
8277 tracing_buffers_read(struct file *filp, char __user *ubuf,
8278 size_t count, loff_t *ppos)
8280 struct ftrace_buffer_info *info = filp->private_data;
8281 struct trace_iterator *iter = &info->iter;
8288 #ifdef CONFIG_TRACER_MAX_TRACE
8289 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8294 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8296 if (IS_ERR(info->spare)) {
8297 ret = PTR_ERR(info->spare);
8300 info->spare_cpu = iter->cpu_file;
8306 /* Do we have previous read data to read? */
8307 if (info->read < PAGE_SIZE)
8311 trace_access_lock(iter->cpu_file);
8312 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8316 trace_access_unlock(iter->cpu_file);
8319 if (trace_empty(iter)) {
8320 if ((filp->f_flags & O_NONBLOCK))
8323 ret = wait_on_pipe(iter, 0);
8334 size = PAGE_SIZE - info->read;
8338 ret = copy_to_user(ubuf, info->spare + info->read, size);
8350 static int tracing_buffers_release(struct inode *inode, struct file *file)
8352 struct ftrace_buffer_info *info = file->private_data;
8353 struct trace_iterator *iter = &info->iter;
8355 mutex_lock(&trace_types_lock);
8357 iter->tr->trace_ref--;
8359 __trace_array_put(iter->tr);
8362 /* Make sure the waiters see the new wait_index */
8365 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8368 ring_buffer_free_read_page(iter->array_buffer->buffer,
8369 info->spare_cpu, info->spare);
8372 mutex_unlock(&trace_types_lock);
8378 struct trace_buffer *buffer;
8381 refcount_t refcount;
8384 static void buffer_ref_release(struct buffer_ref *ref)
8386 if (!refcount_dec_and_test(&ref->refcount))
8388 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8392 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8393 struct pipe_buffer *buf)
8395 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8397 buffer_ref_release(ref);
8401 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8402 struct pipe_buffer *buf)
8404 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8406 if (refcount_read(&ref->refcount) > INT_MAX/2)
8409 refcount_inc(&ref->refcount);
8413 /* Pipe buffer operations for a buffer. */
8414 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8415 .release = buffer_pipe_buf_release,
8416 .get = buffer_pipe_buf_get,
8420 * Callback from splice_to_pipe(), if we need to release some pages
8421 * at the end of the spd in case we error'ed out in filling the pipe.
8423 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8425 struct buffer_ref *ref =
8426 (struct buffer_ref *)spd->partial[i].private;
8428 buffer_ref_release(ref);
8429 spd->partial[i].private = 0;
8433 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8434 struct pipe_inode_info *pipe, size_t len,
8437 struct ftrace_buffer_info *info = file->private_data;
8438 struct trace_iterator *iter = &info->iter;
8439 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8440 struct page *pages_def[PIPE_DEF_BUFFERS];
8441 struct splice_pipe_desc spd = {
8443 .partial = partial_def,
8444 .nr_pages_max = PIPE_DEF_BUFFERS,
8445 .ops = &buffer_pipe_buf_ops,
8446 .spd_release = buffer_spd_release,
8448 struct buffer_ref *ref;
8452 #ifdef CONFIG_TRACER_MAX_TRACE
8453 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8457 if (*ppos & (PAGE_SIZE - 1))
8460 if (len & (PAGE_SIZE - 1)) {
8461 if (len < PAGE_SIZE)
8466 if (splice_grow_spd(pipe, &spd))
8470 trace_access_lock(iter->cpu_file);
8471 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8473 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
8477 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8483 refcount_set(&ref->refcount, 1);
8484 ref->buffer = iter->array_buffer->buffer;
8485 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8486 if (IS_ERR(ref->page)) {
8487 ret = PTR_ERR(ref->page);
8492 ref->cpu = iter->cpu_file;
8494 r = ring_buffer_read_page(ref->buffer, &ref->page,
8495 len, iter->cpu_file, 1);
8497 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8503 page = virt_to_page(ref->page);
8505 spd.pages[i] = page;
8506 spd.partial[i].len = PAGE_SIZE;
8507 spd.partial[i].offset = 0;
8508 spd.partial[i].private = (unsigned long)ref;
8512 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8515 trace_access_unlock(iter->cpu_file);
8518 /* did we read anything? */
8519 if (!spd.nr_pages) {
8526 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8529 wait_index = READ_ONCE(iter->wait_index);
8531 ret = wait_on_pipe(iter, iter->snapshot ? 0 : iter->tr->buffer_percent);
8535 /* No need to wait after waking up when tracing is off */
8536 if (!tracer_tracing_is_on(iter->tr))
8539 /* Make sure we see the new wait_index */
8541 if (wait_index != iter->wait_index)
8547 ret = splice_to_pipe(pipe, &spd);
8549 splice_shrink_spd(&spd);
8554 /* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8555 static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8557 struct ftrace_buffer_info *info = file->private_data;
8558 struct trace_iterator *iter = &info->iter;
8561 return -ENOIOCTLCMD;
8563 mutex_lock(&trace_types_lock);
8566 /* Make sure the waiters see the new wait_index */
8569 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8571 mutex_unlock(&trace_types_lock);
8575 static const struct file_operations tracing_buffers_fops = {
8576 .open = tracing_buffers_open,
8577 .read = tracing_buffers_read,
8578 .poll = tracing_buffers_poll,
8579 .release = tracing_buffers_release,
8580 .splice_read = tracing_buffers_splice_read,
8581 .unlocked_ioctl = tracing_buffers_ioctl,
8582 .llseek = no_llseek,
8586 tracing_stats_read(struct file *filp, char __user *ubuf,
8587 size_t count, loff_t *ppos)
8589 struct inode *inode = file_inode(filp);
8590 struct trace_array *tr = inode->i_private;
8591 struct array_buffer *trace_buf = &tr->array_buffer;
8592 int cpu = tracing_get_cpu(inode);
8593 struct trace_seq *s;
8595 unsigned long long t;
8596 unsigned long usec_rem;
8598 s = kmalloc(sizeof(*s), GFP_KERNEL);
8604 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8605 trace_seq_printf(s, "entries: %ld\n", cnt);
8607 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8608 trace_seq_printf(s, "overrun: %ld\n", cnt);
8610 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8611 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8613 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8614 trace_seq_printf(s, "bytes: %ld\n", cnt);
8616 if (trace_clocks[tr->clock_id].in_ns) {
8617 /* local or global for trace_clock */
8618 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8619 usec_rem = do_div(t, USEC_PER_SEC);
8620 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8623 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8624 usec_rem = do_div(t, USEC_PER_SEC);
8625 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8627 /* counter or tsc mode for trace_clock */
8628 trace_seq_printf(s, "oldest event ts: %llu\n",
8629 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8631 trace_seq_printf(s, "now ts: %llu\n",
8632 ring_buffer_time_stamp(trace_buf->buffer));
8635 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8636 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8638 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8639 trace_seq_printf(s, "read events: %ld\n", cnt);
8641 count = simple_read_from_buffer(ubuf, count, ppos,
8642 s->buffer, trace_seq_used(s));
8649 static const struct file_operations tracing_stats_fops = {
8650 .open = tracing_open_generic_tr,
8651 .read = tracing_stats_read,
8652 .llseek = generic_file_llseek,
8653 .release = tracing_release_generic_tr,
8656 #ifdef CONFIG_DYNAMIC_FTRACE
8659 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8660 size_t cnt, loff_t *ppos)
8666 /* 256 should be plenty to hold the amount needed */
8667 buf = kmalloc(256, GFP_KERNEL);
8671 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8672 ftrace_update_tot_cnt,
8673 ftrace_number_of_pages,
8674 ftrace_number_of_groups);
8676 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8681 static const struct file_operations tracing_dyn_info_fops = {
8682 .open = tracing_open_generic,
8683 .read = tracing_read_dyn_info,
8684 .llseek = generic_file_llseek,
8686 #endif /* CONFIG_DYNAMIC_FTRACE */
8688 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8690 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8691 struct trace_array *tr, struct ftrace_probe_ops *ops,
8694 tracing_snapshot_instance(tr);
8698 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8699 struct trace_array *tr, struct ftrace_probe_ops *ops,
8702 struct ftrace_func_mapper *mapper = data;
8706 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8716 tracing_snapshot_instance(tr);
8720 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8721 struct ftrace_probe_ops *ops, void *data)
8723 struct ftrace_func_mapper *mapper = data;
8726 seq_printf(m, "%ps:", (void *)ip);
8728 seq_puts(m, "snapshot");
8731 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8734 seq_printf(m, ":count=%ld\n", *count);
8736 seq_puts(m, ":unlimited\n");
8742 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8743 unsigned long ip, void *init_data, void **data)
8745 struct ftrace_func_mapper *mapper = *data;
8748 mapper = allocate_ftrace_func_mapper();
8754 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8758 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8759 unsigned long ip, void *data)
8761 struct ftrace_func_mapper *mapper = data;
8766 free_ftrace_func_mapper(mapper, NULL);
8770 ftrace_func_mapper_remove_ip(mapper, ip);
8773 static struct ftrace_probe_ops snapshot_probe_ops = {
8774 .func = ftrace_snapshot,
8775 .print = ftrace_snapshot_print,
8778 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8779 .func = ftrace_count_snapshot,
8780 .print = ftrace_snapshot_print,
8781 .init = ftrace_snapshot_init,
8782 .free = ftrace_snapshot_free,
8786 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8787 char *glob, char *cmd, char *param, int enable)
8789 struct ftrace_probe_ops *ops;
8790 void *count = (void *)-1;
8797 /* hash funcs only work with set_ftrace_filter */
8801 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8804 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8809 number = strsep(¶m, ":");
8811 if (!strlen(number))
8815 * We use the callback data field (which is a pointer)
8818 ret = kstrtoul(number, 0, (unsigned long *)&count);
8823 ret = tracing_alloc_snapshot_instance(tr);
8827 ret = register_ftrace_function_probe(glob, tr, ops, count);
8830 return ret < 0 ? ret : 0;
8833 static struct ftrace_func_command ftrace_snapshot_cmd = {
8835 .func = ftrace_trace_snapshot_callback,
8838 static __init int register_snapshot_cmd(void)
8840 return register_ftrace_command(&ftrace_snapshot_cmd);
8843 static inline __init int register_snapshot_cmd(void) { return 0; }
8844 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8846 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8848 if (WARN_ON(!tr->dir))
8849 return ERR_PTR(-ENODEV);
8851 /* Top directory uses NULL as the parent */
8852 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8855 /* All sub buffers have a descriptor */
8859 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8861 struct dentry *d_tracer;
8864 return tr->percpu_dir;
8866 d_tracer = tracing_get_dentry(tr);
8867 if (IS_ERR(d_tracer))
8870 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8872 MEM_FAIL(!tr->percpu_dir,
8873 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8875 return tr->percpu_dir;
8878 static struct dentry *
8879 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8880 void *data, long cpu, const struct file_operations *fops)
8882 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8884 if (ret) /* See tracing_get_cpu() */
8885 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8890 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8892 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8893 struct dentry *d_cpu;
8894 char cpu_dir[30]; /* 30 characters should be more than enough */
8899 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8900 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8902 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8906 /* per cpu trace_pipe */
8907 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8908 tr, cpu, &tracing_pipe_fops);
8911 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8912 tr, cpu, &tracing_fops);
8914 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8915 tr, cpu, &tracing_buffers_fops);
8917 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8918 tr, cpu, &tracing_stats_fops);
8920 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8921 tr, cpu, &tracing_entries_fops);
8923 #ifdef CONFIG_TRACER_SNAPSHOT
8924 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8925 tr, cpu, &snapshot_fops);
8927 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8928 tr, cpu, &snapshot_raw_fops);
8932 #ifdef CONFIG_FTRACE_SELFTEST
8933 /* Let selftest have access to static functions in this file */
8934 #include "trace_selftest.c"
8938 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8941 struct trace_option_dentry *topt = filp->private_data;
8944 if (topt->flags->val & topt->opt->bit)
8949 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8953 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8956 struct trace_option_dentry *topt = filp->private_data;
8960 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8964 if (val != 0 && val != 1)
8967 if (!!(topt->flags->val & topt->opt->bit) != val) {
8968 mutex_lock(&trace_types_lock);
8969 ret = __set_tracer_option(topt->tr, topt->flags,
8971 mutex_unlock(&trace_types_lock);
8981 static int tracing_open_options(struct inode *inode, struct file *filp)
8983 struct trace_option_dentry *topt = inode->i_private;
8986 ret = tracing_check_open_get_tr(topt->tr);
8990 filp->private_data = inode->i_private;
8994 static int tracing_release_options(struct inode *inode, struct file *file)
8996 struct trace_option_dentry *topt = file->private_data;
8998 trace_array_put(topt->tr);
9002 static const struct file_operations trace_options_fops = {
9003 .open = tracing_open_options,
9004 .read = trace_options_read,
9005 .write = trace_options_write,
9006 .llseek = generic_file_llseek,
9007 .release = tracing_release_options,
9011 * In order to pass in both the trace_array descriptor as well as the index
9012 * to the flag that the trace option file represents, the trace_array
9013 * has a character array of trace_flags_index[], which holds the index
9014 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
9015 * The address of this character array is passed to the flag option file
9016 * read/write callbacks.
9018 * In order to extract both the index and the trace_array descriptor,
9019 * get_tr_index() uses the following algorithm.
9023 * As the pointer itself contains the address of the index (remember
9026 * Then to get the trace_array descriptor, by subtracting that index
9027 * from the ptr, we get to the start of the index itself.
9029 * ptr - idx == &index[0]
9031 * Then a simple container_of() from that pointer gets us to the
9032 * trace_array descriptor.
9034 static void get_tr_index(void *data, struct trace_array **ptr,
9035 unsigned int *pindex)
9037 *pindex = *(unsigned char *)data;
9039 *ptr = container_of(data - *pindex, struct trace_array,
9044 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
9047 void *tr_index = filp->private_data;
9048 struct trace_array *tr;
9052 get_tr_index(tr_index, &tr, &index);
9054 if (tr->trace_flags & (1 << index))
9059 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9063 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
9066 void *tr_index = filp->private_data;
9067 struct trace_array *tr;
9072 get_tr_index(tr_index, &tr, &index);
9074 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9078 if (val != 0 && val != 1)
9081 mutex_lock(&event_mutex);
9082 mutex_lock(&trace_types_lock);
9083 ret = set_tracer_flag(tr, 1 << index, val);
9084 mutex_unlock(&trace_types_lock);
9085 mutex_unlock(&event_mutex);
9095 static const struct file_operations trace_options_core_fops = {
9096 .open = tracing_open_generic,
9097 .read = trace_options_core_read,
9098 .write = trace_options_core_write,
9099 .llseek = generic_file_llseek,
9102 struct dentry *trace_create_file(const char *name,
9104 struct dentry *parent,
9106 const struct file_operations *fops)
9110 ret = tracefs_create_file(name, mode, parent, data, fops);
9112 pr_warn("Could not create tracefs '%s' entry\n", name);
9118 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
9120 struct dentry *d_tracer;
9125 d_tracer = tracing_get_dentry(tr);
9126 if (IS_ERR(d_tracer))
9129 tr->options = tracefs_create_dir("options", d_tracer);
9131 pr_warn("Could not create tracefs directory 'options'\n");
9139 create_trace_option_file(struct trace_array *tr,
9140 struct trace_option_dentry *topt,
9141 struct tracer_flags *flags,
9142 struct tracer_opt *opt)
9144 struct dentry *t_options;
9146 t_options = trace_options_init_dentry(tr);
9150 topt->flags = flags;
9154 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
9155 t_options, topt, &trace_options_fops);
9160 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
9162 struct trace_option_dentry *topts;
9163 struct trace_options *tr_topts;
9164 struct tracer_flags *flags;
9165 struct tracer_opt *opts;
9172 flags = tracer->flags;
9174 if (!flags || !flags->opts)
9178 * If this is an instance, only create flags for tracers
9179 * the instance may have.
9181 if (!trace_ok_for_array(tracer, tr))
9184 for (i = 0; i < tr->nr_topts; i++) {
9185 /* Make sure there's no duplicate flags. */
9186 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
9192 for (cnt = 0; opts[cnt].name; cnt++)
9195 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
9199 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9206 tr->topts = tr_topts;
9207 tr->topts[tr->nr_topts].tracer = tracer;
9208 tr->topts[tr->nr_topts].topts = topts;
9211 for (cnt = 0; opts[cnt].name; cnt++) {
9212 create_trace_option_file(tr, &topts[cnt], flags,
9214 MEM_FAIL(topts[cnt].entry == NULL,
9215 "Failed to create trace option: %s",
9220 static struct dentry *
9221 create_trace_option_core_file(struct trace_array *tr,
9222 const char *option, long index)
9224 struct dentry *t_options;
9226 t_options = trace_options_init_dentry(tr);
9230 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9231 (void *)&tr->trace_flags_index[index],
9232 &trace_options_core_fops);
9235 static void create_trace_options_dir(struct trace_array *tr)
9237 struct dentry *t_options;
9238 bool top_level = tr == &global_trace;
9241 t_options = trace_options_init_dentry(tr);
9245 for (i = 0; trace_options[i]; i++) {
9247 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9248 create_trace_option_core_file(tr, trace_options[i], i);
9253 rb_simple_read(struct file *filp, char __user *ubuf,
9254 size_t cnt, loff_t *ppos)
9256 struct trace_array *tr = filp->private_data;
9260 r = tracer_tracing_is_on(tr);
9261 r = sprintf(buf, "%d\n", r);
9263 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9267 rb_simple_write(struct file *filp, const char __user *ubuf,
9268 size_t cnt, loff_t *ppos)
9270 struct trace_array *tr = filp->private_data;
9271 struct trace_buffer *buffer = tr->array_buffer.buffer;
9275 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9280 mutex_lock(&trace_types_lock);
9281 if (!!val == tracer_tracing_is_on(tr)) {
9282 val = 0; /* do nothing */
9284 tracer_tracing_on(tr);
9285 if (tr->current_trace->start)
9286 tr->current_trace->start(tr);
9288 tracer_tracing_off(tr);
9289 if (tr->current_trace->stop)
9290 tr->current_trace->stop(tr);
9291 /* Wake up any waiters */
9292 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9294 mutex_unlock(&trace_types_lock);
9302 static const struct file_operations rb_simple_fops = {
9303 .open = tracing_open_generic_tr,
9304 .read = rb_simple_read,
9305 .write = rb_simple_write,
9306 .release = tracing_release_generic_tr,
9307 .llseek = default_llseek,
9311 buffer_percent_read(struct file *filp, char __user *ubuf,
9312 size_t cnt, loff_t *ppos)
9314 struct trace_array *tr = filp->private_data;
9318 r = tr->buffer_percent;
9319 r = sprintf(buf, "%d\n", r);
9321 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9325 buffer_percent_write(struct file *filp, const char __user *ubuf,
9326 size_t cnt, loff_t *ppos)
9328 struct trace_array *tr = filp->private_data;
9332 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9339 tr->buffer_percent = val;
9346 static const struct file_operations buffer_percent_fops = {
9347 .open = tracing_open_generic_tr,
9348 .read = buffer_percent_read,
9349 .write = buffer_percent_write,
9350 .release = tracing_release_generic_tr,
9351 .llseek = default_llseek,
9354 static struct dentry *trace_instance_dir;
9357 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9360 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9362 enum ring_buffer_flags rb_flags;
9364 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9368 buf->buffer = ring_buffer_alloc(size, rb_flags);
9372 buf->data = alloc_percpu(struct trace_array_cpu);
9374 ring_buffer_free(buf->buffer);
9379 /* Allocate the first page for all buffers */
9380 set_buffer_entries(&tr->array_buffer,
9381 ring_buffer_size(tr->array_buffer.buffer, 0));
9386 static void free_trace_buffer(struct array_buffer *buf)
9389 ring_buffer_free(buf->buffer);
9391 free_percpu(buf->data);
9396 static int allocate_trace_buffers(struct trace_array *tr, int size)
9400 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9404 #ifdef CONFIG_TRACER_MAX_TRACE
9405 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9406 allocate_snapshot ? size : 1);
9407 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9408 free_trace_buffer(&tr->array_buffer);
9411 tr->allocated_snapshot = allocate_snapshot;
9413 allocate_snapshot = false;
9419 static void free_trace_buffers(struct trace_array *tr)
9424 free_trace_buffer(&tr->array_buffer);
9426 #ifdef CONFIG_TRACER_MAX_TRACE
9427 free_trace_buffer(&tr->max_buffer);
9431 static void init_trace_flags_index(struct trace_array *tr)
9435 /* Used by the trace options files */
9436 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9437 tr->trace_flags_index[i] = i;
9440 static void __update_tracer_options(struct trace_array *tr)
9444 for (t = trace_types; t; t = t->next)
9445 add_tracer_options(tr, t);
9448 static void update_tracer_options(struct trace_array *tr)
9450 mutex_lock(&trace_types_lock);
9451 tracer_options_updated = true;
9452 __update_tracer_options(tr);
9453 mutex_unlock(&trace_types_lock);
9456 /* Must have trace_types_lock held */
9457 struct trace_array *trace_array_find(const char *instance)
9459 struct trace_array *tr, *found = NULL;
9461 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9462 if (tr->name && strcmp(tr->name, instance) == 0) {
9471 struct trace_array *trace_array_find_get(const char *instance)
9473 struct trace_array *tr;
9475 mutex_lock(&trace_types_lock);
9476 tr = trace_array_find(instance);
9479 mutex_unlock(&trace_types_lock);
9484 static int trace_array_create_dir(struct trace_array *tr)
9488 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9492 ret = event_trace_add_tracer(tr->dir, tr);
9494 tracefs_remove(tr->dir);
9498 init_tracer_tracefs(tr, tr->dir);
9499 __update_tracer_options(tr);
9504 static struct trace_array *trace_array_create(const char *name)
9506 struct trace_array *tr;
9510 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9512 return ERR_PTR(ret);
9514 tr->name = kstrdup(name, GFP_KERNEL);
9518 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9521 if (!zalloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9524 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9526 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9528 raw_spin_lock_init(&tr->start_lock);
9530 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9532 tr->current_trace = &nop_trace;
9534 INIT_LIST_HEAD(&tr->systems);
9535 INIT_LIST_HEAD(&tr->events);
9536 INIT_LIST_HEAD(&tr->hist_vars);
9537 INIT_LIST_HEAD(&tr->err_log);
9539 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9542 if (ftrace_allocate_ftrace_ops(tr) < 0)
9545 ftrace_init_trace_array(tr);
9547 init_trace_flags_index(tr);
9549 if (trace_instance_dir) {
9550 ret = trace_array_create_dir(tr);
9554 __trace_early_add_events(tr);
9556 list_add(&tr->list, &ftrace_trace_arrays);
9563 ftrace_free_ftrace_ops(tr);
9564 free_trace_buffers(tr);
9565 free_cpumask_var(tr->pipe_cpumask);
9566 free_cpumask_var(tr->tracing_cpumask);
9570 return ERR_PTR(ret);
9573 static int instance_mkdir(const char *name)
9575 struct trace_array *tr;
9578 mutex_lock(&event_mutex);
9579 mutex_lock(&trace_types_lock);
9582 if (trace_array_find(name))
9585 tr = trace_array_create(name);
9587 ret = PTR_ERR_OR_ZERO(tr);
9590 mutex_unlock(&trace_types_lock);
9591 mutex_unlock(&event_mutex);
9596 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9597 * @name: The name of the trace array to be looked up/created.
9599 * Returns pointer to trace array with given name.
9600 * NULL, if it cannot be created.
9602 * NOTE: This function increments the reference counter associated with the
9603 * trace array returned. This makes sure it cannot be freed while in use.
9604 * Use trace_array_put() once the trace array is no longer needed.
9605 * If the trace_array is to be freed, trace_array_destroy() needs to
9606 * be called after the trace_array_put(), or simply let user space delete
9607 * it from the tracefs instances directory. But until the
9608 * trace_array_put() is called, user space can not delete it.
9611 struct trace_array *trace_array_get_by_name(const char *name)
9613 struct trace_array *tr;
9615 mutex_lock(&event_mutex);
9616 mutex_lock(&trace_types_lock);
9618 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9619 if (tr->name && strcmp(tr->name, name) == 0)
9623 tr = trace_array_create(name);
9631 mutex_unlock(&trace_types_lock);
9632 mutex_unlock(&event_mutex);
9635 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9637 static int __remove_instance(struct trace_array *tr)
9641 /* Reference counter for a newly created trace array = 1. */
9642 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9645 list_del(&tr->list);
9647 /* Disable all the flags that were enabled coming in */
9648 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9649 if ((1 << i) & ZEROED_TRACE_FLAGS)
9650 set_tracer_flag(tr, 1 << i, 0);
9653 tracing_set_nop(tr);
9654 clear_ftrace_function_probes(tr);
9655 event_trace_del_tracer(tr);
9656 ftrace_clear_pids(tr);
9657 ftrace_destroy_function_files(tr);
9658 tracefs_remove(tr->dir);
9659 free_percpu(tr->last_func_repeats);
9660 free_trace_buffers(tr);
9661 clear_tracing_err_log(tr);
9663 for (i = 0; i < tr->nr_topts; i++) {
9664 kfree(tr->topts[i].topts);
9668 free_cpumask_var(tr->pipe_cpumask);
9669 free_cpumask_var(tr->tracing_cpumask);
9676 int trace_array_destroy(struct trace_array *this_tr)
9678 struct trace_array *tr;
9684 mutex_lock(&event_mutex);
9685 mutex_lock(&trace_types_lock);
9689 /* Making sure trace array exists before destroying it. */
9690 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9691 if (tr == this_tr) {
9692 ret = __remove_instance(tr);
9697 mutex_unlock(&trace_types_lock);
9698 mutex_unlock(&event_mutex);
9702 EXPORT_SYMBOL_GPL(trace_array_destroy);
9704 static int instance_rmdir(const char *name)
9706 struct trace_array *tr;
9709 mutex_lock(&event_mutex);
9710 mutex_lock(&trace_types_lock);
9713 tr = trace_array_find(name);
9715 ret = __remove_instance(tr);
9717 mutex_unlock(&trace_types_lock);
9718 mutex_unlock(&event_mutex);
9723 static __init void create_trace_instances(struct dentry *d_tracer)
9725 struct trace_array *tr;
9727 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9730 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9733 mutex_lock(&event_mutex);
9734 mutex_lock(&trace_types_lock);
9736 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9739 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9740 "Failed to create instance directory\n"))
9744 mutex_unlock(&trace_types_lock);
9745 mutex_unlock(&event_mutex);
9749 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9751 struct trace_event_file *file;
9754 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9755 tr, &show_traces_fops);
9757 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9758 tr, &set_tracer_fops);
9760 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9761 tr, &tracing_cpumask_fops);
9763 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9764 tr, &tracing_iter_fops);
9766 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9769 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9770 tr, &tracing_pipe_fops);
9772 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9773 tr, &tracing_entries_fops);
9775 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9776 tr, &tracing_total_entries_fops);
9778 trace_create_file("free_buffer", 0200, d_tracer,
9779 tr, &tracing_free_buffer_fops);
9781 trace_create_file("trace_marker", 0220, d_tracer,
9782 tr, &tracing_mark_fops);
9784 file = __find_event_file(tr, "ftrace", "print");
9785 if (file && file->ef)
9786 eventfs_add_file("trigger", TRACE_MODE_WRITE, file->ef,
9787 file, &event_trigger_fops);
9788 tr->trace_marker_file = file;
9790 trace_create_file("trace_marker_raw", 0220, d_tracer,
9791 tr, &tracing_mark_raw_fops);
9793 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9796 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9797 tr, &rb_simple_fops);
9799 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9800 &trace_time_stamp_mode_fops);
9802 tr->buffer_percent = 50;
9804 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9805 tr, &buffer_percent_fops);
9807 create_trace_options_dir(tr);
9809 #ifdef CONFIG_TRACER_MAX_TRACE
9810 trace_create_maxlat_file(tr, d_tracer);
9813 if (ftrace_create_function_files(tr, d_tracer))
9814 MEM_FAIL(1, "Could not allocate function filter files");
9816 #ifdef CONFIG_TRACER_SNAPSHOT
9817 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9818 tr, &snapshot_fops);
9821 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9822 tr, &tracing_err_log_fops);
9824 for_each_tracing_cpu(cpu)
9825 tracing_init_tracefs_percpu(tr, cpu);
9827 ftrace_init_tracefs(tr, d_tracer);
9830 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9832 struct vfsmount *mnt;
9833 struct file_system_type *type;
9836 * To maintain backward compatibility for tools that mount
9837 * debugfs to get to the tracing facility, tracefs is automatically
9838 * mounted to the debugfs/tracing directory.
9840 type = get_fs_type("tracefs");
9843 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9844 put_filesystem(type);
9853 * tracing_init_dentry - initialize top level trace array
9855 * This is called when creating files or directories in the tracing
9856 * directory. It is called via fs_initcall() by any of the boot up code
9857 * and expects to return the dentry of the top level tracing directory.
9859 int tracing_init_dentry(void)
9861 struct trace_array *tr = &global_trace;
9863 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9864 pr_warn("Tracing disabled due to lockdown\n");
9868 /* The top level trace array uses NULL as parent */
9872 if (WARN_ON(!tracefs_initialized()))
9876 * As there may still be users that expect the tracing
9877 * files to exist in debugfs/tracing, we must automount
9878 * the tracefs file system there, so older tools still
9879 * work with the newer kernel.
9881 tr->dir = debugfs_create_automount("tracing", NULL,
9882 trace_automount, NULL);
9887 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9888 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9890 static struct workqueue_struct *eval_map_wq __initdata;
9891 static struct work_struct eval_map_work __initdata;
9892 static struct work_struct tracerfs_init_work __initdata;
9894 static void __init eval_map_work_func(struct work_struct *work)
9898 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9899 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9902 static int __init trace_eval_init(void)
9904 INIT_WORK(&eval_map_work, eval_map_work_func);
9906 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9908 pr_err("Unable to allocate eval_map_wq\n");
9910 eval_map_work_func(&eval_map_work);
9914 queue_work(eval_map_wq, &eval_map_work);
9918 subsys_initcall(trace_eval_init);
9920 static int __init trace_eval_sync(void)
9922 /* Make sure the eval map updates are finished */
9924 destroy_workqueue(eval_map_wq);
9928 late_initcall_sync(trace_eval_sync);
9931 #ifdef CONFIG_MODULES
9932 static void trace_module_add_evals(struct module *mod)
9934 if (!mod->num_trace_evals)
9938 * Modules with bad taint do not have events created, do
9939 * not bother with enums either.
9941 if (trace_module_has_bad_taint(mod))
9944 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9947 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
9948 static void trace_module_remove_evals(struct module *mod)
9950 union trace_eval_map_item *map;
9951 union trace_eval_map_item **last = &trace_eval_maps;
9953 if (!mod->num_trace_evals)
9956 mutex_lock(&trace_eval_mutex);
9958 map = trace_eval_maps;
9961 if (map->head.mod == mod)
9963 map = trace_eval_jmp_to_tail(map);
9964 last = &map->tail.next;
9965 map = map->tail.next;
9970 *last = trace_eval_jmp_to_tail(map)->tail.next;
9973 mutex_unlock(&trace_eval_mutex);
9976 static inline void trace_module_remove_evals(struct module *mod) { }
9977 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9979 static int trace_module_notify(struct notifier_block *self,
9980 unsigned long val, void *data)
9982 struct module *mod = data;
9985 case MODULE_STATE_COMING:
9986 trace_module_add_evals(mod);
9988 case MODULE_STATE_GOING:
9989 trace_module_remove_evals(mod);
9996 static struct notifier_block trace_module_nb = {
9997 .notifier_call = trace_module_notify,
10000 #endif /* CONFIG_MODULES */
10002 static __init void tracer_init_tracefs_work_func(struct work_struct *work)
10005 event_trace_init();
10007 init_tracer_tracefs(&global_trace, NULL);
10008 ftrace_init_tracefs_toplevel(&global_trace, NULL);
10010 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
10011 &global_trace, &tracing_thresh_fops);
10013 trace_create_file("README", TRACE_MODE_READ, NULL,
10014 NULL, &tracing_readme_fops);
10016 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
10017 NULL, &tracing_saved_cmdlines_fops);
10019 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
10020 NULL, &tracing_saved_cmdlines_size_fops);
10022 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
10023 NULL, &tracing_saved_tgids_fops);
10025 trace_create_eval_file(NULL);
10027 #ifdef CONFIG_MODULES
10028 register_module_notifier(&trace_module_nb);
10031 #ifdef CONFIG_DYNAMIC_FTRACE
10032 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
10033 NULL, &tracing_dyn_info_fops);
10036 create_trace_instances(NULL);
10038 update_tracer_options(&global_trace);
10041 static __init int tracer_init_tracefs(void)
10045 trace_access_lock_init();
10047 ret = tracing_init_dentry();
10052 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
10053 queue_work(eval_map_wq, &tracerfs_init_work);
10055 tracer_init_tracefs_work_func(NULL);
10058 rv_init_interface();
10063 fs_initcall(tracer_init_tracefs);
10065 static int trace_die_panic_handler(struct notifier_block *self,
10066 unsigned long ev, void *unused);
10068 static struct notifier_block trace_panic_notifier = {
10069 .notifier_call = trace_die_panic_handler,
10070 .priority = INT_MAX - 1,
10073 static struct notifier_block trace_die_notifier = {
10074 .notifier_call = trace_die_panic_handler,
10075 .priority = INT_MAX - 1,
10079 * The idea is to execute the following die/panic callback early, in order
10080 * to avoid showing irrelevant information in the trace (like other panic
10081 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
10082 * warnings get disabled (to prevent potential log flooding).
10084 static int trace_die_panic_handler(struct notifier_block *self,
10085 unsigned long ev, void *unused)
10087 if (!ftrace_dump_on_oops)
10088 return NOTIFY_DONE;
10090 /* The die notifier requires DIE_OOPS to trigger */
10091 if (self == &trace_die_notifier && ev != DIE_OOPS)
10092 return NOTIFY_DONE;
10094 ftrace_dump(ftrace_dump_on_oops);
10096 return NOTIFY_DONE;
10100 * printk is set to max of 1024, we really don't need it that big.
10101 * Nothing should be printing 1000 characters anyway.
10103 #define TRACE_MAX_PRINT 1000
10106 * Define here KERN_TRACE so that we have one place to modify
10107 * it if we decide to change what log level the ftrace dump
10110 #define KERN_TRACE KERN_EMERG
10113 trace_printk_seq(struct trace_seq *s)
10115 /* Probably should print a warning here. */
10116 if (s->seq.len >= TRACE_MAX_PRINT)
10117 s->seq.len = TRACE_MAX_PRINT;
10120 * More paranoid code. Although the buffer size is set to
10121 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
10122 * an extra layer of protection.
10124 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
10125 s->seq.len = s->seq.size - 1;
10127 /* should be zero ended, but we are paranoid. */
10128 s->buffer[s->seq.len] = 0;
10130 printk(KERN_TRACE "%s", s->buffer);
10135 void trace_init_global_iter(struct trace_iterator *iter)
10137 iter->tr = &global_trace;
10138 iter->trace = iter->tr->current_trace;
10139 iter->cpu_file = RING_BUFFER_ALL_CPUS;
10140 iter->array_buffer = &global_trace.array_buffer;
10142 if (iter->trace && iter->trace->open)
10143 iter->trace->open(iter);
10145 /* Annotate start of buffers if we had overruns */
10146 if (ring_buffer_overruns(iter->array_buffer->buffer))
10147 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10149 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10150 if (trace_clocks[iter->tr->clock_id].in_ns)
10151 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10153 /* Can not use kmalloc for iter.temp and iter.fmt */
10154 iter->temp = static_temp_buf;
10155 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10156 iter->fmt = static_fmt_buf;
10157 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10160 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10162 /* use static because iter can be a bit big for the stack */
10163 static struct trace_iterator iter;
10164 static atomic_t dump_running;
10165 struct trace_array *tr = &global_trace;
10166 unsigned int old_userobj;
10167 unsigned long flags;
10170 /* Only allow one dump user at a time. */
10171 if (atomic_inc_return(&dump_running) != 1) {
10172 atomic_dec(&dump_running);
10177 * Always turn off tracing when we dump.
10178 * We don't need to show trace output of what happens
10179 * between multiple crashes.
10181 * If the user does a sysrq-z, then they can re-enable
10182 * tracing with echo 1 > tracing_on.
10186 local_irq_save(flags);
10188 /* Simulate the iterator */
10189 trace_init_global_iter(&iter);
10191 for_each_tracing_cpu(cpu) {
10192 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10195 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10197 /* don't look at user memory in panic mode */
10198 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10200 switch (oops_dump_mode) {
10202 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10205 iter.cpu_file = raw_smp_processor_id();
10210 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10211 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10214 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10216 /* Did function tracer already get disabled? */
10217 if (ftrace_is_dead()) {
10218 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10219 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10223 * We need to stop all tracing on all CPUS to read
10224 * the next buffer. This is a bit expensive, but is
10225 * not done often. We fill all what we can read,
10226 * and then release the locks again.
10229 while (!trace_empty(&iter)) {
10232 printk(KERN_TRACE "---------------------------------\n");
10236 trace_iterator_reset(&iter);
10237 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10239 if (trace_find_next_entry_inc(&iter) != NULL) {
10242 ret = print_trace_line(&iter);
10243 if (ret != TRACE_TYPE_NO_CONSUME)
10244 trace_consume(&iter);
10246 touch_nmi_watchdog();
10248 trace_printk_seq(&iter.seq);
10252 printk(KERN_TRACE " (ftrace buffer empty)\n");
10254 printk(KERN_TRACE "---------------------------------\n");
10257 tr->trace_flags |= old_userobj;
10259 for_each_tracing_cpu(cpu) {
10260 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10262 atomic_dec(&dump_running);
10263 local_irq_restore(flags);
10265 EXPORT_SYMBOL_GPL(ftrace_dump);
10267 #define WRITE_BUFSIZE 4096
10269 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10270 size_t count, loff_t *ppos,
10271 int (*createfn)(const char *))
10273 char *kbuf, *buf, *tmp;
10278 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10282 while (done < count) {
10283 size = count - done;
10285 if (size >= WRITE_BUFSIZE)
10286 size = WRITE_BUFSIZE - 1;
10288 if (copy_from_user(kbuf, buffer + done, size)) {
10295 tmp = strchr(buf, '\n');
10298 size = tmp - buf + 1;
10300 size = strlen(buf);
10301 if (done + size < count) {
10304 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10305 pr_warn("Line length is too long: Should be less than %d\n",
10306 WRITE_BUFSIZE - 2);
10313 /* Remove comments */
10314 tmp = strchr(buf, '#');
10319 ret = createfn(buf);
10324 } while (done < count);
10334 #ifdef CONFIG_TRACER_MAX_TRACE
10335 __init static bool tr_needs_alloc_snapshot(const char *name)
10338 int len = strlen(name);
10341 if (!boot_snapshot_index)
10344 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10345 boot_snapshot_info[len] == '\t')
10348 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10352 sprintf(test, "\t%s\t", name);
10353 ret = strstr(boot_snapshot_info, test) == NULL;
10358 __init static void do_allocate_snapshot(const char *name)
10360 if (!tr_needs_alloc_snapshot(name))
10364 * When allocate_snapshot is set, the next call to
10365 * allocate_trace_buffers() (called by trace_array_get_by_name())
10366 * will allocate the snapshot buffer. That will alse clear
10369 allocate_snapshot = true;
10372 static inline void do_allocate_snapshot(const char *name) { }
10375 __init static void enable_instances(void)
10377 struct trace_array *tr;
10382 /* A tab is always appended */
10383 boot_instance_info[boot_instance_index - 1] = '\0';
10384 str = boot_instance_info;
10386 while ((curr_str = strsep(&str, "\t"))) {
10388 tok = strsep(&curr_str, ",");
10390 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10391 do_allocate_snapshot(tok);
10393 tr = trace_array_get_by_name(tok);
10395 pr_warn("Failed to create instance buffer %s\n", curr_str);
10398 /* Allow user space to delete it */
10399 trace_array_put(tr);
10401 while ((tok = strsep(&curr_str, ","))) {
10402 early_enable_events(tr, tok, true);
10407 __init static int tracer_alloc_buffers(void)
10413 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10414 pr_warn("Tracing disabled due to lockdown\n");
10419 * Make sure we don't accidentally add more trace options
10420 * than we have bits for.
10422 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10424 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10427 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10428 goto out_free_buffer_mask;
10430 /* Only allocate trace_printk buffers if a trace_printk exists */
10431 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10432 /* Must be called before global_trace.buffer is allocated */
10433 trace_printk_init_buffers();
10435 /* To save memory, keep the ring buffer size to its minimum */
10436 if (ring_buffer_expanded)
10437 ring_buf_size = trace_buf_size;
10441 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10442 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10444 raw_spin_lock_init(&global_trace.start_lock);
10447 * The prepare callbacks allocates some memory for the ring buffer. We
10448 * don't free the buffer if the CPU goes down. If we were to free
10449 * the buffer, then the user would lose any trace that was in the
10450 * buffer. The memory will be removed once the "instance" is removed.
10452 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10453 "trace/RB:prepare", trace_rb_cpu_prepare,
10456 goto out_free_cpumask;
10457 /* Used for event triggers */
10459 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10461 goto out_rm_hp_state;
10463 if (trace_create_savedcmd() < 0)
10464 goto out_free_temp_buffer;
10466 if (!zalloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10467 goto out_free_savedcmd;
10469 /* TODO: make the number of buffers hot pluggable with CPUS */
10470 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10471 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10472 goto out_free_pipe_cpumask;
10474 if (global_trace.buffer_disabled)
10477 if (trace_boot_clock) {
10478 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10480 pr_warn("Trace clock %s not defined, going back to default\n",
10485 * register_tracer() might reference current_trace, so it
10486 * needs to be set before we register anything. This is
10487 * just a bootstrap of current_trace anyway.
10489 global_trace.current_trace = &nop_trace;
10491 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10493 ftrace_init_global_array_ops(&global_trace);
10495 init_trace_flags_index(&global_trace);
10497 register_tracer(&nop_trace);
10499 /* Function tracing may start here (via kernel command line) */
10500 init_function_trace();
10502 /* All seems OK, enable tracing */
10503 tracing_disabled = 0;
10505 atomic_notifier_chain_register(&panic_notifier_list,
10506 &trace_panic_notifier);
10508 register_die_notifier(&trace_die_notifier);
10510 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10512 INIT_LIST_HEAD(&global_trace.systems);
10513 INIT_LIST_HEAD(&global_trace.events);
10514 INIT_LIST_HEAD(&global_trace.hist_vars);
10515 INIT_LIST_HEAD(&global_trace.err_log);
10516 list_add(&global_trace.list, &ftrace_trace_arrays);
10518 apply_trace_boot_options();
10520 register_snapshot_cmd();
10526 out_free_pipe_cpumask:
10527 free_cpumask_var(global_trace.pipe_cpumask);
10529 free_saved_cmdlines_buffer(savedcmd);
10530 out_free_temp_buffer:
10531 ring_buffer_free(temp_buffer);
10533 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10535 free_cpumask_var(global_trace.tracing_cpumask);
10536 out_free_buffer_mask:
10537 free_cpumask_var(tracing_buffer_mask);
10542 void __init ftrace_boot_snapshot(void)
10544 #ifdef CONFIG_TRACER_MAX_TRACE
10545 struct trace_array *tr;
10547 if (!snapshot_at_boot)
10550 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10551 if (!tr->allocated_snapshot)
10554 tracing_snapshot_instance(tr);
10555 trace_array_puts(tr, "** Boot snapshot taken **\n");
10560 void __init early_trace_init(void)
10562 if (tracepoint_printk) {
10563 tracepoint_print_iter =
10564 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10565 if (MEM_FAIL(!tracepoint_print_iter,
10566 "Failed to allocate trace iterator\n"))
10567 tracepoint_printk = 0;
10569 static_key_enable(&tracepoint_printk_key.key);
10571 tracer_alloc_buffers();
10576 void __init trace_init(void)
10578 trace_event_init();
10580 if (boot_instance_index)
10581 enable_instances();
10584 __init static void clear_boot_tracer(void)
10587 * The default tracer at boot buffer is an init section.
10588 * This function is called in lateinit. If we did not
10589 * find the boot tracer, then clear it out, to prevent
10590 * later registration from accessing the buffer that is
10591 * about to be freed.
10593 if (!default_bootup_tracer)
10596 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10597 default_bootup_tracer);
10598 default_bootup_tracer = NULL;
10601 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10602 __init static void tracing_set_default_clock(void)
10604 /* sched_clock_stable() is determined in late_initcall */
10605 if (!trace_boot_clock && !sched_clock_stable()) {
10606 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10607 pr_warn("Can not set tracing clock due to lockdown\n");
10611 printk(KERN_WARNING
10612 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10613 "If you want to keep using the local clock, then add:\n"
10614 " \"trace_clock=local\"\n"
10615 "on the kernel command line\n");
10616 tracing_set_clock(&global_trace, "global");
10620 static inline void tracing_set_default_clock(void) { }
10623 __init static int late_trace_init(void)
10625 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10626 static_key_disable(&tracepoint_printk_key.key);
10627 tracepoint_printk = 0;
10630 tracing_set_default_clock();
10631 clear_boot_tracer();
10635 late_initcall_sync(late_trace_init);