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/kmemleak.h>
43 #include <linux/poll.h>
44 #include <linux/nmi.h>
46 #include <linux/trace.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/rt.h>
49 #include <linux/fsnotify.h>
50 #include <linux/irq_work.h>
51 #include <linux/workqueue.h>
53 #include <asm/setup.h> /* COMMAND_LINE_SIZE */
56 #include "trace_output.h"
58 #ifdef CONFIG_FTRACE_STARTUP_TEST
60 * We need to change this state when a selftest is running.
61 * A selftest will lurk into the ring-buffer to count the
62 * entries inserted during the selftest although some concurrent
63 * insertions into the ring-buffer such as trace_printk could occurred
64 * at the same time, giving false positive or negative results.
66 static bool __read_mostly tracing_selftest_running;
69 * If boot-time tracing including tracers/events via kernel cmdline
70 * is running, we do not want to run SELFTEST.
72 bool __read_mostly tracing_selftest_disabled;
74 void __init disable_tracing_selftest(const char *reason)
76 if (!tracing_selftest_disabled) {
77 tracing_selftest_disabled = true;
78 pr_info("Ftrace startup test is disabled due to %s\n", reason);
82 #define tracing_selftest_running 0
83 #define tracing_selftest_disabled 0
86 /* Pipe tracepoints to printk */
87 static struct trace_iterator *tracepoint_print_iter;
88 int tracepoint_printk;
89 static bool tracepoint_printk_stop_on_boot __initdata;
90 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
92 /* For tracers that don't implement custom flags */
93 static struct tracer_opt dummy_tracer_opt[] = {
98 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
104 * To prevent the comm cache from being overwritten when no
105 * tracing is active, only save the comm when a trace event
108 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
111 * Kill all tracing for good (never come back).
112 * It is initialized to 1 but will turn to zero if the initialization
113 * of the tracer is successful. But that is the only place that sets
116 static int tracing_disabled = 1;
118 cpumask_var_t __read_mostly tracing_buffer_mask;
121 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
123 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
124 * is set, then ftrace_dump is called. This will output the contents
125 * of the ftrace buffers to the console. This is very useful for
126 * capturing traces that lead to crashes and outputing it to a
129 * It is default off, but you can enable it with either specifying
130 * "ftrace_dump_on_oops" in the kernel command line, or setting
131 * /proc/sys/kernel/ftrace_dump_on_oops
132 * Set 1 if you want to dump buffers of all CPUs
133 * Set 2 if you want to dump the buffer of the CPU that triggered oops
136 enum ftrace_dump_mode ftrace_dump_on_oops;
138 /* When set, tracing will stop when a WARN*() is hit */
139 int __disable_trace_on_warning;
141 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
142 /* Map of enums to their values, for "eval_map" file */
143 struct trace_eval_map_head {
145 unsigned long length;
148 union trace_eval_map_item;
150 struct trace_eval_map_tail {
152 * "end" is first and points to NULL as it must be different
153 * than "mod" or "eval_string"
155 union trace_eval_map_item *next;
156 const char *end; /* points to NULL */
159 static DEFINE_MUTEX(trace_eval_mutex);
162 * The trace_eval_maps are saved in an array with two extra elements,
163 * one at the beginning, and one at the end. The beginning item contains
164 * the count of the saved maps (head.length), and the module they
165 * belong to if not built in (head.mod). The ending item contains a
166 * pointer to the next array of saved eval_map items.
168 union trace_eval_map_item {
169 struct trace_eval_map map;
170 struct trace_eval_map_head head;
171 struct trace_eval_map_tail tail;
174 static union trace_eval_map_item *trace_eval_maps;
175 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
177 int tracing_set_tracer(struct trace_array *tr, const char *buf);
178 static void ftrace_trace_userstack(struct trace_array *tr,
179 struct trace_buffer *buffer,
180 unsigned int trace_ctx);
182 #define MAX_TRACER_SIZE 100
183 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
184 static char *default_bootup_tracer;
186 static bool allocate_snapshot;
187 static bool snapshot_at_boot;
189 static char boot_instance_info[COMMAND_LINE_SIZE] __initdata;
190 static int boot_instance_index;
192 static char boot_snapshot_info[COMMAND_LINE_SIZE] __initdata;
193 static int boot_snapshot_index;
195 static int __init set_cmdline_ftrace(char *str)
197 strscpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
198 default_bootup_tracer = bootup_tracer_buf;
199 /* We are using ftrace early, expand it */
200 trace_set_ring_buffer_expanded(NULL);
203 __setup("ftrace=", set_cmdline_ftrace);
205 static int __init set_ftrace_dump_on_oops(char *str)
207 if (*str++ != '=' || !*str || !strcmp("1", str)) {
208 ftrace_dump_on_oops = DUMP_ALL;
212 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
213 ftrace_dump_on_oops = DUMP_ORIG;
219 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
221 static int __init stop_trace_on_warning(char *str)
223 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
224 __disable_trace_on_warning = 1;
227 __setup("traceoff_on_warning", stop_trace_on_warning);
229 static int __init boot_alloc_snapshot(char *str)
231 char *slot = boot_snapshot_info + boot_snapshot_index;
232 int left = sizeof(boot_snapshot_info) - boot_snapshot_index;
237 if (strlen(str) >= left)
240 ret = snprintf(slot, left, "%s\t", str);
241 boot_snapshot_index += ret;
243 allocate_snapshot = true;
244 /* We also need the main ring buffer expanded */
245 trace_set_ring_buffer_expanded(NULL);
249 __setup("alloc_snapshot", boot_alloc_snapshot);
252 static int __init boot_snapshot(char *str)
254 snapshot_at_boot = true;
255 boot_alloc_snapshot(str);
258 __setup("ftrace_boot_snapshot", boot_snapshot);
261 static int __init boot_instance(char *str)
263 char *slot = boot_instance_info + boot_instance_index;
264 int left = sizeof(boot_instance_info) - boot_instance_index;
267 if (strlen(str) >= left)
270 ret = snprintf(slot, left, "%s\t", str);
271 boot_instance_index += ret;
275 __setup("trace_instance=", boot_instance);
278 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
280 static int __init set_trace_boot_options(char *str)
282 strscpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
285 __setup("trace_options=", set_trace_boot_options);
287 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
288 static char *trace_boot_clock __initdata;
290 static int __init set_trace_boot_clock(char *str)
292 strscpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
293 trace_boot_clock = trace_boot_clock_buf;
296 __setup("trace_clock=", set_trace_boot_clock);
298 static int __init set_tracepoint_printk(char *str)
300 /* Ignore the "tp_printk_stop_on_boot" param */
304 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
305 tracepoint_printk = 1;
308 __setup("tp_printk", set_tracepoint_printk);
310 static int __init set_tracepoint_printk_stop(char *str)
312 tracepoint_printk_stop_on_boot = true;
315 __setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
317 unsigned long long ns2usecs(u64 nsec)
325 trace_process_export(struct trace_export *export,
326 struct ring_buffer_event *event, int flag)
328 struct trace_entry *entry;
329 unsigned int size = 0;
331 if (export->flags & flag) {
332 entry = ring_buffer_event_data(event);
333 size = ring_buffer_event_length(event);
334 export->write(export, entry, size);
338 static DEFINE_MUTEX(ftrace_export_lock);
340 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
342 static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
343 static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
344 static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
346 static inline void ftrace_exports_enable(struct trace_export *export)
348 if (export->flags & TRACE_EXPORT_FUNCTION)
349 static_branch_inc(&trace_function_exports_enabled);
351 if (export->flags & TRACE_EXPORT_EVENT)
352 static_branch_inc(&trace_event_exports_enabled);
354 if (export->flags & TRACE_EXPORT_MARKER)
355 static_branch_inc(&trace_marker_exports_enabled);
358 static inline void ftrace_exports_disable(struct trace_export *export)
360 if (export->flags & TRACE_EXPORT_FUNCTION)
361 static_branch_dec(&trace_function_exports_enabled);
363 if (export->flags & TRACE_EXPORT_EVENT)
364 static_branch_dec(&trace_event_exports_enabled);
366 if (export->flags & TRACE_EXPORT_MARKER)
367 static_branch_dec(&trace_marker_exports_enabled);
370 static void ftrace_exports(struct ring_buffer_event *event, int flag)
372 struct trace_export *export;
374 preempt_disable_notrace();
376 export = rcu_dereference_raw_check(ftrace_exports_list);
378 trace_process_export(export, event, flag);
379 export = rcu_dereference_raw_check(export->next);
382 preempt_enable_notrace();
386 add_trace_export(struct trace_export **list, struct trace_export *export)
388 rcu_assign_pointer(export->next, *list);
390 * We are entering export into the list but another
391 * CPU might be walking that list. We need to make sure
392 * the export->next pointer is valid before another CPU sees
393 * the export pointer included into the list.
395 rcu_assign_pointer(*list, export);
399 rm_trace_export(struct trace_export **list, struct trace_export *export)
401 struct trace_export **p;
403 for (p = list; *p != NULL; p = &(*p)->next)
410 rcu_assign_pointer(*p, (*p)->next);
416 add_ftrace_export(struct trace_export **list, struct trace_export *export)
418 ftrace_exports_enable(export);
420 add_trace_export(list, export);
424 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
428 ret = rm_trace_export(list, export);
429 ftrace_exports_disable(export);
434 int register_ftrace_export(struct trace_export *export)
436 if (WARN_ON_ONCE(!export->write))
439 mutex_lock(&ftrace_export_lock);
441 add_ftrace_export(&ftrace_exports_list, export);
443 mutex_unlock(&ftrace_export_lock);
447 EXPORT_SYMBOL_GPL(register_ftrace_export);
449 int unregister_ftrace_export(struct trace_export *export)
453 mutex_lock(&ftrace_export_lock);
455 ret = rm_ftrace_export(&ftrace_exports_list, export);
457 mutex_unlock(&ftrace_export_lock);
461 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
463 /* trace_flags holds trace_options default values */
464 #define TRACE_DEFAULT_FLAGS \
465 (FUNCTION_DEFAULT_FLAGS | \
466 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
467 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
468 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
469 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
472 /* trace_options that are only supported by global_trace */
473 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
474 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
476 /* trace_flags that are default zero for instances */
477 #define ZEROED_TRACE_FLAGS \
478 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
481 * The global_trace is the descriptor that holds the top-level tracing
482 * buffers for the live tracing.
484 static struct trace_array global_trace = {
485 .trace_flags = TRACE_DEFAULT_FLAGS,
488 void trace_set_ring_buffer_expanded(struct trace_array *tr)
492 tr->ring_buffer_expanded = true;
495 LIST_HEAD(ftrace_trace_arrays);
497 int trace_array_get(struct trace_array *this_tr)
499 struct trace_array *tr;
502 mutex_lock(&trace_types_lock);
503 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
510 mutex_unlock(&trace_types_lock);
515 static void __trace_array_put(struct trace_array *this_tr)
517 WARN_ON(!this_tr->ref);
522 * trace_array_put - Decrement the reference counter for this trace array.
523 * @this_tr : pointer to the trace array
525 * NOTE: Use this when we no longer need the trace array returned by
526 * trace_array_get_by_name(). This ensures the trace array can be later
530 void trace_array_put(struct trace_array *this_tr)
535 mutex_lock(&trace_types_lock);
536 __trace_array_put(this_tr);
537 mutex_unlock(&trace_types_lock);
539 EXPORT_SYMBOL_GPL(trace_array_put);
541 int tracing_check_open_get_tr(struct trace_array *tr)
545 ret = security_locked_down(LOCKDOWN_TRACEFS);
549 if (tracing_disabled)
552 if (tr && trace_array_get(tr) < 0)
558 int call_filter_check_discard(struct trace_event_call *call, void *rec,
559 struct trace_buffer *buffer,
560 struct ring_buffer_event *event)
562 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
563 !filter_match_preds(call->filter, rec)) {
564 __trace_event_discard_commit(buffer, event);
572 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
573 * @filtered_pids: The list of pids to check
574 * @search_pid: The PID to find in @filtered_pids
576 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
579 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
581 return trace_pid_list_is_set(filtered_pids, search_pid);
585 * trace_ignore_this_task - should a task be ignored for tracing
586 * @filtered_pids: The list of pids to check
587 * @filtered_no_pids: The list of pids not to be traced
588 * @task: The task that should be ignored if not filtered
590 * Checks if @task should be traced or not from @filtered_pids.
591 * Returns true if @task should *NOT* be traced.
592 * Returns false if @task should be traced.
595 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
596 struct trace_pid_list *filtered_no_pids,
597 struct task_struct *task)
600 * If filtered_no_pids is not empty, and the task's pid is listed
601 * in filtered_no_pids, then return true.
602 * Otherwise, if filtered_pids is empty, that means we can
603 * trace all tasks. If it has content, then only trace pids
604 * within filtered_pids.
607 return (filtered_pids &&
608 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
610 trace_find_filtered_pid(filtered_no_pids, task->pid));
614 * trace_filter_add_remove_task - Add or remove a task from a pid_list
615 * @pid_list: The list to modify
616 * @self: The current task for fork or NULL for exit
617 * @task: The task to add or remove
619 * If adding a task, if @self is defined, the task is only added if @self
620 * is also included in @pid_list. This happens on fork and tasks should
621 * only be added when the parent is listed. If @self is NULL, then the
622 * @task pid will be removed from the list, which would happen on exit
625 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
626 struct task_struct *self,
627 struct task_struct *task)
632 /* For forks, we only add if the forking task is listed */
634 if (!trace_find_filtered_pid(pid_list, self->pid))
638 /* "self" is set for forks, and NULL for exits */
640 trace_pid_list_set(pid_list, task->pid);
642 trace_pid_list_clear(pid_list, task->pid);
646 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
647 * @pid_list: The pid list to show
648 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
649 * @pos: The position of the file
651 * This is used by the seq_file "next" operation to iterate the pids
652 * listed in a trace_pid_list structure.
654 * Returns the pid+1 as we want to display pid of zero, but NULL would
655 * stop the iteration.
657 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
659 long pid = (unsigned long)v;
664 /* pid already is +1 of the actual previous bit */
665 if (trace_pid_list_next(pid_list, pid, &next) < 0)
670 /* Return pid + 1 to allow zero to be represented */
671 return (void *)(pid + 1);
675 * trace_pid_start - Used for seq_file to start reading pid lists
676 * @pid_list: The pid list to show
677 * @pos: The position of the file
679 * This is used by seq_file "start" operation to start the iteration
682 * Returns the pid+1 as we want to display pid of zero, but NULL would
683 * stop the iteration.
685 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
691 if (trace_pid_list_first(pid_list, &first) < 0)
696 /* Return pid + 1 so that zero can be the exit value */
697 for (pid++; pid && l < *pos;
698 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
704 * trace_pid_show - show the current pid in seq_file processing
705 * @m: The seq_file structure to write into
706 * @v: A void pointer of the pid (+1) value to display
708 * Can be directly used by seq_file operations to display the current
711 int trace_pid_show(struct seq_file *m, void *v)
713 unsigned long pid = (unsigned long)v - 1;
715 seq_printf(m, "%lu\n", pid);
719 /* 128 should be much more than enough */
720 #define PID_BUF_SIZE 127
722 int trace_pid_write(struct trace_pid_list *filtered_pids,
723 struct trace_pid_list **new_pid_list,
724 const char __user *ubuf, size_t cnt)
726 struct trace_pid_list *pid_list;
727 struct trace_parser parser;
735 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
739 * Always recreate a new array. The write is an all or nothing
740 * operation. Always create a new array when adding new pids by
741 * the user. If the operation fails, then the current list is
744 pid_list = trace_pid_list_alloc();
746 trace_parser_put(&parser);
751 /* copy the current bits to the new max */
752 ret = trace_pid_list_first(filtered_pids, &pid);
754 trace_pid_list_set(pid_list, pid);
755 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
765 ret = trace_get_user(&parser, ubuf, cnt, &pos);
773 if (!trace_parser_loaded(&parser))
777 if (kstrtoul(parser.buffer, 0, &val))
782 if (trace_pid_list_set(pid_list, pid) < 0) {
788 trace_parser_clear(&parser);
791 trace_parser_put(&parser);
794 trace_pid_list_free(pid_list);
799 /* Cleared the list of pids */
800 trace_pid_list_free(pid_list);
804 *new_pid_list = pid_list;
809 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
813 /* Early boot up does not have a buffer yet */
815 return trace_clock_local();
817 ts = ring_buffer_time_stamp(buf->buffer);
818 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
823 u64 ftrace_now(int cpu)
825 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
829 * tracing_is_enabled - Show if global_trace has been enabled
831 * Shows if the global trace has been enabled or not. It uses the
832 * mirror flag "buffer_disabled" to be used in fast paths such as for
833 * the irqsoff tracer. But it may be inaccurate due to races. If you
834 * need to know the accurate state, use tracing_is_on() which is a little
835 * slower, but accurate.
837 int tracing_is_enabled(void)
840 * For quick access (irqsoff uses this in fast path), just
841 * return the mirror variable of the state of the ring buffer.
842 * It's a little racy, but we don't really care.
845 return !global_trace.buffer_disabled;
849 * trace_buf_size is the size in bytes that is allocated
850 * for a buffer. Note, the number of bytes is always rounded
853 * This number is purposely set to a low number of 16384.
854 * If the dump on oops happens, it will be much appreciated
855 * to not have to wait for all that output. Anyway this can be
856 * boot time and run time configurable.
858 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
860 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
862 /* trace_types holds a link list of available tracers. */
863 static struct tracer *trace_types __read_mostly;
866 * trace_types_lock is used to protect the trace_types list.
868 DEFINE_MUTEX(trace_types_lock);
871 * serialize the access of the ring buffer
873 * ring buffer serializes readers, but it is low level protection.
874 * The validity of the events (which returns by ring_buffer_peek() ..etc)
875 * are not protected by ring buffer.
877 * The content of events may become garbage if we allow other process consumes
878 * these events concurrently:
879 * A) the page of the consumed events may become a normal page
880 * (not reader page) in ring buffer, and this page will be rewritten
881 * by events producer.
882 * B) The page of the consumed events may become a page for splice_read,
883 * and this page will be returned to system.
885 * These primitives allow multi process access to different cpu ring buffer
888 * These primitives don't distinguish read-only and read-consume access.
889 * Multi read-only access are also serialized.
893 static DECLARE_RWSEM(all_cpu_access_lock);
894 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
896 static inline void trace_access_lock(int cpu)
898 if (cpu == RING_BUFFER_ALL_CPUS) {
899 /* gain it for accessing the whole ring buffer. */
900 down_write(&all_cpu_access_lock);
902 /* gain it for accessing a cpu ring buffer. */
904 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
905 down_read(&all_cpu_access_lock);
907 /* Secondly block other access to this @cpu ring buffer. */
908 mutex_lock(&per_cpu(cpu_access_lock, cpu));
912 static inline void trace_access_unlock(int cpu)
914 if (cpu == RING_BUFFER_ALL_CPUS) {
915 up_write(&all_cpu_access_lock);
917 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
918 up_read(&all_cpu_access_lock);
922 static inline void trace_access_lock_init(void)
926 for_each_possible_cpu(cpu)
927 mutex_init(&per_cpu(cpu_access_lock, cpu));
932 static DEFINE_MUTEX(access_lock);
934 static inline void trace_access_lock(int cpu)
937 mutex_lock(&access_lock);
940 static inline void trace_access_unlock(int cpu)
943 mutex_unlock(&access_lock);
946 static inline void trace_access_lock_init(void)
952 #ifdef CONFIG_STACKTRACE
953 static void __ftrace_trace_stack(struct trace_buffer *buffer,
954 unsigned int trace_ctx,
955 int skip, struct pt_regs *regs);
956 static inline void ftrace_trace_stack(struct trace_array *tr,
957 struct trace_buffer *buffer,
958 unsigned int trace_ctx,
959 int skip, struct pt_regs *regs);
962 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
963 unsigned int trace_ctx,
964 int skip, struct pt_regs *regs)
967 static inline void ftrace_trace_stack(struct trace_array *tr,
968 struct trace_buffer *buffer,
969 unsigned long trace_ctx,
970 int skip, struct pt_regs *regs)
976 static __always_inline void
977 trace_event_setup(struct ring_buffer_event *event,
978 int type, unsigned int trace_ctx)
980 struct trace_entry *ent = ring_buffer_event_data(event);
982 tracing_generic_entry_update(ent, type, trace_ctx);
985 static __always_inline struct ring_buffer_event *
986 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
989 unsigned int trace_ctx)
991 struct ring_buffer_event *event;
993 event = ring_buffer_lock_reserve(buffer, len);
995 trace_event_setup(event, type, trace_ctx);
1000 void tracer_tracing_on(struct trace_array *tr)
1002 if (tr->array_buffer.buffer)
1003 ring_buffer_record_on(tr->array_buffer.buffer);
1005 * This flag is looked at when buffers haven't been allocated
1006 * yet, or by some tracers (like irqsoff), that just want to
1007 * know if the ring buffer has been disabled, but it can handle
1008 * races of where it gets disabled but we still do a record.
1009 * As the check is in the fast path of the tracers, it is more
1010 * important to be fast than accurate.
1012 tr->buffer_disabled = 0;
1013 /* Make the flag seen by readers */
1018 * tracing_on - enable tracing buffers
1020 * This function enables tracing buffers that may have been
1021 * disabled with tracing_off.
1023 void tracing_on(void)
1025 tracer_tracing_on(&global_trace);
1027 EXPORT_SYMBOL_GPL(tracing_on);
1030 static __always_inline void
1031 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
1033 __this_cpu_write(trace_taskinfo_save, true);
1035 /* If this is the temp buffer, we need to commit fully */
1036 if (this_cpu_read(trace_buffered_event) == event) {
1037 /* Length is in event->array[0] */
1038 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1039 /* Release the temp buffer */
1040 this_cpu_dec(trace_buffered_event_cnt);
1041 /* ring_buffer_unlock_commit() enables preemption */
1042 preempt_enable_notrace();
1044 ring_buffer_unlock_commit(buffer);
1047 int __trace_array_puts(struct trace_array *tr, unsigned long ip,
1048 const char *str, int size)
1050 struct ring_buffer_event *event;
1051 struct trace_buffer *buffer;
1052 struct print_entry *entry;
1053 unsigned int trace_ctx;
1056 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1059 if (unlikely(tracing_selftest_running && tr == &global_trace))
1062 if (unlikely(tracing_disabled))
1065 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1067 trace_ctx = tracing_gen_ctx();
1068 buffer = tr->array_buffer.buffer;
1069 ring_buffer_nest_start(buffer);
1070 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1077 entry = ring_buffer_event_data(event);
1080 memcpy(&entry->buf, str, size);
1082 /* Add a newline if necessary */
1083 if (entry->buf[size - 1] != '\n') {
1084 entry->buf[size] = '\n';
1085 entry->buf[size + 1] = '\0';
1087 entry->buf[size] = '\0';
1089 __buffer_unlock_commit(buffer, event);
1090 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1092 ring_buffer_nest_end(buffer);
1095 EXPORT_SYMBOL_GPL(__trace_array_puts);
1098 * __trace_puts - write a constant string into the trace buffer.
1099 * @ip: The address of the caller
1100 * @str: The constant string to write
1101 * @size: The size of the string.
1103 int __trace_puts(unsigned long ip, const char *str, int size)
1105 return __trace_array_puts(&global_trace, ip, str, size);
1107 EXPORT_SYMBOL_GPL(__trace_puts);
1110 * __trace_bputs - write the pointer to a constant string into trace buffer
1111 * @ip: The address of the caller
1112 * @str: The constant string to write to the buffer to
1114 int __trace_bputs(unsigned long ip, const char *str)
1116 struct ring_buffer_event *event;
1117 struct trace_buffer *buffer;
1118 struct bputs_entry *entry;
1119 unsigned int trace_ctx;
1120 int size = sizeof(struct bputs_entry);
1123 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1126 if (unlikely(tracing_selftest_running || tracing_disabled))
1129 trace_ctx = tracing_gen_ctx();
1130 buffer = global_trace.array_buffer.buffer;
1132 ring_buffer_nest_start(buffer);
1133 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1138 entry = ring_buffer_event_data(event);
1142 __buffer_unlock_commit(buffer, event);
1143 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1147 ring_buffer_nest_end(buffer);
1150 EXPORT_SYMBOL_GPL(__trace_bputs);
1152 #ifdef CONFIG_TRACER_SNAPSHOT
1153 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1156 struct tracer *tracer = tr->current_trace;
1157 unsigned long flags;
1160 trace_array_puts(tr, "*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1161 trace_array_puts(tr, "*** snapshot is being ignored ***\n");
1165 if (!tr->allocated_snapshot) {
1166 trace_array_puts(tr, "*** SNAPSHOT NOT ALLOCATED ***\n");
1167 trace_array_puts(tr, "*** stopping trace here! ***\n");
1168 tracer_tracing_off(tr);
1172 /* Note, snapshot can not be used when the tracer uses it */
1173 if (tracer->use_max_tr) {
1174 trace_array_puts(tr, "*** LATENCY TRACER ACTIVE ***\n");
1175 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1179 local_irq_save(flags);
1180 update_max_tr(tr, current, smp_processor_id(), cond_data);
1181 local_irq_restore(flags);
1184 void tracing_snapshot_instance(struct trace_array *tr)
1186 tracing_snapshot_instance_cond(tr, NULL);
1190 * tracing_snapshot - take a snapshot of the current buffer.
1192 * This causes a swap between the snapshot buffer and the current live
1193 * tracing buffer. You can use this to take snapshots of the live
1194 * trace when some condition is triggered, but continue to trace.
1196 * Note, make sure to allocate the snapshot with either
1197 * a tracing_snapshot_alloc(), or by doing it manually
1198 * with: echo 1 > /sys/kernel/tracing/snapshot
1200 * If the snapshot buffer is not allocated, it will stop tracing.
1201 * Basically making a permanent snapshot.
1203 void tracing_snapshot(void)
1205 struct trace_array *tr = &global_trace;
1207 tracing_snapshot_instance(tr);
1209 EXPORT_SYMBOL_GPL(tracing_snapshot);
1212 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1213 * @tr: The tracing instance to snapshot
1214 * @cond_data: The data to be tested conditionally, and possibly saved
1216 * This is the same as tracing_snapshot() except that the snapshot is
1217 * conditional - the snapshot will only happen if the
1218 * cond_snapshot.update() implementation receiving the cond_data
1219 * returns true, which means that the trace array's cond_snapshot
1220 * update() operation used the cond_data to determine whether the
1221 * snapshot should be taken, and if it was, presumably saved it along
1222 * with the snapshot.
1224 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1226 tracing_snapshot_instance_cond(tr, cond_data);
1228 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1231 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1232 * @tr: The tracing instance
1234 * When the user enables a conditional snapshot using
1235 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1236 * with the snapshot. This accessor is used to retrieve it.
1238 * Should not be called from cond_snapshot.update(), since it takes
1239 * the tr->max_lock lock, which the code calling
1240 * cond_snapshot.update() has already done.
1242 * Returns the cond_data associated with the trace array's snapshot.
1244 void *tracing_cond_snapshot_data(struct trace_array *tr)
1246 void *cond_data = NULL;
1248 local_irq_disable();
1249 arch_spin_lock(&tr->max_lock);
1251 if (tr->cond_snapshot)
1252 cond_data = tr->cond_snapshot->cond_data;
1254 arch_spin_unlock(&tr->max_lock);
1259 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1261 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1262 struct array_buffer *size_buf, int cpu_id);
1263 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1265 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1270 if (!tr->allocated_snapshot) {
1272 /* Make the snapshot buffer have the same order as main buffer */
1273 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
1274 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
1278 /* allocate spare buffer */
1279 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1280 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1284 tr->allocated_snapshot = true;
1290 static void free_snapshot(struct trace_array *tr)
1293 * We don't free the ring buffer. instead, resize it because
1294 * The max_tr ring buffer has some state (e.g. ring->clock) and
1295 * we want preserve it.
1297 ring_buffer_subbuf_order_set(tr->max_buffer.buffer, 0);
1298 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1299 set_buffer_entries(&tr->max_buffer, 1);
1300 tracing_reset_online_cpus(&tr->max_buffer);
1301 tr->allocated_snapshot = false;
1305 * tracing_alloc_snapshot - allocate snapshot buffer.
1307 * This only allocates the snapshot buffer if it isn't already
1308 * allocated - it doesn't also take a snapshot.
1310 * This is meant to be used in cases where the snapshot buffer needs
1311 * to be set up for events that can't sleep but need to be able to
1312 * trigger a snapshot.
1314 int tracing_alloc_snapshot(void)
1316 struct trace_array *tr = &global_trace;
1319 ret = tracing_alloc_snapshot_instance(tr);
1324 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1327 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1329 * This is similar to tracing_snapshot(), but it will allocate the
1330 * snapshot buffer if it isn't already allocated. Use this only
1331 * where it is safe to sleep, as the allocation may sleep.
1333 * This causes a swap between the snapshot buffer and the current live
1334 * tracing buffer. You can use this to take snapshots of the live
1335 * trace when some condition is triggered, but continue to trace.
1337 void tracing_snapshot_alloc(void)
1341 ret = tracing_alloc_snapshot();
1347 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1350 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1351 * @tr: The tracing instance
1352 * @cond_data: User data to associate with the snapshot
1353 * @update: Implementation of the cond_snapshot update function
1355 * Check whether the conditional snapshot for the given instance has
1356 * already been enabled, or if the current tracer is already using a
1357 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1358 * save the cond_data and update function inside.
1360 * Returns 0 if successful, error otherwise.
1362 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1363 cond_update_fn_t update)
1365 struct cond_snapshot *cond_snapshot;
1368 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1372 cond_snapshot->cond_data = cond_data;
1373 cond_snapshot->update = update;
1375 mutex_lock(&trace_types_lock);
1377 ret = tracing_alloc_snapshot_instance(tr);
1381 if (tr->current_trace->use_max_tr) {
1387 * The cond_snapshot can only change to NULL without the
1388 * trace_types_lock. We don't care if we race with it going
1389 * to NULL, but we want to make sure that it's not set to
1390 * something other than NULL when we get here, which we can
1391 * do safely with only holding the trace_types_lock and not
1392 * having to take the max_lock.
1394 if (tr->cond_snapshot) {
1399 local_irq_disable();
1400 arch_spin_lock(&tr->max_lock);
1401 tr->cond_snapshot = cond_snapshot;
1402 arch_spin_unlock(&tr->max_lock);
1405 mutex_unlock(&trace_types_lock);
1410 mutex_unlock(&trace_types_lock);
1411 kfree(cond_snapshot);
1414 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1417 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1418 * @tr: The tracing instance
1420 * Check whether the conditional snapshot for the given instance is
1421 * enabled; if so, free the cond_snapshot associated with it,
1422 * otherwise return -EINVAL.
1424 * Returns 0 if successful, error otherwise.
1426 int tracing_snapshot_cond_disable(struct trace_array *tr)
1430 local_irq_disable();
1431 arch_spin_lock(&tr->max_lock);
1433 if (!tr->cond_snapshot)
1436 kfree(tr->cond_snapshot);
1437 tr->cond_snapshot = NULL;
1440 arch_spin_unlock(&tr->max_lock);
1445 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1447 void tracing_snapshot(void)
1449 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1451 EXPORT_SYMBOL_GPL(tracing_snapshot);
1452 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1454 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1456 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1457 int tracing_alloc_snapshot(void)
1459 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1462 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1463 void tracing_snapshot_alloc(void)
1468 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1469 void *tracing_cond_snapshot_data(struct trace_array *tr)
1473 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1474 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1478 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1479 int tracing_snapshot_cond_disable(struct trace_array *tr)
1483 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1484 #define free_snapshot(tr) do { } while (0)
1485 #endif /* CONFIG_TRACER_SNAPSHOT */
1487 void tracer_tracing_off(struct trace_array *tr)
1489 if (tr->array_buffer.buffer)
1490 ring_buffer_record_off(tr->array_buffer.buffer);
1492 * This flag is looked at when buffers haven't been allocated
1493 * yet, or by some tracers (like irqsoff), that just want to
1494 * know if the ring buffer has been disabled, but it can handle
1495 * races of where it gets disabled but we still do a record.
1496 * As the check is in the fast path of the tracers, it is more
1497 * important to be fast than accurate.
1499 tr->buffer_disabled = 1;
1500 /* Make the flag seen by readers */
1505 * tracing_off - turn off tracing buffers
1507 * This function stops the tracing buffers from recording data.
1508 * It does not disable any overhead the tracers themselves may
1509 * be causing. This function simply causes all recording to
1510 * the ring buffers to fail.
1512 void tracing_off(void)
1514 tracer_tracing_off(&global_trace);
1516 EXPORT_SYMBOL_GPL(tracing_off);
1518 void disable_trace_on_warning(void)
1520 if (__disable_trace_on_warning) {
1521 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1522 "Disabling tracing due to warning\n");
1528 * tracer_tracing_is_on - show real state of ring buffer enabled
1529 * @tr : the trace array to know if ring buffer is enabled
1531 * Shows real state of the ring buffer if it is enabled or not.
1533 bool tracer_tracing_is_on(struct trace_array *tr)
1535 if (tr->array_buffer.buffer)
1536 return ring_buffer_record_is_set_on(tr->array_buffer.buffer);
1537 return !tr->buffer_disabled;
1541 * tracing_is_on - show state of ring buffers enabled
1543 int tracing_is_on(void)
1545 return tracer_tracing_is_on(&global_trace);
1547 EXPORT_SYMBOL_GPL(tracing_is_on);
1549 static int __init set_buf_size(char *str)
1551 unsigned long buf_size;
1555 buf_size = memparse(str, &str);
1557 * nr_entries can not be zero and the startup
1558 * tests require some buffer space. Therefore
1559 * ensure we have at least 4096 bytes of buffer.
1561 trace_buf_size = max(4096UL, buf_size);
1564 __setup("trace_buf_size=", set_buf_size);
1566 static int __init set_tracing_thresh(char *str)
1568 unsigned long threshold;
1573 ret = kstrtoul(str, 0, &threshold);
1576 tracing_thresh = threshold * 1000;
1579 __setup("tracing_thresh=", set_tracing_thresh);
1581 unsigned long nsecs_to_usecs(unsigned long nsecs)
1583 return nsecs / 1000;
1587 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1588 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1589 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1590 * of strings in the order that the evals (enum) were defined.
1595 /* These must match the bit positions in trace_iterator_flags */
1596 static const char *trace_options[] = {
1604 int in_ns; /* is this clock in nanoseconds? */
1605 } trace_clocks[] = {
1606 { trace_clock_local, "local", 1 },
1607 { trace_clock_global, "global", 1 },
1608 { trace_clock_counter, "counter", 0 },
1609 { trace_clock_jiffies, "uptime", 0 },
1610 { trace_clock, "perf", 1 },
1611 { ktime_get_mono_fast_ns, "mono", 1 },
1612 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1613 { ktime_get_boot_fast_ns, "boot", 1 },
1614 { ktime_get_tai_fast_ns, "tai", 1 },
1618 bool trace_clock_in_ns(struct trace_array *tr)
1620 if (trace_clocks[tr->clock_id].in_ns)
1627 * trace_parser_get_init - gets the buffer for trace parser
1629 int trace_parser_get_init(struct trace_parser *parser, int size)
1631 memset(parser, 0, sizeof(*parser));
1633 parser->buffer = kmalloc(size, GFP_KERNEL);
1634 if (!parser->buffer)
1637 parser->size = size;
1642 * trace_parser_put - frees the buffer for trace parser
1644 void trace_parser_put(struct trace_parser *parser)
1646 kfree(parser->buffer);
1647 parser->buffer = NULL;
1651 * trace_get_user - reads the user input string separated by space
1652 * (matched by isspace(ch))
1654 * For each string found the 'struct trace_parser' is updated,
1655 * and the function returns.
1657 * Returns number of bytes read.
1659 * See kernel/trace/trace.h for 'struct trace_parser' details.
1661 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1662 size_t cnt, loff_t *ppos)
1669 trace_parser_clear(parser);
1671 ret = get_user(ch, ubuf++);
1679 * The parser is not finished with the last write,
1680 * continue reading the user input without skipping spaces.
1682 if (!parser->cont) {
1683 /* skip white space */
1684 while (cnt && isspace(ch)) {
1685 ret = get_user(ch, ubuf++);
1694 /* only spaces were written */
1695 if (isspace(ch) || !ch) {
1702 /* read the non-space input */
1703 while (cnt && !isspace(ch) && ch) {
1704 if (parser->idx < parser->size - 1)
1705 parser->buffer[parser->idx++] = ch;
1710 ret = get_user(ch, ubuf++);
1717 /* We either got finished input or we have to wait for another call. */
1718 if (isspace(ch) || !ch) {
1719 parser->buffer[parser->idx] = 0;
1720 parser->cont = false;
1721 } else if (parser->idx < parser->size - 1) {
1722 parser->cont = true;
1723 parser->buffer[parser->idx++] = ch;
1724 /* Make sure the parsed string always terminates with '\0'. */
1725 parser->buffer[parser->idx] = 0;
1738 /* TODO add a seq_buf_to_buffer() */
1739 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1743 if (trace_seq_used(s) <= s->readpos)
1746 len = trace_seq_used(s) - s->readpos;
1749 memcpy(buf, s->buffer + s->readpos, cnt);
1755 unsigned long __read_mostly tracing_thresh;
1757 #ifdef CONFIG_TRACER_MAX_TRACE
1758 static const struct file_operations tracing_max_lat_fops;
1760 #ifdef LATENCY_FS_NOTIFY
1762 static struct workqueue_struct *fsnotify_wq;
1764 static void latency_fsnotify_workfn(struct work_struct *work)
1766 struct trace_array *tr = container_of(work, struct trace_array,
1768 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1771 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1773 struct trace_array *tr = container_of(iwork, struct trace_array,
1775 queue_work(fsnotify_wq, &tr->fsnotify_work);
1778 static void trace_create_maxlat_file(struct trace_array *tr,
1779 struct dentry *d_tracer)
1781 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1782 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1783 tr->d_max_latency = trace_create_file("tracing_max_latency",
1786 &tracing_max_lat_fops);
1789 __init static int latency_fsnotify_init(void)
1791 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1792 WQ_UNBOUND | WQ_HIGHPRI, 0);
1794 pr_err("Unable to allocate tr_max_lat_wq\n");
1800 late_initcall_sync(latency_fsnotify_init);
1802 void latency_fsnotify(struct trace_array *tr)
1807 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1808 * possible that we are called from __schedule() or do_idle(), which
1809 * could cause a deadlock.
1811 irq_work_queue(&tr->fsnotify_irqwork);
1814 #else /* !LATENCY_FS_NOTIFY */
1816 #define trace_create_maxlat_file(tr, d_tracer) \
1817 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1818 d_tracer, tr, &tracing_max_lat_fops)
1823 * Copy the new maximum trace into the separate maximum-trace
1824 * structure. (this way the maximum trace is permanently saved,
1825 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1828 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1830 struct array_buffer *trace_buf = &tr->array_buffer;
1831 struct array_buffer *max_buf = &tr->max_buffer;
1832 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1833 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1836 max_buf->time_start = data->preempt_timestamp;
1838 max_data->saved_latency = tr->max_latency;
1839 max_data->critical_start = data->critical_start;
1840 max_data->critical_end = data->critical_end;
1842 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1843 max_data->pid = tsk->pid;
1845 * If tsk == current, then use current_uid(), as that does not use
1846 * RCU. The irq tracer can be called out of RCU scope.
1849 max_data->uid = current_uid();
1851 max_data->uid = task_uid(tsk);
1853 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1854 max_data->policy = tsk->policy;
1855 max_data->rt_priority = tsk->rt_priority;
1857 /* record this tasks comm */
1858 tracing_record_cmdline(tsk);
1859 latency_fsnotify(tr);
1863 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1865 * @tsk: the task with the latency
1866 * @cpu: The cpu that initiated the trace.
1867 * @cond_data: User data associated with a conditional snapshot
1869 * Flip the buffers between the @tr and the max_tr and record information
1870 * about which task was the cause of this latency.
1873 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1879 WARN_ON_ONCE(!irqs_disabled());
1881 if (!tr->allocated_snapshot) {
1882 /* Only the nop tracer should hit this when disabling */
1883 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1887 arch_spin_lock(&tr->max_lock);
1889 /* Inherit the recordable setting from array_buffer */
1890 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1891 ring_buffer_record_on(tr->max_buffer.buffer);
1893 ring_buffer_record_off(tr->max_buffer.buffer);
1895 #ifdef CONFIG_TRACER_SNAPSHOT
1896 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1897 arch_spin_unlock(&tr->max_lock);
1901 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1903 __update_max_tr(tr, tsk, cpu);
1905 arch_spin_unlock(&tr->max_lock);
1907 /* Any waiters on the old snapshot buffer need to wake up */
1908 ring_buffer_wake_waiters(tr->array_buffer.buffer, RING_BUFFER_ALL_CPUS);
1912 * update_max_tr_single - only copy one trace over, and reset the rest
1914 * @tsk: task with the latency
1915 * @cpu: the cpu of the buffer to copy.
1917 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1920 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1927 WARN_ON_ONCE(!irqs_disabled());
1928 if (!tr->allocated_snapshot) {
1929 /* Only the nop tracer should hit this when disabling */
1930 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1934 arch_spin_lock(&tr->max_lock);
1936 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1938 if (ret == -EBUSY) {
1940 * We failed to swap the buffer due to a commit taking
1941 * place on this CPU. We fail to record, but we reset
1942 * the max trace buffer (no one writes directly to it)
1943 * and flag that it failed.
1944 * Another reason is resize is in progress.
1946 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1947 "Failed to swap buffers due to commit or resize in progress\n");
1950 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1952 __update_max_tr(tr, tsk, cpu);
1953 arch_spin_unlock(&tr->max_lock);
1956 #endif /* CONFIG_TRACER_MAX_TRACE */
1959 struct trace_iterator *iter;
1963 static bool wait_pipe_cond(void *data)
1965 struct pipe_wait *pwait = data;
1966 struct trace_iterator *iter = pwait->iter;
1968 if (atomic_read_acquire(&iter->wait_index) != pwait->wait_index)
1971 return iter->closed;
1974 static int wait_on_pipe(struct trace_iterator *iter, int full)
1976 struct pipe_wait pwait;
1979 /* Iterators are static, they should be filled or empty */
1980 if (trace_buffer_iter(iter, iter->cpu_file))
1983 pwait.wait_index = atomic_read_acquire(&iter->wait_index);
1986 ret = ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file, full,
1987 wait_pipe_cond, &pwait);
1989 #ifdef CONFIG_TRACER_MAX_TRACE
1991 * Make sure this is still the snapshot buffer, as if a snapshot were
1992 * to happen, this would now be the main buffer.
1995 iter->array_buffer = &iter->tr->max_buffer;
2000 #ifdef CONFIG_FTRACE_STARTUP_TEST
2001 static bool selftests_can_run;
2003 struct trace_selftests {
2004 struct list_head list;
2005 struct tracer *type;
2008 static LIST_HEAD(postponed_selftests);
2010 static int save_selftest(struct tracer *type)
2012 struct trace_selftests *selftest;
2014 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
2018 selftest->type = type;
2019 list_add(&selftest->list, &postponed_selftests);
2023 static int run_tracer_selftest(struct tracer *type)
2025 struct trace_array *tr = &global_trace;
2026 struct tracer *saved_tracer = tr->current_trace;
2029 if (!type->selftest || tracing_selftest_disabled)
2033 * If a tracer registers early in boot up (before scheduling is
2034 * initialized and such), then do not run its selftests yet.
2035 * Instead, run it a little later in the boot process.
2037 if (!selftests_can_run)
2038 return save_selftest(type);
2040 if (!tracing_is_on()) {
2041 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2047 * Run a selftest on this tracer.
2048 * Here we reset the trace buffer, and set the current
2049 * tracer to be this tracer. The tracer can then run some
2050 * internal tracing to verify that everything is in order.
2051 * If we fail, we do not register this tracer.
2053 tracing_reset_online_cpus(&tr->array_buffer);
2055 tr->current_trace = type;
2057 #ifdef CONFIG_TRACER_MAX_TRACE
2058 if (type->use_max_tr) {
2059 /* If we expanded the buffers, make sure the max is expanded too */
2060 if (tr->ring_buffer_expanded)
2061 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2062 RING_BUFFER_ALL_CPUS);
2063 tr->allocated_snapshot = true;
2067 /* the test is responsible for initializing and enabling */
2068 pr_info("Testing tracer %s: ", type->name);
2069 ret = type->selftest(type, tr);
2070 /* the test is responsible for resetting too */
2071 tr->current_trace = saved_tracer;
2073 printk(KERN_CONT "FAILED!\n");
2074 /* Add the warning after printing 'FAILED' */
2078 /* Only reset on passing, to avoid touching corrupted buffers */
2079 tracing_reset_online_cpus(&tr->array_buffer);
2081 #ifdef CONFIG_TRACER_MAX_TRACE
2082 if (type->use_max_tr) {
2083 tr->allocated_snapshot = false;
2085 /* Shrink the max buffer again */
2086 if (tr->ring_buffer_expanded)
2087 ring_buffer_resize(tr->max_buffer.buffer, 1,
2088 RING_BUFFER_ALL_CPUS);
2092 printk(KERN_CONT "PASSED\n");
2096 static int do_run_tracer_selftest(struct tracer *type)
2101 * Tests can take a long time, especially if they are run one after the
2102 * other, as does happen during bootup when all the tracers are
2103 * registered. This could cause the soft lockup watchdog to trigger.
2107 tracing_selftest_running = true;
2108 ret = run_tracer_selftest(type);
2109 tracing_selftest_running = false;
2114 static __init int init_trace_selftests(void)
2116 struct trace_selftests *p, *n;
2117 struct tracer *t, **last;
2120 selftests_can_run = true;
2122 mutex_lock(&trace_types_lock);
2124 if (list_empty(&postponed_selftests))
2127 pr_info("Running postponed tracer tests:\n");
2129 tracing_selftest_running = true;
2130 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2131 /* This loop can take minutes when sanitizers are enabled, so
2132 * lets make sure we allow RCU processing.
2135 ret = run_tracer_selftest(p->type);
2136 /* If the test fails, then warn and remove from available_tracers */
2138 WARN(1, "tracer: %s failed selftest, disabling\n",
2140 last = &trace_types;
2141 for (t = trace_types; t; t = t->next) {
2152 tracing_selftest_running = false;
2155 mutex_unlock(&trace_types_lock);
2159 core_initcall(init_trace_selftests);
2161 static inline int run_tracer_selftest(struct tracer *type)
2165 static inline int do_run_tracer_selftest(struct tracer *type)
2169 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2171 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2173 static void __init apply_trace_boot_options(void);
2176 * register_tracer - register a tracer with the ftrace system.
2177 * @type: the plugin for the tracer
2179 * Register a new plugin tracer.
2181 int __init register_tracer(struct tracer *type)
2187 pr_info("Tracer must have a name\n");
2191 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2192 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2196 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2197 pr_warn("Can not register tracer %s due to lockdown\n",
2202 mutex_lock(&trace_types_lock);
2204 for (t = trace_types; t; t = t->next) {
2205 if (strcmp(type->name, t->name) == 0) {
2207 pr_info("Tracer %s already registered\n",
2214 if (!type->set_flag)
2215 type->set_flag = &dummy_set_flag;
2217 /*allocate a dummy tracer_flags*/
2218 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2223 type->flags->val = 0;
2224 type->flags->opts = dummy_tracer_opt;
2226 if (!type->flags->opts)
2227 type->flags->opts = dummy_tracer_opt;
2229 /* store the tracer for __set_tracer_option */
2230 type->flags->trace = type;
2232 ret = do_run_tracer_selftest(type);
2236 type->next = trace_types;
2238 add_tracer_options(&global_trace, type);
2241 mutex_unlock(&trace_types_lock);
2243 if (ret || !default_bootup_tracer)
2246 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2249 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2250 /* Do we want this tracer to start on bootup? */
2251 tracing_set_tracer(&global_trace, type->name);
2252 default_bootup_tracer = NULL;
2254 apply_trace_boot_options();
2256 /* disable other selftests, since this will break it. */
2257 disable_tracing_selftest("running a tracer");
2263 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2265 struct trace_buffer *buffer = buf->buffer;
2270 ring_buffer_record_disable(buffer);
2272 /* Make sure all commits have finished */
2274 ring_buffer_reset_cpu(buffer, cpu);
2276 ring_buffer_record_enable(buffer);
2279 void tracing_reset_online_cpus(struct array_buffer *buf)
2281 struct trace_buffer *buffer = buf->buffer;
2286 ring_buffer_record_disable(buffer);
2288 /* Make sure all commits have finished */
2291 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2293 ring_buffer_reset_online_cpus(buffer);
2295 ring_buffer_record_enable(buffer);
2298 /* Must have trace_types_lock held */
2299 void tracing_reset_all_online_cpus_unlocked(void)
2301 struct trace_array *tr;
2303 lockdep_assert_held(&trace_types_lock);
2305 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2306 if (!tr->clear_trace)
2308 tr->clear_trace = false;
2309 tracing_reset_online_cpus(&tr->array_buffer);
2310 #ifdef CONFIG_TRACER_MAX_TRACE
2311 tracing_reset_online_cpus(&tr->max_buffer);
2316 void tracing_reset_all_online_cpus(void)
2318 mutex_lock(&trace_types_lock);
2319 tracing_reset_all_online_cpus_unlocked();
2320 mutex_unlock(&trace_types_lock);
2324 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2325 * is the tgid last observed corresponding to pid=i.
2327 static int *tgid_map;
2329 /* The maximum valid index into tgid_map. */
2330 static size_t tgid_map_max;
2332 #define SAVED_CMDLINES_DEFAULT 128
2333 #define NO_CMDLINE_MAP UINT_MAX
2335 * Preemption must be disabled before acquiring trace_cmdline_lock.
2336 * The various trace_arrays' max_lock must be acquired in a context
2337 * where interrupt is disabled.
2339 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2340 struct saved_cmdlines_buffer {
2341 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2342 unsigned *map_cmdline_to_pid;
2343 unsigned cmdline_num;
2345 char saved_cmdlines[];
2347 static struct saved_cmdlines_buffer *savedcmd;
2349 static inline char *get_saved_cmdlines(int idx)
2351 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2354 static inline void set_cmdline(int idx, const char *cmdline)
2356 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2359 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
2361 int order = get_order(sizeof(*s) + s->cmdline_num * TASK_COMM_LEN);
2363 kfree(s->map_cmdline_to_pid);
2365 free_pages((unsigned long)s, order);
2368 static struct saved_cmdlines_buffer *allocate_cmdlines_buffer(unsigned int val)
2370 struct saved_cmdlines_buffer *s;
2372 int orig_size, size;
2375 /* Figure out how much is needed to hold the given number of cmdlines */
2376 orig_size = sizeof(*s) + val * TASK_COMM_LEN;
2377 order = get_order(orig_size);
2378 size = 1 << (order + PAGE_SHIFT);
2379 page = alloc_pages(GFP_KERNEL, order);
2383 s = page_address(page);
2384 kmemleak_alloc(s, size, 1, GFP_KERNEL);
2385 memset(s, 0, sizeof(*s));
2387 /* Round up to actual allocation */
2388 val = (size - sizeof(*s)) / TASK_COMM_LEN;
2389 s->cmdline_num = val;
2391 s->map_cmdline_to_pid = kmalloc_array(val,
2392 sizeof(*s->map_cmdline_to_pid),
2394 if (!s->map_cmdline_to_pid) {
2395 free_saved_cmdlines_buffer(s);
2400 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2401 sizeof(s->map_pid_to_cmdline));
2402 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2403 val * sizeof(*s->map_cmdline_to_pid));
2408 static int trace_create_savedcmd(void)
2410 savedcmd = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT);
2412 return savedcmd ? 0 : -ENOMEM;
2415 int is_tracing_stopped(void)
2417 return global_trace.stop_count;
2420 static void tracing_start_tr(struct trace_array *tr)
2422 struct trace_buffer *buffer;
2423 unsigned long flags;
2425 if (tracing_disabled)
2428 raw_spin_lock_irqsave(&tr->start_lock, flags);
2429 if (--tr->stop_count) {
2430 if (WARN_ON_ONCE(tr->stop_count < 0)) {
2431 /* Someone screwed up their debugging */
2437 /* Prevent the buffers from switching */
2438 arch_spin_lock(&tr->max_lock);
2440 buffer = tr->array_buffer.buffer;
2442 ring_buffer_record_enable(buffer);
2444 #ifdef CONFIG_TRACER_MAX_TRACE
2445 buffer = tr->max_buffer.buffer;
2447 ring_buffer_record_enable(buffer);
2450 arch_spin_unlock(&tr->max_lock);
2453 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2457 * tracing_start - quick start of the tracer
2459 * If tracing is enabled but was stopped by tracing_stop,
2460 * this will start the tracer back up.
2462 void tracing_start(void)
2465 return tracing_start_tr(&global_trace);
2468 static void tracing_stop_tr(struct trace_array *tr)
2470 struct trace_buffer *buffer;
2471 unsigned long flags;
2473 raw_spin_lock_irqsave(&tr->start_lock, flags);
2474 if (tr->stop_count++)
2477 /* Prevent the buffers from switching */
2478 arch_spin_lock(&tr->max_lock);
2480 buffer = tr->array_buffer.buffer;
2482 ring_buffer_record_disable(buffer);
2484 #ifdef CONFIG_TRACER_MAX_TRACE
2485 buffer = tr->max_buffer.buffer;
2487 ring_buffer_record_disable(buffer);
2490 arch_spin_unlock(&tr->max_lock);
2493 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2497 * tracing_stop - quick stop of the tracer
2499 * Light weight way to stop tracing. Use in conjunction with
2502 void tracing_stop(void)
2504 return tracing_stop_tr(&global_trace);
2507 static int trace_save_cmdline(struct task_struct *tsk)
2511 /* treat recording of idle task as a success */
2515 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2518 * It's not the end of the world if we don't get
2519 * the lock, but we also don't want to spin
2520 * nor do we want to disable interrupts,
2521 * so if we miss here, then better luck next time.
2523 * This is called within the scheduler and wake up, so interrupts
2524 * had better been disabled and run queue lock been held.
2526 lockdep_assert_preemption_disabled();
2527 if (!arch_spin_trylock(&trace_cmdline_lock))
2530 idx = savedcmd->map_pid_to_cmdline[tpid];
2531 if (idx == NO_CMDLINE_MAP) {
2532 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2534 savedcmd->map_pid_to_cmdline[tpid] = idx;
2535 savedcmd->cmdline_idx = idx;
2538 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2539 set_cmdline(idx, tsk->comm);
2541 arch_spin_unlock(&trace_cmdline_lock);
2546 static void __trace_find_cmdline(int pid, char comm[])
2552 strcpy(comm, "<idle>");
2556 if (WARN_ON_ONCE(pid < 0)) {
2557 strcpy(comm, "<XXX>");
2561 tpid = pid & (PID_MAX_DEFAULT - 1);
2562 map = savedcmd->map_pid_to_cmdline[tpid];
2563 if (map != NO_CMDLINE_MAP) {
2564 tpid = savedcmd->map_cmdline_to_pid[map];
2566 strscpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2570 strcpy(comm, "<...>");
2573 void trace_find_cmdline(int pid, char comm[])
2576 arch_spin_lock(&trace_cmdline_lock);
2578 __trace_find_cmdline(pid, comm);
2580 arch_spin_unlock(&trace_cmdline_lock);
2584 static int *trace_find_tgid_ptr(int pid)
2587 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2588 * if we observe a non-NULL tgid_map then we also observe the correct
2591 int *map = smp_load_acquire(&tgid_map);
2593 if (unlikely(!map || pid > tgid_map_max))
2599 int trace_find_tgid(int pid)
2601 int *ptr = trace_find_tgid_ptr(pid);
2603 return ptr ? *ptr : 0;
2606 static int trace_save_tgid(struct task_struct *tsk)
2610 /* treat recording of idle task as a success */
2614 ptr = trace_find_tgid_ptr(tsk->pid);
2622 static bool tracing_record_taskinfo_skip(int flags)
2624 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2626 if (!__this_cpu_read(trace_taskinfo_save))
2632 * tracing_record_taskinfo - record the task info of a task
2634 * @task: task to record
2635 * @flags: TRACE_RECORD_CMDLINE for recording comm
2636 * TRACE_RECORD_TGID for recording tgid
2638 void tracing_record_taskinfo(struct task_struct *task, int flags)
2642 if (tracing_record_taskinfo_skip(flags))
2646 * Record as much task information as possible. If some fail, continue
2647 * to try to record the others.
2649 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2650 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2652 /* If recording any information failed, retry again soon. */
2656 __this_cpu_write(trace_taskinfo_save, false);
2660 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2662 * @prev: previous task during sched_switch
2663 * @next: next task during sched_switch
2664 * @flags: TRACE_RECORD_CMDLINE for recording comm
2665 * TRACE_RECORD_TGID for recording tgid
2667 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2668 struct task_struct *next, int flags)
2672 if (tracing_record_taskinfo_skip(flags))
2676 * Record as much task information as possible. If some fail, continue
2677 * to try to record the others.
2679 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2680 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2681 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2682 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2684 /* If recording any information failed, retry again soon. */
2688 __this_cpu_write(trace_taskinfo_save, false);
2691 /* Helpers to record a specific task information */
2692 void tracing_record_cmdline(struct task_struct *task)
2694 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2697 void tracing_record_tgid(struct task_struct *task)
2699 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2703 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2704 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2705 * simplifies those functions and keeps them in sync.
2707 enum print_line_t trace_handle_return(struct trace_seq *s)
2709 return trace_seq_has_overflowed(s) ?
2710 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2712 EXPORT_SYMBOL_GPL(trace_handle_return);
2714 static unsigned short migration_disable_value(void)
2716 #if defined(CONFIG_SMP)
2717 return current->migration_disabled;
2723 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2725 unsigned int trace_flags = irqs_status;
2728 pc = preempt_count();
2731 trace_flags |= TRACE_FLAG_NMI;
2732 if (pc & HARDIRQ_MASK)
2733 trace_flags |= TRACE_FLAG_HARDIRQ;
2734 if (in_serving_softirq())
2735 trace_flags |= TRACE_FLAG_SOFTIRQ;
2736 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2737 trace_flags |= TRACE_FLAG_BH_OFF;
2739 if (tif_need_resched())
2740 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2741 if (test_preempt_need_resched())
2742 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2743 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2744 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2747 struct ring_buffer_event *
2748 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2751 unsigned int trace_ctx)
2753 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2756 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2757 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2758 static int trace_buffered_event_ref;
2761 * trace_buffered_event_enable - enable buffering events
2763 * When events are being filtered, it is quicker to use a temporary
2764 * buffer to write the event data into if there's a likely chance
2765 * that it will not be committed. The discard of the ring buffer
2766 * is not as fast as committing, and is much slower than copying
2769 * When an event is to be filtered, allocate per cpu buffers to
2770 * write the event data into, and if the event is filtered and discarded
2771 * it is simply dropped, otherwise, the entire data is to be committed
2774 void trace_buffered_event_enable(void)
2776 struct ring_buffer_event *event;
2780 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2782 if (trace_buffered_event_ref++)
2785 for_each_tracing_cpu(cpu) {
2786 page = alloc_pages_node(cpu_to_node(cpu),
2787 GFP_KERNEL | __GFP_NORETRY, 0);
2788 /* This is just an optimization and can handle failures */
2790 pr_err("Failed to allocate event buffer\n");
2794 event = page_address(page);
2795 memset(event, 0, sizeof(*event));
2797 per_cpu(trace_buffered_event, cpu) = event;
2800 if (cpu == smp_processor_id() &&
2801 __this_cpu_read(trace_buffered_event) !=
2802 per_cpu(trace_buffered_event, cpu))
2808 static void enable_trace_buffered_event(void *data)
2810 /* Probably not needed, but do it anyway */
2812 this_cpu_dec(trace_buffered_event_cnt);
2815 static void disable_trace_buffered_event(void *data)
2817 this_cpu_inc(trace_buffered_event_cnt);
2821 * trace_buffered_event_disable - disable buffering events
2823 * When a filter is removed, it is faster to not use the buffered
2824 * events, and to commit directly into the ring buffer. Free up
2825 * the temp buffers when there are no more users. This requires
2826 * special synchronization with current events.
2828 void trace_buffered_event_disable(void)
2832 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2834 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2837 if (--trace_buffered_event_ref)
2840 /* For each CPU, set the buffer as used. */
2841 on_each_cpu_mask(tracing_buffer_mask, disable_trace_buffered_event,
2844 /* Wait for all current users to finish */
2847 for_each_tracing_cpu(cpu) {
2848 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2849 per_cpu(trace_buffered_event, cpu) = NULL;
2853 * Wait for all CPUs that potentially started checking if they can use
2854 * their event buffer only after the previous synchronize_rcu() call and
2855 * they still read a valid pointer from trace_buffered_event. It must be
2856 * ensured they don't see cleared trace_buffered_event_cnt else they
2857 * could wrongly decide to use the pointed-to buffer which is now freed.
2861 /* For each CPU, relinquish the buffer */
2862 on_each_cpu_mask(tracing_buffer_mask, enable_trace_buffered_event, NULL,
2866 static struct trace_buffer *temp_buffer;
2868 struct ring_buffer_event *
2869 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2870 struct trace_event_file *trace_file,
2871 int type, unsigned long len,
2872 unsigned int trace_ctx)
2874 struct ring_buffer_event *entry;
2875 struct trace_array *tr = trace_file->tr;
2878 *current_rb = tr->array_buffer.buffer;
2880 if (!tr->no_filter_buffering_ref &&
2881 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2882 preempt_disable_notrace();
2884 * Filtering is on, so try to use the per cpu buffer first.
2885 * This buffer will simulate a ring_buffer_event,
2886 * where the type_len is zero and the array[0] will
2887 * hold the full length.
2888 * (see include/linux/ring-buffer.h for details on
2889 * how the ring_buffer_event is structured).
2891 * Using a temp buffer during filtering and copying it
2892 * on a matched filter is quicker than writing directly
2893 * into the ring buffer and then discarding it when
2894 * it doesn't match. That is because the discard
2895 * requires several atomic operations to get right.
2896 * Copying on match and doing nothing on a failed match
2897 * is still quicker than no copy on match, but having
2898 * to discard out of the ring buffer on a failed match.
2900 if ((entry = __this_cpu_read(trace_buffered_event))) {
2901 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2903 val = this_cpu_inc_return(trace_buffered_event_cnt);
2906 * Preemption is disabled, but interrupts and NMIs
2907 * can still come in now. If that happens after
2908 * the above increment, then it will have to go
2909 * back to the old method of allocating the event
2910 * on the ring buffer, and if the filter fails, it
2911 * will have to call ring_buffer_discard_commit()
2914 * Need to also check the unlikely case that the
2915 * length is bigger than the temp buffer size.
2916 * If that happens, then the reserve is pretty much
2917 * guaranteed to fail, as the ring buffer currently
2918 * only allows events less than a page. But that may
2919 * change in the future, so let the ring buffer reserve
2920 * handle the failure in that case.
2922 if (val == 1 && likely(len <= max_len)) {
2923 trace_event_setup(entry, type, trace_ctx);
2924 entry->array[0] = len;
2925 /* Return with preemption disabled */
2928 this_cpu_dec(trace_buffered_event_cnt);
2930 /* __trace_buffer_lock_reserve() disables preemption */
2931 preempt_enable_notrace();
2934 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2937 * If tracing is off, but we have triggers enabled
2938 * we still need to look at the event data. Use the temp_buffer
2939 * to store the trace event for the trigger to use. It's recursive
2940 * safe and will not be recorded anywhere.
2942 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2943 *current_rb = temp_buffer;
2944 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2949 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2951 static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2952 static DEFINE_MUTEX(tracepoint_printk_mutex);
2954 static void output_printk(struct trace_event_buffer *fbuffer)
2956 struct trace_event_call *event_call;
2957 struct trace_event_file *file;
2958 struct trace_event *event;
2959 unsigned long flags;
2960 struct trace_iterator *iter = tracepoint_print_iter;
2962 /* We should never get here if iter is NULL */
2963 if (WARN_ON_ONCE(!iter))
2966 event_call = fbuffer->trace_file->event_call;
2967 if (!event_call || !event_call->event.funcs ||
2968 !event_call->event.funcs->trace)
2971 file = fbuffer->trace_file;
2972 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2973 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2974 !filter_match_preds(file->filter, fbuffer->entry)))
2977 event = &fbuffer->trace_file->event_call->event;
2979 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2980 trace_seq_init(&iter->seq);
2981 iter->ent = fbuffer->entry;
2982 event_call->event.funcs->trace(iter, 0, event);
2983 trace_seq_putc(&iter->seq, 0);
2984 printk("%s", iter->seq.buffer);
2986 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2989 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2990 void *buffer, size_t *lenp,
2993 int save_tracepoint_printk;
2996 mutex_lock(&tracepoint_printk_mutex);
2997 save_tracepoint_printk = tracepoint_printk;
2999 ret = proc_dointvec(table, write, buffer, lenp, ppos);
3002 * This will force exiting early, as tracepoint_printk
3003 * is always zero when tracepoint_printk_iter is not allocated
3005 if (!tracepoint_print_iter)
3006 tracepoint_printk = 0;
3008 if (save_tracepoint_printk == tracepoint_printk)
3011 if (tracepoint_printk)
3012 static_key_enable(&tracepoint_printk_key.key);
3014 static_key_disable(&tracepoint_printk_key.key);
3017 mutex_unlock(&tracepoint_printk_mutex);
3022 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
3024 enum event_trigger_type tt = ETT_NONE;
3025 struct trace_event_file *file = fbuffer->trace_file;
3027 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
3028 fbuffer->entry, &tt))
3031 if (static_key_false(&tracepoint_printk_key.key))
3032 output_printk(fbuffer);
3034 if (static_branch_unlikely(&trace_event_exports_enabled))
3035 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
3037 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
3038 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
3042 event_triggers_post_call(file, tt);
3045 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
3050 * trace_buffer_unlock_commit_regs()
3051 * trace_event_buffer_commit()
3052 * trace_event_raw_event_xxx()
3054 # define STACK_SKIP 3
3056 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
3057 struct trace_buffer *buffer,
3058 struct ring_buffer_event *event,
3059 unsigned int trace_ctx,
3060 struct pt_regs *regs)
3062 __buffer_unlock_commit(buffer, event);
3065 * If regs is not set, then skip the necessary functions.
3066 * Note, we can still get here via blktrace, wakeup tracer
3067 * and mmiotrace, but that's ok if they lose a function or
3068 * two. They are not that meaningful.
3070 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
3071 ftrace_trace_userstack(tr, buffer, trace_ctx);
3075 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
3078 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
3079 struct ring_buffer_event *event)
3081 __buffer_unlock_commit(buffer, event);
3085 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
3086 parent_ip, unsigned int trace_ctx)
3088 struct trace_event_call *call = &event_function;
3089 struct trace_buffer *buffer = tr->array_buffer.buffer;
3090 struct ring_buffer_event *event;
3091 struct ftrace_entry *entry;
3093 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
3097 entry = ring_buffer_event_data(event);
3099 entry->parent_ip = parent_ip;
3101 if (!call_filter_check_discard(call, entry, buffer, event)) {
3102 if (static_branch_unlikely(&trace_function_exports_enabled))
3103 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
3104 __buffer_unlock_commit(buffer, event);
3108 #ifdef CONFIG_STACKTRACE
3110 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
3111 #define FTRACE_KSTACK_NESTING 4
3113 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3115 struct ftrace_stack {
3116 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3120 struct ftrace_stacks {
3121 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3124 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3125 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3127 static void __ftrace_trace_stack(struct trace_buffer *buffer,
3128 unsigned int trace_ctx,
3129 int skip, struct pt_regs *regs)
3131 struct trace_event_call *call = &event_kernel_stack;
3132 struct ring_buffer_event *event;
3133 unsigned int size, nr_entries;
3134 struct ftrace_stack *fstack;
3135 struct stack_entry *entry;
3139 * Add one, for this function and the call to save_stack_trace()
3140 * If regs is set, then these functions will not be in the way.
3142 #ifndef CONFIG_UNWINDER_ORC
3147 preempt_disable_notrace();
3149 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3151 /* This should never happen. If it does, yell once and skip */
3152 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3156 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3157 * interrupt will either see the value pre increment or post
3158 * increment. If the interrupt happens pre increment it will have
3159 * restored the counter when it returns. We just need a barrier to
3160 * keep gcc from moving things around.
3164 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3165 size = ARRAY_SIZE(fstack->calls);
3168 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3171 nr_entries = stack_trace_save(fstack->calls, size, skip);
3174 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3175 struct_size(entry, caller, nr_entries),
3179 entry = ring_buffer_event_data(event);
3181 entry->size = nr_entries;
3182 memcpy(&entry->caller, fstack->calls,
3183 flex_array_size(entry, caller, nr_entries));
3185 if (!call_filter_check_discard(call, entry, buffer, event))
3186 __buffer_unlock_commit(buffer, event);
3189 /* Again, don't let gcc optimize things here */
3191 __this_cpu_dec(ftrace_stack_reserve);
3192 preempt_enable_notrace();
3196 static inline void ftrace_trace_stack(struct trace_array *tr,
3197 struct trace_buffer *buffer,
3198 unsigned int trace_ctx,
3199 int skip, struct pt_regs *regs)
3201 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3204 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3207 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3210 struct trace_buffer *buffer = tr->array_buffer.buffer;
3212 if (rcu_is_watching()) {
3213 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3217 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3221 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3222 * but if the above rcu_is_watching() failed, then the NMI
3223 * triggered someplace critical, and ct_irq_enter() should
3224 * not be called from NMI.
3226 if (unlikely(in_nmi()))
3229 ct_irq_enter_irqson();
3230 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3231 ct_irq_exit_irqson();
3235 * trace_dump_stack - record a stack back trace in the trace buffer
3236 * @skip: Number of functions to skip (helper handlers)
3238 void trace_dump_stack(int skip)
3240 if (tracing_disabled || tracing_selftest_running)
3243 #ifndef CONFIG_UNWINDER_ORC
3244 /* Skip 1 to skip this function. */
3247 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3248 tracing_gen_ctx(), skip, NULL);
3250 EXPORT_SYMBOL_GPL(trace_dump_stack);
3252 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3253 static DEFINE_PER_CPU(int, user_stack_count);
3256 ftrace_trace_userstack(struct trace_array *tr,
3257 struct trace_buffer *buffer, unsigned int trace_ctx)
3259 struct trace_event_call *call = &event_user_stack;
3260 struct ring_buffer_event *event;
3261 struct userstack_entry *entry;
3263 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3267 * NMIs can not handle page faults, even with fix ups.
3268 * The save user stack can (and often does) fault.
3270 if (unlikely(in_nmi()))
3274 * prevent recursion, since the user stack tracing may
3275 * trigger other kernel events.
3278 if (__this_cpu_read(user_stack_count))
3281 __this_cpu_inc(user_stack_count);
3283 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3284 sizeof(*entry), trace_ctx);
3286 goto out_drop_count;
3287 entry = ring_buffer_event_data(event);
3289 entry->tgid = current->tgid;
3290 memset(&entry->caller, 0, sizeof(entry->caller));
3292 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3293 if (!call_filter_check_discard(call, entry, buffer, event))
3294 __buffer_unlock_commit(buffer, event);
3297 __this_cpu_dec(user_stack_count);
3301 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3302 static void ftrace_trace_userstack(struct trace_array *tr,
3303 struct trace_buffer *buffer,
3304 unsigned int trace_ctx)
3307 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3309 #endif /* CONFIG_STACKTRACE */
3312 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3313 unsigned long long delta)
3315 entry->bottom_delta_ts = delta & U32_MAX;
3316 entry->top_delta_ts = (delta >> 32);
3319 void trace_last_func_repeats(struct trace_array *tr,
3320 struct trace_func_repeats *last_info,
3321 unsigned int trace_ctx)
3323 struct trace_buffer *buffer = tr->array_buffer.buffer;
3324 struct func_repeats_entry *entry;
3325 struct ring_buffer_event *event;
3328 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3329 sizeof(*entry), trace_ctx);
3333 delta = ring_buffer_event_time_stamp(buffer, event) -
3334 last_info->ts_last_call;
3336 entry = ring_buffer_event_data(event);
3337 entry->ip = last_info->ip;
3338 entry->parent_ip = last_info->parent_ip;
3339 entry->count = last_info->count;
3340 func_repeats_set_delta_ts(entry, delta);
3342 __buffer_unlock_commit(buffer, event);
3345 /* created for use with alloc_percpu */
3346 struct trace_buffer_struct {
3348 char buffer[4][TRACE_BUF_SIZE];
3351 static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3354 * This allows for lockless recording. If we're nested too deeply, then
3355 * this returns NULL.
3357 static char *get_trace_buf(void)
3359 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3361 if (!trace_percpu_buffer || buffer->nesting >= 4)
3366 /* Interrupts must see nesting incremented before we use the buffer */
3368 return &buffer->buffer[buffer->nesting - 1][0];
3371 static void put_trace_buf(void)
3373 /* Don't let the decrement of nesting leak before this */
3375 this_cpu_dec(trace_percpu_buffer->nesting);
3378 static int alloc_percpu_trace_buffer(void)
3380 struct trace_buffer_struct __percpu *buffers;
3382 if (trace_percpu_buffer)
3385 buffers = alloc_percpu(struct trace_buffer_struct);
3386 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3389 trace_percpu_buffer = buffers;
3393 static int buffers_allocated;
3395 void trace_printk_init_buffers(void)
3397 if (buffers_allocated)
3400 if (alloc_percpu_trace_buffer())
3403 /* trace_printk() is for debug use only. Don't use it in production. */
3406 pr_warn("**********************************************************\n");
3407 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3409 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3411 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3412 pr_warn("** unsafe for production use. **\n");
3414 pr_warn("** If you see this message and you are not debugging **\n");
3415 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3417 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3418 pr_warn("**********************************************************\n");
3420 /* Expand the buffers to set size */
3421 tracing_update_buffers(&global_trace);
3423 buffers_allocated = 1;
3426 * trace_printk_init_buffers() can be called by modules.
3427 * If that happens, then we need to start cmdline recording
3428 * directly here. If the global_trace.buffer is already
3429 * allocated here, then this was called by module code.
3431 if (global_trace.array_buffer.buffer)
3432 tracing_start_cmdline_record();
3434 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3436 void trace_printk_start_comm(void)
3438 /* Start tracing comms if trace printk is set */
3439 if (!buffers_allocated)
3441 tracing_start_cmdline_record();
3444 static void trace_printk_start_stop_comm(int enabled)
3446 if (!buffers_allocated)
3450 tracing_start_cmdline_record();
3452 tracing_stop_cmdline_record();
3456 * trace_vbprintk - write binary msg to tracing buffer
3457 * @ip: The address of the caller
3458 * @fmt: The string format to write to the buffer
3459 * @args: Arguments for @fmt
3461 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3463 struct trace_event_call *call = &event_bprint;
3464 struct ring_buffer_event *event;
3465 struct trace_buffer *buffer;
3466 struct trace_array *tr = &global_trace;
3467 struct bprint_entry *entry;
3468 unsigned int trace_ctx;
3472 if (unlikely(tracing_selftest_running || tracing_disabled))
3475 /* Don't pollute graph traces with trace_vprintk internals */
3476 pause_graph_tracing();
3478 trace_ctx = tracing_gen_ctx();
3479 preempt_disable_notrace();
3481 tbuffer = get_trace_buf();
3487 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3489 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3492 size = sizeof(*entry) + sizeof(u32) * len;
3493 buffer = tr->array_buffer.buffer;
3494 ring_buffer_nest_start(buffer);
3495 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3499 entry = ring_buffer_event_data(event);
3503 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3504 if (!call_filter_check_discard(call, entry, buffer, event)) {
3505 __buffer_unlock_commit(buffer, event);
3506 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3510 ring_buffer_nest_end(buffer);
3515 preempt_enable_notrace();
3516 unpause_graph_tracing();
3520 EXPORT_SYMBOL_GPL(trace_vbprintk);
3524 __trace_array_vprintk(struct trace_buffer *buffer,
3525 unsigned long ip, const char *fmt, va_list args)
3527 struct trace_event_call *call = &event_print;
3528 struct ring_buffer_event *event;
3530 struct print_entry *entry;
3531 unsigned int trace_ctx;
3534 if (tracing_disabled)
3537 /* Don't pollute graph traces with trace_vprintk internals */
3538 pause_graph_tracing();
3540 trace_ctx = tracing_gen_ctx();
3541 preempt_disable_notrace();
3544 tbuffer = get_trace_buf();
3550 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3552 size = sizeof(*entry) + len + 1;
3553 ring_buffer_nest_start(buffer);
3554 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3558 entry = ring_buffer_event_data(event);
3561 memcpy(&entry->buf, tbuffer, len + 1);
3562 if (!call_filter_check_discard(call, entry, buffer, event)) {
3563 __buffer_unlock_commit(buffer, event);
3564 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3568 ring_buffer_nest_end(buffer);
3572 preempt_enable_notrace();
3573 unpause_graph_tracing();
3579 int trace_array_vprintk(struct trace_array *tr,
3580 unsigned long ip, const char *fmt, va_list args)
3582 if (tracing_selftest_running && tr == &global_trace)
3585 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3589 * trace_array_printk - Print a message to a specific instance
3590 * @tr: The instance trace_array descriptor
3591 * @ip: The instruction pointer that this is called from.
3592 * @fmt: The format to print (printf format)
3594 * If a subsystem sets up its own instance, they have the right to
3595 * printk strings into their tracing instance buffer using this
3596 * function. Note, this function will not write into the top level
3597 * buffer (use trace_printk() for that), as writing into the top level
3598 * buffer should only have events that can be individually disabled.
3599 * trace_printk() is only used for debugging a kernel, and should not
3600 * be ever incorporated in normal use.
3602 * trace_array_printk() can be used, as it will not add noise to the
3603 * top level tracing buffer.
3605 * Note, trace_array_init_printk() must be called on @tr before this
3609 int trace_array_printk(struct trace_array *tr,
3610 unsigned long ip, const char *fmt, ...)
3618 /* This is only allowed for created instances */
3619 if (tr == &global_trace)
3622 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3626 ret = trace_array_vprintk(tr, ip, fmt, ap);
3630 EXPORT_SYMBOL_GPL(trace_array_printk);
3633 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3634 * @tr: The trace array to initialize the buffers for
3636 * As trace_array_printk() only writes into instances, they are OK to
3637 * have in the kernel (unlike trace_printk()). This needs to be called
3638 * before trace_array_printk() can be used on a trace_array.
3640 int trace_array_init_printk(struct trace_array *tr)
3645 /* This is only allowed for created instances */
3646 if (tr == &global_trace)
3649 return alloc_percpu_trace_buffer();
3651 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3654 int trace_array_printk_buf(struct trace_buffer *buffer,
3655 unsigned long ip, const char *fmt, ...)
3660 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3664 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3670 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3672 return trace_array_vprintk(&global_trace, ip, fmt, args);
3674 EXPORT_SYMBOL_GPL(trace_vprintk);
3676 static void trace_iterator_increment(struct trace_iterator *iter)
3678 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3682 ring_buffer_iter_advance(buf_iter);
3685 static struct trace_entry *
3686 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3687 unsigned long *lost_events)
3689 struct ring_buffer_event *event;
3690 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3693 event = ring_buffer_iter_peek(buf_iter, ts);
3695 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3696 (unsigned long)-1 : 0;
3698 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3703 iter->ent_size = ring_buffer_event_length(event);
3704 return ring_buffer_event_data(event);
3710 static struct trace_entry *
3711 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3712 unsigned long *missing_events, u64 *ent_ts)
3714 struct trace_buffer *buffer = iter->array_buffer->buffer;
3715 struct trace_entry *ent, *next = NULL;
3716 unsigned long lost_events = 0, next_lost = 0;
3717 int cpu_file = iter->cpu_file;
3718 u64 next_ts = 0, ts;
3724 * If we are in a per_cpu trace file, don't bother by iterating over
3725 * all cpu and peek directly.
3727 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3728 if (ring_buffer_empty_cpu(buffer, cpu_file))
3730 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3732 *ent_cpu = cpu_file;
3737 for_each_tracing_cpu(cpu) {
3739 if (ring_buffer_empty_cpu(buffer, cpu))
3742 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3745 * Pick the entry with the smallest timestamp:
3747 if (ent && (!next || ts < next_ts)) {
3751 next_lost = lost_events;
3752 next_size = iter->ent_size;
3756 iter->ent_size = next_size;
3759 *ent_cpu = next_cpu;
3765 *missing_events = next_lost;
3770 #define STATIC_FMT_BUF_SIZE 128
3771 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3773 char *trace_iter_expand_format(struct trace_iterator *iter)
3778 * iter->tr is NULL when used with tp_printk, which makes
3779 * this get called where it is not safe to call krealloc().
3781 if (!iter->tr || iter->fmt == static_fmt_buf)
3784 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3787 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3794 /* Returns true if the string is safe to dereference from an event */
3795 static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3798 unsigned long addr = (unsigned long)str;
3799 struct trace_event *trace_event;
3800 struct trace_event_call *event;
3802 /* Ignore strings with no length */
3806 /* OK if part of the event data */
3807 if ((addr >= (unsigned long)iter->ent) &&
3808 (addr < (unsigned long)iter->ent + iter->ent_size))
3811 /* OK if part of the temp seq buffer */
3812 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3813 (addr < (unsigned long)iter->tmp_seq.buffer + TRACE_SEQ_BUFFER_SIZE))
3816 /* Core rodata can not be freed */
3817 if (is_kernel_rodata(addr))
3820 if (trace_is_tracepoint_string(str))
3824 * Now this could be a module event, referencing core module
3825 * data, which is OK.
3830 trace_event = ftrace_find_event(iter->ent->type);
3834 event = container_of(trace_event, struct trace_event_call, event);
3835 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3838 /* Would rather have rodata, but this will suffice */
3839 if (within_module_core(addr, event->module))
3845 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3847 static int test_can_verify_check(const char *fmt, ...)
3854 * The verifier is dependent on vsnprintf() modifies the va_list
3855 * passed to it, where it is sent as a reference. Some architectures
3856 * (like x86_32) passes it by value, which means that vsnprintf()
3857 * does not modify the va_list passed to it, and the verifier
3858 * would then need to be able to understand all the values that
3859 * vsnprintf can use. If it is passed by value, then the verifier
3863 vsnprintf(buf, 16, "%d", ap);
3864 ret = va_arg(ap, int);
3870 static void test_can_verify(void)
3872 if (!test_can_verify_check("%d %d", 0, 1)) {
3873 pr_info("trace event string verifier disabled\n");
3874 static_branch_inc(&trace_no_verify);
3879 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3880 * @iter: The iterator that holds the seq buffer and the event being printed
3881 * @fmt: The format used to print the event
3882 * @ap: The va_list holding the data to print from @fmt.
3884 * This writes the data into the @iter->seq buffer using the data from
3885 * @fmt and @ap. If the format has a %s, then the source of the string
3886 * is examined to make sure it is safe to print, otherwise it will
3887 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3890 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3893 const char *p = fmt;
3897 if (WARN_ON_ONCE(!fmt))
3900 if (static_branch_unlikely(&trace_no_verify))
3903 /* Don't bother checking when doing a ftrace_dump() */
3904 if (iter->fmt == static_fmt_buf)
3913 /* We only care about %s and variants */
3914 for (i = 0; p[i]; i++) {
3915 if (i + 1 >= iter->fmt_size) {
3917 * If we can't expand the copy buffer,
3920 if (!trace_iter_expand_format(iter))
3924 if (p[i] == '\\' && p[i+1]) {
3929 /* Need to test cases like %08.*s */
3930 for (j = 1; p[i+j]; j++) {
3931 if (isdigit(p[i+j]) ||
3934 if (p[i+j] == '*') {
3946 /* If no %s found then just print normally */
3950 /* Copy up to the %s, and print that */
3951 strncpy(iter->fmt, p, i);
3952 iter->fmt[i] = '\0';
3953 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3956 * If iter->seq is full, the above call no longer guarantees
3957 * that ap is in sync with fmt processing, and further calls
3958 * to va_arg() can return wrong positional arguments.
3960 * Ensure that ap is no longer used in this case.
3962 if (iter->seq.full) {
3968 len = va_arg(ap, int);
3970 /* The ap now points to the string data of the %s */
3971 str = va_arg(ap, const char *);
3974 * If you hit this warning, it is likely that the
3975 * trace event in question used %s on a string that
3976 * was saved at the time of the event, but may not be
3977 * around when the trace is read. Use __string(),
3978 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3979 * instead. See samples/trace_events/trace-events-sample.h
3982 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3983 "fmt: '%s' current_buffer: '%s'",
3984 fmt, seq_buf_str(&iter->seq.seq))) {
3987 /* Try to safely read the string */
3989 if (len + 1 > iter->fmt_size)
3990 len = iter->fmt_size - 1;
3993 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3997 ret = strncpy_from_kernel_nofault(iter->fmt, str,
4001 trace_seq_printf(&iter->seq, "(0x%px)", str);
4003 trace_seq_printf(&iter->seq, "(0x%px:%s)",
4005 str = "[UNSAFE-MEMORY]";
4006 strcpy(iter->fmt, "%s");
4008 strncpy(iter->fmt, p + i, j + 1);
4009 iter->fmt[j+1] = '\0';
4012 trace_seq_printf(&iter->seq, iter->fmt, len, str);
4014 trace_seq_printf(&iter->seq, iter->fmt, str);
4020 trace_seq_vprintf(&iter->seq, p, ap);
4023 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
4025 const char *p, *new_fmt;
4028 if (WARN_ON_ONCE(!fmt))
4031 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
4035 new_fmt = q = iter->fmt;
4037 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
4038 if (!trace_iter_expand_format(iter))
4041 q += iter->fmt - new_fmt;
4042 new_fmt = iter->fmt;
4047 /* Replace %p with %px */
4051 } else if (p[0] == 'p' && !isalnum(p[1])) {
4062 #define STATIC_TEMP_BUF_SIZE 128
4063 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
4065 /* Find the next real entry, without updating the iterator itself */
4066 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
4067 int *ent_cpu, u64 *ent_ts)
4069 /* __find_next_entry will reset ent_size */
4070 int ent_size = iter->ent_size;
4071 struct trace_entry *entry;
4074 * If called from ftrace_dump(), then the iter->temp buffer
4075 * will be the static_temp_buf and not created from kmalloc.
4076 * If the entry size is greater than the buffer, we can
4077 * not save it. Just return NULL in that case. This is only
4078 * used to add markers when two consecutive events' time
4079 * stamps have a large delta. See trace_print_lat_context()
4081 if (iter->temp == static_temp_buf &&
4082 STATIC_TEMP_BUF_SIZE < ent_size)
4086 * The __find_next_entry() may call peek_next_entry(), which may
4087 * call ring_buffer_peek() that may make the contents of iter->ent
4088 * undefined. Need to copy iter->ent now.
4090 if (iter->ent && iter->ent != iter->temp) {
4091 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4092 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4094 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4099 iter->temp_size = iter->ent_size;
4101 memcpy(iter->temp, iter->ent, iter->ent_size);
4102 iter->ent = iter->temp;
4104 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4105 /* Put back the original ent_size */
4106 iter->ent_size = ent_size;
4111 /* Find the next real entry, and increment the iterator to the next entry */
4112 void *trace_find_next_entry_inc(struct trace_iterator *iter)
4114 iter->ent = __find_next_entry(iter, &iter->cpu,
4115 &iter->lost_events, &iter->ts);
4118 trace_iterator_increment(iter);
4120 return iter->ent ? iter : NULL;
4123 static void trace_consume(struct trace_iterator *iter)
4125 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4126 &iter->lost_events);
4129 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4131 struct trace_iterator *iter = m->private;
4135 WARN_ON_ONCE(iter->leftover);
4139 /* can't go backwards */
4144 ent = trace_find_next_entry_inc(iter);
4148 while (ent && iter->idx < i)
4149 ent = trace_find_next_entry_inc(iter);
4156 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4158 struct ring_buffer_iter *buf_iter;
4159 unsigned long entries = 0;
4162 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4164 buf_iter = trace_buffer_iter(iter, cpu);
4168 ring_buffer_iter_reset(buf_iter);
4171 * We could have the case with the max latency tracers
4172 * that a reset never took place on a cpu. This is evident
4173 * by the timestamp being before the start of the buffer.
4175 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4176 if (ts >= iter->array_buffer->time_start)
4179 ring_buffer_iter_advance(buf_iter);
4182 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4186 * The current tracer is copied to avoid a global locking
4189 static void *s_start(struct seq_file *m, loff_t *pos)
4191 struct trace_iterator *iter = m->private;
4192 struct trace_array *tr = iter->tr;
4193 int cpu_file = iter->cpu_file;
4198 mutex_lock(&trace_types_lock);
4199 if (unlikely(tr->current_trace != iter->trace)) {
4200 /* Close iter->trace before switching to the new current tracer */
4201 if (iter->trace->close)
4202 iter->trace->close(iter);
4203 iter->trace = tr->current_trace;
4204 /* Reopen the new current tracer */
4205 if (iter->trace->open)
4206 iter->trace->open(iter);
4208 mutex_unlock(&trace_types_lock);
4210 #ifdef CONFIG_TRACER_MAX_TRACE
4211 if (iter->snapshot && iter->trace->use_max_tr)
4212 return ERR_PTR(-EBUSY);
4215 if (*pos != iter->pos) {
4220 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4221 for_each_tracing_cpu(cpu)
4222 tracing_iter_reset(iter, cpu);
4224 tracing_iter_reset(iter, cpu_file);
4227 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4232 * If we overflowed the seq_file before, then we want
4233 * to just reuse the trace_seq buffer again.
4239 p = s_next(m, p, &l);
4243 trace_event_read_lock();
4244 trace_access_lock(cpu_file);
4248 static void s_stop(struct seq_file *m, void *p)
4250 struct trace_iterator *iter = m->private;
4252 #ifdef CONFIG_TRACER_MAX_TRACE
4253 if (iter->snapshot && iter->trace->use_max_tr)
4257 trace_access_unlock(iter->cpu_file);
4258 trace_event_read_unlock();
4262 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4263 unsigned long *entries, int cpu)
4265 unsigned long count;
4267 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4269 * If this buffer has skipped entries, then we hold all
4270 * entries for the trace and we need to ignore the
4271 * ones before the time stamp.
4273 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4274 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4275 /* total is the same as the entries */
4279 ring_buffer_overrun_cpu(buf->buffer, cpu);
4284 get_total_entries(struct array_buffer *buf,
4285 unsigned long *total, unsigned long *entries)
4293 for_each_tracing_cpu(cpu) {
4294 get_total_entries_cpu(buf, &t, &e, cpu);
4300 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4302 unsigned long total, entries;
4307 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4312 unsigned long trace_total_entries(struct trace_array *tr)
4314 unsigned long total, entries;
4319 get_total_entries(&tr->array_buffer, &total, &entries);
4324 static void print_lat_help_header(struct seq_file *m)
4326 seq_puts(m, "# _------=> CPU# \n"
4327 "# / _-----=> irqs-off/BH-disabled\n"
4328 "# | / _----=> need-resched \n"
4329 "# || / _---=> hardirq/softirq \n"
4330 "# ||| / _--=> preempt-depth \n"
4331 "# |||| / _-=> migrate-disable \n"
4332 "# ||||| / delay \n"
4333 "# cmd pid |||||| time | caller \n"
4334 "# \\ / |||||| \\ | / \n");
4337 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4339 unsigned long total;
4340 unsigned long entries;
4342 get_total_entries(buf, &total, &entries);
4343 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4344 entries, total, num_online_cpus());
4348 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4351 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4353 print_event_info(buf, m);
4355 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4356 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4359 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4362 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4363 static const char space[] = " ";
4364 int prec = tgid ? 12 : 2;
4366 print_event_info(buf, m);
4368 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4369 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4370 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4371 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4372 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4373 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4374 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4375 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4379 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4381 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4382 struct array_buffer *buf = iter->array_buffer;
4383 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4384 struct tracer *type = iter->trace;
4385 unsigned long entries;
4386 unsigned long total;
4387 const char *name = type->name;
4389 get_total_entries(buf, &total, &entries);
4391 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4393 seq_puts(m, "# -----------------------------------"
4394 "---------------------------------\n");
4395 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4396 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4397 nsecs_to_usecs(data->saved_latency),
4401 preempt_model_none() ? "server" :
4402 preempt_model_voluntary() ? "desktop" :
4403 preempt_model_full() ? "preempt" :
4404 preempt_model_rt() ? "preempt_rt" :
4406 /* These are reserved for later use */
4409 seq_printf(m, " #P:%d)\n", num_online_cpus());
4413 seq_puts(m, "# -----------------\n");
4414 seq_printf(m, "# | task: %.16s-%d "
4415 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4416 data->comm, data->pid,
4417 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4418 data->policy, data->rt_priority);
4419 seq_puts(m, "# -----------------\n");
4421 if (data->critical_start) {
4422 seq_puts(m, "# => started at: ");
4423 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4424 trace_print_seq(m, &iter->seq);
4425 seq_puts(m, "\n# => ended at: ");
4426 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4427 trace_print_seq(m, &iter->seq);
4428 seq_puts(m, "\n#\n");
4434 static void test_cpu_buff_start(struct trace_iterator *iter)
4436 struct trace_seq *s = &iter->seq;
4437 struct trace_array *tr = iter->tr;
4439 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4442 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4445 if (cpumask_available(iter->started) &&
4446 cpumask_test_cpu(iter->cpu, iter->started))
4449 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4452 if (cpumask_available(iter->started))
4453 cpumask_set_cpu(iter->cpu, iter->started);
4455 /* Don't print started cpu buffer for the first entry of the trace */
4457 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4461 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4463 struct trace_array *tr = iter->tr;
4464 struct trace_seq *s = &iter->seq;
4465 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4466 struct trace_entry *entry;
4467 struct trace_event *event;
4471 test_cpu_buff_start(iter);
4473 event = ftrace_find_event(entry->type);
4475 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4476 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4477 trace_print_lat_context(iter);
4479 trace_print_context(iter);
4482 if (trace_seq_has_overflowed(s))
4483 return TRACE_TYPE_PARTIAL_LINE;
4486 if (tr->trace_flags & TRACE_ITER_FIELDS)
4487 return print_event_fields(iter, event);
4488 return event->funcs->trace(iter, sym_flags, event);
4491 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4493 return trace_handle_return(s);
4496 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4498 struct trace_array *tr = iter->tr;
4499 struct trace_seq *s = &iter->seq;
4500 struct trace_entry *entry;
4501 struct trace_event *event;
4505 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4506 trace_seq_printf(s, "%d %d %llu ",
4507 entry->pid, iter->cpu, iter->ts);
4509 if (trace_seq_has_overflowed(s))
4510 return TRACE_TYPE_PARTIAL_LINE;
4512 event = ftrace_find_event(entry->type);
4514 return event->funcs->raw(iter, 0, event);
4516 trace_seq_printf(s, "%d ?\n", entry->type);
4518 return trace_handle_return(s);
4521 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4523 struct trace_array *tr = iter->tr;
4524 struct trace_seq *s = &iter->seq;
4525 unsigned char newline = '\n';
4526 struct trace_entry *entry;
4527 struct trace_event *event;
4531 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4532 SEQ_PUT_HEX_FIELD(s, entry->pid);
4533 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4534 SEQ_PUT_HEX_FIELD(s, iter->ts);
4535 if (trace_seq_has_overflowed(s))
4536 return TRACE_TYPE_PARTIAL_LINE;
4539 event = ftrace_find_event(entry->type);
4541 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4542 if (ret != TRACE_TYPE_HANDLED)
4546 SEQ_PUT_FIELD(s, newline);
4548 return trace_handle_return(s);
4551 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4553 struct trace_array *tr = iter->tr;
4554 struct trace_seq *s = &iter->seq;
4555 struct trace_entry *entry;
4556 struct trace_event *event;
4560 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4561 SEQ_PUT_FIELD(s, entry->pid);
4562 SEQ_PUT_FIELD(s, iter->cpu);
4563 SEQ_PUT_FIELD(s, iter->ts);
4564 if (trace_seq_has_overflowed(s))
4565 return TRACE_TYPE_PARTIAL_LINE;
4568 event = ftrace_find_event(entry->type);
4569 return event ? event->funcs->binary(iter, 0, event) :
4573 int trace_empty(struct trace_iterator *iter)
4575 struct ring_buffer_iter *buf_iter;
4578 /* If we are looking at one CPU buffer, only check that one */
4579 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4580 cpu = iter->cpu_file;
4581 buf_iter = trace_buffer_iter(iter, cpu);
4583 if (!ring_buffer_iter_empty(buf_iter))
4586 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4592 for_each_tracing_cpu(cpu) {
4593 buf_iter = trace_buffer_iter(iter, cpu);
4595 if (!ring_buffer_iter_empty(buf_iter))
4598 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4606 /* Called with trace_event_read_lock() held. */
4607 enum print_line_t print_trace_line(struct trace_iterator *iter)
4609 struct trace_array *tr = iter->tr;
4610 unsigned long trace_flags = tr->trace_flags;
4611 enum print_line_t ret;
4613 if (iter->lost_events) {
4614 if (iter->lost_events == (unsigned long)-1)
4615 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4618 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4619 iter->cpu, iter->lost_events);
4620 if (trace_seq_has_overflowed(&iter->seq))
4621 return TRACE_TYPE_PARTIAL_LINE;
4624 if (iter->trace && iter->trace->print_line) {
4625 ret = iter->trace->print_line(iter);
4626 if (ret != TRACE_TYPE_UNHANDLED)
4630 if (iter->ent->type == TRACE_BPUTS &&
4631 trace_flags & TRACE_ITER_PRINTK &&
4632 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4633 return trace_print_bputs_msg_only(iter);
4635 if (iter->ent->type == TRACE_BPRINT &&
4636 trace_flags & TRACE_ITER_PRINTK &&
4637 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4638 return trace_print_bprintk_msg_only(iter);
4640 if (iter->ent->type == TRACE_PRINT &&
4641 trace_flags & TRACE_ITER_PRINTK &&
4642 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4643 return trace_print_printk_msg_only(iter);
4645 if (trace_flags & TRACE_ITER_BIN)
4646 return print_bin_fmt(iter);
4648 if (trace_flags & TRACE_ITER_HEX)
4649 return print_hex_fmt(iter);
4651 if (trace_flags & TRACE_ITER_RAW)
4652 return print_raw_fmt(iter);
4654 return print_trace_fmt(iter);
4657 void trace_latency_header(struct seq_file *m)
4659 struct trace_iterator *iter = m->private;
4660 struct trace_array *tr = iter->tr;
4662 /* print nothing if the buffers are empty */
4663 if (trace_empty(iter))
4666 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4667 print_trace_header(m, iter);
4669 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4670 print_lat_help_header(m);
4673 void trace_default_header(struct seq_file *m)
4675 struct trace_iterator *iter = m->private;
4676 struct trace_array *tr = iter->tr;
4677 unsigned long trace_flags = tr->trace_flags;
4679 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4682 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4683 /* print nothing if the buffers are empty */
4684 if (trace_empty(iter))
4686 print_trace_header(m, iter);
4687 if (!(trace_flags & TRACE_ITER_VERBOSE))
4688 print_lat_help_header(m);
4690 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4691 if (trace_flags & TRACE_ITER_IRQ_INFO)
4692 print_func_help_header_irq(iter->array_buffer,
4695 print_func_help_header(iter->array_buffer, m,
4701 static void test_ftrace_alive(struct seq_file *m)
4703 if (!ftrace_is_dead())
4705 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4706 "# MAY BE MISSING FUNCTION EVENTS\n");
4709 #ifdef CONFIG_TRACER_MAX_TRACE
4710 static void show_snapshot_main_help(struct seq_file *m)
4712 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4713 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4714 "# Takes a snapshot of the main buffer.\n"
4715 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4716 "# (Doesn't have to be '2' works with any number that\n"
4717 "# is not a '0' or '1')\n");
4720 static void show_snapshot_percpu_help(struct seq_file *m)
4722 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4723 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4724 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4725 "# Takes a snapshot of the main buffer for this cpu.\n");
4727 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4728 "# Must use main snapshot file to allocate.\n");
4730 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4731 "# (Doesn't have to be '2' works with any number that\n"
4732 "# is not a '0' or '1')\n");
4735 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4737 if (iter->tr->allocated_snapshot)
4738 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4740 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4742 seq_puts(m, "# Snapshot commands:\n");
4743 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4744 show_snapshot_main_help(m);
4746 show_snapshot_percpu_help(m);
4749 /* Should never be called */
4750 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4753 static int s_show(struct seq_file *m, void *v)
4755 struct trace_iterator *iter = v;
4758 if (iter->ent == NULL) {
4760 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4762 test_ftrace_alive(m);
4764 if (iter->snapshot && trace_empty(iter))
4765 print_snapshot_help(m, iter);
4766 else if (iter->trace && iter->trace->print_header)
4767 iter->trace->print_header(m);
4769 trace_default_header(m);
4771 } else if (iter->leftover) {
4773 * If we filled the seq_file buffer earlier, we
4774 * want to just show it now.
4776 ret = trace_print_seq(m, &iter->seq);
4778 /* ret should this time be zero, but you never know */
4779 iter->leftover = ret;
4782 ret = print_trace_line(iter);
4783 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4785 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
4787 ret = trace_print_seq(m, &iter->seq);
4789 * If we overflow the seq_file buffer, then it will
4790 * ask us for this data again at start up.
4792 * ret is 0 if seq_file write succeeded.
4795 iter->leftover = ret;
4802 * Should be used after trace_array_get(), trace_types_lock
4803 * ensures that i_cdev was already initialized.
4805 static inline int tracing_get_cpu(struct inode *inode)
4807 if (inode->i_cdev) /* See trace_create_cpu_file() */
4808 return (long)inode->i_cdev - 1;
4809 return RING_BUFFER_ALL_CPUS;
4812 static const struct seq_operations tracer_seq_ops = {
4820 * Note, as iter itself can be allocated and freed in different
4821 * ways, this function is only used to free its content, and not
4822 * the iterator itself. The only requirement to all the allocations
4823 * is that it must zero all fields (kzalloc), as freeing works with
4824 * ethier allocated content or NULL.
4826 static void free_trace_iter_content(struct trace_iterator *iter)
4828 /* The fmt is either NULL, allocated or points to static_fmt_buf */
4829 if (iter->fmt != static_fmt_buf)
4833 kfree(iter->buffer_iter);
4834 mutex_destroy(&iter->mutex);
4835 free_cpumask_var(iter->started);
4838 static struct trace_iterator *
4839 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4841 struct trace_array *tr = inode->i_private;
4842 struct trace_iterator *iter;
4845 if (tracing_disabled)
4846 return ERR_PTR(-ENODEV);
4848 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4850 return ERR_PTR(-ENOMEM);
4852 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4854 if (!iter->buffer_iter)
4858 * trace_find_next_entry() may need to save off iter->ent.
4859 * It will place it into the iter->temp buffer. As most
4860 * events are less than 128, allocate a buffer of that size.
4861 * If one is greater, then trace_find_next_entry() will
4862 * allocate a new buffer to adjust for the bigger iter->ent.
4863 * It's not critical if it fails to get allocated here.
4865 iter->temp = kmalloc(128, GFP_KERNEL);
4867 iter->temp_size = 128;
4870 * trace_event_printf() may need to modify given format
4871 * string to replace %p with %px so that it shows real address
4872 * instead of hash value. However, that is only for the event
4873 * tracing, other tracer may not need. Defer the allocation
4874 * until it is needed.
4879 mutex_lock(&trace_types_lock);
4880 iter->trace = tr->current_trace;
4882 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4887 #ifdef CONFIG_TRACER_MAX_TRACE
4888 /* Currently only the top directory has a snapshot */
4889 if (tr->current_trace->print_max || snapshot)
4890 iter->array_buffer = &tr->max_buffer;
4893 iter->array_buffer = &tr->array_buffer;
4894 iter->snapshot = snapshot;
4896 iter->cpu_file = tracing_get_cpu(inode);
4897 mutex_init(&iter->mutex);
4899 /* Notify the tracer early; before we stop tracing. */
4900 if (iter->trace->open)
4901 iter->trace->open(iter);
4903 /* Annotate start of buffers if we had overruns */
4904 if (ring_buffer_overruns(iter->array_buffer->buffer))
4905 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4907 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4908 if (trace_clocks[tr->clock_id].in_ns)
4909 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4912 * If pause-on-trace is enabled, then stop the trace while
4913 * dumping, unless this is the "snapshot" file
4915 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4916 tracing_stop_tr(tr);
4918 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4919 for_each_tracing_cpu(cpu) {
4920 iter->buffer_iter[cpu] =
4921 ring_buffer_read_prepare(iter->array_buffer->buffer,
4924 ring_buffer_read_prepare_sync();
4925 for_each_tracing_cpu(cpu) {
4926 ring_buffer_read_start(iter->buffer_iter[cpu]);
4927 tracing_iter_reset(iter, cpu);
4930 cpu = iter->cpu_file;
4931 iter->buffer_iter[cpu] =
4932 ring_buffer_read_prepare(iter->array_buffer->buffer,
4934 ring_buffer_read_prepare_sync();
4935 ring_buffer_read_start(iter->buffer_iter[cpu]);
4936 tracing_iter_reset(iter, cpu);
4939 mutex_unlock(&trace_types_lock);
4944 mutex_unlock(&trace_types_lock);
4945 free_trace_iter_content(iter);
4947 seq_release_private(inode, file);
4948 return ERR_PTR(-ENOMEM);
4951 int tracing_open_generic(struct inode *inode, struct file *filp)
4955 ret = tracing_check_open_get_tr(NULL);
4959 filp->private_data = inode->i_private;
4963 bool tracing_is_disabled(void)
4965 return (tracing_disabled) ? true: false;
4969 * Open and update trace_array ref count.
4970 * Must have the current trace_array passed to it.
4972 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4974 struct trace_array *tr = inode->i_private;
4977 ret = tracing_check_open_get_tr(tr);
4981 filp->private_data = inode->i_private;
4987 * The private pointer of the inode is the trace_event_file.
4988 * Update the tr ref count associated to it.
4990 int tracing_open_file_tr(struct inode *inode, struct file *filp)
4992 struct trace_event_file *file = inode->i_private;
4995 ret = tracing_check_open_get_tr(file->tr);
4999 mutex_lock(&event_mutex);
5001 /* Fail if the file is marked for removal */
5002 if (file->flags & EVENT_FILE_FL_FREED) {
5003 trace_array_put(file->tr);
5006 event_file_get(file);
5009 mutex_unlock(&event_mutex);
5013 filp->private_data = inode->i_private;
5018 int tracing_release_file_tr(struct inode *inode, struct file *filp)
5020 struct trace_event_file *file = inode->i_private;
5022 trace_array_put(file->tr);
5023 event_file_put(file);
5028 int tracing_single_release_file_tr(struct inode *inode, struct file *filp)
5030 tracing_release_file_tr(inode, filp);
5031 return single_release(inode, filp);
5034 static int tracing_mark_open(struct inode *inode, struct file *filp)
5036 stream_open(inode, filp);
5037 return tracing_open_generic_tr(inode, filp);
5040 static int tracing_release(struct inode *inode, struct file *file)
5042 struct trace_array *tr = inode->i_private;
5043 struct seq_file *m = file->private_data;
5044 struct trace_iterator *iter;
5047 if (!(file->f_mode & FMODE_READ)) {
5048 trace_array_put(tr);
5052 /* Writes do not use seq_file */
5054 mutex_lock(&trace_types_lock);
5056 for_each_tracing_cpu(cpu) {
5057 if (iter->buffer_iter[cpu])
5058 ring_buffer_read_finish(iter->buffer_iter[cpu]);
5061 if (iter->trace && iter->trace->close)
5062 iter->trace->close(iter);
5064 if (!iter->snapshot && tr->stop_count)
5065 /* reenable tracing if it was previously enabled */
5066 tracing_start_tr(tr);
5068 __trace_array_put(tr);
5070 mutex_unlock(&trace_types_lock);
5072 free_trace_iter_content(iter);
5073 seq_release_private(inode, file);
5078 int tracing_release_generic_tr(struct inode *inode, struct file *file)
5080 struct trace_array *tr = inode->i_private;
5082 trace_array_put(tr);
5086 static int tracing_single_release_tr(struct inode *inode, struct file *file)
5088 struct trace_array *tr = inode->i_private;
5090 trace_array_put(tr);
5092 return single_release(inode, file);
5095 static int tracing_open(struct inode *inode, struct file *file)
5097 struct trace_array *tr = inode->i_private;
5098 struct trace_iterator *iter;
5101 ret = tracing_check_open_get_tr(tr);
5105 /* If this file was open for write, then erase contents */
5106 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
5107 int cpu = tracing_get_cpu(inode);
5108 struct array_buffer *trace_buf = &tr->array_buffer;
5110 #ifdef CONFIG_TRACER_MAX_TRACE
5111 if (tr->current_trace->print_max)
5112 trace_buf = &tr->max_buffer;
5115 if (cpu == RING_BUFFER_ALL_CPUS)
5116 tracing_reset_online_cpus(trace_buf);
5118 tracing_reset_cpu(trace_buf, cpu);
5121 if (file->f_mode & FMODE_READ) {
5122 iter = __tracing_open(inode, file, false);
5124 ret = PTR_ERR(iter);
5125 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
5126 iter->iter_flags |= TRACE_FILE_LAT_FMT;
5130 trace_array_put(tr);
5136 * Some tracers are not suitable for instance buffers.
5137 * A tracer is always available for the global array (toplevel)
5138 * or if it explicitly states that it is.
5141 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
5143 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5146 /* Find the next tracer that this trace array may use */
5147 static struct tracer *
5148 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5150 while (t && !trace_ok_for_array(t, tr))
5157 t_next(struct seq_file *m, void *v, loff_t *pos)
5159 struct trace_array *tr = m->private;
5160 struct tracer *t = v;
5165 t = get_tracer_for_array(tr, t->next);
5170 static void *t_start(struct seq_file *m, loff_t *pos)
5172 struct trace_array *tr = m->private;
5176 mutex_lock(&trace_types_lock);
5178 t = get_tracer_for_array(tr, trace_types);
5179 for (; t && l < *pos; t = t_next(m, t, &l))
5185 static void t_stop(struct seq_file *m, void *p)
5187 mutex_unlock(&trace_types_lock);
5190 static int t_show(struct seq_file *m, void *v)
5192 struct tracer *t = v;
5197 seq_puts(m, t->name);
5206 static const struct seq_operations show_traces_seq_ops = {
5213 static int show_traces_open(struct inode *inode, struct file *file)
5215 struct trace_array *tr = inode->i_private;
5219 ret = tracing_check_open_get_tr(tr);
5223 ret = seq_open(file, &show_traces_seq_ops);
5225 trace_array_put(tr);
5229 m = file->private_data;
5235 static int show_traces_release(struct inode *inode, struct file *file)
5237 struct trace_array *tr = inode->i_private;
5239 trace_array_put(tr);
5240 return seq_release(inode, file);
5244 tracing_write_stub(struct file *filp, const char __user *ubuf,
5245 size_t count, loff_t *ppos)
5250 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5254 if (file->f_mode & FMODE_READ)
5255 ret = seq_lseek(file, offset, whence);
5257 file->f_pos = ret = 0;
5262 static const struct file_operations tracing_fops = {
5263 .open = tracing_open,
5265 .read_iter = seq_read_iter,
5266 .splice_read = copy_splice_read,
5267 .write = tracing_write_stub,
5268 .llseek = tracing_lseek,
5269 .release = tracing_release,
5272 static const struct file_operations show_traces_fops = {
5273 .open = show_traces_open,
5275 .llseek = seq_lseek,
5276 .release = show_traces_release,
5280 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5281 size_t count, loff_t *ppos)
5283 struct trace_array *tr = file_inode(filp)->i_private;
5287 len = snprintf(NULL, 0, "%*pb\n",
5288 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5289 mask_str = kmalloc(len, GFP_KERNEL);
5293 len = snprintf(mask_str, len, "%*pb\n",
5294 cpumask_pr_args(tr->tracing_cpumask));
5299 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5307 int tracing_set_cpumask(struct trace_array *tr,
5308 cpumask_var_t tracing_cpumask_new)
5315 local_irq_disable();
5316 arch_spin_lock(&tr->max_lock);
5317 for_each_tracing_cpu(cpu) {
5319 * Increase/decrease the disabled counter if we are
5320 * about to flip a bit in the cpumask:
5322 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5323 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5324 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5325 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5326 #ifdef CONFIG_TRACER_MAX_TRACE
5327 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5330 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5331 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5332 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5333 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5334 #ifdef CONFIG_TRACER_MAX_TRACE
5335 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5339 arch_spin_unlock(&tr->max_lock);
5342 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5348 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5349 size_t count, loff_t *ppos)
5351 struct trace_array *tr = file_inode(filp)->i_private;
5352 cpumask_var_t tracing_cpumask_new;
5355 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5358 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5362 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5366 free_cpumask_var(tracing_cpumask_new);
5371 free_cpumask_var(tracing_cpumask_new);
5376 static const struct file_operations tracing_cpumask_fops = {
5377 .open = tracing_open_generic_tr,
5378 .read = tracing_cpumask_read,
5379 .write = tracing_cpumask_write,
5380 .release = tracing_release_generic_tr,
5381 .llseek = generic_file_llseek,
5384 static int tracing_trace_options_show(struct seq_file *m, void *v)
5386 struct tracer_opt *trace_opts;
5387 struct trace_array *tr = m->private;
5391 mutex_lock(&trace_types_lock);
5392 tracer_flags = tr->current_trace->flags->val;
5393 trace_opts = tr->current_trace->flags->opts;
5395 for (i = 0; trace_options[i]; i++) {
5396 if (tr->trace_flags & (1 << i))
5397 seq_printf(m, "%s\n", trace_options[i]);
5399 seq_printf(m, "no%s\n", trace_options[i]);
5402 for (i = 0; trace_opts[i].name; i++) {
5403 if (tracer_flags & trace_opts[i].bit)
5404 seq_printf(m, "%s\n", trace_opts[i].name);
5406 seq_printf(m, "no%s\n", trace_opts[i].name);
5408 mutex_unlock(&trace_types_lock);
5413 static int __set_tracer_option(struct trace_array *tr,
5414 struct tracer_flags *tracer_flags,
5415 struct tracer_opt *opts, int neg)
5417 struct tracer *trace = tracer_flags->trace;
5420 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5425 tracer_flags->val &= ~opts->bit;
5427 tracer_flags->val |= opts->bit;
5431 /* Try to assign a tracer specific option */
5432 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5434 struct tracer *trace = tr->current_trace;
5435 struct tracer_flags *tracer_flags = trace->flags;
5436 struct tracer_opt *opts = NULL;
5439 for (i = 0; tracer_flags->opts[i].name; i++) {
5440 opts = &tracer_flags->opts[i];
5442 if (strcmp(cmp, opts->name) == 0)
5443 return __set_tracer_option(tr, trace->flags, opts, neg);
5449 /* Some tracers require overwrite to stay enabled */
5450 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5452 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5458 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5462 if ((mask == TRACE_ITER_RECORD_TGID) ||
5463 (mask == TRACE_ITER_RECORD_CMD))
5464 lockdep_assert_held(&event_mutex);
5466 /* do nothing if flag is already set */
5467 if (!!(tr->trace_flags & mask) == !!enabled)
5470 /* Give the tracer a chance to approve the change */
5471 if (tr->current_trace->flag_changed)
5472 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5476 tr->trace_flags |= mask;
5478 tr->trace_flags &= ~mask;
5480 if (mask == TRACE_ITER_RECORD_CMD)
5481 trace_event_enable_cmd_record(enabled);
5483 if (mask == TRACE_ITER_RECORD_TGID) {
5485 tgid_map_max = pid_max;
5486 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5490 * Pairs with smp_load_acquire() in
5491 * trace_find_tgid_ptr() to ensure that if it observes
5492 * the tgid_map we just allocated then it also observes
5493 * the corresponding tgid_map_max value.
5495 smp_store_release(&tgid_map, map);
5498 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5502 trace_event_enable_tgid_record(enabled);
5505 if (mask == TRACE_ITER_EVENT_FORK)
5506 trace_event_follow_fork(tr, enabled);
5508 if (mask == TRACE_ITER_FUNC_FORK)
5509 ftrace_pid_follow_fork(tr, enabled);
5511 if (mask == TRACE_ITER_OVERWRITE) {
5512 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5513 #ifdef CONFIG_TRACER_MAX_TRACE
5514 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5518 if (mask == TRACE_ITER_PRINTK) {
5519 trace_printk_start_stop_comm(enabled);
5520 trace_printk_control(enabled);
5526 int trace_set_options(struct trace_array *tr, char *option)
5531 size_t orig_len = strlen(option);
5534 cmp = strstrip(option);
5536 len = str_has_prefix(cmp, "no");
5542 mutex_lock(&event_mutex);
5543 mutex_lock(&trace_types_lock);
5545 ret = match_string(trace_options, -1, cmp);
5546 /* If no option could be set, test the specific tracer options */
5548 ret = set_tracer_option(tr, cmp, neg);
5550 ret = set_tracer_flag(tr, 1 << ret, !neg);
5552 mutex_unlock(&trace_types_lock);
5553 mutex_unlock(&event_mutex);
5556 * If the first trailing whitespace is replaced with '\0' by strstrip,
5557 * turn it back into a space.
5559 if (orig_len > strlen(option))
5560 option[strlen(option)] = ' ';
5565 static void __init apply_trace_boot_options(void)
5567 char *buf = trace_boot_options_buf;
5571 option = strsep(&buf, ",");
5577 trace_set_options(&global_trace, option);
5579 /* Put back the comma to allow this to be called again */
5586 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5587 size_t cnt, loff_t *ppos)
5589 struct seq_file *m = filp->private_data;
5590 struct trace_array *tr = m->private;
5594 if (cnt >= sizeof(buf))
5597 if (copy_from_user(buf, ubuf, cnt))
5602 ret = trace_set_options(tr, buf);
5611 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5613 struct trace_array *tr = inode->i_private;
5616 ret = tracing_check_open_get_tr(tr);
5620 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5622 trace_array_put(tr);
5627 static const struct file_operations tracing_iter_fops = {
5628 .open = tracing_trace_options_open,
5630 .llseek = seq_lseek,
5631 .release = tracing_single_release_tr,
5632 .write = tracing_trace_options_write,
5635 static const char readme_msg[] =
5636 "tracing mini-HOWTO:\n\n"
5637 "# echo 0 > tracing_on : quick way to disable tracing\n"
5638 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5639 " Important files:\n"
5640 " trace\t\t\t- The static contents of the buffer\n"
5641 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5642 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5643 " current_tracer\t- function and latency tracers\n"
5644 " available_tracers\t- list of configured tracers for current_tracer\n"
5645 " error_log\t- error log for failed commands (that support it)\n"
5646 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5647 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5648 " trace_clock\t\t- change the clock used to order events\n"
5649 " local: Per cpu clock but may not be synced across CPUs\n"
5650 " global: Synced across CPUs but slows tracing down.\n"
5651 " counter: Not a clock, but just an increment\n"
5652 " uptime: Jiffy counter from time of boot\n"
5653 " perf: Same clock that perf events use\n"
5654 #ifdef CONFIG_X86_64
5655 " x86-tsc: TSC cycle counter\n"
5657 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5658 " delta: Delta difference against a buffer-wide timestamp\n"
5659 " absolute: Absolute (standalone) timestamp\n"
5660 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5661 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5662 " tracing_cpumask\t- Limit which CPUs to trace\n"
5663 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5664 "\t\t\t Remove sub-buffer with rmdir\n"
5665 " trace_options\t\t- Set format or modify how tracing happens\n"
5666 "\t\t\t Disable an option by prefixing 'no' to the\n"
5667 "\t\t\t option name\n"
5668 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5669 #ifdef CONFIG_DYNAMIC_FTRACE
5670 "\n available_filter_functions - list of functions that can be filtered on\n"
5671 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5672 "\t\t\t functions\n"
5673 "\t accepts: func_full_name or glob-matching-pattern\n"
5674 "\t modules: Can select a group via module\n"
5675 "\t Format: :mod:<module-name>\n"
5676 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5677 "\t triggers: a command to perform when function is hit\n"
5678 "\t Format: <function>:<trigger>[:count]\n"
5679 "\t trigger: traceon, traceoff\n"
5680 "\t\t enable_event:<system>:<event>\n"
5681 "\t\t disable_event:<system>:<event>\n"
5682 #ifdef CONFIG_STACKTRACE
5685 #ifdef CONFIG_TRACER_SNAPSHOT
5690 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5691 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5692 "\t The first one will disable tracing every time do_fault is hit\n"
5693 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5694 "\t The first time do trap is hit and it disables tracing, the\n"
5695 "\t counter will decrement to 2. If tracing is already disabled,\n"
5696 "\t the counter will not decrement. It only decrements when the\n"
5697 "\t trigger did work\n"
5698 "\t To remove trigger without count:\n"
5699 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5700 "\t To remove trigger with a count:\n"
5701 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5702 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5703 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5704 "\t modules: Can select a group via module command :mod:\n"
5705 "\t Does not accept triggers\n"
5706 #endif /* CONFIG_DYNAMIC_FTRACE */
5707 #ifdef CONFIG_FUNCTION_TRACER
5708 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5710 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5713 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5714 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5715 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5716 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5718 #ifdef CONFIG_TRACER_SNAPSHOT
5719 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5720 "\t\t\t snapshot buffer. Read the contents for more\n"
5721 "\t\t\t information\n"
5723 #ifdef CONFIG_STACK_TRACER
5724 " stack_trace\t\t- Shows the max stack trace when active\n"
5725 " stack_max_size\t- Shows current max stack size that was traced\n"
5726 "\t\t\t Write into this file to reset the max size (trigger a\n"
5727 "\t\t\t new trace)\n"
5728 #ifdef CONFIG_DYNAMIC_FTRACE
5729 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5732 #endif /* CONFIG_STACK_TRACER */
5733 #ifdef CONFIG_DYNAMIC_EVENTS
5734 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5735 "\t\t\t Write into this file to define/undefine new trace events.\n"
5737 #ifdef CONFIG_KPROBE_EVENTS
5738 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5739 "\t\t\t Write into this file to define/undefine new trace events.\n"
5741 #ifdef CONFIG_UPROBE_EVENTS
5742 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5743 "\t\t\t Write into this file to define/undefine new trace events.\n"
5745 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5746 defined(CONFIG_FPROBE_EVENTS)
5747 "\t accepts: event-definitions (one definition per line)\n"
5748 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5749 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5750 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5752 #ifdef CONFIG_FPROBE_EVENTS
5753 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5754 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5756 #ifdef CONFIG_HIST_TRIGGERS
5757 "\t s:[synthetic/]<event> <field> [<field>]\n"
5759 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5760 "\t -:[<group>/][<event>]\n"
5761 #ifdef CONFIG_KPROBE_EVENTS
5762 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5763 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5765 #ifdef CONFIG_UPROBE_EVENTS
5766 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5768 "\t args: <name>=fetcharg[:type]\n"
5769 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5770 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5771 #ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5772 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5773 "\t <argname>[->field[->field|.field...]],\n"
5775 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5778 "\t $stack<index>, $stack, $retval, $comm,\n"
5780 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5781 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5782 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5783 "\t symstr, <type>\\[<array-size>\\]\n"
5784 #ifdef CONFIG_HIST_TRIGGERS
5785 "\t field: <stype> <name>;\n"
5786 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5787 "\t [unsigned] char/int/long\n"
5789 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5790 "\t of the <attached-group>/<attached-event>.\n"
5792 " events/\t\t- Directory containing all trace event subsystems:\n"
5793 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5794 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5795 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5797 " filter\t\t- If set, only events passing filter are traced\n"
5798 " events/<system>/<event>/\t- Directory containing control files for\n"
5800 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5801 " filter\t\t- If set, only events passing filter are traced\n"
5802 " trigger\t\t- If set, a command to perform when event is hit\n"
5803 "\t Format: <trigger>[:count][if <filter>]\n"
5804 "\t trigger: traceon, traceoff\n"
5805 "\t enable_event:<system>:<event>\n"
5806 "\t disable_event:<system>:<event>\n"
5807 #ifdef CONFIG_HIST_TRIGGERS
5808 "\t enable_hist:<system>:<event>\n"
5809 "\t disable_hist:<system>:<event>\n"
5811 #ifdef CONFIG_STACKTRACE
5814 #ifdef CONFIG_TRACER_SNAPSHOT
5817 #ifdef CONFIG_HIST_TRIGGERS
5818 "\t\t hist (see below)\n"
5820 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5821 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5822 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5823 "\t events/block/block_unplug/trigger\n"
5824 "\t The first disables tracing every time block_unplug is hit.\n"
5825 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5826 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5827 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5828 "\t Like function triggers, the counter is only decremented if it\n"
5829 "\t enabled or disabled tracing.\n"
5830 "\t To remove a trigger without a count:\n"
5831 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5832 "\t To remove a trigger with a count:\n"
5833 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5834 "\t Filters can be ignored when removing a trigger.\n"
5835 #ifdef CONFIG_HIST_TRIGGERS
5836 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5837 "\t Format: hist:keys=<field1[,field2,...]>\n"
5838 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5839 "\t [:values=<field1[,field2,...]>]\n"
5840 "\t [:sort=<field1[,field2,...]>]\n"
5841 "\t [:size=#entries]\n"
5842 "\t [:pause][:continue][:clear]\n"
5843 "\t [:name=histname1]\n"
5844 "\t [:nohitcount]\n"
5845 "\t [:<handler>.<action>]\n"
5846 "\t [if <filter>]\n\n"
5847 "\t Note, special fields can be used as well:\n"
5848 "\t common_timestamp - to record current timestamp\n"
5849 "\t common_cpu - to record the CPU the event happened on\n"
5851 "\t A hist trigger variable can be:\n"
5852 "\t - a reference to a field e.g. x=current_timestamp,\n"
5853 "\t - a reference to another variable e.g. y=$x,\n"
5854 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5855 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5857 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5858 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5859 "\t variable reference, field or numeric literal.\n"
5861 "\t When a matching event is hit, an entry is added to a hash\n"
5862 "\t table using the key(s) and value(s) named, and the value of a\n"
5863 "\t sum called 'hitcount' is incremented. Keys and values\n"
5864 "\t correspond to fields in the event's format description. Keys\n"
5865 "\t can be any field, or the special string 'common_stacktrace'.\n"
5866 "\t Compound keys consisting of up to two fields can be specified\n"
5867 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5868 "\t fields. Sort keys consisting of up to two fields can be\n"
5869 "\t specified using the 'sort' keyword. The sort direction can\n"
5870 "\t be modified by appending '.descending' or '.ascending' to a\n"
5871 "\t sort field. The 'size' parameter can be used to specify more\n"
5872 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5873 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5874 "\t its histogram data will be shared with other triggers of the\n"
5875 "\t same name, and trigger hits will update this common data.\n\n"
5876 "\t Reading the 'hist' file for the event will dump the hash\n"
5877 "\t table in its entirety to stdout. If there are multiple hist\n"
5878 "\t triggers attached to an event, there will be a table for each\n"
5879 "\t trigger in the output. The table displayed for a named\n"
5880 "\t trigger will be the same as any other instance having the\n"
5881 "\t same name. The default format used to display a given field\n"
5882 "\t can be modified by appending any of the following modifiers\n"
5883 "\t to the field name, as applicable:\n\n"
5884 "\t .hex display a number as a hex value\n"
5885 "\t .sym display an address as a symbol\n"
5886 "\t .sym-offset display an address as a symbol and offset\n"
5887 "\t .execname display a common_pid as a program name\n"
5888 "\t .syscall display a syscall id as a syscall name\n"
5889 "\t .log2 display log2 value rather than raw number\n"
5890 "\t .buckets=size display values in groups of size rather than raw number\n"
5891 "\t .usecs display a common_timestamp in microseconds\n"
5892 "\t .percent display a number of percentage value\n"
5893 "\t .graph display a bar-graph of a value\n\n"
5894 "\t The 'pause' parameter can be used to pause an existing hist\n"
5895 "\t trigger or to start a hist trigger but not log any events\n"
5896 "\t until told to do so. 'continue' can be used to start or\n"
5897 "\t restart a paused hist trigger.\n\n"
5898 "\t The 'clear' parameter will clear the contents of a running\n"
5899 "\t hist trigger and leave its current paused/active state\n"
5901 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5902 "\t raw hitcount in the histogram.\n\n"
5903 "\t The enable_hist and disable_hist triggers can be used to\n"
5904 "\t have one event conditionally start and stop another event's\n"
5905 "\t already-attached hist trigger. The syntax is analogous to\n"
5906 "\t the enable_event and disable_event triggers.\n\n"
5907 "\t Hist trigger handlers and actions are executed whenever a\n"
5908 "\t a histogram entry is added or updated. They take the form:\n\n"
5909 "\t <handler>.<action>\n\n"
5910 "\t The available handlers are:\n\n"
5911 "\t onmatch(matching.event) - invoke on addition or update\n"
5912 "\t onmax(var) - invoke if var exceeds current max\n"
5913 "\t onchange(var) - invoke action if var changes\n\n"
5914 "\t The available actions are:\n\n"
5915 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5916 "\t save(field,...) - save current event fields\n"
5917 #ifdef CONFIG_TRACER_SNAPSHOT
5918 "\t snapshot() - snapshot the trace buffer\n\n"
5920 #ifdef CONFIG_SYNTH_EVENTS
5921 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5922 "\t Write into this file to define/undefine new synthetic events.\n"
5923 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5929 tracing_readme_read(struct file *filp, char __user *ubuf,
5930 size_t cnt, loff_t *ppos)
5932 return simple_read_from_buffer(ubuf, cnt, ppos,
5933 readme_msg, strlen(readme_msg));
5936 static const struct file_operations tracing_readme_fops = {
5937 .open = tracing_open_generic,
5938 .read = tracing_readme_read,
5939 .llseek = generic_file_llseek,
5942 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5946 return trace_find_tgid_ptr(pid);
5949 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5953 return trace_find_tgid_ptr(pid);
5956 static void saved_tgids_stop(struct seq_file *m, void *v)
5960 static int saved_tgids_show(struct seq_file *m, void *v)
5962 int *entry = (int *)v;
5963 int pid = entry - tgid_map;
5969 seq_printf(m, "%d %d\n", pid, tgid);
5973 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5974 .start = saved_tgids_start,
5975 .stop = saved_tgids_stop,
5976 .next = saved_tgids_next,
5977 .show = saved_tgids_show,
5980 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5984 ret = tracing_check_open_get_tr(NULL);
5988 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5992 static const struct file_operations tracing_saved_tgids_fops = {
5993 .open = tracing_saved_tgids_open,
5995 .llseek = seq_lseek,
5996 .release = seq_release,
5999 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
6001 unsigned int *ptr = v;
6003 if (*pos || m->count)
6008 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
6010 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
6019 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
6025 arch_spin_lock(&trace_cmdline_lock);
6027 v = &savedcmd->map_cmdline_to_pid[0];
6029 v = saved_cmdlines_next(m, v, &l);
6037 static void saved_cmdlines_stop(struct seq_file *m, void *v)
6039 arch_spin_unlock(&trace_cmdline_lock);
6043 static int saved_cmdlines_show(struct seq_file *m, void *v)
6045 char buf[TASK_COMM_LEN];
6046 unsigned int *pid = v;
6048 __trace_find_cmdline(*pid, buf);
6049 seq_printf(m, "%d %s\n", *pid, buf);
6053 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
6054 .start = saved_cmdlines_start,
6055 .next = saved_cmdlines_next,
6056 .stop = saved_cmdlines_stop,
6057 .show = saved_cmdlines_show,
6060 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
6064 ret = tracing_check_open_get_tr(NULL);
6068 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
6071 static const struct file_operations tracing_saved_cmdlines_fops = {
6072 .open = tracing_saved_cmdlines_open,
6074 .llseek = seq_lseek,
6075 .release = seq_release,
6079 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
6080 size_t cnt, loff_t *ppos)
6086 arch_spin_lock(&trace_cmdline_lock);
6087 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
6088 arch_spin_unlock(&trace_cmdline_lock);
6091 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6094 static int tracing_resize_saved_cmdlines(unsigned int val)
6096 struct saved_cmdlines_buffer *s, *savedcmd_temp;
6098 s = allocate_cmdlines_buffer(val);
6103 arch_spin_lock(&trace_cmdline_lock);
6104 savedcmd_temp = savedcmd;
6106 arch_spin_unlock(&trace_cmdline_lock);
6108 free_saved_cmdlines_buffer(savedcmd_temp);
6114 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
6115 size_t cnt, loff_t *ppos)
6120 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6124 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
6125 if (!val || val > PID_MAX_DEFAULT)
6128 ret = tracing_resize_saved_cmdlines((unsigned int)val);
6137 static const struct file_operations tracing_saved_cmdlines_size_fops = {
6138 .open = tracing_open_generic,
6139 .read = tracing_saved_cmdlines_size_read,
6140 .write = tracing_saved_cmdlines_size_write,
6143 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
6144 static union trace_eval_map_item *
6145 update_eval_map(union trace_eval_map_item *ptr)
6147 if (!ptr->map.eval_string) {
6148 if (ptr->tail.next) {
6149 ptr = ptr->tail.next;
6150 /* Set ptr to the next real item (skip head) */
6158 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6160 union trace_eval_map_item *ptr = v;
6163 * Paranoid! If ptr points to end, we don't want to increment past it.
6164 * This really should never happen.
6167 ptr = update_eval_map(ptr);
6168 if (WARN_ON_ONCE(!ptr))
6172 ptr = update_eval_map(ptr);
6177 static void *eval_map_start(struct seq_file *m, loff_t *pos)
6179 union trace_eval_map_item *v;
6182 mutex_lock(&trace_eval_mutex);
6184 v = trace_eval_maps;
6188 while (v && l < *pos) {
6189 v = eval_map_next(m, v, &l);
6195 static void eval_map_stop(struct seq_file *m, void *v)
6197 mutex_unlock(&trace_eval_mutex);
6200 static int eval_map_show(struct seq_file *m, void *v)
6202 union trace_eval_map_item *ptr = v;
6204 seq_printf(m, "%s %ld (%s)\n",
6205 ptr->map.eval_string, ptr->map.eval_value,
6211 static const struct seq_operations tracing_eval_map_seq_ops = {
6212 .start = eval_map_start,
6213 .next = eval_map_next,
6214 .stop = eval_map_stop,
6215 .show = eval_map_show,
6218 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6222 ret = tracing_check_open_get_tr(NULL);
6226 return seq_open(filp, &tracing_eval_map_seq_ops);
6229 static const struct file_operations tracing_eval_map_fops = {
6230 .open = tracing_eval_map_open,
6232 .llseek = seq_lseek,
6233 .release = seq_release,
6236 static inline union trace_eval_map_item *
6237 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6239 /* Return tail of array given the head */
6240 return ptr + ptr->head.length + 1;
6244 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6247 struct trace_eval_map **stop;
6248 struct trace_eval_map **map;
6249 union trace_eval_map_item *map_array;
6250 union trace_eval_map_item *ptr;
6255 * The trace_eval_maps contains the map plus a head and tail item,
6256 * where the head holds the module and length of array, and the
6257 * tail holds a pointer to the next list.
6259 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6261 pr_warn("Unable to allocate trace eval mapping\n");
6265 mutex_lock(&trace_eval_mutex);
6267 if (!trace_eval_maps)
6268 trace_eval_maps = map_array;
6270 ptr = trace_eval_maps;
6272 ptr = trace_eval_jmp_to_tail(ptr);
6273 if (!ptr->tail.next)
6275 ptr = ptr->tail.next;
6278 ptr->tail.next = map_array;
6280 map_array->head.mod = mod;
6281 map_array->head.length = len;
6284 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6285 map_array->map = **map;
6288 memset(map_array, 0, sizeof(*map_array));
6290 mutex_unlock(&trace_eval_mutex);
6293 static void trace_create_eval_file(struct dentry *d_tracer)
6295 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6296 NULL, &tracing_eval_map_fops);
6299 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
6300 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6301 static inline void trace_insert_eval_map_file(struct module *mod,
6302 struct trace_eval_map **start, int len) { }
6303 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6305 static void trace_insert_eval_map(struct module *mod,
6306 struct trace_eval_map **start, int len)
6308 struct trace_eval_map **map;
6315 trace_event_eval_update(map, len);
6317 trace_insert_eval_map_file(mod, start, len);
6321 tracing_set_trace_read(struct file *filp, char __user *ubuf,
6322 size_t cnt, loff_t *ppos)
6324 struct trace_array *tr = filp->private_data;
6325 char buf[MAX_TRACER_SIZE+2];
6328 mutex_lock(&trace_types_lock);
6329 r = sprintf(buf, "%s\n", tr->current_trace->name);
6330 mutex_unlock(&trace_types_lock);
6332 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6335 int tracer_init(struct tracer *t, struct trace_array *tr)
6337 tracing_reset_online_cpus(&tr->array_buffer);
6341 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6345 for_each_tracing_cpu(cpu)
6346 per_cpu_ptr(buf->data, cpu)->entries = val;
6349 static void update_buffer_entries(struct array_buffer *buf, int cpu)
6351 if (cpu == RING_BUFFER_ALL_CPUS) {
6352 set_buffer_entries(buf, ring_buffer_size(buf->buffer, 0));
6354 per_cpu_ptr(buf->data, cpu)->entries = ring_buffer_size(buf->buffer, cpu);
6358 #ifdef CONFIG_TRACER_MAX_TRACE
6359 /* resize @tr's buffer to the size of @size_tr's entries */
6360 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6361 struct array_buffer *size_buf, int cpu_id)
6365 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6366 for_each_tracing_cpu(cpu) {
6367 ret = ring_buffer_resize(trace_buf->buffer,
6368 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6371 per_cpu_ptr(trace_buf->data, cpu)->entries =
6372 per_cpu_ptr(size_buf->data, cpu)->entries;
6375 ret = ring_buffer_resize(trace_buf->buffer,
6376 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6378 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6379 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6384 #endif /* CONFIG_TRACER_MAX_TRACE */
6386 static int __tracing_resize_ring_buffer(struct trace_array *tr,
6387 unsigned long size, int cpu)
6392 * If kernel or user changes the size of the ring buffer
6393 * we use the size that was given, and we can forget about
6394 * expanding it later.
6396 trace_set_ring_buffer_expanded(tr);
6398 /* May be called before buffers are initialized */
6399 if (!tr->array_buffer.buffer)
6402 /* Do not allow tracing while resizing ring buffer */
6403 tracing_stop_tr(tr);
6405 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6409 #ifdef CONFIG_TRACER_MAX_TRACE
6410 if (!tr->allocated_snapshot)
6413 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6415 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6416 &tr->array_buffer, cpu);
6419 * AARGH! We are left with different
6420 * size max buffer!!!!
6421 * The max buffer is our "snapshot" buffer.
6422 * When a tracer needs a snapshot (one of the
6423 * latency tracers), it swaps the max buffer
6424 * with the saved snap shot. We succeeded to
6425 * update the size of the main buffer, but failed to
6426 * update the size of the max buffer. But when we tried
6427 * to reset the main buffer to the original size, we
6428 * failed there too. This is very unlikely to
6429 * happen, but if it does, warn and kill all
6433 tracing_disabled = 1;
6438 update_buffer_entries(&tr->max_buffer, cpu);
6441 #endif /* CONFIG_TRACER_MAX_TRACE */
6443 update_buffer_entries(&tr->array_buffer, cpu);
6445 tracing_start_tr(tr);
6449 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6450 unsigned long size, int cpu_id)
6454 mutex_lock(&trace_types_lock);
6456 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6457 /* make sure, this cpu is enabled in the mask */
6458 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6464 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6469 mutex_unlock(&trace_types_lock);
6476 * tracing_update_buffers - used by tracing facility to expand ring buffers
6477 * @tr: The tracing instance
6479 * To save on memory when the tracing is never used on a system with it
6480 * configured in. The ring buffers are set to a minimum size. But once
6481 * a user starts to use the tracing facility, then they need to grow
6482 * to their default size.
6484 * This function is to be called when a tracer is about to be used.
6486 int tracing_update_buffers(struct trace_array *tr)
6490 mutex_lock(&trace_types_lock);
6491 if (!tr->ring_buffer_expanded)
6492 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6493 RING_BUFFER_ALL_CPUS);
6494 mutex_unlock(&trace_types_lock);
6499 struct trace_option_dentry;
6502 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6505 * Used to clear out the tracer before deletion of an instance.
6506 * Must have trace_types_lock held.
6508 static void tracing_set_nop(struct trace_array *tr)
6510 if (tr->current_trace == &nop_trace)
6513 tr->current_trace->enabled--;
6515 if (tr->current_trace->reset)
6516 tr->current_trace->reset(tr);
6518 tr->current_trace = &nop_trace;
6521 static bool tracer_options_updated;
6523 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6525 /* Only enable if the directory has been created already. */
6529 /* Only create trace option files after update_tracer_options finish */
6530 if (!tracer_options_updated)
6533 create_trace_option_files(tr, t);
6536 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6539 #ifdef CONFIG_TRACER_MAX_TRACE
6544 mutex_lock(&trace_types_lock);
6546 if (!tr->ring_buffer_expanded) {
6547 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6548 RING_BUFFER_ALL_CPUS);
6554 for (t = trace_types; t; t = t->next) {
6555 if (strcmp(t->name, buf) == 0)
6562 if (t == tr->current_trace)
6565 #ifdef CONFIG_TRACER_SNAPSHOT
6566 if (t->use_max_tr) {
6567 local_irq_disable();
6568 arch_spin_lock(&tr->max_lock);
6569 if (tr->cond_snapshot)
6571 arch_spin_unlock(&tr->max_lock);
6577 /* Some tracers won't work on kernel command line */
6578 if (system_state < SYSTEM_RUNNING && t->noboot) {
6579 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6584 /* Some tracers are only allowed for the top level buffer */
6585 if (!trace_ok_for_array(t, tr)) {
6590 /* If trace pipe files are being read, we can't change the tracer */
6591 if (tr->trace_ref) {
6596 trace_branch_disable();
6598 tr->current_trace->enabled--;
6600 if (tr->current_trace->reset)
6601 tr->current_trace->reset(tr);
6603 #ifdef CONFIG_TRACER_MAX_TRACE
6604 had_max_tr = tr->current_trace->use_max_tr;
6606 /* Current trace needs to be nop_trace before synchronize_rcu */
6607 tr->current_trace = &nop_trace;
6609 if (had_max_tr && !t->use_max_tr) {
6611 * We need to make sure that the update_max_tr sees that
6612 * current_trace changed to nop_trace to keep it from
6613 * swapping the buffers after we resize it.
6614 * The update_max_tr is called from interrupts disabled
6615 * so a synchronized_sched() is sufficient.
6621 if (t->use_max_tr && !tr->allocated_snapshot) {
6622 ret = tracing_alloc_snapshot_instance(tr);
6627 tr->current_trace = &nop_trace;
6631 ret = tracer_init(t, tr);
6636 tr->current_trace = t;
6637 tr->current_trace->enabled++;
6638 trace_branch_enable(tr);
6640 mutex_unlock(&trace_types_lock);
6646 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6647 size_t cnt, loff_t *ppos)
6649 struct trace_array *tr = filp->private_data;
6650 char buf[MAX_TRACER_SIZE+1];
6657 if (cnt > MAX_TRACER_SIZE)
6658 cnt = MAX_TRACER_SIZE;
6660 if (copy_from_user(buf, ubuf, cnt))
6667 err = tracing_set_tracer(tr, name);
6677 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6678 size_t cnt, loff_t *ppos)
6683 r = snprintf(buf, sizeof(buf), "%ld\n",
6684 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6685 if (r > sizeof(buf))
6687 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6691 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6692 size_t cnt, loff_t *ppos)
6697 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6707 tracing_thresh_read(struct file *filp, char __user *ubuf,
6708 size_t cnt, loff_t *ppos)
6710 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6714 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6715 size_t cnt, loff_t *ppos)
6717 struct trace_array *tr = filp->private_data;
6720 mutex_lock(&trace_types_lock);
6721 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6725 if (tr->current_trace->update_thresh) {
6726 ret = tr->current_trace->update_thresh(tr);
6733 mutex_unlock(&trace_types_lock);
6738 #ifdef CONFIG_TRACER_MAX_TRACE
6741 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6742 size_t cnt, loff_t *ppos)
6744 struct trace_array *tr = filp->private_data;
6746 return tracing_nsecs_read(&tr->max_latency, ubuf, cnt, ppos);
6750 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6751 size_t cnt, loff_t *ppos)
6753 struct trace_array *tr = filp->private_data;
6755 return tracing_nsecs_write(&tr->max_latency, ubuf, cnt, ppos);
6760 static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6762 if (cpu == RING_BUFFER_ALL_CPUS) {
6763 if (cpumask_empty(tr->pipe_cpumask)) {
6764 cpumask_setall(tr->pipe_cpumask);
6767 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6768 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6774 static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6776 if (cpu == RING_BUFFER_ALL_CPUS) {
6777 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6778 cpumask_clear(tr->pipe_cpumask);
6780 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6781 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6785 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6787 struct trace_array *tr = inode->i_private;
6788 struct trace_iterator *iter;
6792 ret = tracing_check_open_get_tr(tr);
6796 mutex_lock(&trace_types_lock);
6797 cpu = tracing_get_cpu(inode);
6798 ret = open_pipe_on_cpu(tr, cpu);
6800 goto fail_pipe_on_cpu;
6802 /* create a buffer to store the information to pass to userspace */
6803 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6806 goto fail_alloc_iter;
6809 trace_seq_init(&iter->seq);
6810 iter->trace = tr->current_trace;
6812 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6817 /* trace pipe does not show start of buffer */
6818 cpumask_setall(iter->started);
6820 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6821 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6823 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6824 if (trace_clocks[tr->clock_id].in_ns)
6825 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6828 iter->array_buffer = &tr->array_buffer;
6829 iter->cpu_file = cpu;
6830 mutex_init(&iter->mutex);
6831 filp->private_data = iter;
6833 if (iter->trace->pipe_open)
6834 iter->trace->pipe_open(iter);
6836 nonseekable_open(inode, filp);
6840 mutex_unlock(&trace_types_lock);
6846 close_pipe_on_cpu(tr, cpu);
6848 __trace_array_put(tr);
6849 mutex_unlock(&trace_types_lock);
6853 static int tracing_release_pipe(struct inode *inode, struct file *file)
6855 struct trace_iterator *iter = file->private_data;
6856 struct trace_array *tr = inode->i_private;
6858 mutex_lock(&trace_types_lock);
6862 if (iter->trace->pipe_close)
6863 iter->trace->pipe_close(iter);
6864 close_pipe_on_cpu(tr, iter->cpu_file);
6865 mutex_unlock(&trace_types_lock);
6867 free_trace_iter_content(iter);
6870 trace_array_put(tr);
6876 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6878 struct trace_array *tr = iter->tr;
6880 /* Iterators are static, they should be filled or empty */
6881 if (trace_buffer_iter(iter, iter->cpu_file))
6882 return EPOLLIN | EPOLLRDNORM;
6884 if (tr->trace_flags & TRACE_ITER_BLOCK)
6886 * Always select as readable when in blocking mode
6888 return EPOLLIN | EPOLLRDNORM;
6890 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6891 filp, poll_table, iter->tr->buffer_percent);
6895 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6897 struct trace_iterator *iter = filp->private_data;
6899 return trace_poll(iter, filp, poll_table);
6902 /* Must be called with iter->mutex held. */
6903 static int tracing_wait_pipe(struct file *filp)
6905 struct trace_iterator *iter = filp->private_data;
6908 while (trace_empty(iter)) {
6910 if ((filp->f_flags & O_NONBLOCK)) {
6915 * We block until we read something and tracing is disabled.
6916 * We still block if tracing is disabled, but we have never
6917 * read anything. This allows a user to cat this file, and
6918 * then enable tracing. But after we have read something,
6919 * we give an EOF when tracing is again disabled.
6921 * iter->pos will be 0 if we haven't read anything.
6923 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6926 mutex_unlock(&iter->mutex);
6928 ret = wait_on_pipe(iter, 0);
6930 mutex_lock(&iter->mutex);
6943 tracing_read_pipe(struct file *filp, char __user *ubuf,
6944 size_t cnt, loff_t *ppos)
6946 struct trace_iterator *iter = filp->private_data;
6950 * Avoid more than one consumer on a single file descriptor
6951 * This is just a matter of traces coherency, the ring buffer itself
6954 mutex_lock(&iter->mutex);
6956 /* return any leftover data */
6957 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6961 trace_seq_init(&iter->seq);
6963 if (iter->trace->read) {
6964 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6970 sret = tracing_wait_pipe(filp);
6974 /* stop when tracing is finished */
6975 if (trace_empty(iter)) {
6980 if (cnt >= TRACE_SEQ_BUFFER_SIZE)
6981 cnt = TRACE_SEQ_BUFFER_SIZE - 1;
6983 /* reset all but tr, trace, and overruns */
6984 trace_iterator_reset(iter);
6985 cpumask_clear(iter->started);
6986 trace_seq_init(&iter->seq);
6988 trace_event_read_lock();
6989 trace_access_lock(iter->cpu_file);
6990 while (trace_find_next_entry_inc(iter) != NULL) {
6991 enum print_line_t ret;
6992 int save_len = iter->seq.seq.len;
6994 ret = print_trace_line(iter);
6995 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6997 * If one print_trace_line() fills entire trace_seq in one shot,
6998 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6999 * In this case, we need to consume it, otherwise, loop will peek
7000 * this event next time, resulting in an infinite loop.
7002 if (save_len == 0) {
7004 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
7005 trace_consume(iter);
7009 /* In other cases, don't print partial lines */
7010 iter->seq.seq.len = save_len;
7013 if (ret != TRACE_TYPE_NO_CONSUME)
7014 trace_consume(iter);
7016 if (trace_seq_used(&iter->seq) >= cnt)
7020 * Setting the full flag means we reached the trace_seq buffer
7021 * size and we should leave by partial output condition above.
7022 * One of the trace_seq_* functions is not used properly.
7024 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
7027 trace_access_unlock(iter->cpu_file);
7028 trace_event_read_unlock();
7030 /* Now copy what we have to the user */
7031 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
7032 if (iter->seq.readpos >= trace_seq_used(&iter->seq))
7033 trace_seq_init(&iter->seq);
7036 * If there was nothing to send to user, in spite of consuming trace
7037 * entries, go back to wait for more entries.
7043 mutex_unlock(&iter->mutex);
7048 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
7051 __free_page(spd->pages[idx]);
7055 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
7061 /* Seq buffer is page-sized, exactly what we need. */
7063 save_len = iter->seq.seq.len;
7064 ret = print_trace_line(iter);
7066 if (trace_seq_has_overflowed(&iter->seq)) {
7067 iter->seq.seq.len = save_len;
7072 * This should not be hit, because it should only
7073 * be set if the iter->seq overflowed. But check it
7074 * anyway to be safe.
7076 if (ret == TRACE_TYPE_PARTIAL_LINE) {
7077 iter->seq.seq.len = save_len;
7081 count = trace_seq_used(&iter->seq) - save_len;
7084 iter->seq.seq.len = save_len;
7088 if (ret != TRACE_TYPE_NO_CONSUME)
7089 trace_consume(iter);
7091 if (!trace_find_next_entry_inc(iter)) {
7101 static ssize_t tracing_splice_read_pipe(struct file *filp,
7103 struct pipe_inode_info *pipe,
7107 struct page *pages_def[PIPE_DEF_BUFFERS];
7108 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7109 struct trace_iterator *iter = filp->private_data;
7110 struct splice_pipe_desc spd = {
7112 .partial = partial_def,
7113 .nr_pages = 0, /* This gets updated below. */
7114 .nr_pages_max = PIPE_DEF_BUFFERS,
7115 .ops = &default_pipe_buf_ops,
7116 .spd_release = tracing_spd_release_pipe,
7122 if (splice_grow_spd(pipe, &spd))
7125 mutex_lock(&iter->mutex);
7127 if (iter->trace->splice_read) {
7128 ret = iter->trace->splice_read(iter, filp,
7129 ppos, pipe, len, flags);
7134 ret = tracing_wait_pipe(filp);
7138 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
7143 trace_event_read_lock();
7144 trace_access_lock(iter->cpu_file);
7146 /* Fill as many pages as possible. */
7147 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
7148 spd.pages[i] = alloc_page(GFP_KERNEL);
7152 rem = tracing_fill_pipe_page(rem, iter);
7154 /* Copy the data into the page, so we can start over. */
7155 ret = trace_seq_to_buffer(&iter->seq,
7156 page_address(spd.pages[i]),
7157 trace_seq_used(&iter->seq));
7159 __free_page(spd.pages[i]);
7162 spd.partial[i].offset = 0;
7163 spd.partial[i].len = trace_seq_used(&iter->seq);
7165 trace_seq_init(&iter->seq);
7168 trace_access_unlock(iter->cpu_file);
7169 trace_event_read_unlock();
7170 mutex_unlock(&iter->mutex);
7175 ret = splice_to_pipe(pipe, &spd);
7179 splice_shrink_spd(&spd);
7183 mutex_unlock(&iter->mutex);
7188 tracing_entries_read(struct file *filp, char __user *ubuf,
7189 size_t cnt, loff_t *ppos)
7191 struct inode *inode = file_inode(filp);
7192 struct trace_array *tr = inode->i_private;
7193 int cpu = tracing_get_cpu(inode);
7198 mutex_lock(&trace_types_lock);
7200 if (cpu == RING_BUFFER_ALL_CPUS) {
7201 int cpu, buf_size_same;
7206 /* check if all cpu sizes are same */
7207 for_each_tracing_cpu(cpu) {
7208 /* fill in the size from first enabled cpu */
7210 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7211 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7217 if (buf_size_same) {
7218 if (!tr->ring_buffer_expanded)
7219 r = sprintf(buf, "%lu (expanded: %lu)\n",
7221 trace_buf_size >> 10);
7223 r = sprintf(buf, "%lu\n", size >> 10);
7225 r = sprintf(buf, "X\n");
7227 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7229 mutex_unlock(&trace_types_lock);
7231 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7236 tracing_entries_write(struct file *filp, const char __user *ubuf,
7237 size_t cnt, loff_t *ppos)
7239 struct inode *inode = file_inode(filp);
7240 struct trace_array *tr = inode->i_private;
7244 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7248 /* must have at least 1 entry */
7252 /* value is in KB */
7254 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7264 tracing_total_entries_read(struct file *filp, char __user *ubuf,
7265 size_t cnt, loff_t *ppos)
7267 struct trace_array *tr = filp->private_data;
7270 unsigned long size = 0, expanded_size = 0;
7272 mutex_lock(&trace_types_lock);
7273 for_each_tracing_cpu(cpu) {
7274 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7275 if (!tr->ring_buffer_expanded)
7276 expanded_size += trace_buf_size >> 10;
7278 if (tr->ring_buffer_expanded)
7279 r = sprintf(buf, "%lu\n", size);
7281 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7282 mutex_unlock(&trace_types_lock);
7284 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7288 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7289 size_t cnt, loff_t *ppos)
7292 * There is no need to read what the user has written, this function
7293 * is just to make sure that there is no error when "echo" is used
7302 tracing_free_buffer_release(struct inode *inode, struct file *filp)
7304 struct trace_array *tr = inode->i_private;
7306 /* disable tracing ? */
7307 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7308 tracer_tracing_off(tr);
7309 /* resize the ring buffer to 0 */
7310 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7312 trace_array_put(tr);
7317 #define TRACE_MARKER_MAX_SIZE 4096
7320 tracing_mark_write(struct file *filp, const char __user *ubuf,
7321 size_t cnt, loff_t *fpos)
7323 struct trace_array *tr = filp->private_data;
7324 struct ring_buffer_event *event;
7325 enum event_trigger_type tt = ETT_NONE;
7326 struct trace_buffer *buffer;
7327 struct print_entry *entry;
7333 /* Used in tracing_mark_raw_write() as well */
7334 #define FAULTED_STR "<faulted>"
7335 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7337 if (tracing_disabled)
7340 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7343 if ((ssize_t)cnt < 0)
7346 if (cnt > TRACE_MARKER_MAX_SIZE)
7347 cnt = TRACE_MARKER_MAX_SIZE;
7349 meta_size = sizeof(*entry) + 2; /* add '\0' and possible '\n' */
7351 size = cnt + meta_size;
7353 /* If less than "<faulted>", then make sure we can still add that */
7354 if (cnt < FAULTED_SIZE)
7355 size += FAULTED_SIZE - cnt;
7357 buffer = tr->array_buffer.buffer;
7358 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7360 if (unlikely(!event)) {
7362 * If the size was greater than what was allowed, then
7363 * make it smaller and try again.
7365 if (size > ring_buffer_max_event_size(buffer)) {
7366 /* cnt < FAULTED size should never be bigger than max */
7367 if (WARN_ON_ONCE(cnt < FAULTED_SIZE))
7369 cnt = ring_buffer_max_event_size(buffer) - meta_size;
7370 /* The above should only happen once */
7371 if (WARN_ON_ONCE(cnt + meta_size == size))
7376 /* Ring buffer disabled, return as if not open for write */
7380 entry = ring_buffer_event_data(event);
7381 entry->ip = _THIS_IP_;
7383 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7385 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7391 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7392 /* do not add \n before testing triggers, but add \0 */
7393 entry->buf[cnt] = '\0';
7394 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7397 if (entry->buf[cnt - 1] != '\n') {
7398 entry->buf[cnt] = '\n';
7399 entry->buf[cnt + 1] = '\0';
7401 entry->buf[cnt] = '\0';
7403 if (static_branch_unlikely(&trace_marker_exports_enabled))
7404 ftrace_exports(event, TRACE_EXPORT_MARKER);
7405 __buffer_unlock_commit(buffer, event);
7408 event_triggers_post_call(tr->trace_marker_file, tt);
7414 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7415 size_t cnt, loff_t *fpos)
7417 struct trace_array *tr = filp->private_data;
7418 struct ring_buffer_event *event;
7419 struct trace_buffer *buffer;
7420 struct raw_data_entry *entry;
7425 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7427 if (tracing_disabled)
7430 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7433 /* The marker must at least have a tag id */
7434 if (cnt < sizeof(unsigned int))
7437 size = sizeof(*entry) + cnt;
7438 if (cnt < FAULT_SIZE_ID)
7439 size += FAULT_SIZE_ID - cnt;
7441 buffer = tr->array_buffer.buffer;
7443 if (size > ring_buffer_max_event_size(buffer))
7446 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7449 /* Ring buffer disabled, return as if not open for write */
7452 entry = ring_buffer_event_data(event);
7454 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7457 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7462 __buffer_unlock_commit(buffer, event);
7467 static int tracing_clock_show(struct seq_file *m, void *v)
7469 struct trace_array *tr = m->private;
7472 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7474 "%s%s%s%s", i ? " " : "",
7475 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7476 i == tr->clock_id ? "]" : "");
7482 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7486 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7487 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7490 if (i == ARRAY_SIZE(trace_clocks))
7493 mutex_lock(&trace_types_lock);
7497 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7500 * New clock may not be consistent with the previous clock.
7501 * Reset the buffer so that it doesn't have incomparable timestamps.
7503 tracing_reset_online_cpus(&tr->array_buffer);
7505 #ifdef CONFIG_TRACER_MAX_TRACE
7506 if (tr->max_buffer.buffer)
7507 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7508 tracing_reset_online_cpus(&tr->max_buffer);
7511 mutex_unlock(&trace_types_lock);
7516 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7517 size_t cnt, loff_t *fpos)
7519 struct seq_file *m = filp->private_data;
7520 struct trace_array *tr = m->private;
7522 const char *clockstr;
7525 if (cnt >= sizeof(buf))
7528 if (copy_from_user(buf, ubuf, cnt))
7533 clockstr = strstrip(buf);
7535 ret = tracing_set_clock(tr, clockstr);
7544 static int tracing_clock_open(struct inode *inode, struct file *file)
7546 struct trace_array *tr = inode->i_private;
7549 ret = tracing_check_open_get_tr(tr);
7553 ret = single_open(file, tracing_clock_show, inode->i_private);
7555 trace_array_put(tr);
7560 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7562 struct trace_array *tr = m->private;
7564 mutex_lock(&trace_types_lock);
7566 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7567 seq_puts(m, "delta [absolute]\n");
7569 seq_puts(m, "[delta] absolute\n");
7571 mutex_unlock(&trace_types_lock);
7576 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7578 struct trace_array *tr = inode->i_private;
7581 ret = tracing_check_open_get_tr(tr);
7585 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7587 trace_array_put(tr);
7592 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7594 if (rbe == this_cpu_read(trace_buffered_event))
7595 return ring_buffer_time_stamp(buffer);
7597 return ring_buffer_event_time_stamp(buffer, rbe);
7601 * Set or disable using the per CPU trace_buffer_event when possible.
7603 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7607 mutex_lock(&trace_types_lock);
7609 if (set && tr->no_filter_buffering_ref++)
7613 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7618 --tr->no_filter_buffering_ref;
7621 mutex_unlock(&trace_types_lock);
7626 struct ftrace_buffer_info {
7627 struct trace_iterator iter;
7629 unsigned int spare_cpu;
7630 unsigned int spare_size;
7634 #ifdef CONFIG_TRACER_SNAPSHOT
7635 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7637 struct trace_array *tr = inode->i_private;
7638 struct trace_iterator *iter;
7642 ret = tracing_check_open_get_tr(tr);
7646 if (file->f_mode & FMODE_READ) {
7647 iter = __tracing_open(inode, file, true);
7649 ret = PTR_ERR(iter);
7651 /* Writes still need the seq_file to hold the private data */
7653 m = kzalloc(sizeof(*m), GFP_KERNEL);
7656 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7664 iter->array_buffer = &tr->max_buffer;
7665 iter->cpu_file = tracing_get_cpu(inode);
7667 file->private_data = m;
7671 trace_array_put(tr);
7676 static void tracing_swap_cpu_buffer(void *tr)
7678 update_max_tr_single((struct trace_array *)tr, current, smp_processor_id());
7682 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7685 struct seq_file *m = filp->private_data;
7686 struct trace_iterator *iter = m->private;
7687 struct trace_array *tr = iter->tr;
7691 ret = tracing_update_buffers(tr);
7695 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7699 mutex_lock(&trace_types_lock);
7701 if (tr->current_trace->use_max_tr) {
7706 local_irq_disable();
7707 arch_spin_lock(&tr->max_lock);
7708 if (tr->cond_snapshot)
7710 arch_spin_unlock(&tr->max_lock);
7717 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7721 if (tr->allocated_snapshot)
7725 /* Only allow per-cpu swap if the ring buffer supports it */
7726 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7727 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7732 if (tr->allocated_snapshot)
7733 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7734 &tr->array_buffer, iter->cpu_file);
7736 ret = tracing_alloc_snapshot_instance(tr);
7739 /* Now, we're going to swap */
7740 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
7741 local_irq_disable();
7742 update_max_tr(tr, current, smp_processor_id(), NULL);
7745 smp_call_function_single(iter->cpu_file, tracing_swap_cpu_buffer,
7750 if (tr->allocated_snapshot) {
7751 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7752 tracing_reset_online_cpus(&tr->max_buffer);
7754 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7764 mutex_unlock(&trace_types_lock);
7768 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7770 struct seq_file *m = file->private_data;
7773 ret = tracing_release(inode, file);
7775 if (file->f_mode & FMODE_READ)
7778 /* If write only, the seq_file is just a stub */
7786 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7787 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7788 size_t count, loff_t *ppos);
7789 static int tracing_buffers_release(struct inode *inode, struct file *file);
7790 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7791 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7793 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7795 struct ftrace_buffer_info *info;
7798 /* The following checks for tracefs lockdown */
7799 ret = tracing_buffers_open(inode, filp);
7803 info = filp->private_data;
7805 if (info->iter.trace->use_max_tr) {
7806 tracing_buffers_release(inode, filp);
7810 info->iter.snapshot = true;
7811 info->iter.array_buffer = &info->iter.tr->max_buffer;
7816 #endif /* CONFIG_TRACER_SNAPSHOT */
7819 static const struct file_operations tracing_thresh_fops = {
7820 .open = tracing_open_generic,
7821 .read = tracing_thresh_read,
7822 .write = tracing_thresh_write,
7823 .llseek = generic_file_llseek,
7826 #ifdef CONFIG_TRACER_MAX_TRACE
7827 static const struct file_operations tracing_max_lat_fops = {
7828 .open = tracing_open_generic_tr,
7829 .read = tracing_max_lat_read,
7830 .write = tracing_max_lat_write,
7831 .llseek = generic_file_llseek,
7832 .release = tracing_release_generic_tr,
7836 static const struct file_operations set_tracer_fops = {
7837 .open = tracing_open_generic_tr,
7838 .read = tracing_set_trace_read,
7839 .write = tracing_set_trace_write,
7840 .llseek = generic_file_llseek,
7841 .release = tracing_release_generic_tr,
7844 static const struct file_operations tracing_pipe_fops = {
7845 .open = tracing_open_pipe,
7846 .poll = tracing_poll_pipe,
7847 .read = tracing_read_pipe,
7848 .splice_read = tracing_splice_read_pipe,
7849 .release = tracing_release_pipe,
7850 .llseek = no_llseek,
7853 static const struct file_operations tracing_entries_fops = {
7854 .open = tracing_open_generic_tr,
7855 .read = tracing_entries_read,
7856 .write = tracing_entries_write,
7857 .llseek = generic_file_llseek,
7858 .release = tracing_release_generic_tr,
7861 static const struct file_operations tracing_total_entries_fops = {
7862 .open = tracing_open_generic_tr,
7863 .read = tracing_total_entries_read,
7864 .llseek = generic_file_llseek,
7865 .release = tracing_release_generic_tr,
7868 static const struct file_operations tracing_free_buffer_fops = {
7869 .open = tracing_open_generic_tr,
7870 .write = tracing_free_buffer_write,
7871 .release = tracing_free_buffer_release,
7874 static const struct file_operations tracing_mark_fops = {
7875 .open = tracing_mark_open,
7876 .write = tracing_mark_write,
7877 .release = tracing_release_generic_tr,
7880 static const struct file_operations tracing_mark_raw_fops = {
7881 .open = tracing_mark_open,
7882 .write = tracing_mark_raw_write,
7883 .release = tracing_release_generic_tr,
7886 static const struct file_operations trace_clock_fops = {
7887 .open = tracing_clock_open,
7889 .llseek = seq_lseek,
7890 .release = tracing_single_release_tr,
7891 .write = tracing_clock_write,
7894 static const struct file_operations trace_time_stamp_mode_fops = {
7895 .open = tracing_time_stamp_mode_open,
7897 .llseek = seq_lseek,
7898 .release = tracing_single_release_tr,
7901 #ifdef CONFIG_TRACER_SNAPSHOT
7902 static const struct file_operations snapshot_fops = {
7903 .open = tracing_snapshot_open,
7905 .write = tracing_snapshot_write,
7906 .llseek = tracing_lseek,
7907 .release = tracing_snapshot_release,
7910 static const struct file_operations snapshot_raw_fops = {
7911 .open = snapshot_raw_open,
7912 .read = tracing_buffers_read,
7913 .release = tracing_buffers_release,
7914 .splice_read = tracing_buffers_splice_read,
7915 .llseek = no_llseek,
7918 #endif /* CONFIG_TRACER_SNAPSHOT */
7921 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7922 * @filp: The active open file structure
7923 * @ubuf: The userspace provided buffer to read value into
7924 * @cnt: The maximum number of bytes to read
7925 * @ppos: The current "file" position
7927 * This function implements the write interface for a struct trace_min_max_param.
7928 * The filp->private_data must point to a trace_min_max_param structure that
7929 * defines where to write the value, the min and the max acceptable values,
7930 * and a lock to protect the write.
7933 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7935 struct trace_min_max_param *param = filp->private_data;
7942 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7947 mutex_lock(param->lock);
7949 if (param->min && val < *param->min)
7952 if (param->max && val > *param->max)
7959 mutex_unlock(param->lock);
7968 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7969 * @filp: The active open file structure
7970 * @ubuf: The userspace provided buffer to read value into
7971 * @cnt: The maximum number of bytes to read
7972 * @ppos: The current "file" position
7974 * This function implements the read interface for a struct trace_min_max_param.
7975 * The filp->private_data must point to a trace_min_max_param struct with valid
7979 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7981 struct trace_min_max_param *param = filp->private_data;
7982 char buf[U64_STR_SIZE];
7991 if (cnt > sizeof(buf))
7994 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7996 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7999 const struct file_operations trace_min_max_fops = {
8000 .open = tracing_open_generic,
8001 .read = trace_min_max_read,
8002 .write = trace_min_max_write,
8005 #define TRACING_LOG_ERRS_MAX 8
8006 #define TRACING_LOG_LOC_MAX 128
8008 #define CMD_PREFIX " Command: "
8011 const char **errs; /* ptr to loc-specific array of err strings */
8012 u8 type; /* index into errs -> specific err string */
8013 u16 pos; /* caret position */
8017 struct tracing_log_err {
8018 struct list_head list;
8019 struct err_info info;
8020 char loc[TRACING_LOG_LOC_MAX]; /* err location */
8021 char *cmd; /* what caused err */
8024 static DEFINE_MUTEX(tracing_err_log_lock);
8026 static struct tracing_log_err *alloc_tracing_log_err(int len)
8028 struct tracing_log_err *err;
8030 err = kzalloc(sizeof(*err), GFP_KERNEL);
8032 return ERR_PTR(-ENOMEM);
8034 err->cmd = kzalloc(len, GFP_KERNEL);
8037 return ERR_PTR(-ENOMEM);
8043 static void free_tracing_log_err(struct tracing_log_err *err)
8049 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
8052 struct tracing_log_err *err;
8055 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
8056 err = alloc_tracing_log_err(len);
8057 if (PTR_ERR(err) != -ENOMEM)
8058 tr->n_err_log_entries++;
8062 cmd = kzalloc(len, GFP_KERNEL);
8064 return ERR_PTR(-ENOMEM);
8065 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
8068 list_del(&err->list);
8074 * err_pos - find the position of a string within a command for error careting
8075 * @cmd: The tracing command that caused the error
8076 * @str: The string to position the caret at within @cmd
8078 * Finds the position of the first occurrence of @str within @cmd. The
8079 * return value can be passed to tracing_log_err() for caret placement
8082 * Returns the index within @cmd of the first occurrence of @str or 0
8083 * if @str was not found.
8085 unsigned int err_pos(char *cmd, const char *str)
8089 if (WARN_ON(!strlen(cmd)))
8092 found = strstr(cmd, str);
8100 * tracing_log_err - write an error to the tracing error log
8101 * @tr: The associated trace array for the error (NULL for top level array)
8102 * @loc: A string describing where the error occurred
8103 * @cmd: The tracing command that caused the error
8104 * @errs: The array of loc-specific static error strings
8105 * @type: The index into errs[], which produces the specific static err string
8106 * @pos: The position the caret should be placed in the cmd
8108 * Writes an error into tracing/error_log of the form:
8110 * <loc>: error: <text>
8114 * tracing/error_log is a small log file containing the last
8115 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
8116 * unless there has been a tracing error, and the error log can be
8117 * cleared and have its memory freed by writing the empty string in
8118 * truncation mode to it i.e. echo > tracing/error_log.
8120 * NOTE: the @errs array along with the @type param are used to
8121 * produce a static error string - this string is not copied and saved
8122 * when the error is logged - only a pointer to it is saved. See
8123 * existing callers for examples of how static strings are typically
8124 * defined for use with tracing_log_err().
8126 void tracing_log_err(struct trace_array *tr,
8127 const char *loc, const char *cmd,
8128 const char **errs, u8 type, u16 pos)
8130 struct tracing_log_err *err;
8136 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
8138 mutex_lock(&tracing_err_log_lock);
8139 err = get_tracing_log_err(tr, len);
8140 if (PTR_ERR(err) == -ENOMEM) {
8141 mutex_unlock(&tracing_err_log_lock);
8145 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
8146 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
8148 err->info.errs = errs;
8149 err->info.type = type;
8150 err->info.pos = pos;
8151 err->info.ts = local_clock();
8153 list_add_tail(&err->list, &tr->err_log);
8154 mutex_unlock(&tracing_err_log_lock);
8157 static void clear_tracing_err_log(struct trace_array *tr)
8159 struct tracing_log_err *err, *next;
8161 mutex_lock(&tracing_err_log_lock);
8162 list_for_each_entry_safe(err, next, &tr->err_log, list) {
8163 list_del(&err->list);
8164 free_tracing_log_err(err);
8167 tr->n_err_log_entries = 0;
8168 mutex_unlock(&tracing_err_log_lock);
8171 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
8173 struct trace_array *tr = m->private;
8175 mutex_lock(&tracing_err_log_lock);
8177 return seq_list_start(&tr->err_log, *pos);
8180 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
8182 struct trace_array *tr = m->private;
8184 return seq_list_next(v, &tr->err_log, pos);
8187 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
8189 mutex_unlock(&tracing_err_log_lock);
8192 static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
8196 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
8198 for (i = 0; i < pos; i++)
8203 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
8205 struct tracing_log_err *err = v;
8208 const char *err_text = err->info.errs[err->info.type];
8209 u64 sec = err->info.ts;
8212 nsec = do_div(sec, NSEC_PER_SEC);
8213 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
8214 err->loc, err_text);
8215 seq_printf(m, "%s", err->cmd);
8216 tracing_err_log_show_pos(m, err->info.pos);
8222 static const struct seq_operations tracing_err_log_seq_ops = {
8223 .start = tracing_err_log_seq_start,
8224 .next = tracing_err_log_seq_next,
8225 .stop = tracing_err_log_seq_stop,
8226 .show = tracing_err_log_seq_show
8229 static int tracing_err_log_open(struct inode *inode, struct file *file)
8231 struct trace_array *tr = inode->i_private;
8234 ret = tracing_check_open_get_tr(tr);
8238 /* If this file was opened for write, then erase contents */
8239 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8240 clear_tracing_err_log(tr);
8242 if (file->f_mode & FMODE_READ) {
8243 ret = seq_open(file, &tracing_err_log_seq_ops);
8245 struct seq_file *m = file->private_data;
8248 trace_array_put(tr);
8254 static ssize_t tracing_err_log_write(struct file *file,
8255 const char __user *buffer,
8256 size_t count, loff_t *ppos)
8261 static int tracing_err_log_release(struct inode *inode, struct file *file)
8263 struct trace_array *tr = inode->i_private;
8265 trace_array_put(tr);
8267 if (file->f_mode & FMODE_READ)
8268 seq_release(inode, file);
8273 static const struct file_operations tracing_err_log_fops = {
8274 .open = tracing_err_log_open,
8275 .write = tracing_err_log_write,
8277 .llseek = tracing_lseek,
8278 .release = tracing_err_log_release,
8281 static int tracing_buffers_open(struct inode *inode, struct file *filp)
8283 struct trace_array *tr = inode->i_private;
8284 struct ftrace_buffer_info *info;
8287 ret = tracing_check_open_get_tr(tr);
8291 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8293 trace_array_put(tr);
8297 mutex_lock(&trace_types_lock);
8300 info->iter.cpu_file = tracing_get_cpu(inode);
8301 info->iter.trace = tr->current_trace;
8302 info->iter.array_buffer = &tr->array_buffer;
8304 /* Force reading ring buffer for first read */
8305 info->read = (unsigned int)-1;
8307 filp->private_data = info;
8311 mutex_unlock(&trace_types_lock);
8313 ret = nonseekable_open(inode, filp);
8315 trace_array_put(tr);
8321 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8323 struct ftrace_buffer_info *info = filp->private_data;
8324 struct trace_iterator *iter = &info->iter;
8326 return trace_poll(iter, filp, poll_table);
8330 tracing_buffers_read(struct file *filp, char __user *ubuf,
8331 size_t count, loff_t *ppos)
8333 struct ftrace_buffer_info *info = filp->private_data;
8334 struct trace_iterator *iter = &info->iter;
8343 #ifdef CONFIG_TRACER_MAX_TRACE
8344 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8348 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8350 /* Make sure the spare matches the current sub buffer size */
8352 if (page_size != info->spare_size) {
8353 ring_buffer_free_read_page(iter->array_buffer->buffer,
8354 info->spare_cpu, info->spare);
8360 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8362 if (IS_ERR(info->spare)) {
8363 ret = PTR_ERR(info->spare);
8366 info->spare_cpu = iter->cpu_file;
8367 info->spare_size = page_size;
8373 /* Do we have previous read data to read? */
8374 if (info->read < page_size)
8378 trace_access_lock(iter->cpu_file);
8379 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8383 trace_access_unlock(iter->cpu_file);
8386 if (trace_empty(iter)) {
8387 if ((filp->f_flags & O_NONBLOCK))
8390 ret = wait_on_pipe(iter, 0);
8401 size = page_size - info->read;
8404 trace_data = ring_buffer_read_page_data(info->spare);
8405 ret = copy_to_user(ubuf, trace_data + info->read, size);
8417 static int tracing_buffers_flush(struct file *file, fl_owner_t id)
8419 struct ftrace_buffer_info *info = file->private_data;
8420 struct trace_iterator *iter = &info->iter;
8422 iter->closed = true;
8423 /* Make sure the waiters see the new wait_index */
8424 (void)atomic_fetch_inc_release(&iter->wait_index);
8426 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8431 static int tracing_buffers_release(struct inode *inode, struct file *file)
8433 struct ftrace_buffer_info *info = file->private_data;
8434 struct trace_iterator *iter = &info->iter;
8436 mutex_lock(&trace_types_lock);
8438 iter->tr->trace_ref--;
8440 __trace_array_put(iter->tr);
8443 ring_buffer_free_read_page(iter->array_buffer->buffer,
8444 info->spare_cpu, info->spare);
8447 mutex_unlock(&trace_types_lock);
8453 struct trace_buffer *buffer;
8456 refcount_t refcount;
8459 static void buffer_ref_release(struct buffer_ref *ref)
8461 if (!refcount_dec_and_test(&ref->refcount))
8463 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8467 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8468 struct pipe_buffer *buf)
8470 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8472 buffer_ref_release(ref);
8476 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8477 struct pipe_buffer *buf)
8479 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8481 if (refcount_read(&ref->refcount) > INT_MAX/2)
8484 refcount_inc(&ref->refcount);
8488 /* Pipe buffer operations for a buffer. */
8489 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8490 .release = buffer_pipe_buf_release,
8491 .get = buffer_pipe_buf_get,
8495 * Callback from splice_to_pipe(), if we need to release some pages
8496 * at the end of the spd in case we error'ed out in filling the pipe.
8498 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8500 struct buffer_ref *ref =
8501 (struct buffer_ref *)spd->partial[i].private;
8503 buffer_ref_release(ref);
8504 spd->partial[i].private = 0;
8508 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8509 struct pipe_inode_info *pipe, size_t len,
8512 struct ftrace_buffer_info *info = file->private_data;
8513 struct trace_iterator *iter = &info->iter;
8514 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8515 struct page *pages_def[PIPE_DEF_BUFFERS];
8516 struct splice_pipe_desc spd = {
8518 .partial = partial_def,
8519 .nr_pages_max = PIPE_DEF_BUFFERS,
8520 .ops = &buffer_pipe_buf_ops,
8521 .spd_release = buffer_spd_release,
8523 struct buffer_ref *ref;
8529 #ifdef CONFIG_TRACER_MAX_TRACE
8530 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8534 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8535 if (*ppos & (page_size - 1))
8538 if (len & (page_size - 1)) {
8539 if (len < page_size)
8541 len &= (~(page_size - 1));
8544 if (splice_grow_spd(pipe, &spd))
8548 trace_access_lock(iter->cpu_file);
8549 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8551 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= page_size) {
8555 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8561 refcount_set(&ref->refcount, 1);
8562 ref->buffer = iter->array_buffer->buffer;
8563 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8564 if (IS_ERR(ref->page)) {
8565 ret = PTR_ERR(ref->page);
8570 ref->cpu = iter->cpu_file;
8572 r = ring_buffer_read_page(ref->buffer, ref->page,
8573 len, iter->cpu_file, 1);
8575 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8581 page = virt_to_page(ring_buffer_read_page_data(ref->page));
8583 spd.pages[i] = page;
8584 spd.partial[i].len = page_size;
8585 spd.partial[i].offset = 0;
8586 spd.partial[i].private = (unsigned long)ref;
8590 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8593 trace_access_unlock(iter->cpu_file);
8596 /* did we read anything? */
8597 if (!spd.nr_pages) {
8606 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8609 ret = wait_on_pipe(iter, iter->snapshot ? 0 : iter->tr->buffer_percent);
8613 /* No need to wait after waking up when tracing is off */
8614 if (!tracer_tracing_is_on(iter->tr))
8617 /* Iterate one more time to collect any new data then exit */
8623 ret = splice_to_pipe(pipe, &spd);
8625 splice_shrink_spd(&spd);
8630 /* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8631 static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8633 struct ftrace_buffer_info *info = file->private_data;
8634 struct trace_iterator *iter = &info->iter;
8637 return -ENOIOCTLCMD;
8639 mutex_lock(&trace_types_lock);
8641 /* Make sure the waiters see the new wait_index */
8642 (void)atomic_fetch_inc_release(&iter->wait_index);
8644 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8646 mutex_unlock(&trace_types_lock);
8650 static const struct file_operations tracing_buffers_fops = {
8651 .open = tracing_buffers_open,
8652 .read = tracing_buffers_read,
8653 .poll = tracing_buffers_poll,
8654 .release = tracing_buffers_release,
8655 .flush = tracing_buffers_flush,
8656 .splice_read = tracing_buffers_splice_read,
8657 .unlocked_ioctl = tracing_buffers_ioctl,
8658 .llseek = no_llseek,
8662 tracing_stats_read(struct file *filp, char __user *ubuf,
8663 size_t count, loff_t *ppos)
8665 struct inode *inode = file_inode(filp);
8666 struct trace_array *tr = inode->i_private;
8667 struct array_buffer *trace_buf = &tr->array_buffer;
8668 int cpu = tracing_get_cpu(inode);
8669 struct trace_seq *s;
8671 unsigned long long t;
8672 unsigned long usec_rem;
8674 s = kmalloc(sizeof(*s), GFP_KERNEL);
8680 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8681 trace_seq_printf(s, "entries: %ld\n", cnt);
8683 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8684 trace_seq_printf(s, "overrun: %ld\n", cnt);
8686 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8687 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8689 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8690 trace_seq_printf(s, "bytes: %ld\n", cnt);
8692 if (trace_clocks[tr->clock_id].in_ns) {
8693 /* local or global for trace_clock */
8694 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8695 usec_rem = do_div(t, USEC_PER_SEC);
8696 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8699 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8700 usec_rem = do_div(t, USEC_PER_SEC);
8701 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8703 /* counter or tsc mode for trace_clock */
8704 trace_seq_printf(s, "oldest event ts: %llu\n",
8705 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8707 trace_seq_printf(s, "now ts: %llu\n",
8708 ring_buffer_time_stamp(trace_buf->buffer));
8711 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8712 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8714 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8715 trace_seq_printf(s, "read events: %ld\n", cnt);
8717 count = simple_read_from_buffer(ubuf, count, ppos,
8718 s->buffer, trace_seq_used(s));
8725 static const struct file_operations tracing_stats_fops = {
8726 .open = tracing_open_generic_tr,
8727 .read = tracing_stats_read,
8728 .llseek = generic_file_llseek,
8729 .release = tracing_release_generic_tr,
8732 #ifdef CONFIG_DYNAMIC_FTRACE
8735 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8736 size_t cnt, loff_t *ppos)
8742 /* 256 should be plenty to hold the amount needed */
8743 buf = kmalloc(256, GFP_KERNEL);
8747 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8748 ftrace_update_tot_cnt,
8749 ftrace_number_of_pages,
8750 ftrace_number_of_groups);
8752 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8757 static const struct file_operations tracing_dyn_info_fops = {
8758 .open = tracing_open_generic,
8759 .read = tracing_read_dyn_info,
8760 .llseek = generic_file_llseek,
8762 #endif /* CONFIG_DYNAMIC_FTRACE */
8764 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8766 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8767 struct trace_array *tr, struct ftrace_probe_ops *ops,
8770 tracing_snapshot_instance(tr);
8774 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8775 struct trace_array *tr, struct ftrace_probe_ops *ops,
8778 struct ftrace_func_mapper *mapper = data;
8782 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8792 tracing_snapshot_instance(tr);
8796 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8797 struct ftrace_probe_ops *ops, void *data)
8799 struct ftrace_func_mapper *mapper = data;
8802 seq_printf(m, "%ps:", (void *)ip);
8804 seq_puts(m, "snapshot");
8807 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8810 seq_printf(m, ":count=%ld\n", *count);
8812 seq_puts(m, ":unlimited\n");
8818 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8819 unsigned long ip, void *init_data, void **data)
8821 struct ftrace_func_mapper *mapper = *data;
8824 mapper = allocate_ftrace_func_mapper();
8830 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8834 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8835 unsigned long ip, void *data)
8837 struct ftrace_func_mapper *mapper = data;
8842 free_ftrace_func_mapper(mapper, NULL);
8846 ftrace_func_mapper_remove_ip(mapper, ip);
8849 static struct ftrace_probe_ops snapshot_probe_ops = {
8850 .func = ftrace_snapshot,
8851 .print = ftrace_snapshot_print,
8854 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8855 .func = ftrace_count_snapshot,
8856 .print = ftrace_snapshot_print,
8857 .init = ftrace_snapshot_init,
8858 .free = ftrace_snapshot_free,
8862 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8863 char *glob, char *cmd, char *param, int enable)
8865 struct ftrace_probe_ops *ops;
8866 void *count = (void *)-1;
8873 /* hash funcs only work with set_ftrace_filter */
8877 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8880 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8885 number = strsep(¶m, ":");
8887 if (!strlen(number))
8891 * We use the callback data field (which is a pointer)
8894 ret = kstrtoul(number, 0, (unsigned long *)&count);
8899 ret = tracing_alloc_snapshot_instance(tr);
8903 ret = register_ftrace_function_probe(glob, tr, ops, count);
8906 return ret < 0 ? ret : 0;
8909 static struct ftrace_func_command ftrace_snapshot_cmd = {
8911 .func = ftrace_trace_snapshot_callback,
8914 static __init int register_snapshot_cmd(void)
8916 return register_ftrace_command(&ftrace_snapshot_cmd);
8919 static inline __init int register_snapshot_cmd(void) { return 0; }
8920 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8922 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8924 if (WARN_ON(!tr->dir))
8925 return ERR_PTR(-ENODEV);
8927 /* Top directory uses NULL as the parent */
8928 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8931 /* All sub buffers have a descriptor */
8935 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8937 struct dentry *d_tracer;
8940 return tr->percpu_dir;
8942 d_tracer = tracing_get_dentry(tr);
8943 if (IS_ERR(d_tracer))
8946 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8948 MEM_FAIL(!tr->percpu_dir,
8949 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8951 return tr->percpu_dir;
8954 static struct dentry *
8955 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8956 void *data, long cpu, const struct file_operations *fops)
8958 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8960 if (ret) /* See tracing_get_cpu() */
8961 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8966 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8968 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8969 struct dentry *d_cpu;
8970 char cpu_dir[30]; /* 30 characters should be more than enough */
8975 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8976 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8978 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8982 /* per cpu trace_pipe */
8983 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8984 tr, cpu, &tracing_pipe_fops);
8987 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8988 tr, cpu, &tracing_fops);
8990 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8991 tr, cpu, &tracing_buffers_fops);
8993 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8994 tr, cpu, &tracing_stats_fops);
8996 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8997 tr, cpu, &tracing_entries_fops);
8999 #ifdef CONFIG_TRACER_SNAPSHOT
9000 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
9001 tr, cpu, &snapshot_fops);
9003 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
9004 tr, cpu, &snapshot_raw_fops);
9008 #ifdef CONFIG_FTRACE_SELFTEST
9009 /* Let selftest have access to static functions in this file */
9010 #include "trace_selftest.c"
9014 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
9017 struct trace_option_dentry *topt = filp->private_data;
9020 if (topt->flags->val & topt->opt->bit)
9025 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9029 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
9032 struct trace_option_dentry *topt = filp->private_data;
9036 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9040 if (val != 0 && val != 1)
9043 if (!!(topt->flags->val & topt->opt->bit) != val) {
9044 mutex_lock(&trace_types_lock);
9045 ret = __set_tracer_option(topt->tr, topt->flags,
9047 mutex_unlock(&trace_types_lock);
9057 static int tracing_open_options(struct inode *inode, struct file *filp)
9059 struct trace_option_dentry *topt = inode->i_private;
9062 ret = tracing_check_open_get_tr(topt->tr);
9066 filp->private_data = inode->i_private;
9070 static int tracing_release_options(struct inode *inode, struct file *file)
9072 struct trace_option_dentry *topt = file->private_data;
9074 trace_array_put(topt->tr);
9078 static const struct file_operations trace_options_fops = {
9079 .open = tracing_open_options,
9080 .read = trace_options_read,
9081 .write = trace_options_write,
9082 .llseek = generic_file_llseek,
9083 .release = tracing_release_options,
9087 * In order to pass in both the trace_array descriptor as well as the index
9088 * to the flag that the trace option file represents, the trace_array
9089 * has a character array of trace_flags_index[], which holds the index
9090 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
9091 * The address of this character array is passed to the flag option file
9092 * read/write callbacks.
9094 * In order to extract both the index and the trace_array descriptor,
9095 * get_tr_index() uses the following algorithm.
9099 * As the pointer itself contains the address of the index (remember
9102 * Then to get the trace_array descriptor, by subtracting that index
9103 * from the ptr, we get to the start of the index itself.
9105 * ptr - idx == &index[0]
9107 * Then a simple container_of() from that pointer gets us to the
9108 * trace_array descriptor.
9110 static void get_tr_index(void *data, struct trace_array **ptr,
9111 unsigned int *pindex)
9113 *pindex = *(unsigned char *)data;
9115 *ptr = container_of(data - *pindex, struct trace_array,
9120 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
9123 void *tr_index = filp->private_data;
9124 struct trace_array *tr;
9128 get_tr_index(tr_index, &tr, &index);
9130 if (tr->trace_flags & (1 << index))
9135 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9139 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
9142 void *tr_index = filp->private_data;
9143 struct trace_array *tr;
9148 get_tr_index(tr_index, &tr, &index);
9150 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9154 if (val != 0 && val != 1)
9157 mutex_lock(&event_mutex);
9158 mutex_lock(&trace_types_lock);
9159 ret = set_tracer_flag(tr, 1 << index, val);
9160 mutex_unlock(&trace_types_lock);
9161 mutex_unlock(&event_mutex);
9171 static const struct file_operations trace_options_core_fops = {
9172 .open = tracing_open_generic,
9173 .read = trace_options_core_read,
9174 .write = trace_options_core_write,
9175 .llseek = generic_file_llseek,
9178 struct dentry *trace_create_file(const char *name,
9180 struct dentry *parent,
9182 const struct file_operations *fops)
9186 ret = tracefs_create_file(name, mode, parent, data, fops);
9188 pr_warn("Could not create tracefs '%s' entry\n", name);
9194 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
9196 struct dentry *d_tracer;
9201 d_tracer = tracing_get_dentry(tr);
9202 if (IS_ERR(d_tracer))
9205 tr->options = tracefs_create_dir("options", d_tracer);
9207 pr_warn("Could not create tracefs directory 'options'\n");
9215 create_trace_option_file(struct trace_array *tr,
9216 struct trace_option_dentry *topt,
9217 struct tracer_flags *flags,
9218 struct tracer_opt *opt)
9220 struct dentry *t_options;
9222 t_options = trace_options_init_dentry(tr);
9226 topt->flags = flags;
9230 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
9231 t_options, topt, &trace_options_fops);
9236 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
9238 struct trace_option_dentry *topts;
9239 struct trace_options *tr_topts;
9240 struct tracer_flags *flags;
9241 struct tracer_opt *opts;
9248 flags = tracer->flags;
9250 if (!flags || !flags->opts)
9254 * If this is an instance, only create flags for tracers
9255 * the instance may have.
9257 if (!trace_ok_for_array(tracer, tr))
9260 for (i = 0; i < tr->nr_topts; i++) {
9261 /* Make sure there's no duplicate flags. */
9262 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
9268 for (cnt = 0; opts[cnt].name; cnt++)
9271 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
9275 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9282 tr->topts = tr_topts;
9283 tr->topts[tr->nr_topts].tracer = tracer;
9284 tr->topts[tr->nr_topts].topts = topts;
9287 for (cnt = 0; opts[cnt].name; cnt++) {
9288 create_trace_option_file(tr, &topts[cnt], flags,
9290 MEM_FAIL(topts[cnt].entry == NULL,
9291 "Failed to create trace option: %s",
9296 static struct dentry *
9297 create_trace_option_core_file(struct trace_array *tr,
9298 const char *option, long index)
9300 struct dentry *t_options;
9302 t_options = trace_options_init_dentry(tr);
9306 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9307 (void *)&tr->trace_flags_index[index],
9308 &trace_options_core_fops);
9311 static void create_trace_options_dir(struct trace_array *tr)
9313 struct dentry *t_options;
9314 bool top_level = tr == &global_trace;
9317 t_options = trace_options_init_dentry(tr);
9321 for (i = 0; trace_options[i]; i++) {
9323 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9324 create_trace_option_core_file(tr, trace_options[i], i);
9329 rb_simple_read(struct file *filp, char __user *ubuf,
9330 size_t cnt, loff_t *ppos)
9332 struct trace_array *tr = filp->private_data;
9336 r = tracer_tracing_is_on(tr);
9337 r = sprintf(buf, "%d\n", r);
9339 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9343 rb_simple_write(struct file *filp, const char __user *ubuf,
9344 size_t cnt, loff_t *ppos)
9346 struct trace_array *tr = filp->private_data;
9347 struct trace_buffer *buffer = tr->array_buffer.buffer;
9351 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9356 mutex_lock(&trace_types_lock);
9357 if (!!val == tracer_tracing_is_on(tr)) {
9358 val = 0; /* do nothing */
9360 tracer_tracing_on(tr);
9361 if (tr->current_trace->start)
9362 tr->current_trace->start(tr);
9364 tracer_tracing_off(tr);
9365 if (tr->current_trace->stop)
9366 tr->current_trace->stop(tr);
9367 /* Wake up any waiters */
9368 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9370 mutex_unlock(&trace_types_lock);
9378 static const struct file_operations rb_simple_fops = {
9379 .open = tracing_open_generic_tr,
9380 .read = rb_simple_read,
9381 .write = rb_simple_write,
9382 .release = tracing_release_generic_tr,
9383 .llseek = default_llseek,
9387 buffer_percent_read(struct file *filp, char __user *ubuf,
9388 size_t cnt, loff_t *ppos)
9390 struct trace_array *tr = filp->private_data;
9394 r = tr->buffer_percent;
9395 r = sprintf(buf, "%d\n", r);
9397 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9401 buffer_percent_write(struct file *filp, const char __user *ubuf,
9402 size_t cnt, loff_t *ppos)
9404 struct trace_array *tr = filp->private_data;
9408 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9415 tr->buffer_percent = val;
9422 static const struct file_operations buffer_percent_fops = {
9423 .open = tracing_open_generic_tr,
9424 .read = buffer_percent_read,
9425 .write = buffer_percent_write,
9426 .release = tracing_release_generic_tr,
9427 .llseek = default_llseek,
9431 buffer_subbuf_size_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
9433 struct trace_array *tr = filp->private_data;
9439 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9440 size = (PAGE_SIZE << order) / 1024;
9442 r = sprintf(buf, "%zd\n", size);
9444 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9448 buffer_subbuf_size_write(struct file *filp, const char __user *ubuf,
9449 size_t cnt, loff_t *ppos)
9451 struct trace_array *tr = filp->private_data;
9458 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9462 val *= 1024; /* value passed in is in KB */
9464 pages = DIV_ROUND_UP(val, PAGE_SIZE);
9465 order = fls(pages - 1);
9467 /* limit between 1 and 128 system pages */
9468 if (order < 0 || order > 7)
9471 /* Do not allow tracing while changing the order of the ring buffer */
9472 tracing_stop_tr(tr);
9474 old_order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9475 if (old_order == order)
9478 ret = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, order);
9482 #ifdef CONFIG_TRACER_MAX_TRACE
9484 if (!tr->allocated_snapshot)
9487 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
9489 /* Put back the old order */
9490 cnt = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, old_order);
9491 if (WARN_ON_ONCE(cnt)) {
9493 * AARGH! We are left with different orders!
9494 * The max buffer is our "snapshot" buffer.
9495 * When a tracer needs a snapshot (one of the
9496 * latency tracers), it swaps the max buffer
9497 * with the saved snap shot. We succeeded to
9498 * update the order of the main buffer, but failed to
9499 * update the order of the max buffer. But when we tried
9500 * to reset the main buffer to the original size, we
9501 * failed there too. This is very unlikely to
9502 * happen, but if it does, warn and kill all
9505 tracing_disabled = 1;
9515 tracing_start_tr(tr);
9519 static const struct file_operations buffer_subbuf_size_fops = {
9520 .open = tracing_open_generic_tr,
9521 .read = buffer_subbuf_size_read,
9522 .write = buffer_subbuf_size_write,
9523 .release = tracing_release_generic_tr,
9524 .llseek = default_llseek,
9527 static struct dentry *trace_instance_dir;
9530 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9533 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9535 enum ring_buffer_flags rb_flags;
9537 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9541 buf->buffer = ring_buffer_alloc(size, rb_flags);
9545 buf->data = alloc_percpu(struct trace_array_cpu);
9547 ring_buffer_free(buf->buffer);
9552 /* Allocate the first page for all buffers */
9553 set_buffer_entries(&tr->array_buffer,
9554 ring_buffer_size(tr->array_buffer.buffer, 0));
9559 static void free_trace_buffer(struct array_buffer *buf)
9562 ring_buffer_free(buf->buffer);
9564 free_percpu(buf->data);
9569 static int allocate_trace_buffers(struct trace_array *tr, int size)
9573 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9577 #ifdef CONFIG_TRACER_MAX_TRACE
9578 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9579 allocate_snapshot ? size : 1);
9580 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9581 free_trace_buffer(&tr->array_buffer);
9584 tr->allocated_snapshot = allocate_snapshot;
9586 allocate_snapshot = false;
9592 static void free_trace_buffers(struct trace_array *tr)
9597 free_trace_buffer(&tr->array_buffer);
9599 #ifdef CONFIG_TRACER_MAX_TRACE
9600 free_trace_buffer(&tr->max_buffer);
9604 static void init_trace_flags_index(struct trace_array *tr)
9608 /* Used by the trace options files */
9609 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9610 tr->trace_flags_index[i] = i;
9613 static void __update_tracer_options(struct trace_array *tr)
9617 for (t = trace_types; t; t = t->next)
9618 add_tracer_options(tr, t);
9621 static void update_tracer_options(struct trace_array *tr)
9623 mutex_lock(&trace_types_lock);
9624 tracer_options_updated = true;
9625 __update_tracer_options(tr);
9626 mutex_unlock(&trace_types_lock);
9629 /* Must have trace_types_lock held */
9630 struct trace_array *trace_array_find(const char *instance)
9632 struct trace_array *tr, *found = NULL;
9634 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9635 if (tr->name && strcmp(tr->name, instance) == 0) {
9644 struct trace_array *trace_array_find_get(const char *instance)
9646 struct trace_array *tr;
9648 mutex_lock(&trace_types_lock);
9649 tr = trace_array_find(instance);
9652 mutex_unlock(&trace_types_lock);
9657 static int trace_array_create_dir(struct trace_array *tr)
9661 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9665 ret = event_trace_add_tracer(tr->dir, tr);
9667 tracefs_remove(tr->dir);
9671 init_tracer_tracefs(tr, tr->dir);
9672 __update_tracer_options(tr);
9677 static struct trace_array *
9678 trace_array_create_systems(const char *name, const char *systems)
9680 struct trace_array *tr;
9684 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9686 return ERR_PTR(ret);
9688 tr->name = kstrdup(name, GFP_KERNEL);
9692 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9695 if (!zalloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9699 tr->system_names = kstrdup_const(systems, GFP_KERNEL);
9700 if (!tr->system_names)
9704 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9706 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9708 raw_spin_lock_init(&tr->start_lock);
9710 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9712 tr->current_trace = &nop_trace;
9714 INIT_LIST_HEAD(&tr->systems);
9715 INIT_LIST_HEAD(&tr->events);
9716 INIT_LIST_HEAD(&tr->hist_vars);
9717 INIT_LIST_HEAD(&tr->err_log);
9719 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9722 /* The ring buffer is defaultly expanded */
9723 trace_set_ring_buffer_expanded(tr);
9725 if (ftrace_allocate_ftrace_ops(tr) < 0)
9728 ftrace_init_trace_array(tr);
9730 init_trace_flags_index(tr);
9732 if (trace_instance_dir) {
9733 ret = trace_array_create_dir(tr);
9737 __trace_early_add_events(tr);
9739 list_add(&tr->list, &ftrace_trace_arrays);
9746 ftrace_free_ftrace_ops(tr);
9747 free_trace_buffers(tr);
9748 free_cpumask_var(tr->pipe_cpumask);
9749 free_cpumask_var(tr->tracing_cpumask);
9750 kfree_const(tr->system_names);
9754 return ERR_PTR(ret);
9757 static struct trace_array *trace_array_create(const char *name)
9759 return trace_array_create_systems(name, NULL);
9762 static int instance_mkdir(const char *name)
9764 struct trace_array *tr;
9767 mutex_lock(&event_mutex);
9768 mutex_lock(&trace_types_lock);
9771 if (trace_array_find(name))
9774 tr = trace_array_create(name);
9776 ret = PTR_ERR_OR_ZERO(tr);
9779 mutex_unlock(&trace_types_lock);
9780 mutex_unlock(&event_mutex);
9785 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9786 * @name: The name of the trace array to be looked up/created.
9787 * @systems: A list of systems to create event directories for (NULL for all)
9789 * Returns pointer to trace array with given name.
9790 * NULL, if it cannot be created.
9792 * NOTE: This function increments the reference counter associated with the
9793 * trace array returned. This makes sure it cannot be freed while in use.
9794 * Use trace_array_put() once the trace array is no longer needed.
9795 * If the trace_array is to be freed, trace_array_destroy() needs to
9796 * be called after the trace_array_put(), or simply let user space delete
9797 * it from the tracefs instances directory. But until the
9798 * trace_array_put() is called, user space can not delete it.
9801 struct trace_array *trace_array_get_by_name(const char *name, const char *systems)
9803 struct trace_array *tr;
9805 mutex_lock(&event_mutex);
9806 mutex_lock(&trace_types_lock);
9808 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9809 if (tr->name && strcmp(tr->name, name) == 0)
9813 tr = trace_array_create_systems(name, systems);
9821 mutex_unlock(&trace_types_lock);
9822 mutex_unlock(&event_mutex);
9825 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9827 static int __remove_instance(struct trace_array *tr)
9831 /* Reference counter for a newly created trace array = 1. */
9832 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9835 list_del(&tr->list);
9837 /* Disable all the flags that were enabled coming in */
9838 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9839 if ((1 << i) & ZEROED_TRACE_FLAGS)
9840 set_tracer_flag(tr, 1 << i, 0);
9843 tracing_set_nop(tr);
9844 clear_ftrace_function_probes(tr);
9845 event_trace_del_tracer(tr);
9846 ftrace_clear_pids(tr);
9847 ftrace_destroy_function_files(tr);
9848 tracefs_remove(tr->dir);
9849 free_percpu(tr->last_func_repeats);
9850 free_trace_buffers(tr);
9851 clear_tracing_err_log(tr);
9853 for (i = 0; i < tr->nr_topts; i++) {
9854 kfree(tr->topts[i].topts);
9858 free_cpumask_var(tr->pipe_cpumask);
9859 free_cpumask_var(tr->tracing_cpumask);
9860 kfree_const(tr->system_names);
9867 int trace_array_destroy(struct trace_array *this_tr)
9869 struct trace_array *tr;
9875 mutex_lock(&event_mutex);
9876 mutex_lock(&trace_types_lock);
9880 /* Making sure trace array exists before destroying it. */
9881 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9882 if (tr == this_tr) {
9883 ret = __remove_instance(tr);
9888 mutex_unlock(&trace_types_lock);
9889 mutex_unlock(&event_mutex);
9893 EXPORT_SYMBOL_GPL(trace_array_destroy);
9895 static int instance_rmdir(const char *name)
9897 struct trace_array *tr;
9900 mutex_lock(&event_mutex);
9901 mutex_lock(&trace_types_lock);
9904 tr = trace_array_find(name);
9906 ret = __remove_instance(tr);
9908 mutex_unlock(&trace_types_lock);
9909 mutex_unlock(&event_mutex);
9914 static __init void create_trace_instances(struct dentry *d_tracer)
9916 struct trace_array *tr;
9918 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9921 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9924 mutex_lock(&event_mutex);
9925 mutex_lock(&trace_types_lock);
9927 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9930 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9931 "Failed to create instance directory\n"))
9935 mutex_unlock(&trace_types_lock);
9936 mutex_unlock(&event_mutex);
9940 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9944 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9945 tr, &show_traces_fops);
9947 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9948 tr, &set_tracer_fops);
9950 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9951 tr, &tracing_cpumask_fops);
9953 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9954 tr, &tracing_iter_fops);
9956 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9959 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9960 tr, &tracing_pipe_fops);
9962 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9963 tr, &tracing_entries_fops);
9965 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9966 tr, &tracing_total_entries_fops);
9968 trace_create_file("free_buffer", 0200, d_tracer,
9969 tr, &tracing_free_buffer_fops);
9971 trace_create_file("trace_marker", 0220, d_tracer,
9972 tr, &tracing_mark_fops);
9974 tr->trace_marker_file = __find_event_file(tr, "ftrace", "print");
9976 trace_create_file("trace_marker_raw", 0220, d_tracer,
9977 tr, &tracing_mark_raw_fops);
9979 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9982 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9983 tr, &rb_simple_fops);
9985 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9986 &trace_time_stamp_mode_fops);
9988 tr->buffer_percent = 50;
9990 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9991 tr, &buffer_percent_fops);
9993 trace_create_file("buffer_subbuf_size_kb", TRACE_MODE_WRITE, d_tracer,
9994 tr, &buffer_subbuf_size_fops);
9996 create_trace_options_dir(tr);
9998 #ifdef CONFIG_TRACER_MAX_TRACE
9999 trace_create_maxlat_file(tr, d_tracer);
10002 if (ftrace_create_function_files(tr, d_tracer))
10003 MEM_FAIL(1, "Could not allocate function filter files");
10005 #ifdef CONFIG_TRACER_SNAPSHOT
10006 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
10007 tr, &snapshot_fops);
10010 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
10011 tr, &tracing_err_log_fops);
10013 for_each_tracing_cpu(cpu)
10014 tracing_init_tracefs_percpu(tr, cpu);
10016 ftrace_init_tracefs(tr, d_tracer);
10019 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
10021 struct vfsmount *mnt;
10022 struct file_system_type *type;
10025 * To maintain backward compatibility for tools that mount
10026 * debugfs to get to the tracing facility, tracefs is automatically
10027 * mounted to the debugfs/tracing directory.
10029 type = get_fs_type("tracefs");
10032 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
10033 put_filesystem(type);
10042 * tracing_init_dentry - initialize top level trace array
10044 * This is called when creating files or directories in the tracing
10045 * directory. It is called via fs_initcall() by any of the boot up code
10046 * and expects to return the dentry of the top level tracing directory.
10048 int tracing_init_dentry(void)
10050 struct trace_array *tr = &global_trace;
10052 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10053 pr_warn("Tracing disabled due to lockdown\n");
10057 /* The top level trace array uses NULL as parent */
10061 if (WARN_ON(!tracefs_initialized()))
10065 * As there may still be users that expect the tracing
10066 * files to exist in debugfs/tracing, we must automount
10067 * the tracefs file system there, so older tools still
10068 * work with the newer kernel.
10070 tr->dir = debugfs_create_automount("tracing", NULL,
10071 trace_automount, NULL);
10076 extern struct trace_eval_map *__start_ftrace_eval_maps[];
10077 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
10079 static struct workqueue_struct *eval_map_wq __initdata;
10080 static struct work_struct eval_map_work __initdata;
10081 static struct work_struct tracerfs_init_work __initdata;
10083 static void __init eval_map_work_func(struct work_struct *work)
10087 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
10088 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
10091 static int __init trace_eval_init(void)
10093 INIT_WORK(&eval_map_work, eval_map_work_func);
10095 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
10096 if (!eval_map_wq) {
10097 pr_err("Unable to allocate eval_map_wq\n");
10099 eval_map_work_func(&eval_map_work);
10103 queue_work(eval_map_wq, &eval_map_work);
10107 subsys_initcall(trace_eval_init);
10109 static int __init trace_eval_sync(void)
10111 /* Make sure the eval map updates are finished */
10113 destroy_workqueue(eval_map_wq);
10117 late_initcall_sync(trace_eval_sync);
10120 #ifdef CONFIG_MODULES
10121 static void trace_module_add_evals(struct module *mod)
10123 if (!mod->num_trace_evals)
10127 * Modules with bad taint do not have events created, do
10128 * not bother with enums either.
10130 if (trace_module_has_bad_taint(mod))
10133 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
10136 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
10137 static void trace_module_remove_evals(struct module *mod)
10139 union trace_eval_map_item *map;
10140 union trace_eval_map_item **last = &trace_eval_maps;
10142 if (!mod->num_trace_evals)
10145 mutex_lock(&trace_eval_mutex);
10147 map = trace_eval_maps;
10150 if (map->head.mod == mod)
10152 map = trace_eval_jmp_to_tail(map);
10153 last = &map->tail.next;
10154 map = map->tail.next;
10159 *last = trace_eval_jmp_to_tail(map)->tail.next;
10162 mutex_unlock(&trace_eval_mutex);
10165 static inline void trace_module_remove_evals(struct module *mod) { }
10166 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
10168 static int trace_module_notify(struct notifier_block *self,
10169 unsigned long val, void *data)
10171 struct module *mod = data;
10174 case MODULE_STATE_COMING:
10175 trace_module_add_evals(mod);
10177 case MODULE_STATE_GOING:
10178 trace_module_remove_evals(mod);
10185 static struct notifier_block trace_module_nb = {
10186 .notifier_call = trace_module_notify,
10189 #endif /* CONFIG_MODULES */
10191 static __init void tracer_init_tracefs_work_func(struct work_struct *work)
10194 event_trace_init();
10196 init_tracer_tracefs(&global_trace, NULL);
10197 ftrace_init_tracefs_toplevel(&global_trace, NULL);
10199 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
10200 &global_trace, &tracing_thresh_fops);
10202 trace_create_file("README", TRACE_MODE_READ, NULL,
10203 NULL, &tracing_readme_fops);
10205 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
10206 NULL, &tracing_saved_cmdlines_fops);
10208 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
10209 NULL, &tracing_saved_cmdlines_size_fops);
10211 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
10212 NULL, &tracing_saved_tgids_fops);
10214 trace_create_eval_file(NULL);
10216 #ifdef CONFIG_MODULES
10217 register_module_notifier(&trace_module_nb);
10220 #ifdef CONFIG_DYNAMIC_FTRACE
10221 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
10222 NULL, &tracing_dyn_info_fops);
10225 create_trace_instances(NULL);
10227 update_tracer_options(&global_trace);
10230 static __init int tracer_init_tracefs(void)
10234 trace_access_lock_init();
10236 ret = tracing_init_dentry();
10241 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
10242 queue_work(eval_map_wq, &tracerfs_init_work);
10244 tracer_init_tracefs_work_func(NULL);
10247 rv_init_interface();
10252 fs_initcall(tracer_init_tracefs);
10254 static int trace_die_panic_handler(struct notifier_block *self,
10255 unsigned long ev, void *unused);
10257 static struct notifier_block trace_panic_notifier = {
10258 .notifier_call = trace_die_panic_handler,
10259 .priority = INT_MAX - 1,
10262 static struct notifier_block trace_die_notifier = {
10263 .notifier_call = trace_die_panic_handler,
10264 .priority = INT_MAX - 1,
10268 * The idea is to execute the following die/panic callback early, in order
10269 * to avoid showing irrelevant information in the trace (like other panic
10270 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
10271 * warnings get disabled (to prevent potential log flooding).
10273 static int trace_die_panic_handler(struct notifier_block *self,
10274 unsigned long ev, void *unused)
10276 if (!ftrace_dump_on_oops)
10277 return NOTIFY_DONE;
10279 /* The die notifier requires DIE_OOPS to trigger */
10280 if (self == &trace_die_notifier && ev != DIE_OOPS)
10281 return NOTIFY_DONE;
10283 ftrace_dump(ftrace_dump_on_oops);
10285 return NOTIFY_DONE;
10289 * printk is set to max of 1024, we really don't need it that big.
10290 * Nothing should be printing 1000 characters anyway.
10292 #define TRACE_MAX_PRINT 1000
10295 * Define here KERN_TRACE so that we have one place to modify
10296 * it if we decide to change what log level the ftrace dump
10299 #define KERN_TRACE KERN_EMERG
10302 trace_printk_seq(struct trace_seq *s)
10304 /* Probably should print a warning here. */
10305 if (s->seq.len >= TRACE_MAX_PRINT)
10306 s->seq.len = TRACE_MAX_PRINT;
10309 * More paranoid code. Although the buffer size is set to
10310 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
10311 * an extra layer of protection.
10313 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
10314 s->seq.len = s->seq.size - 1;
10316 /* should be zero ended, but we are paranoid. */
10317 s->buffer[s->seq.len] = 0;
10319 printk(KERN_TRACE "%s", s->buffer);
10324 void trace_init_global_iter(struct trace_iterator *iter)
10326 iter->tr = &global_trace;
10327 iter->trace = iter->tr->current_trace;
10328 iter->cpu_file = RING_BUFFER_ALL_CPUS;
10329 iter->array_buffer = &global_trace.array_buffer;
10331 if (iter->trace && iter->trace->open)
10332 iter->trace->open(iter);
10334 /* Annotate start of buffers if we had overruns */
10335 if (ring_buffer_overruns(iter->array_buffer->buffer))
10336 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10338 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10339 if (trace_clocks[iter->tr->clock_id].in_ns)
10340 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10342 /* Can not use kmalloc for iter.temp and iter.fmt */
10343 iter->temp = static_temp_buf;
10344 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10345 iter->fmt = static_fmt_buf;
10346 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10349 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10351 /* use static because iter can be a bit big for the stack */
10352 static struct trace_iterator iter;
10353 static atomic_t dump_running;
10354 struct trace_array *tr = &global_trace;
10355 unsigned int old_userobj;
10356 unsigned long flags;
10359 /* Only allow one dump user at a time. */
10360 if (atomic_inc_return(&dump_running) != 1) {
10361 atomic_dec(&dump_running);
10366 * Always turn off tracing when we dump.
10367 * We don't need to show trace output of what happens
10368 * between multiple crashes.
10370 * If the user does a sysrq-z, then they can re-enable
10371 * tracing with echo 1 > tracing_on.
10375 local_irq_save(flags);
10377 /* Simulate the iterator */
10378 trace_init_global_iter(&iter);
10380 for_each_tracing_cpu(cpu) {
10381 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10384 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10386 /* don't look at user memory in panic mode */
10387 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10389 switch (oops_dump_mode) {
10391 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10394 iter.cpu_file = raw_smp_processor_id();
10399 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10400 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10403 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10405 /* Did function tracer already get disabled? */
10406 if (ftrace_is_dead()) {
10407 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10408 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10412 * We need to stop all tracing on all CPUS to read
10413 * the next buffer. This is a bit expensive, but is
10414 * not done often. We fill all what we can read,
10415 * and then release the locks again.
10418 while (!trace_empty(&iter)) {
10421 printk(KERN_TRACE "---------------------------------\n");
10425 trace_iterator_reset(&iter);
10426 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10428 if (trace_find_next_entry_inc(&iter) != NULL) {
10431 ret = print_trace_line(&iter);
10432 if (ret != TRACE_TYPE_NO_CONSUME)
10433 trace_consume(&iter);
10435 touch_nmi_watchdog();
10437 trace_printk_seq(&iter.seq);
10441 printk(KERN_TRACE " (ftrace buffer empty)\n");
10443 printk(KERN_TRACE "---------------------------------\n");
10446 tr->trace_flags |= old_userobj;
10448 for_each_tracing_cpu(cpu) {
10449 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10451 atomic_dec(&dump_running);
10452 local_irq_restore(flags);
10454 EXPORT_SYMBOL_GPL(ftrace_dump);
10456 #define WRITE_BUFSIZE 4096
10458 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10459 size_t count, loff_t *ppos,
10460 int (*createfn)(const char *))
10462 char *kbuf, *buf, *tmp;
10467 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10471 while (done < count) {
10472 size = count - done;
10474 if (size >= WRITE_BUFSIZE)
10475 size = WRITE_BUFSIZE - 1;
10477 if (copy_from_user(kbuf, buffer + done, size)) {
10484 tmp = strchr(buf, '\n');
10487 size = tmp - buf + 1;
10489 size = strlen(buf);
10490 if (done + size < count) {
10493 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10494 pr_warn("Line length is too long: Should be less than %d\n",
10495 WRITE_BUFSIZE - 2);
10502 /* Remove comments */
10503 tmp = strchr(buf, '#');
10508 ret = createfn(buf);
10513 } while (done < count);
10523 #ifdef CONFIG_TRACER_MAX_TRACE
10524 __init static bool tr_needs_alloc_snapshot(const char *name)
10527 int len = strlen(name);
10530 if (!boot_snapshot_index)
10533 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10534 boot_snapshot_info[len] == '\t')
10537 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10541 sprintf(test, "\t%s\t", name);
10542 ret = strstr(boot_snapshot_info, test) == NULL;
10547 __init static void do_allocate_snapshot(const char *name)
10549 if (!tr_needs_alloc_snapshot(name))
10553 * When allocate_snapshot is set, the next call to
10554 * allocate_trace_buffers() (called by trace_array_get_by_name())
10555 * will allocate the snapshot buffer. That will alse clear
10558 allocate_snapshot = true;
10561 static inline void do_allocate_snapshot(const char *name) { }
10564 __init static void enable_instances(void)
10566 struct trace_array *tr;
10571 /* A tab is always appended */
10572 boot_instance_info[boot_instance_index - 1] = '\0';
10573 str = boot_instance_info;
10575 while ((curr_str = strsep(&str, "\t"))) {
10577 tok = strsep(&curr_str, ",");
10579 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10580 do_allocate_snapshot(tok);
10582 tr = trace_array_get_by_name(tok, NULL);
10584 pr_warn("Failed to create instance buffer %s\n", curr_str);
10587 /* Allow user space to delete it */
10588 trace_array_put(tr);
10590 while ((tok = strsep(&curr_str, ","))) {
10591 early_enable_events(tr, tok, true);
10596 __init static int tracer_alloc_buffers(void)
10602 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10603 pr_warn("Tracing disabled due to lockdown\n");
10608 * Make sure we don't accidentally add more trace options
10609 * than we have bits for.
10611 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10613 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10616 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10617 goto out_free_buffer_mask;
10619 /* Only allocate trace_printk buffers if a trace_printk exists */
10620 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10621 /* Must be called before global_trace.buffer is allocated */
10622 trace_printk_init_buffers();
10624 /* To save memory, keep the ring buffer size to its minimum */
10625 if (global_trace.ring_buffer_expanded)
10626 ring_buf_size = trace_buf_size;
10630 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10631 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10633 raw_spin_lock_init(&global_trace.start_lock);
10636 * The prepare callbacks allocates some memory for the ring buffer. We
10637 * don't free the buffer if the CPU goes down. If we were to free
10638 * the buffer, then the user would lose any trace that was in the
10639 * buffer. The memory will be removed once the "instance" is removed.
10641 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10642 "trace/RB:prepare", trace_rb_cpu_prepare,
10645 goto out_free_cpumask;
10646 /* Used for event triggers */
10648 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10650 goto out_rm_hp_state;
10652 if (trace_create_savedcmd() < 0)
10653 goto out_free_temp_buffer;
10655 if (!zalloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10656 goto out_free_savedcmd;
10658 /* TODO: make the number of buffers hot pluggable with CPUS */
10659 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10660 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10661 goto out_free_pipe_cpumask;
10663 if (global_trace.buffer_disabled)
10666 if (trace_boot_clock) {
10667 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10669 pr_warn("Trace clock %s not defined, going back to default\n",
10674 * register_tracer() might reference current_trace, so it
10675 * needs to be set before we register anything. This is
10676 * just a bootstrap of current_trace anyway.
10678 global_trace.current_trace = &nop_trace;
10680 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10682 ftrace_init_global_array_ops(&global_trace);
10684 init_trace_flags_index(&global_trace);
10686 register_tracer(&nop_trace);
10688 /* Function tracing may start here (via kernel command line) */
10689 init_function_trace();
10691 /* All seems OK, enable tracing */
10692 tracing_disabled = 0;
10694 atomic_notifier_chain_register(&panic_notifier_list,
10695 &trace_panic_notifier);
10697 register_die_notifier(&trace_die_notifier);
10699 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10701 INIT_LIST_HEAD(&global_trace.systems);
10702 INIT_LIST_HEAD(&global_trace.events);
10703 INIT_LIST_HEAD(&global_trace.hist_vars);
10704 INIT_LIST_HEAD(&global_trace.err_log);
10705 list_add(&global_trace.list, &ftrace_trace_arrays);
10707 apply_trace_boot_options();
10709 register_snapshot_cmd();
10715 out_free_pipe_cpumask:
10716 free_cpumask_var(global_trace.pipe_cpumask);
10718 free_saved_cmdlines_buffer(savedcmd);
10719 out_free_temp_buffer:
10720 ring_buffer_free(temp_buffer);
10722 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10724 free_cpumask_var(global_trace.tracing_cpumask);
10725 out_free_buffer_mask:
10726 free_cpumask_var(tracing_buffer_mask);
10731 void __init ftrace_boot_snapshot(void)
10733 #ifdef CONFIG_TRACER_MAX_TRACE
10734 struct trace_array *tr;
10736 if (!snapshot_at_boot)
10739 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10740 if (!tr->allocated_snapshot)
10743 tracing_snapshot_instance(tr);
10744 trace_array_puts(tr, "** Boot snapshot taken **\n");
10749 void __init early_trace_init(void)
10751 if (tracepoint_printk) {
10752 tracepoint_print_iter =
10753 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10754 if (MEM_FAIL(!tracepoint_print_iter,
10755 "Failed to allocate trace iterator\n"))
10756 tracepoint_printk = 0;
10758 static_key_enable(&tracepoint_printk_key.key);
10760 tracer_alloc_buffers();
10765 void __init trace_init(void)
10767 trace_event_init();
10769 if (boot_instance_index)
10770 enable_instances();
10773 __init static void clear_boot_tracer(void)
10776 * The default tracer at boot buffer is an init section.
10777 * This function is called in lateinit. If we did not
10778 * find the boot tracer, then clear it out, to prevent
10779 * later registration from accessing the buffer that is
10780 * about to be freed.
10782 if (!default_bootup_tracer)
10785 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10786 default_bootup_tracer);
10787 default_bootup_tracer = NULL;
10790 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10791 __init static void tracing_set_default_clock(void)
10793 /* sched_clock_stable() is determined in late_initcall */
10794 if (!trace_boot_clock && !sched_clock_stable()) {
10795 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10796 pr_warn("Can not set tracing clock due to lockdown\n");
10800 printk(KERN_WARNING
10801 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10802 "If you want to keep using the local clock, then add:\n"
10803 " \"trace_clock=local\"\n"
10804 "on the kernel command line\n");
10805 tracing_set_clock(&global_trace, "global");
10809 static inline void tracing_set_default_clock(void) { }
10812 __init static int late_trace_init(void)
10814 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10815 static_key_disable(&tracepoint_printk_key.key);
10816 tracepoint_printk = 0;
10819 tracing_set_default_clock();
10820 clear_boot_tracer();
10824 late_initcall_sync(late_trace_init);