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 */
1958 static int wait_on_pipe(struct trace_iterator *iter, int full)
1962 /* Iterators are static, they should be filled or empty */
1963 if (trace_buffer_iter(iter, iter->cpu_file))
1966 ret = ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file, full);
1968 #ifdef CONFIG_TRACER_MAX_TRACE
1970 * Make sure this is still the snapshot buffer, as if a snapshot were
1971 * to happen, this would now be the main buffer.
1974 iter->array_buffer = &iter->tr->max_buffer;
1979 #ifdef CONFIG_FTRACE_STARTUP_TEST
1980 static bool selftests_can_run;
1982 struct trace_selftests {
1983 struct list_head list;
1984 struct tracer *type;
1987 static LIST_HEAD(postponed_selftests);
1989 static int save_selftest(struct tracer *type)
1991 struct trace_selftests *selftest;
1993 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1997 selftest->type = type;
1998 list_add(&selftest->list, &postponed_selftests);
2002 static int run_tracer_selftest(struct tracer *type)
2004 struct trace_array *tr = &global_trace;
2005 struct tracer *saved_tracer = tr->current_trace;
2008 if (!type->selftest || tracing_selftest_disabled)
2012 * If a tracer registers early in boot up (before scheduling is
2013 * initialized and such), then do not run its selftests yet.
2014 * Instead, run it a little later in the boot process.
2016 if (!selftests_can_run)
2017 return save_selftest(type);
2019 if (!tracing_is_on()) {
2020 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2026 * Run a selftest on this tracer.
2027 * Here we reset the trace buffer, and set the current
2028 * tracer to be this tracer. The tracer can then run some
2029 * internal tracing to verify that everything is in order.
2030 * If we fail, we do not register this tracer.
2032 tracing_reset_online_cpus(&tr->array_buffer);
2034 tr->current_trace = type;
2036 #ifdef CONFIG_TRACER_MAX_TRACE
2037 if (type->use_max_tr) {
2038 /* If we expanded the buffers, make sure the max is expanded too */
2039 if (tr->ring_buffer_expanded)
2040 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2041 RING_BUFFER_ALL_CPUS);
2042 tr->allocated_snapshot = true;
2046 /* the test is responsible for initializing and enabling */
2047 pr_info("Testing tracer %s: ", type->name);
2048 ret = type->selftest(type, tr);
2049 /* the test is responsible for resetting too */
2050 tr->current_trace = saved_tracer;
2052 printk(KERN_CONT "FAILED!\n");
2053 /* Add the warning after printing 'FAILED' */
2057 /* Only reset on passing, to avoid touching corrupted buffers */
2058 tracing_reset_online_cpus(&tr->array_buffer);
2060 #ifdef CONFIG_TRACER_MAX_TRACE
2061 if (type->use_max_tr) {
2062 tr->allocated_snapshot = false;
2064 /* Shrink the max buffer again */
2065 if (tr->ring_buffer_expanded)
2066 ring_buffer_resize(tr->max_buffer.buffer, 1,
2067 RING_BUFFER_ALL_CPUS);
2071 printk(KERN_CONT "PASSED\n");
2075 static int do_run_tracer_selftest(struct tracer *type)
2080 * Tests can take a long time, especially if they are run one after the
2081 * other, as does happen during bootup when all the tracers are
2082 * registered. This could cause the soft lockup watchdog to trigger.
2086 tracing_selftest_running = true;
2087 ret = run_tracer_selftest(type);
2088 tracing_selftest_running = false;
2093 static __init int init_trace_selftests(void)
2095 struct trace_selftests *p, *n;
2096 struct tracer *t, **last;
2099 selftests_can_run = true;
2101 mutex_lock(&trace_types_lock);
2103 if (list_empty(&postponed_selftests))
2106 pr_info("Running postponed tracer tests:\n");
2108 tracing_selftest_running = true;
2109 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2110 /* This loop can take minutes when sanitizers are enabled, so
2111 * lets make sure we allow RCU processing.
2114 ret = run_tracer_selftest(p->type);
2115 /* If the test fails, then warn and remove from available_tracers */
2117 WARN(1, "tracer: %s failed selftest, disabling\n",
2119 last = &trace_types;
2120 for (t = trace_types; t; t = t->next) {
2131 tracing_selftest_running = false;
2134 mutex_unlock(&trace_types_lock);
2138 core_initcall(init_trace_selftests);
2140 static inline int run_tracer_selftest(struct tracer *type)
2144 static inline int do_run_tracer_selftest(struct tracer *type)
2148 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2150 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2152 static void __init apply_trace_boot_options(void);
2155 * register_tracer - register a tracer with the ftrace system.
2156 * @type: the plugin for the tracer
2158 * Register a new plugin tracer.
2160 int __init register_tracer(struct tracer *type)
2166 pr_info("Tracer must have a name\n");
2170 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2171 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2175 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2176 pr_warn("Can not register tracer %s due to lockdown\n",
2181 mutex_lock(&trace_types_lock);
2183 for (t = trace_types; t; t = t->next) {
2184 if (strcmp(type->name, t->name) == 0) {
2186 pr_info("Tracer %s already registered\n",
2193 if (!type->set_flag)
2194 type->set_flag = &dummy_set_flag;
2196 /*allocate a dummy tracer_flags*/
2197 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2202 type->flags->val = 0;
2203 type->flags->opts = dummy_tracer_opt;
2205 if (!type->flags->opts)
2206 type->flags->opts = dummy_tracer_opt;
2208 /* store the tracer for __set_tracer_option */
2209 type->flags->trace = type;
2211 ret = do_run_tracer_selftest(type);
2215 type->next = trace_types;
2217 add_tracer_options(&global_trace, type);
2220 mutex_unlock(&trace_types_lock);
2222 if (ret || !default_bootup_tracer)
2225 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2228 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2229 /* Do we want this tracer to start on bootup? */
2230 tracing_set_tracer(&global_trace, type->name);
2231 default_bootup_tracer = NULL;
2233 apply_trace_boot_options();
2235 /* disable other selftests, since this will break it. */
2236 disable_tracing_selftest("running a tracer");
2242 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2244 struct trace_buffer *buffer = buf->buffer;
2249 ring_buffer_record_disable(buffer);
2251 /* Make sure all commits have finished */
2253 ring_buffer_reset_cpu(buffer, cpu);
2255 ring_buffer_record_enable(buffer);
2258 void tracing_reset_online_cpus(struct array_buffer *buf)
2260 struct trace_buffer *buffer = buf->buffer;
2265 ring_buffer_record_disable(buffer);
2267 /* Make sure all commits have finished */
2270 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2272 ring_buffer_reset_online_cpus(buffer);
2274 ring_buffer_record_enable(buffer);
2277 /* Must have trace_types_lock held */
2278 void tracing_reset_all_online_cpus_unlocked(void)
2280 struct trace_array *tr;
2282 lockdep_assert_held(&trace_types_lock);
2284 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2285 if (!tr->clear_trace)
2287 tr->clear_trace = false;
2288 tracing_reset_online_cpus(&tr->array_buffer);
2289 #ifdef CONFIG_TRACER_MAX_TRACE
2290 tracing_reset_online_cpus(&tr->max_buffer);
2295 void tracing_reset_all_online_cpus(void)
2297 mutex_lock(&trace_types_lock);
2298 tracing_reset_all_online_cpus_unlocked();
2299 mutex_unlock(&trace_types_lock);
2303 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2304 * is the tgid last observed corresponding to pid=i.
2306 static int *tgid_map;
2308 /* The maximum valid index into tgid_map. */
2309 static size_t tgid_map_max;
2311 #define SAVED_CMDLINES_DEFAULT 128
2312 #define NO_CMDLINE_MAP UINT_MAX
2314 * Preemption must be disabled before acquiring trace_cmdline_lock.
2315 * The various trace_arrays' max_lock must be acquired in a context
2316 * where interrupt is disabled.
2318 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2319 struct saved_cmdlines_buffer {
2320 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2321 unsigned *map_cmdline_to_pid;
2322 unsigned cmdline_num;
2324 char saved_cmdlines[];
2326 static struct saved_cmdlines_buffer *savedcmd;
2328 static inline char *get_saved_cmdlines(int idx)
2330 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2333 static inline void set_cmdline(int idx, const char *cmdline)
2335 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2338 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
2340 int order = get_order(sizeof(*s) + s->cmdline_num * TASK_COMM_LEN);
2342 kfree(s->map_cmdline_to_pid);
2344 free_pages((unsigned long)s, order);
2347 static struct saved_cmdlines_buffer *allocate_cmdlines_buffer(unsigned int val)
2349 struct saved_cmdlines_buffer *s;
2351 int orig_size, size;
2354 /* Figure out how much is needed to hold the given number of cmdlines */
2355 orig_size = sizeof(*s) + val * TASK_COMM_LEN;
2356 order = get_order(orig_size);
2357 size = 1 << (order + PAGE_SHIFT);
2358 page = alloc_pages(GFP_KERNEL, order);
2362 s = page_address(page);
2363 kmemleak_alloc(s, size, 1, GFP_KERNEL);
2364 memset(s, 0, sizeof(*s));
2366 /* Round up to actual allocation */
2367 val = (size - sizeof(*s)) / TASK_COMM_LEN;
2368 s->cmdline_num = val;
2370 s->map_cmdline_to_pid = kmalloc_array(val,
2371 sizeof(*s->map_cmdline_to_pid),
2373 if (!s->map_cmdline_to_pid) {
2374 free_saved_cmdlines_buffer(s);
2379 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2380 sizeof(s->map_pid_to_cmdline));
2381 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2382 val * sizeof(*s->map_cmdline_to_pid));
2387 static int trace_create_savedcmd(void)
2389 savedcmd = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT);
2391 return savedcmd ? 0 : -ENOMEM;
2394 int is_tracing_stopped(void)
2396 return global_trace.stop_count;
2399 static void tracing_start_tr(struct trace_array *tr)
2401 struct trace_buffer *buffer;
2402 unsigned long flags;
2404 if (tracing_disabled)
2407 raw_spin_lock_irqsave(&tr->start_lock, flags);
2408 if (--tr->stop_count) {
2409 if (WARN_ON_ONCE(tr->stop_count < 0)) {
2410 /* Someone screwed up their debugging */
2416 /* Prevent the buffers from switching */
2417 arch_spin_lock(&tr->max_lock);
2419 buffer = tr->array_buffer.buffer;
2421 ring_buffer_record_enable(buffer);
2423 #ifdef CONFIG_TRACER_MAX_TRACE
2424 buffer = tr->max_buffer.buffer;
2426 ring_buffer_record_enable(buffer);
2429 arch_spin_unlock(&tr->max_lock);
2432 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2436 * tracing_start - quick start of the tracer
2438 * If tracing is enabled but was stopped by tracing_stop,
2439 * this will start the tracer back up.
2441 void tracing_start(void)
2444 return tracing_start_tr(&global_trace);
2447 static void tracing_stop_tr(struct trace_array *tr)
2449 struct trace_buffer *buffer;
2450 unsigned long flags;
2452 raw_spin_lock_irqsave(&tr->start_lock, flags);
2453 if (tr->stop_count++)
2456 /* Prevent the buffers from switching */
2457 arch_spin_lock(&tr->max_lock);
2459 buffer = tr->array_buffer.buffer;
2461 ring_buffer_record_disable(buffer);
2463 #ifdef CONFIG_TRACER_MAX_TRACE
2464 buffer = tr->max_buffer.buffer;
2466 ring_buffer_record_disable(buffer);
2469 arch_spin_unlock(&tr->max_lock);
2472 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2476 * tracing_stop - quick stop of the tracer
2478 * Light weight way to stop tracing. Use in conjunction with
2481 void tracing_stop(void)
2483 return tracing_stop_tr(&global_trace);
2486 static int trace_save_cmdline(struct task_struct *tsk)
2490 /* treat recording of idle task as a success */
2494 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2497 * It's not the end of the world if we don't get
2498 * the lock, but we also don't want to spin
2499 * nor do we want to disable interrupts,
2500 * so if we miss here, then better luck next time.
2502 * This is called within the scheduler and wake up, so interrupts
2503 * had better been disabled and run queue lock been held.
2505 lockdep_assert_preemption_disabled();
2506 if (!arch_spin_trylock(&trace_cmdline_lock))
2509 idx = savedcmd->map_pid_to_cmdline[tpid];
2510 if (idx == NO_CMDLINE_MAP) {
2511 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2513 savedcmd->map_pid_to_cmdline[tpid] = idx;
2514 savedcmd->cmdline_idx = idx;
2517 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2518 set_cmdline(idx, tsk->comm);
2520 arch_spin_unlock(&trace_cmdline_lock);
2525 static void __trace_find_cmdline(int pid, char comm[])
2531 strcpy(comm, "<idle>");
2535 if (WARN_ON_ONCE(pid < 0)) {
2536 strcpy(comm, "<XXX>");
2540 tpid = pid & (PID_MAX_DEFAULT - 1);
2541 map = savedcmd->map_pid_to_cmdline[tpid];
2542 if (map != NO_CMDLINE_MAP) {
2543 tpid = savedcmd->map_cmdline_to_pid[map];
2545 strscpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2549 strcpy(comm, "<...>");
2552 void trace_find_cmdline(int pid, char comm[])
2555 arch_spin_lock(&trace_cmdline_lock);
2557 __trace_find_cmdline(pid, comm);
2559 arch_spin_unlock(&trace_cmdline_lock);
2563 static int *trace_find_tgid_ptr(int pid)
2566 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2567 * if we observe a non-NULL tgid_map then we also observe the correct
2570 int *map = smp_load_acquire(&tgid_map);
2572 if (unlikely(!map || pid > tgid_map_max))
2578 int trace_find_tgid(int pid)
2580 int *ptr = trace_find_tgid_ptr(pid);
2582 return ptr ? *ptr : 0;
2585 static int trace_save_tgid(struct task_struct *tsk)
2589 /* treat recording of idle task as a success */
2593 ptr = trace_find_tgid_ptr(tsk->pid);
2601 static bool tracing_record_taskinfo_skip(int flags)
2603 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2605 if (!__this_cpu_read(trace_taskinfo_save))
2611 * tracing_record_taskinfo - record the task info of a task
2613 * @task: task to record
2614 * @flags: TRACE_RECORD_CMDLINE for recording comm
2615 * TRACE_RECORD_TGID for recording tgid
2617 void tracing_record_taskinfo(struct task_struct *task, int flags)
2621 if (tracing_record_taskinfo_skip(flags))
2625 * Record as much task information as possible. If some fail, continue
2626 * to try to record the others.
2628 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2629 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2631 /* If recording any information failed, retry again soon. */
2635 __this_cpu_write(trace_taskinfo_save, false);
2639 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2641 * @prev: previous task during sched_switch
2642 * @next: next task during sched_switch
2643 * @flags: TRACE_RECORD_CMDLINE for recording comm
2644 * TRACE_RECORD_TGID for recording tgid
2646 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2647 struct task_struct *next, int flags)
2651 if (tracing_record_taskinfo_skip(flags))
2655 * Record as much task information as possible. If some fail, continue
2656 * to try to record the others.
2658 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2659 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2660 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2661 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2663 /* If recording any information failed, retry again soon. */
2667 __this_cpu_write(trace_taskinfo_save, false);
2670 /* Helpers to record a specific task information */
2671 void tracing_record_cmdline(struct task_struct *task)
2673 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2676 void tracing_record_tgid(struct task_struct *task)
2678 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2682 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2683 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2684 * simplifies those functions and keeps them in sync.
2686 enum print_line_t trace_handle_return(struct trace_seq *s)
2688 return trace_seq_has_overflowed(s) ?
2689 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2691 EXPORT_SYMBOL_GPL(trace_handle_return);
2693 static unsigned short migration_disable_value(void)
2695 #if defined(CONFIG_SMP)
2696 return current->migration_disabled;
2702 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2704 unsigned int trace_flags = irqs_status;
2707 pc = preempt_count();
2710 trace_flags |= TRACE_FLAG_NMI;
2711 if (pc & HARDIRQ_MASK)
2712 trace_flags |= TRACE_FLAG_HARDIRQ;
2713 if (in_serving_softirq())
2714 trace_flags |= TRACE_FLAG_SOFTIRQ;
2715 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2716 trace_flags |= TRACE_FLAG_BH_OFF;
2718 if (tif_need_resched())
2719 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2720 if (test_preempt_need_resched())
2721 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2722 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2723 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2726 struct ring_buffer_event *
2727 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2730 unsigned int trace_ctx)
2732 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2735 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2736 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2737 static int trace_buffered_event_ref;
2740 * trace_buffered_event_enable - enable buffering events
2742 * When events are being filtered, it is quicker to use a temporary
2743 * buffer to write the event data into if there's a likely chance
2744 * that it will not be committed. The discard of the ring buffer
2745 * is not as fast as committing, and is much slower than copying
2748 * When an event is to be filtered, allocate per cpu buffers to
2749 * write the event data into, and if the event is filtered and discarded
2750 * it is simply dropped, otherwise, the entire data is to be committed
2753 void trace_buffered_event_enable(void)
2755 struct ring_buffer_event *event;
2759 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2761 if (trace_buffered_event_ref++)
2764 for_each_tracing_cpu(cpu) {
2765 page = alloc_pages_node(cpu_to_node(cpu),
2766 GFP_KERNEL | __GFP_NORETRY, 0);
2767 /* This is just an optimization and can handle failures */
2769 pr_err("Failed to allocate event buffer\n");
2773 event = page_address(page);
2774 memset(event, 0, sizeof(*event));
2776 per_cpu(trace_buffered_event, cpu) = event;
2779 if (cpu == smp_processor_id() &&
2780 __this_cpu_read(trace_buffered_event) !=
2781 per_cpu(trace_buffered_event, cpu))
2787 static void enable_trace_buffered_event(void *data)
2789 /* Probably not needed, but do it anyway */
2791 this_cpu_dec(trace_buffered_event_cnt);
2794 static void disable_trace_buffered_event(void *data)
2796 this_cpu_inc(trace_buffered_event_cnt);
2800 * trace_buffered_event_disable - disable buffering events
2802 * When a filter is removed, it is faster to not use the buffered
2803 * events, and to commit directly into the ring buffer. Free up
2804 * the temp buffers when there are no more users. This requires
2805 * special synchronization with current events.
2807 void trace_buffered_event_disable(void)
2811 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2813 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2816 if (--trace_buffered_event_ref)
2819 /* For each CPU, set the buffer as used. */
2820 on_each_cpu_mask(tracing_buffer_mask, disable_trace_buffered_event,
2823 /* Wait for all current users to finish */
2826 for_each_tracing_cpu(cpu) {
2827 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2828 per_cpu(trace_buffered_event, cpu) = NULL;
2832 * Wait for all CPUs that potentially started checking if they can use
2833 * their event buffer only after the previous synchronize_rcu() call and
2834 * they still read a valid pointer from trace_buffered_event. It must be
2835 * ensured they don't see cleared trace_buffered_event_cnt else they
2836 * could wrongly decide to use the pointed-to buffer which is now freed.
2840 /* For each CPU, relinquish the buffer */
2841 on_each_cpu_mask(tracing_buffer_mask, enable_trace_buffered_event, NULL,
2845 static struct trace_buffer *temp_buffer;
2847 struct ring_buffer_event *
2848 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2849 struct trace_event_file *trace_file,
2850 int type, unsigned long len,
2851 unsigned int trace_ctx)
2853 struct ring_buffer_event *entry;
2854 struct trace_array *tr = trace_file->tr;
2857 *current_rb = tr->array_buffer.buffer;
2859 if (!tr->no_filter_buffering_ref &&
2860 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2861 preempt_disable_notrace();
2863 * Filtering is on, so try to use the per cpu buffer first.
2864 * This buffer will simulate a ring_buffer_event,
2865 * where the type_len is zero and the array[0] will
2866 * hold the full length.
2867 * (see include/linux/ring-buffer.h for details on
2868 * how the ring_buffer_event is structured).
2870 * Using a temp buffer during filtering and copying it
2871 * on a matched filter is quicker than writing directly
2872 * into the ring buffer and then discarding it when
2873 * it doesn't match. That is because the discard
2874 * requires several atomic operations to get right.
2875 * Copying on match and doing nothing on a failed match
2876 * is still quicker than no copy on match, but having
2877 * to discard out of the ring buffer on a failed match.
2879 if ((entry = __this_cpu_read(trace_buffered_event))) {
2880 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2882 val = this_cpu_inc_return(trace_buffered_event_cnt);
2885 * Preemption is disabled, but interrupts and NMIs
2886 * can still come in now. If that happens after
2887 * the above increment, then it will have to go
2888 * back to the old method of allocating the event
2889 * on the ring buffer, and if the filter fails, it
2890 * will have to call ring_buffer_discard_commit()
2893 * Need to also check the unlikely case that the
2894 * length is bigger than the temp buffer size.
2895 * If that happens, then the reserve is pretty much
2896 * guaranteed to fail, as the ring buffer currently
2897 * only allows events less than a page. But that may
2898 * change in the future, so let the ring buffer reserve
2899 * handle the failure in that case.
2901 if (val == 1 && likely(len <= max_len)) {
2902 trace_event_setup(entry, type, trace_ctx);
2903 entry->array[0] = len;
2904 /* Return with preemption disabled */
2907 this_cpu_dec(trace_buffered_event_cnt);
2909 /* __trace_buffer_lock_reserve() disables preemption */
2910 preempt_enable_notrace();
2913 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2916 * If tracing is off, but we have triggers enabled
2917 * we still need to look at the event data. Use the temp_buffer
2918 * to store the trace event for the trigger to use. It's recursive
2919 * safe and will not be recorded anywhere.
2921 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2922 *current_rb = temp_buffer;
2923 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2928 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2930 static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2931 static DEFINE_MUTEX(tracepoint_printk_mutex);
2933 static void output_printk(struct trace_event_buffer *fbuffer)
2935 struct trace_event_call *event_call;
2936 struct trace_event_file *file;
2937 struct trace_event *event;
2938 unsigned long flags;
2939 struct trace_iterator *iter = tracepoint_print_iter;
2941 /* We should never get here if iter is NULL */
2942 if (WARN_ON_ONCE(!iter))
2945 event_call = fbuffer->trace_file->event_call;
2946 if (!event_call || !event_call->event.funcs ||
2947 !event_call->event.funcs->trace)
2950 file = fbuffer->trace_file;
2951 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2952 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2953 !filter_match_preds(file->filter, fbuffer->entry)))
2956 event = &fbuffer->trace_file->event_call->event;
2958 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2959 trace_seq_init(&iter->seq);
2960 iter->ent = fbuffer->entry;
2961 event_call->event.funcs->trace(iter, 0, event);
2962 trace_seq_putc(&iter->seq, 0);
2963 printk("%s", iter->seq.buffer);
2965 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2968 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2969 void *buffer, size_t *lenp,
2972 int save_tracepoint_printk;
2975 mutex_lock(&tracepoint_printk_mutex);
2976 save_tracepoint_printk = tracepoint_printk;
2978 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2981 * This will force exiting early, as tracepoint_printk
2982 * is always zero when tracepoint_printk_iter is not allocated
2984 if (!tracepoint_print_iter)
2985 tracepoint_printk = 0;
2987 if (save_tracepoint_printk == tracepoint_printk)
2990 if (tracepoint_printk)
2991 static_key_enable(&tracepoint_printk_key.key);
2993 static_key_disable(&tracepoint_printk_key.key);
2996 mutex_unlock(&tracepoint_printk_mutex);
3001 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
3003 enum event_trigger_type tt = ETT_NONE;
3004 struct trace_event_file *file = fbuffer->trace_file;
3006 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
3007 fbuffer->entry, &tt))
3010 if (static_key_false(&tracepoint_printk_key.key))
3011 output_printk(fbuffer);
3013 if (static_branch_unlikely(&trace_event_exports_enabled))
3014 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
3016 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
3017 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
3021 event_triggers_post_call(file, tt);
3024 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
3029 * trace_buffer_unlock_commit_regs()
3030 * trace_event_buffer_commit()
3031 * trace_event_raw_event_xxx()
3033 # define STACK_SKIP 3
3035 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
3036 struct trace_buffer *buffer,
3037 struct ring_buffer_event *event,
3038 unsigned int trace_ctx,
3039 struct pt_regs *regs)
3041 __buffer_unlock_commit(buffer, event);
3044 * If regs is not set, then skip the necessary functions.
3045 * Note, we can still get here via blktrace, wakeup tracer
3046 * and mmiotrace, but that's ok if they lose a function or
3047 * two. They are not that meaningful.
3049 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
3050 ftrace_trace_userstack(tr, buffer, trace_ctx);
3054 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
3057 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
3058 struct ring_buffer_event *event)
3060 __buffer_unlock_commit(buffer, event);
3064 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
3065 parent_ip, unsigned int trace_ctx)
3067 struct trace_event_call *call = &event_function;
3068 struct trace_buffer *buffer = tr->array_buffer.buffer;
3069 struct ring_buffer_event *event;
3070 struct ftrace_entry *entry;
3072 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
3076 entry = ring_buffer_event_data(event);
3078 entry->parent_ip = parent_ip;
3080 if (!call_filter_check_discard(call, entry, buffer, event)) {
3081 if (static_branch_unlikely(&trace_function_exports_enabled))
3082 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
3083 __buffer_unlock_commit(buffer, event);
3087 #ifdef CONFIG_STACKTRACE
3089 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
3090 #define FTRACE_KSTACK_NESTING 4
3092 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3094 struct ftrace_stack {
3095 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3099 struct ftrace_stacks {
3100 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3103 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3104 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3106 static void __ftrace_trace_stack(struct trace_buffer *buffer,
3107 unsigned int trace_ctx,
3108 int skip, struct pt_regs *regs)
3110 struct trace_event_call *call = &event_kernel_stack;
3111 struct ring_buffer_event *event;
3112 unsigned int size, nr_entries;
3113 struct ftrace_stack *fstack;
3114 struct stack_entry *entry;
3118 * Add one, for this function and the call to save_stack_trace()
3119 * If regs is set, then these functions will not be in the way.
3121 #ifndef CONFIG_UNWINDER_ORC
3126 preempt_disable_notrace();
3128 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3130 /* This should never happen. If it does, yell once and skip */
3131 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3135 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3136 * interrupt will either see the value pre increment or post
3137 * increment. If the interrupt happens pre increment it will have
3138 * restored the counter when it returns. We just need a barrier to
3139 * keep gcc from moving things around.
3143 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3144 size = ARRAY_SIZE(fstack->calls);
3147 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3150 nr_entries = stack_trace_save(fstack->calls, size, skip);
3153 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3154 struct_size(entry, caller, nr_entries),
3158 entry = ring_buffer_event_data(event);
3160 entry->size = nr_entries;
3161 memcpy(&entry->caller, fstack->calls,
3162 flex_array_size(entry, caller, nr_entries));
3164 if (!call_filter_check_discard(call, entry, buffer, event))
3165 __buffer_unlock_commit(buffer, event);
3168 /* Again, don't let gcc optimize things here */
3170 __this_cpu_dec(ftrace_stack_reserve);
3171 preempt_enable_notrace();
3175 static inline void ftrace_trace_stack(struct trace_array *tr,
3176 struct trace_buffer *buffer,
3177 unsigned int trace_ctx,
3178 int skip, struct pt_regs *regs)
3180 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3183 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3186 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3189 struct trace_buffer *buffer = tr->array_buffer.buffer;
3191 if (rcu_is_watching()) {
3192 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3196 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3200 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3201 * but if the above rcu_is_watching() failed, then the NMI
3202 * triggered someplace critical, and ct_irq_enter() should
3203 * not be called from NMI.
3205 if (unlikely(in_nmi()))
3208 ct_irq_enter_irqson();
3209 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3210 ct_irq_exit_irqson();
3214 * trace_dump_stack - record a stack back trace in the trace buffer
3215 * @skip: Number of functions to skip (helper handlers)
3217 void trace_dump_stack(int skip)
3219 if (tracing_disabled || tracing_selftest_running)
3222 #ifndef CONFIG_UNWINDER_ORC
3223 /* Skip 1 to skip this function. */
3226 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3227 tracing_gen_ctx(), skip, NULL);
3229 EXPORT_SYMBOL_GPL(trace_dump_stack);
3231 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3232 static DEFINE_PER_CPU(int, user_stack_count);
3235 ftrace_trace_userstack(struct trace_array *tr,
3236 struct trace_buffer *buffer, unsigned int trace_ctx)
3238 struct trace_event_call *call = &event_user_stack;
3239 struct ring_buffer_event *event;
3240 struct userstack_entry *entry;
3242 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3246 * NMIs can not handle page faults, even with fix ups.
3247 * The save user stack can (and often does) fault.
3249 if (unlikely(in_nmi()))
3253 * prevent recursion, since the user stack tracing may
3254 * trigger other kernel events.
3257 if (__this_cpu_read(user_stack_count))
3260 __this_cpu_inc(user_stack_count);
3262 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3263 sizeof(*entry), trace_ctx);
3265 goto out_drop_count;
3266 entry = ring_buffer_event_data(event);
3268 entry->tgid = current->tgid;
3269 memset(&entry->caller, 0, sizeof(entry->caller));
3271 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3272 if (!call_filter_check_discard(call, entry, buffer, event))
3273 __buffer_unlock_commit(buffer, event);
3276 __this_cpu_dec(user_stack_count);
3280 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3281 static void ftrace_trace_userstack(struct trace_array *tr,
3282 struct trace_buffer *buffer,
3283 unsigned int trace_ctx)
3286 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3288 #endif /* CONFIG_STACKTRACE */
3291 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3292 unsigned long long delta)
3294 entry->bottom_delta_ts = delta & U32_MAX;
3295 entry->top_delta_ts = (delta >> 32);
3298 void trace_last_func_repeats(struct trace_array *tr,
3299 struct trace_func_repeats *last_info,
3300 unsigned int trace_ctx)
3302 struct trace_buffer *buffer = tr->array_buffer.buffer;
3303 struct func_repeats_entry *entry;
3304 struct ring_buffer_event *event;
3307 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3308 sizeof(*entry), trace_ctx);
3312 delta = ring_buffer_event_time_stamp(buffer, event) -
3313 last_info->ts_last_call;
3315 entry = ring_buffer_event_data(event);
3316 entry->ip = last_info->ip;
3317 entry->parent_ip = last_info->parent_ip;
3318 entry->count = last_info->count;
3319 func_repeats_set_delta_ts(entry, delta);
3321 __buffer_unlock_commit(buffer, event);
3324 /* created for use with alloc_percpu */
3325 struct trace_buffer_struct {
3327 char buffer[4][TRACE_BUF_SIZE];
3330 static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3333 * This allows for lockless recording. If we're nested too deeply, then
3334 * this returns NULL.
3336 static char *get_trace_buf(void)
3338 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3340 if (!trace_percpu_buffer || buffer->nesting >= 4)
3345 /* Interrupts must see nesting incremented before we use the buffer */
3347 return &buffer->buffer[buffer->nesting - 1][0];
3350 static void put_trace_buf(void)
3352 /* Don't let the decrement of nesting leak before this */
3354 this_cpu_dec(trace_percpu_buffer->nesting);
3357 static int alloc_percpu_trace_buffer(void)
3359 struct trace_buffer_struct __percpu *buffers;
3361 if (trace_percpu_buffer)
3364 buffers = alloc_percpu(struct trace_buffer_struct);
3365 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3368 trace_percpu_buffer = buffers;
3372 static int buffers_allocated;
3374 void trace_printk_init_buffers(void)
3376 if (buffers_allocated)
3379 if (alloc_percpu_trace_buffer())
3382 /* trace_printk() is for debug use only. Don't use it in production. */
3385 pr_warn("**********************************************************\n");
3386 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3388 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3390 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3391 pr_warn("** unsafe for production use. **\n");
3393 pr_warn("** If you see this message and you are not debugging **\n");
3394 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3396 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3397 pr_warn("**********************************************************\n");
3399 /* Expand the buffers to set size */
3400 tracing_update_buffers(&global_trace);
3402 buffers_allocated = 1;
3405 * trace_printk_init_buffers() can be called by modules.
3406 * If that happens, then we need to start cmdline recording
3407 * directly here. If the global_trace.buffer is already
3408 * allocated here, then this was called by module code.
3410 if (global_trace.array_buffer.buffer)
3411 tracing_start_cmdline_record();
3413 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3415 void trace_printk_start_comm(void)
3417 /* Start tracing comms if trace printk is set */
3418 if (!buffers_allocated)
3420 tracing_start_cmdline_record();
3423 static void trace_printk_start_stop_comm(int enabled)
3425 if (!buffers_allocated)
3429 tracing_start_cmdline_record();
3431 tracing_stop_cmdline_record();
3435 * trace_vbprintk - write binary msg to tracing buffer
3436 * @ip: The address of the caller
3437 * @fmt: The string format to write to the buffer
3438 * @args: Arguments for @fmt
3440 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3442 struct trace_event_call *call = &event_bprint;
3443 struct ring_buffer_event *event;
3444 struct trace_buffer *buffer;
3445 struct trace_array *tr = &global_trace;
3446 struct bprint_entry *entry;
3447 unsigned int trace_ctx;
3451 if (unlikely(tracing_selftest_running || tracing_disabled))
3454 /* Don't pollute graph traces with trace_vprintk internals */
3455 pause_graph_tracing();
3457 trace_ctx = tracing_gen_ctx();
3458 preempt_disable_notrace();
3460 tbuffer = get_trace_buf();
3466 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3468 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3471 size = sizeof(*entry) + sizeof(u32) * len;
3472 buffer = tr->array_buffer.buffer;
3473 ring_buffer_nest_start(buffer);
3474 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3478 entry = ring_buffer_event_data(event);
3482 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3483 if (!call_filter_check_discard(call, entry, buffer, event)) {
3484 __buffer_unlock_commit(buffer, event);
3485 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3489 ring_buffer_nest_end(buffer);
3494 preempt_enable_notrace();
3495 unpause_graph_tracing();
3499 EXPORT_SYMBOL_GPL(trace_vbprintk);
3503 __trace_array_vprintk(struct trace_buffer *buffer,
3504 unsigned long ip, const char *fmt, va_list args)
3506 struct trace_event_call *call = &event_print;
3507 struct ring_buffer_event *event;
3509 struct print_entry *entry;
3510 unsigned int trace_ctx;
3513 if (tracing_disabled)
3516 /* Don't pollute graph traces with trace_vprintk internals */
3517 pause_graph_tracing();
3519 trace_ctx = tracing_gen_ctx();
3520 preempt_disable_notrace();
3523 tbuffer = get_trace_buf();
3529 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3531 size = sizeof(*entry) + len + 1;
3532 ring_buffer_nest_start(buffer);
3533 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3537 entry = ring_buffer_event_data(event);
3540 memcpy(&entry->buf, tbuffer, len + 1);
3541 if (!call_filter_check_discard(call, entry, buffer, event)) {
3542 __buffer_unlock_commit(buffer, event);
3543 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3547 ring_buffer_nest_end(buffer);
3551 preempt_enable_notrace();
3552 unpause_graph_tracing();
3558 int trace_array_vprintk(struct trace_array *tr,
3559 unsigned long ip, const char *fmt, va_list args)
3561 if (tracing_selftest_running && tr == &global_trace)
3564 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3568 * trace_array_printk - Print a message to a specific instance
3569 * @tr: The instance trace_array descriptor
3570 * @ip: The instruction pointer that this is called from.
3571 * @fmt: The format to print (printf format)
3573 * If a subsystem sets up its own instance, they have the right to
3574 * printk strings into their tracing instance buffer using this
3575 * function. Note, this function will not write into the top level
3576 * buffer (use trace_printk() for that), as writing into the top level
3577 * buffer should only have events that can be individually disabled.
3578 * trace_printk() is only used for debugging a kernel, and should not
3579 * be ever incorporated in normal use.
3581 * trace_array_printk() can be used, as it will not add noise to the
3582 * top level tracing buffer.
3584 * Note, trace_array_init_printk() must be called on @tr before this
3588 int trace_array_printk(struct trace_array *tr,
3589 unsigned long ip, const char *fmt, ...)
3597 /* This is only allowed for created instances */
3598 if (tr == &global_trace)
3601 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3605 ret = trace_array_vprintk(tr, ip, fmt, ap);
3609 EXPORT_SYMBOL_GPL(trace_array_printk);
3612 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3613 * @tr: The trace array to initialize the buffers for
3615 * As trace_array_printk() only writes into instances, they are OK to
3616 * have in the kernel (unlike trace_printk()). This needs to be called
3617 * before trace_array_printk() can be used on a trace_array.
3619 int trace_array_init_printk(struct trace_array *tr)
3624 /* This is only allowed for created instances */
3625 if (tr == &global_trace)
3628 return alloc_percpu_trace_buffer();
3630 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3633 int trace_array_printk_buf(struct trace_buffer *buffer,
3634 unsigned long ip, const char *fmt, ...)
3639 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3643 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3649 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3651 return trace_array_vprintk(&global_trace, ip, fmt, args);
3653 EXPORT_SYMBOL_GPL(trace_vprintk);
3655 static void trace_iterator_increment(struct trace_iterator *iter)
3657 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3661 ring_buffer_iter_advance(buf_iter);
3664 static struct trace_entry *
3665 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3666 unsigned long *lost_events)
3668 struct ring_buffer_event *event;
3669 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3672 event = ring_buffer_iter_peek(buf_iter, ts);
3674 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3675 (unsigned long)-1 : 0;
3677 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3682 iter->ent_size = ring_buffer_event_length(event);
3683 return ring_buffer_event_data(event);
3689 static struct trace_entry *
3690 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3691 unsigned long *missing_events, u64 *ent_ts)
3693 struct trace_buffer *buffer = iter->array_buffer->buffer;
3694 struct trace_entry *ent, *next = NULL;
3695 unsigned long lost_events = 0, next_lost = 0;
3696 int cpu_file = iter->cpu_file;
3697 u64 next_ts = 0, ts;
3703 * If we are in a per_cpu trace file, don't bother by iterating over
3704 * all cpu and peek directly.
3706 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3707 if (ring_buffer_empty_cpu(buffer, cpu_file))
3709 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3711 *ent_cpu = cpu_file;
3716 for_each_tracing_cpu(cpu) {
3718 if (ring_buffer_empty_cpu(buffer, cpu))
3721 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3724 * Pick the entry with the smallest timestamp:
3726 if (ent && (!next || ts < next_ts)) {
3730 next_lost = lost_events;
3731 next_size = iter->ent_size;
3735 iter->ent_size = next_size;
3738 *ent_cpu = next_cpu;
3744 *missing_events = next_lost;
3749 #define STATIC_FMT_BUF_SIZE 128
3750 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3752 char *trace_iter_expand_format(struct trace_iterator *iter)
3757 * iter->tr is NULL when used with tp_printk, which makes
3758 * this get called where it is not safe to call krealloc().
3760 if (!iter->tr || iter->fmt == static_fmt_buf)
3763 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3766 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3773 /* Returns true if the string is safe to dereference from an event */
3774 static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3777 unsigned long addr = (unsigned long)str;
3778 struct trace_event *trace_event;
3779 struct trace_event_call *event;
3781 /* Ignore strings with no length */
3785 /* OK if part of the event data */
3786 if ((addr >= (unsigned long)iter->ent) &&
3787 (addr < (unsigned long)iter->ent + iter->ent_size))
3790 /* OK if part of the temp seq buffer */
3791 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3792 (addr < (unsigned long)iter->tmp_seq.buffer + TRACE_SEQ_BUFFER_SIZE))
3795 /* Core rodata can not be freed */
3796 if (is_kernel_rodata(addr))
3799 if (trace_is_tracepoint_string(str))
3803 * Now this could be a module event, referencing core module
3804 * data, which is OK.
3809 trace_event = ftrace_find_event(iter->ent->type);
3813 event = container_of(trace_event, struct trace_event_call, event);
3814 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3817 /* Would rather have rodata, but this will suffice */
3818 if (within_module_core(addr, event->module))
3824 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3826 static int test_can_verify_check(const char *fmt, ...)
3833 * The verifier is dependent on vsnprintf() modifies the va_list
3834 * passed to it, where it is sent as a reference. Some architectures
3835 * (like x86_32) passes it by value, which means that vsnprintf()
3836 * does not modify the va_list passed to it, and the verifier
3837 * would then need to be able to understand all the values that
3838 * vsnprintf can use. If it is passed by value, then the verifier
3842 vsnprintf(buf, 16, "%d", ap);
3843 ret = va_arg(ap, int);
3849 static void test_can_verify(void)
3851 if (!test_can_verify_check("%d %d", 0, 1)) {
3852 pr_info("trace event string verifier disabled\n");
3853 static_branch_inc(&trace_no_verify);
3858 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3859 * @iter: The iterator that holds the seq buffer and the event being printed
3860 * @fmt: The format used to print the event
3861 * @ap: The va_list holding the data to print from @fmt.
3863 * This writes the data into the @iter->seq buffer using the data from
3864 * @fmt and @ap. If the format has a %s, then the source of the string
3865 * is examined to make sure it is safe to print, otherwise it will
3866 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3869 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3872 const char *p = fmt;
3876 if (WARN_ON_ONCE(!fmt))
3879 if (static_branch_unlikely(&trace_no_verify))
3882 /* Don't bother checking when doing a ftrace_dump() */
3883 if (iter->fmt == static_fmt_buf)
3892 /* We only care about %s and variants */
3893 for (i = 0; p[i]; i++) {
3894 if (i + 1 >= iter->fmt_size) {
3896 * If we can't expand the copy buffer,
3899 if (!trace_iter_expand_format(iter))
3903 if (p[i] == '\\' && p[i+1]) {
3908 /* Need to test cases like %08.*s */
3909 for (j = 1; p[i+j]; j++) {
3910 if (isdigit(p[i+j]) ||
3913 if (p[i+j] == '*') {
3925 /* If no %s found then just print normally */
3929 /* Copy up to the %s, and print that */
3930 strncpy(iter->fmt, p, i);
3931 iter->fmt[i] = '\0';
3932 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3935 * If iter->seq is full, the above call no longer guarantees
3936 * that ap is in sync with fmt processing, and further calls
3937 * to va_arg() can return wrong positional arguments.
3939 * Ensure that ap is no longer used in this case.
3941 if (iter->seq.full) {
3947 len = va_arg(ap, int);
3949 /* The ap now points to the string data of the %s */
3950 str = va_arg(ap, const char *);
3953 * If you hit this warning, it is likely that the
3954 * trace event in question used %s on a string that
3955 * was saved at the time of the event, but may not be
3956 * around when the trace is read. Use __string(),
3957 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3958 * instead. See samples/trace_events/trace-events-sample.h
3961 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3962 "fmt: '%s' current_buffer: '%s'",
3963 fmt, seq_buf_str(&iter->seq.seq))) {
3966 /* Try to safely read the string */
3968 if (len + 1 > iter->fmt_size)
3969 len = iter->fmt_size - 1;
3972 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3976 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3980 trace_seq_printf(&iter->seq, "(0x%px)", str);
3982 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3984 str = "[UNSAFE-MEMORY]";
3985 strcpy(iter->fmt, "%s");
3987 strncpy(iter->fmt, p + i, j + 1);
3988 iter->fmt[j+1] = '\0';
3991 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3993 trace_seq_printf(&iter->seq, iter->fmt, str);
3999 trace_seq_vprintf(&iter->seq, p, ap);
4002 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
4004 const char *p, *new_fmt;
4007 if (WARN_ON_ONCE(!fmt))
4010 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
4014 new_fmt = q = iter->fmt;
4016 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
4017 if (!trace_iter_expand_format(iter))
4020 q += iter->fmt - new_fmt;
4021 new_fmt = iter->fmt;
4026 /* Replace %p with %px */
4030 } else if (p[0] == 'p' && !isalnum(p[1])) {
4041 #define STATIC_TEMP_BUF_SIZE 128
4042 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
4044 /* Find the next real entry, without updating the iterator itself */
4045 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
4046 int *ent_cpu, u64 *ent_ts)
4048 /* __find_next_entry will reset ent_size */
4049 int ent_size = iter->ent_size;
4050 struct trace_entry *entry;
4053 * If called from ftrace_dump(), then the iter->temp buffer
4054 * will be the static_temp_buf and not created from kmalloc.
4055 * If the entry size is greater than the buffer, we can
4056 * not save it. Just return NULL in that case. This is only
4057 * used to add markers when two consecutive events' time
4058 * stamps have a large delta. See trace_print_lat_context()
4060 if (iter->temp == static_temp_buf &&
4061 STATIC_TEMP_BUF_SIZE < ent_size)
4065 * The __find_next_entry() may call peek_next_entry(), which may
4066 * call ring_buffer_peek() that may make the contents of iter->ent
4067 * undefined. Need to copy iter->ent now.
4069 if (iter->ent && iter->ent != iter->temp) {
4070 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4071 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4073 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4078 iter->temp_size = iter->ent_size;
4080 memcpy(iter->temp, iter->ent, iter->ent_size);
4081 iter->ent = iter->temp;
4083 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4084 /* Put back the original ent_size */
4085 iter->ent_size = ent_size;
4090 /* Find the next real entry, and increment the iterator to the next entry */
4091 void *trace_find_next_entry_inc(struct trace_iterator *iter)
4093 iter->ent = __find_next_entry(iter, &iter->cpu,
4094 &iter->lost_events, &iter->ts);
4097 trace_iterator_increment(iter);
4099 return iter->ent ? iter : NULL;
4102 static void trace_consume(struct trace_iterator *iter)
4104 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4105 &iter->lost_events);
4108 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4110 struct trace_iterator *iter = m->private;
4114 WARN_ON_ONCE(iter->leftover);
4118 /* can't go backwards */
4123 ent = trace_find_next_entry_inc(iter);
4127 while (ent && iter->idx < i)
4128 ent = trace_find_next_entry_inc(iter);
4135 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4137 struct ring_buffer_iter *buf_iter;
4138 unsigned long entries = 0;
4141 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4143 buf_iter = trace_buffer_iter(iter, cpu);
4147 ring_buffer_iter_reset(buf_iter);
4150 * We could have the case with the max latency tracers
4151 * that a reset never took place on a cpu. This is evident
4152 * by the timestamp being before the start of the buffer.
4154 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4155 if (ts >= iter->array_buffer->time_start)
4158 ring_buffer_iter_advance(buf_iter);
4161 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4165 * The current tracer is copied to avoid a global locking
4168 static void *s_start(struct seq_file *m, loff_t *pos)
4170 struct trace_iterator *iter = m->private;
4171 struct trace_array *tr = iter->tr;
4172 int cpu_file = iter->cpu_file;
4177 mutex_lock(&trace_types_lock);
4178 if (unlikely(tr->current_trace != iter->trace)) {
4179 /* Close iter->trace before switching to the new current tracer */
4180 if (iter->trace->close)
4181 iter->trace->close(iter);
4182 iter->trace = tr->current_trace;
4183 /* Reopen the new current tracer */
4184 if (iter->trace->open)
4185 iter->trace->open(iter);
4187 mutex_unlock(&trace_types_lock);
4189 #ifdef CONFIG_TRACER_MAX_TRACE
4190 if (iter->snapshot && iter->trace->use_max_tr)
4191 return ERR_PTR(-EBUSY);
4194 if (*pos != iter->pos) {
4199 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4200 for_each_tracing_cpu(cpu)
4201 tracing_iter_reset(iter, cpu);
4203 tracing_iter_reset(iter, cpu_file);
4206 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4211 * If we overflowed the seq_file before, then we want
4212 * to just reuse the trace_seq buffer again.
4218 p = s_next(m, p, &l);
4222 trace_event_read_lock();
4223 trace_access_lock(cpu_file);
4227 static void s_stop(struct seq_file *m, void *p)
4229 struct trace_iterator *iter = m->private;
4231 #ifdef CONFIG_TRACER_MAX_TRACE
4232 if (iter->snapshot && iter->trace->use_max_tr)
4236 trace_access_unlock(iter->cpu_file);
4237 trace_event_read_unlock();
4241 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4242 unsigned long *entries, int cpu)
4244 unsigned long count;
4246 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4248 * If this buffer has skipped entries, then we hold all
4249 * entries for the trace and we need to ignore the
4250 * ones before the time stamp.
4252 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4253 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4254 /* total is the same as the entries */
4258 ring_buffer_overrun_cpu(buf->buffer, cpu);
4263 get_total_entries(struct array_buffer *buf,
4264 unsigned long *total, unsigned long *entries)
4272 for_each_tracing_cpu(cpu) {
4273 get_total_entries_cpu(buf, &t, &e, cpu);
4279 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4281 unsigned long total, entries;
4286 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4291 unsigned long trace_total_entries(struct trace_array *tr)
4293 unsigned long total, entries;
4298 get_total_entries(&tr->array_buffer, &total, &entries);
4303 static void print_lat_help_header(struct seq_file *m)
4305 seq_puts(m, "# _------=> CPU# \n"
4306 "# / _-----=> irqs-off/BH-disabled\n"
4307 "# | / _----=> need-resched \n"
4308 "# || / _---=> hardirq/softirq \n"
4309 "# ||| / _--=> preempt-depth \n"
4310 "# |||| / _-=> migrate-disable \n"
4311 "# ||||| / delay \n"
4312 "# cmd pid |||||| time | caller \n"
4313 "# \\ / |||||| \\ | / \n");
4316 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4318 unsigned long total;
4319 unsigned long entries;
4321 get_total_entries(buf, &total, &entries);
4322 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4323 entries, total, num_online_cpus());
4327 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4330 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4332 print_event_info(buf, m);
4334 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4335 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4338 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4341 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4342 static const char space[] = " ";
4343 int prec = tgid ? 12 : 2;
4345 print_event_info(buf, m);
4347 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4348 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4349 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4350 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4351 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4352 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4353 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4354 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4358 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4360 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4361 struct array_buffer *buf = iter->array_buffer;
4362 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4363 struct tracer *type = iter->trace;
4364 unsigned long entries;
4365 unsigned long total;
4366 const char *name = type->name;
4368 get_total_entries(buf, &total, &entries);
4370 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4372 seq_puts(m, "# -----------------------------------"
4373 "---------------------------------\n");
4374 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4375 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4376 nsecs_to_usecs(data->saved_latency),
4380 preempt_model_none() ? "server" :
4381 preempt_model_voluntary() ? "desktop" :
4382 preempt_model_full() ? "preempt" :
4383 preempt_model_rt() ? "preempt_rt" :
4385 /* These are reserved for later use */
4388 seq_printf(m, " #P:%d)\n", num_online_cpus());
4392 seq_puts(m, "# -----------------\n");
4393 seq_printf(m, "# | task: %.16s-%d "
4394 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4395 data->comm, data->pid,
4396 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4397 data->policy, data->rt_priority);
4398 seq_puts(m, "# -----------------\n");
4400 if (data->critical_start) {
4401 seq_puts(m, "# => started at: ");
4402 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4403 trace_print_seq(m, &iter->seq);
4404 seq_puts(m, "\n# => ended at: ");
4405 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4406 trace_print_seq(m, &iter->seq);
4407 seq_puts(m, "\n#\n");
4413 static void test_cpu_buff_start(struct trace_iterator *iter)
4415 struct trace_seq *s = &iter->seq;
4416 struct trace_array *tr = iter->tr;
4418 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4421 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4424 if (cpumask_available(iter->started) &&
4425 cpumask_test_cpu(iter->cpu, iter->started))
4428 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4431 if (cpumask_available(iter->started))
4432 cpumask_set_cpu(iter->cpu, iter->started);
4434 /* Don't print started cpu buffer for the first entry of the trace */
4436 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4440 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4442 struct trace_array *tr = iter->tr;
4443 struct trace_seq *s = &iter->seq;
4444 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4445 struct trace_entry *entry;
4446 struct trace_event *event;
4450 test_cpu_buff_start(iter);
4452 event = ftrace_find_event(entry->type);
4454 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4455 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4456 trace_print_lat_context(iter);
4458 trace_print_context(iter);
4461 if (trace_seq_has_overflowed(s))
4462 return TRACE_TYPE_PARTIAL_LINE;
4465 if (tr->trace_flags & TRACE_ITER_FIELDS)
4466 return print_event_fields(iter, event);
4467 return event->funcs->trace(iter, sym_flags, event);
4470 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4472 return trace_handle_return(s);
4475 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4477 struct trace_array *tr = iter->tr;
4478 struct trace_seq *s = &iter->seq;
4479 struct trace_entry *entry;
4480 struct trace_event *event;
4484 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4485 trace_seq_printf(s, "%d %d %llu ",
4486 entry->pid, iter->cpu, iter->ts);
4488 if (trace_seq_has_overflowed(s))
4489 return TRACE_TYPE_PARTIAL_LINE;
4491 event = ftrace_find_event(entry->type);
4493 return event->funcs->raw(iter, 0, event);
4495 trace_seq_printf(s, "%d ?\n", entry->type);
4497 return trace_handle_return(s);
4500 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4502 struct trace_array *tr = iter->tr;
4503 struct trace_seq *s = &iter->seq;
4504 unsigned char newline = '\n';
4505 struct trace_entry *entry;
4506 struct trace_event *event;
4510 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4511 SEQ_PUT_HEX_FIELD(s, entry->pid);
4512 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4513 SEQ_PUT_HEX_FIELD(s, iter->ts);
4514 if (trace_seq_has_overflowed(s))
4515 return TRACE_TYPE_PARTIAL_LINE;
4518 event = ftrace_find_event(entry->type);
4520 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4521 if (ret != TRACE_TYPE_HANDLED)
4525 SEQ_PUT_FIELD(s, newline);
4527 return trace_handle_return(s);
4530 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4532 struct trace_array *tr = iter->tr;
4533 struct trace_seq *s = &iter->seq;
4534 struct trace_entry *entry;
4535 struct trace_event *event;
4539 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4540 SEQ_PUT_FIELD(s, entry->pid);
4541 SEQ_PUT_FIELD(s, iter->cpu);
4542 SEQ_PUT_FIELD(s, iter->ts);
4543 if (trace_seq_has_overflowed(s))
4544 return TRACE_TYPE_PARTIAL_LINE;
4547 event = ftrace_find_event(entry->type);
4548 return event ? event->funcs->binary(iter, 0, event) :
4552 int trace_empty(struct trace_iterator *iter)
4554 struct ring_buffer_iter *buf_iter;
4557 /* If we are looking at one CPU buffer, only check that one */
4558 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4559 cpu = iter->cpu_file;
4560 buf_iter = trace_buffer_iter(iter, cpu);
4562 if (!ring_buffer_iter_empty(buf_iter))
4565 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4571 for_each_tracing_cpu(cpu) {
4572 buf_iter = trace_buffer_iter(iter, cpu);
4574 if (!ring_buffer_iter_empty(buf_iter))
4577 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4585 /* Called with trace_event_read_lock() held. */
4586 enum print_line_t print_trace_line(struct trace_iterator *iter)
4588 struct trace_array *tr = iter->tr;
4589 unsigned long trace_flags = tr->trace_flags;
4590 enum print_line_t ret;
4592 if (iter->lost_events) {
4593 if (iter->lost_events == (unsigned long)-1)
4594 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4597 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4598 iter->cpu, iter->lost_events);
4599 if (trace_seq_has_overflowed(&iter->seq))
4600 return TRACE_TYPE_PARTIAL_LINE;
4603 if (iter->trace && iter->trace->print_line) {
4604 ret = iter->trace->print_line(iter);
4605 if (ret != TRACE_TYPE_UNHANDLED)
4609 if (iter->ent->type == TRACE_BPUTS &&
4610 trace_flags & TRACE_ITER_PRINTK &&
4611 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4612 return trace_print_bputs_msg_only(iter);
4614 if (iter->ent->type == TRACE_BPRINT &&
4615 trace_flags & TRACE_ITER_PRINTK &&
4616 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4617 return trace_print_bprintk_msg_only(iter);
4619 if (iter->ent->type == TRACE_PRINT &&
4620 trace_flags & TRACE_ITER_PRINTK &&
4621 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4622 return trace_print_printk_msg_only(iter);
4624 if (trace_flags & TRACE_ITER_BIN)
4625 return print_bin_fmt(iter);
4627 if (trace_flags & TRACE_ITER_HEX)
4628 return print_hex_fmt(iter);
4630 if (trace_flags & TRACE_ITER_RAW)
4631 return print_raw_fmt(iter);
4633 return print_trace_fmt(iter);
4636 void trace_latency_header(struct seq_file *m)
4638 struct trace_iterator *iter = m->private;
4639 struct trace_array *tr = iter->tr;
4641 /* print nothing if the buffers are empty */
4642 if (trace_empty(iter))
4645 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4646 print_trace_header(m, iter);
4648 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4649 print_lat_help_header(m);
4652 void trace_default_header(struct seq_file *m)
4654 struct trace_iterator *iter = m->private;
4655 struct trace_array *tr = iter->tr;
4656 unsigned long trace_flags = tr->trace_flags;
4658 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4661 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4662 /* print nothing if the buffers are empty */
4663 if (trace_empty(iter))
4665 print_trace_header(m, iter);
4666 if (!(trace_flags & TRACE_ITER_VERBOSE))
4667 print_lat_help_header(m);
4669 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4670 if (trace_flags & TRACE_ITER_IRQ_INFO)
4671 print_func_help_header_irq(iter->array_buffer,
4674 print_func_help_header(iter->array_buffer, m,
4680 static void test_ftrace_alive(struct seq_file *m)
4682 if (!ftrace_is_dead())
4684 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4685 "# MAY BE MISSING FUNCTION EVENTS\n");
4688 #ifdef CONFIG_TRACER_MAX_TRACE
4689 static void show_snapshot_main_help(struct seq_file *m)
4691 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4692 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4693 "# Takes a snapshot of the main buffer.\n"
4694 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4695 "# (Doesn't have to be '2' works with any number that\n"
4696 "# is not a '0' or '1')\n");
4699 static void show_snapshot_percpu_help(struct seq_file *m)
4701 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4702 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4703 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4704 "# Takes a snapshot of the main buffer for this cpu.\n");
4706 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4707 "# Must use main snapshot file to allocate.\n");
4709 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4710 "# (Doesn't have to be '2' works with any number that\n"
4711 "# is not a '0' or '1')\n");
4714 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4716 if (iter->tr->allocated_snapshot)
4717 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4719 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4721 seq_puts(m, "# Snapshot commands:\n");
4722 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4723 show_snapshot_main_help(m);
4725 show_snapshot_percpu_help(m);
4728 /* Should never be called */
4729 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4732 static int s_show(struct seq_file *m, void *v)
4734 struct trace_iterator *iter = v;
4737 if (iter->ent == NULL) {
4739 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4741 test_ftrace_alive(m);
4743 if (iter->snapshot && trace_empty(iter))
4744 print_snapshot_help(m, iter);
4745 else if (iter->trace && iter->trace->print_header)
4746 iter->trace->print_header(m);
4748 trace_default_header(m);
4750 } else if (iter->leftover) {
4752 * If we filled the seq_file buffer earlier, we
4753 * want to just show it now.
4755 ret = trace_print_seq(m, &iter->seq);
4757 /* ret should this time be zero, but you never know */
4758 iter->leftover = ret;
4761 ret = print_trace_line(iter);
4762 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4764 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
4766 ret = trace_print_seq(m, &iter->seq);
4768 * If we overflow the seq_file buffer, then it will
4769 * ask us for this data again at start up.
4771 * ret is 0 if seq_file write succeeded.
4774 iter->leftover = ret;
4781 * Should be used after trace_array_get(), trace_types_lock
4782 * ensures that i_cdev was already initialized.
4784 static inline int tracing_get_cpu(struct inode *inode)
4786 if (inode->i_cdev) /* See trace_create_cpu_file() */
4787 return (long)inode->i_cdev - 1;
4788 return RING_BUFFER_ALL_CPUS;
4791 static const struct seq_operations tracer_seq_ops = {
4799 * Note, as iter itself can be allocated and freed in different
4800 * ways, this function is only used to free its content, and not
4801 * the iterator itself. The only requirement to all the allocations
4802 * is that it must zero all fields (kzalloc), as freeing works with
4803 * ethier allocated content or NULL.
4805 static void free_trace_iter_content(struct trace_iterator *iter)
4807 /* The fmt is either NULL, allocated or points to static_fmt_buf */
4808 if (iter->fmt != static_fmt_buf)
4812 kfree(iter->buffer_iter);
4813 mutex_destroy(&iter->mutex);
4814 free_cpumask_var(iter->started);
4817 static struct trace_iterator *
4818 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4820 struct trace_array *tr = inode->i_private;
4821 struct trace_iterator *iter;
4824 if (tracing_disabled)
4825 return ERR_PTR(-ENODEV);
4827 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4829 return ERR_PTR(-ENOMEM);
4831 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4833 if (!iter->buffer_iter)
4837 * trace_find_next_entry() may need to save off iter->ent.
4838 * It will place it into the iter->temp buffer. As most
4839 * events are less than 128, allocate a buffer of that size.
4840 * If one is greater, then trace_find_next_entry() will
4841 * allocate a new buffer to adjust for the bigger iter->ent.
4842 * It's not critical if it fails to get allocated here.
4844 iter->temp = kmalloc(128, GFP_KERNEL);
4846 iter->temp_size = 128;
4849 * trace_event_printf() may need to modify given format
4850 * string to replace %p with %px so that it shows real address
4851 * instead of hash value. However, that is only for the event
4852 * tracing, other tracer may not need. Defer the allocation
4853 * until it is needed.
4858 mutex_lock(&trace_types_lock);
4859 iter->trace = tr->current_trace;
4861 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4866 #ifdef CONFIG_TRACER_MAX_TRACE
4867 /* Currently only the top directory has a snapshot */
4868 if (tr->current_trace->print_max || snapshot)
4869 iter->array_buffer = &tr->max_buffer;
4872 iter->array_buffer = &tr->array_buffer;
4873 iter->snapshot = snapshot;
4875 iter->cpu_file = tracing_get_cpu(inode);
4876 mutex_init(&iter->mutex);
4878 /* Notify the tracer early; before we stop tracing. */
4879 if (iter->trace->open)
4880 iter->trace->open(iter);
4882 /* Annotate start of buffers if we had overruns */
4883 if (ring_buffer_overruns(iter->array_buffer->buffer))
4884 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4886 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4887 if (trace_clocks[tr->clock_id].in_ns)
4888 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4891 * If pause-on-trace is enabled, then stop the trace while
4892 * dumping, unless this is the "snapshot" file
4894 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4895 tracing_stop_tr(tr);
4897 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4898 for_each_tracing_cpu(cpu) {
4899 iter->buffer_iter[cpu] =
4900 ring_buffer_read_prepare(iter->array_buffer->buffer,
4903 ring_buffer_read_prepare_sync();
4904 for_each_tracing_cpu(cpu) {
4905 ring_buffer_read_start(iter->buffer_iter[cpu]);
4906 tracing_iter_reset(iter, cpu);
4909 cpu = iter->cpu_file;
4910 iter->buffer_iter[cpu] =
4911 ring_buffer_read_prepare(iter->array_buffer->buffer,
4913 ring_buffer_read_prepare_sync();
4914 ring_buffer_read_start(iter->buffer_iter[cpu]);
4915 tracing_iter_reset(iter, cpu);
4918 mutex_unlock(&trace_types_lock);
4923 mutex_unlock(&trace_types_lock);
4924 free_trace_iter_content(iter);
4926 seq_release_private(inode, file);
4927 return ERR_PTR(-ENOMEM);
4930 int tracing_open_generic(struct inode *inode, struct file *filp)
4934 ret = tracing_check_open_get_tr(NULL);
4938 filp->private_data = inode->i_private;
4942 bool tracing_is_disabled(void)
4944 return (tracing_disabled) ? true: false;
4948 * Open and update trace_array ref count.
4949 * Must have the current trace_array passed to it.
4951 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4953 struct trace_array *tr = inode->i_private;
4956 ret = tracing_check_open_get_tr(tr);
4960 filp->private_data = inode->i_private;
4966 * The private pointer of the inode is the trace_event_file.
4967 * Update the tr ref count associated to it.
4969 int tracing_open_file_tr(struct inode *inode, struct file *filp)
4971 struct trace_event_file *file = inode->i_private;
4974 ret = tracing_check_open_get_tr(file->tr);
4978 mutex_lock(&event_mutex);
4980 /* Fail if the file is marked for removal */
4981 if (file->flags & EVENT_FILE_FL_FREED) {
4982 trace_array_put(file->tr);
4985 event_file_get(file);
4988 mutex_unlock(&event_mutex);
4992 filp->private_data = inode->i_private;
4997 int tracing_release_file_tr(struct inode *inode, struct file *filp)
4999 struct trace_event_file *file = inode->i_private;
5001 trace_array_put(file->tr);
5002 event_file_put(file);
5007 int tracing_single_release_file_tr(struct inode *inode, struct file *filp)
5009 tracing_release_file_tr(inode, filp);
5010 return single_release(inode, filp);
5013 static int tracing_mark_open(struct inode *inode, struct file *filp)
5015 stream_open(inode, filp);
5016 return tracing_open_generic_tr(inode, filp);
5019 static int tracing_release(struct inode *inode, struct file *file)
5021 struct trace_array *tr = inode->i_private;
5022 struct seq_file *m = file->private_data;
5023 struct trace_iterator *iter;
5026 if (!(file->f_mode & FMODE_READ)) {
5027 trace_array_put(tr);
5031 /* Writes do not use seq_file */
5033 mutex_lock(&trace_types_lock);
5035 for_each_tracing_cpu(cpu) {
5036 if (iter->buffer_iter[cpu])
5037 ring_buffer_read_finish(iter->buffer_iter[cpu]);
5040 if (iter->trace && iter->trace->close)
5041 iter->trace->close(iter);
5043 if (!iter->snapshot && tr->stop_count)
5044 /* reenable tracing if it was previously enabled */
5045 tracing_start_tr(tr);
5047 __trace_array_put(tr);
5049 mutex_unlock(&trace_types_lock);
5051 free_trace_iter_content(iter);
5052 seq_release_private(inode, file);
5057 int tracing_release_generic_tr(struct inode *inode, struct file *file)
5059 struct trace_array *tr = inode->i_private;
5061 trace_array_put(tr);
5065 static int tracing_single_release_tr(struct inode *inode, struct file *file)
5067 struct trace_array *tr = inode->i_private;
5069 trace_array_put(tr);
5071 return single_release(inode, file);
5074 static int tracing_open(struct inode *inode, struct file *file)
5076 struct trace_array *tr = inode->i_private;
5077 struct trace_iterator *iter;
5080 ret = tracing_check_open_get_tr(tr);
5084 /* If this file was open for write, then erase contents */
5085 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
5086 int cpu = tracing_get_cpu(inode);
5087 struct array_buffer *trace_buf = &tr->array_buffer;
5089 #ifdef CONFIG_TRACER_MAX_TRACE
5090 if (tr->current_trace->print_max)
5091 trace_buf = &tr->max_buffer;
5094 if (cpu == RING_BUFFER_ALL_CPUS)
5095 tracing_reset_online_cpus(trace_buf);
5097 tracing_reset_cpu(trace_buf, cpu);
5100 if (file->f_mode & FMODE_READ) {
5101 iter = __tracing_open(inode, file, false);
5103 ret = PTR_ERR(iter);
5104 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
5105 iter->iter_flags |= TRACE_FILE_LAT_FMT;
5109 trace_array_put(tr);
5115 * Some tracers are not suitable for instance buffers.
5116 * A tracer is always available for the global array (toplevel)
5117 * or if it explicitly states that it is.
5120 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
5122 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5125 /* Find the next tracer that this trace array may use */
5126 static struct tracer *
5127 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5129 while (t && !trace_ok_for_array(t, tr))
5136 t_next(struct seq_file *m, void *v, loff_t *pos)
5138 struct trace_array *tr = m->private;
5139 struct tracer *t = v;
5144 t = get_tracer_for_array(tr, t->next);
5149 static void *t_start(struct seq_file *m, loff_t *pos)
5151 struct trace_array *tr = m->private;
5155 mutex_lock(&trace_types_lock);
5157 t = get_tracer_for_array(tr, trace_types);
5158 for (; t && l < *pos; t = t_next(m, t, &l))
5164 static void t_stop(struct seq_file *m, void *p)
5166 mutex_unlock(&trace_types_lock);
5169 static int t_show(struct seq_file *m, void *v)
5171 struct tracer *t = v;
5176 seq_puts(m, t->name);
5185 static const struct seq_operations show_traces_seq_ops = {
5192 static int show_traces_open(struct inode *inode, struct file *file)
5194 struct trace_array *tr = inode->i_private;
5198 ret = tracing_check_open_get_tr(tr);
5202 ret = seq_open(file, &show_traces_seq_ops);
5204 trace_array_put(tr);
5208 m = file->private_data;
5214 static int show_traces_release(struct inode *inode, struct file *file)
5216 struct trace_array *tr = inode->i_private;
5218 trace_array_put(tr);
5219 return seq_release(inode, file);
5223 tracing_write_stub(struct file *filp, const char __user *ubuf,
5224 size_t count, loff_t *ppos)
5229 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5233 if (file->f_mode & FMODE_READ)
5234 ret = seq_lseek(file, offset, whence);
5236 file->f_pos = ret = 0;
5241 static const struct file_operations tracing_fops = {
5242 .open = tracing_open,
5244 .read_iter = seq_read_iter,
5245 .splice_read = copy_splice_read,
5246 .write = tracing_write_stub,
5247 .llseek = tracing_lseek,
5248 .release = tracing_release,
5251 static const struct file_operations show_traces_fops = {
5252 .open = show_traces_open,
5254 .llseek = seq_lseek,
5255 .release = show_traces_release,
5259 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5260 size_t count, loff_t *ppos)
5262 struct trace_array *tr = file_inode(filp)->i_private;
5266 len = snprintf(NULL, 0, "%*pb\n",
5267 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5268 mask_str = kmalloc(len, GFP_KERNEL);
5272 len = snprintf(mask_str, len, "%*pb\n",
5273 cpumask_pr_args(tr->tracing_cpumask));
5278 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5286 int tracing_set_cpumask(struct trace_array *tr,
5287 cpumask_var_t tracing_cpumask_new)
5294 local_irq_disable();
5295 arch_spin_lock(&tr->max_lock);
5296 for_each_tracing_cpu(cpu) {
5298 * Increase/decrease the disabled counter if we are
5299 * about to flip a bit in the cpumask:
5301 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5302 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5303 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5304 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5305 #ifdef CONFIG_TRACER_MAX_TRACE
5306 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5309 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5310 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5311 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5312 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5313 #ifdef CONFIG_TRACER_MAX_TRACE
5314 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5318 arch_spin_unlock(&tr->max_lock);
5321 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5327 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5328 size_t count, loff_t *ppos)
5330 struct trace_array *tr = file_inode(filp)->i_private;
5331 cpumask_var_t tracing_cpumask_new;
5334 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5337 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5341 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5345 free_cpumask_var(tracing_cpumask_new);
5350 free_cpumask_var(tracing_cpumask_new);
5355 static const struct file_operations tracing_cpumask_fops = {
5356 .open = tracing_open_generic_tr,
5357 .read = tracing_cpumask_read,
5358 .write = tracing_cpumask_write,
5359 .release = tracing_release_generic_tr,
5360 .llseek = generic_file_llseek,
5363 static int tracing_trace_options_show(struct seq_file *m, void *v)
5365 struct tracer_opt *trace_opts;
5366 struct trace_array *tr = m->private;
5370 mutex_lock(&trace_types_lock);
5371 tracer_flags = tr->current_trace->flags->val;
5372 trace_opts = tr->current_trace->flags->opts;
5374 for (i = 0; trace_options[i]; i++) {
5375 if (tr->trace_flags & (1 << i))
5376 seq_printf(m, "%s\n", trace_options[i]);
5378 seq_printf(m, "no%s\n", trace_options[i]);
5381 for (i = 0; trace_opts[i].name; i++) {
5382 if (tracer_flags & trace_opts[i].bit)
5383 seq_printf(m, "%s\n", trace_opts[i].name);
5385 seq_printf(m, "no%s\n", trace_opts[i].name);
5387 mutex_unlock(&trace_types_lock);
5392 static int __set_tracer_option(struct trace_array *tr,
5393 struct tracer_flags *tracer_flags,
5394 struct tracer_opt *opts, int neg)
5396 struct tracer *trace = tracer_flags->trace;
5399 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5404 tracer_flags->val &= ~opts->bit;
5406 tracer_flags->val |= opts->bit;
5410 /* Try to assign a tracer specific option */
5411 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5413 struct tracer *trace = tr->current_trace;
5414 struct tracer_flags *tracer_flags = trace->flags;
5415 struct tracer_opt *opts = NULL;
5418 for (i = 0; tracer_flags->opts[i].name; i++) {
5419 opts = &tracer_flags->opts[i];
5421 if (strcmp(cmp, opts->name) == 0)
5422 return __set_tracer_option(tr, trace->flags, opts, neg);
5428 /* Some tracers require overwrite to stay enabled */
5429 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5431 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5437 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5441 if ((mask == TRACE_ITER_RECORD_TGID) ||
5442 (mask == TRACE_ITER_RECORD_CMD))
5443 lockdep_assert_held(&event_mutex);
5445 /* do nothing if flag is already set */
5446 if (!!(tr->trace_flags & mask) == !!enabled)
5449 /* Give the tracer a chance to approve the change */
5450 if (tr->current_trace->flag_changed)
5451 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5455 tr->trace_flags |= mask;
5457 tr->trace_flags &= ~mask;
5459 if (mask == TRACE_ITER_RECORD_CMD)
5460 trace_event_enable_cmd_record(enabled);
5462 if (mask == TRACE_ITER_RECORD_TGID) {
5464 tgid_map_max = pid_max;
5465 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5469 * Pairs with smp_load_acquire() in
5470 * trace_find_tgid_ptr() to ensure that if it observes
5471 * the tgid_map we just allocated then it also observes
5472 * the corresponding tgid_map_max value.
5474 smp_store_release(&tgid_map, map);
5477 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5481 trace_event_enable_tgid_record(enabled);
5484 if (mask == TRACE_ITER_EVENT_FORK)
5485 trace_event_follow_fork(tr, enabled);
5487 if (mask == TRACE_ITER_FUNC_FORK)
5488 ftrace_pid_follow_fork(tr, enabled);
5490 if (mask == TRACE_ITER_OVERWRITE) {
5491 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5492 #ifdef CONFIG_TRACER_MAX_TRACE
5493 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5497 if (mask == TRACE_ITER_PRINTK) {
5498 trace_printk_start_stop_comm(enabled);
5499 trace_printk_control(enabled);
5505 int trace_set_options(struct trace_array *tr, char *option)
5510 size_t orig_len = strlen(option);
5513 cmp = strstrip(option);
5515 len = str_has_prefix(cmp, "no");
5521 mutex_lock(&event_mutex);
5522 mutex_lock(&trace_types_lock);
5524 ret = match_string(trace_options, -1, cmp);
5525 /* If no option could be set, test the specific tracer options */
5527 ret = set_tracer_option(tr, cmp, neg);
5529 ret = set_tracer_flag(tr, 1 << ret, !neg);
5531 mutex_unlock(&trace_types_lock);
5532 mutex_unlock(&event_mutex);
5535 * If the first trailing whitespace is replaced with '\0' by strstrip,
5536 * turn it back into a space.
5538 if (orig_len > strlen(option))
5539 option[strlen(option)] = ' ';
5544 static void __init apply_trace_boot_options(void)
5546 char *buf = trace_boot_options_buf;
5550 option = strsep(&buf, ",");
5556 trace_set_options(&global_trace, option);
5558 /* Put back the comma to allow this to be called again */
5565 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5566 size_t cnt, loff_t *ppos)
5568 struct seq_file *m = filp->private_data;
5569 struct trace_array *tr = m->private;
5573 if (cnt >= sizeof(buf))
5576 if (copy_from_user(buf, ubuf, cnt))
5581 ret = trace_set_options(tr, buf);
5590 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5592 struct trace_array *tr = inode->i_private;
5595 ret = tracing_check_open_get_tr(tr);
5599 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5601 trace_array_put(tr);
5606 static const struct file_operations tracing_iter_fops = {
5607 .open = tracing_trace_options_open,
5609 .llseek = seq_lseek,
5610 .release = tracing_single_release_tr,
5611 .write = tracing_trace_options_write,
5614 static const char readme_msg[] =
5615 "tracing mini-HOWTO:\n\n"
5616 "# echo 0 > tracing_on : quick way to disable tracing\n"
5617 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5618 " Important files:\n"
5619 " trace\t\t\t- The static contents of the buffer\n"
5620 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5621 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5622 " current_tracer\t- function and latency tracers\n"
5623 " available_tracers\t- list of configured tracers for current_tracer\n"
5624 " error_log\t- error log for failed commands (that support it)\n"
5625 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5626 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5627 " trace_clock\t\t- change the clock used to order events\n"
5628 " local: Per cpu clock but may not be synced across CPUs\n"
5629 " global: Synced across CPUs but slows tracing down.\n"
5630 " counter: Not a clock, but just an increment\n"
5631 " uptime: Jiffy counter from time of boot\n"
5632 " perf: Same clock that perf events use\n"
5633 #ifdef CONFIG_X86_64
5634 " x86-tsc: TSC cycle counter\n"
5636 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5637 " delta: Delta difference against a buffer-wide timestamp\n"
5638 " absolute: Absolute (standalone) timestamp\n"
5639 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5640 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5641 " tracing_cpumask\t- Limit which CPUs to trace\n"
5642 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5643 "\t\t\t Remove sub-buffer with rmdir\n"
5644 " trace_options\t\t- Set format or modify how tracing happens\n"
5645 "\t\t\t Disable an option by prefixing 'no' to the\n"
5646 "\t\t\t option name\n"
5647 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5648 #ifdef CONFIG_DYNAMIC_FTRACE
5649 "\n available_filter_functions - list of functions that can be filtered on\n"
5650 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5651 "\t\t\t functions\n"
5652 "\t accepts: func_full_name or glob-matching-pattern\n"
5653 "\t modules: Can select a group via module\n"
5654 "\t Format: :mod:<module-name>\n"
5655 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5656 "\t triggers: a command to perform when function is hit\n"
5657 "\t Format: <function>:<trigger>[:count]\n"
5658 "\t trigger: traceon, traceoff\n"
5659 "\t\t enable_event:<system>:<event>\n"
5660 "\t\t disable_event:<system>:<event>\n"
5661 #ifdef CONFIG_STACKTRACE
5664 #ifdef CONFIG_TRACER_SNAPSHOT
5669 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5670 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5671 "\t The first one will disable tracing every time do_fault is hit\n"
5672 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5673 "\t The first time do trap is hit and it disables tracing, the\n"
5674 "\t counter will decrement to 2. If tracing is already disabled,\n"
5675 "\t the counter will not decrement. It only decrements when the\n"
5676 "\t trigger did work\n"
5677 "\t To remove trigger without count:\n"
5678 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5679 "\t To remove trigger with a count:\n"
5680 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5681 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5682 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5683 "\t modules: Can select a group via module command :mod:\n"
5684 "\t Does not accept triggers\n"
5685 #endif /* CONFIG_DYNAMIC_FTRACE */
5686 #ifdef CONFIG_FUNCTION_TRACER
5687 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5689 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5692 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5693 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5694 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5695 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5697 #ifdef CONFIG_TRACER_SNAPSHOT
5698 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5699 "\t\t\t snapshot buffer. Read the contents for more\n"
5700 "\t\t\t information\n"
5702 #ifdef CONFIG_STACK_TRACER
5703 " stack_trace\t\t- Shows the max stack trace when active\n"
5704 " stack_max_size\t- Shows current max stack size that was traced\n"
5705 "\t\t\t Write into this file to reset the max size (trigger a\n"
5706 "\t\t\t new trace)\n"
5707 #ifdef CONFIG_DYNAMIC_FTRACE
5708 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5711 #endif /* CONFIG_STACK_TRACER */
5712 #ifdef CONFIG_DYNAMIC_EVENTS
5713 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5714 "\t\t\t Write into this file to define/undefine new trace events.\n"
5716 #ifdef CONFIG_KPROBE_EVENTS
5717 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5718 "\t\t\t Write into this file to define/undefine new trace events.\n"
5720 #ifdef CONFIG_UPROBE_EVENTS
5721 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5722 "\t\t\t Write into this file to define/undefine new trace events.\n"
5724 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5725 defined(CONFIG_FPROBE_EVENTS)
5726 "\t accepts: event-definitions (one definition per line)\n"
5727 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5728 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5729 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5731 #ifdef CONFIG_FPROBE_EVENTS
5732 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5733 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5735 #ifdef CONFIG_HIST_TRIGGERS
5736 "\t s:[synthetic/]<event> <field> [<field>]\n"
5738 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5739 "\t -:[<group>/][<event>]\n"
5740 #ifdef CONFIG_KPROBE_EVENTS
5741 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5742 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5744 #ifdef CONFIG_UPROBE_EVENTS
5745 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5747 "\t args: <name>=fetcharg[:type]\n"
5748 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5749 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5750 #ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5751 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5752 "\t <argname>[->field[->field|.field...]],\n"
5754 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5757 "\t $stack<index>, $stack, $retval, $comm,\n"
5759 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5760 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5761 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5762 "\t symstr, <type>\\[<array-size>\\]\n"
5763 #ifdef CONFIG_HIST_TRIGGERS
5764 "\t field: <stype> <name>;\n"
5765 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5766 "\t [unsigned] char/int/long\n"
5768 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5769 "\t of the <attached-group>/<attached-event>.\n"
5771 " events/\t\t- Directory containing all trace event subsystems:\n"
5772 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5773 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5774 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5776 " filter\t\t- If set, only events passing filter are traced\n"
5777 " events/<system>/<event>/\t- Directory containing control files for\n"
5779 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5780 " filter\t\t- If set, only events passing filter are traced\n"
5781 " trigger\t\t- If set, a command to perform when event is hit\n"
5782 "\t Format: <trigger>[:count][if <filter>]\n"
5783 "\t trigger: traceon, traceoff\n"
5784 "\t enable_event:<system>:<event>\n"
5785 "\t disable_event:<system>:<event>\n"
5786 #ifdef CONFIG_HIST_TRIGGERS
5787 "\t enable_hist:<system>:<event>\n"
5788 "\t disable_hist:<system>:<event>\n"
5790 #ifdef CONFIG_STACKTRACE
5793 #ifdef CONFIG_TRACER_SNAPSHOT
5796 #ifdef CONFIG_HIST_TRIGGERS
5797 "\t\t hist (see below)\n"
5799 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5800 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5801 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5802 "\t events/block/block_unplug/trigger\n"
5803 "\t The first disables tracing every time block_unplug is hit.\n"
5804 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5805 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5806 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5807 "\t Like function triggers, the counter is only decremented if it\n"
5808 "\t enabled or disabled tracing.\n"
5809 "\t To remove a trigger without a count:\n"
5810 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5811 "\t To remove a trigger with a count:\n"
5812 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5813 "\t Filters can be ignored when removing a trigger.\n"
5814 #ifdef CONFIG_HIST_TRIGGERS
5815 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5816 "\t Format: hist:keys=<field1[,field2,...]>\n"
5817 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5818 "\t [:values=<field1[,field2,...]>]\n"
5819 "\t [:sort=<field1[,field2,...]>]\n"
5820 "\t [:size=#entries]\n"
5821 "\t [:pause][:continue][:clear]\n"
5822 "\t [:name=histname1]\n"
5823 "\t [:nohitcount]\n"
5824 "\t [:<handler>.<action>]\n"
5825 "\t [if <filter>]\n\n"
5826 "\t Note, special fields can be used as well:\n"
5827 "\t common_timestamp - to record current timestamp\n"
5828 "\t common_cpu - to record the CPU the event happened on\n"
5830 "\t A hist trigger variable can be:\n"
5831 "\t - a reference to a field e.g. x=current_timestamp,\n"
5832 "\t - a reference to another variable e.g. y=$x,\n"
5833 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5834 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5836 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5837 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5838 "\t variable reference, field or numeric literal.\n"
5840 "\t When a matching event is hit, an entry is added to a hash\n"
5841 "\t table using the key(s) and value(s) named, and the value of a\n"
5842 "\t sum called 'hitcount' is incremented. Keys and values\n"
5843 "\t correspond to fields in the event's format description. Keys\n"
5844 "\t can be any field, or the special string 'common_stacktrace'.\n"
5845 "\t Compound keys consisting of up to two fields can be specified\n"
5846 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5847 "\t fields. Sort keys consisting of up to two fields can be\n"
5848 "\t specified using the 'sort' keyword. The sort direction can\n"
5849 "\t be modified by appending '.descending' or '.ascending' to a\n"
5850 "\t sort field. The 'size' parameter can be used to specify more\n"
5851 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5852 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5853 "\t its histogram data will be shared with other triggers of the\n"
5854 "\t same name, and trigger hits will update this common data.\n\n"
5855 "\t Reading the 'hist' file for the event will dump the hash\n"
5856 "\t table in its entirety to stdout. If there are multiple hist\n"
5857 "\t triggers attached to an event, there will be a table for each\n"
5858 "\t trigger in the output. The table displayed for a named\n"
5859 "\t trigger will be the same as any other instance having the\n"
5860 "\t same name. The default format used to display a given field\n"
5861 "\t can be modified by appending any of the following modifiers\n"
5862 "\t to the field name, as applicable:\n\n"
5863 "\t .hex display a number as a hex value\n"
5864 "\t .sym display an address as a symbol\n"
5865 "\t .sym-offset display an address as a symbol and offset\n"
5866 "\t .execname display a common_pid as a program name\n"
5867 "\t .syscall display a syscall id as a syscall name\n"
5868 "\t .log2 display log2 value rather than raw number\n"
5869 "\t .buckets=size display values in groups of size rather than raw number\n"
5870 "\t .usecs display a common_timestamp in microseconds\n"
5871 "\t .percent display a number of percentage value\n"
5872 "\t .graph display a bar-graph of a value\n\n"
5873 "\t The 'pause' parameter can be used to pause an existing hist\n"
5874 "\t trigger or to start a hist trigger but not log any events\n"
5875 "\t until told to do so. 'continue' can be used to start or\n"
5876 "\t restart a paused hist trigger.\n\n"
5877 "\t The 'clear' parameter will clear the contents of a running\n"
5878 "\t hist trigger and leave its current paused/active state\n"
5880 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5881 "\t raw hitcount in the histogram.\n\n"
5882 "\t The enable_hist and disable_hist triggers can be used to\n"
5883 "\t have one event conditionally start and stop another event's\n"
5884 "\t already-attached hist trigger. The syntax is analogous to\n"
5885 "\t the enable_event and disable_event triggers.\n\n"
5886 "\t Hist trigger handlers and actions are executed whenever a\n"
5887 "\t a histogram entry is added or updated. They take the form:\n\n"
5888 "\t <handler>.<action>\n\n"
5889 "\t The available handlers are:\n\n"
5890 "\t onmatch(matching.event) - invoke on addition or update\n"
5891 "\t onmax(var) - invoke if var exceeds current max\n"
5892 "\t onchange(var) - invoke action if var changes\n\n"
5893 "\t The available actions are:\n\n"
5894 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5895 "\t save(field,...) - save current event fields\n"
5896 #ifdef CONFIG_TRACER_SNAPSHOT
5897 "\t snapshot() - snapshot the trace buffer\n\n"
5899 #ifdef CONFIG_SYNTH_EVENTS
5900 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5901 "\t Write into this file to define/undefine new synthetic events.\n"
5902 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5908 tracing_readme_read(struct file *filp, char __user *ubuf,
5909 size_t cnt, loff_t *ppos)
5911 return simple_read_from_buffer(ubuf, cnt, ppos,
5912 readme_msg, strlen(readme_msg));
5915 static const struct file_operations tracing_readme_fops = {
5916 .open = tracing_open_generic,
5917 .read = tracing_readme_read,
5918 .llseek = generic_file_llseek,
5921 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5925 return trace_find_tgid_ptr(pid);
5928 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5932 return trace_find_tgid_ptr(pid);
5935 static void saved_tgids_stop(struct seq_file *m, void *v)
5939 static int saved_tgids_show(struct seq_file *m, void *v)
5941 int *entry = (int *)v;
5942 int pid = entry - tgid_map;
5948 seq_printf(m, "%d %d\n", pid, tgid);
5952 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5953 .start = saved_tgids_start,
5954 .stop = saved_tgids_stop,
5955 .next = saved_tgids_next,
5956 .show = saved_tgids_show,
5959 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5963 ret = tracing_check_open_get_tr(NULL);
5967 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5971 static const struct file_operations tracing_saved_tgids_fops = {
5972 .open = tracing_saved_tgids_open,
5974 .llseek = seq_lseek,
5975 .release = seq_release,
5978 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5980 unsigned int *ptr = v;
5982 if (*pos || m->count)
5987 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5989 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5998 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
6004 arch_spin_lock(&trace_cmdline_lock);
6006 v = &savedcmd->map_cmdline_to_pid[0];
6008 v = saved_cmdlines_next(m, v, &l);
6016 static void saved_cmdlines_stop(struct seq_file *m, void *v)
6018 arch_spin_unlock(&trace_cmdline_lock);
6022 static int saved_cmdlines_show(struct seq_file *m, void *v)
6024 char buf[TASK_COMM_LEN];
6025 unsigned int *pid = v;
6027 __trace_find_cmdline(*pid, buf);
6028 seq_printf(m, "%d %s\n", *pid, buf);
6032 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
6033 .start = saved_cmdlines_start,
6034 .next = saved_cmdlines_next,
6035 .stop = saved_cmdlines_stop,
6036 .show = saved_cmdlines_show,
6039 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
6043 ret = tracing_check_open_get_tr(NULL);
6047 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
6050 static const struct file_operations tracing_saved_cmdlines_fops = {
6051 .open = tracing_saved_cmdlines_open,
6053 .llseek = seq_lseek,
6054 .release = seq_release,
6058 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
6059 size_t cnt, loff_t *ppos)
6065 arch_spin_lock(&trace_cmdline_lock);
6066 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
6067 arch_spin_unlock(&trace_cmdline_lock);
6070 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6073 static int tracing_resize_saved_cmdlines(unsigned int val)
6075 struct saved_cmdlines_buffer *s, *savedcmd_temp;
6077 s = allocate_cmdlines_buffer(val);
6082 arch_spin_lock(&trace_cmdline_lock);
6083 savedcmd_temp = savedcmd;
6085 arch_spin_unlock(&trace_cmdline_lock);
6087 free_saved_cmdlines_buffer(savedcmd_temp);
6093 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
6094 size_t cnt, loff_t *ppos)
6099 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6103 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
6104 if (!val || val > PID_MAX_DEFAULT)
6107 ret = tracing_resize_saved_cmdlines((unsigned int)val);
6116 static const struct file_operations tracing_saved_cmdlines_size_fops = {
6117 .open = tracing_open_generic,
6118 .read = tracing_saved_cmdlines_size_read,
6119 .write = tracing_saved_cmdlines_size_write,
6122 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
6123 static union trace_eval_map_item *
6124 update_eval_map(union trace_eval_map_item *ptr)
6126 if (!ptr->map.eval_string) {
6127 if (ptr->tail.next) {
6128 ptr = ptr->tail.next;
6129 /* Set ptr to the next real item (skip head) */
6137 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6139 union trace_eval_map_item *ptr = v;
6142 * Paranoid! If ptr points to end, we don't want to increment past it.
6143 * This really should never happen.
6146 ptr = update_eval_map(ptr);
6147 if (WARN_ON_ONCE(!ptr))
6151 ptr = update_eval_map(ptr);
6156 static void *eval_map_start(struct seq_file *m, loff_t *pos)
6158 union trace_eval_map_item *v;
6161 mutex_lock(&trace_eval_mutex);
6163 v = trace_eval_maps;
6167 while (v && l < *pos) {
6168 v = eval_map_next(m, v, &l);
6174 static void eval_map_stop(struct seq_file *m, void *v)
6176 mutex_unlock(&trace_eval_mutex);
6179 static int eval_map_show(struct seq_file *m, void *v)
6181 union trace_eval_map_item *ptr = v;
6183 seq_printf(m, "%s %ld (%s)\n",
6184 ptr->map.eval_string, ptr->map.eval_value,
6190 static const struct seq_operations tracing_eval_map_seq_ops = {
6191 .start = eval_map_start,
6192 .next = eval_map_next,
6193 .stop = eval_map_stop,
6194 .show = eval_map_show,
6197 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6201 ret = tracing_check_open_get_tr(NULL);
6205 return seq_open(filp, &tracing_eval_map_seq_ops);
6208 static const struct file_operations tracing_eval_map_fops = {
6209 .open = tracing_eval_map_open,
6211 .llseek = seq_lseek,
6212 .release = seq_release,
6215 static inline union trace_eval_map_item *
6216 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6218 /* Return tail of array given the head */
6219 return ptr + ptr->head.length + 1;
6223 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6226 struct trace_eval_map **stop;
6227 struct trace_eval_map **map;
6228 union trace_eval_map_item *map_array;
6229 union trace_eval_map_item *ptr;
6234 * The trace_eval_maps contains the map plus a head and tail item,
6235 * where the head holds the module and length of array, and the
6236 * tail holds a pointer to the next list.
6238 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6240 pr_warn("Unable to allocate trace eval mapping\n");
6244 mutex_lock(&trace_eval_mutex);
6246 if (!trace_eval_maps)
6247 trace_eval_maps = map_array;
6249 ptr = trace_eval_maps;
6251 ptr = trace_eval_jmp_to_tail(ptr);
6252 if (!ptr->tail.next)
6254 ptr = ptr->tail.next;
6257 ptr->tail.next = map_array;
6259 map_array->head.mod = mod;
6260 map_array->head.length = len;
6263 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6264 map_array->map = **map;
6267 memset(map_array, 0, sizeof(*map_array));
6269 mutex_unlock(&trace_eval_mutex);
6272 static void trace_create_eval_file(struct dentry *d_tracer)
6274 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6275 NULL, &tracing_eval_map_fops);
6278 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
6279 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6280 static inline void trace_insert_eval_map_file(struct module *mod,
6281 struct trace_eval_map **start, int len) { }
6282 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6284 static void trace_insert_eval_map(struct module *mod,
6285 struct trace_eval_map **start, int len)
6287 struct trace_eval_map **map;
6294 trace_event_eval_update(map, len);
6296 trace_insert_eval_map_file(mod, start, len);
6300 tracing_set_trace_read(struct file *filp, char __user *ubuf,
6301 size_t cnt, loff_t *ppos)
6303 struct trace_array *tr = filp->private_data;
6304 char buf[MAX_TRACER_SIZE+2];
6307 mutex_lock(&trace_types_lock);
6308 r = sprintf(buf, "%s\n", tr->current_trace->name);
6309 mutex_unlock(&trace_types_lock);
6311 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6314 int tracer_init(struct tracer *t, struct trace_array *tr)
6316 tracing_reset_online_cpus(&tr->array_buffer);
6320 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6324 for_each_tracing_cpu(cpu)
6325 per_cpu_ptr(buf->data, cpu)->entries = val;
6328 static void update_buffer_entries(struct array_buffer *buf, int cpu)
6330 if (cpu == RING_BUFFER_ALL_CPUS) {
6331 set_buffer_entries(buf, ring_buffer_size(buf->buffer, 0));
6333 per_cpu_ptr(buf->data, cpu)->entries = ring_buffer_size(buf->buffer, cpu);
6337 #ifdef CONFIG_TRACER_MAX_TRACE
6338 /* resize @tr's buffer to the size of @size_tr's entries */
6339 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6340 struct array_buffer *size_buf, int cpu_id)
6344 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6345 for_each_tracing_cpu(cpu) {
6346 ret = ring_buffer_resize(trace_buf->buffer,
6347 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6350 per_cpu_ptr(trace_buf->data, cpu)->entries =
6351 per_cpu_ptr(size_buf->data, cpu)->entries;
6354 ret = ring_buffer_resize(trace_buf->buffer,
6355 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6357 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6358 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6363 #endif /* CONFIG_TRACER_MAX_TRACE */
6365 static int __tracing_resize_ring_buffer(struct trace_array *tr,
6366 unsigned long size, int cpu)
6371 * If kernel or user changes the size of the ring buffer
6372 * we use the size that was given, and we can forget about
6373 * expanding it later.
6375 trace_set_ring_buffer_expanded(tr);
6377 /* May be called before buffers are initialized */
6378 if (!tr->array_buffer.buffer)
6381 /* Do not allow tracing while resizing ring buffer */
6382 tracing_stop_tr(tr);
6384 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6388 #ifdef CONFIG_TRACER_MAX_TRACE
6389 if (!tr->allocated_snapshot)
6392 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6394 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6395 &tr->array_buffer, cpu);
6398 * AARGH! We are left with different
6399 * size max buffer!!!!
6400 * The max buffer is our "snapshot" buffer.
6401 * When a tracer needs a snapshot (one of the
6402 * latency tracers), it swaps the max buffer
6403 * with the saved snap shot. We succeeded to
6404 * update the size of the main buffer, but failed to
6405 * update the size of the max buffer. But when we tried
6406 * to reset the main buffer to the original size, we
6407 * failed there too. This is very unlikely to
6408 * happen, but if it does, warn and kill all
6412 tracing_disabled = 1;
6417 update_buffer_entries(&tr->max_buffer, cpu);
6420 #endif /* CONFIG_TRACER_MAX_TRACE */
6422 update_buffer_entries(&tr->array_buffer, cpu);
6424 tracing_start_tr(tr);
6428 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6429 unsigned long size, int cpu_id)
6433 mutex_lock(&trace_types_lock);
6435 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6436 /* make sure, this cpu is enabled in the mask */
6437 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6443 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6448 mutex_unlock(&trace_types_lock);
6455 * tracing_update_buffers - used by tracing facility to expand ring buffers
6456 * @tr: The tracing instance
6458 * To save on memory when the tracing is never used on a system with it
6459 * configured in. The ring buffers are set to a minimum size. But once
6460 * a user starts to use the tracing facility, then they need to grow
6461 * to their default size.
6463 * This function is to be called when a tracer is about to be used.
6465 int tracing_update_buffers(struct trace_array *tr)
6469 mutex_lock(&trace_types_lock);
6470 if (!tr->ring_buffer_expanded)
6471 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6472 RING_BUFFER_ALL_CPUS);
6473 mutex_unlock(&trace_types_lock);
6478 struct trace_option_dentry;
6481 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6484 * Used to clear out the tracer before deletion of an instance.
6485 * Must have trace_types_lock held.
6487 static void tracing_set_nop(struct trace_array *tr)
6489 if (tr->current_trace == &nop_trace)
6492 tr->current_trace->enabled--;
6494 if (tr->current_trace->reset)
6495 tr->current_trace->reset(tr);
6497 tr->current_trace = &nop_trace;
6500 static bool tracer_options_updated;
6502 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6504 /* Only enable if the directory has been created already. */
6508 /* Only create trace option files after update_tracer_options finish */
6509 if (!tracer_options_updated)
6512 create_trace_option_files(tr, t);
6515 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6518 #ifdef CONFIG_TRACER_MAX_TRACE
6523 mutex_lock(&trace_types_lock);
6525 if (!tr->ring_buffer_expanded) {
6526 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6527 RING_BUFFER_ALL_CPUS);
6533 for (t = trace_types; t; t = t->next) {
6534 if (strcmp(t->name, buf) == 0)
6541 if (t == tr->current_trace)
6544 #ifdef CONFIG_TRACER_SNAPSHOT
6545 if (t->use_max_tr) {
6546 local_irq_disable();
6547 arch_spin_lock(&tr->max_lock);
6548 if (tr->cond_snapshot)
6550 arch_spin_unlock(&tr->max_lock);
6556 /* Some tracers won't work on kernel command line */
6557 if (system_state < SYSTEM_RUNNING && t->noboot) {
6558 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6563 /* Some tracers are only allowed for the top level buffer */
6564 if (!trace_ok_for_array(t, tr)) {
6569 /* If trace pipe files are being read, we can't change the tracer */
6570 if (tr->trace_ref) {
6575 trace_branch_disable();
6577 tr->current_trace->enabled--;
6579 if (tr->current_trace->reset)
6580 tr->current_trace->reset(tr);
6582 #ifdef CONFIG_TRACER_MAX_TRACE
6583 had_max_tr = tr->current_trace->use_max_tr;
6585 /* Current trace needs to be nop_trace before synchronize_rcu */
6586 tr->current_trace = &nop_trace;
6588 if (had_max_tr && !t->use_max_tr) {
6590 * We need to make sure that the update_max_tr sees that
6591 * current_trace changed to nop_trace to keep it from
6592 * swapping the buffers after we resize it.
6593 * The update_max_tr is called from interrupts disabled
6594 * so a synchronized_sched() is sufficient.
6600 if (t->use_max_tr && !tr->allocated_snapshot) {
6601 ret = tracing_alloc_snapshot_instance(tr);
6606 tr->current_trace = &nop_trace;
6610 ret = tracer_init(t, tr);
6615 tr->current_trace = t;
6616 tr->current_trace->enabled++;
6617 trace_branch_enable(tr);
6619 mutex_unlock(&trace_types_lock);
6625 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6626 size_t cnt, loff_t *ppos)
6628 struct trace_array *tr = filp->private_data;
6629 char buf[MAX_TRACER_SIZE+1];
6636 if (cnt > MAX_TRACER_SIZE)
6637 cnt = MAX_TRACER_SIZE;
6639 if (copy_from_user(buf, ubuf, cnt))
6646 err = tracing_set_tracer(tr, name);
6656 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6657 size_t cnt, loff_t *ppos)
6662 r = snprintf(buf, sizeof(buf), "%ld\n",
6663 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6664 if (r > sizeof(buf))
6666 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6670 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6671 size_t cnt, loff_t *ppos)
6676 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6686 tracing_thresh_read(struct file *filp, char __user *ubuf,
6687 size_t cnt, loff_t *ppos)
6689 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6693 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6694 size_t cnt, loff_t *ppos)
6696 struct trace_array *tr = filp->private_data;
6699 mutex_lock(&trace_types_lock);
6700 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6704 if (tr->current_trace->update_thresh) {
6705 ret = tr->current_trace->update_thresh(tr);
6712 mutex_unlock(&trace_types_lock);
6717 #ifdef CONFIG_TRACER_MAX_TRACE
6720 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6721 size_t cnt, loff_t *ppos)
6723 struct trace_array *tr = filp->private_data;
6725 return tracing_nsecs_read(&tr->max_latency, ubuf, cnt, ppos);
6729 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6730 size_t cnt, loff_t *ppos)
6732 struct trace_array *tr = filp->private_data;
6734 return tracing_nsecs_write(&tr->max_latency, ubuf, cnt, ppos);
6739 static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6741 if (cpu == RING_BUFFER_ALL_CPUS) {
6742 if (cpumask_empty(tr->pipe_cpumask)) {
6743 cpumask_setall(tr->pipe_cpumask);
6746 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6747 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6753 static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6755 if (cpu == RING_BUFFER_ALL_CPUS) {
6756 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6757 cpumask_clear(tr->pipe_cpumask);
6759 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6760 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6764 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6766 struct trace_array *tr = inode->i_private;
6767 struct trace_iterator *iter;
6771 ret = tracing_check_open_get_tr(tr);
6775 mutex_lock(&trace_types_lock);
6776 cpu = tracing_get_cpu(inode);
6777 ret = open_pipe_on_cpu(tr, cpu);
6779 goto fail_pipe_on_cpu;
6781 /* create a buffer to store the information to pass to userspace */
6782 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6785 goto fail_alloc_iter;
6788 trace_seq_init(&iter->seq);
6789 iter->trace = tr->current_trace;
6791 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6796 /* trace pipe does not show start of buffer */
6797 cpumask_setall(iter->started);
6799 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6800 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6802 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6803 if (trace_clocks[tr->clock_id].in_ns)
6804 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6807 iter->array_buffer = &tr->array_buffer;
6808 iter->cpu_file = cpu;
6809 mutex_init(&iter->mutex);
6810 filp->private_data = iter;
6812 if (iter->trace->pipe_open)
6813 iter->trace->pipe_open(iter);
6815 nonseekable_open(inode, filp);
6819 mutex_unlock(&trace_types_lock);
6825 close_pipe_on_cpu(tr, cpu);
6827 __trace_array_put(tr);
6828 mutex_unlock(&trace_types_lock);
6832 static int tracing_release_pipe(struct inode *inode, struct file *file)
6834 struct trace_iterator *iter = file->private_data;
6835 struct trace_array *tr = inode->i_private;
6837 mutex_lock(&trace_types_lock);
6841 if (iter->trace->pipe_close)
6842 iter->trace->pipe_close(iter);
6843 close_pipe_on_cpu(tr, iter->cpu_file);
6844 mutex_unlock(&trace_types_lock);
6846 free_trace_iter_content(iter);
6849 trace_array_put(tr);
6855 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6857 struct trace_array *tr = iter->tr;
6859 /* Iterators are static, they should be filled or empty */
6860 if (trace_buffer_iter(iter, iter->cpu_file))
6861 return EPOLLIN | EPOLLRDNORM;
6863 if (tr->trace_flags & TRACE_ITER_BLOCK)
6865 * Always select as readable when in blocking mode
6867 return EPOLLIN | EPOLLRDNORM;
6869 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6870 filp, poll_table, iter->tr->buffer_percent);
6874 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6876 struct trace_iterator *iter = filp->private_data;
6878 return trace_poll(iter, filp, poll_table);
6881 /* Must be called with iter->mutex held. */
6882 static int tracing_wait_pipe(struct file *filp)
6884 struct trace_iterator *iter = filp->private_data;
6887 while (trace_empty(iter)) {
6889 if ((filp->f_flags & O_NONBLOCK)) {
6894 * We block until we read something and tracing is disabled.
6895 * We still block if tracing is disabled, but we have never
6896 * read anything. This allows a user to cat this file, and
6897 * then enable tracing. But after we have read something,
6898 * we give an EOF when tracing is again disabled.
6900 * iter->pos will be 0 if we haven't read anything.
6902 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6905 mutex_unlock(&iter->mutex);
6907 ret = wait_on_pipe(iter, 0);
6909 mutex_lock(&iter->mutex);
6922 tracing_read_pipe(struct file *filp, char __user *ubuf,
6923 size_t cnt, loff_t *ppos)
6925 struct trace_iterator *iter = filp->private_data;
6929 * Avoid more than one consumer on a single file descriptor
6930 * This is just a matter of traces coherency, the ring buffer itself
6933 mutex_lock(&iter->mutex);
6935 /* return any leftover data */
6936 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6940 trace_seq_init(&iter->seq);
6942 if (iter->trace->read) {
6943 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6949 sret = tracing_wait_pipe(filp);
6953 /* stop when tracing is finished */
6954 if (trace_empty(iter)) {
6959 if (cnt >= TRACE_SEQ_BUFFER_SIZE)
6960 cnt = TRACE_SEQ_BUFFER_SIZE - 1;
6962 /* reset all but tr, trace, and overruns */
6963 trace_iterator_reset(iter);
6964 cpumask_clear(iter->started);
6965 trace_seq_init(&iter->seq);
6967 trace_event_read_lock();
6968 trace_access_lock(iter->cpu_file);
6969 while (trace_find_next_entry_inc(iter) != NULL) {
6970 enum print_line_t ret;
6971 int save_len = iter->seq.seq.len;
6973 ret = print_trace_line(iter);
6974 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6976 * If one print_trace_line() fills entire trace_seq in one shot,
6977 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6978 * In this case, we need to consume it, otherwise, loop will peek
6979 * this event next time, resulting in an infinite loop.
6981 if (save_len == 0) {
6983 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6984 trace_consume(iter);
6988 /* In other cases, don't print partial lines */
6989 iter->seq.seq.len = save_len;
6992 if (ret != TRACE_TYPE_NO_CONSUME)
6993 trace_consume(iter);
6995 if (trace_seq_used(&iter->seq) >= cnt)
6999 * Setting the full flag means we reached the trace_seq buffer
7000 * size and we should leave by partial output condition above.
7001 * One of the trace_seq_* functions is not used properly.
7003 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
7006 trace_access_unlock(iter->cpu_file);
7007 trace_event_read_unlock();
7009 /* Now copy what we have to the user */
7010 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
7011 if (iter->seq.readpos >= trace_seq_used(&iter->seq))
7012 trace_seq_init(&iter->seq);
7015 * If there was nothing to send to user, in spite of consuming trace
7016 * entries, go back to wait for more entries.
7022 mutex_unlock(&iter->mutex);
7027 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
7030 __free_page(spd->pages[idx]);
7034 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
7040 /* Seq buffer is page-sized, exactly what we need. */
7042 save_len = iter->seq.seq.len;
7043 ret = print_trace_line(iter);
7045 if (trace_seq_has_overflowed(&iter->seq)) {
7046 iter->seq.seq.len = save_len;
7051 * This should not be hit, because it should only
7052 * be set if the iter->seq overflowed. But check it
7053 * anyway to be safe.
7055 if (ret == TRACE_TYPE_PARTIAL_LINE) {
7056 iter->seq.seq.len = save_len;
7060 count = trace_seq_used(&iter->seq) - save_len;
7063 iter->seq.seq.len = save_len;
7067 if (ret != TRACE_TYPE_NO_CONSUME)
7068 trace_consume(iter);
7070 if (!trace_find_next_entry_inc(iter)) {
7080 static ssize_t tracing_splice_read_pipe(struct file *filp,
7082 struct pipe_inode_info *pipe,
7086 struct page *pages_def[PIPE_DEF_BUFFERS];
7087 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7088 struct trace_iterator *iter = filp->private_data;
7089 struct splice_pipe_desc spd = {
7091 .partial = partial_def,
7092 .nr_pages = 0, /* This gets updated below. */
7093 .nr_pages_max = PIPE_DEF_BUFFERS,
7094 .ops = &default_pipe_buf_ops,
7095 .spd_release = tracing_spd_release_pipe,
7101 if (splice_grow_spd(pipe, &spd))
7104 mutex_lock(&iter->mutex);
7106 if (iter->trace->splice_read) {
7107 ret = iter->trace->splice_read(iter, filp,
7108 ppos, pipe, len, flags);
7113 ret = tracing_wait_pipe(filp);
7117 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
7122 trace_event_read_lock();
7123 trace_access_lock(iter->cpu_file);
7125 /* Fill as many pages as possible. */
7126 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
7127 spd.pages[i] = alloc_page(GFP_KERNEL);
7131 rem = tracing_fill_pipe_page(rem, iter);
7133 /* Copy the data into the page, so we can start over. */
7134 ret = trace_seq_to_buffer(&iter->seq,
7135 page_address(spd.pages[i]),
7136 trace_seq_used(&iter->seq));
7138 __free_page(spd.pages[i]);
7141 spd.partial[i].offset = 0;
7142 spd.partial[i].len = trace_seq_used(&iter->seq);
7144 trace_seq_init(&iter->seq);
7147 trace_access_unlock(iter->cpu_file);
7148 trace_event_read_unlock();
7149 mutex_unlock(&iter->mutex);
7154 ret = splice_to_pipe(pipe, &spd);
7158 splice_shrink_spd(&spd);
7162 mutex_unlock(&iter->mutex);
7167 tracing_entries_read(struct file *filp, char __user *ubuf,
7168 size_t cnt, loff_t *ppos)
7170 struct inode *inode = file_inode(filp);
7171 struct trace_array *tr = inode->i_private;
7172 int cpu = tracing_get_cpu(inode);
7177 mutex_lock(&trace_types_lock);
7179 if (cpu == RING_BUFFER_ALL_CPUS) {
7180 int cpu, buf_size_same;
7185 /* check if all cpu sizes are same */
7186 for_each_tracing_cpu(cpu) {
7187 /* fill in the size from first enabled cpu */
7189 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7190 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7196 if (buf_size_same) {
7197 if (!tr->ring_buffer_expanded)
7198 r = sprintf(buf, "%lu (expanded: %lu)\n",
7200 trace_buf_size >> 10);
7202 r = sprintf(buf, "%lu\n", size >> 10);
7204 r = sprintf(buf, "X\n");
7206 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7208 mutex_unlock(&trace_types_lock);
7210 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7215 tracing_entries_write(struct file *filp, const char __user *ubuf,
7216 size_t cnt, loff_t *ppos)
7218 struct inode *inode = file_inode(filp);
7219 struct trace_array *tr = inode->i_private;
7223 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7227 /* must have at least 1 entry */
7231 /* value is in KB */
7233 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7243 tracing_total_entries_read(struct file *filp, char __user *ubuf,
7244 size_t cnt, loff_t *ppos)
7246 struct trace_array *tr = filp->private_data;
7249 unsigned long size = 0, expanded_size = 0;
7251 mutex_lock(&trace_types_lock);
7252 for_each_tracing_cpu(cpu) {
7253 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7254 if (!tr->ring_buffer_expanded)
7255 expanded_size += trace_buf_size >> 10;
7257 if (tr->ring_buffer_expanded)
7258 r = sprintf(buf, "%lu\n", size);
7260 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7261 mutex_unlock(&trace_types_lock);
7263 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7267 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7268 size_t cnt, loff_t *ppos)
7271 * There is no need to read what the user has written, this function
7272 * is just to make sure that there is no error when "echo" is used
7281 tracing_free_buffer_release(struct inode *inode, struct file *filp)
7283 struct trace_array *tr = inode->i_private;
7285 /* disable tracing ? */
7286 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7287 tracer_tracing_off(tr);
7288 /* resize the ring buffer to 0 */
7289 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7291 trace_array_put(tr);
7296 #define TRACE_MARKER_MAX_SIZE 4096
7299 tracing_mark_write(struct file *filp, const char __user *ubuf,
7300 size_t cnt, loff_t *fpos)
7302 struct trace_array *tr = filp->private_data;
7303 struct ring_buffer_event *event;
7304 enum event_trigger_type tt = ETT_NONE;
7305 struct trace_buffer *buffer;
7306 struct print_entry *entry;
7312 /* Used in tracing_mark_raw_write() as well */
7313 #define FAULTED_STR "<faulted>"
7314 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7316 if (tracing_disabled)
7319 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7322 if ((ssize_t)cnt < 0)
7325 if (cnt > TRACE_MARKER_MAX_SIZE)
7326 cnt = TRACE_MARKER_MAX_SIZE;
7328 meta_size = sizeof(*entry) + 2; /* add '\0' and possible '\n' */
7330 size = cnt + meta_size;
7332 /* If less than "<faulted>", then make sure we can still add that */
7333 if (cnt < FAULTED_SIZE)
7334 size += FAULTED_SIZE - cnt;
7336 buffer = tr->array_buffer.buffer;
7337 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7339 if (unlikely(!event)) {
7341 * If the size was greater than what was allowed, then
7342 * make it smaller and try again.
7344 if (size > ring_buffer_max_event_size(buffer)) {
7345 /* cnt < FAULTED size should never be bigger than max */
7346 if (WARN_ON_ONCE(cnt < FAULTED_SIZE))
7348 cnt = ring_buffer_max_event_size(buffer) - meta_size;
7349 /* The above should only happen once */
7350 if (WARN_ON_ONCE(cnt + meta_size == size))
7355 /* Ring buffer disabled, return as if not open for write */
7359 entry = ring_buffer_event_data(event);
7360 entry->ip = _THIS_IP_;
7362 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7364 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7370 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7371 /* do not add \n before testing triggers, but add \0 */
7372 entry->buf[cnt] = '\0';
7373 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7376 if (entry->buf[cnt - 1] != '\n') {
7377 entry->buf[cnt] = '\n';
7378 entry->buf[cnt + 1] = '\0';
7380 entry->buf[cnt] = '\0';
7382 if (static_branch_unlikely(&trace_marker_exports_enabled))
7383 ftrace_exports(event, TRACE_EXPORT_MARKER);
7384 __buffer_unlock_commit(buffer, event);
7387 event_triggers_post_call(tr->trace_marker_file, tt);
7393 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7394 size_t cnt, loff_t *fpos)
7396 struct trace_array *tr = filp->private_data;
7397 struct ring_buffer_event *event;
7398 struct trace_buffer *buffer;
7399 struct raw_data_entry *entry;
7404 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7406 if (tracing_disabled)
7409 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7412 /* The marker must at least have a tag id */
7413 if (cnt < sizeof(unsigned int))
7416 size = sizeof(*entry) + cnt;
7417 if (cnt < FAULT_SIZE_ID)
7418 size += FAULT_SIZE_ID - cnt;
7420 buffer = tr->array_buffer.buffer;
7422 if (size > ring_buffer_max_event_size(buffer))
7425 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7428 /* Ring buffer disabled, return as if not open for write */
7431 entry = ring_buffer_event_data(event);
7433 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7436 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7441 __buffer_unlock_commit(buffer, event);
7446 static int tracing_clock_show(struct seq_file *m, void *v)
7448 struct trace_array *tr = m->private;
7451 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7453 "%s%s%s%s", i ? " " : "",
7454 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7455 i == tr->clock_id ? "]" : "");
7461 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7465 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7466 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7469 if (i == ARRAY_SIZE(trace_clocks))
7472 mutex_lock(&trace_types_lock);
7476 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7479 * New clock may not be consistent with the previous clock.
7480 * Reset the buffer so that it doesn't have incomparable timestamps.
7482 tracing_reset_online_cpus(&tr->array_buffer);
7484 #ifdef CONFIG_TRACER_MAX_TRACE
7485 if (tr->max_buffer.buffer)
7486 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7487 tracing_reset_online_cpus(&tr->max_buffer);
7490 mutex_unlock(&trace_types_lock);
7495 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7496 size_t cnt, loff_t *fpos)
7498 struct seq_file *m = filp->private_data;
7499 struct trace_array *tr = m->private;
7501 const char *clockstr;
7504 if (cnt >= sizeof(buf))
7507 if (copy_from_user(buf, ubuf, cnt))
7512 clockstr = strstrip(buf);
7514 ret = tracing_set_clock(tr, clockstr);
7523 static int tracing_clock_open(struct inode *inode, struct file *file)
7525 struct trace_array *tr = inode->i_private;
7528 ret = tracing_check_open_get_tr(tr);
7532 ret = single_open(file, tracing_clock_show, inode->i_private);
7534 trace_array_put(tr);
7539 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7541 struct trace_array *tr = m->private;
7543 mutex_lock(&trace_types_lock);
7545 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7546 seq_puts(m, "delta [absolute]\n");
7548 seq_puts(m, "[delta] absolute\n");
7550 mutex_unlock(&trace_types_lock);
7555 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7557 struct trace_array *tr = inode->i_private;
7560 ret = tracing_check_open_get_tr(tr);
7564 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7566 trace_array_put(tr);
7571 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7573 if (rbe == this_cpu_read(trace_buffered_event))
7574 return ring_buffer_time_stamp(buffer);
7576 return ring_buffer_event_time_stamp(buffer, rbe);
7580 * Set or disable using the per CPU trace_buffer_event when possible.
7582 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7586 mutex_lock(&trace_types_lock);
7588 if (set && tr->no_filter_buffering_ref++)
7592 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7597 --tr->no_filter_buffering_ref;
7600 mutex_unlock(&trace_types_lock);
7605 struct ftrace_buffer_info {
7606 struct trace_iterator iter;
7608 unsigned int spare_cpu;
7609 unsigned int spare_size;
7613 #ifdef CONFIG_TRACER_SNAPSHOT
7614 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7616 struct trace_array *tr = inode->i_private;
7617 struct trace_iterator *iter;
7621 ret = tracing_check_open_get_tr(tr);
7625 if (file->f_mode & FMODE_READ) {
7626 iter = __tracing_open(inode, file, true);
7628 ret = PTR_ERR(iter);
7630 /* Writes still need the seq_file to hold the private data */
7632 m = kzalloc(sizeof(*m), GFP_KERNEL);
7635 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7643 iter->array_buffer = &tr->max_buffer;
7644 iter->cpu_file = tracing_get_cpu(inode);
7646 file->private_data = m;
7650 trace_array_put(tr);
7655 static void tracing_swap_cpu_buffer(void *tr)
7657 update_max_tr_single((struct trace_array *)tr, current, smp_processor_id());
7661 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7664 struct seq_file *m = filp->private_data;
7665 struct trace_iterator *iter = m->private;
7666 struct trace_array *tr = iter->tr;
7670 ret = tracing_update_buffers(tr);
7674 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7678 mutex_lock(&trace_types_lock);
7680 if (tr->current_trace->use_max_tr) {
7685 local_irq_disable();
7686 arch_spin_lock(&tr->max_lock);
7687 if (tr->cond_snapshot)
7689 arch_spin_unlock(&tr->max_lock);
7696 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7700 if (tr->allocated_snapshot)
7704 /* Only allow per-cpu swap if the ring buffer supports it */
7705 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7706 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7711 if (tr->allocated_snapshot)
7712 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7713 &tr->array_buffer, iter->cpu_file);
7715 ret = tracing_alloc_snapshot_instance(tr);
7718 /* Now, we're going to swap */
7719 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
7720 local_irq_disable();
7721 update_max_tr(tr, current, smp_processor_id(), NULL);
7724 smp_call_function_single(iter->cpu_file, tracing_swap_cpu_buffer,
7729 if (tr->allocated_snapshot) {
7730 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7731 tracing_reset_online_cpus(&tr->max_buffer);
7733 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7743 mutex_unlock(&trace_types_lock);
7747 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7749 struct seq_file *m = file->private_data;
7752 ret = tracing_release(inode, file);
7754 if (file->f_mode & FMODE_READ)
7757 /* If write only, the seq_file is just a stub */
7765 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7766 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7767 size_t count, loff_t *ppos);
7768 static int tracing_buffers_release(struct inode *inode, struct file *file);
7769 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7770 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7772 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7774 struct ftrace_buffer_info *info;
7777 /* The following checks for tracefs lockdown */
7778 ret = tracing_buffers_open(inode, filp);
7782 info = filp->private_data;
7784 if (info->iter.trace->use_max_tr) {
7785 tracing_buffers_release(inode, filp);
7789 info->iter.snapshot = true;
7790 info->iter.array_buffer = &info->iter.tr->max_buffer;
7795 #endif /* CONFIG_TRACER_SNAPSHOT */
7798 static const struct file_operations tracing_thresh_fops = {
7799 .open = tracing_open_generic,
7800 .read = tracing_thresh_read,
7801 .write = tracing_thresh_write,
7802 .llseek = generic_file_llseek,
7805 #ifdef CONFIG_TRACER_MAX_TRACE
7806 static const struct file_operations tracing_max_lat_fops = {
7807 .open = tracing_open_generic_tr,
7808 .read = tracing_max_lat_read,
7809 .write = tracing_max_lat_write,
7810 .llseek = generic_file_llseek,
7811 .release = tracing_release_generic_tr,
7815 static const struct file_operations set_tracer_fops = {
7816 .open = tracing_open_generic_tr,
7817 .read = tracing_set_trace_read,
7818 .write = tracing_set_trace_write,
7819 .llseek = generic_file_llseek,
7820 .release = tracing_release_generic_tr,
7823 static const struct file_operations tracing_pipe_fops = {
7824 .open = tracing_open_pipe,
7825 .poll = tracing_poll_pipe,
7826 .read = tracing_read_pipe,
7827 .splice_read = tracing_splice_read_pipe,
7828 .release = tracing_release_pipe,
7829 .llseek = no_llseek,
7832 static const struct file_operations tracing_entries_fops = {
7833 .open = tracing_open_generic_tr,
7834 .read = tracing_entries_read,
7835 .write = tracing_entries_write,
7836 .llseek = generic_file_llseek,
7837 .release = tracing_release_generic_tr,
7840 static const struct file_operations tracing_total_entries_fops = {
7841 .open = tracing_open_generic_tr,
7842 .read = tracing_total_entries_read,
7843 .llseek = generic_file_llseek,
7844 .release = tracing_release_generic_tr,
7847 static const struct file_operations tracing_free_buffer_fops = {
7848 .open = tracing_open_generic_tr,
7849 .write = tracing_free_buffer_write,
7850 .release = tracing_free_buffer_release,
7853 static const struct file_operations tracing_mark_fops = {
7854 .open = tracing_mark_open,
7855 .write = tracing_mark_write,
7856 .release = tracing_release_generic_tr,
7859 static const struct file_operations tracing_mark_raw_fops = {
7860 .open = tracing_mark_open,
7861 .write = tracing_mark_raw_write,
7862 .release = tracing_release_generic_tr,
7865 static const struct file_operations trace_clock_fops = {
7866 .open = tracing_clock_open,
7868 .llseek = seq_lseek,
7869 .release = tracing_single_release_tr,
7870 .write = tracing_clock_write,
7873 static const struct file_operations trace_time_stamp_mode_fops = {
7874 .open = tracing_time_stamp_mode_open,
7876 .llseek = seq_lseek,
7877 .release = tracing_single_release_tr,
7880 #ifdef CONFIG_TRACER_SNAPSHOT
7881 static const struct file_operations snapshot_fops = {
7882 .open = tracing_snapshot_open,
7884 .write = tracing_snapshot_write,
7885 .llseek = tracing_lseek,
7886 .release = tracing_snapshot_release,
7889 static const struct file_operations snapshot_raw_fops = {
7890 .open = snapshot_raw_open,
7891 .read = tracing_buffers_read,
7892 .release = tracing_buffers_release,
7893 .splice_read = tracing_buffers_splice_read,
7894 .llseek = no_llseek,
7897 #endif /* CONFIG_TRACER_SNAPSHOT */
7900 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7901 * @filp: The active open file structure
7902 * @ubuf: The userspace provided buffer to read value into
7903 * @cnt: The maximum number of bytes to read
7904 * @ppos: The current "file" position
7906 * This function implements the write interface for a struct trace_min_max_param.
7907 * The filp->private_data must point to a trace_min_max_param structure that
7908 * defines where to write the value, the min and the max acceptable values,
7909 * and a lock to protect the write.
7912 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7914 struct trace_min_max_param *param = filp->private_data;
7921 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7926 mutex_lock(param->lock);
7928 if (param->min && val < *param->min)
7931 if (param->max && val > *param->max)
7938 mutex_unlock(param->lock);
7947 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7948 * @filp: The active open file structure
7949 * @ubuf: The userspace provided buffer to read value into
7950 * @cnt: The maximum number of bytes to read
7951 * @ppos: The current "file" position
7953 * This function implements the read interface for a struct trace_min_max_param.
7954 * The filp->private_data must point to a trace_min_max_param struct with valid
7958 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7960 struct trace_min_max_param *param = filp->private_data;
7961 char buf[U64_STR_SIZE];
7970 if (cnt > sizeof(buf))
7973 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7975 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7978 const struct file_operations trace_min_max_fops = {
7979 .open = tracing_open_generic,
7980 .read = trace_min_max_read,
7981 .write = trace_min_max_write,
7984 #define TRACING_LOG_ERRS_MAX 8
7985 #define TRACING_LOG_LOC_MAX 128
7987 #define CMD_PREFIX " Command: "
7990 const char **errs; /* ptr to loc-specific array of err strings */
7991 u8 type; /* index into errs -> specific err string */
7992 u16 pos; /* caret position */
7996 struct tracing_log_err {
7997 struct list_head list;
7998 struct err_info info;
7999 char loc[TRACING_LOG_LOC_MAX]; /* err location */
8000 char *cmd; /* what caused err */
8003 static DEFINE_MUTEX(tracing_err_log_lock);
8005 static struct tracing_log_err *alloc_tracing_log_err(int len)
8007 struct tracing_log_err *err;
8009 err = kzalloc(sizeof(*err), GFP_KERNEL);
8011 return ERR_PTR(-ENOMEM);
8013 err->cmd = kzalloc(len, GFP_KERNEL);
8016 return ERR_PTR(-ENOMEM);
8022 static void free_tracing_log_err(struct tracing_log_err *err)
8028 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
8031 struct tracing_log_err *err;
8034 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
8035 err = alloc_tracing_log_err(len);
8036 if (PTR_ERR(err) != -ENOMEM)
8037 tr->n_err_log_entries++;
8041 cmd = kzalloc(len, GFP_KERNEL);
8043 return ERR_PTR(-ENOMEM);
8044 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
8047 list_del(&err->list);
8053 * err_pos - find the position of a string within a command for error careting
8054 * @cmd: The tracing command that caused the error
8055 * @str: The string to position the caret at within @cmd
8057 * Finds the position of the first occurrence of @str within @cmd. The
8058 * return value can be passed to tracing_log_err() for caret placement
8061 * Returns the index within @cmd of the first occurrence of @str or 0
8062 * if @str was not found.
8064 unsigned int err_pos(char *cmd, const char *str)
8068 if (WARN_ON(!strlen(cmd)))
8071 found = strstr(cmd, str);
8079 * tracing_log_err - write an error to the tracing error log
8080 * @tr: The associated trace array for the error (NULL for top level array)
8081 * @loc: A string describing where the error occurred
8082 * @cmd: The tracing command that caused the error
8083 * @errs: The array of loc-specific static error strings
8084 * @type: The index into errs[], which produces the specific static err string
8085 * @pos: The position the caret should be placed in the cmd
8087 * Writes an error into tracing/error_log of the form:
8089 * <loc>: error: <text>
8093 * tracing/error_log is a small log file containing the last
8094 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
8095 * unless there has been a tracing error, and the error log can be
8096 * cleared and have its memory freed by writing the empty string in
8097 * truncation mode to it i.e. echo > tracing/error_log.
8099 * NOTE: the @errs array along with the @type param are used to
8100 * produce a static error string - this string is not copied and saved
8101 * when the error is logged - only a pointer to it is saved. See
8102 * existing callers for examples of how static strings are typically
8103 * defined for use with tracing_log_err().
8105 void tracing_log_err(struct trace_array *tr,
8106 const char *loc, const char *cmd,
8107 const char **errs, u8 type, u16 pos)
8109 struct tracing_log_err *err;
8115 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
8117 mutex_lock(&tracing_err_log_lock);
8118 err = get_tracing_log_err(tr, len);
8119 if (PTR_ERR(err) == -ENOMEM) {
8120 mutex_unlock(&tracing_err_log_lock);
8124 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
8125 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
8127 err->info.errs = errs;
8128 err->info.type = type;
8129 err->info.pos = pos;
8130 err->info.ts = local_clock();
8132 list_add_tail(&err->list, &tr->err_log);
8133 mutex_unlock(&tracing_err_log_lock);
8136 static void clear_tracing_err_log(struct trace_array *tr)
8138 struct tracing_log_err *err, *next;
8140 mutex_lock(&tracing_err_log_lock);
8141 list_for_each_entry_safe(err, next, &tr->err_log, list) {
8142 list_del(&err->list);
8143 free_tracing_log_err(err);
8146 tr->n_err_log_entries = 0;
8147 mutex_unlock(&tracing_err_log_lock);
8150 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
8152 struct trace_array *tr = m->private;
8154 mutex_lock(&tracing_err_log_lock);
8156 return seq_list_start(&tr->err_log, *pos);
8159 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
8161 struct trace_array *tr = m->private;
8163 return seq_list_next(v, &tr->err_log, pos);
8166 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
8168 mutex_unlock(&tracing_err_log_lock);
8171 static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
8175 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
8177 for (i = 0; i < pos; i++)
8182 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
8184 struct tracing_log_err *err = v;
8187 const char *err_text = err->info.errs[err->info.type];
8188 u64 sec = err->info.ts;
8191 nsec = do_div(sec, NSEC_PER_SEC);
8192 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
8193 err->loc, err_text);
8194 seq_printf(m, "%s", err->cmd);
8195 tracing_err_log_show_pos(m, err->info.pos);
8201 static const struct seq_operations tracing_err_log_seq_ops = {
8202 .start = tracing_err_log_seq_start,
8203 .next = tracing_err_log_seq_next,
8204 .stop = tracing_err_log_seq_stop,
8205 .show = tracing_err_log_seq_show
8208 static int tracing_err_log_open(struct inode *inode, struct file *file)
8210 struct trace_array *tr = inode->i_private;
8213 ret = tracing_check_open_get_tr(tr);
8217 /* If this file was opened for write, then erase contents */
8218 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8219 clear_tracing_err_log(tr);
8221 if (file->f_mode & FMODE_READ) {
8222 ret = seq_open(file, &tracing_err_log_seq_ops);
8224 struct seq_file *m = file->private_data;
8227 trace_array_put(tr);
8233 static ssize_t tracing_err_log_write(struct file *file,
8234 const char __user *buffer,
8235 size_t count, loff_t *ppos)
8240 static int tracing_err_log_release(struct inode *inode, struct file *file)
8242 struct trace_array *tr = inode->i_private;
8244 trace_array_put(tr);
8246 if (file->f_mode & FMODE_READ)
8247 seq_release(inode, file);
8252 static const struct file_operations tracing_err_log_fops = {
8253 .open = tracing_err_log_open,
8254 .write = tracing_err_log_write,
8256 .llseek = tracing_lseek,
8257 .release = tracing_err_log_release,
8260 static int tracing_buffers_open(struct inode *inode, struct file *filp)
8262 struct trace_array *tr = inode->i_private;
8263 struct ftrace_buffer_info *info;
8266 ret = tracing_check_open_get_tr(tr);
8270 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8272 trace_array_put(tr);
8276 mutex_lock(&trace_types_lock);
8279 info->iter.cpu_file = tracing_get_cpu(inode);
8280 info->iter.trace = tr->current_trace;
8281 info->iter.array_buffer = &tr->array_buffer;
8283 /* Force reading ring buffer for first read */
8284 info->read = (unsigned int)-1;
8286 filp->private_data = info;
8290 mutex_unlock(&trace_types_lock);
8292 ret = nonseekable_open(inode, filp);
8294 trace_array_put(tr);
8300 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8302 struct ftrace_buffer_info *info = filp->private_data;
8303 struct trace_iterator *iter = &info->iter;
8305 return trace_poll(iter, filp, poll_table);
8309 tracing_buffers_read(struct file *filp, char __user *ubuf,
8310 size_t count, loff_t *ppos)
8312 struct ftrace_buffer_info *info = filp->private_data;
8313 struct trace_iterator *iter = &info->iter;
8322 #ifdef CONFIG_TRACER_MAX_TRACE
8323 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8327 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8329 /* Make sure the spare matches the current sub buffer size */
8331 if (page_size != info->spare_size) {
8332 ring_buffer_free_read_page(iter->array_buffer->buffer,
8333 info->spare_cpu, info->spare);
8339 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8341 if (IS_ERR(info->spare)) {
8342 ret = PTR_ERR(info->spare);
8345 info->spare_cpu = iter->cpu_file;
8346 info->spare_size = page_size;
8352 /* Do we have previous read data to read? */
8353 if (info->read < page_size)
8357 trace_access_lock(iter->cpu_file);
8358 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8362 trace_access_unlock(iter->cpu_file);
8365 if (trace_empty(iter)) {
8366 if ((filp->f_flags & O_NONBLOCK))
8369 ret = wait_on_pipe(iter, 0);
8380 size = page_size - info->read;
8383 trace_data = ring_buffer_read_page_data(info->spare);
8384 ret = copy_to_user(ubuf, trace_data + info->read, size);
8396 static int tracing_buffers_flush(struct file *file, fl_owner_t id)
8398 struct ftrace_buffer_info *info = file->private_data;
8399 struct trace_iterator *iter = &info->iter;
8402 /* Make sure the waiters see the new wait_index */
8405 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8410 static int tracing_buffers_release(struct inode *inode, struct file *file)
8412 struct ftrace_buffer_info *info = file->private_data;
8413 struct trace_iterator *iter = &info->iter;
8415 mutex_lock(&trace_types_lock);
8417 iter->tr->trace_ref--;
8419 __trace_array_put(iter->tr);
8422 ring_buffer_free_read_page(iter->array_buffer->buffer,
8423 info->spare_cpu, info->spare);
8426 mutex_unlock(&trace_types_lock);
8432 struct trace_buffer *buffer;
8435 refcount_t refcount;
8438 static void buffer_ref_release(struct buffer_ref *ref)
8440 if (!refcount_dec_and_test(&ref->refcount))
8442 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8446 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8447 struct pipe_buffer *buf)
8449 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8451 buffer_ref_release(ref);
8455 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8456 struct pipe_buffer *buf)
8458 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8460 if (refcount_read(&ref->refcount) > INT_MAX/2)
8463 refcount_inc(&ref->refcount);
8467 /* Pipe buffer operations for a buffer. */
8468 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8469 .release = buffer_pipe_buf_release,
8470 .get = buffer_pipe_buf_get,
8474 * Callback from splice_to_pipe(), if we need to release some pages
8475 * at the end of the spd in case we error'ed out in filling the pipe.
8477 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8479 struct buffer_ref *ref =
8480 (struct buffer_ref *)spd->partial[i].private;
8482 buffer_ref_release(ref);
8483 spd->partial[i].private = 0;
8487 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8488 struct pipe_inode_info *pipe, size_t len,
8491 struct ftrace_buffer_info *info = file->private_data;
8492 struct trace_iterator *iter = &info->iter;
8493 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8494 struct page *pages_def[PIPE_DEF_BUFFERS];
8495 struct splice_pipe_desc spd = {
8497 .partial = partial_def,
8498 .nr_pages_max = PIPE_DEF_BUFFERS,
8499 .ops = &buffer_pipe_buf_ops,
8500 .spd_release = buffer_spd_release,
8502 struct buffer_ref *ref;
8507 #ifdef CONFIG_TRACER_MAX_TRACE
8508 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8512 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8513 if (*ppos & (page_size - 1))
8516 if (len & (page_size - 1)) {
8517 if (len < page_size)
8519 len &= (~(page_size - 1));
8522 if (splice_grow_spd(pipe, &spd))
8526 trace_access_lock(iter->cpu_file);
8527 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8529 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= page_size) {
8533 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8539 refcount_set(&ref->refcount, 1);
8540 ref->buffer = iter->array_buffer->buffer;
8541 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8542 if (IS_ERR(ref->page)) {
8543 ret = PTR_ERR(ref->page);
8548 ref->cpu = iter->cpu_file;
8550 r = ring_buffer_read_page(ref->buffer, ref->page,
8551 len, iter->cpu_file, 1);
8553 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8559 page = virt_to_page(ring_buffer_read_page_data(ref->page));
8561 spd.pages[i] = page;
8562 spd.partial[i].len = page_size;
8563 spd.partial[i].offset = 0;
8564 spd.partial[i].private = (unsigned long)ref;
8568 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8571 trace_access_unlock(iter->cpu_file);
8574 /* did we read anything? */
8575 if (!spd.nr_pages) {
8582 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8585 wait_index = READ_ONCE(iter->wait_index);
8587 ret = wait_on_pipe(iter, iter->snapshot ? 0 : iter->tr->buffer_percent);
8591 /* No need to wait after waking up when tracing is off */
8592 if (!tracer_tracing_is_on(iter->tr))
8595 /* Make sure we see the new wait_index */
8597 if (wait_index != iter->wait_index)
8603 ret = splice_to_pipe(pipe, &spd);
8605 splice_shrink_spd(&spd);
8610 /* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8611 static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8613 struct ftrace_buffer_info *info = file->private_data;
8614 struct trace_iterator *iter = &info->iter;
8617 return -ENOIOCTLCMD;
8619 mutex_lock(&trace_types_lock);
8622 /* Make sure the waiters see the new wait_index */
8625 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8627 mutex_unlock(&trace_types_lock);
8631 static const struct file_operations tracing_buffers_fops = {
8632 .open = tracing_buffers_open,
8633 .read = tracing_buffers_read,
8634 .poll = tracing_buffers_poll,
8635 .release = tracing_buffers_release,
8636 .flush = tracing_buffers_flush,
8637 .splice_read = tracing_buffers_splice_read,
8638 .unlocked_ioctl = tracing_buffers_ioctl,
8639 .llseek = no_llseek,
8643 tracing_stats_read(struct file *filp, char __user *ubuf,
8644 size_t count, loff_t *ppos)
8646 struct inode *inode = file_inode(filp);
8647 struct trace_array *tr = inode->i_private;
8648 struct array_buffer *trace_buf = &tr->array_buffer;
8649 int cpu = tracing_get_cpu(inode);
8650 struct trace_seq *s;
8652 unsigned long long t;
8653 unsigned long usec_rem;
8655 s = kmalloc(sizeof(*s), GFP_KERNEL);
8661 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8662 trace_seq_printf(s, "entries: %ld\n", cnt);
8664 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8665 trace_seq_printf(s, "overrun: %ld\n", cnt);
8667 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8668 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8670 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8671 trace_seq_printf(s, "bytes: %ld\n", cnt);
8673 if (trace_clocks[tr->clock_id].in_ns) {
8674 /* local or global for trace_clock */
8675 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8676 usec_rem = do_div(t, USEC_PER_SEC);
8677 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8680 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8681 usec_rem = do_div(t, USEC_PER_SEC);
8682 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8684 /* counter or tsc mode for trace_clock */
8685 trace_seq_printf(s, "oldest event ts: %llu\n",
8686 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8688 trace_seq_printf(s, "now ts: %llu\n",
8689 ring_buffer_time_stamp(trace_buf->buffer));
8692 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8693 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8695 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8696 trace_seq_printf(s, "read events: %ld\n", cnt);
8698 count = simple_read_from_buffer(ubuf, count, ppos,
8699 s->buffer, trace_seq_used(s));
8706 static const struct file_operations tracing_stats_fops = {
8707 .open = tracing_open_generic_tr,
8708 .read = tracing_stats_read,
8709 .llseek = generic_file_llseek,
8710 .release = tracing_release_generic_tr,
8713 #ifdef CONFIG_DYNAMIC_FTRACE
8716 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8717 size_t cnt, loff_t *ppos)
8723 /* 256 should be plenty to hold the amount needed */
8724 buf = kmalloc(256, GFP_KERNEL);
8728 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8729 ftrace_update_tot_cnt,
8730 ftrace_number_of_pages,
8731 ftrace_number_of_groups);
8733 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8738 static const struct file_operations tracing_dyn_info_fops = {
8739 .open = tracing_open_generic,
8740 .read = tracing_read_dyn_info,
8741 .llseek = generic_file_llseek,
8743 #endif /* CONFIG_DYNAMIC_FTRACE */
8745 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8747 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8748 struct trace_array *tr, struct ftrace_probe_ops *ops,
8751 tracing_snapshot_instance(tr);
8755 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8756 struct trace_array *tr, struct ftrace_probe_ops *ops,
8759 struct ftrace_func_mapper *mapper = data;
8763 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8773 tracing_snapshot_instance(tr);
8777 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8778 struct ftrace_probe_ops *ops, void *data)
8780 struct ftrace_func_mapper *mapper = data;
8783 seq_printf(m, "%ps:", (void *)ip);
8785 seq_puts(m, "snapshot");
8788 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8791 seq_printf(m, ":count=%ld\n", *count);
8793 seq_puts(m, ":unlimited\n");
8799 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8800 unsigned long ip, void *init_data, void **data)
8802 struct ftrace_func_mapper *mapper = *data;
8805 mapper = allocate_ftrace_func_mapper();
8811 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8815 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8816 unsigned long ip, void *data)
8818 struct ftrace_func_mapper *mapper = data;
8823 free_ftrace_func_mapper(mapper, NULL);
8827 ftrace_func_mapper_remove_ip(mapper, ip);
8830 static struct ftrace_probe_ops snapshot_probe_ops = {
8831 .func = ftrace_snapshot,
8832 .print = ftrace_snapshot_print,
8835 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8836 .func = ftrace_count_snapshot,
8837 .print = ftrace_snapshot_print,
8838 .init = ftrace_snapshot_init,
8839 .free = ftrace_snapshot_free,
8843 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8844 char *glob, char *cmd, char *param, int enable)
8846 struct ftrace_probe_ops *ops;
8847 void *count = (void *)-1;
8854 /* hash funcs only work with set_ftrace_filter */
8858 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8861 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8866 number = strsep(¶m, ":");
8868 if (!strlen(number))
8872 * We use the callback data field (which is a pointer)
8875 ret = kstrtoul(number, 0, (unsigned long *)&count);
8880 ret = tracing_alloc_snapshot_instance(tr);
8884 ret = register_ftrace_function_probe(glob, tr, ops, count);
8887 return ret < 0 ? ret : 0;
8890 static struct ftrace_func_command ftrace_snapshot_cmd = {
8892 .func = ftrace_trace_snapshot_callback,
8895 static __init int register_snapshot_cmd(void)
8897 return register_ftrace_command(&ftrace_snapshot_cmd);
8900 static inline __init int register_snapshot_cmd(void) { return 0; }
8901 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8903 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8905 if (WARN_ON(!tr->dir))
8906 return ERR_PTR(-ENODEV);
8908 /* Top directory uses NULL as the parent */
8909 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8912 /* All sub buffers have a descriptor */
8916 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8918 struct dentry *d_tracer;
8921 return tr->percpu_dir;
8923 d_tracer = tracing_get_dentry(tr);
8924 if (IS_ERR(d_tracer))
8927 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8929 MEM_FAIL(!tr->percpu_dir,
8930 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8932 return tr->percpu_dir;
8935 static struct dentry *
8936 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8937 void *data, long cpu, const struct file_operations *fops)
8939 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8941 if (ret) /* See tracing_get_cpu() */
8942 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8947 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8949 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8950 struct dentry *d_cpu;
8951 char cpu_dir[30]; /* 30 characters should be more than enough */
8956 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8957 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8959 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8963 /* per cpu trace_pipe */
8964 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8965 tr, cpu, &tracing_pipe_fops);
8968 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8969 tr, cpu, &tracing_fops);
8971 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8972 tr, cpu, &tracing_buffers_fops);
8974 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8975 tr, cpu, &tracing_stats_fops);
8977 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8978 tr, cpu, &tracing_entries_fops);
8980 #ifdef CONFIG_TRACER_SNAPSHOT
8981 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8982 tr, cpu, &snapshot_fops);
8984 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8985 tr, cpu, &snapshot_raw_fops);
8989 #ifdef CONFIG_FTRACE_SELFTEST
8990 /* Let selftest have access to static functions in this file */
8991 #include "trace_selftest.c"
8995 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8998 struct trace_option_dentry *topt = filp->private_data;
9001 if (topt->flags->val & topt->opt->bit)
9006 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9010 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
9013 struct trace_option_dentry *topt = filp->private_data;
9017 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9021 if (val != 0 && val != 1)
9024 if (!!(topt->flags->val & topt->opt->bit) != val) {
9025 mutex_lock(&trace_types_lock);
9026 ret = __set_tracer_option(topt->tr, topt->flags,
9028 mutex_unlock(&trace_types_lock);
9038 static int tracing_open_options(struct inode *inode, struct file *filp)
9040 struct trace_option_dentry *topt = inode->i_private;
9043 ret = tracing_check_open_get_tr(topt->tr);
9047 filp->private_data = inode->i_private;
9051 static int tracing_release_options(struct inode *inode, struct file *file)
9053 struct trace_option_dentry *topt = file->private_data;
9055 trace_array_put(topt->tr);
9059 static const struct file_operations trace_options_fops = {
9060 .open = tracing_open_options,
9061 .read = trace_options_read,
9062 .write = trace_options_write,
9063 .llseek = generic_file_llseek,
9064 .release = tracing_release_options,
9068 * In order to pass in both the trace_array descriptor as well as the index
9069 * to the flag that the trace option file represents, the trace_array
9070 * has a character array of trace_flags_index[], which holds the index
9071 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
9072 * The address of this character array is passed to the flag option file
9073 * read/write callbacks.
9075 * In order to extract both the index and the trace_array descriptor,
9076 * get_tr_index() uses the following algorithm.
9080 * As the pointer itself contains the address of the index (remember
9083 * Then to get the trace_array descriptor, by subtracting that index
9084 * from the ptr, we get to the start of the index itself.
9086 * ptr - idx == &index[0]
9088 * Then a simple container_of() from that pointer gets us to the
9089 * trace_array descriptor.
9091 static void get_tr_index(void *data, struct trace_array **ptr,
9092 unsigned int *pindex)
9094 *pindex = *(unsigned char *)data;
9096 *ptr = container_of(data - *pindex, struct trace_array,
9101 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
9104 void *tr_index = filp->private_data;
9105 struct trace_array *tr;
9109 get_tr_index(tr_index, &tr, &index);
9111 if (tr->trace_flags & (1 << index))
9116 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9120 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
9123 void *tr_index = filp->private_data;
9124 struct trace_array *tr;
9129 get_tr_index(tr_index, &tr, &index);
9131 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9135 if (val != 0 && val != 1)
9138 mutex_lock(&event_mutex);
9139 mutex_lock(&trace_types_lock);
9140 ret = set_tracer_flag(tr, 1 << index, val);
9141 mutex_unlock(&trace_types_lock);
9142 mutex_unlock(&event_mutex);
9152 static const struct file_operations trace_options_core_fops = {
9153 .open = tracing_open_generic,
9154 .read = trace_options_core_read,
9155 .write = trace_options_core_write,
9156 .llseek = generic_file_llseek,
9159 struct dentry *trace_create_file(const char *name,
9161 struct dentry *parent,
9163 const struct file_operations *fops)
9167 ret = tracefs_create_file(name, mode, parent, data, fops);
9169 pr_warn("Could not create tracefs '%s' entry\n", name);
9175 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
9177 struct dentry *d_tracer;
9182 d_tracer = tracing_get_dentry(tr);
9183 if (IS_ERR(d_tracer))
9186 tr->options = tracefs_create_dir("options", d_tracer);
9188 pr_warn("Could not create tracefs directory 'options'\n");
9196 create_trace_option_file(struct trace_array *tr,
9197 struct trace_option_dentry *topt,
9198 struct tracer_flags *flags,
9199 struct tracer_opt *opt)
9201 struct dentry *t_options;
9203 t_options = trace_options_init_dentry(tr);
9207 topt->flags = flags;
9211 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
9212 t_options, topt, &trace_options_fops);
9217 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
9219 struct trace_option_dentry *topts;
9220 struct trace_options *tr_topts;
9221 struct tracer_flags *flags;
9222 struct tracer_opt *opts;
9229 flags = tracer->flags;
9231 if (!flags || !flags->opts)
9235 * If this is an instance, only create flags for tracers
9236 * the instance may have.
9238 if (!trace_ok_for_array(tracer, tr))
9241 for (i = 0; i < tr->nr_topts; i++) {
9242 /* Make sure there's no duplicate flags. */
9243 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
9249 for (cnt = 0; opts[cnt].name; cnt++)
9252 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
9256 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9263 tr->topts = tr_topts;
9264 tr->topts[tr->nr_topts].tracer = tracer;
9265 tr->topts[tr->nr_topts].topts = topts;
9268 for (cnt = 0; opts[cnt].name; cnt++) {
9269 create_trace_option_file(tr, &topts[cnt], flags,
9271 MEM_FAIL(topts[cnt].entry == NULL,
9272 "Failed to create trace option: %s",
9277 static struct dentry *
9278 create_trace_option_core_file(struct trace_array *tr,
9279 const char *option, long index)
9281 struct dentry *t_options;
9283 t_options = trace_options_init_dentry(tr);
9287 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9288 (void *)&tr->trace_flags_index[index],
9289 &trace_options_core_fops);
9292 static void create_trace_options_dir(struct trace_array *tr)
9294 struct dentry *t_options;
9295 bool top_level = tr == &global_trace;
9298 t_options = trace_options_init_dentry(tr);
9302 for (i = 0; trace_options[i]; i++) {
9304 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9305 create_trace_option_core_file(tr, trace_options[i], i);
9310 rb_simple_read(struct file *filp, char __user *ubuf,
9311 size_t cnt, loff_t *ppos)
9313 struct trace_array *tr = filp->private_data;
9317 r = tracer_tracing_is_on(tr);
9318 r = sprintf(buf, "%d\n", r);
9320 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9324 rb_simple_write(struct file *filp, const char __user *ubuf,
9325 size_t cnt, loff_t *ppos)
9327 struct trace_array *tr = filp->private_data;
9328 struct trace_buffer *buffer = tr->array_buffer.buffer;
9332 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9337 mutex_lock(&trace_types_lock);
9338 if (!!val == tracer_tracing_is_on(tr)) {
9339 val = 0; /* do nothing */
9341 tracer_tracing_on(tr);
9342 if (tr->current_trace->start)
9343 tr->current_trace->start(tr);
9345 tracer_tracing_off(tr);
9346 if (tr->current_trace->stop)
9347 tr->current_trace->stop(tr);
9348 /* Wake up any waiters */
9349 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9351 mutex_unlock(&trace_types_lock);
9359 static const struct file_operations rb_simple_fops = {
9360 .open = tracing_open_generic_tr,
9361 .read = rb_simple_read,
9362 .write = rb_simple_write,
9363 .release = tracing_release_generic_tr,
9364 .llseek = default_llseek,
9368 buffer_percent_read(struct file *filp, char __user *ubuf,
9369 size_t cnt, loff_t *ppos)
9371 struct trace_array *tr = filp->private_data;
9375 r = tr->buffer_percent;
9376 r = sprintf(buf, "%d\n", r);
9378 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9382 buffer_percent_write(struct file *filp, const char __user *ubuf,
9383 size_t cnt, loff_t *ppos)
9385 struct trace_array *tr = filp->private_data;
9389 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9396 tr->buffer_percent = val;
9403 static const struct file_operations buffer_percent_fops = {
9404 .open = tracing_open_generic_tr,
9405 .read = buffer_percent_read,
9406 .write = buffer_percent_write,
9407 .release = tracing_release_generic_tr,
9408 .llseek = default_llseek,
9412 buffer_subbuf_size_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
9414 struct trace_array *tr = filp->private_data;
9420 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9421 size = (PAGE_SIZE << order) / 1024;
9423 r = sprintf(buf, "%zd\n", size);
9425 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9429 buffer_subbuf_size_write(struct file *filp, const char __user *ubuf,
9430 size_t cnt, loff_t *ppos)
9432 struct trace_array *tr = filp->private_data;
9439 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9443 val *= 1024; /* value passed in is in KB */
9445 pages = DIV_ROUND_UP(val, PAGE_SIZE);
9446 order = fls(pages - 1);
9448 /* limit between 1 and 128 system pages */
9449 if (order < 0 || order > 7)
9452 /* Do not allow tracing while changing the order of the ring buffer */
9453 tracing_stop_tr(tr);
9455 old_order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9456 if (old_order == order)
9459 ret = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, order);
9463 #ifdef CONFIG_TRACER_MAX_TRACE
9465 if (!tr->allocated_snapshot)
9468 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
9470 /* Put back the old order */
9471 cnt = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, old_order);
9472 if (WARN_ON_ONCE(cnt)) {
9474 * AARGH! We are left with different orders!
9475 * The max buffer is our "snapshot" buffer.
9476 * When a tracer needs a snapshot (one of the
9477 * latency tracers), it swaps the max buffer
9478 * with the saved snap shot. We succeeded to
9479 * update the order of the main buffer, but failed to
9480 * update the order of the max buffer. But when we tried
9481 * to reset the main buffer to the original size, we
9482 * failed there too. This is very unlikely to
9483 * happen, but if it does, warn and kill all
9486 tracing_disabled = 1;
9496 tracing_start_tr(tr);
9500 static const struct file_operations buffer_subbuf_size_fops = {
9501 .open = tracing_open_generic_tr,
9502 .read = buffer_subbuf_size_read,
9503 .write = buffer_subbuf_size_write,
9504 .release = tracing_release_generic_tr,
9505 .llseek = default_llseek,
9508 static struct dentry *trace_instance_dir;
9511 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9514 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9516 enum ring_buffer_flags rb_flags;
9518 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9522 buf->buffer = ring_buffer_alloc(size, rb_flags);
9526 buf->data = alloc_percpu(struct trace_array_cpu);
9528 ring_buffer_free(buf->buffer);
9533 /* Allocate the first page for all buffers */
9534 set_buffer_entries(&tr->array_buffer,
9535 ring_buffer_size(tr->array_buffer.buffer, 0));
9540 static void free_trace_buffer(struct array_buffer *buf)
9543 ring_buffer_free(buf->buffer);
9545 free_percpu(buf->data);
9550 static int allocate_trace_buffers(struct trace_array *tr, int size)
9554 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9558 #ifdef CONFIG_TRACER_MAX_TRACE
9559 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9560 allocate_snapshot ? size : 1);
9561 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9562 free_trace_buffer(&tr->array_buffer);
9565 tr->allocated_snapshot = allocate_snapshot;
9567 allocate_snapshot = false;
9573 static void free_trace_buffers(struct trace_array *tr)
9578 free_trace_buffer(&tr->array_buffer);
9580 #ifdef CONFIG_TRACER_MAX_TRACE
9581 free_trace_buffer(&tr->max_buffer);
9585 static void init_trace_flags_index(struct trace_array *tr)
9589 /* Used by the trace options files */
9590 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9591 tr->trace_flags_index[i] = i;
9594 static void __update_tracer_options(struct trace_array *tr)
9598 for (t = trace_types; t; t = t->next)
9599 add_tracer_options(tr, t);
9602 static void update_tracer_options(struct trace_array *tr)
9604 mutex_lock(&trace_types_lock);
9605 tracer_options_updated = true;
9606 __update_tracer_options(tr);
9607 mutex_unlock(&trace_types_lock);
9610 /* Must have trace_types_lock held */
9611 struct trace_array *trace_array_find(const char *instance)
9613 struct trace_array *tr, *found = NULL;
9615 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9616 if (tr->name && strcmp(tr->name, instance) == 0) {
9625 struct trace_array *trace_array_find_get(const char *instance)
9627 struct trace_array *tr;
9629 mutex_lock(&trace_types_lock);
9630 tr = trace_array_find(instance);
9633 mutex_unlock(&trace_types_lock);
9638 static int trace_array_create_dir(struct trace_array *tr)
9642 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9646 ret = event_trace_add_tracer(tr->dir, tr);
9648 tracefs_remove(tr->dir);
9652 init_tracer_tracefs(tr, tr->dir);
9653 __update_tracer_options(tr);
9658 static struct trace_array *
9659 trace_array_create_systems(const char *name, const char *systems)
9661 struct trace_array *tr;
9665 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9667 return ERR_PTR(ret);
9669 tr->name = kstrdup(name, GFP_KERNEL);
9673 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9676 if (!zalloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9680 tr->system_names = kstrdup_const(systems, GFP_KERNEL);
9681 if (!tr->system_names)
9685 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9687 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9689 raw_spin_lock_init(&tr->start_lock);
9691 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9693 tr->current_trace = &nop_trace;
9695 INIT_LIST_HEAD(&tr->systems);
9696 INIT_LIST_HEAD(&tr->events);
9697 INIT_LIST_HEAD(&tr->hist_vars);
9698 INIT_LIST_HEAD(&tr->err_log);
9700 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9703 /* The ring buffer is defaultly expanded */
9704 trace_set_ring_buffer_expanded(tr);
9706 if (ftrace_allocate_ftrace_ops(tr) < 0)
9709 ftrace_init_trace_array(tr);
9711 init_trace_flags_index(tr);
9713 if (trace_instance_dir) {
9714 ret = trace_array_create_dir(tr);
9718 __trace_early_add_events(tr);
9720 list_add(&tr->list, &ftrace_trace_arrays);
9727 ftrace_free_ftrace_ops(tr);
9728 free_trace_buffers(tr);
9729 free_cpumask_var(tr->pipe_cpumask);
9730 free_cpumask_var(tr->tracing_cpumask);
9731 kfree_const(tr->system_names);
9735 return ERR_PTR(ret);
9738 static struct trace_array *trace_array_create(const char *name)
9740 return trace_array_create_systems(name, NULL);
9743 static int instance_mkdir(const char *name)
9745 struct trace_array *tr;
9748 mutex_lock(&event_mutex);
9749 mutex_lock(&trace_types_lock);
9752 if (trace_array_find(name))
9755 tr = trace_array_create(name);
9757 ret = PTR_ERR_OR_ZERO(tr);
9760 mutex_unlock(&trace_types_lock);
9761 mutex_unlock(&event_mutex);
9766 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9767 * @name: The name of the trace array to be looked up/created.
9768 * @systems: A list of systems to create event directories for (NULL for all)
9770 * Returns pointer to trace array with given name.
9771 * NULL, if it cannot be created.
9773 * NOTE: This function increments the reference counter associated with the
9774 * trace array returned. This makes sure it cannot be freed while in use.
9775 * Use trace_array_put() once the trace array is no longer needed.
9776 * If the trace_array is to be freed, trace_array_destroy() needs to
9777 * be called after the trace_array_put(), or simply let user space delete
9778 * it from the tracefs instances directory. But until the
9779 * trace_array_put() is called, user space can not delete it.
9782 struct trace_array *trace_array_get_by_name(const char *name, const char *systems)
9784 struct trace_array *tr;
9786 mutex_lock(&event_mutex);
9787 mutex_lock(&trace_types_lock);
9789 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9790 if (tr->name && strcmp(tr->name, name) == 0)
9794 tr = trace_array_create_systems(name, systems);
9802 mutex_unlock(&trace_types_lock);
9803 mutex_unlock(&event_mutex);
9806 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9808 static int __remove_instance(struct trace_array *tr)
9812 /* Reference counter for a newly created trace array = 1. */
9813 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9816 list_del(&tr->list);
9818 /* Disable all the flags that were enabled coming in */
9819 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9820 if ((1 << i) & ZEROED_TRACE_FLAGS)
9821 set_tracer_flag(tr, 1 << i, 0);
9824 tracing_set_nop(tr);
9825 clear_ftrace_function_probes(tr);
9826 event_trace_del_tracer(tr);
9827 ftrace_clear_pids(tr);
9828 ftrace_destroy_function_files(tr);
9829 tracefs_remove(tr->dir);
9830 free_percpu(tr->last_func_repeats);
9831 free_trace_buffers(tr);
9832 clear_tracing_err_log(tr);
9834 for (i = 0; i < tr->nr_topts; i++) {
9835 kfree(tr->topts[i].topts);
9839 free_cpumask_var(tr->pipe_cpumask);
9840 free_cpumask_var(tr->tracing_cpumask);
9841 kfree_const(tr->system_names);
9848 int trace_array_destroy(struct trace_array *this_tr)
9850 struct trace_array *tr;
9856 mutex_lock(&event_mutex);
9857 mutex_lock(&trace_types_lock);
9861 /* Making sure trace array exists before destroying it. */
9862 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9863 if (tr == this_tr) {
9864 ret = __remove_instance(tr);
9869 mutex_unlock(&trace_types_lock);
9870 mutex_unlock(&event_mutex);
9874 EXPORT_SYMBOL_GPL(trace_array_destroy);
9876 static int instance_rmdir(const char *name)
9878 struct trace_array *tr;
9881 mutex_lock(&event_mutex);
9882 mutex_lock(&trace_types_lock);
9885 tr = trace_array_find(name);
9887 ret = __remove_instance(tr);
9889 mutex_unlock(&trace_types_lock);
9890 mutex_unlock(&event_mutex);
9895 static __init void create_trace_instances(struct dentry *d_tracer)
9897 struct trace_array *tr;
9899 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9902 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9905 mutex_lock(&event_mutex);
9906 mutex_lock(&trace_types_lock);
9908 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9911 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9912 "Failed to create instance directory\n"))
9916 mutex_unlock(&trace_types_lock);
9917 mutex_unlock(&event_mutex);
9921 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9925 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9926 tr, &show_traces_fops);
9928 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9929 tr, &set_tracer_fops);
9931 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9932 tr, &tracing_cpumask_fops);
9934 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9935 tr, &tracing_iter_fops);
9937 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9940 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9941 tr, &tracing_pipe_fops);
9943 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9944 tr, &tracing_entries_fops);
9946 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9947 tr, &tracing_total_entries_fops);
9949 trace_create_file("free_buffer", 0200, d_tracer,
9950 tr, &tracing_free_buffer_fops);
9952 trace_create_file("trace_marker", 0220, d_tracer,
9953 tr, &tracing_mark_fops);
9955 tr->trace_marker_file = __find_event_file(tr, "ftrace", "print");
9957 trace_create_file("trace_marker_raw", 0220, d_tracer,
9958 tr, &tracing_mark_raw_fops);
9960 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9963 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9964 tr, &rb_simple_fops);
9966 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9967 &trace_time_stamp_mode_fops);
9969 tr->buffer_percent = 50;
9971 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9972 tr, &buffer_percent_fops);
9974 trace_create_file("buffer_subbuf_size_kb", TRACE_MODE_WRITE, d_tracer,
9975 tr, &buffer_subbuf_size_fops);
9977 create_trace_options_dir(tr);
9979 #ifdef CONFIG_TRACER_MAX_TRACE
9980 trace_create_maxlat_file(tr, d_tracer);
9983 if (ftrace_create_function_files(tr, d_tracer))
9984 MEM_FAIL(1, "Could not allocate function filter files");
9986 #ifdef CONFIG_TRACER_SNAPSHOT
9987 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9988 tr, &snapshot_fops);
9991 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9992 tr, &tracing_err_log_fops);
9994 for_each_tracing_cpu(cpu)
9995 tracing_init_tracefs_percpu(tr, cpu);
9997 ftrace_init_tracefs(tr, d_tracer);
10000 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
10002 struct vfsmount *mnt;
10003 struct file_system_type *type;
10006 * To maintain backward compatibility for tools that mount
10007 * debugfs to get to the tracing facility, tracefs is automatically
10008 * mounted to the debugfs/tracing directory.
10010 type = get_fs_type("tracefs");
10013 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
10014 put_filesystem(type);
10023 * tracing_init_dentry - initialize top level trace array
10025 * This is called when creating files or directories in the tracing
10026 * directory. It is called via fs_initcall() by any of the boot up code
10027 * and expects to return the dentry of the top level tracing directory.
10029 int tracing_init_dentry(void)
10031 struct trace_array *tr = &global_trace;
10033 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10034 pr_warn("Tracing disabled due to lockdown\n");
10038 /* The top level trace array uses NULL as parent */
10042 if (WARN_ON(!tracefs_initialized()))
10046 * As there may still be users that expect the tracing
10047 * files to exist in debugfs/tracing, we must automount
10048 * the tracefs file system there, so older tools still
10049 * work with the newer kernel.
10051 tr->dir = debugfs_create_automount("tracing", NULL,
10052 trace_automount, NULL);
10057 extern struct trace_eval_map *__start_ftrace_eval_maps[];
10058 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
10060 static struct workqueue_struct *eval_map_wq __initdata;
10061 static struct work_struct eval_map_work __initdata;
10062 static struct work_struct tracerfs_init_work __initdata;
10064 static void __init eval_map_work_func(struct work_struct *work)
10068 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
10069 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
10072 static int __init trace_eval_init(void)
10074 INIT_WORK(&eval_map_work, eval_map_work_func);
10076 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
10077 if (!eval_map_wq) {
10078 pr_err("Unable to allocate eval_map_wq\n");
10080 eval_map_work_func(&eval_map_work);
10084 queue_work(eval_map_wq, &eval_map_work);
10088 subsys_initcall(trace_eval_init);
10090 static int __init trace_eval_sync(void)
10092 /* Make sure the eval map updates are finished */
10094 destroy_workqueue(eval_map_wq);
10098 late_initcall_sync(trace_eval_sync);
10101 #ifdef CONFIG_MODULES
10102 static void trace_module_add_evals(struct module *mod)
10104 if (!mod->num_trace_evals)
10108 * Modules with bad taint do not have events created, do
10109 * not bother with enums either.
10111 if (trace_module_has_bad_taint(mod))
10114 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
10117 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
10118 static void trace_module_remove_evals(struct module *mod)
10120 union trace_eval_map_item *map;
10121 union trace_eval_map_item **last = &trace_eval_maps;
10123 if (!mod->num_trace_evals)
10126 mutex_lock(&trace_eval_mutex);
10128 map = trace_eval_maps;
10131 if (map->head.mod == mod)
10133 map = trace_eval_jmp_to_tail(map);
10134 last = &map->tail.next;
10135 map = map->tail.next;
10140 *last = trace_eval_jmp_to_tail(map)->tail.next;
10143 mutex_unlock(&trace_eval_mutex);
10146 static inline void trace_module_remove_evals(struct module *mod) { }
10147 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
10149 static int trace_module_notify(struct notifier_block *self,
10150 unsigned long val, void *data)
10152 struct module *mod = data;
10155 case MODULE_STATE_COMING:
10156 trace_module_add_evals(mod);
10158 case MODULE_STATE_GOING:
10159 trace_module_remove_evals(mod);
10166 static struct notifier_block trace_module_nb = {
10167 .notifier_call = trace_module_notify,
10170 #endif /* CONFIG_MODULES */
10172 static __init void tracer_init_tracefs_work_func(struct work_struct *work)
10175 event_trace_init();
10177 init_tracer_tracefs(&global_trace, NULL);
10178 ftrace_init_tracefs_toplevel(&global_trace, NULL);
10180 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
10181 &global_trace, &tracing_thresh_fops);
10183 trace_create_file("README", TRACE_MODE_READ, NULL,
10184 NULL, &tracing_readme_fops);
10186 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
10187 NULL, &tracing_saved_cmdlines_fops);
10189 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
10190 NULL, &tracing_saved_cmdlines_size_fops);
10192 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
10193 NULL, &tracing_saved_tgids_fops);
10195 trace_create_eval_file(NULL);
10197 #ifdef CONFIG_MODULES
10198 register_module_notifier(&trace_module_nb);
10201 #ifdef CONFIG_DYNAMIC_FTRACE
10202 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
10203 NULL, &tracing_dyn_info_fops);
10206 create_trace_instances(NULL);
10208 update_tracer_options(&global_trace);
10211 static __init int tracer_init_tracefs(void)
10215 trace_access_lock_init();
10217 ret = tracing_init_dentry();
10222 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
10223 queue_work(eval_map_wq, &tracerfs_init_work);
10225 tracer_init_tracefs_work_func(NULL);
10228 rv_init_interface();
10233 fs_initcall(tracer_init_tracefs);
10235 static int trace_die_panic_handler(struct notifier_block *self,
10236 unsigned long ev, void *unused);
10238 static struct notifier_block trace_panic_notifier = {
10239 .notifier_call = trace_die_panic_handler,
10240 .priority = INT_MAX - 1,
10243 static struct notifier_block trace_die_notifier = {
10244 .notifier_call = trace_die_panic_handler,
10245 .priority = INT_MAX - 1,
10249 * The idea is to execute the following die/panic callback early, in order
10250 * to avoid showing irrelevant information in the trace (like other panic
10251 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
10252 * warnings get disabled (to prevent potential log flooding).
10254 static int trace_die_panic_handler(struct notifier_block *self,
10255 unsigned long ev, void *unused)
10257 if (!ftrace_dump_on_oops)
10258 return NOTIFY_DONE;
10260 /* The die notifier requires DIE_OOPS to trigger */
10261 if (self == &trace_die_notifier && ev != DIE_OOPS)
10262 return NOTIFY_DONE;
10264 ftrace_dump(ftrace_dump_on_oops);
10266 return NOTIFY_DONE;
10270 * printk is set to max of 1024, we really don't need it that big.
10271 * Nothing should be printing 1000 characters anyway.
10273 #define TRACE_MAX_PRINT 1000
10276 * Define here KERN_TRACE so that we have one place to modify
10277 * it if we decide to change what log level the ftrace dump
10280 #define KERN_TRACE KERN_EMERG
10283 trace_printk_seq(struct trace_seq *s)
10285 /* Probably should print a warning here. */
10286 if (s->seq.len >= TRACE_MAX_PRINT)
10287 s->seq.len = TRACE_MAX_PRINT;
10290 * More paranoid code. Although the buffer size is set to
10291 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
10292 * an extra layer of protection.
10294 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
10295 s->seq.len = s->seq.size - 1;
10297 /* should be zero ended, but we are paranoid. */
10298 s->buffer[s->seq.len] = 0;
10300 printk(KERN_TRACE "%s", s->buffer);
10305 void trace_init_global_iter(struct trace_iterator *iter)
10307 iter->tr = &global_trace;
10308 iter->trace = iter->tr->current_trace;
10309 iter->cpu_file = RING_BUFFER_ALL_CPUS;
10310 iter->array_buffer = &global_trace.array_buffer;
10312 if (iter->trace && iter->trace->open)
10313 iter->trace->open(iter);
10315 /* Annotate start of buffers if we had overruns */
10316 if (ring_buffer_overruns(iter->array_buffer->buffer))
10317 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10319 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10320 if (trace_clocks[iter->tr->clock_id].in_ns)
10321 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10323 /* Can not use kmalloc for iter.temp and iter.fmt */
10324 iter->temp = static_temp_buf;
10325 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10326 iter->fmt = static_fmt_buf;
10327 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10330 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10332 /* use static because iter can be a bit big for the stack */
10333 static struct trace_iterator iter;
10334 static atomic_t dump_running;
10335 struct trace_array *tr = &global_trace;
10336 unsigned int old_userobj;
10337 unsigned long flags;
10340 /* Only allow one dump user at a time. */
10341 if (atomic_inc_return(&dump_running) != 1) {
10342 atomic_dec(&dump_running);
10347 * Always turn off tracing when we dump.
10348 * We don't need to show trace output of what happens
10349 * between multiple crashes.
10351 * If the user does a sysrq-z, then they can re-enable
10352 * tracing with echo 1 > tracing_on.
10356 local_irq_save(flags);
10358 /* Simulate the iterator */
10359 trace_init_global_iter(&iter);
10361 for_each_tracing_cpu(cpu) {
10362 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10365 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10367 /* don't look at user memory in panic mode */
10368 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10370 switch (oops_dump_mode) {
10372 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10375 iter.cpu_file = raw_smp_processor_id();
10380 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10381 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10384 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10386 /* Did function tracer already get disabled? */
10387 if (ftrace_is_dead()) {
10388 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10389 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10393 * We need to stop all tracing on all CPUS to read
10394 * the next buffer. This is a bit expensive, but is
10395 * not done often. We fill all what we can read,
10396 * and then release the locks again.
10399 while (!trace_empty(&iter)) {
10402 printk(KERN_TRACE "---------------------------------\n");
10406 trace_iterator_reset(&iter);
10407 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10409 if (trace_find_next_entry_inc(&iter) != NULL) {
10412 ret = print_trace_line(&iter);
10413 if (ret != TRACE_TYPE_NO_CONSUME)
10414 trace_consume(&iter);
10416 touch_nmi_watchdog();
10418 trace_printk_seq(&iter.seq);
10422 printk(KERN_TRACE " (ftrace buffer empty)\n");
10424 printk(KERN_TRACE "---------------------------------\n");
10427 tr->trace_flags |= old_userobj;
10429 for_each_tracing_cpu(cpu) {
10430 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10432 atomic_dec(&dump_running);
10433 local_irq_restore(flags);
10435 EXPORT_SYMBOL_GPL(ftrace_dump);
10437 #define WRITE_BUFSIZE 4096
10439 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10440 size_t count, loff_t *ppos,
10441 int (*createfn)(const char *))
10443 char *kbuf, *buf, *tmp;
10448 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10452 while (done < count) {
10453 size = count - done;
10455 if (size >= WRITE_BUFSIZE)
10456 size = WRITE_BUFSIZE - 1;
10458 if (copy_from_user(kbuf, buffer + done, size)) {
10465 tmp = strchr(buf, '\n');
10468 size = tmp - buf + 1;
10470 size = strlen(buf);
10471 if (done + size < count) {
10474 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10475 pr_warn("Line length is too long: Should be less than %d\n",
10476 WRITE_BUFSIZE - 2);
10483 /* Remove comments */
10484 tmp = strchr(buf, '#');
10489 ret = createfn(buf);
10494 } while (done < count);
10504 #ifdef CONFIG_TRACER_MAX_TRACE
10505 __init static bool tr_needs_alloc_snapshot(const char *name)
10508 int len = strlen(name);
10511 if (!boot_snapshot_index)
10514 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10515 boot_snapshot_info[len] == '\t')
10518 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10522 sprintf(test, "\t%s\t", name);
10523 ret = strstr(boot_snapshot_info, test) == NULL;
10528 __init static void do_allocate_snapshot(const char *name)
10530 if (!tr_needs_alloc_snapshot(name))
10534 * When allocate_snapshot is set, the next call to
10535 * allocate_trace_buffers() (called by trace_array_get_by_name())
10536 * will allocate the snapshot buffer. That will alse clear
10539 allocate_snapshot = true;
10542 static inline void do_allocate_snapshot(const char *name) { }
10545 __init static void enable_instances(void)
10547 struct trace_array *tr;
10552 /* A tab is always appended */
10553 boot_instance_info[boot_instance_index - 1] = '\0';
10554 str = boot_instance_info;
10556 while ((curr_str = strsep(&str, "\t"))) {
10558 tok = strsep(&curr_str, ",");
10560 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10561 do_allocate_snapshot(tok);
10563 tr = trace_array_get_by_name(tok, NULL);
10565 pr_warn("Failed to create instance buffer %s\n", curr_str);
10568 /* Allow user space to delete it */
10569 trace_array_put(tr);
10571 while ((tok = strsep(&curr_str, ","))) {
10572 early_enable_events(tr, tok, true);
10577 __init static int tracer_alloc_buffers(void)
10583 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10584 pr_warn("Tracing disabled due to lockdown\n");
10589 * Make sure we don't accidentally add more trace options
10590 * than we have bits for.
10592 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10594 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10597 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10598 goto out_free_buffer_mask;
10600 /* Only allocate trace_printk buffers if a trace_printk exists */
10601 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10602 /* Must be called before global_trace.buffer is allocated */
10603 trace_printk_init_buffers();
10605 /* To save memory, keep the ring buffer size to its minimum */
10606 if (global_trace.ring_buffer_expanded)
10607 ring_buf_size = trace_buf_size;
10611 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10612 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10614 raw_spin_lock_init(&global_trace.start_lock);
10617 * The prepare callbacks allocates some memory for the ring buffer. We
10618 * don't free the buffer if the CPU goes down. If we were to free
10619 * the buffer, then the user would lose any trace that was in the
10620 * buffer. The memory will be removed once the "instance" is removed.
10622 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10623 "trace/RB:prepare", trace_rb_cpu_prepare,
10626 goto out_free_cpumask;
10627 /* Used for event triggers */
10629 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10631 goto out_rm_hp_state;
10633 if (trace_create_savedcmd() < 0)
10634 goto out_free_temp_buffer;
10636 if (!zalloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10637 goto out_free_savedcmd;
10639 /* TODO: make the number of buffers hot pluggable with CPUS */
10640 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10641 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10642 goto out_free_pipe_cpumask;
10644 if (global_trace.buffer_disabled)
10647 if (trace_boot_clock) {
10648 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10650 pr_warn("Trace clock %s not defined, going back to default\n",
10655 * register_tracer() might reference current_trace, so it
10656 * needs to be set before we register anything. This is
10657 * just a bootstrap of current_trace anyway.
10659 global_trace.current_trace = &nop_trace;
10661 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10663 ftrace_init_global_array_ops(&global_trace);
10665 init_trace_flags_index(&global_trace);
10667 register_tracer(&nop_trace);
10669 /* Function tracing may start here (via kernel command line) */
10670 init_function_trace();
10672 /* All seems OK, enable tracing */
10673 tracing_disabled = 0;
10675 atomic_notifier_chain_register(&panic_notifier_list,
10676 &trace_panic_notifier);
10678 register_die_notifier(&trace_die_notifier);
10680 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10682 INIT_LIST_HEAD(&global_trace.systems);
10683 INIT_LIST_HEAD(&global_trace.events);
10684 INIT_LIST_HEAD(&global_trace.hist_vars);
10685 INIT_LIST_HEAD(&global_trace.err_log);
10686 list_add(&global_trace.list, &ftrace_trace_arrays);
10688 apply_trace_boot_options();
10690 register_snapshot_cmd();
10696 out_free_pipe_cpumask:
10697 free_cpumask_var(global_trace.pipe_cpumask);
10699 free_saved_cmdlines_buffer(savedcmd);
10700 out_free_temp_buffer:
10701 ring_buffer_free(temp_buffer);
10703 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10705 free_cpumask_var(global_trace.tracing_cpumask);
10706 out_free_buffer_mask:
10707 free_cpumask_var(tracing_buffer_mask);
10712 void __init ftrace_boot_snapshot(void)
10714 #ifdef CONFIG_TRACER_MAX_TRACE
10715 struct trace_array *tr;
10717 if (!snapshot_at_boot)
10720 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10721 if (!tr->allocated_snapshot)
10724 tracing_snapshot_instance(tr);
10725 trace_array_puts(tr, "** Boot snapshot taken **\n");
10730 void __init early_trace_init(void)
10732 if (tracepoint_printk) {
10733 tracepoint_print_iter =
10734 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10735 if (MEM_FAIL(!tracepoint_print_iter,
10736 "Failed to allocate trace iterator\n"))
10737 tracepoint_printk = 0;
10739 static_key_enable(&tracepoint_printk_key.key);
10741 tracer_alloc_buffers();
10746 void __init trace_init(void)
10748 trace_event_init();
10750 if (boot_instance_index)
10751 enable_instances();
10754 __init static void clear_boot_tracer(void)
10757 * The default tracer at boot buffer is an init section.
10758 * This function is called in lateinit. If we did not
10759 * find the boot tracer, then clear it out, to prevent
10760 * later registration from accessing the buffer that is
10761 * about to be freed.
10763 if (!default_bootup_tracer)
10766 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10767 default_bootup_tracer);
10768 default_bootup_tracer = NULL;
10771 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10772 __init static void tracing_set_default_clock(void)
10774 /* sched_clock_stable() is determined in late_initcall */
10775 if (!trace_boot_clock && !sched_clock_stable()) {
10776 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10777 pr_warn("Can not set tracing clock due to lockdown\n");
10781 printk(KERN_WARNING
10782 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10783 "If you want to keep using the local clock, then add:\n"
10784 " \"trace_clock=local\"\n"
10785 "on the kernel command line\n");
10786 tracing_set_clock(&global_trace, "global");
10790 static inline void tracing_set_default_clock(void) { }
10793 __init static int late_trace_init(void)
10795 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10796 static_key_disable(&tracepoint_printk_key.key);
10797 tracepoint_printk = 0;
10800 tracing_set_default_clock();
10801 clear_boot_tracer();
10805 late_initcall_sync(late_trace_init);