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/notifier.h>
23 #include <linux/irqflags.h>
24 #include <linux/debugfs.h>
25 #include <linux/tracefs.h>
26 #include <linux/pagemap.h>
27 #include <linux/hardirq.h>
28 #include <linux/linkage.h>
29 #include <linux/uaccess.h>
30 #include <linux/vmalloc.h>
31 #include <linux/ftrace.h>
32 #include <linux/module.h>
33 #include <linux/percpu.h>
34 #include <linux/splice.h>
35 #include <linux/kdebug.h>
36 #include <linux/string.h>
37 #include <linux/mount.h>
38 #include <linux/rwsem.h>
39 #include <linux/slab.h>
40 #include <linux/ctype.h>
41 #include <linux/init.h>
42 #include <linux/panic_notifier.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>
54 #include "trace_output.h"
57 * On boot up, the ring buffer is set to the minimum size, so that
58 * we do not waste memory on systems that are not using tracing.
60 bool ring_buffer_expanded;
63 * We need to change this state when a selftest is running.
64 * A selftest will lurk into the ring-buffer to count the
65 * entries inserted during the selftest although some concurrent
66 * insertions into the ring-buffer such as trace_printk could occurred
67 * at the same time, giving false positive or negative results.
69 static bool __read_mostly tracing_selftest_running;
72 * If boot-time tracing including tracers/events via kernel cmdline
73 * is running, we do not want to run SELFTEST.
75 bool __read_mostly tracing_selftest_disabled;
77 #ifdef CONFIG_FTRACE_STARTUP_TEST
78 void __init disable_tracing_selftest(const char *reason)
80 if (!tracing_selftest_disabled) {
81 tracing_selftest_disabled = true;
82 pr_info("Ftrace startup test is disabled due to %s\n", reason);
87 /* Pipe tracepoints to printk */
88 struct trace_iterator *tracepoint_print_iter;
89 int tracepoint_printk;
90 static bool tracepoint_printk_stop_on_boot __initdata;
91 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
93 /* For tracers that don't implement custom flags */
94 static struct tracer_opt dummy_tracer_opt[] = {
99 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
105 * To prevent the comm cache from being overwritten when no
106 * tracing is active, only save the comm when a trace event
109 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
112 * Kill all tracing for good (never come back).
113 * It is initialized to 1 but will turn to zero if the initialization
114 * of the tracer is successful. But that is the only place that sets
117 static int tracing_disabled = 1;
119 cpumask_var_t __read_mostly tracing_buffer_mask;
122 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
124 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
125 * is set, then ftrace_dump is called. This will output the contents
126 * of the ftrace buffers to the console. This is very useful for
127 * capturing traces that lead to crashes and outputing it to a
130 * It is default off, but you can enable it with either specifying
131 * "ftrace_dump_on_oops" in the kernel command line, or setting
132 * /proc/sys/kernel/ftrace_dump_on_oops
133 * Set 1 if you want to dump buffers of all CPUs
134 * Set 2 if you want to dump the buffer of the CPU that triggered oops
137 enum ftrace_dump_mode ftrace_dump_on_oops;
139 /* When set, tracing will stop when a WARN*() is hit */
140 int __disable_trace_on_warning;
142 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
143 /* Map of enums to their values, for "eval_map" file */
144 struct trace_eval_map_head {
146 unsigned long length;
149 union trace_eval_map_item;
151 struct trace_eval_map_tail {
153 * "end" is first and points to NULL as it must be different
154 * than "mod" or "eval_string"
156 union trace_eval_map_item *next;
157 const char *end; /* points to NULL */
160 static DEFINE_MUTEX(trace_eval_mutex);
163 * The trace_eval_maps are saved in an array with two extra elements,
164 * one at the beginning, and one at the end. The beginning item contains
165 * the count of the saved maps (head.length), and the module they
166 * belong to if not built in (head.mod). The ending item contains a
167 * pointer to the next array of saved eval_map items.
169 union trace_eval_map_item {
170 struct trace_eval_map map;
171 struct trace_eval_map_head head;
172 struct trace_eval_map_tail tail;
175 static union trace_eval_map_item *trace_eval_maps;
176 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
178 int tracing_set_tracer(struct trace_array *tr, const char *buf);
179 static void ftrace_trace_userstack(struct trace_array *tr,
180 struct trace_buffer *buffer,
181 unsigned int trace_ctx);
183 #define MAX_TRACER_SIZE 100
184 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
185 static char *default_bootup_tracer;
187 static bool allocate_snapshot;
189 static int __init set_cmdline_ftrace(char *str)
191 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
192 default_bootup_tracer = bootup_tracer_buf;
193 /* We are using ftrace early, expand it */
194 ring_buffer_expanded = true;
197 __setup("ftrace=", set_cmdline_ftrace);
199 static int __init set_ftrace_dump_on_oops(char *str)
201 if (*str++ != '=' || !*str || !strcmp("1", str)) {
202 ftrace_dump_on_oops = DUMP_ALL;
206 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
207 ftrace_dump_on_oops = DUMP_ORIG;
213 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
215 static int __init stop_trace_on_warning(char *str)
217 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
218 __disable_trace_on_warning = 1;
221 __setup("traceoff_on_warning", stop_trace_on_warning);
223 static int __init boot_alloc_snapshot(char *str)
225 allocate_snapshot = true;
226 /* We also need the main ring buffer expanded */
227 ring_buffer_expanded = true;
230 __setup("alloc_snapshot", boot_alloc_snapshot);
233 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
235 static int __init set_trace_boot_options(char *str)
237 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
240 __setup("trace_options=", set_trace_boot_options);
242 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
243 static char *trace_boot_clock __initdata;
245 static int __init set_trace_boot_clock(char *str)
247 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
248 trace_boot_clock = trace_boot_clock_buf;
251 __setup("trace_clock=", set_trace_boot_clock);
253 static int __init set_tracepoint_printk(char *str)
255 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
256 tracepoint_printk = 1;
259 __setup("tp_printk", set_tracepoint_printk);
261 static int __init set_tracepoint_printk_stop(char *str)
263 tracepoint_printk_stop_on_boot = true;
266 __setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
268 unsigned long long ns2usecs(u64 nsec)
276 trace_process_export(struct trace_export *export,
277 struct ring_buffer_event *event, int flag)
279 struct trace_entry *entry;
280 unsigned int size = 0;
282 if (export->flags & flag) {
283 entry = ring_buffer_event_data(event);
284 size = ring_buffer_event_length(event);
285 export->write(export, entry, size);
289 static DEFINE_MUTEX(ftrace_export_lock);
291 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
293 static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
294 static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
295 static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
297 static inline void ftrace_exports_enable(struct trace_export *export)
299 if (export->flags & TRACE_EXPORT_FUNCTION)
300 static_branch_inc(&trace_function_exports_enabled);
302 if (export->flags & TRACE_EXPORT_EVENT)
303 static_branch_inc(&trace_event_exports_enabled);
305 if (export->flags & TRACE_EXPORT_MARKER)
306 static_branch_inc(&trace_marker_exports_enabled);
309 static inline void ftrace_exports_disable(struct trace_export *export)
311 if (export->flags & TRACE_EXPORT_FUNCTION)
312 static_branch_dec(&trace_function_exports_enabled);
314 if (export->flags & TRACE_EXPORT_EVENT)
315 static_branch_dec(&trace_event_exports_enabled);
317 if (export->flags & TRACE_EXPORT_MARKER)
318 static_branch_dec(&trace_marker_exports_enabled);
321 static void ftrace_exports(struct ring_buffer_event *event, int flag)
323 struct trace_export *export;
325 preempt_disable_notrace();
327 export = rcu_dereference_raw_check(ftrace_exports_list);
329 trace_process_export(export, event, flag);
330 export = rcu_dereference_raw_check(export->next);
333 preempt_enable_notrace();
337 add_trace_export(struct trace_export **list, struct trace_export *export)
339 rcu_assign_pointer(export->next, *list);
341 * We are entering export into the list but another
342 * CPU might be walking that list. We need to make sure
343 * the export->next pointer is valid before another CPU sees
344 * the export pointer included into the list.
346 rcu_assign_pointer(*list, export);
350 rm_trace_export(struct trace_export **list, struct trace_export *export)
352 struct trace_export **p;
354 for (p = list; *p != NULL; p = &(*p)->next)
361 rcu_assign_pointer(*p, (*p)->next);
367 add_ftrace_export(struct trace_export **list, struct trace_export *export)
369 ftrace_exports_enable(export);
371 add_trace_export(list, export);
375 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
379 ret = rm_trace_export(list, export);
380 ftrace_exports_disable(export);
385 int register_ftrace_export(struct trace_export *export)
387 if (WARN_ON_ONCE(!export->write))
390 mutex_lock(&ftrace_export_lock);
392 add_ftrace_export(&ftrace_exports_list, export);
394 mutex_unlock(&ftrace_export_lock);
398 EXPORT_SYMBOL_GPL(register_ftrace_export);
400 int unregister_ftrace_export(struct trace_export *export)
404 mutex_lock(&ftrace_export_lock);
406 ret = rm_ftrace_export(&ftrace_exports_list, export);
408 mutex_unlock(&ftrace_export_lock);
412 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
414 /* trace_flags holds trace_options default values */
415 #define TRACE_DEFAULT_FLAGS \
416 (FUNCTION_DEFAULT_FLAGS | \
417 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
418 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
419 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
420 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
423 /* trace_options that are only supported by global_trace */
424 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
425 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
427 /* trace_flags that are default zero for instances */
428 #define ZEROED_TRACE_FLAGS \
429 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
432 * The global_trace is the descriptor that holds the top-level tracing
433 * buffers for the live tracing.
435 static struct trace_array global_trace = {
436 .trace_flags = TRACE_DEFAULT_FLAGS,
439 LIST_HEAD(ftrace_trace_arrays);
441 int trace_array_get(struct trace_array *this_tr)
443 struct trace_array *tr;
446 mutex_lock(&trace_types_lock);
447 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
454 mutex_unlock(&trace_types_lock);
459 static void __trace_array_put(struct trace_array *this_tr)
461 WARN_ON(!this_tr->ref);
466 * trace_array_put - Decrement the reference counter for this trace array.
467 * @this_tr : pointer to the trace array
469 * NOTE: Use this when we no longer need the trace array returned by
470 * trace_array_get_by_name(). This ensures the trace array can be later
474 void trace_array_put(struct trace_array *this_tr)
479 mutex_lock(&trace_types_lock);
480 __trace_array_put(this_tr);
481 mutex_unlock(&trace_types_lock);
483 EXPORT_SYMBOL_GPL(trace_array_put);
485 int tracing_check_open_get_tr(struct trace_array *tr)
489 ret = security_locked_down(LOCKDOWN_TRACEFS);
493 if (tracing_disabled)
496 if (tr && trace_array_get(tr) < 0)
502 int call_filter_check_discard(struct trace_event_call *call, void *rec,
503 struct trace_buffer *buffer,
504 struct ring_buffer_event *event)
506 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
507 !filter_match_preds(call->filter, rec)) {
508 __trace_event_discard_commit(buffer, event);
515 void trace_free_pid_list(struct trace_pid_list *pid_list)
517 vfree(pid_list->pids);
522 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
523 * @filtered_pids: The list of pids to check
524 * @search_pid: The PID to find in @filtered_pids
526 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
529 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
532 * If pid_max changed after filtered_pids was created, we
533 * by default ignore all pids greater than the previous pid_max.
535 if (search_pid >= filtered_pids->pid_max)
538 return test_bit(search_pid, filtered_pids->pids);
542 * trace_ignore_this_task - should a task be ignored for tracing
543 * @filtered_pids: The list of pids to check
544 * @filtered_no_pids: The list of pids not to be traced
545 * @task: The task that should be ignored if not filtered
547 * Checks if @task should be traced or not from @filtered_pids.
548 * Returns true if @task should *NOT* be traced.
549 * Returns false if @task should be traced.
552 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
553 struct trace_pid_list *filtered_no_pids,
554 struct task_struct *task)
557 * If filtered_no_pids is not empty, and the task's pid is listed
558 * in filtered_no_pids, then return true.
559 * Otherwise, if filtered_pids is empty, that means we can
560 * trace all tasks. If it has content, then only trace pids
561 * within filtered_pids.
564 return (filtered_pids &&
565 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
567 trace_find_filtered_pid(filtered_no_pids, task->pid));
571 * trace_filter_add_remove_task - Add or remove a task from a pid_list
572 * @pid_list: The list to modify
573 * @self: The current task for fork or NULL for exit
574 * @task: The task to add or remove
576 * If adding a task, if @self is defined, the task is only added if @self
577 * is also included in @pid_list. This happens on fork and tasks should
578 * only be added when the parent is listed. If @self is NULL, then the
579 * @task pid will be removed from the list, which would happen on exit
582 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
583 struct task_struct *self,
584 struct task_struct *task)
589 /* For forks, we only add if the forking task is listed */
591 if (!trace_find_filtered_pid(pid_list, self->pid))
595 /* Sorry, but we don't support pid_max changing after setting */
596 if (task->pid >= pid_list->pid_max)
599 /* "self" is set for forks, and NULL for exits */
601 set_bit(task->pid, pid_list->pids);
603 clear_bit(task->pid, pid_list->pids);
607 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
608 * @pid_list: The pid list to show
609 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
610 * @pos: The position of the file
612 * This is used by the seq_file "next" operation to iterate the pids
613 * listed in a trace_pid_list structure.
615 * Returns the pid+1 as we want to display pid of zero, but NULL would
616 * stop the iteration.
618 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
620 unsigned long pid = (unsigned long)v;
624 /* pid already is +1 of the actual previous bit */
625 pid = find_next_bit(pid_list->pids, pid_list->pid_max, pid);
627 /* Return pid + 1 to allow zero to be represented */
628 if (pid < pid_list->pid_max)
629 return (void *)(pid + 1);
635 * trace_pid_start - Used for seq_file to start reading pid lists
636 * @pid_list: The pid list to show
637 * @pos: The position of the file
639 * This is used by seq_file "start" operation to start the iteration
642 * Returns the pid+1 as we want to display pid of zero, but NULL would
643 * stop the iteration.
645 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
650 pid = find_first_bit(pid_list->pids, pid_list->pid_max);
651 if (pid >= pid_list->pid_max)
654 /* Return pid + 1 so that zero can be the exit value */
655 for (pid++; pid && l < *pos;
656 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
662 * trace_pid_show - show the current pid in seq_file processing
663 * @m: The seq_file structure to write into
664 * @v: A void pointer of the pid (+1) value to display
666 * Can be directly used by seq_file operations to display the current
669 int trace_pid_show(struct seq_file *m, void *v)
671 unsigned long pid = (unsigned long)v - 1;
673 seq_printf(m, "%lu\n", pid);
677 /* 128 should be much more than enough */
678 #define PID_BUF_SIZE 127
680 int trace_pid_write(struct trace_pid_list *filtered_pids,
681 struct trace_pid_list **new_pid_list,
682 const char __user *ubuf, size_t cnt)
684 struct trace_pid_list *pid_list;
685 struct trace_parser parser;
693 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
697 * Always recreate a new array. The write is an all or nothing
698 * operation. Always create a new array when adding new pids by
699 * the user. If the operation fails, then the current list is
702 pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
704 trace_parser_put(&parser);
708 pid_list->pid_max = READ_ONCE(pid_max);
710 /* Only truncating will shrink pid_max */
711 if (filtered_pids && filtered_pids->pid_max > pid_list->pid_max)
712 pid_list->pid_max = filtered_pids->pid_max;
714 pid_list->pids = vzalloc((pid_list->pid_max + 7) >> 3);
715 if (!pid_list->pids) {
716 trace_parser_put(&parser);
722 /* copy the current bits to the new max */
723 for_each_set_bit(pid, filtered_pids->pids,
724 filtered_pids->pid_max) {
725 set_bit(pid, pid_list->pids);
734 ret = trace_get_user(&parser, ubuf, cnt, &pos);
735 if (ret < 0 || !trace_parser_loaded(&parser))
743 if (kstrtoul(parser.buffer, 0, &val))
745 if (val >= pid_list->pid_max)
750 set_bit(pid, pid_list->pids);
753 trace_parser_clear(&parser);
756 trace_parser_put(&parser);
759 trace_free_pid_list(pid_list);
764 /* Cleared the list of pids */
765 trace_free_pid_list(pid_list);
770 *new_pid_list = pid_list;
775 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
779 /* Early boot up does not have a buffer yet */
781 return trace_clock_local();
783 ts = ring_buffer_time_stamp(buf->buffer);
784 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
789 u64 ftrace_now(int cpu)
791 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
795 * tracing_is_enabled - Show if global_trace has been enabled
797 * Shows if the global trace has been enabled or not. It uses the
798 * mirror flag "buffer_disabled" to be used in fast paths such as for
799 * the irqsoff tracer. But it may be inaccurate due to races. If you
800 * need to know the accurate state, use tracing_is_on() which is a little
801 * slower, but accurate.
803 int tracing_is_enabled(void)
806 * For quick access (irqsoff uses this in fast path), just
807 * return the mirror variable of the state of the ring buffer.
808 * It's a little racy, but we don't really care.
811 return !global_trace.buffer_disabled;
815 * trace_buf_size is the size in bytes that is allocated
816 * for a buffer. Note, the number of bytes is always rounded
819 * This number is purposely set to a low number of 16384.
820 * If the dump on oops happens, it will be much appreciated
821 * to not have to wait for all that output. Anyway this can be
822 * boot time and run time configurable.
824 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
826 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
828 /* trace_types holds a link list of available tracers. */
829 static struct tracer *trace_types __read_mostly;
832 * trace_types_lock is used to protect the trace_types list.
834 DEFINE_MUTEX(trace_types_lock);
837 * serialize the access of the ring buffer
839 * ring buffer serializes readers, but it is low level protection.
840 * The validity of the events (which returns by ring_buffer_peek() ..etc)
841 * are not protected by ring buffer.
843 * The content of events may become garbage if we allow other process consumes
844 * these events concurrently:
845 * A) the page of the consumed events may become a normal page
846 * (not reader page) in ring buffer, and this page will be rewritten
847 * by events producer.
848 * B) The page of the consumed events may become a page for splice_read,
849 * and this page will be returned to system.
851 * These primitives allow multi process access to different cpu ring buffer
854 * These primitives don't distinguish read-only and read-consume access.
855 * Multi read-only access are also serialized.
859 static DECLARE_RWSEM(all_cpu_access_lock);
860 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
862 static inline void trace_access_lock(int cpu)
864 if (cpu == RING_BUFFER_ALL_CPUS) {
865 /* gain it for accessing the whole ring buffer. */
866 down_write(&all_cpu_access_lock);
868 /* gain it for accessing a cpu ring buffer. */
870 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
871 down_read(&all_cpu_access_lock);
873 /* Secondly block other access to this @cpu ring buffer. */
874 mutex_lock(&per_cpu(cpu_access_lock, cpu));
878 static inline void trace_access_unlock(int cpu)
880 if (cpu == RING_BUFFER_ALL_CPUS) {
881 up_write(&all_cpu_access_lock);
883 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
884 up_read(&all_cpu_access_lock);
888 static inline void trace_access_lock_init(void)
892 for_each_possible_cpu(cpu)
893 mutex_init(&per_cpu(cpu_access_lock, cpu));
898 static DEFINE_MUTEX(access_lock);
900 static inline void trace_access_lock(int cpu)
903 mutex_lock(&access_lock);
906 static inline void trace_access_unlock(int cpu)
909 mutex_unlock(&access_lock);
912 static inline void trace_access_lock_init(void)
918 #ifdef CONFIG_STACKTRACE
919 static void __ftrace_trace_stack(struct trace_buffer *buffer,
920 unsigned int trace_ctx,
921 int skip, struct pt_regs *regs);
922 static inline void ftrace_trace_stack(struct trace_array *tr,
923 struct trace_buffer *buffer,
924 unsigned int trace_ctx,
925 int skip, struct pt_regs *regs);
928 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
929 unsigned int trace_ctx,
930 int skip, struct pt_regs *regs)
933 static inline void ftrace_trace_stack(struct trace_array *tr,
934 struct trace_buffer *buffer,
935 unsigned long trace_ctx,
936 int skip, struct pt_regs *regs)
942 static __always_inline void
943 trace_event_setup(struct ring_buffer_event *event,
944 int type, unsigned int trace_ctx)
946 struct trace_entry *ent = ring_buffer_event_data(event);
948 tracing_generic_entry_update(ent, type, trace_ctx);
951 static __always_inline struct ring_buffer_event *
952 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
955 unsigned int trace_ctx)
957 struct ring_buffer_event *event;
959 event = ring_buffer_lock_reserve(buffer, len);
961 trace_event_setup(event, type, trace_ctx);
966 void tracer_tracing_on(struct trace_array *tr)
968 if (tr->array_buffer.buffer)
969 ring_buffer_record_on(tr->array_buffer.buffer);
971 * This flag is looked at when buffers haven't been allocated
972 * yet, or by some tracers (like irqsoff), that just want to
973 * know if the ring buffer has been disabled, but it can handle
974 * races of where it gets disabled but we still do a record.
975 * As the check is in the fast path of the tracers, it is more
976 * important to be fast than accurate.
978 tr->buffer_disabled = 0;
979 /* Make the flag seen by readers */
984 * tracing_on - enable tracing buffers
986 * This function enables tracing buffers that may have been
987 * disabled with tracing_off.
989 void tracing_on(void)
991 tracer_tracing_on(&global_trace);
993 EXPORT_SYMBOL_GPL(tracing_on);
996 static __always_inline void
997 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
999 __this_cpu_write(trace_taskinfo_save, true);
1001 /* If this is the temp buffer, we need to commit fully */
1002 if (this_cpu_read(trace_buffered_event) == event) {
1003 /* Length is in event->array[0] */
1004 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1005 /* Release the temp buffer */
1006 this_cpu_dec(trace_buffered_event_cnt);
1008 ring_buffer_unlock_commit(buffer, event);
1012 * __trace_puts - write a constant string into the trace buffer.
1013 * @ip: The address of the caller
1014 * @str: The constant string to write
1015 * @size: The size of the string.
1017 int __trace_puts(unsigned long ip, const char *str, int size)
1019 struct ring_buffer_event *event;
1020 struct trace_buffer *buffer;
1021 struct print_entry *entry;
1022 unsigned int trace_ctx;
1025 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1028 if (unlikely(tracing_selftest_running || tracing_disabled))
1031 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1033 trace_ctx = tracing_gen_ctx();
1034 buffer = global_trace.array_buffer.buffer;
1035 ring_buffer_nest_start(buffer);
1036 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1043 entry = ring_buffer_event_data(event);
1046 memcpy(&entry->buf, str, size);
1048 /* Add a newline if necessary */
1049 if (entry->buf[size - 1] != '\n') {
1050 entry->buf[size] = '\n';
1051 entry->buf[size + 1] = '\0';
1053 entry->buf[size] = '\0';
1055 __buffer_unlock_commit(buffer, event);
1056 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1058 ring_buffer_nest_end(buffer);
1061 EXPORT_SYMBOL_GPL(__trace_puts);
1064 * __trace_bputs - write the pointer to a constant string into trace buffer
1065 * @ip: The address of the caller
1066 * @str: The constant string to write to the buffer to
1068 int __trace_bputs(unsigned long ip, const char *str)
1070 struct ring_buffer_event *event;
1071 struct trace_buffer *buffer;
1072 struct bputs_entry *entry;
1073 unsigned int trace_ctx;
1074 int size = sizeof(struct bputs_entry);
1077 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1080 if (unlikely(tracing_selftest_running || tracing_disabled))
1083 trace_ctx = tracing_gen_ctx();
1084 buffer = global_trace.array_buffer.buffer;
1086 ring_buffer_nest_start(buffer);
1087 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1092 entry = ring_buffer_event_data(event);
1096 __buffer_unlock_commit(buffer, event);
1097 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1101 ring_buffer_nest_end(buffer);
1104 EXPORT_SYMBOL_GPL(__trace_bputs);
1106 #ifdef CONFIG_TRACER_SNAPSHOT
1107 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1110 struct tracer *tracer = tr->current_trace;
1111 unsigned long flags;
1114 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1115 internal_trace_puts("*** snapshot is being ignored ***\n");
1119 if (!tr->allocated_snapshot) {
1120 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
1121 internal_trace_puts("*** stopping trace here! ***\n");
1126 /* Note, snapshot can not be used when the tracer uses it */
1127 if (tracer->use_max_tr) {
1128 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
1129 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
1133 local_irq_save(flags);
1134 update_max_tr(tr, current, smp_processor_id(), cond_data);
1135 local_irq_restore(flags);
1138 void tracing_snapshot_instance(struct trace_array *tr)
1140 tracing_snapshot_instance_cond(tr, NULL);
1144 * tracing_snapshot - take a snapshot of the current buffer.
1146 * This causes a swap between the snapshot buffer and the current live
1147 * tracing buffer. You can use this to take snapshots of the live
1148 * trace when some condition is triggered, but continue to trace.
1150 * Note, make sure to allocate the snapshot with either
1151 * a tracing_snapshot_alloc(), or by doing it manually
1152 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
1154 * If the snapshot buffer is not allocated, it will stop tracing.
1155 * Basically making a permanent snapshot.
1157 void tracing_snapshot(void)
1159 struct trace_array *tr = &global_trace;
1161 tracing_snapshot_instance(tr);
1163 EXPORT_SYMBOL_GPL(tracing_snapshot);
1166 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1167 * @tr: The tracing instance to snapshot
1168 * @cond_data: The data to be tested conditionally, and possibly saved
1170 * This is the same as tracing_snapshot() except that the snapshot is
1171 * conditional - the snapshot will only happen if the
1172 * cond_snapshot.update() implementation receiving the cond_data
1173 * returns true, which means that the trace array's cond_snapshot
1174 * update() operation used the cond_data to determine whether the
1175 * snapshot should be taken, and if it was, presumably saved it along
1176 * with the snapshot.
1178 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1180 tracing_snapshot_instance_cond(tr, cond_data);
1182 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1185 * tracing_snapshot_cond_data - get the user data associated with a snapshot
1186 * @tr: The tracing instance
1188 * When the user enables a conditional snapshot using
1189 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1190 * with the snapshot. This accessor is used to retrieve it.
1192 * Should not be called from cond_snapshot.update(), since it takes
1193 * the tr->max_lock lock, which the code calling
1194 * cond_snapshot.update() has already done.
1196 * Returns the cond_data associated with the trace array's snapshot.
1198 void *tracing_cond_snapshot_data(struct trace_array *tr)
1200 void *cond_data = NULL;
1202 arch_spin_lock(&tr->max_lock);
1204 if (tr->cond_snapshot)
1205 cond_data = tr->cond_snapshot->cond_data;
1207 arch_spin_unlock(&tr->max_lock);
1211 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1213 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1214 struct array_buffer *size_buf, int cpu_id);
1215 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1217 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1221 if (!tr->allocated_snapshot) {
1223 /* allocate spare buffer */
1224 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1225 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1229 tr->allocated_snapshot = true;
1235 static void free_snapshot(struct trace_array *tr)
1238 * We don't free the ring buffer. instead, resize it because
1239 * The max_tr ring buffer has some state (e.g. ring->clock) and
1240 * we want preserve it.
1242 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1243 set_buffer_entries(&tr->max_buffer, 1);
1244 tracing_reset_online_cpus(&tr->max_buffer);
1245 tr->allocated_snapshot = false;
1249 * tracing_alloc_snapshot - allocate snapshot buffer.
1251 * This only allocates the snapshot buffer if it isn't already
1252 * allocated - it doesn't also take a snapshot.
1254 * This is meant to be used in cases where the snapshot buffer needs
1255 * to be set up for events that can't sleep but need to be able to
1256 * trigger a snapshot.
1258 int tracing_alloc_snapshot(void)
1260 struct trace_array *tr = &global_trace;
1263 ret = tracing_alloc_snapshot_instance(tr);
1268 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1271 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1273 * This is similar to tracing_snapshot(), but it will allocate the
1274 * snapshot buffer if it isn't already allocated. Use this only
1275 * where it is safe to sleep, as the allocation may sleep.
1277 * This causes a swap between the snapshot buffer and the current live
1278 * tracing buffer. You can use this to take snapshots of the live
1279 * trace when some condition is triggered, but continue to trace.
1281 void tracing_snapshot_alloc(void)
1285 ret = tracing_alloc_snapshot();
1291 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1294 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1295 * @tr: The tracing instance
1296 * @cond_data: User data to associate with the snapshot
1297 * @update: Implementation of the cond_snapshot update function
1299 * Check whether the conditional snapshot for the given instance has
1300 * already been enabled, or if the current tracer is already using a
1301 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1302 * save the cond_data and update function inside.
1304 * Returns 0 if successful, error otherwise.
1306 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1307 cond_update_fn_t update)
1309 struct cond_snapshot *cond_snapshot;
1312 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1316 cond_snapshot->cond_data = cond_data;
1317 cond_snapshot->update = update;
1319 mutex_lock(&trace_types_lock);
1321 ret = tracing_alloc_snapshot_instance(tr);
1325 if (tr->current_trace->use_max_tr) {
1331 * The cond_snapshot can only change to NULL without the
1332 * trace_types_lock. We don't care if we race with it going
1333 * to NULL, but we want to make sure that it's not set to
1334 * something other than NULL when we get here, which we can
1335 * do safely with only holding the trace_types_lock and not
1336 * having to take the max_lock.
1338 if (tr->cond_snapshot) {
1343 arch_spin_lock(&tr->max_lock);
1344 tr->cond_snapshot = cond_snapshot;
1345 arch_spin_unlock(&tr->max_lock);
1347 mutex_unlock(&trace_types_lock);
1352 mutex_unlock(&trace_types_lock);
1353 kfree(cond_snapshot);
1356 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1359 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1360 * @tr: The tracing instance
1362 * Check whether the conditional snapshot for the given instance is
1363 * enabled; if so, free the cond_snapshot associated with it,
1364 * otherwise return -EINVAL.
1366 * Returns 0 if successful, error otherwise.
1368 int tracing_snapshot_cond_disable(struct trace_array *tr)
1372 arch_spin_lock(&tr->max_lock);
1374 if (!tr->cond_snapshot)
1377 kfree(tr->cond_snapshot);
1378 tr->cond_snapshot = NULL;
1381 arch_spin_unlock(&tr->max_lock);
1385 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1387 void tracing_snapshot(void)
1389 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1391 EXPORT_SYMBOL_GPL(tracing_snapshot);
1392 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1394 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1396 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1397 int tracing_alloc_snapshot(void)
1399 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1402 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1403 void tracing_snapshot_alloc(void)
1408 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1409 void *tracing_cond_snapshot_data(struct trace_array *tr)
1413 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1414 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1418 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1419 int tracing_snapshot_cond_disable(struct trace_array *tr)
1423 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1424 #endif /* CONFIG_TRACER_SNAPSHOT */
1426 void tracer_tracing_off(struct trace_array *tr)
1428 if (tr->array_buffer.buffer)
1429 ring_buffer_record_off(tr->array_buffer.buffer);
1431 * This flag is looked at when buffers haven't been allocated
1432 * yet, or by some tracers (like irqsoff), that just want to
1433 * know if the ring buffer has been disabled, but it can handle
1434 * races of where it gets disabled but we still do a record.
1435 * As the check is in the fast path of the tracers, it is more
1436 * important to be fast than accurate.
1438 tr->buffer_disabled = 1;
1439 /* Make the flag seen by readers */
1444 * tracing_off - turn off tracing buffers
1446 * This function stops the tracing buffers from recording data.
1447 * It does not disable any overhead the tracers themselves may
1448 * be causing. This function simply causes all recording to
1449 * the ring buffers to fail.
1451 void tracing_off(void)
1453 tracer_tracing_off(&global_trace);
1455 EXPORT_SYMBOL_GPL(tracing_off);
1457 void disable_trace_on_warning(void)
1459 if (__disable_trace_on_warning) {
1460 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1461 "Disabling tracing due to warning\n");
1467 * tracer_tracing_is_on - show real state of ring buffer enabled
1468 * @tr : the trace array to know if ring buffer is enabled
1470 * Shows real state of the ring buffer if it is enabled or not.
1472 bool tracer_tracing_is_on(struct trace_array *tr)
1474 if (tr->array_buffer.buffer)
1475 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1476 return !tr->buffer_disabled;
1480 * tracing_is_on - show state of ring buffers enabled
1482 int tracing_is_on(void)
1484 return tracer_tracing_is_on(&global_trace);
1486 EXPORT_SYMBOL_GPL(tracing_is_on);
1488 static int __init set_buf_size(char *str)
1490 unsigned long buf_size;
1494 buf_size = memparse(str, &str);
1495 /* nr_entries can not be zero */
1498 trace_buf_size = buf_size;
1501 __setup("trace_buf_size=", set_buf_size);
1503 static int __init set_tracing_thresh(char *str)
1505 unsigned long threshold;
1510 ret = kstrtoul(str, 0, &threshold);
1513 tracing_thresh = threshold * 1000;
1516 __setup("tracing_thresh=", set_tracing_thresh);
1518 unsigned long nsecs_to_usecs(unsigned long nsecs)
1520 return nsecs / 1000;
1524 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1525 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1526 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1527 * of strings in the order that the evals (enum) were defined.
1532 /* These must match the bit positions in trace_iterator_flags */
1533 static const char *trace_options[] = {
1541 int in_ns; /* is this clock in nanoseconds? */
1542 } trace_clocks[] = {
1543 { trace_clock_local, "local", 1 },
1544 { trace_clock_global, "global", 1 },
1545 { trace_clock_counter, "counter", 0 },
1546 { trace_clock_jiffies, "uptime", 0 },
1547 { trace_clock, "perf", 1 },
1548 { ktime_get_mono_fast_ns, "mono", 1 },
1549 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1550 { ktime_get_boot_fast_ns, "boot", 1 },
1554 bool trace_clock_in_ns(struct trace_array *tr)
1556 if (trace_clocks[tr->clock_id].in_ns)
1563 * trace_parser_get_init - gets the buffer for trace parser
1565 int trace_parser_get_init(struct trace_parser *parser, int size)
1567 memset(parser, 0, sizeof(*parser));
1569 parser->buffer = kmalloc(size, GFP_KERNEL);
1570 if (!parser->buffer)
1573 parser->size = size;
1578 * trace_parser_put - frees the buffer for trace parser
1580 void trace_parser_put(struct trace_parser *parser)
1582 kfree(parser->buffer);
1583 parser->buffer = NULL;
1587 * trace_get_user - reads the user input string separated by space
1588 * (matched by isspace(ch))
1590 * For each string found the 'struct trace_parser' is updated,
1591 * and the function returns.
1593 * Returns number of bytes read.
1595 * See kernel/trace/trace.h for 'struct trace_parser' details.
1597 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1598 size_t cnt, loff_t *ppos)
1605 trace_parser_clear(parser);
1607 ret = get_user(ch, ubuf++);
1615 * The parser is not finished with the last write,
1616 * continue reading the user input without skipping spaces.
1618 if (!parser->cont) {
1619 /* skip white space */
1620 while (cnt && isspace(ch)) {
1621 ret = get_user(ch, ubuf++);
1630 /* only spaces were written */
1631 if (isspace(ch) || !ch) {
1638 /* read the non-space input */
1639 while (cnt && !isspace(ch) && ch) {
1640 if (parser->idx < parser->size - 1)
1641 parser->buffer[parser->idx++] = ch;
1646 ret = get_user(ch, ubuf++);
1653 /* We either got finished input or we have to wait for another call. */
1654 if (isspace(ch) || !ch) {
1655 parser->buffer[parser->idx] = 0;
1656 parser->cont = false;
1657 } else if (parser->idx < parser->size - 1) {
1658 parser->cont = true;
1659 parser->buffer[parser->idx++] = ch;
1660 /* Make sure the parsed string always terminates with '\0'. */
1661 parser->buffer[parser->idx] = 0;
1674 /* TODO add a seq_buf_to_buffer() */
1675 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1679 if (trace_seq_used(s) <= s->seq.readpos)
1682 len = trace_seq_used(s) - s->seq.readpos;
1685 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1687 s->seq.readpos += cnt;
1691 unsigned long __read_mostly tracing_thresh;
1692 static const struct file_operations tracing_max_lat_fops;
1694 #ifdef LATENCY_FS_NOTIFY
1696 static struct workqueue_struct *fsnotify_wq;
1698 static void latency_fsnotify_workfn(struct work_struct *work)
1700 struct trace_array *tr = container_of(work, struct trace_array,
1702 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1705 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1707 struct trace_array *tr = container_of(iwork, struct trace_array,
1709 queue_work(fsnotify_wq, &tr->fsnotify_work);
1712 static void trace_create_maxlat_file(struct trace_array *tr,
1713 struct dentry *d_tracer)
1715 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1716 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1717 tr->d_max_latency = trace_create_file("tracing_max_latency", 0644,
1718 d_tracer, &tr->max_latency,
1719 &tracing_max_lat_fops);
1722 __init static int latency_fsnotify_init(void)
1724 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1725 WQ_UNBOUND | WQ_HIGHPRI, 0);
1727 pr_err("Unable to allocate tr_max_lat_wq\n");
1733 late_initcall_sync(latency_fsnotify_init);
1735 void latency_fsnotify(struct trace_array *tr)
1740 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1741 * possible that we are called from __schedule() or do_idle(), which
1742 * could cause a deadlock.
1744 irq_work_queue(&tr->fsnotify_irqwork);
1748 * (defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)) && \
1749 * defined(CONFIG_FSNOTIFY)
1753 #define trace_create_maxlat_file(tr, d_tracer) \
1754 trace_create_file("tracing_max_latency", 0644, d_tracer, \
1755 &tr->max_latency, &tracing_max_lat_fops)
1759 #ifdef CONFIG_TRACER_MAX_TRACE
1761 * Copy the new maximum trace into the separate maximum-trace
1762 * structure. (this way the maximum trace is permanently saved,
1763 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1766 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1768 struct array_buffer *trace_buf = &tr->array_buffer;
1769 struct array_buffer *max_buf = &tr->max_buffer;
1770 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1771 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1774 max_buf->time_start = data->preempt_timestamp;
1776 max_data->saved_latency = tr->max_latency;
1777 max_data->critical_start = data->critical_start;
1778 max_data->critical_end = data->critical_end;
1780 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1781 max_data->pid = tsk->pid;
1783 * If tsk == current, then use current_uid(), as that does not use
1784 * RCU. The irq tracer can be called out of RCU scope.
1787 max_data->uid = current_uid();
1789 max_data->uid = task_uid(tsk);
1791 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1792 max_data->policy = tsk->policy;
1793 max_data->rt_priority = tsk->rt_priority;
1795 /* record this tasks comm */
1796 tracing_record_cmdline(tsk);
1797 latency_fsnotify(tr);
1801 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1803 * @tsk: the task with the latency
1804 * @cpu: The cpu that initiated the trace.
1805 * @cond_data: User data associated with a conditional snapshot
1807 * Flip the buffers between the @tr and the max_tr and record information
1808 * about which task was the cause of this latency.
1811 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1817 WARN_ON_ONCE(!irqs_disabled());
1819 if (!tr->allocated_snapshot) {
1820 /* Only the nop tracer should hit this when disabling */
1821 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1825 arch_spin_lock(&tr->max_lock);
1827 /* Inherit the recordable setting from array_buffer */
1828 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1829 ring_buffer_record_on(tr->max_buffer.buffer);
1831 ring_buffer_record_off(tr->max_buffer.buffer);
1833 #ifdef CONFIG_TRACER_SNAPSHOT
1834 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data))
1837 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1839 __update_max_tr(tr, tsk, cpu);
1842 arch_spin_unlock(&tr->max_lock);
1846 * update_max_tr_single - only copy one trace over, and reset the rest
1848 * @tsk: task with the latency
1849 * @cpu: the cpu of the buffer to copy.
1851 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1854 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1861 WARN_ON_ONCE(!irqs_disabled());
1862 if (!tr->allocated_snapshot) {
1863 /* Only the nop tracer should hit this when disabling */
1864 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1868 arch_spin_lock(&tr->max_lock);
1870 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1872 if (ret == -EBUSY) {
1874 * We failed to swap the buffer due to a commit taking
1875 * place on this CPU. We fail to record, but we reset
1876 * the max trace buffer (no one writes directly to it)
1877 * and flag that it failed.
1879 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1880 "Failed to swap buffers due to commit in progress\n");
1883 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1885 __update_max_tr(tr, tsk, cpu);
1886 arch_spin_unlock(&tr->max_lock);
1888 #endif /* CONFIG_TRACER_MAX_TRACE */
1890 static int wait_on_pipe(struct trace_iterator *iter, int full)
1892 /* Iterators are static, they should be filled or empty */
1893 if (trace_buffer_iter(iter, iter->cpu_file))
1896 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1900 #ifdef CONFIG_FTRACE_STARTUP_TEST
1901 static bool selftests_can_run;
1903 struct trace_selftests {
1904 struct list_head list;
1905 struct tracer *type;
1908 static LIST_HEAD(postponed_selftests);
1910 static int save_selftest(struct tracer *type)
1912 struct trace_selftests *selftest;
1914 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1918 selftest->type = type;
1919 list_add(&selftest->list, &postponed_selftests);
1923 static int run_tracer_selftest(struct tracer *type)
1925 struct trace_array *tr = &global_trace;
1926 struct tracer *saved_tracer = tr->current_trace;
1929 if (!type->selftest || tracing_selftest_disabled)
1933 * If a tracer registers early in boot up (before scheduling is
1934 * initialized and such), then do not run its selftests yet.
1935 * Instead, run it a little later in the boot process.
1937 if (!selftests_can_run)
1938 return save_selftest(type);
1940 if (!tracing_is_on()) {
1941 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
1947 * Run a selftest on this tracer.
1948 * Here we reset the trace buffer, and set the current
1949 * tracer to be this tracer. The tracer can then run some
1950 * internal tracing to verify that everything is in order.
1951 * If we fail, we do not register this tracer.
1953 tracing_reset_online_cpus(&tr->array_buffer);
1955 tr->current_trace = type;
1957 #ifdef CONFIG_TRACER_MAX_TRACE
1958 if (type->use_max_tr) {
1959 /* If we expanded the buffers, make sure the max is expanded too */
1960 if (ring_buffer_expanded)
1961 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1962 RING_BUFFER_ALL_CPUS);
1963 tr->allocated_snapshot = true;
1967 /* the test is responsible for initializing and enabling */
1968 pr_info("Testing tracer %s: ", type->name);
1969 ret = type->selftest(type, tr);
1970 /* the test is responsible for resetting too */
1971 tr->current_trace = saved_tracer;
1973 printk(KERN_CONT "FAILED!\n");
1974 /* Add the warning after printing 'FAILED' */
1978 /* Only reset on passing, to avoid touching corrupted buffers */
1979 tracing_reset_online_cpus(&tr->array_buffer);
1981 #ifdef CONFIG_TRACER_MAX_TRACE
1982 if (type->use_max_tr) {
1983 tr->allocated_snapshot = false;
1985 /* Shrink the max buffer again */
1986 if (ring_buffer_expanded)
1987 ring_buffer_resize(tr->max_buffer.buffer, 1,
1988 RING_BUFFER_ALL_CPUS);
1992 printk(KERN_CONT "PASSED\n");
1996 static __init int init_trace_selftests(void)
1998 struct trace_selftests *p, *n;
1999 struct tracer *t, **last;
2002 selftests_can_run = true;
2004 mutex_lock(&trace_types_lock);
2006 if (list_empty(&postponed_selftests))
2009 pr_info("Running postponed tracer tests:\n");
2011 tracing_selftest_running = true;
2012 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2013 /* This loop can take minutes when sanitizers are enabled, so
2014 * lets make sure we allow RCU processing.
2017 ret = run_tracer_selftest(p->type);
2018 /* If the test fails, then warn and remove from available_tracers */
2020 WARN(1, "tracer: %s failed selftest, disabling\n",
2022 last = &trace_types;
2023 for (t = trace_types; t; t = t->next) {
2034 tracing_selftest_running = false;
2037 mutex_unlock(&trace_types_lock);
2041 core_initcall(init_trace_selftests);
2043 static inline int run_tracer_selftest(struct tracer *type)
2047 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2049 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2051 static void __init apply_trace_boot_options(void);
2054 * register_tracer - register a tracer with the ftrace system.
2055 * @type: the plugin for the tracer
2057 * Register a new plugin tracer.
2059 int __init register_tracer(struct tracer *type)
2065 pr_info("Tracer must have a name\n");
2069 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2070 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2074 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2075 pr_warn("Can not register tracer %s due to lockdown\n",
2080 mutex_lock(&trace_types_lock);
2082 tracing_selftest_running = true;
2084 for (t = trace_types; t; t = t->next) {
2085 if (strcmp(type->name, t->name) == 0) {
2087 pr_info("Tracer %s already registered\n",
2094 if (!type->set_flag)
2095 type->set_flag = &dummy_set_flag;
2097 /*allocate a dummy tracer_flags*/
2098 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2103 type->flags->val = 0;
2104 type->flags->opts = dummy_tracer_opt;
2106 if (!type->flags->opts)
2107 type->flags->opts = dummy_tracer_opt;
2109 /* store the tracer for __set_tracer_option */
2110 type->flags->trace = type;
2112 ret = run_tracer_selftest(type);
2116 type->next = trace_types;
2118 add_tracer_options(&global_trace, type);
2121 tracing_selftest_running = false;
2122 mutex_unlock(&trace_types_lock);
2124 if (ret || !default_bootup_tracer)
2127 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2130 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2131 /* Do we want this tracer to start on bootup? */
2132 tracing_set_tracer(&global_trace, type->name);
2133 default_bootup_tracer = NULL;
2135 apply_trace_boot_options();
2137 /* disable other selftests, since this will break it. */
2138 disable_tracing_selftest("running a tracer");
2144 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2146 struct trace_buffer *buffer = buf->buffer;
2151 ring_buffer_record_disable(buffer);
2153 /* Make sure all commits have finished */
2155 ring_buffer_reset_cpu(buffer, cpu);
2157 ring_buffer_record_enable(buffer);
2160 void tracing_reset_online_cpus(struct array_buffer *buf)
2162 struct trace_buffer *buffer = buf->buffer;
2167 ring_buffer_record_disable(buffer);
2169 /* Make sure all commits have finished */
2172 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2174 ring_buffer_reset_online_cpus(buffer);
2176 ring_buffer_record_enable(buffer);
2179 /* Must have trace_types_lock held */
2180 void tracing_reset_all_online_cpus(void)
2182 struct trace_array *tr;
2184 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2185 if (!tr->clear_trace)
2187 tr->clear_trace = false;
2188 tracing_reset_online_cpus(&tr->array_buffer);
2189 #ifdef CONFIG_TRACER_MAX_TRACE
2190 tracing_reset_online_cpus(&tr->max_buffer);
2196 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2197 * is the tgid last observed corresponding to pid=i.
2199 static int *tgid_map;
2201 /* The maximum valid index into tgid_map. */
2202 static size_t tgid_map_max;
2204 #define SAVED_CMDLINES_DEFAULT 128
2205 #define NO_CMDLINE_MAP UINT_MAX
2206 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2207 struct saved_cmdlines_buffer {
2208 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2209 unsigned *map_cmdline_to_pid;
2210 unsigned cmdline_num;
2212 char *saved_cmdlines;
2214 static struct saved_cmdlines_buffer *savedcmd;
2216 static inline char *get_saved_cmdlines(int idx)
2218 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2221 static inline void set_cmdline(int idx, const char *cmdline)
2223 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2226 static int allocate_cmdlines_buffer(unsigned int val,
2227 struct saved_cmdlines_buffer *s)
2229 s->map_cmdline_to_pid = kmalloc_array(val,
2230 sizeof(*s->map_cmdline_to_pid),
2232 if (!s->map_cmdline_to_pid)
2235 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2236 if (!s->saved_cmdlines) {
2237 kfree(s->map_cmdline_to_pid);
2242 s->cmdline_num = val;
2243 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2244 sizeof(s->map_pid_to_cmdline));
2245 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2246 val * sizeof(*s->map_cmdline_to_pid));
2251 static int trace_create_savedcmd(void)
2255 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2259 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2269 int is_tracing_stopped(void)
2271 return global_trace.stop_count;
2275 * tracing_start - quick start of the tracer
2277 * If tracing is enabled but was stopped by tracing_stop,
2278 * this will start the tracer back up.
2280 void tracing_start(void)
2282 struct trace_buffer *buffer;
2283 unsigned long flags;
2285 if (tracing_disabled)
2288 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2289 if (--global_trace.stop_count) {
2290 if (global_trace.stop_count < 0) {
2291 /* Someone screwed up their debugging */
2293 global_trace.stop_count = 0;
2298 /* Prevent the buffers from switching */
2299 arch_spin_lock(&global_trace.max_lock);
2301 buffer = global_trace.array_buffer.buffer;
2303 ring_buffer_record_enable(buffer);
2305 #ifdef CONFIG_TRACER_MAX_TRACE
2306 buffer = global_trace.max_buffer.buffer;
2308 ring_buffer_record_enable(buffer);
2311 arch_spin_unlock(&global_trace.max_lock);
2314 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2317 static void tracing_start_tr(struct trace_array *tr)
2319 struct trace_buffer *buffer;
2320 unsigned long flags;
2322 if (tracing_disabled)
2325 /* If global, we need to also start the max tracer */
2326 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2327 return tracing_start();
2329 raw_spin_lock_irqsave(&tr->start_lock, flags);
2331 if (--tr->stop_count) {
2332 if (tr->stop_count < 0) {
2333 /* Someone screwed up their debugging */
2340 buffer = tr->array_buffer.buffer;
2342 ring_buffer_record_enable(buffer);
2345 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2349 * tracing_stop - quick stop of the tracer
2351 * Light weight way to stop tracing. Use in conjunction with
2354 void tracing_stop(void)
2356 struct trace_buffer *buffer;
2357 unsigned long flags;
2359 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2360 if (global_trace.stop_count++)
2363 /* Prevent the buffers from switching */
2364 arch_spin_lock(&global_trace.max_lock);
2366 buffer = global_trace.array_buffer.buffer;
2368 ring_buffer_record_disable(buffer);
2370 #ifdef CONFIG_TRACER_MAX_TRACE
2371 buffer = global_trace.max_buffer.buffer;
2373 ring_buffer_record_disable(buffer);
2376 arch_spin_unlock(&global_trace.max_lock);
2379 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2382 static void tracing_stop_tr(struct trace_array *tr)
2384 struct trace_buffer *buffer;
2385 unsigned long flags;
2387 /* If global, we need to also stop the max tracer */
2388 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2389 return tracing_stop();
2391 raw_spin_lock_irqsave(&tr->start_lock, flags);
2392 if (tr->stop_count++)
2395 buffer = tr->array_buffer.buffer;
2397 ring_buffer_record_disable(buffer);
2400 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2403 static int trace_save_cmdline(struct task_struct *tsk)
2407 /* treat recording of idle task as a success */
2411 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2414 * It's not the end of the world if we don't get
2415 * the lock, but we also don't want to spin
2416 * nor do we want to disable interrupts,
2417 * so if we miss here, then better luck next time.
2419 if (!arch_spin_trylock(&trace_cmdline_lock))
2422 idx = savedcmd->map_pid_to_cmdline[tpid];
2423 if (idx == NO_CMDLINE_MAP) {
2424 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2426 savedcmd->map_pid_to_cmdline[tpid] = idx;
2427 savedcmd->cmdline_idx = idx;
2430 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2431 set_cmdline(idx, tsk->comm);
2433 arch_spin_unlock(&trace_cmdline_lock);
2438 static void __trace_find_cmdline(int pid, char comm[])
2444 strcpy(comm, "<idle>");
2448 if (WARN_ON_ONCE(pid < 0)) {
2449 strcpy(comm, "<XXX>");
2453 tpid = pid & (PID_MAX_DEFAULT - 1);
2454 map = savedcmd->map_pid_to_cmdline[tpid];
2455 if (map != NO_CMDLINE_MAP) {
2456 tpid = savedcmd->map_cmdline_to_pid[map];
2458 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2462 strcpy(comm, "<...>");
2465 void trace_find_cmdline(int pid, char comm[])
2468 arch_spin_lock(&trace_cmdline_lock);
2470 __trace_find_cmdline(pid, comm);
2472 arch_spin_unlock(&trace_cmdline_lock);
2476 static int *trace_find_tgid_ptr(int pid)
2479 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2480 * if we observe a non-NULL tgid_map then we also observe the correct
2483 int *map = smp_load_acquire(&tgid_map);
2485 if (unlikely(!map || pid > tgid_map_max))
2491 int trace_find_tgid(int pid)
2493 int *ptr = trace_find_tgid_ptr(pid);
2495 return ptr ? *ptr : 0;
2498 static int trace_save_tgid(struct task_struct *tsk)
2502 /* treat recording of idle task as a success */
2506 ptr = trace_find_tgid_ptr(tsk->pid);
2514 static bool tracing_record_taskinfo_skip(int flags)
2516 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2518 if (!__this_cpu_read(trace_taskinfo_save))
2524 * tracing_record_taskinfo - record the task info of a task
2526 * @task: task to record
2527 * @flags: TRACE_RECORD_CMDLINE for recording comm
2528 * TRACE_RECORD_TGID for recording tgid
2530 void tracing_record_taskinfo(struct task_struct *task, int flags)
2534 if (tracing_record_taskinfo_skip(flags))
2538 * Record as much task information as possible. If some fail, continue
2539 * to try to record the others.
2541 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2542 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2544 /* If recording any information failed, retry again soon. */
2548 __this_cpu_write(trace_taskinfo_save, false);
2552 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2554 * @prev: previous task during sched_switch
2555 * @next: next task during sched_switch
2556 * @flags: TRACE_RECORD_CMDLINE for recording comm
2557 * TRACE_RECORD_TGID for recording tgid
2559 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2560 struct task_struct *next, int flags)
2564 if (tracing_record_taskinfo_skip(flags))
2568 * Record as much task information as possible. If some fail, continue
2569 * to try to record the others.
2571 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2572 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2573 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2574 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2576 /* If recording any information failed, retry again soon. */
2580 __this_cpu_write(trace_taskinfo_save, false);
2583 /* Helpers to record a specific task information */
2584 void tracing_record_cmdline(struct task_struct *task)
2586 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2589 void tracing_record_tgid(struct task_struct *task)
2591 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2595 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2596 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2597 * simplifies those functions and keeps them in sync.
2599 enum print_line_t trace_handle_return(struct trace_seq *s)
2601 return trace_seq_has_overflowed(s) ?
2602 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2604 EXPORT_SYMBOL_GPL(trace_handle_return);
2606 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2608 unsigned int trace_flags = irqs_status;
2611 pc = preempt_count();
2614 trace_flags |= TRACE_FLAG_NMI;
2615 if (pc & HARDIRQ_MASK)
2616 trace_flags |= TRACE_FLAG_HARDIRQ;
2617 if (in_serving_softirq())
2618 trace_flags |= TRACE_FLAG_SOFTIRQ;
2620 if (tif_need_resched())
2621 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2622 if (test_preempt_need_resched())
2623 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2624 return (trace_flags << 16) | (pc & 0xff);
2627 struct ring_buffer_event *
2628 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2631 unsigned int trace_ctx)
2633 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2636 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2637 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2638 static int trace_buffered_event_ref;
2641 * trace_buffered_event_enable - enable buffering events
2643 * When events are being filtered, it is quicker to use a temporary
2644 * buffer to write the event data into if there's a likely chance
2645 * that it will not be committed. The discard of the ring buffer
2646 * is not as fast as committing, and is much slower than copying
2649 * When an event is to be filtered, allocate per cpu buffers to
2650 * write the event data into, and if the event is filtered and discarded
2651 * it is simply dropped, otherwise, the entire data is to be committed
2654 void trace_buffered_event_enable(void)
2656 struct ring_buffer_event *event;
2660 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2662 if (trace_buffered_event_ref++)
2665 for_each_tracing_cpu(cpu) {
2666 page = alloc_pages_node(cpu_to_node(cpu),
2667 GFP_KERNEL | __GFP_NORETRY, 0);
2671 event = page_address(page);
2672 memset(event, 0, sizeof(*event));
2674 per_cpu(trace_buffered_event, cpu) = event;
2677 if (cpu == smp_processor_id() &&
2678 __this_cpu_read(trace_buffered_event) !=
2679 per_cpu(trace_buffered_event, cpu))
2686 trace_buffered_event_disable();
2689 static void enable_trace_buffered_event(void *data)
2691 /* Probably not needed, but do it anyway */
2693 this_cpu_dec(trace_buffered_event_cnt);
2696 static void disable_trace_buffered_event(void *data)
2698 this_cpu_inc(trace_buffered_event_cnt);
2702 * trace_buffered_event_disable - disable buffering events
2704 * When a filter is removed, it is faster to not use the buffered
2705 * events, and to commit directly into the ring buffer. Free up
2706 * the temp buffers when there are no more users. This requires
2707 * special synchronization with current events.
2709 void trace_buffered_event_disable(void)
2713 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2715 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2718 if (--trace_buffered_event_ref)
2722 /* For each CPU, set the buffer as used. */
2723 smp_call_function_many(tracing_buffer_mask,
2724 disable_trace_buffered_event, NULL, 1);
2727 /* Wait for all current users to finish */
2730 for_each_tracing_cpu(cpu) {
2731 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2732 per_cpu(trace_buffered_event, cpu) = NULL;
2735 * Make sure trace_buffered_event is NULL before clearing
2736 * trace_buffered_event_cnt.
2741 /* Do the work on each cpu */
2742 smp_call_function_many(tracing_buffer_mask,
2743 enable_trace_buffered_event, NULL, 1);
2747 static struct trace_buffer *temp_buffer;
2749 struct ring_buffer_event *
2750 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2751 struct trace_event_file *trace_file,
2752 int type, unsigned long len,
2753 unsigned int trace_ctx)
2755 struct ring_buffer_event *entry;
2756 struct trace_array *tr = trace_file->tr;
2759 *current_rb = tr->array_buffer.buffer;
2761 if (!tr->no_filter_buffering_ref &&
2762 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED)) &&
2763 (entry = this_cpu_read(trace_buffered_event))) {
2765 * Filtering is on, so try to use the per cpu buffer first.
2766 * This buffer will simulate a ring_buffer_event,
2767 * where the type_len is zero and the array[0] will
2768 * hold the full length.
2769 * (see include/linux/ring-buffer.h for details on
2770 * how the ring_buffer_event is structured).
2772 * Using a temp buffer during filtering and copying it
2773 * on a matched filter is quicker than writing directly
2774 * into the ring buffer and then discarding it when
2775 * it doesn't match. That is because the discard
2776 * requires several atomic operations to get right.
2777 * Copying on match and doing nothing on a failed match
2778 * is still quicker than no copy on match, but having
2779 * to discard out of the ring buffer on a failed match.
2781 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2783 val = this_cpu_inc_return(trace_buffered_event_cnt);
2786 * Preemption is disabled, but interrupts and NMIs
2787 * can still come in now. If that happens after
2788 * the above increment, then it will have to go
2789 * back to the old method of allocating the event
2790 * on the ring buffer, and if the filter fails, it
2791 * will have to call ring_buffer_discard_commit()
2794 * Need to also check the unlikely case that the
2795 * length is bigger than the temp buffer size.
2796 * If that happens, then the reserve is pretty much
2797 * guaranteed to fail, as the ring buffer currently
2798 * only allows events less than a page. But that may
2799 * change in the future, so let the ring buffer reserve
2800 * handle the failure in that case.
2802 if (val == 1 && likely(len <= max_len)) {
2803 trace_event_setup(entry, type, trace_ctx);
2804 entry->array[0] = len;
2807 this_cpu_dec(trace_buffered_event_cnt);
2810 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2813 * If tracing is off, but we have triggers enabled
2814 * we still need to look at the event data. Use the temp_buffer
2815 * to store the trace event for the trigger to use. It's recursive
2816 * safe and will not be recorded anywhere.
2818 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2819 *current_rb = temp_buffer;
2820 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2825 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2827 static DEFINE_SPINLOCK(tracepoint_iter_lock);
2828 static DEFINE_MUTEX(tracepoint_printk_mutex);
2830 static void output_printk(struct trace_event_buffer *fbuffer)
2832 struct trace_event_call *event_call;
2833 struct trace_event_file *file;
2834 struct trace_event *event;
2835 unsigned long flags;
2836 struct trace_iterator *iter = tracepoint_print_iter;
2838 /* We should never get here if iter is NULL */
2839 if (WARN_ON_ONCE(!iter))
2842 event_call = fbuffer->trace_file->event_call;
2843 if (!event_call || !event_call->event.funcs ||
2844 !event_call->event.funcs->trace)
2847 file = fbuffer->trace_file;
2848 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2849 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2850 !filter_match_preds(file->filter, fbuffer->entry)))
2853 event = &fbuffer->trace_file->event_call->event;
2855 spin_lock_irqsave(&tracepoint_iter_lock, flags);
2856 trace_seq_init(&iter->seq);
2857 iter->ent = fbuffer->entry;
2858 event_call->event.funcs->trace(iter, 0, event);
2859 trace_seq_putc(&iter->seq, 0);
2860 printk("%s", iter->seq.buffer);
2862 spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2865 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2866 void *buffer, size_t *lenp,
2869 int save_tracepoint_printk;
2872 mutex_lock(&tracepoint_printk_mutex);
2873 save_tracepoint_printk = tracepoint_printk;
2875 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2878 * This will force exiting early, as tracepoint_printk
2879 * is always zero when tracepoint_printk_iter is not allocated
2881 if (!tracepoint_print_iter)
2882 tracepoint_printk = 0;
2884 if (save_tracepoint_printk == tracepoint_printk)
2887 if (tracepoint_printk)
2888 static_key_enable(&tracepoint_printk_key.key);
2890 static_key_disable(&tracepoint_printk_key.key);
2893 mutex_unlock(&tracepoint_printk_mutex);
2898 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2900 enum event_trigger_type tt = ETT_NONE;
2901 struct trace_event_file *file = fbuffer->trace_file;
2903 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2904 fbuffer->entry, &tt))
2907 if (static_key_false(&tracepoint_printk_key.key))
2908 output_printk(fbuffer);
2910 if (static_branch_unlikely(&trace_event_exports_enabled))
2911 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2913 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2914 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
2918 event_triggers_post_call(file, tt);
2921 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2926 * trace_buffer_unlock_commit_regs()
2927 * trace_event_buffer_commit()
2928 * trace_event_raw_event_xxx()
2930 # define STACK_SKIP 3
2932 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2933 struct trace_buffer *buffer,
2934 struct ring_buffer_event *event,
2935 unsigned int trace_ctx,
2936 struct pt_regs *regs)
2938 __buffer_unlock_commit(buffer, event);
2941 * If regs is not set, then skip the necessary functions.
2942 * Note, we can still get here via blktrace, wakeup tracer
2943 * and mmiotrace, but that's ok if they lose a function or
2944 * two. They are not that meaningful.
2946 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
2947 ftrace_trace_userstack(tr, buffer, trace_ctx);
2951 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2954 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2955 struct ring_buffer_event *event)
2957 __buffer_unlock_commit(buffer, event);
2961 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
2962 parent_ip, unsigned int trace_ctx)
2964 struct trace_event_call *call = &event_function;
2965 struct trace_buffer *buffer = tr->array_buffer.buffer;
2966 struct ring_buffer_event *event;
2967 struct ftrace_entry *entry;
2969 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
2973 entry = ring_buffer_event_data(event);
2975 entry->parent_ip = parent_ip;
2977 if (!call_filter_check_discard(call, entry, buffer, event)) {
2978 if (static_branch_unlikely(&trace_function_exports_enabled))
2979 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
2980 __buffer_unlock_commit(buffer, event);
2984 #ifdef CONFIG_STACKTRACE
2986 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
2987 #define FTRACE_KSTACK_NESTING 4
2989 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
2991 struct ftrace_stack {
2992 unsigned long calls[FTRACE_KSTACK_ENTRIES];
2996 struct ftrace_stacks {
2997 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3000 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3001 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3003 static void __ftrace_trace_stack(struct trace_buffer *buffer,
3004 unsigned int trace_ctx,
3005 int skip, struct pt_regs *regs)
3007 struct trace_event_call *call = &event_kernel_stack;
3008 struct ring_buffer_event *event;
3009 unsigned int size, nr_entries;
3010 struct ftrace_stack *fstack;
3011 struct stack_entry *entry;
3015 * Add one, for this function and the call to save_stack_trace()
3016 * If regs is set, then these functions will not be in the way.
3018 #ifndef CONFIG_UNWINDER_ORC
3023 preempt_disable_notrace();
3025 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3027 /* This should never happen. If it does, yell once and skip */
3028 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3032 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3033 * interrupt will either see the value pre increment or post
3034 * increment. If the interrupt happens pre increment it will have
3035 * restored the counter when it returns. We just need a barrier to
3036 * keep gcc from moving things around.
3040 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3041 size = ARRAY_SIZE(fstack->calls);
3044 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3047 nr_entries = stack_trace_save(fstack->calls, size, skip);
3050 size = nr_entries * sizeof(unsigned long);
3051 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3052 (sizeof(*entry) - sizeof(entry->caller)) + size,
3056 entry = ring_buffer_event_data(event);
3058 memcpy(&entry->caller, fstack->calls, size);
3059 entry->size = nr_entries;
3061 if (!call_filter_check_discard(call, entry, buffer, event))
3062 __buffer_unlock_commit(buffer, event);
3065 /* Again, don't let gcc optimize things here */
3067 __this_cpu_dec(ftrace_stack_reserve);
3068 preempt_enable_notrace();
3072 static inline void ftrace_trace_stack(struct trace_array *tr,
3073 struct trace_buffer *buffer,
3074 unsigned int trace_ctx,
3075 int skip, struct pt_regs *regs)
3077 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3080 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3083 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3086 struct trace_buffer *buffer = tr->array_buffer.buffer;
3088 if (rcu_is_watching()) {
3089 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3094 * When an NMI triggers, RCU is enabled via rcu_nmi_enter(),
3095 * but if the above rcu_is_watching() failed, then the NMI
3096 * triggered someplace critical, and rcu_irq_enter() should
3097 * not be called from NMI.
3099 if (unlikely(in_nmi()))
3102 rcu_irq_enter_irqson();
3103 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3104 rcu_irq_exit_irqson();
3108 * trace_dump_stack - record a stack back trace in the trace buffer
3109 * @skip: Number of functions to skip (helper handlers)
3111 void trace_dump_stack(int skip)
3113 if (tracing_disabled || tracing_selftest_running)
3116 #ifndef CONFIG_UNWINDER_ORC
3117 /* Skip 1 to skip this function. */
3120 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3121 tracing_gen_ctx(), skip, NULL);
3123 EXPORT_SYMBOL_GPL(trace_dump_stack);
3125 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3126 static DEFINE_PER_CPU(int, user_stack_count);
3129 ftrace_trace_userstack(struct trace_array *tr,
3130 struct trace_buffer *buffer, unsigned int trace_ctx)
3132 struct trace_event_call *call = &event_user_stack;
3133 struct ring_buffer_event *event;
3134 struct userstack_entry *entry;
3136 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3140 * NMIs can not handle page faults, even with fix ups.
3141 * The save user stack can (and often does) fault.
3143 if (unlikely(in_nmi()))
3147 * prevent recursion, since the user stack tracing may
3148 * trigger other kernel events.
3151 if (__this_cpu_read(user_stack_count))
3154 __this_cpu_inc(user_stack_count);
3156 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3157 sizeof(*entry), trace_ctx);
3159 goto out_drop_count;
3160 entry = ring_buffer_event_data(event);
3162 entry->tgid = current->tgid;
3163 memset(&entry->caller, 0, sizeof(entry->caller));
3165 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3166 if (!call_filter_check_discard(call, entry, buffer, event))
3167 __buffer_unlock_commit(buffer, event);
3170 __this_cpu_dec(user_stack_count);
3174 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3175 static void ftrace_trace_userstack(struct trace_array *tr,
3176 struct trace_buffer *buffer,
3177 unsigned int trace_ctx)
3180 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3182 #endif /* CONFIG_STACKTRACE */
3185 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3186 unsigned long long delta)
3188 entry->bottom_delta_ts = delta & U32_MAX;
3189 entry->top_delta_ts = (delta >> 32);
3192 void trace_last_func_repeats(struct trace_array *tr,
3193 struct trace_func_repeats *last_info,
3194 unsigned int trace_ctx)
3196 struct trace_buffer *buffer = tr->array_buffer.buffer;
3197 struct func_repeats_entry *entry;
3198 struct ring_buffer_event *event;
3201 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3202 sizeof(*entry), trace_ctx);
3206 delta = ring_buffer_event_time_stamp(buffer, event) -
3207 last_info->ts_last_call;
3209 entry = ring_buffer_event_data(event);
3210 entry->ip = last_info->ip;
3211 entry->parent_ip = last_info->parent_ip;
3212 entry->count = last_info->count;
3213 func_repeats_set_delta_ts(entry, delta);
3215 __buffer_unlock_commit(buffer, event);
3218 /* created for use with alloc_percpu */
3219 struct trace_buffer_struct {
3221 char buffer[4][TRACE_BUF_SIZE];
3224 static struct trace_buffer_struct *trace_percpu_buffer;
3227 * This allows for lockless recording. If we're nested too deeply, then
3228 * this returns NULL.
3230 static char *get_trace_buf(void)
3232 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3234 if (!buffer || buffer->nesting >= 4)
3239 /* Interrupts must see nesting incremented before we use the buffer */
3241 return &buffer->buffer[buffer->nesting - 1][0];
3244 static void put_trace_buf(void)
3246 /* Don't let the decrement of nesting leak before this */
3248 this_cpu_dec(trace_percpu_buffer->nesting);
3251 static int alloc_percpu_trace_buffer(void)
3253 struct trace_buffer_struct *buffers;
3255 if (trace_percpu_buffer)
3258 buffers = alloc_percpu(struct trace_buffer_struct);
3259 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3262 trace_percpu_buffer = buffers;
3266 static int buffers_allocated;
3268 void trace_printk_init_buffers(void)
3270 if (buffers_allocated)
3273 if (alloc_percpu_trace_buffer())
3276 /* trace_printk() is for debug use only. Don't use it in production. */
3279 pr_warn("**********************************************************\n");
3280 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3282 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3284 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3285 pr_warn("** unsafe for production use. **\n");
3287 pr_warn("** If you see this message and you are not debugging **\n");
3288 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3290 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3291 pr_warn("**********************************************************\n");
3293 /* Expand the buffers to set size */
3294 tracing_update_buffers();
3296 buffers_allocated = 1;
3299 * trace_printk_init_buffers() can be called by modules.
3300 * If that happens, then we need to start cmdline recording
3301 * directly here. If the global_trace.buffer is already
3302 * allocated here, then this was called by module code.
3304 if (global_trace.array_buffer.buffer)
3305 tracing_start_cmdline_record();
3307 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3309 void trace_printk_start_comm(void)
3311 /* Start tracing comms if trace printk is set */
3312 if (!buffers_allocated)
3314 tracing_start_cmdline_record();
3317 static void trace_printk_start_stop_comm(int enabled)
3319 if (!buffers_allocated)
3323 tracing_start_cmdline_record();
3325 tracing_stop_cmdline_record();
3329 * trace_vbprintk - write binary msg to tracing buffer
3330 * @ip: The address of the caller
3331 * @fmt: The string format to write to the buffer
3332 * @args: Arguments for @fmt
3334 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3336 struct trace_event_call *call = &event_bprint;
3337 struct ring_buffer_event *event;
3338 struct trace_buffer *buffer;
3339 struct trace_array *tr = &global_trace;
3340 struct bprint_entry *entry;
3341 unsigned int trace_ctx;
3345 if (unlikely(tracing_selftest_running || tracing_disabled))
3348 /* Don't pollute graph traces with trace_vprintk internals */
3349 pause_graph_tracing();
3351 trace_ctx = tracing_gen_ctx();
3352 preempt_disable_notrace();
3354 tbuffer = get_trace_buf();
3360 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3362 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3365 size = sizeof(*entry) + sizeof(u32) * len;
3366 buffer = tr->array_buffer.buffer;
3367 ring_buffer_nest_start(buffer);
3368 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3372 entry = ring_buffer_event_data(event);
3376 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3377 if (!call_filter_check_discard(call, entry, buffer, event)) {
3378 __buffer_unlock_commit(buffer, event);
3379 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3383 ring_buffer_nest_end(buffer);
3388 preempt_enable_notrace();
3389 unpause_graph_tracing();
3393 EXPORT_SYMBOL_GPL(trace_vbprintk);
3397 __trace_array_vprintk(struct trace_buffer *buffer,
3398 unsigned long ip, const char *fmt, va_list args)
3400 struct trace_event_call *call = &event_print;
3401 struct ring_buffer_event *event;
3403 struct print_entry *entry;
3404 unsigned int trace_ctx;
3407 if (tracing_disabled || tracing_selftest_running)
3410 /* Don't pollute graph traces with trace_vprintk internals */
3411 pause_graph_tracing();
3413 trace_ctx = tracing_gen_ctx();
3414 preempt_disable_notrace();
3417 tbuffer = get_trace_buf();
3423 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3425 size = sizeof(*entry) + len + 1;
3426 ring_buffer_nest_start(buffer);
3427 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3431 entry = ring_buffer_event_data(event);
3434 memcpy(&entry->buf, tbuffer, len + 1);
3435 if (!call_filter_check_discard(call, entry, buffer, event)) {
3436 __buffer_unlock_commit(buffer, event);
3437 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3441 ring_buffer_nest_end(buffer);
3445 preempt_enable_notrace();
3446 unpause_graph_tracing();
3452 int trace_array_vprintk(struct trace_array *tr,
3453 unsigned long ip, const char *fmt, va_list args)
3455 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3459 * trace_array_printk - Print a message to a specific instance
3460 * @tr: The instance trace_array descriptor
3461 * @ip: The instruction pointer that this is called from.
3462 * @fmt: The format to print (printf format)
3464 * If a subsystem sets up its own instance, they have the right to
3465 * printk strings into their tracing instance buffer using this
3466 * function. Note, this function will not write into the top level
3467 * buffer (use trace_printk() for that), as writing into the top level
3468 * buffer should only have events that can be individually disabled.
3469 * trace_printk() is only used for debugging a kernel, and should not
3470 * be ever incorporated in normal use.
3472 * trace_array_printk() can be used, as it will not add noise to the
3473 * top level tracing buffer.
3475 * Note, trace_array_init_printk() must be called on @tr before this
3479 int trace_array_printk(struct trace_array *tr,
3480 unsigned long ip, const char *fmt, ...)
3488 /* This is only allowed for created instances */
3489 if (tr == &global_trace)
3492 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3496 ret = trace_array_vprintk(tr, ip, fmt, ap);
3500 EXPORT_SYMBOL_GPL(trace_array_printk);
3503 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3504 * @tr: The trace array to initialize the buffers for
3506 * As trace_array_printk() only writes into instances, they are OK to
3507 * have in the kernel (unlike trace_printk()). This needs to be called
3508 * before trace_array_printk() can be used on a trace_array.
3510 int trace_array_init_printk(struct trace_array *tr)
3515 /* This is only allowed for created instances */
3516 if (tr == &global_trace)
3519 return alloc_percpu_trace_buffer();
3521 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3524 int trace_array_printk_buf(struct trace_buffer *buffer,
3525 unsigned long ip, const char *fmt, ...)
3530 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3534 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3540 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3542 return trace_array_vprintk(&global_trace, ip, fmt, args);
3544 EXPORT_SYMBOL_GPL(trace_vprintk);
3546 static void trace_iterator_increment(struct trace_iterator *iter)
3548 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3552 ring_buffer_iter_advance(buf_iter);
3555 static struct trace_entry *
3556 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3557 unsigned long *lost_events)
3559 struct ring_buffer_event *event;
3560 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3563 event = ring_buffer_iter_peek(buf_iter, ts);
3565 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3566 (unsigned long)-1 : 0;
3568 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3573 iter->ent_size = ring_buffer_event_length(event);
3574 return ring_buffer_event_data(event);
3580 static struct trace_entry *
3581 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3582 unsigned long *missing_events, u64 *ent_ts)
3584 struct trace_buffer *buffer = iter->array_buffer->buffer;
3585 struct trace_entry *ent, *next = NULL;
3586 unsigned long lost_events = 0, next_lost = 0;
3587 int cpu_file = iter->cpu_file;
3588 u64 next_ts = 0, ts;
3594 * If we are in a per_cpu trace file, don't bother by iterating over
3595 * all cpu and peek directly.
3597 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3598 if (ring_buffer_empty_cpu(buffer, cpu_file))
3600 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3602 *ent_cpu = cpu_file;
3607 for_each_tracing_cpu(cpu) {
3609 if (ring_buffer_empty_cpu(buffer, cpu))
3612 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3615 * Pick the entry with the smallest timestamp:
3617 if (ent && (!next || ts < next_ts)) {
3621 next_lost = lost_events;
3622 next_size = iter->ent_size;
3626 iter->ent_size = next_size;
3629 *ent_cpu = next_cpu;
3635 *missing_events = next_lost;
3640 #define STATIC_FMT_BUF_SIZE 128
3641 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3643 static char *trace_iter_expand_format(struct trace_iterator *iter)
3648 * iter->tr is NULL when used with tp_printk, which makes
3649 * this get called where it is not safe to call krealloc().
3651 if (!iter->tr || iter->fmt == static_fmt_buf)
3654 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3657 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3664 /* Returns true if the string is safe to dereference from an event */
3665 static bool trace_safe_str(struct trace_iterator *iter, const char *str)
3667 unsigned long addr = (unsigned long)str;
3668 struct trace_event *trace_event;
3669 struct trace_event_call *event;
3671 /* OK if part of the event data */
3672 if ((addr >= (unsigned long)iter->ent) &&
3673 (addr < (unsigned long)iter->ent + iter->ent_size))
3676 /* OK if part of the temp seq buffer */
3677 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3678 (addr < (unsigned long)iter->tmp_seq.buffer + PAGE_SIZE))
3681 /* Core rodata can not be freed */
3682 if (is_kernel_rodata(addr))
3685 if (trace_is_tracepoint_string(str))
3689 * Now this could be a module event, referencing core module
3690 * data, which is OK.
3695 trace_event = ftrace_find_event(iter->ent->type);
3699 event = container_of(trace_event, struct trace_event_call, event);
3700 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3703 /* Would rather have rodata, but this will suffice */
3704 if (within_module_core(addr, event->module))
3710 static const char *show_buffer(struct trace_seq *s)
3712 struct seq_buf *seq = &s->seq;
3714 seq_buf_terminate(seq);
3719 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3721 static int test_can_verify_check(const char *fmt, ...)
3728 * The verifier is dependent on vsnprintf() modifies the va_list
3729 * passed to it, where it is sent as a reference. Some architectures
3730 * (like x86_32) passes it by value, which means that vsnprintf()
3731 * does not modify the va_list passed to it, and the verifier
3732 * would then need to be able to understand all the values that
3733 * vsnprintf can use. If it is passed by value, then the verifier
3737 vsnprintf(buf, 16, "%d", ap);
3738 ret = va_arg(ap, int);
3744 static void test_can_verify(void)
3746 if (!test_can_verify_check("%d %d", 0, 1)) {
3747 pr_info("trace event string verifier disabled\n");
3748 static_branch_inc(&trace_no_verify);
3753 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3754 * @iter: The iterator that holds the seq buffer and the event being printed
3755 * @fmt: The format used to print the event
3756 * @ap: The va_list holding the data to print from @fmt.
3758 * This writes the data into the @iter->seq buffer using the data from
3759 * @fmt and @ap. If the format has a %s, then the source of the string
3760 * is examined to make sure it is safe to print, otherwise it will
3761 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3764 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3767 const char *p = fmt;
3771 if (WARN_ON_ONCE(!fmt))
3774 if (static_branch_unlikely(&trace_no_verify))
3777 /* Don't bother checking when doing a ftrace_dump() */
3778 if (iter->fmt == static_fmt_buf)
3787 /* We only care about %s and variants */
3788 for (i = 0; p[i]; i++) {
3789 if (i + 1 >= iter->fmt_size) {
3791 * If we can't expand the copy buffer,
3794 if (!trace_iter_expand_format(iter))
3798 if (p[i] == '\\' && p[i+1]) {
3803 /* Need to test cases like %08.*s */
3804 for (j = 1; p[i+j]; j++) {
3805 if (isdigit(p[i+j]) ||
3808 if (p[i+j] == '*') {
3820 /* If no %s found then just print normally */
3824 /* Copy up to the %s, and print that */
3825 strncpy(iter->fmt, p, i);
3826 iter->fmt[i] = '\0';
3827 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3830 len = va_arg(ap, int);
3832 /* The ap now points to the string data of the %s */
3833 str = va_arg(ap, const char *);
3836 * If you hit this warning, it is likely that the
3837 * trace event in question used %s on a string that
3838 * was saved at the time of the event, but may not be
3839 * around when the trace is read. Use __string(),
3840 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3841 * instead. See samples/trace_events/trace-events-sample.h
3844 if (WARN_ONCE(!trace_safe_str(iter, str),
3845 "fmt: '%s' current_buffer: '%s'",
3846 fmt, show_buffer(&iter->seq))) {
3849 /* Try to safely read the string */
3851 if (len + 1 > iter->fmt_size)
3852 len = iter->fmt_size - 1;
3855 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3859 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3863 trace_seq_printf(&iter->seq, "(0x%px)", str);
3865 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3867 str = "[UNSAFE-MEMORY]";
3868 strcpy(iter->fmt, "%s");
3870 strncpy(iter->fmt, p + i, j + 1);
3871 iter->fmt[j+1] = '\0';
3874 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3876 trace_seq_printf(&iter->seq, iter->fmt, str);
3882 trace_seq_vprintf(&iter->seq, p, ap);
3885 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3887 const char *p, *new_fmt;
3890 if (WARN_ON_ONCE(!fmt))
3893 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
3897 new_fmt = q = iter->fmt;
3899 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
3900 if (!trace_iter_expand_format(iter))
3903 q += iter->fmt - new_fmt;
3904 new_fmt = iter->fmt;
3909 /* Replace %p with %px */
3913 } else if (p[0] == 'p' && !isalnum(p[1])) {
3924 #define STATIC_TEMP_BUF_SIZE 128
3925 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3927 /* Find the next real entry, without updating the iterator itself */
3928 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3929 int *ent_cpu, u64 *ent_ts)
3931 /* __find_next_entry will reset ent_size */
3932 int ent_size = iter->ent_size;
3933 struct trace_entry *entry;
3936 * If called from ftrace_dump(), then the iter->temp buffer
3937 * will be the static_temp_buf and not created from kmalloc.
3938 * If the entry size is greater than the buffer, we can
3939 * not save it. Just return NULL in that case. This is only
3940 * used to add markers when two consecutive events' time
3941 * stamps have a large delta. See trace_print_lat_context()
3943 if (iter->temp == static_temp_buf &&
3944 STATIC_TEMP_BUF_SIZE < ent_size)
3948 * The __find_next_entry() may call peek_next_entry(), which may
3949 * call ring_buffer_peek() that may make the contents of iter->ent
3950 * undefined. Need to copy iter->ent now.
3952 if (iter->ent && iter->ent != iter->temp) {
3953 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
3954 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
3956 temp = kmalloc(iter->ent_size, GFP_KERNEL);
3961 iter->temp_size = iter->ent_size;
3963 memcpy(iter->temp, iter->ent, iter->ent_size);
3964 iter->ent = iter->temp;
3966 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
3967 /* Put back the original ent_size */
3968 iter->ent_size = ent_size;
3973 /* Find the next real entry, and increment the iterator to the next entry */
3974 void *trace_find_next_entry_inc(struct trace_iterator *iter)
3976 iter->ent = __find_next_entry(iter, &iter->cpu,
3977 &iter->lost_events, &iter->ts);
3980 trace_iterator_increment(iter);
3982 return iter->ent ? iter : NULL;
3985 static void trace_consume(struct trace_iterator *iter)
3987 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
3988 &iter->lost_events);
3991 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
3993 struct trace_iterator *iter = m->private;
3997 WARN_ON_ONCE(iter->leftover);
4001 /* can't go backwards */
4006 ent = trace_find_next_entry_inc(iter);
4010 while (ent && iter->idx < i)
4011 ent = trace_find_next_entry_inc(iter);
4018 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4020 struct ring_buffer_iter *buf_iter;
4021 unsigned long entries = 0;
4024 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4026 buf_iter = trace_buffer_iter(iter, cpu);
4030 ring_buffer_iter_reset(buf_iter);
4033 * We could have the case with the max latency tracers
4034 * that a reset never took place on a cpu. This is evident
4035 * by the timestamp being before the start of the buffer.
4037 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4038 if (ts >= iter->array_buffer->time_start)
4041 ring_buffer_iter_advance(buf_iter);
4044 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4048 * The current tracer is copied to avoid a global locking
4051 static void *s_start(struct seq_file *m, loff_t *pos)
4053 struct trace_iterator *iter = m->private;
4054 struct trace_array *tr = iter->tr;
4055 int cpu_file = iter->cpu_file;
4061 * copy the tracer to avoid using a global lock all around.
4062 * iter->trace is a copy of current_trace, the pointer to the
4063 * name may be used instead of a strcmp(), as iter->trace->name
4064 * will point to the same string as current_trace->name.
4066 mutex_lock(&trace_types_lock);
4067 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
4068 *iter->trace = *tr->current_trace;
4069 mutex_unlock(&trace_types_lock);
4071 #ifdef CONFIG_TRACER_MAX_TRACE
4072 if (iter->snapshot && iter->trace->use_max_tr)
4073 return ERR_PTR(-EBUSY);
4076 if (*pos != iter->pos) {
4081 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4082 for_each_tracing_cpu(cpu)
4083 tracing_iter_reset(iter, cpu);
4085 tracing_iter_reset(iter, cpu_file);
4088 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4093 * If we overflowed the seq_file before, then we want
4094 * to just reuse the trace_seq buffer again.
4100 p = s_next(m, p, &l);
4104 trace_event_read_lock();
4105 trace_access_lock(cpu_file);
4109 static void s_stop(struct seq_file *m, void *p)
4111 struct trace_iterator *iter = m->private;
4113 #ifdef CONFIG_TRACER_MAX_TRACE
4114 if (iter->snapshot && iter->trace->use_max_tr)
4118 trace_access_unlock(iter->cpu_file);
4119 trace_event_read_unlock();
4123 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4124 unsigned long *entries, int cpu)
4126 unsigned long count;
4128 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4130 * If this buffer has skipped entries, then we hold all
4131 * entries for the trace and we need to ignore the
4132 * ones before the time stamp.
4134 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4135 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4136 /* total is the same as the entries */
4140 ring_buffer_overrun_cpu(buf->buffer, cpu);
4145 get_total_entries(struct array_buffer *buf,
4146 unsigned long *total, unsigned long *entries)
4154 for_each_tracing_cpu(cpu) {
4155 get_total_entries_cpu(buf, &t, &e, cpu);
4161 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4163 unsigned long total, entries;
4168 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4173 unsigned long trace_total_entries(struct trace_array *tr)
4175 unsigned long total, entries;
4180 get_total_entries(&tr->array_buffer, &total, &entries);
4185 static void print_lat_help_header(struct seq_file *m)
4187 seq_puts(m, "# _------=> CPU# \n"
4188 "# / _-----=> irqs-off \n"
4189 "# | / _----=> need-resched \n"
4190 "# || / _---=> hardirq/softirq \n"
4191 "# ||| / _--=> preempt-depth \n"
4193 "# cmd pid ||||| time | caller \n"
4194 "# \\ / ||||| \\ | / \n");
4197 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4199 unsigned long total;
4200 unsigned long entries;
4202 get_total_entries(buf, &total, &entries);
4203 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4204 entries, total, num_online_cpus());
4208 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4211 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4213 print_event_info(buf, m);
4215 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4216 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4219 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4222 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4223 const char *space = " ";
4224 int prec = tgid ? 12 : 2;
4226 print_event_info(buf, m);
4228 seq_printf(m, "# %.*s _-----=> irqs-off\n", prec, space);
4229 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4230 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4231 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4232 seq_printf(m, "# %.*s||| / delay\n", prec, space);
4233 seq_printf(m, "# TASK-PID %.*s CPU# |||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4234 seq_printf(m, "# | | %.*s | |||| | |\n", prec, " | ");
4238 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4240 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4241 struct array_buffer *buf = iter->array_buffer;
4242 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4243 struct tracer *type = iter->trace;
4244 unsigned long entries;
4245 unsigned long total;
4246 const char *name = "preemption";
4250 get_total_entries(buf, &total, &entries);
4252 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4254 seq_puts(m, "# -----------------------------------"
4255 "---------------------------------\n");
4256 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4257 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4258 nsecs_to_usecs(data->saved_latency),
4262 #if defined(CONFIG_PREEMPT_NONE)
4264 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
4266 #elif defined(CONFIG_PREEMPT)
4268 #elif defined(CONFIG_PREEMPT_RT)
4273 /* These are reserved for later use */
4276 seq_printf(m, " #P:%d)\n", num_online_cpus());
4280 seq_puts(m, "# -----------------\n");
4281 seq_printf(m, "# | task: %.16s-%d "
4282 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4283 data->comm, data->pid,
4284 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4285 data->policy, data->rt_priority);
4286 seq_puts(m, "# -----------------\n");
4288 if (data->critical_start) {
4289 seq_puts(m, "# => started at: ");
4290 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4291 trace_print_seq(m, &iter->seq);
4292 seq_puts(m, "\n# => ended at: ");
4293 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4294 trace_print_seq(m, &iter->seq);
4295 seq_puts(m, "\n#\n");
4301 static void test_cpu_buff_start(struct trace_iterator *iter)
4303 struct trace_seq *s = &iter->seq;
4304 struct trace_array *tr = iter->tr;
4306 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4309 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4312 if (cpumask_available(iter->started) &&
4313 cpumask_test_cpu(iter->cpu, iter->started))
4316 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4319 if (cpumask_available(iter->started))
4320 cpumask_set_cpu(iter->cpu, iter->started);
4322 /* Don't print started cpu buffer for the first entry of the trace */
4324 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4328 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4330 struct trace_array *tr = iter->tr;
4331 struct trace_seq *s = &iter->seq;
4332 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4333 struct trace_entry *entry;
4334 struct trace_event *event;
4338 test_cpu_buff_start(iter);
4340 event = ftrace_find_event(entry->type);
4342 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4343 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4344 trace_print_lat_context(iter);
4346 trace_print_context(iter);
4349 if (trace_seq_has_overflowed(s))
4350 return TRACE_TYPE_PARTIAL_LINE;
4353 return event->funcs->trace(iter, sym_flags, event);
4355 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4357 return trace_handle_return(s);
4360 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4362 struct trace_array *tr = iter->tr;
4363 struct trace_seq *s = &iter->seq;
4364 struct trace_entry *entry;
4365 struct trace_event *event;
4369 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4370 trace_seq_printf(s, "%d %d %llu ",
4371 entry->pid, iter->cpu, iter->ts);
4373 if (trace_seq_has_overflowed(s))
4374 return TRACE_TYPE_PARTIAL_LINE;
4376 event = ftrace_find_event(entry->type);
4378 return event->funcs->raw(iter, 0, event);
4380 trace_seq_printf(s, "%d ?\n", entry->type);
4382 return trace_handle_return(s);
4385 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4387 struct trace_array *tr = iter->tr;
4388 struct trace_seq *s = &iter->seq;
4389 unsigned char newline = '\n';
4390 struct trace_entry *entry;
4391 struct trace_event *event;
4395 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4396 SEQ_PUT_HEX_FIELD(s, entry->pid);
4397 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4398 SEQ_PUT_HEX_FIELD(s, iter->ts);
4399 if (trace_seq_has_overflowed(s))
4400 return TRACE_TYPE_PARTIAL_LINE;
4403 event = ftrace_find_event(entry->type);
4405 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4406 if (ret != TRACE_TYPE_HANDLED)
4410 SEQ_PUT_FIELD(s, newline);
4412 return trace_handle_return(s);
4415 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4417 struct trace_array *tr = iter->tr;
4418 struct trace_seq *s = &iter->seq;
4419 struct trace_entry *entry;
4420 struct trace_event *event;
4424 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4425 SEQ_PUT_FIELD(s, entry->pid);
4426 SEQ_PUT_FIELD(s, iter->cpu);
4427 SEQ_PUT_FIELD(s, iter->ts);
4428 if (trace_seq_has_overflowed(s))
4429 return TRACE_TYPE_PARTIAL_LINE;
4432 event = ftrace_find_event(entry->type);
4433 return event ? event->funcs->binary(iter, 0, event) :
4437 int trace_empty(struct trace_iterator *iter)
4439 struct ring_buffer_iter *buf_iter;
4442 /* If we are looking at one CPU buffer, only check that one */
4443 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4444 cpu = iter->cpu_file;
4445 buf_iter = trace_buffer_iter(iter, cpu);
4447 if (!ring_buffer_iter_empty(buf_iter))
4450 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4456 for_each_tracing_cpu(cpu) {
4457 buf_iter = trace_buffer_iter(iter, cpu);
4459 if (!ring_buffer_iter_empty(buf_iter))
4462 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4470 /* Called with trace_event_read_lock() held. */
4471 enum print_line_t print_trace_line(struct trace_iterator *iter)
4473 struct trace_array *tr = iter->tr;
4474 unsigned long trace_flags = tr->trace_flags;
4475 enum print_line_t ret;
4477 if (iter->lost_events) {
4478 if (iter->lost_events == (unsigned long)-1)
4479 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4482 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4483 iter->cpu, iter->lost_events);
4484 if (trace_seq_has_overflowed(&iter->seq))
4485 return TRACE_TYPE_PARTIAL_LINE;
4488 if (iter->trace && iter->trace->print_line) {
4489 ret = iter->trace->print_line(iter);
4490 if (ret != TRACE_TYPE_UNHANDLED)
4494 if (iter->ent->type == TRACE_BPUTS &&
4495 trace_flags & TRACE_ITER_PRINTK &&
4496 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4497 return trace_print_bputs_msg_only(iter);
4499 if (iter->ent->type == TRACE_BPRINT &&
4500 trace_flags & TRACE_ITER_PRINTK &&
4501 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4502 return trace_print_bprintk_msg_only(iter);
4504 if (iter->ent->type == TRACE_PRINT &&
4505 trace_flags & TRACE_ITER_PRINTK &&
4506 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4507 return trace_print_printk_msg_only(iter);
4509 if (trace_flags & TRACE_ITER_BIN)
4510 return print_bin_fmt(iter);
4512 if (trace_flags & TRACE_ITER_HEX)
4513 return print_hex_fmt(iter);
4515 if (trace_flags & TRACE_ITER_RAW)
4516 return print_raw_fmt(iter);
4518 return print_trace_fmt(iter);
4521 void trace_latency_header(struct seq_file *m)
4523 struct trace_iterator *iter = m->private;
4524 struct trace_array *tr = iter->tr;
4526 /* print nothing if the buffers are empty */
4527 if (trace_empty(iter))
4530 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4531 print_trace_header(m, iter);
4533 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4534 print_lat_help_header(m);
4537 void trace_default_header(struct seq_file *m)
4539 struct trace_iterator *iter = m->private;
4540 struct trace_array *tr = iter->tr;
4541 unsigned long trace_flags = tr->trace_flags;
4543 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4546 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4547 /* print nothing if the buffers are empty */
4548 if (trace_empty(iter))
4550 print_trace_header(m, iter);
4551 if (!(trace_flags & TRACE_ITER_VERBOSE))
4552 print_lat_help_header(m);
4554 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4555 if (trace_flags & TRACE_ITER_IRQ_INFO)
4556 print_func_help_header_irq(iter->array_buffer,
4559 print_func_help_header(iter->array_buffer, m,
4565 static void test_ftrace_alive(struct seq_file *m)
4567 if (!ftrace_is_dead())
4569 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4570 "# MAY BE MISSING FUNCTION EVENTS\n");
4573 #ifdef CONFIG_TRACER_MAX_TRACE
4574 static void show_snapshot_main_help(struct seq_file *m)
4576 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4577 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4578 "# Takes a snapshot of the main buffer.\n"
4579 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4580 "# (Doesn't have to be '2' works with any number that\n"
4581 "# is not a '0' or '1')\n");
4584 static void show_snapshot_percpu_help(struct seq_file *m)
4586 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4587 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4588 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4589 "# Takes a snapshot of the main buffer for this cpu.\n");
4591 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4592 "# Must use main snapshot file to allocate.\n");
4594 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4595 "# (Doesn't have to be '2' works with any number that\n"
4596 "# is not a '0' or '1')\n");
4599 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4601 if (iter->tr->allocated_snapshot)
4602 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4604 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4606 seq_puts(m, "# Snapshot commands:\n");
4607 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4608 show_snapshot_main_help(m);
4610 show_snapshot_percpu_help(m);
4613 /* Should never be called */
4614 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4617 static int s_show(struct seq_file *m, void *v)
4619 struct trace_iterator *iter = v;
4622 if (iter->ent == NULL) {
4624 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4626 test_ftrace_alive(m);
4628 if (iter->snapshot && trace_empty(iter))
4629 print_snapshot_help(m, iter);
4630 else if (iter->trace && iter->trace->print_header)
4631 iter->trace->print_header(m);
4633 trace_default_header(m);
4635 } else if (iter->leftover) {
4637 * If we filled the seq_file buffer earlier, we
4638 * want to just show it now.
4640 ret = trace_print_seq(m, &iter->seq);
4642 /* ret should this time be zero, but you never know */
4643 iter->leftover = ret;
4646 print_trace_line(iter);
4647 ret = trace_print_seq(m, &iter->seq);
4649 * If we overflow the seq_file buffer, then it will
4650 * ask us for this data again at start up.
4652 * ret is 0 if seq_file write succeeded.
4655 iter->leftover = ret;
4662 * Should be used after trace_array_get(), trace_types_lock
4663 * ensures that i_cdev was already initialized.
4665 static inline int tracing_get_cpu(struct inode *inode)
4667 if (inode->i_cdev) /* See trace_create_cpu_file() */
4668 return (long)inode->i_cdev - 1;
4669 return RING_BUFFER_ALL_CPUS;
4672 static const struct seq_operations tracer_seq_ops = {
4679 static struct trace_iterator *
4680 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4682 struct trace_array *tr = inode->i_private;
4683 struct trace_iterator *iter;
4686 if (tracing_disabled)
4687 return ERR_PTR(-ENODEV);
4689 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4691 return ERR_PTR(-ENOMEM);
4693 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4695 if (!iter->buffer_iter)
4699 * trace_find_next_entry() may need to save off iter->ent.
4700 * It will place it into the iter->temp buffer. As most
4701 * events are less than 128, allocate a buffer of that size.
4702 * If one is greater, then trace_find_next_entry() will
4703 * allocate a new buffer to adjust for the bigger iter->ent.
4704 * It's not critical if it fails to get allocated here.
4706 iter->temp = kmalloc(128, GFP_KERNEL);
4708 iter->temp_size = 128;
4711 * trace_event_printf() may need to modify given format
4712 * string to replace %p with %px so that it shows real address
4713 * instead of hash value. However, that is only for the event
4714 * tracing, other tracer may not need. Defer the allocation
4715 * until it is needed.
4721 * We make a copy of the current tracer to avoid concurrent
4722 * changes on it while we are reading.
4724 mutex_lock(&trace_types_lock);
4725 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4729 *iter->trace = *tr->current_trace;
4731 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4736 #ifdef CONFIG_TRACER_MAX_TRACE
4737 /* Currently only the top directory has a snapshot */
4738 if (tr->current_trace->print_max || snapshot)
4739 iter->array_buffer = &tr->max_buffer;
4742 iter->array_buffer = &tr->array_buffer;
4743 iter->snapshot = snapshot;
4745 iter->cpu_file = tracing_get_cpu(inode);
4746 mutex_init(&iter->mutex);
4748 /* Notify the tracer early; before we stop tracing. */
4749 if (iter->trace->open)
4750 iter->trace->open(iter);
4752 /* Annotate start of buffers if we had overruns */
4753 if (ring_buffer_overruns(iter->array_buffer->buffer))
4754 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4756 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4757 if (trace_clocks[tr->clock_id].in_ns)
4758 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4761 * If pause-on-trace is enabled, then stop the trace while
4762 * dumping, unless this is the "snapshot" file
4764 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4765 tracing_stop_tr(tr);
4767 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4768 for_each_tracing_cpu(cpu) {
4769 iter->buffer_iter[cpu] =
4770 ring_buffer_read_prepare(iter->array_buffer->buffer,
4773 ring_buffer_read_prepare_sync();
4774 for_each_tracing_cpu(cpu) {
4775 ring_buffer_read_start(iter->buffer_iter[cpu]);
4776 tracing_iter_reset(iter, cpu);
4779 cpu = iter->cpu_file;
4780 iter->buffer_iter[cpu] =
4781 ring_buffer_read_prepare(iter->array_buffer->buffer,
4783 ring_buffer_read_prepare_sync();
4784 ring_buffer_read_start(iter->buffer_iter[cpu]);
4785 tracing_iter_reset(iter, cpu);
4788 mutex_unlock(&trace_types_lock);
4793 mutex_unlock(&trace_types_lock);
4796 kfree(iter->buffer_iter);
4798 seq_release_private(inode, file);
4799 return ERR_PTR(-ENOMEM);
4802 int tracing_open_generic(struct inode *inode, struct file *filp)
4806 ret = tracing_check_open_get_tr(NULL);
4810 filp->private_data = inode->i_private;
4814 bool tracing_is_disabled(void)
4816 return (tracing_disabled) ? true: false;
4820 * Open and update trace_array ref count.
4821 * Must have the current trace_array passed to it.
4823 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4825 struct trace_array *tr = inode->i_private;
4828 ret = tracing_check_open_get_tr(tr);
4832 filp->private_data = inode->i_private;
4837 static int tracing_release(struct inode *inode, struct file *file)
4839 struct trace_array *tr = inode->i_private;
4840 struct seq_file *m = file->private_data;
4841 struct trace_iterator *iter;
4844 if (!(file->f_mode & FMODE_READ)) {
4845 trace_array_put(tr);
4849 /* Writes do not use seq_file */
4851 mutex_lock(&trace_types_lock);
4853 for_each_tracing_cpu(cpu) {
4854 if (iter->buffer_iter[cpu])
4855 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4858 if (iter->trace && iter->trace->close)
4859 iter->trace->close(iter);
4861 if (!iter->snapshot && tr->stop_count)
4862 /* reenable tracing if it was previously enabled */
4863 tracing_start_tr(tr);
4865 __trace_array_put(tr);
4867 mutex_unlock(&trace_types_lock);
4869 mutex_destroy(&iter->mutex);
4870 free_cpumask_var(iter->started);
4874 kfree(iter->buffer_iter);
4875 seq_release_private(inode, file);
4880 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4882 struct trace_array *tr = inode->i_private;
4884 trace_array_put(tr);
4888 static int tracing_single_release_tr(struct inode *inode, struct file *file)
4890 struct trace_array *tr = inode->i_private;
4892 trace_array_put(tr);
4894 return single_release(inode, file);
4897 static int tracing_open(struct inode *inode, struct file *file)
4899 struct trace_array *tr = inode->i_private;
4900 struct trace_iterator *iter;
4903 ret = tracing_check_open_get_tr(tr);
4907 /* If this file was open for write, then erase contents */
4908 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4909 int cpu = tracing_get_cpu(inode);
4910 struct array_buffer *trace_buf = &tr->array_buffer;
4912 #ifdef CONFIG_TRACER_MAX_TRACE
4913 if (tr->current_trace->print_max)
4914 trace_buf = &tr->max_buffer;
4917 if (cpu == RING_BUFFER_ALL_CPUS)
4918 tracing_reset_online_cpus(trace_buf);
4920 tracing_reset_cpu(trace_buf, cpu);
4923 if (file->f_mode & FMODE_READ) {
4924 iter = __tracing_open(inode, file, false);
4926 ret = PTR_ERR(iter);
4927 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4928 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4932 trace_array_put(tr);
4938 * Some tracers are not suitable for instance buffers.
4939 * A tracer is always available for the global array (toplevel)
4940 * or if it explicitly states that it is.
4943 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
4945 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
4948 /* Find the next tracer that this trace array may use */
4949 static struct tracer *
4950 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
4952 while (t && !trace_ok_for_array(t, tr))
4959 t_next(struct seq_file *m, void *v, loff_t *pos)
4961 struct trace_array *tr = m->private;
4962 struct tracer *t = v;
4967 t = get_tracer_for_array(tr, t->next);
4972 static void *t_start(struct seq_file *m, loff_t *pos)
4974 struct trace_array *tr = m->private;
4978 mutex_lock(&trace_types_lock);
4980 t = get_tracer_for_array(tr, trace_types);
4981 for (; t && l < *pos; t = t_next(m, t, &l))
4987 static void t_stop(struct seq_file *m, void *p)
4989 mutex_unlock(&trace_types_lock);
4992 static int t_show(struct seq_file *m, void *v)
4994 struct tracer *t = v;
4999 seq_puts(m, t->name);
5008 static const struct seq_operations show_traces_seq_ops = {
5015 static int show_traces_open(struct inode *inode, struct file *file)
5017 struct trace_array *tr = inode->i_private;
5021 ret = tracing_check_open_get_tr(tr);
5025 ret = seq_open(file, &show_traces_seq_ops);
5027 trace_array_put(tr);
5031 m = file->private_data;
5037 static int show_traces_release(struct inode *inode, struct file *file)
5039 struct trace_array *tr = inode->i_private;
5041 trace_array_put(tr);
5042 return seq_release(inode, file);
5046 tracing_write_stub(struct file *filp, const char __user *ubuf,
5047 size_t count, loff_t *ppos)
5052 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5056 if (file->f_mode & FMODE_READ)
5057 ret = seq_lseek(file, offset, whence);
5059 file->f_pos = ret = 0;
5064 static const struct file_operations tracing_fops = {
5065 .open = tracing_open,
5067 .write = tracing_write_stub,
5068 .llseek = tracing_lseek,
5069 .release = tracing_release,
5072 static const struct file_operations show_traces_fops = {
5073 .open = show_traces_open,
5075 .llseek = seq_lseek,
5076 .release = show_traces_release,
5080 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5081 size_t count, loff_t *ppos)
5083 struct trace_array *tr = file_inode(filp)->i_private;
5087 len = snprintf(NULL, 0, "%*pb\n",
5088 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5089 mask_str = kmalloc(len, GFP_KERNEL);
5093 len = snprintf(mask_str, len, "%*pb\n",
5094 cpumask_pr_args(tr->tracing_cpumask));
5099 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5107 int tracing_set_cpumask(struct trace_array *tr,
5108 cpumask_var_t tracing_cpumask_new)
5115 local_irq_disable();
5116 arch_spin_lock(&tr->max_lock);
5117 for_each_tracing_cpu(cpu) {
5119 * Increase/decrease the disabled counter if we are
5120 * about to flip a bit in the cpumask:
5122 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5123 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5124 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5125 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5127 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5128 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5129 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5130 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5133 arch_spin_unlock(&tr->max_lock);
5136 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5142 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5143 size_t count, loff_t *ppos)
5145 struct trace_array *tr = file_inode(filp)->i_private;
5146 cpumask_var_t tracing_cpumask_new;
5149 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5152 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5156 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5160 free_cpumask_var(tracing_cpumask_new);
5165 free_cpumask_var(tracing_cpumask_new);
5170 static const struct file_operations tracing_cpumask_fops = {
5171 .open = tracing_open_generic_tr,
5172 .read = tracing_cpumask_read,
5173 .write = tracing_cpumask_write,
5174 .release = tracing_release_generic_tr,
5175 .llseek = generic_file_llseek,
5178 static int tracing_trace_options_show(struct seq_file *m, void *v)
5180 struct tracer_opt *trace_opts;
5181 struct trace_array *tr = m->private;
5185 mutex_lock(&trace_types_lock);
5186 tracer_flags = tr->current_trace->flags->val;
5187 trace_opts = tr->current_trace->flags->opts;
5189 for (i = 0; trace_options[i]; i++) {
5190 if (tr->trace_flags & (1 << i))
5191 seq_printf(m, "%s\n", trace_options[i]);
5193 seq_printf(m, "no%s\n", trace_options[i]);
5196 for (i = 0; trace_opts[i].name; i++) {
5197 if (tracer_flags & trace_opts[i].bit)
5198 seq_printf(m, "%s\n", trace_opts[i].name);
5200 seq_printf(m, "no%s\n", trace_opts[i].name);
5202 mutex_unlock(&trace_types_lock);
5207 static int __set_tracer_option(struct trace_array *tr,
5208 struct tracer_flags *tracer_flags,
5209 struct tracer_opt *opts, int neg)
5211 struct tracer *trace = tracer_flags->trace;
5214 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5219 tracer_flags->val &= ~opts->bit;
5221 tracer_flags->val |= opts->bit;
5225 /* Try to assign a tracer specific option */
5226 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5228 struct tracer *trace = tr->current_trace;
5229 struct tracer_flags *tracer_flags = trace->flags;
5230 struct tracer_opt *opts = NULL;
5233 for (i = 0; tracer_flags->opts[i].name; i++) {
5234 opts = &tracer_flags->opts[i];
5236 if (strcmp(cmp, opts->name) == 0)
5237 return __set_tracer_option(tr, trace->flags, opts, neg);
5243 /* Some tracers require overwrite to stay enabled */
5244 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5246 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5252 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5256 if ((mask == TRACE_ITER_RECORD_TGID) ||
5257 (mask == TRACE_ITER_RECORD_CMD))
5258 lockdep_assert_held(&event_mutex);
5260 /* do nothing if flag is already set */
5261 if (!!(tr->trace_flags & mask) == !!enabled)
5264 /* Give the tracer a chance to approve the change */
5265 if (tr->current_trace->flag_changed)
5266 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5270 tr->trace_flags |= mask;
5272 tr->trace_flags &= ~mask;
5274 if (mask == TRACE_ITER_RECORD_CMD)
5275 trace_event_enable_cmd_record(enabled);
5277 if (mask == TRACE_ITER_RECORD_TGID) {
5279 tgid_map_max = pid_max;
5280 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5284 * Pairs with smp_load_acquire() in
5285 * trace_find_tgid_ptr() to ensure that if it observes
5286 * the tgid_map we just allocated then it also observes
5287 * the corresponding tgid_map_max value.
5289 smp_store_release(&tgid_map, map);
5292 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5296 trace_event_enable_tgid_record(enabled);
5299 if (mask == TRACE_ITER_EVENT_FORK)
5300 trace_event_follow_fork(tr, enabled);
5302 if (mask == TRACE_ITER_FUNC_FORK)
5303 ftrace_pid_follow_fork(tr, enabled);
5305 if (mask == TRACE_ITER_OVERWRITE) {
5306 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5307 #ifdef CONFIG_TRACER_MAX_TRACE
5308 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5312 if (mask == TRACE_ITER_PRINTK) {
5313 trace_printk_start_stop_comm(enabled);
5314 trace_printk_control(enabled);
5320 int trace_set_options(struct trace_array *tr, char *option)
5325 size_t orig_len = strlen(option);
5328 cmp = strstrip(option);
5330 len = str_has_prefix(cmp, "no");
5336 mutex_lock(&event_mutex);
5337 mutex_lock(&trace_types_lock);
5339 ret = match_string(trace_options, -1, cmp);
5340 /* If no option could be set, test the specific tracer options */
5342 ret = set_tracer_option(tr, cmp, neg);
5344 ret = set_tracer_flag(tr, 1 << ret, !neg);
5346 mutex_unlock(&trace_types_lock);
5347 mutex_unlock(&event_mutex);
5350 * If the first trailing whitespace is replaced with '\0' by strstrip,
5351 * turn it back into a space.
5353 if (orig_len > strlen(option))
5354 option[strlen(option)] = ' ';
5359 static void __init apply_trace_boot_options(void)
5361 char *buf = trace_boot_options_buf;
5365 option = strsep(&buf, ",");
5371 trace_set_options(&global_trace, option);
5373 /* Put back the comma to allow this to be called again */
5380 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5381 size_t cnt, loff_t *ppos)
5383 struct seq_file *m = filp->private_data;
5384 struct trace_array *tr = m->private;
5388 if (cnt >= sizeof(buf))
5391 if (copy_from_user(buf, ubuf, cnt))
5396 ret = trace_set_options(tr, buf);
5405 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5407 struct trace_array *tr = inode->i_private;
5410 ret = tracing_check_open_get_tr(tr);
5414 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5416 trace_array_put(tr);
5421 static const struct file_operations tracing_iter_fops = {
5422 .open = tracing_trace_options_open,
5424 .llseek = seq_lseek,
5425 .release = tracing_single_release_tr,
5426 .write = tracing_trace_options_write,
5429 static const char readme_msg[] =
5430 "tracing mini-HOWTO:\n\n"
5431 "# echo 0 > tracing_on : quick way to disable tracing\n"
5432 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5433 " Important files:\n"
5434 " trace\t\t\t- The static contents of the buffer\n"
5435 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5436 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5437 " current_tracer\t- function and latency tracers\n"
5438 " available_tracers\t- list of configured tracers for current_tracer\n"
5439 " error_log\t- error log for failed commands (that support it)\n"
5440 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5441 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5442 " trace_clock\t\t-change the clock used to order events\n"
5443 " local: Per cpu clock but may not be synced across CPUs\n"
5444 " global: Synced across CPUs but slows tracing down.\n"
5445 " counter: Not a clock, but just an increment\n"
5446 " uptime: Jiffy counter from time of boot\n"
5447 " perf: Same clock that perf events use\n"
5448 #ifdef CONFIG_X86_64
5449 " x86-tsc: TSC cycle counter\n"
5451 "\n timestamp_mode\t-view the mode used to timestamp events\n"
5452 " delta: Delta difference against a buffer-wide timestamp\n"
5453 " absolute: Absolute (standalone) timestamp\n"
5454 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5455 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5456 " tracing_cpumask\t- Limit which CPUs to trace\n"
5457 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5458 "\t\t\t Remove sub-buffer with rmdir\n"
5459 " trace_options\t\t- Set format or modify how tracing happens\n"
5460 "\t\t\t Disable an option by prefixing 'no' to the\n"
5461 "\t\t\t option name\n"
5462 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5463 #ifdef CONFIG_DYNAMIC_FTRACE
5464 "\n available_filter_functions - list of functions that can be filtered on\n"
5465 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5466 "\t\t\t functions\n"
5467 "\t accepts: func_full_name or glob-matching-pattern\n"
5468 "\t modules: Can select a group via module\n"
5469 "\t Format: :mod:<module-name>\n"
5470 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5471 "\t triggers: a command to perform when function is hit\n"
5472 "\t Format: <function>:<trigger>[:count]\n"
5473 "\t trigger: traceon, traceoff\n"
5474 "\t\t enable_event:<system>:<event>\n"
5475 "\t\t disable_event:<system>:<event>\n"
5476 #ifdef CONFIG_STACKTRACE
5479 #ifdef CONFIG_TRACER_SNAPSHOT
5484 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5485 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5486 "\t The first one will disable tracing every time do_fault is hit\n"
5487 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5488 "\t The first time do trap is hit and it disables tracing, the\n"
5489 "\t counter will decrement to 2. If tracing is already disabled,\n"
5490 "\t the counter will not decrement. It only decrements when the\n"
5491 "\t trigger did work\n"
5492 "\t To remove trigger without count:\n"
5493 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5494 "\t To remove trigger with a count:\n"
5495 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5496 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5497 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5498 "\t modules: Can select a group via module command :mod:\n"
5499 "\t Does not accept triggers\n"
5500 #endif /* CONFIG_DYNAMIC_FTRACE */
5501 #ifdef CONFIG_FUNCTION_TRACER
5502 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5504 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5507 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5508 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5509 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5510 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5512 #ifdef CONFIG_TRACER_SNAPSHOT
5513 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5514 "\t\t\t snapshot buffer. Read the contents for more\n"
5515 "\t\t\t information\n"
5517 #ifdef CONFIG_STACK_TRACER
5518 " stack_trace\t\t- Shows the max stack trace when active\n"
5519 " stack_max_size\t- Shows current max stack size that was traced\n"
5520 "\t\t\t Write into this file to reset the max size (trigger a\n"
5521 "\t\t\t new trace)\n"
5522 #ifdef CONFIG_DYNAMIC_FTRACE
5523 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5526 #endif /* CONFIG_STACK_TRACER */
5527 #ifdef CONFIG_DYNAMIC_EVENTS
5528 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5529 "\t\t\t Write into this file to define/undefine new trace events.\n"
5531 #ifdef CONFIG_KPROBE_EVENTS
5532 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5533 "\t\t\t Write into this file to define/undefine new trace events.\n"
5535 #ifdef CONFIG_UPROBE_EVENTS
5536 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5537 "\t\t\t Write into this file to define/undefine new trace events.\n"
5539 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5540 "\t accepts: event-definitions (one definition per line)\n"
5541 "\t Format: p[:[<group>/]<event>] <place> [<args>]\n"
5542 "\t r[maxactive][:[<group>/]<event>] <place> [<args>]\n"
5543 #ifdef CONFIG_HIST_TRIGGERS
5544 "\t s:[synthetic/]<event> <field> [<field>]\n"
5546 "\t e[:[<group>/]<event>] <attached-group>.<attached-event> [<args>]\n"
5547 "\t -:[<group>/]<event>\n"
5548 #ifdef CONFIG_KPROBE_EVENTS
5549 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5550 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5552 #ifdef CONFIG_UPROBE_EVENTS
5553 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5555 "\t args: <name>=fetcharg[:type]\n"
5556 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5557 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5558 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5560 "\t $stack<index>, $stack, $retval, $comm,\n"
5562 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5563 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string, symbol,\n"
5564 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5565 "\t <type>\\[<array-size>\\]\n"
5566 #ifdef CONFIG_HIST_TRIGGERS
5567 "\t field: <stype> <name>;\n"
5568 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5569 "\t [unsigned] char/int/long\n"
5571 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5572 "\t of the <attached-group>/<attached-event>.\n"
5574 " events/\t\t- Directory containing all trace event subsystems:\n"
5575 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5576 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5577 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5579 " filter\t\t- If set, only events passing filter are traced\n"
5580 " events/<system>/<event>/\t- Directory containing control files for\n"
5582 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5583 " filter\t\t- If set, only events passing filter are traced\n"
5584 " trigger\t\t- If set, a command to perform when event is hit\n"
5585 "\t Format: <trigger>[:count][if <filter>]\n"
5586 "\t trigger: traceon, traceoff\n"
5587 "\t enable_event:<system>:<event>\n"
5588 "\t disable_event:<system>:<event>\n"
5589 #ifdef CONFIG_HIST_TRIGGERS
5590 "\t enable_hist:<system>:<event>\n"
5591 "\t disable_hist:<system>:<event>\n"
5593 #ifdef CONFIG_STACKTRACE
5596 #ifdef CONFIG_TRACER_SNAPSHOT
5599 #ifdef CONFIG_HIST_TRIGGERS
5600 "\t\t hist (see below)\n"
5602 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5603 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5604 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5605 "\t events/block/block_unplug/trigger\n"
5606 "\t The first disables tracing every time block_unplug is hit.\n"
5607 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5608 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5609 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5610 "\t Like function triggers, the counter is only decremented if it\n"
5611 "\t enabled or disabled tracing.\n"
5612 "\t To remove a trigger without a count:\n"
5613 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5614 "\t To remove a trigger with a count:\n"
5615 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5616 "\t Filters can be ignored when removing a trigger.\n"
5617 #ifdef CONFIG_HIST_TRIGGERS
5618 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5619 "\t Format: hist:keys=<field1[,field2,...]>\n"
5620 "\t [:values=<field1[,field2,...]>]\n"
5621 "\t [:sort=<field1[,field2,...]>]\n"
5622 "\t [:size=#entries]\n"
5623 "\t [:pause][:continue][:clear]\n"
5624 "\t [:name=histname1]\n"
5625 "\t [:<handler>.<action>]\n"
5626 "\t [if <filter>]\n\n"
5627 "\t Note, special fields can be used as well:\n"
5628 "\t common_timestamp - to record current timestamp\n"
5629 "\t common_cpu - to record the CPU the event happened on\n"
5631 "\t When a matching event is hit, an entry is added to a hash\n"
5632 "\t table using the key(s) and value(s) named, and the value of a\n"
5633 "\t sum called 'hitcount' is incremented. Keys and values\n"
5634 "\t correspond to fields in the event's format description. Keys\n"
5635 "\t can be any field, or the special string 'stacktrace'.\n"
5636 "\t Compound keys consisting of up to two fields can be specified\n"
5637 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5638 "\t fields. Sort keys consisting of up to two fields can be\n"
5639 "\t specified using the 'sort' keyword. The sort direction can\n"
5640 "\t be modified by appending '.descending' or '.ascending' to a\n"
5641 "\t sort field. The 'size' parameter can be used to specify more\n"
5642 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5643 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5644 "\t its histogram data will be shared with other triggers of the\n"
5645 "\t same name, and trigger hits will update this common data.\n\n"
5646 "\t Reading the 'hist' file for the event will dump the hash\n"
5647 "\t table in its entirety to stdout. If there are multiple hist\n"
5648 "\t triggers attached to an event, there will be a table for each\n"
5649 "\t trigger in the output. The table displayed for a named\n"
5650 "\t trigger will be the same as any other instance having the\n"
5651 "\t same name. The default format used to display a given field\n"
5652 "\t can be modified by appending any of the following modifiers\n"
5653 "\t to the field name, as applicable:\n\n"
5654 "\t .hex display a number as a hex value\n"
5655 "\t .sym display an address as a symbol\n"
5656 "\t .sym-offset display an address as a symbol and offset\n"
5657 "\t .execname display a common_pid as a program name\n"
5658 "\t .syscall display a syscall id as a syscall name\n"
5659 "\t .log2 display log2 value rather than raw number\n"
5660 "\t .buckets=size display values in groups of size rather than raw number\n"
5661 "\t .usecs display a common_timestamp in microseconds\n\n"
5662 "\t The 'pause' parameter can be used to pause an existing hist\n"
5663 "\t trigger or to start a hist trigger but not log any events\n"
5664 "\t until told to do so. 'continue' can be used to start or\n"
5665 "\t restart a paused hist trigger.\n\n"
5666 "\t The 'clear' parameter will clear the contents of a running\n"
5667 "\t hist trigger and leave its current paused/active state\n"
5669 "\t The enable_hist and disable_hist triggers can be used to\n"
5670 "\t have one event conditionally start and stop another event's\n"
5671 "\t already-attached hist trigger. The syntax is analogous to\n"
5672 "\t the enable_event and disable_event triggers.\n\n"
5673 "\t Hist trigger handlers and actions are executed whenever a\n"
5674 "\t a histogram entry is added or updated. They take the form:\n\n"
5675 "\t <handler>.<action>\n\n"
5676 "\t The available handlers are:\n\n"
5677 "\t onmatch(matching.event) - invoke on addition or update\n"
5678 "\t onmax(var) - invoke if var exceeds current max\n"
5679 "\t onchange(var) - invoke action if var changes\n\n"
5680 "\t The available actions are:\n\n"
5681 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5682 "\t save(field,...) - save current event fields\n"
5683 #ifdef CONFIG_TRACER_SNAPSHOT
5684 "\t snapshot() - snapshot the trace buffer\n\n"
5686 #ifdef CONFIG_SYNTH_EVENTS
5687 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5688 "\t Write into this file to define/undefine new synthetic events.\n"
5689 "\t example: echo 'myevent u64 lat; char name[]' >> synthetic_events\n"
5695 tracing_readme_read(struct file *filp, char __user *ubuf,
5696 size_t cnt, loff_t *ppos)
5698 return simple_read_from_buffer(ubuf, cnt, ppos,
5699 readme_msg, strlen(readme_msg));
5702 static const struct file_operations tracing_readme_fops = {
5703 .open = tracing_open_generic,
5704 .read = tracing_readme_read,
5705 .llseek = generic_file_llseek,
5708 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5712 return trace_find_tgid_ptr(pid);
5715 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5719 return trace_find_tgid_ptr(pid);
5722 static void saved_tgids_stop(struct seq_file *m, void *v)
5726 static int saved_tgids_show(struct seq_file *m, void *v)
5728 int *entry = (int *)v;
5729 int pid = entry - tgid_map;
5735 seq_printf(m, "%d %d\n", pid, tgid);
5739 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5740 .start = saved_tgids_start,
5741 .stop = saved_tgids_stop,
5742 .next = saved_tgids_next,
5743 .show = saved_tgids_show,
5746 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5750 ret = tracing_check_open_get_tr(NULL);
5754 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5758 static const struct file_operations tracing_saved_tgids_fops = {
5759 .open = tracing_saved_tgids_open,
5761 .llseek = seq_lseek,
5762 .release = seq_release,
5765 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5767 unsigned int *ptr = v;
5769 if (*pos || m->count)
5774 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5776 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5785 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5791 arch_spin_lock(&trace_cmdline_lock);
5793 v = &savedcmd->map_cmdline_to_pid[0];
5795 v = saved_cmdlines_next(m, v, &l);
5803 static void saved_cmdlines_stop(struct seq_file *m, void *v)
5805 arch_spin_unlock(&trace_cmdline_lock);
5809 static int saved_cmdlines_show(struct seq_file *m, void *v)
5811 char buf[TASK_COMM_LEN];
5812 unsigned int *pid = v;
5814 __trace_find_cmdline(*pid, buf);
5815 seq_printf(m, "%d %s\n", *pid, buf);
5819 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5820 .start = saved_cmdlines_start,
5821 .next = saved_cmdlines_next,
5822 .stop = saved_cmdlines_stop,
5823 .show = saved_cmdlines_show,
5826 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5830 ret = tracing_check_open_get_tr(NULL);
5834 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5837 static const struct file_operations tracing_saved_cmdlines_fops = {
5838 .open = tracing_saved_cmdlines_open,
5840 .llseek = seq_lseek,
5841 .release = seq_release,
5845 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5846 size_t cnt, loff_t *ppos)
5851 arch_spin_lock(&trace_cmdline_lock);
5852 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5853 arch_spin_unlock(&trace_cmdline_lock);
5855 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5858 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5860 kfree(s->saved_cmdlines);
5861 kfree(s->map_cmdline_to_pid);
5865 static int tracing_resize_saved_cmdlines(unsigned int val)
5867 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5869 s = kmalloc(sizeof(*s), GFP_KERNEL);
5873 if (allocate_cmdlines_buffer(val, s) < 0) {
5878 arch_spin_lock(&trace_cmdline_lock);
5879 savedcmd_temp = savedcmd;
5881 arch_spin_unlock(&trace_cmdline_lock);
5882 free_saved_cmdlines_buffer(savedcmd_temp);
5888 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
5889 size_t cnt, loff_t *ppos)
5894 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5898 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
5899 if (!val || val > PID_MAX_DEFAULT)
5902 ret = tracing_resize_saved_cmdlines((unsigned int)val);
5911 static const struct file_operations tracing_saved_cmdlines_size_fops = {
5912 .open = tracing_open_generic,
5913 .read = tracing_saved_cmdlines_size_read,
5914 .write = tracing_saved_cmdlines_size_write,
5917 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
5918 static union trace_eval_map_item *
5919 update_eval_map(union trace_eval_map_item *ptr)
5921 if (!ptr->map.eval_string) {
5922 if (ptr->tail.next) {
5923 ptr = ptr->tail.next;
5924 /* Set ptr to the next real item (skip head) */
5932 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
5934 union trace_eval_map_item *ptr = v;
5937 * Paranoid! If ptr points to end, we don't want to increment past it.
5938 * This really should never happen.
5941 ptr = update_eval_map(ptr);
5942 if (WARN_ON_ONCE(!ptr))
5946 ptr = update_eval_map(ptr);
5951 static void *eval_map_start(struct seq_file *m, loff_t *pos)
5953 union trace_eval_map_item *v;
5956 mutex_lock(&trace_eval_mutex);
5958 v = trace_eval_maps;
5962 while (v && l < *pos) {
5963 v = eval_map_next(m, v, &l);
5969 static void eval_map_stop(struct seq_file *m, void *v)
5971 mutex_unlock(&trace_eval_mutex);
5974 static int eval_map_show(struct seq_file *m, void *v)
5976 union trace_eval_map_item *ptr = v;
5978 seq_printf(m, "%s %ld (%s)\n",
5979 ptr->map.eval_string, ptr->map.eval_value,
5985 static const struct seq_operations tracing_eval_map_seq_ops = {
5986 .start = eval_map_start,
5987 .next = eval_map_next,
5988 .stop = eval_map_stop,
5989 .show = eval_map_show,
5992 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
5996 ret = tracing_check_open_get_tr(NULL);
6000 return seq_open(filp, &tracing_eval_map_seq_ops);
6003 static const struct file_operations tracing_eval_map_fops = {
6004 .open = tracing_eval_map_open,
6006 .llseek = seq_lseek,
6007 .release = seq_release,
6010 static inline union trace_eval_map_item *
6011 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6013 /* Return tail of array given the head */
6014 return ptr + ptr->head.length + 1;
6018 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6021 struct trace_eval_map **stop;
6022 struct trace_eval_map **map;
6023 union trace_eval_map_item *map_array;
6024 union trace_eval_map_item *ptr;
6029 * The trace_eval_maps contains the map plus a head and tail item,
6030 * where the head holds the module and length of array, and the
6031 * tail holds a pointer to the next list.
6033 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6035 pr_warn("Unable to allocate trace eval mapping\n");
6039 mutex_lock(&trace_eval_mutex);
6041 if (!trace_eval_maps)
6042 trace_eval_maps = map_array;
6044 ptr = trace_eval_maps;
6046 ptr = trace_eval_jmp_to_tail(ptr);
6047 if (!ptr->tail.next)
6049 ptr = ptr->tail.next;
6052 ptr->tail.next = map_array;
6054 map_array->head.mod = mod;
6055 map_array->head.length = len;
6058 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6059 map_array->map = **map;
6062 memset(map_array, 0, sizeof(*map_array));
6064 mutex_unlock(&trace_eval_mutex);
6067 static void trace_create_eval_file(struct dentry *d_tracer)
6069 trace_create_file("eval_map", 0444, d_tracer,
6070 NULL, &tracing_eval_map_fops);
6073 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
6074 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6075 static inline void trace_insert_eval_map_file(struct module *mod,
6076 struct trace_eval_map **start, int len) { }
6077 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6079 static void trace_insert_eval_map(struct module *mod,
6080 struct trace_eval_map **start, int len)
6082 struct trace_eval_map **map;
6089 trace_event_eval_update(map, len);
6091 trace_insert_eval_map_file(mod, start, len);
6095 tracing_set_trace_read(struct file *filp, char __user *ubuf,
6096 size_t cnt, loff_t *ppos)
6098 struct trace_array *tr = filp->private_data;
6099 char buf[MAX_TRACER_SIZE+2];
6102 mutex_lock(&trace_types_lock);
6103 r = sprintf(buf, "%s\n", tr->current_trace->name);
6104 mutex_unlock(&trace_types_lock);
6106 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6109 int tracer_init(struct tracer *t, struct trace_array *tr)
6111 tracing_reset_online_cpus(&tr->array_buffer);
6115 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6119 for_each_tracing_cpu(cpu)
6120 per_cpu_ptr(buf->data, cpu)->entries = val;
6123 #ifdef CONFIG_TRACER_MAX_TRACE
6124 /* resize @tr's buffer to the size of @size_tr's entries */
6125 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6126 struct array_buffer *size_buf, int cpu_id)
6130 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6131 for_each_tracing_cpu(cpu) {
6132 ret = ring_buffer_resize(trace_buf->buffer,
6133 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6136 per_cpu_ptr(trace_buf->data, cpu)->entries =
6137 per_cpu_ptr(size_buf->data, cpu)->entries;
6140 ret = ring_buffer_resize(trace_buf->buffer,
6141 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6143 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6144 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6149 #endif /* CONFIG_TRACER_MAX_TRACE */
6151 static int __tracing_resize_ring_buffer(struct trace_array *tr,
6152 unsigned long size, int cpu)
6157 * If kernel or user changes the size of the ring buffer
6158 * we use the size that was given, and we can forget about
6159 * expanding it later.
6161 ring_buffer_expanded = true;
6163 /* May be called before buffers are initialized */
6164 if (!tr->array_buffer.buffer)
6167 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6171 #ifdef CONFIG_TRACER_MAX_TRACE
6172 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
6173 !tr->current_trace->use_max_tr)
6176 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6178 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6179 &tr->array_buffer, cpu);
6182 * AARGH! We are left with different
6183 * size max buffer!!!!
6184 * The max buffer is our "snapshot" buffer.
6185 * When a tracer needs a snapshot (one of the
6186 * latency tracers), it swaps the max buffer
6187 * with the saved snap shot. We succeeded to
6188 * update the size of the main buffer, but failed to
6189 * update the size of the max buffer. But when we tried
6190 * to reset the main buffer to the original size, we
6191 * failed there too. This is very unlikely to
6192 * happen, but if it does, warn and kill all
6196 tracing_disabled = 1;
6201 if (cpu == RING_BUFFER_ALL_CPUS)
6202 set_buffer_entries(&tr->max_buffer, size);
6204 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
6207 #endif /* CONFIG_TRACER_MAX_TRACE */
6209 if (cpu == RING_BUFFER_ALL_CPUS)
6210 set_buffer_entries(&tr->array_buffer, size);
6212 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
6217 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6218 unsigned long size, int cpu_id)
6222 mutex_lock(&trace_types_lock);
6224 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6225 /* make sure, this cpu is enabled in the mask */
6226 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6232 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6237 mutex_unlock(&trace_types_lock);
6244 * tracing_update_buffers - used by tracing facility to expand ring buffers
6246 * To save on memory when the tracing is never used on a system with it
6247 * configured in. The ring buffers are set to a minimum size. But once
6248 * a user starts to use the tracing facility, then they need to grow
6249 * to their default size.
6251 * This function is to be called when a tracer is about to be used.
6253 int tracing_update_buffers(void)
6257 mutex_lock(&trace_types_lock);
6258 if (!ring_buffer_expanded)
6259 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
6260 RING_BUFFER_ALL_CPUS);
6261 mutex_unlock(&trace_types_lock);
6266 struct trace_option_dentry;
6269 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6272 * Used to clear out the tracer before deletion of an instance.
6273 * Must have trace_types_lock held.
6275 static void tracing_set_nop(struct trace_array *tr)
6277 if (tr->current_trace == &nop_trace)
6280 tr->current_trace->enabled--;
6282 if (tr->current_trace->reset)
6283 tr->current_trace->reset(tr);
6285 tr->current_trace = &nop_trace;
6288 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6290 /* Only enable if the directory has been created already. */
6294 create_trace_option_files(tr, t);
6297 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6300 #ifdef CONFIG_TRACER_MAX_TRACE
6305 mutex_lock(&trace_types_lock);
6307 if (!ring_buffer_expanded) {
6308 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6309 RING_BUFFER_ALL_CPUS);
6315 for (t = trace_types; t; t = t->next) {
6316 if (strcmp(t->name, buf) == 0)
6323 if (t == tr->current_trace)
6326 #ifdef CONFIG_TRACER_SNAPSHOT
6327 if (t->use_max_tr) {
6328 arch_spin_lock(&tr->max_lock);
6329 if (tr->cond_snapshot)
6331 arch_spin_unlock(&tr->max_lock);
6336 /* Some tracers won't work on kernel command line */
6337 if (system_state < SYSTEM_RUNNING && t->noboot) {
6338 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6343 /* Some tracers are only allowed for the top level buffer */
6344 if (!trace_ok_for_array(t, tr)) {
6349 /* If trace pipe files are being read, we can't change the tracer */
6350 if (tr->trace_ref) {
6355 trace_branch_disable();
6357 tr->current_trace->enabled--;
6359 if (tr->current_trace->reset)
6360 tr->current_trace->reset(tr);
6362 /* Current trace needs to be nop_trace before synchronize_rcu */
6363 tr->current_trace = &nop_trace;
6365 #ifdef CONFIG_TRACER_MAX_TRACE
6366 had_max_tr = tr->allocated_snapshot;
6368 if (had_max_tr && !t->use_max_tr) {
6370 * We need to make sure that the update_max_tr sees that
6371 * current_trace changed to nop_trace to keep it from
6372 * swapping the buffers after we resize it.
6373 * The update_max_tr is called from interrupts disabled
6374 * so a synchronized_sched() is sufficient.
6381 #ifdef CONFIG_TRACER_MAX_TRACE
6382 if (t->use_max_tr && !had_max_tr) {
6383 ret = tracing_alloc_snapshot_instance(tr);
6390 ret = tracer_init(t, tr);
6395 tr->current_trace = t;
6396 tr->current_trace->enabled++;
6397 trace_branch_enable(tr);
6399 mutex_unlock(&trace_types_lock);
6405 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6406 size_t cnt, loff_t *ppos)
6408 struct trace_array *tr = filp->private_data;
6409 char buf[MAX_TRACER_SIZE+1];
6416 if (cnt > MAX_TRACER_SIZE)
6417 cnt = MAX_TRACER_SIZE;
6419 if (copy_from_user(buf, ubuf, cnt))
6424 /* strip ending whitespace. */
6425 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
6428 err = tracing_set_tracer(tr, buf);
6438 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6439 size_t cnt, loff_t *ppos)
6444 r = snprintf(buf, sizeof(buf), "%ld\n",
6445 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6446 if (r > sizeof(buf))
6448 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6452 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6453 size_t cnt, loff_t *ppos)
6458 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6468 tracing_thresh_read(struct file *filp, char __user *ubuf,
6469 size_t cnt, loff_t *ppos)
6471 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6475 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6476 size_t cnt, loff_t *ppos)
6478 struct trace_array *tr = filp->private_data;
6481 mutex_lock(&trace_types_lock);
6482 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6486 if (tr->current_trace->update_thresh) {
6487 ret = tr->current_trace->update_thresh(tr);
6494 mutex_unlock(&trace_types_lock);
6499 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
6502 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6503 size_t cnt, loff_t *ppos)
6505 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6509 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6510 size_t cnt, loff_t *ppos)
6512 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6517 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6519 struct trace_array *tr = inode->i_private;
6520 struct trace_iterator *iter;
6523 ret = tracing_check_open_get_tr(tr);
6527 mutex_lock(&trace_types_lock);
6529 /* create a buffer to store the information to pass to userspace */
6530 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6533 __trace_array_put(tr);
6537 trace_seq_init(&iter->seq);
6538 iter->trace = tr->current_trace;
6540 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6545 /* trace pipe does not show start of buffer */
6546 cpumask_setall(iter->started);
6548 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6549 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6551 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6552 if (trace_clocks[tr->clock_id].in_ns)
6553 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6556 iter->array_buffer = &tr->array_buffer;
6557 iter->cpu_file = tracing_get_cpu(inode);
6558 mutex_init(&iter->mutex);
6559 filp->private_data = iter;
6561 if (iter->trace->pipe_open)
6562 iter->trace->pipe_open(iter);
6564 nonseekable_open(inode, filp);
6568 mutex_unlock(&trace_types_lock);
6573 __trace_array_put(tr);
6574 mutex_unlock(&trace_types_lock);
6578 static int tracing_release_pipe(struct inode *inode, struct file *file)
6580 struct trace_iterator *iter = file->private_data;
6581 struct trace_array *tr = inode->i_private;
6583 mutex_lock(&trace_types_lock);
6587 if (iter->trace->pipe_close)
6588 iter->trace->pipe_close(iter);
6590 mutex_unlock(&trace_types_lock);
6592 free_cpumask_var(iter->started);
6593 mutex_destroy(&iter->mutex);
6596 trace_array_put(tr);
6602 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6604 struct trace_array *tr = iter->tr;
6606 /* Iterators are static, they should be filled or empty */
6607 if (trace_buffer_iter(iter, iter->cpu_file))
6608 return EPOLLIN | EPOLLRDNORM;
6610 if (tr->trace_flags & TRACE_ITER_BLOCK)
6612 * Always select as readable when in blocking mode
6614 return EPOLLIN | EPOLLRDNORM;
6616 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6621 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6623 struct trace_iterator *iter = filp->private_data;
6625 return trace_poll(iter, filp, poll_table);
6628 /* Must be called with iter->mutex held. */
6629 static int tracing_wait_pipe(struct file *filp)
6631 struct trace_iterator *iter = filp->private_data;
6634 while (trace_empty(iter)) {
6636 if ((filp->f_flags & O_NONBLOCK)) {
6641 * We block until we read something and tracing is disabled.
6642 * We still block if tracing is disabled, but we have never
6643 * read anything. This allows a user to cat this file, and
6644 * then enable tracing. But after we have read something,
6645 * we give an EOF when tracing is again disabled.
6647 * iter->pos will be 0 if we haven't read anything.
6649 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6652 mutex_unlock(&iter->mutex);
6654 ret = wait_on_pipe(iter, 0);
6656 mutex_lock(&iter->mutex);
6669 tracing_read_pipe(struct file *filp, char __user *ubuf,
6670 size_t cnt, loff_t *ppos)
6672 struct trace_iterator *iter = filp->private_data;
6676 * Avoid more than one consumer on a single file descriptor
6677 * This is just a matter of traces coherency, the ring buffer itself
6680 mutex_lock(&iter->mutex);
6682 /* return any leftover data */
6683 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6687 trace_seq_init(&iter->seq);
6689 if (iter->trace->read) {
6690 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6696 sret = tracing_wait_pipe(filp);
6700 /* stop when tracing is finished */
6701 if (trace_empty(iter)) {
6706 if (cnt >= PAGE_SIZE)
6707 cnt = PAGE_SIZE - 1;
6709 /* reset all but tr, trace, and overruns */
6710 memset(&iter->seq, 0,
6711 sizeof(struct trace_iterator) -
6712 offsetof(struct trace_iterator, seq));
6713 cpumask_clear(iter->started);
6714 trace_seq_init(&iter->seq);
6717 trace_event_read_lock();
6718 trace_access_lock(iter->cpu_file);
6719 while (trace_find_next_entry_inc(iter) != NULL) {
6720 enum print_line_t ret;
6721 int save_len = iter->seq.seq.len;
6723 ret = print_trace_line(iter);
6724 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6725 /* don't print partial lines */
6726 iter->seq.seq.len = save_len;
6729 if (ret != TRACE_TYPE_NO_CONSUME)
6730 trace_consume(iter);
6732 if (trace_seq_used(&iter->seq) >= cnt)
6736 * Setting the full flag means we reached the trace_seq buffer
6737 * size and we should leave by partial output condition above.
6738 * One of the trace_seq_* functions is not used properly.
6740 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6743 trace_access_unlock(iter->cpu_file);
6744 trace_event_read_unlock();
6746 /* Now copy what we have to the user */
6747 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6748 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6749 trace_seq_init(&iter->seq);
6752 * If there was nothing to send to user, in spite of consuming trace
6753 * entries, go back to wait for more entries.
6759 mutex_unlock(&iter->mutex);
6764 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6767 __free_page(spd->pages[idx]);
6771 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6777 /* Seq buffer is page-sized, exactly what we need. */
6779 save_len = iter->seq.seq.len;
6780 ret = print_trace_line(iter);
6782 if (trace_seq_has_overflowed(&iter->seq)) {
6783 iter->seq.seq.len = save_len;
6788 * This should not be hit, because it should only
6789 * be set if the iter->seq overflowed. But check it
6790 * anyway to be safe.
6792 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6793 iter->seq.seq.len = save_len;
6797 count = trace_seq_used(&iter->seq) - save_len;
6800 iter->seq.seq.len = save_len;
6804 if (ret != TRACE_TYPE_NO_CONSUME)
6805 trace_consume(iter);
6807 if (!trace_find_next_entry_inc(iter)) {
6817 static ssize_t tracing_splice_read_pipe(struct file *filp,
6819 struct pipe_inode_info *pipe,
6823 struct page *pages_def[PIPE_DEF_BUFFERS];
6824 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6825 struct trace_iterator *iter = filp->private_data;
6826 struct splice_pipe_desc spd = {
6828 .partial = partial_def,
6829 .nr_pages = 0, /* This gets updated below. */
6830 .nr_pages_max = PIPE_DEF_BUFFERS,
6831 .ops = &default_pipe_buf_ops,
6832 .spd_release = tracing_spd_release_pipe,
6838 if (splice_grow_spd(pipe, &spd))
6841 mutex_lock(&iter->mutex);
6843 if (iter->trace->splice_read) {
6844 ret = iter->trace->splice_read(iter, filp,
6845 ppos, pipe, len, flags);
6850 ret = tracing_wait_pipe(filp);
6854 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6859 trace_event_read_lock();
6860 trace_access_lock(iter->cpu_file);
6862 /* Fill as many pages as possible. */
6863 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6864 spd.pages[i] = alloc_page(GFP_KERNEL);
6868 rem = tracing_fill_pipe_page(rem, iter);
6870 /* Copy the data into the page, so we can start over. */
6871 ret = trace_seq_to_buffer(&iter->seq,
6872 page_address(spd.pages[i]),
6873 trace_seq_used(&iter->seq));
6875 __free_page(spd.pages[i]);
6878 spd.partial[i].offset = 0;
6879 spd.partial[i].len = trace_seq_used(&iter->seq);
6881 trace_seq_init(&iter->seq);
6884 trace_access_unlock(iter->cpu_file);
6885 trace_event_read_unlock();
6886 mutex_unlock(&iter->mutex);
6891 ret = splice_to_pipe(pipe, &spd);
6895 splice_shrink_spd(&spd);
6899 mutex_unlock(&iter->mutex);
6904 tracing_entries_read(struct file *filp, char __user *ubuf,
6905 size_t cnt, loff_t *ppos)
6907 struct inode *inode = file_inode(filp);
6908 struct trace_array *tr = inode->i_private;
6909 int cpu = tracing_get_cpu(inode);
6914 mutex_lock(&trace_types_lock);
6916 if (cpu == RING_BUFFER_ALL_CPUS) {
6917 int cpu, buf_size_same;
6922 /* check if all cpu sizes are same */
6923 for_each_tracing_cpu(cpu) {
6924 /* fill in the size from first enabled cpu */
6926 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
6927 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
6933 if (buf_size_same) {
6934 if (!ring_buffer_expanded)
6935 r = sprintf(buf, "%lu (expanded: %lu)\n",
6937 trace_buf_size >> 10);
6939 r = sprintf(buf, "%lu\n", size >> 10);
6941 r = sprintf(buf, "X\n");
6943 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
6945 mutex_unlock(&trace_types_lock);
6947 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6952 tracing_entries_write(struct file *filp, const char __user *ubuf,
6953 size_t cnt, loff_t *ppos)
6955 struct inode *inode = file_inode(filp);
6956 struct trace_array *tr = inode->i_private;
6960 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6964 /* must have at least 1 entry */
6968 /* value is in KB */
6970 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
6980 tracing_total_entries_read(struct file *filp, char __user *ubuf,
6981 size_t cnt, loff_t *ppos)
6983 struct trace_array *tr = filp->private_data;
6986 unsigned long size = 0, expanded_size = 0;
6988 mutex_lock(&trace_types_lock);
6989 for_each_tracing_cpu(cpu) {
6990 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
6991 if (!ring_buffer_expanded)
6992 expanded_size += trace_buf_size >> 10;
6994 if (ring_buffer_expanded)
6995 r = sprintf(buf, "%lu\n", size);
6997 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
6998 mutex_unlock(&trace_types_lock);
7000 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7004 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7005 size_t cnt, loff_t *ppos)
7008 * There is no need to read what the user has written, this function
7009 * is just to make sure that there is no error when "echo" is used
7018 tracing_free_buffer_release(struct inode *inode, struct file *filp)
7020 struct trace_array *tr = inode->i_private;
7022 /* disable tracing ? */
7023 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7024 tracer_tracing_off(tr);
7025 /* resize the ring buffer to 0 */
7026 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7028 trace_array_put(tr);
7034 tracing_mark_write(struct file *filp, const char __user *ubuf,
7035 size_t cnt, loff_t *fpos)
7037 struct trace_array *tr = filp->private_data;
7038 struct ring_buffer_event *event;
7039 enum event_trigger_type tt = ETT_NONE;
7040 struct trace_buffer *buffer;
7041 struct print_entry *entry;
7046 /* Used in tracing_mark_raw_write() as well */
7047 #define FAULTED_STR "<faulted>"
7048 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7050 if (tracing_disabled)
7053 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7056 if (cnt > TRACE_BUF_SIZE)
7057 cnt = TRACE_BUF_SIZE;
7059 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7061 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
7063 /* If less than "<faulted>", then make sure we can still add that */
7064 if (cnt < FAULTED_SIZE)
7065 size += FAULTED_SIZE - cnt;
7067 buffer = tr->array_buffer.buffer;
7068 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7070 if (unlikely(!event))
7071 /* Ring buffer disabled, return as if not open for write */
7074 entry = ring_buffer_event_data(event);
7075 entry->ip = _THIS_IP_;
7077 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7079 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7085 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7086 /* do not add \n before testing triggers, but add \0 */
7087 entry->buf[cnt] = '\0';
7088 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7091 if (entry->buf[cnt - 1] != '\n') {
7092 entry->buf[cnt] = '\n';
7093 entry->buf[cnt + 1] = '\0';
7095 entry->buf[cnt] = '\0';
7097 if (static_branch_unlikely(&trace_marker_exports_enabled))
7098 ftrace_exports(event, TRACE_EXPORT_MARKER);
7099 __buffer_unlock_commit(buffer, event);
7102 event_triggers_post_call(tr->trace_marker_file, tt);
7110 /* Limit it for now to 3K (including tag) */
7111 #define RAW_DATA_MAX_SIZE (1024*3)
7114 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7115 size_t cnt, loff_t *fpos)
7117 struct trace_array *tr = filp->private_data;
7118 struct ring_buffer_event *event;
7119 struct trace_buffer *buffer;
7120 struct raw_data_entry *entry;
7125 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7127 if (tracing_disabled)
7130 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7133 /* The marker must at least have a tag id */
7134 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
7137 if (cnt > TRACE_BUF_SIZE)
7138 cnt = TRACE_BUF_SIZE;
7140 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7142 size = sizeof(*entry) + cnt;
7143 if (cnt < FAULT_SIZE_ID)
7144 size += FAULT_SIZE_ID - cnt;
7146 buffer = tr->array_buffer.buffer;
7147 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7150 /* Ring buffer disabled, return as if not open for write */
7153 entry = ring_buffer_event_data(event);
7155 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7158 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7163 __buffer_unlock_commit(buffer, event);
7171 static int tracing_clock_show(struct seq_file *m, void *v)
7173 struct trace_array *tr = m->private;
7176 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7178 "%s%s%s%s", i ? " " : "",
7179 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7180 i == tr->clock_id ? "]" : "");
7186 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7190 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7191 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7194 if (i == ARRAY_SIZE(trace_clocks))
7197 mutex_lock(&trace_types_lock);
7201 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7204 * New clock may not be consistent with the previous clock.
7205 * Reset the buffer so that it doesn't have incomparable timestamps.
7207 tracing_reset_online_cpus(&tr->array_buffer);
7209 #ifdef CONFIG_TRACER_MAX_TRACE
7210 if (tr->max_buffer.buffer)
7211 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7212 tracing_reset_online_cpus(&tr->max_buffer);
7215 mutex_unlock(&trace_types_lock);
7220 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7221 size_t cnt, loff_t *fpos)
7223 struct seq_file *m = filp->private_data;
7224 struct trace_array *tr = m->private;
7226 const char *clockstr;
7229 if (cnt >= sizeof(buf))
7232 if (copy_from_user(buf, ubuf, cnt))
7237 clockstr = strstrip(buf);
7239 ret = tracing_set_clock(tr, clockstr);
7248 static int tracing_clock_open(struct inode *inode, struct file *file)
7250 struct trace_array *tr = inode->i_private;
7253 ret = tracing_check_open_get_tr(tr);
7257 ret = single_open(file, tracing_clock_show, inode->i_private);
7259 trace_array_put(tr);
7264 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7266 struct trace_array *tr = m->private;
7268 mutex_lock(&trace_types_lock);
7270 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7271 seq_puts(m, "delta [absolute]\n");
7273 seq_puts(m, "[delta] absolute\n");
7275 mutex_unlock(&trace_types_lock);
7280 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7282 struct trace_array *tr = inode->i_private;
7285 ret = tracing_check_open_get_tr(tr);
7289 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7291 trace_array_put(tr);
7296 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7298 if (rbe == this_cpu_read(trace_buffered_event))
7299 return ring_buffer_time_stamp(buffer);
7301 return ring_buffer_event_time_stamp(buffer, rbe);
7305 * Set or disable using the per CPU trace_buffer_event when possible.
7307 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7311 mutex_lock(&trace_types_lock);
7313 if (set && tr->no_filter_buffering_ref++)
7317 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7322 --tr->no_filter_buffering_ref;
7325 mutex_unlock(&trace_types_lock);
7330 struct ftrace_buffer_info {
7331 struct trace_iterator iter;
7333 unsigned int spare_cpu;
7337 #ifdef CONFIG_TRACER_SNAPSHOT
7338 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7340 struct trace_array *tr = inode->i_private;
7341 struct trace_iterator *iter;
7345 ret = tracing_check_open_get_tr(tr);
7349 if (file->f_mode & FMODE_READ) {
7350 iter = __tracing_open(inode, file, true);
7352 ret = PTR_ERR(iter);
7354 /* Writes still need the seq_file to hold the private data */
7356 m = kzalloc(sizeof(*m), GFP_KERNEL);
7359 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7367 iter->array_buffer = &tr->max_buffer;
7368 iter->cpu_file = tracing_get_cpu(inode);
7370 file->private_data = m;
7374 trace_array_put(tr);
7380 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7383 struct seq_file *m = filp->private_data;
7384 struct trace_iterator *iter = m->private;
7385 struct trace_array *tr = iter->tr;
7389 ret = tracing_update_buffers();
7393 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7397 mutex_lock(&trace_types_lock);
7399 if (tr->current_trace->use_max_tr) {
7404 arch_spin_lock(&tr->max_lock);
7405 if (tr->cond_snapshot)
7407 arch_spin_unlock(&tr->max_lock);
7413 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7417 if (tr->allocated_snapshot)
7421 /* Only allow per-cpu swap if the ring buffer supports it */
7422 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7423 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7428 if (tr->allocated_snapshot)
7429 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7430 &tr->array_buffer, iter->cpu_file);
7432 ret = tracing_alloc_snapshot_instance(tr);
7435 local_irq_disable();
7436 /* Now, we're going to swap */
7437 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7438 update_max_tr(tr, current, smp_processor_id(), NULL);
7440 update_max_tr_single(tr, current, iter->cpu_file);
7444 if (tr->allocated_snapshot) {
7445 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7446 tracing_reset_online_cpus(&tr->max_buffer);
7448 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7458 mutex_unlock(&trace_types_lock);
7462 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7464 struct seq_file *m = file->private_data;
7467 ret = tracing_release(inode, file);
7469 if (file->f_mode & FMODE_READ)
7472 /* If write only, the seq_file is just a stub */
7480 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7481 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7482 size_t count, loff_t *ppos);
7483 static int tracing_buffers_release(struct inode *inode, struct file *file);
7484 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7485 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7487 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7489 struct ftrace_buffer_info *info;
7492 /* The following checks for tracefs lockdown */
7493 ret = tracing_buffers_open(inode, filp);
7497 info = filp->private_data;
7499 if (info->iter.trace->use_max_tr) {
7500 tracing_buffers_release(inode, filp);
7504 info->iter.snapshot = true;
7505 info->iter.array_buffer = &info->iter.tr->max_buffer;
7510 #endif /* CONFIG_TRACER_SNAPSHOT */
7513 static const struct file_operations tracing_thresh_fops = {
7514 .open = tracing_open_generic,
7515 .read = tracing_thresh_read,
7516 .write = tracing_thresh_write,
7517 .llseek = generic_file_llseek,
7520 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
7521 static const struct file_operations tracing_max_lat_fops = {
7522 .open = tracing_open_generic,
7523 .read = tracing_max_lat_read,
7524 .write = tracing_max_lat_write,
7525 .llseek = generic_file_llseek,
7529 static const struct file_operations set_tracer_fops = {
7530 .open = tracing_open_generic,
7531 .read = tracing_set_trace_read,
7532 .write = tracing_set_trace_write,
7533 .llseek = generic_file_llseek,
7536 static const struct file_operations tracing_pipe_fops = {
7537 .open = tracing_open_pipe,
7538 .poll = tracing_poll_pipe,
7539 .read = tracing_read_pipe,
7540 .splice_read = tracing_splice_read_pipe,
7541 .release = tracing_release_pipe,
7542 .llseek = no_llseek,
7545 static const struct file_operations tracing_entries_fops = {
7546 .open = tracing_open_generic_tr,
7547 .read = tracing_entries_read,
7548 .write = tracing_entries_write,
7549 .llseek = generic_file_llseek,
7550 .release = tracing_release_generic_tr,
7553 static const struct file_operations tracing_total_entries_fops = {
7554 .open = tracing_open_generic_tr,
7555 .read = tracing_total_entries_read,
7556 .llseek = generic_file_llseek,
7557 .release = tracing_release_generic_tr,
7560 static const struct file_operations tracing_free_buffer_fops = {
7561 .open = tracing_open_generic_tr,
7562 .write = tracing_free_buffer_write,
7563 .release = tracing_free_buffer_release,
7566 static const struct file_operations tracing_mark_fops = {
7567 .open = tracing_open_generic_tr,
7568 .write = tracing_mark_write,
7569 .llseek = generic_file_llseek,
7570 .release = tracing_release_generic_tr,
7573 static const struct file_operations tracing_mark_raw_fops = {
7574 .open = tracing_open_generic_tr,
7575 .write = tracing_mark_raw_write,
7576 .llseek = generic_file_llseek,
7577 .release = tracing_release_generic_tr,
7580 static const struct file_operations trace_clock_fops = {
7581 .open = tracing_clock_open,
7583 .llseek = seq_lseek,
7584 .release = tracing_single_release_tr,
7585 .write = tracing_clock_write,
7588 static const struct file_operations trace_time_stamp_mode_fops = {
7589 .open = tracing_time_stamp_mode_open,
7591 .llseek = seq_lseek,
7592 .release = tracing_single_release_tr,
7595 #ifdef CONFIG_TRACER_SNAPSHOT
7596 static const struct file_operations snapshot_fops = {
7597 .open = tracing_snapshot_open,
7599 .write = tracing_snapshot_write,
7600 .llseek = tracing_lseek,
7601 .release = tracing_snapshot_release,
7604 static const struct file_operations snapshot_raw_fops = {
7605 .open = snapshot_raw_open,
7606 .read = tracing_buffers_read,
7607 .release = tracing_buffers_release,
7608 .splice_read = tracing_buffers_splice_read,
7609 .llseek = no_llseek,
7612 #endif /* CONFIG_TRACER_SNAPSHOT */
7615 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7616 * @filp: The active open file structure
7617 * @ubuf: The userspace provided buffer to read value into
7618 * @cnt: The maximum number of bytes to read
7619 * @ppos: The current "file" position
7621 * This function implements the write interface for a struct trace_min_max_param.
7622 * The filp->private_data must point to a trace_min_max_param structure that
7623 * defines where to write the value, the min and the max acceptable values,
7624 * and a lock to protect the write.
7627 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7629 struct trace_min_max_param *param = filp->private_data;
7636 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7641 mutex_lock(param->lock);
7643 if (param->min && val < *param->min)
7646 if (param->max && val > *param->max)
7653 mutex_unlock(param->lock);
7662 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7663 * @filp: The active open file structure
7664 * @ubuf: The userspace provided buffer to read value into
7665 * @cnt: The maximum number of bytes to read
7666 * @ppos: The current "file" position
7668 * This function implements the read interface for a struct trace_min_max_param.
7669 * The filp->private_data must point to a trace_min_max_param struct with valid
7673 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7675 struct trace_min_max_param *param = filp->private_data;
7676 char buf[U64_STR_SIZE];
7685 if (cnt > sizeof(buf))
7688 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7690 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7693 const struct file_operations trace_min_max_fops = {
7694 .open = tracing_open_generic,
7695 .read = trace_min_max_read,
7696 .write = trace_min_max_write,
7699 #define TRACING_LOG_ERRS_MAX 8
7700 #define TRACING_LOG_LOC_MAX 128
7702 #define CMD_PREFIX " Command: "
7705 const char **errs; /* ptr to loc-specific array of err strings */
7706 u8 type; /* index into errs -> specific err string */
7707 u8 pos; /* MAX_FILTER_STR_VAL = 256 */
7711 struct tracing_log_err {
7712 struct list_head list;
7713 struct err_info info;
7714 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7715 char cmd[MAX_FILTER_STR_VAL]; /* what caused err */
7718 static DEFINE_MUTEX(tracing_err_log_lock);
7720 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr)
7722 struct tracing_log_err *err;
7724 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7725 err = kzalloc(sizeof(*err), GFP_KERNEL);
7727 err = ERR_PTR(-ENOMEM);
7728 tr->n_err_log_entries++;
7733 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7734 list_del(&err->list);
7740 * err_pos - find the position of a string within a command for error careting
7741 * @cmd: The tracing command that caused the error
7742 * @str: The string to position the caret at within @cmd
7744 * Finds the position of the first occurrence of @str within @cmd. The
7745 * return value can be passed to tracing_log_err() for caret placement
7748 * Returns the index within @cmd of the first occurrence of @str or 0
7749 * if @str was not found.
7751 unsigned int err_pos(char *cmd, const char *str)
7755 if (WARN_ON(!strlen(cmd)))
7758 found = strstr(cmd, str);
7766 * tracing_log_err - write an error to the tracing error log
7767 * @tr: The associated trace array for the error (NULL for top level array)
7768 * @loc: A string describing where the error occurred
7769 * @cmd: The tracing command that caused the error
7770 * @errs: The array of loc-specific static error strings
7771 * @type: The index into errs[], which produces the specific static err string
7772 * @pos: The position the caret should be placed in the cmd
7774 * Writes an error into tracing/error_log of the form:
7776 * <loc>: error: <text>
7780 * tracing/error_log is a small log file containing the last
7781 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7782 * unless there has been a tracing error, and the error log can be
7783 * cleared and have its memory freed by writing the empty string in
7784 * truncation mode to it i.e. echo > tracing/error_log.
7786 * NOTE: the @errs array along with the @type param are used to
7787 * produce a static error string - this string is not copied and saved
7788 * when the error is logged - only a pointer to it is saved. See
7789 * existing callers for examples of how static strings are typically
7790 * defined for use with tracing_log_err().
7792 void tracing_log_err(struct trace_array *tr,
7793 const char *loc, const char *cmd,
7794 const char **errs, u8 type, u8 pos)
7796 struct tracing_log_err *err;
7801 mutex_lock(&tracing_err_log_lock);
7802 err = get_tracing_log_err(tr);
7803 if (PTR_ERR(err) == -ENOMEM) {
7804 mutex_unlock(&tracing_err_log_lock);
7808 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7809 snprintf(err->cmd, MAX_FILTER_STR_VAL,"\n" CMD_PREFIX "%s\n", cmd);
7811 err->info.errs = errs;
7812 err->info.type = type;
7813 err->info.pos = pos;
7814 err->info.ts = local_clock();
7816 list_add_tail(&err->list, &tr->err_log);
7817 mutex_unlock(&tracing_err_log_lock);
7820 static void clear_tracing_err_log(struct trace_array *tr)
7822 struct tracing_log_err *err, *next;
7824 mutex_lock(&tracing_err_log_lock);
7825 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7826 list_del(&err->list);
7830 tr->n_err_log_entries = 0;
7831 mutex_unlock(&tracing_err_log_lock);
7834 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7836 struct trace_array *tr = m->private;
7838 mutex_lock(&tracing_err_log_lock);
7840 return seq_list_start(&tr->err_log, *pos);
7843 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7845 struct trace_array *tr = m->private;
7847 return seq_list_next(v, &tr->err_log, pos);
7850 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7852 mutex_unlock(&tracing_err_log_lock);
7855 static void tracing_err_log_show_pos(struct seq_file *m, u8 pos)
7859 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7861 for (i = 0; i < pos; i++)
7866 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7868 struct tracing_log_err *err = v;
7871 const char *err_text = err->info.errs[err->info.type];
7872 u64 sec = err->info.ts;
7875 nsec = do_div(sec, NSEC_PER_SEC);
7876 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7877 err->loc, err_text);
7878 seq_printf(m, "%s", err->cmd);
7879 tracing_err_log_show_pos(m, err->info.pos);
7885 static const struct seq_operations tracing_err_log_seq_ops = {
7886 .start = tracing_err_log_seq_start,
7887 .next = tracing_err_log_seq_next,
7888 .stop = tracing_err_log_seq_stop,
7889 .show = tracing_err_log_seq_show
7892 static int tracing_err_log_open(struct inode *inode, struct file *file)
7894 struct trace_array *tr = inode->i_private;
7897 ret = tracing_check_open_get_tr(tr);
7901 /* If this file was opened for write, then erase contents */
7902 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
7903 clear_tracing_err_log(tr);
7905 if (file->f_mode & FMODE_READ) {
7906 ret = seq_open(file, &tracing_err_log_seq_ops);
7908 struct seq_file *m = file->private_data;
7911 trace_array_put(tr);
7917 static ssize_t tracing_err_log_write(struct file *file,
7918 const char __user *buffer,
7919 size_t count, loff_t *ppos)
7924 static int tracing_err_log_release(struct inode *inode, struct file *file)
7926 struct trace_array *tr = inode->i_private;
7928 trace_array_put(tr);
7930 if (file->f_mode & FMODE_READ)
7931 seq_release(inode, file);
7936 static const struct file_operations tracing_err_log_fops = {
7937 .open = tracing_err_log_open,
7938 .write = tracing_err_log_write,
7940 .llseek = seq_lseek,
7941 .release = tracing_err_log_release,
7944 static int tracing_buffers_open(struct inode *inode, struct file *filp)
7946 struct trace_array *tr = inode->i_private;
7947 struct ftrace_buffer_info *info;
7950 ret = tracing_check_open_get_tr(tr);
7954 info = kvzalloc(sizeof(*info), GFP_KERNEL);
7956 trace_array_put(tr);
7960 mutex_lock(&trace_types_lock);
7963 info->iter.cpu_file = tracing_get_cpu(inode);
7964 info->iter.trace = tr->current_trace;
7965 info->iter.array_buffer = &tr->array_buffer;
7967 /* Force reading ring buffer for first read */
7968 info->read = (unsigned int)-1;
7970 filp->private_data = info;
7974 mutex_unlock(&trace_types_lock);
7976 ret = nonseekable_open(inode, filp);
7978 trace_array_put(tr);
7984 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
7986 struct ftrace_buffer_info *info = filp->private_data;
7987 struct trace_iterator *iter = &info->iter;
7989 return trace_poll(iter, filp, poll_table);
7993 tracing_buffers_read(struct file *filp, char __user *ubuf,
7994 size_t count, loff_t *ppos)
7996 struct ftrace_buffer_info *info = filp->private_data;
7997 struct trace_iterator *iter = &info->iter;
8004 #ifdef CONFIG_TRACER_MAX_TRACE
8005 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8010 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8012 if (IS_ERR(info->spare)) {
8013 ret = PTR_ERR(info->spare);
8016 info->spare_cpu = iter->cpu_file;
8022 /* Do we have previous read data to read? */
8023 if (info->read < PAGE_SIZE)
8027 trace_access_lock(iter->cpu_file);
8028 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8032 trace_access_unlock(iter->cpu_file);
8035 if (trace_empty(iter)) {
8036 if ((filp->f_flags & O_NONBLOCK))
8039 ret = wait_on_pipe(iter, 0);
8050 size = PAGE_SIZE - info->read;
8054 ret = copy_to_user(ubuf, info->spare + info->read, size);
8066 static int tracing_buffers_release(struct inode *inode, struct file *file)
8068 struct ftrace_buffer_info *info = file->private_data;
8069 struct trace_iterator *iter = &info->iter;
8071 mutex_lock(&trace_types_lock);
8073 iter->tr->trace_ref--;
8075 __trace_array_put(iter->tr);
8078 ring_buffer_free_read_page(iter->array_buffer->buffer,
8079 info->spare_cpu, info->spare);
8082 mutex_unlock(&trace_types_lock);
8088 struct trace_buffer *buffer;
8091 refcount_t refcount;
8094 static void buffer_ref_release(struct buffer_ref *ref)
8096 if (!refcount_dec_and_test(&ref->refcount))
8098 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8102 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8103 struct pipe_buffer *buf)
8105 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8107 buffer_ref_release(ref);
8111 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8112 struct pipe_buffer *buf)
8114 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8116 if (refcount_read(&ref->refcount) > INT_MAX/2)
8119 refcount_inc(&ref->refcount);
8123 /* Pipe buffer operations for a buffer. */
8124 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8125 .release = buffer_pipe_buf_release,
8126 .get = buffer_pipe_buf_get,
8130 * Callback from splice_to_pipe(), if we need to release some pages
8131 * at the end of the spd in case we error'ed out in filling the pipe.
8133 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8135 struct buffer_ref *ref =
8136 (struct buffer_ref *)spd->partial[i].private;
8138 buffer_ref_release(ref);
8139 spd->partial[i].private = 0;
8143 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8144 struct pipe_inode_info *pipe, size_t len,
8147 struct ftrace_buffer_info *info = file->private_data;
8148 struct trace_iterator *iter = &info->iter;
8149 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8150 struct page *pages_def[PIPE_DEF_BUFFERS];
8151 struct splice_pipe_desc spd = {
8153 .partial = partial_def,
8154 .nr_pages_max = PIPE_DEF_BUFFERS,
8155 .ops = &buffer_pipe_buf_ops,
8156 .spd_release = buffer_spd_release,
8158 struct buffer_ref *ref;
8162 #ifdef CONFIG_TRACER_MAX_TRACE
8163 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8167 if (*ppos & (PAGE_SIZE - 1))
8170 if (len & (PAGE_SIZE - 1)) {
8171 if (len < PAGE_SIZE)
8176 if (splice_grow_spd(pipe, &spd))
8180 trace_access_lock(iter->cpu_file);
8181 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8183 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
8187 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8193 refcount_set(&ref->refcount, 1);
8194 ref->buffer = iter->array_buffer->buffer;
8195 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8196 if (IS_ERR(ref->page)) {
8197 ret = PTR_ERR(ref->page);
8202 ref->cpu = iter->cpu_file;
8204 r = ring_buffer_read_page(ref->buffer, &ref->page,
8205 len, iter->cpu_file, 1);
8207 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8213 page = virt_to_page(ref->page);
8215 spd.pages[i] = page;
8216 spd.partial[i].len = PAGE_SIZE;
8217 spd.partial[i].offset = 0;
8218 spd.partial[i].private = (unsigned long)ref;
8222 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8225 trace_access_unlock(iter->cpu_file);
8228 /* did we read anything? */
8229 if (!spd.nr_pages) {
8234 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8237 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
8244 ret = splice_to_pipe(pipe, &spd);
8246 splice_shrink_spd(&spd);
8251 static const struct file_operations tracing_buffers_fops = {
8252 .open = tracing_buffers_open,
8253 .read = tracing_buffers_read,
8254 .poll = tracing_buffers_poll,
8255 .release = tracing_buffers_release,
8256 .splice_read = tracing_buffers_splice_read,
8257 .llseek = no_llseek,
8261 tracing_stats_read(struct file *filp, char __user *ubuf,
8262 size_t count, loff_t *ppos)
8264 struct inode *inode = file_inode(filp);
8265 struct trace_array *tr = inode->i_private;
8266 struct array_buffer *trace_buf = &tr->array_buffer;
8267 int cpu = tracing_get_cpu(inode);
8268 struct trace_seq *s;
8270 unsigned long long t;
8271 unsigned long usec_rem;
8273 s = kmalloc(sizeof(*s), GFP_KERNEL);
8279 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8280 trace_seq_printf(s, "entries: %ld\n", cnt);
8282 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8283 trace_seq_printf(s, "overrun: %ld\n", cnt);
8285 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8286 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8288 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8289 trace_seq_printf(s, "bytes: %ld\n", cnt);
8291 if (trace_clocks[tr->clock_id].in_ns) {
8292 /* local or global for trace_clock */
8293 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8294 usec_rem = do_div(t, USEC_PER_SEC);
8295 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8298 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8299 usec_rem = do_div(t, USEC_PER_SEC);
8300 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8302 /* counter or tsc mode for trace_clock */
8303 trace_seq_printf(s, "oldest event ts: %llu\n",
8304 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8306 trace_seq_printf(s, "now ts: %llu\n",
8307 ring_buffer_time_stamp(trace_buf->buffer));
8310 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8311 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8313 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8314 trace_seq_printf(s, "read events: %ld\n", cnt);
8316 count = simple_read_from_buffer(ubuf, count, ppos,
8317 s->buffer, trace_seq_used(s));
8324 static const struct file_operations tracing_stats_fops = {
8325 .open = tracing_open_generic_tr,
8326 .read = tracing_stats_read,
8327 .llseek = generic_file_llseek,
8328 .release = tracing_release_generic_tr,
8331 #ifdef CONFIG_DYNAMIC_FTRACE
8334 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8335 size_t cnt, loff_t *ppos)
8341 /* 256 should be plenty to hold the amount needed */
8342 buf = kmalloc(256, GFP_KERNEL);
8346 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8347 ftrace_update_tot_cnt,
8348 ftrace_number_of_pages,
8349 ftrace_number_of_groups);
8351 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8356 static const struct file_operations tracing_dyn_info_fops = {
8357 .open = tracing_open_generic,
8358 .read = tracing_read_dyn_info,
8359 .llseek = generic_file_llseek,
8361 #endif /* CONFIG_DYNAMIC_FTRACE */
8363 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8365 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8366 struct trace_array *tr, struct ftrace_probe_ops *ops,
8369 tracing_snapshot_instance(tr);
8373 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8374 struct trace_array *tr, struct ftrace_probe_ops *ops,
8377 struct ftrace_func_mapper *mapper = data;
8381 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8391 tracing_snapshot_instance(tr);
8395 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8396 struct ftrace_probe_ops *ops, void *data)
8398 struct ftrace_func_mapper *mapper = data;
8401 seq_printf(m, "%ps:", (void *)ip);
8403 seq_puts(m, "snapshot");
8406 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8409 seq_printf(m, ":count=%ld\n", *count);
8411 seq_puts(m, ":unlimited\n");
8417 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8418 unsigned long ip, void *init_data, void **data)
8420 struct ftrace_func_mapper *mapper = *data;
8423 mapper = allocate_ftrace_func_mapper();
8429 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8433 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8434 unsigned long ip, void *data)
8436 struct ftrace_func_mapper *mapper = data;
8441 free_ftrace_func_mapper(mapper, NULL);
8445 ftrace_func_mapper_remove_ip(mapper, ip);
8448 static struct ftrace_probe_ops snapshot_probe_ops = {
8449 .func = ftrace_snapshot,
8450 .print = ftrace_snapshot_print,
8453 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8454 .func = ftrace_count_snapshot,
8455 .print = ftrace_snapshot_print,
8456 .init = ftrace_snapshot_init,
8457 .free = ftrace_snapshot_free,
8461 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8462 char *glob, char *cmd, char *param, int enable)
8464 struct ftrace_probe_ops *ops;
8465 void *count = (void *)-1;
8472 /* hash funcs only work with set_ftrace_filter */
8476 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8479 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8484 number = strsep(¶m, ":");
8486 if (!strlen(number))
8490 * We use the callback data field (which is a pointer)
8493 ret = kstrtoul(number, 0, (unsigned long *)&count);
8498 ret = tracing_alloc_snapshot_instance(tr);
8502 ret = register_ftrace_function_probe(glob, tr, ops, count);
8505 return ret < 0 ? ret : 0;
8508 static struct ftrace_func_command ftrace_snapshot_cmd = {
8510 .func = ftrace_trace_snapshot_callback,
8513 static __init int register_snapshot_cmd(void)
8515 return register_ftrace_command(&ftrace_snapshot_cmd);
8518 static inline __init int register_snapshot_cmd(void) { return 0; }
8519 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8521 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8523 if (WARN_ON(!tr->dir))
8524 return ERR_PTR(-ENODEV);
8526 /* Top directory uses NULL as the parent */
8527 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8530 /* All sub buffers have a descriptor */
8534 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8536 struct dentry *d_tracer;
8539 return tr->percpu_dir;
8541 d_tracer = tracing_get_dentry(tr);
8542 if (IS_ERR(d_tracer))
8545 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8547 MEM_FAIL(!tr->percpu_dir,
8548 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8550 return tr->percpu_dir;
8553 static struct dentry *
8554 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8555 void *data, long cpu, const struct file_operations *fops)
8557 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8559 if (ret) /* See tracing_get_cpu() */
8560 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8565 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8567 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8568 struct dentry *d_cpu;
8569 char cpu_dir[30]; /* 30 characters should be more than enough */
8574 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8575 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8577 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8581 /* per cpu trace_pipe */
8582 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
8583 tr, cpu, &tracing_pipe_fops);
8586 trace_create_cpu_file("trace", 0644, d_cpu,
8587 tr, cpu, &tracing_fops);
8589 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
8590 tr, cpu, &tracing_buffers_fops);
8592 trace_create_cpu_file("stats", 0444, d_cpu,
8593 tr, cpu, &tracing_stats_fops);
8595 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
8596 tr, cpu, &tracing_entries_fops);
8598 #ifdef CONFIG_TRACER_SNAPSHOT
8599 trace_create_cpu_file("snapshot", 0644, d_cpu,
8600 tr, cpu, &snapshot_fops);
8602 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
8603 tr, cpu, &snapshot_raw_fops);
8607 #ifdef CONFIG_FTRACE_SELFTEST
8608 /* Let selftest have access to static functions in this file */
8609 #include "trace_selftest.c"
8613 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8616 struct trace_option_dentry *topt = filp->private_data;
8619 if (topt->flags->val & topt->opt->bit)
8624 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8628 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8631 struct trace_option_dentry *topt = filp->private_data;
8635 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8639 if (val != 0 && val != 1)
8642 if (!!(topt->flags->val & topt->opt->bit) != val) {
8643 mutex_lock(&trace_types_lock);
8644 ret = __set_tracer_option(topt->tr, topt->flags,
8646 mutex_unlock(&trace_types_lock);
8657 static const struct file_operations trace_options_fops = {
8658 .open = tracing_open_generic,
8659 .read = trace_options_read,
8660 .write = trace_options_write,
8661 .llseek = generic_file_llseek,
8665 * In order to pass in both the trace_array descriptor as well as the index
8666 * to the flag that the trace option file represents, the trace_array
8667 * has a character array of trace_flags_index[], which holds the index
8668 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8669 * The address of this character array is passed to the flag option file
8670 * read/write callbacks.
8672 * In order to extract both the index and the trace_array descriptor,
8673 * get_tr_index() uses the following algorithm.
8677 * As the pointer itself contains the address of the index (remember
8680 * Then to get the trace_array descriptor, by subtracting that index
8681 * from the ptr, we get to the start of the index itself.
8683 * ptr - idx == &index[0]
8685 * Then a simple container_of() from that pointer gets us to the
8686 * trace_array descriptor.
8688 static void get_tr_index(void *data, struct trace_array **ptr,
8689 unsigned int *pindex)
8691 *pindex = *(unsigned char *)data;
8693 *ptr = container_of(data - *pindex, struct trace_array,
8698 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8701 void *tr_index = filp->private_data;
8702 struct trace_array *tr;
8706 get_tr_index(tr_index, &tr, &index);
8708 if (tr->trace_flags & (1 << index))
8713 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8717 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8720 void *tr_index = filp->private_data;
8721 struct trace_array *tr;
8726 get_tr_index(tr_index, &tr, &index);
8728 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8732 if (val != 0 && val != 1)
8735 mutex_lock(&event_mutex);
8736 mutex_lock(&trace_types_lock);
8737 ret = set_tracer_flag(tr, 1 << index, val);
8738 mutex_unlock(&trace_types_lock);
8739 mutex_unlock(&event_mutex);
8749 static const struct file_operations trace_options_core_fops = {
8750 .open = tracing_open_generic,
8751 .read = trace_options_core_read,
8752 .write = trace_options_core_write,
8753 .llseek = generic_file_llseek,
8756 struct dentry *trace_create_file(const char *name,
8758 struct dentry *parent,
8760 const struct file_operations *fops)
8764 ret = tracefs_create_file(name, mode, parent, data, fops);
8766 pr_warn("Could not create tracefs '%s' entry\n", name);
8772 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8774 struct dentry *d_tracer;
8779 d_tracer = tracing_get_dentry(tr);
8780 if (IS_ERR(d_tracer))
8783 tr->options = tracefs_create_dir("options", d_tracer);
8785 pr_warn("Could not create tracefs directory 'options'\n");
8793 create_trace_option_file(struct trace_array *tr,
8794 struct trace_option_dentry *topt,
8795 struct tracer_flags *flags,
8796 struct tracer_opt *opt)
8798 struct dentry *t_options;
8800 t_options = trace_options_init_dentry(tr);
8804 topt->flags = flags;
8808 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
8809 &trace_options_fops);
8814 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8816 struct trace_option_dentry *topts;
8817 struct trace_options *tr_topts;
8818 struct tracer_flags *flags;
8819 struct tracer_opt *opts;
8826 flags = tracer->flags;
8828 if (!flags || !flags->opts)
8832 * If this is an instance, only create flags for tracers
8833 * the instance may have.
8835 if (!trace_ok_for_array(tracer, tr))
8838 for (i = 0; i < tr->nr_topts; i++) {
8839 /* Make sure there's no duplicate flags. */
8840 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8846 for (cnt = 0; opts[cnt].name; cnt++)
8849 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8853 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
8860 tr->topts = tr_topts;
8861 tr->topts[tr->nr_topts].tracer = tracer;
8862 tr->topts[tr->nr_topts].topts = topts;
8865 for (cnt = 0; opts[cnt].name; cnt++) {
8866 create_trace_option_file(tr, &topts[cnt], flags,
8868 MEM_FAIL(topts[cnt].entry == NULL,
8869 "Failed to create trace option: %s",
8874 static struct dentry *
8875 create_trace_option_core_file(struct trace_array *tr,
8876 const char *option, long index)
8878 struct dentry *t_options;
8880 t_options = trace_options_init_dentry(tr);
8884 return trace_create_file(option, 0644, t_options,
8885 (void *)&tr->trace_flags_index[index],
8886 &trace_options_core_fops);
8889 static void create_trace_options_dir(struct trace_array *tr)
8891 struct dentry *t_options;
8892 bool top_level = tr == &global_trace;
8895 t_options = trace_options_init_dentry(tr);
8899 for (i = 0; trace_options[i]; i++) {
8901 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
8902 create_trace_option_core_file(tr, trace_options[i], i);
8907 rb_simple_read(struct file *filp, char __user *ubuf,
8908 size_t cnt, loff_t *ppos)
8910 struct trace_array *tr = filp->private_data;
8914 r = tracer_tracing_is_on(tr);
8915 r = sprintf(buf, "%d\n", r);
8917 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8921 rb_simple_write(struct file *filp, const char __user *ubuf,
8922 size_t cnt, loff_t *ppos)
8924 struct trace_array *tr = filp->private_data;
8925 struct trace_buffer *buffer = tr->array_buffer.buffer;
8929 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8934 mutex_lock(&trace_types_lock);
8935 if (!!val == tracer_tracing_is_on(tr)) {
8936 val = 0; /* do nothing */
8938 tracer_tracing_on(tr);
8939 if (tr->current_trace->start)
8940 tr->current_trace->start(tr);
8942 tracer_tracing_off(tr);
8943 if (tr->current_trace->stop)
8944 tr->current_trace->stop(tr);
8946 mutex_unlock(&trace_types_lock);
8954 static const struct file_operations rb_simple_fops = {
8955 .open = tracing_open_generic_tr,
8956 .read = rb_simple_read,
8957 .write = rb_simple_write,
8958 .release = tracing_release_generic_tr,
8959 .llseek = default_llseek,
8963 buffer_percent_read(struct file *filp, char __user *ubuf,
8964 size_t cnt, loff_t *ppos)
8966 struct trace_array *tr = filp->private_data;
8970 r = tr->buffer_percent;
8971 r = sprintf(buf, "%d\n", r);
8973 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8977 buffer_percent_write(struct file *filp, const char __user *ubuf,
8978 size_t cnt, loff_t *ppos)
8980 struct trace_array *tr = filp->private_data;
8984 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8994 tr->buffer_percent = val;
9001 static const struct file_operations buffer_percent_fops = {
9002 .open = tracing_open_generic_tr,
9003 .read = buffer_percent_read,
9004 .write = buffer_percent_write,
9005 .release = tracing_release_generic_tr,
9006 .llseek = default_llseek,
9009 static struct dentry *trace_instance_dir;
9012 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9015 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9017 enum ring_buffer_flags rb_flags;
9019 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9023 buf->buffer = ring_buffer_alloc(size, rb_flags);
9027 buf->data = alloc_percpu(struct trace_array_cpu);
9029 ring_buffer_free(buf->buffer);
9034 /* Allocate the first page for all buffers */
9035 set_buffer_entries(&tr->array_buffer,
9036 ring_buffer_size(tr->array_buffer.buffer, 0));
9041 static int allocate_trace_buffers(struct trace_array *tr, int size)
9045 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9049 #ifdef CONFIG_TRACER_MAX_TRACE
9050 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9051 allocate_snapshot ? size : 1);
9052 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9053 ring_buffer_free(tr->array_buffer.buffer);
9054 tr->array_buffer.buffer = NULL;
9055 free_percpu(tr->array_buffer.data);
9056 tr->array_buffer.data = NULL;
9059 tr->allocated_snapshot = allocate_snapshot;
9062 * Only the top level trace array gets its snapshot allocated
9063 * from the kernel command line.
9065 allocate_snapshot = false;
9071 static void free_trace_buffer(struct array_buffer *buf)
9074 ring_buffer_free(buf->buffer);
9076 free_percpu(buf->data);
9081 static void free_trace_buffers(struct trace_array *tr)
9086 free_trace_buffer(&tr->array_buffer);
9088 #ifdef CONFIG_TRACER_MAX_TRACE
9089 free_trace_buffer(&tr->max_buffer);
9093 static void init_trace_flags_index(struct trace_array *tr)
9097 /* Used by the trace options files */
9098 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9099 tr->trace_flags_index[i] = i;
9102 static void __update_tracer_options(struct trace_array *tr)
9106 for (t = trace_types; t; t = t->next)
9107 add_tracer_options(tr, t);
9110 static void update_tracer_options(struct trace_array *tr)
9112 mutex_lock(&trace_types_lock);
9113 __update_tracer_options(tr);
9114 mutex_unlock(&trace_types_lock);
9117 /* Must have trace_types_lock held */
9118 struct trace_array *trace_array_find(const char *instance)
9120 struct trace_array *tr, *found = NULL;
9122 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9123 if (tr->name && strcmp(tr->name, instance) == 0) {
9132 struct trace_array *trace_array_find_get(const char *instance)
9134 struct trace_array *tr;
9136 mutex_lock(&trace_types_lock);
9137 tr = trace_array_find(instance);
9140 mutex_unlock(&trace_types_lock);
9145 static int trace_array_create_dir(struct trace_array *tr)
9149 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9153 ret = event_trace_add_tracer(tr->dir, tr);
9155 tracefs_remove(tr->dir);
9159 init_tracer_tracefs(tr, tr->dir);
9160 __update_tracer_options(tr);
9165 static struct trace_array *trace_array_create(const char *name)
9167 struct trace_array *tr;
9171 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9173 return ERR_PTR(ret);
9175 tr->name = kstrdup(name, GFP_KERNEL);
9179 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9182 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9184 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9186 raw_spin_lock_init(&tr->start_lock);
9188 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9190 tr->current_trace = &nop_trace;
9192 INIT_LIST_HEAD(&tr->systems);
9193 INIT_LIST_HEAD(&tr->events);
9194 INIT_LIST_HEAD(&tr->hist_vars);
9195 INIT_LIST_HEAD(&tr->err_log);
9197 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9200 if (ftrace_allocate_ftrace_ops(tr) < 0)
9203 ftrace_init_trace_array(tr);
9205 init_trace_flags_index(tr);
9207 if (trace_instance_dir) {
9208 ret = trace_array_create_dir(tr);
9212 __trace_early_add_events(tr);
9214 list_add(&tr->list, &ftrace_trace_arrays);
9221 ftrace_free_ftrace_ops(tr);
9222 free_trace_buffers(tr);
9223 free_cpumask_var(tr->tracing_cpumask);
9227 return ERR_PTR(ret);
9230 static int instance_mkdir(const char *name)
9232 struct trace_array *tr;
9235 mutex_lock(&event_mutex);
9236 mutex_lock(&trace_types_lock);
9239 if (trace_array_find(name))
9242 tr = trace_array_create(name);
9244 ret = PTR_ERR_OR_ZERO(tr);
9247 mutex_unlock(&trace_types_lock);
9248 mutex_unlock(&event_mutex);
9253 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9254 * @name: The name of the trace array to be looked up/created.
9256 * Returns pointer to trace array with given name.
9257 * NULL, if it cannot be created.
9259 * NOTE: This function increments the reference counter associated with the
9260 * trace array returned. This makes sure it cannot be freed while in use.
9261 * Use trace_array_put() once the trace array is no longer needed.
9262 * If the trace_array is to be freed, trace_array_destroy() needs to
9263 * be called after the trace_array_put(), or simply let user space delete
9264 * it from the tracefs instances directory. But until the
9265 * trace_array_put() is called, user space can not delete it.
9268 struct trace_array *trace_array_get_by_name(const char *name)
9270 struct trace_array *tr;
9272 mutex_lock(&event_mutex);
9273 mutex_lock(&trace_types_lock);
9275 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9276 if (tr->name && strcmp(tr->name, name) == 0)
9280 tr = trace_array_create(name);
9288 mutex_unlock(&trace_types_lock);
9289 mutex_unlock(&event_mutex);
9292 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9294 static int __remove_instance(struct trace_array *tr)
9298 /* Reference counter for a newly created trace array = 1. */
9299 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9302 list_del(&tr->list);
9304 /* Disable all the flags that were enabled coming in */
9305 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9306 if ((1 << i) & ZEROED_TRACE_FLAGS)
9307 set_tracer_flag(tr, 1 << i, 0);
9310 tracing_set_nop(tr);
9311 clear_ftrace_function_probes(tr);
9312 event_trace_del_tracer(tr);
9313 ftrace_clear_pids(tr);
9314 ftrace_destroy_function_files(tr);
9315 tracefs_remove(tr->dir);
9316 free_percpu(tr->last_func_repeats);
9317 free_trace_buffers(tr);
9319 for (i = 0; i < tr->nr_topts; i++) {
9320 kfree(tr->topts[i].topts);
9324 free_cpumask_var(tr->tracing_cpumask);
9331 int trace_array_destroy(struct trace_array *this_tr)
9333 struct trace_array *tr;
9339 mutex_lock(&event_mutex);
9340 mutex_lock(&trace_types_lock);
9344 /* Making sure trace array exists before destroying it. */
9345 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9346 if (tr == this_tr) {
9347 ret = __remove_instance(tr);
9352 mutex_unlock(&trace_types_lock);
9353 mutex_unlock(&event_mutex);
9357 EXPORT_SYMBOL_GPL(trace_array_destroy);
9359 static int instance_rmdir(const char *name)
9361 struct trace_array *tr;
9364 mutex_lock(&event_mutex);
9365 mutex_lock(&trace_types_lock);
9368 tr = trace_array_find(name);
9370 ret = __remove_instance(tr);
9372 mutex_unlock(&trace_types_lock);
9373 mutex_unlock(&event_mutex);
9378 static __init void create_trace_instances(struct dentry *d_tracer)
9380 struct trace_array *tr;
9382 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9385 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9388 mutex_lock(&event_mutex);
9389 mutex_lock(&trace_types_lock);
9391 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9394 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9395 "Failed to create instance directory\n"))
9399 mutex_unlock(&trace_types_lock);
9400 mutex_unlock(&event_mutex);
9404 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9406 struct trace_event_file *file;
9409 trace_create_file("available_tracers", 0444, d_tracer,
9410 tr, &show_traces_fops);
9412 trace_create_file("current_tracer", 0644, d_tracer,
9413 tr, &set_tracer_fops);
9415 trace_create_file("tracing_cpumask", 0644, d_tracer,
9416 tr, &tracing_cpumask_fops);
9418 trace_create_file("trace_options", 0644, d_tracer,
9419 tr, &tracing_iter_fops);
9421 trace_create_file("trace", 0644, d_tracer,
9424 trace_create_file("trace_pipe", 0444, d_tracer,
9425 tr, &tracing_pipe_fops);
9427 trace_create_file("buffer_size_kb", 0644, d_tracer,
9428 tr, &tracing_entries_fops);
9430 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
9431 tr, &tracing_total_entries_fops);
9433 trace_create_file("free_buffer", 0200, d_tracer,
9434 tr, &tracing_free_buffer_fops);
9436 trace_create_file("trace_marker", 0220, d_tracer,
9437 tr, &tracing_mark_fops);
9439 file = __find_event_file(tr, "ftrace", "print");
9440 if (file && file->dir)
9441 trace_create_file("trigger", 0644, file->dir, file,
9442 &event_trigger_fops);
9443 tr->trace_marker_file = file;
9445 trace_create_file("trace_marker_raw", 0220, d_tracer,
9446 tr, &tracing_mark_raw_fops);
9448 trace_create_file("trace_clock", 0644, d_tracer, tr,
9451 trace_create_file("tracing_on", 0644, d_tracer,
9452 tr, &rb_simple_fops);
9454 trace_create_file("timestamp_mode", 0444, d_tracer, tr,
9455 &trace_time_stamp_mode_fops);
9457 tr->buffer_percent = 50;
9459 trace_create_file("buffer_percent", 0444, d_tracer,
9460 tr, &buffer_percent_fops);
9462 create_trace_options_dir(tr);
9464 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
9465 trace_create_maxlat_file(tr, d_tracer);
9468 if (ftrace_create_function_files(tr, d_tracer))
9469 MEM_FAIL(1, "Could not allocate function filter files");
9471 #ifdef CONFIG_TRACER_SNAPSHOT
9472 trace_create_file("snapshot", 0644, d_tracer,
9473 tr, &snapshot_fops);
9476 trace_create_file("error_log", 0644, d_tracer,
9477 tr, &tracing_err_log_fops);
9479 for_each_tracing_cpu(cpu)
9480 tracing_init_tracefs_percpu(tr, cpu);
9482 ftrace_init_tracefs(tr, d_tracer);
9485 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9487 struct vfsmount *mnt;
9488 struct file_system_type *type;
9491 * To maintain backward compatibility for tools that mount
9492 * debugfs to get to the tracing facility, tracefs is automatically
9493 * mounted to the debugfs/tracing directory.
9495 type = get_fs_type("tracefs");
9498 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9499 put_filesystem(type);
9508 * tracing_init_dentry - initialize top level trace array
9510 * This is called when creating files or directories in the tracing
9511 * directory. It is called via fs_initcall() by any of the boot up code
9512 * and expects to return the dentry of the top level tracing directory.
9514 int tracing_init_dentry(void)
9516 struct trace_array *tr = &global_trace;
9518 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9519 pr_warn("Tracing disabled due to lockdown\n");
9523 /* The top level trace array uses NULL as parent */
9527 if (WARN_ON(!tracefs_initialized()))
9531 * As there may still be users that expect the tracing
9532 * files to exist in debugfs/tracing, we must automount
9533 * the tracefs file system there, so older tools still
9534 * work with the newer kernel.
9536 tr->dir = debugfs_create_automount("tracing", NULL,
9537 trace_automount, NULL);
9542 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9543 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9545 static struct workqueue_struct *eval_map_wq __initdata;
9546 static struct work_struct eval_map_work __initdata;
9548 static void __init eval_map_work_func(struct work_struct *work)
9552 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9553 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9556 static int __init trace_eval_init(void)
9558 INIT_WORK(&eval_map_work, eval_map_work_func);
9560 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9562 pr_err("Unable to allocate eval_map_wq\n");
9564 eval_map_work_func(&eval_map_work);
9568 queue_work(eval_map_wq, &eval_map_work);
9572 static int __init trace_eval_sync(void)
9574 /* Make sure the eval map updates are finished */
9576 destroy_workqueue(eval_map_wq);
9580 late_initcall_sync(trace_eval_sync);
9583 #ifdef CONFIG_MODULES
9584 static void trace_module_add_evals(struct module *mod)
9586 if (!mod->num_trace_evals)
9590 * Modules with bad taint do not have events created, do
9591 * not bother with enums either.
9593 if (trace_module_has_bad_taint(mod))
9596 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9599 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
9600 static void trace_module_remove_evals(struct module *mod)
9602 union trace_eval_map_item *map;
9603 union trace_eval_map_item **last = &trace_eval_maps;
9605 if (!mod->num_trace_evals)
9608 mutex_lock(&trace_eval_mutex);
9610 map = trace_eval_maps;
9613 if (map->head.mod == mod)
9615 map = trace_eval_jmp_to_tail(map);
9616 last = &map->tail.next;
9617 map = map->tail.next;
9622 *last = trace_eval_jmp_to_tail(map)->tail.next;
9625 mutex_unlock(&trace_eval_mutex);
9628 static inline void trace_module_remove_evals(struct module *mod) { }
9629 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9631 static int trace_module_notify(struct notifier_block *self,
9632 unsigned long val, void *data)
9634 struct module *mod = data;
9637 case MODULE_STATE_COMING:
9638 trace_module_add_evals(mod);
9640 case MODULE_STATE_GOING:
9641 trace_module_remove_evals(mod);
9648 static struct notifier_block trace_module_nb = {
9649 .notifier_call = trace_module_notify,
9652 #endif /* CONFIG_MODULES */
9654 static __init int tracer_init_tracefs(void)
9658 trace_access_lock_init();
9660 ret = tracing_init_dentry();
9666 init_tracer_tracefs(&global_trace, NULL);
9667 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9669 trace_create_file("tracing_thresh", 0644, NULL,
9670 &global_trace, &tracing_thresh_fops);
9672 trace_create_file("README", 0444, NULL,
9673 NULL, &tracing_readme_fops);
9675 trace_create_file("saved_cmdlines", 0444, NULL,
9676 NULL, &tracing_saved_cmdlines_fops);
9678 trace_create_file("saved_cmdlines_size", 0644, NULL,
9679 NULL, &tracing_saved_cmdlines_size_fops);
9681 trace_create_file("saved_tgids", 0444, NULL,
9682 NULL, &tracing_saved_tgids_fops);
9686 trace_create_eval_file(NULL);
9688 #ifdef CONFIG_MODULES
9689 register_module_notifier(&trace_module_nb);
9692 #ifdef CONFIG_DYNAMIC_FTRACE
9693 trace_create_file("dyn_ftrace_total_info", 0444, NULL,
9694 NULL, &tracing_dyn_info_fops);
9697 create_trace_instances(NULL);
9699 update_tracer_options(&global_trace);
9704 fs_initcall(tracer_init_tracefs);
9706 static int trace_panic_handler(struct notifier_block *this,
9707 unsigned long event, void *unused)
9709 if (ftrace_dump_on_oops)
9710 ftrace_dump(ftrace_dump_on_oops);
9714 static struct notifier_block trace_panic_notifier = {
9715 .notifier_call = trace_panic_handler,
9717 .priority = 150 /* priority: INT_MAX >= x >= 0 */
9720 static int trace_die_handler(struct notifier_block *self,
9726 if (ftrace_dump_on_oops)
9727 ftrace_dump(ftrace_dump_on_oops);
9735 static struct notifier_block trace_die_notifier = {
9736 .notifier_call = trace_die_handler,
9741 * printk is set to max of 1024, we really don't need it that big.
9742 * Nothing should be printing 1000 characters anyway.
9744 #define TRACE_MAX_PRINT 1000
9747 * Define here KERN_TRACE so that we have one place to modify
9748 * it if we decide to change what log level the ftrace dump
9751 #define KERN_TRACE KERN_EMERG
9754 trace_printk_seq(struct trace_seq *s)
9756 /* Probably should print a warning here. */
9757 if (s->seq.len >= TRACE_MAX_PRINT)
9758 s->seq.len = TRACE_MAX_PRINT;
9761 * More paranoid code. Although the buffer size is set to
9762 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9763 * an extra layer of protection.
9765 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9766 s->seq.len = s->seq.size - 1;
9768 /* should be zero ended, but we are paranoid. */
9769 s->buffer[s->seq.len] = 0;
9771 printk(KERN_TRACE "%s", s->buffer);
9776 void trace_init_global_iter(struct trace_iterator *iter)
9778 iter->tr = &global_trace;
9779 iter->trace = iter->tr->current_trace;
9780 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9781 iter->array_buffer = &global_trace.array_buffer;
9783 if (iter->trace && iter->trace->open)
9784 iter->trace->open(iter);
9786 /* Annotate start of buffers if we had overruns */
9787 if (ring_buffer_overruns(iter->array_buffer->buffer))
9788 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9790 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9791 if (trace_clocks[iter->tr->clock_id].in_ns)
9792 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9795 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
9797 /* use static because iter can be a bit big for the stack */
9798 static struct trace_iterator iter;
9799 static atomic_t dump_running;
9800 struct trace_array *tr = &global_trace;
9801 unsigned int old_userobj;
9802 unsigned long flags;
9805 /* Only allow one dump user at a time. */
9806 if (atomic_inc_return(&dump_running) != 1) {
9807 atomic_dec(&dump_running);
9812 * Always turn off tracing when we dump.
9813 * We don't need to show trace output of what happens
9814 * between multiple crashes.
9816 * If the user does a sysrq-z, then they can re-enable
9817 * tracing with echo 1 > tracing_on.
9821 local_irq_save(flags);
9823 /* Simulate the iterator */
9824 trace_init_global_iter(&iter);
9825 /* Can not use kmalloc for iter.temp and iter.fmt */
9826 iter.temp = static_temp_buf;
9827 iter.temp_size = STATIC_TEMP_BUF_SIZE;
9828 iter.fmt = static_fmt_buf;
9829 iter.fmt_size = STATIC_FMT_BUF_SIZE;
9831 for_each_tracing_cpu(cpu) {
9832 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9835 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
9837 /* don't look at user memory in panic mode */
9838 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
9840 switch (oops_dump_mode) {
9842 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9845 iter.cpu_file = raw_smp_processor_id();
9850 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
9851 iter.cpu_file = RING_BUFFER_ALL_CPUS;
9854 printk(KERN_TRACE "Dumping ftrace buffer:\n");
9856 /* Did function tracer already get disabled? */
9857 if (ftrace_is_dead()) {
9858 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
9859 printk("# MAY BE MISSING FUNCTION EVENTS\n");
9863 * We need to stop all tracing on all CPUS to read
9864 * the next buffer. This is a bit expensive, but is
9865 * not done often. We fill all what we can read,
9866 * and then release the locks again.
9869 while (!trace_empty(&iter)) {
9872 printk(KERN_TRACE "---------------------------------\n");
9876 trace_iterator_reset(&iter);
9877 iter.iter_flags |= TRACE_FILE_LAT_FMT;
9879 if (trace_find_next_entry_inc(&iter) != NULL) {
9882 ret = print_trace_line(&iter);
9883 if (ret != TRACE_TYPE_NO_CONSUME)
9884 trace_consume(&iter);
9886 touch_nmi_watchdog();
9888 trace_printk_seq(&iter.seq);
9892 printk(KERN_TRACE " (ftrace buffer empty)\n");
9894 printk(KERN_TRACE "---------------------------------\n");
9897 tr->trace_flags |= old_userobj;
9899 for_each_tracing_cpu(cpu) {
9900 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9902 atomic_dec(&dump_running);
9903 local_irq_restore(flags);
9905 EXPORT_SYMBOL_GPL(ftrace_dump);
9907 #define WRITE_BUFSIZE 4096
9909 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
9910 size_t count, loff_t *ppos,
9911 int (*createfn)(const char *))
9913 char *kbuf, *buf, *tmp;
9918 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
9922 while (done < count) {
9923 size = count - done;
9925 if (size >= WRITE_BUFSIZE)
9926 size = WRITE_BUFSIZE - 1;
9928 if (copy_from_user(kbuf, buffer + done, size)) {
9935 tmp = strchr(buf, '\n');
9938 size = tmp - buf + 1;
9941 if (done + size < count) {
9944 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
9945 pr_warn("Line length is too long: Should be less than %d\n",
9953 /* Remove comments */
9954 tmp = strchr(buf, '#');
9959 ret = createfn(buf);
9964 } while (done < count);
9974 __init static int tracer_alloc_buffers(void)
9980 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9981 pr_warn("Tracing disabled due to lockdown\n");
9986 * Make sure we don't accidentally add more trace options
9987 * than we have bits for.
9989 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
9991 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
9994 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
9995 goto out_free_buffer_mask;
9997 /* Only allocate trace_printk buffers if a trace_printk exists */
9998 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
9999 /* Must be called before global_trace.buffer is allocated */
10000 trace_printk_init_buffers();
10002 /* To save memory, keep the ring buffer size to its minimum */
10003 if (ring_buffer_expanded)
10004 ring_buf_size = trace_buf_size;
10008 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10009 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10011 raw_spin_lock_init(&global_trace.start_lock);
10014 * The prepare callbacks allocates some memory for the ring buffer. We
10015 * don't free the buffer if the CPU goes down. If we were to free
10016 * the buffer, then the user would lose any trace that was in the
10017 * buffer. The memory will be removed once the "instance" is removed.
10019 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10020 "trace/RB:preapre", trace_rb_cpu_prepare,
10023 goto out_free_cpumask;
10024 /* Used for event triggers */
10026 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10028 goto out_rm_hp_state;
10030 if (trace_create_savedcmd() < 0)
10031 goto out_free_temp_buffer;
10033 /* TODO: make the number of buffers hot pluggable with CPUS */
10034 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10035 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10036 goto out_free_savedcmd;
10039 if (global_trace.buffer_disabled)
10042 if (trace_boot_clock) {
10043 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10045 pr_warn("Trace clock %s not defined, going back to default\n",
10050 * register_tracer() might reference current_trace, so it
10051 * needs to be set before we register anything. This is
10052 * just a bootstrap of current_trace anyway.
10054 global_trace.current_trace = &nop_trace;
10056 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10058 ftrace_init_global_array_ops(&global_trace);
10060 init_trace_flags_index(&global_trace);
10062 register_tracer(&nop_trace);
10064 /* Function tracing may start here (via kernel command line) */
10065 init_function_trace();
10067 /* All seems OK, enable tracing */
10068 tracing_disabled = 0;
10070 atomic_notifier_chain_register(&panic_notifier_list,
10071 &trace_panic_notifier);
10073 register_die_notifier(&trace_die_notifier);
10075 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10077 INIT_LIST_HEAD(&global_trace.systems);
10078 INIT_LIST_HEAD(&global_trace.events);
10079 INIT_LIST_HEAD(&global_trace.hist_vars);
10080 INIT_LIST_HEAD(&global_trace.err_log);
10081 list_add(&global_trace.list, &ftrace_trace_arrays);
10083 apply_trace_boot_options();
10085 register_snapshot_cmd();
10092 free_saved_cmdlines_buffer(savedcmd);
10093 out_free_temp_buffer:
10094 ring_buffer_free(temp_buffer);
10096 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10098 free_cpumask_var(global_trace.tracing_cpumask);
10099 out_free_buffer_mask:
10100 free_cpumask_var(tracing_buffer_mask);
10105 void __init early_trace_init(void)
10107 if (tracepoint_printk) {
10108 tracepoint_print_iter =
10109 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10110 if (MEM_FAIL(!tracepoint_print_iter,
10111 "Failed to allocate trace iterator\n"))
10112 tracepoint_printk = 0;
10114 static_key_enable(&tracepoint_printk_key.key);
10116 tracer_alloc_buffers();
10119 void __init trace_init(void)
10121 trace_event_init();
10124 __init static void clear_boot_tracer(void)
10127 * The default tracer at boot buffer is an init section.
10128 * This function is called in lateinit. If we did not
10129 * find the boot tracer, then clear it out, to prevent
10130 * later registration from accessing the buffer that is
10131 * about to be freed.
10133 if (!default_bootup_tracer)
10136 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10137 default_bootup_tracer);
10138 default_bootup_tracer = NULL;
10141 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10142 __init static void tracing_set_default_clock(void)
10144 /* sched_clock_stable() is determined in late_initcall */
10145 if (!trace_boot_clock && !sched_clock_stable()) {
10146 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10147 pr_warn("Can not set tracing clock due to lockdown\n");
10151 printk(KERN_WARNING
10152 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10153 "If you want to keep using the local clock, then add:\n"
10154 " \"trace_clock=local\"\n"
10155 "on the kernel command line\n");
10156 tracing_set_clock(&global_trace, "global");
10160 static inline void tracing_set_default_clock(void) { }
10163 __init static int late_trace_init(void)
10165 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10166 static_key_disable(&tracepoint_printk_key.key);
10167 tracepoint_printk = 0;
10170 tracing_set_default_clock();
10171 clear_boot_tracer();
10175 late_initcall_sync(late_trace_init);