1 // SPDX-License-Identifier: GPL-2.0
3 * ring buffer based function tracer
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
15 #include <linux/ring_buffer.h>
16 #include <generated/utsrelease.h>
17 #include <linux/stacktrace.h>
18 #include <linux/writeback.h>
19 #include <linux/kallsyms.h>
20 #include <linux/security.h>
21 #include <linux/seq_file.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/tracefs.h>
25 #include <linux/pagemap.h>
26 #include <linux/hardirq.h>
27 #include <linux/linkage.h>
28 #include <linux/uaccess.h>
29 #include <linux/vmalloc.h>
30 #include <linux/ftrace.h>
31 #include <linux/module.h>
32 #include <linux/percpu.h>
33 #include <linux/splice.h>
34 #include <linux/kdebug.h>
35 #include <linux/string.h>
36 #include <linux/mount.h>
37 #include <linux/rwsem.h>
38 #include <linux/slab.h>
39 #include <linux/ctype.h>
40 #include <linux/init.h>
41 #include <linux/panic_notifier.h>
42 #include <linux/poll.h>
43 #include <linux/nmi.h>
45 #include <linux/trace.h>
46 #include <linux/sched/clock.h>
47 #include <linux/sched/rt.h>
48 #include <linux/fsnotify.h>
49 #include <linux/irq_work.h>
50 #include <linux/workqueue.h>
52 #include <asm/setup.h> /* COMMAND_LINE_SIZE */
55 #include "trace_output.h"
58 * On boot up, the ring buffer is set to the minimum size, so that
59 * we do not waste memory on systems that are not using tracing.
61 bool ring_buffer_expanded;
63 #ifdef CONFIG_FTRACE_STARTUP_TEST
65 * We need to change this state when a selftest is running.
66 * A selftest will lurk into the ring-buffer to count the
67 * entries inserted during the selftest although some concurrent
68 * insertions into the ring-buffer such as trace_printk could occurred
69 * at the same time, giving false positive or negative results.
71 static bool __read_mostly tracing_selftest_running;
74 * If boot-time tracing including tracers/events via kernel cmdline
75 * is running, we do not want to run SELFTEST.
77 bool __read_mostly tracing_selftest_disabled;
79 void __init disable_tracing_selftest(const char *reason)
81 if (!tracing_selftest_disabled) {
82 tracing_selftest_disabled = true;
83 pr_info("Ftrace startup test is disabled due to %s\n", reason);
87 #define tracing_selftest_running 0
88 #define tracing_selftest_disabled 0
91 /* Pipe tracepoints to printk */
92 static struct trace_iterator *tracepoint_print_iter;
93 int tracepoint_printk;
94 static bool tracepoint_printk_stop_on_boot __initdata;
95 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
97 /* For tracers that don't implement custom flags */
98 static struct tracer_opt dummy_tracer_opt[] = {
103 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
109 * To prevent the comm cache from being overwritten when no
110 * tracing is active, only save the comm when a trace event
113 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
116 * Kill all tracing for good (never come back).
117 * It is initialized to 1 but will turn to zero if the initialization
118 * of the tracer is successful. But that is the only place that sets
121 static int tracing_disabled = 1;
123 cpumask_var_t __read_mostly tracing_buffer_mask;
126 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
128 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
129 * is set, then ftrace_dump is called. This will output the contents
130 * of the ftrace buffers to the console. This is very useful for
131 * capturing traces that lead to crashes and outputing it to a
134 * It is default off, but you can enable it with either specifying
135 * "ftrace_dump_on_oops" in the kernel command line, or setting
136 * /proc/sys/kernel/ftrace_dump_on_oops
137 * Set 1 if you want to dump buffers of all CPUs
138 * Set 2 if you want to dump the buffer of the CPU that triggered oops
141 enum ftrace_dump_mode ftrace_dump_on_oops;
143 /* When set, tracing will stop when a WARN*() is hit */
144 int __disable_trace_on_warning;
146 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
147 /* Map of enums to their values, for "eval_map" file */
148 struct trace_eval_map_head {
150 unsigned long length;
153 union trace_eval_map_item;
155 struct trace_eval_map_tail {
157 * "end" is first and points to NULL as it must be different
158 * than "mod" or "eval_string"
160 union trace_eval_map_item *next;
161 const char *end; /* points to NULL */
164 static DEFINE_MUTEX(trace_eval_mutex);
167 * The trace_eval_maps are saved in an array with two extra elements,
168 * one at the beginning, and one at the end. The beginning item contains
169 * the count of the saved maps (head.length), and the module they
170 * belong to if not built in (head.mod). The ending item contains a
171 * pointer to the next array of saved eval_map items.
173 union trace_eval_map_item {
174 struct trace_eval_map map;
175 struct trace_eval_map_head head;
176 struct trace_eval_map_tail tail;
179 static union trace_eval_map_item *trace_eval_maps;
180 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
182 int tracing_set_tracer(struct trace_array *tr, const char *buf);
183 static void ftrace_trace_userstack(struct trace_array *tr,
184 struct trace_buffer *buffer,
185 unsigned int trace_ctx);
187 #define MAX_TRACER_SIZE 100
188 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
189 static char *default_bootup_tracer;
191 static bool allocate_snapshot;
192 static bool snapshot_at_boot;
194 static char boot_instance_info[COMMAND_LINE_SIZE] __initdata;
195 static int boot_instance_index;
197 static char boot_snapshot_info[COMMAND_LINE_SIZE] __initdata;
198 static int boot_snapshot_index;
200 static int __init set_cmdline_ftrace(char *str)
202 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
203 default_bootup_tracer = bootup_tracer_buf;
204 /* We are using ftrace early, expand it */
205 ring_buffer_expanded = true;
208 __setup("ftrace=", set_cmdline_ftrace);
210 static int __init set_ftrace_dump_on_oops(char *str)
212 if (*str++ != '=' || !*str || !strcmp("1", str)) {
213 ftrace_dump_on_oops = DUMP_ALL;
217 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
218 ftrace_dump_on_oops = DUMP_ORIG;
224 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
226 static int __init stop_trace_on_warning(char *str)
228 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
229 __disable_trace_on_warning = 1;
232 __setup("traceoff_on_warning", stop_trace_on_warning);
234 static int __init boot_alloc_snapshot(char *str)
236 char *slot = boot_snapshot_info + boot_snapshot_index;
237 int left = sizeof(boot_snapshot_info) - boot_snapshot_index;
242 if (strlen(str) >= left)
245 ret = snprintf(slot, left, "%s\t", str);
246 boot_snapshot_index += ret;
248 allocate_snapshot = true;
249 /* We also need the main ring buffer expanded */
250 ring_buffer_expanded = true;
254 __setup("alloc_snapshot", boot_alloc_snapshot);
257 static int __init boot_snapshot(char *str)
259 snapshot_at_boot = true;
260 boot_alloc_snapshot(str);
263 __setup("ftrace_boot_snapshot", boot_snapshot);
266 static int __init boot_instance(char *str)
268 char *slot = boot_instance_info + boot_instance_index;
269 int left = sizeof(boot_instance_info) - boot_instance_index;
272 if (strlen(str) >= left)
275 ret = snprintf(slot, left, "%s\t", str);
276 boot_instance_index += ret;
280 __setup("trace_instance=", boot_instance);
283 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
285 static int __init set_trace_boot_options(char *str)
287 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
290 __setup("trace_options=", set_trace_boot_options);
292 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
293 static char *trace_boot_clock __initdata;
295 static int __init set_trace_boot_clock(char *str)
297 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
298 trace_boot_clock = trace_boot_clock_buf;
301 __setup("trace_clock=", set_trace_boot_clock);
303 static int __init set_tracepoint_printk(char *str)
305 /* Ignore the "tp_printk_stop_on_boot" param */
309 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
310 tracepoint_printk = 1;
313 __setup("tp_printk", set_tracepoint_printk);
315 static int __init set_tracepoint_printk_stop(char *str)
317 tracepoint_printk_stop_on_boot = true;
320 __setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
322 unsigned long long ns2usecs(u64 nsec)
330 trace_process_export(struct trace_export *export,
331 struct ring_buffer_event *event, int flag)
333 struct trace_entry *entry;
334 unsigned int size = 0;
336 if (export->flags & flag) {
337 entry = ring_buffer_event_data(event);
338 size = ring_buffer_event_length(event);
339 export->write(export, entry, size);
343 static DEFINE_MUTEX(ftrace_export_lock);
345 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
347 static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
348 static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
349 static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
351 static inline void ftrace_exports_enable(struct trace_export *export)
353 if (export->flags & TRACE_EXPORT_FUNCTION)
354 static_branch_inc(&trace_function_exports_enabled);
356 if (export->flags & TRACE_EXPORT_EVENT)
357 static_branch_inc(&trace_event_exports_enabled);
359 if (export->flags & TRACE_EXPORT_MARKER)
360 static_branch_inc(&trace_marker_exports_enabled);
363 static inline void ftrace_exports_disable(struct trace_export *export)
365 if (export->flags & TRACE_EXPORT_FUNCTION)
366 static_branch_dec(&trace_function_exports_enabled);
368 if (export->flags & TRACE_EXPORT_EVENT)
369 static_branch_dec(&trace_event_exports_enabled);
371 if (export->flags & TRACE_EXPORT_MARKER)
372 static_branch_dec(&trace_marker_exports_enabled);
375 static void ftrace_exports(struct ring_buffer_event *event, int flag)
377 struct trace_export *export;
379 preempt_disable_notrace();
381 export = rcu_dereference_raw_check(ftrace_exports_list);
383 trace_process_export(export, event, flag);
384 export = rcu_dereference_raw_check(export->next);
387 preempt_enable_notrace();
391 add_trace_export(struct trace_export **list, struct trace_export *export)
393 rcu_assign_pointer(export->next, *list);
395 * We are entering export into the list but another
396 * CPU might be walking that list. We need to make sure
397 * the export->next pointer is valid before another CPU sees
398 * the export pointer included into the list.
400 rcu_assign_pointer(*list, export);
404 rm_trace_export(struct trace_export **list, struct trace_export *export)
406 struct trace_export **p;
408 for (p = list; *p != NULL; p = &(*p)->next)
415 rcu_assign_pointer(*p, (*p)->next);
421 add_ftrace_export(struct trace_export **list, struct trace_export *export)
423 ftrace_exports_enable(export);
425 add_trace_export(list, export);
429 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
433 ret = rm_trace_export(list, export);
434 ftrace_exports_disable(export);
439 int register_ftrace_export(struct trace_export *export)
441 if (WARN_ON_ONCE(!export->write))
444 mutex_lock(&ftrace_export_lock);
446 add_ftrace_export(&ftrace_exports_list, export);
448 mutex_unlock(&ftrace_export_lock);
452 EXPORT_SYMBOL_GPL(register_ftrace_export);
454 int unregister_ftrace_export(struct trace_export *export)
458 mutex_lock(&ftrace_export_lock);
460 ret = rm_ftrace_export(&ftrace_exports_list, export);
462 mutex_unlock(&ftrace_export_lock);
466 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
468 /* trace_flags holds trace_options default values */
469 #define TRACE_DEFAULT_FLAGS \
470 (FUNCTION_DEFAULT_FLAGS | \
471 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
472 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
473 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
474 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
477 /* trace_options that are only supported by global_trace */
478 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
479 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
481 /* trace_flags that are default zero for instances */
482 #define ZEROED_TRACE_FLAGS \
483 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
486 * The global_trace is the descriptor that holds the top-level tracing
487 * buffers for the live tracing.
489 static struct trace_array global_trace = {
490 .trace_flags = TRACE_DEFAULT_FLAGS,
493 LIST_HEAD(ftrace_trace_arrays);
495 int trace_array_get(struct trace_array *this_tr)
497 struct trace_array *tr;
500 mutex_lock(&trace_types_lock);
501 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
508 mutex_unlock(&trace_types_lock);
513 static void __trace_array_put(struct trace_array *this_tr)
515 WARN_ON(!this_tr->ref);
520 * trace_array_put - Decrement the reference counter for this trace array.
521 * @this_tr : pointer to the trace array
523 * NOTE: Use this when we no longer need the trace array returned by
524 * trace_array_get_by_name(). This ensures the trace array can be later
528 void trace_array_put(struct trace_array *this_tr)
533 mutex_lock(&trace_types_lock);
534 __trace_array_put(this_tr);
535 mutex_unlock(&trace_types_lock);
537 EXPORT_SYMBOL_GPL(trace_array_put);
539 int tracing_check_open_get_tr(struct trace_array *tr)
543 ret = security_locked_down(LOCKDOWN_TRACEFS);
547 if (tracing_disabled)
550 if (tr && trace_array_get(tr) < 0)
556 int call_filter_check_discard(struct trace_event_call *call, void *rec,
557 struct trace_buffer *buffer,
558 struct ring_buffer_event *event)
560 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
561 !filter_match_preds(call->filter, rec)) {
562 __trace_event_discard_commit(buffer, event);
570 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
571 * @filtered_pids: The list of pids to check
572 * @search_pid: The PID to find in @filtered_pids
574 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
577 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
579 return trace_pid_list_is_set(filtered_pids, search_pid);
583 * trace_ignore_this_task - should a task be ignored for tracing
584 * @filtered_pids: The list of pids to check
585 * @filtered_no_pids: The list of pids not to be traced
586 * @task: The task that should be ignored if not filtered
588 * Checks if @task should be traced or not from @filtered_pids.
589 * Returns true if @task should *NOT* be traced.
590 * Returns false if @task should be traced.
593 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
594 struct trace_pid_list *filtered_no_pids,
595 struct task_struct *task)
598 * If filtered_no_pids is not empty, and the task's pid is listed
599 * in filtered_no_pids, then return true.
600 * Otherwise, if filtered_pids is empty, that means we can
601 * trace all tasks. If it has content, then only trace pids
602 * within filtered_pids.
605 return (filtered_pids &&
606 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
608 trace_find_filtered_pid(filtered_no_pids, task->pid));
612 * trace_filter_add_remove_task - Add or remove a task from a pid_list
613 * @pid_list: The list to modify
614 * @self: The current task for fork or NULL for exit
615 * @task: The task to add or remove
617 * If adding a task, if @self is defined, the task is only added if @self
618 * is also included in @pid_list. This happens on fork and tasks should
619 * only be added when the parent is listed. If @self is NULL, then the
620 * @task pid will be removed from the list, which would happen on exit
623 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
624 struct task_struct *self,
625 struct task_struct *task)
630 /* For forks, we only add if the forking task is listed */
632 if (!trace_find_filtered_pid(pid_list, self->pid))
636 /* "self" is set for forks, and NULL for exits */
638 trace_pid_list_set(pid_list, task->pid);
640 trace_pid_list_clear(pid_list, task->pid);
644 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
645 * @pid_list: The pid list to show
646 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
647 * @pos: The position of the file
649 * This is used by the seq_file "next" operation to iterate the pids
650 * listed in a trace_pid_list structure.
652 * Returns the pid+1 as we want to display pid of zero, but NULL would
653 * stop the iteration.
655 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
657 long pid = (unsigned long)v;
662 /* pid already is +1 of the actual previous bit */
663 if (trace_pid_list_next(pid_list, pid, &next) < 0)
668 /* Return pid + 1 to allow zero to be represented */
669 return (void *)(pid + 1);
673 * trace_pid_start - Used for seq_file to start reading pid lists
674 * @pid_list: The pid list to show
675 * @pos: The position of the file
677 * This is used by seq_file "start" operation to start the iteration
680 * Returns the pid+1 as we want to display pid of zero, but NULL would
681 * stop the iteration.
683 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
689 if (trace_pid_list_first(pid_list, &first) < 0)
694 /* Return pid + 1 so that zero can be the exit value */
695 for (pid++; pid && l < *pos;
696 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
702 * trace_pid_show - show the current pid in seq_file processing
703 * @m: The seq_file structure to write into
704 * @v: A void pointer of the pid (+1) value to display
706 * Can be directly used by seq_file operations to display the current
709 int trace_pid_show(struct seq_file *m, void *v)
711 unsigned long pid = (unsigned long)v - 1;
713 seq_printf(m, "%lu\n", pid);
717 /* 128 should be much more than enough */
718 #define PID_BUF_SIZE 127
720 int trace_pid_write(struct trace_pid_list *filtered_pids,
721 struct trace_pid_list **new_pid_list,
722 const char __user *ubuf, size_t cnt)
724 struct trace_pid_list *pid_list;
725 struct trace_parser parser;
733 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
737 * Always recreate a new array. The write is an all or nothing
738 * operation. Always create a new array when adding new pids by
739 * the user. If the operation fails, then the current list is
742 pid_list = trace_pid_list_alloc();
744 trace_parser_put(&parser);
749 /* copy the current bits to the new max */
750 ret = trace_pid_list_first(filtered_pids, &pid);
752 trace_pid_list_set(pid_list, pid);
753 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
763 ret = trace_get_user(&parser, ubuf, cnt, &pos);
771 if (!trace_parser_loaded(&parser))
775 if (kstrtoul(parser.buffer, 0, &val))
780 if (trace_pid_list_set(pid_list, pid) < 0) {
786 trace_parser_clear(&parser);
789 trace_parser_put(&parser);
792 trace_pid_list_free(pid_list);
797 /* Cleared the list of pids */
798 trace_pid_list_free(pid_list);
802 *new_pid_list = pid_list;
807 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
811 /* Early boot up does not have a buffer yet */
813 return trace_clock_local();
815 ts = ring_buffer_time_stamp(buf->buffer);
816 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
821 u64 ftrace_now(int cpu)
823 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
827 * tracing_is_enabled - Show if global_trace has been enabled
829 * Shows if the global trace has been enabled or not. It uses the
830 * mirror flag "buffer_disabled" to be used in fast paths such as for
831 * the irqsoff tracer. But it may be inaccurate due to races. If you
832 * need to know the accurate state, use tracing_is_on() which is a little
833 * slower, but accurate.
835 int tracing_is_enabled(void)
838 * For quick access (irqsoff uses this in fast path), just
839 * return the mirror variable of the state of the ring buffer.
840 * It's a little racy, but we don't really care.
843 return !global_trace.buffer_disabled;
847 * trace_buf_size is the size in bytes that is allocated
848 * for a buffer. Note, the number of bytes is always rounded
851 * This number is purposely set to a low number of 16384.
852 * If the dump on oops happens, it will be much appreciated
853 * to not have to wait for all that output. Anyway this can be
854 * boot time and run time configurable.
856 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
858 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
860 /* trace_types holds a link list of available tracers. */
861 static struct tracer *trace_types __read_mostly;
864 * trace_types_lock is used to protect the trace_types list.
866 DEFINE_MUTEX(trace_types_lock);
869 * serialize the access of the ring buffer
871 * ring buffer serializes readers, but it is low level protection.
872 * The validity of the events (which returns by ring_buffer_peek() ..etc)
873 * are not protected by ring buffer.
875 * The content of events may become garbage if we allow other process consumes
876 * these events concurrently:
877 * A) the page of the consumed events may become a normal page
878 * (not reader page) in ring buffer, and this page will be rewritten
879 * by events producer.
880 * B) The page of the consumed events may become a page for splice_read,
881 * and this page will be returned to system.
883 * These primitives allow multi process access to different cpu ring buffer
886 * These primitives don't distinguish read-only and read-consume access.
887 * Multi read-only access are also serialized.
891 static DECLARE_RWSEM(all_cpu_access_lock);
892 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
894 static inline void trace_access_lock(int cpu)
896 if (cpu == RING_BUFFER_ALL_CPUS) {
897 /* gain it for accessing the whole ring buffer. */
898 down_write(&all_cpu_access_lock);
900 /* gain it for accessing a cpu ring buffer. */
902 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
903 down_read(&all_cpu_access_lock);
905 /* Secondly block other access to this @cpu ring buffer. */
906 mutex_lock(&per_cpu(cpu_access_lock, cpu));
910 static inline void trace_access_unlock(int cpu)
912 if (cpu == RING_BUFFER_ALL_CPUS) {
913 up_write(&all_cpu_access_lock);
915 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
916 up_read(&all_cpu_access_lock);
920 static inline void trace_access_lock_init(void)
924 for_each_possible_cpu(cpu)
925 mutex_init(&per_cpu(cpu_access_lock, cpu));
930 static DEFINE_MUTEX(access_lock);
932 static inline void trace_access_lock(int cpu)
935 mutex_lock(&access_lock);
938 static inline void trace_access_unlock(int cpu)
941 mutex_unlock(&access_lock);
944 static inline void trace_access_lock_init(void)
950 #ifdef CONFIG_STACKTRACE
951 static void __ftrace_trace_stack(struct trace_buffer *buffer,
952 unsigned int trace_ctx,
953 int skip, struct pt_regs *regs);
954 static inline void ftrace_trace_stack(struct trace_array *tr,
955 struct trace_buffer *buffer,
956 unsigned int trace_ctx,
957 int skip, struct pt_regs *regs);
960 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
961 unsigned int trace_ctx,
962 int skip, struct pt_regs *regs)
965 static inline void ftrace_trace_stack(struct trace_array *tr,
966 struct trace_buffer *buffer,
967 unsigned long trace_ctx,
968 int skip, struct pt_regs *regs)
974 static __always_inline void
975 trace_event_setup(struct ring_buffer_event *event,
976 int type, unsigned int trace_ctx)
978 struct trace_entry *ent = ring_buffer_event_data(event);
980 tracing_generic_entry_update(ent, type, trace_ctx);
983 static __always_inline struct ring_buffer_event *
984 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
987 unsigned int trace_ctx)
989 struct ring_buffer_event *event;
991 event = ring_buffer_lock_reserve(buffer, len);
993 trace_event_setup(event, type, trace_ctx);
998 void tracer_tracing_on(struct trace_array *tr)
1000 if (tr->array_buffer.buffer)
1001 ring_buffer_record_on(tr->array_buffer.buffer);
1003 * This flag is looked at when buffers haven't been allocated
1004 * yet, or by some tracers (like irqsoff), that just want to
1005 * know if the ring buffer has been disabled, but it can handle
1006 * races of where it gets disabled but we still do a record.
1007 * As the check is in the fast path of the tracers, it is more
1008 * important to be fast than accurate.
1010 tr->buffer_disabled = 0;
1011 /* Make the flag seen by readers */
1016 * tracing_on - enable tracing buffers
1018 * This function enables tracing buffers that may have been
1019 * disabled with tracing_off.
1021 void tracing_on(void)
1023 tracer_tracing_on(&global_trace);
1025 EXPORT_SYMBOL_GPL(tracing_on);
1028 static __always_inline void
1029 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
1031 __this_cpu_write(trace_taskinfo_save, true);
1033 /* If this is the temp buffer, we need to commit fully */
1034 if (this_cpu_read(trace_buffered_event) == event) {
1035 /* Length is in event->array[0] */
1036 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1037 /* Release the temp buffer */
1038 this_cpu_dec(trace_buffered_event_cnt);
1039 /* ring_buffer_unlock_commit() enables preemption */
1040 preempt_enable_notrace();
1042 ring_buffer_unlock_commit(buffer);
1045 int __trace_array_puts(struct trace_array *tr, unsigned long ip,
1046 const char *str, int size)
1048 struct ring_buffer_event *event;
1049 struct trace_buffer *buffer;
1050 struct print_entry *entry;
1051 unsigned int trace_ctx;
1054 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1057 if (unlikely(tracing_selftest_running && tr == &global_trace))
1060 if (unlikely(tracing_disabled))
1063 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1065 trace_ctx = tracing_gen_ctx();
1066 buffer = tr->array_buffer.buffer;
1067 ring_buffer_nest_start(buffer);
1068 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1075 entry = ring_buffer_event_data(event);
1078 memcpy(&entry->buf, str, size);
1080 /* Add a newline if necessary */
1081 if (entry->buf[size - 1] != '\n') {
1082 entry->buf[size] = '\n';
1083 entry->buf[size + 1] = '\0';
1085 entry->buf[size] = '\0';
1087 __buffer_unlock_commit(buffer, event);
1088 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1090 ring_buffer_nest_end(buffer);
1093 EXPORT_SYMBOL_GPL(__trace_array_puts);
1096 * __trace_puts - write a constant string into the trace buffer.
1097 * @ip: The address of the caller
1098 * @str: The constant string to write
1099 * @size: The size of the string.
1101 int __trace_puts(unsigned long ip, const char *str, int size)
1103 return __trace_array_puts(&global_trace, ip, str, size);
1105 EXPORT_SYMBOL_GPL(__trace_puts);
1108 * __trace_bputs - write the pointer to a constant string into trace buffer
1109 * @ip: The address of the caller
1110 * @str: The constant string to write to the buffer to
1112 int __trace_bputs(unsigned long ip, const char *str)
1114 struct ring_buffer_event *event;
1115 struct trace_buffer *buffer;
1116 struct bputs_entry *entry;
1117 unsigned int trace_ctx;
1118 int size = sizeof(struct bputs_entry);
1121 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1124 if (unlikely(tracing_selftest_running || tracing_disabled))
1127 trace_ctx = tracing_gen_ctx();
1128 buffer = global_trace.array_buffer.buffer;
1130 ring_buffer_nest_start(buffer);
1131 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1136 entry = ring_buffer_event_data(event);
1140 __buffer_unlock_commit(buffer, event);
1141 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1145 ring_buffer_nest_end(buffer);
1148 EXPORT_SYMBOL_GPL(__trace_bputs);
1150 #ifdef CONFIG_TRACER_SNAPSHOT
1151 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1154 struct tracer *tracer = tr->current_trace;
1155 unsigned long flags;
1158 trace_array_puts(tr, "*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1159 trace_array_puts(tr, "*** snapshot is being ignored ***\n");
1163 if (!tr->allocated_snapshot) {
1164 trace_array_puts(tr, "*** SNAPSHOT NOT ALLOCATED ***\n");
1165 trace_array_puts(tr, "*** stopping trace here! ***\n");
1166 tracer_tracing_off(tr);
1170 /* Note, snapshot can not be used when the tracer uses it */
1171 if (tracer->use_max_tr) {
1172 trace_array_puts(tr, "*** LATENCY TRACER ACTIVE ***\n");
1173 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1177 local_irq_save(flags);
1178 update_max_tr(tr, current, smp_processor_id(), cond_data);
1179 local_irq_restore(flags);
1182 void tracing_snapshot_instance(struct trace_array *tr)
1184 tracing_snapshot_instance_cond(tr, NULL);
1188 * tracing_snapshot - take a snapshot of the current buffer.
1190 * This causes a swap between the snapshot buffer and the current live
1191 * tracing buffer. You can use this to take snapshots of the live
1192 * trace when some condition is triggered, but continue to trace.
1194 * Note, make sure to allocate the snapshot with either
1195 * a tracing_snapshot_alloc(), or by doing it manually
1196 * with: echo 1 > /sys/kernel/tracing/snapshot
1198 * If the snapshot buffer is not allocated, it will stop tracing.
1199 * Basically making a permanent snapshot.
1201 void tracing_snapshot(void)
1203 struct trace_array *tr = &global_trace;
1205 tracing_snapshot_instance(tr);
1207 EXPORT_SYMBOL_GPL(tracing_snapshot);
1210 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1211 * @tr: The tracing instance to snapshot
1212 * @cond_data: The data to be tested conditionally, and possibly saved
1214 * This is the same as tracing_snapshot() except that the snapshot is
1215 * conditional - the snapshot will only happen if the
1216 * cond_snapshot.update() implementation receiving the cond_data
1217 * returns true, which means that the trace array's cond_snapshot
1218 * update() operation used the cond_data to determine whether the
1219 * snapshot should be taken, and if it was, presumably saved it along
1220 * with the snapshot.
1222 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1224 tracing_snapshot_instance_cond(tr, cond_data);
1226 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1229 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1230 * @tr: The tracing instance
1232 * When the user enables a conditional snapshot using
1233 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1234 * with the snapshot. This accessor is used to retrieve it.
1236 * Should not be called from cond_snapshot.update(), since it takes
1237 * the tr->max_lock lock, which the code calling
1238 * cond_snapshot.update() has already done.
1240 * Returns the cond_data associated with the trace array's snapshot.
1242 void *tracing_cond_snapshot_data(struct trace_array *tr)
1244 void *cond_data = NULL;
1246 local_irq_disable();
1247 arch_spin_lock(&tr->max_lock);
1249 if (tr->cond_snapshot)
1250 cond_data = tr->cond_snapshot->cond_data;
1252 arch_spin_unlock(&tr->max_lock);
1257 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1259 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1260 struct array_buffer *size_buf, int cpu_id);
1261 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1263 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1267 if (!tr->allocated_snapshot) {
1269 /* allocate spare buffer */
1270 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1271 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1275 tr->allocated_snapshot = true;
1281 static void free_snapshot(struct trace_array *tr)
1284 * We don't free the ring buffer. instead, resize it because
1285 * The max_tr ring buffer has some state (e.g. ring->clock) and
1286 * we want preserve it.
1288 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1289 set_buffer_entries(&tr->max_buffer, 1);
1290 tracing_reset_online_cpus(&tr->max_buffer);
1291 tr->allocated_snapshot = false;
1295 * tracing_alloc_snapshot - allocate snapshot buffer.
1297 * This only allocates the snapshot buffer if it isn't already
1298 * allocated - it doesn't also take a snapshot.
1300 * This is meant to be used in cases where the snapshot buffer needs
1301 * to be set up for events that can't sleep but need to be able to
1302 * trigger a snapshot.
1304 int tracing_alloc_snapshot(void)
1306 struct trace_array *tr = &global_trace;
1309 ret = tracing_alloc_snapshot_instance(tr);
1314 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1317 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1319 * This is similar to tracing_snapshot(), but it will allocate the
1320 * snapshot buffer if it isn't already allocated. Use this only
1321 * where it is safe to sleep, as the allocation may sleep.
1323 * This causes a swap between the snapshot buffer and the current live
1324 * tracing buffer. You can use this to take snapshots of the live
1325 * trace when some condition is triggered, but continue to trace.
1327 void tracing_snapshot_alloc(void)
1331 ret = tracing_alloc_snapshot();
1337 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1340 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1341 * @tr: The tracing instance
1342 * @cond_data: User data to associate with the snapshot
1343 * @update: Implementation of the cond_snapshot update function
1345 * Check whether the conditional snapshot for the given instance has
1346 * already been enabled, or if the current tracer is already using a
1347 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1348 * save the cond_data and update function inside.
1350 * Returns 0 if successful, error otherwise.
1352 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1353 cond_update_fn_t update)
1355 struct cond_snapshot *cond_snapshot;
1358 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1362 cond_snapshot->cond_data = cond_data;
1363 cond_snapshot->update = update;
1365 mutex_lock(&trace_types_lock);
1367 ret = tracing_alloc_snapshot_instance(tr);
1371 if (tr->current_trace->use_max_tr) {
1377 * The cond_snapshot can only change to NULL without the
1378 * trace_types_lock. We don't care if we race with it going
1379 * to NULL, but we want to make sure that it's not set to
1380 * something other than NULL when we get here, which we can
1381 * do safely with only holding the trace_types_lock and not
1382 * having to take the max_lock.
1384 if (tr->cond_snapshot) {
1389 local_irq_disable();
1390 arch_spin_lock(&tr->max_lock);
1391 tr->cond_snapshot = cond_snapshot;
1392 arch_spin_unlock(&tr->max_lock);
1395 mutex_unlock(&trace_types_lock);
1400 mutex_unlock(&trace_types_lock);
1401 kfree(cond_snapshot);
1404 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1407 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1408 * @tr: The tracing instance
1410 * Check whether the conditional snapshot for the given instance is
1411 * enabled; if so, free the cond_snapshot associated with it,
1412 * otherwise return -EINVAL.
1414 * Returns 0 if successful, error otherwise.
1416 int tracing_snapshot_cond_disable(struct trace_array *tr)
1420 local_irq_disable();
1421 arch_spin_lock(&tr->max_lock);
1423 if (!tr->cond_snapshot)
1426 kfree(tr->cond_snapshot);
1427 tr->cond_snapshot = NULL;
1430 arch_spin_unlock(&tr->max_lock);
1435 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1437 void tracing_snapshot(void)
1439 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1441 EXPORT_SYMBOL_GPL(tracing_snapshot);
1442 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1444 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1446 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1447 int tracing_alloc_snapshot(void)
1449 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1452 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1453 void tracing_snapshot_alloc(void)
1458 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1459 void *tracing_cond_snapshot_data(struct trace_array *tr)
1463 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1464 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1468 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1469 int tracing_snapshot_cond_disable(struct trace_array *tr)
1473 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1474 #define free_snapshot(tr) do { } while (0)
1475 #endif /* CONFIG_TRACER_SNAPSHOT */
1477 void tracer_tracing_off(struct trace_array *tr)
1479 if (tr->array_buffer.buffer)
1480 ring_buffer_record_off(tr->array_buffer.buffer);
1482 * This flag is looked at when buffers haven't been allocated
1483 * yet, or by some tracers (like irqsoff), that just want to
1484 * know if the ring buffer has been disabled, but it can handle
1485 * races of where it gets disabled but we still do a record.
1486 * As the check is in the fast path of the tracers, it is more
1487 * important to be fast than accurate.
1489 tr->buffer_disabled = 1;
1490 /* Make the flag seen by readers */
1495 * tracing_off - turn off tracing buffers
1497 * This function stops the tracing buffers from recording data.
1498 * It does not disable any overhead the tracers themselves may
1499 * be causing. This function simply causes all recording to
1500 * the ring buffers to fail.
1502 void tracing_off(void)
1504 tracer_tracing_off(&global_trace);
1506 EXPORT_SYMBOL_GPL(tracing_off);
1508 void disable_trace_on_warning(void)
1510 if (__disable_trace_on_warning) {
1511 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1512 "Disabling tracing due to warning\n");
1518 * tracer_tracing_is_on - show real state of ring buffer enabled
1519 * @tr : the trace array to know if ring buffer is enabled
1521 * Shows real state of the ring buffer if it is enabled or not.
1523 bool tracer_tracing_is_on(struct trace_array *tr)
1525 if (tr->array_buffer.buffer)
1526 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1527 return !tr->buffer_disabled;
1531 * tracing_is_on - show state of ring buffers enabled
1533 int tracing_is_on(void)
1535 return tracer_tracing_is_on(&global_trace);
1537 EXPORT_SYMBOL_GPL(tracing_is_on);
1539 static int __init set_buf_size(char *str)
1541 unsigned long buf_size;
1545 buf_size = memparse(str, &str);
1547 * nr_entries can not be zero and the startup
1548 * tests require some buffer space. Therefore
1549 * ensure we have at least 4096 bytes of buffer.
1551 trace_buf_size = max(4096UL, buf_size);
1554 __setup("trace_buf_size=", set_buf_size);
1556 static int __init set_tracing_thresh(char *str)
1558 unsigned long threshold;
1563 ret = kstrtoul(str, 0, &threshold);
1566 tracing_thresh = threshold * 1000;
1569 __setup("tracing_thresh=", set_tracing_thresh);
1571 unsigned long nsecs_to_usecs(unsigned long nsecs)
1573 return nsecs / 1000;
1577 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1578 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1579 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1580 * of strings in the order that the evals (enum) were defined.
1585 /* These must match the bit positions in trace_iterator_flags */
1586 static const char *trace_options[] = {
1594 int in_ns; /* is this clock in nanoseconds? */
1595 } trace_clocks[] = {
1596 { trace_clock_local, "local", 1 },
1597 { trace_clock_global, "global", 1 },
1598 { trace_clock_counter, "counter", 0 },
1599 { trace_clock_jiffies, "uptime", 0 },
1600 { trace_clock, "perf", 1 },
1601 { ktime_get_mono_fast_ns, "mono", 1 },
1602 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1603 { ktime_get_boot_fast_ns, "boot", 1 },
1604 { ktime_get_tai_fast_ns, "tai", 1 },
1608 bool trace_clock_in_ns(struct trace_array *tr)
1610 if (trace_clocks[tr->clock_id].in_ns)
1617 * trace_parser_get_init - gets the buffer for trace parser
1619 int trace_parser_get_init(struct trace_parser *parser, int size)
1621 memset(parser, 0, sizeof(*parser));
1623 parser->buffer = kmalloc(size, GFP_KERNEL);
1624 if (!parser->buffer)
1627 parser->size = size;
1632 * trace_parser_put - frees the buffer for trace parser
1634 void trace_parser_put(struct trace_parser *parser)
1636 kfree(parser->buffer);
1637 parser->buffer = NULL;
1641 * trace_get_user - reads the user input string separated by space
1642 * (matched by isspace(ch))
1644 * For each string found the 'struct trace_parser' is updated,
1645 * and the function returns.
1647 * Returns number of bytes read.
1649 * See kernel/trace/trace.h for 'struct trace_parser' details.
1651 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1652 size_t cnt, loff_t *ppos)
1659 trace_parser_clear(parser);
1661 ret = get_user(ch, ubuf++);
1669 * The parser is not finished with the last write,
1670 * continue reading the user input without skipping spaces.
1672 if (!parser->cont) {
1673 /* skip white space */
1674 while (cnt && isspace(ch)) {
1675 ret = get_user(ch, ubuf++);
1684 /* only spaces were written */
1685 if (isspace(ch) || !ch) {
1692 /* read the non-space input */
1693 while (cnt && !isspace(ch) && ch) {
1694 if (parser->idx < parser->size - 1)
1695 parser->buffer[parser->idx++] = ch;
1700 ret = get_user(ch, ubuf++);
1707 /* We either got finished input or we have to wait for another call. */
1708 if (isspace(ch) || !ch) {
1709 parser->buffer[parser->idx] = 0;
1710 parser->cont = false;
1711 } else if (parser->idx < parser->size - 1) {
1712 parser->cont = true;
1713 parser->buffer[parser->idx++] = ch;
1714 /* Make sure the parsed string always terminates with '\0'. */
1715 parser->buffer[parser->idx] = 0;
1728 /* TODO add a seq_buf_to_buffer() */
1729 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1733 if (trace_seq_used(s) <= s->seq.readpos)
1736 len = trace_seq_used(s) - s->seq.readpos;
1739 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1741 s->seq.readpos += cnt;
1745 unsigned long __read_mostly tracing_thresh;
1747 #ifdef CONFIG_TRACER_MAX_TRACE
1748 static const struct file_operations tracing_max_lat_fops;
1750 #ifdef LATENCY_FS_NOTIFY
1752 static struct workqueue_struct *fsnotify_wq;
1754 static void latency_fsnotify_workfn(struct work_struct *work)
1756 struct trace_array *tr = container_of(work, struct trace_array,
1758 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1761 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1763 struct trace_array *tr = container_of(iwork, struct trace_array,
1765 queue_work(fsnotify_wq, &tr->fsnotify_work);
1768 static void trace_create_maxlat_file(struct trace_array *tr,
1769 struct dentry *d_tracer)
1771 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1772 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1773 tr->d_max_latency = trace_create_file("tracing_max_latency",
1775 d_tracer, &tr->max_latency,
1776 &tracing_max_lat_fops);
1779 __init static int latency_fsnotify_init(void)
1781 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1782 WQ_UNBOUND | WQ_HIGHPRI, 0);
1784 pr_err("Unable to allocate tr_max_lat_wq\n");
1790 late_initcall_sync(latency_fsnotify_init);
1792 void latency_fsnotify(struct trace_array *tr)
1797 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1798 * possible that we are called from __schedule() or do_idle(), which
1799 * could cause a deadlock.
1801 irq_work_queue(&tr->fsnotify_irqwork);
1804 #else /* !LATENCY_FS_NOTIFY */
1806 #define trace_create_maxlat_file(tr, d_tracer) \
1807 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1808 d_tracer, &tr->max_latency, &tracing_max_lat_fops)
1813 * Copy the new maximum trace into the separate maximum-trace
1814 * structure. (this way the maximum trace is permanently saved,
1815 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1818 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1820 struct array_buffer *trace_buf = &tr->array_buffer;
1821 struct array_buffer *max_buf = &tr->max_buffer;
1822 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1823 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1826 max_buf->time_start = data->preempt_timestamp;
1828 max_data->saved_latency = tr->max_latency;
1829 max_data->critical_start = data->critical_start;
1830 max_data->critical_end = data->critical_end;
1832 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1833 max_data->pid = tsk->pid;
1835 * If tsk == current, then use current_uid(), as that does not use
1836 * RCU. The irq tracer can be called out of RCU scope.
1839 max_data->uid = current_uid();
1841 max_data->uid = task_uid(tsk);
1843 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1844 max_data->policy = tsk->policy;
1845 max_data->rt_priority = tsk->rt_priority;
1847 /* record this tasks comm */
1848 tracing_record_cmdline(tsk);
1849 latency_fsnotify(tr);
1853 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1855 * @tsk: the task with the latency
1856 * @cpu: The cpu that initiated the trace.
1857 * @cond_data: User data associated with a conditional snapshot
1859 * Flip the buffers between the @tr and the max_tr and record information
1860 * about which task was the cause of this latency.
1863 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1869 WARN_ON_ONCE(!irqs_disabled());
1871 if (!tr->allocated_snapshot) {
1872 /* Only the nop tracer should hit this when disabling */
1873 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1877 arch_spin_lock(&tr->max_lock);
1879 /* Inherit the recordable setting from array_buffer */
1880 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1881 ring_buffer_record_on(tr->max_buffer.buffer);
1883 ring_buffer_record_off(tr->max_buffer.buffer);
1885 #ifdef CONFIG_TRACER_SNAPSHOT
1886 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1887 arch_spin_unlock(&tr->max_lock);
1891 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1893 __update_max_tr(tr, tsk, cpu);
1895 arch_spin_unlock(&tr->max_lock);
1899 * update_max_tr_single - only copy one trace over, and reset the rest
1901 * @tsk: task with the latency
1902 * @cpu: the cpu of the buffer to copy.
1904 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1907 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1914 WARN_ON_ONCE(!irqs_disabled());
1915 if (!tr->allocated_snapshot) {
1916 /* Only the nop tracer should hit this when disabling */
1917 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1921 arch_spin_lock(&tr->max_lock);
1923 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1925 if (ret == -EBUSY) {
1927 * We failed to swap the buffer due to a commit taking
1928 * place on this CPU. We fail to record, but we reset
1929 * the max trace buffer (no one writes directly to it)
1930 * and flag that it failed.
1932 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1933 "Failed to swap buffers due to commit in progress\n");
1936 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1938 __update_max_tr(tr, tsk, cpu);
1939 arch_spin_unlock(&tr->max_lock);
1942 #endif /* CONFIG_TRACER_MAX_TRACE */
1944 static int wait_on_pipe(struct trace_iterator *iter, int full)
1946 /* Iterators are static, they should be filled or empty */
1947 if (trace_buffer_iter(iter, iter->cpu_file))
1950 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1954 #ifdef CONFIG_FTRACE_STARTUP_TEST
1955 static bool selftests_can_run;
1957 struct trace_selftests {
1958 struct list_head list;
1959 struct tracer *type;
1962 static LIST_HEAD(postponed_selftests);
1964 static int save_selftest(struct tracer *type)
1966 struct trace_selftests *selftest;
1968 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1972 selftest->type = type;
1973 list_add(&selftest->list, &postponed_selftests);
1977 static int run_tracer_selftest(struct tracer *type)
1979 struct trace_array *tr = &global_trace;
1980 struct tracer *saved_tracer = tr->current_trace;
1983 if (!type->selftest || tracing_selftest_disabled)
1987 * If a tracer registers early in boot up (before scheduling is
1988 * initialized and such), then do not run its selftests yet.
1989 * Instead, run it a little later in the boot process.
1991 if (!selftests_can_run)
1992 return save_selftest(type);
1994 if (!tracing_is_on()) {
1995 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2001 * Run a selftest on this tracer.
2002 * Here we reset the trace buffer, and set the current
2003 * tracer to be this tracer. The tracer can then run some
2004 * internal tracing to verify that everything is in order.
2005 * If we fail, we do not register this tracer.
2007 tracing_reset_online_cpus(&tr->array_buffer);
2009 tr->current_trace = type;
2011 #ifdef CONFIG_TRACER_MAX_TRACE
2012 if (type->use_max_tr) {
2013 /* If we expanded the buffers, make sure the max is expanded too */
2014 if (ring_buffer_expanded)
2015 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2016 RING_BUFFER_ALL_CPUS);
2017 tr->allocated_snapshot = true;
2021 /* the test is responsible for initializing and enabling */
2022 pr_info("Testing tracer %s: ", type->name);
2023 ret = type->selftest(type, tr);
2024 /* the test is responsible for resetting too */
2025 tr->current_trace = saved_tracer;
2027 printk(KERN_CONT "FAILED!\n");
2028 /* Add the warning after printing 'FAILED' */
2032 /* Only reset on passing, to avoid touching corrupted buffers */
2033 tracing_reset_online_cpus(&tr->array_buffer);
2035 #ifdef CONFIG_TRACER_MAX_TRACE
2036 if (type->use_max_tr) {
2037 tr->allocated_snapshot = false;
2039 /* Shrink the max buffer again */
2040 if (ring_buffer_expanded)
2041 ring_buffer_resize(tr->max_buffer.buffer, 1,
2042 RING_BUFFER_ALL_CPUS);
2046 printk(KERN_CONT "PASSED\n");
2050 static int do_run_tracer_selftest(struct tracer *type)
2055 * Tests can take a long time, especially if they are run one after the
2056 * other, as does happen during bootup when all the tracers are
2057 * registered. This could cause the soft lockup watchdog to trigger.
2061 tracing_selftest_running = true;
2062 ret = run_tracer_selftest(type);
2063 tracing_selftest_running = false;
2068 static __init int init_trace_selftests(void)
2070 struct trace_selftests *p, *n;
2071 struct tracer *t, **last;
2074 selftests_can_run = true;
2076 mutex_lock(&trace_types_lock);
2078 if (list_empty(&postponed_selftests))
2081 pr_info("Running postponed tracer tests:\n");
2083 tracing_selftest_running = true;
2084 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2085 /* This loop can take minutes when sanitizers are enabled, so
2086 * lets make sure we allow RCU processing.
2089 ret = run_tracer_selftest(p->type);
2090 /* If the test fails, then warn and remove from available_tracers */
2092 WARN(1, "tracer: %s failed selftest, disabling\n",
2094 last = &trace_types;
2095 for (t = trace_types; t; t = t->next) {
2106 tracing_selftest_running = false;
2109 mutex_unlock(&trace_types_lock);
2113 core_initcall(init_trace_selftests);
2115 static inline int run_tracer_selftest(struct tracer *type)
2119 static inline int do_run_tracer_selftest(struct tracer *type)
2123 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2125 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2127 static void __init apply_trace_boot_options(void);
2130 * register_tracer - register a tracer with the ftrace system.
2131 * @type: the plugin for the tracer
2133 * Register a new plugin tracer.
2135 int __init register_tracer(struct tracer *type)
2141 pr_info("Tracer must have a name\n");
2145 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2146 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2150 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2151 pr_warn("Can not register tracer %s due to lockdown\n",
2156 mutex_lock(&trace_types_lock);
2158 for (t = trace_types; t; t = t->next) {
2159 if (strcmp(type->name, t->name) == 0) {
2161 pr_info("Tracer %s already registered\n",
2168 if (!type->set_flag)
2169 type->set_flag = &dummy_set_flag;
2171 /*allocate a dummy tracer_flags*/
2172 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2177 type->flags->val = 0;
2178 type->flags->opts = dummy_tracer_opt;
2180 if (!type->flags->opts)
2181 type->flags->opts = dummy_tracer_opt;
2183 /* store the tracer for __set_tracer_option */
2184 type->flags->trace = type;
2186 ret = do_run_tracer_selftest(type);
2190 type->next = trace_types;
2192 add_tracer_options(&global_trace, type);
2195 mutex_unlock(&trace_types_lock);
2197 if (ret || !default_bootup_tracer)
2200 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2203 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2204 /* Do we want this tracer to start on bootup? */
2205 tracing_set_tracer(&global_trace, type->name);
2206 default_bootup_tracer = NULL;
2208 apply_trace_boot_options();
2210 /* disable other selftests, since this will break it. */
2211 disable_tracing_selftest("running a tracer");
2217 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2219 struct trace_buffer *buffer = buf->buffer;
2224 ring_buffer_record_disable(buffer);
2226 /* Make sure all commits have finished */
2228 ring_buffer_reset_cpu(buffer, cpu);
2230 ring_buffer_record_enable(buffer);
2233 void tracing_reset_online_cpus(struct array_buffer *buf)
2235 struct trace_buffer *buffer = buf->buffer;
2240 ring_buffer_record_disable(buffer);
2242 /* Make sure all commits have finished */
2245 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2247 ring_buffer_reset_online_cpus(buffer);
2249 ring_buffer_record_enable(buffer);
2252 /* Must have trace_types_lock held */
2253 void tracing_reset_all_online_cpus_unlocked(void)
2255 struct trace_array *tr;
2257 lockdep_assert_held(&trace_types_lock);
2259 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2260 if (!tr->clear_trace)
2262 tr->clear_trace = false;
2263 tracing_reset_online_cpus(&tr->array_buffer);
2264 #ifdef CONFIG_TRACER_MAX_TRACE
2265 tracing_reset_online_cpus(&tr->max_buffer);
2270 void tracing_reset_all_online_cpus(void)
2272 mutex_lock(&trace_types_lock);
2273 tracing_reset_all_online_cpus_unlocked();
2274 mutex_unlock(&trace_types_lock);
2278 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2279 * is the tgid last observed corresponding to pid=i.
2281 static int *tgid_map;
2283 /* The maximum valid index into tgid_map. */
2284 static size_t tgid_map_max;
2286 #define SAVED_CMDLINES_DEFAULT 128
2287 #define NO_CMDLINE_MAP UINT_MAX
2289 * Preemption must be disabled before acquiring trace_cmdline_lock.
2290 * The various trace_arrays' max_lock must be acquired in a context
2291 * where interrupt is disabled.
2293 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2294 struct saved_cmdlines_buffer {
2295 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2296 unsigned *map_cmdline_to_pid;
2297 unsigned cmdline_num;
2299 char *saved_cmdlines;
2301 static struct saved_cmdlines_buffer *savedcmd;
2303 static inline char *get_saved_cmdlines(int idx)
2305 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2308 static inline void set_cmdline(int idx, const char *cmdline)
2310 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2313 static int allocate_cmdlines_buffer(unsigned int val,
2314 struct saved_cmdlines_buffer *s)
2316 s->map_cmdline_to_pid = kmalloc_array(val,
2317 sizeof(*s->map_cmdline_to_pid),
2319 if (!s->map_cmdline_to_pid)
2322 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2323 if (!s->saved_cmdlines) {
2324 kfree(s->map_cmdline_to_pid);
2329 s->cmdline_num = val;
2330 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2331 sizeof(s->map_pid_to_cmdline));
2332 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2333 val * sizeof(*s->map_cmdline_to_pid));
2338 static int trace_create_savedcmd(void)
2342 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2346 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2356 int is_tracing_stopped(void)
2358 return global_trace.stop_count;
2362 * tracing_start - quick start of the tracer
2364 * If tracing is enabled but was stopped by tracing_stop,
2365 * this will start the tracer back up.
2367 void tracing_start(void)
2369 struct trace_buffer *buffer;
2370 unsigned long flags;
2372 if (tracing_disabled)
2375 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2376 if (--global_trace.stop_count) {
2377 if (global_trace.stop_count < 0) {
2378 /* Someone screwed up their debugging */
2380 global_trace.stop_count = 0;
2385 /* Prevent the buffers from switching */
2386 arch_spin_lock(&global_trace.max_lock);
2388 buffer = global_trace.array_buffer.buffer;
2390 ring_buffer_record_enable(buffer);
2392 #ifdef CONFIG_TRACER_MAX_TRACE
2393 buffer = global_trace.max_buffer.buffer;
2395 ring_buffer_record_enable(buffer);
2398 arch_spin_unlock(&global_trace.max_lock);
2401 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2404 static void tracing_start_tr(struct trace_array *tr)
2406 struct trace_buffer *buffer;
2407 unsigned long flags;
2409 if (tracing_disabled)
2412 /* If global, we need to also start the max tracer */
2413 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2414 return tracing_start();
2416 raw_spin_lock_irqsave(&tr->start_lock, flags);
2418 if (--tr->stop_count) {
2419 if (tr->stop_count < 0) {
2420 /* Someone screwed up their debugging */
2427 buffer = tr->array_buffer.buffer;
2429 ring_buffer_record_enable(buffer);
2432 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2436 * tracing_stop - quick stop of the tracer
2438 * Light weight way to stop tracing. Use in conjunction with
2441 void tracing_stop(void)
2443 struct trace_buffer *buffer;
2444 unsigned long flags;
2446 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2447 if (global_trace.stop_count++)
2450 /* Prevent the buffers from switching */
2451 arch_spin_lock(&global_trace.max_lock);
2453 buffer = global_trace.array_buffer.buffer;
2455 ring_buffer_record_disable(buffer);
2457 #ifdef CONFIG_TRACER_MAX_TRACE
2458 buffer = global_trace.max_buffer.buffer;
2460 ring_buffer_record_disable(buffer);
2463 arch_spin_unlock(&global_trace.max_lock);
2466 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2469 static void tracing_stop_tr(struct trace_array *tr)
2471 struct trace_buffer *buffer;
2472 unsigned long flags;
2474 /* If global, we need to also stop the max tracer */
2475 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2476 return tracing_stop();
2478 raw_spin_lock_irqsave(&tr->start_lock, flags);
2479 if (tr->stop_count++)
2482 buffer = tr->array_buffer.buffer;
2484 ring_buffer_record_disable(buffer);
2487 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2490 static int trace_save_cmdline(struct task_struct *tsk)
2494 /* treat recording of idle task as a success */
2498 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2501 * It's not the end of the world if we don't get
2502 * the lock, but we also don't want to spin
2503 * nor do we want to disable interrupts,
2504 * so if we miss here, then better luck next time.
2506 * This is called within the scheduler and wake up, so interrupts
2507 * had better been disabled and run queue lock been held.
2509 lockdep_assert_preemption_disabled();
2510 if (!arch_spin_trylock(&trace_cmdline_lock))
2513 idx = savedcmd->map_pid_to_cmdline[tpid];
2514 if (idx == NO_CMDLINE_MAP) {
2515 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2517 savedcmd->map_pid_to_cmdline[tpid] = idx;
2518 savedcmd->cmdline_idx = idx;
2521 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2522 set_cmdline(idx, tsk->comm);
2524 arch_spin_unlock(&trace_cmdline_lock);
2529 static void __trace_find_cmdline(int pid, char comm[])
2535 strcpy(comm, "<idle>");
2539 if (WARN_ON_ONCE(pid < 0)) {
2540 strcpy(comm, "<XXX>");
2544 tpid = pid & (PID_MAX_DEFAULT - 1);
2545 map = savedcmd->map_pid_to_cmdline[tpid];
2546 if (map != NO_CMDLINE_MAP) {
2547 tpid = savedcmd->map_cmdline_to_pid[map];
2549 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2553 strcpy(comm, "<...>");
2556 void trace_find_cmdline(int pid, char comm[])
2559 arch_spin_lock(&trace_cmdline_lock);
2561 __trace_find_cmdline(pid, comm);
2563 arch_spin_unlock(&trace_cmdline_lock);
2567 static int *trace_find_tgid_ptr(int pid)
2570 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2571 * if we observe a non-NULL tgid_map then we also observe the correct
2574 int *map = smp_load_acquire(&tgid_map);
2576 if (unlikely(!map || pid > tgid_map_max))
2582 int trace_find_tgid(int pid)
2584 int *ptr = trace_find_tgid_ptr(pid);
2586 return ptr ? *ptr : 0;
2589 static int trace_save_tgid(struct task_struct *tsk)
2593 /* treat recording of idle task as a success */
2597 ptr = trace_find_tgid_ptr(tsk->pid);
2605 static bool tracing_record_taskinfo_skip(int flags)
2607 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2609 if (!__this_cpu_read(trace_taskinfo_save))
2615 * tracing_record_taskinfo - record the task info of a task
2617 * @task: task to record
2618 * @flags: TRACE_RECORD_CMDLINE for recording comm
2619 * TRACE_RECORD_TGID for recording tgid
2621 void tracing_record_taskinfo(struct task_struct *task, int flags)
2625 if (tracing_record_taskinfo_skip(flags))
2629 * Record as much task information as possible. If some fail, continue
2630 * to try to record the others.
2632 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2633 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2635 /* If recording any information failed, retry again soon. */
2639 __this_cpu_write(trace_taskinfo_save, false);
2643 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2645 * @prev: previous task during sched_switch
2646 * @next: next task during sched_switch
2647 * @flags: TRACE_RECORD_CMDLINE for recording comm
2648 * TRACE_RECORD_TGID for recording tgid
2650 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2651 struct task_struct *next, int flags)
2655 if (tracing_record_taskinfo_skip(flags))
2659 * Record as much task information as possible. If some fail, continue
2660 * to try to record the others.
2662 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2663 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2664 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2665 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2667 /* If recording any information failed, retry again soon. */
2671 __this_cpu_write(trace_taskinfo_save, false);
2674 /* Helpers to record a specific task information */
2675 void tracing_record_cmdline(struct task_struct *task)
2677 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2680 void tracing_record_tgid(struct task_struct *task)
2682 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2686 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2687 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2688 * simplifies those functions and keeps them in sync.
2690 enum print_line_t trace_handle_return(struct trace_seq *s)
2692 return trace_seq_has_overflowed(s) ?
2693 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2695 EXPORT_SYMBOL_GPL(trace_handle_return);
2697 static unsigned short migration_disable_value(void)
2699 #if defined(CONFIG_SMP)
2700 return current->migration_disabled;
2706 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2708 unsigned int trace_flags = irqs_status;
2711 pc = preempt_count();
2714 trace_flags |= TRACE_FLAG_NMI;
2715 if (pc & HARDIRQ_MASK)
2716 trace_flags |= TRACE_FLAG_HARDIRQ;
2717 if (in_serving_softirq())
2718 trace_flags |= TRACE_FLAG_SOFTIRQ;
2719 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2720 trace_flags |= TRACE_FLAG_BH_OFF;
2722 if (tif_need_resched())
2723 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2724 if (test_preempt_need_resched())
2725 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2726 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2727 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2730 struct ring_buffer_event *
2731 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2734 unsigned int trace_ctx)
2736 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2739 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2740 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2741 static int trace_buffered_event_ref;
2744 * trace_buffered_event_enable - enable buffering events
2746 * When events are being filtered, it is quicker to use a temporary
2747 * buffer to write the event data into if there's a likely chance
2748 * that it will not be committed. The discard of the ring buffer
2749 * is not as fast as committing, and is much slower than copying
2752 * When an event is to be filtered, allocate per cpu buffers to
2753 * write the event data into, and if the event is filtered and discarded
2754 * it is simply dropped, otherwise, the entire data is to be committed
2757 void trace_buffered_event_enable(void)
2759 struct ring_buffer_event *event;
2763 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2765 if (trace_buffered_event_ref++)
2768 for_each_tracing_cpu(cpu) {
2769 page = alloc_pages_node(cpu_to_node(cpu),
2770 GFP_KERNEL | __GFP_NORETRY, 0);
2774 event = page_address(page);
2775 memset(event, 0, sizeof(*event));
2777 per_cpu(trace_buffered_event, cpu) = event;
2780 if (cpu == smp_processor_id() &&
2781 __this_cpu_read(trace_buffered_event) !=
2782 per_cpu(trace_buffered_event, cpu))
2789 trace_buffered_event_disable();
2792 static void enable_trace_buffered_event(void *data)
2794 /* Probably not needed, but do it anyway */
2796 this_cpu_dec(trace_buffered_event_cnt);
2799 static void disable_trace_buffered_event(void *data)
2801 this_cpu_inc(trace_buffered_event_cnt);
2805 * trace_buffered_event_disable - disable buffering events
2807 * When a filter is removed, it is faster to not use the buffered
2808 * events, and to commit directly into the ring buffer. Free up
2809 * the temp buffers when there are no more users. This requires
2810 * special synchronization with current events.
2812 void trace_buffered_event_disable(void)
2816 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2818 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2821 if (--trace_buffered_event_ref)
2825 /* For each CPU, set the buffer as used. */
2826 smp_call_function_many(tracing_buffer_mask,
2827 disable_trace_buffered_event, NULL, 1);
2830 /* Wait for all current users to finish */
2833 for_each_tracing_cpu(cpu) {
2834 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2835 per_cpu(trace_buffered_event, cpu) = NULL;
2838 * Make sure trace_buffered_event is NULL before clearing
2839 * trace_buffered_event_cnt.
2844 /* Do the work on each cpu */
2845 smp_call_function_many(tracing_buffer_mask,
2846 enable_trace_buffered_event, NULL, 1);
2850 static struct trace_buffer *temp_buffer;
2852 struct ring_buffer_event *
2853 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2854 struct trace_event_file *trace_file,
2855 int type, unsigned long len,
2856 unsigned int trace_ctx)
2858 struct ring_buffer_event *entry;
2859 struct trace_array *tr = trace_file->tr;
2862 *current_rb = tr->array_buffer.buffer;
2864 if (!tr->no_filter_buffering_ref &&
2865 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2866 preempt_disable_notrace();
2868 * Filtering is on, so try to use the per cpu buffer first.
2869 * This buffer will simulate a ring_buffer_event,
2870 * where the type_len is zero and the array[0] will
2871 * hold the full length.
2872 * (see include/linux/ring-buffer.h for details on
2873 * how the ring_buffer_event is structured).
2875 * Using a temp buffer during filtering and copying it
2876 * on a matched filter is quicker than writing directly
2877 * into the ring buffer and then discarding it when
2878 * it doesn't match. That is because the discard
2879 * requires several atomic operations to get right.
2880 * Copying on match and doing nothing on a failed match
2881 * is still quicker than no copy on match, but having
2882 * to discard out of the ring buffer on a failed match.
2884 if ((entry = __this_cpu_read(trace_buffered_event))) {
2885 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2887 val = this_cpu_inc_return(trace_buffered_event_cnt);
2890 * Preemption is disabled, but interrupts and NMIs
2891 * can still come in now. If that happens after
2892 * the above increment, then it will have to go
2893 * back to the old method of allocating the event
2894 * on the ring buffer, and if the filter fails, it
2895 * will have to call ring_buffer_discard_commit()
2898 * Need to also check the unlikely case that the
2899 * length is bigger than the temp buffer size.
2900 * If that happens, then the reserve is pretty much
2901 * guaranteed to fail, as the ring buffer currently
2902 * only allows events less than a page. But that may
2903 * change in the future, so let the ring buffer reserve
2904 * handle the failure in that case.
2906 if (val == 1 && likely(len <= max_len)) {
2907 trace_event_setup(entry, type, trace_ctx);
2908 entry->array[0] = len;
2909 /* Return with preemption disabled */
2912 this_cpu_dec(trace_buffered_event_cnt);
2914 /* __trace_buffer_lock_reserve() disables preemption */
2915 preempt_enable_notrace();
2918 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2921 * If tracing is off, but we have triggers enabled
2922 * we still need to look at the event data. Use the temp_buffer
2923 * to store the trace event for the trigger to use. It's recursive
2924 * safe and will not be recorded anywhere.
2926 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2927 *current_rb = temp_buffer;
2928 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2933 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2935 static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2936 static DEFINE_MUTEX(tracepoint_printk_mutex);
2938 static void output_printk(struct trace_event_buffer *fbuffer)
2940 struct trace_event_call *event_call;
2941 struct trace_event_file *file;
2942 struct trace_event *event;
2943 unsigned long flags;
2944 struct trace_iterator *iter = tracepoint_print_iter;
2946 /* We should never get here if iter is NULL */
2947 if (WARN_ON_ONCE(!iter))
2950 event_call = fbuffer->trace_file->event_call;
2951 if (!event_call || !event_call->event.funcs ||
2952 !event_call->event.funcs->trace)
2955 file = fbuffer->trace_file;
2956 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2957 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2958 !filter_match_preds(file->filter, fbuffer->entry)))
2961 event = &fbuffer->trace_file->event_call->event;
2963 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2964 trace_seq_init(&iter->seq);
2965 iter->ent = fbuffer->entry;
2966 event_call->event.funcs->trace(iter, 0, event);
2967 trace_seq_putc(&iter->seq, 0);
2968 printk("%s", iter->seq.buffer);
2970 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2973 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2974 void *buffer, size_t *lenp,
2977 int save_tracepoint_printk;
2980 mutex_lock(&tracepoint_printk_mutex);
2981 save_tracepoint_printk = tracepoint_printk;
2983 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2986 * This will force exiting early, as tracepoint_printk
2987 * is always zero when tracepoint_printk_iter is not allocated
2989 if (!tracepoint_print_iter)
2990 tracepoint_printk = 0;
2992 if (save_tracepoint_printk == tracepoint_printk)
2995 if (tracepoint_printk)
2996 static_key_enable(&tracepoint_printk_key.key);
2998 static_key_disable(&tracepoint_printk_key.key);
3001 mutex_unlock(&tracepoint_printk_mutex);
3006 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
3008 enum event_trigger_type tt = ETT_NONE;
3009 struct trace_event_file *file = fbuffer->trace_file;
3011 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
3012 fbuffer->entry, &tt))
3015 if (static_key_false(&tracepoint_printk_key.key))
3016 output_printk(fbuffer);
3018 if (static_branch_unlikely(&trace_event_exports_enabled))
3019 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
3021 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
3022 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
3026 event_triggers_post_call(file, tt);
3029 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
3034 * trace_buffer_unlock_commit_regs()
3035 * trace_event_buffer_commit()
3036 * trace_event_raw_event_xxx()
3038 # define STACK_SKIP 3
3040 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
3041 struct trace_buffer *buffer,
3042 struct ring_buffer_event *event,
3043 unsigned int trace_ctx,
3044 struct pt_regs *regs)
3046 __buffer_unlock_commit(buffer, event);
3049 * If regs is not set, then skip the necessary functions.
3050 * Note, we can still get here via blktrace, wakeup tracer
3051 * and mmiotrace, but that's ok if they lose a function or
3052 * two. They are not that meaningful.
3054 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
3055 ftrace_trace_userstack(tr, buffer, trace_ctx);
3059 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
3062 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
3063 struct ring_buffer_event *event)
3065 __buffer_unlock_commit(buffer, event);
3069 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
3070 parent_ip, unsigned int trace_ctx)
3072 struct trace_event_call *call = &event_function;
3073 struct trace_buffer *buffer = tr->array_buffer.buffer;
3074 struct ring_buffer_event *event;
3075 struct ftrace_entry *entry;
3077 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
3081 entry = ring_buffer_event_data(event);
3083 entry->parent_ip = parent_ip;
3085 if (!call_filter_check_discard(call, entry, buffer, event)) {
3086 if (static_branch_unlikely(&trace_function_exports_enabled))
3087 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
3088 __buffer_unlock_commit(buffer, event);
3092 #ifdef CONFIG_STACKTRACE
3094 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
3095 #define FTRACE_KSTACK_NESTING 4
3097 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3099 struct ftrace_stack {
3100 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3104 struct ftrace_stacks {
3105 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3108 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3109 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3111 static void __ftrace_trace_stack(struct trace_buffer *buffer,
3112 unsigned int trace_ctx,
3113 int skip, struct pt_regs *regs)
3115 struct trace_event_call *call = &event_kernel_stack;
3116 struct ring_buffer_event *event;
3117 unsigned int size, nr_entries;
3118 struct ftrace_stack *fstack;
3119 struct stack_entry *entry;
3123 * Add one, for this function and the call to save_stack_trace()
3124 * If regs is set, then these functions will not be in the way.
3126 #ifndef CONFIG_UNWINDER_ORC
3131 preempt_disable_notrace();
3133 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3135 /* This should never happen. If it does, yell once and skip */
3136 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3140 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3141 * interrupt will either see the value pre increment or post
3142 * increment. If the interrupt happens pre increment it will have
3143 * restored the counter when it returns. We just need a barrier to
3144 * keep gcc from moving things around.
3148 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3149 size = ARRAY_SIZE(fstack->calls);
3152 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3155 nr_entries = stack_trace_save(fstack->calls, size, skip);
3158 size = nr_entries * sizeof(unsigned long);
3159 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3160 (sizeof(*entry) - sizeof(entry->caller)) + size,
3164 entry = ring_buffer_event_data(event);
3166 memcpy(&entry->caller, fstack->calls, size);
3167 entry->size = nr_entries;
3169 if (!call_filter_check_discard(call, entry, buffer, event))
3170 __buffer_unlock_commit(buffer, event);
3173 /* Again, don't let gcc optimize things here */
3175 __this_cpu_dec(ftrace_stack_reserve);
3176 preempt_enable_notrace();
3180 static inline void ftrace_trace_stack(struct trace_array *tr,
3181 struct trace_buffer *buffer,
3182 unsigned int trace_ctx,
3183 int skip, struct pt_regs *regs)
3185 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3188 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3191 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3194 struct trace_buffer *buffer = tr->array_buffer.buffer;
3196 if (rcu_is_watching()) {
3197 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3201 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3205 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3206 * but if the above rcu_is_watching() failed, then the NMI
3207 * triggered someplace critical, and ct_irq_enter() should
3208 * not be called from NMI.
3210 if (unlikely(in_nmi()))
3213 ct_irq_enter_irqson();
3214 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3215 ct_irq_exit_irqson();
3219 * trace_dump_stack - record a stack back trace in the trace buffer
3220 * @skip: Number of functions to skip (helper handlers)
3222 void trace_dump_stack(int skip)
3224 if (tracing_disabled || tracing_selftest_running)
3227 #ifndef CONFIG_UNWINDER_ORC
3228 /* Skip 1 to skip this function. */
3231 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3232 tracing_gen_ctx(), skip, NULL);
3234 EXPORT_SYMBOL_GPL(trace_dump_stack);
3236 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3237 static DEFINE_PER_CPU(int, user_stack_count);
3240 ftrace_trace_userstack(struct trace_array *tr,
3241 struct trace_buffer *buffer, unsigned int trace_ctx)
3243 struct trace_event_call *call = &event_user_stack;
3244 struct ring_buffer_event *event;
3245 struct userstack_entry *entry;
3247 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3251 * NMIs can not handle page faults, even with fix ups.
3252 * The save user stack can (and often does) fault.
3254 if (unlikely(in_nmi()))
3258 * prevent recursion, since the user stack tracing may
3259 * trigger other kernel events.
3262 if (__this_cpu_read(user_stack_count))
3265 __this_cpu_inc(user_stack_count);
3267 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3268 sizeof(*entry), trace_ctx);
3270 goto out_drop_count;
3271 entry = ring_buffer_event_data(event);
3273 entry->tgid = current->tgid;
3274 memset(&entry->caller, 0, sizeof(entry->caller));
3276 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3277 if (!call_filter_check_discard(call, entry, buffer, event))
3278 __buffer_unlock_commit(buffer, event);
3281 __this_cpu_dec(user_stack_count);
3285 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3286 static void ftrace_trace_userstack(struct trace_array *tr,
3287 struct trace_buffer *buffer,
3288 unsigned int trace_ctx)
3291 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3293 #endif /* CONFIG_STACKTRACE */
3296 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3297 unsigned long long delta)
3299 entry->bottom_delta_ts = delta & U32_MAX;
3300 entry->top_delta_ts = (delta >> 32);
3303 void trace_last_func_repeats(struct trace_array *tr,
3304 struct trace_func_repeats *last_info,
3305 unsigned int trace_ctx)
3307 struct trace_buffer *buffer = tr->array_buffer.buffer;
3308 struct func_repeats_entry *entry;
3309 struct ring_buffer_event *event;
3312 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3313 sizeof(*entry), trace_ctx);
3317 delta = ring_buffer_event_time_stamp(buffer, event) -
3318 last_info->ts_last_call;
3320 entry = ring_buffer_event_data(event);
3321 entry->ip = last_info->ip;
3322 entry->parent_ip = last_info->parent_ip;
3323 entry->count = last_info->count;
3324 func_repeats_set_delta_ts(entry, delta);
3326 __buffer_unlock_commit(buffer, event);
3329 /* created for use with alloc_percpu */
3330 struct trace_buffer_struct {
3332 char buffer[4][TRACE_BUF_SIZE];
3335 static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3338 * This allows for lockless recording. If we're nested too deeply, then
3339 * this returns NULL.
3341 static char *get_trace_buf(void)
3343 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3345 if (!trace_percpu_buffer || buffer->nesting >= 4)
3350 /* Interrupts must see nesting incremented before we use the buffer */
3352 return &buffer->buffer[buffer->nesting - 1][0];
3355 static void put_trace_buf(void)
3357 /* Don't let the decrement of nesting leak before this */
3359 this_cpu_dec(trace_percpu_buffer->nesting);
3362 static int alloc_percpu_trace_buffer(void)
3364 struct trace_buffer_struct __percpu *buffers;
3366 if (trace_percpu_buffer)
3369 buffers = alloc_percpu(struct trace_buffer_struct);
3370 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3373 trace_percpu_buffer = buffers;
3377 static int buffers_allocated;
3379 void trace_printk_init_buffers(void)
3381 if (buffers_allocated)
3384 if (alloc_percpu_trace_buffer())
3387 /* trace_printk() is for debug use only. Don't use it in production. */
3390 pr_warn("**********************************************************\n");
3391 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3393 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3395 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3396 pr_warn("** unsafe for production use. **\n");
3398 pr_warn("** If you see this message and you are not debugging **\n");
3399 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3401 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3402 pr_warn("**********************************************************\n");
3404 /* Expand the buffers to set size */
3405 tracing_update_buffers();
3407 buffers_allocated = 1;
3410 * trace_printk_init_buffers() can be called by modules.
3411 * If that happens, then we need to start cmdline recording
3412 * directly here. If the global_trace.buffer is already
3413 * allocated here, then this was called by module code.
3415 if (global_trace.array_buffer.buffer)
3416 tracing_start_cmdline_record();
3418 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3420 void trace_printk_start_comm(void)
3422 /* Start tracing comms if trace printk is set */
3423 if (!buffers_allocated)
3425 tracing_start_cmdline_record();
3428 static void trace_printk_start_stop_comm(int enabled)
3430 if (!buffers_allocated)
3434 tracing_start_cmdline_record();
3436 tracing_stop_cmdline_record();
3440 * trace_vbprintk - write binary msg to tracing buffer
3441 * @ip: The address of the caller
3442 * @fmt: The string format to write to the buffer
3443 * @args: Arguments for @fmt
3445 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3447 struct trace_event_call *call = &event_bprint;
3448 struct ring_buffer_event *event;
3449 struct trace_buffer *buffer;
3450 struct trace_array *tr = &global_trace;
3451 struct bprint_entry *entry;
3452 unsigned int trace_ctx;
3456 if (unlikely(tracing_selftest_running || tracing_disabled))
3459 /* Don't pollute graph traces with trace_vprintk internals */
3460 pause_graph_tracing();
3462 trace_ctx = tracing_gen_ctx();
3463 preempt_disable_notrace();
3465 tbuffer = get_trace_buf();
3471 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3473 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3476 size = sizeof(*entry) + sizeof(u32) * len;
3477 buffer = tr->array_buffer.buffer;
3478 ring_buffer_nest_start(buffer);
3479 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3483 entry = ring_buffer_event_data(event);
3487 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3488 if (!call_filter_check_discard(call, entry, buffer, event)) {
3489 __buffer_unlock_commit(buffer, event);
3490 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3494 ring_buffer_nest_end(buffer);
3499 preempt_enable_notrace();
3500 unpause_graph_tracing();
3504 EXPORT_SYMBOL_GPL(trace_vbprintk);
3508 __trace_array_vprintk(struct trace_buffer *buffer,
3509 unsigned long ip, const char *fmt, va_list args)
3511 struct trace_event_call *call = &event_print;
3512 struct ring_buffer_event *event;
3514 struct print_entry *entry;
3515 unsigned int trace_ctx;
3518 if (tracing_disabled)
3521 /* Don't pollute graph traces with trace_vprintk internals */
3522 pause_graph_tracing();
3524 trace_ctx = tracing_gen_ctx();
3525 preempt_disable_notrace();
3528 tbuffer = get_trace_buf();
3534 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3536 size = sizeof(*entry) + len + 1;
3537 ring_buffer_nest_start(buffer);
3538 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3542 entry = ring_buffer_event_data(event);
3545 memcpy(&entry->buf, tbuffer, len + 1);
3546 if (!call_filter_check_discard(call, entry, buffer, event)) {
3547 __buffer_unlock_commit(buffer, event);
3548 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3552 ring_buffer_nest_end(buffer);
3556 preempt_enable_notrace();
3557 unpause_graph_tracing();
3563 int trace_array_vprintk(struct trace_array *tr,
3564 unsigned long ip, const char *fmt, va_list args)
3566 if (tracing_selftest_running && tr == &global_trace)
3569 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3573 * trace_array_printk - Print a message to a specific instance
3574 * @tr: The instance trace_array descriptor
3575 * @ip: The instruction pointer that this is called from.
3576 * @fmt: The format to print (printf format)
3578 * If a subsystem sets up its own instance, they have the right to
3579 * printk strings into their tracing instance buffer using this
3580 * function. Note, this function will not write into the top level
3581 * buffer (use trace_printk() for that), as writing into the top level
3582 * buffer should only have events that can be individually disabled.
3583 * trace_printk() is only used for debugging a kernel, and should not
3584 * be ever incorporated in normal use.
3586 * trace_array_printk() can be used, as it will not add noise to the
3587 * top level tracing buffer.
3589 * Note, trace_array_init_printk() must be called on @tr before this
3593 int trace_array_printk(struct trace_array *tr,
3594 unsigned long ip, const char *fmt, ...)
3602 /* This is only allowed for created instances */
3603 if (tr == &global_trace)
3606 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3610 ret = trace_array_vprintk(tr, ip, fmt, ap);
3614 EXPORT_SYMBOL_GPL(trace_array_printk);
3617 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3618 * @tr: The trace array to initialize the buffers for
3620 * As trace_array_printk() only writes into instances, they are OK to
3621 * have in the kernel (unlike trace_printk()). This needs to be called
3622 * before trace_array_printk() can be used on a trace_array.
3624 int trace_array_init_printk(struct trace_array *tr)
3629 /* This is only allowed for created instances */
3630 if (tr == &global_trace)
3633 return alloc_percpu_trace_buffer();
3635 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3638 int trace_array_printk_buf(struct trace_buffer *buffer,
3639 unsigned long ip, const char *fmt, ...)
3644 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3648 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3654 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3656 return trace_array_vprintk(&global_trace, ip, fmt, args);
3658 EXPORT_SYMBOL_GPL(trace_vprintk);
3660 static void trace_iterator_increment(struct trace_iterator *iter)
3662 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3666 ring_buffer_iter_advance(buf_iter);
3669 static struct trace_entry *
3670 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3671 unsigned long *lost_events)
3673 struct ring_buffer_event *event;
3674 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3677 event = ring_buffer_iter_peek(buf_iter, ts);
3679 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3680 (unsigned long)-1 : 0;
3682 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3687 iter->ent_size = ring_buffer_event_length(event);
3688 return ring_buffer_event_data(event);
3694 static struct trace_entry *
3695 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3696 unsigned long *missing_events, u64 *ent_ts)
3698 struct trace_buffer *buffer = iter->array_buffer->buffer;
3699 struct trace_entry *ent, *next = NULL;
3700 unsigned long lost_events = 0, next_lost = 0;
3701 int cpu_file = iter->cpu_file;
3702 u64 next_ts = 0, ts;
3708 * If we are in a per_cpu trace file, don't bother by iterating over
3709 * all cpu and peek directly.
3711 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3712 if (ring_buffer_empty_cpu(buffer, cpu_file))
3714 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3716 *ent_cpu = cpu_file;
3721 for_each_tracing_cpu(cpu) {
3723 if (ring_buffer_empty_cpu(buffer, cpu))
3726 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3729 * Pick the entry with the smallest timestamp:
3731 if (ent && (!next || ts < next_ts)) {
3735 next_lost = lost_events;
3736 next_size = iter->ent_size;
3740 iter->ent_size = next_size;
3743 *ent_cpu = next_cpu;
3749 *missing_events = next_lost;
3754 #define STATIC_FMT_BUF_SIZE 128
3755 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3757 char *trace_iter_expand_format(struct trace_iterator *iter)
3762 * iter->tr is NULL when used with tp_printk, which makes
3763 * this get called where it is not safe to call krealloc().
3765 if (!iter->tr || iter->fmt == static_fmt_buf)
3768 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3771 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3778 /* Returns true if the string is safe to dereference from an event */
3779 static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3782 unsigned long addr = (unsigned long)str;
3783 struct trace_event *trace_event;
3784 struct trace_event_call *event;
3786 /* Ignore strings with no length */
3790 /* OK if part of the event data */
3791 if ((addr >= (unsigned long)iter->ent) &&
3792 (addr < (unsigned long)iter->ent + iter->ent_size))
3795 /* OK if part of the temp seq buffer */
3796 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3797 (addr < (unsigned long)iter->tmp_seq.buffer + PAGE_SIZE))
3800 /* Core rodata can not be freed */
3801 if (is_kernel_rodata(addr))
3804 if (trace_is_tracepoint_string(str))
3808 * Now this could be a module event, referencing core module
3809 * data, which is OK.
3814 trace_event = ftrace_find_event(iter->ent->type);
3818 event = container_of(trace_event, struct trace_event_call, event);
3819 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3822 /* Would rather have rodata, but this will suffice */
3823 if (within_module_core(addr, event->module))
3829 static const char *show_buffer(struct trace_seq *s)
3831 struct seq_buf *seq = &s->seq;
3833 seq_buf_terminate(seq);
3838 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3840 static int test_can_verify_check(const char *fmt, ...)
3847 * The verifier is dependent on vsnprintf() modifies the va_list
3848 * passed to it, where it is sent as a reference. Some architectures
3849 * (like x86_32) passes it by value, which means that vsnprintf()
3850 * does not modify the va_list passed to it, and the verifier
3851 * would then need to be able to understand all the values that
3852 * vsnprintf can use. If it is passed by value, then the verifier
3856 vsnprintf(buf, 16, "%d", ap);
3857 ret = va_arg(ap, int);
3863 static void test_can_verify(void)
3865 if (!test_can_verify_check("%d %d", 0, 1)) {
3866 pr_info("trace event string verifier disabled\n");
3867 static_branch_inc(&trace_no_verify);
3872 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3873 * @iter: The iterator that holds the seq buffer and the event being printed
3874 * @fmt: The format used to print the event
3875 * @ap: The va_list holding the data to print from @fmt.
3877 * This writes the data into the @iter->seq buffer using the data from
3878 * @fmt and @ap. If the format has a %s, then the source of the string
3879 * is examined to make sure it is safe to print, otherwise it will
3880 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3883 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3886 const char *p = fmt;
3890 if (WARN_ON_ONCE(!fmt))
3893 if (static_branch_unlikely(&trace_no_verify))
3896 /* Don't bother checking when doing a ftrace_dump() */
3897 if (iter->fmt == static_fmt_buf)
3906 /* We only care about %s and variants */
3907 for (i = 0; p[i]; i++) {
3908 if (i + 1 >= iter->fmt_size) {
3910 * If we can't expand the copy buffer,
3913 if (!trace_iter_expand_format(iter))
3917 if (p[i] == '\\' && p[i+1]) {
3922 /* Need to test cases like %08.*s */
3923 for (j = 1; p[i+j]; j++) {
3924 if (isdigit(p[i+j]) ||
3927 if (p[i+j] == '*') {
3939 /* If no %s found then just print normally */
3943 /* Copy up to the %s, and print that */
3944 strncpy(iter->fmt, p, i);
3945 iter->fmt[i] = '\0';
3946 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3949 * If iter->seq is full, the above call no longer guarantees
3950 * that ap is in sync with fmt processing, and further calls
3951 * to va_arg() can return wrong positional arguments.
3953 * Ensure that ap is no longer used in this case.
3955 if (iter->seq.full) {
3961 len = va_arg(ap, int);
3963 /* The ap now points to the string data of the %s */
3964 str = va_arg(ap, const char *);
3967 * If you hit this warning, it is likely that the
3968 * trace event in question used %s on a string that
3969 * was saved at the time of the event, but may not be
3970 * around when the trace is read. Use __string(),
3971 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3972 * instead. See samples/trace_events/trace-events-sample.h
3975 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3976 "fmt: '%s' current_buffer: '%s'",
3977 fmt, show_buffer(&iter->seq))) {
3980 /* Try to safely read the string */
3982 if (len + 1 > iter->fmt_size)
3983 len = iter->fmt_size - 1;
3986 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3990 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3994 trace_seq_printf(&iter->seq, "(0x%px)", str);
3996 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3998 str = "[UNSAFE-MEMORY]";
3999 strcpy(iter->fmt, "%s");
4001 strncpy(iter->fmt, p + i, j + 1);
4002 iter->fmt[j+1] = '\0';
4005 trace_seq_printf(&iter->seq, iter->fmt, len, str);
4007 trace_seq_printf(&iter->seq, iter->fmt, str);
4013 trace_seq_vprintf(&iter->seq, p, ap);
4016 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
4018 const char *p, *new_fmt;
4021 if (WARN_ON_ONCE(!fmt))
4024 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
4028 new_fmt = q = iter->fmt;
4030 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
4031 if (!trace_iter_expand_format(iter))
4034 q += iter->fmt - new_fmt;
4035 new_fmt = iter->fmt;
4040 /* Replace %p with %px */
4044 } else if (p[0] == 'p' && !isalnum(p[1])) {
4055 #define STATIC_TEMP_BUF_SIZE 128
4056 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
4058 /* Find the next real entry, without updating the iterator itself */
4059 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
4060 int *ent_cpu, u64 *ent_ts)
4062 /* __find_next_entry will reset ent_size */
4063 int ent_size = iter->ent_size;
4064 struct trace_entry *entry;
4067 * If called from ftrace_dump(), then the iter->temp buffer
4068 * will be the static_temp_buf and not created from kmalloc.
4069 * If the entry size is greater than the buffer, we can
4070 * not save it. Just return NULL in that case. This is only
4071 * used to add markers when two consecutive events' time
4072 * stamps have a large delta. See trace_print_lat_context()
4074 if (iter->temp == static_temp_buf &&
4075 STATIC_TEMP_BUF_SIZE < ent_size)
4079 * The __find_next_entry() may call peek_next_entry(), which may
4080 * call ring_buffer_peek() that may make the contents of iter->ent
4081 * undefined. Need to copy iter->ent now.
4083 if (iter->ent && iter->ent != iter->temp) {
4084 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4085 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4087 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4092 iter->temp_size = iter->ent_size;
4094 memcpy(iter->temp, iter->ent, iter->ent_size);
4095 iter->ent = iter->temp;
4097 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4098 /* Put back the original ent_size */
4099 iter->ent_size = ent_size;
4104 /* Find the next real entry, and increment the iterator to the next entry */
4105 void *trace_find_next_entry_inc(struct trace_iterator *iter)
4107 iter->ent = __find_next_entry(iter, &iter->cpu,
4108 &iter->lost_events, &iter->ts);
4111 trace_iterator_increment(iter);
4113 return iter->ent ? iter : NULL;
4116 static void trace_consume(struct trace_iterator *iter)
4118 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4119 &iter->lost_events);
4122 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4124 struct trace_iterator *iter = m->private;
4128 WARN_ON_ONCE(iter->leftover);
4132 /* can't go backwards */
4137 ent = trace_find_next_entry_inc(iter);
4141 while (ent && iter->idx < i)
4142 ent = trace_find_next_entry_inc(iter);
4149 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4151 struct ring_buffer_iter *buf_iter;
4152 unsigned long entries = 0;
4155 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4157 buf_iter = trace_buffer_iter(iter, cpu);
4161 ring_buffer_iter_reset(buf_iter);
4164 * We could have the case with the max latency tracers
4165 * that a reset never took place on a cpu. This is evident
4166 * by the timestamp being before the start of the buffer.
4168 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4169 if (ts >= iter->array_buffer->time_start)
4172 ring_buffer_iter_advance(buf_iter);
4175 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4179 * The current tracer is copied to avoid a global locking
4182 static void *s_start(struct seq_file *m, loff_t *pos)
4184 struct trace_iterator *iter = m->private;
4185 struct trace_array *tr = iter->tr;
4186 int cpu_file = iter->cpu_file;
4192 * copy the tracer to avoid using a global lock all around.
4193 * iter->trace is a copy of current_trace, the pointer to the
4194 * name may be used instead of a strcmp(), as iter->trace->name
4195 * will point to the same string as current_trace->name.
4197 mutex_lock(&trace_types_lock);
4198 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
4199 *iter->trace = *tr->current_trace;
4200 mutex_unlock(&trace_types_lock);
4202 #ifdef CONFIG_TRACER_MAX_TRACE
4203 if (iter->snapshot && iter->trace->use_max_tr)
4204 return ERR_PTR(-EBUSY);
4207 if (*pos != iter->pos) {
4212 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4213 for_each_tracing_cpu(cpu)
4214 tracing_iter_reset(iter, cpu);
4216 tracing_iter_reset(iter, cpu_file);
4219 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4224 * If we overflowed the seq_file before, then we want
4225 * to just reuse the trace_seq buffer again.
4231 p = s_next(m, p, &l);
4235 trace_event_read_lock();
4236 trace_access_lock(cpu_file);
4240 static void s_stop(struct seq_file *m, void *p)
4242 struct trace_iterator *iter = m->private;
4244 #ifdef CONFIG_TRACER_MAX_TRACE
4245 if (iter->snapshot && iter->trace->use_max_tr)
4249 trace_access_unlock(iter->cpu_file);
4250 trace_event_read_unlock();
4254 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4255 unsigned long *entries, int cpu)
4257 unsigned long count;
4259 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4261 * If this buffer has skipped entries, then we hold all
4262 * entries for the trace and we need to ignore the
4263 * ones before the time stamp.
4265 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4266 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4267 /* total is the same as the entries */
4271 ring_buffer_overrun_cpu(buf->buffer, cpu);
4276 get_total_entries(struct array_buffer *buf,
4277 unsigned long *total, unsigned long *entries)
4285 for_each_tracing_cpu(cpu) {
4286 get_total_entries_cpu(buf, &t, &e, cpu);
4292 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4294 unsigned long total, entries;
4299 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4304 unsigned long trace_total_entries(struct trace_array *tr)
4306 unsigned long total, entries;
4311 get_total_entries(&tr->array_buffer, &total, &entries);
4316 static void print_lat_help_header(struct seq_file *m)
4318 seq_puts(m, "# _------=> CPU# \n"
4319 "# / _-----=> irqs-off/BH-disabled\n"
4320 "# | / _----=> need-resched \n"
4321 "# || / _---=> hardirq/softirq \n"
4322 "# ||| / _--=> preempt-depth \n"
4323 "# |||| / _-=> migrate-disable \n"
4324 "# ||||| / delay \n"
4325 "# cmd pid |||||| time | caller \n"
4326 "# \\ / |||||| \\ | / \n");
4329 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4331 unsigned long total;
4332 unsigned long entries;
4334 get_total_entries(buf, &total, &entries);
4335 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4336 entries, total, num_online_cpus());
4340 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4343 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4345 print_event_info(buf, m);
4347 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4348 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4351 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4354 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4355 static const char space[] = " ";
4356 int prec = tgid ? 12 : 2;
4358 print_event_info(buf, m);
4360 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4361 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4362 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4363 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4364 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4365 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4366 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4367 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4371 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4373 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4374 struct array_buffer *buf = iter->array_buffer;
4375 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4376 struct tracer *type = iter->trace;
4377 unsigned long entries;
4378 unsigned long total;
4379 const char *name = type->name;
4381 get_total_entries(buf, &total, &entries);
4383 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4385 seq_puts(m, "# -----------------------------------"
4386 "---------------------------------\n");
4387 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4388 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4389 nsecs_to_usecs(data->saved_latency),
4393 preempt_model_none() ? "server" :
4394 preempt_model_voluntary() ? "desktop" :
4395 preempt_model_full() ? "preempt" :
4396 preempt_model_rt() ? "preempt_rt" :
4398 /* These are reserved for later use */
4401 seq_printf(m, " #P:%d)\n", num_online_cpus());
4405 seq_puts(m, "# -----------------\n");
4406 seq_printf(m, "# | task: %.16s-%d "
4407 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4408 data->comm, data->pid,
4409 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4410 data->policy, data->rt_priority);
4411 seq_puts(m, "# -----------------\n");
4413 if (data->critical_start) {
4414 seq_puts(m, "# => started at: ");
4415 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4416 trace_print_seq(m, &iter->seq);
4417 seq_puts(m, "\n# => ended at: ");
4418 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4419 trace_print_seq(m, &iter->seq);
4420 seq_puts(m, "\n#\n");
4426 static void test_cpu_buff_start(struct trace_iterator *iter)
4428 struct trace_seq *s = &iter->seq;
4429 struct trace_array *tr = iter->tr;
4431 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4434 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4437 if (cpumask_available(iter->started) &&
4438 cpumask_test_cpu(iter->cpu, iter->started))
4441 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4444 if (cpumask_available(iter->started))
4445 cpumask_set_cpu(iter->cpu, iter->started);
4447 /* Don't print started cpu buffer for the first entry of the trace */
4449 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4453 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4455 struct trace_array *tr = iter->tr;
4456 struct trace_seq *s = &iter->seq;
4457 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4458 struct trace_entry *entry;
4459 struct trace_event *event;
4463 test_cpu_buff_start(iter);
4465 event = ftrace_find_event(entry->type);
4467 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4468 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4469 trace_print_lat_context(iter);
4471 trace_print_context(iter);
4474 if (trace_seq_has_overflowed(s))
4475 return TRACE_TYPE_PARTIAL_LINE;
4478 if (tr->trace_flags & TRACE_ITER_FIELDS)
4479 return print_event_fields(iter, event);
4480 return event->funcs->trace(iter, sym_flags, event);
4483 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4485 return trace_handle_return(s);
4488 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4490 struct trace_array *tr = iter->tr;
4491 struct trace_seq *s = &iter->seq;
4492 struct trace_entry *entry;
4493 struct trace_event *event;
4497 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4498 trace_seq_printf(s, "%d %d %llu ",
4499 entry->pid, iter->cpu, iter->ts);
4501 if (trace_seq_has_overflowed(s))
4502 return TRACE_TYPE_PARTIAL_LINE;
4504 event = ftrace_find_event(entry->type);
4506 return event->funcs->raw(iter, 0, event);
4508 trace_seq_printf(s, "%d ?\n", entry->type);
4510 return trace_handle_return(s);
4513 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4515 struct trace_array *tr = iter->tr;
4516 struct trace_seq *s = &iter->seq;
4517 unsigned char newline = '\n';
4518 struct trace_entry *entry;
4519 struct trace_event *event;
4523 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4524 SEQ_PUT_HEX_FIELD(s, entry->pid);
4525 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4526 SEQ_PUT_HEX_FIELD(s, iter->ts);
4527 if (trace_seq_has_overflowed(s))
4528 return TRACE_TYPE_PARTIAL_LINE;
4531 event = ftrace_find_event(entry->type);
4533 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4534 if (ret != TRACE_TYPE_HANDLED)
4538 SEQ_PUT_FIELD(s, newline);
4540 return trace_handle_return(s);
4543 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4545 struct trace_array *tr = iter->tr;
4546 struct trace_seq *s = &iter->seq;
4547 struct trace_entry *entry;
4548 struct trace_event *event;
4552 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4553 SEQ_PUT_FIELD(s, entry->pid);
4554 SEQ_PUT_FIELD(s, iter->cpu);
4555 SEQ_PUT_FIELD(s, iter->ts);
4556 if (trace_seq_has_overflowed(s))
4557 return TRACE_TYPE_PARTIAL_LINE;
4560 event = ftrace_find_event(entry->type);
4561 return event ? event->funcs->binary(iter, 0, event) :
4565 int trace_empty(struct trace_iterator *iter)
4567 struct ring_buffer_iter *buf_iter;
4570 /* If we are looking at one CPU buffer, only check that one */
4571 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4572 cpu = iter->cpu_file;
4573 buf_iter = trace_buffer_iter(iter, cpu);
4575 if (!ring_buffer_iter_empty(buf_iter))
4578 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4584 for_each_tracing_cpu(cpu) {
4585 buf_iter = trace_buffer_iter(iter, cpu);
4587 if (!ring_buffer_iter_empty(buf_iter))
4590 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4598 /* Called with trace_event_read_lock() held. */
4599 enum print_line_t print_trace_line(struct trace_iterator *iter)
4601 struct trace_array *tr = iter->tr;
4602 unsigned long trace_flags = tr->trace_flags;
4603 enum print_line_t ret;
4605 if (iter->lost_events) {
4606 if (iter->lost_events == (unsigned long)-1)
4607 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4610 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4611 iter->cpu, iter->lost_events);
4612 if (trace_seq_has_overflowed(&iter->seq))
4613 return TRACE_TYPE_PARTIAL_LINE;
4616 if (iter->trace && iter->trace->print_line) {
4617 ret = iter->trace->print_line(iter);
4618 if (ret != TRACE_TYPE_UNHANDLED)
4622 if (iter->ent->type == TRACE_BPUTS &&
4623 trace_flags & TRACE_ITER_PRINTK &&
4624 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4625 return trace_print_bputs_msg_only(iter);
4627 if (iter->ent->type == TRACE_BPRINT &&
4628 trace_flags & TRACE_ITER_PRINTK &&
4629 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4630 return trace_print_bprintk_msg_only(iter);
4632 if (iter->ent->type == TRACE_PRINT &&
4633 trace_flags & TRACE_ITER_PRINTK &&
4634 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4635 return trace_print_printk_msg_only(iter);
4637 if (trace_flags & TRACE_ITER_BIN)
4638 return print_bin_fmt(iter);
4640 if (trace_flags & TRACE_ITER_HEX)
4641 return print_hex_fmt(iter);
4643 if (trace_flags & TRACE_ITER_RAW)
4644 return print_raw_fmt(iter);
4646 return print_trace_fmt(iter);
4649 void trace_latency_header(struct seq_file *m)
4651 struct trace_iterator *iter = m->private;
4652 struct trace_array *tr = iter->tr;
4654 /* print nothing if the buffers are empty */
4655 if (trace_empty(iter))
4658 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4659 print_trace_header(m, iter);
4661 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4662 print_lat_help_header(m);
4665 void trace_default_header(struct seq_file *m)
4667 struct trace_iterator *iter = m->private;
4668 struct trace_array *tr = iter->tr;
4669 unsigned long trace_flags = tr->trace_flags;
4671 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4674 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4675 /* print nothing if the buffers are empty */
4676 if (trace_empty(iter))
4678 print_trace_header(m, iter);
4679 if (!(trace_flags & TRACE_ITER_VERBOSE))
4680 print_lat_help_header(m);
4682 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4683 if (trace_flags & TRACE_ITER_IRQ_INFO)
4684 print_func_help_header_irq(iter->array_buffer,
4687 print_func_help_header(iter->array_buffer, m,
4693 static void test_ftrace_alive(struct seq_file *m)
4695 if (!ftrace_is_dead())
4697 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4698 "# MAY BE MISSING FUNCTION EVENTS\n");
4701 #ifdef CONFIG_TRACER_MAX_TRACE
4702 static void show_snapshot_main_help(struct seq_file *m)
4704 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4705 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4706 "# Takes a snapshot of the main buffer.\n"
4707 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4708 "# (Doesn't have to be '2' works with any number that\n"
4709 "# is not a '0' or '1')\n");
4712 static void show_snapshot_percpu_help(struct seq_file *m)
4714 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4715 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4716 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4717 "# Takes a snapshot of the main buffer for this cpu.\n");
4719 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4720 "# Must use main snapshot file to allocate.\n");
4722 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4723 "# (Doesn't have to be '2' works with any number that\n"
4724 "# is not a '0' or '1')\n");
4727 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4729 if (iter->tr->allocated_snapshot)
4730 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4732 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4734 seq_puts(m, "# Snapshot commands:\n");
4735 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4736 show_snapshot_main_help(m);
4738 show_snapshot_percpu_help(m);
4741 /* Should never be called */
4742 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4745 static int s_show(struct seq_file *m, void *v)
4747 struct trace_iterator *iter = v;
4750 if (iter->ent == NULL) {
4752 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4754 test_ftrace_alive(m);
4756 if (iter->snapshot && trace_empty(iter))
4757 print_snapshot_help(m, iter);
4758 else if (iter->trace && iter->trace->print_header)
4759 iter->trace->print_header(m);
4761 trace_default_header(m);
4763 } else if (iter->leftover) {
4765 * If we filled the seq_file buffer earlier, we
4766 * want to just show it now.
4768 ret = trace_print_seq(m, &iter->seq);
4770 /* ret should this time be zero, but you never know */
4771 iter->leftover = ret;
4774 print_trace_line(iter);
4775 ret = trace_print_seq(m, &iter->seq);
4777 * If we overflow the seq_file buffer, then it will
4778 * ask us for this data again at start up.
4780 * ret is 0 if seq_file write succeeded.
4783 iter->leftover = ret;
4790 * Should be used after trace_array_get(), trace_types_lock
4791 * ensures that i_cdev was already initialized.
4793 static inline int tracing_get_cpu(struct inode *inode)
4795 if (inode->i_cdev) /* See trace_create_cpu_file() */
4796 return (long)inode->i_cdev - 1;
4797 return RING_BUFFER_ALL_CPUS;
4800 static const struct seq_operations tracer_seq_ops = {
4807 static struct trace_iterator *
4808 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4810 struct trace_array *tr = inode->i_private;
4811 struct trace_iterator *iter;
4814 if (tracing_disabled)
4815 return ERR_PTR(-ENODEV);
4817 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4819 return ERR_PTR(-ENOMEM);
4821 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4823 if (!iter->buffer_iter)
4827 * trace_find_next_entry() may need to save off iter->ent.
4828 * It will place it into the iter->temp buffer. As most
4829 * events are less than 128, allocate a buffer of that size.
4830 * If one is greater, then trace_find_next_entry() will
4831 * allocate a new buffer to adjust for the bigger iter->ent.
4832 * It's not critical if it fails to get allocated here.
4834 iter->temp = kmalloc(128, GFP_KERNEL);
4836 iter->temp_size = 128;
4839 * trace_event_printf() may need to modify given format
4840 * string to replace %p with %px so that it shows real address
4841 * instead of hash value. However, that is only for the event
4842 * tracing, other tracer may not need. Defer the allocation
4843 * until it is needed.
4849 * We make a copy of the current tracer to avoid concurrent
4850 * changes on it while we are reading.
4852 mutex_lock(&trace_types_lock);
4853 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4857 *iter->trace = *tr->current_trace;
4859 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4864 #ifdef CONFIG_TRACER_MAX_TRACE
4865 /* Currently only the top directory has a snapshot */
4866 if (tr->current_trace->print_max || snapshot)
4867 iter->array_buffer = &tr->max_buffer;
4870 iter->array_buffer = &tr->array_buffer;
4871 iter->snapshot = snapshot;
4873 iter->cpu_file = tracing_get_cpu(inode);
4874 mutex_init(&iter->mutex);
4876 /* Notify the tracer early; before we stop tracing. */
4877 if (iter->trace->open)
4878 iter->trace->open(iter);
4880 /* Annotate start of buffers if we had overruns */
4881 if (ring_buffer_overruns(iter->array_buffer->buffer))
4882 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4884 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4885 if (trace_clocks[tr->clock_id].in_ns)
4886 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4889 * If pause-on-trace is enabled, then stop the trace while
4890 * dumping, unless this is the "snapshot" file
4892 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4893 tracing_stop_tr(tr);
4895 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4896 for_each_tracing_cpu(cpu) {
4897 iter->buffer_iter[cpu] =
4898 ring_buffer_read_prepare(iter->array_buffer->buffer,
4901 ring_buffer_read_prepare_sync();
4902 for_each_tracing_cpu(cpu) {
4903 ring_buffer_read_start(iter->buffer_iter[cpu]);
4904 tracing_iter_reset(iter, cpu);
4907 cpu = iter->cpu_file;
4908 iter->buffer_iter[cpu] =
4909 ring_buffer_read_prepare(iter->array_buffer->buffer,
4911 ring_buffer_read_prepare_sync();
4912 ring_buffer_read_start(iter->buffer_iter[cpu]);
4913 tracing_iter_reset(iter, cpu);
4916 mutex_unlock(&trace_types_lock);
4921 mutex_unlock(&trace_types_lock);
4924 kfree(iter->buffer_iter);
4926 seq_release_private(inode, file);
4927 return ERR_PTR(-ENOMEM);
4930 int tracing_open_generic(struct inode *inode, struct file *filp)
4934 ret = tracing_check_open_get_tr(NULL);
4938 filp->private_data = inode->i_private;
4942 bool tracing_is_disabled(void)
4944 return (tracing_disabled) ? true: false;
4948 * Open and update trace_array ref count.
4949 * Must have the current trace_array passed to it.
4951 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4953 struct trace_array *tr = inode->i_private;
4956 ret = tracing_check_open_get_tr(tr);
4960 filp->private_data = inode->i_private;
4965 static int tracing_mark_open(struct inode *inode, struct file *filp)
4967 stream_open(inode, filp);
4968 return tracing_open_generic_tr(inode, filp);
4971 static int tracing_release(struct inode *inode, struct file *file)
4973 struct trace_array *tr = inode->i_private;
4974 struct seq_file *m = file->private_data;
4975 struct trace_iterator *iter;
4978 if (!(file->f_mode & FMODE_READ)) {
4979 trace_array_put(tr);
4983 /* Writes do not use seq_file */
4985 mutex_lock(&trace_types_lock);
4987 for_each_tracing_cpu(cpu) {
4988 if (iter->buffer_iter[cpu])
4989 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4992 if (iter->trace && iter->trace->close)
4993 iter->trace->close(iter);
4995 if (!iter->snapshot && tr->stop_count)
4996 /* reenable tracing if it was previously enabled */
4997 tracing_start_tr(tr);
4999 __trace_array_put(tr);
5001 mutex_unlock(&trace_types_lock);
5003 mutex_destroy(&iter->mutex);
5004 free_cpumask_var(iter->started);
5008 kfree(iter->buffer_iter);
5009 seq_release_private(inode, file);
5014 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
5016 struct trace_array *tr = inode->i_private;
5018 trace_array_put(tr);
5022 static int tracing_single_release_tr(struct inode *inode, struct file *file)
5024 struct trace_array *tr = inode->i_private;
5026 trace_array_put(tr);
5028 return single_release(inode, file);
5031 static int tracing_open(struct inode *inode, struct file *file)
5033 struct trace_array *tr = inode->i_private;
5034 struct trace_iterator *iter;
5037 ret = tracing_check_open_get_tr(tr);
5041 /* If this file was open for write, then erase contents */
5042 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
5043 int cpu = tracing_get_cpu(inode);
5044 struct array_buffer *trace_buf = &tr->array_buffer;
5046 #ifdef CONFIG_TRACER_MAX_TRACE
5047 if (tr->current_trace->print_max)
5048 trace_buf = &tr->max_buffer;
5051 if (cpu == RING_BUFFER_ALL_CPUS)
5052 tracing_reset_online_cpus(trace_buf);
5054 tracing_reset_cpu(trace_buf, cpu);
5057 if (file->f_mode & FMODE_READ) {
5058 iter = __tracing_open(inode, file, false);
5060 ret = PTR_ERR(iter);
5061 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
5062 iter->iter_flags |= TRACE_FILE_LAT_FMT;
5066 trace_array_put(tr);
5072 * Some tracers are not suitable for instance buffers.
5073 * A tracer is always available for the global array (toplevel)
5074 * or if it explicitly states that it is.
5077 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
5079 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5082 /* Find the next tracer that this trace array may use */
5083 static struct tracer *
5084 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5086 while (t && !trace_ok_for_array(t, tr))
5093 t_next(struct seq_file *m, void *v, loff_t *pos)
5095 struct trace_array *tr = m->private;
5096 struct tracer *t = v;
5101 t = get_tracer_for_array(tr, t->next);
5106 static void *t_start(struct seq_file *m, loff_t *pos)
5108 struct trace_array *tr = m->private;
5112 mutex_lock(&trace_types_lock);
5114 t = get_tracer_for_array(tr, trace_types);
5115 for (; t && l < *pos; t = t_next(m, t, &l))
5121 static void t_stop(struct seq_file *m, void *p)
5123 mutex_unlock(&trace_types_lock);
5126 static int t_show(struct seq_file *m, void *v)
5128 struct tracer *t = v;
5133 seq_puts(m, t->name);
5142 static const struct seq_operations show_traces_seq_ops = {
5149 static int show_traces_open(struct inode *inode, struct file *file)
5151 struct trace_array *tr = inode->i_private;
5155 ret = tracing_check_open_get_tr(tr);
5159 ret = seq_open(file, &show_traces_seq_ops);
5161 trace_array_put(tr);
5165 m = file->private_data;
5171 static int show_traces_release(struct inode *inode, struct file *file)
5173 struct trace_array *tr = inode->i_private;
5175 trace_array_put(tr);
5176 return seq_release(inode, file);
5180 tracing_write_stub(struct file *filp, const char __user *ubuf,
5181 size_t count, loff_t *ppos)
5186 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5190 if (file->f_mode & FMODE_READ)
5191 ret = seq_lseek(file, offset, whence);
5193 file->f_pos = ret = 0;
5198 static const struct file_operations tracing_fops = {
5199 .open = tracing_open,
5201 .read_iter = seq_read_iter,
5202 .splice_read = generic_file_splice_read,
5203 .write = tracing_write_stub,
5204 .llseek = tracing_lseek,
5205 .release = tracing_release,
5208 static const struct file_operations show_traces_fops = {
5209 .open = show_traces_open,
5211 .llseek = seq_lseek,
5212 .release = show_traces_release,
5216 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5217 size_t count, loff_t *ppos)
5219 struct trace_array *tr = file_inode(filp)->i_private;
5223 len = snprintf(NULL, 0, "%*pb\n",
5224 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5225 mask_str = kmalloc(len, GFP_KERNEL);
5229 len = snprintf(mask_str, len, "%*pb\n",
5230 cpumask_pr_args(tr->tracing_cpumask));
5235 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5243 int tracing_set_cpumask(struct trace_array *tr,
5244 cpumask_var_t tracing_cpumask_new)
5251 local_irq_disable();
5252 arch_spin_lock(&tr->max_lock);
5253 for_each_tracing_cpu(cpu) {
5255 * Increase/decrease the disabled counter if we are
5256 * about to flip a bit in the cpumask:
5258 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5259 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5260 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5261 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5263 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5264 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5265 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5266 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5269 arch_spin_unlock(&tr->max_lock);
5272 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5278 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5279 size_t count, loff_t *ppos)
5281 struct trace_array *tr = file_inode(filp)->i_private;
5282 cpumask_var_t tracing_cpumask_new;
5285 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5288 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5292 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5296 free_cpumask_var(tracing_cpumask_new);
5301 free_cpumask_var(tracing_cpumask_new);
5306 static const struct file_operations tracing_cpumask_fops = {
5307 .open = tracing_open_generic_tr,
5308 .read = tracing_cpumask_read,
5309 .write = tracing_cpumask_write,
5310 .release = tracing_release_generic_tr,
5311 .llseek = generic_file_llseek,
5314 static int tracing_trace_options_show(struct seq_file *m, void *v)
5316 struct tracer_opt *trace_opts;
5317 struct trace_array *tr = m->private;
5321 mutex_lock(&trace_types_lock);
5322 tracer_flags = tr->current_trace->flags->val;
5323 trace_opts = tr->current_trace->flags->opts;
5325 for (i = 0; trace_options[i]; i++) {
5326 if (tr->trace_flags & (1 << i))
5327 seq_printf(m, "%s\n", trace_options[i]);
5329 seq_printf(m, "no%s\n", trace_options[i]);
5332 for (i = 0; trace_opts[i].name; i++) {
5333 if (tracer_flags & trace_opts[i].bit)
5334 seq_printf(m, "%s\n", trace_opts[i].name);
5336 seq_printf(m, "no%s\n", trace_opts[i].name);
5338 mutex_unlock(&trace_types_lock);
5343 static int __set_tracer_option(struct trace_array *tr,
5344 struct tracer_flags *tracer_flags,
5345 struct tracer_opt *opts, int neg)
5347 struct tracer *trace = tracer_flags->trace;
5350 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5355 tracer_flags->val &= ~opts->bit;
5357 tracer_flags->val |= opts->bit;
5361 /* Try to assign a tracer specific option */
5362 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5364 struct tracer *trace = tr->current_trace;
5365 struct tracer_flags *tracer_flags = trace->flags;
5366 struct tracer_opt *opts = NULL;
5369 for (i = 0; tracer_flags->opts[i].name; i++) {
5370 opts = &tracer_flags->opts[i];
5372 if (strcmp(cmp, opts->name) == 0)
5373 return __set_tracer_option(tr, trace->flags, opts, neg);
5379 /* Some tracers require overwrite to stay enabled */
5380 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5382 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5388 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5392 if ((mask == TRACE_ITER_RECORD_TGID) ||
5393 (mask == TRACE_ITER_RECORD_CMD))
5394 lockdep_assert_held(&event_mutex);
5396 /* do nothing if flag is already set */
5397 if (!!(tr->trace_flags & mask) == !!enabled)
5400 /* Give the tracer a chance to approve the change */
5401 if (tr->current_trace->flag_changed)
5402 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5406 tr->trace_flags |= mask;
5408 tr->trace_flags &= ~mask;
5410 if (mask == TRACE_ITER_RECORD_CMD)
5411 trace_event_enable_cmd_record(enabled);
5413 if (mask == TRACE_ITER_RECORD_TGID) {
5415 tgid_map_max = pid_max;
5416 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5420 * Pairs with smp_load_acquire() in
5421 * trace_find_tgid_ptr() to ensure that if it observes
5422 * the tgid_map we just allocated then it also observes
5423 * the corresponding tgid_map_max value.
5425 smp_store_release(&tgid_map, map);
5428 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5432 trace_event_enable_tgid_record(enabled);
5435 if (mask == TRACE_ITER_EVENT_FORK)
5436 trace_event_follow_fork(tr, enabled);
5438 if (mask == TRACE_ITER_FUNC_FORK)
5439 ftrace_pid_follow_fork(tr, enabled);
5441 if (mask == TRACE_ITER_OVERWRITE) {
5442 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5443 #ifdef CONFIG_TRACER_MAX_TRACE
5444 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5448 if (mask == TRACE_ITER_PRINTK) {
5449 trace_printk_start_stop_comm(enabled);
5450 trace_printk_control(enabled);
5456 int trace_set_options(struct trace_array *tr, char *option)
5461 size_t orig_len = strlen(option);
5464 cmp = strstrip(option);
5466 len = str_has_prefix(cmp, "no");
5472 mutex_lock(&event_mutex);
5473 mutex_lock(&trace_types_lock);
5475 ret = match_string(trace_options, -1, cmp);
5476 /* If no option could be set, test the specific tracer options */
5478 ret = set_tracer_option(tr, cmp, neg);
5480 ret = set_tracer_flag(tr, 1 << ret, !neg);
5482 mutex_unlock(&trace_types_lock);
5483 mutex_unlock(&event_mutex);
5486 * If the first trailing whitespace is replaced with '\0' by strstrip,
5487 * turn it back into a space.
5489 if (orig_len > strlen(option))
5490 option[strlen(option)] = ' ';
5495 static void __init apply_trace_boot_options(void)
5497 char *buf = trace_boot_options_buf;
5501 option = strsep(&buf, ",");
5507 trace_set_options(&global_trace, option);
5509 /* Put back the comma to allow this to be called again */
5516 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5517 size_t cnt, loff_t *ppos)
5519 struct seq_file *m = filp->private_data;
5520 struct trace_array *tr = m->private;
5524 if (cnt >= sizeof(buf))
5527 if (copy_from_user(buf, ubuf, cnt))
5532 ret = trace_set_options(tr, buf);
5541 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5543 struct trace_array *tr = inode->i_private;
5546 ret = tracing_check_open_get_tr(tr);
5550 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5552 trace_array_put(tr);
5557 static const struct file_operations tracing_iter_fops = {
5558 .open = tracing_trace_options_open,
5560 .llseek = seq_lseek,
5561 .release = tracing_single_release_tr,
5562 .write = tracing_trace_options_write,
5565 static const char readme_msg[] =
5566 "tracing mini-HOWTO:\n\n"
5567 "# echo 0 > tracing_on : quick way to disable tracing\n"
5568 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5569 " Important files:\n"
5570 " trace\t\t\t- The static contents of the buffer\n"
5571 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5572 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5573 " current_tracer\t- function and latency tracers\n"
5574 " available_tracers\t- list of configured tracers for current_tracer\n"
5575 " error_log\t- error log for failed commands (that support it)\n"
5576 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5577 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5578 " trace_clock\t\t- change the clock used to order events\n"
5579 " local: Per cpu clock but may not be synced across CPUs\n"
5580 " global: Synced across CPUs but slows tracing down.\n"
5581 " counter: Not a clock, but just an increment\n"
5582 " uptime: Jiffy counter from time of boot\n"
5583 " perf: Same clock that perf events use\n"
5584 #ifdef CONFIG_X86_64
5585 " x86-tsc: TSC cycle counter\n"
5587 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5588 " delta: Delta difference against a buffer-wide timestamp\n"
5589 " absolute: Absolute (standalone) timestamp\n"
5590 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5591 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5592 " tracing_cpumask\t- Limit which CPUs to trace\n"
5593 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5594 "\t\t\t Remove sub-buffer with rmdir\n"
5595 " trace_options\t\t- Set format or modify how tracing happens\n"
5596 "\t\t\t Disable an option by prefixing 'no' to the\n"
5597 "\t\t\t option name\n"
5598 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5599 #ifdef CONFIG_DYNAMIC_FTRACE
5600 "\n available_filter_functions - list of functions that can be filtered on\n"
5601 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5602 "\t\t\t functions\n"
5603 "\t accepts: func_full_name or glob-matching-pattern\n"
5604 "\t modules: Can select a group via module\n"
5605 "\t Format: :mod:<module-name>\n"
5606 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5607 "\t triggers: a command to perform when function is hit\n"
5608 "\t Format: <function>:<trigger>[:count]\n"
5609 "\t trigger: traceon, traceoff\n"
5610 "\t\t enable_event:<system>:<event>\n"
5611 "\t\t disable_event:<system>:<event>\n"
5612 #ifdef CONFIG_STACKTRACE
5615 #ifdef CONFIG_TRACER_SNAPSHOT
5620 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5621 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5622 "\t The first one will disable tracing every time do_fault is hit\n"
5623 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5624 "\t The first time do trap is hit and it disables tracing, the\n"
5625 "\t counter will decrement to 2. If tracing is already disabled,\n"
5626 "\t the counter will not decrement. It only decrements when the\n"
5627 "\t trigger did work\n"
5628 "\t To remove trigger without count:\n"
5629 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5630 "\t To remove trigger with a count:\n"
5631 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5632 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5633 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5634 "\t modules: Can select a group via module command :mod:\n"
5635 "\t Does not accept triggers\n"
5636 #endif /* CONFIG_DYNAMIC_FTRACE */
5637 #ifdef CONFIG_FUNCTION_TRACER
5638 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5640 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5643 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5644 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5645 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5646 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5648 #ifdef CONFIG_TRACER_SNAPSHOT
5649 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5650 "\t\t\t snapshot buffer. Read the contents for more\n"
5651 "\t\t\t information\n"
5653 #ifdef CONFIG_STACK_TRACER
5654 " stack_trace\t\t- Shows the max stack trace when active\n"
5655 " stack_max_size\t- Shows current max stack size that was traced\n"
5656 "\t\t\t Write into this file to reset the max size (trigger a\n"
5657 "\t\t\t new trace)\n"
5658 #ifdef CONFIG_DYNAMIC_FTRACE
5659 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5662 #endif /* CONFIG_STACK_TRACER */
5663 #ifdef CONFIG_DYNAMIC_EVENTS
5664 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5665 "\t\t\t Write into this file to define/undefine new trace events.\n"
5667 #ifdef CONFIG_KPROBE_EVENTS
5668 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5669 "\t\t\t Write into this file to define/undefine new trace events.\n"
5671 #ifdef CONFIG_UPROBE_EVENTS
5672 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5673 "\t\t\t Write into this file to define/undefine new trace events.\n"
5675 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5676 "\t accepts: event-definitions (one definition per line)\n"
5677 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5678 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5679 #ifdef CONFIG_HIST_TRIGGERS
5680 "\t s:[synthetic/]<event> <field> [<field>]\n"
5682 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5683 "\t -:[<group>/][<event>]\n"
5684 #ifdef CONFIG_KPROBE_EVENTS
5685 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5686 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5688 #ifdef CONFIG_UPROBE_EVENTS
5689 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5691 "\t args: <name>=fetcharg[:type]\n"
5692 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5693 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5694 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5696 "\t $stack<index>, $stack, $retval, $comm,\n"
5698 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5699 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5700 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5701 "\t symstr, <type>\\[<array-size>\\]\n"
5702 #ifdef CONFIG_HIST_TRIGGERS
5703 "\t field: <stype> <name>;\n"
5704 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5705 "\t [unsigned] char/int/long\n"
5707 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5708 "\t of the <attached-group>/<attached-event>.\n"
5710 " events/\t\t- Directory containing all trace event subsystems:\n"
5711 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5712 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5713 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5715 " filter\t\t- If set, only events passing filter are traced\n"
5716 " events/<system>/<event>/\t- Directory containing control files for\n"
5718 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5719 " filter\t\t- If set, only events passing filter are traced\n"
5720 " trigger\t\t- If set, a command to perform when event is hit\n"
5721 "\t Format: <trigger>[:count][if <filter>]\n"
5722 "\t trigger: traceon, traceoff\n"
5723 "\t enable_event:<system>:<event>\n"
5724 "\t disable_event:<system>:<event>\n"
5725 #ifdef CONFIG_HIST_TRIGGERS
5726 "\t enable_hist:<system>:<event>\n"
5727 "\t disable_hist:<system>:<event>\n"
5729 #ifdef CONFIG_STACKTRACE
5732 #ifdef CONFIG_TRACER_SNAPSHOT
5735 #ifdef CONFIG_HIST_TRIGGERS
5736 "\t\t hist (see below)\n"
5738 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5739 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5740 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5741 "\t events/block/block_unplug/trigger\n"
5742 "\t The first disables tracing every time block_unplug is hit.\n"
5743 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5744 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5745 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5746 "\t Like function triggers, the counter is only decremented if it\n"
5747 "\t enabled or disabled tracing.\n"
5748 "\t To remove a trigger without a count:\n"
5749 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5750 "\t To remove a trigger with a count:\n"
5751 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5752 "\t Filters can be ignored when removing a trigger.\n"
5753 #ifdef CONFIG_HIST_TRIGGERS
5754 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5755 "\t Format: hist:keys=<field1[,field2,...]>\n"
5756 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5757 "\t [:values=<field1[,field2,...]>]\n"
5758 "\t [:sort=<field1[,field2,...]>]\n"
5759 "\t [:size=#entries]\n"
5760 "\t [:pause][:continue][:clear]\n"
5761 "\t [:name=histname1]\n"
5762 "\t [:nohitcount]\n"
5763 "\t [:<handler>.<action>]\n"
5764 "\t [if <filter>]\n\n"
5765 "\t Note, special fields can be used as well:\n"
5766 "\t common_timestamp - to record current timestamp\n"
5767 "\t common_cpu - to record the CPU the event happened on\n"
5769 "\t A hist trigger variable can be:\n"
5770 "\t - a reference to a field e.g. x=current_timestamp,\n"
5771 "\t - a reference to another variable e.g. y=$x,\n"
5772 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5773 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5775 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5776 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5777 "\t variable reference, field or numeric literal.\n"
5779 "\t When a matching event is hit, an entry is added to a hash\n"
5780 "\t table using the key(s) and value(s) named, and the value of a\n"
5781 "\t sum called 'hitcount' is incremented. Keys and values\n"
5782 "\t correspond to fields in the event's format description. Keys\n"
5783 "\t can be any field, or the special string 'common_stacktrace'.\n"
5784 "\t Compound keys consisting of up to two fields can be specified\n"
5785 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5786 "\t fields. Sort keys consisting of up to two fields can be\n"
5787 "\t specified using the 'sort' keyword. The sort direction can\n"
5788 "\t be modified by appending '.descending' or '.ascending' to a\n"
5789 "\t sort field. The 'size' parameter can be used to specify more\n"
5790 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5791 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5792 "\t its histogram data will be shared with other triggers of the\n"
5793 "\t same name, and trigger hits will update this common data.\n\n"
5794 "\t Reading the 'hist' file for the event will dump the hash\n"
5795 "\t table in its entirety to stdout. If there are multiple hist\n"
5796 "\t triggers attached to an event, there will be a table for each\n"
5797 "\t trigger in the output. The table displayed for a named\n"
5798 "\t trigger will be the same as any other instance having the\n"
5799 "\t same name. The default format used to display a given field\n"
5800 "\t can be modified by appending any of the following modifiers\n"
5801 "\t to the field name, as applicable:\n\n"
5802 "\t .hex display a number as a hex value\n"
5803 "\t .sym display an address as a symbol\n"
5804 "\t .sym-offset display an address as a symbol and offset\n"
5805 "\t .execname display a common_pid as a program name\n"
5806 "\t .syscall display a syscall id as a syscall name\n"
5807 "\t .log2 display log2 value rather than raw number\n"
5808 "\t .buckets=size display values in groups of size rather than raw number\n"
5809 "\t .usecs display a common_timestamp in microseconds\n"
5810 "\t .percent display a number of percentage value\n"
5811 "\t .graph display a bar-graph of a value\n\n"
5812 "\t The 'pause' parameter can be used to pause an existing hist\n"
5813 "\t trigger or to start a hist trigger but not log any events\n"
5814 "\t until told to do so. 'continue' can be used to start or\n"
5815 "\t restart a paused hist trigger.\n\n"
5816 "\t The 'clear' parameter will clear the contents of a running\n"
5817 "\t hist trigger and leave its current paused/active state\n"
5819 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5820 "\t raw hitcount in the histogram.\n\n"
5821 "\t The enable_hist and disable_hist triggers can be used to\n"
5822 "\t have one event conditionally start and stop another event's\n"
5823 "\t already-attached hist trigger. The syntax is analogous to\n"
5824 "\t the enable_event and disable_event triggers.\n\n"
5825 "\t Hist trigger handlers and actions are executed whenever a\n"
5826 "\t a histogram entry is added or updated. They take the form:\n\n"
5827 "\t <handler>.<action>\n\n"
5828 "\t The available handlers are:\n\n"
5829 "\t onmatch(matching.event) - invoke on addition or update\n"
5830 "\t onmax(var) - invoke if var exceeds current max\n"
5831 "\t onchange(var) - invoke action if var changes\n\n"
5832 "\t The available actions are:\n\n"
5833 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5834 "\t save(field,...) - save current event fields\n"
5835 #ifdef CONFIG_TRACER_SNAPSHOT
5836 "\t snapshot() - snapshot the trace buffer\n\n"
5838 #ifdef CONFIG_SYNTH_EVENTS
5839 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5840 "\t Write into this file to define/undefine new synthetic events.\n"
5841 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5847 tracing_readme_read(struct file *filp, char __user *ubuf,
5848 size_t cnt, loff_t *ppos)
5850 return simple_read_from_buffer(ubuf, cnt, ppos,
5851 readme_msg, strlen(readme_msg));
5854 static const struct file_operations tracing_readme_fops = {
5855 .open = tracing_open_generic,
5856 .read = tracing_readme_read,
5857 .llseek = generic_file_llseek,
5860 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5864 return trace_find_tgid_ptr(pid);
5867 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5871 return trace_find_tgid_ptr(pid);
5874 static void saved_tgids_stop(struct seq_file *m, void *v)
5878 static int saved_tgids_show(struct seq_file *m, void *v)
5880 int *entry = (int *)v;
5881 int pid = entry - tgid_map;
5887 seq_printf(m, "%d %d\n", pid, tgid);
5891 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5892 .start = saved_tgids_start,
5893 .stop = saved_tgids_stop,
5894 .next = saved_tgids_next,
5895 .show = saved_tgids_show,
5898 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5902 ret = tracing_check_open_get_tr(NULL);
5906 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5910 static const struct file_operations tracing_saved_tgids_fops = {
5911 .open = tracing_saved_tgids_open,
5913 .llseek = seq_lseek,
5914 .release = seq_release,
5917 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5919 unsigned int *ptr = v;
5921 if (*pos || m->count)
5926 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5928 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5937 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5943 arch_spin_lock(&trace_cmdline_lock);
5945 v = &savedcmd->map_cmdline_to_pid[0];
5947 v = saved_cmdlines_next(m, v, &l);
5955 static void saved_cmdlines_stop(struct seq_file *m, void *v)
5957 arch_spin_unlock(&trace_cmdline_lock);
5961 static int saved_cmdlines_show(struct seq_file *m, void *v)
5963 char buf[TASK_COMM_LEN];
5964 unsigned int *pid = v;
5966 __trace_find_cmdline(*pid, buf);
5967 seq_printf(m, "%d %s\n", *pid, buf);
5971 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5972 .start = saved_cmdlines_start,
5973 .next = saved_cmdlines_next,
5974 .stop = saved_cmdlines_stop,
5975 .show = saved_cmdlines_show,
5978 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5982 ret = tracing_check_open_get_tr(NULL);
5986 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5989 static const struct file_operations tracing_saved_cmdlines_fops = {
5990 .open = tracing_saved_cmdlines_open,
5992 .llseek = seq_lseek,
5993 .release = seq_release,
5997 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5998 size_t cnt, loff_t *ppos)
6004 arch_spin_lock(&trace_cmdline_lock);
6005 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
6006 arch_spin_unlock(&trace_cmdline_lock);
6009 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6012 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
6014 kfree(s->saved_cmdlines);
6015 kfree(s->map_cmdline_to_pid);
6019 static int tracing_resize_saved_cmdlines(unsigned int val)
6021 struct saved_cmdlines_buffer *s, *savedcmd_temp;
6023 s = kmalloc(sizeof(*s), GFP_KERNEL);
6027 if (allocate_cmdlines_buffer(val, s) < 0) {
6033 arch_spin_lock(&trace_cmdline_lock);
6034 savedcmd_temp = savedcmd;
6036 arch_spin_unlock(&trace_cmdline_lock);
6038 free_saved_cmdlines_buffer(savedcmd_temp);
6044 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
6045 size_t cnt, loff_t *ppos)
6050 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6054 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
6055 if (!val || val > PID_MAX_DEFAULT)
6058 ret = tracing_resize_saved_cmdlines((unsigned int)val);
6067 static const struct file_operations tracing_saved_cmdlines_size_fops = {
6068 .open = tracing_open_generic,
6069 .read = tracing_saved_cmdlines_size_read,
6070 .write = tracing_saved_cmdlines_size_write,
6073 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
6074 static union trace_eval_map_item *
6075 update_eval_map(union trace_eval_map_item *ptr)
6077 if (!ptr->map.eval_string) {
6078 if (ptr->tail.next) {
6079 ptr = ptr->tail.next;
6080 /* Set ptr to the next real item (skip head) */
6088 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6090 union trace_eval_map_item *ptr = v;
6093 * Paranoid! If ptr points to end, we don't want to increment past it.
6094 * This really should never happen.
6097 ptr = update_eval_map(ptr);
6098 if (WARN_ON_ONCE(!ptr))
6102 ptr = update_eval_map(ptr);
6107 static void *eval_map_start(struct seq_file *m, loff_t *pos)
6109 union trace_eval_map_item *v;
6112 mutex_lock(&trace_eval_mutex);
6114 v = trace_eval_maps;
6118 while (v && l < *pos) {
6119 v = eval_map_next(m, v, &l);
6125 static void eval_map_stop(struct seq_file *m, void *v)
6127 mutex_unlock(&trace_eval_mutex);
6130 static int eval_map_show(struct seq_file *m, void *v)
6132 union trace_eval_map_item *ptr = v;
6134 seq_printf(m, "%s %ld (%s)\n",
6135 ptr->map.eval_string, ptr->map.eval_value,
6141 static const struct seq_operations tracing_eval_map_seq_ops = {
6142 .start = eval_map_start,
6143 .next = eval_map_next,
6144 .stop = eval_map_stop,
6145 .show = eval_map_show,
6148 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6152 ret = tracing_check_open_get_tr(NULL);
6156 return seq_open(filp, &tracing_eval_map_seq_ops);
6159 static const struct file_operations tracing_eval_map_fops = {
6160 .open = tracing_eval_map_open,
6162 .llseek = seq_lseek,
6163 .release = seq_release,
6166 static inline union trace_eval_map_item *
6167 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6169 /* Return tail of array given the head */
6170 return ptr + ptr->head.length + 1;
6174 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6177 struct trace_eval_map **stop;
6178 struct trace_eval_map **map;
6179 union trace_eval_map_item *map_array;
6180 union trace_eval_map_item *ptr;
6185 * The trace_eval_maps contains the map plus a head and tail item,
6186 * where the head holds the module and length of array, and the
6187 * tail holds a pointer to the next list.
6189 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6191 pr_warn("Unable to allocate trace eval mapping\n");
6195 mutex_lock(&trace_eval_mutex);
6197 if (!trace_eval_maps)
6198 trace_eval_maps = map_array;
6200 ptr = trace_eval_maps;
6202 ptr = trace_eval_jmp_to_tail(ptr);
6203 if (!ptr->tail.next)
6205 ptr = ptr->tail.next;
6208 ptr->tail.next = map_array;
6210 map_array->head.mod = mod;
6211 map_array->head.length = len;
6214 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6215 map_array->map = **map;
6218 memset(map_array, 0, sizeof(*map_array));
6220 mutex_unlock(&trace_eval_mutex);
6223 static void trace_create_eval_file(struct dentry *d_tracer)
6225 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6226 NULL, &tracing_eval_map_fops);
6229 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
6230 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6231 static inline void trace_insert_eval_map_file(struct module *mod,
6232 struct trace_eval_map **start, int len) { }
6233 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6235 static void trace_insert_eval_map(struct module *mod,
6236 struct trace_eval_map **start, int len)
6238 struct trace_eval_map **map;
6245 trace_event_eval_update(map, len);
6247 trace_insert_eval_map_file(mod, start, len);
6251 tracing_set_trace_read(struct file *filp, char __user *ubuf,
6252 size_t cnt, loff_t *ppos)
6254 struct trace_array *tr = filp->private_data;
6255 char buf[MAX_TRACER_SIZE+2];
6258 mutex_lock(&trace_types_lock);
6259 r = sprintf(buf, "%s\n", tr->current_trace->name);
6260 mutex_unlock(&trace_types_lock);
6262 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6265 int tracer_init(struct tracer *t, struct trace_array *tr)
6267 tracing_reset_online_cpus(&tr->array_buffer);
6271 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6275 for_each_tracing_cpu(cpu)
6276 per_cpu_ptr(buf->data, cpu)->entries = val;
6279 #ifdef CONFIG_TRACER_MAX_TRACE
6280 /* resize @tr's buffer to the size of @size_tr's entries */
6281 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6282 struct array_buffer *size_buf, int cpu_id)
6286 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6287 for_each_tracing_cpu(cpu) {
6288 ret = ring_buffer_resize(trace_buf->buffer,
6289 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6292 per_cpu_ptr(trace_buf->data, cpu)->entries =
6293 per_cpu_ptr(size_buf->data, cpu)->entries;
6296 ret = ring_buffer_resize(trace_buf->buffer,
6297 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6299 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6300 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6305 #endif /* CONFIG_TRACER_MAX_TRACE */
6307 static int __tracing_resize_ring_buffer(struct trace_array *tr,
6308 unsigned long size, int cpu)
6313 * If kernel or user changes the size of the ring buffer
6314 * we use the size that was given, and we can forget about
6315 * expanding it later.
6317 ring_buffer_expanded = true;
6319 /* May be called before buffers are initialized */
6320 if (!tr->array_buffer.buffer)
6323 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6327 #ifdef CONFIG_TRACER_MAX_TRACE
6328 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
6329 !tr->current_trace->use_max_tr)
6332 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6334 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6335 &tr->array_buffer, cpu);
6338 * AARGH! We are left with different
6339 * size max buffer!!!!
6340 * The max buffer is our "snapshot" buffer.
6341 * When a tracer needs a snapshot (one of the
6342 * latency tracers), it swaps the max buffer
6343 * with the saved snap shot. We succeeded to
6344 * update the size of the main buffer, but failed to
6345 * update the size of the max buffer. But when we tried
6346 * to reset the main buffer to the original size, we
6347 * failed there too. This is very unlikely to
6348 * happen, but if it does, warn and kill all
6352 tracing_disabled = 1;
6357 if (cpu == RING_BUFFER_ALL_CPUS)
6358 set_buffer_entries(&tr->max_buffer, size);
6360 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
6363 #endif /* CONFIG_TRACER_MAX_TRACE */
6365 if (cpu == RING_BUFFER_ALL_CPUS)
6366 set_buffer_entries(&tr->array_buffer, size);
6368 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
6373 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6374 unsigned long size, int cpu_id)
6378 mutex_lock(&trace_types_lock);
6380 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6381 /* make sure, this cpu is enabled in the mask */
6382 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6388 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6393 mutex_unlock(&trace_types_lock);
6400 * tracing_update_buffers - used by tracing facility to expand ring buffers
6402 * To save on memory when the tracing is never used on a system with it
6403 * configured in. The ring buffers are set to a minimum size. But once
6404 * a user starts to use the tracing facility, then they need to grow
6405 * to their default size.
6407 * This function is to be called when a tracer is about to be used.
6409 int tracing_update_buffers(void)
6413 mutex_lock(&trace_types_lock);
6414 if (!ring_buffer_expanded)
6415 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
6416 RING_BUFFER_ALL_CPUS);
6417 mutex_unlock(&trace_types_lock);
6422 struct trace_option_dentry;
6425 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6428 * Used to clear out the tracer before deletion of an instance.
6429 * Must have trace_types_lock held.
6431 static void tracing_set_nop(struct trace_array *tr)
6433 if (tr->current_trace == &nop_trace)
6436 tr->current_trace->enabled--;
6438 if (tr->current_trace->reset)
6439 tr->current_trace->reset(tr);
6441 tr->current_trace = &nop_trace;
6444 static bool tracer_options_updated;
6446 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6448 /* Only enable if the directory has been created already. */
6452 /* Only create trace option files after update_tracer_options finish */
6453 if (!tracer_options_updated)
6456 create_trace_option_files(tr, t);
6459 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6462 #ifdef CONFIG_TRACER_MAX_TRACE
6467 mutex_lock(&trace_types_lock);
6469 if (!ring_buffer_expanded) {
6470 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6471 RING_BUFFER_ALL_CPUS);
6477 for (t = trace_types; t; t = t->next) {
6478 if (strcmp(t->name, buf) == 0)
6485 if (t == tr->current_trace)
6488 #ifdef CONFIG_TRACER_SNAPSHOT
6489 if (t->use_max_tr) {
6490 local_irq_disable();
6491 arch_spin_lock(&tr->max_lock);
6492 if (tr->cond_snapshot)
6494 arch_spin_unlock(&tr->max_lock);
6500 /* Some tracers won't work on kernel command line */
6501 if (system_state < SYSTEM_RUNNING && t->noboot) {
6502 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6507 /* Some tracers are only allowed for the top level buffer */
6508 if (!trace_ok_for_array(t, tr)) {
6513 /* If trace pipe files are being read, we can't change the tracer */
6514 if (tr->trace_ref) {
6519 trace_branch_disable();
6521 tr->current_trace->enabled--;
6523 if (tr->current_trace->reset)
6524 tr->current_trace->reset(tr);
6526 #ifdef CONFIG_TRACER_MAX_TRACE
6527 had_max_tr = tr->current_trace->use_max_tr;
6529 /* Current trace needs to be nop_trace before synchronize_rcu */
6530 tr->current_trace = &nop_trace;
6532 if (had_max_tr && !t->use_max_tr) {
6534 * We need to make sure that the update_max_tr sees that
6535 * current_trace changed to nop_trace to keep it from
6536 * swapping the buffers after we resize it.
6537 * The update_max_tr is called from interrupts disabled
6538 * so a synchronized_sched() is sufficient.
6544 if (t->use_max_tr && !tr->allocated_snapshot) {
6545 ret = tracing_alloc_snapshot_instance(tr);
6550 tr->current_trace = &nop_trace;
6554 ret = tracer_init(t, tr);
6559 tr->current_trace = t;
6560 tr->current_trace->enabled++;
6561 trace_branch_enable(tr);
6563 mutex_unlock(&trace_types_lock);
6569 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6570 size_t cnt, loff_t *ppos)
6572 struct trace_array *tr = filp->private_data;
6573 char buf[MAX_TRACER_SIZE+1];
6580 if (cnt > MAX_TRACER_SIZE)
6581 cnt = MAX_TRACER_SIZE;
6583 if (copy_from_user(buf, ubuf, cnt))
6590 err = tracing_set_tracer(tr, name);
6600 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6601 size_t cnt, loff_t *ppos)
6606 r = snprintf(buf, sizeof(buf), "%ld\n",
6607 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6608 if (r > sizeof(buf))
6610 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6614 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6615 size_t cnt, loff_t *ppos)
6620 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6630 tracing_thresh_read(struct file *filp, char __user *ubuf,
6631 size_t cnt, loff_t *ppos)
6633 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6637 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6638 size_t cnt, loff_t *ppos)
6640 struct trace_array *tr = filp->private_data;
6643 mutex_lock(&trace_types_lock);
6644 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6648 if (tr->current_trace->update_thresh) {
6649 ret = tr->current_trace->update_thresh(tr);
6656 mutex_unlock(&trace_types_lock);
6661 #ifdef CONFIG_TRACER_MAX_TRACE
6664 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6665 size_t cnt, loff_t *ppos)
6667 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6671 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6672 size_t cnt, loff_t *ppos)
6674 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6679 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6681 struct trace_array *tr = inode->i_private;
6682 struct trace_iterator *iter;
6685 ret = tracing_check_open_get_tr(tr);
6689 mutex_lock(&trace_types_lock);
6691 /* create a buffer to store the information to pass to userspace */
6692 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6695 __trace_array_put(tr);
6699 trace_seq_init(&iter->seq);
6700 iter->trace = tr->current_trace;
6702 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6707 /* trace pipe does not show start of buffer */
6708 cpumask_setall(iter->started);
6710 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6711 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6713 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6714 if (trace_clocks[tr->clock_id].in_ns)
6715 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6718 iter->array_buffer = &tr->array_buffer;
6719 iter->cpu_file = tracing_get_cpu(inode);
6720 mutex_init(&iter->mutex);
6721 filp->private_data = iter;
6723 if (iter->trace->pipe_open)
6724 iter->trace->pipe_open(iter);
6726 nonseekable_open(inode, filp);
6730 mutex_unlock(&trace_types_lock);
6735 __trace_array_put(tr);
6736 mutex_unlock(&trace_types_lock);
6740 static int tracing_release_pipe(struct inode *inode, struct file *file)
6742 struct trace_iterator *iter = file->private_data;
6743 struct trace_array *tr = inode->i_private;
6745 mutex_lock(&trace_types_lock);
6749 if (iter->trace->pipe_close)
6750 iter->trace->pipe_close(iter);
6752 mutex_unlock(&trace_types_lock);
6754 free_cpumask_var(iter->started);
6756 mutex_destroy(&iter->mutex);
6759 trace_array_put(tr);
6765 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6767 struct trace_array *tr = iter->tr;
6769 /* Iterators are static, they should be filled or empty */
6770 if (trace_buffer_iter(iter, iter->cpu_file))
6771 return EPOLLIN | EPOLLRDNORM;
6773 if (tr->trace_flags & TRACE_ITER_BLOCK)
6775 * Always select as readable when in blocking mode
6777 return EPOLLIN | EPOLLRDNORM;
6779 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6780 filp, poll_table, iter->tr->buffer_percent);
6784 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6786 struct trace_iterator *iter = filp->private_data;
6788 return trace_poll(iter, filp, poll_table);
6791 /* Must be called with iter->mutex held. */
6792 static int tracing_wait_pipe(struct file *filp)
6794 struct trace_iterator *iter = filp->private_data;
6797 while (trace_empty(iter)) {
6799 if ((filp->f_flags & O_NONBLOCK)) {
6804 * We block until we read something and tracing is disabled.
6805 * We still block if tracing is disabled, but we have never
6806 * read anything. This allows a user to cat this file, and
6807 * then enable tracing. But after we have read something,
6808 * we give an EOF when tracing is again disabled.
6810 * iter->pos will be 0 if we haven't read anything.
6812 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6815 mutex_unlock(&iter->mutex);
6817 ret = wait_on_pipe(iter, 0);
6819 mutex_lock(&iter->mutex);
6832 tracing_read_pipe(struct file *filp, char __user *ubuf,
6833 size_t cnt, loff_t *ppos)
6835 struct trace_iterator *iter = filp->private_data;
6839 * Avoid more than one consumer on a single file descriptor
6840 * This is just a matter of traces coherency, the ring buffer itself
6843 mutex_lock(&iter->mutex);
6845 /* return any leftover data */
6846 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6850 trace_seq_init(&iter->seq);
6852 if (iter->trace->read) {
6853 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6859 sret = tracing_wait_pipe(filp);
6863 /* stop when tracing is finished */
6864 if (trace_empty(iter)) {
6869 if (cnt >= PAGE_SIZE)
6870 cnt = PAGE_SIZE - 1;
6872 /* reset all but tr, trace, and overruns */
6873 trace_iterator_reset(iter);
6874 cpumask_clear(iter->started);
6875 trace_seq_init(&iter->seq);
6877 trace_event_read_lock();
6878 trace_access_lock(iter->cpu_file);
6879 while (trace_find_next_entry_inc(iter) != NULL) {
6880 enum print_line_t ret;
6881 int save_len = iter->seq.seq.len;
6883 ret = print_trace_line(iter);
6884 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6886 * If one print_trace_line() fills entire trace_seq in one shot,
6887 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6888 * In this case, we need to consume it, otherwise, loop will peek
6889 * this event next time, resulting in an infinite loop.
6891 if (save_len == 0) {
6893 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6894 trace_consume(iter);
6898 /* In other cases, don't print partial lines */
6899 iter->seq.seq.len = save_len;
6902 if (ret != TRACE_TYPE_NO_CONSUME)
6903 trace_consume(iter);
6905 if (trace_seq_used(&iter->seq) >= cnt)
6909 * Setting the full flag means we reached the trace_seq buffer
6910 * size and we should leave by partial output condition above.
6911 * One of the trace_seq_* functions is not used properly.
6913 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6916 trace_access_unlock(iter->cpu_file);
6917 trace_event_read_unlock();
6919 /* Now copy what we have to the user */
6920 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6921 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6922 trace_seq_init(&iter->seq);
6925 * If there was nothing to send to user, in spite of consuming trace
6926 * entries, go back to wait for more entries.
6932 mutex_unlock(&iter->mutex);
6937 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6940 __free_page(spd->pages[idx]);
6944 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6950 /* Seq buffer is page-sized, exactly what we need. */
6952 save_len = iter->seq.seq.len;
6953 ret = print_trace_line(iter);
6955 if (trace_seq_has_overflowed(&iter->seq)) {
6956 iter->seq.seq.len = save_len;
6961 * This should not be hit, because it should only
6962 * be set if the iter->seq overflowed. But check it
6963 * anyway to be safe.
6965 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6966 iter->seq.seq.len = save_len;
6970 count = trace_seq_used(&iter->seq) - save_len;
6973 iter->seq.seq.len = save_len;
6977 if (ret != TRACE_TYPE_NO_CONSUME)
6978 trace_consume(iter);
6980 if (!trace_find_next_entry_inc(iter)) {
6990 static ssize_t tracing_splice_read_pipe(struct file *filp,
6992 struct pipe_inode_info *pipe,
6996 struct page *pages_def[PIPE_DEF_BUFFERS];
6997 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6998 struct trace_iterator *iter = filp->private_data;
6999 struct splice_pipe_desc spd = {
7001 .partial = partial_def,
7002 .nr_pages = 0, /* This gets updated below. */
7003 .nr_pages_max = PIPE_DEF_BUFFERS,
7004 .ops = &default_pipe_buf_ops,
7005 .spd_release = tracing_spd_release_pipe,
7011 if (splice_grow_spd(pipe, &spd))
7014 mutex_lock(&iter->mutex);
7016 if (iter->trace->splice_read) {
7017 ret = iter->trace->splice_read(iter, filp,
7018 ppos, pipe, len, flags);
7023 ret = tracing_wait_pipe(filp);
7027 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
7032 trace_event_read_lock();
7033 trace_access_lock(iter->cpu_file);
7035 /* Fill as many pages as possible. */
7036 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
7037 spd.pages[i] = alloc_page(GFP_KERNEL);
7041 rem = tracing_fill_pipe_page(rem, iter);
7043 /* Copy the data into the page, so we can start over. */
7044 ret = trace_seq_to_buffer(&iter->seq,
7045 page_address(spd.pages[i]),
7046 trace_seq_used(&iter->seq));
7048 __free_page(spd.pages[i]);
7051 spd.partial[i].offset = 0;
7052 spd.partial[i].len = trace_seq_used(&iter->seq);
7054 trace_seq_init(&iter->seq);
7057 trace_access_unlock(iter->cpu_file);
7058 trace_event_read_unlock();
7059 mutex_unlock(&iter->mutex);
7064 ret = splice_to_pipe(pipe, &spd);
7068 splice_shrink_spd(&spd);
7072 mutex_unlock(&iter->mutex);
7077 tracing_entries_read(struct file *filp, char __user *ubuf,
7078 size_t cnt, loff_t *ppos)
7080 struct inode *inode = file_inode(filp);
7081 struct trace_array *tr = inode->i_private;
7082 int cpu = tracing_get_cpu(inode);
7087 mutex_lock(&trace_types_lock);
7089 if (cpu == RING_BUFFER_ALL_CPUS) {
7090 int cpu, buf_size_same;
7095 /* check if all cpu sizes are same */
7096 for_each_tracing_cpu(cpu) {
7097 /* fill in the size from first enabled cpu */
7099 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7100 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7106 if (buf_size_same) {
7107 if (!ring_buffer_expanded)
7108 r = sprintf(buf, "%lu (expanded: %lu)\n",
7110 trace_buf_size >> 10);
7112 r = sprintf(buf, "%lu\n", size >> 10);
7114 r = sprintf(buf, "X\n");
7116 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7118 mutex_unlock(&trace_types_lock);
7120 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7125 tracing_entries_write(struct file *filp, const char __user *ubuf,
7126 size_t cnt, loff_t *ppos)
7128 struct inode *inode = file_inode(filp);
7129 struct trace_array *tr = inode->i_private;
7133 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7137 /* must have at least 1 entry */
7141 /* value is in KB */
7143 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7153 tracing_total_entries_read(struct file *filp, char __user *ubuf,
7154 size_t cnt, loff_t *ppos)
7156 struct trace_array *tr = filp->private_data;
7159 unsigned long size = 0, expanded_size = 0;
7161 mutex_lock(&trace_types_lock);
7162 for_each_tracing_cpu(cpu) {
7163 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7164 if (!ring_buffer_expanded)
7165 expanded_size += trace_buf_size >> 10;
7167 if (ring_buffer_expanded)
7168 r = sprintf(buf, "%lu\n", size);
7170 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7171 mutex_unlock(&trace_types_lock);
7173 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7177 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7178 size_t cnt, loff_t *ppos)
7181 * There is no need to read what the user has written, this function
7182 * is just to make sure that there is no error when "echo" is used
7191 tracing_free_buffer_release(struct inode *inode, struct file *filp)
7193 struct trace_array *tr = inode->i_private;
7195 /* disable tracing ? */
7196 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7197 tracer_tracing_off(tr);
7198 /* resize the ring buffer to 0 */
7199 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7201 trace_array_put(tr);
7207 tracing_mark_write(struct file *filp, const char __user *ubuf,
7208 size_t cnt, loff_t *fpos)
7210 struct trace_array *tr = filp->private_data;
7211 struct ring_buffer_event *event;
7212 enum event_trigger_type tt = ETT_NONE;
7213 struct trace_buffer *buffer;
7214 struct print_entry *entry;
7219 /* Used in tracing_mark_raw_write() as well */
7220 #define FAULTED_STR "<faulted>"
7221 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7223 if (tracing_disabled)
7226 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7229 if (cnt > TRACE_BUF_SIZE)
7230 cnt = TRACE_BUF_SIZE;
7232 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7234 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
7236 /* If less than "<faulted>", then make sure we can still add that */
7237 if (cnt < FAULTED_SIZE)
7238 size += FAULTED_SIZE - cnt;
7240 buffer = tr->array_buffer.buffer;
7241 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7243 if (unlikely(!event))
7244 /* Ring buffer disabled, return as if not open for write */
7247 entry = ring_buffer_event_data(event);
7248 entry->ip = _THIS_IP_;
7250 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7252 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7258 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7259 /* do not add \n before testing triggers, but add \0 */
7260 entry->buf[cnt] = '\0';
7261 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7264 if (entry->buf[cnt - 1] != '\n') {
7265 entry->buf[cnt] = '\n';
7266 entry->buf[cnt + 1] = '\0';
7268 entry->buf[cnt] = '\0';
7270 if (static_branch_unlikely(&trace_marker_exports_enabled))
7271 ftrace_exports(event, TRACE_EXPORT_MARKER);
7272 __buffer_unlock_commit(buffer, event);
7275 event_triggers_post_call(tr->trace_marker_file, tt);
7280 /* Limit it for now to 3K (including tag) */
7281 #define RAW_DATA_MAX_SIZE (1024*3)
7284 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7285 size_t cnt, loff_t *fpos)
7287 struct trace_array *tr = filp->private_data;
7288 struct ring_buffer_event *event;
7289 struct trace_buffer *buffer;
7290 struct raw_data_entry *entry;
7295 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7297 if (tracing_disabled)
7300 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7303 /* The marker must at least have a tag id */
7304 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
7307 if (cnt > TRACE_BUF_SIZE)
7308 cnt = TRACE_BUF_SIZE;
7310 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7312 size = sizeof(*entry) + cnt;
7313 if (cnt < FAULT_SIZE_ID)
7314 size += FAULT_SIZE_ID - cnt;
7316 buffer = tr->array_buffer.buffer;
7317 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7320 /* Ring buffer disabled, return as if not open for write */
7323 entry = ring_buffer_event_data(event);
7325 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7328 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7333 __buffer_unlock_commit(buffer, event);
7338 static int tracing_clock_show(struct seq_file *m, void *v)
7340 struct trace_array *tr = m->private;
7343 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7345 "%s%s%s%s", i ? " " : "",
7346 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7347 i == tr->clock_id ? "]" : "");
7353 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7357 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7358 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7361 if (i == ARRAY_SIZE(trace_clocks))
7364 mutex_lock(&trace_types_lock);
7368 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7371 * New clock may not be consistent with the previous clock.
7372 * Reset the buffer so that it doesn't have incomparable timestamps.
7374 tracing_reset_online_cpus(&tr->array_buffer);
7376 #ifdef CONFIG_TRACER_MAX_TRACE
7377 if (tr->max_buffer.buffer)
7378 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7379 tracing_reset_online_cpus(&tr->max_buffer);
7382 mutex_unlock(&trace_types_lock);
7387 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7388 size_t cnt, loff_t *fpos)
7390 struct seq_file *m = filp->private_data;
7391 struct trace_array *tr = m->private;
7393 const char *clockstr;
7396 if (cnt >= sizeof(buf))
7399 if (copy_from_user(buf, ubuf, cnt))
7404 clockstr = strstrip(buf);
7406 ret = tracing_set_clock(tr, clockstr);
7415 static int tracing_clock_open(struct inode *inode, struct file *file)
7417 struct trace_array *tr = inode->i_private;
7420 ret = tracing_check_open_get_tr(tr);
7424 ret = single_open(file, tracing_clock_show, inode->i_private);
7426 trace_array_put(tr);
7431 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7433 struct trace_array *tr = m->private;
7435 mutex_lock(&trace_types_lock);
7437 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7438 seq_puts(m, "delta [absolute]\n");
7440 seq_puts(m, "[delta] absolute\n");
7442 mutex_unlock(&trace_types_lock);
7447 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7449 struct trace_array *tr = inode->i_private;
7452 ret = tracing_check_open_get_tr(tr);
7456 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7458 trace_array_put(tr);
7463 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7465 if (rbe == this_cpu_read(trace_buffered_event))
7466 return ring_buffer_time_stamp(buffer);
7468 return ring_buffer_event_time_stamp(buffer, rbe);
7472 * Set or disable using the per CPU trace_buffer_event when possible.
7474 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7478 mutex_lock(&trace_types_lock);
7480 if (set && tr->no_filter_buffering_ref++)
7484 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7489 --tr->no_filter_buffering_ref;
7492 mutex_unlock(&trace_types_lock);
7497 struct ftrace_buffer_info {
7498 struct trace_iterator iter;
7500 unsigned int spare_cpu;
7504 #ifdef CONFIG_TRACER_SNAPSHOT
7505 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7507 struct trace_array *tr = inode->i_private;
7508 struct trace_iterator *iter;
7512 ret = tracing_check_open_get_tr(tr);
7516 if (file->f_mode & FMODE_READ) {
7517 iter = __tracing_open(inode, file, true);
7519 ret = PTR_ERR(iter);
7521 /* Writes still need the seq_file to hold the private data */
7523 m = kzalloc(sizeof(*m), GFP_KERNEL);
7526 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7534 iter->array_buffer = &tr->max_buffer;
7535 iter->cpu_file = tracing_get_cpu(inode);
7537 file->private_data = m;
7541 trace_array_put(tr);
7547 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7550 struct seq_file *m = filp->private_data;
7551 struct trace_iterator *iter = m->private;
7552 struct trace_array *tr = iter->tr;
7556 ret = tracing_update_buffers();
7560 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7564 mutex_lock(&trace_types_lock);
7566 if (tr->current_trace->use_max_tr) {
7571 local_irq_disable();
7572 arch_spin_lock(&tr->max_lock);
7573 if (tr->cond_snapshot)
7575 arch_spin_unlock(&tr->max_lock);
7582 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7586 if (tr->allocated_snapshot)
7590 /* Only allow per-cpu swap if the ring buffer supports it */
7591 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7592 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7597 if (tr->allocated_snapshot)
7598 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7599 &tr->array_buffer, iter->cpu_file);
7601 ret = tracing_alloc_snapshot_instance(tr);
7604 local_irq_disable();
7605 /* Now, we're going to swap */
7606 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7607 update_max_tr(tr, current, smp_processor_id(), NULL);
7609 update_max_tr_single(tr, current, iter->cpu_file);
7613 if (tr->allocated_snapshot) {
7614 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7615 tracing_reset_online_cpus(&tr->max_buffer);
7617 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7627 mutex_unlock(&trace_types_lock);
7631 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7633 struct seq_file *m = file->private_data;
7636 ret = tracing_release(inode, file);
7638 if (file->f_mode & FMODE_READ)
7641 /* If write only, the seq_file is just a stub */
7649 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7650 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7651 size_t count, loff_t *ppos);
7652 static int tracing_buffers_release(struct inode *inode, struct file *file);
7653 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7654 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7656 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7658 struct ftrace_buffer_info *info;
7661 /* The following checks for tracefs lockdown */
7662 ret = tracing_buffers_open(inode, filp);
7666 info = filp->private_data;
7668 if (info->iter.trace->use_max_tr) {
7669 tracing_buffers_release(inode, filp);
7673 info->iter.snapshot = true;
7674 info->iter.array_buffer = &info->iter.tr->max_buffer;
7679 #endif /* CONFIG_TRACER_SNAPSHOT */
7682 static const struct file_operations tracing_thresh_fops = {
7683 .open = tracing_open_generic,
7684 .read = tracing_thresh_read,
7685 .write = tracing_thresh_write,
7686 .llseek = generic_file_llseek,
7689 #ifdef CONFIG_TRACER_MAX_TRACE
7690 static const struct file_operations tracing_max_lat_fops = {
7691 .open = tracing_open_generic,
7692 .read = tracing_max_lat_read,
7693 .write = tracing_max_lat_write,
7694 .llseek = generic_file_llseek,
7698 static const struct file_operations set_tracer_fops = {
7699 .open = tracing_open_generic,
7700 .read = tracing_set_trace_read,
7701 .write = tracing_set_trace_write,
7702 .llseek = generic_file_llseek,
7705 static const struct file_operations tracing_pipe_fops = {
7706 .open = tracing_open_pipe,
7707 .poll = tracing_poll_pipe,
7708 .read = tracing_read_pipe,
7709 .splice_read = tracing_splice_read_pipe,
7710 .release = tracing_release_pipe,
7711 .llseek = no_llseek,
7714 static const struct file_operations tracing_entries_fops = {
7715 .open = tracing_open_generic_tr,
7716 .read = tracing_entries_read,
7717 .write = tracing_entries_write,
7718 .llseek = generic_file_llseek,
7719 .release = tracing_release_generic_tr,
7722 static const struct file_operations tracing_total_entries_fops = {
7723 .open = tracing_open_generic_tr,
7724 .read = tracing_total_entries_read,
7725 .llseek = generic_file_llseek,
7726 .release = tracing_release_generic_tr,
7729 static const struct file_operations tracing_free_buffer_fops = {
7730 .open = tracing_open_generic_tr,
7731 .write = tracing_free_buffer_write,
7732 .release = tracing_free_buffer_release,
7735 static const struct file_operations tracing_mark_fops = {
7736 .open = tracing_mark_open,
7737 .write = tracing_mark_write,
7738 .release = tracing_release_generic_tr,
7741 static const struct file_operations tracing_mark_raw_fops = {
7742 .open = tracing_mark_open,
7743 .write = tracing_mark_raw_write,
7744 .release = tracing_release_generic_tr,
7747 static const struct file_operations trace_clock_fops = {
7748 .open = tracing_clock_open,
7750 .llseek = seq_lseek,
7751 .release = tracing_single_release_tr,
7752 .write = tracing_clock_write,
7755 static const struct file_operations trace_time_stamp_mode_fops = {
7756 .open = tracing_time_stamp_mode_open,
7758 .llseek = seq_lseek,
7759 .release = tracing_single_release_tr,
7762 #ifdef CONFIG_TRACER_SNAPSHOT
7763 static const struct file_operations snapshot_fops = {
7764 .open = tracing_snapshot_open,
7766 .write = tracing_snapshot_write,
7767 .llseek = tracing_lseek,
7768 .release = tracing_snapshot_release,
7771 static const struct file_operations snapshot_raw_fops = {
7772 .open = snapshot_raw_open,
7773 .read = tracing_buffers_read,
7774 .release = tracing_buffers_release,
7775 .splice_read = tracing_buffers_splice_read,
7776 .llseek = no_llseek,
7779 #endif /* CONFIG_TRACER_SNAPSHOT */
7782 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7783 * @filp: The active open file structure
7784 * @ubuf: The userspace provided buffer to read value into
7785 * @cnt: The maximum number of bytes to read
7786 * @ppos: The current "file" position
7788 * This function implements the write interface for a struct trace_min_max_param.
7789 * The filp->private_data must point to a trace_min_max_param structure that
7790 * defines where to write the value, the min and the max acceptable values,
7791 * and a lock to protect the write.
7794 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7796 struct trace_min_max_param *param = filp->private_data;
7803 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7808 mutex_lock(param->lock);
7810 if (param->min && val < *param->min)
7813 if (param->max && val > *param->max)
7820 mutex_unlock(param->lock);
7829 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7830 * @filp: The active open file structure
7831 * @ubuf: The userspace provided buffer to read value into
7832 * @cnt: The maximum number of bytes to read
7833 * @ppos: The current "file" position
7835 * This function implements the read interface for a struct trace_min_max_param.
7836 * The filp->private_data must point to a trace_min_max_param struct with valid
7840 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7842 struct trace_min_max_param *param = filp->private_data;
7843 char buf[U64_STR_SIZE];
7852 if (cnt > sizeof(buf))
7855 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7857 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7860 const struct file_operations trace_min_max_fops = {
7861 .open = tracing_open_generic,
7862 .read = trace_min_max_read,
7863 .write = trace_min_max_write,
7866 #define TRACING_LOG_ERRS_MAX 8
7867 #define TRACING_LOG_LOC_MAX 128
7869 #define CMD_PREFIX " Command: "
7872 const char **errs; /* ptr to loc-specific array of err strings */
7873 u8 type; /* index into errs -> specific err string */
7874 u16 pos; /* caret position */
7878 struct tracing_log_err {
7879 struct list_head list;
7880 struct err_info info;
7881 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7882 char *cmd; /* what caused err */
7885 static DEFINE_MUTEX(tracing_err_log_lock);
7887 static struct tracing_log_err *alloc_tracing_log_err(int len)
7889 struct tracing_log_err *err;
7891 err = kzalloc(sizeof(*err), GFP_KERNEL);
7893 return ERR_PTR(-ENOMEM);
7895 err->cmd = kzalloc(len, GFP_KERNEL);
7898 return ERR_PTR(-ENOMEM);
7904 static void free_tracing_log_err(struct tracing_log_err *err)
7910 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7913 struct tracing_log_err *err;
7916 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7917 err = alloc_tracing_log_err(len);
7918 if (PTR_ERR(err) != -ENOMEM)
7919 tr->n_err_log_entries++;
7923 cmd = kzalloc(len, GFP_KERNEL);
7925 return ERR_PTR(-ENOMEM);
7926 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7929 list_del(&err->list);
7935 * err_pos - find the position of a string within a command for error careting
7936 * @cmd: The tracing command that caused the error
7937 * @str: The string to position the caret at within @cmd
7939 * Finds the position of the first occurrence of @str within @cmd. The
7940 * return value can be passed to tracing_log_err() for caret placement
7943 * Returns the index within @cmd of the first occurrence of @str or 0
7944 * if @str was not found.
7946 unsigned int err_pos(char *cmd, const char *str)
7950 if (WARN_ON(!strlen(cmd)))
7953 found = strstr(cmd, str);
7961 * tracing_log_err - write an error to the tracing error log
7962 * @tr: The associated trace array for the error (NULL for top level array)
7963 * @loc: A string describing where the error occurred
7964 * @cmd: The tracing command that caused the error
7965 * @errs: The array of loc-specific static error strings
7966 * @type: The index into errs[], which produces the specific static err string
7967 * @pos: The position the caret should be placed in the cmd
7969 * Writes an error into tracing/error_log of the form:
7971 * <loc>: error: <text>
7975 * tracing/error_log is a small log file containing the last
7976 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7977 * unless there has been a tracing error, and the error log can be
7978 * cleared and have its memory freed by writing the empty string in
7979 * truncation mode to it i.e. echo > tracing/error_log.
7981 * NOTE: the @errs array along with the @type param are used to
7982 * produce a static error string - this string is not copied and saved
7983 * when the error is logged - only a pointer to it is saved. See
7984 * existing callers for examples of how static strings are typically
7985 * defined for use with tracing_log_err().
7987 void tracing_log_err(struct trace_array *tr,
7988 const char *loc, const char *cmd,
7989 const char **errs, u8 type, u16 pos)
7991 struct tracing_log_err *err;
7997 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
7999 mutex_lock(&tracing_err_log_lock);
8000 err = get_tracing_log_err(tr, len);
8001 if (PTR_ERR(err) == -ENOMEM) {
8002 mutex_unlock(&tracing_err_log_lock);
8006 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
8007 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
8009 err->info.errs = errs;
8010 err->info.type = type;
8011 err->info.pos = pos;
8012 err->info.ts = local_clock();
8014 list_add_tail(&err->list, &tr->err_log);
8015 mutex_unlock(&tracing_err_log_lock);
8018 static void clear_tracing_err_log(struct trace_array *tr)
8020 struct tracing_log_err *err, *next;
8022 mutex_lock(&tracing_err_log_lock);
8023 list_for_each_entry_safe(err, next, &tr->err_log, list) {
8024 list_del(&err->list);
8025 free_tracing_log_err(err);
8028 tr->n_err_log_entries = 0;
8029 mutex_unlock(&tracing_err_log_lock);
8032 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
8034 struct trace_array *tr = m->private;
8036 mutex_lock(&tracing_err_log_lock);
8038 return seq_list_start(&tr->err_log, *pos);
8041 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
8043 struct trace_array *tr = m->private;
8045 return seq_list_next(v, &tr->err_log, pos);
8048 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
8050 mutex_unlock(&tracing_err_log_lock);
8053 static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
8057 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
8059 for (i = 0; i < pos; i++)
8064 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
8066 struct tracing_log_err *err = v;
8069 const char *err_text = err->info.errs[err->info.type];
8070 u64 sec = err->info.ts;
8073 nsec = do_div(sec, NSEC_PER_SEC);
8074 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
8075 err->loc, err_text);
8076 seq_printf(m, "%s", err->cmd);
8077 tracing_err_log_show_pos(m, err->info.pos);
8083 static const struct seq_operations tracing_err_log_seq_ops = {
8084 .start = tracing_err_log_seq_start,
8085 .next = tracing_err_log_seq_next,
8086 .stop = tracing_err_log_seq_stop,
8087 .show = tracing_err_log_seq_show
8090 static int tracing_err_log_open(struct inode *inode, struct file *file)
8092 struct trace_array *tr = inode->i_private;
8095 ret = tracing_check_open_get_tr(tr);
8099 /* If this file was opened for write, then erase contents */
8100 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8101 clear_tracing_err_log(tr);
8103 if (file->f_mode & FMODE_READ) {
8104 ret = seq_open(file, &tracing_err_log_seq_ops);
8106 struct seq_file *m = file->private_data;
8109 trace_array_put(tr);
8115 static ssize_t tracing_err_log_write(struct file *file,
8116 const char __user *buffer,
8117 size_t count, loff_t *ppos)
8122 static int tracing_err_log_release(struct inode *inode, struct file *file)
8124 struct trace_array *tr = inode->i_private;
8126 trace_array_put(tr);
8128 if (file->f_mode & FMODE_READ)
8129 seq_release(inode, file);
8134 static const struct file_operations tracing_err_log_fops = {
8135 .open = tracing_err_log_open,
8136 .write = tracing_err_log_write,
8138 .llseek = seq_lseek,
8139 .release = tracing_err_log_release,
8142 static int tracing_buffers_open(struct inode *inode, struct file *filp)
8144 struct trace_array *tr = inode->i_private;
8145 struct ftrace_buffer_info *info;
8148 ret = tracing_check_open_get_tr(tr);
8152 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8154 trace_array_put(tr);
8158 mutex_lock(&trace_types_lock);
8161 info->iter.cpu_file = tracing_get_cpu(inode);
8162 info->iter.trace = tr->current_trace;
8163 info->iter.array_buffer = &tr->array_buffer;
8165 /* Force reading ring buffer for first read */
8166 info->read = (unsigned int)-1;
8168 filp->private_data = info;
8172 mutex_unlock(&trace_types_lock);
8174 ret = nonseekable_open(inode, filp);
8176 trace_array_put(tr);
8182 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8184 struct ftrace_buffer_info *info = filp->private_data;
8185 struct trace_iterator *iter = &info->iter;
8187 return trace_poll(iter, filp, poll_table);
8191 tracing_buffers_read(struct file *filp, char __user *ubuf,
8192 size_t count, loff_t *ppos)
8194 struct ftrace_buffer_info *info = filp->private_data;
8195 struct trace_iterator *iter = &info->iter;
8202 #ifdef CONFIG_TRACER_MAX_TRACE
8203 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8208 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8210 if (IS_ERR(info->spare)) {
8211 ret = PTR_ERR(info->spare);
8214 info->spare_cpu = iter->cpu_file;
8220 /* Do we have previous read data to read? */
8221 if (info->read < PAGE_SIZE)
8225 trace_access_lock(iter->cpu_file);
8226 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8230 trace_access_unlock(iter->cpu_file);
8233 if (trace_empty(iter)) {
8234 if ((filp->f_flags & O_NONBLOCK))
8237 ret = wait_on_pipe(iter, 0);
8248 size = PAGE_SIZE - info->read;
8252 ret = copy_to_user(ubuf, info->spare + info->read, size);
8264 static int tracing_buffers_release(struct inode *inode, struct file *file)
8266 struct ftrace_buffer_info *info = file->private_data;
8267 struct trace_iterator *iter = &info->iter;
8269 mutex_lock(&trace_types_lock);
8271 iter->tr->trace_ref--;
8273 __trace_array_put(iter->tr);
8276 /* Make sure the waiters see the new wait_index */
8279 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8282 ring_buffer_free_read_page(iter->array_buffer->buffer,
8283 info->spare_cpu, info->spare);
8286 mutex_unlock(&trace_types_lock);
8292 struct trace_buffer *buffer;
8295 refcount_t refcount;
8298 static void buffer_ref_release(struct buffer_ref *ref)
8300 if (!refcount_dec_and_test(&ref->refcount))
8302 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8306 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8307 struct pipe_buffer *buf)
8309 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8311 buffer_ref_release(ref);
8315 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8316 struct pipe_buffer *buf)
8318 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8320 if (refcount_read(&ref->refcount) > INT_MAX/2)
8323 refcount_inc(&ref->refcount);
8327 /* Pipe buffer operations for a buffer. */
8328 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8329 .release = buffer_pipe_buf_release,
8330 .get = buffer_pipe_buf_get,
8334 * Callback from splice_to_pipe(), if we need to release some pages
8335 * at the end of the spd in case we error'ed out in filling the pipe.
8337 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8339 struct buffer_ref *ref =
8340 (struct buffer_ref *)spd->partial[i].private;
8342 buffer_ref_release(ref);
8343 spd->partial[i].private = 0;
8347 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8348 struct pipe_inode_info *pipe, size_t len,
8351 struct ftrace_buffer_info *info = file->private_data;
8352 struct trace_iterator *iter = &info->iter;
8353 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8354 struct page *pages_def[PIPE_DEF_BUFFERS];
8355 struct splice_pipe_desc spd = {
8357 .partial = partial_def,
8358 .nr_pages_max = PIPE_DEF_BUFFERS,
8359 .ops = &buffer_pipe_buf_ops,
8360 .spd_release = buffer_spd_release,
8362 struct buffer_ref *ref;
8366 #ifdef CONFIG_TRACER_MAX_TRACE
8367 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8371 if (*ppos & (PAGE_SIZE - 1))
8374 if (len & (PAGE_SIZE - 1)) {
8375 if (len < PAGE_SIZE)
8380 if (splice_grow_spd(pipe, &spd))
8384 trace_access_lock(iter->cpu_file);
8385 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8387 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
8391 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8397 refcount_set(&ref->refcount, 1);
8398 ref->buffer = iter->array_buffer->buffer;
8399 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8400 if (IS_ERR(ref->page)) {
8401 ret = PTR_ERR(ref->page);
8406 ref->cpu = iter->cpu_file;
8408 r = ring_buffer_read_page(ref->buffer, &ref->page,
8409 len, iter->cpu_file, 1);
8411 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8417 page = virt_to_page(ref->page);
8419 spd.pages[i] = page;
8420 spd.partial[i].len = PAGE_SIZE;
8421 spd.partial[i].offset = 0;
8422 spd.partial[i].private = (unsigned long)ref;
8426 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8429 trace_access_unlock(iter->cpu_file);
8432 /* did we read anything? */
8433 if (!spd.nr_pages) {
8440 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8443 wait_index = READ_ONCE(iter->wait_index);
8445 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
8449 /* No need to wait after waking up when tracing is off */
8450 if (!tracer_tracing_is_on(iter->tr))
8453 /* Make sure we see the new wait_index */
8455 if (wait_index != iter->wait_index)
8461 ret = splice_to_pipe(pipe, &spd);
8463 splice_shrink_spd(&spd);
8468 /* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8469 static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8471 struct ftrace_buffer_info *info = file->private_data;
8472 struct trace_iterator *iter = &info->iter;
8475 return -ENOIOCTLCMD;
8477 mutex_lock(&trace_types_lock);
8480 /* Make sure the waiters see the new wait_index */
8483 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8485 mutex_unlock(&trace_types_lock);
8489 static const struct file_operations tracing_buffers_fops = {
8490 .open = tracing_buffers_open,
8491 .read = tracing_buffers_read,
8492 .poll = tracing_buffers_poll,
8493 .release = tracing_buffers_release,
8494 .splice_read = tracing_buffers_splice_read,
8495 .unlocked_ioctl = tracing_buffers_ioctl,
8496 .llseek = no_llseek,
8500 tracing_stats_read(struct file *filp, char __user *ubuf,
8501 size_t count, loff_t *ppos)
8503 struct inode *inode = file_inode(filp);
8504 struct trace_array *tr = inode->i_private;
8505 struct array_buffer *trace_buf = &tr->array_buffer;
8506 int cpu = tracing_get_cpu(inode);
8507 struct trace_seq *s;
8509 unsigned long long t;
8510 unsigned long usec_rem;
8512 s = kmalloc(sizeof(*s), GFP_KERNEL);
8518 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8519 trace_seq_printf(s, "entries: %ld\n", cnt);
8521 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8522 trace_seq_printf(s, "overrun: %ld\n", cnt);
8524 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8525 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8527 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8528 trace_seq_printf(s, "bytes: %ld\n", cnt);
8530 if (trace_clocks[tr->clock_id].in_ns) {
8531 /* local or global for trace_clock */
8532 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8533 usec_rem = do_div(t, USEC_PER_SEC);
8534 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8537 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8538 usec_rem = do_div(t, USEC_PER_SEC);
8539 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8541 /* counter or tsc mode for trace_clock */
8542 trace_seq_printf(s, "oldest event ts: %llu\n",
8543 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8545 trace_seq_printf(s, "now ts: %llu\n",
8546 ring_buffer_time_stamp(trace_buf->buffer));
8549 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8550 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8552 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8553 trace_seq_printf(s, "read events: %ld\n", cnt);
8555 count = simple_read_from_buffer(ubuf, count, ppos,
8556 s->buffer, trace_seq_used(s));
8563 static const struct file_operations tracing_stats_fops = {
8564 .open = tracing_open_generic_tr,
8565 .read = tracing_stats_read,
8566 .llseek = generic_file_llseek,
8567 .release = tracing_release_generic_tr,
8570 #ifdef CONFIG_DYNAMIC_FTRACE
8573 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8574 size_t cnt, loff_t *ppos)
8580 /* 256 should be plenty to hold the amount needed */
8581 buf = kmalloc(256, GFP_KERNEL);
8585 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8586 ftrace_update_tot_cnt,
8587 ftrace_number_of_pages,
8588 ftrace_number_of_groups);
8590 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8595 static const struct file_operations tracing_dyn_info_fops = {
8596 .open = tracing_open_generic,
8597 .read = tracing_read_dyn_info,
8598 .llseek = generic_file_llseek,
8600 #endif /* CONFIG_DYNAMIC_FTRACE */
8602 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8604 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8605 struct trace_array *tr, struct ftrace_probe_ops *ops,
8608 tracing_snapshot_instance(tr);
8612 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8613 struct trace_array *tr, struct ftrace_probe_ops *ops,
8616 struct ftrace_func_mapper *mapper = data;
8620 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8630 tracing_snapshot_instance(tr);
8634 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8635 struct ftrace_probe_ops *ops, void *data)
8637 struct ftrace_func_mapper *mapper = data;
8640 seq_printf(m, "%ps:", (void *)ip);
8642 seq_puts(m, "snapshot");
8645 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8648 seq_printf(m, ":count=%ld\n", *count);
8650 seq_puts(m, ":unlimited\n");
8656 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8657 unsigned long ip, void *init_data, void **data)
8659 struct ftrace_func_mapper *mapper = *data;
8662 mapper = allocate_ftrace_func_mapper();
8668 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8672 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8673 unsigned long ip, void *data)
8675 struct ftrace_func_mapper *mapper = data;
8680 free_ftrace_func_mapper(mapper, NULL);
8684 ftrace_func_mapper_remove_ip(mapper, ip);
8687 static struct ftrace_probe_ops snapshot_probe_ops = {
8688 .func = ftrace_snapshot,
8689 .print = ftrace_snapshot_print,
8692 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8693 .func = ftrace_count_snapshot,
8694 .print = ftrace_snapshot_print,
8695 .init = ftrace_snapshot_init,
8696 .free = ftrace_snapshot_free,
8700 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8701 char *glob, char *cmd, char *param, int enable)
8703 struct ftrace_probe_ops *ops;
8704 void *count = (void *)-1;
8711 /* hash funcs only work with set_ftrace_filter */
8715 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8718 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8723 number = strsep(¶m, ":");
8725 if (!strlen(number))
8729 * We use the callback data field (which is a pointer)
8732 ret = kstrtoul(number, 0, (unsigned long *)&count);
8737 ret = tracing_alloc_snapshot_instance(tr);
8741 ret = register_ftrace_function_probe(glob, tr, ops, count);
8744 return ret < 0 ? ret : 0;
8747 static struct ftrace_func_command ftrace_snapshot_cmd = {
8749 .func = ftrace_trace_snapshot_callback,
8752 static __init int register_snapshot_cmd(void)
8754 return register_ftrace_command(&ftrace_snapshot_cmd);
8757 static inline __init int register_snapshot_cmd(void) { return 0; }
8758 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8760 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8762 if (WARN_ON(!tr->dir))
8763 return ERR_PTR(-ENODEV);
8765 /* Top directory uses NULL as the parent */
8766 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8769 /* All sub buffers have a descriptor */
8773 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8775 struct dentry *d_tracer;
8778 return tr->percpu_dir;
8780 d_tracer = tracing_get_dentry(tr);
8781 if (IS_ERR(d_tracer))
8784 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8786 MEM_FAIL(!tr->percpu_dir,
8787 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8789 return tr->percpu_dir;
8792 static struct dentry *
8793 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8794 void *data, long cpu, const struct file_operations *fops)
8796 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8798 if (ret) /* See tracing_get_cpu() */
8799 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8804 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8806 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8807 struct dentry *d_cpu;
8808 char cpu_dir[30]; /* 30 characters should be more than enough */
8813 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8814 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8816 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8820 /* per cpu trace_pipe */
8821 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8822 tr, cpu, &tracing_pipe_fops);
8825 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8826 tr, cpu, &tracing_fops);
8828 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8829 tr, cpu, &tracing_buffers_fops);
8831 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8832 tr, cpu, &tracing_stats_fops);
8834 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8835 tr, cpu, &tracing_entries_fops);
8837 #ifdef CONFIG_TRACER_SNAPSHOT
8838 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8839 tr, cpu, &snapshot_fops);
8841 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8842 tr, cpu, &snapshot_raw_fops);
8846 #ifdef CONFIG_FTRACE_SELFTEST
8847 /* Let selftest have access to static functions in this file */
8848 #include "trace_selftest.c"
8852 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8855 struct trace_option_dentry *topt = filp->private_data;
8858 if (topt->flags->val & topt->opt->bit)
8863 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8867 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8870 struct trace_option_dentry *topt = filp->private_data;
8874 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8878 if (val != 0 && val != 1)
8881 if (!!(topt->flags->val & topt->opt->bit) != val) {
8882 mutex_lock(&trace_types_lock);
8883 ret = __set_tracer_option(topt->tr, topt->flags,
8885 mutex_unlock(&trace_types_lock);
8896 static const struct file_operations trace_options_fops = {
8897 .open = tracing_open_generic,
8898 .read = trace_options_read,
8899 .write = trace_options_write,
8900 .llseek = generic_file_llseek,
8904 * In order to pass in both the trace_array descriptor as well as the index
8905 * to the flag that the trace option file represents, the trace_array
8906 * has a character array of trace_flags_index[], which holds the index
8907 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8908 * The address of this character array is passed to the flag option file
8909 * read/write callbacks.
8911 * In order to extract both the index and the trace_array descriptor,
8912 * get_tr_index() uses the following algorithm.
8916 * As the pointer itself contains the address of the index (remember
8919 * Then to get the trace_array descriptor, by subtracting that index
8920 * from the ptr, we get to the start of the index itself.
8922 * ptr - idx == &index[0]
8924 * Then a simple container_of() from that pointer gets us to the
8925 * trace_array descriptor.
8927 static void get_tr_index(void *data, struct trace_array **ptr,
8928 unsigned int *pindex)
8930 *pindex = *(unsigned char *)data;
8932 *ptr = container_of(data - *pindex, struct trace_array,
8937 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8940 void *tr_index = filp->private_data;
8941 struct trace_array *tr;
8945 get_tr_index(tr_index, &tr, &index);
8947 if (tr->trace_flags & (1 << index))
8952 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8956 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8959 void *tr_index = filp->private_data;
8960 struct trace_array *tr;
8965 get_tr_index(tr_index, &tr, &index);
8967 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8971 if (val != 0 && val != 1)
8974 mutex_lock(&event_mutex);
8975 mutex_lock(&trace_types_lock);
8976 ret = set_tracer_flag(tr, 1 << index, val);
8977 mutex_unlock(&trace_types_lock);
8978 mutex_unlock(&event_mutex);
8988 static const struct file_operations trace_options_core_fops = {
8989 .open = tracing_open_generic,
8990 .read = trace_options_core_read,
8991 .write = trace_options_core_write,
8992 .llseek = generic_file_llseek,
8995 struct dentry *trace_create_file(const char *name,
8997 struct dentry *parent,
8999 const struct file_operations *fops)
9003 ret = tracefs_create_file(name, mode, parent, data, fops);
9005 pr_warn("Could not create tracefs '%s' entry\n", name);
9011 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
9013 struct dentry *d_tracer;
9018 d_tracer = tracing_get_dentry(tr);
9019 if (IS_ERR(d_tracer))
9022 tr->options = tracefs_create_dir("options", d_tracer);
9024 pr_warn("Could not create tracefs directory 'options'\n");
9032 create_trace_option_file(struct trace_array *tr,
9033 struct trace_option_dentry *topt,
9034 struct tracer_flags *flags,
9035 struct tracer_opt *opt)
9037 struct dentry *t_options;
9039 t_options = trace_options_init_dentry(tr);
9043 topt->flags = flags;
9047 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
9048 t_options, topt, &trace_options_fops);
9053 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
9055 struct trace_option_dentry *topts;
9056 struct trace_options *tr_topts;
9057 struct tracer_flags *flags;
9058 struct tracer_opt *opts;
9065 flags = tracer->flags;
9067 if (!flags || !flags->opts)
9071 * If this is an instance, only create flags for tracers
9072 * the instance may have.
9074 if (!trace_ok_for_array(tracer, tr))
9077 for (i = 0; i < tr->nr_topts; i++) {
9078 /* Make sure there's no duplicate flags. */
9079 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
9085 for (cnt = 0; opts[cnt].name; cnt++)
9088 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
9092 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9099 tr->topts = tr_topts;
9100 tr->topts[tr->nr_topts].tracer = tracer;
9101 tr->topts[tr->nr_topts].topts = topts;
9104 for (cnt = 0; opts[cnt].name; cnt++) {
9105 create_trace_option_file(tr, &topts[cnt], flags,
9107 MEM_FAIL(topts[cnt].entry == NULL,
9108 "Failed to create trace option: %s",
9113 static struct dentry *
9114 create_trace_option_core_file(struct trace_array *tr,
9115 const char *option, long index)
9117 struct dentry *t_options;
9119 t_options = trace_options_init_dentry(tr);
9123 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9124 (void *)&tr->trace_flags_index[index],
9125 &trace_options_core_fops);
9128 static void create_trace_options_dir(struct trace_array *tr)
9130 struct dentry *t_options;
9131 bool top_level = tr == &global_trace;
9134 t_options = trace_options_init_dentry(tr);
9138 for (i = 0; trace_options[i]; i++) {
9140 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9141 create_trace_option_core_file(tr, trace_options[i], i);
9146 rb_simple_read(struct file *filp, char __user *ubuf,
9147 size_t cnt, loff_t *ppos)
9149 struct trace_array *tr = filp->private_data;
9153 r = tracer_tracing_is_on(tr);
9154 r = sprintf(buf, "%d\n", r);
9156 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9160 rb_simple_write(struct file *filp, const char __user *ubuf,
9161 size_t cnt, loff_t *ppos)
9163 struct trace_array *tr = filp->private_data;
9164 struct trace_buffer *buffer = tr->array_buffer.buffer;
9168 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9173 mutex_lock(&trace_types_lock);
9174 if (!!val == tracer_tracing_is_on(tr)) {
9175 val = 0; /* do nothing */
9177 tracer_tracing_on(tr);
9178 if (tr->current_trace->start)
9179 tr->current_trace->start(tr);
9181 tracer_tracing_off(tr);
9182 if (tr->current_trace->stop)
9183 tr->current_trace->stop(tr);
9184 /* Wake up any waiters */
9185 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9187 mutex_unlock(&trace_types_lock);
9195 static const struct file_operations rb_simple_fops = {
9196 .open = tracing_open_generic_tr,
9197 .read = rb_simple_read,
9198 .write = rb_simple_write,
9199 .release = tracing_release_generic_tr,
9200 .llseek = default_llseek,
9204 buffer_percent_read(struct file *filp, char __user *ubuf,
9205 size_t cnt, loff_t *ppos)
9207 struct trace_array *tr = filp->private_data;
9211 r = tr->buffer_percent;
9212 r = sprintf(buf, "%d\n", r);
9214 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9218 buffer_percent_write(struct file *filp, const char __user *ubuf,
9219 size_t cnt, loff_t *ppos)
9221 struct trace_array *tr = filp->private_data;
9225 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9232 tr->buffer_percent = val;
9239 static const struct file_operations buffer_percent_fops = {
9240 .open = tracing_open_generic_tr,
9241 .read = buffer_percent_read,
9242 .write = buffer_percent_write,
9243 .release = tracing_release_generic_tr,
9244 .llseek = default_llseek,
9247 static struct dentry *trace_instance_dir;
9250 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9253 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9255 enum ring_buffer_flags rb_flags;
9257 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9261 buf->buffer = ring_buffer_alloc(size, rb_flags);
9265 buf->data = alloc_percpu(struct trace_array_cpu);
9267 ring_buffer_free(buf->buffer);
9272 /* Allocate the first page for all buffers */
9273 set_buffer_entries(&tr->array_buffer,
9274 ring_buffer_size(tr->array_buffer.buffer, 0));
9279 static void free_trace_buffer(struct array_buffer *buf)
9282 ring_buffer_free(buf->buffer);
9284 free_percpu(buf->data);
9289 static int allocate_trace_buffers(struct trace_array *tr, int size)
9293 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9297 #ifdef CONFIG_TRACER_MAX_TRACE
9298 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9299 allocate_snapshot ? size : 1);
9300 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9301 free_trace_buffer(&tr->array_buffer);
9304 tr->allocated_snapshot = allocate_snapshot;
9306 allocate_snapshot = false;
9312 static void free_trace_buffers(struct trace_array *tr)
9317 free_trace_buffer(&tr->array_buffer);
9319 #ifdef CONFIG_TRACER_MAX_TRACE
9320 free_trace_buffer(&tr->max_buffer);
9324 static void init_trace_flags_index(struct trace_array *tr)
9328 /* Used by the trace options files */
9329 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9330 tr->trace_flags_index[i] = i;
9333 static void __update_tracer_options(struct trace_array *tr)
9337 for (t = trace_types; t; t = t->next)
9338 add_tracer_options(tr, t);
9341 static void update_tracer_options(struct trace_array *tr)
9343 mutex_lock(&trace_types_lock);
9344 tracer_options_updated = true;
9345 __update_tracer_options(tr);
9346 mutex_unlock(&trace_types_lock);
9349 /* Must have trace_types_lock held */
9350 struct trace_array *trace_array_find(const char *instance)
9352 struct trace_array *tr, *found = NULL;
9354 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9355 if (tr->name && strcmp(tr->name, instance) == 0) {
9364 struct trace_array *trace_array_find_get(const char *instance)
9366 struct trace_array *tr;
9368 mutex_lock(&trace_types_lock);
9369 tr = trace_array_find(instance);
9372 mutex_unlock(&trace_types_lock);
9377 static int trace_array_create_dir(struct trace_array *tr)
9381 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9385 ret = event_trace_add_tracer(tr->dir, tr);
9387 tracefs_remove(tr->dir);
9391 init_tracer_tracefs(tr, tr->dir);
9392 __update_tracer_options(tr);
9397 static struct trace_array *trace_array_create(const char *name)
9399 struct trace_array *tr;
9403 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9405 return ERR_PTR(ret);
9407 tr->name = kstrdup(name, GFP_KERNEL);
9411 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9414 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9416 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9418 raw_spin_lock_init(&tr->start_lock);
9420 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9422 tr->current_trace = &nop_trace;
9424 INIT_LIST_HEAD(&tr->systems);
9425 INIT_LIST_HEAD(&tr->events);
9426 INIT_LIST_HEAD(&tr->hist_vars);
9427 INIT_LIST_HEAD(&tr->err_log);
9429 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9432 if (ftrace_allocate_ftrace_ops(tr) < 0)
9435 ftrace_init_trace_array(tr);
9437 init_trace_flags_index(tr);
9439 if (trace_instance_dir) {
9440 ret = trace_array_create_dir(tr);
9444 __trace_early_add_events(tr);
9446 list_add(&tr->list, &ftrace_trace_arrays);
9453 ftrace_free_ftrace_ops(tr);
9454 free_trace_buffers(tr);
9455 free_cpumask_var(tr->tracing_cpumask);
9459 return ERR_PTR(ret);
9462 static int instance_mkdir(const char *name)
9464 struct trace_array *tr;
9467 mutex_lock(&event_mutex);
9468 mutex_lock(&trace_types_lock);
9471 if (trace_array_find(name))
9474 tr = trace_array_create(name);
9476 ret = PTR_ERR_OR_ZERO(tr);
9479 mutex_unlock(&trace_types_lock);
9480 mutex_unlock(&event_mutex);
9485 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9486 * @name: The name of the trace array to be looked up/created.
9488 * Returns pointer to trace array with given name.
9489 * NULL, if it cannot be created.
9491 * NOTE: This function increments the reference counter associated with the
9492 * trace array returned. This makes sure it cannot be freed while in use.
9493 * Use trace_array_put() once the trace array is no longer needed.
9494 * If the trace_array is to be freed, trace_array_destroy() needs to
9495 * be called after the trace_array_put(), or simply let user space delete
9496 * it from the tracefs instances directory. But until the
9497 * trace_array_put() is called, user space can not delete it.
9500 struct trace_array *trace_array_get_by_name(const char *name)
9502 struct trace_array *tr;
9504 mutex_lock(&event_mutex);
9505 mutex_lock(&trace_types_lock);
9507 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9508 if (tr->name && strcmp(tr->name, name) == 0)
9512 tr = trace_array_create(name);
9520 mutex_unlock(&trace_types_lock);
9521 mutex_unlock(&event_mutex);
9524 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9526 static int __remove_instance(struct trace_array *tr)
9530 /* Reference counter for a newly created trace array = 1. */
9531 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9534 list_del(&tr->list);
9536 /* Disable all the flags that were enabled coming in */
9537 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9538 if ((1 << i) & ZEROED_TRACE_FLAGS)
9539 set_tracer_flag(tr, 1 << i, 0);
9542 tracing_set_nop(tr);
9543 clear_ftrace_function_probes(tr);
9544 event_trace_del_tracer(tr);
9545 ftrace_clear_pids(tr);
9546 ftrace_destroy_function_files(tr);
9547 tracefs_remove(tr->dir);
9548 free_percpu(tr->last_func_repeats);
9549 free_trace_buffers(tr);
9550 clear_tracing_err_log(tr);
9552 for (i = 0; i < tr->nr_topts; i++) {
9553 kfree(tr->topts[i].topts);
9557 free_cpumask_var(tr->tracing_cpumask);
9564 int trace_array_destroy(struct trace_array *this_tr)
9566 struct trace_array *tr;
9572 mutex_lock(&event_mutex);
9573 mutex_lock(&trace_types_lock);
9577 /* Making sure trace array exists before destroying it. */
9578 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9579 if (tr == this_tr) {
9580 ret = __remove_instance(tr);
9585 mutex_unlock(&trace_types_lock);
9586 mutex_unlock(&event_mutex);
9590 EXPORT_SYMBOL_GPL(trace_array_destroy);
9592 static int instance_rmdir(const char *name)
9594 struct trace_array *tr;
9597 mutex_lock(&event_mutex);
9598 mutex_lock(&trace_types_lock);
9601 tr = trace_array_find(name);
9603 ret = __remove_instance(tr);
9605 mutex_unlock(&trace_types_lock);
9606 mutex_unlock(&event_mutex);
9611 static __init void create_trace_instances(struct dentry *d_tracer)
9613 struct trace_array *tr;
9615 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9618 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9621 mutex_lock(&event_mutex);
9622 mutex_lock(&trace_types_lock);
9624 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9627 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9628 "Failed to create instance directory\n"))
9632 mutex_unlock(&trace_types_lock);
9633 mutex_unlock(&event_mutex);
9637 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9639 struct trace_event_file *file;
9642 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9643 tr, &show_traces_fops);
9645 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9646 tr, &set_tracer_fops);
9648 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9649 tr, &tracing_cpumask_fops);
9651 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9652 tr, &tracing_iter_fops);
9654 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9657 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9658 tr, &tracing_pipe_fops);
9660 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9661 tr, &tracing_entries_fops);
9663 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9664 tr, &tracing_total_entries_fops);
9666 trace_create_file("free_buffer", 0200, d_tracer,
9667 tr, &tracing_free_buffer_fops);
9669 trace_create_file("trace_marker", 0220, d_tracer,
9670 tr, &tracing_mark_fops);
9672 file = __find_event_file(tr, "ftrace", "print");
9673 if (file && file->dir)
9674 trace_create_file("trigger", TRACE_MODE_WRITE, file->dir,
9675 file, &event_trigger_fops);
9676 tr->trace_marker_file = file;
9678 trace_create_file("trace_marker_raw", 0220, d_tracer,
9679 tr, &tracing_mark_raw_fops);
9681 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9684 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9685 tr, &rb_simple_fops);
9687 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9688 &trace_time_stamp_mode_fops);
9690 tr->buffer_percent = 50;
9692 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9693 tr, &buffer_percent_fops);
9695 create_trace_options_dir(tr);
9697 #ifdef CONFIG_TRACER_MAX_TRACE
9698 trace_create_maxlat_file(tr, d_tracer);
9701 if (ftrace_create_function_files(tr, d_tracer))
9702 MEM_FAIL(1, "Could not allocate function filter files");
9704 #ifdef CONFIG_TRACER_SNAPSHOT
9705 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9706 tr, &snapshot_fops);
9709 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9710 tr, &tracing_err_log_fops);
9712 for_each_tracing_cpu(cpu)
9713 tracing_init_tracefs_percpu(tr, cpu);
9715 ftrace_init_tracefs(tr, d_tracer);
9718 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9720 struct vfsmount *mnt;
9721 struct file_system_type *type;
9724 * To maintain backward compatibility for tools that mount
9725 * debugfs to get to the tracing facility, tracefs is automatically
9726 * mounted to the debugfs/tracing directory.
9728 type = get_fs_type("tracefs");
9731 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9732 put_filesystem(type);
9741 * tracing_init_dentry - initialize top level trace array
9743 * This is called when creating files or directories in the tracing
9744 * directory. It is called via fs_initcall() by any of the boot up code
9745 * and expects to return the dentry of the top level tracing directory.
9747 int tracing_init_dentry(void)
9749 struct trace_array *tr = &global_trace;
9751 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9752 pr_warn("Tracing disabled due to lockdown\n");
9756 /* The top level trace array uses NULL as parent */
9760 if (WARN_ON(!tracefs_initialized()))
9764 * As there may still be users that expect the tracing
9765 * files to exist in debugfs/tracing, we must automount
9766 * the tracefs file system there, so older tools still
9767 * work with the newer kernel.
9769 tr->dir = debugfs_create_automount("tracing", NULL,
9770 trace_automount, NULL);
9775 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9776 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9778 static struct workqueue_struct *eval_map_wq __initdata;
9779 static struct work_struct eval_map_work __initdata;
9780 static struct work_struct tracerfs_init_work __initdata;
9782 static void __init eval_map_work_func(struct work_struct *work)
9786 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9787 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9790 static int __init trace_eval_init(void)
9792 INIT_WORK(&eval_map_work, eval_map_work_func);
9794 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9796 pr_err("Unable to allocate eval_map_wq\n");
9798 eval_map_work_func(&eval_map_work);
9802 queue_work(eval_map_wq, &eval_map_work);
9806 subsys_initcall(trace_eval_init);
9808 static int __init trace_eval_sync(void)
9810 /* Make sure the eval map updates are finished */
9812 destroy_workqueue(eval_map_wq);
9816 late_initcall_sync(trace_eval_sync);
9819 #ifdef CONFIG_MODULES
9820 static void trace_module_add_evals(struct module *mod)
9822 if (!mod->num_trace_evals)
9826 * Modules with bad taint do not have events created, do
9827 * not bother with enums either.
9829 if (trace_module_has_bad_taint(mod))
9832 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9835 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
9836 static void trace_module_remove_evals(struct module *mod)
9838 union trace_eval_map_item *map;
9839 union trace_eval_map_item **last = &trace_eval_maps;
9841 if (!mod->num_trace_evals)
9844 mutex_lock(&trace_eval_mutex);
9846 map = trace_eval_maps;
9849 if (map->head.mod == mod)
9851 map = trace_eval_jmp_to_tail(map);
9852 last = &map->tail.next;
9853 map = map->tail.next;
9858 *last = trace_eval_jmp_to_tail(map)->tail.next;
9861 mutex_unlock(&trace_eval_mutex);
9864 static inline void trace_module_remove_evals(struct module *mod) { }
9865 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9867 static int trace_module_notify(struct notifier_block *self,
9868 unsigned long val, void *data)
9870 struct module *mod = data;
9873 case MODULE_STATE_COMING:
9874 trace_module_add_evals(mod);
9876 case MODULE_STATE_GOING:
9877 trace_module_remove_evals(mod);
9884 static struct notifier_block trace_module_nb = {
9885 .notifier_call = trace_module_notify,
9888 #endif /* CONFIG_MODULES */
9890 static __init void tracer_init_tracefs_work_func(struct work_struct *work)
9895 init_tracer_tracefs(&global_trace, NULL);
9896 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9898 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9899 &global_trace, &tracing_thresh_fops);
9901 trace_create_file("README", TRACE_MODE_READ, NULL,
9902 NULL, &tracing_readme_fops);
9904 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9905 NULL, &tracing_saved_cmdlines_fops);
9907 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9908 NULL, &tracing_saved_cmdlines_size_fops);
9910 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9911 NULL, &tracing_saved_tgids_fops);
9913 trace_create_eval_file(NULL);
9915 #ifdef CONFIG_MODULES
9916 register_module_notifier(&trace_module_nb);
9919 #ifdef CONFIG_DYNAMIC_FTRACE
9920 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
9921 NULL, &tracing_dyn_info_fops);
9924 create_trace_instances(NULL);
9926 update_tracer_options(&global_trace);
9929 static __init int tracer_init_tracefs(void)
9933 trace_access_lock_init();
9935 ret = tracing_init_dentry();
9940 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
9941 queue_work(eval_map_wq, &tracerfs_init_work);
9943 tracer_init_tracefs_work_func(NULL);
9946 rv_init_interface();
9951 fs_initcall(tracer_init_tracefs);
9953 static int trace_die_panic_handler(struct notifier_block *self,
9954 unsigned long ev, void *unused);
9956 static struct notifier_block trace_panic_notifier = {
9957 .notifier_call = trace_die_panic_handler,
9958 .priority = INT_MAX - 1,
9961 static struct notifier_block trace_die_notifier = {
9962 .notifier_call = trace_die_panic_handler,
9963 .priority = INT_MAX - 1,
9967 * The idea is to execute the following die/panic callback early, in order
9968 * to avoid showing irrelevant information in the trace (like other panic
9969 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
9970 * warnings get disabled (to prevent potential log flooding).
9972 static int trace_die_panic_handler(struct notifier_block *self,
9973 unsigned long ev, void *unused)
9975 if (!ftrace_dump_on_oops)
9978 /* The die notifier requires DIE_OOPS to trigger */
9979 if (self == &trace_die_notifier && ev != DIE_OOPS)
9982 ftrace_dump(ftrace_dump_on_oops);
9988 * printk is set to max of 1024, we really don't need it that big.
9989 * Nothing should be printing 1000 characters anyway.
9991 #define TRACE_MAX_PRINT 1000
9994 * Define here KERN_TRACE so that we have one place to modify
9995 * it if we decide to change what log level the ftrace dump
9998 #define KERN_TRACE KERN_EMERG
10001 trace_printk_seq(struct trace_seq *s)
10003 /* Probably should print a warning here. */
10004 if (s->seq.len >= TRACE_MAX_PRINT)
10005 s->seq.len = TRACE_MAX_PRINT;
10008 * More paranoid code. Although the buffer size is set to
10009 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
10010 * an extra layer of protection.
10012 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
10013 s->seq.len = s->seq.size - 1;
10015 /* should be zero ended, but we are paranoid. */
10016 s->buffer[s->seq.len] = 0;
10018 printk(KERN_TRACE "%s", s->buffer);
10023 void trace_init_global_iter(struct trace_iterator *iter)
10025 iter->tr = &global_trace;
10026 iter->trace = iter->tr->current_trace;
10027 iter->cpu_file = RING_BUFFER_ALL_CPUS;
10028 iter->array_buffer = &global_trace.array_buffer;
10030 if (iter->trace && iter->trace->open)
10031 iter->trace->open(iter);
10033 /* Annotate start of buffers if we had overruns */
10034 if (ring_buffer_overruns(iter->array_buffer->buffer))
10035 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10037 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10038 if (trace_clocks[iter->tr->clock_id].in_ns)
10039 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10041 /* Can not use kmalloc for iter.temp and iter.fmt */
10042 iter->temp = static_temp_buf;
10043 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10044 iter->fmt = static_fmt_buf;
10045 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10048 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10050 /* use static because iter can be a bit big for the stack */
10051 static struct trace_iterator iter;
10052 static atomic_t dump_running;
10053 struct trace_array *tr = &global_trace;
10054 unsigned int old_userobj;
10055 unsigned long flags;
10058 /* Only allow one dump user at a time. */
10059 if (atomic_inc_return(&dump_running) != 1) {
10060 atomic_dec(&dump_running);
10065 * Always turn off tracing when we dump.
10066 * We don't need to show trace output of what happens
10067 * between multiple crashes.
10069 * If the user does a sysrq-z, then they can re-enable
10070 * tracing with echo 1 > tracing_on.
10074 local_irq_save(flags);
10076 /* Simulate the iterator */
10077 trace_init_global_iter(&iter);
10079 for_each_tracing_cpu(cpu) {
10080 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10083 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10085 /* don't look at user memory in panic mode */
10086 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10088 switch (oops_dump_mode) {
10090 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10093 iter.cpu_file = raw_smp_processor_id();
10098 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10099 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10102 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10104 /* Did function tracer already get disabled? */
10105 if (ftrace_is_dead()) {
10106 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10107 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10111 * We need to stop all tracing on all CPUS to read
10112 * the next buffer. This is a bit expensive, but is
10113 * not done often. We fill all what we can read,
10114 * and then release the locks again.
10117 while (!trace_empty(&iter)) {
10120 printk(KERN_TRACE "---------------------------------\n");
10124 trace_iterator_reset(&iter);
10125 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10127 if (trace_find_next_entry_inc(&iter) != NULL) {
10130 ret = print_trace_line(&iter);
10131 if (ret != TRACE_TYPE_NO_CONSUME)
10132 trace_consume(&iter);
10134 touch_nmi_watchdog();
10136 trace_printk_seq(&iter.seq);
10140 printk(KERN_TRACE " (ftrace buffer empty)\n");
10142 printk(KERN_TRACE "---------------------------------\n");
10145 tr->trace_flags |= old_userobj;
10147 for_each_tracing_cpu(cpu) {
10148 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10150 atomic_dec(&dump_running);
10151 local_irq_restore(flags);
10153 EXPORT_SYMBOL_GPL(ftrace_dump);
10155 #define WRITE_BUFSIZE 4096
10157 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10158 size_t count, loff_t *ppos,
10159 int (*createfn)(const char *))
10161 char *kbuf, *buf, *tmp;
10166 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10170 while (done < count) {
10171 size = count - done;
10173 if (size >= WRITE_BUFSIZE)
10174 size = WRITE_BUFSIZE - 1;
10176 if (copy_from_user(kbuf, buffer + done, size)) {
10183 tmp = strchr(buf, '\n');
10186 size = tmp - buf + 1;
10188 size = strlen(buf);
10189 if (done + size < count) {
10192 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10193 pr_warn("Line length is too long: Should be less than %d\n",
10194 WRITE_BUFSIZE - 2);
10201 /* Remove comments */
10202 tmp = strchr(buf, '#');
10207 ret = createfn(buf);
10212 } while (done < count);
10222 #ifdef CONFIG_TRACER_MAX_TRACE
10223 __init static bool tr_needs_alloc_snapshot(const char *name)
10226 int len = strlen(name);
10229 if (!boot_snapshot_index)
10232 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10233 boot_snapshot_info[len] == '\t')
10236 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10240 sprintf(test, "\t%s\t", name);
10241 ret = strstr(boot_snapshot_info, test) == NULL;
10246 __init static void do_allocate_snapshot(const char *name)
10248 if (!tr_needs_alloc_snapshot(name))
10252 * When allocate_snapshot is set, the next call to
10253 * allocate_trace_buffers() (called by trace_array_get_by_name())
10254 * will allocate the snapshot buffer. That will alse clear
10257 allocate_snapshot = true;
10260 static inline void do_allocate_snapshot(const char *name) { }
10263 __init static void enable_instances(void)
10265 struct trace_array *tr;
10270 /* A tab is always appended */
10271 boot_instance_info[boot_instance_index - 1] = '\0';
10272 str = boot_instance_info;
10274 while ((curr_str = strsep(&str, "\t"))) {
10276 tok = strsep(&curr_str, ",");
10278 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10279 do_allocate_snapshot(tok);
10281 tr = trace_array_get_by_name(tok);
10283 pr_warn("Failed to create instance buffer %s\n", curr_str);
10286 /* Allow user space to delete it */
10287 trace_array_put(tr);
10289 while ((tok = strsep(&curr_str, ","))) {
10290 early_enable_events(tr, tok, true);
10295 __init static int tracer_alloc_buffers(void)
10301 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10302 pr_warn("Tracing disabled due to lockdown\n");
10307 * Make sure we don't accidentally add more trace options
10308 * than we have bits for.
10310 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10312 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10315 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10316 goto out_free_buffer_mask;
10318 /* Only allocate trace_printk buffers if a trace_printk exists */
10319 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10320 /* Must be called before global_trace.buffer is allocated */
10321 trace_printk_init_buffers();
10323 /* To save memory, keep the ring buffer size to its minimum */
10324 if (ring_buffer_expanded)
10325 ring_buf_size = trace_buf_size;
10329 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10330 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10332 raw_spin_lock_init(&global_trace.start_lock);
10335 * The prepare callbacks allocates some memory for the ring buffer. We
10336 * don't free the buffer if the CPU goes down. If we were to free
10337 * the buffer, then the user would lose any trace that was in the
10338 * buffer. The memory will be removed once the "instance" is removed.
10340 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10341 "trace/RB:prepare", trace_rb_cpu_prepare,
10344 goto out_free_cpumask;
10345 /* Used for event triggers */
10347 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10349 goto out_rm_hp_state;
10351 if (trace_create_savedcmd() < 0)
10352 goto out_free_temp_buffer;
10354 /* TODO: make the number of buffers hot pluggable with CPUS */
10355 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10356 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10357 goto out_free_savedcmd;
10360 if (global_trace.buffer_disabled)
10363 if (trace_boot_clock) {
10364 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10366 pr_warn("Trace clock %s not defined, going back to default\n",
10371 * register_tracer() might reference current_trace, so it
10372 * needs to be set before we register anything. This is
10373 * just a bootstrap of current_trace anyway.
10375 global_trace.current_trace = &nop_trace;
10377 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10379 ftrace_init_global_array_ops(&global_trace);
10381 init_trace_flags_index(&global_trace);
10383 register_tracer(&nop_trace);
10385 /* Function tracing may start here (via kernel command line) */
10386 init_function_trace();
10388 /* All seems OK, enable tracing */
10389 tracing_disabled = 0;
10391 atomic_notifier_chain_register(&panic_notifier_list,
10392 &trace_panic_notifier);
10394 register_die_notifier(&trace_die_notifier);
10396 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10398 INIT_LIST_HEAD(&global_trace.systems);
10399 INIT_LIST_HEAD(&global_trace.events);
10400 INIT_LIST_HEAD(&global_trace.hist_vars);
10401 INIT_LIST_HEAD(&global_trace.err_log);
10402 list_add(&global_trace.list, &ftrace_trace_arrays);
10404 apply_trace_boot_options();
10406 register_snapshot_cmd();
10413 free_saved_cmdlines_buffer(savedcmd);
10414 out_free_temp_buffer:
10415 ring_buffer_free(temp_buffer);
10417 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10419 free_cpumask_var(global_trace.tracing_cpumask);
10420 out_free_buffer_mask:
10421 free_cpumask_var(tracing_buffer_mask);
10426 void __init ftrace_boot_snapshot(void)
10428 #ifdef CONFIG_TRACER_MAX_TRACE
10429 struct trace_array *tr;
10431 if (!snapshot_at_boot)
10434 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10435 if (!tr->allocated_snapshot)
10438 tracing_snapshot_instance(tr);
10439 trace_array_puts(tr, "** Boot snapshot taken **\n");
10444 void __init early_trace_init(void)
10446 if (tracepoint_printk) {
10447 tracepoint_print_iter =
10448 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10449 if (MEM_FAIL(!tracepoint_print_iter,
10450 "Failed to allocate trace iterator\n"))
10451 tracepoint_printk = 0;
10453 static_key_enable(&tracepoint_printk_key.key);
10455 tracer_alloc_buffers();
10460 void __init trace_init(void)
10462 trace_event_init();
10464 if (boot_instance_index)
10465 enable_instances();
10468 __init static void clear_boot_tracer(void)
10471 * The default tracer at boot buffer is an init section.
10472 * This function is called in lateinit. If we did not
10473 * find the boot tracer, then clear it out, to prevent
10474 * later registration from accessing the buffer that is
10475 * about to be freed.
10477 if (!default_bootup_tracer)
10480 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10481 default_bootup_tracer);
10482 default_bootup_tracer = NULL;
10485 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10486 __init static void tracing_set_default_clock(void)
10488 /* sched_clock_stable() is determined in late_initcall */
10489 if (!trace_boot_clock && !sched_clock_stable()) {
10490 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10491 pr_warn("Can not set tracing clock due to lockdown\n");
10495 printk(KERN_WARNING
10496 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10497 "If you want to keep using the local clock, then add:\n"
10498 " \"trace_clock=local\"\n"
10499 "on the kernel command line\n");
10500 tracing_set_clock(&global_trace, "global");
10504 static inline void tracing_set_default_clock(void) { }
10507 __init static int late_trace_init(void)
10509 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10510 static_key_disable(&tracepoint_printk_key.key);
10511 tracepoint_printk = 0;
10514 tracing_set_default_clock();
10515 clear_boot_tracer();
10519 late_initcall_sync(late_trace_init);