1 // SPDX-License-Identifier: GPL-2.0
3 * ring buffer based function tracer
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
15 #include <linux/ring_buffer.h>
16 #include <generated/utsrelease.h>
17 #include <linux/stacktrace.h>
18 #include <linux/writeback.h>
19 #include <linux/kallsyms.h>
20 #include <linux/security.h>
21 #include <linux/seq_file.h>
22 #include <linux/notifier.h>
23 #include <linux/irqflags.h>
24 #include <linux/debugfs.h>
25 #include <linux/tracefs.h>
26 #include <linux/pagemap.h>
27 #include <linux/hardirq.h>
28 #include <linux/linkage.h>
29 #include <linux/uaccess.h>
30 #include <linux/vmalloc.h>
31 #include <linux/ftrace.h>
32 #include <linux/module.h>
33 #include <linux/percpu.h>
34 #include <linux/splice.h>
35 #include <linux/kdebug.h>
36 #include <linux/string.h>
37 #include <linux/mount.h>
38 #include <linux/rwsem.h>
39 #include <linux/slab.h>
40 #include <linux/ctype.h>
41 #include <linux/init.h>
42 #include <linux/panic_notifier.h>
43 #include <linux/poll.h>
44 #include <linux/nmi.h>
46 #include <linux/trace.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/rt.h>
49 #include <linux/fsnotify.h>
50 #include <linux/irq_work.h>
51 #include <linux/workqueue.h>
54 #include "trace_output.h"
57 * On boot up, the ring buffer is set to the minimum size, so that
58 * we do not waste memory on systems that are not using tracing.
60 bool ring_buffer_expanded;
63 * We need to change this state when a selftest is running.
64 * A selftest will lurk into the ring-buffer to count the
65 * entries inserted during the selftest although some concurrent
66 * insertions into the ring-buffer such as trace_printk could occurred
67 * at the same time, giving false positive or negative results.
69 static bool __read_mostly tracing_selftest_running;
72 * If boot-time tracing including tracers/events via kernel cmdline
73 * is running, we do not want to run SELFTEST.
75 bool __read_mostly tracing_selftest_disabled;
77 #ifdef CONFIG_FTRACE_STARTUP_TEST
78 void __init disable_tracing_selftest(const char *reason)
80 if (!tracing_selftest_disabled) {
81 tracing_selftest_disabled = true;
82 pr_info("Ftrace startup test is disabled due to %s\n", reason);
87 /* Pipe tracepoints to printk */
88 static struct trace_iterator *tracepoint_print_iter;
89 int tracepoint_printk;
90 static bool tracepoint_printk_stop_on_boot __initdata;
91 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
93 /* For tracers that don't implement custom flags */
94 static struct tracer_opt dummy_tracer_opt[] = {
99 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
105 * To prevent the comm cache from being overwritten when no
106 * tracing is active, only save the comm when a trace event
109 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
112 * Kill all tracing for good (never come back).
113 * It is initialized to 1 but will turn to zero if the initialization
114 * of the tracer is successful. But that is the only place that sets
117 static int tracing_disabled = 1;
119 cpumask_var_t __read_mostly tracing_buffer_mask;
122 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
124 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
125 * is set, then ftrace_dump is called. This will output the contents
126 * of the ftrace buffers to the console. This is very useful for
127 * capturing traces that lead to crashes and outputing it to a
130 * It is default off, but you can enable it with either specifying
131 * "ftrace_dump_on_oops" in the kernel command line, or setting
132 * /proc/sys/kernel/ftrace_dump_on_oops
133 * Set 1 if you want to dump buffers of all CPUs
134 * Set 2 if you want to dump the buffer of the CPU that triggered oops
137 enum ftrace_dump_mode ftrace_dump_on_oops;
139 /* When set, tracing will stop when a WARN*() is hit */
140 int __disable_trace_on_warning;
142 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
143 /* Map of enums to their values, for "eval_map" file */
144 struct trace_eval_map_head {
146 unsigned long length;
149 union trace_eval_map_item;
151 struct trace_eval_map_tail {
153 * "end" is first and points to NULL as it must be different
154 * than "mod" or "eval_string"
156 union trace_eval_map_item *next;
157 const char *end; /* points to NULL */
160 static DEFINE_MUTEX(trace_eval_mutex);
163 * The trace_eval_maps are saved in an array with two extra elements,
164 * one at the beginning, and one at the end. The beginning item contains
165 * the count of the saved maps (head.length), and the module they
166 * belong to if not built in (head.mod). The ending item contains a
167 * pointer to the next array of saved eval_map items.
169 union trace_eval_map_item {
170 struct trace_eval_map map;
171 struct trace_eval_map_head head;
172 struct trace_eval_map_tail tail;
175 static union trace_eval_map_item *trace_eval_maps;
176 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
178 int tracing_set_tracer(struct trace_array *tr, const char *buf);
179 static void ftrace_trace_userstack(struct trace_array *tr,
180 struct trace_buffer *buffer,
181 unsigned int trace_ctx);
183 #define MAX_TRACER_SIZE 100
184 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
185 static char *default_bootup_tracer;
187 static bool allocate_snapshot;
188 static bool snapshot_at_boot;
190 static int __init set_cmdline_ftrace(char *str)
192 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
193 default_bootup_tracer = bootup_tracer_buf;
194 /* We are using ftrace early, expand it */
195 ring_buffer_expanded = true;
198 __setup("ftrace=", set_cmdline_ftrace);
200 static int __init set_ftrace_dump_on_oops(char *str)
202 if (*str++ != '=' || !*str || !strcmp("1", str)) {
203 ftrace_dump_on_oops = DUMP_ALL;
207 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
208 ftrace_dump_on_oops = DUMP_ORIG;
214 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
216 static int __init stop_trace_on_warning(char *str)
218 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
219 __disable_trace_on_warning = 1;
222 __setup("traceoff_on_warning", stop_trace_on_warning);
224 static int __init boot_alloc_snapshot(char *str)
226 allocate_snapshot = true;
227 /* We also need the main ring buffer expanded */
228 ring_buffer_expanded = true;
231 __setup("alloc_snapshot", boot_alloc_snapshot);
234 static int __init boot_snapshot(char *str)
236 snapshot_at_boot = true;
237 boot_alloc_snapshot(str);
240 __setup("ftrace_boot_snapshot", boot_snapshot);
243 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
245 static int __init set_trace_boot_options(char *str)
247 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
250 __setup("trace_options=", set_trace_boot_options);
252 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
253 static char *trace_boot_clock __initdata;
255 static int __init set_trace_boot_clock(char *str)
257 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
258 trace_boot_clock = trace_boot_clock_buf;
261 __setup("trace_clock=", set_trace_boot_clock);
263 static int __init set_tracepoint_printk(char *str)
265 /* Ignore the "tp_printk_stop_on_boot" param */
269 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
270 tracepoint_printk = 1;
273 __setup("tp_printk", set_tracepoint_printk);
275 static int __init set_tracepoint_printk_stop(char *str)
277 tracepoint_printk_stop_on_boot = true;
280 __setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
282 unsigned long long ns2usecs(u64 nsec)
290 trace_process_export(struct trace_export *export,
291 struct ring_buffer_event *event, int flag)
293 struct trace_entry *entry;
294 unsigned int size = 0;
296 if (export->flags & flag) {
297 entry = ring_buffer_event_data(event);
298 size = ring_buffer_event_length(event);
299 export->write(export, entry, size);
303 static DEFINE_MUTEX(ftrace_export_lock);
305 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
307 static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
308 static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
309 static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
311 static inline void ftrace_exports_enable(struct trace_export *export)
313 if (export->flags & TRACE_EXPORT_FUNCTION)
314 static_branch_inc(&trace_function_exports_enabled);
316 if (export->flags & TRACE_EXPORT_EVENT)
317 static_branch_inc(&trace_event_exports_enabled);
319 if (export->flags & TRACE_EXPORT_MARKER)
320 static_branch_inc(&trace_marker_exports_enabled);
323 static inline void ftrace_exports_disable(struct trace_export *export)
325 if (export->flags & TRACE_EXPORT_FUNCTION)
326 static_branch_dec(&trace_function_exports_enabled);
328 if (export->flags & TRACE_EXPORT_EVENT)
329 static_branch_dec(&trace_event_exports_enabled);
331 if (export->flags & TRACE_EXPORT_MARKER)
332 static_branch_dec(&trace_marker_exports_enabled);
335 static void ftrace_exports(struct ring_buffer_event *event, int flag)
337 struct trace_export *export;
339 preempt_disable_notrace();
341 export = rcu_dereference_raw_check(ftrace_exports_list);
343 trace_process_export(export, event, flag);
344 export = rcu_dereference_raw_check(export->next);
347 preempt_enable_notrace();
351 add_trace_export(struct trace_export **list, struct trace_export *export)
353 rcu_assign_pointer(export->next, *list);
355 * We are entering export into the list but another
356 * CPU might be walking that list. We need to make sure
357 * the export->next pointer is valid before another CPU sees
358 * the export pointer included into the list.
360 rcu_assign_pointer(*list, export);
364 rm_trace_export(struct trace_export **list, struct trace_export *export)
366 struct trace_export **p;
368 for (p = list; *p != NULL; p = &(*p)->next)
375 rcu_assign_pointer(*p, (*p)->next);
381 add_ftrace_export(struct trace_export **list, struct trace_export *export)
383 ftrace_exports_enable(export);
385 add_trace_export(list, export);
389 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
393 ret = rm_trace_export(list, export);
394 ftrace_exports_disable(export);
399 int register_ftrace_export(struct trace_export *export)
401 if (WARN_ON_ONCE(!export->write))
404 mutex_lock(&ftrace_export_lock);
406 add_ftrace_export(&ftrace_exports_list, export);
408 mutex_unlock(&ftrace_export_lock);
412 EXPORT_SYMBOL_GPL(register_ftrace_export);
414 int unregister_ftrace_export(struct trace_export *export)
418 mutex_lock(&ftrace_export_lock);
420 ret = rm_ftrace_export(&ftrace_exports_list, export);
422 mutex_unlock(&ftrace_export_lock);
426 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
428 /* trace_flags holds trace_options default values */
429 #define TRACE_DEFAULT_FLAGS \
430 (FUNCTION_DEFAULT_FLAGS | \
431 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
432 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
433 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
434 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
437 /* trace_options that are only supported by global_trace */
438 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
439 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
441 /* trace_flags that are default zero for instances */
442 #define ZEROED_TRACE_FLAGS \
443 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
446 * The global_trace is the descriptor that holds the top-level tracing
447 * buffers for the live tracing.
449 static struct trace_array global_trace = {
450 .trace_flags = TRACE_DEFAULT_FLAGS,
453 LIST_HEAD(ftrace_trace_arrays);
455 int trace_array_get(struct trace_array *this_tr)
457 struct trace_array *tr;
460 mutex_lock(&trace_types_lock);
461 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
468 mutex_unlock(&trace_types_lock);
473 static void __trace_array_put(struct trace_array *this_tr)
475 WARN_ON(!this_tr->ref);
480 * trace_array_put - Decrement the reference counter for this trace array.
481 * @this_tr : pointer to the trace array
483 * NOTE: Use this when we no longer need the trace array returned by
484 * trace_array_get_by_name(). This ensures the trace array can be later
488 void trace_array_put(struct trace_array *this_tr)
493 mutex_lock(&trace_types_lock);
494 __trace_array_put(this_tr);
495 mutex_unlock(&trace_types_lock);
497 EXPORT_SYMBOL_GPL(trace_array_put);
499 int tracing_check_open_get_tr(struct trace_array *tr)
503 ret = security_locked_down(LOCKDOWN_TRACEFS);
507 if (tracing_disabled)
510 if (tr && trace_array_get(tr) < 0)
516 int call_filter_check_discard(struct trace_event_call *call, void *rec,
517 struct trace_buffer *buffer,
518 struct ring_buffer_event *event)
520 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
521 !filter_match_preds(call->filter, rec)) {
522 __trace_event_discard_commit(buffer, event);
530 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
531 * @filtered_pids: The list of pids to check
532 * @search_pid: The PID to find in @filtered_pids
534 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
537 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
539 return trace_pid_list_is_set(filtered_pids, search_pid);
543 * trace_ignore_this_task - should a task be ignored for tracing
544 * @filtered_pids: The list of pids to check
545 * @filtered_no_pids: The list of pids not to be traced
546 * @task: The task that should be ignored if not filtered
548 * Checks if @task should be traced or not from @filtered_pids.
549 * Returns true if @task should *NOT* be traced.
550 * Returns false if @task should be traced.
553 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
554 struct trace_pid_list *filtered_no_pids,
555 struct task_struct *task)
558 * If filtered_no_pids is not empty, and the task's pid is listed
559 * in filtered_no_pids, then return true.
560 * Otherwise, if filtered_pids is empty, that means we can
561 * trace all tasks. If it has content, then only trace pids
562 * within filtered_pids.
565 return (filtered_pids &&
566 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
568 trace_find_filtered_pid(filtered_no_pids, task->pid));
572 * trace_filter_add_remove_task - Add or remove a task from a pid_list
573 * @pid_list: The list to modify
574 * @self: The current task for fork or NULL for exit
575 * @task: The task to add or remove
577 * If adding a task, if @self is defined, the task is only added if @self
578 * is also included in @pid_list. This happens on fork and tasks should
579 * only be added when the parent is listed. If @self is NULL, then the
580 * @task pid will be removed from the list, which would happen on exit
583 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
584 struct task_struct *self,
585 struct task_struct *task)
590 /* For forks, we only add if the forking task is listed */
592 if (!trace_find_filtered_pid(pid_list, self->pid))
596 /* "self" is set for forks, and NULL for exits */
598 trace_pid_list_set(pid_list, task->pid);
600 trace_pid_list_clear(pid_list, task->pid);
604 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
605 * @pid_list: The pid list to show
606 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
607 * @pos: The position of the file
609 * This is used by the seq_file "next" operation to iterate the pids
610 * listed in a trace_pid_list structure.
612 * Returns the pid+1 as we want to display pid of zero, but NULL would
613 * stop the iteration.
615 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
617 long pid = (unsigned long)v;
622 /* pid already is +1 of the actual previous bit */
623 if (trace_pid_list_next(pid_list, pid, &next) < 0)
628 /* Return pid + 1 to allow zero to be represented */
629 return (void *)(pid + 1);
633 * trace_pid_start - Used for seq_file to start reading pid lists
634 * @pid_list: The pid list to show
635 * @pos: The position of the file
637 * This is used by seq_file "start" operation to start the iteration
640 * Returns the pid+1 as we want to display pid of zero, but NULL would
641 * stop the iteration.
643 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
649 if (trace_pid_list_first(pid_list, &first) < 0)
654 /* Return pid + 1 so that zero can be the exit value */
655 for (pid++; pid && l < *pos;
656 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
662 * trace_pid_show - show the current pid in seq_file processing
663 * @m: The seq_file structure to write into
664 * @v: A void pointer of the pid (+1) value to display
666 * Can be directly used by seq_file operations to display the current
669 int trace_pid_show(struct seq_file *m, void *v)
671 unsigned long pid = (unsigned long)v - 1;
673 seq_printf(m, "%lu\n", pid);
677 /* 128 should be much more than enough */
678 #define PID_BUF_SIZE 127
680 int trace_pid_write(struct trace_pid_list *filtered_pids,
681 struct trace_pid_list **new_pid_list,
682 const char __user *ubuf, size_t cnt)
684 struct trace_pid_list *pid_list;
685 struct trace_parser parser;
693 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
697 * Always recreate a new array. The write is an all or nothing
698 * operation. Always create a new array when adding new pids by
699 * the user. If the operation fails, then the current list is
702 pid_list = trace_pid_list_alloc();
704 trace_parser_put(&parser);
709 /* copy the current bits to the new max */
710 ret = trace_pid_list_first(filtered_pids, &pid);
712 trace_pid_list_set(pid_list, pid);
713 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
723 ret = trace_get_user(&parser, ubuf, cnt, &pos);
731 if (!trace_parser_loaded(&parser))
735 if (kstrtoul(parser.buffer, 0, &val))
740 if (trace_pid_list_set(pid_list, pid) < 0) {
746 trace_parser_clear(&parser);
749 trace_parser_put(&parser);
752 trace_pid_list_free(pid_list);
757 /* Cleared the list of pids */
758 trace_pid_list_free(pid_list);
762 *new_pid_list = pid_list;
767 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
771 /* Early boot up does not have a buffer yet */
773 return trace_clock_local();
775 ts = ring_buffer_time_stamp(buf->buffer);
776 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
781 u64 ftrace_now(int cpu)
783 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
787 * tracing_is_enabled - Show if global_trace has been enabled
789 * Shows if the global trace has been enabled or not. It uses the
790 * mirror flag "buffer_disabled" to be used in fast paths such as for
791 * the irqsoff tracer. But it may be inaccurate due to races. If you
792 * need to know the accurate state, use tracing_is_on() which is a little
793 * slower, but accurate.
795 int tracing_is_enabled(void)
798 * For quick access (irqsoff uses this in fast path), just
799 * return the mirror variable of the state of the ring buffer.
800 * It's a little racy, but we don't really care.
803 return !global_trace.buffer_disabled;
807 * trace_buf_size is the size in bytes that is allocated
808 * for a buffer. Note, the number of bytes is always rounded
811 * This number is purposely set to a low number of 16384.
812 * If the dump on oops happens, it will be much appreciated
813 * to not have to wait for all that output. Anyway this can be
814 * boot time and run time configurable.
816 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
818 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
820 /* trace_types holds a link list of available tracers. */
821 static struct tracer *trace_types __read_mostly;
824 * trace_types_lock is used to protect the trace_types list.
826 DEFINE_MUTEX(trace_types_lock);
829 * serialize the access of the ring buffer
831 * ring buffer serializes readers, but it is low level protection.
832 * The validity of the events (which returns by ring_buffer_peek() ..etc)
833 * are not protected by ring buffer.
835 * The content of events may become garbage if we allow other process consumes
836 * these events concurrently:
837 * A) the page of the consumed events may become a normal page
838 * (not reader page) in ring buffer, and this page will be rewritten
839 * by events producer.
840 * B) The page of the consumed events may become a page for splice_read,
841 * and this page will be returned to system.
843 * These primitives allow multi process access to different cpu ring buffer
846 * These primitives don't distinguish read-only and read-consume access.
847 * Multi read-only access are also serialized.
851 static DECLARE_RWSEM(all_cpu_access_lock);
852 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
854 static inline void trace_access_lock(int cpu)
856 if (cpu == RING_BUFFER_ALL_CPUS) {
857 /* gain it for accessing the whole ring buffer. */
858 down_write(&all_cpu_access_lock);
860 /* gain it for accessing a cpu ring buffer. */
862 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
863 down_read(&all_cpu_access_lock);
865 /* Secondly block other access to this @cpu ring buffer. */
866 mutex_lock(&per_cpu(cpu_access_lock, cpu));
870 static inline void trace_access_unlock(int cpu)
872 if (cpu == RING_BUFFER_ALL_CPUS) {
873 up_write(&all_cpu_access_lock);
875 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
876 up_read(&all_cpu_access_lock);
880 static inline void trace_access_lock_init(void)
884 for_each_possible_cpu(cpu)
885 mutex_init(&per_cpu(cpu_access_lock, cpu));
890 static DEFINE_MUTEX(access_lock);
892 static inline void trace_access_lock(int cpu)
895 mutex_lock(&access_lock);
898 static inline void trace_access_unlock(int cpu)
901 mutex_unlock(&access_lock);
904 static inline void trace_access_lock_init(void)
910 #ifdef CONFIG_STACKTRACE
911 static void __ftrace_trace_stack(struct trace_buffer *buffer,
912 unsigned int trace_ctx,
913 int skip, struct pt_regs *regs);
914 static inline void ftrace_trace_stack(struct trace_array *tr,
915 struct trace_buffer *buffer,
916 unsigned int trace_ctx,
917 int skip, struct pt_regs *regs);
920 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
921 unsigned int trace_ctx,
922 int skip, struct pt_regs *regs)
925 static inline void ftrace_trace_stack(struct trace_array *tr,
926 struct trace_buffer *buffer,
927 unsigned long trace_ctx,
928 int skip, struct pt_regs *regs)
934 static __always_inline void
935 trace_event_setup(struct ring_buffer_event *event,
936 int type, unsigned int trace_ctx)
938 struct trace_entry *ent = ring_buffer_event_data(event);
940 tracing_generic_entry_update(ent, type, trace_ctx);
943 static __always_inline struct ring_buffer_event *
944 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
947 unsigned int trace_ctx)
949 struct ring_buffer_event *event;
951 event = ring_buffer_lock_reserve(buffer, len);
953 trace_event_setup(event, type, trace_ctx);
958 void tracer_tracing_on(struct trace_array *tr)
960 if (tr->array_buffer.buffer)
961 ring_buffer_record_on(tr->array_buffer.buffer);
963 * This flag is looked at when buffers haven't been allocated
964 * yet, or by some tracers (like irqsoff), that just want to
965 * know if the ring buffer has been disabled, but it can handle
966 * races of where it gets disabled but we still do a record.
967 * As the check is in the fast path of the tracers, it is more
968 * important to be fast than accurate.
970 tr->buffer_disabled = 0;
971 /* Make the flag seen by readers */
976 * tracing_on - enable tracing buffers
978 * This function enables tracing buffers that may have been
979 * disabled with tracing_off.
981 void tracing_on(void)
983 tracer_tracing_on(&global_trace);
985 EXPORT_SYMBOL_GPL(tracing_on);
988 static __always_inline void
989 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
991 __this_cpu_write(trace_taskinfo_save, true);
993 /* If this is the temp buffer, we need to commit fully */
994 if (this_cpu_read(trace_buffered_event) == event) {
995 /* Length is in event->array[0] */
996 ring_buffer_write(buffer, event->array[0], &event->array[1]);
997 /* Release the temp buffer */
998 this_cpu_dec(trace_buffered_event_cnt);
999 /* ring_buffer_unlock_commit() enables preemption */
1000 preempt_enable_notrace();
1002 ring_buffer_unlock_commit(buffer);
1006 * __trace_puts - write a constant string into the trace buffer.
1007 * @ip: The address of the caller
1008 * @str: The constant string to write
1009 * @size: The size of the string.
1011 int __trace_puts(unsigned long ip, const char *str, int size)
1013 struct ring_buffer_event *event;
1014 struct trace_buffer *buffer;
1015 struct print_entry *entry;
1016 unsigned int trace_ctx;
1019 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1022 if (unlikely(tracing_selftest_running || tracing_disabled))
1025 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1027 trace_ctx = tracing_gen_ctx();
1028 buffer = global_trace.array_buffer.buffer;
1029 ring_buffer_nest_start(buffer);
1030 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1037 entry = ring_buffer_event_data(event);
1040 memcpy(&entry->buf, str, size);
1042 /* Add a newline if necessary */
1043 if (entry->buf[size - 1] != '\n') {
1044 entry->buf[size] = '\n';
1045 entry->buf[size + 1] = '\0';
1047 entry->buf[size] = '\0';
1049 __buffer_unlock_commit(buffer, event);
1050 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1052 ring_buffer_nest_end(buffer);
1055 EXPORT_SYMBOL_GPL(__trace_puts);
1058 * __trace_bputs - write the pointer to a constant string into trace buffer
1059 * @ip: The address of the caller
1060 * @str: The constant string to write to the buffer to
1062 int __trace_bputs(unsigned long ip, const char *str)
1064 struct ring_buffer_event *event;
1065 struct trace_buffer *buffer;
1066 struct bputs_entry *entry;
1067 unsigned int trace_ctx;
1068 int size = sizeof(struct bputs_entry);
1071 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1074 if (unlikely(tracing_selftest_running || tracing_disabled))
1077 trace_ctx = tracing_gen_ctx();
1078 buffer = global_trace.array_buffer.buffer;
1080 ring_buffer_nest_start(buffer);
1081 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1086 entry = ring_buffer_event_data(event);
1090 __buffer_unlock_commit(buffer, event);
1091 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1095 ring_buffer_nest_end(buffer);
1098 EXPORT_SYMBOL_GPL(__trace_bputs);
1100 #ifdef CONFIG_TRACER_SNAPSHOT
1101 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1104 struct tracer *tracer = tr->current_trace;
1105 unsigned long flags;
1108 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1109 internal_trace_puts("*** snapshot is being ignored ***\n");
1113 if (!tr->allocated_snapshot) {
1114 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
1115 internal_trace_puts("*** stopping trace here! ***\n");
1120 /* Note, snapshot can not be used when the tracer uses it */
1121 if (tracer->use_max_tr) {
1122 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
1123 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
1127 local_irq_save(flags);
1128 update_max_tr(tr, current, smp_processor_id(), cond_data);
1129 local_irq_restore(flags);
1132 void tracing_snapshot_instance(struct trace_array *tr)
1134 tracing_snapshot_instance_cond(tr, NULL);
1138 * tracing_snapshot - take a snapshot of the current buffer.
1140 * This causes a swap between the snapshot buffer and the current live
1141 * tracing buffer. You can use this to take snapshots of the live
1142 * trace when some condition is triggered, but continue to trace.
1144 * Note, make sure to allocate the snapshot with either
1145 * a tracing_snapshot_alloc(), or by doing it manually
1146 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
1148 * If the snapshot buffer is not allocated, it will stop tracing.
1149 * Basically making a permanent snapshot.
1151 void tracing_snapshot(void)
1153 struct trace_array *tr = &global_trace;
1155 tracing_snapshot_instance(tr);
1157 EXPORT_SYMBOL_GPL(tracing_snapshot);
1160 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1161 * @tr: The tracing instance to snapshot
1162 * @cond_data: The data to be tested conditionally, and possibly saved
1164 * This is the same as tracing_snapshot() except that the snapshot is
1165 * conditional - the snapshot will only happen if the
1166 * cond_snapshot.update() implementation receiving the cond_data
1167 * returns true, which means that the trace array's cond_snapshot
1168 * update() operation used the cond_data to determine whether the
1169 * snapshot should be taken, and if it was, presumably saved it along
1170 * with the snapshot.
1172 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1174 tracing_snapshot_instance_cond(tr, cond_data);
1176 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1179 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1180 * @tr: The tracing instance
1182 * When the user enables a conditional snapshot using
1183 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1184 * with the snapshot. This accessor is used to retrieve it.
1186 * Should not be called from cond_snapshot.update(), since it takes
1187 * the tr->max_lock lock, which the code calling
1188 * cond_snapshot.update() has already done.
1190 * Returns the cond_data associated with the trace array's snapshot.
1192 void *tracing_cond_snapshot_data(struct trace_array *tr)
1194 void *cond_data = NULL;
1196 local_irq_disable();
1197 arch_spin_lock(&tr->max_lock);
1199 if (tr->cond_snapshot)
1200 cond_data = tr->cond_snapshot->cond_data;
1202 arch_spin_unlock(&tr->max_lock);
1207 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1209 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1210 struct array_buffer *size_buf, int cpu_id);
1211 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1213 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1217 if (!tr->allocated_snapshot) {
1219 /* allocate spare buffer */
1220 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1221 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1225 tr->allocated_snapshot = true;
1231 static void free_snapshot(struct trace_array *tr)
1234 * We don't free the ring buffer. instead, resize it because
1235 * The max_tr ring buffer has some state (e.g. ring->clock) and
1236 * we want preserve it.
1238 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1239 set_buffer_entries(&tr->max_buffer, 1);
1240 tracing_reset_online_cpus(&tr->max_buffer);
1241 tr->allocated_snapshot = false;
1245 * tracing_alloc_snapshot - allocate snapshot buffer.
1247 * This only allocates the snapshot buffer if it isn't already
1248 * allocated - it doesn't also take a snapshot.
1250 * This is meant to be used in cases where the snapshot buffer needs
1251 * to be set up for events that can't sleep but need to be able to
1252 * trigger a snapshot.
1254 int tracing_alloc_snapshot(void)
1256 struct trace_array *tr = &global_trace;
1259 ret = tracing_alloc_snapshot_instance(tr);
1264 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1267 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1269 * This is similar to tracing_snapshot(), but it will allocate the
1270 * snapshot buffer if it isn't already allocated. Use this only
1271 * where it is safe to sleep, as the allocation may sleep.
1273 * This causes a swap between the snapshot buffer and the current live
1274 * tracing buffer. You can use this to take snapshots of the live
1275 * trace when some condition is triggered, but continue to trace.
1277 void tracing_snapshot_alloc(void)
1281 ret = tracing_alloc_snapshot();
1287 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1290 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1291 * @tr: The tracing instance
1292 * @cond_data: User data to associate with the snapshot
1293 * @update: Implementation of the cond_snapshot update function
1295 * Check whether the conditional snapshot for the given instance has
1296 * already been enabled, or if the current tracer is already using a
1297 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1298 * save the cond_data and update function inside.
1300 * Returns 0 if successful, error otherwise.
1302 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1303 cond_update_fn_t update)
1305 struct cond_snapshot *cond_snapshot;
1308 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1312 cond_snapshot->cond_data = cond_data;
1313 cond_snapshot->update = update;
1315 mutex_lock(&trace_types_lock);
1317 ret = tracing_alloc_snapshot_instance(tr);
1321 if (tr->current_trace->use_max_tr) {
1327 * The cond_snapshot can only change to NULL without the
1328 * trace_types_lock. We don't care if we race with it going
1329 * to NULL, but we want to make sure that it's not set to
1330 * something other than NULL when we get here, which we can
1331 * do safely with only holding the trace_types_lock and not
1332 * having to take the max_lock.
1334 if (tr->cond_snapshot) {
1339 local_irq_disable();
1340 arch_spin_lock(&tr->max_lock);
1341 tr->cond_snapshot = cond_snapshot;
1342 arch_spin_unlock(&tr->max_lock);
1345 mutex_unlock(&trace_types_lock);
1350 mutex_unlock(&trace_types_lock);
1351 kfree(cond_snapshot);
1354 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1357 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1358 * @tr: The tracing instance
1360 * Check whether the conditional snapshot for the given instance is
1361 * enabled; if so, free the cond_snapshot associated with it,
1362 * otherwise return -EINVAL.
1364 * Returns 0 if successful, error otherwise.
1366 int tracing_snapshot_cond_disable(struct trace_array *tr)
1370 local_irq_disable();
1371 arch_spin_lock(&tr->max_lock);
1373 if (!tr->cond_snapshot)
1376 kfree(tr->cond_snapshot);
1377 tr->cond_snapshot = NULL;
1380 arch_spin_unlock(&tr->max_lock);
1385 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1387 void tracing_snapshot(void)
1389 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1391 EXPORT_SYMBOL_GPL(tracing_snapshot);
1392 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1394 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1396 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1397 int tracing_alloc_snapshot(void)
1399 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1402 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1403 void tracing_snapshot_alloc(void)
1408 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1409 void *tracing_cond_snapshot_data(struct trace_array *tr)
1413 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1414 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1418 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1419 int tracing_snapshot_cond_disable(struct trace_array *tr)
1423 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1424 #endif /* CONFIG_TRACER_SNAPSHOT */
1426 void tracer_tracing_off(struct trace_array *tr)
1428 if (tr->array_buffer.buffer)
1429 ring_buffer_record_off(tr->array_buffer.buffer);
1431 * This flag is looked at when buffers haven't been allocated
1432 * yet, or by some tracers (like irqsoff), that just want to
1433 * know if the ring buffer has been disabled, but it can handle
1434 * races of where it gets disabled but we still do a record.
1435 * As the check is in the fast path of the tracers, it is more
1436 * important to be fast than accurate.
1438 tr->buffer_disabled = 1;
1439 /* Make the flag seen by readers */
1444 * tracing_off - turn off tracing buffers
1446 * This function stops the tracing buffers from recording data.
1447 * It does not disable any overhead the tracers themselves may
1448 * be causing. This function simply causes all recording to
1449 * the ring buffers to fail.
1451 void tracing_off(void)
1453 tracer_tracing_off(&global_trace);
1455 EXPORT_SYMBOL_GPL(tracing_off);
1457 void disable_trace_on_warning(void)
1459 if (__disable_trace_on_warning) {
1460 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1461 "Disabling tracing due to warning\n");
1467 * tracer_tracing_is_on - show real state of ring buffer enabled
1468 * @tr : the trace array to know if ring buffer is enabled
1470 * Shows real state of the ring buffer if it is enabled or not.
1472 bool tracer_tracing_is_on(struct trace_array *tr)
1474 if (tr->array_buffer.buffer)
1475 return ring_buffer_record_is_on(tr->array_buffer.buffer);
1476 return !tr->buffer_disabled;
1480 * tracing_is_on - show state of ring buffers enabled
1482 int tracing_is_on(void)
1484 return tracer_tracing_is_on(&global_trace);
1486 EXPORT_SYMBOL_GPL(tracing_is_on);
1488 static int __init set_buf_size(char *str)
1490 unsigned long buf_size;
1494 buf_size = memparse(str, &str);
1496 * nr_entries can not be zero and the startup
1497 * tests require some buffer space. Therefore
1498 * ensure we have at least 4096 bytes of buffer.
1500 trace_buf_size = max(4096UL, buf_size);
1503 __setup("trace_buf_size=", set_buf_size);
1505 static int __init set_tracing_thresh(char *str)
1507 unsigned long threshold;
1512 ret = kstrtoul(str, 0, &threshold);
1515 tracing_thresh = threshold * 1000;
1518 __setup("tracing_thresh=", set_tracing_thresh);
1520 unsigned long nsecs_to_usecs(unsigned long nsecs)
1522 return nsecs / 1000;
1526 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1527 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1528 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1529 * of strings in the order that the evals (enum) were defined.
1534 /* These must match the bit positions in trace_iterator_flags */
1535 static const char *trace_options[] = {
1543 int in_ns; /* is this clock in nanoseconds? */
1544 } trace_clocks[] = {
1545 { trace_clock_local, "local", 1 },
1546 { trace_clock_global, "global", 1 },
1547 { trace_clock_counter, "counter", 0 },
1548 { trace_clock_jiffies, "uptime", 0 },
1549 { trace_clock, "perf", 1 },
1550 { ktime_get_mono_fast_ns, "mono", 1 },
1551 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1552 { ktime_get_boot_fast_ns, "boot", 1 },
1553 { ktime_get_tai_fast_ns, "tai", 1 },
1557 bool trace_clock_in_ns(struct trace_array *tr)
1559 if (trace_clocks[tr->clock_id].in_ns)
1566 * trace_parser_get_init - gets the buffer for trace parser
1568 int trace_parser_get_init(struct trace_parser *parser, int size)
1570 memset(parser, 0, sizeof(*parser));
1572 parser->buffer = kmalloc(size, GFP_KERNEL);
1573 if (!parser->buffer)
1576 parser->size = size;
1581 * trace_parser_put - frees the buffer for trace parser
1583 void trace_parser_put(struct trace_parser *parser)
1585 kfree(parser->buffer);
1586 parser->buffer = NULL;
1590 * trace_get_user - reads the user input string separated by space
1591 * (matched by isspace(ch))
1593 * For each string found the 'struct trace_parser' is updated,
1594 * and the function returns.
1596 * Returns number of bytes read.
1598 * See kernel/trace/trace.h for 'struct trace_parser' details.
1600 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1601 size_t cnt, loff_t *ppos)
1608 trace_parser_clear(parser);
1610 ret = get_user(ch, ubuf++);
1618 * The parser is not finished with the last write,
1619 * continue reading the user input without skipping spaces.
1621 if (!parser->cont) {
1622 /* skip white space */
1623 while (cnt && isspace(ch)) {
1624 ret = get_user(ch, ubuf++);
1633 /* only spaces were written */
1634 if (isspace(ch) || !ch) {
1641 /* read the non-space input */
1642 while (cnt && !isspace(ch) && ch) {
1643 if (parser->idx < parser->size - 1)
1644 parser->buffer[parser->idx++] = ch;
1649 ret = get_user(ch, ubuf++);
1656 /* We either got finished input or we have to wait for another call. */
1657 if (isspace(ch) || !ch) {
1658 parser->buffer[parser->idx] = 0;
1659 parser->cont = false;
1660 } else if (parser->idx < parser->size - 1) {
1661 parser->cont = true;
1662 parser->buffer[parser->idx++] = ch;
1663 /* Make sure the parsed string always terminates with '\0'. */
1664 parser->buffer[parser->idx] = 0;
1677 /* TODO add a seq_buf_to_buffer() */
1678 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1682 if (trace_seq_used(s) <= s->seq.readpos)
1685 len = trace_seq_used(s) - s->seq.readpos;
1688 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1690 s->seq.readpos += cnt;
1694 unsigned long __read_mostly tracing_thresh;
1695 static const struct file_operations tracing_max_lat_fops;
1697 #ifdef LATENCY_FS_NOTIFY
1699 static struct workqueue_struct *fsnotify_wq;
1701 static void latency_fsnotify_workfn(struct work_struct *work)
1703 struct trace_array *tr = container_of(work, struct trace_array,
1705 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1708 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1710 struct trace_array *tr = container_of(iwork, struct trace_array,
1712 queue_work(fsnotify_wq, &tr->fsnotify_work);
1715 static void trace_create_maxlat_file(struct trace_array *tr,
1716 struct dentry *d_tracer)
1718 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1719 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1720 tr->d_max_latency = trace_create_file("tracing_max_latency",
1722 d_tracer, &tr->max_latency,
1723 &tracing_max_lat_fops);
1726 __init static int latency_fsnotify_init(void)
1728 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1729 WQ_UNBOUND | WQ_HIGHPRI, 0);
1731 pr_err("Unable to allocate tr_max_lat_wq\n");
1737 late_initcall_sync(latency_fsnotify_init);
1739 void latency_fsnotify(struct trace_array *tr)
1744 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1745 * possible that we are called from __schedule() or do_idle(), which
1746 * could cause a deadlock.
1748 irq_work_queue(&tr->fsnotify_irqwork);
1751 #elif defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER) \
1752 || defined(CONFIG_OSNOISE_TRACER)
1754 #define trace_create_maxlat_file(tr, d_tracer) \
1755 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1756 d_tracer, &tr->max_latency, &tracing_max_lat_fops)
1759 #define trace_create_maxlat_file(tr, d_tracer) do { } while (0)
1762 #ifdef CONFIG_TRACER_MAX_TRACE
1764 * Copy the new maximum trace into the separate maximum-trace
1765 * structure. (this way the maximum trace is permanently saved,
1766 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1769 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1771 struct array_buffer *trace_buf = &tr->array_buffer;
1772 struct array_buffer *max_buf = &tr->max_buffer;
1773 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1774 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1777 max_buf->time_start = data->preempt_timestamp;
1779 max_data->saved_latency = tr->max_latency;
1780 max_data->critical_start = data->critical_start;
1781 max_data->critical_end = data->critical_end;
1783 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1784 max_data->pid = tsk->pid;
1786 * If tsk == current, then use current_uid(), as that does not use
1787 * RCU. The irq tracer can be called out of RCU scope.
1790 max_data->uid = current_uid();
1792 max_data->uid = task_uid(tsk);
1794 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1795 max_data->policy = tsk->policy;
1796 max_data->rt_priority = tsk->rt_priority;
1798 /* record this tasks comm */
1799 tracing_record_cmdline(tsk);
1800 latency_fsnotify(tr);
1804 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1806 * @tsk: the task with the latency
1807 * @cpu: The cpu that initiated the trace.
1808 * @cond_data: User data associated with a conditional snapshot
1810 * Flip the buffers between the @tr and the max_tr and record information
1811 * about which task was the cause of this latency.
1814 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1820 WARN_ON_ONCE(!irqs_disabled());
1822 if (!tr->allocated_snapshot) {
1823 /* Only the nop tracer should hit this when disabling */
1824 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1828 arch_spin_lock(&tr->max_lock);
1830 /* Inherit the recordable setting from array_buffer */
1831 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1832 ring_buffer_record_on(tr->max_buffer.buffer);
1834 ring_buffer_record_off(tr->max_buffer.buffer);
1836 #ifdef CONFIG_TRACER_SNAPSHOT
1837 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data))
1840 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1842 __update_max_tr(tr, tsk, cpu);
1845 arch_spin_unlock(&tr->max_lock);
1849 * update_max_tr_single - only copy one trace over, and reset the rest
1851 * @tsk: task with the latency
1852 * @cpu: the cpu of the buffer to copy.
1854 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1857 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1864 WARN_ON_ONCE(!irqs_disabled());
1865 if (!tr->allocated_snapshot) {
1866 /* Only the nop tracer should hit this when disabling */
1867 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1871 arch_spin_lock(&tr->max_lock);
1873 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1875 if (ret == -EBUSY) {
1877 * We failed to swap the buffer due to a commit taking
1878 * place on this CPU. We fail to record, but we reset
1879 * the max trace buffer (no one writes directly to it)
1880 * and flag that it failed.
1882 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1883 "Failed to swap buffers due to commit in progress\n");
1886 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1888 __update_max_tr(tr, tsk, cpu);
1889 arch_spin_unlock(&tr->max_lock);
1891 #endif /* CONFIG_TRACER_MAX_TRACE */
1893 static int wait_on_pipe(struct trace_iterator *iter, int full)
1895 /* Iterators are static, they should be filled or empty */
1896 if (trace_buffer_iter(iter, iter->cpu_file))
1899 return ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file,
1903 #ifdef CONFIG_FTRACE_STARTUP_TEST
1904 static bool selftests_can_run;
1906 struct trace_selftests {
1907 struct list_head list;
1908 struct tracer *type;
1911 static LIST_HEAD(postponed_selftests);
1913 static int save_selftest(struct tracer *type)
1915 struct trace_selftests *selftest;
1917 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1921 selftest->type = type;
1922 list_add(&selftest->list, &postponed_selftests);
1926 static int run_tracer_selftest(struct tracer *type)
1928 struct trace_array *tr = &global_trace;
1929 struct tracer *saved_tracer = tr->current_trace;
1932 if (!type->selftest || tracing_selftest_disabled)
1936 * If a tracer registers early in boot up (before scheduling is
1937 * initialized and such), then do not run its selftests yet.
1938 * Instead, run it a little later in the boot process.
1940 if (!selftests_can_run)
1941 return save_selftest(type);
1943 if (!tracing_is_on()) {
1944 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
1950 * Run a selftest on this tracer.
1951 * Here we reset the trace buffer, and set the current
1952 * tracer to be this tracer. The tracer can then run some
1953 * internal tracing to verify that everything is in order.
1954 * If we fail, we do not register this tracer.
1956 tracing_reset_online_cpus(&tr->array_buffer);
1958 tr->current_trace = type;
1960 #ifdef CONFIG_TRACER_MAX_TRACE
1961 if (type->use_max_tr) {
1962 /* If we expanded the buffers, make sure the max is expanded too */
1963 if (ring_buffer_expanded)
1964 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1965 RING_BUFFER_ALL_CPUS);
1966 tr->allocated_snapshot = true;
1970 /* the test is responsible for initializing and enabling */
1971 pr_info("Testing tracer %s: ", type->name);
1972 ret = type->selftest(type, tr);
1973 /* the test is responsible for resetting too */
1974 tr->current_trace = saved_tracer;
1976 printk(KERN_CONT "FAILED!\n");
1977 /* Add the warning after printing 'FAILED' */
1981 /* Only reset on passing, to avoid touching corrupted buffers */
1982 tracing_reset_online_cpus(&tr->array_buffer);
1984 #ifdef CONFIG_TRACER_MAX_TRACE
1985 if (type->use_max_tr) {
1986 tr->allocated_snapshot = false;
1988 /* Shrink the max buffer again */
1989 if (ring_buffer_expanded)
1990 ring_buffer_resize(tr->max_buffer.buffer, 1,
1991 RING_BUFFER_ALL_CPUS);
1995 printk(KERN_CONT "PASSED\n");
1999 static __init int init_trace_selftests(void)
2001 struct trace_selftests *p, *n;
2002 struct tracer *t, **last;
2005 selftests_can_run = true;
2007 mutex_lock(&trace_types_lock);
2009 if (list_empty(&postponed_selftests))
2012 pr_info("Running postponed tracer tests:\n");
2014 tracing_selftest_running = true;
2015 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2016 /* This loop can take minutes when sanitizers are enabled, so
2017 * lets make sure we allow RCU processing.
2020 ret = run_tracer_selftest(p->type);
2021 /* If the test fails, then warn and remove from available_tracers */
2023 WARN(1, "tracer: %s failed selftest, disabling\n",
2025 last = &trace_types;
2026 for (t = trace_types; t; t = t->next) {
2037 tracing_selftest_running = false;
2040 mutex_unlock(&trace_types_lock);
2044 core_initcall(init_trace_selftests);
2046 static inline int run_tracer_selftest(struct tracer *type)
2050 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2052 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2054 static void __init apply_trace_boot_options(void);
2057 * register_tracer - register a tracer with the ftrace system.
2058 * @type: the plugin for the tracer
2060 * Register a new plugin tracer.
2062 int __init register_tracer(struct tracer *type)
2068 pr_info("Tracer must have a name\n");
2072 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2073 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2077 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2078 pr_warn("Can not register tracer %s due to lockdown\n",
2083 mutex_lock(&trace_types_lock);
2085 tracing_selftest_running = true;
2087 for (t = trace_types; t; t = t->next) {
2088 if (strcmp(type->name, t->name) == 0) {
2090 pr_info("Tracer %s already registered\n",
2097 if (!type->set_flag)
2098 type->set_flag = &dummy_set_flag;
2100 /*allocate a dummy tracer_flags*/
2101 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2106 type->flags->val = 0;
2107 type->flags->opts = dummy_tracer_opt;
2109 if (!type->flags->opts)
2110 type->flags->opts = dummy_tracer_opt;
2112 /* store the tracer for __set_tracer_option */
2113 type->flags->trace = type;
2115 ret = run_tracer_selftest(type);
2119 type->next = trace_types;
2121 add_tracer_options(&global_trace, type);
2124 tracing_selftest_running = false;
2125 mutex_unlock(&trace_types_lock);
2127 if (ret || !default_bootup_tracer)
2130 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2133 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2134 /* Do we want this tracer to start on bootup? */
2135 tracing_set_tracer(&global_trace, type->name);
2136 default_bootup_tracer = NULL;
2138 apply_trace_boot_options();
2140 /* disable other selftests, since this will break it. */
2141 disable_tracing_selftest("running a tracer");
2147 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2149 struct trace_buffer *buffer = buf->buffer;
2154 ring_buffer_record_disable(buffer);
2156 /* Make sure all commits have finished */
2158 ring_buffer_reset_cpu(buffer, cpu);
2160 ring_buffer_record_enable(buffer);
2163 void tracing_reset_online_cpus(struct array_buffer *buf)
2165 struct trace_buffer *buffer = buf->buffer;
2170 ring_buffer_record_disable(buffer);
2172 /* Make sure all commits have finished */
2175 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2177 ring_buffer_reset_online_cpus(buffer);
2179 ring_buffer_record_enable(buffer);
2182 /* Must have trace_types_lock held */
2183 void tracing_reset_all_online_cpus_unlocked(void)
2185 struct trace_array *tr;
2187 lockdep_assert_held(&trace_types_lock);
2189 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2190 if (!tr->clear_trace)
2192 tr->clear_trace = false;
2193 tracing_reset_online_cpus(&tr->array_buffer);
2194 #ifdef CONFIG_TRACER_MAX_TRACE
2195 tracing_reset_online_cpus(&tr->max_buffer);
2200 void tracing_reset_all_online_cpus(void)
2202 mutex_lock(&trace_types_lock);
2203 tracing_reset_all_online_cpus_unlocked();
2204 mutex_unlock(&trace_types_lock);
2208 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2209 * is the tgid last observed corresponding to pid=i.
2211 static int *tgid_map;
2213 /* The maximum valid index into tgid_map. */
2214 static size_t tgid_map_max;
2216 #define SAVED_CMDLINES_DEFAULT 128
2217 #define NO_CMDLINE_MAP UINT_MAX
2219 * Preemption must be disabled before acquiring trace_cmdline_lock.
2220 * The various trace_arrays' max_lock must be acquired in a context
2221 * where interrupt is disabled.
2223 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2224 struct saved_cmdlines_buffer {
2225 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2226 unsigned *map_cmdline_to_pid;
2227 unsigned cmdline_num;
2229 char *saved_cmdlines;
2231 static struct saved_cmdlines_buffer *savedcmd;
2233 static inline char *get_saved_cmdlines(int idx)
2235 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2238 static inline void set_cmdline(int idx, const char *cmdline)
2240 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2243 static int allocate_cmdlines_buffer(unsigned int val,
2244 struct saved_cmdlines_buffer *s)
2246 s->map_cmdline_to_pid = kmalloc_array(val,
2247 sizeof(*s->map_cmdline_to_pid),
2249 if (!s->map_cmdline_to_pid)
2252 s->saved_cmdlines = kmalloc_array(TASK_COMM_LEN, val, GFP_KERNEL);
2253 if (!s->saved_cmdlines) {
2254 kfree(s->map_cmdline_to_pid);
2259 s->cmdline_num = val;
2260 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2261 sizeof(s->map_pid_to_cmdline));
2262 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2263 val * sizeof(*s->map_cmdline_to_pid));
2268 static int trace_create_savedcmd(void)
2272 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
2276 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
2286 int is_tracing_stopped(void)
2288 return global_trace.stop_count;
2292 * tracing_start - quick start of the tracer
2294 * If tracing is enabled but was stopped by tracing_stop,
2295 * this will start the tracer back up.
2297 void tracing_start(void)
2299 struct trace_buffer *buffer;
2300 unsigned long flags;
2302 if (tracing_disabled)
2305 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2306 if (--global_trace.stop_count) {
2307 if (global_trace.stop_count < 0) {
2308 /* Someone screwed up their debugging */
2310 global_trace.stop_count = 0;
2315 /* Prevent the buffers from switching */
2316 arch_spin_lock(&global_trace.max_lock);
2318 buffer = global_trace.array_buffer.buffer;
2320 ring_buffer_record_enable(buffer);
2322 #ifdef CONFIG_TRACER_MAX_TRACE
2323 buffer = global_trace.max_buffer.buffer;
2325 ring_buffer_record_enable(buffer);
2328 arch_spin_unlock(&global_trace.max_lock);
2331 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2334 static void tracing_start_tr(struct trace_array *tr)
2336 struct trace_buffer *buffer;
2337 unsigned long flags;
2339 if (tracing_disabled)
2342 /* If global, we need to also start the max tracer */
2343 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2344 return tracing_start();
2346 raw_spin_lock_irqsave(&tr->start_lock, flags);
2348 if (--tr->stop_count) {
2349 if (tr->stop_count < 0) {
2350 /* Someone screwed up their debugging */
2357 buffer = tr->array_buffer.buffer;
2359 ring_buffer_record_enable(buffer);
2362 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2366 * tracing_stop - quick stop of the tracer
2368 * Light weight way to stop tracing. Use in conjunction with
2371 void tracing_stop(void)
2373 struct trace_buffer *buffer;
2374 unsigned long flags;
2376 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
2377 if (global_trace.stop_count++)
2380 /* Prevent the buffers from switching */
2381 arch_spin_lock(&global_trace.max_lock);
2383 buffer = global_trace.array_buffer.buffer;
2385 ring_buffer_record_disable(buffer);
2387 #ifdef CONFIG_TRACER_MAX_TRACE
2388 buffer = global_trace.max_buffer.buffer;
2390 ring_buffer_record_disable(buffer);
2393 arch_spin_unlock(&global_trace.max_lock);
2396 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
2399 static void tracing_stop_tr(struct trace_array *tr)
2401 struct trace_buffer *buffer;
2402 unsigned long flags;
2404 /* If global, we need to also stop the max tracer */
2405 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
2406 return tracing_stop();
2408 raw_spin_lock_irqsave(&tr->start_lock, flags);
2409 if (tr->stop_count++)
2412 buffer = tr->array_buffer.buffer;
2414 ring_buffer_record_disable(buffer);
2417 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2420 static int trace_save_cmdline(struct task_struct *tsk)
2424 /* treat recording of idle task as a success */
2428 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2431 * It's not the end of the world if we don't get
2432 * the lock, but we also don't want to spin
2433 * nor do we want to disable interrupts,
2434 * so if we miss here, then better luck next time.
2436 * This is called within the scheduler and wake up, so interrupts
2437 * had better been disabled and run queue lock been held.
2439 lockdep_assert_preemption_disabled();
2440 if (!arch_spin_trylock(&trace_cmdline_lock))
2443 idx = savedcmd->map_pid_to_cmdline[tpid];
2444 if (idx == NO_CMDLINE_MAP) {
2445 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2447 savedcmd->map_pid_to_cmdline[tpid] = idx;
2448 savedcmd->cmdline_idx = idx;
2451 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2452 set_cmdline(idx, tsk->comm);
2454 arch_spin_unlock(&trace_cmdline_lock);
2459 static void __trace_find_cmdline(int pid, char comm[])
2465 strcpy(comm, "<idle>");
2469 if (WARN_ON_ONCE(pid < 0)) {
2470 strcpy(comm, "<XXX>");
2474 tpid = pid & (PID_MAX_DEFAULT - 1);
2475 map = savedcmd->map_pid_to_cmdline[tpid];
2476 if (map != NO_CMDLINE_MAP) {
2477 tpid = savedcmd->map_cmdline_to_pid[map];
2479 strlcpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2483 strcpy(comm, "<...>");
2486 void trace_find_cmdline(int pid, char comm[])
2489 arch_spin_lock(&trace_cmdline_lock);
2491 __trace_find_cmdline(pid, comm);
2493 arch_spin_unlock(&trace_cmdline_lock);
2497 static int *trace_find_tgid_ptr(int pid)
2500 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2501 * if we observe a non-NULL tgid_map then we also observe the correct
2504 int *map = smp_load_acquire(&tgid_map);
2506 if (unlikely(!map || pid > tgid_map_max))
2512 int trace_find_tgid(int pid)
2514 int *ptr = trace_find_tgid_ptr(pid);
2516 return ptr ? *ptr : 0;
2519 static int trace_save_tgid(struct task_struct *tsk)
2523 /* treat recording of idle task as a success */
2527 ptr = trace_find_tgid_ptr(tsk->pid);
2535 static bool tracing_record_taskinfo_skip(int flags)
2537 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2539 if (!__this_cpu_read(trace_taskinfo_save))
2545 * tracing_record_taskinfo - record the task info of a task
2547 * @task: task to record
2548 * @flags: TRACE_RECORD_CMDLINE for recording comm
2549 * TRACE_RECORD_TGID for recording tgid
2551 void tracing_record_taskinfo(struct task_struct *task, int flags)
2555 if (tracing_record_taskinfo_skip(flags))
2559 * Record as much task information as possible. If some fail, continue
2560 * to try to record the others.
2562 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2563 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2565 /* If recording any information failed, retry again soon. */
2569 __this_cpu_write(trace_taskinfo_save, false);
2573 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2575 * @prev: previous task during sched_switch
2576 * @next: next task during sched_switch
2577 * @flags: TRACE_RECORD_CMDLINE for recording comm
2578 * TRACE_RECORD_TGID for recording tgid
2580 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2581 struct task_struct *next, int flags)
2585 if (tracing_record_taskinfo_skip(flags))
2589 * Record as much task information as possible. If some fail, continue
2590 * to try to record the others.
2592 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2593 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2594 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2595 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2597 /* If recording any information failed, retry again soon. */
2601 __this_cpu_write(trace_taskinfo_save, false);
2604 /* Helpers to record a specific task information */
2605 void tracing_record_cmdline(struct task_struct *task)
2607 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2610 void tracing_record_tgid(struct task_struct *task)
2612 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2616 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2617 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2618 * simplifies those functions and keeps them in sync.
2620 enum print_line_t trace_handle_return(struct trace_seq *s)
2622 return trace_seq_has_overflowed(s) ?
2623 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2625 EXPORT_SYMBOL_GPL(trace_handle_return);
2627 static unsigned short migration_disable_value(void)
2629 #if defined(CONFIG_SMP)
2630 return current->migration_disabled;
2636 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2638 unsigned int trace_flags = irqs_status;
2641 pc = preempt_count();
2644 trace_flags |= TRACE_FLAG_NMI;
2645 if (pc & HARDIRQ_MASK)
2646 trace_flags |= TRACE_FLAG_HARDIRQ;
2647 if (in_serving_softirq())
2648 trace_flags |= TRACE_FLAG_SOFTIRQ;
2649 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2650 trace_flags |= TRACE_FLAG_BH_OFF;
2652 if (tif_need_resched())
2653 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2654 if (test_preempt_need_resched())
2655 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2656 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2657 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2660 struct ring_buffer_event *
2661 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2664 unsigned int trace_ctx)
2666 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2669 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2670 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2671 static int trace_buffered_event_ref;
2674 * trace_buffered_event_enable - enable buffering events
2676 * When events are being filtered, it is quicker to use a temporary
2677 * buffer to write the event data into if there's a likely chance
2678 * that it will not be committed. The discard of the ring buffer
2679 * is not as fast as committing, and is much slower than copying
2682 * When an event is to be filtered, allocate per cpu buffers to
2683 * write the event data into, and if the event is filtered and discarded
2684 * it is simply dropped, otherwise, the entire data is to be committed
2687 void trace_buffered_event_enable(void)
2689 struct ring_buffer_event *event;
2693 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2695 if (trace_buffered_event_ref++)
2698 for_each_tracing_cpu(cpu) {
2699 page = alloc_pages_node(cpu_to_node(cpu),
2700 GFP_KERNEL | __GFP_NORETRY, 0);
2704 event = page_address(page);
2705 memset(event, 0, sizeof(*event));
2707 per_cpu(trace_buffered_event, cpu) = event;
2710 if (cpu == smp_processor_id() &&
2711 __this_cpu_read(trace_buffered_event) !=
2712 per_cpu(trace_buffered_event, cpu))
2719 trace_buffered_event_disable();
2722 static void enable_trace_buffered_event(void *data)
2724 /* Probably not needed, but do it anyway */
2726 this_cpu_dec(trace_buffered_event_cnt);
2729 static void disable_trace_buffered_event(void *data)
2731 this_cpu_inc(trace_buffered_event_cnt);
2735 * trace_buffered_event_disable - disable buffering events
2737 * When a filter is removed, it is faster to not use the buffered
2738 * events, and to commit directly into the ring buffer. Free up
2739 * the temp buffers when there are no more users. This requires
2740 * special synchronization with current events.
2742 void trace_buffered_event_disable(void)
2746 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2748 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2751 if (--trace_buffered_event_ref)
2755 /* For each CPU, set the buffer as used. */
2756 smp_call_function_many(tracing_buffer_mask,
2757 disable_trace_buffered_event, NULL, 1);
2760 /* Wait for all current users to finish */
2763 for_each_tracing_cpu(cpu) {
2764 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2765 per_cpu(trace_buffered_event, cpu) = NULL;
2768 * Make sure trace_buffered_event is NULL before clearing
2769 * trace_buffered_event_cnt.
2774 /* Do the work on each cpu */
2775 smp_call_function_many(tracing_buffer_mask,
2776 enable_trace_buffered_event, NULL, 1);
2780 static struct trace_buffer *temp_buffer;
2782 struct ring_buffer_event *
2783 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2784 struct trace_event_file *trace_file,
2785 int type, unsigned long len,
2786 unsigned int trace_ctx)
2788 struct ring_buffer_event *entry;
2789 struct trace_array *tr = trace_file->tr;
2792 *current_rb = tr->array_buffer.buffer;
2794 if (!tr->no_filter_buffering_ref &&
2795 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2796 preempt_disable_notrace();
2798 * Filtering is on, so try to use the per cpu buffer first.
2799 * This buffer will simulate a ring_buffer_event,
2800 * where the type_len is zero and the array[0] will
2801 * hold the full length.
2802 * (see include/linux/ring-buffer.h for details on
2803 * how the ring_buffer_event is structured).
2805 * Using a temp buffer during filtering and copying it
2806 * on a matched filter is quicker than writing directly
2807 * into the ring buffer and then discarding it when
2808 * it doesn't match. That is because the discard
2809 * requires several atomic operations to get right.
2810 * Copying on match and doing nothing on a failed match
2811 * is still quicker than no copy on match, but having
2812 * to discard out of the ring buffer on a failed match.
2814 if ((entry = __this_cpu_read(trace_buffered_event))) {
2815 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2817 val = this_cpu_inc_return(trace_buffered_event_cnt);
2820 * Preemption is disabled, but interrupts and NMIs
2821 * can still come in now. If that happens after
2822 * the above increment, then it will have to go
2823 * back to the old method of allocating the event
2824 * on the ring buffer, and if the filter fails, it
2825 * will have to call ring_buffer_discard_commit()
2828 * Need to also check the unlikely case that the
2829 * length is bigger than the temp buffer size.
2830 * If that happens, then the reserve is pretty much
2831 * guaranteed to fail, as the ring buffer currently
2832 * only allows events less than a page. But that may
2833 * change in the future, so let the ring buffer reserve
2834 * handle the failure in that case.
2836 if (val == 1 && likely(len <= max_len)) {
2837 trace_event_setup(entry, type, trace_ctx);
2838 entry->array[0] = len;
2839 /* Return with preemption disabled */
2842 this_cpu_dec(trace_buffered_event_cnt);
2844 /* __trace_buffer_lock_reserve() disables preemption */
2845 preempt_enable_notrace();
2848 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2851 * If tracing is off, but we have triggers enabled
2852 * we still need to look at the event data. Use the temp_buffer
2853 * to store the trace event for the trigger to use. It's recursive
2854 * safe and will not be recorded anywhere.
2856 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2857 *current_rb = temp_buffer;
2858 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2863 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2865 static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2866 static DEFINE_MUTEX(tracepoint_printk_mutex);
2868 static void output_printk(struct trace_event_buffer *fbuffer)
2870 struct trace_event_call *event_call;
2871 struct trace_event_file *file;
2872 struct trace_event *event;
2873 unsigned long flags;
2874 struct trace_iterator *iter = tracepoint_print_iter;
2876 /* We should never get here if iter is NULL */
2877 if (WARN_ON_ONCE(!iter))
2880 event_call = fbuffer->trace_file->event_call;
2881 if (!event_call || !event_call->event.funcs ||
2882 !event_call->event.funcs->trace)
2885 file = fbuffer->trace_file;
2886 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2887 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2888 !filter_match_preds(file->filter, fbuffer->entry)))
2891 event = &fbuffer->trace_file->event_call->event;
2893 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2894 trace_seq_init(&iter->seq);
2895 iter->ent = fbuffer->entry;
2896 event_call->event.funcs->trace(iter, 0, event);
2897 trace_seq_putc(&iter->seq, 0);
2898 printk("%s", iter->seq.buffer);
2900 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2903 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2904 void *buffer, size_t *lenp,
2907 int save_tracepoint_printk;
2910 mutex_lock(&tracepoint_printk_mutex);
2911 save_tracepoint_printk = tracepoint_printk;
2913 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2916 * This will force exiting early, as tracepoint_printk
2917 * is always zero when tracepoint_printk_iter is not allocated
2919 if (!tracepoint_print_iter)
2920 tracepoint_printk = 0;
2922 if (save_tracepoint_printk == tracepoint_printk)
2925 if (tracepoint_printk)
2926 static_key_enable(&tracepoint_printk_key.key);
2928 static_key_disable(&tracepoint_printk_key.key);
2931 mutex_unlock(&tracepoint_printk_mutex);
2936 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
2938 enum event_trigger_type tt = ETT_NONE;
2939 struct trace_event_file *file = fbuffer->trace_file;
2941 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
2942 fbuffer->entry, &tt))
2945 if (static_key_false(&tracepoint_printk_key.key))
2946 output_printk(fbuffer);
2948 if (static_branch_unlikely(&trace_event_exports_enabled))
2949 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
2951 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
2952 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
2956 event_triggers_post_call(file, tt);
2959 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
2964 * trace_buffer_unlock_commit_regs()
2965 * trace_event_buffer_commit()
2966 * trace_event_raw_event_xxx()
2968 # define STACK_SKIP 3
2970 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
2971 struct trace_buffer *buffer,
2972 struct ring_buffer_event *event,
2973 unsigned int trace_ctx,
2974 struct pt_regs *regs)
2976 __buffer_unlock_commit(buffer, event);
2979 * If regs is not set, then skip the necessary functions.
2980 * Note, we can still get here via blktrace, wakeup tracer
2981 * and mmiotrace, but that's ok if they lose a function or
2982 * two. They are not that meaningful.
2984 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
2985 ftrace_trace_userstack(tr, buffer, trace_ctx);
2989 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
2992 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
2993 struct ring_buffer_event *event)
2995 __buffer_unlock_commit(buffer, event);
2999 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
3000 parent_ip, unsigned int trace_ctx)
3002 struct trace_event_call *call = &event_function;
3003 struct trace_buffer *buffer = tr->array_buffer.buffer;
3004 struct ring_buffer_event *event;
3005 struct ftrace_entry *entry;
3007 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
3011 entry = ring_buffer_event_data(event);
3013 entry->parent_ip = parent_ip;
3015 if (!call_filter_check_discard(call, entry, buffer, event)) {
3016 if (static_branch_unlikely(&trace_function_exports_enabled))
3017 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
3018 __buffer_unlock_commit(buffer, event);
3022 #ifdef CONFIG_STACKTRACE
3024 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
3025 #define FTRACE_KSTACK_NESTING 4
3027 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3029 struct ftrace_stack {
3030 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3034 struct ftrace_stacks {
3035 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3038 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3039 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3041 static void __ftrace_trace_stack(struct trace_buffer *buffer,
3042 unsigned int trace_ctx,
3043 int skip, struct pt_regs *regs)
3045 struct trace_event_call *call = &event_kernel_stack;
3046 struct ring_buffer_event *event;
3047 unsigned int size, nr_entries;
3048 struct ftrace_stack *fstack;
3049 struct stack_entry *entry;
3053 * Add one, for this function and the call to save_stack_trace()
3054 * If regs is set, then these functions will not be in the way.
3056 #ifndef CONFIG_UNWINDER_ORC
3061 preempt_disable_notrace();
3063 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3065 /* This should never happen. If it does, yell once and skip */
3066 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3070 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3071 * interrupt will either see the value pre increment or post
3072 * increment. If the interrupt happens pre increment it will have
3073 * restored the counter when it returns. We just need a barrier to
3074 * keep gcc from moving things around.
3078 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3079 size = ARRAY_SIZE(fstack->calls);
3082 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3085 nr_entries = stack_trace_save(fstack->calls, size, skip);
3088 size = nr_entries * sizeof(unsigned long);
3089 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3090 (sizeof(*entry) - sizeof(entry->caller)) + size,
3094 entry = ring_buffer_event_data(event);
3096 memcpy(&entry->caller, fstack->calls, size);
3097 entry->size = nr_entries;
3099 if (!call_filter_check_discard(call, entry, buffer, event))
3100 __buffer_unlock_commit(buffer, event);
3103 /* Again, don't let gcc optimize things here */
3105 __this_cpu_dec(ftrace_stack_reserve);
3106 preempt_enable_notrace();
3110 static inline void ftrace_trace_stack(struct trace_array *tr,
3111 struct trace_buffer *buffer,
3112 unsigned int trace_ctx,
3113 int skip, struct pt_regs *regs)
3115 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3118 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3121 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3124 struct trace_buffer *buffer = tr->array_buffer.buffer;
3126 if (rcu_is_watching()) {
3127 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3132 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3133 * but if the above rcu_is_watching() failed, then the NMI
3134 * triggered someplace critical, and ct_irq_enter() should
3135 * not be called from NMI.
3137 if (unlikely(in_nmi()))
3140 ct_irq_enter_irqson();
3141 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3142 ct_irq_exit_irqson();
3146 * trace_dump_stack - record a stack back trace in the trace buffer
3147 * @skip: Number of functions to skip (helper handlers)
3149 void trace_dump_stack(int skip)
3151 if (tracing_disabled || tracing_selftest_running)
3154 #ifndef CONFIG_UNWINDER_ORC
3155 /* Skip 1 to skip this function. */
3158 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3159 tracing_gen_ctx(), skip, NULL);
3161 EXPORT_SYMBOL_GPL(trace_dump_stack);
3163 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3164 static DEFINE_PER_CPU(int, user_stack_count);
3167 ftrace_trace_userstack(struct trace_array *tr,
3168 struct trace_buffer *buffer, unsigned int trace_ctx)
3170 struct trace_event_call *call = &event_user_stack;
3171 struct ring_buffer_event *event;
3172 struct userstack_entry *entry;
3174 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3178 * NMIs can not handle page faults, even with fix ups.
3179 * The save user stack can (and often does) fault.
3181 if (unlikely(in_nmi()))
3185 * prevent recursion, since the user stack tracing may
3186 * trigger other kernel events.
3189 if (__this_cpu_read(user_stack_count))
3192 __this_cpu_inc(user_stack_count);
3194 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3195 sizeof(*entry), trace_ctx);
3197 goto out_drop_count;
3198 entry = ring_buffer_event_data(event);
3200 entry->tgid = current->tgid;
3201 memset(&entry->caller, 0, sizeof(entry->caller));
3203 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3204 if (!call_filter_check_discard(call, entry, buffer, event))
3205 __buffer_unlock_commit(buffer, event);
3208 __this_cpu_dec(user_stack_count);
3212 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
3213 static void ftrace_trace_userstack(struct trace_array *tr,
3214 struct trace_buffer *buffer,
3215 unsigned int trace_ctx)
3218 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3220 #endif /* CONFIG_STACKTRACE */
3223 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3224 unsigned long long delta)
3226 entry->bottom_delta_ts = delta & U32_MAX;
3227 entry->top_delta_ts = (delta >> 32);
3230 void trace_last_func_repeats(struct trace_array *tr,
3231 struct trace_func_repeats *last_info,
3232 unsigned int trace_ctx)
3234 struct trace_buffer *buffer = tr->array_buffer.buffer;
3235 struct func_repeats_entry *entry;
3236 struct ring_buffer_event *event;
3239 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3240 sizeof(*entry), trace_ctx);
3244 delta = ring_buffer_event_time_stamp(buffer, event) -
3245 last_info->ts_last_call;
3247 entry = ring_buffer_event_data(event);
3248 entry->ip = last_info->ip;
3249 entry->parent_ip = last_info->parent_ip;
3250 entry->count = last_info->count;
3251 func_repeats_set_delta_ts(entry, delta);
3253 __buffer_unlock_commit(buffer, event);
3256 /* created for use with alloc_percpu */
3257 struct trace_buffer_struct {
3259 char buffer[4][TRACE_BUF_SIZE];
3262 static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3265 * This allows for lockless recording. If we're nested too deeply, then
3266 * this returns NULL.
3268 static char *get_trace_buf(void)
3270 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3272 if (!trace_percpu_buffer || buffer->nesting >= 4)
3277 /* Interrupts must see nesting incremented before we use the buffer */
3279 return &buffer->buffer[buffer->nesting - 1][0];
3282 static void put_trace_buf(void)
3284 /* Don't let the decrement of nesting leak before this */
3286 this_cpu_dec(trace_percpu_buffer->nesting);
3289 static int alloc_percpu_trace_buffer(void)
3291 struct trace_buffer_struct __percpu *buffers;
3293 if (trace_percpu_buffer)
3296 buffers = alloc_percpu(struct trace_buffer_struct);
3297 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3300 trace_percpu_buffer = buffers;
3304 static int buffers_allocated;
3306 void trace_printk_init_buffers(void)
3308 if (buffers_allocated)
3311 if (alloc_percpu_trace_buffer())
3314 /* trace_printk() is for debug use only. Don't use it in production. */
3317 pr_warn("**********************************************************\n");
3318 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3320 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3322 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3323 pr_warn("** unsafe for production use. **\n");
3325 pr_warn("** If you see this message and you are not debugging **\n");
3326 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3328 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3329 pr_warn("**********************************************************\n");
3331 /* Expand the buffers to set size */
3332 tracing_update_buffers();
3334 buffers_allocated = 1;
3337 * trace_printk_init_buffers() can be called by modules.
3338 * If that happens, then we need to start cmdline recording
3339 * directly here. If the global_trace.buffer is already
3340 * allocated here, then this was called by module code.
3342 if (global_trace.array_buffer.buffer)
3343 tracing_start_cmdline_record();
3345 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3347 void trace_printk_start_comm(void)
3349 /* Start tracing comms if trace printk is set */
3350 if (!buffers_allocated)
3352 tracing_start_cmdline_record();
3355 static void trace_printk_start_stop_comm(int enabled)
3357 if (!buffers_allocated)
3361 tracing_start_cmdline_record();
3363 tracing_stop_cmdline_record();
3367 * trace_vbprintk - write binary msg to tracing buffer
3368 * @ip: The address of the caller
3369 * @fmt: The string format to write to the buffer
3370 * @args: Arguments for @fmt
3372 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3374 struct trace_event_call *call = &event_bprint;
3375 struct ring_buffer_event *event;
3376 struct trace_buffer *buffer;
3377 struct trace_array *tr = &global_trace;
3378 struct bprint_entry *entry;
3379 unsigned int trace_ctx;
3383 if (unlikely(tracing_selftest_running || tracing_disabled))
3386 /* Don't pollute graph traces with trace_vprintk internals */
3387 pause_graph_tracing();
3389 trace_ctx = tracing_gen_ctx();
3390 preempt_disable_notrace();
3392 tbuffer = get_trace_buf();
3398 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3400 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3403 size = sizeof(*entry) + sizeof(u32) * len;
3404 buffer = tr->array_buffer.buffer;
3405 ring_buffer_nest_start(buffer);
3406 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3410 entry = ring_buffer_event_data(event);
3414 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3415 if (!call_filter_check_discard(call, entry, buffer, event)) {
3416 __buffer_unlock_commit(buffer, event);
3417 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3421 ring_buffer_nest_end(buffer);
3426 preempt_enable_notrace();
3427 unpause_graph_tracing();
3431 EXPORT_SYMBOL_GPL(trace_vbprintk);
3435 __trace_array_vprintk(struct trace_buffer *buffer,
3436 unsigned long ip, const char *fmt, va_list args)
3438 struct trace_event_call *call = &event_print;
3439 struct ring_buffer_event *event;
3441 struct print_entry *entry;
3442 unsigned int trace_ctx;
3445 if (tracing_disabled || tracing_selftest_running)
3448 /* Don't pollute graph traces with trace_vprintk internals */
3449 pause_graph_tracing();
3451 trace_ctx = tracing_gen_ctx();
3452 preempt_disable_notrace();
3455 tbuffer = get_trace_buf();
3461 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3463 size = sizeof(*entry) + len + 1;
3464 ring_buffer_nest_start(buffer);
3465 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3469 entry = ring_buffer_event_data(event);
3472 memcpy(&entry->buf, tbuffer, len + 1);
3473 if (!call_filter_check_discard(call, entry, buffer, event)) {
3474 __buffer_unlock_commit(buffer, event);
3475 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3479 ring_buffer_nest_end(buffer);
3483 preempt_enable_notrace();
3484 unpause_graph_tracing();
3490 int trace_array_vprintk(struct trace_array *tr,
3491 unsigned long ip, const char *fmt, va_list args)
3493 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3497 * trace_array_printk - Print a message to a specific instance
3498 * @tr: The instance trace_array descriptor
3499 * @ip: The instruction pointer that this is called from.
3500 * @fmt: The format to print (printf format)
3502 * If a subsystem sets up its own instance, they have the right to
3503 * printk strings into their tracing instance buffer using this
3504 * function. Note, this function will not write into the top level
3505 * buffer (use trace_printk() for that), as writing into the top level
3506 * buffer should only have events that can be individually disabled.
3507 * trace_printk() is only used for debugging a kernel, and should not
3508 * be ever incorporated in normal use.
3510 * trace_array_printk() can be used, as it will not add noise to the
3511 * top level tracing buffer.
3513 * Note, trace_array_init_printk() must be called on @tr before this
3517 int trace_array_printk(struct trace_array *tr,
3518 unsigned long ip, const char *fmt, ...)
3526 /* This is only allowed for created instances */
3527 if (tr == &global_trace)
3530 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3534 ret = trace_array_vprintk(tr, ip, fmt, ap);
3538 EXPORT_SYMBOL_GPL(trace_array_printk);
3541 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3542 * @tr: The trace array to initialize the buffers for
3544 * As trace_array_printk() only writes into instances, they are OK to
3545 * have in the kernel (unlike trace_printk()). This needs to be called
3546 * before trace_array_printk() can be used on a trace_array.
3548 int trace_array_init_printk(struct trace_array *tr)
3553 /* This is only allowed for created instances */
3554 if (tr == &global_trace)
3557 return alloc_percpu_trace_buffer();
3559 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3562 int trace_array_printk_buf(struct trace_buffer *buffer,
3563 unsigned long ip, const char *fmt, ...)
3568 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3572 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3578 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3580 return trace_array_vprintk(&global_trace, ip, fmt, args);
3582 EXPORT_SYMBOL_GPL(trace_vprintk);
3584 static void trace_iterator_increment(struct trace_iterator *iter)
3586 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3590 ring_buffer_iter_advance(buf_iter);
3593 static struct trace_entry *
3594 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3595 unsigned long *lost_events)
3597 struct ring_buffer_event *event;
3598 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3601 event = ring_buffer_iter_peek(buf_iter, ts);
3603 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3604 (unsigned long)-1 : 0;
3606 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3611 iter->ent_size = ring_buffer_event_length(event);
3612 return ring_buffer_event_data(event);
3618 static struct trace_entry *
3619 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3620 unsigned long *missing_events, u64 *ent_ts)
3622 struct trace_buffer *buffer = iter->array_buffer->buffer;
3623 struct trace_entry *ent, *next = NULL;
3624 unsigned long lost_events = 0, next_lost = 0;
3625 int cpu_file = iter->cpu_file;
3626 u64 next_ts = 0, ts;
3632 * If we are in a per_cpu trace file, don't bother by iterating over
3633 * all cpu and peek directly.
3635 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3636 if (ring_buffer_empty_cpu(buffer, cpu_file))
3638 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3640 *ent_cpu = cpu_file;
3645 for_each_tracing_cpu(cpu) {
3647 if (ring_buffer_empty_cpu(buffer, cpu))
3650 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3653 * Pick the entry with the smallest timestamp:
3655 if (ent && (!next || ts < next_ts)) {
3659 next_lost = lost_events;
3660 next_size = iter->ent_size;
3664 iter->ent_size = next_size;
3667 *ent_cpu = next_cpu;
3673 *missing_events = next_lost;
3678 #define STATIC_FMT_BUF_SIZE 128
3679 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3681 static char *trace_iter_expand_format(struct trace_iterator *iter)
3686 * iter->tr is NULL when used with tp_printk, which makes
3687 * this get called where it is not safe to call krealloc().
3689 if (!iter->tr || iter->fmt == static_fmt_buf)
3692 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3695 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3702 /* Returns true if the string is safe to dereference from an event */
3703 static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3706 unsigned long addr = (unsigned long)str;
3707 struct trace_event *trace_event;
3708 struct trace_event_call *event;
3710 /* Ignore strings with no length */
3714 /* OK if part of the event data */
3715 if ((addr >= (unsigned long)iter->ent) &&
3716 (addr < (unsigned long)iter->ent + iter->ent_size))
3719 /* OK if part of the temp seq buffer */
3720 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3721 (addr < (unsigned long)iter->tmp_seq.buffer + PAGE_SIZE))
3724 /* Core rodata can not be freed */
3725 if (is_kernel_rodata(addr))
3728 if (trace_is_tracepoint_string(str))
3732 * Now this could be a module event, referencing core module
3733 * data, which is OK.
3738 trace_event = ftrace_find_event(iter->ent->type);
3742 event = container_of(trace_event, struct trace_event_call, event);
3743 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3746 /* Would rather have rodata, but this will suffice */
3747 if (within_module_core(addr, event->module))
3753 static const char *show_buffer(struct trace_seq *s)
3755 struct seq_buf *seq = &s->seq;
3757 seq_buf_terminate(seq);
3762 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3764 static int test_can_verify_check(const char *fmt, ...)
3771 * The verifier is dependent on vsnprintf() modifies the va_list
3772 * passed to it, where it is sent as a reference. Some architectures
3773 * (like x86_32) passes it by value, which means that vsnprintf()
3774 * does not modify the va_list passed to it, and the verifier
3775 * would then need to be able to understand all the values that
3776 * vsnprintf can use. If it is passed by value, then the verifier
3780 vsnprintf(buf, 16, "%d", ap);
3781 ret = va_arg(ap, int);
3787 static void test_can_verify(void)
3789 if (!test_can_verify_check("%d %d", 0, 1)) {
3790 pr_info("trace event string verifier disabled\n");
3791 static_branch_inc(&trace_no_verify);
3796 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3797 * @iter: The iterator that holds the seq buffer and the event being printed
3798 * @fmt: The format used to print the event
3799 * @ap: The va_list holding the data to print from @fmt.
3801 * This writes the data into the @iter->seq buffer using the data from
3802 * @fmt and @ap. If the format has a %s, then the source of the string
3803 * is examined to make sure it is safe to print, otherwise it will
3804 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3807 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3810 const char *p = fmt;
3814 if (WARN_ON_ONCE(!fmt))
3817 if (static_branch_unlikely(&trace_no_verify))
3820 /* Don't bother checking when doing a ftrace_dump() */
3821 if (iter->fmt == static_fmt_buf)
3830 /* We only care about %s and variants */
3831 for (i = 0; p[i]; i++) {
3832 if (i + 1 >= iter->fmt_size) {
3834 * If we can't expand the copy buffer,
3837 if (!trace_iter_expand_format(iter))
3841 if (p[i] == '\\' && p[i+1]) {
3846 /* Need to test cases like %08.*s */
3847 for (j = 1; p[i+j]; j++) {
3848 if (isdigit(p[i+j]) ||
3851 if (p[i+j] == '*') {
3863 /* If no %s found then just print normally */
3867 /* Copy up to the %s, and print that */
3868 strncpy(iter->fmt, p, i);
3869 iter->fmt[i] = '\0';
3870 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3873 * If iter->seq is full, the above call no longer guarantees
3874 * that ap is in sync with fmt processing, and further calls
3875 * to va_arg() can return wrong positional arguments.
3877 * Ensure that ap is no longer used in this case.
3879 if (iter->seq.full) {
3885 len = va_arg(ap, int);
3887 /* The ap now points to the string data of the %s */
3888 str = va_arg(ap, const char *);
3891 * If you hit this warning, it is likely that the
3892 * trace event in question used %s on a string that
3893 * was saved at the time of the event, but may not be
3894 * around when the trace is read. Use __string(),
3895 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3896 * instead. See samples/trace_events/trace-events-sample.h
3899 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3900 "fmt: '%s' current_buffer: '%s'",
3901 fmt, show_buffer(&iter->seq))) {
3904 /* Try to safely read the string */
3906 if (len + 1 > iter->fmt_size)
3907 len = iter->fmt_size - 1;
3910 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3914 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3918 trace_seq_printf(&iter->seq, "(0x%px)", str);
3920 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3922 str = "[UNSAFE-MEMORY]";
3923 strcpy(iter->fmt, "%s");
3925 strncpy(iter->fmt, p + i, j + 1);
3926 iter->fmt[j+1] = '\0';
3929 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3931 trace_seq_printf(&iter->seq, iter->fmt, str);
3937 trace_seq_vprintf(&iter->seq, p, ap);
3940 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
3942 const char *p, *new_fmt;
3945 if (WARN_ON_ONCE(!fmt))
3948 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
3952 new_fmt = q = iter->fmt;
3954 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
3955 if (!trace_iter_expand_format(iter))
3958 q += iter->fmt - new_fmt;
3959 new_fmt = iter->fmt;
3964 /* Replace %p with %px */
3968 } else if (p[0] == 'p' && !isalnum(p[1])) {
3979 #define STATIC_TEMP_BUF_SIZE 128
3980 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
3982 /* Find the next real entry, without updating the iterator itself */
3983 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
3984 int *ent_cpu, u64 *ent_ts)
3986 /* __find_next_entry will reset ent_size */
3987 int ent_size = iter->ent_size;
3988 struct trace_entry *entry;
3991 * If called from ftrace_dump(), then the iter->temp buffer
3992 * will be the static_temp_buf and not created from kmalloc.
3993 * If the entry size is greater than the buffer, we can
3994 * not save it. Just return NULL in that case. This is only
3995 * used to add markers when two consecutive events' time
3996 * stamps have a large delta. See trace_print_lat_context()
3998 if (iter->temp == static_temp_buf &&
3999 STATIC_TEMP_BUF_SIZE < ent_size)
4003 * The __find_next_entry() may call peek_next_entry(), which may
4004 * call ring_buffer_peek() that may make the contents of iter->ent
4005 * undefined. Need to copy iter->ent now.
4007 if (iter->ent && iter->ent != iter->temp) {
4008 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4009 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4011 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4016 iter->temp_size = iter->ent_size;
4018 memcpy(iter->temp, iter->ent, iter->ent_size);
4019 iter->ent = iter->temp;
4021 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4022 /* Put back the original ent_size */
4023 iter->ent_size = ent_size;
4028 /* Find the next real entry, and increment the iterator to the next entry */
4029 void *trace_find_next_entry_inc(struct trace_iterator *iter)
4031 iter->ent = __find_next_entry(iter, &iter->cpu,
4032 &iter->lost_events, &iter->ts);
4035 trace_iterator_increment(iter);
4037 return iter->ent ? iter : NULL;
4040 static void trace_consume(struct trace_iterator *iter)
4042 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4043 &iter->lost_events);
4046 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4048 struct trace_iterator *iter = m->private;
4052 WARN_ON_ONCE(iter->leftover);
4056 /* can't go backwards */
4061 ent = trace_find_next_entry_inc(iter);
4065 while (ent && iter->idx < i)
4066 ent = trace_find_next_entry_inc(iter);
4073 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4075 struct ring_buffer_iter *buf_iter;
4076 unsigned long entries = 0;
4079 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4081 buf_iter = trace_buffer_iter(iter, cpu);
4085 ring_buffer_iter_reset(buf_iter);
4088 * We could have the case with the max latency tracers
4089 * that a reset never took place on a cpu. This is evident
4090 * by the timestamp being before the start of the buffer.
4092 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4093 if (ts >= iter->array_buffer->time_start)
4096 ring_buffer_iter_advance(buf_iter);
4099 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4103 * The current tracer is copied to avoid a global locking
4106 static void *s_start(struct seq_file *m, loff_t *pos)
4108 struct trace_iterator *iter = m->private;
4109 struct trace_array *tr = iter->tr;
4110 int cpu_file = iter->cpu_file;
4116 * copy the tracer to avoid using a global lock all around.
4117 * iter->trace is a copy of current_trace, the pointer to the
4118 * name may be used instead of a strcmp(), as iter->trace->name
4119 * will point to the same string as current_trace->name.
4121 mutex_lock(&trace_types_lock);
4122 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
4123 *iter->trace = *tr->current_trace;
4124 mutex_unlock(&trace_types_lock);
4126 #ifdef CONFIG_TRACER_MAX_TRACE
4127 if (iter->snapshot && iter->trace->use_max_tr)
4128 return ERR_PTR(-EBUSY);
4131 if (*pos != iter->pos) {
4136 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4137 for_each_tracing_cpu(cpu)
4138 tracing_iter_reset(iter, cpu);
4140 tracing_iter_reset(iter, cpu_file);
4143 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4148 * If we overflowed the seq_file before, then we want
4149 * to just reuse the trace_seq buffer again.
4155 p = s_next(m, p, &l);
4159 trace_event_read_lock();
4160 trace_access_lock(cpu_file);
4164 static void s_stop(struct seq_file *m, void *p)
4166 struct trace_iterator *iter = m->private;
4168 #ifdef CONFIG_TRACER_MAX_TRACE
4169 if (iter->snapshot && iter->trace->use_max_tr)
4173 trace_access_unlock(iter->cpu_file);
4174 trace_event_read_unlock();
4178 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4179 unsigned long *entries, int cpu)
4181 unsigned long count;
4183 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4185 * If this buffer has skipped entries, then we hold all
4186 * entries for the trace and we need to ignore the
4187 * ones before the time stamp.
4189 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4190 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4191 /* total is the same as the entries */
4195 ring_buffer_overrun_cpu(buf->buffer, cpu);
4200 get_total_entries(struct array_buffer *buf,
4201 unsigned long *total, unsigned long *entries)
4209 for_each_tracing_cpu(cpu) {
4210 get_total_entries_cpu(buf, &t, &e, cpu);
4216 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4218 unsigned long total, entries;
4223 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4228 unsigned long trace_total_entries(struct trace_array *tr)
4230 unsigned long total, entries;
4235 get_total_entries(&tr->array_buffer, &total, &entries);
4240 static void print_lat_help_header(struct seq_file *m)
4242 seq_puts(m, "# _------=> CPU# \n"
4243 "# / _-----=> irqs-off/BH-disabled\n"
4244 "# | / _----=> need-resched \n"
4245 "# || / _---=> hardirq/softirq \n"
4246 "# ||| / _--=> preempt-depth \n"
4247 "# |||| / _-=> migrate-disable \n"
4248 "# ||||| / delay \n"
4249 "# cmd pid |||||| time | caller \n"
4250 "# \\ / |||||| \\ | / \n");
4253 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4255 unsigned long total;
4256 unsigned long entries;
4258 get_total_entries(buf, &total, &entries);
4259 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4260 entries, total, num_online_cpus());
4264 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4267 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4269 print_event_info(buf, m);
4271 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4272 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4275 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4278 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4279 static const char space[] = " ";
4280 int prec = tgid ? 12 : 2;
4282 print_event_info(buf, m);
4284 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4285 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4286 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4287 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4288 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4289 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4290 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4291 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4295 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4297 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4298 struct array_buffer *buf = iter->array_buffer;
4299 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4300 struct tracer *type = iter->trace;
4301 unsigned long entries;
4302 unsigned long total;
4303 const char *name = type->name;
4305 get_total_entries(buf, &total, &entries);
4307 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4309 seq_puts(m, "# -----------------------------------"
4310 "---------------------------------\n");
4311 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4312 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4313 nsecs_to_usecs(data->saved_latency),
4317 preempt_model_none() ? "server" :
4318 preempt_model_voluntary() ? "desktop" :
4319 preempt_model_full() ? "preempt" :
4320 preempt_model_rt() ? "preempt_rt" :
4322 /* These are reserved for later use */
4325 seq_printf(m, " #P:%d)\n", num_online_cpus());
4329 seq_puts(m, "# -----------------\n");
4330 seq_printf(m, "# | task: %.16s-%d "
4331 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4332 data->comm, data->pid,
4333 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4334 data->policy, data->rt_priority);
4335 seq_puts(m, "# -----------------\n");
4337 if (data->critical_start) {
4338 seq_puts(m, "# => started at: ");
4339 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4340 trace_print_seq(m, &iter->seq);
4341 seq_puts(m, "\n# => ended at: ");
4342 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4343 trace_print_seq(m, &iter->seq);
4344 seq_puts(m, "\n#\n");
4350 static void test_cpu_buff_start(struct trace_iterator *iter)
4352 struct trace_seq *s = &iter->seq;
4353 struct trace_array *tr = iter->tr;
4355 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4358 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4361 if (cpumask_available(iter->started) &&
4362 cpumask_test_cpu(iter->cpu, iter->started))
4365 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4368 if (cpumask_available(iter->started))
4369 cpumask_set_cpu(iter->cpu, iter->started);
4371 /* Don't print started cpu buffer for the first entry of the trace */
4373 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4377 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4379 struct trace_array *tr = iter->tr;
4380 struct trace_seq *s = &iter->seq;
4381 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4382 struct trace_entry *entry;
4383 struct trace_event *event;
4387 test_cpu_buff_start(iter);
4389 event = ftrace_find_event(entry->type);
4391 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4392 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4393 trace_print_lat_context(iter);
4395 trace_print_context(iter);
4398 if (trace_seq_has_overflowed(s))
4399 return TRACE_TYPE_PARTIAL_LINE;
4402 return event->funcs->trace(iter, sym_flags, event);
4404 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4406 return trace_handle_return(s);
4409 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4411 struct trace_array *tr = iter->tr;
4412 struct trace_seq *s = &iter->seq;
4413 struct trace_entry *entry;
4414 struct trace_event *event;
4418 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4419 trace_seq_printf(s, "%d %d %llu ",
4420 entry->pid, iter->cpu, iter->ts);
4422 if (trace_seq_has_overflowed(s))
4423 return TRACE_TYPE_PARTIAL_LINE;
4425 event = ftrace_find_event(entry->type);
4427 return event->funcs->raw(iter, 0, event);
4429 trace_seq_printf(s, "%d ?\n", entry->type);
4431 return trace_handle_return(s);
4434 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4436 struct trace_array *tr = iter->tr;
4437 struct trace_seq *s = &iter->seq;
4438 unsigned char newline = '\n';
4439 struct trace_entry *entry;
4440 struct trace_event *event;
4444 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4445 SEQ_PUT_HEX_FIELD(s, entry->pid);
4446 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4447 SEQ_PUT_HEX_FIELD(s, iter->ts);
4448 if (trace_seq_has_overflowed(s))
4449 return TRACE_TYPE_PARTIAL_LINE;
4452 event = ftrace_find_event(entry->type);
4454 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4455 if (ret != TRACE_TYPE_HANDLED)
4459 SEQ_PUT_FIELD(s, newline);
4461 return trace_handle_return(s);
4464 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4466 struct trace_array *tr = iter->tr;
4467 struct trace_seq *s = &iter->seq;
4468 struct trace_entry *entry;
4469 struct trace_event *event;
4473 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4474 SEQ_PUT_FIELD(s, entry->pid);
4475 SEQ_PUT_FIELD(s, iter->cpu);
4476 SEQ_PUT_FIELD(s, iter->ts);
4477 if (trace_seq_has_overflowed(s))
4478 return TRACE_TYPE_PARTIAL_LINE;
4481 event = ftrace_find_event(entry->type);
4482 return event ? event->funcs->binary(iter, 0, event) :
4486 int trace_empty(struct trace_iterator *iter)
4488 struct ring_buffer_iter *buf_iter;
4491 /* If we are looking at one CPU buffer, only check that one */
4492 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4493 cpu = iter->cpu_file;
4494 buf_iter = trace_buffer_iter(iter, cpu);
4496 if (!ring_buffer_iter_empty(buf_iter))
4499 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4505 for_each_tracing_cpu(cpu) {
4506 buf_iter = trace_buffer_iter(iter, cpu);
4508 if (!ring_buffer_iter_empty(buf_iter))
4511 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4519 /* Called with trace_event_read_lock() held. */
4520 enum print_line_t print_trace_line(struct trace_iterator *iter)
4522 struct trace_array *tr = iter->tr;
4523 unsigned long trace_flags = tr->trace_flags;
4524 enum print_line_t ret;
4526 if (iter->lost_events) {
4527 if (iter->lost_events == (unsigned long)-1)
4528 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4531 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4532 iter->cpu, iter->lost_events);
4533 if (trace_seq_has_overflowed(&iter->seq))
4534 return TRACE_TYPE_PARTIAL_LINE;
4537 if (iter->trace && iter->trace->print_line) {
4538 ret = iter->trace->print_line(iter);
4539 if (ret != TRACE_TYPE_UNHANDLED)
4543 if (iter->ent->type == TRACE_BPUTS &&
4544 trace_flags & TRACE_ITER_PRINTK &&
4545 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4546 return trace_print_bputs_msg_only(iter);
4548 if (iter->ent->type == TRACE_BPRINT &&
4549 trace_flags & TRACE_ITER_PRINTK &&
4550 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4551 return trace_print_bprintk_msg_only(iter);
4553 if (iter->ent->type == TRACE_PRINT &&
4554 trace_flags & TRACE_ITER_PRINTK &&
4555 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4556 return trace_print_printk_msg_only(iter);
4558 if (trace_flags & TRACE_ITER_BIN)
4559 return print_bin_fmt(iter);
4561 if (trace_flags & TRACE_ITER_HEX)
4562 return print_hex_fmt(iter);
4564 if (trace_flags & TRACE_ITER_RAW)
4565 return print_raw_fmt(iter);
4567 return print_trace_fmt(iter);
4570 void trace_latency_header(struct seq_file *m)
4572 struct trace_iterator *iter = m->private;
4573 struct trace_array *tr = iter->tr;
4575 /* print nothing if the buffers are empty */
4576 if (trace_empty(iter))
4579 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4580 print_trace_header(m, iter);
4582 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4583 print_lat_help_header(m);
4586 void trace_default_header(struct seq_file *m)
4588 struct trace_iterator *iter = m->private;
4589 struct trace_array *tr = iter->tr;
4590 unsigned long trace_flags = tr->trace_flags;
4592 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4595 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4596 /* print nothing if the buffers are empty */
4597 if (trace_empty(iter))
4599 print_trace_header(m, iter);
4600 if (!(trace_flags & TRACE_ITER_VERBOSE))
4601 print_lat_help_header(m);
4603 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4604 if (trace_flags & TRACE_ITER_IRQ_INFO)
4605 print_func_help_header_irq(iter->array_buffer,
4608 print_func_help_header(iter->array_buffer, m,
4614 static void test_ftrace_alive(struct seq_file *m)
4616 if (!ftrace_is_dead())
4618 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4619 "# MAY BE MISSING FUNCTION EVENTS\n");
4622 #ifdef CONFIG_TRACER_MAX_TRACE
4623 static void show_snapshot_main_help(struct seq_file *m)
4625 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4626 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4627 "# Takes a snapshot of the main buffer.\n"
4628 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4629 "# (Doesn't have to be '2' works with any number that\n"
4630 "# is not a '0' or '1')\n");
4633 static void show_snapshot_percpu_help(struct seq_file *m)
4635 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4636 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4637 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4638 "# Takes a snapshot of the main buffer for this cpu.\n");
4640 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4641 "# Must use main snapshot file to allocate.\n");
4643 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4644 "# (Doesn't have to be '2' works with any number that\n"
4645 "# is not a '0' or '1')\n");
4648 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4650 if (iter->tr->allocated_snapshot)
4651 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4653 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4655 seq_puts(m, "# Snapshot commands:\n");
4656 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4657 show_snapshot_main_help(m);
4659 show_snapshot_percpu_help(m);
4662 /* Should never be called */
4663 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4666 static int s_show(struct seq_file *m, void *v)
4668 struct trace_iterator *iter = v;
4671 if (iter->ent == NULL) {
4673 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4675 test_ftrace_alive(m);
4677 if (iter->snapshot && trace_empty(iter))
4678 print_snapshot_help(m, iter);
4679 else if (iter->trace && iter->trace->print_header)
4680 iter->trace->print_header(m);
4682 trace_default_header(m);
4684 } else if (iter->leftover) {
4686 * If we filled the seq_file buffer earlier, we
4687 * want to just show it now.
4689 ret = trace_print_seq(m, &iter->seq);
4691 /* ret should this time be zero, but you never know */
4692 iter->leftover = ret;
4695 print_trace_line(iter);
4696 ret = trace_print_seq(m, &iter->seq);
4698 * If we overflow the seq_file buffer, then it will
4699 * ask us for this data again at start up.
4701 * ret is 0 if seq_file write succeeded.
4704 iter->leftover = ret;
4711 * Should be used after trace_array_get(), trace_types_lock
4712 * ensures that i_cdev was already initialized.
4714 static inline int tracing_get_cpu(struct inode *inode)
4716 if (inode->i_cdev) /* See trace_create_cpu_file() */
4717 return (long)inode->i_cdev - 1;
4718 return RING_BUFFER_ALL_CPUS;
4721 static const struct seq_operations tracer_seq_ops = {
4728 static struct trace_iterator *
4729 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4731 struct trace_array *tr = inode->i_private;
4732 struct trace_iterator *iter;
4735 if (tracing_disabled)
4736 return ERR_PTR(-ENODEV);
4738 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4740 return ERR_PTR(-ENOMEM);
4742 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4744 if (!iter->buffer_iter)
4748 * trace_find_next_entry() may need to save off iter->ent.
4749 * It will place it into the iter->temp buffer. As most
4750 * events are less than 128, allocate a buffer of that size.
4751 * If one is greater, then trace_find_next_entry() will
4752 * allocate a new buffer to adjust for the bigger iter->ent.
4753 * It's not critical if it fails to get allocated here.
4755 iter->temp = kmalloc(128, GFP_KERNEL);
4757 iter->temp_size = 128;
4760 * trace_event_printf() may need to modify given format
4761 * string to replace %p with %px so that it shows real address
4762 * instead of hash value. However, that is only for the event
4763 * tracing, other tracer may not need. Defer the allocation
4764 * until it is needed.
4770 * We make a copy of the current tracer to avoid concurrent
4771 * changes on it while we are reading.
4773 mutex_lock(&trace_types_lock);
4774 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
4778 *iter->trace = *tr->current_trace;
4780 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4785 #ifdef CONFIG_TRACER_MAX_TRACE
4786 /* Currently only the top directory has a snapshot */
4787 if (tr->current_trace->print_max || snapshot)
4788 iter->array_buffer = &tr->max_buffer;
4791 iter->array_buffer = &tr->array_buffer;
4792 iter->snapshot = snapshot;
4794 iter->cpu_file = tracing_get_cpu(inode);
4795 mutex_init(&iter->mutex);
4797 /* Notify the tracer early; before we stop tracing. */
4798 if (iter->trace->open)
4799 iter->trace->open(iter);
4801 /* Annotate start of buffers if we had overruns */
4802 if (ring_buffer_overruns(iter->array_buffer->buffer))
4803 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4805 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4806 if (trace_clocks[tr->clock_id].in_ns)
4807 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4810 * If pause-on-trace is enabled, then stop the trace while
4811 * dumping, unless this is the "snapshot" file
4813 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4814 tracing_stop_tr(tr);
4816 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4817 for_each_tracing_cpu(cpu) {
4818 iter->buffer_iter[cpu] =
4819 ring_buffer_read_prepare(iter->array_buffer->buffer,
4822 ring_buffer_read_prepare_sync();
4823 for_each_tracing_cpu(cpu) {
4824 ring_buffer_read_start(iter->buffer_iter[cpu]);
4825 tracing_iter_reset(iter, cpu);
4828 cpu = iter->cpu_file;
4829 iter->buffer_iter[cpu] =
4830 ring_buffer_read_prepare(iter->array_buffer->buffer,
4832 ring_buffer_read_prepare_sync();
4833 ring_buffer_read_start(iter->buffer_iter[cpu]);
4834 tracing_iter_reset(iter, cpu);
4837 mutex_unlock(&trace_types_lock);
4842 mutex_unlock(&trace_types_lock);
4845 kfree(iter->buffer_iter);
4847 seq_release_private(inode, file);
4848 return ERR_PTR(-ENOMEM);
4851 int tracing_open_generic(struct inode *inode, struct file *filp)
4855 ret = tracing_check_open_get_tr(NULL);
4859 filp->private_data = inode->i_private;
4863 bool tracing_is_disabled(void)
4865 return (tracing_disabled) ? true: false;
4869 * Open and update trace_array ref count.
4870 * Must have the current trace_array passed to it.
4872 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4874 struct trace_array *tr = inode->i_private;
4877 ret = tracing_check_open_get_tr(tr);
4881 filp->private_data = inode->i_private;
4886 static int tracing_mark_open(struct inode *inode, struct file *filp)
4888 stream_open(inode, filp);
4889 return tracing_open_generic_tr(inode, filp);
4892 static int tracing_release(struct inode *inode, struct file *file)
4894 struct trace_array *tr = inode->i_private;
4895 struct seq_file *m = file->private_data;
4896 struct trace_iterator *iter;
4899 if (!(file->f_mode & FMODE_READ)) {
4900 trace_array_put(tr);
4904 /* Writes do not use seq_file */
4906 mutex_lock(&trace_types_lock);
4908 for_each_tracing_cpu(cpu) {
4909 if (iter->buffer_iter[cpu])
4910 ring_buffer_read_finish(iter->buffer_iter[cpu]);
4913 if (iter->trace && iter->trace->close)
4914 iter->trace->close(iter);
4916 if (!iter->snapshot && tr->stop_count)
4917 /* reenable tracing if it was previously enabled */
4918 tracing_start_tr(tr);
4920 __trace_array_put(tr);
4922 mutex_unlock(&trace_types_lock);
4924 mutex_destroy(&iter->mutex);
4925 free_cpumask_var(iter->started);
4929 kfree(iter->buffer_iter);
4930 seq_release_private(inode, file);
4935 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
4937 struct trace_array *tr = inode->i_private;
4939 trace_array_put(tr);
4943 static int tracing_single_release_tr(struct inode *inode, struct file *file)
4945 struct trace_array *tr = inode->i_private;
4947 trace_array_put(tr);
4949 return single_release(inode, file);
4952 static int tracing_open(struct inode *inode, struct file *file)
4954 struct trace_array *tr = inode->i_private;
4955 struct trace_iterator *iter;
4958 ret = tracing_check_open_get_tr(tr);
4962 /* If this file was open for write, then erase contents */
4963 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
4964 int cpu = tracing_get_cpu(inode);
4965 struct array_buffer *trace_buf = &tr->array_buffer;
4967 #ifdef CONFIG_TRACER_MAX_TRACE
4968 if (tr->current_trace->print_max)
4969 trace_buf = &tr->max_buffer;
4972 if (cpu == RING_BUFFER_ALL_CPUS)
4973 tracing_reset_online_cpus(trace_buf);
4975 tracing_reset_cpu(trace_buf, cpu);
4978 if (file->f_mode & FMODE_READ) {
4979 iter = __tracing_open(inode, file, false);
4981 ret = PTR_ERR(iter);
4982 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4983 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4987 trace_array_put(tr);
4993 * Some tracers are not suitable for instance buffers.
4994 * A tracer is always available for the global array (toplevel)
4995 * or if it explicitly states that it is.
4998 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
5000 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5003 /* Find the next tracer that this trace array may use */
5004 static struct tracer *
5005 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5007 while (t && !trace_ok_for_array(t, tr))
5014 t_next(struct seq_file *m, void *v, loff_t *pos)
5016 struct trace_array *tr = m->private;
5017 struct tracer *t = v;
5022 t = get_tracer_for_array(tr, t->next);
5027 static void *t_start(struct seq_file *m, loff_t *pos)
5029 struct trace_array *tr = m->private;
5033 mutex_lock(&trace_types_lock);
5035 t = get_tracer_for_array(tr, trace_types);
5036 for (; t && l < *pos; t = t_next(m, t, &l))
5042 static void t_stop(struct seq_file *m, void *p)
5044 mutex_unlock(&trace_types_lock);
5047 static int t_show(struct seq_file *m, void *v)
5049 struct tracer *t = v;
5054 seq_puts(m, t->name);
5063 static const struct seq_operations show_traces_seq_ops = {
5070 static int show_traces_open(struct inode *inode, struct file *file)
5072 struct trace_array *tr = inode->i_private;
5076 ret = tracing_check_open_get_tr(tr);
5080 ret = seq_open(file, &show_traces_seq_ops);
5082 trace_array_put(tr);
5086 m = file->private_data;
5092 static int show_traces_release(struct inode *inode, struct file *file)
5094 struct trace_array *tr = inode->i_private;
5096 trace_array_put(tr);
5097 return seq_release(inode, file);
5101 tracing_write_stub(struct file *filp, const char __user *ubuf,
5102 size_t count, loff_t *ppos)
5107 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5111 if (file->f_mode & FMODE_READ)
5112 ret = seq_lseek(file, offset, whence);
5114 file->f_pos = ret = 0;
5119 static const struct file_operations tracing_fops = {
5120 .open = tracing_open,
5122 .write = tracing_write_stub,
5123 .llseek = tracing_lseek,
5124 .release = tracing_release,
5127 static const struct file_operations show_traces_fops = {
5128 .open = show_traces_open,
5130 .llseek = seq_lseek,
5131 .release = show_traces_release,
5135 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5136 size_t count, loff_t *ppos)
5138 struct trace_array *tr = file_inode(filp)->i_private;
5142 len = snprintf(NULL, 0, "%*pb\n",
5143 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5144 mask_str = kmalloc(len, GFP_KERNEL);
5148 len = snprintf(mask_str, len, "%*pb\n",
5149 cpumask_pr_args(tr->tracing_cpumask));
5154 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5162 int tracing_set_cpumask(struct trace_array *tr,
5163 cpumask_var_t tracing_cpumask_new)
5170 local_irq_disable();
5171 arch_spin_lock(&tr->max_lock);
5172 for_each_tracing_cpu(cpu) {
5174 * Increase/decrease the disabled counter if we are
5175 * about to flip a bit in the cpumask:
5177 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5178 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5179 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5180 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5182 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5183 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5184 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5185 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5188 arch_spin_unlock(&tr->max_lock);
5191 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5197 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5198 size_t count, loff_t *ppos)
5200 struct trace_array *tr = file_inode(filp)->i_private;
5201 cpumask_var_t tracing_cpumask_new;
5204 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5207 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5211 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5215 free_cpumask_var(tracing_cpumask_new);
5220 free_cpumask_var(tracing_cpumask_new);
5225 static const struct file_operations tracing_cpumask_fops = {
5226 .open = tracing_open_generic_tr,
5227 .read = tracing_cpumask_read,
5228 .write = tracing_cpumask_write,
5229 .release = tracing_release_generic_tr,
5230 .llseek = generic_file_llseek,
5233 static int tracing_trace_options_show(struct seq_file *m, void *v)
5235 struct tracer_opt *trace_opts;
5236 struct trace_array *tr = m->private;
5240 mutex_lock(&trace_types_lock);
5241 tracer_flags = tr->current_trace->flags->val;
5242 trace_opts = tr->current_trace->flags->opts;
5244 for (i = 0; trace_options[i]; i++) {
5245 if (tr->trace_flags & (1 << i))
5246 seq_printf(m, "%s\n", trace_options[i]);
5248 seq_printf(m, "no%s\n", trace_options[i]);
5251 for (i = 0; trace_opts[i].name; i++) {
5252 if (tracer_flags & trace_opts[i].bit)
5253 seq_printf(m, "%s\n", trace_opts[i].name);
5255 seq_printf(m, "no%s\n", trace_opts[i].name);
5257 mutex_unlock(&trace_types_lock);
5262 static int __set_tracer_option(struct trace_array *tr,
5263 struct tracer_flags *tracer_flags,
5264 struct tracer_opt *opts, int neg)
5266 struct tracer *trace = tracer_flags->trace;
5269 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5274 tracer_flags->val &= ~opts->bit;
5276 tracer_flags->val |= opts->bit;
5280 /* Try to assign a tracer specific option */
5281 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5283 struct tracer *trace = tr->current_trace;
5284 struct tracer_flags *tracer_flags = trace->flags;
5285 struct tracer_opt *opts = NULL;
5288 for (i = 0; tracer_flags->opts[i].name; i++) {
5289 opts = &tracer_flags->opts[i];
5291 if (strcmp(cmp, opts->name) == 0)
5292 return __set_tracer_option(tr, trace->flags, opts, neg);
5298 /* Some tracers require overwrite to stay enabled */
5299 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5301 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5307 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5311 if ((mask == TRACE_ITER_RECORD_TGID) ||
5312 (mask == TRACE_ITER_RECORD_CMD))
5313 lockdep_assert_held(&event_mutex);
5315 /* do nothing if flag is already set */
5316 if (!!(tr->trace_flags & mask) == !!enabled)
5319 /* Give the tracer a chance to approve the change */
5320 if (tr->current_trace->flag_changed)
5321 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5325 tr->trace_flags |= mask;
5327 tr->trace_flags &= ~mask;
5329 if (mask == TRACE_ITER_RECORD_CMD)
5330 trace_event_enable_cmd_record(enabled);
5332 if (mask == TRACE_ITER_RECORD_TGID) {
5334 tgid_map_max = pid_max;
5335 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5339 * Pairs with smp_load_acquire() in
5340 * trace_find_tgid_ptr() to ensure that if it observes
5341 * the tgid_map we just allocated then it also observes
5342 * the corresponding tgid_map_max value.
5344 smp_store_release(&tgid_map, map);
5347 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5351 trace_event_enable_tgid_record(enabled);
5354 if (mask == TRACE_ITER_EVENT_FORK)
5355 trace_event_follow_fork(tr, enabled);
5357 if (mask == TRACE_ITER_FUNC_FORK)
5358 ftrace_pid_follow_fork(tr, enabled);
5360 if (mask == TRACE_ITER_OVERWRITE) {
5361 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5362 #ifdef CONFIG_TRACER_MAX_TRACE
5363 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5367 if (mask == TRACE_ITER_PRINTK) {
5368 trace_printk_start_stop_comm(enabled);
5369 trace_printk_control(enabled);
5375 int trace_set_options(struct trace_array *tr, char *option)
5380 size_t orig_len = strlen(option);
5383 cmp = strstrip(option);
5385 len = str_has_prefix(cmp, "no");
5391 mutex_lock(&event_mutex);
5392 mutex_lock(&trace_types_lock);
5394 ret = match_string(trace_options, -1, cmp);
5395 /* If no option could be set, test the specific tracer options */
5397 ret = set_tracer_option(tr, cmp, neg);
5399 ret = set_tracer_flag(tr, 1 << ret, !neg);
5401 mutex_unlock(&trace_types_lock);
5402 mutex_unlock(&event_mutex);
5405 * If the first trailing whitespace is replaced with '\0' by strstrip,
5406 * turn it back into a space.
5408 if (orig_len > strlen(option))
5409 option[strlen(option)] = ' ';
5414 static void __init apply_trace_boot_options(void)
5416 char *buf = trace_boot_options_buf;
5420 option = strsep(&buf, ",");
5426 trace_set_options(&global_trace, option);
5428 /* Put back the comma to allow this to be called again */
5435 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5436 size_t cnt, loff_t *ppos)
5438 struct seq_file *m = filp->private_data;
5439 struct trace_array *tr = m->private;
5443 if (cnt >= sizeof(buf))
5446 if (copy_from_user(buf, ubuf, cnt))
5451 ret = trace_set_options(tr, buf);
5460 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5462 struct trace_array *tr = inode->i_private;
5465 ret = tracing_check_open_get_tr(tr);
5469 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5471 trace_array_put(tr);
5476 static const struct file_operations tracing_iter_fops = {
5477 .open = tracing_trace_options_open,
5479 .llseek = seq_lseek,
5480 .release = tracing_single_release_tr,
5481 .write = tracing_trace_options_write,
5484 static const char readme_msg[] =
5485 "tracing mini-HOWTO:\n\n"
5486 "# echo 0 > tracing_on : quick way to disable tracing\n"
5487 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5488 " Important files:\n"
5489 " trace\t\t\t- The static contents of the buffer\n"
5490 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5491 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5492 " current_tracer\t- function and latency tracers\n"
5493 " available_tracers\t- list of configured tracers for current_tracer\n"
5494 " error_log\t- error log for failed commands (that support it)\n"
5495 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5496 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5497 " trace_clock\t\t- change the clock used to order events\n"
5498 " local: Per cpu clock but may not be synced across CPUs\n"
5499 " global: Synced across CPUs but slows tracing down.\n"
5500 " counter: Not a clock, but just an increment\n"
5501 " uptime: Jiffy counter from time of boot\n"
5502 " perf: Same clock that perf events use\n"
5503 #ifdef CONFIG_X86_64
5504 " x86-tsc: TSC cycle counter\n"
5506 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5507 " delta: Delta difference against a buffer-wide timestamp\n"
5508 " absolute: Absolute (standalone) timestamp\n"
5509 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5510 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5511 " tracing_cpumask\t- Limit which CPUs to trace\n"
5512 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5513 "\t\t\t Remove sub-buffer with rmdir\n"
5514 " trace_options\t\t- Set format or modify how tracing happens\n"
5515 "\t\t\t Disable an option by prefixing 'no' to the\n"
5516 "\t\t\t option name\n"
5517 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5518 #ifdef CONFIG_DYNAMIC_FTRACE
5519 "\n available_filter_functions - list of functions that can be filtered on\n"
5520 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5521 "\t\t\t functions\n"
5522 "\t accepts: func_full_name or glob-matching-pattern\n"
5523 "\t modules: Can select a group via module\n"
5524 "\t Format: :mod:<module-name>\n"
5525 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5526 "\t triggers: a command to perform when function is hit\n"
5527 "\t Format: <function>:<trigger>[:count]\n"
5528 "\t trigger: traceon, traceoff\n"
5529 "\t\t enable_event:<system>:<event>\n"
5530 "\t\t disable_event:<system>:<event>\n"
5531 #ifdef CONFIG_STACKTRACE
5534 #ifdef CONFIG_TRACER_SNAPSHOT
5539 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5540 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5541 "\t The first one will disable tracing every time do_fault is hit\n"
5542 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5543 "\t The first time do trap is hit and it disables tracing, the\n"
5544 "\t counter will decrement to 2. If tracing is already disabled,\n"
5545 "\t the counter will not decrement. It only decrements when the\n"
5546 "\t trigger did work\n"
5547 "\t To remove trigger without count:\n"
5548 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5549 "\t To remove trigger with a count:\n"
5550 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5551 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5552 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5553 "\t modules: Can select a group via module command :mod:\n"
5554 "\t Does not accept triggers\n"
5555 #endif /* CONFIG_DYNAMIC_FTRACE */
5556 #ifdef CONFIG_FUNCTION_TRACER
5557 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5559 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5562 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5563 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5564 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5565 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5567 #ifdef CONFIG_TRACER_SNAPSHOT
5568 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5569 "\t\t\t snapshot buffer. Read the contents for more\n"
5570 "\t\t\t information\n"
5572 #ifdef CONFIG_STACK_TRACER
5573 " stack_trace\t\t- Shows the max stack trace when active\n"
5574 " stack_max_size\t- Shows current max stack size that was traced\n"
5575 "\t\t\t Write into this file to reset the max size (trigger a\n"
5576 "\t\t\t new trace)\n"
5577 #ifdef CONFIG_DYNAMIC_FTRACE
5578 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5581 #endif /* CONFIG_STACK_TRACER */
5582 #ifdef CONFIG_DYNAMIC_EVENTS
5583 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5584 "\t\t\t Write into this file to define/undefine new trace events.\n"
5586 #ifdef CONFIG_KPROBE_EVENTS
5587 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5588 "\t\t\t Write into this file to define/undefine new trace events.\n"
5590 #ifdef CONFIG_UPROBE_EVENTS
5591 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5592 "\t\t\t Write into this file to define/undefine new trace events.\n"
5594 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5595 "\t accepts: event-definitions (one definition per line)\n"
5596 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5597 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5598 #ifdef CONFIG_HIST_TRIGGERS
5599 "\t s:[synthetic/]<event> <field> [<field>]\n"
5601 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>]\n"
5602 "\t -:[<group>/][<event>]\n"
5603 #ifdef CONFIG_KPROBE_EVENTS
5604 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5605 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5607 #ifdef CONFIG_UPROBE_EVENTS
5608 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5610 "\t args: <name>=fetcharg[:type]\n"
5611 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5612 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5613 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5615 "\t $stack<index>, $stack, $retval, $comm,\n"
5617 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5618 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, string, symbol,\n"
5619 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5620 "\t <type>\\[<array-size>\\]\n"
5621 #ifdef CONFIG_HIST_TRIGGERS
5622 "\t field: <stype> <name>;\n"
5623 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5624 "\t [unsigned] char/int/long\n"
5626 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5627 "\t of the <attached-group>/<attached-event>.\n"
5629 " events/\t\t- Directory containing all trace event subsystems:\n"
5630 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5631 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5632 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5634 " filter\t\t- If set, only events passing filter are traced\n"
5635 " events/<system>/<event>/\t- Directory containing control files for\n"
5637 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5638 " filter\t\t- If set, only events passing filter are traced\n"
5639 " trigger\t\t- If set, a command to perform when event is hit\n"
5640 "\t Format: <trigger>[:count][if <filter>]\n"
5641 "\t trigger: traceon, traceoff\n"
5642 "\t enable_event:<system>:<event>\n"
5643 "\t disable_event:<system>:<event>\n"
5644 #ifdef CONFIG_HIST_TRIGGERS
5645 "\t enable_hist:<system>:<event>\n"
5646 "\t disable_hist:<system>:<event>\n"
5648 #ifdef CONFIG_STACKTRACE
5651 #ifdef CONFIG_TRACER_SNAPSHOT
5654 #ifdef CONFIG_HIST_TRIGGERS
5655 "\t\t hist (see below)\n"
5657 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5658 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5659 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5660 "\t events/block/block_unplug/trigger\n"
5661 "\t The first disables tracing every time block_unplug is hit.\n"
5662 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5663 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5664 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5665 "\t Like function triggers, the counter is only decremented if it\n"
5666 "\t enabled or disabled tracing.\n"
5667 "\t To remove a trigger without a count:\n"
5668 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5669 "\t To remove a trigger with a count:\n"
5670 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5671 "\t Filters can be ignored when removing a trigger.\n"
5672 #ifdef CONFIG_HIST_TRIGGERS
5673 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5674 "\t Format: hist:keys=<field1[,field2,...]>\n"
5675 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5676 "\t [:values=<field1[,field2,...]>]\n"
5677 "\t [:sort=<field1[,field2,...]>]\n"
5678 "\t [:size=#entries]\n"
5679 "\t [:pause][:continue][:clear]\n"
5680 "\t [:name=histname1]\n"
5681 "\t [:<handler>.<action>]\n"
5682 "\t [if <filter>]\n\n"
5683 "\t Note, special fields can be used as well:\n"
5684 "\t common_timestamp - to record current timestamp\n"
5685 "\t common_cpu - to record the CPU the event happened on\n"
5687 "\t A hist trigger variable can be:\n"
5688 "\t - a reference to a field e.g. x=current_timestamp,\n"
5689 "\t - a reference to another variable e.g. y=$x,\n"
5690 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5691 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5693 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5694 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5695 "\t variable reference, field or numeric literal.\n"
5697 "\t When a matching event is hit, an entry is added to a hash\n"
5698 "\t table using the key(s) and value(s) named, and the value of a\n"
5699 "\t sum called 'hitcount' is incremented. Keys and values\n"
5700 "\t correspond to fields in the event's format description. Keys\n"
5701 "\t can be any field, or the special string 'stacktrace'.\n"
5702 "\t Compound keys consisting of up to two fields can be specified\n"
5703 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5704 "\t fields. Sort keys consisting of up to two fields can be\n"
5705 "\t specified using the 'sort' keyword. The sort direction can\n"
5706 "\t be modified by appending '.descending' or '.ascending' to a\n"
5707 "\t sort field. The 'size' parameter can be used to specify more\n"
5708 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5709 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5710 "\t its histogram data will be shared with other triggers of the\n"
5711 "\t same name, and trigger hits will update this common data.\n\n"
5712 "\t Reading the 'hist' file for the event will dump the hash\n"
5713 "\t table in its entirety to stdout. If there are multiple hist\n"
5714 "\t triggers attached to an event, there will be a table for each\n"
5715 "\t trigger in the output. The table displayed for a named\n"
5716 "\t trigger will be the same as any other instance having the\n"
5717 "\t same name. The default format used to display a given field\n"
5718 "\t can be modified by appending any of the following modifiers\n"
5719 "\t to the field name, as applicable:\n\n"
5720 "\t .hex display a number as a hex value\n"
5721 "\t .sym display an address as a symbol\n"
5722 "\t .sym-offset display an address as a symbol and offset\n"
5723 "\t .execname display a common_pid as a program name\n"
5724 "\t .syscall display a syscall id as a syscall name\n"
5725 "\t .log2 display log2 value rather than raw number\n"
5726 "\t .buckets=size display values in groups of size rather than raw number\n"
5727 "\t .usecs display a common_timestamp in microseconds\n"
5728 "\t .percent display a number of percentage value\n"
5729 "\t .graph display a bar-graph of a value\n\n"
5730 "\t The 'pause' parameter can be used to pause an existing hist\n"
5731 "\t trigger or to start a hist trigger but not log any events\n"
5732 "\t until told to do so. 'continue' can be used to start or\n"
5733 "\t restart a paused hist trigger.\n\n"
5734 "\t The 'clear' parameter will clear the contents of a running\n"
5735 "\t hist trigger and leave its current paused/active state\n"
5737 "\t The enable_hist and disable_hist triggers can be used to\n"
5738 "\t have one event conditionally start and stop another event's\n"
5739 "\t already-attached hist trigger. The syntax is analogous to\n"
5740 "\t the enable_event and disable_event triggers.\n\n"
5741 "\t Hist trigger handlers and actions are executed whenever a\n"
5742 "\t a histogram entry is added or updated. They take the form:\n\n"
5743 "\t <handler>.<action>\n\n"
5744 "\t The available handlers are:\n\n"
5745 "\t onmatch(matching.event) - invoke on addition or update\n"
5746 "\t onmax(var) - invoke if var exceeds current max\n"
5747 "\t onchange(var) - invoke action if var changes\n\n"
5748 "\t The available actions are:\n\n"
5749 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5750 "\t save(field,...) - save current event fields\n"
5751 #ifdef CONFIG_TRACER_SNAPSHOT
5752 "\t snapshot() - snapshot the trace buffer\n\n"
5754 #ifdef CONFIG_SYNTH_EVENTS
5755 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5756 "\t Write into this file to define/undefine new synthetic events.\n"
5757 "\t example: echo 'myevent u64 lat; char name[]' >> synthetic_events\n"
5763 tracing_readme_read(struct file *filp, char __user *ubuf,
5764 size_t cnt, loff_t *ppos)
5766 return simple_read_from_buffer(ubuf, cnt, ppos,
5767 readme_msg, strlen(readme_msg));
5770 static const struct file_operations tracing_readme_fops = {
5771 .open = tracing_open_generic,
5772 .read = tracing_readme_read,
5773 .llseek = generic_file_llseek,
5776 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5780 return trace_find_tgid_ptr(pid);
5783 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5787 return trace_find_tgid_ptr(pid);
5790 static void saved_tgids_stop(struct seq_file *m, void *v)
5794 static int saved_tgids_show(struct seq_file *m, void *v)
5796 int *entry = (int *)v;
5797 int pid = entry - tgid_map;
5803 seq_printf(m, "%d %d\n", pid, tgid);
5807 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5808 .start = saved_tgids_start,
5809 .stop = saved_tgids_stop,
5810 .next = saved_tgids_next,
5811 .show = saved_tgids_show,
5814 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5818 ret = tracing_check_open_get_tr(NULL);
5822 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5826 static const struct file_operations tracing_saved_tgids_fops = {
5827 .open = tracing_saved_tgids_open,
5829 .llseek = seq_lseek,
5830 .release = seq_release,
5833 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5835 unsigned int *ptr = v;
5837 if (*pos || m->count)
5842 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5844 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5853 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5859 arch_spin_lock(&trace_cmdline_lock);
5861 v = &savedcmd->map_cmdline_to_pid[0];
5863 v = saved_cmdlines_next(m, v, &l);
5871 static void saved_cmdlines_stop(struct seq_file *m, void *v)
5873 arch_spin_unlock(&trace_cmdline_lock);
5877 static int saved_cmdlines_show(struct seq_file *m, void *v)
5879 char buf[TASK_COMM_LEN];
5880 unsigned int *pid = v;
5882 __trace_find_cmdline(*pid, buf);
5883 seq_printf(m, "%d %s\n", *pid, buf);
5887 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
5888 .start = saved_cmdlines_start,
5889 .next = saved_cmdlines_next,
5890 .stop = saved_cmdlines_stop,
5891 .show = saved_cmdlines_show,
5894 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
5898 ret = tracing_check_open_get_tr(NULL);
5902 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
5905 static const struct file_operations tracing_saved_cmdlines_fops = {
5906 .open = tracing_saved_cmdlines_open,
5908 .llseek = seq_lseek,
5909 .release = seq_release,
5913 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
5914 size_t cnt, loff_t *ppos)
5920 arch_spin_lock(&trace_cmdline_lock);
5921 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
5922 arch_spin_unlock(&trace_cmdline_lock);
5925 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5928 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
5930 kfree(s->saved_cmdlines);
5931 kfree(s->map_cmdline_to_pid);
5935 static int tracing_resize_saved_cmdlines(unsigned int val)
5937 struct saved_cmdlines_buffer *s, *savedcmd_temp;
5939 s = kmalloc(sizeof(*s), GFP_KERNEL);
5943 if (allocate_cmdlines_buffer(val, s) < 0) {
5949 arch_spin_lock(&trace_cmdline_lock);
5950 savedcmd_temp = savedcmd;
5952 arch_spin_unlock(&trace_cmdline_lock);
5954 free_saved_cmdlines_buffer(savedcmd_temp);
5960 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
5961 size_t cnt, loff_t *ppos)
5966 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5970 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
5971 if (!val || val > PID_MAX_DEFAULT)
5974 ret = tracing_resize_saved_cmdlines((unsigned int)val);
5983 static const struct file_operations tracing_saved_cmdlines_size_fops = {
5984 .open = tracing_open_generic,
5985 .read = tracing_saved_cmdlines_size_read,
5986 .write = tracing_saved_cmdlines_size_write,
5989 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
5990 static union trace_eval_map_item *
5991 update_eval_map(union trace_eval_map_item *ptr)
5993 if (!ptr->map.eval_string) {
5994 if (ptr->tail.next) {
5995 ptr = ptr->tail.next;
5996 /* Set ptr to the next real item (skip head) */
6004 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6006 union trace_eval_map_item *ptr = v;
6009 * Paranoid! If ptr points to end, we don't want to increment past it.
6010 * This really should never happen.
6013 ptr = update_eval_map(ptr);
6014 if (WARN_ON_ONCE(!ptr))
6018 ptr = update_eval_map(ptr);
6023 static void *eval_map_start(struct seq_file *m, loff_t *pos)
6025 union trace_eval_map_item *v;
6028 mutex_lock(&trace_eval_mutex);
6030 v = trace_eval_maps;
6034 while (v && l < *pos) {
6035 v = eval_map_next(m, v, &l);
6041 static void eval_map_stop(struct seq_file *m, void *v)
6043 mutex_unlock(&trace_eval_mutex);
6046 static int eval_map_show(struct seq_file *m, void *v)
6048 union trace_eval_map_item *ptr = v;
6050 seq_printf(m, "%s %ld (%s)\n",
6051 ptr->map.eval_string, ptr->map.eval_value,
6057 static const struct seq_operations tracing_eval_map_seq_ops = {
6058 .start = eval_map_start,
6059 .next = eval_map_next,
6060 .stop = eval_map_stop,
6061 .show = eval_map_show,
6064 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6068 ret = tracing_check_open_get_tr(NULL);
6072 return seq_open(filp, &tracing_eval_map_seq_ops);
6075 static const struct file_operations tracing_eval_map_fops = {
6076 .open = tracing_eval_map_open,
6078 .llseek = seq_lseek,
6079 .release = seq_release,
6082 static inline union trace_eval_map_item *
6083 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6085 /* Return tail of array given the head */
6086 return ptr + ptr->head.length + 1;
6090 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6093 struct trace_eval_map **stop;
6094 struct trace_eval_map **map;
6095 union trace_eval_map_item *map_array;
6096 union trace_eval_map_item *ptr;
6101 * The trace_eval_maps contains the map plus a head and tail item,
6102 * where the head holds the module and length of array, and the
6103 * tail holds a pointer to the next list.
6105 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6107 pr_warn("Unable to allocate trace eval mapping\n");
6111 mutex_lock(&trace_eval_mutex);
6113 if (!trace_eval_maps)
6114 trace_eval_maps = map_array;
6116 ptr = trace_eval_maps;
6118 ptr = trace_eval_jmp_to_tail(ptr);
6119 if (!ptr->tail.next)
6121 ptr = ptr->tail.next;
6124 ptr->tail.next = map_array;
6126 map_array->head.mod = mod;
6127 map_array->head.length = len;
6130 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6131 map_array->map = **map;
6134 memset(map_array, 0, sizeof(*map_array));
6136 mutex_unlock(&trace_eval_mutex);
6139 static void trace_create_eval_file(struct dentry *d_tracer)
6141 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6142 NULL, &tracing_eval_map_fops);
6145 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
6146 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
6147 static inline void trace_insert_eval_map_file(struct module *mod,
6148 struct trace_eval_map **start, int len) { }
6149 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6151 static void trace_insert_eval_map(struct module *mod,
6152 struct trace_eval_map **start, int len)
6154 struct trace_eval_map **map;
6161 trace_event_eval_update(map, len);
6163 trace_insert_eval_map_file(mod, start, len);
6167 tracing_set_trace_read(struct file *filp, char __user *ubuf,
6168 size_t cnt, loff_t *ppos)
6170 struct trace_array *tr = filp->private_data;
6171 char buf[MAX_TRACER_SIZE+2];
6174 mutex_lock(&trace_types_lock);
6175 r = sprintf(buf, "%s\n", tr->current_trace->name);
6176 mutex_unlock(&trace_types_lock);
6178 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6181 int tracer_init(struct tracer *t, struct trace_array *tr)
6183 tracing_reset_online_cpus(&tr->array_buffer);
6187 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6191 for_each_tracing_cpu(cpu)
6192 per_cpu_ptr(buf->data, cpu)->entries = val;
6195 #ifdef CONFIG_TRACER_MAX_TRACE
6196 /* resize @tr's buffer to the size of @size_tr's entries */
6197 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6198 struct array_buffer *size_buf, int cpu_id)
6202 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6203 for_each_tracing_cpu(cpu) {
6204 ret = ring_buffer_resize(trace_buf->buffer,
6205 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6208 per_cpu_ptr(trace_buf->data, cpu)->entries =
6209 per_cpu_ptr(size_buf->data, cpu)->entries;
6212 ret = ring_buffer_resize(trace_buf->buffer,
6213 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6215 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6216 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6221 #endif /* CONFIG_TRACER_MAX_TRACE */
6223 static int __tracing_resize_ring_buffer(struct trace_array *tr,
6224 unsigned long size, int cpu)
6229 * If kernel or user changes the size of the ring buffer
6230 * we use the size that was given, and we can forget about
6231 * expanding it later.
6233 ring_buffer_expanded = true;
6235 /* May be called before buffers are initialized */
6236 if (!tr->array_buffer.buffer)
6239 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6243 #ifdef CONFIG_TRACER_MAX_TRACE
6244 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
6245 !tr->current_trace->use_max_tr)
6248 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6250 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6251 &tr->array_buffer, cpu);
6254 * AARGH! We are left with different
6255 * size max buffer!!!!
6256 * The max buffer is our "snapshot" buffer.
6257 * When a tracer needs a snapshot (one of the
6258 * latency tracers), it swaps the max buffer
6259 * with the saved snap shot. We succeeded to
6260 * update the size of the main buffer, but failed to
6261 * update the size of the max buffer. But when we tried
6262 * to reset the main buffer to the original size, we
6263 * failed there too. This is very unlikely to
6264 * happen, but if it does, warn and kill all
6268 tracing_disabled = 1;
6273 if (cpu == RING_BUFFER_ALL_CPUS)
6274 set_buffer_entries(&tr->max_buffer, size);
6276 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
6279 #endif /* CONFIG_TRACER_MAX_TRACE */
6281 if (cpu == RING_BUFFER_ALL_CPUS)
6282 set_buffer_entries(&tr->array_buffer, size);
6284 per_cpu_ptr(tr->array_buffer.data, cpu)->entries = size;
6289 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6290 unsigned long size, int cpu_id)
6294 mutex_lock(&trace_types_lock);
6296 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6297 /* make sure, this cpu is enabled in the mask */
6298 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6304 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6309 mutex_unlock(&trace_types_lock);
6316 * tracing_update_buffers - used by tracing facility to expand ring buffers
6318 * To save on memory when the tracing is never used on a system with it
6319 * configured in. The ring buffers are set to a minimum size. But once
6320 * a user starts to use the tracing facility, then they need to grow
6321 * to their default size.
6323 * This function is to be called when a tracer is about to be used.
6325 int tracing_update_buffers(void)
6329 mutex_lock(&trace_types_lock);
6330 if (!ring_buffer_expanded)
6331 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
6332 RING_BUFFER_ALL_CPUS);
6333 mutex_unlock(&trace_types_lock);
6338 struct trace_option_dentry;
6341 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6344 * Used to clear out the tracer before deletion of an instance.
6345 * Must have trace_types_lock held.
6347 static void tracing_set_nop(struct trace_array *tr)
6349 if (tr->current_trace == &nop_trace)
6352 tr->current_trace->enabled--;
6354 if (tr->current_trace->reset)
6355 tr->current_trace->reset(tr);
6357 tr->current_trace = &nop_trace;
6360 static bool tracer_options_updated;
6362 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6364 /* Only enable if the directory has been created already. */
6368 /* Only create trace option files after update_tracer_options finish */
6369 if (!tracer_options_updated)
6372 create_trace_option_files(tr, t);
6375 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6378 #ifdef CONFIG_TRACER_MAX_TRACE
6383 mutex_lock(&trace_types_lock);
6385 if (!ring_buffer_expanded) {
6386 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6387 RING_BUFFER_ALL_CPUS);
6393 for (t = trace_types; t; t = t->next) {
6394 if (strcmp(t->name, buf) == 0)
6401 if (t == tr->current_trace)
6404 #ifdef CONFIG_TRACER_SNAPSHOT
6405 if (t->use_max_tr) {
6406 local_irq_disable();
6407 arch_spin_lock(&tr->max_lock);
6408 if (tr->cond_snapshot)
6410 arch_spin_unlock(&tr->max_lock);
6416 /* Some tracers won't work on kernel command line */
6417 if (system_state < SYSTEM_RUNNING && t->noboot) {
6418 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6423 /* Some tracers are only allowed for the top level buffer */
6424 if (!trace_ok_for_array(t, tr)) {
6429 /* If trace pipe files are being read, we can't change the tracer */
6430 if (tr->trace_ref) {
6435 trace_branch_disable();
6437 tr->current_trace->enabled--;
6439 if (tr->current_trace->reset)
6440 tr->current_trace->reset(tr);
6442 #ifdef CONFIG_TRACER_MAX_TRACE
6443 had_max_tr = tr->current_trace->use_max_tr;
6445 /* Current trace needs to be nop_trace before synchronize_rcu */
6446 tr->current_trace = &nop_trace;
6448 if (had_max_tr && !t->use_max_tr) {
6450 * We need to make sure that the update_max_tr sees that
6451 * current_trace changed to nop_trace to keep it from
6452 * swapping the buffers after we resize it.
6453 * The update_max_tr is called from interrupts disabled
6454 * so a synchronized_sched() is sufficient.
6460 if (t->use_max_tr && !tr->allocated_snapshot) {
6461 ret = tracing_alloc_snapshot_instance(tr);
6466 tr->current_trace = &nop_trace;
6470 ret = tracer_init(t, tr);
6475 tr->current_trace = t;
6476 tr->current_trace->enabled++;
6477 trace_branch_enable(tr);
6479 mutex_unlock(&trace_types_lock);
6485 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6486 size_t cnt, loff_t *ppos)
6488 struct trace_array *tr = filp->private_data;
6489 char buf[MAX_TRACER_SIZE+1];
6496 if (cnt > MAX_TRACER_SIZE)
6497 cnt = MAX_TRACER_SIZE;
6499 if (copy_from_user(buf, ubuf, cnt))
6506 err = tracing_set_tracer(tr, name);
6516 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6517 size_t cnt, loff_t *ppos)
6522 r = snprintf(buf, sizeof(buf), "%ld\n",
6523 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6524 if (r > sizeof(buf))
6526 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6530 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6531 size_t cnt, loff_t *ppos)
6536 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6546 tracing_thresh_read(struct file *filp, char __user *ubuf,
6547 size_t cnt, loff_t *ppos)
6549 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6553 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6554 size_t cnt, loff_t *ppos)
6556 struct trace_array *tr = filp->private_data;
6559 mutex_lock(&trace_types_lock);
6560 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6564 if (tr->current_trace->update_thresh) {
6565 ret = tr->current_trace->update_thresh(tr);
6572 mutex_unlock(&trace_types_lock);
6577 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
6580 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6581 size_t cnt, loff_t *ppos)
6583 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
6587 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6588 size_t cnt, loff_t *ppos)
6590 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
6595 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6597 struct trace_array *tr = inode->i_private;
6598 struct trace_iterator *iter;
6601 ret = tracing_check_open_get_tr(tr);
6605 mutex_lock(&trace_types_lock);
6607 /* create a buffer to store the information to pass to userspace */
6608 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6611 __trace_array_put(tr);
6615 trace_seq_init(&iter->seq);
6616 iter->trace = tr->current_trace;
6618 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6623 /* trace pipe does not show start of buffer */
6624 cpumask_setall(iter->started);
6626 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6627 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6629 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6630 if (trace_clocks[tr->clock_id].in_ns)
6631 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6634 iter->array_buffer = &tr->array_buffer;
6635 iter->cpu_file = tracing_get_cpu(inode);
6636 mutex_init(&iter->mutex);
6637 filp->private_data = iter;
6639 if (iter->trace->pipe_open)
6640 iter->trace->pipe_open(iter);
6642 nonseekable_open(inode, filp);
6646 mutex_unlock(&trace_types_lock);
6651 __trace_array_put(tr);
6652 mutex_unlock(&trace_types_lock);
6656 static int tracing_release_pipe(struct inode *inode, struct file *file)
6658 struct trace_iterator *iter = file->private_data;
6659 struct trace_array *tr = inode->i_private;
6661 mutex_lock(&trace_types_lock);
6665 if (iter->trace->pipe_close)
6666 iter->trace->pipe_close(iter);
6668 mutex_unlock(&trace_types_lock);
6670 free_cpumask_var(iter->started);
6672 mutex_destroy(&iter->mutex);
6675 trace_array_put(tr);
6681 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6683 struct trace_array *tr = iter->tr;
6685 /* Iterators are static, they should be filled or empty */
6686 if (trace_buffer_iter(iter, iter->cpu_file))
6687 return EPOLLIN | EPOLLRDNORM;
6689 if (tr->trace_flags & TRACE_ITER_BLOCK)
6691 * Always select as readable when in blocking mode
6693 return EPOLLIN | EPOLLRDNORM;
6695 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6696 filp, poll_table, iter->tr->buffer_percent);
6700 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6702 struct trace_iterator *iter = filp->private_data;
6704 return trace_poll(iter, filp, poll_table);
6707 /* Must be called with iter->mutex held. */
6708 static int tracing_wait_pipe(struct file *filp)
6710 struct trace_iterator *iter = filp->private_data;
6713 while (trace_empty(iter)) {
6715 if ((filp->f_flags & O_NONBLOCK)) {
6720 * We block until we read something and tracing is disabled.
6721 * We still block if tracing is disabled, but we have never
6722 * read anything. This allows a user to cat this file, and
6723 * then enable tracing. But after we have read something,
6724 * we give an EOF when tracing is again disabled.
6726 * iter->pos will be 0 if we haven't read anything.
6728 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6731 mutex_unlock(&iter->mutex);
6733 ret = wait_on_pipe(iter, 0);
6735 mutex_lock(&iter->mutex);
6748 tracing_read_pipe(struct file *filp, char __user *ubuf,
6749 size_t cnt, loff_t *ppos)
6751 struct trace_iterator *iter = filp->private_data;
6755 * Avoid more than one consumer on a single file descriptor
6756 * This is just a matter of traces coherency, the ring buffer itself
6759 mutex_lock(&iter->mutex);
6761 /* return any leftover data */
6762 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6766 trace_seq_init(&iter->seq);
6768 if (iter->trace->read) {
6769 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6775 sret = tracing_wait_pipe(filp);
6779 /* stop when tracing is finished */
6780 if (trace_empty(iter)) {
6785 if (cnt >= PAGE_SIZE)
6786 cnt = PAGE_SIZE - 1;
6788 /* reset all but tr, trace, and overruns */
6789 trace_iterator_reset(iter);
6790 cpumask_clear(iter->started);
6791 trace_seq_init(&iter->seq);
6793 trace_event_read_lock();
6794 trace_access_lock(iter->cpu_file);
6795 while (trace_find_next_entry_inc(iter) != NULL) {
6796 enum print_line_t ret;
6797 int save_len = iter->seq.seq.len;
6799 ret = print_trace_line(iter);
6800 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6801 /* don't print partial lines */
6802 iter->seq.seq.len = save_len;
6805 if (ret != TRACE_TYPE_NO_CONSUME)
6806 trace_consume(iter);
6808 if (trace_seq_used(&iter->seq) >= cnt)
6812 * Setting the full flag means we reached the trace_seq buffer
6813 * size and we should leave by partial output condition above.
6814 * One of the trace_seq_* functions is not used properly.
6816 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
6819 trace_access_unlock(iter->cpu_file);
6820 trace_event_read_unlock();
6822 /* Now copy what we have to the user */
6823 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6824 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
6825 trace_seq_init(&iter->seq);
6828 * If there was nothing to send to user, in spite of consuming trace
6829 * entries, go back to wait for more entries.
6835 mutex_unlock(&iter->mutex);
6840 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
6843 __free_page(spd->pages[idx]);
6847 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
6853 /* Seq buffer is page-sized, exactly what we need. */
6855 save_len = iter->seq.seq.len;
6856 ret = print_trace_line(iter);
6858 if (trace_seq_has_overflowed(&iter->seq)) {
6859 iter->seq.seq.len = save_len;
6864 * This should not be hit, because it should only
6865 * be set if the iter->seq overflowed. But check it
6866 * anyway to be safe.
6868 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6869 iter->seq.seq.len = save_len;
6873 count = trace_seq_used(&iter->seq) - save_len;
6876 iter->seq.seq.len = save_len;
6880 if (ret != TRACE_TYPE_NO_CONSUME)
6881 trace_consume(iter);
6883 if (!trace_find_next_entry_inc(iter)) {
6893 static ssize_t tracing_splice_read_pipe(struct file *filp,
6895 struct pipe_inode_info *pipe,
6899 struct page *pages_def[PIPE_DEF_BUFFERS];
6900 struct partial_page partial_def[PIPE_DEF_BUFFERS];
6901 struct trace_iterator *iter = filp->private_data;
6902 struct splice_pipe_desc spd = {
6904 .partial = partial_def,
6905 .nr_pages = 0, /* This gets updated below. */
6906 .nr_pages_max = PIPE_DEF_BUFFERS,
6907 .ops = &default_pipe_buf_ops,
6908 .spd_release = tracing_spd_release_pipe,
6914 if (splice_grow_spd(pipe, &spd))
6917 mutex_lock(&iter->mutex);
6919 if (iter->trace->splice_read) {
6920 ret = iter->trace->splice_read(iter, filp,
6921 ppos, pipe, len, flags);
6926 ret = tracing_wait_pipe(filp);
6930 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
6935 trace_event_read_lock();
6936 trace_access_lock(iter->cpu_file);
6938 /* Fill as many pages as possible. */
6939 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
6940 spd.pages[i] = alloc_page(GFP_KERNEL);
6944 rem = tracing_fill_pipe_page(rem, iter);
6946 /* Copy the data into the page, so we can start over. */
6947 ret = trace_seq_to_buffer(&iter->seq,
6948 page_address(spd.pages[i]),
6949 trace_seq_used(&iter->seq));
6951 __free_page(spd.pages[i]);
6954 spd.partial[i].offset = 0;
6955 spd.partial[i].len = trace_seq_used(&iter->seq);
6957 trace_seq_init(&iter->seq);
6960 trace_access_unlock(iter->cpu_file);
6961 trace_event_read_unlock();
6962 mutex_unlock(&iter->mutex);
6967 ret = splice_to_pipe(pipe, &spd);
6971 splice_shrink_spd(&spd);
6975 mutex_unlock(&iter->mutex);
6980 tracing_entries_read(struct file *filp, char __user *ubuf,
6981 size_t cnt, loff_t *ppos)
6983 struct inode *inode = file_inode(filp);
6984 struct trace_array *tr = inode->i_private;
6985 int cpu = tracing_get_cpu(inode);
6990 mutex_lock(&trace_types_lock);
6992 if (cpu == RING_BUFFER_ALL_CPUS) {
6993 int cpu, buf_size_same;
6998 /* check if all cpu sizes are same */
6999 for_each_tracing_cpu(cpu) {
7000 /* fill in the size from first enabled cpu */
7002 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7003 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7009 if (buf_size_same) {
7010 if (!ring_buffer_expanded)
7011 r = sprintf(buf, "%lu (expanded: %lu)\n",
7013 trace_buf_size >> 10);
7015 r = sprintf(buf, "%lu\n", size >> 10);
7017 r = sprintf(buf, "X\n");
7019 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7021 mutex_unlock(&trace_types_lock);
7023 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7028 tracing_entries_write(struct file *filp, const char __user *ubuf,
7029 size_t cnt, loff_t *ppos)
7031 struct inode *inode = file_inode(filp);
7032 struct trace_array *tr = inode->i_private;
7036 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7040 /* must have at least 1 entry */
7044 /* value is in KB */
7046 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7056 tracing_total_entries_read(struct file *filp, char __user *ubuf,
7057 size_t cnt, loff_t *ppos)
7059 struct trace_array *tr = filp->private_data;
7062 unsigned long size = 0, expanded_size = 0;
7064 mutex_lock(&trace_types_lock);
7065 for_each_tracing_cpu(cpu) {
7066 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7067 if (!ring_buffer_expanded)
7068 expanded_size += trace_buf_size >> 10;
7070 if (ring_buffer_expanded)
7071 r = sprintf(buf, "%lu\n", size);
7073 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7074 mutex_unlock(&trace_types_lock);
7076 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7080 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7081 size_t cnt, loff_t *ppos)
7084 * There is no need to read what the user has written, this function
7085 * is just to make sure that there is no error when "echo" is used
7094 tracing_free_buffer_release(struct inode *inode, struct file *filp)
7096 struct trace_array *tr = inode->i_private;
7098 /* disable tracing ? */
7099 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7100 tracer_tracing_off(tr);
7101 /* resize the ring buffer to 0 */
7102 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7104 trace_array_put(tr);
7110 tracing_mark_write(struct file *filp, const char __user *ubuf,
7111 size_t cnt, loff_t *fpos)
7113 struct trace_array *tr = filp->private_data;
7114 struct ring_buffer_event *event;
7115 enum event_trigger_type tt = ETT_NONE;
7116 struct trace_buffer *buffer;
7117 struct print_entry *entry;
7122 /* Used in tracing_mark_raw_write() as well */
7123 #define FAULTED_STR "<faulted>"
7124 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7126 if (tracing_disabled)
7129 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7132 if (cnt > TRACE_BUF_SIZE)
7133 cnt = TRACE_BUF_SIZE;
7135 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7137 size = sizeof(*entry) + cnt + 2; /* add '\0' and possible '\n' */
7139 /* If less than "<faulted>", then make sure we can still add that */
7140 if (cnt < FAULTED_SIZE)
7141 size += FAULTED_SIZE - cnt;
7143 buffer = tr->array_buffer.buffer;
7144 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7146 if (unlikely(!event))
7147 /* Ring buffer disabled, return as if not open for write */
7150 entry = ring_buffer_event_data(event);
7151 entry->ip = _THIS_IP_;
7153 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7155 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7161 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7162 /* do not add \n before testing triggers, but add \0 */
7163 entry->buf[cnt] = '\0';
7164 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7167 if (entry->buf[cnt - 1] != '\n') {
7168 entry->buf[cnt] = '\n';
7169 entry->buf[cnt + 1] = '\0';
7171 entry->buf[cnt] = '\0';
7173 if (static_branch_unlikely(&trace_marker_exports_enabled))
7174 ftrace_exports(event, TRACE_EXPORT_MARKER);
7175 __buffer_unlock_commit(buffer, event);
7178 event_triggers_post_call(tr->trace_marker_file, tt);
7183 /* Limit it for now to 3K (including tag) */
7184 #define RAW_DATA_MAX_SIZE (1024*3)
7187 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7188 size_t cnt, loff_t *fpos)
7190 struct trace_array *tr = filp->private_data;
7191 struct ring_buffer_event *event;
7192 struct trace_buffer *buffer;
7193 struct raw_data_entry *entry;
7198 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7200 if (tracing_disabled)
7203 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7206 /* The marker must at least have a tag id */
7207 if (cnt < sizeof(unsigned int) || cnt > RAW_DATA_MAX_SIZE)
7210 if (cnt > TRACE_BUF_SIZE)
7211 cnt = TRACE_BUF_SIZE;
7213 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
7215 size = sizeof(*entry) + cnt;
7216 if (cnt < FAULT_SIZE_ID)
7217 size += FAULT_SIZE_ID - cnt;
7219 buffer = tr->array_buffer.buffer;
7220 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7223 /* Ring buffer disabled, return as if not open for write */
7226 entry = ring_buffer_event_data(event);
7228 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7231 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7236 __buffer_unlock_commit(buffer, event);
7241 static int tracing_clock_show(struct seq_file *m, void *v)
7243 struct trace_array *tr = m->private;
7246 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7248 "%s%s%s%s", i ? " " : "",
7249 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7250 i == tr->clock_id ? "]" : "");
7256 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7260 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7261 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7264 if (i == ARRAY_SIZE(trace_clocks))
7267 mutex_lock(&trace_types_lock);
7271 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7274 * New clock may not be consistent with the previous clock.
7275 * Reset the buffer so that it doesn't have incomparable timestamps.
7277 tracing_reset_online_cpus(&tr->array_buffer);
7279 #ifdef CONFIG_TRACER_MAX_TRACE
7280 if (tr->max_buffer.buffer)
7281 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7282 tracing_reset_online_cpus(&tr->max_buffer);
7285 mutex_unlock(&trace_types_lock);
7290 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7291 size_t cnt, loff_t *fpos)
7293 struct seq_file *m = filp->private_data;
7294 struct trace_array *tr = m->private;
7296 const char *clockstr;
7299 if (cnt >= sizeof(buf))
7302 if (copy_from_user(buf, ubuf, cnt))
7307 clockstr = strstrip(buf);
7309 ret = tracing_set_clock(tr, clockstr);
7318 static int tracing_clock_open(struct inode *inode, struct file *file)
7320 struct trace_array *tr = inode->i_private;
7323 ret = tracing_check_open_get_tr(tr);
7327 ret = single_open(file, tracing_clock_show, inode->i_private);
7329 trace_array_put(tr);
7334 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7336 struct trace_array *tr = m->private;
7338 mutex_lock(&trace_types_lock);
7340 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7341 seq_puts(m, "delta [absolute]\n");
7343 seq_puts(m, "[delta] absolute\n");
7345 mutex_unlock(&trace_types_lock);
7350 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7352 struct trace_array *tr = inode->i_private;
7355 ret = tracing_check_open_get_tr(tr);
7359 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7361 trace_array_put(tr);
7366 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7368 if (rbe == this_cpu_read(trace_buffered_event))
7369 return ring_buffer_time_stamp(buffer);
7371 return ring_buffer_event_time_stamp(buffer, rbe);
7375 * Set or disable using the per CPU trace_buffer_event when possible.
7377 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7381 mutex_lock(&trace_types_lock);
7383 if (set && tr->no_filter_buffering_ref++)
7387 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7392 --tr->no_filter_buffering_ref;
7395 mutex_unlock(&trace_types_lock);
7400 struct ftrace_buffer_info {
7401 struct trace_iterator iter;
7403 unsigned int spare_cpu;
7407 #ifdef CONFIG_TRACER_SNAPSHOT
7408 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7410 struct trace_array *tr = inode->i_private;
7411 struct trace_iterator *iter;
7415 ret = tracing_check_open_get_tr(tr);
7419 if (file->f_mode & FMODE_READ) {
7420 iter = __tracing_open(inode, file, true);
7422 ret = PTR_ERR(iter);
7424 /* Writes still need the seq_file to hold the private data */
7426 m = kzalloc(sizeof(*m), GFP_KERNEL);
7429 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7437 iter->array_buffer = &tr->max_buffer;
7438 iter->cpu_file = tracing_get_cpu(inode);
7440 file->private_data = m;
7444 trace_array_put(tr);
7450 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7453 struct seq_file *m = filp->private_data;
7454 struct trace_iterator *iter = m->private;
7455 struct trace_array *tr = iter->tr;
7459 ret = tracing_update_buffers();
7463 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7467 mutex_lock(&trace_types_lock);
7469 if (tr->current_trace->use_max_tr) {
7474 local_irq_disable();
7475 arch_spin_lock(&tr->max_lock);
7476 if (tr->cond_snapshot)
7478 arch_spin_unlock(&tr->max_lock);
7485 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7489 if (tr->allocated_snapshot)
7493 /* Only allow per-cpu swap if the ring buffer supports it */
7494 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7495 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7500 if (tr->allocated_snapshot)
7501 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7502 &tr->array_buffer, iter->cpu_file);
7504 ret = tracing_alloc_snapshot_instance(tr);
7507 local_irq_disable();
7508 /* Now, we're going to swap */
7509 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7510 update_max_tr(tr, current, smp_processor_id(), NULL);
7512 update_max_tr_single(tr, current, iter->cpu_file);
7516 if (tr->allocated_snapshot) {
7517 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7518 tracing_reset_online_cpus(&tr->max_buffer);
7520 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7530 mutex_unlock(&trace_types_lock);
7534 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7536 struct seq_file *m = file->private_data;
7539 ret = tracing_release(inode, file);
7541 if (file->f_mode & FMODE_READ)
7544 /* If write only, the seq_file is just a stub */
7552 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7553 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7554 size_t count, loff_t *ppos);
7555 static int tracing_buffers_release(struct inode *inode, struct file *file);
7556 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7557 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7559 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7561 struct ftrace_buffer_info *info;
7564 /* The following checks for tracefs lockdown */
7565 ret = tracing_buffers_open(inode, filp);
7569 info = filp->private_data;
7571 if (info->iter.trace->use_max_tr) {
7572 tracing_buffers_release(inode, filp);
7576 info->iter.snapshot = true;
7577 info->iter.array_buffer = &info->iter.tr->max_buffer;
7582 #endif /* CONFIG_TRACER_SNAPSHOT */
7585 static const struct file_operations tracing_thresh_fops = {
7586 .open = tracing_open_generic,
7587 .read = tracing_thresh_read,
7588 .write = tracing_thresh_write,
7589 .llseek = generic_file_llseek,
7592 #if defined(CONFIG_TRACER_MAX_TRACE) || defined(CONFIG_HWLAT_TRACER)
7593 static const struct file_operations tracing_max_lat_fops = {
7594 .open = tracing_open_generic,
7595 .read = tracing_max_lat_read,
7596 .write = tracing_max_lat_write,
7597 .llseek = generic_file_llseek,
7601 static const struct file_operations set_tracer_fops = {
7602 .open = tracing_open_generic,
7603 .read = tracing_set_trace_read,
7604 .write = tracing_set_trace_write,
7605 .llseek = generic_file_llseek,
7608 static const struct file_operations tracing_pipe_fops = {
7609 .open = tracing_open_pipe,
7610 .poll = tracing_poll_pipe,
7611 .read = tracing_read_pipe,
7612 .splice_read = tracing_splice_read_pipe,
7613 .release = tracing_release_pipe,
7614 .llseek = no_llseek,
7617 static const struct file_operations tracing_entries_fops = {
7618 .open = tracing_open_generic_tr,
7619 .read = tracing_entries_read,
7620 .write = tracing_entries_write,
7621 .llseek = generic_file_llseek,
7622 .release = tracing_release_generic_tr,
7625 static const struct file_operations tracing_total_entries_fops = {
7626 .open = tracing_open_generic_tr,
7627 .read = tracing_total_entries_read,
7628 .llseek = generic_file_llseek,
7629 .release = tracing_release_generic_tr,
7632 static const struct file_operations tracing_free_buffer_fops = {
7633 .open = tracing_open_generic_tr,
7634 .write = tracing_free_buffer_write,
7635 .release = tracing_free_buffer_release,
7638 static const struct file_operations tracing_mark_fops = {
7639 .open = tracing_mark_open,
7640 .write = tracing_mark_write,
7641 .release = tracing_release_generic_tr,
7644 static const struct file_operations tracing_mark_raw_fops = {
7645 .open = tracing_mark_open,
7646 .write = tracing_mark_raw_write,
7647 .release = tracing_release_generic_tr,
7650 static const struct file_operations trace_clock_fops = {
7651 .open = tracing_clock_open,
7653 .llseek = seq_lseek,
7654 .release = tracing_single_release_tr,
7655 .write = tracing_clock_write,
7658 static const struct file_operations trace_time_stamp_mode_fops = {
7659 .open = tracing_time_stamp_mode_open,
7661 .llseek = seq_lseek,
7662 .release = tracing_single_release_tr,
7665 #ifdef CONFIG_TRACER_SNAPSHOT
7666 static const struct file_operations snapshot_fops = {
7667 .open = tracing_snapshot_open,
7669 .write = tracing_snapshot_write,
7670 .llseek = tracing_lseek,
7671 .release = tracing_snapshot_release,
7674 static const struct file_operations snapshot_raw_fops = {
7675 .open = snapshot_raw_open,
7676 .read = tracing_buffers_read,
7677 .release = tracing_buffers_release,
7678 .splice_read = tracing_buffers_splice_read,
7679 .llseek = no_llseek,
7682 #endif /* CONFIG_TRACER_SNAPSHOT */
7685 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7686 * @filp: The active open file structure
7687 * @ubuf: The userspace provided buffer to read value into
7688 * @cnt: The maximum number of bytes to read
7689 * @ppos: The current "file" position
7691 * This function implements the write interface for a struct trace_min_max_param.
7692 * The filp->private_data must point to a trace_min_max_param structure that
7693 * defines where to write the value, the min and the max acceptable values,
7694 * and a lock to protect the write.
7697 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7699 struct trace_min_max_param *param = filp->private_data;
7706 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7711 mutex_lock(param->lock);
7713 if (param->min && val < *param->min)
7716 if (param->max && val > *param->max)
7723 mutex_unlock(param->lock);
7732 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7733 * @filp: The active open file structure
7734 * @ubuf: The userspace provided buffer to read value into
7735 * @cnt: The maximum number of bytes to read
7736 * @ppos: The current "file" position
7738 * This function implements the read interface for a struct trace_min_max_param.
7739 * The filp->private_data must point to a trace_min_max_param struct with valid
7743 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7745 struct trace_min_max_param *param = filp->private_data;
7746 char buf[U64_STR_SIZE];
7755 if (cnt > sizeof(buf))
7758 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7760 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7763 const struct file_operations trace_min_max_fops = {
7764 .open = tracing_open_generic,
7765 .read = trace_min_max_read,
7766 .write = trace_min_max_write,
7769 #define TRACING_LOG_ERRS_MAX 8
7770 #define TRACING_LOG_LOC_MAX 128
7772 #define CMD_PREFIX " Command: "
7775 const char **errs; /* ptr to loc-specific array of err strings */
7776 u8 type; /* index into errs -> specific err string */
7777 u16 pos; /* caret position */
7781 struct tracing_log_err {
7782 struct list_head list;
7783 struct err_info info;
7784 char loc[TRACING_LOG_LOC_MAX]; /* err location */
7785 char *cmd; /* what caused err */
7788 static DEFINE_MUTEX(tracing_err_log_lock);
7790 static struct tracing_log_err *alloc_tracing_log_err(int len)
7792 struct tracing_log_err *err;
7794 err = kzalloc(sizeof(*err), GFP_KERNEL);
7796 return ERR_PTR(-ENOMEM);
7798 err->cmd = kzalloc(len, GFP_KERNEL);
7801 return ERR_PTR(-ENOMEM);
7807 static void free_tracing_log_err(struct tracing_log_err *err)
7813 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
7816 struct tracing_log_err *err;
7819 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
7820 err = alloc_tracing_log_err(len);
7821 if (PTR_ERR(err) != -ENOMEM)
7822 tr->n_err_log_entries++;
7826 cmd = kzalloc(len, GFP_KERNEL);
7828 return ERR_PTR(-ENOMEM);
7829 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
7832 list_del(&err->list);
7838 * err_pos - find the position of a string within a command for error careting
7839 * @cmd: The tracing command that caused the error
7840 * @str: The string to position the caret at within @cmd
7842 * Finds the position of the first occurrence of @str within @cmd. The
7843 * return value can be passed to tracing_log_err() for caret placement
7846 * Returns the index within @cmd of the first occurrence of @str or 0
7847 * if @str was not found.
7849 unsigned int err_pos(char *cmd, const char *str)
7853 if (WARN_ON(!strlen(cmd)))
7856 found = strstr(cmd, str);
7864 * tracing_log_err - write an error to the tracing error log
7865 * @tr: The associated trace array for the error (NULL for top level array)
7866 * @loc: A string describing where the error occurred
7867 * @cmd: The tracing command that caused the error
7868 * @errs: The array of loc-specific static error strings
7869 * @type: The index into errs[], which produces the specific static err string
7870 * @pos: The position the caret should be placed in the cmd
7872 * Writes an error into tracing/error_log of the form:
7874 * <loc>: error: <text>
7878 * tracing/error_log is a small log file containing the last
7879 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
7880 * unless there has been a tracing error, and the error log can be
7881 * cleared and have its memory freed by writing the empty string in
7882 * truncation mode to it i.e. echo > tracing/error_log.
7884 * NOTE: the @errs array along with the @type param are used to
7885 * produce a static error string - this string is not copied and saved
7886 * when the error is logged - only a pointer to it is saved. See
7887 * existing callers for examples of how static strings are typically
7888 * defined for use with tracing_log_err().
7890 void tracing_log_err(struct trace_array *tr,
7891 const char *loc, const char *cmd,
7892 const char **errs, u8 type, u16 pos)
7894 struct tracing_log_err *err;
7900 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
7902 mutex_lock(&tracing_err_log_lock);
7903 err = get_tracing_log_err(tr, len);
7904 if (PTR_ERR(err) == -ENOMEM) {
7905 mutex_unlock(&tracing_err_log_lock);
7909 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
7910 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
7912 err->info.errs = errs;
7913 err->info.type = type;
7914 err->info.pos = pos;
7915 err->info.ts = local_clock();
7917 list_add_tail(&err->list, &tr->err_log);
7918 mutex_unlock(&tracing_err_log_lock);
7921 static void clear_tracing_err_log(struct trace_array *tr)
7923 struct tracing_log_err *err, *next;
7925 mutex_lock(&tracing_err_log_lock);
7926 list_for_each_entry_safe(err, next, &tr->err_log, list) {
7927 list_del(&err->list);
7928 free_tracing_log_err(err);
7931 tr->n_err_log_entries = 0;
7932 mutex_unlock(&tracing_err_log_lock);
7935 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
7937 struct trace_array *tr = m->private;
7939 mutex_lock(&tracing_err_log_lock);
7941 return seq_list_start(&tr->err_log, *pos);
7944 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
7946 struct trace_array *tr = m->private;
7948 return seq_list_next(v, &tr->err_log, pos);
7951 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
7953 mutex_unlock(&tracing_err_log_lock);
7956 static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
7960 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
7962 for (i = 0; i < pos; i++)
7967 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
7969 struct tracing_log_err *err = v;
7972 const char *err_text = err->info.errs[err->info.type];
7973 u64 sec = err->info.ts;
7976 nsec = do_div(sec, NSEC_PER_SEC);
7977 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
7978 err->loc, err_text);
7979 seq_printf(m, "%s", err->cmd);
7980 tracing_err_log_show_pos(m, err->info.pos);
7986 static const struct seq_operations tracing_err_log_seq_ops = {
7987 .start = tracing_err_log_seq_start,
7988 .next = tracing_err_log_seq_next,
7989 .stop = tracing_err_log_seq_stop,
7990 .show = tracing_err_log_seq_show
7993 static int tracing_err_log_open(struct inode *inode, struct file *file)
7995 struct trace_array *tr = inode->i_private;
7998 ret = tracing_check_open_get_tr(tr);
8002 /* If this file was opened for write, then erase contents */
8003 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8004 clear_tracing_err_log(tr);
8006 if (file->f_mode & FMODE_READ) {
8007 ret = seq_open(file, &tracing_err_log_seq_ops);
8009 struct seq_file *m = file->private_data;
8012 trace_array_put(tr);
8018 static ssize_t tracing_err_log_write(struct file *file,
8019 const char __user *buffer,
8020 size_t count, loff_t *ppos)
8025 static int tracing_err_log_release(struct inode *inode, struct file *file)
8027 struct trace_array *tr = inode->i_private;
8029 trace_array_put(tr);
8031 if (file->f_mode & FMODE_READ)
8032 seq_release(inode, file);
8037 static const struct file_operations tracing_err_log_fops = {
8038 .open = tracing_err_log_open,
8039 .write = tracing_err_log_write,
8041 .llseek = seq_lseek,
8042 .release = tracing_err_log_release,
8045 static int tracing_buffers_open(struct inode *inode, struct file *filp)
8047 struct trace_array *tr = inode->i_private;
8048 struct ftrace_buffer_info *info;
8051 ret = tracing_check_open_get_tr(tr);
8055 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8057 trace_array_put(tr);
8061 mutex_lock(&trace_types_lock);
8064 info->iter.cpu_file = tracing_get_cpu(inode);
8065 info->iter.trace = tr->current_trace;
8066 info->iter.array_buffer = &tr->array_buffer;
8068 /* Force reading ring buffer for first read */
8069 info->read = (unsigned int)-1;
8071 filp->private_data = info;
8075 mutex_unlock(&trace_types_lock);
8077 ret = nonseekable_open(inode, filp);
8079 trace_array_put(tr);
8085 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8087 struct ftrace_buffer_info *info = filp->private_data;
8088 struct trace_iterator *iter = &info->iter;
8090 return trace_poll(iter, filp, poll_table);
8094 tracing_buffers_read(struct file *filp, char __user *ubuf,
8095 size_t count, loff_t *ppos)
8097 struct ftrace_buffer_info *info = filp->private_data;
8098 struct trace_iterator *iter = &info->iter;
8105 #ifdef CONFIG_TRACER_MAX_TRACE
8106 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8111 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8113 if (IS_ERR(info->spare)) {
8114 ret = PTR_ERR(info->spare);
8117 info->spare_cpu = iter->cpu_file;
8123 /* Do we have previous read data to read? */
8124 if (info->read < PAGE_SIZE)
8128 trace_access_lock(iter->cpu_file);
8129 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8133 trace_access_unlock(iter->cpu_file);
8136 if (trace_empty(iter)) {
8137 if ((filp->f_flags & O_NONBLOCK))
8140 ret = wait_on_pipe(iter, 0);
8151 size = PAGE_SIZE - info->read;
8155 ret = copy_to_user(ubuf, info->spare + info->read, size);
8167 static int tracing_buffers_release(struct inode *inode, struct file *file)
8169 struct ftrace_buffer_info *info = file->private_data;
8170 struct trace_iterator *iter = &info->iter;
8172 mutex_lock(&trace_types_lock);
8174 iter->tr->trace_ref--;
8176 __trace_array_put(iter->tr);
8179 /* Make sure the waiters see the new wait_index */
8182 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8185 ring_buffer_free_read_page(iter->array_buffer->buffer,
8186 info->spare_cpu, info->spare);
8189 mutex_unlock(&trace_types_lock);
8195 struct trace_buffer *buffer;
8198 refcount_t refcount;
8201 static void buffer_ref_release(struct buffer_ref *ref)
8203 if (!refcount_dec_and_test(&ref->refcount))
8205 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8209 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8210 struct pipe_buffer *buf)
8212 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8214 buffer_ref_release(ref);
8218 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8219 struct pipe_buffer *buf)
8221 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8223 if (refcount_read(&ref->refcount) > INT_MAX/2)
8226 refcount_inc(&ref->refcount);
8230 /* Pipe buffer operations for a buffer. */
8231 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8232 .release = buffer_pipe_buf_release,
8233 .get = buffer_pipe_buf_get,
8237 * Callback from splice_to_pipe(), if we need to release some pages
8238 * at the end of the spd in case we error'ed out in filling the pipe.
8240 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8242 struct buffer_ref *ref =
8243 (struct buffer_ref *)spd->partial[i].private;
8245 buffer_ref_release(ref);
8246 spd->partial[i].private = 0;
8250 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8251 struct pipe_inode_info *pipe, size_t len,
8254 struct ftrace_buffer_info *info = file->private_data;
8255 struct trace_iterator *iter = &info->iter;
8256 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8257 struct page *pages_def[PIPE_DEF_BUFFERS];
8258 struct splice_pipe_desc spd = {
8260 .partial = partial_def,
8261 .nr_pages_max = PIPE_DEF_BUFFERS,
8262 .ops = &buffer_pipe_buf_ops,
8263 .spd_release = buffer_spd_release,
8265 struct buffer_ref *ref;
8269 #ifdef CONFIG_TRACER_MAX_TRACE
8270 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8274 if (*ppos & (PAGE_SIZE - 1))
8277 if (len & (PAGE_SIZE - 1)) {
8278 if (len < PAGE_SIZE)
8283 if (splice_grow_spd(pipe, &spd))
8287 trace_access_lock(iter->cpu_file);
8288 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8290 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
8294 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8300 refcount_set(&ref->refcount, 1);
8301 ref->buffer = iter->array_buffer->buffer;
8302 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8303 if (IS_ERR(ref->page)) {
8304 ret = PTR_ERR(ref->page);
8309 ref->cpu = iter->cpu_file;
8311 r = ring_buffer_read_page(ref->buffer, &ref->page,
8312 len, iter->cpu_file, 1);
8314 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8320 page = virt_to_page(ref->page);
8322 spd.pages[i] = page;
8323 spd.partial[i].len = PAGE_SIZE;
8324 spd.partial[i].offset = 0;
8325 spd.partial[i].private = (unsigned long)ref;
8329 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8332 trace_access_unlock(iter->cpu_file);
8335 /* did we read anything? */
8336 if (!spd.nr_pages) {
8343 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8346 wait_index = READ_ONCE(iter->wait_index);
8348 ret = wait_on_pipe(iter, iter->tr->buffer_percent);
8352 /* No need to wait after waking up when tracing is off */
8353 if (!tracer_tracing_is_on(iter->tr))
8356 /* Make sure we see the new wait_index */
8358 if (wait_index != iter->wait_index)
8364 ret = splice_to_pipe(pipe, &spd);
8366 splice_shrink_spd(&spd);
8371 /* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
8372 static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8374 struct ftrace_buffer_info *info = file->private_data;
8375 struct trace_iterator *iter = &info->iter;
8378 return -ENOIOCTLCMD;
8380 mutex_lock(&trace_types_lock);
8383 /* Make sure the waiters see the new wait_index */
8386 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8388 mutex_unlock(&trace_types_lock);
8392 static const struct file_operations tracing_buffers_fops = {
8393 .open = tracing_buffers_open,
8394 .read = tracing_buffers_read,
8395 .poll = tracing_buffers_poll,
8396 .release = tracing_buffers_release,
8397 .splice_read = tracing_buffers_splice_read,
8398 .unlocked_ioctl = tracing_buffers_ioctl,
8399 .llseek = no_llseek,
8403 tracing_stats_read(struct file *filp, char __user *ubuf,
8404 size_t count, loff_t *ppos)
8406 struct inode *inode = file_inode(filp);
8407 struct trace_array *tr = inode->i_private;
8408 struct array_buffer *trace_buf = &tr->array_buffer;
8409 int cpu = tracing_get_cpu(inode);
8410 struct trace_seq *s;
8412 unsigned long long t;
8413 unsigned long usec_rem;
8415 s = kmalloc(sizeof(*s), GFP_KERNEL);
8421 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8422 trace_seq_printf(s, "entries: %ld\n", cnt);
8424 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8425 trace_seq_printf(s, "overrun: %ld\n", cnt);
8427 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8428 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8430 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8431 trace_seq_printf(s, "bytes: %ld\n", cnt);
8433 if (trace_clocks[tr->clock_id].in_ns) {
8434 /* local or global for trace_clock */
8435 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8436 usec_rem = do_div(t, USEC_PER_SEC);
8437 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8440 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8441 usec_rem = do_div(t, USEC_PER_SEC);
8442 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8444 /* counter or tsc mode for trace_clock */
8445 trace_seq_printf(s, "oldest event ts: %llu\n",
8446 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8448 trace_seq_printf(s, "now ts: %llu\n",
8449 ring_buffer_time_stamp(trace_buf->buffer));
8452 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8453 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8455 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8456 trace_seq_printf(s, "read events: %ld\n", cnt);
8458 count = simple_read_from_buffer(ubuf, count, ppos,
8459 s->buffer, trace_seq_used(s));
8466 static const struct file_operations tracing_stats_fops = {
8467 .open = tracing_open_generic_tr,
8468 .read = tracing_stats_read,
8469 .llseek = generic_file_llseek,
8470 .release = tracing_release_generic_tr,
8473 #ifdef CONFIG_DYNAMIC_FTRACE
8476 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8477 size_t cnt, loff_t *ppos)
8483 /* 256 should be plenty to hold the amount needed */
8484 buf = kmalloc(256, GFP_KERNEL);
8488 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8489 ftrace_update_tot_cnt,
8490 ftrace_number_of_pages,
8491 ftrace_number_of_groups);
8493 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8498 static const struct file_operations tracing_dyn_info_fops = {
8499 .open = tracing_open_generic,
8500 .read = tracing_read_dyn_info,
8501 .llseek = generic_file_llseek,
8503 #endif /* CONFIG_DYNAMIC_FTRACE */
8505 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8507 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8508 struct trace_array *tr, struct ftrace_probe_ops *ops,
8511 tracing_snapshot_instance(tr);
8515 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8516 struct trace_array *tr, struct ftrace_probe_ops *ops,
8519 struct ftrace_func_mapper *mapper = data;
8523 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8533 tracing_snapshot_instance(tr);
8537 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8538 struct ftrace_probe_ops *ops, void *data)
8540 struct ftrace_func_mapper *mapper = data;
8543 seq_printf(m, "%ps:", (void *)ip);
8545 seq_puts(m, "snapshot");
8548 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8551 seq_printf(m, ":count=%ld\n", *count);
8553 seq_puts(m, ":unlimited\n");
8559 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8560 unsigned long ip, void *init_data, void **data)
8562 struct ftrace_func_mapper *mapper = *data;
8565 mapper = allocate_ftrace_func_mapper();
8571 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8575 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8576 unsigned long ip, void *data)
8578 struct ftrace_func_mapper *mapper = data;
8583 free_ftrace_func_mapper(mapper, NULL);
8587 ftrace_func_mapper_remove_ip(mapper, ip);
8590 static struct ftrace_probe_ops snapshot_probe_ops = {
8591 .func = ftrace_snapshot,
8592 .print = ftrace_snapshot_print,
8595 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8596 .func = ftrace_count_snapshot,
8597 .print = ftrace_snapshot_print,
8598 .init = ftrace_snapshot_init,
8599 .free = ftrace_snapshot_free,
8603 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8604 char *glob, char *cmd, char *param, int enable)
8606 struct ftrace_probe_ops *ops;
8607 void *count = (void *)-1;
8614 /* hash funcs only work with set_ftrace_filter */
8618 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8621 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8626 number = strsep(¶m, ":");
8628 if (!strlen(number))
8632 * We use the callback data field (which is a pointer)
8635 ret = kstrtoul(number, 0, (unsigned long *)&count);
8640 ret = tracing_alloc_snapshot_instance(tr);
8644 ret = register_ftrace_function_probe(glob, tr, ops, count);
8647 return ret < 0 ? ret : 0;
8650 static struct ftrace_func_command ftrace_snapshot_cmd = {
8652 .func = ftrace_trace_snapshot_callback,
8655 static __init int register_snapshot_cmd(void)
8657 return register_ftrace_command(&ftrace_snapshot_cmd);
8660 static inline __init int register_snapshot_cmd(void) { return 0; }
8661 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8663 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8665 if (WARN_ON(!tr->dir))
8666 return ERR_PTR(-ENODEV);
8668 /* Top directory uses NULL as the parent */
8669 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8672 /* All sub buffers have a descriptor */
8676 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8678 struct dentry *d_tracer;
8681 return tr->percpu_dir;
8683 d_tracer = tracing_get_dentry(tr);
8684 if (IS_ERR(d_tracer))
8687 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8689 MEM_FAIL(!tr->percpu_dir,
8690 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8692 return tr->percpu_dir;
8695 static struct dentry *
8696 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8697 void *data, long cpu, const struct file_operations *fops)
8699 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8701 if (ret) /* See tracing_get_cpu() */
8702 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8707 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8709 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8710 struct dentry *d_cpu;
8711 char cpu_dir[30]; /* 30 characters should be more than enough */
8716 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8717 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8719 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8723 /* per cpu trace_pipe */
8724 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8725 tr, cpu, &tracing_pipe_fops);
8728 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8729 tr, cpu, &tracing_fops);
8731 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8732 tr, cpu, &tracing_buffers_fops);
8734 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8735 tr, cpu, &tracing_stats_fops);
8737 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8738 tr, cpu, &tracing_entries_fops);
8740 #ifdef CONFIG_TRACER_SNAPSHOT
8741 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8742 tr, cpu, &snapshot_fops);
8744 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8745 tr, cpu, &snapshot_raw_fops);
8749 #ifdef CONFIG_FTRACE_SELFTEST
8750 /* Let selftest have access to static functions in this file */
8751 #include "trace_selftest.c"
8755 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8758 struct trace_option_dentry *topt = filp->private_data;
8761 if (topt->flags->val & topt->opt->bit)
8766 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8770 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
8773 struct trace_option_dentry *topt = filp->private_data;
8777 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8781 if (val != 0 && val != 1)
8784 if (!!(topt->flags->val & topt->opt->bit) != val) {
8785 mutex_lock(&trace_types_lock);
8786 ret = __set_tracer_option(topt->tr, topt->flags,
8788 mutex_unlock(&trace_types_lock);
8799 static const struct file_operations trace_options_fops = {
8800 .open = tracing_open_generic,
8801 .read = trace_options_read,
8802 .write = trace_options_write,
8803 .llseek = generic_file_llseek,
8807 * In order to pass in both the trace_array descriptor as well as the index
8808 * to the flag that the trace option file represents, the trace_array
8809 * has a character array of trace_flags_index[], which holds the index
8810 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
8811 * The address of this character array is passed to the flag option file
8812 * read/write callbacks.
8814 * In order to extract both the index and the trace_array descriptor,
8815 * get_tr_index() uses the following algorithm.
8819 * As the pointer itself contains the address of the index (remember
8822 * Then to get the trace_array descriptor, by subtracting that index
8823 * from the ptr, we get to the start of the index itself.
8825 * ptr - idx == &index[0]
8827 * Then a simple container_of() from that pointer gets us to the
8828 * trace_array descriptor.
8830 static void get_tr_index(void *data, struct trace_array **ptr,
8831 unsigned int *pindex)
8833 *pindex = *(unsigned char *)data;
8835 *ptr = container_of(data - *pindex, struct trace_array,
8840 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
8843 void *tr_index = filp->private_data;
8844 struct trace_array *tr;
8848 get_tr_index(tr_index, &tr, &index);
8850 if (tr->trace_flags & (1 << index))
8855 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
8859 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
8862 void *tr_index = filp->private_data;
8863 struct trace_array *tr;
8868 get_tr_index(tr_index, &tr, &index);
8870 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
8874 if (val != 0 && val != 1)
8877 mutex_lock(&event_mutex);
8878 mutex_lock(&trace_types_lock);
8879 ret = set_tracer_flag(tr, 1 << index, val);
8880 mutex_unlock(&trace_types_lock);
8881 mutex_unlock(&event_mutex);
8891 static const struct file_operations trace_options_core_fops = {
8892 .open = tracing_open_generic,
8893 .read = trace_options_core_read,
8894 .write = trace_options_core_write,
8895 .llseek = generic_file_llseek,
8898 struct dentry *trace_create_file(const char *name,
8900 struct dentry *parent,
8902 const struct file_operations *fops)
8906 ret = tracefs_create_file(name, mode, parent, data, fops);
8908 pr_warn("Could not create tracefs '%s' entry\n", name);
8914 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
8916 struct dentry *d_tracer;
8921 d_tracer = tracing_get_dentry(tr);
8922 if (IS_ERR(d_tracer))
8925 tr->options = tracefs_create_dir("options", d_tracer);
8927 pr_warn("Could not create tracefs directory 'options'\n");
8935 create_trace_option_file(struct trace_array *tr,
8936 struct trace_option_dentry *topt,
8937 struct tracer_flags *flags,
8938 struct tracer_opt *opt)
8940 struct dentry *t_options;
8942 t_options = trace_options_init_dentry(tr);
8946 topt->flags = flags;
8950 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
8951 t_options, topt, &trace_options_fops);
8956 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
8958 struct trace_option_dentry *topts;
8959 struct trace_options *tr_topts;
8960 struct tracer_flags *flags;
8961 struct tracer_opt *opts;
8968 flags = tracer->flags;
8970 if (!flags || !flags->opts)
8974 * If this is an instance, only create flags for tracers
8975 * the instance may have.
8977 if (!trace_ok_for_array(tracer, tr))
8980 for (i = 0; i < tr->nr_topts; i++) {
8981 /* Make sure there's no duplicate flags. */
8982 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
8988 for (cnt = 0; opts[cnt].name; cnt++)
8991 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
8995 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9002 tr->topts = tr_topts;
9003 tr->topts[tr->nr_topts].tracer = tracer;
9004 tr->topts[tr->nr_topts].topts = topts;
9007 for (cnt = 0; opts[cnt].name; cnt++) {
9008 create_trace_option_file(tr, &topts[cnt], flags,
9010 MEM_FAIL(topts[cnt].entry == NULL,
9011 "Failed to create trace option: %s",
9016 static struct dentry *
9017 create_trace_option_core_file(struct trace_array *tr,
9018 const char *option, long index)
9020 struct dentry *t_options;
9022 t_options = trace_options_init_dentry(tr);
9026 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9027 (void *)&tr->trace_flags_index[index],
9028 &trace_options_core_fops);
9031 static void create_trace_options_dir(struct trace_array *tr)
9033 struct dentry *t_options;
9034 bool top_level = tr == &global_trace;
9037 t_options = trace_options_init_dentry(tr);
9041 for (i = 0; trace_options[i]; i++) {
9043 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9044 create_trace_option_core_file(tr, trace_options[i], i);
9049 rb_simple_read(struct file *filp, char __user *ubuf,
9050 size_t cnt, loff_t *ppos)
9052 struct trace_array *tr = filp->private_data;
9056 r = tracer_tracing_is_on(tr);
9057 r = sprintf(buf, "%d\n", r);
9059 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9063 rb_simple_write(struct file *filp, const char __user *ubuf,
9064 size_t cnt, loff_t *ppos)
9066 struct trace_array *tr = filp->private_data;
9067 struct trace_buffer *buffer = tr->array_buffer.buffer;
9071 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9076 mutex_lock(&trace_types_lock);
9077 if (!!val == tracer_tracing_is_on(tr)) {
9078 val = 0; /* do nothing */
9080 tracer_tracing_on(tr);
9081 if (tr->current_trace->start)
9082 tr->current_trace->start(tr);
9084 tracer_tracing_off(tr);
9085 if (tr->current_trace->stop)
9086 tr->current_trace->stop(tr);
9087 /* Wake up any waiters */
9088 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9090 mutex_unlock(&trace_types_lock);
9098 static const struct file_operations rb_simple_fops = {
9099 .open = tracing_open_generic_tr,
9100 .read = rb_simple_read,
9101 .write = rb_simple_write,
9102 .release = tracing_release_generic_tr,
9103 .llseek = default_llseek,
9107 buffer_percent_read(struct file *filp, char __user *ubuf,
9108 size_t cnt, loff_t *ppos)
9110 struct trace_array *tr = filp->private_data;
9114 r = tr->buffer_percent;
9115 r = sprintf(buf, "%d\n", r);
9117 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9121 buffer_percent_write(struct file *filp, const char __user *ubuf,
9122 size_t cnt, loff_t *ppos)
9124 struct trace_array *tr = filp->private_data;
9128 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9138 tr->buffer_percent = val;
9145 static const struct file_operations buffer_percent_fops = {
9146 .open = tracing_open_generic_tr,
9147 .read = buffer_percent_read,
9148 .write = buffer_percent_write,
9149 .release = tracing_release_generic_tr,
9150 .llseek = default_llseek,
9153 static struct dentry *trace_instance_dir;
9156 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9159 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9161 enum ring_buffer_flags rb_flags;
9163 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9167 buf->buffer = ring_buffer_alloc(size, rb_flags);
9171 buf->data = alloc_percpu(struct trace_array_cpu);
9173 ring_buffer_free(buf->buffer);
9178 /* Allocate the first page for all buffers */
9179 set_buffer_entries(&tr->array_buffer,
9180 ring_buffer_size(tr->array_buffer.buffer, 0));
9185 static void free_trace_buffer(struct array_buffer *buf)
9188 ring_buffer_free(buf->buffer);
9190 free_percpu(buf->data);
9195 static int allocate_trace_buffers(struct trace_array *tr, int size)
9199 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9203 #ifdef CONFIG_TRACER_MAX_TRACE
9204 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9205 allocate_snapshot ? size : 1);
9206 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9207 free_trace_buffer(&tr->array_buffer);
9210 tr->allocated_snapshot = allocate_snapshot;
9213 * Only the top level trace array gets its snapshot allocated
9214 * from the kernel command line.
9216 allocate_snapshot = false;
9222 static void free_trace_buffers(struct trace_array *tr)
9227 free_trace_buffer(&tr->array_buffer);
9229 #ifdef CONFIG_TRACER_MAX_TRACE
9230 free_trace_buffer(&tr->max_buffer);
9234 static void init_trace_flags_index(struct trace_array *tr)
9238 /* Used by the trace options files */
9239 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9240 tr->trace_flags_index[i] = i;
9243 static void __update_tracer_options(struct trace_array *tr)
9247 for (t = trace_types; t; t = t->next)
9248 add_tracer_options(tr, t);
9251 static void update_tracer_options(struct trace_array *tr)
9253 mutex_lock(&trace_types_lock);
9254 tracer_options_updated = true;
9255 __update_tracer_options(tr);
9256 mutex_unlock(&trace_types_lock);
9259 /* Must have trace_types_lock held */
9260 struct trace_array *trace_array_find(const char *instance)
9262 struct trace_array *tr, *found = NULL;
9264 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9265 if (tr->name && strcmp(tr->name, instance) == 0) {
9274 struct trace_array *trace_array_find_get(const char *instance)
9276 struct trace_array *tr;
9278 mutex_lock(&trace_types_lock);
9279 tr = trace_array_find(instance);
9282 mutex_unlock(&trace_types_lock);
9287 static int trace_array_create_dir(struct trace_array *tr)
9291 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9295 ret = event_trace_add_tracer(tr->dir, tr);
9297 tracefs_remove(tr->dir);
9301 init_tracer_tracefs(tr, tr->dir);
9302 __update_tracer_options(tr);
9307 static struct trace_array *trace_array_create(const char *name)
9309 struct trace_array *tr;
9313 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9315 return ERR_PTR(ret);
9317 tr->name = kstrdup(name, GFP_KERNEL);
9321 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9324 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9326 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9328 raw_spin_lock_init(&tr->start_lock);
9330 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9332 tr->current_trace = &nop_trace;
9334 INIT_LIST_HEAD(&tr->systems);
9335 INIT_LIST_HEAD(&tr->events);
9336 INIT_LIST_HEAD(&tr->hist_vars);
9337 INIT_LIST_HEAD(&tr->err_log);
9339 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9342 if (ftrace_allocate_ftrace_ops(tr) < 0)
9345 ftrace_init_trace_array(tr);
9347 init_trace_flags_index(tr);
9349 if (trace_instance_dir) {
9350 ret = trace_array_create_dir(tr);
9354 __trace_early_add_events(tr);
9356 list_add(&tr->list, &ftrace_trace_arrays);
9363 ftrace_free_ftrace_ops(tr);
9364 free_trace_buffers(tr);
9365 free_cpumask_var(tr->tracing_cpumask);
9369 return ERR_PTR(ret);
9372 static int instance_mkdir(const char *name)
9374 struct trace_array *tr;
9377 mutex_lock(&event_mutex);
9378 mutex_lock(&trace_types_lock);
9381 if (trace_array_find(name))
9384 tr = trace_array_create(name);
9386 ret = PTR_ERR_OR_ZERO(tr);
9389 mutex_unlock(&trace_types_lock);
9390 mutex_unlock(&event_mutex);
9395 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9396 * @name: The name of the trace array to be looked up/created.
9398 * Returns pointer to trace array with given name.
9399 * NULL, if it cannot be created.
9401 * NOTE: This function increments the reference counter associated with the
9402 * trace array returned. This makes sure it cannot be freed while in use.
9403 * Use trace_array_put() once the trace array is no longer needed.
9404 * If the trace_array is to be freed, trace_array_destroy() needs to
9405 * be called after the trace_array_put(), or simply let user space delete
9406 * it from the tracefs instances directory. But until the
9407 * trace_array_put() is called, user space can not delete it.
9410 struct trace_array *trace_array_get_by_name(const char *name)
9412 struct trace_array *tr;
9414 mutex_lock(&event_mutex);
9415 mutex_lock(&trace_types_lock);
9417 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9418 if (tr->name && strcmp(tr->name, name) == 0)
9422 tr = trace_array_create(name);
9430 mutex_unlock(&trace_types_lock);
9431 mutex_unlock(&event_mutex);
9434 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9436 static int __remove_instance(struct trace_array *tr)
9440 /* Reference counter for a newly created trace array = 1. */
9441 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9444 list_del(&tr->list);
9446 /* Disable all the flags that were enabled coming in */
9447 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9448 if ((1 << i) & ZEROED_TRACE_FLAGS)
9449 set_tracer_flag(tr, 1 << i, 0);
9452 tracing_set_nop(tr);
9453 clear_ftrace_function_probes(tr);
9454 event_trace_del_tracer(tr);
9455 ftrace_clear_pids(tr);
9456 ftrace_destroy_function_files(tr);
9457 tracefs_remove(tr->dir);
9458 free_percpu(tr->last_func_repeats);
9459 free_trace_buffers(tr);
9461 for (i = 0; i < tr->nr_topts; i++) {
9462 kfree(tr->topts[i].topts);
9466 free_cpumask_var(tr->tracing_cpumask);
9473 int trace_array_destroy(struct trace_array *this_tr)
9475 struct trace_array *tr;
9481 mutex_lock(&event_mutex);
9482 mutex_lock(&trace_types_lock);
9486 /* Making sure trace array exists before destroying it. */
9487 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9488 if (tr == this_tr) {
9489 ret = __remove_instance(tr);
9494 mutex_unlock(&trace_types_lock);
9495 mutex_unlock(&event_mutex);
9499 EXPORT_SYMBOL_GPL(trace_array_destroy);
9501 static int instance_rmdir(const char *name)
9503 struct trace_array *tr;
9506 mutex_lock(&event_mutex);
9507 mutex_lock(&trace_types_lock);
9510 tr = trace_array_find(name);
9512 ret = __remove_instance(tr);
9514 mutex_unlock(&trace_types_lock);
9515 mutex_unlock(&event_mutex);
9520 static __init void create_trace_instances(struct dentry *d_tracer)
9522 struct trace_array *tr;
9524 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9527 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9530 mutex_lock(&event_mutex);
9531 mutex_lock(&trace_types_lock);
9533 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9536 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9537 "Failed to create instance directory\n"))
9541 mutex_unlock(&trace_types_lock);
9542 mutex_unlock(&event_mutex);
9546 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9548 struct trace_event_file *file;
9551 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9552 tr, &show_traces_fops);
9554 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9555 tr, &set_tracer_fops);
9557 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9558 tr, &tracing_cpumask_fops);
9560 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9561 tr, &tracing_iter_fops);
9563 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9566 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9567 tr, &tracing_pipe_fops);
9569 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9570 tr, &tracing_entries_fops);
9572 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9573 tr, &tracing_total_entries_fops);
9575 trace_create_file("free_buffer", 0200, d_tracer,
9576 tr, &tracing_free_buffer_fops);
9578 trace_create_file("trace_marker", 0220, d_tracer,
9579 tr, &tracing_mark_fops);
9581 file = __find_event_file(tr, "ftrace", "print");
9582 if (file && file->dir)
9583 trace_create_file("trigger", TRACE_MODE_WRITE, file->dir,
9584 file, &event_trigger_fops);
9585 tr->trace_marker_file = file;
9587 trace_create_file("trace_marker_raw", 0220, d_tracer,
9588 tr, &tracing_mark_raw_fops);
9590 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9593 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9594 tr, &rb_simple_fops);
9596 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9597 &trace_time_stamp_mode_fops);
9599 tr->buffer_percent = 50;
9601 trace_create_file("buffer_percent", TRACE_MODE_READ, d_tracer,
9602 tr, &buffer_percent_fops);
9604 create_trace_options_dir(tr);
9606 trace_create_maxlat_file(tr, d_tracer);
9608 if (ftrace_create_function_files(tr, d_tracer))
9609 MEM_FAIL(1, "Could not allocate function filter files");
9611 #ifdef CONFIG_TRACER_SNAPSHOT
9612 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9613 tr, &snapshot_fops);
9616 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9617 tr, &tracing_err_log_fops);
9619 for_each_tracing_cpu(cpu)
9620 tracing_init_tracefs_percpu(tr, cpu);
9622 ftrace_init_tracefs(tr, d_tracer);
9625 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
9627 struct vfsmount *mnt;
9628 struct file_system_type *type;
9631 * To maintain backward compatibility for tools that mount
9632 * debugfs to get to the tracing facility, tracefs is automatically
9633 * mounted to the debugfs/tracing directory.
9635 type = get_fs_type("tracefs");
9638 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
9639 put_filesystem(type);
9648 * tracing_init_dentry - initialize top level trace array
9650 * This is called when creating files or directories in the tracing
9651 * directory. It is called via fs_initcall() by any of the boot up code
9652 * and expects to return the dentry of the top level tracing directory.
9654 int tracing_init_dentry(void)
9656 struct trace_array *tr = &global_trace;
9658 if (security_locked_down(LOCKDOWN_TRACEFS)) {
9659 pr_warn("Tracing disabled due to lockdown\n");
9663 /* The top level trace array uses NULL as parent */
9667 if (WARN_ON(!tracefs_initialized()))
9671 * As there may still be users that expect the tracing
9672 * files to exist in debugfs/tracing, we must automount
9673 * the tracefs file system there, so older tools still
9674 * work with the newer kernel.
9676 tr->dir = debugfs_create_automount("tracing", NULL,
9677 trace_automount, NULL);
9682 extern struct trace_eval_map *__start_ftrace_eval_maps[];
9683 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
9685 static struct workqueue_struct *eval_map_wq __initdata;
9686 static struct work_struct eval_map_work __initdata;
9687 static struct work_struct tracerfs_init_work __initdata;
9689 static void __init eval_map_work_func(struct work_struct *work)
9693 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
9694 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
9697 static int __init trace_eval_init(void)
9699 INIT_WORK(&eval_map_work, eval_map_work_func);
9701 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
9703 pr_err("Unable to allocate eval_map_wq\n");
9705 eval_map_work_func(&eval_map_work);
9709 queue_work(eval_map_wq, &eval_map_work);
9713 subsys_initcall(trace_eval_init);
9715 static int __init trace_eval_sync(void)
9717 /* Make sure the eval map updates are finished */
9719 destroy_workqueue(eval_map_wq);
9723 late_initcall_sync(trace_eval_sync);
9726 #ifdef CONFIG_MODULES
9727 static void trace_module_add_evals(struct module *mod)
9729 if (!mod->num_trace_evals)
9733 * Modules with bad taint do not have events created, do
9734 * not bother with enums either.
9736 if (trace_module_has_bad_taint(mod))
9739 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
9742 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
9743 static void trace_module_remove_evals(struct module *mod)
9745 union trace_eval_map_item *map;
9746 union trace_eval_map_item **last = &trace_eval_maps;
9748 if (!mod->num_trace_evals)
9751 mutex_lock(&trace_eval_mutex);
9753 map = trace_eval_maps;
9756 if (map->head.mod == mod)
9758 map = trace_eval_jmp_to_tail(map);
9759 last = &map->tail.next;
9760 map = map->tail.next;
9765 *last = trace_eval_jmp_to_tail(map)->tail.next;
9768 mutex_unlock(&trace_eval_mutex);
9771 static inline void trace_module_remove_evals(struct module *mod) { }
9772 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
9774 static int trace_module_notify(struct notifier_block *self,
9775 unsigned long val, void *data)
9777 struct module *mod = data;
9780 case MODULE_STATE_COMING:
9781 trace_module_add_evals(mod);
9783 case MODULE_STATE_GOING:
9784 trace_module_remove_evals(mod);
9791 static struct notifier_block trace_module_nb = {
9792 .notifier_call = trace_module_notify,
9795 #endif /* CONFIG_MODULES */
9797 static __init void tracer_init_tracefs_work_func(struct work_struct *work)
9802 init_tracer_tracefs(&global_trace, NULL);
9803 ftrace_init_tracefs_toplevel(&global_trace, NULL);
9805 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
9806 &global_trace, &tracing_thresh_fops);
9808 trace_create_file("README", TRACE_MODE_READ, NULL,
9809 NULL, &tracing_readme_fops);
9811 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
9812 NULL, &tracing_saved_cmdlines_fops);
9814 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
9815 NULL, &tracing_saved_cmdlines_size_fops);
9817 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
9818 NULL, &tracing_saved_tgids_fops);
9820 trace_create_eval_file(NULL);
9822 #ifdef CONFIG_MODULES
9823 register_module_notifier(&trace_module_nb);
9826 #ifdef CONFIG_DYNAMIC_FTRACE
9827 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
9828 NULL, &tracing_dyn_info_fops);
9831 create_trace_instances(NULL);
9833 update_tracer_options(&global_trace);
9836 static __init int tracer_init_tracefs(void)
9840 trace_access_lock_init();
9842 ret = tracing_init_dentry();
9847 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
9848 queue_work(eval_map_wq, &tracerfs_init_work);
9850 tracer_init_tracefs_work_func(NULL);
9853 rv_init_interface();
9858 fs_initcall(tracer_init_tracefs);
9860 static int trace_panic_handler(struct notifier_block *this,
9861 unsigned long event, void *unused)
9863 if (ftrace_dump_on_oops)
9864 ftrace_dump(ftrace_dump_on_oops);
9868 static struct notifier_block trace_panic_notifier = {
9869 .notifier_call = trace_panic_handler,
9871 .priority = 150 /* priority: INT_MAX >= x >= 0 */
9874 static int trace_die_handler(struct notifier_block *self,
9880 if (ftrace_dump_on_oops)
9881 ftrace_dump(ftrace_dump_on_oops);
9889 static struct notifier_block trace_die_notifier = {
9890 .notifier_call = trace_die_handler,
9895 * printk is set to max of 1024, we really don't need it that big.
9896 * Nothing should be printing 1000 characters anyway.
9898 #define TRACE_MAX_PRINT 1000
9901 * Define here KERN_TRACE so that we have one place to modify
9902 * it if we decide to change what log level the ftrace dump
9905 #define KERN_TRACE KERN_EMERG
9908 trace_printk_seq(struct trace_seq *s)
9910 /* Probably should print a warning here. */
9911 if (s->seq.len >= TRACE_MAX_PRINT)
9912 s->seq.len = TRACE_MAX_PRINT;
9915 * More paranoid code. Although the buffer size is set to
9916 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
9917 * an extra layer of protection.
9919 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
9920 s->seq.len = s->seq.size - 1;
9922 /* should be zero ended, but we are paranoid. */
9923 s->buffer[s->seq.len] = 0;
9925 printk(KERN_TRACE "%s", s->buffer);
9930 void trace_init_global_iter(struct trace_iterator *iter)
9932 iter->tr = &global_trace;
9933 iter->trace = iter->tr->current_trace;
9934 iter->cpu_file = RING_BUFFER_ALL_CPUS;
9935 iter->array_buffer = &global_trace.array_buffer;
9937 if (iter->trace && iter->trace->open)
9938 iter->trace->open(iter);
9940 /* Annotate start of buffers if we had overruns */
9941 if (ring_buffer_overruns(iter->array_buffer->buffer))
9942 iter->iter_flags |= TRACE_FILE_ANNOTATE;
9944 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
9945 if (trace_clocks[iter->tr->clock_id].in_ns)
9946 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
9948 /* Can not use kmalloc for iter.temp and iter.fmt */
9949 iter->temp = static_temp_buf;
9950 iter->temp_size = STATIC_TEMP_BUF_SIZE;
9951 iter->fmt = static_fmt_buf;
9952 iter->fmt_size = STATIC_FMT_BUF_SIZE;
9955 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
9957 /* use static because iter can be a bit big for the stack */
9958 static struct trace_iterator iter;
9959 static atomic_t dump_running;
9960 struct trace_array *tr = &global_trace;
9961 unsigned int old_userobj;
9962 unsigned long flags;
9965 /* Only allow one dump user at a time. */
9966 if (atomic_inc_return(&dump_running) != 1) {
9967 atomic_dec(&dump_running);
9972 * Always turn off tracing when we dump.
9973 * We don't need to show trace output of what happens
9974 * between multiple crashes.
9976 * If the user does a sysrq-z, then they can re-enable
9977 * tracing with echo 1 > tracing_on.
9981 local_irq_save(flags);
9983 /* Simulate the iterator */
9984 trace_init_global_iter(&iter);
9986 for_each_tracing_cpu(cpu) {
9987 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
9990 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
9992 /* don't look at user memory in panic mode */
9993 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
9995 switch (oops_dump_mode) {
9997 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10000 iter.cpu_file = raw_smp_processor_id();
10005 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10006 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10009 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10011 /* Did function tracer already get disabled? */
10012 if (ftrace_is_dead()) {
10013 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10014 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10018 * We need to stop all tracing on all CPUS to read
10019 * the next buffer. This is a bit expensive, but is
10020 * not done often. We fill all what we can read,
10021 * and then release the locks again.
10024 while (!trace_empty(&iter)) {
10027 printk(KERN_TRACE "---------------------------------\n");
10031 trace_iterator_reset(&iter);
10032 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10034 if (trace_find_next_entry_inc(&iter) != NULL) {
10037 ret = print_trace_line(&iter);
10038 if (ret != TRACE_TYPE_NO_CONSUME)
10039 trace_consume(&iter);
10041 touch_nmi_watchdog();
10043 trace_printk_seq(&iter.seq);
10047 printk(KERN_TRACE " (ftrace buffer empty)\n");
10049 printk(KERN_TRACE "---------------------------------\n");
10052 tr->trace_flags |= old_userobj;
10054 for_each_tracing_cpu(cpu) {
10055 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10057 atomic_dec(&dump_running);
10058 local_irq_restore(flags);
10060 EXPORT_SYMBOL_GPL(ftrace_dump);
10062 #define WRITE_BUFSIZE 4096
10064 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10065 size_t count, loff_t *ppos,
10066 int (*createfn)(const char *))
10068 char *kbuf, *buf, *tmp;
10073 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10077 while (done < count) {
10078 size = count - done;
10080 if (size >= WRITE_BUFSIZE)
10081 size = WRITE_BUFSIZE - 1;
10083 if (copy_from_user(kbuf, buffer + done, size)) {
10090 tmp = strchr(buf, '\n');
10093 size = tmp - buf + 1;
10095 size = strlen(buf);
10096 if (done + size < count) {
10099 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10100 pr_warn("Line length is too long: Should be less than %d\n",
10101 WRITE_BUFSIZE - 2);
10108 /* Remove comments */
10109 tmp = strchr(buf, '#');
10114 ret = createfn(buf);
10119 } while (done < count);
10129 __init static int tracer_alloc_buffers(void)
10135 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10136 pr_warn("Tracing disabled due to lockdown\n");
10141 * Make sure we don't accidentally add more trace options
10142 * than we have bits for.
10144 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10146 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10149 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10150 goto out_free_buffer_mask;
10152 /* Only allocate trace_printk buffers if a trace_printk exists */
10153 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10154 /* Must be called before global_trace.buffer is allocated */
10155 trace_printk_init_buffers();
10157 /* To save memory, keep the ring buffer size to its minimum */
10158 if (ring_buffer_expanded)
10159 ring_buf_size = trace_buf_size;
10163 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10164 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10166 raw_spin_lock_init(&global_trace.start_lock);
10169 * The prepare callbacks allocates some memory for the ring buffer. We
10170 * don't free the buffer if the CPU goes down. If we were to free
10171 * the buffer, then the user would lose any trace that was in the
10172 * buffer. The memory will be removed once the "instance" is removed.
10174 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10175 "trace/RB:prepare", trace_rb_cpu_prepare,
10178 goto out_free_cpumask;
10179 /* Used for event triggers */
10181 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10183 goto out_rm_hp_state;
10185 if (trace_create_savedcmd() < 0)
10186 goto out_free_temp_buffer;
10188 /* TODO: make the number of buffers hot pluggable with CPUS */
10189 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10190 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10191 goto out_free_savedcmd;
10194 if (global_trace.buffer_disabled)
10197 if (trace_boot_clock) {
10198 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10200 pr_warn("Trace clock %s not defined, going back to default\n",
10205 * register_tracer() might reference current_trace, so it
10206 * needs to be set before we register anything. This is
10207 * just a bootstrap of current_trace anyway.
10209 global_trace.current_trace = &nop_trace;
10211 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10213 ftrace_init_global_array_ops(&global_trace);
10215 init_trace_flags_index(&global_trace);
10217 register_tracer(&nop_trace);
10219 /* Function tracing may start here (via kernel command line) */
10220 init_function_trace();
10222 /* All seems OK, enable tracing */
10223 tracing_disabled = 0;
10225 atomic_notifier_chain_register(&panic_notifier_list,
10226 &trace_panic_notifier);
10228 register_die_notifier(&trace_die_notifier);
10230 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10232 INIT_LIST_HEAD(&global_trace.systems);
10233 INIT_LIST_HEAD(&global_trace.events);
10234 INIT_LIST_HEAD(&global_trace.hist_vars);
10235 INIT_LIST_HEAD(&global_trace.err_log);
10236 list_add(&global_trace.list, &ftrace_trace_arrays);
10238 apply_trace_boot_options();
10240 register_snapshot_cmd();
10247 free_saved_cmdlines_buffer(savedcmd);
10248 out_free_temp_buffer:
10249 ring_buffer_free(temp_buffer);
10251 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10253 free_cpumask_var(global_trace.tracing_cpumask);
10254 out_free_buffer_mask:
10255 free_cpumask_var(tracing_buffer_mask);
10260 void __init ftrace_boot_snapshot(void)
10262 if (snapshot_at_boot) {
10263 tracing_snapshot();
10264 internal_trace_puts("** Boot snapshot taken **\n");
10268 void __init early_trace_init(void)
10270 if (tracepoint_printk) {
10271 tracepoint_print_iter =
10272 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10273 if (MEM_FAIL(!tracepoint_print_iter,
10274 "Failed to allocate trace iterator\n"))
10275 tracepoint_printk = 0;
10277 static_key_enable(&tracepoint_printk_key.key);
10279 tracer_alloc_buffers();
10282 void __init trace_init(void)
10284 trace_event_init();
10287 __init static void clear_boot_tracer(void)
10290 * The default tracer at boot buffer is an init section.
10291 * This function is called in lateinit. If we did not
10292 * find the boot tracer, then clear it out, to prevent
10293 * later registration from accessing the buffer that is
10294 * about to be freed.
10296 if (!default_bootup_tracer)
10299 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10300 default_bootup_tracer);
10301 default_bootup_tracer = NULL;
10304 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
10305 __init static void tracing_set_default_clock(void)
10307 /* sched_clock_stable() is determined in late_initcall */
10308 if (!trace_boot_clock && !sched_clock_stable()) {
10309 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10310 pr_warn("Can not set tracing clock due to lockdown\n");
10314 printk(KERN_WARNING
10315 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10316 "If you want to keep using the local clock, then add:\n"
10317 " \"trace_clock=local\"\n"
10318 "on the kernel command line\n");
10319 tracing_set_clock(&global_trace, "global");
10323 static inline void tracing_set_default_clock(void) { }
10326 __init static int late_trace_init(void)
10328 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10329 static_key_disable(&tracepoint_printk_key.key);
10330 tracepoint_printk = 0;
10333 tracing_set_default_clock();
10334 clear_boot_tracer();
10338 late_initcall_sync(late_trace_init);