2 * ring buffer based function tracer
4 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 Nadia Yvette Chambers
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/nmi.h>
41 #include <linux/sched/rt.h>
44 #include "trace_output.h"
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 int ring_buffer_expanded;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
71 static struct tracer_flags dummy_tracer_flags = {
73 .opts = dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
82 * To prevent the comm cache from being overwritten when no
83 * tracing is active, only save the comm when a trace event
86 static DEFINE_PER_CPU(bool, trace_cmdline_save);
89 * Kill all tracing for good (never come back).
90 * It is initialized to 1 but will turn to zero if the initialization
91 * of the tracer is successful. But that is the only place that sets
94 static int tracing_disabled = 1;
96 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
98 cpumask_var_t __read_mostly tracing_buffer_mask;
101 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
103 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
104 * is set, then ftrace_dump is called. This will output the contents
105 * of the ftrace buffers to the console. This is very useful for
106 * capturing traces that lead to crashes and outputing it to a
109 * It is default off, but you can enable it with either specifying
110 * "ftrace_dump_on_oops" in the kernel command line, or setting
111 * /proc/sys/kernel/ftrace_dump_on_oops
112 * Set 1 if you want to dump buffers of all CPUs
113 * Set 2 if you want to dump the buffer of the CPU that triggered oops
116 enum ftrace_dump_mode ftrace_dump_on_oops;
118 static int tracing_set_tracer(const char *buf);
120 #define MAX_TRACER_SIZE 100
121 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
122 static char *default_bootup_tracer;
124 static int __init set_cmdline_ftrace(char *str)
126 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
127 default_bootup_tracer = bootup_tracer_buf;
128 /* We are using ftrace early, expand it */
129 ring_buffer_expanded = 1;
132 __setup("ftrace=", set_cmdline_ftrace);
134 static int __init set_ftrace_dump_on_oops(char *str)
136 if (*str++ != '=' || !*str) {
137 ftrace_dump_on_oops = DUMP_ALL;
141 if (!strcmp("orig_cpu", str)) {
142 ftrace_dump_on_oops = DUMP_ORIG;
148 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
151 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
152 static char *trace_boot_options __initdata;
154 static int __init set_trace_boot_options(char *str)
156 strncpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
157 trace_boot_options = trace_boot_options_buf;
160 __setup("trace_options=", set_trace_boot_options);
162 unsigned long long ns2usecs(cycle_t nsec)
170 * The global_trace is the descriptor that holds the tracing
171 * buffers for the live tracing. For each CPU, it contains
172 * a link list of pages that will store trace entries. The
173 * page descriptor of the pages in the memory is used to hold
174 * the link list by linking the lru item in the page descriptor
175 * to each of the pages in the buffer per CPU.
177 * For each active CPU there is a data field that holds the
178 * pages for the buffer for that CPU. Each CPU has the same number
179 * of pages allocated for its buffer.
181 static struct trace_array global_trace;
183 LIST_HEAD(ftrace_trace_arrays);
185 int filter_current_check_discard(struct ring_buffer *buffer,
186 struct ftrace_event_call *call, void *rec,
187 struct ring_buffer_event *event)
189 return filter_check_discard(call, rec, buffer, event);
191 EXPORT_SYMBOL_GPL(filter_current_check_discard);
193 cycle_t ftrace_now(int cpu)
197 /* Early boot up does not have a buffer yet */
198 if (!global_trace.trace_buffer.buffer)
199 return trace_clock_local();
201 ts = ring_buffer_time_stamp(global_trace.trace_buffer.buffer, cpu);
202 ring_buffer_normalize_time_stamp(global_trace.trace_buffer.buffer, cpu, &ts);
207 int tracing_is_enabled(void)
209 return tracing_is_on();
213 * trace_buf_size is the size in bytes that is allocated
214 * for a buffer. Note, the number of bytes is always rounded
217 * This number is purposely set to a low number of 16384.
218 * If the dump on oops happens, it will be much appreciated
219 * to not have to wait for all that output. Anyway this can be
220 * boot time and run time configurable.
222 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
224 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
226 /* trace_types holds a link list of available tracers. */
227 static struct tracer *trace_types __read_mostly;
230 * trace_types_lock is used to protect the trace_types list.
232 static DEFINE_MUTEX(trace_types_lock);
235 * serialize the access of the ring buffer
237 * ring buffer serializes readers, but it is low level protection.
238 * The validity of the events (which returns by ring_buffer_peek() ..etc)
239 * are not protected by ring buffer.
241 * The content of events may become garbage if we allow other process consumes
242 * these events concurrently:
243 * A) the page of the consumed events may become a normal page
244 * (not reader page) in ring buffer, and this page will be rewrited
245 * by events producer.
246 * B) The page of the consumed events may become a page for splice_read,
247 * and this page will be returned to system.
249 * These primitives allow multi process access to different cpu ring buffer
252 * These primitives don't distinguish read-only and read-consume access.
253 * Multi read-only access are also serialized.
257 static DECLARE_RWSEM(all_cpu_access_lock);
258 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
260 static inline void trace_access_lock(int cpu)
262 if (cpu == RING_BUFFER_ALL_CPUS) {
263 /* gain it for accessing the whole ring buffer. */
264 down_write(&all_cpu_access_lock);
266 /* gain it for accessing a cpu ring buffer. */
268 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
269 down_read(&all_cpu_access_lock);
271 /* Secondly block other access to this @cpu ring buffer. */
272 mutex_lock(&per_cpu(cpu_access_lock, cpu));
276 static inline void trace_access_unlock(int cpu)
278 if (cpu == RING_BUFFER_ALL_CPUS) {
279 up_write(&all_cpu_access_lock);
281 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
282 up_read(&all_cpu_access_lock);
286 static inline void trace_access_lock_init(void)
290 for_each_possible_cpu(cpu)
291 mutex_init(&per_cpu(cpu_access_lock, cpu));
296 static DEFINE_MUTEX(access_lock);
298 static inline void trace_access_lock(int cpu)
301 mutex_lock(&access_lock);
304 static inline void trace_access_unlock(int cpu)
307 mutex_unlock(&access_lock);
310 static inline void trace_access_lock_init(void)
316 /* trace_flags holds trace_options default values */
317 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
318 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
319 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
320 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS;
323 * tracing_on - enable tracing buffers
325 * This function enables tracing buffers that may have been
326 * disabled with tracing_off.
328 void tracing_on(void)
330 if (global_trace.trace_buffer.buffer)
331 ring_buffer_record_on(global_trace.trace_buffer.buffer);
333 * This flag is only looked at when buffers haven't been
334 * allocated yet. We don't really care about the race
335 * between setting this flag and actually turning
338 global_trace.buffer_disabled = 0;
340 EXPORT_SYMBOL_GPL(tracing_on);
343 * tracing_off - turn off tracing buffers
345 * This function stops the tracing buffers from recording data.
346 * It does not disable any overhead the tracers themselves may
347 * be causing. This function simply causes all recording to
348 * the ring buffers to fail.
350 void tracing_off(void)
352 if (global_trace.trace_buffer.buffer)
353 ring_buffer_record_off(global_trace.trace_buffer.buffer);
355 * This flag is only looked at when buffers haven't been
356 * allocated yet. We don't really care about the race
357 * between setting this flag and actually turning
360 global_trace.buffer_disabled = 1;
362 EXPORT_SYMBOL_GPL(tracing_off);
365 * tracing_is_on - show state of ring buffers enabled
367 int tracing_is_on(void)
369 if (global_trace.trace_buffer.buffer)
370 return ring_buffer_record_is_on(global_trace.trace_buffer.buffer);
371 return !global_trace.buffer_disabled;
373 EXPORT_SYMBOL_GPL(tracing_is_on);
375 static int __init set_buf_size(char *str)
377 unsigned long buf_size;
381 buf_size = memparse(str, &str);
382 /* nr_entries can not be zero */
385 trace_buf_size = buf_size;
388 __setup("trace_buf_size=", set_buf_size);
390 static int __init set_tracing_thresh(char *str)
392 unsigned long threshold;
397 ret = kstrtoul(str, 0, &threshold);
400 tracing_thresh = threshold * 1000;
403 __setup("tracing_thresh=", set_tracing_thresh);
405 unsigned long nsecs_to_usecs(unsigned long nsecs)
410 /* These must match the bit postions in trace_iterator_flags */
411 static const char *trace_options[] = {
443 int in_ns; /* is this clock in nanoseconds? */
445 { trace_clock_local, "local", 1 },
446 { trace_clock_global, "global", 1 },
447 { trace_clock_counter, "counter", 0 },
454 * trace_parser_get_init - gets the buffer for trace parser
456 int trace_parser_get_init(struct trace_parser *parser, int size)
458 memset(parser, 0, sizeof(*parser));
460 parser->buffer = kmalloc(size, GFP_KERNEL);
469 * trace_parser_put - frees the buffer for trace parser
471 void trace_parser_put(struct trace_parser *parser)
473 kfree(parser->buffer);
477 * trace_get_user - reads the user input string separated by space
478 * (matched by isspace(ch))
480 * For each string found the 'struct trace_parser' is updated,
481 * and the function returns.
483 * Returns number of bytes read.
485 * See kernel/trace/trace.h for 'struct trace_parser' details.
487 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
488 size_t cnt, loff_t *ppos)
495 trace_parser_clear(parser);
497 ret = get_user(ch, ubuf++);
505 * The parser is not finished with the last write,
506 * continue reading the user input without skipping spaces.
509 /* skip white space */
510 while (cnt && isspace(ch)) {
511 ret = get_user(ch, ubuf++);
518 /* only spaces were written */
528 /* read the non-space input */
529 while (cnt && !isspace(ch)) {
530 if (parser->idx < parser->size - 1)
531 parser->buffer[parser->idx++] = ch;
536 ret = get_user(ch, ubuf++);
543 /* We either got finished input or we have to wait for another call. */
545 parser->buffer[parser->idx] = 0;
546 parser->cont = false;
549 parser->buffer[parser->idx++] = ch;
559 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
567 if (s->len <= s->readpos)
570 len = s->len - s->readpos;
573 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
583 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
587 if (s->len <= s->readpos)
590 len = s->len - s->readpos;
593 memcpy(buf, s->buffer + s->readpos, cnt);
600 * ftrace_max_lock is used to protect the swapping of buffers
601 * when taking a max snapshot. The buffers themselves are
602 * protected by per_cpu spinlocks. But the action of the swap
603 * needs its own lock.
605 * This is defined as a arch_spinlock_t in order to help
606 * with performance when lockdep debugging is enabled.
608 * It is also used in other places outside the update_max_tr
609 * so it needs to be defined outside of the
610 * CONFIG_TRACER_MAX_TRACE.
612 static arch_spinlock_t ftrace_max_lock =
613 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
615 unsigned long __read_mostly tracing_thresh;
617 #ifdef CONFIG_TRACER_MAX_TRACE
618 unsigned long __read_mostly tracing_max_latency;
621 * Copy the new maximum trace into the separate maximum-trace
622 * structure. (this way the maximum trace is permanently saved,
623 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
626 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
628 struct trace_buffer *trace_buf = &tr->trace_buffer;
629 struct trace_buffer *max_buf = &tr->max_buffer;
630 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
631 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
634 max_buf->time_start = data->preempt_timestamp;
636 max_data->saved_latency = tracing_max_latency;
637 max_data->critical_start = data->critical_start;
638 max_data->critical_end = data->critical_end;
640 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
641 max_data->pid = tsk->pid;
642 max_data->uid = task_uid(tsk);
643 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
644 max_data->policy = tsk->policy;
645 max_data->rt_priority = tsk->rt_priority;
647 /* record this tasks comm */
648 tracing_record_cmdline(tsk);
652 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
654 * @tsk: the task with the latency
655 * @cpu: The cpu that initiated the trace.
657 * Flip the buffers between the @tr and the max_tr and record information
658 * about which task was the cause of this latency.
661 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
663 struct ring_buffer *buf;
668 WARN_ON_ONCE(!irqs_disabled());
670 if (!tr->allocated_snapshot) {
671 /* Only the nop tracer should hit this when disabling */
672 WARN_ON_ONCE(tr->current_trace != &nop_trace);
676 arch_spin_lock(&ftrace_max_lock);
678 buf = tr->trace_buffer.buffer;
679 tr->trace_buffer.buffer = tr->max_buffer.buffer;
680 tr->max_buffer.buffer = buf;
682 __update_max_tr(tr, tsk, cpu);
683 arch_spin_unlock(&ftrace_max_lock);
687 * update_max_tr_single - only copy one trace over, and reset the rest
689 * @tsk - task with the latency
690 * @cpu - the cpu of the buffer to copy.
692 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
695 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
702 WARN_ON_ONCE(!irqs_disabled());
703 if (WARN_ON_ONCE(!tr->allocated_snapshot))
706 arch_spin_lock(&ftrace_max_lock);
708 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
712 * We failed to swap the buffer due to a commit taking
713 * place on this CPU. We fail to record, but we reset
714 * the max trace buffer (no one writes directly to it)
715 * and flag that it failed.
717 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
718 "Failed to swap buffers due to commit in progress\n");
721 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
723 __update_max_tr(tr, tsk, cpu);
724 arch_spin_unlock(&ftrace_max_lock);
726 #endif /* CONFIG_TRACER_MAX_TRACE */
728 static void default_wait_pipe(struct trace_iterator *iter)
730 /* Iterators are static, they should be filled or empty */
731 if (trace_buffer_iter(iter, iter->cpu_file))
734 ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
738 * register_tracer - register a tracer with the ftrace system.
739 * @type - the plugin for the tracer
741 * Register a new plugin tracer.
743 int register_tracer(struct tracer *type)
749 pr_info("Tracer must have a name\n");
753 if (strlen(type->name) >= MAX_TRACER_SIZE) {
754 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
758 mutex_lock(&trace_types_lock);
760 tracing_selftest_running = true;
762 for (t = trace_types; t; t = t->next) {
763 if (strcmp(type->name, t->name) == 0) {
765 pr_info("Tracer %s already registered\n",
773 type->set_flag = &dummy_set_flag;
775 type->flags = &dummy_tracer_flags;
777 if (!type->flags->opts)
778 type->flags->opts = dummy_tracer_opt;
779 if (!type->wait_pipe)
780 type->wait_pipe = default_wait_pipe;
783 #ifdef CONFIG_FTRACE_STARTUP_TEST
784 if (type->selftest && !tracing_selftest_disabled) {
785 struct trace_array *tr = &global_trace;
786 struct tracer *saved_tracer = tr->current_trace;
789 * Run a selftest on this tracer.
790 * Here we reset the trace buffer, and set the current
791 * tracer to be this tracer. The tracer can then run some
792 * internal tracing to verify that everything is in order.
793 * If we fail, we do not register this tracer.
795 tracing_reset_online_cpus(&tr->trace_buffer);
797 tr->current_trace = type;
799 #ifdef CONFIG_TRACER_MAX_TRACE
800 if (type->use_max_tr) {
801 /* If we expanded the buffers, make sure the max is expanded too */
802 if (ring_buffer_expanded)
803 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
804 RING_BUFFER_ALL_CPUS);
805 tr->allocated_snapshot = true;
809 /* the test is responsible for initializing and enabling */
810 pr_info("Testing tracer %s: ", type->name);
811 ret = type->selftest(type, tr);
812 /* the test is responsible for resetting too */
813 tr->current_trace = saved_tracer;
815 printk(KERN_CONT "FAILED!\n");
816 /* Add the warning after printing 'FAILED' */
820 /* Only reset on passing, to avoid touching corrupted buffers */
821 tracing_reset_online_cpus(&tr->trace_buffer);
823 #ifdef CONFIG_TRACER_MAX_TRACE
824 if (type->use_max_tr) {
825 tr->allocated_snapshot = false;
827 /* Shrink the max buffer again */
828 if (ring_buffer_expanded)
829 ring_buffer_resize(tr->max_buffer.buffer, 1,
830 RING_BUFFER_ALL_CPUS);
834 printk(KERN_CONT "PASSED\n");
838 type->next = trace_types;
842 tracing_selftest_running = false;
843 mutex_unlock(&trace_types_lock);
845 if (ret || !default_bootup_tracer)
848 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
851 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
852 /* Do we want this tracer to start on bootup? */
853 tracing_set_tracer(type->name);
854 default_bootup_tracer = NULL;
855 /* disable other selftests, since this will break it. */
856 tracing_selftest_disabled = 1;
857 #ifdef CONFIG_FTRACE_STARTUP_TEST
858 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
866 void tracing_reset(struct trace_buffer *buf, int cpu)
868 struct ring_buffer *buffer = buf->buffer;
873 ring_buffer_record_disable(buffer);
875 /* Make sure all commits have finished */
877 ring_buffer_reset_cpu(buffer, cpu);
879 ring_buffer_record_enable(buffer);
882 void tracing_reset_online_cpus(struct trace_buffer *buf)
884 struct ring_buffer *buffer = buf->buffer;
890 ring_buffer_record_disable(buffer);
892 /* Make sure all commits have finished */
895 buf->time_start = ftrace_now(buf->cpu);
897 for_each_online_cpu(cpu)
898 ring_buffer_reset_cpu(buffer, cpu);
900 ring_buffer_record_enable(buffer);
903 void tracing_reset_current(int cpu)
905 tracing_reset(&global_trace.trace_buffer, cpu);
908 void tracing_reset_all_online_cpus(void)
910 struct trace_array *tr;
912 mutex_lock(&trace_types_lock);
913 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
914 tracing_reset_online_cpus(&tr->trace_buffer);
915 #ifdef CONFIG_TRACER_MAX_TRACE
916 tracing_reset_online_cpus(&tr->max_buffer);
919 mutex_unlock(&trace_types_lock);
922 #define SAVED_CMDLINES 128
923 #define NO_CMDLINE_MAP UINT_MAX
924 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
925 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
926 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
927 static int cmdline_idx;
928 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
930 /* temporary disable recording */
931 static atomic_t trace_record_cmdline_disabled __read_mostly;
933 static void trace_init_cmdlines(void)
935 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
936 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
940 int is_tracing_stopped(void)
942 return global_trace.stop_count;
946 * ftrace_off_permanent - disable all ftrace code permanently
948 * This should only be called when a serious anomally has
949 * been detected. This will turn off the function tracing,
950 * ring buffers, and other tracing utilites. It takes no
951 * locks and can be called from any context.
953 void ftrace_off_permanent(void)
955 tracing_disabled = 1;
957 tracing_off_permanent();
961 * tracing_start - quick start of the tracer
963 * If tracing is enabled but was stopped by tracing_stop,
964 * this will start the tracer back up.
966 void tracing_start(void)
968 struct ring_buffer *buffer;
971 if (tracing_disabled)
974 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
975 if (--global_trace.stop_count) {
976 if (global_trace.stop_count < 0) {
977 /* Someone screwed up their debugging */
979 global_trace.stop_count = 0;
984 /* Prevent the buffers from switching */
985 arch_spin_lock(&ftrace_max_lock);
987 buffer = global_trace.trace_buffer.buffer;
989 ring_buffer_record_enable(buffer);
991 #ifdef CONFIG_TRACER_MAX_TRACE
992 buffer = global_trace.max_buffer.buffer;
994 ring_buffer_record_enable(buffer);
997 arch_spin_unlock(&ftrace_max_lock);
1001 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1004 static void tracing_start_tr(struct trace_array *tr)
1006 struct ring_buffer *buffer;
1007 unsigned long flags;
1009 if (tracing_disabled)
1012 /* If global, we need to also start the max tracer */
1013 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1014 return tracing_start();
1016 raw_spin_lock_irqsave(&tr->start_lock, flags);
1018 if (--tr->stop_count) {
1019 if (tr->stop_count < 0) {
1020 /* Someone screwed up their debugging */
1027 buffer = tr->trace_buffer.buffer;
1029 ring_buffer_record_enable(buffer);
1032 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1036 * tracing_stop - quick stop of the tracer
1038 * Light weight way to stop tracing. Use in conjunction with
1041 void tracing_stop(void)
1043 struct ring_buffer *buffer;
1044 unsigned long flags;
1047 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1048 if (global_trace.stop_count++)
1051 /* Prevent the buffers from switching */
1052 arch_spin_lock(&ftrace_max_lock);
1054 buffer = global_trace.trace_buffer.buffer;
1056 ring_buffer_record_disable(buffer);
1058 #ifdef CONFIG_TRACER_MAX_TRACE
1059 buffer = global_trace.max_buffer.buffer;
1061 ring_buffer_record_disable(buffer);
1064 arch_spin_unlock(&ftrace_max_lock);
1067 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1070 static void tracing_stop_tr(struct trace_array *tr)
1072 struct ring_buffer *buffer;
1073 unsigned long flags;
1075 /* If global, we need to also stop the max tracer */
1076 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1077 return tracing_stop();
1079 raw_spin_lock_irqsave(&tr->start_lock, flags);
1080 if (tr->stop_count++)
1083 buffer = tr->trace_buffer.buffer;
1085 ring_buffer_record_disable(buffer);
1088 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1091 void trace_stop_cmdline_recording(void);
1093 static void trace_save_cmdline(struct task_struct *tsk)
1097 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1101 * It's not the end of the world if we don't get
1102 * the lock, but we also don't want to spin
1103 * nor do we want to disable interrupts,
1104 * so if we miss here, then better luck next time.
1106 if (!arch_spin_trylock(&trace_cmdline_lock))
1109 idx = map_pid_to_cmdline[tsk->pid];
1110 if (idx == NO_CMDLINE_MAP) {
1111 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1114 * Check whether the cmdline buffer at idx has a pid
1115 * mapped. We are going to overwrite that entry so we
1116 * need to clear the map_pid_to_cmdline. Otherwise we
1117 * would read the new comm for the old pid.
1119 pid = map_cmdline_to_pid[idx];
1120 if (pid != NO_CMDLINE_MAP)
1121 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1123 map_cmdline_to_pid[idx] = tsk->pid;
1124 map_pid_to_cmdline[tsk->pid] = idx;
1129 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1131 arch_spin_unlock(&trace_cmdline_lock);
1134 void trace_find_cmdline(int pid, char comm[])
1139 strcpy(comm, "<idle>");
1143 if (WARN_ON_ONCE(pid < 0)) {
1144 strcpy(comm, "<XXX>");
1148 if (pid > PID_MAX_DEFAULT) {
1149 strcpy(comm, "<...>");
1154 arch_spin_lock(&trace_cmdline_lock);
1155 map = map_pid_to_cmdline[pid];
1156 if (map != NO_CMDLINE_MAP)
1157 strcpy(comm, saved_cmdlines[map]);
1159 strcpy(comm, "<...>");
1161 arch_spin_unlock(&trace_cmdline_lock);
1165 void tracing_record_cmdline(struct task_struct *tsk)
1167 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1170 if (!__this_cpu_read(trace_cmdline_save))
1173 __this_cpu_write(trace_cmdline_save, false);
1175 trace_save_cmdline(tsk);
1179 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1182 struct task_struct *tsk = current;
1184 entry->preempt_count = pc & 0xff;
1185 entry->pid = (tsk) ? tsk->pid : 0;
1187 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1188 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1190 TRACE_FLAG_IRQS_NOSUPPORT |
1192 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1193 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1194 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1196 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1198 struct ring_buffer_event *
1199 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1202 unsigned long flags, int pc)
1204 struct ring_buffer_event *event;
1206 event = ring_buffer_lock_reserve(buffer, len);
1207 if (event != NULL) {
1208 struct trace_entry *ent = ring_buffer_event_data(event);
1210 tracing_generic_entry_update(ent, flags, pc);
1218 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1220 __this_cpu_write(trace_cmdline_save, true);
1221 ring_buffer_unlock_commit(buffer, event);
1225 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1226 struct ring_buffer_event *event,
1227 unsigned long flags, int pc)
1229 __buffer_unlock_commit(buffer, event);
1231 ftrace_trace_stack(buffer, flags, 6, pc);
1232 ftrace_trace_userstack(buffer, flags, pc);
1235 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1236 struct ring_buffer_event *event,
1237 unsigned long flags, int pc)
1239 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1241 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1243 struct ring_buffer_event *
1244 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1245 struct ftrace_event_file *ftrace_file,
1246 int type, unsigned long len,
1247 unsigned long flags, int pc)
1249 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1250 return trace_buffer_lock_reserve(*current_rb,
1251 type, len, flags, pc);
1253 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1255 struct ring_buffer_event *
1256 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1257 int type, unsigned long len,
1258 unsigned long flags, int pc)
1260 *current_rb = global_trace.trace_buffer.buffer;
1261 return trace_buffer_lock_reserve(*current_rb,
1262 type, len, flags, pc);
1264 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1266 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1267 struct ring_buffer_event *event,
1268 unsigned long flags, int pc)
1270 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1272 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1274 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1275 struct ring_buffer_event *event,
1276 unsigned long flags, int pc,
1277 struct pt_regs *regs)
1279 __buffer_unlock_commit(buffer, event);
1281 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1282 ftrace_trace_userstack(buffer, flags, pc);
1284 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1286 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1287 struct ring_buffer_event *event)
1289 ring_buffer_discard_commit(buffer, event);
1291 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1294 trace_function(struct trace_array *tr,
1295 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1298 struct ftrace_event_call *call = &event_function;
1299 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1300 struct ring_buffer_event *event;
1301 struct ftrace_entry *entry;
1303 /* If we are reading the ring buffer, don't trace */
1304 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1307 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1311 entry = ring_buffer_event_data(event);
1313 entry->parent_ip = parent_ip;
1315 if (!filter_check_discard(call, entry, buffer, event))
1316 __buffer_unlock_commit(buffer, event);
1320 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1321 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1324 if (likely(!atomic_read(&data->disabled)))
1325 trace_function(tr, ip, parent_ip, flags, pc);
1328 #ifdef CONFIG_STACKTRACE
1330 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1331 struct ftrace_stack {
1332 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1335 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1336 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1338 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1339 unsigned long flags,
1340 int skip, int pc, struct pt_regs *regs)
1342 struct ftrace_event_call *call = &event_kernel_stack;
1343 struct ring_buffer_event *event;
1344 struct stack_entry *entry;
1345 struct stack_trace trace;
1347 int size = FTRACE_STACK_ENTRIES;
1349 trace.nr_entries = 0;
1353 * Since events can happen in NMIs there's no safe way to
1354 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1355 * or NMI comes in, it will just have to use the default
1356 * FTRACE_STACK_SIZE.
1358 preempt_disable_notrace();
1360 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1362 * We don't need any atomic variables, just a barrier.
1363 * If an interrupt comes in, we don't care, because it would
1364 * have exited and put the counter back to what we want.
1365 * We just need a barrier to keep gcc from moving things
1369 if (use_stack == 1) {
1370 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1371 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1374 save_stack_trace_regs(regs, &trace);
1376 save_stack_trace(&trace);
1378 if (trace.nr_entries > size)
1379 size = trace.nr_entries;
1381 /* From now on, use_stack is a boolean */
1384 size *= sizeof(unsigned long);
1386 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1387 sizeof(*entry) + size, flags, pc);
1390 entry = ring_buffer_event_data(event);
1392 memset(&entry->caller, 0, size);
1395 memcpy(&entry->caller, trace.entries,
1396 trace.nr_entries * sizeof(unsigned long));
1398 trace.max_entries = FTRACE_STACK_ENTRIES;
1399 trace.entries = entry->caller;
1401 save_stack_trace_regs(regs, &trace);
1403 save_stack_trace(&trace);
1406 entry->size = trace.nr_entries;
1408 if (!filter_check_discard(call, entry, buffer, event))
1409 __buffer_unlock_commit(buffer, event);
1412 /* Again, don't let gcc optimize things here */
1414 __this_cpu_dec(ftrace_stack_reserve);
1415 preempt_enable_notrace();
1419 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1420 int skip, int pc, struct pt_regs *regs)
1422 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1425 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1428 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1431 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1434 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1437 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1440 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1444 * trace_dump_stack - record a stack back trace in the trace buffer
1446 void trace_dump_stack(void)
1448 unsigned long flags;
1450 if (tracing_disabled || tracing_selftest_running)
1453 local_save_flags(flags);
1455 /* skipping 3 traces, seems to get us at the caller of this function */
1456 __ftrace_trace_stack(global_trace.trace_buffer.buffer, flags, 3,
1457 preempt_count(), NULL);
1460 static DEFINE_PER_CPU(int, user_stack_count);
1463 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1465 struct ftrace_event_call *call = &event_user_stack;
1466 struct ring_buffer_event *event;
1467 struct userstack_entry *entry;
1468 struct stack_trace trace;
1470 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1474 * NMIs can not handle page faults, even with fix ups.
1475 * The save user stack can (and often does) fault.
1477 if (unlikely(in_nmi()))
1481 * prevent recursion, since the user stack tracing may
1482 * trigger other kernel events.
1485 if (__this_cpu_read(user_stack_count))
1488 __this_cpu_inc(user_stack_count);
1490 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1491 sizeof(*entry), flags, pc);
1493 goto out_drop_count;
1494 entry = ring_buffer_event_data(event);
1496 entry->tgid = current->tgid;
1497 memset(&entry->caller, 0, sizeof(entry->caller));
1499 trace.nr_entries = 0;
1500 trace.max_entries = FTRACE_STACK_ENTRIES;
1502 trace.entries = entry->caller;
1504 save_stack_trace_user(&trace);
1505 if (!filter_check_discard(call, entry, buffer, event))
1506 __buffer_unlock_commit(buffer, event);
1509 __this_cpu_dec(user_stack_count);
1515 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1517 ftrace_trace_userstack(tr, flags, preempt_count());
1521 #endif /* CONFIG_STACKTRACE */
1523 /* created for use with alloc_percpu */
1524 struct trace_buffer_struct {
1525 char buffer[TRACE_BUF_SIZE];
1528 static struct trace_buffer_struct *trace_percpu_buffer;
1529 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1530 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1531 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1534 * The buffer used is dependent on the context. There is a per cpu
1535 * buffer for normal context, softirq contex, hard irq context and
1536 * for NMI context. Thise allows for lockless recording.
1538 * Note, if the buffers failed to be allocated, then this returns NULL
1540 static char *get_trace_buf(void)
1542 struct trace_buffer_struct *percpu_buffer;
1545 * If we have allocated per cpu buffers, then we do not
1546 * need to do any locking.
1549 percpu_buffer = trace_percpu_nmi_buffer;
1551 percpu_buffer = trace_percpu_irq_buffer;
1552 else if (in_softirq())
1553 percpu_buffer = trace_percpu_sirq_buffer;
1555 percpu_buffer = trace_percpu_buffer;
1560 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1563 static int alloc_percpu_trace_buffer(void)
1565 struct trace_buffer_struct *buffers;
1566 struct trace_buffer_struct *sirq_buffers;
1567 struct trace_buffer_struct *irq_buffers;
1568 struct trace_buffer_struct *nmi_buffers;
1570 buffers = alloc_percpu(struct trace_buffer_struct);
1574 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1578 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1582 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1586 trace_percpu_buffer = buffers;
1587 trace_percpu_sirq_buffer = sirq_buffers;
1588 trace_percpu_irq_buffer = irq_buffers;
1589 trace_percpu_nmi_buffer = nmi_buffers;
1594 free_percpu(irq_buffers);
1596 free_percpu(sirq_buffers);
1598 free_percpu(buffers);
1600 WARN(1, "Could not allocate percpu trace_printk buffer");
1604 static int buffers_allocated;
1606 void trace_printk_init_buffers(void)
1608 if (buffers_allocated)
1611 if (alloc_percpu_trace_buffer())
1614 pr_info("ftrace: Allocated trace_printk buffers\n");
1616 /* Expand the buffers to set size */
1617 tracing_update_buffers();
1619 buffers_allocated = 1;
1622 * trace_printk_init_buffers() can be called by modules.
1623 * If that happens, then we need to start cmdline recording
1624 * directly here. If the global_trace.buffer is already
1625 * allocated here, then this was called by module code.
1627 if (global_trace.trace_buffer.buffer)
1628 tracing_start_cmdline_record();
1631 void trace_printk_start_comm(void)
1633 /* Start tracing comms if trace printk is set */
1634 if (!buffers_allocated)
1636 tracing_start_cmdline_record();
1639 static void trace_printk_start_stop_comm(int enabled)
1641 if (!buffers_allocated)
1645 tracing_start_cmdline_record();
1647 tracing_stop_cmdline_record();
1651 * trace_vbprintk - write binary msg to tracing buffer
1654 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1656 struct ftrace_event_call *call = &event_bprint;
1657 struct ring_buffer_event *event;
1658 struct ring_buffer *buffer;
1659 struct trace_array *tr = &global_trace;
1660 struct bprint_entry *entry;
1661 unsigned long flags;
1663 int len = 0, size, pc;
1665 if (unlikely(tracing_selftest_running || tracing_disabled))
1668 /* Don't pollute graph traces with trace_vprintk internals */
1669 pause_graph_tracing();
1671 pc = preempt_count();
1672 preempt_disable_notrace();
1674 tbuffer = get_trace_buf();
1680 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1682 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1685 local_save_flags(flags);
1686 size = sizeof(*entry) + sizeof(u32) * len;
1687 buffer = tr->trace_buffer.buffer;
1688 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1692 entry = ring_buffer_event_data(event);
1696 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1697 if (!filter_check_discard(call, entry, buffer, event)) {
1698 __buffer_unlock_commit(buffer, event);
1699 ftrace_trace_stack(buffer, flags, 6, pc);
1703 preempt_enable_notrace();
1704 unpause_graph_tracing();
1708 EXPORT_SYMBOL_GPL(trace_vbprintk);
1711 __trace_array_vprintk(struct ring_buffer *buffer,
1712 unsigned long ip, const char *fmt, va_list args)
1714 struct ftrace_event_call *call = &event_print;
1715 struct ring_buffer_event *event;
1716 int len = 0, size, pc;
1717 struct print_entry *entry;
1718 unsigned long flags;
1721 if (tracing_disabled || tracing_selftest_running)
1724 /* Don't pollute graph traces with trace_vprintk internals */
1725 pause_graph_tracing();
1727 pc = preempt_count();
1728 preempt_disable_notrace();
1731 tbuffer = get_trace_buf();
1737 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1738 if (len > TRACE_BUF_SIZE)
1741 local_save_flags(flags);
1742 size = sizeof(*entry) + len + 1;
1743 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1747 entry = ring_buffer_event_data(event);
1750 memcpy(&entry->buf, tbuffer, len);
1751 entry->buf[len] = '\0';
1752 if (!filter_check_discard(call, entry, buffer, event)) {
1753 __buffer_unlock_commit(buffer, event);
1754 ftrace_trace_stack(buffer, flags, 6, pc);
1757 preempt_enable_notrace();
1758 unpause_graph_tracing();
1763 int trace_array_vprintk(struct trace_array *tr,
1764 unsigned long ip, const char *fmt, va_list args)
1766 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
1769 int trace_array_printk(struct trace_array *tr,
1770 unsigned long ip, const char *fmt, ...)
1775 if (!(trace_flags & TRACE_ITER_PRINTK))
1779 ret = trace_array_vprintk(tr, ip, fmt, ap);
1784 int trace_array_printk_buf(struct ring_buffer *buffer,
1785 unsigned long ip, const char *fmt, ...)
1790 if (!(trace_flags & TRACE_ITER_PRINTK))
1794 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
1799 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1801 return trace_array_vprintk(&global_trace, ip, fmt, args);
1803 EXPORT_SYMBOL_GPL(trace_vprintk);
1805 static void trace_iterator_increment(struct trace_iterator *iter)
1807 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
1811 ring_buffer_read(buf_iter, NULL);
1814 static struct trace_entry *
1815 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1816 unsigned long *lost_events)
1818 struct ring_buffer_event *event;
1819 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
1822 event = ring_buffer_iter_peek(buf_iter, ts);
1824 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
1828 iter->ent_size = ring_buffer_event_length(event);
1829 return ring_buffer_event_data(event);
1835 static struct trace_entry *
1836 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1837 unsigned long *missing_events, u64 *ent_ts)
1839 struct ring_buffer *buffer = iter->trace_buffer->buffer;
1840 struct trace_entry *ent, *next = NULL;
1841 unsigned long lost_events = 0, next_lost = 0;
1842 int cpu_file = iter->cpu_file;
1843 u64 next_ts = 0, ts;
1849 * If we are in a per_cpu trace file, don't bother by iterating over
1850 * all cpu and peek directly.
1852 if (cpu_file > RING_BUFFER_ALL_CPUS) {
1853 if (ring_buffer_empty_cpu(buffer, cpu_file))
1855 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1857 *ent_cpu = cpu_file;
1862 for_each_tracing_cpu(cpu) {
1864 if (ring_buffer_empty_cpu(buffer, cpu))
1867 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1870 * Pick the entry with the smallest timestamp:
1872 if (ent && (!next || ts < next_ts)) {
1876 next_lost = lost_events;
1877 next_size = iter->ent_size;
1881 iter->ent_size = next_size;
1884 *ent_cpu = next_cpu;
1890 *missing_events = next_lost;
1895 /* Find the next real entry, without updating the iterator itself */
1896 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1897 int *ent_cpu, u64 *ent_ts)
1899 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1902 /* Find the next real entry, and increment the iterator to the next entry */
1903 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1905 iter->ent = __find_next_entry(iter, &iter->cpu,
1906 &iter->lost_events, &iter->ts);
1909 trace_iterator_increment(iter);
1911 return iter->ent ? iter : NULL;
1914 static void trace_consume(struct trace_iterator *iter)
1916 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
1917 &iter->lost_events);
1920 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1922 struct trace_iterator *iter = m->private;
1926 WARN_ON_ONCE(iter->leftover);
1930 /* can't go backwards */
1935 ent = trace_find_next_entry_inc(iter);
1939 while (ent && iter->idx < i)
1940 ent = trace_find_next_entry_inc(iter);
1947 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1949 struct ring_buffer_event *event;
1950 struct ring_buffer_iter *buf_iter;
1951 unsigned long entries = 0;
1954 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
1956 buf_iter = trace_buffer_iter(iter, cpu);
1960 ring_buffer_iter_reset(buf_iter);
1963 * We could have the case with the max latency tracers
1964 * that a reset never took place on a cpu. This is evident
1965 * by the timestamp being before the start of the buffer.
1967 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1968 if (ts >= iter->trace_buffer->time_start)
1971 ring_buffer_read(buf_iter, NULL);
1974 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
1978 * The current tracer is copied to avoid a global locking
1981 static void *s_start(struct seq_file *m, loff_t *pos)
1983 struct trace_iterator *iter = m->private;
1984 struct trace_array *tr = iter->tr;
1985 int cpu_file = iter->cpu_file;
1991 * copy the tracer to avoid using a global lock all around.
1992 * iter->trace is a copy of current_trace, the pointer to the
1993 * name may be used instead of a strcmp(), as iter->trace->name
1994 * will point to the same string as current_trace->name.
1996 mutex_lock(&trace_types_lock);
1997 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
1998 *iter->trace = *tr->current_trace;
1999 mutex_unlock(&trace_types_lock);
2001 #ifdef CONFIG_TRACER_MAX_TRACE
2002 if (iter->snapshot && iter->trace->use_max_tr)
2003 return ERR_PTR(-EBUSY);
2006 if (!iter->snapshot)
2007 atomic_inc(&trace_record_cmdline_disabled);
2009 if (*pos != iter->pos) {
2014 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2015 for_each_tracing_cpu(cpu)
2016 tracing_iter_reset(iter, cpu);
2018 tracing_iter_reset(iter, cpu_file);
2021 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2026 * If we overflowed the seq_file before, then we want
2027 * to just reuse the trace_seq buffer again.
2033 p = s_next(m, p, &l);
2037 trace_event_read_lock();
2038 trace_access_lock(cpu_file);
2042 static void s_stop(struct seq_file *m, void *p)
2044 struct trace_iterator *iter = m->private;
2046 #ifdef CONFIG_TRACER_MAX_TRACE
2047 if (iter->snapshot && iter->trace->use_max_tr)
2051 if (!iter->snapshot)
2052 atomic_dec(&trace_record_cmdline_disabled);
2054 trace_access_unlock(iter->cpu_file);
2055 trace_event_read_unlock();
2059 get_total_entries(struct trace_buffer *buf,
2060 unsigned long *total, unsigned long *entries)
2062 unsigned long count;
2068 for_each_tracing_cpu(cpu) {
2069 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2071 * If this buffer has skipped entries, then we hold all
2072 * entries for the trace and we need to ignore the
2073 * ones before the time stamp.
2075 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2076 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2077 /* total is the same as the entries */
2081 ring_buffer_overrun_cpu(buf->buffer, cpu);
2086 static void print_lat_help_header(struct seq_file *m)
2088 seq_puts(m, "# _------=> CPU# \n");
2089 seq_puts(m, "# / _-----=> irqs-off \n");
2090 seq_puts(m, "# | / _----=> need-resched \n");
2091 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2092 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2093 seq_puts(m, "# |||| / delay \n");
2094 seq_puts(m, "# cmd pid ||||| time | caller \n");
2095 seq_puts(m, "# \\ / ||||| \\ | / \n");
2098 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2100 unsigned long total;
2101 unsigned long entries;
2103 get_total_entries(buf, &total, &entries);
2104 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2105 entries, total, num_online_cpus());
2109 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2111 print_event_info(buf, m);
2112 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2113 seq_puts(m, "# | | | | |\n");
2116 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2118 print_event_info(buf, m);
2119 seq_puts(m, "# _-----=> irqs-off\n");
2120 seq_puts(m, "# / _----=> need-resched\n");
2121 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2122 seq_puts(m, "# || / _--=> preempt-depth\n");
2123 seq_puts(m, "# ||| / delay\n");
2124 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2125 seq_puts(m, "# | | | |||| | |\n");
2129 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2131 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2132 struct trace_buffer *buf = iter->trace_buffer;
2133 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2134 struct tracer *type = iter->trace;
2135 unsigned long entries;
2136 unsigned long total;
2137 const char *name = "preemption";
2141 get_total_entries(buf, &total, &entries);
2143 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2145 seq_puts(m, "# -----------------------------------"
2146 "---------------------------------\n");
2147 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2148 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2149 nsecs_to_usecs(data->saved_latency),
2153 #if defined(CONFIG_PREEMPT_NONE)
2155 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2157 #elif defined(CONFIG_PREEMPT)
2162 /* These are reserved for later use */
2165 seq_printf(m, " #P:%d)\n", num_online_cpus());
2169 seq_puts(m, "# -----------------\n");
2170 seq_printf(m, "# | task: %.16s-%d "
2171 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2172 data->comm, data->pid,
2173 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2174 data->policy, data->rt_priority);
2175 seq_puts(m, "# -----------------\n");
2177 if (data->critical_start) {
2178 seq_puts(m, "# => started at: ");
2179 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2180 trace_print_seq(m, &iter->seq);
2181 seq_puts(m, "\n# => ended at: ");
2182 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2183 trace_print_seq(m, &iter->seq);
2184 seq_puts(m, "\n#\n");
2190 static void test_cpu_buff_start(struct trace_iterator *iter)
2192 struct trace_seq *s = &iter->seq;
2194 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2197 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2200 if (cpumask_test_cpu(iter->cpu, iter->started))
2203 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2206 cpumask_set_cpu(iter->cpu, iter->started);
2208 /* Don't print started cpu buffer for the first entry of the trace */
2210 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2214 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2216 struct trace_seq *s = &iter->seq;
2217 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2218 struct trace_entry *entry;
2219 struct trace_event *event;
2223 test_cpu_buff_start(iter);
2225 event = ftrace_find_event(entry->type);
2227 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2228 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2229 if (!trace_print_lat_context(iter))
2232 if (!trace_print_context(iter))
2238 return event->funcs->trace(iter, sym_flags, event);
2240 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2243 return TRACE_TYPE_HANDLED;
2245 return TRACE_TYPE_PARTIAL_LINE;
2248 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2250 struct trace_seq *s = &iter->seq;
2251 struct trace_entry *entry;
2252 struct trace_event *event;
2256 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2257 if (!trace_seq_printf(s, "%d %d %llu ",
2258 entry->pid, iter->cpu, iter->ts))
2262 event = ftrace_find_event(entry->type);
2264 return event->funcs->raw(iter, 0, event);
2266 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2269 return TRACE_TYPE_HANDLED;
2271 return TRACE_TYPE_PARTIAL_LINE;
2274 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2276 struct trace_seq *s = &iter->seq;
2277 unsigned char newline = '\n';
2278 struct trace_entry *entry;
2279 struct trace_event *event;
2283 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2284 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2285 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2286 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2289 event = ftrace_find_event(entry->type);
2291 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2292 if (ret != TRACE_TYPE_HANDLED)
2296 SEQ_PUT_FIELD_RET(s, newline);
2298 return TRACE_TYPE_HANDLED;
2301 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2303 struct trace_seq *s = &iter->seq;
2304 struct trace_entry *entry;
2305 struct trace_event *event;
2309 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2310 SEQ_PUT_FIELD_RET(s, entry->pid);
2311 SEQ_PUT_FIELD_RET(s, iter->cpu);
2312 SEQ_PUT_FIELD_RET(s, iter->ts);
2315 event = ftrace_find_event(entry->type);
2316 return event ? event->funcs->binary(iter, 0, event) :
2320 int trace_empty(struct trace_iterator *iter)
2322 struct ring_buffer_iter *buf_iter;
2325 /* If we are looking at one CPU buffer, only check that one */
2326 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2327 cpu = iter->cpu_file;
2328 buf_iter = trace_buffer_iter(iter, cpu);
2330 if (!ring_buffer_iter_empty(buf_iter))
2333 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2339 for_each_tracing_cpu(cpu) {
2340 buf_iter = trace_buffer_iter(iter, cpu);
2342 if (!ring_buffer_iter_empty(buf_iter))
2345 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2353 /* Called with trace_event_read_lock() held. */
2354 enum print_line_t print_trace_line(struct trace_iterator *iter)
2356 enum print_line_t ret;
2358 if (iter->lost_events &&
2359 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2360 iter->cpu, iter->lost_events))
2361 return TRACE_TYPE_PARTIAL_LINE;
2363 if (iter->trace && iter->trace->print_line) {
2364 ret = iter->trace->print_line(iter);
2365 if (ret != TRACE_TYPE_UNHANDLED)
2369 if (iter->ent->type == TRACE_BPRINT &&
2370 trace_flags & TRACE_ITER_PRINTK &&
2371 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2372 return trace_print_bprintk_msg_only(iter);
2374 if (iter->ent->type == TRACE_PRINT &&
2375 trace_flags & TRACE_ITER_PRINTK &&
2376 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2377 return trace_print_printk_msg_only(iter);
2379 if (trace_flags & TRACE_ITER_BIN)
2380 return print_bin_fmt(iter);
2382 if (trace_flags & TRACE_ITER_HEX)
2383 return print_hex_fmt(iter);
2385 if (trace_flags & TRACE_ITER_RAW)
2386 return print_raw_fmt(iter);
2388 return print_trace_fmt(iter);
2391 void trace_latency_header(struct seq_file *m)
2393 struct trace_iterator *iter = m->private;
2395 /* print nothing if the buffers are empty */
2396 if (trace_empty(iter))
2399 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2400 print_trace_header(m, iter);
2402 if (!(trace_flags & TRACE_ITER_VERBOSE))
2403 print_lat_help_header(m);
2406 void trace_default_header(struct seq_file *m)
2408 struct trace_iterator *iter = m->private;
2410 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2413 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2414 /* print nothing if the buffers are empty */
2415 if (trace_empty(iter))
2417 print_trace_header(m, iter);
2418 if (!(trace_flags & TRACE_ITER_VERBOSE))
2419 print_lat_help_header(m);
2421 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2422 if (trace_flags & TRACE_ITER_IRQ_INFO)
2423 print_func_help_header_irq(iter->trace_buffer, m);
2425 print_func_help_header(iter->trace_buffer, m);
2430 static void test_ftrace_alive(struct seq_file *m)
2432 if (!ftrace_is_dead())
2434 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2435 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2438 #ifdef CONFIG_TRACER_MAX_TRACE
2439 static void show_snapshot_main_help(struct seq_file *m)
2441 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2442 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2443 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2444 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2445 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2446 seq_printf(m, "# is not a '0' or '1')\n");
2449 static void show_snapshot_percpu_help(struct seq_file *m)
2451 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2452 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2453 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2454 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2456 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2457 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2459 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2460 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2461 seq_printf(m, "# is not a '0' or '1')\n");
2464 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2466 if (iter->tr->allocated_snapshot)
2467 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2469 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2471 seq_printf(m, "# Snapshot commands:\n");
2472 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2473 show_snapshot_main_help(m);
2475 show_snapshot_percpu_help(m);
2478 /* Should never be called */
2479 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2482 static int s_show(struct seq_file *m, void *v)
2484 struct trace_iterator *iter = v;
2487 if (iter->ent == NULL) {
2489 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2491 test_ftrace_alive(m);
2493 if (iter->snapshot && trace_empty(iter))
2494 print_snapshot_help(m, iter);
2495 else if (iter->trace && iter->trace->print_header)
2496 iter->trace->print_header(m);
2498 trace_default_header(m);
2500 } else if (iter->leftover) {
2502 * If we filled the seq_file buffer earlier, we
2503 * want to just show it now.
2505 ret = trace_print_seq(m, &iter->seq);
2507 /* ret should this time be zero, but you never know */
2508 iter->leftover = ret;
2511 print_trace_line(iter);
2512 ret = trace_print_seq(m, &iter->seq);
2514 * If we overflow the seq_file buffer, then it will
2515 * ask us for this data again at start up.
2517 * ret is 0 if seq_file write succeeded.
2520 iter->leftover = ret;
2526 static const struct seq_operations tracer_seq_ops = {
2533 static struct trace_iterator *
2534 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2536 struct trace_cpu *tc = inode->i_private;
2537 struct trace_array *tr = tc->tr;
2538 struct trace_iterator *iter;
2541 if (tracing_disabled)
2542 return ERR_PTR(-ENODEV);
2544 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2546 return ERR_PTR(-ENOMEM);
2548 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2550 if (!iter->buffer_iter)
2554 * We make a copy of the current tracer to avoid concurrent
2555 * changes on it while we are reading.
2557 mutex_lock(&trace_types_lock);
2558 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2562 *iter->trace = *tr->current_trace;
2564 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2569 #ifdef CONFIG_TRACER_MAX_TRACE
2570 /* Currently only the top directory has a snapshot */
2571 if (tr->current_trace->print_max || snapshot)
2572 iter->trace_buffer = &tr->max_buffer;
2575 iter->trace_buffer = &tr->trace_buffer;
2576 iter->snapshot = snapshot;
2578 mutex_init(&iter->mutex);
2579 iter->cpu_file = tc->cpu;
2581 /* Notify the tracer early; before we stop tracing. */
2582 if (iter->trace && iter->trace->open)
2583 iter->trace->open(iter);
2585 /* Annotate start of buffers if we had overruns */
2586 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2587 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2589 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2590 if (trace_clocks[trace_clock_id].in_ns)
2591 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2593 /* stop the trace while dumping if we are not opening "snapshot" */
2594 if (!iter->snapshot)
2595 tracing_stop_tr(tr);
2597 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2598 for_each_tracing_cpu(cpu) {
2599 iter->buffer_iter[cpu] =
2600 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2602 ring_buffer_read_prepare_sync();
2603 for_each_tracing_cpu(cpu) {
2604 ring_buffer_read_start(iter->buffer_iter[cpu]);
2605 tracing_iter_reset(iter, cpu);
2608 cpu = iter->cpu_file;
2609 iter->buffer_iter[cpu] =
2610 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2611 ring_buffer_read_prepare_sync();
2612 ring_buffer_read_start(iter->buffer_iter[cpu]);
2613 tracing_iter_reset(iter, cpu);
2616 mutex_unlock(&trace_types_lock);
2621 mutex_unlock(&trace_types_lock);
2623 kfree(iter->buffer_iter);
2625 seq_release_private(inode, file);
2626 return ERR_PTR(-ENOMEM);
2629 int tracing_open_generic(struct inode *inode, struct file *filp)
2631 if (tracing_disabled)
2634 filp->private_data = inode->i_private;
2638 static int tracing_release(struct inode *inode, struct file *file)
2640 struct seq_file *m = file->private_data;
2641 struct trace_iterator *iter;
2642 struct trace_array *tr;
2645 if (!(file->f_mode & FMODE_READ))
2651 mutex_lock(&trace_types_lock);
2652 for_each_tracing_cpu(cpu) {
2653 if (iter->buffer_iter[cpu])
2654 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2657 if (iter->trace && iter->trace->close)
2658 iter->trace->close(iter);
2660 if (!iter->snapshot)
2661 /* reenable tracing if it was previously enabled */
2662 tracing_start_tr(tr);
2663 mutex_unlock(&trace_types_lock);
2665 mutex_destroy(&iter->mutex);
2666 free_cpumask_var(iter->started);
2668 kfree(iter->buffer_iter);
2669 seq_release_private(inode, file);
2673 static int tracing_open(struct inode *inode, struct file *file)
2675 struct trace_iterator *iter;
2678 /* If this file was open for write, then erase contents */
2679 if ((file->f_mode & FMODE_WRITE) &&
2680 (file->f_flags & O_TRUNC)) {
2681 struct trace_cpu *tc = inode->i_private;
2682 struct trace_array *tr = tc->tr;
2684 if (tc->cpu == RING_BUFFER_ALL_CPUS)
2685 tracing_reset_online_cpus(&tr->trace_buffer);
2687 tracing_reset(&tr->trace_buffer, tc->cpu);
2690 if (file->f_mode & FMODE_READ) {
2691 iter = __tracing_open(inode, file, false);
2693 ret = PTR_ERR(iter);
2694 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2695 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2701 t_next(struct seq_file *m, void *v, loff_t *pos)
2703 struct tracer *t = v;
2713 static void *t_start(struct seq_file *m, loff_t *pos)
2718 mutex_lock(&trace_types_lock);
2719 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2725 static void t_stop(struct seq_file *m, void *p)
2727 mutex_unlock(&trace_types_lock);
2730 static int t_show(struct seq_file *m, void *v)
2732 struct tracer *t = v;
2737 seq_printf(m, "%s", t->name);
2746 static const struct seq_operations show_traces_seq_ops = {
2753 static int show_traces_open(struct inode *inode, struct file *file)
2755 if (tracing_disabled)
2758 return seq_open(file, &show_traces_seq_ops);
2762 tracing_write_stub(struct file *filp, const char __user *ubuf,
2763 size_t count, loff_t *ppos)
2768 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2770 if (file->f_mode & FMODE_READ)
2771 return seq_lseek(file, offset, origin);
2776 static const struct file_operations tracing_fops = {
2777 .open = tracing_open,
2779 .write = tracing_write_stub,
2780 .llseek = tracing_seek,
2781 .release = tracing_release,
2784 static const struct file_operations show_traces_fops = {
2785 .open = show_traces_open,
2787 .release = seq_release,
2788 .llseek = seq_lseek,
2792 * Only trace on a CPU if the bitmask is set:
2794 static cpumask_var_t tracing_cpumask;
2797 * The tracer itself will not take this lock, but still we want
2798 * to provide a consistent cpumask to user-space:
2800 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2803 * Temporary storage for the character representation of the
2804 * CPU bitmask (and one more byte for the newline):
2806 static char mask_str[NR_CPUS + 1];
2809 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2810 size_t count, loff_t *ppos)
2814 mutex_lock(&tracing_cpumask_update_lock);
2816 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2817 if (count - len < 2) {
2821 len += sprintf(mask_str + len, "\n");
2822 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2825 mutex_unlock(&tracing_cpumask_update_lock);
2831 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2832 size_t count, loff_t *ppos)
2834 struct trace_array *tr = filp->private_data;
2835 cpumask_var_t tracing_cpumask_new;
2838 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2841 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2845 mutex_lock(&tracing_cpumask_update_lock);
2847 local_irq_disable();
2848 arch_spin_lock(&ftrace_max_lock);
2849 for_each_tracing_cpu(cpu) {
2851 * Increase/decrease the disabled counter if we are
2852 * about to flip a bit in the cpumask:
2854 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2855 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2856 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
2857 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
2859 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2860 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2861 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
2862 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
2865 arch_spin_unlock(&ftrace_max_lock);
2868 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2870 mutex_unlock(&tracing_cpumask_update_lock);
2871 free_cpumask_var(tracing_cpumask_new);
2876 free_cpumask_var(tracing_cpumask_new);
2881 static const struct file_operations tracing_cpumask_fops = {
2882 .open = tracing_open_generic,
2883 .read = tracing_cpumask_read,
2884 .write = tracing_cpumask_write,
2885 .llseek = generic_file_llseek,
2888 static int tracing_trace_options_show(struct seq_file *m, void *v)
2890 struct tracer_opt *trace_opts;
2891 struct trace_array *tr = m->private;
2895 mutex_lock(&trace_types_lock);
2896 tracer_flags = tr->current_trace->flags->val;
2897 trace_opts = tr->current_trace->flags->opts;
2899 for (i = 0; trace_options[i]; i++) {
2900 if (trace_flags & (1 << i))
2901 seq_printf(m, "%s\n", trace_options[i]);
2903 seq_printf(m, "no%s\n", trace_options[i]);
2906 for (i = 0; trace_opts[i].name; i++) {
2907 if (tracer_flags & trace_opts[i].bit)
2908 seq_printf(m, "%s\n", trace_opts[i].name);
2910 seq_printf(m, "no%s\n", trace_opts[i].name);
2912 mutex_unlock(&trace_types_lock);
2917 static int __set_tracer_option(struct tracer *trace,
2918 struct tracer_flags *tracer_flags,
2919 struct tracer_opt *opts, int neg)
2923 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2928 tracer_flags->val &= ~opts->bit;
2930 tracer_flags->val |= opts->bit;
2934 /* Try to assign a tracer specific option */
2935 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2937 struct tracer_flags *tracer_flags = trace->flags;
2938 struct tracer_opt *opts = NULL;
2941 for (i = 0; tracer_flags->opts[i].name; i++) {
2942 opts = &tracer_flags->opts[i];
2944 if (strcmp(cmp, opts->name) == 0)
2945 return __set_tracer_option(trace, trace->flags,
2952 /* Some tracers require overwrite to stay enabled */
2953 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
2955 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
2961 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
2963 /* do nothing if flag is already set */
2964 if (!!(trace_flags & mask) == !!enabled)
2967 /* Give the tracer a chance to approve the change */
2968 if (tr->current_trace->flag_changed)
2969 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
2973 trace_flags |= mask;
2975 trace_flags &= ~mask;
2977 if (mask == TRACE_ITER_RECORD_CMD)
2978 trace_event_enable_cmd_record(enabled);
2980 if (mask == TRACE_ITER_OVERWRITE) {
2981 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
2982 #ifdef CONFIG_TRACER_MAX_TRACE
2983 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
2987 if (mask == TRACE_ITER_PRINTK)
2988 trace_printk_start_stop_comm(enabled);
2993 static int trace_set_options(struct trace_array *tr, char *option)
3000 cmp = strstrip(option);
3002 if (strncmp(cmp, "no", 2) == 0) {
3007 mutex_lock(&trace_types_lock);
3009 for (i = 0; trace_options[i]; i++) {
3010 if (strcmp(cmp, trace_options[i]) == 0) {
3011 ret = set_tracer_flag(tr, 1 << i, !neg);
3016 /* If no option could be set, test the specific tracer options */
3017 if (!trace_options[i])
3018 ret = set_tracer_option(tr->current_trace, cmp, neg);
3020 mutex_unlock(&trace_types_lock);
3026 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3027 size_t cnt, loff_t *ppos)
3029 struct seq_file *m = filp->private_data;
3030 struct trace_array *tr = m->private;
3034 if (cnt >= sizeof(buf))
3037 if (copy_from_user(&buf, ubuf, cnt))
3042 ret = trace_set_options(tr, buf);
3051 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3053 if (tracing_disabled)
3056 return single_open(file, tracing_trace_options_show, inode->i_private);
3059 static const struct file_operations tracing_iter_fops = {
3060 .open = tracing_trace_options_open,
3062 .llseek = seq_lseek,
3063 .release = single_release,
3064 .write = tracing_trace_options_write,
3067 static const char readme_msg[] =
3068 "tracing mini-HOWTO:\n\n"
3069 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
3070 "# cat /sys/kernel/debug/tracing/available_tracers\n"
3071 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
3072 "# cat /sys/kernel/debug/tracing/current_tracer\n"
3074 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
3075 "# cat /sys/kernel/debug/tracing/current_tracer\n"
3077 "# cat /sys/kernel/debug/tracing/trace_options\n"
3078 "noprint-parent nosym-offset nosym-addr noverbose\n"
3079 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
3080 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
3081 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
3082 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
3086 tracing_readme_read(struct file *filp, char __user *ubuf,
3087 size_t cnt, loff_t *ppos)
3089 return simple_read_from_buffer(ubuf, cnt, ppos,
3090 readme_msg, strlen(readme_msg));
3093 static const struct file_operations tracing_readme_fops = {
3094 .open = tracing_open_generic,
3095 .read = tracing_readme_read,
3096 .llseek = generic_file_llseek,
3100 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3101 size_t cnt, loff_t *ppos)
3110 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3114 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3122 for (i = 0; i < SAVED_CMDLINES; i++) {
3125 pid = map_cmdline_to_pid[i];
3126 if (pid == -1 || pid == NO_CMDLINE_MAP)
3129 trace_find_cmdline(pid, buf_comm);
3130 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3135 len = simple_read_from_buffer(ubuf, cnt, ppos,
3144 static const struct file_operations tracing_saved_cmdlines_fops = {
3145 .open = tracing_open_generic,
3146 .read = tracing_saved_cmdlines_read,
3147 .llseek = generic_file_llseek,
3151 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3152 size_t cnt, loff_t *ppos)
3154 struct trace_array *tr = filp->private_data;
3155 char buf[MAX_TRACER_SIZE+2];
3158 mutex_lock(&trace_types_lock);
3159 r = sprintf(buf, "%s\n", tr->current_trace->name);
3160 mutex_unlock(&trace_types_lock);
3162 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3165 int tracer_init(struct tracer *t, struct trace_array *tr)
3167 tracing_reset_online_cpus(&tr->trace_buffer);
3171 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3175 for_each_tracing_cpu(cpu)
3176 per_cpu_ptr(buf->data, cpu)->entries = val;
3179 #ifdef CONFIG_TRACER_MAX_TRACE
3180 /* resize @tr's buffer to the size of @size_tr's entries */
3181 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3182 struct trace_buffer *size_buf, int cpu_id)
3186 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3187 for_each_tracing_cpu(cpu) {
3188 ret = ring_buffer_resize(trace_buf->buffer,
3189 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3192 per_cpu_ptr(trace_buf->data, cpu)->entries =
3193 per_cpu_ptr(size_buf->data, cpu)->entries;
3196 ret = ring_buffer_resize(trace_buf->buffer,
3197 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3199 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3200 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3205 #endif /* CONFIG_TRACER_MAX_TRACE */
3207 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3208 unsigned long size, int cpu)
3213 * If kernel or user changes the size of the ring buffer
3214 * we use the size that was given, and we can forget about
3215 * expanding it later.
3217 ring_buffer_expanded = 1;
3219 /* May be called before buffers are initialized */
3220 if (!tr->trace_buffer.buffer)
3223 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3227 #ifdef CONFIG_TRACER_MAX_TRACE
3228 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3229 !tr->current_trace->use_max_tr)
3232 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3234 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3235 &tr->trace_buffer, cpu);
3238 * AARGH! We are left with different
3239 * size max buffer!!!!
3240 * The max buffer is our "snapshot" buffer.
3241 * When a tracer needs a snapshot (one of the
3242 * latency tracers), it swaps the max buffer
3243 * with the saved snap shot. We succeeded to
3244 * update the size of the main buffer, but failed to
3245 * update the size of the max buffer. But when we tried
3246 * to reset the main buffer to the original size, we
3247 * failed there too. This is very unlikely to
3248 * happen, but if it does, warn and kill all
3252 tracing_disabled = 1;
3257 if (cpu == RING_BUFFER_ALL_CPUS)
3258 set_buffer_entries(&tr->max_buffer, size);
3260 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3263 #endif /* CONFIG_TRACER_MAX_TRACE */
3265 if (cpu == RING_BUFFER_ALL_CPUS)
3266 set_buffer_entries(&tr->trace_buffer, size);
3268 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3273 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3274 unsigned long size, int cpu_id)
3278 mutex_lock(&trace_types_lock);
3280 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3281 /* make sure, this cpu is enabled in the mask */
3282 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3288 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3293 mutex_unlock(&trace_types_lock);
3300 * tracing_update_buffers - used by tracing facility to expand ring buffers
3302 * To save on memory when the tracing is never used on a system with it
3303 * configured in. The ring buffers are set to a minimum size. But once
3304 * a user starts to use the tracing facility, then they need to grow
3305 * to their default size.
3307 * This function is to be called when a tracer is about to be used.
3309 int tracing_update_buffers(void)
3313 mutex_lock(&trace_types_lock);
3314 if (!ring_buffer_expanded)
3315 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3316 RING_BUFFER_ALL_CPUS);
3317 mutex_unlock(&trace_types_lock);
3322 struct trace_option_dentry;
3324 static struct trace_option_dentry *
3325 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3328 destroy_trace_option_files(struct trace_option_dentry *topts);
3330 static int tracing_set_tracer(const char *buf)
3332 static struct trace_option_dentry *topts;
3333 struct trace_array *tr = &global_trace;
3335 #ifdef CONFIG_TRACER_MAX_TRACE
3340 mutex_lock(&trace_types_lock);
3342 if (!ring_buffer_expanded) {
3343 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3344 RING_BUFFER_ALL_CPUS);
3350 for (t = trace_types; t; t = t->next) {
3351 if (strcmp(t->name, buf) == 0)
3358 if (t == tr->current_trace)
3361 trace_branch_disable();
3363 tr->current_trace->enabled = false;
3365 if (tr->current_trace->reset)
3366 tr->current_trace->reset(tr);
3368 /* Current trace needs to be nop_trace before synchronize_sched */
3369 tr->current_trace = &nop_trace;
3371 #ifdef CONFIG_TRACER_MAX_TRACE
3372 had_max_tr = tr->allocated_snapshot;
3374 if (had_max_tr && !t->use_max_tr) {
3376 * We need to make sure that the update_max_tr sees that
3377 * current_trace changed to nop_trace to keep it from
3378 * swapping the buffers after we resize it.
3379 * The update_max_tr is called from interrupts disabled
3380 * so a synchronized_sched() is sufficient.
3382 synchronize_sched();
3384 * We don't free the ring buffer. instead, resize it because
3385 * The max_tr ring buffer has some state (e.g. ring->clock) and
3386 * we want preserve it.
3388 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
3389 set_buffer_entries(&tr->max_buffer, 1);
3390 tracing_reset_online_cpus(&tr->max_buffer);
3391 tr->allocated_snapshot = false;
3394 destroy_trace_option_files(topts);
3396 topts = create_trace_option_files(tr, t);
3398 #ifdef CONFIG_TRACER_MAX_TRACE
3399 if (t->use_max_tr && !had_max_tr) {
3400 /* we need to make per cpu buffer sizes equivalent */
3401 ret = resize_buffer_duplicate_size(&tr->max_buffer, &tr->trace_buffer,
3402 RING_BUFFER_ALL_CPUS);
3405 tr->allocated_snapshot = true;
3410 ret = tracer_init(t, tr);
3415 tr->current_trace = t;
3416 tr->current_trace->enabled = true;
3417 trace_branch_enable(tr);
3419 mutex_unlock(&trace_types_lock);
3425 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3426 size_t cnt, loff_t *ppos)
3428 char buf[MAX_TRACER_SIZE+1];
3435 if (cnt > MAX_TRACER_SIZE)
3436 cnt = MAX_TRACER_SIZE;
3438 if (copy_from_user(&buf, ubuf, cnt))
3443 /* strip ending whitespace. */
3444 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3447 err = tracing_set_tracer(buf);
3457 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3458 size_t cnt, loff_t *ppos)
3460 unsigned long *ptr = filp->private_data;
3464 r = snprintf(buf, sizeof(buf), "%ld\n",
3465 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3466 if (r > sizeof(buf))
3468 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3472 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3473 size_t cnt, loff_t *ppos)
3475 unsigned long *ptr = filp->private_data;
3479 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3488 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3490 struct trace_cpu *tc = inode->i_private;
3491 struct trace_array *tr = tc->tr;
3492 struct trace_iterator *iter;
3495 if (tracing_disabled)
3498 mutex_lock(&trace_types_lock);
3500 /* create a buffer to store the information to pass to userspace */
3501 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3508 * We make a copy of the current tracer to avoid concurrent
3509 * changes on it while we are reading.
3511 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3516 *iter->trace = *tr->current_trace;
3518 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3523 /* trace pipe does not show start of buffer */
3524 cpumask_setall(iter->started);
3526 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3527 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3529 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3530 if (trace_clocks[trace_clock_id].in_ns)
3531 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3533 iter->cpu_file = tc->cpu;
3535 iter->trace_buffer = &tc->tr->trace_buffer;
3536 mutex_init(&iter->mutex);
3537 filp->private_data = iter;
3539 if (iter->trace->pipe_open)
3540 iter->trace->pipe_open(iter);
3542 nonseekable_open(inode, filp);
3544 mutex_unlock(&trace_types_lock);
3550 mutex_unlock(&trace_types_lock);
3554 static int tracing_release_pipe(struct inode *inode, struct file *file)
3556 struct trace_iterator *iter = file->private_data;
3558 mutex_lock(&trace_types_lock);
3560 if (iter->trace->pipe_close)
3561 iter->trace->pipe_close(iter);
3563 mutex_unlock(&trace_types_lock);
3565 free_cpumask_var(iter->started);
3566 mutex_destroy(&iter->mutex);
3574 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
3576 /* Iterators are static, they should be filled or empty */
3577 if (trace_buffer_iter(iter, iter->cpu_file))
3578 return POLLIN | POLLRDNORM;
3580 if (trace_flags & TRACE_ITER_BLOCK)
3582 * Always select as readable when in blocking mode
3584 return POLLIN | POLLRDNORM;
3586 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
3591 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3593 struct trace_iterator *iter = filp->private_data;
3595 return trace_poll(iter, filp, poll_table);
3599 * This is a make-shift waitqueue.
3600 * A tracer might use this callback on some rare cases:
3602 * 1) the current tracer might hold the runqueue lock when it wakes up
3603 * a reader, hence a deadlock (sched, function, and function graph tracers)
3604 * 2) the function tracers, trace all functions, we don't want
3605 * the overhead of calling wake_up and friends
3606 * (and tracing them too)
3608 * Anyway, this is really very primitive wakeup.
3610 void poll_wait_pipe(struct trace_iterator *iter)
3612 set_current_state(TASK_INTERRUPTIBLE);
3613 /* sleep for 100 msecs, and try again. */
3614 schedule_timeout(HZ / 10);
3617 /* Must be called with trace_types_lock mutex held. */
3618 static int tracing_wait_pipe(struct file *filp)
3620 struct trace_iterator *iter = filp->private_data;
3622 while (trace_empty(iter)) {
3624 if ((filp->f_flags & O_NONBLOCK)) {
3628 mutex_unlock(&iter->mutex);
3630 iter->trace->wait_pipe(iter);
3632 mutex_lock(&iter->mutex);
3634 if (signal_pending(current))
3638 * We block until we read something and tracing is disabled.
3639 * We still block if tracing is disabled, but we have never
3640 * read anything. This allows a user to cat this file, and
3641 * then enable tracing. But after we have read something,
3642 * we give an EOF when tracing is again disabled.
3644 * iter->pos will be 0 if we haven't read anything.
3646 if (!tracing_is_enabled() && iter->pos)
3657 tracing_read_pipe(struct file *filp, char __user *ubuf,
3658 size_t cnt, loff_t *ppos)
3660 struct trace_iterator *iter = filp->private_data;
3661 struct trace_array *tr = iter->tr;
3664 /* return any leftover data */
3665 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3669 trace_seq_init(&iter->seq);
3671 /* copy the tracer to avoid using a global lock all around */
3672 mutex_lock(&trace_types_lock);
3673 if (unlikely(iter->trace->name != tr->current_trace->name))
3674 *iter->trace = *tr->current_trace;
3675 mutex_unlock(&trace_types_lock);
3678 * Avoid more than one consumer on a single file descriptor
3679 * This is just a matter of traces coherency, the ring buffer itself
3682 mutex_lock(&iter->mutex);
3683 if (iter->trace->read) {
3684 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3690 sret = tracing_wait_pipe(filp);
3694 /* stop when tracing is finished */
3695 if (trace_empty(iter)) {
3700 if (cnt >= PAGE_SIZE)
3701 cnt = PAGE_SIZE - 1;
3703 /* reset all but tr, trace, and overruns */
3704 memset(&iter->seq, 0,
3705 sizeof(struct trace_iterator) -
3706 offsetof(struct trace_iterator, seq));
3709 trace_event_read_lock();
3710 trace_access_lock(iter->cpu_file);
3711 while (trace_find_next_entry_inc(iter) != NULL) {
3712 enum print_line_t ret;
3713 int len = iter->seq.len;
3715 ret = print_trace_line(iter);
3716 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3717 /* don't print partial lines */
3718 iter->seq.len = len;
3721 if (ret != TRACE_TYPE_NO_CONSUME)
3722 trace_consume(iter);
3724 if (iter->seq.len >= cnt)
3728 * Setting the full flag means we reached the trace_seq buffer
3729 * size and we should leave by partial output condition above.
3730 * One of the trace_seq_* functions is not used properly.
3732 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3735 trace_access_unlock(iter->cpu_file);
3736 trace_event_read_unlock();
3738 /* Now copy what we have to the user */
3739 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3740 if (iter->seq.readpos >= iter->seq.len)
3741 trace_seq_init(&iter->seq);
3744 * If there was nothing to send to user, in spite of consuming trace
3745 * entries, go back to wait for more entries.
3751 mutex_unlock(&iter->mutex);
3756 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3757 struct pipe_buffer *buf)
3759 __free_page(buf->page);
3762 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3765 __free_page(spd->pages[idx]);
3768 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3770 .map = generic_pipe_buf_map,
3771 .unmap = generic_pipe_buf_unmap,
3772 .confirm = generic_pipe_buf_confirm,
3773 .release = tracing_pipe_buf_release,
3774 .steal = generic_pipe_buf_steal,
3775 .get = generic_pipe_buf_get,
3779 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3784 /* Seq buffer is page-sized, exactly what we need. */
3786 count = iter->seq.len;
3787 ret = print_trace_line(iter);
3788 count = iter->seq.len - count;
3791 iter->seq.len -= count;
3794 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3795 iter->seq.len -= count;
3799 if (ret != TRACE_TYPE_NO_CONSUME)
3800 trace_consume(iter);
3802 if (!trace_find_next_entry_inc(iter)) {
3812 static ssize_t tracing_splice_read_pipe(struct file *filp,
3814 struct pipe_inode_info *pipe,
3818 struct page *pages_def[PIPE_DEF_BUFFERS];
3819 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3820 struct trace_iterator *iter = filp->private_data;
3821 struct splice_pipe_desc spd = {
3823 .partial = partial_def,
3824 .nr_pages = 0, /* This gets updated below. */
3825 .nr_pages_max = PIPE_DEF_BUFFERS,
3827 .ops = &tracing_pipe_buf_ops,
3828 .spd_release = tracing_spd_release_pipe,
3830 struct trace_array *tr = iter->tr;
3835 if (splice_grow_spd(pipe, &spd))
3838 /* copy the tracer to avoid using a global lock all around */
3839 mutex_lock(&trace_types_lock);
3840 if (unlikely(iter->trace->name != tr->current_trace->name))
3841 *iter->trace = *tr->current_trace;
3842 mutex_unlock(&trace_types_lock);
3844 mutex_lock(&iter->mutex);
3846 if (iter->trace->splice_read) {
3847 ret = iter->trace->splice_read(iter, filp,
3848 ppos, pipe, len, flags);
3853 ret = tracing_wait_pipe(filp);
3857 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3862 trace_event_read_lock();
3863 trace_access_lock(iter->cpu_file);
3865 /* Fill as many pages as possible. */
3866 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3867 spd.pages[i] = alloc_page(GFP_KERNEL);
3871 rem = tracing_fill_pipe_page(rem, iter);
3873 /* Copy the data into the page, so we can start over. */
3874 ret = trace_seq_to_buffer(&iter->seq,
3875 page_address(spd.pages[i]),
3878 __free_page(spd.pages[i]);
3881 spd.partial[i].offset = 0;
3882 spd.partial[i].len = iter->seq.len;
3884 trace_seq_init(&iter->seq);
3887 trace_access_unlock(iter->cpu_file);
3888 trace_event_read_unlock();
3889 mutex_unlock(&iter->mutex);
3893 ret = splice_to_pipe(pipe, &spd);
3895 splice_shrink_spd(&spd);
3899 mutex_unlock(&iter->mutex);
3904 tracing_entries_read(struct file *filp, char __user *ubuf,
3905 size_t cnt, loff_t *ppos)
3907 struct trace_cpu *tc = filp->private_data;
3908 struct trace_array *tr = tc->tr;
3913 mutex_lock(&trace_types_lock);
3915 if (tc->cpu == RING_BUFFER_ALL_CPUS) {
3916 int cpu, buf_size_same;
3921 /* check if all cpu sizes are same */
3922 for_each_tracing_cpu(cpu) {
3923 /* fill in the size from first enabled cpu */
3925 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
3926 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
3932 if (buf_size_same) {
3933 if (!ring_buffer_expanded)
3934 r = sprintf(buf, "%lu (expanded: %lu)\n",
3936 trace_buf_size >> 10);
3938 r = sprintf(buf, "%lu\n", size >> 10);
3940 r = sprintf(buf, "X\n");
3942 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, tc->cpu)->entries >> 10);
3944 mutex_unlock(&trace_types_lock);
3946 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3951 tracing_entries_write(struct file *filp, const char __user *ubuf,
3952 size_t cnt, loff_t *ppos)
3954 struct trace_cpu *tc = filp->private_data;
3958 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3962 /* must have at least 1 entry */
3966 /* value is in KB */
3969 ret = tracing_resize_ring_buffer(tc->tr, val, tc->cpu);
3979 tracing_total_entries_read(struct file *filp, char __user *ubuf,
3980 size_t cnt, loff_t *ppos)
3982 struct trace_array *tr = filp->private_data;
3985 unsigned long size = 0, expanded_size = 0;
3987 mutex_lock(&trace_types_lock);
3988 for_each_tracing_cpu(cpu) {
3989 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
3990 if (!ring_buffer_expanded)
3991 expanded_size += trace_buf_size >> 10;
3993 if (ring_buffer_expanded)
3994 r = sprintf(buf, "%lu\n", size);
3996 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3997 mutex_unlock(&trace_types_lock);
3999 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4003 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4004 size_t cnt, loff_t *ppos)
4007 * There is no need to read what the user has written, this function
4008 * is just to make sure that there is no error when "echo" is used
4017 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4019 struct trace_array *tr = inode->i_private;
4021 /* disable tracing ? */
4022 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4024 /* resize the ring buffer to 0 */
4025 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4031 tracing_mark_write(struct file *filp, const char __user *ubuf,
4032 size_t cnt, loff_t *fpos)
4034 unsigned long addr = (unsigned long)ubuf;
4035 struct ring_buffer_event *event;
4036 struct ring_buffer *buffer;
4037 struct print_entry *entry;
4038 unsigned long irq_flags;
4039 struct page *pages[2];
4049 if (tracing_disabled)
4052 if (!(trace_flags & TRACE_ITER_MARKERS))
4055 if (cnt > TRACE_BUF_SIZE)
4056 cnt = TRACE_BUF_SIZE;
4059 * Userspace is injecting traces into the kernel trace buffer.
4060 * We want to be as non intrusive as possible.
4061 * To do so, we do not want to allocate any special buffers
4062 * or take any locks, but instead write the userspace data
4063 * straight into the ring buffer.
4065 * First we need to pin the userspace buffer into memory,
4066 * which, most likely it is, because it just referenced it.
4067 * But there's no guarantee that it is. By using get_user_pages_fast()
4068 * and kmap_atomic/kunmap_atomic() we can get access to the
4069 * pages directly. We then write the data directly into the
4072 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4074 /* check if we cross pages */
4075 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4078 offset = addr & (PAGE_SIZE - 1);
4081 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4082 if (ret < nr_pages) {
4084 put_page(pages[ret]);
4089 for (i = 0; i < nr_pages; i++)
4090 map_page[i] = kmap_atomic(pages[i]);
4092 local_save_flags(irq_flags);
4093 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4094 buffer = global_trace.trace_buffer.buffer;
4095 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4096 irq_flags, preempt_count());
4098 /* Ring buffer disabled, return as if not open for write */
4103 entry = ring_buffer_event_data(event);
4104 entry->ip = _THIS_IP_;
4106 if (nr_pages == 2) {
4107 len = PAGE_SIZE - offset;
4108 memcpy(&entry->buf, map_page[0] + offset, len);
4109 memcpy(&entry->buf[len], map_page[1], cnt - len);
4111 memcpy(&entry->buf, map_page[0] + offset, cnt);
4113 if (entry->buf[cnt - 1] != '\n') {
4114 entry->buf[cnt] = '\n';
4115 entry->buf[cnt + 1] = '\0';
4117 entry->buf[cnt] = '\0';
4119 __buffer_unlock_commit(buffer, event);
4126 for (i = 0; i < nr_pages; i++){
4127 kunmap_atomic(map_page[i]);
4134 static int tracing_clock_show(struct seq_file *m, void *v)
4136 struct trace_array *tr = m->private;
4139 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4141 "%s%s%s%s", i ? " " : "",
4142 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4143 i == tr->clock_id ? "]" : "");
4149 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4150 size_t cnt, loff_t *fpos)
4152 struct seq_file *m = filp->private_data;
4153 struct trace_array *tr = m->private;
4155 const char *clockstr;
4158 if (cnt >= sizeof(buf))
4161 if (copy_from_user(&buf, ubuf, cnt))
4166 clockstr = strstrip(buf);
4168 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4169 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4172 if (i == ARRAY_SIZE(trace_clocks))
4175 mutex_lock(&trace_types_lock);
4179 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4182 * New clock may not be consistent with the previous clock.
4183 * Reset the buffer so that it doesn't have incomparable timestamps.
4185 tracing_reset_online_cpus(&global_trace.trace_buffer);
4187 #ifdef CONFIG_TRACER_MAX_TRACE
4188 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4189 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4190 tracing_reset_online_cpus(&global_trace.max_buffer);
4193 mutex_unlock(&trace_types_lock);
4200 static int tracing_clock_open(struct inode *inode, struct file *file)
4202 if (tracing_disabled)
4205 return single_open(file, tracing_clock_show, inode->i_private);
4208 struct ftrace_buffer_info {
4209 struct trace_iterator iter;
4214 #ifdef CONFIG_TRACER_SNAPSHOT
4215 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4217 struct trace_cpu *tc = inode->i_private;
4218 struct trace_iterator *iter;
4222 if (file->f_mode & FMODE_READ) {
4223 iter = __tracing_open(inode, file, true);
4225 ret = PTR_ERR(iter);
4227 /* Writes still need the seq_file to hold the private data */
4228 m = kzalloc(sizeof(*m), GFP_KERNEL);
4231 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4237 iter->trace_buffer = &tc->tr->max_buffer;
4238 iter->cpu_file = tc->cpu;
4240 file->private_data = m;
4247 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4250 struct seq_file *m = filp->private_data;
4251 struct trace_iterator *iter = m->private;
4252 struct trace_array *tr = iter->tr;
4256 ret = tracing_update_buffers();
4260 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4264 mutex_lock(&trace_types_lock);
4266 if (tr->current_trace->use_max_tr) {
4273 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4277 if (tr->allocated_snapshot) {
4278 /* free spare buffer */
4279 ring_buffer_resize(tr->max_buffer.buffer, 1,
4280 RING_BUFFER_ALL_CPUS);
4281 set_buffer_entries(&tr->max_buffer, 1);
4282 tracing_reset_online_cpus(&tr->max_buffer);
4283 tr->allocated_snapshot = false;
4287 /* Only allow per-cpu swap if the ring buffer supports it */
4288 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4289 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4294 if (!tr->allocated_snapshot) {
4295 /* allocate spare buffer */
4296 ret = resize_buffer_duplicate_size(&tr->max_buffer,
4297 &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
4300 tr->allocated_snapshot = true;
4302 local_irq_disable();
4303 /* Now, we're going to swap */
4304 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4305 update_max_tr(&global_trace, current, smp_processor_id());
4307 update_max_tr_single(&global_trace, current, iter->cpu_file);
4311 if (tr->allocated_snapshot) {
4312 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4313 tracing_reset_online_cpus(&tr->max_buffer);
4315 tracing_reset(&tr->max_buffer, iter->cpu_file);
4325 mutex_unlock(&trace_types_lock);
4329 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4331 struct seq_file *m = file->private_data;
4333 if (file->f_mode & FMODE_READ)
4334 return tracing_release(inode, file);
4336 /* If write only, the seq_file is just a stub */
4344 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4345 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4346 size_t count, loff_t *ppos);
4347 static int tracing_buffers_release(struct inode *inode, struct file *file);
4348 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4349 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4351 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4353 struct ftrace_buffer_info *info;
4356 ret = tracing_buffers_open(inode, filp);
4360 info = filp->private_data;
4362 if (info->iter.trace->use_max_tr) {
4363 tracing_buffers_release(inode, filp);
4367 info->iter.snapshot = true;
4368 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4373 #endif /* CONFIG_TRACER_SNAPSHOT */
4376 static const struct file_operations tracing_max_lat_fops = {
4377 .open = tracing_open_generic,
4378 .read = tracing_max_lat_read,
4379 .write = tracing_max_lat_write,
4380 .llseek = generic_file_llseek,
4383 static const struct file_operations set_tracer_fops = {
4384 .open = tracing_open_generic,
4385 .read = tracing_set_trace_read,
4386 .write = tracing_set_trace_write,
4387 .llseek = generic_file_llseek,
4390 static const struct file_operations tracing_pipe_fops = {
4391 .open = tracing_open_pipe,
4392 .poll = tracing_poll_pipe,
4393 .read = tracing_read_pipe,
4394 .splice_read = tracing_splice_read_pipe,
4395 .release = tracing_release_pipe,
4396 .llseek = no_llseek,
4399 static const struct file_operations tracing_entries_fops = {
4400 .open = tracing_open_generic,
4401 .read = tracing_entries_read,
4402 .write = tracing_entries_write,
4403 .llseek = generic_file_llseek,
4406 static const struct file_operations tracing_total_entries_fops = {
4407 .open = tracing_open_generic,
4408 .read = tracing_total_entries_read,
4409 .llseek = generic_file_llseek,
4412 static const struct file_operations tracing_free_buffer_fops = {
4413 .write = tracing_free_buffer_write,
4414 .release = tracing_free_buffer_release,
4417 static const struct file_operations tracing_mark_fops = {
4418 .open = tracing_open_generic,
4419 .write = tracing_mark_write,
4420 .llseek = generic_file_llseek,
4423 static const struct file_operations trace_clock_fops = {
4424 .open = tracing_clock_open,
4426 .llseek = seq_lseek,
4427 .release = single_release,
4428 .write = tracing_clock_write,
4431 #ifdef CONFIG_TRACER_SNAPSHOT
4432 static const struct file_operations snapshot_fops = {
4433 .open = tracing_snapshot_open,
4435 .write = tracing_snapshot_write,
4436 .llseek = tracing_seek,
4437 .release = tracing_snapshot_release,
4440 static const struct file_operations snapshot_raw_fops = {
4441 .open = snapshot_raw_open,
4442 .read = tracing_buffers_read,
4443 .release = tracing_buffers_release,
4444 .splice_read = tracing_buffers_splice_read,
4445 .llseek = no_llseek,
4448 #endif /* CONFIG_TRACER_SNAPSHOT */
4450 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4452 struct trace_cpu *tc = inode->i_private;
4453 struct trace_array *tr = tc->tr;
4454 struct ftrace_buffer_info *info;
4456 if (tracing_disabled)
4459 info = kzalloc(sizeof(*info), GFP_KERNEL);
4464 info->iter.cpu_file = tc->cpu;
4465 info->iter.trace = tr->current_trace;
4466 info->iter.trace_buffer = &tr->trace_buffer;
4468 /* Force reading ring buffer for first read */
4469 info->read = (unsigned int)-1;
4471 filp->private_data = info;
4473 return nonseekable_open(inode, filp);
4477 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4479 struct ftrace_buffer_info *info = filp->private_data;
4480 struct trace_iterator *iter = &info->iter;
4482 return trace_poll(iter, filp, poll_table);
4486 tracing_buffers_read(struct file *filp, char __user *ubuf,
4487 size_t count, loff_t *ppos)
4489 struct ftrace_buffer_info *info = filp->private_data;
4490 struct trace_iterator *iter = &info->iter;
4497 mutex_lock(&trace_types_lock);
4499 #ifdef CONFIG_TRACER_MAX_TRACE
4500 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4507 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
4513 /* Do we have previous read data to read? */
4514 if (info->read < PAGE_SIZE)
4518 trace_access_lock(iter->cpu_file);
4519 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
4523 trace_access_unlock(iter->cpu_file);
4526 if (trace_empty(iter)) {
4527 if ((filp->f_flags & O_NONBLOCK)) {
4531 mutex_unlock(&trace_types_lock);
4532 iter->trace->wait_pipe(iter);
4533 mutex_lock(&trace_types_lock);
4534 if (signal_pending(current)) {
4546 size = PAGE_SIZE - info->read;
4550 ret = copy_to_user(ubuf, info->spare + info->read, size);
4561 mutex_unlock(&trace_types_lock);
4566 static int tracing_buffers_release(struct inode *inode, struct file *file)
4568 struct ftrace_buffer_info *info = file->private_data;
4569 struct trace_iterator *iter = &info->iter;
4572 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
4579 struct ring_buffer *buffer;
4584 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4585 struct pipe_buffer *buf)
4587 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4592 ring_buffer_free_read_page(ref->buffer, ref->page);
4597 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4598 struct pipe_buffer *buf)
4600 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4605 /* Pipe buffer operations for a buffer. */
4606 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4608 .map = generic_pipe_buf_map,
4609 .unmap = generic_pipe_buf_unmap,
4610 .confirm = generic_pipe_buf_confirm,
4611 .release = buffer_pipe_buf_release,
4612 .steal = generic_pipe_buf_steal,
4613 .get = buffer_pipe_buf_get,
4617 * Callback from splice_to_pipe(), if we need to release some pages
4618 * at the end of the spd in case we error'ed out in filling the pipe.
4620 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4622 struct buffer_ref *ref =
4623 (struct buffer_ref *)spd->partial[i].private;
4628 ring_buffer_free_read_page(ref->buffer, ref->page);
4630 spd->partial[i].private = 0;
4634 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4635 struct pipe_inode_info *pipe, size_t len,
4638 struct ftrace_buffer_info *info = file->private_data;
4639 struct trace_iterator *iter = &info->iter;
4640 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4641 struct page *pages_def[PIPE_DEF_BUFFERS];
4642 struct splice_pipe_desc spd = {
4644 .partial = partial_def,
4645 .nr_pages_max = PIPE_DEF_BUFFERS,
4647 .ops = &buffer_pipe_buf_ops,
4648 .spd_release = buffer_spd_release,
4650 struct buffer_ref *ref;
4651 int entries, size, i;
4654 mutex_lock(&trace_types_lock);
4656 #ifdef CONFIG_TRACER_MAX_TRACE
4657 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4663 if (splice_grow_spd(pipe, &spd)) {
4668 if (*ppos & (PAGE_SIZE - 1)) {
4673 if (len & (PAGE_SIZE - 1)) {
4674 if (len < PAGE_SIZE) {
4682 trace_access_lock(iter->cpu_file);
4683 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
4685 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4689 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4694 ref->buffer = iter->trace_buffer->buffer;
4695 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
4701 r = ring_buffer_read_page(ref->buffer, &ref->page,
4702 len, iter->cpu_file, 1);
4704 ring_buffer_free_read_page(ref->buffer, ref->page);
4710 * zero out any left over data, this is going to
4713 size = ring_buffer_page_len(ref->page);
4714 if (size < PAGE_SIZE)
4715 memset(ref->page + size, 0, PAGE_SIZE - size);
4717 page = virt_to_page(ref->page);
4719 spd.pages[i] = page;
4720 spd.partial[i].len = PAGE_SIZE;
4721 spd.partial[i].offset = 0;
4722 spd.partial[i].private = (unsigned long)ref;
4726 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
4729 trace_access_unlock(iter->cpu_file);
4732 /* did we read anything? */
4733 if (!spd.nr_pages) {
4734 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
4738 mutex_unlock(&trace_types_lock);
4739 iter->trace->wait_pipe(iter);
4740 mutex_lock(&trace_types_lock);
4741 if (signal_pending(current)) {
4748 ret = splice_to_pipe(pipe, &spd);
4749 splice_shrink_spd(&spd);
4751 mutex_unlock(&trace_types_lock);
4756 static const struct file_operations tracing_buffers_fops = {
4757 .open = tracing_buffers_open,
4758 .read = tracing_buffers_read,
4759 .poll = tracing_buffers_poll,
4760 .release = tracing_buffers_release,
4761 .splice_read = tracing_buffers_splice_read,
4762 .llseek = no_llseek,
4766 tracing_stats_read(struct file *filp, char __user *ubuf,
4767 size_t count, loff_t *ppos)
4769 struct trace_cpu *tc = filp->private_data;
4770 struct trace_array *tr = tc->tr;
4771 struct trace_buffer *trace_buf = &tr->trace_buffer;
4772 struct trace_seq *s;
4774 unsigned long long t;
4775 unsigned long usec_rem;
4778 s = kmalloc(sizeof(*s), GFP_KERNEL);
4784 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
4785 trace_seq_printf(s, "entries: %ld\n", cnt);
4787 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
4788 trace_seq_printf(s, "overrun: %ld\n", cnt);
4790 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
4791 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4793 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
4794 trace_seq_printf(s, "bytes: %ld\n", cnt);
4796 if (trace_clocks[trace_clock_id].in_ns) {
4797 /* local or global for trace_clock */
4798 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
4799 usec_rem = do_div(t, USEC_PER_SEC);
4800 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
4803 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
4804 usec_rem = do_div(t, USEC_PER_SEC);
4805 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4807 /* counter or tsc mode for trace_clock */
4808 trace_seq_printf(s, "oldest event ts: %llu\n",
4809 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
4811 trace_seq_printf(s, "now ts: %llu\n",
4812 ring_buffer_time_stamp(trace_buf->buffer, cpu));
4815 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
4816 trace_seq_printf(s, "dropped events: %ld\n", cnt);
4818 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
4819 trace_seq_printf(s, "read events: %ld\n", cnt);
4821 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4828 static const struct file_operations tracing_stats_fops = {
4829 .open = tracing_open_generic,
4830 .read = tracing_stats_read,
4831 .llseek = generic_file_llseek,
4834 #ifdef CONFIG_DYNAMIC_FTRACE
4836 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4842 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4843 size_t cnt, loff_t *ppos)
4845 static char ftrace_dyn_info_buffer[1024];
4846 static DEFINE_MUTEX(dyn_info_mutex);
4847 unsigned long *p = filp->private_data;
4848 char *buf = ftrace_dyn_info_buffer;
4849 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4852 mutex_lock(&dyn_info_mutex);
4853 r = sprintf(buf, "%ld ", *p);
4855 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4858 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4860 mutex_unlock(&dyn_info_mutex);
4865 static const struct file_operations tracing_dyn_info_fops = {
4866 .open = tracing_open_generic,
4867 .read = tracing_read_dyn_info,
4868 .llseek = generic_file_llseek,
4872 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
4879 if (!debugfs_initialized())
4882 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
4883 tr->dir = debugfs_create_dir("tracing", NULL);
4885 if (!tr->dir && !once) {
4887 pr_warning("Could not create debugfs directory 'tracing'\n");
4894 struct dentry *tracing_init_dentry(void)
4896 return tracing_init_dentry_tr(&global_trace);
4899 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
4901 struct dentry *d_tracer;
4904 return tr->percpu_dir;
4906 d_tracer = tracing_init_dentry_tr(tr);
4910 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
4912 WARN_ONCE(!tr->percpu_dir,
4913 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
4915 return tr->percpu_dir;
4919 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
4921 struct trace_array_cpu *data = per_cpu_ptr(tr->trace_buffer.data, cpu);
4922 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
4923 struct dentry *d_cpu;
4924 char cpu_dir[30]; /* 30 characters should be more than enough */
4929 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4930 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4932 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4936 /* per cpu trace_pipe */
4937 trace_create_file("trace_pipe", 0444, d_cpu,
4938 (void *)&data->trace_cpu, &tracing_pipe_fops);
4941 trace_create_file("trace", 0644, d_cpu,
4942 (void *)&data->trace_cpu, &tracing_fops);
4944 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4945 (void *)&data->trace_cpu, &tracing_buffers_fops);
4947 trace_create_file("stats", 0444, d_cpu,
4948 (void *)&data->trace_cpu, &tracing_stats_fops);
4950 trace_create_file("buffer_size_kb", 0444, d_cpu,
4951 (void *)&data->trace_cpu, &tracing_entries_fops);
4953 #ifdef CONFIG_TRACER_SNAPSHOT
4954 trace_create_file("snapshot", 0644, d_cpu,
4955 (void *)&data->trace_cpu, &snapshot_fops);
4957 trace_create_file("snapshot_raw", 0444, d_cpu,
4958 (void *)&data->trace_cpu, &snapshot_raw_fops);
4962 #ifdef CONFIG_FTRACE_SELFTEST
4963 /* Let selftest have access to static functions in this file */
4964 #include "trace_selftest.c"
4967 struct trace_option_dentry {
4968 struct tracer_opt *opt;
4969 struct tracer_flags *flags;
4970 struct trace_array *tr;
4971 struct dentry *entry;
4975 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4978 struct trace_option_dentry *topt = filp->private_data;
4981 if (topt->flags->val & topt->opt->bit)
4986 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4990 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4993 struct trace_option_dentry *topt = filp->private_data;
4997 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5001 if (val != 0 && val != 1)
5004 if (!!(topt->flags->val & topt->opt->bit) != val) {
5005 mutex_lock(&trace_types_lock);
5006 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5008 mutex_unlock(&trace_types_lock);
5019 static const struct file_operations trace_options_fops = {
5020 .open = tracing_open_generic,
5021 .read = trace_options_read,
5022 .write = trace_options_write,
5023 .llseek = generic_file_llseek,
5027 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5030 long index = (long)filp->private_data;
5033 if (trace_flags & (1 << index))
5038 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5042 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5045 struct trace_array *tr = &global_trace;
5046 long index = (long)filp->private_data;
5050 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5054 if (val != 0 && val != 1)
5057 mutex_lock(&trace_types_lock);
5058 ret = set_tracer_flag(tr, 1 << index, val);
5059 mutex_unlock(&trace_types_lock);
5069 static const struct file_operations trace_options_core_fops = {
5070 .open = tracing_open_generic,
5071 .read = trace_options_core_read,
5072 .write = trace_options_core_write,
5073 .llseek = generic_file_llseek,
5076 struct dentry *trace_create_file(const char *name,
5078 struct dentry *parent,
5080 const struct file_operations *fops)
5084 ret = debugfs_create_file(name, mode, parent, data, fops);
5086 pr_warning("Could not create debugfs '%s' entry\n", name);
5092 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5094 struct dentry *d_tracer;
5099 d_tracer = tracing_init_dentry_tr(tr);
5103 tr->options = debugfs_create_dir("options", d_tracer);
5105 pr_warning("Could not create debugfs directory 'options'\n");
5113 create_trace_option_file(struct trace_array *tr,
5114 struct trace_option_dentry *topt,
5115 struct tracer_flags *flags,
5116 struct tracer_opt *opt)
5118 struct dentry *t_options;
5120 t_options = trace_options_init_dentry(tr);
5124 topt->flags = flags;
5128 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5129 &trace_options_fops);
5133 static struct trace_option_dentry *
5134 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5136 struct trace_option_dentry *topts;
5137 struct tracer_flags *flags;
5138 struct tracer_opt *opts;
5144 flags = tracer->flags;
5146 if (!flags || !flags->opts)
5151 for (cnt = 0; opts[cnt].name; cnt++)
5154 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5158 for (cnt = 0; opts[cnt].name; cnt++)
5159 create_trace_option_file(tr, &topts[cnt], flags,
5166 destroy_trace_option_files(struct trace_option_dentry *topts)
5173 for (cnt = 0; topts[cnt].opt; cnt++) {
5174 if (topts[cnt].entry)
5175 debugfs_remove(topts[cnt].entry);
5181 static struct dentry *
5182 create_trace_option_core_file(struct trace_array *tr,
5183 const char *option, long index)
5185 struct dentry *t_options;
5187 t_options = trace_options_init_dentry(tr);
5191 return trace_create_file(option, 0644, t_options, (void *)index,
5192 &trace_options_core_fops);
5195 static __init void create_trace_options_dir(struct trace_array *tr)
5197 struct dentry *t_options;
5200 t_options = trace_options_init_dentry(tr);
5204 for (i = 0; trace_options[i]; i++)
5205 create_trace_option_core_file(tr, trace_options[i], i);
5209 rb_simple_read(struct file *filp, char __user *ubuf,
5210 size_t cnt, loff_t *ppos)
5212 struct trace_array *tr = filp->private_data;
5213 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5218 r = ring_buffer_record_is_on(buffer);
5222 r = sprintf(buf, "%d\n", r);
5224 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5228 rb_simple_write(struct file *filp, const char __user *ubuf,
5229 size_t cnt, loff_t *ppos)
5231 struct trace_array *tr = filp->private_data;
5232 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5236 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5241 mutex_lock(&trace_types_lock);
5243 ring_buffer_record_on(buffer);
5244 if (tr->current_trace->start)
5245 tr->current_trace->start(tr);
5247 ring_buffer_record_off(buffer);
5248 if (tr->current_trace->stop)
5249 tr->current_trace->stop(tr);
5251 mutex_unlock(&trace_types_lock);
5259 static const struct file_operations rb_simple_fops = {
5260 .open = tracing_open_generic,
5261 .read = rb_simple_read,
5262 .write = rb_simple_write,
5263 .llseek = default_llseek,
5266 struct dentry *trace_instance_dir;
5269 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5271 static void init_trace_buffers(struct trace_array *tr, struct trace_buffer *buf)
5275 for_each_tracing_cpu(cpu) {
5276 memset(per_cpu_ptr(buf->data, cpu), 0, sizeof(struct trace_array_cpu));
5277 per_cpu_ptr(buf->data, cpu)->trace_cpu.cpu = cpu;
5278 per_cpu_ptr(buf->data, cpu)->trace_cpu.tr = tr;
5282 static int allocate_trace_buffers(struct trace_array *tr, int size)
5284 enum ring_buffer_flags rb_flags;
5286 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5288 tr->trace_buffer.buffer = ring_buffer_alloc(size, rb_flags);
5289 if (!tr->trace_buffer.buffer)
5292 tr->trace_buffer.data = alloc_percpu(struct trace_array_cpu);
5293 if (!tr->trace_buffer.data)
5296 init_trace_buffers(tr, &tr->trace_buffer);
5298 /* Allocate the first page for all buffers */
5299 set_buffer_entries(&tr->trace_buffer,
5300 ring_buffer_size(tr->trace_buffer.buffer, 0));
5302 #ifdef CONFIG_TRACER_MAX_TRACE
5304 tr->max_buffer.buffer = ring_buffer_alloc(1, rb_flags);
5305 if (!tr->max_buffer.buffer)
5308 tr->max_buffer.data = alloc_percpu(struct trace_array_cpu);
5309 if (!tr->max_buffer.data)
5312 init_trace_buffers(tr, &tr->max_buffer);
5314 set_buffer_entries(&tr->max_buffer, 1);
5319 if (tr->trace_buffer.buffer)
5320 ring_buffer_free(tr->trace_buffer.buffer);
5321 free_percpu(tr->trace_buffer.data);
5323 #ifdef CONFIG_TRACER_MAX_TRACE
5324 if (tr->max_buffer.buffer)
5325 ring_buffer_free(tr->max_buffer.buffer);
5326 free_percpu(tr->max_buffer.data);
5331 static int new_instance_create(const char *name)
5333 struct trace_array *tr;
5336 mutex_lock(&trace_types_lock);
5339 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5340 if (tr->name && strcmp(tr->name, name) == 0)
5345 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5349 tr->name = kstrdup(name, GFP_KERNEL);
5353 raw_spin_lock_init(&tr->start_lock);
5355 tr->current_trace = &nop_trace;
5357 INIT_LIST_HEAD(&tr->systems);
5358 INIT_LIST_HEAD(&tr->events);
5360 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5363 /* Holder for file callbacks */
5364 tr->trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
5365 tr->trace_cpu.tr = tr;
5367 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5371 ret = event_trace_add_tracer(tr->dir, tr);
5375 init_tracer_debugfs(tr, tr->dir);
5377 list_add(&tr->list, &ftrace_trace_arrays);
5379 mutex_unlock(&trace_types_lock);
5384 if (tr->trace_buffer.buffer)
5385 ring_buffer_free(tr->trace_buffer.buffer);
5390 mutex_unlock(&trace_types_lock);
5396 static int instance_delete(const char *name)
5398 struct trace_array *tr;
5402 mutex_lock(&trace_types_lock);
5405 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5406 if (tr->name && strcmp(tr->name, name) == 0) {
5414 list_del(&tr->list);
5416 event_trace_del_tracer(tr);
5417 debugfs_remove_recursive(tr->dir);
5418 free_percpu(tr->trace_buffer.data);
5419 ring_buffer_free(tr->trace_buffer.buffer);
5427 mutex_unlock(&trace_types_lock);
5432 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
5434 struct dentry *parent;
5437 /* Paranoid: Make sure the parent is the "instances" directory */
5438 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
5439 if (WARN_ON_ONCE(parent != trace_instance_dir))
5443 * The inode mutex is locked, but debugfs_create_dir() will also
5444 * take the mutex. As the instances directory can not be destroyed
5445 * or changed in any other way, it is safe to unlock it, and
5446 * let the dentry try. If two users try to make the same dir at
5447 * the same time, then the new_instance_create() will determine the
5450 mutex_unlock(&inode->i_mutex);
5452 ret = new_instance_create(dentry->d_iname);
5454 mutex_lock(&inode->i_mutex);
5459 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
5461 struct dentry *parent;
5464 /* Paranoid: Make sure the parent is the "instances" directory */
5465 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
5466 if (WARN_ON_ONCE(parent != trace_instance_dir))
5469 /* The caller did a dget() on dentry */
5470 mutex_unlock(&dentry->d_inode->i_mutex);
5473 * The inode mutex is locked, but debugfs_create_dir() will also
5474 * take the mutex. As the instances directory can not be destroyed
5475 * or changed in any other way, it is safe to unlock it, and
5476 * let the dentry try. If two users try to make the same dir at
5477 * the same time, then the instance_delete() will determine the
5480 mutex_unlock(&inode->i_mutex);
5482 ret = instance_delete(dentry->d_iname);
5484 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
5485 mutex_lock(&dentry->d_inode->i_mutex);
5490 static const struct inode_operations instance_dir_inode_operations = {
5491 .lookup = simple_lookup,
5492 .mkdir = instance_mkdir,
5493 .rmdir = instance_rmdir,
5496 static __init void create_trace_instances(struct dentry *d_tracer)
5498 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
5499 if (WARN_ON(!trace_instance_dir))
5502 /* Hijack the dir inode operations, to allow mkdir */
5503 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
5507 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
5510 trace_create_file("trace_options", 0644, d_tracer,
5511 tr, &tracing_iter_fops);
5513 trace_create_file("trace", 0644, d_tracer,
5514 (void *)&tr->trace_cpu, &tracing_fops);
5516 trace_create_file("trace_pipe", 0444, d_tracer,
5517 (void *)&tr->trace_cpu, &tracing_pipe_fops);
5519 trace_create_file("buffer_size_kb", 0644, d_tracer,
5520 (void *)&tr->trace_cpu, &tracing_entries_fops);
5522 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
5523 tr, &tracing_total_entries_fops);
5525 trace_create_file("free_buffer", 0644, d_tracer,
5526 tr, &tracing_free_buffer_fops);
5528 trace_create_file("trace_marker", 0220, d_tracer,
5529 tr, &tracing_mark_fops);
5531 trace_create_file("trace_clock", 0644, d_tracer, tr,
5534 trace_create_file("tracing_on", 0644, d_tracer,
5535 tr, &rb_simple_fops);
5538 static __init int tracer_init_debugfs(void)
5540 struct dentry *d_tracer;
5543 trace_access_lock_init();
5545 d_tracer = tracing_init_dentry();
5547 init_tracer_debugfs(&global_trace, d_tracer);
5549 trace_create_file("tracing_cpumask", 0644, d_tracer,
5550 &global_trace, &tracing_cpumask_fops);
5552 trace_create_file("available_tracers", 0444, d_tracer,
5553 &global_trace, &show_traces_fops);
5555 trace_create_file("current_tracer", 0644, d_tracer,
5556 &global_trace, &set_tracer_fops);
5558 #ifdef CONFIG_TRACER_MAX_TRACE
5559 trace_create_file("tracing_max_latency", 0644, d_tracer,
5560 &tracing_max_latency, &tracing_max_lat_fops);
5563 trace_create_file("tracing_thresh", 0644, d_tracer,
5564 &tracing_thresh, &tracing_max_lat_fops);
5566 trace_create_file("README", 0444, d_tracer,
5567 NULL, &tracing_readme_fops);
5569 trace_create_file("saved_cmdlines", 0444, d_tracer,
5570 NULL, &tracing_saved_cmdlines_fops);
5572 #ifdef CONFIG_DYNAMIC_FTRACE
5573 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
5574 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
5577 #ifdef CONFIG_TRACER_SNAPSHOT
5578 trace_create_file("snapshot", 0644, d_tracer,
5579 (void *)&global_trace.trace_cpu, &snapshot_fops);
5582 create_trace_instances(d_tracer);
5584 create_trace_options_dir(&global_trace);
5586 for_each_tracing_cpu(cpu)
5587 tracing_init_debugfs_percpu(&global_trace, cpu);
5592 static int trace_panic_handler(struct notifier_block *this,
5593 unsigned long event, void *unused)
5595 if (ftrace_dump_on_oops)
5596 ftrace_dump(ftrace_dump_on_oops);
5600 static struct notifier_block trace_panic_notifier = {
5601 .notifier_call = trace_panic_handler,
5603 .priority = 150 /* priority: INT_MAX >= x >= 0 */
5606 static int trace_die_handler(struct notifier_block *self,
5612 if (ftrace_dump_on_oops)
5613 ftrace_dump(ftrace_dump_on_oops);
5621 static struct notifier_block trace_die_notifier = {
5622 .notifier_call = trace_die_handler,
5627 * printk is set to max of 1024, we really don't need it that big.
5628 * Nothing should be printing 1000 characters anyway.
5630 #define TRACE_MAX_PRINT 1000
5633 * Define here KERN_TRACE so that we have one place to modify
5634 * it if we decide to change what log level the ftrace dump
5637 #define KERN_TRACE KERN_EMERG
5640 trace_printk_seq(struct trace_seq *s)
5642 /* Probably should print a warning here. */
5646 /* should be zero ended, but we are paranoid. */
5647 s->buffer[s->len] = 0;
5649 printk(KERN_TRACE "%s", s->buffer);
5654 void trace_init_global_iter(struct trace_iterator *iter)
5656 iter->tr = &global_trace;
5657 iter->trace = iter->tr->current_trace;
5658 iter->cpu_file = RING_BUFFER_ALL_CPUS;
5659 iter->trace_buffer = &global_trace.trace_buffer;
5663 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
5665 static arch_spinlock_t ftrace_dump_lock =
5666 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
5667 /* use static because iter can be a bit big for the stack */
5668 static struct trace_iterator iter;
5669 unsigned int old_userobj;
5670 static int dump_ran;
5671 unsigned long flags;
5675 local_irq_save(flags);
5676 arch_spin_lock(&ftrace_dump_lock);
5684 /* Did function tracer already get disabled? */
5685 if (ftrace_is_dead()) {
5686 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
5687 printk("# MAY BE MISSING FUNCTION EVENTS\n");
5690 if (disable_tracing)
5693 /* Simulate the iterator */
5694 trace_init_global_iter(&iter);
5696 for_each_tracing_cpu(cpu) {
5697 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
5700 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
5702 /* don't look at user memory in panic mode */
5703 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
5705 switch (oops_dump_mode) {
5707 iter.cpu_file = RING_BUFFER_ALL_CPUS;
5710 iter.cpu_file = raw_smp_processor_id();
5715 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
5716 iter.cpu_file = RING_BUFFER_ALL_CPUS;
5719 printk(KERN_TRACE "Dumping ftrace buffer:\n");
5722 * We need to stop all tracing on all CPUS to read the
5723 * the next buffer. This is a bit expensive, but is
5724 * not done often. We fill all what we can read,
5725 * and then release the locks again.
5728 while (!trace_empty(&iter)) {
5731 printk(KERN_TRACE "---------------------------------\n");
5735 /* reset all but tr, trace, and overruns */
5736 memset(&iter.seq, 0,
5737 sizeof(struct trace_iterator) -
5738 offsetof(struct trace_iterator, seq));
5739 iter.iter_flags |= TRACE_FILE_LAT_FMT;
5742 if (trace_find_next_entry_inc(&iter) != NULL) {
5745 ret = print_trace_line(&iter);
5746 if (ret != TRACE_TYPE_NO_CONSUME)
5747 trace_consume(&iter);
5749 touch_nmi_watchdog();
5751 trace_printk_seq(&iter.seq);
5755 printk(KERN_TRACE " (ftrace buffer empty)\n");
5757 printk(KERN_TRACE "---------------------------------\n");
5760 /* Re-enable tracing if requested */
5761 if (!disable_tracing) {
5762 trace_flags |= old_userobj;
5764 for_each_tracing_cpu(cpu) {
5765 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
5771 arch_spin_unlock(&ftrace_dump_lock);
5772 local_irq_restore(flags);
5775 /* By default: disable tracing after the dump */
5776 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
5778 __ftrace_dump(true, oops_dump_mode);
5780 EXPORT_SYMBOL_GPL(ftrace_dump);
5782 __init static int tracer_alloc_buffers(void)
5788 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
5791 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
5792 goto out_free_buffer_mask;
5794 /* Only allocate trace_printk buffers if a trace_printk exists */
5795 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
5796 /* Must be called before global_trace.buffer is allocated */
5797 trace_printk_init_buffers();
5799 /* To save memory, keep the ring buffer size to its minimum */
5800 if (ring_buffer_expanded)
5801 ring_buf_size = trace_buf_size;
5805 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
5806 cpumask_copy(tracing_cpumask, cpu_all_mask);
5808 raw_spin_lock_init(&global_trace.start_lock);
5810 /* TODO: make the number of buffers hot pluggable with CPUS */
5811 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
5812 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
5814 goto out_free_cpumask;
5817 if (global_trace.buffer_disabled)
5820 trace_init_cmdlines();
5822 register_tracer(&nop_trace);
5824 global_trace.current_trace = &nop_trace;
5826 /* All seems OK, enable tracing */
5827 tracing_disabled = 0;
5829 atomic_notifier_chain_register(&panic_notifier_list,
5830 &trace_panic_notifier);
5832 register_die_notifier(&trace_die_notifier);
5834 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
5836 /* Holder for file callbacks */
5837 global_trace.trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
5838 global_trace.trace_cpu.tr = &global_trace;
5840 INIT_LIST_HEAD(&global_trace.systems);
5841 INIT_LIST_HEAD(&global_trace.events);
5842 list_add(&global_trace.list, &ftrace_trace_arrays);
5844 while (trace_boot_options) {
5847 option = strsep(&trace_boot_options, ",");
5848 trace_set_options(&global_trace, option);
5854 free_percpu(global_trace.trace_buffer.data);
5855 #ifdef CONFIG_TRACER_MAX_TRACE
5856 free_percpu(global_trace.max_buffer.data);
5858 free_cpumask_var(tracing_cpumask);
5859 out_free_buffer_mask:
5860 free_cpumask_var(tracing_buffer_mask);
5865 __init static int clear_boot_tracer(void)
5868 * The default tracer at boot buffer is an init section.
5869 * This function is called in lateinit. If we did not
5870 * find the boot tracer, then clear it out, to prevent
5871 * later registration from accessing the buffer that is
5872 * about to be freed.
5874 if (!default_bootup_tracer)
5877 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
5878 default_bootup_tracer);
5879 default_bootup_tracer = NULL;
5884 early_initcall(tracer_alloc_buffers);
5885 fs_initcall(tracer_init_debugfs);
5886 late_initcall(clear_boot_tracer);