2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 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 William Lee Irwin III
14 #include <linux/ring_buffer.h>
15 #include <linux/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/ctype.h>
34 #include <linux/init.h>
35 #include <linux/poll.h>
36 #include <linux/gfp.h>
40 #include "trace_output.h"
42 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
44 unsigned long __read_mostly tracing_max_latency;
45 unsigned long __read_mostly tracing_thresh;
48 * On boot up, the ring buffer is set to the minimum size, so that
49 * we do not waste memory on systems that are not using tracing.
51 static int ring_buffer_expanded;
54 * We need to change this state when a selftest is running.
55 * A selftest will lurk into the ring-buffer to count the
56 * entries inserted during the selftest although some concurrent
57 * insertions into the ring-buffer such as trace_printk could occurred
58 * at the same time, giving false positive or negative results.
60 static bool __read_mostly tracing_selftest_running;
63 * If a tracer is running, we do not want to run SELFTEST.
65 static bool __read_mostly tracing_selftest_disabled;
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt[] = {
72 static struct tracer_flags dummy_tracer_flags = {
74 .opts = dummy_tracer_opt
77 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
83 * Kill all tracing for good (never come back).
84 * It is initialized to 1 but will turn to zero if the initialization
85 * of the tracer is successful. But that is the only place that sets
88 static int tracing_disabled = 1;
90 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
92 static inline void ftrace_disable_cpu(void)
95 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
98 static inline void ftrace_enable_cpu(void)
100 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
104 static cpumask_var_t __read_mostly tracing_buffer_mask;
106 /* Define which cpu buffers are currently read in trace_pipe */
107 static cpumask_var_t tracing_reader_cpumask;
109 #define for_each_tracing_cpu(cpu) \
110 for_each_cpu(cpu, tracing_buffer_mask)
113 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
115 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
116 * is set, then ftrace_dump is called. This will output the contents
117 * of the ftrace buffers to the console. This is very useful for
118 * capturing traces that lead to crashes and outputing it to a
121 * It is default off, but you can enable it with either specifying
122 * "ftrace_dump_on_oops" in the kernel command line, or setting
123 * /proc/sys/kernel/ftrace_dump_on_oops to true.
125 int ftrace_dump_on_oops;
127 static int tracing_set_tracer(const char *buf);
129 #define BOOTUP_TRACER_SIZE 100
130 static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata;
131 static char *default_bootup_tracer;
133 static int __init set_ftrace(char *str)
135 strncpy(bootup_tracer_buf, str, BOOTUP_TRACER_SIZE);
136 default_bootup_tracer = bootup_tracer_buf;
137 /* We are using ftrace early, expand it */
138 ring_buffer_expanded = 1;
141 __setup("ftrace=", set_ftrace);
143 static int __init set_ftrace_dump_on_oops(char *str)
145 ftrace_dump_on_oops = 1;
148 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
151 ns2usecs(cycle_t nsec)
159 * The global_trace is the descriptor that holds the tracing
160 * buffers for the live tracing. For each CPU, it contains
161 * a link list of pages that will store trace entries. The
162 * page descriptor of the pages in the memory is used to hold
163 * the link list by linking the lru item in the page descriptor
164 * to each of the pages in the buffer per CPU.
166 * For each active CPU there is a data field that holds the
167 * pages for the buffer for that CPU. Each CPU has the same number
168 * of pages allocated for its buffer.
170 static struct trace_array global_trace;
172 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
174 cycle_t ftrace_now(int cpu)
178 /* Early boot up does not have a buffer yet */
179 if (!global_trace.buffer)
180 return trace_clock_local();
182 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
183 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
189 * The max_tr is used to snapshot the global_trace when a maximum
190 * latency is reached. Some tracers will use this to store a maximum
191 * trace while it continues examining live traces.
193 * The buffers for the max_tr are set up the same as the global_trace.
194 * When a snapshot is taken, the link list of the max_tr is swapped
195 * with the link list of the global_trace and the buffers are reset for
196 * the global_trace so the tracing can continue.
198 static struct trace_array max_tr;
200 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
202 /* tracer_enabled is used to toggle activation of a tracer */
203 static int tracer_enabled = 1;
206 * tracing_is_enabled - return tracer_enabled status
208 * This function is used by other tracers to know the status
209 * of the tracer_enabled flag. Tracers may use this function
210 * to know if it should enable their features when starting
211 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
213 int tracing_is_enabled(void)
215 return tracer_enabled;
219 * trace_buf_size is the size in bytes that is allocated
220 * for a buffer. Note, the number of bytes is always rounded
223 * This number is purposely set to a low number of 16384.
224 * If the dump on oops happens, it will be much appreciated
225 * to not have to wait for all that output. Anyway this can be
226 * boot time and run time configurable.
228 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
230 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
232 /* trace_types holds a link list of available tracers. */
233 static struct tracer *trace_types __read_mostly;
235 /* current_trace points to the tracer that is currently active */
236 static struct tracer *current_trace __read_mostly;
239 * max_tracer_type_len is used to simplify the allocating of
240 * buffers to read userspace tracer names. We keep track of
241 * the longest tracer name registered.
243 static int max_tracer_type_len;
246 * trace_types_lock is used to protect the trace_types list.
247 * This lock is also used to keep user access serialized.
248 * Accesses from userspace will grab this lock while userspace
249 * activities happen inside the kernel.
251 static DEFINE_MUTEX(trace_types_lock);
253 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
254 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
256 /* trace_flags holds trace_options default values */
257 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
258 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO;
261 * trace_wake_up - wake up tasks waiting for trace input
263 * Simply wakes up any task that is blocked on the trace_wait
264 * queue. These is used with trace_poll for tasks polling the trace.
266 void trace_wake_up(void)
269 * The runqueue_is_locked() can fail, but this is the best we
272 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
273 wake_up(&trace_wait);
276 static int __init set_buf_size(char *str)
278 unsigned long buf_size;
283 ret = strict_strtoul(str, 0, &buf_size);
284 /* nr_entries can not be zero */
285 if (ret < 0 || buf_size == 0)
287 trace_buf_size = buf_size;
290 __setup("trace_buf_size=", set_buf_size);
292 unsigned long nsecs_to_usecs(unsigned long nsecs)
297 /* These must match the bit postions in trace_iterator_flags */
298 static const char *trace_options[] = {
323 * ftrace_max_lock is used to protect the swapping of buffers
324 * when taking a max snapshot. The buffers themselves are
325 * protected by per_cpu spinlocks. But the action of the swap
326 * needs its own lock.
328 * This is defined as a raw_spinlock_t in order to help
329 * with performance when lockdep debugging is enabled.
331 static raw_spinlock_t ftrace_max_lock =
332 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
335 * Copy the new maximum trace into the separate maximum-trace
336 * structure. (this way the maximum trace is permanently saved,
337 * for later retrieval via /debugfs/tracing/latency_trace)
340 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
342 struct trace_array_cpu *data = tr->data[cpu];
345 max_tr.time_start = data->preempt_timestamp;
347 data = max_tr.data[cpu];
348 data->saved_latency = tracing_max_latency;
350 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
351 data->pid = tsk->pid;
352 data->uid = task_uid(tsk);
353 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
354 data->policy = tsk->policy;
355 data->rt_priority = tsk->rt_priority;
357 /* record this tasks comm */
358 tracing_record_cmdline(tsk);
361 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
369 if (s->len <= s->readpos)
372 len = s->len - s->readpos;
375 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
385 ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
390 if (s->len <= s->readpos)
393 len = s->len - s->readpos;
396 ret = memcpy(buf, s->buffer + s->readpos, cnt);
405 trace_print_seq(struct seq_file *m, struct trace_seq *s)
407 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
410 seq_puts(m, s->buffer);
416 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
418 * @tsk: the task with the latency
419 * @cpu: The cpu that initiated the trace.
421 * Flip the buffers between the @tr and the max_tr and record information
422 * about which task was the cause of this latency.
425 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
427 struct ring_buffer *buf = tr->buffer;
429 WARN_ON_ONCE(!irqs_disabled());
430 __raw_spin_lock(&ftrace_max_lock);
432 tr->buffer = max_tr.buffer;
435 ftrace_disable_cpu();
436 ring_buffer_reset(tr->buffer);
439 __update_max_tr(tr, tsk, cpu);
440 __raw_spin_unlock(&ftrace_max_lock);
444 * update_max_tr_single - only copy one trace over, and reset the rest
446 * @tsk - task with the latency
447 * @cpu - the cpu of the buffer to copy.
449 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
452 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
456 WARN_ON_ONCE(!irqs_disabled());
457 __raw_spin_lock(&ftrace_max_lock);
459 ftrace_disable_cpu();
461 ring_buffer_reset(max_tr.buffer);
462 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
466 WARN_ON_ONCE(ret && ret != -EAGAIN);
468 __update_max_tr(tr, tsk, cpu);
469 __raw_spin_unlock(&ftrace_max_lock);
473 * register_tracer - register a tracer with the ftrace system.
474 * @type - the plugin for the tracer
476 * Register a new plugin tracer.
478 int register_tracer(struct tracer *type)
479 __releases(kernel_lock)
480 __acquires(kernel_lock)
487 pr_info("Tracer must have a name\n");
492 * When this gets called we hold the BKL which means that
493 * preemption is disabled. Various trace selftests however
494 * need to disable and enable preemption for successful tests.
495 * So we drop the BKL here and grab it after the tests again.
498 mutex_lock(&trace_types_lock);
500 tracing_selftest_running = true;
502 for (t = trace_types; t; t = t->next) {
503 if (strcmp(type->name, t->name) == 0) {
505 pr_info("Trace %s already registered\n",
513 type->set_flag = &dummy_set_flag;
515 type->flags = &dummy_tracer_flags;
517 if (!type->flags->opts)
518 type->flags->opts = dummy_tracer_opt;
519 if (!type->wait_pipe)
520 type->wait_pipe = default_wait_pipe;
523 #ifdef CONFIG_FTRACE_STARTUP_TEST
524 if (type->selftest && !tracing_selftest_disabled) {
525 struct tracer *saved_tracer = current_trace;
526 struct trace_array *tr = &global_trace;
530 * Run a selftest on this tracer.
531 * Here we reset the trace buffer, and set the current
532 * tracer to be this tracer. The tracer can then run some
533 * internal tracing to verify that everything is in order.
534 * If we fail, we do not register this tracer.
536 for_each_tracing_cpu(i)
537 tracing_reset(tr, i);
539 current_trace = type;
540 /* the test is responsible for initializing and enabling */
541 pr_info("Testing tracer %s: ", type->name);
542 ret = type->selftest(type, tr);
543 /* the test is responsible for resetting too */
544 current_trace = saved_tracer;
546 printk(KERN_CONT "FAILED!\n");
549 /* Only reset on passing, to avoid touching corrupted buffers */
550 for_each_tracing_cpu(i)
551 tracing_reset(tr, i);
553 printk(KERN_CONT "PASSED\n");
557 type->next = trace_types;
559 len = strlen(type->name);
560 if (len > max_tracer_type_len)
561 max_tracer_type_len = len;
564 tracing_selftest_running = false;
565 mutex_unlock(&trace_types_lock);
567 if (ret || !default_bootup_tracer)
570 if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE))
573 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
574 /* Do we want this tracer to start on bootup? */
575 tracing_set_tracer(type->name);
576 default_bootup_tracer = NULL;
577 /* disable other selftests, since this will break it. */
578 tracing_selftest_disabled = 1;
579 #ifdef CONFIG_FTRACE_STARTUP_TEST
580 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
589 void unregister_tracer(struct tracer *type)
594 mutex_lock(&trace_types_lock);
595 for (t = &trace_types; *t; t = &(*t)->next) {
599 pr_info("Trace %s not registered\n", type->name);
605 if (type == current_trace && tracer_enabled) {
608 if (current_trace->stop)
609 current_trace->stop(&global_trace);
610 current_trace = &nop_trace;
613 if (strlen(type->name) != max_tracer_type_len)
616 max_tracer_type_len = 0;
617 for (t = &trace_types; *t; t = &(*t)->next) {
618 len = strlen((*t)->name);
619 if (len > max_tracer_type_len)
620 max_tracer_type_len = len;
623 mutex_unlock(&trace_types_lock);
626 void tracing_reset(struct trace_array *tr, int cpu)
628 ftrace_disable_cpu();
629 ring_buffer_reset_cpu(tr->buffer, cpu);
633 void tracing_reset_online_cpus(struct trace_array *tr)
637 tr->time_start = ftrace_now(tr->cpu);
639 for_each_online_cpu(cpu)
640 tracing_reset(tr, cpu);
643 #define SAVED_CMDLINES 128
644 #define NO_CMDLINE_MAP UINT_MAX
645 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
646 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
647 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
648 static int cmdline_idx;
649 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
651 /* temporary disable recording */
652 static atomic_t trace_record_cmdline_disabled __read_mostly;
654 static void trace_init_cmdlines(void)
656 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
657 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
661 static int trace_stop_count;
662 static DEFINE_SPINLOCK(tracing_start_lock);
665 * ftrace_off_permanent - disable all ftrace code permanently
667 * This should only be called when a serious anomally has
668 * been detected. This will turn off the function tracing,
669 * ring buffers, and other tracing utilites. It takes no
670 * locks and can be called from any context.
672 void ftrace_off_permanent(void)
674 tracing_disabled = 1;
676 tracing_off_permanent();
680 * tracing_start - quick start of the tracer
682 * If tracing is enabled but was stopped by tracing_stop,
683 * this will start the tracer back up.
685 void tracing_start(void)
687 struct ring_buffer *buffer;
690 if (tracing_disabled)
693 spin_lock_irqsave(&tracing_start_lock, flags);
694 if (--trace_stop_count) {
695 if (trace_stop_count < 0) {
696 /* Someone screwed up their debugging */
698 trace_stop_count = 0;
704 buffer = global_trace.buffer;
706 ring_buffer_record_enable(buffer);
708 buffer = max_tr.buffer;
710 ring_buffer_record_enable(buffer);
714 spin_unlock_irqrestore(&tracing_start_lock, flags);
718 * tracing_stop - quick stop of the tracer
720 * Light weight way to stop tracing. Use in conjunction with
723 void tracing_stop(void)
725 struct ring_buffer *buffer;
729 spin_lock_irqsave(&tracing_start_lock, flags);
730 if (trace_stop_count++)
733 buffer = global_trace.buffer;
735 ring_buffer_record_disable(buffer);
737 buffer = max_tr.buffer;
739 ring_buffer_record_disable(buffer);
742 spin_unlock_irqrestore(&tracing_start_lock, flags);
745 void trace_stop_cmdline_recording(void);
747 static void trace_save_cmdline(struct task_struct *tsk)
751 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
755 * It's not the end of the world if we don't get
756 * the lock, but we also don't want to spin
757 * nor do we want to disable interrupts,
758 * so if we miss here, then better luck next time.
760 if (!__raw_spin_trylock(&trace_cmdline_lock))
763 idx = map_pid_to_cmdline[tsk->pid];
764 if (idx == NO_CMDLINE_MAP) {
765 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
768 * Check whether the cmdline buffer at idx has a pid
769 * mapped. We are going to overwrite that entry so we
770 * need to clear the map_pid_to_cmdline. Otherwise we
771 * would read the new comm for the old pid.
773 pid = map_cmdline_to_pid[idx];
774 if (pid != NO_CMDLINE_MAP)
775 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
777 map_cmdline_to_pid[idx] = tsk->pid;
778 map_pid_to_cmdline[tsk->pid] = idx;
783 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
785 __raw_spin_unlock(&trace_cmdline_lock);
788 void trace_find_cmdline(int pid, char comm[])
793 strcpy(comm, "<idle>");
797 if (pid > PID_MAX_DEFAULT) {
798 strcpy(comm, "<...>");
802 __raw_spin_lock(&trace_cmdline_lock);
803 map = map_pid_to_cmdline[pid];
804 if (map != NO_CMDLINE_MAP)
805 strcpy(comm, saved_cmdlines[map]);
807 strcpy(comm, "<...>");
809 __raw_spin_unlock(&trace_cmdline_lock);
812 void tracing_record_cmdline(struct task_struct *tsk)
814 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
818 trace_save_cmdline(tsk);
822 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
825 struct task_struct *tsk = current;
827 entry->preempt_count = pc & 0xff;
828 entry->pid = (tsk) ? tsk->pid : 0;
829 entry->tgid = (tsk) ? tsk->tgid : 0;
831 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
832 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
834 TRACE_FLAG_IRQS_NOSUPPORT |
836 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
837 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
838 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
841 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
844 unsigned long flags, int pc)
846 struct ring_buffer_event *event;
848 event = ring_buffer_lock_reserve(tr->buffer, len);
850 struct trace_entry *ent = ring_buffer_event_data(event);
852 tracing_generic_entry_update(ent, flags, pc);
858 static void ftrace_trace_stack(struct trace_array *tr,
859 unsigned long flags, int skip, int pc);
860 static void ftrace_trace_userstack(struct trace_array *tr,
861 unsigned long flags, int pc);
863 void trace_buffer_unlock_commit(struct trace_array *tr,
864 struct ring_buffer_event *event,
865 unsigned long flags, int pc)
867 ring_buffer_unlock_commit(tr->buffer, event);
869 ftrace_trace_stack(tr, flags, 6, pc);
870 ftrace_trace_userstack(tr, flags, pc);
874 struct ring_buffer_event *
875 trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
876 unsigned long flags, int pc)
878 return trace_buffer_lock_reserve(&global_trace,
879 type, len, flags, pc);
882 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
883 unsigned long flags, int pc)
885 return trace_buffer_unlock_commit(&global_trace, event, flags, pc);
889 trace_function(struct trace_array *tr,
890 unsigned long ip, unsigned long parent_ip, unsigned long flags,
893 struct ring_buffer_event *event;
894 struct ftrace_entry *entry;
896 /* If we are reading the ring buffer, don't trace */
897 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
900 event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
904 entry = ring_buffer_event_data(event);
906 entry->parent_ip = parent_ip;
907 ring_buffer_unlock_commit(tr->buffer, event);
910 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
911 static void __trace_graph_entry(struct trace_array *tr,
912 struct ftrace_graph_ent *trace,
916 struct ring_buffer_event *event;
917 struct ftrace_graph_ent_entry *entry;
919 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
922 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
923 sizeof(*entry), flags, pc);
926 entry = ring_buffer_event_data(event);
927 entry->graph_ent = *trace;
928 ring_buffer_unlock_commit(global_trace.buffer, event);
931 static void __trace_graph_return(struct trace_array *tr,
932 struct ftrace_graph_ret *trace,
936 struct ring_buffer_event *event;
937 struct ftrace_graph_ret_entry *entry;
939 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
942 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
943 sizeof(*entry), flags, pc);
946 entry = ring_buffer_event_data(event);
948 ring_buffer_unlock_commit(global_trace.buffer, event);
953 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
954 unsigned long ip, unsigned long parent_ip, unsigned long flags,
957 if (likely(!atomic_read(&data->disabled)))
958 trace_function(tr, ip, parent_ip, flags, pc);
961 static void __ftrace_trace_stack(struct trace_array *tr,
965 #ifdef CONFIG_STACKTRACE
966 struct ring_buffer_event *event;
967 struct stack_entry *entry;
968 struct stack_trace trace;
970 event = trace_buffer_lock_reserve(tr, TRACE_STACK,
971 sizeof(*entry), flags, pc);
974 entry = ring_buffer_event_data(event);
975 memset(&entry->caller, 0, sizeof(entry->caller));
977 trace.nr_entries = 0;
978 trace.max_entries = FTRACE_STACK_ENTRIES;
980 trace.entries = entry->caller;
982 save_stack_trace(&trace);
983 ring_buffer_unlock_commit(tr->buffer, event);
987 static void ftrace_trace_stack(struct trace_array *tr,
991 if (!(trace_flags & TRACE_ITER_STACKTRACE))
994 __ftrace_trace_stack(tr, flags, skip, pc);
997 void __trace_stack(struct trace_array *tr,
1001 __ftrace_trace_stack(tr, flags, skip, pc);
1004 static void ftrace_trace_userstack(struct trace_array *tr,
1005 unsigned long flags, int pc)
1007 #ifdef CONFIG_STACKTRACE
1008 struct ring_buffer_event *event;
1009 struct userstack_entry *entry;
1010 struct stack_trace trace;
1012 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1015 event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1016 sizeof(*entry), flags, pc);
1019 entry = ring_buffer_event_data(event);
1021 memset(&entry->caller, 0, sizeof(entry->caller));
1023 trace.nr_entries = 0;
1024 trace.max_entries = FTRACE_STACK_ENTRIES;
1026 trace.entries = entry->caller;
1028 save_stack_trace_user(&trace);
1029 ring_buffer_unlock_commit(tr->buffer, event);
1034 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1036 ftrace_trace_userstack(tr, flags, preempt_count());
1041 ftrace_trace_special(void *__tr,
1042 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1045 struct ring_buffer_event *event;
1046 struct trace_array *tr = __tr;
1047 struct special_entry *entry;
1049 event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1050 sizeof(*entry), 0, pc);
1053 entry = ring_buffer_event_data(event);
1057 trace_buffer_unlock_commit(tr, event, 0, pc);
1061 __trace_special(void *__tr, void *__data,
1062 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1064 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1068 tracing_sched_switch_trace(struct trace_array *tr,
1069 struct task_struct *prev,
1070 struct task_struct *next,
1071 unsigned long flags, int pc)
1073 struct ring_buffer_event *event;
1074 struct ctx_switch_entry *entry;
1076 event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1077 sizeof(*entry), flags, pc);
1080 entry = ring_buffer_event_data(event);
1081 entry->prev_pid = prev->pid;
1082 entry->prev_prio = prev->prio;
1083 entry->prev_state = prev->state;
1084 entry->next_pid = next->pid;
1085 entry->next_prio = next->prio;
1086 entry->next_state = next->state;
1087 entry->next_cpu = task_cpu(next);
1088 trace_buffer_unlock_commit(tr, event, flags, pc);
1092 tracing_sched_wakeup_trace(struct trace_array *tr,
1093 struct task_struct *wakee,
1094 struct task_struct *curr,
1095 unsigned long flags, int pc)
1097 struct ring_buffer_event *event;
1098 struct ctx_switch_entry *entry;
1100 event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1101 sizeof(*entry), flags, pc);
1104 entry = ring_buffer_event_data(event);
1105 entry->prev_pid = curr->pid;
1106 entry->prev_prio = curr->prio;
1107 entry->prev_state = curr->state;
1108 entry->next_pid = wakee->pid;
1109 entry->next_prio = wakee->prio;
1110 entry->next_state = wakee->state;
1111 entry->next_cpu = task_cpu(wakee);
1113 ring_buffer_unlock_commit(tr->buffer, event);
1114 ftrace_trace_stack(tr, flags, 6, pc);
1115 ftrace_trace_userstack(tr, flags, pc);
1119 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1121 struct trace_array *tr = &global_trace;
1122 struct trace_array_cpu *data;
1123 unsigned long flags;
1127 if (tracing_disabled)
1130 pc = preempt_count();
1131 local_irq_save(flags);
1132 cpu = raw_smp_processor_id();
1133 data = tr->data[cpu];
1135 if (likely(atomic_inc_return(&data->disabled) == 1))
1136 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1138 atomic_dec(&data->disabled);
1139 local_irq_restore(flags);
1142 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1143 int trace_graph_entry(struct ftrace_graph_ent *trace)
1145 struct trace_array *tr = &global_trace;
1146 struct trace_array_cpu *data;
1147 unsigned long flags;
1152 if (!ftrace_trace_task(current))
1155 if (!ftrace_graph_addr(trace->func))
1158 local_irq_save(flags);
1159 cpu = raw_smp_processor_id();
1160 data = tr->data[cpu];
1161 disabled = atomic_inc_return(&data->disabled);
1162 if (likely(disabled == 1)) {
1163 pc = preempt_count();
1164 __trace_graph_entry(tr, trace, flags, pc);
1166 /* Only do the atomic if it is not already set */
1167 if (!test_tsk_trace_graph(current))
1168 set_tsk_trace_graph(current);
1169 atomic_dec(&data->disabled);
1170 local_irq_restore(flags);
1175 void trace_graph_return(struct ftrace_graph_ret *trace)
1177 struct trace_array *tr = &global_trace;
1178 struct trace_array_cpu *data;
1179 unsigned long flags;
1184 local_irq_save(flags);
1185 cpu = raw_smp_processor_id();
1186 data = tr->data[cpu];
1187 disabled = atomic_inc_return(&data->disabled);
1188 if (likely(disabled == 1)) {
1189 pc = preempt_count();
1190 __trace_graph_return(tr, trace, flags, pc);
1193 clear_tsk_trace_graph(current);
1194 atomic_dec(&data->disabled);
1195 local_irq_restore(flags);
1197 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1201 * trace_vbprintk - write binary msg to tracing buffer
1204 int trace_vbprintk(unsigned long ip, int depth, const char *fmt, va_list args)
1206 static raw_spinlock_t trace_buf_lock =
1207 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1208 static u32 trace_buf[TRACE_BUF_SIZE];
1210 struct ring_buffer_event *event;
1211 struct trace_array *tr = &global_trace;
1212 struct trace_array_cpu *data;
1213 struct bprint_entry *entry;
1214 unsigned long flags;
1216 int cpu, len = 0, size, pc;
1218 if (unlikely(tracing_selftest_running || tracing_disabled))
1221 /* Don't pollute graph traces with trace_vprintk internals */
1222 pause_graph_tracing();
1224 pc = preempt_count();
1225 resched = ftrace_preempt_disable();
1226 cpu = raw_smp_processor_id();
1227 data = tr->data[cpu];
1229 if (unlikely(atomic_read(&data->disabled)))
1232 /* Lockdep uses trace_printk for lock tracing */
1233 local_irq_save(flags);
1234 __raw_spin_lock(&trace_buf_lock);
1235 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1237 if (len > TRACE_BUF_SIZE || len < 0)
1240 size = sizeof(*entry) + sizeof(u32) * len;
1241 event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1244 entry = ring_buffer_event_data(event);
1246 entry->depth = depth;
1249 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1250 ring_buffer_unlock_commit(tr->buffer, event);
1253 __raw_spin_unlock(&trace_buf_lock);
1254 local_irq_restore(flags);
1257 ftrace_preempt_enable(resched);
1258 unpause_graph_tracing();
1262 EXPORT_SYMBOL_GPL(trace_vbprintk);
1264 int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
1266 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1267 static char trace_buf[TRACE_BUF_SIZE];
1269 struct ring_buffer_event *event;
1270 struct trace_array *tr = &global_trace;
1271 struct trace_array_cpu *data;
1272 int cpu, len = 0, size, pc;
1273 struct print_entry *entry;
1274 unsigned long irq_flags;
1276 if (tracing_disabled || tracing_selftest_running)
1279 pc = preempt_count();
1280 preempt_disable_notrace();
1281 cpu = raw_smp_processor_id();
1282 data = tr->data[cpu];
1284 if (unlikely(atomic_read(&data->disabled)))
1287 pause_graph_tracing();
1288 raw_local_irq_save(irq_flags);
1289 __raw_spin_lock(&trace_buf_lock);
1290 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1292 len = min(len, TRACE_BUF_SIZE-1);
1295 size = sizeof(*entry) + len + 1;
1296 event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1299 entry = ring_buffer_event_data(event);
1301 entry->depth = depth;
1303 memcpy(&entry->buf, trace_buf, len);
1304 entry->buf[len] = 0;
1305 ring_buffer_unlock_commit(tr->buffer, event);
1308 __raw_spin_unlock(&trace_buf_lock);
1309 raw_local_irq_restore(irq_flags);
1310 unpause_graph_tracing();
1312 preempt_enable_notrace();
1316 EXPORT_SYMBOL_GPL(trace_vprintk);
1318 enum trace_file_type {
1319 TRACE_FILE_LAT_FMT = 1,
1320 TRACE_FILE_ANNOTATE = 2,
1323 static void trace_iterator_increment(struct trace_iterator *iter)
1325 /* Don't allow ftrace to trace into the ring buffers */
1326 ftrace_disable_cpu();
1329 if (iter->buffer_iter[iter->cpu])
1330 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1332 ftrace_enable_cpu();
1335 static struct trace_entry *
1336 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1338 struct ring_buffer_event *event;
1339 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1341 /* Don't allow ftrace to trace into the ring buffers */
1342 ftrace_disable_cpu();
1345 event = ring_buffer_iter_peek(buf_iter, ts);
1347 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1349 ftrace_enable_cpu();
1351 return event ? ring_buffer_event_data(event) : NULL;
1354 static struct trace_entry *
1355 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1357 struct ring_buffer *buffer = iter->tr->buffer;
1358 struct trace_entry *ent, *next = NULL;
1359 int cpu_file = iter->cpu_file;
1360 u64 next_ts = 0, ts;
1365 * If we are in a per_cpu trace file, don't bother by iterating over
1366 * all cpu and peek directly.
1368 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1369 if (ring_buffer_empty_cpu(buffer, cpu_file))
1371 ent = peek_next_entry(iter, cpu_file, ent_ts);
1373 *ent_cpu = cpu_file;
1378 for_each_tracing_cpu(cpu) {
1380 if (ring_buffer_empty_cpu(buffer, cpu))
1383 ent = peek_next_entry(iter, cpu, &ts);
1386 * Pick the entry with the smallest timestamp:
1388 if (ent && (!next || ts < next_ts)) {
1396 *ent_cpu = next_cpu;
1404 /* Find the next real entry, without updating the iterator itself */
1405 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1406 int *ent_cpu, u64 *ent_ts)
1408 return __find_next_entry(iter, ent_cpu, ent_ts);
1411 /* Find the next real entry, and increment the iterator to the next entry */
1412 static void *find_next_entry_inc(struct trace_iterator *iter)
1414 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1417 trace_iterator_increment(iter);
1419 return iter->ent ? iter : NULL;
1422 static void trace_consume(struct trace_iterator *iter)
1424 /* Don't allow ftrace to trace into the ring buffers */
1425 ftrace_disable_cpu();
1426 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1427 ftrace_enable_cpu();
1430 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1432 struct trace_iterator *iter = m->private;
1438 /* can't go backwards */
1443 ent = find_next_entry_inc(iter);
1447 while (ent && iter->idx < i)
1448 ent = find_next_entry_inc(iter);
1456 * No necessary locking here. The worst thing which can
1457 * happen is loosing events consumed at the same time
1458 * by a trace_pipe reader.
1459 * Other than that, we don't risk to crash the ring buffer
1460 * because it serializes the readers.
1462 * The current tracer is copied to avoid a global locking
1465 static void *s_start(struct seq_file *m, loff_t *pos)
1467 struct trace_iterator *iter = m->private;
1468 static struct tracer *old_tracer;
1469 int cpu_file = iter->cpu_file;
1474 /* copy the tracer to avoid using a global lock all around */
1475 mutex_lock(&trace_types_lock);
1476 if (unlikely(old_tracer != current_trace && current_trace)) {
1477 old_tracer = current_trace;
1478 *iter->trace = *current_trace;
1480 mutex_unlock(&trace_types_lock);
1482 atomic_inc(&trace_record_cmdline_disabled);
1484 if (*pos != iter->pos) {
1489 ftrace_disable_cpu();
1491 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1492 for_each_tracing_cpu(cpu)
1493 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1495 ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1498 ftrace_enable_cpu();
1500 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1505 p = s_next(m, p, &l);
1511 static void s_stop(struct seq_file *m, void *p)
1513 atomic_dec(&trace_record_cmdline_disabled);
1516 static void print_lat_help_header(struct seq_file *m)
1518 seq_puts(m, "# _------=> CPU# \n");
1519 seq_puts(m, "# / _-----=> irqs-off \n");
1520 seq_puts(m, "# | / _----=> need-resched \n");
1521 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1522 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1523 seq_puts(m, "# |||| / \n");
1524 seq_puts(m, "# ||||| delay \n");
1525 seq_puts(m, "# cmd pid ||||| time | caller \n");
1526 seq_puts(m, "# \\ / ||||| \\ | / \n");
1529 static void print_func_help_header(struct seq_file *m)
1531 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1532 seq_puts(m, "# | | | | |\n");
1537 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1539 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1540 struct trace_array *tr = iter->tr;
1541 struct trace_array_cpu *data = tr->data[tr->cpu];
1542 struct tracer *type = current_trace;
1543 unsigned long total;
1544 unsigned long entries;
1545 const char *name = "preemption";
1550 entries = ring_buffer_entries(iter->tr->buffer);
1552 ring_buffer_overruns(iter->tr->buffer);
1554 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1556 seq_puts(m, "# -----------------------------------"
1557 "---------------------------------\n");
1558 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1559 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1560 nsecs_to_usecs(data->saved_latency),
1564 #if defined(CONFIG_PREEMPT_NONE)
1566 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1568 #elif defined(CONFIG_PREEMPT)
1573 /* These are reserved for later use */
1576 seq_printf(m, " #P:%d)\n", num_online_cpus());
1580 seq_puts(m, "# -----------------\n");
1581 seq_printf(m, "# | task: %.16s-%d "
1582 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1583 data->comm, data->pid, data->uid, data->nice,
1584 data->policy, data->rt_priority);
1585 seq_puts(m, "# -----------------\n");
1587 if (data->critical_start) {
1588 seq_puts(m, "# => started at: ");
1589 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1590 trace_print_seq(m, &iter->seq);
1591 seq_puts(m, "\n# => ended at: ");
1592 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1593 trace_print_seq(m, &iter->seq);
1600 static void test_cpu_buff_start(struct trace_iterator *iter)
1602 struct trace_seq *s = &iter->seq;
1604 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1607 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1610 if (cpumask_test_cpu(iter->cpu, iter->started))
1613 cpumask_set_cpu(iter->cpu, iter->started);
1614 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1617 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1619 struct trace_seq *s = &iter->seq;
1620 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1621 struct trace_entry *entry;
1622 struct trace_event *event;
1626 test_cpu_buff_start(iter);
1628 event = ftrace_find_event(entry->type);
1630 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1631 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1632 if (!trace_print_lat_context(iter))
1635 if (!trace_print_context(iter))
1641 return event->trace(iter, sym_flags);
1643 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1646 return TRACE_TYPE_HANDLED;
1648 return TRACE_TYPE_PARTIAL_LINE;
1651 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1653 struct trace_seq *s = &iter->seq;
1654 struct trace_entry *entry;
1655 struct trace_event *event;
1659 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1660 if (!trace_seq_printf(s, "%d %d %llu ",
1661 entry->pid, iter->cpu, iter->ts))
1665 event = ftrace_find_event(entry->type);
1667 return event->raw(iter, 0);
1669 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1672 return TRACE_TYPE_HANDLED;
1674 return TRACE_TYPE_PARTIAL_LINE;
1677 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1679 struct trace_seq *s = &iter->seq;
1680 unsigned char newline = '\n';
1681 struct trace_entry *entry;
1682 struct trace_event *event;
1686 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1687 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1688 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1689 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1692 event = ftrace_find_event(entry->type);
1694 enum print_line_t ret = event->hex(iter, 0);
1695 if (ret != TRACE_TYPE_HANDLED)
1699 SEQ_PUT_FIELD_RET(s, newline);
1701 return TRACE_TYPE_HANDLED;
1704 static enum print_line_t print_bprintk_msg_only(struct trace_iterator *iter)
1706 struct trace_seq *s = &iter->seq;
1707 struct trace_entry *entry = iter->ent;
1708 struct bprint_entry *field;
1711 trace_assign_type(field, entry);
1713 ret = trace_seq_bprintf(s, field->fmt, field->buf);
1715 return TRACE_TYPE_PARTIAL_LINE;
1717 return TRACE_TYPE_HANDLED;
1720 static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
1722 struct trace_seq *s = &iter->seq;
1723 struct trace_entry *entry = iter->ent;
1724 struct print_entry *field;
1727 trace_assign_type(field, entry);
1729 ret = trace_seq_printf(s, "%s", field->buf);
1731 return TRACE_TYPE_PARTIAL_LINE;
1733 return TRACE_TYPE_HANDLED;
1736 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1738 struct trace_seq *s = &iter->seq;
1739 struct trace_entry *entry;
1740 struct trace_event *event;
1744 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1745 SEQ_PUT_FIELD_RET(s, entry->pid);
1746 SEQ_PUT_FIELD_RET(s, iter->cpu);
1747 SEQ_PUT_FIELD_RET(s, iter->ts);
1750 event = ftrace_find_event(entry->type);
1751 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1754 static int trace_empty(struct trace_iterator *iter)
1758 /* If we are looking at one CPU buffer, only check that one */
1759 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1760 cpu = iter->cpu_file;
1761 if (iter->buffer_iter[cpu]) {
1762 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1765 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1771 for_each_tracing_cpu(cpu) {
1772 if (iter->buffer_iter[cpu]) {
1773 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1776 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1784 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1786 enum print_line_t ret;
1788 if (iter->trace && iter->trace->print_line) {
1789 ret = iter->trace->print_line(iter);
1790 if (ret != TRACE_TYPE_UNHANDLED)
1794 if (iter->ent->type == TRACE_BPRINT &&
1795 trace_flags & TRACE_ITER_PRINTK &&
1796 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1797 return print_bprintk_msg_only(iter);
1799 if (iter->ent->type == TRACE_PRINT &&
1800 trace_flags & TRACE_ITER_PRINTK &&
1801 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1802 return print_printk_msg_only(iter);
1804 if (trace_flags & TRACE_ITER_BIN)
1805 return print_bin_fmt(iter);
1807 if (trace_flags & TRACE_ITER_HEX)
1808 return print_hex_fmt(iter);
1810 if (trace_flags & TRACE_ITER_RAW)
1811 return print_raw_fmt(iter);
1813 return print_trace_fmt(iter);
1816 static int s_show(struct seq_file *m, void *v)
1818 struct trace_iterator *iter = v;
1820 if (iter->ent == NULL) {
1822 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1825 if (iter->trace && iter->trace->print_header)
1826 iter->trace->print_header(m);
1827 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1828 /* print nothing if the buffers are empty */
1829 if (trace_empty(iter))
1831 print_trace_header(m, iter);
1832 if (!(trace_flags & TRACE_ITER_VERBOSE))
1833 print_lat_help_header(m);
1835 if (!(trace_flags & TRACE_ITER_VERBOSE))
1836 print_func_help_header(m);
1839 print_trace_line(iter);
1840 trace_print_seq(m, &iter->seq);
1846 static struct seq_operations tracer_seq_ops = {
1853 static struct trace_iterator *
1854 __tracing_open(struct inode *inode, struct file *file)
1856 long cpu_file = (long) inode->i_private;
1857 void *fail_ret = ERR_PTR(-ENOMEM);
1858 struct trace_iterator *iter;
1862 if (tracing_disabled)
1863 return ERR_PTR(-ENODEV);
1865 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1867 return ERR_PTR(-ENOMEM);
1870 * We make a copy of the current tracer to avoid concurrent
1871 * changes on it while we are reading.
1873 mutex_lock(&trace_types_lock);
1874 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1879 *iter->trace = *current_trace;
1881 if (current_trace && current_trace->print_max)
1884 iter->tr = &global_trace;
1886 mutex_init(&iter->mutex);
1887 iter->cpu_file = cpu_file;
1889 /* Notify the tracer early; before we stop tracing. */
1890 if (iter->trace && iter->trace->open)
1891 iter->trace->open(iter);
1893 /* Annotate start of buffers if we had overruns */
1894 if (ring_buffer_overruns(iter->tr->buffer))
1895 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1897 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1898 for_each_tracing_cpu(cpu) {
1900 iter->buffer_iter[cpu] =
1901 ring_buffer_read_start(iter->tr->buffer, cpu);
1904 cpu = iter->cpu_file;
1905 iter->buffer_iter[cpu] =
1906 ring_buffer_read_start(iter->tr->buffer, cpu);
1909 /* TODO stop tracer */
1910 ret = seq_open(file, &tracer_seq_ops);
1912 fail_ret = ERR_PTR(ret);
1916 m = file->private_data;
1919 /* stop the trace while dumping */
1922 mutex_unlock(&trace_types_lock);
1927 for_each_tracing_cpu(cpu) {
1928 if (iter->buffer_iter[cpu])
1929 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1932 mutex_unlock(&trace_types_lock);
1939 int tracing_open_generic(struct inode *inode, struct file *filp)
1941 if (tracing_disabled)
1944 filp->private_data = inode->i_private;
1948 static int tracing_release(struct inode *inode, struct file *file)
1950 struct seq_file *m = (struct seq_file *)file->private_data;
1951 struct trace_iterator *iter = m->private;
1954 mutex_lock(&trace_types_lock);
1955 for_each_tracing_cpu(cpu) {
1956 if (iter->buffer_iter[cpu])
1957 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1960 if (iter->trace && iter->trace->close)
1961 iter->trace->close(iter);
1963 /* reenable tracing if it was previously enabled */
1965 mutex_unlock(&trace_types_lock);
1967 seq_release(inode, file);
1968 mutex_destroy(&iter->mutex);
1974 static int tracing_open(struct inode *inode, struct file *file)
1976 struct trace_iterator *iter;
1979 iter = __tracing_open(inode, file);
1981 ret = PTR_ERR(iter);
1982 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
1983 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1989 t_next(struct seq_file *m, void *v, loff_t *pos)
1991 struct tracer *t = m->private;
2003 static void *t_start(struct seq_file *m, loff_t *pos)
2005 struct tracer *t = m->private;
2008 mutex_lock(&trace_types_lock);
2009 for (; t && l < *pos; t = t_next(m, t, &l))
2015 static void t_stop(struct seq_file *m, void *p)
2017 mutex_unlock(&trace_types_lock);
2020 static int t_show(struct seq_file *m, void *v)
2022 struct tracer *t = v;
2027 seq_printf(m, "%s", t->name);
2036 static struct seq_operations show_traces_seq_ops = {
2043 static int show_traces_open(struct inode *inode, struct file *file)
2047 if (tracing_disabled)
2050 ret = seq_open(file, &show_traces_seq_ops);
2052 struct seq_file *m = file->private_data;
2053 m->private = trace_types;
2059 static const struct file_operations tracing_fops = {
2060 .open = tracing_open,
2062 .llseek = seq_lseek,
2063 .release = tracing_release,
2066 static const struct file_operations show_traces_fops = {
2067 .open = show_traces_open,
2069 .release = seq_release,
2073 * Only trace on a CPU if the bitmask is set:
2075 static cpumask_var_t tracing_cpumask;
2078 * The tracer itself will not take this lock, but still we want
2079 * to provide a consistent cpumask to user-space:
2081 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2084 * Temporary storage for the character representation of the
2085 * CPU bitmask (and one more byte for the newline):
2087 static char mask_str[NR_CPUS + 1];
2090 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2091 size_t count, loff_t *ppos)
2095 mutex_lock(&tracing_cpumask_update_lock);
2097 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2098 if (count - len < 2) {
2102 len += sprintf(mask_str + len, "\n");
2103 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2106 mutex_unlock(&tracing_cpumask_update_lock);
2112 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2113 size_t count, loff_t *ppos)
2116 cpumask_var_t tracing_cpumask_new;
2118 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2121 mutex_lock(&tracing_cpumask_update_lock);
2122 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2126 local_irq_disable();
2127 __raw_spin_lock(&ftrace_max_lock);
2128 for_each_tracing_cpu(cpu) {
2130 * Increase/decrease the disabled counter if we are
2131 * about to flip a bit in the cpumask:
2133 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2134 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2135 atomic_inc(&global_trace.data[cpu]->disabled);
2137 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2138 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2139 atomic_dec(&global_trace.data[cpu]->disabled);
2142 __raw_spin_unlock(&ftrace_max_lock);
2145 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2147 mutex_unlock(&tracing_cpumask_update_lock);
2148 free_cpumask_var(tracing_cpumask_new);
2153 mutex_unlock(&tracing_cpumask_update_lock);
2154 free_cpumask_var(tracing_cpumask);
2159 static const struct file_operations tracing_cpumask_fops = {
2160 .open = tracing_open_generic,
2161 .read = tracing_cpumask_read,
2162 .write = tracing_cpumask_write,
2166 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2167 size_t cnt, loff_t *ppos)
2169 struct tracer_opt *trace_opts;
2177 /* calculate max size */
2178 for (i = 0; trace_options[i]; i++) {
2179 len += strlen(trace_options[i]);
2180 len += 3; /* "no" and newline */
2183 mutex_lock(&trace_types_lock);
2184 tracer_flags = current_trace->flags->val;
2185 trace_opts = current_trace->flags->opts;
2188 * Increase the size with names of options specific
2189 * of the current tracer.
2191 for (i = 0; trace_opts[i].name; i++) {
2192 len += strlen(trace_opts[i].name);
2193 len += 3; /* "no" and newline */
2196 /* +2 for \n and \0 */
2197 buf = kmalloc(len + 2, GFP_KERNEL);
2199 mutex_unlock(&trace_types_lock);
2203 for (i = 0; trace_options[i]; i++) {
2204 if (trace_flags & (1 << i))
2205 r += sprintf(buf + r, "%s\n", trace_options[i]);
2207 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2210 for (i = 0; trace_opts[i].name; i++) {
2211 if (tracer_flags & trace_opts[i].bit)
2212 r += sprintf(buf + r, "%s\n",
2213 trace_opts[i].name);
2215 r += sprintf(buf + r, "no%s\n",
2216 trace_opts[i].name);
2218 mutex_unlock(&trace_types_lock);
2220 WARN_ON(r >= len + 2);
2222 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2228 /* Try to assign a tracer specific option */
2229 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2231 struct tracer_flags *trace_flags = trace->flags;
2232 struct tracer_opt *opts = NULL;
2236 for (i = 0; trace_flags->opts[i].name; i++) {
2237 opts = &trace_flags->opts[i];
2238 len = strlen(opts->name);
2240 if (strncmp(cmp, opts->name, len) == 0) {
2241 ret = trace->set_flag(trace_flags->val,
2247 if (!trace_flags->opts[i].name)
2250 /* Refused to handle */
2255 trace_flags->val &= ~opts->bit;
2257 trace_flags->val |= opts->bit;
2262 static void set_tracer_flags(unsigned int mask, int enabled)
2264 /* do nothing if flag is already set */
2265 if (!!(trace_flags & mask) == !!enabled)
2269 trace_flags |= mask;
2271 trace_flags &= ~mask;
2273 if (mask == TRACE_ITER_GLOBAL_CLK) {
2277 func = trace_clock_global;
2279 func = trace_clock_local;
2281 mutex_lock(&trace_types_lock);
2282 ring_buffer_set_clock(global_trace.buffer, func);
2285 ring_buffer_set_clock(max_tr.buffer, func);
2286 mutex_unlock(&trace_types_lock);
2291 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2292 size_t cnt, loff_t *ppos)
2300 if (cnt >= sizeof(buf))
2303 if (copy_from_user(&buf, ubuf, cnt))
2308 if (strncmp(buf, "no", 2) == 0) {
2313 for (i = 0; trace_options[i]; i++) {
2314 int len = strlen(trace_options[i]);
2316 if (strncmp(cmp, trace_options[i], len) == 0) {
2317 set_tracer_flags(1 << i, !neg);
2322 /* If no option could be set, test the specific tracer options */
2323 if (!trace_options[i]) {
2324 mutex_lock(&trace_types_lock);
2325 ret = set_tracer_option(current_trace, cmp, neg);
2326 mutex_unlock(&trace_types_lock);
2336 static const struct file_operations tracing_iter_fops = {
2337 .open = tracing_open_generic,
2338 .read = tracing_trace_options_read,
2339 .write = tracing_trace_options_write,
2342 static const char readme_msg[] =
2343 "tracing mini-HOWTO:\n\n"
2345 "# mount -t debugfs nodev /debug\n\n"
2346 "# cat /debug/tracing/available_tracers\n"
2347 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2348 "# cat /debug/tracing/current_tracer\n"
2350 "# echo sched_switch > /debug/tracing/current_tracer\n"
2351 "# cat /debug/tracing/current_tracer\n"
2353 "# cat /debug/tracing/trace_options\n"
2354 "noprint-parent nosym-offset nosym-addr noverbose\n"
2355 "# echo print-parent > /debug/tracing/trace_options\n"
2356 "# echo 1 > /debug/tracing/tracing_enabled\n"
2357 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2358 "echo 0 > /debug/tracing/tracing_enabled\n"
2362 tracing_readme_read(struct file *filp, char __user *ubuf,
2363 size_t cnt, loff_t *ppos)
2365 return simple_read_from_buffer(ubuf, cnt, ppos,
2366 readme_msg, strlen(readme_msg));
2369 static const struct file_operations tracing_readme_fops = {
2370 .open = tracing_open_generic,
2371 .read = tracing_readme_read,
2375 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2376 size_t cnt, loff_t *ppos)
2381 r = sprintf(buf, "%u\n", tracer_enabled);
2382 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2386 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2387 size_t cnt, loff_t *ppos)
2389 struct trace_array *tr = filp->private_data;
2394 if (cnt >= sizeof(buf))
2397 if (copy_from_user(&buf, ubuf, cnt))
2402 ret = strict_strtoul(buf, 10, &val);
2408 mutex_lock(&trace_types_lock);
2409 if (tracer_enabled ^ val) {
2412 if (current_trace->start)
2413 current_trace->start(tr);
2418 if (current_trace->stop)
2419 current_trace->stop(tr);
2422 mutex_unlock(&trace_types_lock);
2430 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2431 size_t cnt, loff_t *ppos)
2433 char buf[max_tracer_type_len+2];
2436 mutex_lock(&trace_types_lock);
2438 r = sprintf(buf, "%s\n", current_trace->name);
2440 r = sprintf(buf, "\n");
2441 mutex_unlock(&trace_types_lock);
2443 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2446 int tracer_init(struct tracer *t, struct trace_array *tr)
2448 tracing_reset_online_cpus(tr);
2452 static int tracing_resize_ring_buffer(unsigned long size)
2457 * If kernel or user changes the size of the ring buffer
2458 * we use the size that was given, and we can forget about
2459 * expanding it later.
2461 ring_buffer_expanded = 1;
2463 ret = ring_buffer_resize(global_trace.buffer, size);
2467 ret = ring_buffer_resize(max_tr.buffer, size);
2471 r = ring_buffer_resize(global_trace.buffer,
2472 global_trace.entries);
2475 * AARGH! We are left with different
2476 * size max buffer!!!!
2477 * The max buffer is our "snapshot" buffer.
2478 * When a tracer needs a snapshot (one of the
2479 * latency tracers), it swaps the max buffer
2480 * with the saved snap shot. We succeeded to
2481 * update the size of the main buffer, but failed to
2482 * update the size of the max buffer. But when we tried
2483 * to reset the main buffer to the original size, we
2484 * failed there too. This is very unlikely to
2485 * happen, but if it does, warn and kill all
2489 tracing_disabled = 1;
2494 global_trace.entries = size;
2500 * tracing_update_buffers - used by tracing facility to expand ring buffers
2502 * To save on memory when the tracing is never used on a system with it
2503 * configured in. The ring buffers are set to a minimum size. But once
2504 * a user starts to use the tracing facility, then they need to grow
2505 * to their default size.
2507 * This function is to be called when a tracer is about to be used.
2509 int tracing_update_buffers(void)
2513 mutex_lock(&trace_types_lock);
2514 if (!ring_buffer_expanded)
2515 ret = tracing_resize_ring_buffer(trace_buf_size);
2516 mutex_unlock(&trace_types_lock);
2521 struct trace_option_dentry;
2523 static struct trace_option_dentry *
2524 create_trace_option_files(struct tracer *tracer);
2527 destroy_trace_option_files(struct trace_option_dentry *topts);
2529 static int tracing_set_tracer(const char *buf)
2531 static struct trace_option_dentry *topts;
2532 struct trace_array *tr = &global_trace;
2536 mutex_lock(&trace_types_lock);
2538 if (!ring_buffer_expanded) {
2539 ret = tracing_resize_ring_buffer(trace_buf_size);
2545 for (t = trace_types; t; t = t->next) {
2546 if (strcmp(t->name, buf) == 0)
2553 if (t == current_trace)
2556 trace_branch_disable();
2557 if (current_trace && current_trace->reset)
2558 current_trace->reset(tr);
2560 destroy_trace_option_files(topts);
2564 topts = create_trace_option_files(current_trace);
2567 ret = tracer_init(t, tr);
2572 trace_branch_enable(tr);
2574 mutex_unlock(&trace_types_lock);
2580 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2581 size_t cnt, loff_t *ppos)
2583 char buf[max_tracer_type_len+1];
2590 if (cnt > max_tracer_type_len)
2591 cnt = max_tracer_type_len;
2593 if (copy_from_user(&buf, ubuf, cnt))
2598 /* strip ending whitespace. */
2599 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2602 err = tracing_set_tracer(buf);
2612 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2613 size_t cnt, loff_t *ppos)
2615 unsigned long *ptr = filp->private_data;
2619 r = snprintf(buf, sizeof(buf), "%ld\n",
2620 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2621 if (r > sizeof(buf))
2623 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2627 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2628 size_t cnt, loff_t *ppos)
2630 unsigned long *ptr = filp->private_data;
2635 if (cnt >= sizeof(buf))
2638 if (copy_from_user(&buf, ubuf, cnt))
2643 ret = strict_strtoul(buf, 10, &val);
2652 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2654 long cpu_file = (long) inode->i_private;
2655 struct trace_iterator *iter;
2658 if (tracing_disabled)
2661 mutex_lock(&trace_types_lock);
2663 /* We only allow one reader per cpu */
2664 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2665 if (!cpumask_empty(tracing_reader_cpumask)) {
2669 cpumask_setall(tracing_reader_cpumask);
2671 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2672 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2679 /* create a buffer to store the information to pass to userspace */
2680 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2687 * We make a copy of the current tracer to avoid concurrent
2688 * changes on it while we are reading.
2690 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2696 *iter->trace = *current_trace;
2698 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2703 /* trace pipe does not show start of buffer */
2704 cpumask_setall(iter->started);
2706 iter->cpu_file = cpu_file;
2707 iter->tr = &global_trace;
2708 mutex_init(&iter->mutex);
2709 filp->private_data = iter;
2711 if (iter->trace->pipe_open)
2712 iter->trace->pipe_open(iter);
2715 mutex_unlock(&trace_types_lock);
2721 mutex_unlock(&trace_types_lock);
2725 static int tracing_release_pipe(struct inode *inode, struct file *file)
2727 struct trace_iterator *iter = file->private_data;
2729 mutex_lock(&trace_types_lock);
2731 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2732 cpumask_clear(tracing_reader_cpumask);
2734 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2736 mutex_unlock(&trace_types_lock);
2738 free_cpumask_var(iter->started);
2739 mutex_destroy(&iter->mutex);
2747 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2749 struct trace_iterator *iter = filp->private_data;
2751 if (trace_flags & TRACE_ITER_BLOCK) {
2753 * Always select as readable when in blocking mode
2755 return POLLIN | POLLRDNORM;
2757 if (!trace_empty(iter))
2758 return POLLIN | POLLRDNORM;
2759 poll_wait(filp, &trace_wait, poll_table);
2760 if (!trace_empty(iter))
2761 return POLLIN | POLLRDNORM;
2768 void default_wait_pipe(struct trace_iterator *iter)
2772 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2774 if (trace_empty(iter))
2777 finish_wait(&trace_wait, &wait);
2781 * This is a make-shift waitqueue.
2782 * A tracer might use this callback on some rare cases:
2784 * 1) the current tracer might hold the runqueue lock when it wakes up
2785 * a reader, hence a deadlock (sched, function, and function graph tracers)
2786 * 2) the function tracers, trace all functions, we don't want
2787 * the overhead of calling wake_up and friends
2788 * (and tracing them too)
2790 * Anyway, this is really very primitive wakeup.
2792 void poll_wait_pipe(struct trace_iterator *iter)
2794 set_current_state(TASK_INTERRUPTIBLE);
2795 /* sleep for 100 msecs, and try again. */
2796 schedule_timeout(HZ / 10);
2799 /* Must be called with trace_types_lock mutex held. */
2800 static int tracing_wait_pipe(struct file *filp)
2802 struct trace_iterator *iter = filp->private_data;
2804 while (trace_empty(iter)) {
2806 if ((filp->f_flags & O_NONBLOCK)) {
2810 mutex_unlock(&iter->mutex);
2812 iter->trace->wait_pipe(iter);
2814 mutex_lock(&iter->mutex);
2816 if (signal_pending(current))
2820 * We block until we read something and tracing is disabled.
2821 * We still block if tracing is disabled, but we have never
2822 * read anything. This allows a user to cat this file, and
2823 * then enable tracing. But after we have read something,
2824 * we give an EOF when tracing is again disabled.
2826 * iter->pos will be 0 if we haven't read anything.
2828 if (!tracer_enabled && iter->pos)
2839 tracing_read_pipe(struct file *filp, char __user *ubuf,
2840 size_t cnt, loff_t *ppos)
2842 struct trace_iterator *iter = filp->private_data;
2843 static struct tracer *old_tracer;
2846 /* return any leftover data */
2847 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2851 trace_seq_init(&iter->seq);
2853 /* copy the tracer to avoid using a global lock all around */
2854 mutex_lock(&trace_types_lock);
2855 if (unlikely(old_tracer != current_trace && current_trace)) {
2856 old_tracer = current_trace;
2857 *iter->trace = *current_trace;
2859 mutex_unlock(&trace_types_lock);
2862 * Avoid more than one consumer on a single file descriptor
2863 * This is just a matter of traces coherency, the ring buffer itself
2866 mutex_lock(&iter->mutex);
2867 if (iter->trace->read) {
2868 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2874 sret = tracing_wait_pipe(filp);
2878 /* stop when tracing is finished */
2879 if (trace_empty(iter)) {
2884 if (cnt >= PAGE_SIZE)
2885 cnt = PAGE_SIZE - 1;
2887 /* reset all but tr, trace, and overruns */
2888 memset(&iter->seq, 0,
2889 sizeof(struct trace_iterator) -
2890 offsetof(struct trace_iterator, seq));
2893 while (find_next_entry_inc(iter) != NULL) {
2894 enum print_line_t ret;
2895 int len = iter->seq.len;
2897 ret = print_trace_line(iter);
2898 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2899 /* don't print partial lines */
2900 iter->seq.len = len;
2903 if (ret != TRACE_TYPE_NO_CONSUME)
2904 trace_consume(iter);
2906 if (iter->seq.len >= cnt)
2910 /* Now copy what we have to the user */
2911 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2912 if (iter->seq.readpos >= iter->seq.len)
2913 trace_seq_init(&iter->seq);
2916 * If there was nothing to send to user, inspite of consuming trace
2917 * entries, go back to wait for more entries.
2923 mutex_unlock(&iter->mutex);
2928 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2929 struct pipe_buffer *buf)
2931 __free_page(buf->page);
2934 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2937 __free_page(spd->pages[idx]);
2940 static struct pipe_buf_operations tracing_pipe_buf_ops = {
2942 .map = generic_pipe_buf_map,
2943 .unmap = generic_pipe_buf_unmap,
2944 .confirm = generic_pipe_buf_confirm,
2945 .release = tracing_pipe_buf_release,
2946 .steal = generic_pipe_buf_steal,
2947 .get = generic_pipe_buf_get,
2951 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2956 /* Seq buffer is page-sized, exactly what we need. */
2958 count = iter->seq.len;
2959 ret = print_trace_line(iter);
2960 count = iter->seq.len - count;
2963 iter->seq.len -= count;
2966 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2967 iter->seq.len -= count;
2971 trace_consume(iter);
2973 if (!find_next_entry_inc(iter)) {
2983 static ssize_t tracing_splice_read_pipe(struct file *filp,
2985 struct pipe_inode_info *pipe,
2989 struct page *pages[PIPE_BUFFERS];
2990 struct partial_page partial[PIPE_BUFFERS];
2991 struct trace_iterator *iter = filp->private_data;
2992 struct splice_pipe_desc spd = {
2995 .nr_pages = 0, /* This gets updated below. */
2997 .ops = &tracing_pipe_buf_ops,
2998 .spd_release = tracing_spd_release_pipe,
3000 static struct tracer *old_tracer;
3005 /* copy the tracer to avoid using a global lock all around */
3006 mutex_lock(&trace_types_lock);
3007 if (unlikely(old_tracer != current_trace && current_trace)) {
3008 old_tracer = current_trace;
3009 *iter->trace = *current_trace;
3011 mutex_unlock(&trace_types_lock);
3013 mutex_lock(&iter->mutex);
3015 if (iter->trace->splice_read) {
3016 ret = iter->trace->splice_read(iter, filp,
3017 ppos, pipe, len, flags);
3022 ret = tracing_wait_pipe(filp);
3026 if (!iter->ent && !find_next_entry_inc(iter)) {
3031 /* Fill as many pages as possible. */
3032 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3033 pages[i] = alloc_page(GFP_KERNEL);
3037 rem = tracing_fill_pipe_page(rem, iter);
3039 /* Copy the data into the page, so we can start over. */
3040 ret = trace_seq_to_buffer(&iter->seq,
3041 page_address(pages[i]),
3044 __free_page(pages[i]);
3047 partial[i].offset = 0;
3048 partial[i].len = iter->seq.len;
3050 trace_seq_init(&iter->seq);
3053 mutex_unlock(&iter->mutex);
3057 return splice_to_pipe(pipe, &spd);
3060 mutex_unlock(&iter->mutex);
3066 tracing_entries_read(struct file *filp, char __user *ubuf,
3067 size_t cnt, loff_t *ppos)
3069 struct trace_array *tr = filp->private_data;
3073 mutex_lock(&trace_types_lock);
3074 if (!ring_buffer_expanded)
3075 r = sprintf(buf, "%lu (expanded: %lu)\n",
3077 trace_buf_size >> 10);
3079 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3080 mutex_unlock(&trace_types_lock);
3082 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3086 tracing_entries_write(struct file *filp, const char __user *ubuf,
3087 size_t cnt, loff_t *ppos)
3093 if (cnt >= sizeof(buf))
3096 if (copy_from_user(&buf, ubuf, cnt))
3101 ret = strict_strtoul(buf, 10, &val);
3105 /* must have at least 1 entry */
3109 mutex_lock(&trace_types_lock);
3113 /* disable all cpu buffers */
3114 for_each_tracing_cpu(cpu) {
3115 if (global_trace.data[cpu])
3116 atomic_inc(&global_trace.data[cpu]->disabled);
3117 if (max_tr.data[cpu])
3118 atomic_inc(&max_tr.data[cpu]->disabled);
3121 /* value is in KB */
3124 if (val != global_trace.entries) {
3125 ret = tracing_resize_ring_buffer(val);
3134 /* If check pages failed, return ENOMEM */
3135 if (tracing_disabled)
3138 for_each_tracing_cpu(cpu) {
3139 if (global_trace.data[cpu])
3140 atomic_dec(&global_trace.data[cpu]->disabled);
3141 if (max_tr.data[cpu])
3142 atomic_dec(&max_tr.data[cpu]->disabled);
3146 max_tr.entries = global_trace.entries;
3147 mutex_unlock(&trace_types_lock);
3152 static int mark_printk(const char *fmt, ...)
3156 va_start(args, fmt);
3157 ret = trace_vprintk(0, -1, fmt, args);
3163 tracing_mark_write(struct file *filp, const char __user *ubuf,
3164 size_t cnt, loff_t *fpos)
3169 if (tracing_disabled)
3172 if (cnt > TRACE_BUF_SIZE)
3173 cnt = TRACE_BUF_SIZE;
3175 buf = kmalloc(cnt + 1, GFP_KERNEL);
3179 if (copy_from_user(buf, ubuf, cnt)) {
3184 /* Cut from the first nil or newline. */
3186 end = strchr(buf, '\n');
3190 cnt = mark_printk("%s\n", buf);
3197 static const struct file_operations tracing_max_lat_fops = {
3198 .open = tracing_open_generic,
3199 .read = tracing_max_lat_read,
3200 .write = tracing_max_lat_write,
3203 static const struct file_operations tracing_ctrl_fops = {
3204 .open = tracing_open_generic,
3205 .read = tracing_ctrl_read,
3206 .write = tracing_ctrl_write,
3209 static const struct file_operations set_tracer_fops = {
3210 .open = tracing_open_generic,
3211 .read = tracing_set_trace_read,
3212 .write = tracing_set_trace_write,
3215 static const struct file_operations tracing_pipe_fops = {
3216 .open = tracing_open_pipe,
3217 .poll = tracing_poll_pipe,
3218 .read = tracing_read_pipe,
3219 .splice_read = tracing_splice_read_pipe,
3220 .release = tracing_release_pipe,
3223 static const struct file_operations tracing_entries_fops = {
3224 .open = tracing_open_generic,
3225 .read = tracing_entries_read,
3226 .write = tracing_entries_write,
3229 static const struct file_operations tracing_mark_fops = {
3230 .open = tracing_open_generic,
3231 .write = tracing_mark_write,
3234 struct ftrace_buffer_info {
3235 struct trace_array *tr;
3241 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3243 int cpu = (int)(long)inode->i_private;
3244 struct ftrace_buffer_info *info;
3246 if (tracing_disabled)
3249 info = kzalloc(sizeof(*info), GFP_KERNEL);
3253 info->tr = &global_trace;
3255 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3256 /* Force reading ring buffer for first read */
3257 info->read = (unsigned int)-1;
3261 filp->private_data = info;
3271 tracing_buffers_read(struct file *filp, char __user *ubuf,
3272 size_t count, loff_t *ppos)
3274 struct ftrace_buffer_info *info = filp->private_data;
3282 /* Do we have previous read data to read? */
3283 if (info->read < PAGE_SIZE)
3288 ret = ring_buffer_read_page(info->tr->buffer,
3295 pos = ring_buffer_page_len(info->spare);
3297 if (pos < PAGE_SIZE)
3298 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3301 size = PAGE_SIZE - info->read;
3305 ret = copy_to_user(ubuf, info->spare + info->read, size);
3316 static int tracing_buffers_release(struct inode *inode, struct file *file)
3318 struct ftrace_buffer_info *info = file->private_data;
3320 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3327 struct ring_buffer *buffer;
3332 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3333 struct pipe_buffer *buf)
3335 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3340 ring_buffer_free_read_page(ref->buffer, ref->page);
3345 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3346 struct pipe_buffer *buf)
3351 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3352 struct pipe_buffer *buf)
3354 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3359 /* Pipe buffer operations for a buffer. */
3360 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3362 .map = generic_pipe_buf_map,
3363 .unmap = generic_pipe_buf_unmap,
3364 .confirm = generic_pipe_buf_confirm,
3365 .release = buffer_pipe_buf_release,
3366 .steal = buffer_pipe_buf_steal,
3367 .get = buffer_pipe_buf_get,
3371 * Callback from splice_to_pipe(), if we need to release some pages
3372 * at the end of the spd in case we error'ed out in filling the pipe.
3374 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3376 struct buffer_ref *ref =
3377 (struct buffer_ref *)spd->partial[i].private;
3382 ring_buffer_free_read_page(ref->buffer, ref->page);
3384 spd->partial[i].private = 0;
3388 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3389 struct pipe_inode_info *pipe, size_t len,
3392 struct ftrace_buffer_info *info = file->private_data;
3393 struct partial_page partial[PIPE_BUFFERS];
3394 struct page *pages[PIPE_BUFFERS];
3395 struct splice_pipe_desc spd = {
3399 .ops = &buffer_pipe_buf_ops,
3400 .spd_release = buffer_spd_release,
3402 struct buffer_ref *ref;
3407 * We can't seek on a buffer input
3409 if (unlikely(*ppos))
3413 for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) {
3417 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3421 ref->buffer = info->tr->buffer;
3422 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3428 r = ring_buffer_read_page(ref->buffer, &ref->page,
3431 ring_buffer_free_read_page(ref->buffer,
3438 * zero out any left over data, this is going to
3441 size = ring_buffer_page_len(ref->page);
3442 if (size < PAGE_SIZE)
3443 memset(ref->page + size, 0, PAGE_SIZE - size);
3445 page = virt_to_page(ref->page);
3447 spd.pages[i] = page;
3448 spd.partial[i].len = PAGE_SIZE;
3449 spd.partial[i].offset = 0;
3450 spd.partial[i].private = (unsigned long)ref;
3456 /* did we read anything? */
3457 if (!spd.nr_pages) {
3458 if (flags & SPLICE_F_NONBLOCK)
3466 ret = splice_to_pipe(pipe, &spd);
3471 static const struct file_operations tracing_buffers_fops = {
3472 .open = tracing_buffers_open,
3473 .read = tracing_buffers_read,
3474 .release = tracing_buffers_release,
3475 .splice_read = tracing_buffers_splice_read,
3476 .llseek = no_llseek,
3479 #ifdef CONFIG_DYNAMIC_FTRACE
3481 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3487 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3488 size_t cnt, loff_t *ppos)
3490 static char ftrace_dyn_info_buffer[1024];
3491 static DEFINE_MUTEX(dyn_info_mutex);
3492 unsigned long *p = filp->private_data;
3493 char *buf = ftrace_dyn_info_buffer;
3494 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3497 mutex_lock(&dyn_info_mutex);
3498 r = sprintf(buf, "%ld ", *p);
3500 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3503 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3505 mutex_unlock(&dyn_info_mutex);
3510 static const struct file_operations tracing_dyn_info_fops = {
3511 .open = tracing_open_generic,
3512 .read = tracing_read_dyn_info,
3516 static struct dentry *d_tracer;
3518 struct dentry *tracing_init_dentry(void)
3525 d_tracer = debugfs_create_dir("tracing", NULL);
3527 if (!d_tracer && !once) {
3529 pr_warning("Could not create debugfs directory 'tracing'\n");
3536 static struct dentry *d_percpu;
3538 struct dentry *tracing_dentry_percpu(void)
3541 struct dentry *d_tracer;
3546 d_tracer = tracing_init_dentry();
3551 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3553 if (!d_percpu && !once) {
3555 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3562 static void tracing_init_debugfs_percpu(long cpu)
3564 struct dentry *d_percpu = tracing_dentry_percpu();
3565 struct dentry *entry, *d_cpu;
3566 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3569 if (cpu > 999 || cpu < 0)
3572 sprintf(cpu_dir, "cpu%ld", cpu);
3573 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3575 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3579 /* per cpu trace_pipe */
3580 entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
3581 (void *) cpu, &tracing_pipe_fops);
3583 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3586 entry = debugfs_create_file("trace", 0444, d_cpu,
3587 (void *) cpu, &tracing_fops);
3589 pr_warning("Could not create debugfs 'trace' entry\n");
3591 entry = debugfs_create_file("trace_pipe_raw", 0444, d_cpu,
3592 (void *) cpu, &tracing_buffers_fops);
3594 pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
3597 #ifdef CONFIG_FTRACE_SELFTEST
3598 /* Let selftest have access to static functions in this file */
3599 #include "trace_selftest.c"
3602 struct trace_option_dentry {
3603 struct tracer_opt *opt;
3604 struct tracer_flags *flags;
3605 struct dentry *entry;
3609 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3612 struct trace_option_dentry *topt = filp->private_data;
3615 if (topt->flags->val & topt->opt->bit)
3620 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3624 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3627 struct trace_option_dentry *topt = filp->private_data;
3632 if (cnt >= sizeof(buf))
3635 if (copy_from_user(&buf, ubuf, cnt))
3640 ret = strict_strtoul(buf, 10, &val);
3647 /* do nothing if already cleared */
3648 if (!(topt->flags->val & topt->opt->bit))
3651 mutex_lock(&trace_types_lock);
3652 if (current_trace->set_flag)
3653 ret = current_trace->set_flag(topt->flags->val,
3655 mutex_unlock(&trace_types_lock);
3658 topt->flags->val &= ~topt->opt->bit;
3661 /* do nothing if already set */
3662 if (topt->flags->val & topt->opt->bit)
3665 mutex_lock(&trace_types_lock);
3666 if (current_trace->set_flag)
3667 ret = current_trace->set_flag(topt->flags->val,
3669 mutex_unlock(&trace_types_lock);
3672 topt->flags->val |= topt->opt->bit;
3685 static const struct file_operations trace_options_fops = {
3686 .open = tracing_open_generic,
3687 .read = trace_options_read,
3688 .write = trace_options_write,
3692 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3695 long index = (long)filp->private_data;
3698 if (trace_flags & (1 << index))
3703 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3707 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3710 long index = (long)filp->private_data;
3715 if (cnt >= sizeof(buf))
3718 if (copy_from_user(&buf, ubuf, cnt))
3723 ret = strict_strtoul(buf, 10, &val);
3729 trace_flags &= ~(1 << index);
3732 trace_flags |= 1 << index;
3744 static const struct file_operations trace_options_core_fops = {
3745 .open = tracing_open_generic,
3746 .read = trace_options_core_read,
3747 .write = trace_options_core_write,
3750 static struct dentry *trace_options_init_dentry(void)
3752 struct dentry *d_tracer;
3753 static struct dentry *t_options;
3758 d_tracer = tracing_init_dentry();
3762 t_options = debugfs_create_dir("options", d_tracer);
3764 pr_warning("Could not create debugfs directory 'options'\n");
3772 create_trace_option_file(struct trace_option_dentry *topt,
3773 struct tracer_flags *flags,
3774 struct tracer_opt *opt)
3776 struct dentry *t_options;
3777 struct dentry *entry;
3779 t_options = trace_options_init_dentry();
3783 topt->flags = flags;
3786 entry = debugfs_create_file(opt->name, 0644, t_options, topt,
3787 &trace_options_fops);
3789 topt->entry = entry;
3793 static struct trace_option_dentry *
3794 create_trace_option_files(struct tracer *tracer)
3796 struct trace_option_dentry *topts;
3797 struct tracer_flags *flags;
3798 struct tracer_opt *opts;
3804 flags = tracer->flags;
3806 if (!flags || !flags->opts)
3811 for (cnt = 0; opts[cnt].name; cnt++)
3814 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3818 for (cnt = 0; opts[cnt].name; cnt++)
3819 create_trace_option_file(&topts[cnt], flags,
3826 destroy_trace_option_files(struct trace_option_dentry *topts)
3833 for (cnt = 0; topts[cnt].opt; cnt++) {
3834 if (topts[cnt].entry)
3835 debugfs_remove(topts[cnt].entry);
3841 static struct dentry *
3842 create_trace_option_core_file(const char *option, long index)
3844 struct dentry *t_options;
3845 struct dentry *entry;
3847 t_options = trace_options_init_dentry();
3851 entry = debugfs_create_file(option, 0644, t_options, (void *)index,
3852 &trace_options_core_fops);
3857 static __init void create_trace_options_dir(void)
3859 struct dentry *t_options;
3860 struct dentry *entry;
3863 t_options = trace_options_init_dentry();
3867 for (i = 0; trace_options[i]; i++) {
3868 entry = create_trace_option_core_file(trace_options[i], i);
3870 pr_warning("Could not create debugfs %s entry\n",
3875 static __init int tracer_init_debugfs(void)
3877 struct dentry *d_tracer;
3878 struct dentry *entry;
3881 d_tracer = tracing_init_dentry();
3883 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3884 &global_trace, &tracing_ctrl_fops);
3886 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3888 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3889 NULL, &tracing_iter_fops);
3891 pr_warning("Could not create debugfs 'trace_options' entry\n");
3893 create_trace_options_dir();
3895 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3896 NULL, &tracing_cpumask_fops);
3898 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3900 entry = debugfs_create_file("trace", 0444, d_tracer,
3901 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3903 pr_warning("Could not create debugfs 'trace' entry\n");
3905 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3906 &global_trace, &show_traces_fops);
3908 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3910 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3911 &global_trace, &set_tracer_fops);
3913 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3915 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3916 &tracing_max_latency,
3917 &tracing_max_lat_fops);
3919 pr_warning("Could not create debugfs "
3920 "'tracing_max_latency' entry\n");
3922 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3923 &tracing_thresh, &tracing_max_lat_fops);
3925 pr_warning("Could not create debugfs "
3926 "'tracing_thresh' entry\n");
3927 entry = debugfs_create_file("README", 0644, d_tracer,
3928 NULL, &tracing_readme_fops);
3930 pr_warning("Could not create debugfs 'README' entry\n");
3932 entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
3933 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3935 pr_warning("Could not create debugfs "
3936 "'trace_pipe' entry\n");
3938 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3939 &global_trace, &tracing_entries_fops);
3941 pr_warning("Could not create debugfs "
3942 "'buffer_size_kb' entry\n");
3944 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3945 NULL, &tracing_mark_fops);
3947 pr_warning("Could not create debugfs "
3948 "'trace_marker' entry\n");
3950 #ifdef CONFIG_DYNAMIC_FTRACE
3951 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3952 &ftrace_update_tot_cnt,
3953 &tracing_dyn_info_fops);
3955 pr_warning("Could not create debugfs "
3956 "'dyn_ftrace_total_info' entry\n");
3958 #ifdef CONFIG_SYSPROF_TRACER
3959 init_tracer_sysprof_debugfs(d_tracer);
3962 for_each_tracing_cpu(cpu)
3963 tracing_init_debugfs_percpu(cpu);
3968 static int trace_panic_handler(struct notifier_block *this,
3969 unsigned long event, void *unused)
3971 if (ftrace_dump_on_oops)
3976 static struct notifier_block trace_panic_notifier = {
3977 .notifier_call = trace_panic_handler,
3979 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3982 static int trace_die_handler(struct notifier_block *self,
3988 if (ftrace_dump_on_oops)
3997 static struct notifier_block trace_die_notifier = {
3998 .notifier_call = trace_die_handler,
4003 * printk is set to max of 1024, we really don't need it that big.
4004 * Nothing should be printing 1000 characters anyway.
4006 #define TRACE_MAX_PRINT 1000
4009 * Define here KERN_TRACE so that we have one place to modify
4010 * it if we decide to change what log level the ftrace dump
4013 #define KERN_TRACE KERN_EMERG
4016 trace_printk_seq(struct trace_seq *s)
4018 /* Probably should print a warning here. */
4022 /* should be zero ended, but we are paranoid. */
4023 s->buffer[s->len] = 0;
4025 printk(KERN_TRACE "%s", s->buffer);
4030 void ftrace_dump(void)
4032 static DEFINE_SPINLOCK(ftrace_dump_lock);
4033 /* use static because iter can be a bit big for the stack */
4034 static struct trace_iterator iter;
4035 static int dump_ran;
4036 unsigned long flags;
4040 spin_lock_irqsave(&ftrace_dump_lock, flags);
4046 /* No turning back! */
4050 for_each_tracing_cpu(cpu) {
4051 atomic_inc(&global_trace.data[cpu]->disabled);
4054 /* don't look at user memory in panic mode */
4055 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4057 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4059 /* Simulate the iterator */
4060 iter.tr = &global_trace;
4061 iter.trace = current_trace;
4062 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4065 * We need to stop all tracing on all CPUS to read the
4066 * the next buffer. This is a bit expensive, but is
4067 * not done often. We fill all what we can read,
4068 * and then release the locks again.
4071 while (!trace_empty(&iter)) {
4074 printk(KERN_TRACE "---------------------------------\n");
4078 /* reset all but tr, trace, and overruns */
4079 memset(&iter.seq, 0,
4080 sizeof(struct trace_iterator) -
4081 offsetof(struct trace_iterator, seq));
4082 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4085 if (find_next_entry_inc(&iter) != NULL) {
4086 print_trace_line(&iter);
4087 trace_consume(&iter);
4090 trace_printk_seq(&iter.seq);
4094 printk(KERN_TRACE " (ftrace buffer empty)\n");
4096 printk(KERN_TRACE "---------------------------------\n");
4099 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
4102 __init static int tracer_alloc_buffers(void)
4104 struct trace_array_cpu *data;
4109 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4112 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4113 goto out_free_buffer_mask;
4115 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4116 goto out_free_tracing_cpumask;
4118 /* To save memory, keep the ring buffer size to its minimum */
4119 if (ring_buffer_expanded)
4120 ring_buf_size = trace_buf_size;
4124 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4125 cpumask_copy(tracing_cpumask, cpu_all_mask);
4126 cpumask_clear(tracing_reader_cpumask);
4128 /* TODO: make the number of buffers hot pluggable with CPUS */
4129 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4130 TRACE_BUFFER_FLAGS);
4131 if (!global_trace.buffer) {
4132 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4134 goto out_free_cpumask;
4136 global_trace.entries = ring_buffer_size(global_trace.buffer);
4139 #ifdef CONFIG_TRACER_MAX_TRACE
4140 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4141 TRACE_BUFFER_FLAGS);
4142 if (!max_tr.buffer) {
4143 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4145 ring_buffer_free(global_trace.buffer);
4146 goto out_free_cpumask;
4148 max_tr.entries = ring_buffer_size(max_tr.buffer);
4149 WARN_ON(max_tr.entries != global_trace.entries);
4152 /* Allocate the first page for all buffers */
4153 for_each_tracing_cpu(i) {
4154 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4155 max_tr.data[i] = &per_cpu(max_data, i);
4158 trace_init_cmdlines();
4160 register_tracer(&nop_trace);
4161 current_trace = &nop_trace;
4162 #ifdef CONFIG_BOOT_TRACER
4163 register_tracer(&boot_tracer);
4165 /* All seems OK, enable tracing */
4166 tracing_disabled = 0;
4168 atomic_notifier_chain_register(&panic_notifier_list,
4169 &trace_panic_notifier);
4171 register_die_notifier(&trace_die_notifier);
4176 free_cpumask_var(tracing_reader_cpumask);
4177 out_free_tracing_cpumask:
4178 free_cpumask_var(tracing_cpumask);
4179 out_free_buffer_mask:
4180 free_cpumask_var(tracing_buffer_mask);
4185 __init static int clear_boot_tracer(void)
4188 * The default tracer at boot buffer is an init section.
4189 * This function is called in lateinit. If we did not
4190 * find the boot tracer, then clear it out, to prevent
4191 * later registration from accessing the buffer that is
4192 * about to be freed.
4194 if (!default_bootup_tracer)
4197 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4198 default_bootup_tracer);
4199 default_bootup_tracer = NULL;
4204 early_initcall(tracer_alloc_buffers);
4205 fs_initcall(tracer_init_debugfs);
4206 late_initcall(clear_boot_tracer);