![Steven Rostedt](/assets/img/avatar_default.png)
On Wed, 16 Jul 2008, Vegard Nossum wrote:
> When booting 4d3702b6
, I got this huge thing:
>
> Testing tracer wakeup: <4>------------[ cut here ]------------
> WARNING: at kernel/lockdep.c:2731 check_flags+0x123/0x160()
> Modules linked in:
> Pid: 1, comm: swapper Not tainted 2.6.26-crashing-02127-g4d3702b6 #30
> [<c015c349>] warn_on_slowpath+0x59/0xb0
> [<c01276c6>] ? ftrace_call+0x5/0x8
> [<c012d800>] ? native_read_tsc+0x0/0x20
> [<c0158de2>] ? sub_preempt_count+0x12/0xf0
> [<c01814eb>] ? trace_hardirqs_off+0xb/0x10
> [<c0182fbc>] ? __lock_acquire+0x2cc/0x1120
> [<c01814eb>] ? trace_hardirqs_off+0xb/0x10
> [<c01276af>] ? mcount_call+0x5/0xa
> [<c017ff53>] check_flags+0x123/0x160
> [<c0183e61>] lock_acquire+0x51/0xd0
> [<c01276c6>] ? ftrace_call+0x5/0x8
> [<c0613d4f>] _spin_lock_irqsave+0x5f/0xa0
> [<c01a8d45>] ? ftrace_record_ip+0xf5/0x220
> [<c02d5413>] ? debug_locks_off+0x3/0x50
> [<c01a8d45>] ftrace_record_ip+0xf5/0x220
> [<c01276af>] mcount_call+0x5/0xa
> [<c02d5418>] ? debug_locks_off+0x8/0x50
> [<c017ff27>] check_flags+0xf7/0x160
> [<c0183e61>] lock_acquire+0x51/0xd0
> [<c01276c6>] ? ftrace_call+0x5/0x8
> [<c0613d4f>] _spin_lock_irqsave+0x5f/0xa0
> [<c01affcd>] ? wakeup_tracer_call+0x6d/0xf0
> [<c01625e2>] ? _local_bh_enable+0x62/0xb0
> [<c0158ddd>] ? sub_preempt_count+0xd/0xf0
> [<c01affcd>] wakeup_tracer_call+0x6d/0xf0
> [<c0162724>] ? __do_softirq+0xf4/0x110
> [<c01afff1>] ? wakeup_tracer_call+0x91/0xf0
> [<c01276c6>] ftrace_call+0x5/0x8
> [<c0162724>] ? __do_softirq+0xf4/0x110
> [<c0158de2>] ? sub_preempt_count+0x12/0xf0
> [<c01625e2>] _local_bh_enable+0x62/0xb0
> [<c0162724>] __do_softirq+0xf4/0x110
> [<c01627ed>] do_softirq+0xad/0xb0
> [<c0162a15>] irq_exit+0xa5/0xb0
> [<c013a506>] smp_apic_timer_interrupt+0x66/0xa0
> [<c02d3fac>] ? trace_hardirqs_off_thunk+0xc/0x10
> [<c0127449>] apic_timer_interrupt+0x2d/0x34
> [<c018007b>] ? find_usage_backwards+0xb/0xf0
> [<c0613a09>] ? _spin_unlock_irqrestore+0x69/0x80
> [<c014ef32>] tg_shares_up+0x132/0x1d0
> [<c014d2a2>] walk_tg_tree+0x62/0xa0
> [<c014ee00>] ? tg_shares_up+0x0/0x1d0
> [<c014a860>] ? tg_nop+0x0/0x10
> [<c015499d>] update_shares+0x5d/0x80
> [<c0154a2f>] try_to_wake_up+0x6f/0x280
> [<c01a8b90>] ? __ftrace_modify_code+0x0/0xc0
> [<c01a8b90>] ? __ftrace_modify_code+0x0/0xc0
> [<c0154c94>] wake_up_process+0x14/0x20
> [<c01725f6>] kthread_create+0x66/0xb0
> [<c0195400>] ? do_stop+0x0/0x200
> [<c0195320>] ? __stop_machine_run+0x30/0xb0
> [<c0195340>] __stop_machine_run+0x50/0xb0
> [<c0195400>] ? do_stop+0x0/0x200
> [<c01a8b90>] ? __ftrace_modify_code+0x0/0xc0
> [<c061242d>] ? mutex_unlock+0xd/0x10
> [<c01953cc>] stop_machine_run+0x2c/0x60
> [<c01a94d3>] unregister_ftrace_function+0x103/0x180
> [<c01b0517>] stop_wakeup_tracer+0x17/0x60
> [<c01b056f>] wakeup_tracer_ctrl_update+0xf/0x30
> [<c01ab8d5>] trace_selftest_startup_wakeup+0xb5/0x130
> [<c01ab950>] ? trace_wakeup_test_thread+0x0/0x70
> [<c01aadf5>] register_tracer+0x135/0x1b0
> [<c0877d02>] init_wakeup_tracer+0xd/0xf
> [<c085d437>] kernel_init+0x1a9/0x2ce
> [<c061397b>] ? _spin_unlock_irq+0x3b/0x60
> [<c02d3f9c>] ? trace_hardirqs_on_thunk+0xc/0x10
> [<c0877cf5>] ? init_wakeup_tracer+0x0/0xf
> [<c0182646>] ? trace_hardirqs_on_caller+0x126/0x180
> [<c02d3f9c>] ? trace_hardirqs_on_thunk+0xc/0x10
> [<c01269c8>] ? restore_nocheck_notrace+0x0/0xe
> [<c085d28e>] ? kernel_init+0x0/0x2ce
> [<c085d28e>] ? kernel_init+0x0/0x2ce
> [<c01275fb>] kernel_thread_helper+0x7/0x10
> =======================
> ---[ end trace a7919e7f17c0a725 ]---
> irq event stamp: 579530
> hardirqs last enabled at (579528): [<c01826ab>] trace_hardirqs_on+0xb/0x10
> hardirqs last disabled at (579529): [<c01814eb>] trace_hardirqs_off+0xb/0x10
> softirqs last enabled at (579530): [<c0162724>] __do_softirq+0xf4/0x110
> softirqs last disabled at (579517): [<c01627ed>] do_softirq+0xad/0xb0
> irq event stamp: 579530
> hardirqs last enabled at (579528): [<c01826ab>] trace_hardirqs_on+0xb/0x10
> hardirqs last disabled at (579529): [<c01814eb>] trace_hardirqs_off+0xb/0x10
> softirqs last enabled at (579530): [<c0162724>] __do_softirq+0xf4/0x110
> softirqs last disabled at (579517): [<c01627ed>] do_softirq+0xad/0xb0
> PASSED
>
> Incidentally, the kernel also hung while I was typing in this report.
Things get weird between lockdep and ftrace because ftrace can be called
within lockdep internal code (via the mcount pointer) and lockdep can be
called with ftrace (via spin_locks).
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Tested-by: Vegard Nossum <vegard.nossum@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
454 lines
9.8 KiB
C
454 lines
9.8 KiB
C
/*
|
|
* trace task wakeup timings
|
|
*
|
|
* Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
|
|
* Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
|
|
*
|
|
* Based on code from the latency_tracer, that is:
|
|
*
|
|
* Copyright (C) 2004-2006 Ingo Molnar
|
|
* Copyright (C) 2004 William Lee Irwin III
|
|
*/
|
|
#include <linux/module.h>
|
|
#include <linux/fs.h>
|
|
#include <linux/debugfs.h>
|
|
#include <linux/kallsyms.h>
|
|
#include <linux/uaccess.h>
|
|
#include <linux/ftrace.h>
|
|
#include <linux/marker.h>
|
|
|
|
#include "trace.h"
|
|
|
|
static struct trace_array *wakeup_trace;
|
|
static int __read_mostly tracer_enabled;
|
|
|
|
static struct task_struct *wakeup_task;
|
|
static int wakeup_cpu;
|
|
static unsigned wakeup_prio = -1;
|
|
|
|
static raw_spinlock_t wakeup_lock =
|
|
(raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
|
|
|
|
static void __wakeup_reset(struct trace_array *tr);
|
|
|
|
#ifdef CONFIG_FTRACE
|
|
/*
|
|
* irqsoff uses its own tracer function to keep the overhead down:
|
|
*/
|
|
static void
|
|
wakeup_tracer_call(unsigned long ip, unsigned long parent_ip)
|
|
{
|
|
struct trace_array *tr = wakeup_trace;
|
|
struct trace_array_cpu *data;
|
|
unsigned long flags;
|
|
long disabled;
|
|
int resched;
|
|
int cpu;
|
|
|
|
if (likely(!wakeup_task))
|
|
return;
|
|
|
|
resched = need_resched();
|
|
preempt_disable_notrace();
|
|
|
|
cpu = raw_smp_processor_id();
|
|
data = tr->data[cpu];
|
|
disabled = atomic_inc_return(&data->disabled);
|
|
if (unlikely(disabled != 1))
|
|
goto out;
|
|
|
|
local_irq_save(flags);
|
|
__raw_spin_lock(&wakeup_lock);
|
|
|
|
if (unlikely(!wakeup_task))
|
|
goto unlock;
|
|
|
|
/*
|
|
* The task can't disappear because it needs to
|
|
* wake up first, and we have the wakeup_lock.
|
|
*/
|
|
if (task_cpu(wakeup_task) != cpu)
|
|
goto unlock;
|
|
|
|
trace_function(tr, data, ip, parent_ip, flags);
|
|
|
|
unlock:
|
|
__raw_spin_unlock(&wakeup_lock);
|
|
local_irq_restore(flags);
|
|
|
|
out:
|
|
atomic_dec(&data->disabled);
|
|
|
|
/*
|
|
* To prevent recursion from the scheduler, if the
|
|
* resched flag was set before we entered, then
|
|
* don't reschedule.
|
|
*/
|
|
if (resched)
|
|
preempt_enable_no_resched_notrace();
|
|
else
|
|
preempt_enable_notrace();
|
|
}
|
|
|
|
static struct ftrace_ops trace_ops __read_mostly =
|
|
{
|
|
.func = wakeup_tracer_call,
|
|
};
|
|
#endif /* CONFIG_FTRACE */
|
|
|
|
/*
|
|
* Should this new latency be reported/recorded?
|
|
*/
|
|
static int report_latency(cycle_t delta)
|
|
{
|
|
if (tracing_thresh) {
|
|
if (delta < tracing_thresh)
|
|
return 0;
|
|
} else {
|
|
if (delta <= tracing_max_latency)
|
|
return 0;
|
|
}
|
|
return 1;
|
|
}
|
|
|
|
static void notrace
|
|
wakeup_sched_switch(void *private, void *rq, struct task_struct *prev,
|
|
struct task_struct *next)
|
|
{
|
|
unsigned long latency = 0, t0 = 0, t1 = 0;
|
|
struct trace_array **ptr = private;
|
|
struct trace_array *tr = *ptr;
|
|
struct trace_array_cpu *data;
|
|
cycle_t T0, T1, delta;
|
|
unsigned long flags;
|
|
long disabled;
|
|
int cpu;
|
|
|
|
if (unlikely(!tracer_enabled))
|
|
return;
|
|
|
|
/*
|
|
* When we start a new trace, we set wakeup_task to NULL
|
|
* and then set tracer_enabled = 1. We want to make sure
|
|
* that another CPU does not see the tracer_enabled = 1
|
|
* and the wakeup_task with an older task, that might
|
|
* actually be the same as next.
|
|
*/
|
|
smp_rmb();
|
|
|
|
if (next != wakeup_task)
|
|
return;
|
|
|
|
/* The task we are waiting for is waking up */
|
|
data = tr->data[wakeup_cpu];
|
|
|
|
/* disable local data, not wakeup_cpu data */
|
|
cpu = raw_smp_processor_id();
|
|
disabled = atomic_inc_return(&tr->data[cpu]->disabled);
|
|
if (likely(disabled != 1))
|
|
goto out;
|
|
|
|
local_irq_save(flags);
|
|
__raw_spin_lock(&wakeup_lock);
|
|
|
|
/* We could race with grabbing wakeup_lock */
|
|
if (unlikely(!tracer_enabled || next != wakeup_task))
|
|
goto out_unlock;
|
|
|
|
trace_function(tr, data, CALLER_ADDR1, CALLER_ADDR2, flags);
|
|
|
|
/*
|
|
* usecs conversion is slow so we try to delay the conversion
|
|
* as long as possible:
|
|
*/
|
|
T0 = data->preempt_timestamp;
|
|
T1 = ftrace_now(cpu);
|
|
delta = T1-T0;
|
|
|
|
if (!report_latency(delta))
|
|
goto out_unlock;
|
|
|
|
latency = nsecs_to_usecs(delta);
|
|
|
|
tracing_max_latency = delta;
|
|
t0 = nsecs_to_usecs(T0);
|
|
t1 = nsecs_to_usecs(T1);
|
|
|
|
update_max_tr(tr, wakeup_task, wakeup_cpu);
|
|
|
|
out_unlock:
|
|
__wakeup_reset(tr);
|
|
__raw_spin_unlock(&wakeup_lock);
|
|
local_irq_restore(flags);
|
|
out:
|
|
atomic_dec(&tr->data[cpu]->disabled);
|
|
}
|
|
|
|
static notrace void
|
|
sched_switch_callback(void *probe_data, void *call_data,
|
|
const char *format, va_list *args)
|
|
{
|
|
struct task_struct *prev;
|
|
struct task_struct *next;
|
|
struct rq *__rq;
|
|
|
|
/* skip prev_pid %d next_pid %d prev_state %ld */
|
|
(void)va_arg(*args, int);
|
|
(void)va_arg(*args, int);
|
|
(void)va_arg(*args, long);
|
|
__rq = va_arg(*args, typeof(__rq));
|
|
prev = va_arg(*args, typeof(prev));
|
|
next = va_arg(*args, typeof(next));
|
|
|
|
tracing_record_cmdline(prev);
|
|
|
|
/*
|
|
* If tracer_switch_func only points to the local
|
|
* switch func, it still needs the ptr passed to it.
|
|
*/
|
|
wakeup_sched_switch(probe_data, __rq, prev, next);
|
|
}
|
|
|
|
static void __wakeup_reset(struct trace_array *tr)
|
|
{
|
|
struct trace_array_cpu *data;
|
|
int cpu;
|
|
|
|
for_each_possible_cpu(cpu) {
|
|
data = tr->data[cpu];
|
|
tracing_reset(data);
|
|
}
|
|
|
|
wakeup_cpu = -1;
|
|
wakeup_prio = -1;
|
|
|
|
if (wakeup_task)
|
|
put_task_struct(wakeup_task);
|
|
|
|
wakeup_task = NULL;
|
|
}
|
|
|
|
static void wakeup_reset(struct trace_array *tr)
|
|
{
|
|
unsigned long flags;
|
|
|
|
local_irq_save(flags);
|
|
__raw_spin_lock(&wakeup_lock);
|
|
__wakeup_reset(tr);
|
|
__raw_spin_unlock(&wakeup_lock);
|
|
local_irq_restore(flags);
|
|
}
|
|
|
|
static void
|
|
wakeup_check_start(struct trace_array *tr, struct task_struct *p,
|
|
struct task_struct *curr)
|
|
{
|
|
int cpu = smp_processor_id();
|
|
unsigned long flags;
|
|
long disabled;
|
|
|
|
if (likely(!rt_task(p)) ||
|
|
p->prio >= wakeup_prio ||
|
|
p->prio >= curr->prio)
|
|
return;
|
|
|
|
disabled = atomic_inc_return(&tr->data[cpu]->disabled);
|
|
if (unlikely(disabled != 1))
|
|
goto out;
|
|
|
|
/* interrupts should be off from try_to_wake_up */
|
|
__raw_spin_lock(&wakeup_lock);
|
|
|
|
/* check for races. */
|
|
if (!tracer_enabled || p->prio >= wakeup_prio)
|
|
goto out_locked;
|
|
|
|
/* reset the trace */
|
|
__wakeup_reset(tr);
|
|
|
|
wakeup_cpu = task_cpu(p);
|
|
wakeup_prio = p->prio;
|
|
|
|
wakeup_task = p;
|
|
get_task_struct(wakeup_task);
|
|
|
|
local_save_flags(flags);
|
|
|
|
tr->data[wakeup_cpu]->preempt_timestamp = ftrace_now(cpu);
|
|
trace_function(tr, tr->data[wakeup_cpu],
|
|
CALLER_ADDR1, CALLER_ADDR2, flags);
|
|
|
|
out_locked:
|
|
__raw_spin_unlock(&wakeup_lock);
|
|
out:
|
|
atomic_dec(&tr->data[cpu]->disabled);
|
|
}
|
|
|
|
static notrace void
|
|
wake_up_callback(void *probe_data, void *call_data,
|
|
const char *format, va_list *args)
|
|
{
|
|
struct trace_array **ptr = probe_data;
|
|
struct trace_array *tr = *ptr;
|
|
struct task_struct *curr;
|
|
struct task_struct *task;
|
|
struct rq *__rq;
|
|
|
|
if (likely(!tracer_enabled))
|
|
return;
|
|
|
|
/* Skip pid %d state %ld */
|
|
(void)va_arg(*args, int);
|
|
(void)va_arg(*args, long);
|
|
/* now get the meat: "rq %p task %p rq->curr %p" */
|
|
__rq = va_arg(*args, typeof(__rq));
|
|
task = va_arg(*args, typeof(task));
|
|
curr = va_arg(*args, typeof(curr));
|
|
|
|
tracing_record_cmdline(task);
|
|
tracing_record_cmdline(curr);
|
|
|
|
wakeup_check_start(tr, task, curr);
|
|
}
|
|
|
|
static void start_wakeup_tracer(struct trace_array *tr)
|
|
{
|
|
int ret;
|
|
|
|
ret = marker_probe_register("kernel_sched_wakeup",
|
|
"pid %d state %ld ## rq %p task %p rq->curr %p",
|
|
wake_up_callback,
|
|
&wakeup_trace);
|
|
if (ret) {
|
|
pr_info("wakeup trace: Couldn't add marker"
|
|
" probe to kernel_sched_wakeup\n");
|
|
return;
|
|
}
|
|
|
|
ret = marker_probe_register("kernel_sched_wakeup_new",
|
|
"pid %d state %ld ## rq %p task %p rq->curr %p",
|
|
wake_up_callback,
|
|
&wakeup_trace);
|
|
if (ret) {
|
|
pr_info("wakeup trace: Couldn't add marker"
|
|
" probe to kernel_sched_wakeup_new\n");
|
|
goto fail_deprobe;
|
|
}
|
|
|
|
ret = marker_probe_register("kernel_sched_schedule",
|
|
"prev_pid %d next_pid %d prev_state %ld "
|
|
"## rq %p prev %p next %p",
|
|
sched_switch_callback,
|
|
&wakeup_trace);
|
|
if (ret) {
|
|
pr_info("sched trace: Couldn't add marker"
|
|
" probe to kernel_sched_schedule\n");
|
|
goto fail_deprobe_wake_new;
|
|
}
|
|
|
|
wakeup_reset(tr);
|
|
|
|
/*
|
|
* Don't let the tracer_enabled = 1 show up before
|
|
* the wakeup_task is reset. This may be overkill since
|
|
* wakeup_reset does a spin_unlock after setting the
|
|
* wakeup_task to NULL, but I want to be safe.
|
|
* This is a slow path anyway.
|
|
*/
|
|
smp_wmb();
|
|
|
|
register_ftrace_function(&trace_ops);
|
|
|
|
tracer_enabled = 1;
|
|
|
|
return;
|
|
fail_deprobe_wake_new:
|
|
marker_probe_unregister("kernel_sched_wakeup_new",
|
|
wake_up_callback,
|
|
&wakeup_trace);
|
|
fail_deprobe:
|
|
marker_probe_unregister("kernel_sched_wakeup",
|
|
wake_up_callback,
|
|
&wakeup_trace);
|
|
}
|
|
|
|
static void stop_wakeup_tracer(struct trace_array *tr)
|
|
{
|
|
tracer_enabled = 0;
|
|
unregister_ftrace_function(&trace_ops);
|
|
marker_probe_unregister("kernel_sched_schedule",
|
|
sched_switch_callback,
|
|
&wakeup_trace);
|
|
marker_probe_unregister("kernel_sched_wakeup_new",
|
|
wake_up_callback,
|
|
&wakeup_trace);
|
|
marker_probe_unregister("kernel_sched_wakeup",
|
|
wake_up_callback,
|
|
&wakeup_trace);
|
|
}
|
|
|
|
static void wakeup_tracer_init(struct trace_array *tr)
|
|
{
|
|
wakeup_trace = tr;
|
|
|
|
if (tr->ctrl)
|
|
start_wakeup_tracer(tr);
|
|
}
|
|
|
|
static void wakeup_tracer_reset(struct trace_array *tr)
|
|
{
|
|
if (tr->ctrl) {
|
|
stop_wakeup_tracer(tr);
|
|
/* make sure we put back any tasks we are tracing */
|
|
wakeup_reset(tr);
|
|
}
|
|
}
|
|
|
|
static void wakeup_tracer_ctrl_update(struct trace_array *tr)
|
|
{
|
|
if (tr->ctrl)
|
|
start_wakeup_tracer(tr);
|
|
else
|
|
stop_wakeup_tracer(tr);
|
|
}
|
|
|
|
static void wakeup_tracer_open(struct trace_iterator *iter)
|
|
{
|
|
/* stop the trace while dumping */
|
|
if (iter->tr->ctrl)
|
|
stop_wakeup_tracer(iter->tr);
|
|
}
|
|
|
|
static void wakeup_tracer_close(struct trace_iterator *iter)
|
|
{
|
|
/* forget about any processes we were recording */
|
|
if (iter->tr->ctrl)
|
|
start_wakeup_tracer(iter->tr);
|
|
}
|
|
|
|
static struct tracer wakeup_tracer __read_mostly =
|
|
{
|
|
.name = "wakeup",
|
|
.init = wakeup_tracer_init,
|
|
.reset = wakeup_tracer_reset,
|
|
.open = wakeup_tracer_open,
|
|
.close = wakeup_tracer_close,
|
|
.ctrl_update = wakeup_tracer_ctrl_update,
|
|
.print_max = 1,
|
|
#ifdef CONFIG_FTRACE_SELFTEST
|
|
.selftest = trace_selftest_startup_wakeup,
|
|
#endif
|
|
};
|
|
|
|
__init static int init_wakeup_tracer(void)
|
|
{
|
|
int ret;
|
|
|
|
ret = register_tracer(&wakeup_tracer);
|
|
if (ret)
|
|
return ret;
|
|
|
|
return 0;
|
|
}
|
|
device_initcall(init_wakeup_tracer);
|