The patch titled
sleep profiling
has been removed from the -mm tree. Its filename was
sleep-profiling.patch
This patch was dropped because it was merged into mainline or a subsystem tree
------------------------------------------------------
Subject: sleep profiling
From: Ingo Molnar <mingo@xxxxxxx>
Implement prof=sleep profiling. TASK_UNINTERRUPTIBLE sleeps will be taken
as a profile hit, and every millisecond spent sleeping causes a profile-hit
for the call site that initiated the sleep.
Sample readprofile output on i386:
306 ps2_sendbyte 1.3973
432 call_usermodehelper_keys 1.9548
484 ps2_command 0.6453
790 __driver_attach 4.7879
1593 msleep 44.2500
3976 sync_buffer 64.1290
4076 do_lookup 12.4648
8587 sync_page 122.6714
20820 total 0.0067
(NOTE: architectures need to check whether get_wchan() can be called from
deep within the wakeup path.)
akpm: we need to mark more functions __sched. lock_sock(), msleep(), others..
akpm: the contention in do_lookup() is a surprise. Presumably doing disk
reads for directory contents while holding i_mutex.
[akpm@xxxxxxxx: various fixes]
Signed-off-by: Ingo Molnar <mingo@xxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxx>
---
Documentation/kernel-parameters.txt | 1
include/linux/profile.h | 24 +++++++++++++++-
kernel/profile.c | 39 ++++++++++++++++++++------
kernel/sched.c | 11 +++++++
4 files changed, 65 insertions(+), 10 deletions(-)
diff -puN Documentation/kernel-parameters.txt~sleep-profiling Documentation/kernel-parameters.txt
--- a/Documentation/kernel-parameters.txt~sleep-profiling
+++ a/Documentation/kernel-parameters.txt
@@ -1294,6 +1294,7 @@ and is between 256 and 4096 characters.
Param: "schedule" - profile schedule points.
Param: <number> - step/bucket size as a power of 2 for
statistical time based profiling.
+ Param: "sleep" - profile D-state sleeping (millisecs)
processor.max_cstate= [HW,ACPI]
Limit processor to maximum C-state
diff -puN include/linux/profile.h~sleep-profiling include/linux/profile.h
--- a/include/linux/profile.h~sleep-profiling
+++ a/include/linux/profile.h
@@ -6,10 +6,15 @@
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/cpumask.h>
+#include <linux/cache.h>
+
#include <asm/errno.h>
+extern int prof_on __read_mostly;
+
#define CPU_PROFILING 1
#define SCHED_PROFILING 2
+#define SLEEP_PROFILING 3
struct proc_dir_entry;
struct pt_regs;
@@ -18,7 +23,24 @@ struct notifier_block;
/* init basic kernel profiler */
void __init profile_init(void);
void profile_tick(int);
-void profile_hit(int, void *);
+
+/*
+ * Add multiple profiler hits to a given address:
+ */
+void profile_hits(int, void *ip, unsigned int nr_hits);
+
+/*
+ * Single profiler hit:
+ */
+static inline void profile_hit(int type, void *ip)
+{
+ /*
+ * Speedup for the common (no profiling enabled) case:
+ */
+ if (unlikely(prof_on == type))
+ profile_hits(type, ip, 1);
+}
+
#ifdef CONFIG_PROC_FS
void create_prof_cpu_mask(struct proc_dir_entry *);
#else
diff -puN kernel/profile.c~sleep-profiling kernel/profile.c
--- a/kernel/profile.c~sleep-profiling
+++ a/kernel/profile.c
@@ -40,7 +40,7 @@ int (*timer_hook)(struct pt_regs *) __re
static atomic_t *prof_buffer;
static unsigned long prof_len, prof_shift;
-static int prof_on __read_mostly;
+int prof_on __read_mostly;
static cpumask_t prof_cpu_mask = CPU_MASK_ALL;
#ifdef CONFIG_SMP
static DEFINE_PER_CPU(struct profile_hit *[2], cpu_profile_hits);
@@ -51,9 +51,19 @@ static DEFINE_MUTEX(profile_flip_mutex);
static int __init profile_setup(char * str)
{
static char __initdata schedstr[] = "schedule";
+ static char __initdata sleepstr[] = "sleep";
int par;
- if (!strncmp(str, schedstr, strlen(schedstr))) {
+ if (!strncmp(str, sleepstr, strlen(sleepstr))) {
+ prof_on = SLEEP_PROFILING;
+ if (str[strlen(sleepstr)] == ',')
+ str += strlen(sleepstr) + 1;
+ if (get_option(&str, &par))
+ prof_shift = par;
+ printk(KERN_INFO
+ "kernel sleep profiling enabled (shift: %ld)\n",
+ prof_shift);
+ } else if (!strncmp(str, sleepstr, strlen(sleepstr))) {
prof_on = SCHED_PROFILING;
if (str[strlen(schedstr)] == ',')
str += strlen(schedstr) + 1;
@@ -204,7 +214,8 @@ EXPORT_SYMBOL_GPL(profile_event_unregist
* positions to which hits are accounted during short intervals (e.g.
* several seconds) is usually very small. Exclusion from buffer
* flipping is provided by interrupt disablement (note that for
- * SCHED_PROFILING profile_hit() may be called from process context).
+ * SCHED_PROFILING or SLEEP_PROFILING profile_hit() may be called from
+ * process context).
* The hash function is meant to be lightweight as opposed to strong,
* and was vaguely inspired by ppc64 firmware-supported inverted
* pagetable hash functions, but uses a full hashtable full of finite
@@ -257,7 +268,7 @@ static void profile_discard_flip_buffers
mutex_unlock(&profile_flip_mutex);
}
-void profile_hit(int type, void *__pc)
+void profile_hits(int type, void *__pc, unsigned int nr_hits)
{
unsigned long primary, secondary, flags, pc = (unsigned long)__pc;
int i, j, cpu;
@@ -274,21 +285,31 @@ void profile_hit(int type, void *__pc)
put_cpu();
return;
}
+ /*
+ * We buffer the global profiler buffer into a per-CPU
+ * queue and thus reduce the number of global (and possibly
+ * NUMA-alien) accesses. The write-queue is self-coalescing:
+ */
local_irq_save(flags);
do {
for (j = 0; j < PROFILE_GRPSZ; ++j) {
if (hits[i + j].pc == pc) {
- hits[i + j].hits++;
+ hits[i + j].hits += nr_hits;
goto out;
} else if (!hits[i + j].hits) {
hits[i + j].pc = pc;
- hits[i + j].hits = 1;
+ hits[i + j].hits = nr_hits;
goto out;
}
}
i = (i + secondary) & (NR_PROFILE_HIT - 1);
} while (i != primary);
- atomic_inc(&prof_buffer[pc]);
+
+ /*
+ * Add the current hit(s) and flush the write-queue out
+ * to the global buffer:
+ */
+ atomic_add(nr_hits, &prof_buffer[pc]);
for (i = 0; i < NR_PROFILE_HIT; ++i) {
atomic_add(hits[i].hits, &prof_buffer[hits[i].pc]);
hits[i].pc = hits[i].hits = 0;
@@ -356,14 +377,14 @@ static int __devinit profile_cpu_callbac
#define profile_flip_buffers() do { } while (0)
#define profile_discard_flip_buffers() do { } while (0)
-void profile_hit(int type, void *__pc)
+void profile_hits(int type, void *__pc, unsigned int nr_hits)
{
unsigned long pc;
if (prof_on != type || !prof_buffer)
return;
pc = ((unsigned long)__pc - (unsigned long)_stext) >> prof_shift;
- atomic_inc(&prof_buffer[min(pc, prof_len - 1)]);
+ atomic_add(nr_hits, &prof_buffer[min(pc, prof_len - 1)]);
}
#endif /* !CONFIG_SMP */
diff -puN kernel/sched.c~sleep-profiling kernel/sched.c
--- a/kernel/sched.c~sleep-profiling
+++ a/kernel/sched.c
@@ -948,6 +948,17 @@ static void activate_task(struct task_st
}
#endif
+ /*
+ * Sleep time is in units of nanosecs, so shift by 20 to get a
+ * milliseconds-range estimation of the amount of time that the task
+ * spent sleeping:
+ */
+ if (unlikely(prof_on == SLEEP_PROFILING)) {
+ if (p->state == TASK_UNINTERRUPTIBLE)
+ profile_hits(SLEEP_PROFILING, (void *)get_wchan(p),
+ (now - p->timestamp) >> 20);
+ }
+
if (!rt_task(p))
p->prio = recalc_task_prio(p, now);
_
Patches currently in -mm which might be from mingo@xxxxxxx are
origin.patch
workqueue-dont-hold-workqueue_mutex-in-flush_scheduled_work.patch
acpi-i686-x86_64-fix-laptop-bootup-hang-in-init_acpi.patch
revert-x86_64-mm-add-genapic_force.patch
revert-x86_64-mm-fix-the-irqbalance-quirk-for-e7320-e7520-e7525.patch
convert-i386-pda-code-to-use-%fs.patch
genapic-optimize-fix-apic-mode-setup-2.patch
genapic-always-use-physical-delivery-mode-on-8-cpus.patch
genapic-remove-es7000-workaround.patch
genapic-remove-clustered-apic-mode.patch
genapic-default-to-physical-mode-on-hotplug-cpu-kernels.patch
cpuset-remove-sched-domain-hooks-from-cpusets.patch
cpei-gets-warning-at-kernel-irq-migrationc27-move_masked_irq.patch
time-re-add-verify_pmtmr_rate.patch
debug-add-sysrq_always_enabled-boot-option.patch
lockdep-filter-off-by-default.patch
lockdep-improve-verbose-messages.patch
lockdep-improve-lockdep_reset.patch
lockdep-clean-up-very_verbose-define.patch
lockdep-use-chain-hash-on-config_debug_lockdep-too.patch
lockdep-print-irq-trace-info-on-asserts.patch
lockdep-fix-possible-races-while-disabling-lock-debugging.patch
remove-the-old-bd_mutex-lockdep-annotation.patch
new-bd_mutex-lockdep-annotation.patch
remove-lock_key-approach-to-managing-nested-bd_mutex-locks.patch
simplify-some-aspects-of-bd_mutex-nesting.patch
use-mutex_lock_nested-for-bd_mutex-to-avoid-lockdep-warning.patch
avoid-lockdep-warning-in-md.patch
fix-generic-warn_on-message.patch
schedc-correct-comment-for-this_rq_lock-routine.patch
sched-fix-migration-cost-estimator.patch
sched-domain-move-sched-group-allocations-to-percpu-area.patch
move_task_off_dead_cpu-should-be-called-with-disabled-ints.patch
sched-domain-increase-the-smt-busy-rebalance-interval.patch
sched-avoid-taking-rq-lock-in-wake_priority_sleeper.patch
sched-remove-staggering-of-load-balancing.patch
sched-disable-interrupts-for-locking-in-load_balance.patch
sched-extract-load-calculation-from-rebalance_tick.patch
sched-move-idle-status-calculation-into-rebalance_tick.patch
sched-use-softirq-for-load-balancing.patch
sched-call-tasklet-less-frequently.patch
sched-add-option-to-serialize-load-balancing.patch
sched-add-option-to-serialize-load-balancing-fix.patch
sched-improve-migration-accuracy.patch
sched-improve-migration-accuracy-tidy.patch
sched-decrease-number-of-load-balances.patch
sched-remove-lb_stopbalance-counter.patch
sched-optimize-activate_task-for-rt-task.patch
kernel-schedc-whitespace-cleanups.patch
kernel-schedc-whitespace-cleanups-more.patch
mm-only-sched-add-a-few-scheduler-event-counters.patch
sched-add-above-background-load-function.patch
mm-implement-swap-prefetching.patch
mm-implement-swap-prefetching-use-ctl_unnumbered.patch
sched-cleanup-remove-task_t-convert-to-struct-task_struct-prefetch.patch
gtod-persistent-clock-support-core.patch
gtod-persistent-clock-support-i386.patch
time-uninline-jiffiesh.patch
time-uninline-jiffiesh-fix.patch
time-fix-msecs_to_jiffies-bug.patch
time-fix-timeout-overflow.patch
cleanup-uninline-irq_enter-and-move-it-into-a-function.patch
dynticks-extend-next_timer_interrupt-to-use-a-reference-jiffie.patch
dynticks-extend-next_timer_interrupt-to-use-a-reference-jiffie-remove-incorrect-warning-in-kernel-timerc.patch
hrtimers-namespace-and-enum-cleanup.patch
hrtimers-clean-up-locking.patch
hrtimers-clean-up-locking-fix.patch
updated-hrtimers-state-tracking.patch
updated-hrtimers-clean-up-callback-tracking.patch
updated-hrtimers-move-and-add-documentation.patch
updated-add-a-framework-to-manage-clock-event-devices.patch
updated-acpi-include-apich.patch
updated-acpi-keep-track-of-timer-broadcast.patch
updated-acpi-add-state-propagation-for-dynamic-broadcasting.patch
updated-i386-cleanup-apic-code.patch
updated-i386-convert-to-clock-event-devices.patch
updated-pm_timer-allow-early-access-and-move-externs-to-a-header-file.patch
updated-i386-rework-local-apic-calibration.patch
updated-high-res-timers-core.patch
updated-high-res-timers-core-high-res-timers-do-itimer-rearming-in-process-context.patch
updated-gtod-mark-tsc-unusable-for-highres-timers.patch
high-res-timers-utilize-tsc-clocksource-again.patch
high-res-timers-utilize-tsc-clocksource-again-fix.patch
updated-dynticks-core-code.patch
updated-dynticks-core-code-fix-resume-bug.patch
updated-dyntick-add-nohz-stats-to-proc-stat.patch
updated-dynticks-i386-arch-code.patch
updated-dynticks-fix-nmi-watchdog.patch
updated-high-res-timers-dynticks-enable-i386-support.patch
updated-debugging-feature-timer-stats.patch
clockevents-core-check-for-clock-event-device-handler-being-non-null-before-calling-it.patch
round_jiffies-infrastructure.patch
round_jiffies-infrastructure-fix.patch
clocksource-add-usage-of-config_sysfs.patch
clocksource-small-cleanup-2.patch
clocksource-small-cleanup-2-fix.patch
clocksource-small-acpi_pm-cleanup.patch
kvm-amd-svm-implementation-more-i386-fixes.patch
detect-atomic-counter-underflows.patch
debug-shared-irqs.patch
make-frame_pointer-default=y.patch
mutex-subsystem-synchro-test-module.patch
vdso-print-fatal-signals.patch
vdso-improve-print_fatal_signals-support-by-adding-memory-maps.patch
vdso-print-fatal-signals-use-ctl_unnumbered.patch
lockdep-show-held-locks-when-showing-a-stackdump.patch
lockdep-show-held-locks-when-showing-a-stackdump-fix.patch
lockdep-show-held-locks-when-showing-a-stackdump-fix-2.patch
-
To unsubscribe from this list: send the line "unsubscribe mm-commits" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
