The patch titled
sleep profiling
has been added to the -mm tree. Its filename is
sleep-profiling.patch
See http://www.zip.com.au/~akpm/linux/patches/stuff/added-to-mm.txt to find
out what to do about this
------------------------------------------------------
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.)
Signed-off-by: Ingo Molnar <mingo@xxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxx>
---
include/linux/profile.h | 22 ++++++++++++++++++++-
kernel/profile.c | 39 +++++++++++++++++++++++++++++---------
kernel/sched.c | 11 ++++++++++
3 files changed, 62 insertions(+), 10 deletions(-)
diff -puN include/linux/profile.h~sleep-profiling include/linux/profile.h
--- a/include/linux/profile.h~sleep-profiling
+++ a/include/linux/profile.h
@@ -8,8 +8,11 @@
#include <linux/cpumask.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 +21,24 @@ 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 30
+ * 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
bonding-lockdep-annotation.patch
lockdep-annotate-nfs-nfsd-in-kernel-sockets.patch
lockdep-annotate-nfs-nfsd-in-kernel-sockets-tidy.patch
sysrq-x-show-blocked-tasks.patch
setup_irq-better-mismatch-debugging.patch
lockdep-internal-locking-fixes.patch
lockdep-misc-fixes-in-lockdepc.patch
cpuset-remove-sched-domain-hooks-from-cpusets.patch
binfmt_elf-randomize-pie-binaries.patch
enforce-unsigned-long-flags-when-spinlocking.patch
disable-clone_child_cleartid-for-abnormal-exit.patch
lockdep-spin_lock_irqsave_nested.patch
lockdep-spin_lock_irqsave_nested-fix.patch
lockdep-spin_lock_irqsave_nested-fix-2.patch
lockdep-annotate-bcsp-driver.patch
lockdep-print-current-locks-on-in_atomic-warnings.patch
lockdep-name-some-old-style-locks.patch
debug-workqueue-locking-sanity-v2.patch
sleep-profiling.patch
sleep-profiling-fixes.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-avoid-taking-rq-lock-in-wake_priority_sleeper.patch
sched-disable-interrupts-for-locking-in-load_balance.patch
sched-extract-load-calculation-from-rebalance_tick.patch
sched-stagger-load-balancing-in-build_sched_domains.patch
sched-move-idle-stat-calculation-into-rebalance_tick.patch
sched-use-tasklet-to-call-balancing.patch
sched-call-tasklet-less-frequently.patch
sched-call-tasklet-less-frequently-fix.patch
kernel-schedc-whitespace-cleanups.patch
kernel-schedc-whitespace-cleanups-more.patch
sched-add-above-background-load-function.patch
mm-implement-swap-prefetching.patch
sched-cleanup-remove-task_t-convert-to-struct-task_struct-prefetch.patch
gtod-exponential-update_wall_time.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-state-tracking.patch
hrtimers-clean-up-callback-tracking.patch
hrtimers-move-and-add-documentation.patch
clockevents-core.patch
clockevents-drivers-for-i386.patch
high-res-timers-core.patch
gtod-mark-tsc-unusable-for-highres-timers.patch
dynticks-core.patch
dynticks-add-nohz-stats-to-proc-stat.patch
dynticks-i386-arch-code.patch
high-res-timers-dynticks-enable-i386-support.patch
debugging-feature-timer-stats.patch
round_jiffies-infrastructure.patch
round_jiffies-infrastructure-fix.patch
clocksource-add-usage-of-config_sysfs.patch
clocksource-small-cleanup-2.patch
clocksource-small-acpi_pm-cleanup.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
-
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
