sched: sched_clock_idle_[sleep|wakeup]_event()

construct a more or less wall-clock time out of sched_clock(), by
using ACPI-idle's existing knowledge about how much time we spent
idling. This allows the rq clock to work around TSC-stops-in-C2,
TSC-gets-corrupted-in-C3 type of problems.

( Besides the scheduler's statistics this also benefits blktrace and
  printk-timestamps as well. )

Furthermore, the precise before-C2/C3-sleep and after-C2/C3-wakeup
callbacks allow the scheduler to get out the most of the period where
the CPU has a reliable TSC. This results in slightly more precise
task statistics.

the ACPI bits were acked by Len.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Len Brown <len.brown@intel.com>

5 files changed, 61 insertions, 19 deletions

diff --git a/arch/i386/kernel/tsc.c b/arch/i386/kernel/tsc.c
index debd7dbb4158..a39280b4dd3a 100644
--- a/arch/i386/kernel/tsc.c
+++ b/arch/i386/kernel/tsc.c
@@ -292,7 +292,6 @@ static struct clocksource clocksource_tsc = {
 
 void mark_tsc_unstable(char *reason)
 {
-	sched_clock_unstable_event();
 	if (!tsc_unstable) {
 		tsc_unstable = 1;
 		tsc_enabled = 0;
diff --git a/drivers/acpi/processor_idle.c b/drivers/acpi/processor_idle.c
index a8634a0655fc..d9b8af763e1e 100644
--- a/drivers/acpi/processor_idle.c
+++ b/drivers/acpi/processor_idle.c
@@ -63,6 +63,7 @@
 ACPI_MODULE_NAME("processor_idle");
 #define ACPI_PROCESSOR_FILE_POWER	"power"
 #define US_TO_PM_TIMER_TICKS(t)		((t * (PM_TIMER_FREQUENCY/1000)) / 1000)
+#define PM_TIMER_TICK_NS		(1000000000ULL/PM_TIMER_FREQUENCY)
 #define C2_OVERHEAD			4	/* 1us (3.579 ticks per us) */
 #define C3_OVERHEAD			4	/* 1us (3.579 ticks per us) */
 static void (*pm_idle_save) (void) __read_mostly;
@@ -462,6 +463,9 @@ static void acpi_processor_idle(void)
 		 * TBD: Can't get time duration while in C1, as resumes
 		 *      go to an ISR rather than here.  Need to instrument
 		 *      base interrupt handler.
+		 *
+		 * Note: the TSC better not stop in C1, sched_clock() will
+		 *       skew otherwise.
 		 */
 		sleep_ticks = 0xFFFFFFFF;
 		break;
@@ -469,6 +473,8 @@ static void acpi_processor_idle(void)
 	case ACPI_STATE_C2:
 		/* Get start time (ticks) */
 		t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
+		/* Tell the scheduler that we are going deep-idle: */
+		sched_clock_idle_sleep_event();
 		/* Invoke C2 */
 		acpi_state_timer_broadcast(pr, cx, 1);
 		acpi_cstate_enter(cx);
@@ -479,17 +485,22 @@ static void acpi_processor_idle(void)
 		/* TSC halts in C2, so notify users */
 		mark_tsc_unstable("possible TSC halt in C2");
 #endif
+		/* Compute time (ticks) that we were actually asleep */
+		sleep_ticks = ticks_elapsed(t1, t2);
+
+		/* Tell the scheduler how much we idled: */
+		sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
+
 		/* Re-enable interrupts */
 		local_irq_enable();
+		/* Do not account our idle-switching overhead: */
+		sleep_ticks -= cx->latency_ticks + C2_OVERHEAD;
+
 		current_thread_info()->status |= TS_POLLING;
-		/* Compute time (ticks) that we were actually asleep */
-		sleep_ticks =
-		    ticks_elapsed(t1, t2) - cx->latency_ticks - C2_OVERHEAD;
 		acpi_state_timer_broadcast(pr, cx, 0);
 		break;
 
 	case ACPI_STATE_C3:
-
 		/*
 		 * disable bus master
 		 * bm_check implies we need ARB_DIS
@@ -518,6 +529,8 @@ static void acpi_processor_idle(void)
 		t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 		/* Invoke C3 */
 		acpi_state_timer_broadcast(pr, cx, 1);
+		/* Tell the scheduler that we are going deep-idle: */
+		sched_clock_idle_sleep_event();
 		acpi_cstate_enter(cx);
 		/* Get end time (ticks) */
 		t2 = inl(acpi_gbl_FADT.xpm_timer_block.address);
@@ -531,12 +544,17 @@ static void acpi_processor_idle(void)
 		/* TSC halts in C3, so notify users */
 		mark_tsc_unstable("TSC halts in C3");
 #endif
+		/* Compute time (ticks) that we were actually asleep */
+		sleep_ticks = ticks_elapsed(t1, t2);
+		/* Tell the scheduler how much we idled: */
+		sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
+
 		/* Re-enable interrupts */
 		local_irq_enable();
+		/* Do not account our idle-switching overhead: */
+		sleep_ticks -= cx->latency_ticks + C3_OVERHEAD;
+
 		current_thread_info()->status |= TS_POLLING;
-		/* Compute time (ticks) that we were actually asleep */
-		sleep_ticks =
-		    ticks_elapsed(t1, t2) - cx->latency_ticks - C3_OVERHEAD;
 		acpi_state_timer_broadcast(pr, cx, 0);
 		break;
 
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 682ef87da6eb..1845b2e99a87 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1388,7 +1388,8 @@ extern void sched_exec(void);
 #define sched_exec()   {}
 #endif
 
-extern void sched_clock_unstable_event(void);
+extern void sched_clock_idle_sleep_event(void);
+extern void sched_clock_idle_wakeup_event(u64 delta_ns);
 
 #ifdef CONFIG_HOTPLUG_CPU
 extern void idle_task_exit(void);
diff --git a/kernel/sched.c b/kernel/sched.c
index 45e17b83b7f1..48e7586168ef 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -262,7 +262,8 @@ struct rq {
 	s64 clock_max_delta;
 
 	unsigned int clock_warps, clock_overflows;
-	unsigned int clock_unstable_events;
+	u64 idle_clock;
+	unsigned int clock_deep_idle_events;
 	u64 tick_timestamp;
 
 	atomic_t nr_iowait;
@@ -556,18 +557,40 @@ static inline struct rq *this_rq_lock(void)
 }
 
 /*
- * CPU frequency is/was unstable - start new by setting prev_clock_raw:
+ * We are going deep-idle (irqs are disabled):
  */
-void sched_clock_unstable_event(void)
+void sched_clock_idle_sleep_event(void)
 {
-	unsigned long flags;
-	struct rq *rq;
+	struct rq *rq = cpu_rq(smp_processor_id());
 
-	rq = task_rq_lock(current, &flags);
-	rq->prev_clock_raw = sched_clock();
-	rq->clock_unstable_events++;
-	task_rq_unlock(rq, &flags);
+	spin_lock(&rq->lock);
+	__update_rq_clock(rq);
+	spin_unlock(&rq->lock);
+	rq->clock_deep_idle_events++;
+}
+EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event);
+
+/*
+ * We just idled delta nanoseconds (called with irqs disabled):
+ */
+void sched_clock_idle_wakeup_event(u64 delta_ns)
+{
+	struct rq *rq = cpu_rq(smp_processor_id());
+	u64 now = sched_clock();
+
+	rq->idle_clock += delta_ns;
+	/*
+	 * Override the previous timestamp and ignore all
+	 * sched_clock() deltas that occured while we idled,
+	 * and use the PM-provided delta_ns to advance the
+	 * rq clock:
+	 */
+	spin_lock(&rq->lock);
+	rq->prev_clock_raw = now;
+	rq->clock += delta_ns;
+	spin_unlock(&rq->lock);
 }
+EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event);
 
 /*
  * resched_task - mark a task 'to be rescheduled now'.
diff --git a/kernel/sched_debug.c b/kernel/sched_debug.c
index 87e524762b85..ab18f45f2ab2 100644
--- a/kernel/sched_debug.c
+++ b/kernel/sched_debug.c
@@ -154,10 +154,11 @@ static void print_cpu(struct seq_file *m, int cpu)
 	P(next_balance);
 	P(curr->pid);
 	P(clock);
+	P(idle_clock);
 	P(prev_clock_raw);
 	P(clock_warps);
 	P(clock_overflows);
-	P(clock_unstable_events);
+	P(clock_deep_idle_events);
 	P(clock_max_delta);
 	P(cpu_load[0]);
 	P(cpu_load[1]);