Skip to contentSkip to conversation

3cce50dfec4a arm64: perf: Disable PMU while processing counter overflows

v4.14.y In review Aman Priyadarshi 3 participants 7 messages Jun 16, 2021 – Jun 25, 2021

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

Hi Greg,

Following the conversation with Marc, we discovered that an important fix for ARM PMU is not backported to 4.14.y tree, it affects counter value and give out nonsensical result.

Can you please include the following commit?

```
commit 3cce50dfec4a5b0414c974190940f47dd32c6dee
Author: Suzuki K Poulose <[email protected]>
Date:   Tue Jul 10 09:58:03 2018 +0100

    arm64: perf: Disable PMU while processing counter overflows
    
    The arm64 PMU updates the event counters and reprograms the
    counters in the overflow IRQ handler without disabling the
    PMU. This could potentially cause skews in for group counters,
    where the overflowed counters may potentially loose some event
    counts, while they are reprogrammed. To prevent this, disable
    the PMU while we process the counter overflows and enable it
    right back when we are done.
    
    This patch also moves the PMU stop/start routines to avoid a
    forward declaration.
    
    Suggested-by: Mark Rutland <[email protected]>
    Cc: Will Deacon <[email protected]>
    Acked-by: Mark Rutland <[email protected]>
    Signed-off-by: Suzuki K Poulose <[email protected]>
    Signed-off-by: Will Deacon <[email protected]>
```

For more details:

https://lists.cs.columbia.edu/pipermail/kvmarm/2021-June/047471.html

Regards,
Aman Priyadarshi

Amazon Development Center Germany GmbH
Krausenstr. 38
10117 Berlin
Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss Eingetragen am Amtsgericht Charlottenburg unter HRB 149173 B
Sitz: Berlin
Ust-ID: DE 289 237 879

#1 lore raw
Quoted 23 lines · Aman Priyadarshi

Hi Greg,

Following the conversation with Marc, we discovered that an important fix for ARM PMU is not backported to 4.14.y tree, it affects counter value and give out nonsensical result.

Can you please include the following commit?

```
commit 3cce50dfec4a5b0414c974190940f47dd32c6dee
Author: Suzuki K Poulose <[email protected]>
Date:   Tue Jul 10 09:58:03 2018 +0100

    arm64: perf: Disable PMU while processing counter overflows
    
    The arm64 PMU updates the event counters and reprograms the
    counters in the overflow IRQ handler without disabling the
    PMU. This could potentially cause skews in for group counters,
    where the overflowed counters may potentially loose some event
    counts, while they are reprogrammed. To prevent this, disable
    the PMU while we process the counter overflows and enable it
    right back when we are done.
    
    This patch also moves the PMU stop/start routines to avoid a
Quoted 8 lines
    forward declaration.
    
    Suggested-by: Mark Rutland <[email protected]>
    Cc: Will Deacon <[email protected]>
    Acked-by: Mark Rutland <[email protected]>
    Signed-off-by: Suzuki K Poulose <[email protected]>
    Signed-off-by: Will Deacon <[email protected]>
```

It does not apply cleanly so you will have to provide a working backport for us.

Quoted 3 lines

For more details:

https://lists.cs.columbia.edu/pipermail/kvmarm/2021-June/047471.html

lore.kernel.org is your friend :)

thanks,

greg k-h

#2 lore raw

PATCH 1/1 arm64: perf: Disable PMU while processing counter overflows

From: Suzuki K Poulose <[email protected]>

[ Upstream commit 3cce50dfec4a5b0414c974190940f47dd32c6dee ]

The arm64 PMU updates the event counters and reprograms the counters in the overflow IRQ handler without disabling the PMU. This could potentially cause skews in for group counters, where the overflowed counters may potentially loose some event counts, while they are reprogrammed. To prevent this, disable the PMU while we process the counter overflows and enable it right back when we are done.

This patch also moves the PMU stop/start routines to avoid a forward declaration.

---
 arch/arm64/kernel/perf_event.c | 50 +++++++++++++++++++---------------
 1 file changed, 28 insertions(+), 22 deletions(-)
arch/arm64/kernel/perf_event.c +28 -22
index 53df84b2a07f..4ee1228d29eb 100644
@@ -670,6 +670,28 @@ static void armv8pmu_disable_event(struct perf_event *event)
 	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
 }
 
+static void armv8pmu_start(struct arm_pmu *cpu_pmu)
+{
+	unsigned long flags;
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
+
+	raw_spin_lock_irqsave(&events->pmu_lock, flags);
+	/* Enable all counters */
+	armv8pmu_pmcr_write(armv8pmu_pmcr_read() | ARMV8_PMU_PMCR_E);
+	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+}
+
+static void armv8pmu_stop(struct arm_pmu *cpu_pmu)
+{
+	unsigned long flags;
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
+
+	raw_spin_lock_irqsave(&events->pmu_lock, flags);
+	/* Disable all counters */
+	armv8pmu_pmcr_write(armv8pmu_pmcr_read() & ~ARMV8_PMU_PMCR_E);
+	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+}
+
 static irqreturn_t armv8pmu_handle_irq(int irq_num, void *dev)
 {
 	u32 pmovsr;
@@ -695,6 +717,11 @@ static irqreturn_t armv8pmu_handle_irq(int irq_num, void *dev)
 	 */
 	regs = get_irq_regs();
 
+	/*
+	 * Stop the PMU while processing the counter overflows
+	 * to prevent skews in group events.
+	 */
+	armv8pmu_stop(cpu_pmu);
 	for (idx = 0; idx < cpu_pmu->num_events; ++idx) {
 		struct perf_event *event = cpuc->events[idx];
 		struct hw_perf_event *hwc;
@@ -719,6 +746,7 @@ static irqreturn_t armv8pmu_handle_irq(int irq_num, void *dev)
 		if (perf_event_overflow(event, &data, regs))
 			cpu_pmu->disable(event);
 	}
+	armv8pmu_start(cpu_pmu);
 
 	/*
 	 * Handle the pending perf events.
@@ -732,28 +760,6 @@ static irqreturn_t armv8pmu_handle_irq(int irq_num, void *dev)
 	return IRQ_HANDLED;
 }
 
-static void armv8pmu_start(struct arm_pmu *cpu_pmu)
-{
-	unsigned long flags;
-	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
-
-	raw_spin_lock_irqsave(&events->pmu_lock, flags);
-	/* Enable all counters */
-	armv8pmu_pmcr_write(armv8pmu_pmcr_read() | ARMV8_PMU_PMCR_E);
-	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
-}
-
-static void armv8pmu_stop(struct arm_pmu *cpu_pmu)
-{
-	unsigned long flags;
-	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
-
-	raw_spin_lock_irqsave(&events->pmu_lock, flags);
-	/* Disable all counters */
-	armv8pmu_pmcr_write(armv8pmu_pmcr_read() & ~ARMV8_PMU_PMCR_E);
-	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
-}
-
 static int armv8pmu_get_event_idx(struct pmu_hw_events *cpuc,
 				  struct perf_event *event)
 {
Signature · 12 lines
2.17.1




Amazon Development Center Germany GmbH
Krausenstr. 38
10117 Berlin
Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss
Eingetragen am Amtsgericht Charlottenburg unter HRB 149173 B
Sitz: Berlin
Ust-ID: DE 289 237 879
#3 lore raw

Re: [PATCH] arm64: perf: Disable PMU while processing counter overflows

Quoted 22 lines · Aman Priyadarshi

From: Suzuki K Poulose <[email protected]>

[ Upstream commit 3cce50dfec4a5b0414c974190940f47dd32c6dee ]

The arm64 PMU updates the event counters and reprograms the counters in the overflow IRQ handler without disabling the PMU. This could potentially cause skews in for group counters, where the overflowed counters may potentially loose some event counts, while they are reprogrammed. To prevent this, disable the PMU while we process the counter overflows and enable it right back when we are done.

This patch also moves the PMU stop/start routines to avoid a forward declaration.

Quoted 3 lines
---
 arch/arm64/kernel/perf_event.c | 50 +++++++++++++++++++---------------
 1 file changed, 28 insertions(+), 22 deletions(-)

What stable tree(s) do you want this applied to?

thanks,

greg k-h

#4 lore raw

Re: [PATCH] arm64: perf: Disable PMU while processing counter overflows

Quoted 28 lines · Greg Kroah-Hartman
Quoted 22 lines · Aman Priyadarshi

From: Suzuki K Poulose <[email protected]>

[ Upstream commit 3cce50dfec4a5b0414c974190940f47dd32c6dee ]

The arm64 PMU updates the event counters and reprograms the counters in the overflow IRQ handler without disabling the PMU. This could potentially cause skews in for group counters, where the overflowed counters may potentially loose some event counts, while they are reprogrammed. To prevent this, disable the PMU while we process the counter overflows and enable it right back when we are done.

This patch also moves the PMU stop/start routines to avoid a forward declaration.

Quoted 3 lines
---
 arch/arm64/kernel/perf_event.c | 50 +++++++++++++++++++---------------
 1 file changed, 28 insertions(+), 22 deletions(-)
Quoted 1 line

What stable tree(s) do you want this applied to?

I guess that'd be 4.14 and previous stables if the patch actually applies.

Thanks,

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
#5 lore raw

Re: [PATCH] arm64: perf: Disable PMU while processing counter overflows

Quoted 37 lines · Marc Zyngier

CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.

Quoted 29 lines · Greg Kroah-Hartman
Quoted 26 lines · Aman Priyadarshi

From: Suzuki K Poulose <[email protected]>

[ Upstream commit 3cce50dfec4a5b0414c974190940f47dd32c6dee ]

The arm64 PMU updates the event counters and reprograms the counters in the overflow IRQ handler without disabling the PMU. This could potentially cause skews in for group counters, where the overflowed counters may potentially loose some event counts, while they are reprogrammed. To prevent this, disable the PMU while we process the counter overflows and enable it right back when we are done.

This patch also moves the PMU stop/start routines to avoid a forward declaration.

---
 arch/arm64/kernel/perf_event.c | 50 +++++++++++++++++++-------------
--
 1 file changed, 28 insertions(+), 22 deletions(-)
Quoted 5 lines
Quoted 1 line

What stable tree(s) do you want this applied to?

I guess that'd be 4.14 and previous stables if the patch actually applies.

Correct. I have tested the patch on 4.14.y, can confirm that it applies cleanly on 4.9.y as well.

Thanks,
Aman Priyadarshi

Amazon Development Center Germany GmbH
Krausenstr. 38
10117 Berlin
Geschaeftsfuehrung: Christian Schlaeger, Jonathan Weiss Eingetragen am Amtsgericht Charlottenburg unter HRB 149173 B
Sitz: Berlin
Ust-ID: DE 289 237 879

#6 lore raw

Re: [PATCH] arm64: perf: Disable PMU while processing counter overflows

Quoted 38 lines · Aman Priyadarshi
Quoted 37 lines · Marc Zyngier

CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.

Quoted 29 lines · Greg Kroah-Hartman
Quoted 26 lines · Aman Priyadarshi

From: Suzuki K Poulose <[email protected]>

[ Upstream commit 3cce50dfec4a5b0414c974190940f47dd32c6dee ]

The arm64 PMU updates the event counters and reprograms the counters in the overflow IRQ handler without disabling the PMU. This could potentially cause skews in for group counters, where the overflowed counters may potentially loose some event counts, while they are reprogrammed. To prevent this, disable the PMU while we process the counter overflows and enable it right back when we are done.

This patch also moves the PMU stop/start routines to avoid a forward declaration.

---
 arch/arm64/kernel/perf_event.c | 50 +++++++++++++++++++-------------
--
 1 file changed, 28 insertions(+), 22 deletions(-)
Quoted 8 lines
Quoted 5 lines
Quoted 1 line

What stable tree(s) do you want this applied to?

I guess that'd be 4.14 and previous stables if the patch actually applies.

Correct. I have tested the patch on 4.14.y, can confirm that it applies cleanly on 4.9.y as well.

Now queued up, thanks.

greg k-h