Skip to contentSkip to conversation

KVM: arm64: Assorted PMU emulation fixes

PATCH Superseded Marc Zyngier 2 participants 12 messages Oct 8, 2019 – Oct 11, 2019

Subsystems (5)
KVM perf events arm64 KVM arm64 +1 more subsystems

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
author maintainer
#1 lore raw

PATCH v2 1/5 KVM: arm64: pmu: Fix cycle counter truncation

In reply to Marc Zyngier

When a counter is disabled, its value is sampled before the event is being disabled, and the value written back in the shadow register.

In that process, the value gets truncated to 32bit, which is adequate for any counter but the cycle counter (defined as a 64bit counter).

This obviously results in a corrupted counter, and things like "perf record -e cycles" not working at all when run in a guest... A similar, but less critical bug exists in kvm_pmu_get_counter_value.

Make the truncation conditional on the counter not being the cycle counter, which results in a minor code reorganisation.

Fixes 80f393a23be6 ("KVM: arm/arm64: Support chained PMU counters")
---
 virt/kvm/arm/pmu.c | 22 ++++++++++++----------
 1 file changed, 12 insertions(+), 10 deletions(-)
virt/kvm/arm/pmu.c +12 -10
index 362a01886bab..c30c3a74fc7f 100644
@@ -146,8 +146,7 @@ u64 kvm_pmu_get_counter_value(struct kvm_vcpu *vcpu, u64 select_idx)
 	if (kvm_pmu_pmc_is_chained(pmc) &&
 	    kvm_pmu_idx_is_high_counter(select_idx))
 		counter = upper_32_bits(counter);
-
-	else if (!kvm_pmu_idx_is_64bit(vcpu, select_idx))
+	else if (select_idx != ARMV8_PMU_CYCLE_IDX)
 		counter = lower_32_bits(counter);
 
 	return counter;
@@ -193,7 +192,7 @@ static void kvm_pmu_release_perf_event(struct kvm_pmc *pmc)
  */
 static void kvm_pmu_stop_counter(struct kvm_vcpu *vcpu, struct kvm_pmc *pmc)
 {
-	u64 counter, reg;
+	u64 counter, reg, val;
 
 	pmc = kvm_pmu_get_canonical_pmc(pmc);
 	if (!pmc->perf_event)
@@ -201,16 +200,19 @@ static void kvm_pmu_stop_counter(struct kvm_vcpu *vcpu, struct kvm_pmc *pmc)
 
 	counter = kvm_pmu_get_pair_counter_value(vcpu, pmc);
 
-	if (kvm_pmu_pmc_is_chained(pmc)) {
-		reg = PMEVCNTR0_EL0 + pmc->idx;
-		__vcpu_sys_reg(vcpu, reg) = lower_32_bits(counter);
-		__vcpu_sys_reg(vcpu, reg + 1) = upper_32_bits(counter);
+	if (pmc->idx == ARMV8_PMU_CYCLE_IDX) {
+		reg = PMCCNTR_EL0;
+		val = counter;
 	} else {
-		reg = (pmc->idx == ARMV8_PMU_CYCLE_IDX)
-		       ? PMCCNTR_EL0 : PMEVCNTR0_EL0 + pmc->idx;
-		__vcpu_sys_reg(vcpu, reg) = lower_32_bits(counter);
+		reg = PMEVCNTR0_EL0 + pmc->idx;
+		val = lower_32_bits(counter);
 	}
 
+	__vcpu_sys_reg(vcpu, reg) = val;
+
+	if (kvm_pmu_pmc_is_chained(pmc))
+		__vcpu_sys_reg(vcpu, reg + 1) = upper_32_bits(counter);
+
 	kvm_pmu_release_perf_event(pmc);
 }
 
Signature · 1 line
2.20.1

Marc Zyngier

author maintainer
#0 lore raw

v2 cover KVM: arm64: Assorted PMU emulation fixes

I recently came across a number of PMU emulation bugs, all which can result in unexpected behaviours in an unsuspecting guest. The first two patches already have been discussed on the list, but I'm including them here as part of a slightly longer series.

The third patch is new as of v2, and fixes a bug preventing chained events from ever being used.

The fourth patch is also new as of v2, and is an arm64 PMU change for which I clearly don't know what I'm doing. I'd appreciate some guidance from Will or Mark.

The last patch fixes an issue that has been here from day one, where we confuse architectural overflow of a counter and perf sampling period, and uses patch #4 to fix the issue.

I'l planning to send patches 1 through to 3 as fixes shortly, but I expect the last two patches to require more discussions.

Marc Zyngier (5):
KVM: arm64: pmu: Fix cycle counter truncation
arm64: KVM: Handle PMCR_EL0.LC as RES1 on pure AArch64 systems
KVM: arm64: pmu: Set the CHAINED attribute before creating the
in-kernel event
arm64: perf: Add reload-on-overflow capability
KVM: arm64: pmu: Reset sample period on overflow handling

 arch/arm64/include/asm/perf_event.h |  4 +++
 arch/arm64/kernel/perf_event.c      |  8 ++++-
 arch/arm64/kvm/sys_regs.c           |  4 +++
 virt/kvm/arm/pmu.c                  | 45 +++++++++++++++++++----------
 4 files changed, 45 insertions(+), 16 deletions(-)
Signature · 1 line
2.20.1
author maintainer
#5 lore raw

PATCH v2 4/5 arm64: perf: Add reload-on-overflow capability

As KVM uses perf as a way to emulate an ARMv8 PMU, it needs to be able to change the sample period as part of the overflow handling (once an overflow has taken place, the following overflow point is the overflow of the virtual counter).

Deleting and recreating the in-kernel event is difficult, as we're in interrupt context. Instead, we can teach the PMU driver a new trick, which is to stop the event before the overflow handling, and reprogram it once it has been handled. This would give KVM the opportunity to adjust the next sample period. This feature is gated on a new flag that can get set by KVM in a subsequent patch.

Whilst we're at it, move the CHAINED flag from the KVM emulation to the perf_event.h file and adjust the PMU code accordingly.

---
 arch/arm64/include/asm/perf_event.h | 4 ++++
 arch/arm64/kernel/perf_event.c      | 8 +++++++-
 virt/kvm/arm/pmu.c                  | 4 +---
 3 files changed, 12 insertions(+), 4 deletions(-)
arch/arm64/include/asm/perf_event.h +4 -0
index 2bdbc79bbd01..8b6b38f2db8e 100644
@@ -223,4 +223,8 @@ extern unsigned long perf_misc_flags(struct pt_regs *regs);
 	(regs)->pstate = PSR_MODE_EL1h;	\
 }
 
+/* Flags used by KVM, among others */
+#define PERF_ATTR_CFG1_CHAINED_EVENT	(1U << 0)
+#define PERF_ATTR_CFG1_RELOAD_EVENT	(1U << 1)
+
 #endif
arch/arm64/kernel/perf_event.c +7 -1
index a0b4f1bca491..98907c9e5508 100644
@@ -322,7 +322,7 @@ PMU_FORMAT_ATTR(long, "config1:0");
 
 static inline bool armv8pmu_event_is_64bit(struct perf_event *event)
 {
-	return event->attr.config1 & 0x1;
+	return event->attr.config1 & PERF_ATTR_CFG1_CHAINED_EVENT;
 }
 
 static struct attribute *armv8_pmuv3_format_attrs[] = {
@@ -736,8 +736,14 @@ static irqreturn_t armv8pmu_handle_irq(struct arm_pmu *cpu_pmu)
 		if (!armpmu_event_set_period(event))
 			continue;
 
+		if (event->attr.config1 & PERF_ATTR_CFG1_RELOAD_EVENT)
+			cpu_pmu->pmu.stop(event, PERF_EF_RELOAD);
+
 		if (perf_event_overflow(event, &data, regs))
 			cpu_pmu->disable(event);
+
+		if (event->attr.config1 & PERF_ATTR_CFG1_RELOAD_EVENT)
+			cpu_pmu->pmu.start(event, PERF_EF_RELOAD);
 	}
 	armv8pmu_start(cpu_pmu);
 
virt/kvm/arm/pmu.c +1 -3
index f291d4ac3519..25a483a04beb 100644
@@ -15,8 +15,6 @@
 
 static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx);
 
-#define PERF_ATTR_CFG1_KVM_PMU_CHAINED 0x1
-
 /**
  * kvm_pmu_idx_is_64bit - determine if select_idx is a 64bit counter
  * @vcpu: The vcpu pointer
@@ -570,7 +568,7 @@ static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx)
 		 */
 		attr.sample_period = (-counter) & GENMASK(63, 0);
 		if (kvm_pmu_counter_is_enabled(vcpu, pmc->idx + 1))
-			attr.config1 |= PERF_ATTR_CFG1_KVM_PMU_CHAINED;
+			attr.config1 |= PERF_ATTR_CFG1_CHAINED_EVENT;
 
 		event = perf_event_create_kernel_counter(&attr, -1, current,
 							 kvm_pmu_perf_overflow,
Signature · 1 line
2.20.1
author maintainer
#8 lore raw

PATCH v2 5/5 KVM: arm64: pmu: Reset sample period on overflow handling

In reply to Marc Zyngier

The PMU emulation code uses the perf event sample period to trigger the overflow detection. This works fine for the *first* overflow handling, but results in a huge number of interrupts on the host, unrelated to the number of interrupts handled in the guest (a x20 factor is pretty common for the cycle counter). On a slow system (such as a SW model), this can result in the guest only making forward progress at a glacial pace.

It turns out that the clue is in the name. The sample period is exactly that: a period. And once the an overflow has occured, the following period should be the full width of the associated counter, instead of whatever the guest had initially programed.

Reset the sample period to the architected value in the overflow handler, which now results in a number of host interrupts that is much closer to the number of interrupts in the guest.

Fixes b02386eb7dac ("arm64: KVM: Add PMU overflow interrupt routing")
---
 virt/kvm/arm/pmu.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
virt/kvm/arm/pmu.c +15 -0
index 25a483a04beb..8b524d74c68a 100644
@@ -442,6 +442,20 @@ static void kvm_pmu_perf_overflow(struct perf_event *perf_event,
 	struct kvm_pmc *pmc = perf_event->overflow_handler_context;
 	struct kvm_vcpu *vcpu = kvm_pmc_to_vcpu(pmc);
 	int idx = pmc->idx;
+	u64 period;
+
+	/*
+	 * Reset the sample period to the architectural limit,
+	 * i.e. the point where the counter overflows.
+	 */
+	period = -(local64_read(&pmc->perf_event->count));
+
+	if (!kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
+		period &= GENMASK(31, 0);
+
+	local64_set(&pmc->perf_event->hw.period_left, 0);
+	pmc->perf_event->attr.sample_period = period;
+	pmc->perf_event->hw.sample_period = period;
 
 	__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= BIT(idx);
 
@@ -557,6 +571,7 @@ static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx)
 	attr.exclude_host = 1; /* Don't count host events */
 	attr.config = (pmc->idx == ARMV8_PMU_CYCLE_IDX) ?
 		ARMV8_PMUV3_PERFCTR_CPU_CYCLES : eventsel;
+	attr.config1 = PERF_ATTR_CFG1_RELOAD_EVENT;
 
 	counter = kvm_pmu_get_pair_counter_value(vcpu, pmc);
 
Signature · 1 line
2.20.1
author maintainer
#3 lore raw

PATCH v2 3/5 KVM: arm64: pmu: Set the CHAINED attribute before creating the in-kernel event

In reply to Marc Zyngier

The current convention for KVM to request a chained event from the host PMU is to set bit[0] in attr.config1 (PERF_ATTR_CFG1_KVM_PMU_CHAINED).

But as it turns out, this bit gets set *after* we create the kernel event that backs our virtual counter, meaning that we never get a 64bit counter.

Moving the setting to an earlier point solves the problem.

Fixes 80f393a23be6 ("KVM: arm/arm64: Support chained PMU counters")
---
 virt/kvm/arm/pmu.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
virt/kvm/arm/pmu.c +3 -3
index c30c3a74fc7f..f291d4ac3519 100644
@@ -569,12 +569,12 @@ static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx)
 		 * high counter.
 		 */
 		attr.sample_period = (-counter) & GENMASK(63, 0);
+		if (kvm_pmu_counter_is_enabled(vcpu, pmc->idx + 1))
+			attr.config1 |= PERF_ATTR_CFG1_KVM_PMU_CHAINED;
+
 		event = perf_event_create_kernel_counter(&attr, -1, current,
 							 kvm_pmu_perf_overflow,
 							 pmc + 1);
-
-		if (kvm_pmu_counter_is_enabled(vcpu, pmc->idx + 1))
-			attr.config1 |= PERF_ATTR_CFG1_KVM_PMU_CHAINED;
 	} else {
 		/* The initial sample period (overflow count) of an event. */
 		if (kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
Signature · 6 lines
2.20.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author maintainer
#2 lore raw

PATCH v2 2/5 arm64: KVM: Handle PMCR_EL0.LC as RES1 on pure AArch64 systems

In reply to Marc Zyngier

Of PMCR_EL0.LC, the ARMv8 ARM says:

"In an AArch64 only implementation, this field is RES 1."

So be it.

Fixes ab9468340d2bc ("arm64: KVM: Add access handler for PMCR register")
---
 arch/arm64/kvm/sys_regs.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/arm64/kvm/sys_regs.c +4 -0
index 2071260a275b..46822afc57e0 100644
@@ -632,6 +632,8 @@ static void reset_pmcr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 	 */
 	val = ((pmcr & ~ARMV8_PMU_PMCR_MASK)
 	       | (ARMV8_PMU_PMCR_MASK & 0xdecafbad)) & (~ARMV8_PMU_PMCR_E);
+	if (!system_supports_32bit_el0())
+		val |= ARMV8_PMU_PMCR_LC;
 	__vcpu_sys_reg(vcpu, r->reg) = val;
 }
 
@@ -682,6 +684,8 @@ static bool access_pmcr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
 		val = __vcpu_sys_reg(vcpu, PMCR_EL0);
 		val &= ~ARMV8_PMU_PMCR_MASK;
 		val |= p->regval & ARMV8_PMU_PMCR_MASK;
+		if (!system_supports_32bit_el0())
+			val |= ARMV8_PMU_PMCR_LC;
 		__vcpu_sys_reg(vcpu, PMCR_EL0) = val;
 		kvm_pmu_handle_pmcr(vcpu, val);
 		kvm_vcpu_pmu_restore_guest(vcpu);
Signature · 6 lines
2.20.1

_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
author maintainer
#6 lore raw

Re: [PATCH v2 4/5] arm64: perf: Add reload-on-overflow capability

In reply to Marc Zyngier
Quoted 50 lines · On Tue, 8 Oct 2019 17:01:27 +0100 Marc Zyngier <[email protected]> wrote:

As KVM uses perf as a way to emulate an ARMv8 PMU, it needs to be able to change the sample period as part of the overflow handling (once an overflow has taken place, the following overflow point is the overflow of the virtual counter).

Deleting and recreating the in-kernel event is difficult, as we're in interrupt context. Instead, we can teach the PMU driver a new trick, which is to stop the event before the overflow handling, and reprogram it once it has been handled. This would give KVM the opportunity to adjust the next sample period. This feature is gated on a new flag that can get set by KVM in a subsequent patch.

Whilst we're at it, move the CHAINED flag from the KVM emulation to the perf_event.h file and adjust the PMU code accordingly.

---
 arch/arm64/include/asm/perf_event.h | 4 ++++
 arch/arm64/kernel/perf_event.c      | 8 +++++++-
 virt/kvm/arm/pmu.c                  | 4 +---
 3 files changed, 12 insertions(+), 4 deletions(-)
arch/arm64/include/asm/perf_event.h +4 -0
index 2bdbc79bbd01..8b6b38f2db8e 100644
@@ -223,4 +223,8 @@ extern unsigned long perf_misc_flags(struct pt_regs *regs);
 	(regs)->pstate = PSR_MODE_EL1h;	\
 }
 
+/* Flags used by KVM, among others */
+#define PERF_ATTR_CFG1_CHAINED_EVENT	(1U << 0)
+#define PERF_ATTR_CFG1_RELOAD_EVENT	(1U << 1)
+
 #endif
arch/arm64/kernel/perf_event.c +1 -1
index a0b4f1bca491..98907c9e5508 100644
@@ -322,7 +322,7 @@ PMU_FORMAT_ATTR(long, "config1:0");
 
 static inline bool armv8pmu_event_is_64bit(struct perf_event *event)
 {
-	return event->attr.config1 & 0x1;
+	return event->attr.config1 & PERF_ATTR_CFG1_CHAINED_EVENT;
 }
 
 static struct attribute *armv8_pmuv3_format_attrs[] = {

arch/arm64/kernel/perf_event.c static irqreturn_t armv8pmu_handle_irq(struct arm_pmu *cpu_pmu) · line 741

Quoted 6 lines
arch/arm64/kernel/perf_event.c +3 -0
@@ -736,8 +736,14 @@ static irqreturn_t armv8pmu_handle_irq(struct arm_pmu *cpu_pmu)
 		if (!armpmu_event_set_period(event))
 			continue;
 
+		if (event->attr.config1 & PERF_ATTR_CFG1_RELOAD_EVENT)
+			cpu_pmu->pmu.stop(event, PERF_EF_RELOAD);
+

Actually, I just realized that there is probably no need for this patch as a standalone change. I can perfectly fold the stop() and start() calls into the last patch, as part of the overflow handler.

The question is still whether that's a good idea or not.

Thanks,

	M.
Quoted 30 lines
arch/arm64/kernel/perf_event.c +3 -0
 		if (perf_event_overflow(event, &data, regs))
 			cpu_pmu->disable(event);
+
+		if (event->attr.config1 & PERF_ATTR_CFG1_RELOAD_EVENT)
+			cpu_pmu->pmu.start(event, PERF_EF_RELOAD);
 	}
 	armv8pmu_start(cpu_pmu);
 
virt/kvm/arm/pmu.c +1 -3
index f291d4ac3519..25a483a04beb 100644
@@ -15,8 +15,6 @@
 
 static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx);
 
-#define PERF_ATTR_CFG1_KVM_PMU_CHAINED 0x1
-
 /**
  * kvm_pmu_idx_is_64bit - determine if select_idx is a 64bit counter
  * @vcpu: The vcpu pointer
@@ -570,7 +568,7 @@ static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx)
 		 */
 		attr.sample_period = (-counter) & GENMASK(63, 0);
 		if (kvm_pmu_counter_is_enabled(vcpu, pmc->idx + 1))
-			attr.config1 |= PERF_ATTR_CFG1_KVM_PMU_CHAINED;
+			attr.config1 |= PERF_ATTR_CFG1_CHAINED_EVENT;
 
 		event = perf_event_create_kernel_counter(&attr, -1, current,
 							 kvm_pmu_perf_overflow,
Signature · 1 line
Jazz is not dead. It just smells funny...
Reviewed-by on v2 3/5
#4 lore raw

Re: [PATCH v2 3/5] KVM: arm64: pmu: Set the CHAINED attribute before creating the in-kernel event

In reply to Marc Zyngier
Quoted 11 lines · Marc Zyngier

The current convention for KVM to request a chained event from the host PMU is to set bit[0] in attr.config1 (PERF_ATTR_CFG1_KVM_PMU_CHAINED).

But as it turns out, this bit gets set *after* we create the kernel event that backs our virtual counter, meaning that we never get a 64bit counter.

Moving the setting to an earlier point solves the problem.

Fixes 80f393a23be6 ("KVM: arm/arm64: Support chained PMU counters")
Quoted 27 lines
---
 virt/kvm/arm/pmu.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
virt/kvm/arm/pmu.c +3 -3
index c30c3a74fc7f..f291d4ac3519 100644
@@ -569,12 +569,12 @@ static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx)
 		 * high counter.
 		 */
 		attr.sample_period = (-counter) & GENMASK(63, 0);
+		if (kvm_pmu_counter_is_enabled(vcpu, pmc->idx + 1))
+			attr.config1 |= PERF_ATTR_CFG1_KVM_PMU_CHAINED;
+
 		event = perf_event_create_kernel_counter(&attr, -1, current,
 							 kvm_pmu_perf_overflow,
 							 pmc + 1);
-
-		if (kvm_pmu_counter_is_enabled(vcpu, pmc->idx + 1))
-			attr.config1 |= PERF_ATTR_CFG1_KVM_PMU_CHAINED;
 	} else {
 		/* The initial sample period (overflow count) of an event. */
 		if (kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
Signature · 1 line
2.20.1
#7 lore raw

Re: [PATCH v2 4/5] arm64: perf: Add reload-on-overflow capability

In reply to Marc Zyngier
Quoted 41 lines · Marc Zyngier

As KVM uses perf as a way to emulate an ARMv8 PMU, it needs to be able to change the sample period as part of the overflow handling (once an overflow has taken place, the following overflow point is the overflow of the virtual counter).

Deleting and recreating the in-kernel event is difficult, as we're in interrupt context. Instead, we can teach the PMU driver a new trick, which is to stop the event before the overflow handling, and reprogram it once it has been handled. This would give KVM the opportunity to adjust the next sample period. This feature is gated on a new flag that can get set by KVM in a subsequent patch.

Whilst we're at it, move the CHAINED flag from the KVM emulation to the perf_event.h file and adjust the PMU code accordingly.

---
 arch/arm64/include/asm/perf_event.h | 4 ++++
 arch/arm64/kernel/perf_event.c      | 8 +++++++-
 virt/kvm/arm/pmu.c                  | 4 +---
 3 files changed, 12 insertions(+), 4 deletions(-)
arch/arm64/include/asm/perf_event.h +4 -0
index 2bdbc79bbd01..8b6b38f2db8e 100644
@@ -223,4 +223,8 @@ extern unsigned long perf_misc_flags(struct pt_regs *regs);
 	(regs)->pstate = PSR_MODE_EL1h;	\
 }
 
+/* Flags used by KVM, among others */
+#define PERF_ATTR_CFG1_CHAINED_EVENT	(1U << 0)
+#define PERF_ATTR_CFG1_RELOAD_EVENT	(1U << 1)
+
 #endif
arch/arm64/kernel/perf_event.c +0 -0
index a0b4f1bca491..98907c9e5508 100644

arch/arm64/kernel/perf_event.c PMU_FORMAT_ATTR(long, "config1:0"); · line 325

Quoted 5 lines
arch/arm64/kernel/perf_event.c +1 -1
@@ -322,7 +322,7 @@ PMU_FORMAT_ATTR(long, "config1:0");
 
 static inline bool armv8pmu_event_is_64bit(struct perf_event *event)
 {
-	return event->attr.config1 & 0x1;
+	return event->attr.config1 & PERF_ATTR_CFG1_CHAINED_EVENT;

I'm pleased to see this be replaced with a define, it helps readers see the link between this and the KVM driver.

arch/arm64/kernel/perf_event.c static irqreturn_t armv8pmu_handle_irq(struct arm_pmu *cpu_pmu) · line 740

Quoted 9 lines
arch/arm64/kernel/perf_event.c +2 -0
 }
 
 static struct attribute *armv8_pmuv3_format_attrs[] = {
@@ -736,8 +736,14 @@ static irqreturn_t armv8pmu_handle_irq(struct arm_pmu *cpu_pmu)
 		if (!armpmu_event_set_period(event))
 			continue;
 
+		if (event->attr.config1 & PERF_ATTR_CFG1_RELOAD_EVENT)
+			cpu_pmu->pmu.stop(event, PERF_EF_RELOAD);

I believe PERF_EF_RELOAD is only intended to be used in the stop calls. I'd suggest that you replace it with PERF_EF_UPDATE instead, this tells the PMU to update the counter with the latest value from the hardware. (Though the ARM PMU driver always does this regardless to the flag anyway).

Thanks,

Andrew Murray

Quoted 34 lines
arch/arm64/kernel/perf_event.c +4 -0
+
 		if (perf_event_overflow(event, &data, regs))
 			cpu_pmu->disable(event);
+
+		if (event->attr.config1 & PERF_ATTR_CFG1_RELOAD_EVENT)
+			cpu_pmu->pmu.start(event, PERF_EF_RELOAD);
 	}
 	armv8pmu_start(cpu_pmu);
 
virt/kvm/arm/pmu.c +1 -3
index f291d4ac3519..25a483a04beb 100644
@@ -15,8 +15,6 @@
 
 static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx);
 
-#define PERF_ATTR_CFG1_KVM_PMU_CHAINED 0x1
-
 /**
  * kvm_pmu_idx_is_64bit - determine if select_idx is a 64bit counter
  * @vcpu: The vcpu pointer
@@ -570,7 +568,7 @@ static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx)
 		 */
 		attr.sample_period = (-counter) & GENMASK(63, 0);
 		if (kvm_pmu_counter_is_enabled(vcpu, pmc->idx + 1))
-			attr.config1 |= PERF_ATTR_CFG1_KVM_PMU_CHAINED;
+			attr.config1 |= PERF_ATTR_CFG1_CHAINED_EVENT;
 
 		event = perf_event_create_kernel_counter(&attr, -1, current,
 							 kvm_pmu_perf_overflow,
Signature · 1 line
2.20.1
#9 lore raw

Re: [PATCH v2 5/5] KVM: arm64: pmu: Reset sample period on overflow handling

In reply to Marc Zyngier
Quoted 35 lines · Marc Zyngier

The PMU emulation code uses the perf event sample period to trigger the overflow detection. This works fine for the *first* overflow handling, but results in a huge number of interrupts on the host, unrelated to the number of interrupts handled in the guest (a x20 factor is pretty common for the cycle counter). On a slow system (such as a SW model), this can result in the guest only making forward progress at a glacial pace.

It turns out that the clue is in the name. The sample period is exactly that: a period. And once the an overflow has occured, the following period should be the full width of the associated counter, instead of whatever the guest had initially programed.

Reset the sample period to the architected value in the overflow handler, which now results in a number of host interrupts that is much closer to the number of interrupts in the guest.

Fixes b02386eb7dac ("arm64: KVM: Add PMU overflow interrupt routing")
---
 virt/kvm/arm/pmu.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
virt/kvm/arm/pmu.c +4 -0
index 25a483a04beb..8b524d74c68a 100644
@@ -442,6 +442,20 @@ static void kvm_pmu_perf_overflow(struct perf_event *perf_event,
 	struct kvm_pmc *pmc = perf_event->overflow_handler_context;
 	struct kvm_vcpu *vcpu = kvm_pmc_to_vcpu(pmc);
 	int idx = pmc->idx;
+	u64 period;
+
+	/*
+	 * Reset the sample period to the architectural limit,

virt/kvm/arm/pmu.c static void kvm_pmu_perf_overflow(struct perf_event *perf_event, · line 458

Quoted 10 lines
virt/kvm/arm/pmu.c +10 -0
+	 * i.e. the point where the counter overflows.
+	 */
+	period = -(local64_read(&pmc->perf_event->count));
+
+	if (!kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
+		period &= GENMASK(31, 0);
+
+	local64_set(&pmc->perf_event->hw.period_left, 0);
+	pmc->perf_event->attr.sample_period = period;
+	pmc->perf_event->hw.sample_period = period;

I believe that above, you are reducing the period by the amount period_left would have been - they cancel each other out.

Given that kvm_pmu_perf_overflow is now always called between a cpu_pmu->pmu.stop and a cpu_pmu->pmu.start, it means armpmu_event_update has been called prior to this function, and armpmu_event_set_period will be called after...

Therefore, I think the above could be reduced to:

+ /*
+ * Reset the sample period to the architectural limit,
+ * i.e. the point where the counter overflows.
+ */
+ u64 period = GENMASK(63, 0);
+ if (!kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
+ period = GENMASK(31, 0);
+
+ pmc->perf_event->attr.sample_period = period;
+ pmc->perf_event->hw.sample_period = period;

This is because armpmu_event_set_period takes into account the overflow and the counter wrapping via the "if (unlikely(left <= 0)) {" block.

Though this code confuses me easily, so I may be talking rubbish.

virt/kvm/arm/pmu.c static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx) · line 574

Quoted 8 lines
virt/kvm/arm/pmu.c +1 -0
 	__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= BIT(idx);
 
@@ -557,6 +571,7 @@ static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx)
 	attr.exclude_host = 1; /* Don't count host events */
 	attr.config = (pmc->idx == ARMV8_PMU_CYCLE_IDX) ?
 		ARMV8_PMUV3_PERFCTR_CPU_CYCLES : eventsel;
+	attr.config1 = PERF_ATTR_CFG1_RELOAD_EVENT;

I'm not sure that this flag, or patch 4 is really needed. As the perf events created by KVM are pinned to the task and exclude_(host,hv) are set - I think the perf event is not active at this point. Therefore if you change the sample period, you can wait until the perf event gets scheduled back in (when you return to the guest) where it's call to pmu.start will result in armpmu_event_set_period being called. In other words the pmu.start and pmu.stop you add in patch 4 is effectively being done for you by perf when the KVM task is switched out.

I'd be interested to see if the following works:

+ WARN_ON(pmc->perf_event->state == PERF_EVENT_STATE_ACTIVE)
+
+ /*
+ * Reset the sample period to the architectural limit,
+ * i.e. the point where the counter overflows.
+ */
+ u64 period = GENMASK(63, 0);
+ if (!kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
+ period = GENMASK(31, 0);
+
+ pmc->perf_event->attr.sample_period = period;
+ pmc->perf_event->hw.sample_period = period;

virt/kvm/arm/pmu.c static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx) · line 577

Quoted 3 lines
virt/kvm/arm/pmu.c +0 -0
 	counter = kvm_pmu_get_pair_counter_value(vcpu, pmc);
 

What about ARM 32 bit support for this?

Thanks,

Andrew Murray

Quoted 3 lines
Signature · 1 line
2.20.1
author maintainer
#10 lore raw

Re: [PATCH v2 5/5] KVM: arm64: pmu: Reset sample period on overflow handling

Quoted 41 lines · On Tue, 8 Oct 2019 23:42:22 +0100 Andrew Murray <[email protected]> wrote:
Quoted 40 lines · Marc Zyngier

The PMU emulation code uses the perf event sample period to trigger the overflow detection. This works fine for the *first* overflow handling, but results in a huge number of interrupts on the host, unrelated to the number of interrupts handled in the guest (a x20 factor is pretty common for the cycle counter). On a slow system (such as a SW model), this can result in the guest only making forward progress at a glacial pace.

It turns out that the clue is in the name. The sample period is exactly that: a period. And once the an overflow has occured, the following period should be the full width of the associated counter, instead of whatever the guest had initially programed.

Reset the sample period to the architected value in the overflow handler, which now results in a number of host interrupts that is much closer to the number of interrupts in the guest.

Fixes b02386eb7dac ("arm64: KVM: Add PMU overflow interrupt routing")
---
 virt/kvm/arm/pmu.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
virt/kvm/arm/pmu.c +9 -0
index 25a483a04beb..8b524d74c68a 100644
@@ -442,6 +442,20 @@ static void kvm_pmu_perf_overflow(struct perf_event *perf_event,
 	struct kvm_pmc *pmc = perf_event->overflow_handler_context;
 	struct kvm_vcpu *vcpu = kvm_pmc_to_vcpu(pmc);
 	int idx = pmc->idx;
+	u64 period;
+
+	/*
+	 * Reset the sample period to the architectural limit,
+	 * i.e. the point where the counter overflows.
+	 */
+	period = -(local64_read(&pmc->perf_event->count));
+
+	if (!kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
Quoted 8 lines

virt/kvm/arm/pmu.c static void kvm_pmu_perf_overflow(struct perf_event *perf_event, · line 458

Quoted 5 lines
virt/kvm/arm/pmu.c +5 -0
+		period &= GENMASK(31, 0);
+
+	local64_set(&pmc->perf_event->hw.period_left, 0);
+	pmc->perf_event->attr.sample_period = period;
+	pmc->perf_event->hw.sample_period = period;  

I believe that above, you are reducing the period by the amount period_left would have been - they cancel each other out.

That's not what I see happening, having put some traces:

kvm_pmu_perf_overflow: count = 308 left = 129
kvm_pmu_perf_overflow: count = 409 left = 47
kvm_pmu_perf_overflow: count = 585 left = 223
kvm_pmu_perf_overflow: count = 775 left = 413
kvm_pmu_perf_overflow: count = 1368 left = 986
kvm_pmu_perf_overflow: count = 2086 left = 1716
kvm_pmu_perf_overflow: count = 958 left = 584
kvm_pmu_perf_overflow: count = 1907 left = 1551
kvm_pmu_perf_overflow: count = 7292 left = 6932

although I've now moved the stop/start calls inside the overflow handler so that I don't have to mess with the PMU backend.

Quoted 18 lines

Given that kvm_pmu_perf_overflow is now always called between a cpu_pmu->pmu.stop and a cpu_pmu->pmu.start, it means armpmu_event_update has been called prior to this function, and armpmu_event_set_period will be called after...

Therefore, I think the above could be reduced to:

virt/kvm/arm/pmu.c +10 -0
+	/*
+	 * Reset the sample period to the architectural limit,
+	 * i.e. the point where the counter overflows.
+	 */
+	u64 period = GENMASK(63, 0);
+	if (!kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
+		period = GENMASK(31, 0);
+
+	pmc->perf_event->attr.sample_period = period;
+	pmc->perf_event->hw.sample_period = period;
Quoted 2 lines

This is because armpmu_event_set_period takes into account the overflow and the counter wrapping via the "if (unlikely(left <= 0)) {" block.

I think that's an oversimplification. As shown above, the counter has moved forward, and there is a delta to be accounted for.

Quoted 1 line

Though this code confuses me easily, so I may be talking rubbish.

Same here! ;-)

Quoted 35 lines

virt/kvm/arm/pmu.c static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx) · line 574

Quoted 9 lines
virt/kvm/arm/pmu.c +1 -0
 	__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= BIT(idx);
 
@@ -557,6 +571,7 @@ static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx)
 	attr.exclude_host = 1; /* Don't count host events */
 	attr.config = (pmc->idx == ARMV8_PMU_CYCLE_IDX) ?
 		ARMV8_PMUV3_PERFCTR_CPU_CYCLES : eventsel;
+	attr.config1 = PERF_ATTR_CFG1_RELOAD_EVENT;  

I'm not sure that this flag, or patch 4 is really needed. As the perf events created by KVM are pinned to the task and exclude_(host,hv) are set - I think the perf event is not active at this point. Therefore if you change the sample period, you can wait until the perf event gets scheduled back in (when you return to the guest) where it's call to pmu.start will result in armpmu_event_set_period being called. In other words the pmu.start and pmu.stop you add in patch 4 is effectively being done for you by perf when the KVM task is switched out.

I'd be interested to see if the following works:

virt/kvm/arm/pmu.c +12 -0
+	WARN_ON(pmc->perf_event->state == PERF_EVENT_STATE_ACTIVE)
+
+	/*
+	 * Reset the sample period to the architectural limit,
+	 * i.e. the point where the counter overflows.
+	 */
+	u64 period = GENMASK(63, 0);
+	if (!kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
+		period = GENMASK(31, 0);
+
+	pmc->perf_event->attr.sample_period = period;
+	pmc->perf_event->hw.sample_period = period;
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

counter = kvm_pmu_get_pair_counter_value(vcpu, pmc);

The warning fires, which is expected: for event to be inactive, you need to have the vcpu being scheduled out. When the PMU interrupt fires, it is bound to preempt the vcpu itself, and the event is of course still active.

Quoted 1 line

What about ARM 32 bit support for this?

What about it? 32bit KVM/arm doesn't support the PMU at all. A 32bit guest on a 64bit host could use the PMU just fine (it is just that 32bit Linux doesn't have a PMUv3 driver -- I had patches for that, but they never made it upstream).

Thanks,

	M.
Signature · 1 line
Jazz is not dead. It just smells funny...
#11 lore raw

Re: [PATCH v2 5/5] KVM: arm64: pmu: Reset sample period on overflow handling

Quoted 55 lines · Marc Zyngier
Quoted 41 lines · On Tue, 8 Oct 2019 23:42:22 +0100 Andrew Murray <[email protected]> wrote:
Quoted 40 lines · Marc Zyngier

The PMU emulation code uses the perf event sample period to trigger the overflow detection. This works fine for the *first* overflow handling, but results in a huge number of interrupts on the host, unrelated to the number of interrupts handled in the guest (a x20 factor is pretty common for the cycle counter). On a slow system (such as a SW model), this can result in the guest only making forward progress at a glacial pace.

It turns out that the clue is in the name. The sample period is exactly that: a period. And once the an overflow has occured, the following period should be the full width of the associated counter, instead of whatever the guest had initially programed.

Reset the sample period to the architected value in the overflow handler, which now results in a number of host interrupts that is much closer to the number of interrupts in the guest.

Fixes b02386eb7dac ("arm64: KVM: Add PMU overflow interrupt routing")
---
 virt/kvm/arm/pmu.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
virt/kvm/arm/pmu.c +9 -0
index 25a483a04beb..8b524d74c68a 100644
@@ -442,6 +442,20 @@ static void kvm_pmu_perf_overflow(struct perf_event *perf_event,
 	struct kvm_pmc *pmc = perf_event->overflow_handler_context;
 	struct kvm_vcpu *vcpu = kvm_pmc_to_vcpu(pmc);
 	int idx = pmc->idx;
+	u64 period;
+
+	/*
+	 * Reset the sample period to the architectural limit,
+	 * i.e. the point where the counter overflows.
+	 */
+	period = -(local64_read(&pmc->perf_event->count));
+
+	if (!kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
Quoted 8 lines

virt/kvm/arm/pmu.c static void kvm_pmu_perf_overflow(struct perf_event *perf_event, · line 458

Quoted 5 lines
virt/kvm/arm/pmu.c +5 -0
+		period &= GENMASK(31, 0);
+
+	local64_set(&pmc->perf_event->hw.period_left, 0);
+	pmc->perf_event->attr.sample_period = period;
+	pmc->perf_event->hw.sample_period = period;  

I believe that above, you are reducing the period by the amount period_left would have been - they cancel each other out.

That's not what I see happening, having put some traces:

Quoted 9 lines

kvm_pmu_perf_overflow: count = 308 left = 129
kvm_pmu_perf_overflow: count = 409 left = 47
kvm_pmu_perf_overflow: count = 585 left = 223
kvm_pmu_perf_overflow: count = 775 left = 413
kvm_pmu_perf_overflow: count = 1368 left = 986
kvm_pmu_perf_overflow: count = 2086 left = 1716
kvm_pmu_perf_overflow: count = 958 left = 584
kvm_pmu_perf_overflow: count = 1907 left = 1551
kvm_pmu_perf_overflow: count = 7292 left = 6932

Indeed.

Quoted 22 lines

although I've now moved the stop/start calls inside the overflow handler so that I don't have to mess with the PMU backend.

Quoted 18 lines

Given that kvm_pmu_perf_overflow is now always called between a cpu_pmu->pmu.stop and a cpu_pmu->pmu.start, it means armpmu_event_update has been called prior to this function, and armpmu_event_set_period will be called after...

Therefore, I think the above could be reduced to:

virt/kvm/arm/pmu.c +10 -0
+	/*
+	 * Reset the sample period to the architectural limit,
+	 * i.e. the point where the counter overflows.
+	 */
+	u64 period = GENMASK(63, 0);
+	if (!kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
+		period = GENMASK(31, 0);
+
+	pmc->perf_event->attr.sample_period = period;
+	pmc->perf_event->hw.sample_period = period;
Quoted 6 lines
Quoted 2 lines

This is because armpmu_event_set_period takes into account the overflow and the counter wrapping via the "if (unlikely(left <= 0)) {" block.

I think that's an oversimplification. As shown above, the counter has moved forward, and there is a delta to be accounted for.

Yeah, I probably need to spend more time understanding this...

Quoted 39 lines
Quoted 1 line

Though this code confuses me easily, so I may be talking rubbish.

Same here! ;-)

Quoted 35 lines

virt/kvm/arm/pmu.c static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx) · line 574

Quoted 9 lines
virt/kvm/arm/pmu.c +1 -0
 	__vcpu_sys_reg(vcpu, PMOVSSET_EL0) |= BIT(idx);
 
@@ -557,6 +571,7 @@ static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx)
 	attr.exclude_host = 1; /* Don't count host events */
 	attr.config = (pmc->idx == ARMV8_PMU_CYCLE_IDX) ?
 		ARMV8_PMUV3_PERFCTR_CPU_CYCLES : eventsel;
+	attr.config1 = PERF_ATTR_CFG1_RELOAD_EVENT;  

I'm not sure that this flag, or patch 4 is really needed. As the perf events created by KVM are pinned to the task and exclude_(host,hv) are set - I think the perf event is not active at this point. Therefore if you change the sample period, you can wait until the perf event gets scheduled back in (when you return to the guest) where it's call to pmu.start will result in armpmu_event_set_period being called. In other words the pmu.start and pmu.stop you add in patch 4 is effectively being done for you by perf when the KVM task is switched out.

I'd be interested to see if the following works:

virt/kvm/arm/pmu.c +12 -0
+	WARN_ON(pmc->perf_event->state == PERF_EVENT_STATE_ACTIVE)
+
+	/*
+	 * Reset the sample period to the architectural limit,
+	 * i.e. the point where the counter overflows.
+	 */
+	u64 period = GENMASK(63, 0);
+	if (!kvm_pmu_idx_is_64bit(vcpu, pmc->idx))
+		period = GENMASK(31, 0);
+
+	pmc->perf_event->attr.sample_period = period;
+	pmc->perf_event->hw.sample_period = period;
Quoted 2 lines
Quoted 7 lines
Quoted 2 lines
Quoted 2 lines

counter = kvm_pmu_get_pair_counter_value(vcpu, pmc);

The warning fires, which is expected: for event to be inactive, you need to have the vcpu being scheduled out. When the PMU interrupt fires, it is bound to preempt the vcpu itself, and the event is of course still active.

That makes sense. That also provides a justification for stopping and starting the PMU.

Quoted 4 lines
Quoted 2 lines

What about ARM 32 bit support for this?

What about it? 32bit KVM/arm doesn't support the PMU at all.

Thanks for the clarification.

Andrew Murray

Quoted 10 lines

A 32bit
guest on a 64bit host could use the PMU just fine (it is just that 32bit Linux doesn't have a PMUv3 driver -- I had patches for that, but they never made it upstream).

Thanks,

	M.
Signature · 1 line
Jazz is not dead. It just smells funny...