* [PATCH] kvm/x86: Reduce counter period change overhead and delay the effective time
@ 2020-03-17 7:53 Like Xu
2020-03-17 8:00 ` Like Xu
2020-03-17 8:14 ` [PATCH v2] KVM: x86/pmu: " Like Xu
0 siblings, 2 replies; 6+ messages in thread
From: Like Xu @ 2020-03-17 7:53 UTC (permalink / raw)
To: Paolo Bonzini, Jim Mattson, Eric Hankland, Wanpeng Li
Cc: Sean Christopherson, Vitaly Kuznetsov, Joerg Roedel, kvm,
linux-kernel, Like Xu
The cost of perf_event_period() is unstable, and when the guest samples
multiple events, the overhead increases dramatically (5378 ns on E5-2699).
For a non-running counter, the effective time of the new period is when
its corresponding enable bit is enabled. Calling perf_event_period()
in advance is superfluous. For a running counter, it's safe to delay the
effective time until the KVM_REQ_PMU event is handled. If there are
multiple perf_event_period() calls before handling KVM_REQ_PMU,
it helps to reduce the total cost.
Signed-off-by: Like Xu <like.xu@linux.intel.com>
---
arch/x86/kvm/pmu.c | 11 -----------
arch/x86/kvm/pmu.h | 11 +++++++++++
arch/x86/kvm/vmx/pmu_intel.c | 10 ++++------
3 files changed, 15 insertions(+), 17 deletions(-)
diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index d1f8ca57d354..527a8bb85080 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -437,17 +437,6 @@ void kvm_pmu_init(struct kvm_vcpu *vcpu)
kvm_pmu_refresh(vcpu);
}
-static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
-{
- struct kvm_pmu *pmu = pmc_to_pmu(pmc);
-
- if (pmc_is_fixed(pmc))
- return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
- pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
-
- return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
-}
-
/* Release perf_events for vPMCs that have been unused for a full time slice. */
void kvm_pmu_cleanup(struct kvm_vcpu *vcpu)
{
diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
index d7da2b9e0755..cd112e825d2c 100644
--- a/arch/x86/kvm/pmu.h
+++ b/arch/x86/kvm/pmu.h
@@ -138,6 +138,17 @@ static inline u64 get_sample_period(struct kvm_pmc *pmc, u64 counter_value)
return sample_period;
}
+static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
+{
+ struct kvm_pmu *pmu = pmc_to_pmu(pmc);
+
+ if (pmc_is_fixed(pmc))
+ return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
+ pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
+
+ return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
+}
+
void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel);
void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 ctrl, int fixed_idx);
void reprogram_counter(struct kvm_pmu *pmu, int pmc_idx);
diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
index 7c857737b438..4e689273eb05 100644
--- a/arch/x86/kvm/vmx/pmu_intel.c
+++ b/arch/x86/kvm/vmx/pmu_intel.c
@@ -263,15 +263,13 @@ static int intel_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
if (!msr_info->host_initiated)
data = (s64)(s32)data;
pmc->counter += data - pmc_read_counter(pmc);
- if (pmc->perf_event)
- perf_event_period(pmc->perf_event,
- get_sample_period(pmc, data));
+ if (pmc_speculative_in_use(pmc)) {
+ kvm_make_request(KVM_REQ_PMU, pmc->vcpu);
return 0;
} else if ((pmc = get_fixed_pmc(pmu, msr))) {
pmc->counter += data - pmc_read_counter(pmc);
- if (pmc->perf_event)
- perf_event_period(pmc->perf_event,
- get_sample_period(pmc, data));
+ if (pmc_speculative_in_use(pmc)) {
+ kvm_make_request(KVM_REQ_PMU, pmc->vcpu);
return 0;
} else if ((pmc = get_gp_pmc(pmu, msr, MSR_P6_EVNTSEL0))) {
if (data == pmc->eventsel)
--
2.21.1
^ permalink raw reply related [flat|nested] 6+ messages in thread* Re: [PATCH] kvm/x86: Reduce counter period change overhead and delay the effective time
2020-03-17 7:53 [PATCH] kvm/x86: Reduce counter period change overhead and delay the effective time Like Xu
@ 2020-03-17 8:00 ` Like Xu
2020-03-17 8:14 ` [PATCH v2] KVM: x86/pmu: " Like Xu
1 sibling, 0 replies; 6+ messages in thread
From: Like Xu @ 2020-03-17 8:00 UTC (permalink / raw)
To: Paolo Bonzini, Jim Mattson, Eric Hankland, Wanpeng Li
Cc: Sean Christopherson, Vitaly Kuznetsov, Joerg Roedel, kvm, linux-kernel
On 2020/3/17 15:53, Like Xu wrote:
> The cost of perf_event_period() is unstable, and when the guest samples
> multiple events, the overhead increases dramatically (5378 ns on E5-2699).
>
> For a non-running counter, the effective time of the new period is when
> its corresponding enable bit is enabled. Calling perf_event_period()
> in advance is superfluous. For a running counter, it's safe to delay the
> effective time until the KVM_REQ_PMU event is handled. If there are
> multiple perf_event_period() calls before handling KVM_REQ_PMU,
> it helps to reduce the total cost.
>
> Signed-off-by: Like Xu <like.xu@linux.intel.com>
>
> ---
> arch/x86/kvm/pmu.c | 11 -----------
> arch/x86/kvm/pmu.h | 11 +++++++++++
> arch/x86/kvm/vmx/pmu_intel.c | 10 ++++------
> 3 files changed, 15 insertions(+), 17 deletions(-)
>
> diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
> index d1f8ca57d354..527a8bb85080 100644
> --- a/arch/x86/kvm/pmu.c
> +++ b/arch/x86/kvm/pmu.c
> @@ -437,17 +437,6 @@ void kvm_pmu_init(struct kvm_vcpu *vcpu)
> kvm_pmu_refresh(vcpu);
> }
>
> -static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
> -{
> - struct kvm_pmu *pmu = pmc_to_pmu(pmc);
> -
> - if (pmc_is_fixed(pmc))
> - return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
> - pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
> -
> - return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
> -}
> -
> /* Release perf_events for vPMCs that have been unused for a full time slice. */
> void kvm_pmu_cleanup(struct kvm_vcpu *vcpu)
> {
> diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
> index d7da2b9e0755..cd112e825d2c 100644
> --- a/arch/x86/kvm/pmu.h
> +++ b/arch/x86/kvm/pmu.h
> @@ -138,6 +138,17 @@ static inline u64 get_sample_period(struct kvm_pmc *pmc, u64 counter_value)
> return sample_period;
> }
>
> +static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
> +{
> + struct kvm_pmu *pmu = pmc_to_pmu(pmc);
> +
> + if (pmc_is_fixed(pmc))
> + return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
> + pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
> +
> + return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
> +}
> +
> void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel);
> void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 ctrl, int fixed_idx);
> void reprogram_counter(struct kvm_pmu *pmu, int pmc_idx);
> diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
> index 7c857737b438..4e689273eb05 100644
> --- a/arch/x86/kvm/vmx/pmu_intel.c
> +++ b/arch/x86/kvm/vmx/pmu_intel.c
> @@ -263,15 +263,13 @@ static int intel_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> if (!msr_info->host_initiated)
> data = (s64)(s32)data;
> pmc->counter += data - pmc_read_counter(pmc);
> - if (pmc->perf_event)
> - perf_event_period(pmc->perf_event,
> - get_sample_period(pmc, data));
> + if (pmc_speculative_in_use(pmc)) {
Oops, the "{" is a shameful mistake.
> + kvm_make_request(KVM_REQ_PMU, pmc->vcpu);
> return 0;
> } else if ((pmc = get_fixed_pmc(pmu, msr))) {
> pmc->counter += data - pmc_read_counter(pmc);
> - if (pmc->perf_event)
> - perf_event_period(pmc->perf_event,
> - get_sample_period(pmc, data));
> + if (pmc_speculative_in_use(pmc)) {
> + kvm_make_request(KVM_REQ_PMU, pmc->vcpu);
> return 0;
> } else if ((pmc = get_gp_pmc(pmu, msr, MSR_P6_EVNTSEL0))) {
> if (data == pmc->eventsel)
>
^ permalink raw reply [flat|nested] 6+ messages in thread* [PATCH v2] KVM: x86/pmu: Reduce counter period change overhead and delay the effective time
2020-03-17 7:53 [PATCH] kvm/x86: Reduce counter period change overhead and delay the effective time Like Xu
2020-03-17 8:00 ` Like Xu
@ 2020-03-17 8:14 ` Like Xu
2020-03-26 12:47 ` Like Xu
1 sibling, 1 reply; 6+ messages in thread
From: Like Xu @ 2020-03-17 8:14 UTC (permalink / raw)
To: pbonzini, like.xu
Cc: ehankland, jmattson, joro, kvm, linux-kernel,
sean.j.christopherson, vkuznets, wanpengli
The cost of perf_event_period() is unstable, and when the guest samples
multiple events, the overhead increases dramatically (5378 ns on E5-2699).
For a non-running counter, the effective time of the new period is when
its corresponding enable bit is enabled. Calling perf_event_period()
in advance is superfluous. For a running counter, it's safe to delay the
effective time until the KVM_REQ_PMU event is handled. If there are
multiple perf_event_period() calls before handling KVM_REQ_PMU,
it helps to reduce the total cost.
Signed-off-by: Like Xu <like.xu@linux.intel.com>
---
arch/x86/kvm/pmu.c | 11 -----------
arch/x86/kvm/pmu.h | 11 +++++++++++
arch/x86/kvm/vmx/pmu_intel.c | 10 ++++------
3 files changed, 15 insertions(+), 17 deletions(-)
diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index d1f8ca57d354..527a8bb85080 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -437,17 +437,6 @@ void kvm_pmu_init(struct kvm_vcpu *vcpu)
kvm_pmu_refresh(vcpu);
}
-static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
-{
- struct kvm_pmu *pmu = pmc_to_pmu(pmc);
-
- if (pmc_is_fixed(pmc))
- return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
- pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
-
- return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
-}
-
/* Release perf_events for vPMCs that have been unused for a full time slice. */
void kvm_pmu_cleanup(struct kvm_vcpu *vcpu)
{
diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
index d7da2b9e0755..cd112e825d2c 100644
--- a/arch/x86/kvm/pmu.h
+++ b/arch/x86/kvm/pmu.h
@@ -138,6 +138,17 @@ static inline u64 get_sample_period(struct kvm_pmc *pmc, u64 counter_value)
return sample_period;
}
+static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
+{
+ struct kvm_pmu *pmu = pmc_to_pmu(pmc);
+
+ if (pmc_is_fixed(pmc))
+ return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
+ pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
+
+ return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
+}
+
void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel);
void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 ctrl, int fixed_idx);
void reprogram_counter(struct kvm_pmu *pmu, int pmc_idx);
diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
index 7c857737b438..20f654a0c09b 100644
--- a/arch/x86/kvm/vmx/pmu_intel.c
+++ b/arch/x86/kvm/vmx/pmu_intel.c
@@ -263,15 +263,13 @@ static int intel_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
if (!msr_info->host_initiated)
data = (s64)(s32)data;
pmc->counter += data - pmc_read_counter(pmc);
- if (pmc->perf_event)
- perf_event_period(pmc->perf_event,
- get_sample_period(pmc, data));
+ if (pmc_speculative_in_use(pmc))
+ kvm_make_request(KVM_REQ_PMU, vcpu);
return 0;
} else if ((pmc = get_fixed_pmc(pmu, msr))) {
pmc->counter += data - pmc_read_counter(pmc);
- if (pmc->perf_event)
- perf_event_period(pmc->perf_event,
- get_sample_period(pmc, data));
+ if (pmc_speculative_in_use(pmc))
+ kvm_make_request(KVM_REQ_PMU, vcpu);
return 0;
} else if ((pmc = get_gp_pmc(pmu, msr, MSR_P6_EVNTSEL0))) {
if (data == pmc->eventsel)
--
2.21.1
^ permalink raw reply related [flat|nested] 6+ messages in thread* Re: [PATCH v2] KVM: x86/pmu: Reduce counter period change overhead and delay the effective time
2020-03-17 8:14 ` [PATCH v2] KVM: x86/pmu: " Like Xu
@ 2020-03-26 12:47 ` Like Xu
2020-04-08 14:04 ` Like Xu
0 siblings, 1 reply; 6+ messages in thread
From: Like Xu @ 2020-03-26 12:47 UTC (permalink / raw)
To: pbonzini
Cc: ehankland, jmattson, joro, kvm, linux-kernel,
sean.j.christopherson, vkuznets, wanpengli
Anyone to help review this change?
Thanks,
Like Xu
On 2020/3/17 16:14, Like Xu wrote:
> The cost of perf_event_period() is unstable, and when the guest samples
> multiple events, the overhead increases dramatically (5378 ns on E5-2699).
>
> For a non-running counter, the effective time of the new period is when
> its corresponding enable bit is enabled. Calling perf_event_period()
> in advance is superfluous. For a running counter, it's safe to delay the
> effective time until the KVM_REQ_PMU event is handled. If there are
> multiple perf_event_period() calls before handling KVM_REQ_PMU,
> it helps to reduce the total cost.
>
> Signed-off-by: Like Xu <like.xu@linux.intel.com>
> ---
> arch/x86/kvm/pmu.c | 11 -----------
> arch/x86/kvm/pmu.h | 11 +++++++++++
> arch/x86/kvm/vmx/pmu_intel.c | 10 ++++------
> 3 files changed, 15 insertions(+), 17 deletions(-)
>
> diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
> index d1f8ca57d354..527a8bb85080 100644
> --- a/arch/x86/kvm/pmu.c
> +++ b/arch/x86/kvm/pmu.c
> @@ -437,17 +437,6 @@ void kvm_pmu_init(struct kvm_vcpu *vcpu)
> kvm_pmu_refresh(vcpu);
> }
>
> -static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
> -{
> - struct kvm_pmu *pmu = pmc_to_pmu(pmc);
> -
> - if (pmc_is_fixed(pmc))
> - return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
> - pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
> -
> - return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
> -}
> -
> /* Release perf_events for vPMCs that have been unused for a full time slice. */
> void kvm_pmu_cleanup(struct kvm_vcpu *vcpu)
> {
> diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
> index d7da2b9e0755..cd112e825d2c 100644
> --- a/arch/x86/kvm/pmu.h
> +++ b/arch/x86/kvm/pmu.h
> @@ -138,6 +138,17 @@ static inline u64 get_sample_period(struct kvm_pmc *pmc, u64 counter_value)
> return sample_period;
> }
>
> +static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
> +{
> + struct kvm_pmu *pmu = pmc_to_pmu(pmc);
> +
> + if (pmc_is_fixed(pmc))
> + return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
> + pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
> +
> + return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
> +}
> +
> void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel);
> void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 ctrl, int fixed_idx);
> void reprogram_counter(struct kvm_pmu *pmu, int pmc_idx);
> diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
> index 7c857737b438..20f654a0c09b 100644
> --- a/arch/x86/kvm/vmx/pmu_intel.c
> +++ b/arch/x86/kvm/vmx/pmu_intel.c
> @@ -263,15 +263,13 @@ static int intel_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> if (!msr_info->host_initiated)
> data = (s64)(s32)data;
> pmc->counter += data - pmc_read_counter(pmc);
> - if (pmc->perf_event)
> - perf_event_period(pmc->perf_event,
> - get_sample_period(pmc, data));
> + if (pmc_speculative_in_use(pmc))
> + kvm_make_request(KVM_REQ_PMU, vcpu);
> return 0;
> } else if ((pmc = get_fixed_pmc(pmu, msr))) {
> pmc->counter += data - pmc_read_counter(pmc);
> - if (pmc->perf_event)
> - perf_event_period(pmc->perf_event,
> - get_sample_period(pmc, data));
> + if (pmc_speculative_in_use(pmc))
> + kvm_make_request(KVM_REQ_PMU, vcpu);
> return 0;
> } else if ((pmc = get_gp_pmc(pmu, msr, MSR_P6_EVNTSEL0))) {
> if (data == pmc->eventsel)
>
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [PATCH v2] KVM: x86/pmu: Reduce counter period change overhead and delay the effective time
2020-03-26 12:47 ` Like Xu
@ 2020-04-08 14:04 ` Like Xu
2020-04-16 14:41 ` Like Xu
0 siblings, 1 reply; 6+ messages in thread
From: Like Xu @ 2020-04-08 14:04 UTC (permalink / raw)
To: pbonzini; +Cc: ehankland, jmattson, kvm, linux-kernel, wanpengli
Hi Paolo,
Could you please take a look at this patch?
If there is anything needs to be improved, please let me know.
Thanks,
Like Xu
On 2020/3/26 20:47, Like Xu wrote:
> Anyone to help review this change?
>
> Thanks,
> Like Xu
>
> On 2020/3/17 16:14, Like Xu wrote:
>> The cost of perf_event_period() is unstable, and when the guest samples
>> multiple events, the overhead increases dramatically (5378 ns on E5-2699).
>>
>> For a non-running counter, the effective time of the new period is when
>> its corresponding enable bit is enabled. Calling perf_event_period()
>> in advance is superfluous. For a running counter, it's safe to delay the
>> effective time until the KVM_REQ_PMU event is handled. If there are
>> multiple perf_event_period() calls before handling KVM_REQ_PMU,
>> it helps to reduce the total cost.
>>
>> Signed-off-by: Like Xu <like.xu@linux.intel.com>
>> ---
>> arch/x86/kvm/pmu.c | 11 -----------
>> arch/x86/kvm/pmu.h | 11 +++++++++++
>> arch/x86/kvm/vmx/pmu_intel.c | 10 ++++------
>> 3 files changed, 15 insertions(+), 17 deletions(-)
>>
>> diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
>> index d1f8ca57d354..527a8bb85080 100644
>> --- a/arch/x86/kvm/pmu.c
>> +++ b/arch/x86/kvm/pmu.c
>> @@ -437,17 +437,6 @@ void kvm_pmu_init(struct kvm_vcpu *vcpu)
>> kvm_pmu_refresh(vcpu);
>> }
>> -static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
>> -{
>> - struct kvm_pmu *pmu = pmc_to_pmu(pmc);
>> -
>> - if (pmc_is_fixed(pmc))
>> - return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
>> - pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
>> -
>> - return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
>> -}
>> -
>> /* Release perf_events for vPMCs that have been unused for a full time
>> slice. */
>> void kvm_pmu_cleanup(struct kvm_vcpu *vcpu)
>> {
>> diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
>> index d7da2b9e0755..cd112e825d2c 100644
>> --- a/arch/x86/kvm/pmu.h
>> +++ b/arch/x86/kvm/pmu.h
>> @@ -138,6 +138,17 @@ static inline u64 get_sample_period(struct kvm_pmc
>> *pmc, u64 counter_value)
>> return sample_period;
>> }
>> +static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
>> +{
>> + struct kvm_pmu *pmu = pmc_to_pmu(pmc);
>> +
>> + if (pmc_is_fixed(pmc))
>> + return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
>> + pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
>> +
>> + return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
>> +}
>> +
>> void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel);
>> void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 ctrl, int fixed_idx);
>> void reprogram_counter(struct kvm_pmu *pmu, int pmc_idx);
>> diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
>> index 7c857737b438..20f654a0c09b 100644
>> --- a/arch/x86/kvm/vmx/pmu_intel.c
>> +++ b/arch/x86/kvm/vmx/pmu_intel.c
>> @@ -263,15 +263,13 @@ static int intel_pmu_set_msr(struct kvm_vcpu *vcpu,
>> struct msr_data *msr_info)
>> if (!msr_info->host_initiated)
>> data = (s64)(s32)data;
>> pmc->counter += data - pmc_read_counter(pmc);
>> - if (pmc->perf_event)
>> - perf_event_period(pmc->perf_event,
>> - get_sample_period(pmc, data));
>> + if (pmc_speculative_in_use(pmc))
>> + kvm_make_request(KVM_REQ_PMU, vcpu);
>> return 0;
>> } else if ((pmc = get_fixed_pmc(pmu, msr))) {
>> pmc->counter += data - pmc_read_counter(pmc);
>> - if (pmc->perf_event)
>> - perf_event_period(pmc->perf_event,
>> - get_sample_period(pmc, data));
>> + if (pmc_speculative_in_use(pmc))
>> + kvm_make_request(KVM_REQ_PMU, vcpu);
>> return 0;
>> } else if ((pmc = get_gp_pmc(pmu, msr, MSR_P6_EVNTSEL0))) {
>> if (data == pmc->eventsel)
>>
>
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [PATCH v2] KVM: x86/pmu: Reduce counter period change overhead and delay the effective time
2020-04-08 14:04 ` Like Xu
@ 2020-04-16 14:41 ` Like Xu
0 siblings, 0 replies; 6+ messages in thread
From: Like Xu @ 2020-04-16 14:41 UTC (permalink / raw)
To: pbonzini; +Cc: ehankland, jmattson, kvm, linux-kernel, wanpengli
Ping.
On 2020/4/8 22:04, Like Xu wrote:
> Hi Paolo,
>
> Could you please take a look at this patch?
> If there is anything needs to be improved, please let me know.
>
> Thanks,
> Like Xu
>
> On 2020/3/26 20:47, Like Xu wrote:
>> Anyone to help review this change?
>>
>> Thanks,
>> Like Xu
>>
>> On 2020/3/17 16:14, Like Xu wrote:
>>> The cost of perf_event_period() is unstable, and when the guest samples
>>> multiple events, the overhead increases dramatically (5378 ns on E5-2699).
>>>
>>> For a non-running counter, the effective time of the new period is when
>>> its corresponding enable bit is enabled. Calling perf_event_period()
>>> in advance is superfluous. For a running counter, it's safe to delay the
>>> effective time until the KVM_REQ_PMU event is handled. If there are
>>> multiple perf_event_period() calls before handling KVM_REQ_PMU,
>>> it helps to reduce the total cost.
>>>
>>> Signed-off-by: Like Xu <like.xu@linux.intel.com>
>>> ---
>>> arch/x86/kvm/pmu.c | 11 -----------
>>> arch/x86/kvm/pmu.h | 11 +++++++++++
>>> arch/x86/kvm/vmx/pmu_intel.c | 10 ++++------
>>> 3 files changed, 15 insertions(+), 17 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
>>> index d1f8ca57d354..527a8bb85080 100644
>>> --- a/arch/x86/kvm/pmu.c
>>> +++ b/arch/x86/kvm/pmu.c
>>> @@ -437,17 +437,6 @@ void kvm_pmu_init(struct kvm_vcpu *vcpu)
>>> kvm_pmu_refresh(vcpu);
>>> }
>>> -static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
>>> -{
>>> - struct kvm_pmu *pmu = pmc_to_pmu(pmc);
>>> -
>>> - if (pmc_is_fixed(pmc))
>>> - return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
>>> - pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
>>> -
>>> - return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
>>> -}
>>> -
>>> /* Release perf_events for vPMCs that have been unused for a full
>>> time slice. */
>>> void kvm_pmu_cleanup(struct kvm_vcpu *vcpu)
>>> {
>>> diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
>>> index d7da2b9e0755..cd112e825d2c 100644
>>> --- a/arch/x86/kvm/pmu.h
>>> +++ b/arch/x86/kvm/pmu.h
>>> @@ -138,6 +138,17 @@ static inline u64 get_sample_period(struct kvm_pmc
>>> *pmc, u64 counter_value)
>>> return sample_period;
>>> }
>>> +static inline bool pmc_speculative_in_use(struct kvm_pmc *pmc)
>>> +{
>>> + struct kvm_pmu *pmu = pmc_to_pmu(pmc);
>>> +
>>> + if (pmc_is_fixed(pmc))
>>> + return fixed_ctrl_field(pmu->fixed_ctr_ctrl,
>>> + pmc->idx - INTEL_PMC_IDX_FIXED) & 0x3;
>>> +
>>> + return pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE;
>>> +}
>>> +
>>> void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel);
>>> void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 ctrl, int
>>> fixed_idx);
>>> void reprogram_counter(struct kvm_pmu *pmu, int pmc_idx);
>>> diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c
>>> index 7c857737b438..20f654a0c09b 100644
>>> --- a/arch/x86/kvm/vmx/pmu_intel.c
>>> +++ b/arch/x86/kvm/vmx/pmu_intel.c
>>> @@ -263,15 +263,13 @@ static int intel_pmu_set_msr(struct kvm_vcpu
>>> *vcpu, struct msr_data *msr_info)
>>> if (!msr_info->host_initiated)
>>> data = (s64)(s32)data;
>>> pmc->counter += data - pmc_read_counter(pmc);
>>> - if (pmc->perf_event)
>>> - perf_event_period(pmc->perf_event,
>>> - get_sample_period(pmc, data));
>>> + if (pmc_speculative_in_use(pmc))
>>> + kvm_make_request(KVM_REQ_PMU, vcpu);
>>> return 0;
>>> } else if ((pmc = get_fixed_pmc(pmu, msr))) {
>>> pmc->counter += data - pmc_read_counter(pmc);
>>> - if (pmc->perf_event)
>>> - perf_event_period(pmc->perf_event,
>>> - get_sample_period(pmc, data));
>>> + if (pmc_speculative_in_use(pmc))
>>> + kvm_make_request(KVM_REQ_PMU, vcpu);
>>> return 0;
>>> } else if ((pmc = get_gp_pmc(pmu, msr, MSR_P6_EVNTSEL0))) {
>>> if (data == pmc->eventsel)
>>>
>>
>
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2020-04-16 14:42 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-03-17 7:53 [PATCH] kvm/x86: Reduce counter period change overhead and delay the effective time Like Xu
2020-03-17 8:00 ` Like Xu
2020-03-17 8:14 ` [PATCH v2] KVM: x86/pmu: " Like Xu
2020-03-26 12:47 ` Like Xu
2020-04-08 14:04 ` Like Xu
2020-04-16 14:41 ` Like Xu
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).