Re: [PATCH v2 4/5] KVM: arm/arm64: lazily create perf events on enable

[Suzuki K Poulose <suzuki.poulose@arm.com>]

Hi Andrew,

On 04/02/2019 16:53, Andrew Murray wrote:
> To prevent re-creating perf events everytime the counter registers
> are changed, let's instead lazily create the event when the event
> is first enabled and destroy it when it changes.
> 
> Signed-off-by: Andrew Murray <andrew.murray@arm.com>
> ---
>   virt/kvm/arm/pmu.c | 96 ++++++++++++++++++++++++++++++++++++++----------------
>   1 file changed, 68 insertions(+), 28 deletions(-)
> 
> diff --git a/virt/kvm/arm/pmu.c b/virt/kvm/arm/pmu.c
> index 95d74ec..a64aeb2 100644
> --- a/virt/kvm/arm/pmu.c
> +++ b/virt/kvm/arm/pmu.c
> @@ -24,7 +24,10 @@
>   #include <kvm/arm_pmu.h>
>   #include <kvm/arm_vgic.h>
>   
> +static void kvm_pmu_sync_counter_enable(struct kvm_vcpu *vcpu, u64 select_idx);
>   static void kvm_pmu_create_perf_event(struct kvm_vcpu *vcpu, u64 select_idx);
> +static void kvm_pmu_stop_counter(struct kvm_vcpu *vcpu, struct kvm_pmc *pmc);
> +
>   /**
>    * kvm_pmu_get_counter_value - get PMU counter value
>    * @vcpu: The vcpu pointer
> @@ -59,13 +62,15 @@ u64 kvm_pmu_get_counter_value(struct kvm_vcpu *vcpu, u64 select_idx)
>   void kvm_pmu_set_counter_value(struct kvm_vcpu *vcpu, u64 select_idx, u64 val)
>   {
>   	u64 reg;
> +	struct kvm_pmu *pmu = &vcpu->arch.pmu;
> +	struct kvm_pmc *pmc = &pmu->pmc[select_idx];
>   
>   	reg = (select_idx == ARMV8_PMU_CYCLE_IDX)
>   	      ? PMCCNTR_EL0 : PMEVCNTR0_EL0 + select_idx;
>   	__vcpu_sys_reg(vcpu, reg) += (s64)val - kvm_pmu_get_counter_value(vcpu, select_idx);
>   
> -	/* Recreate the perf event to reflect the updated sample_period */
> -	kvm_pmu_create_perf_event(vcpu, select_idx);
> +	kvm_pmu_stop_counter(vcpu, pmc);
> +	kvm_pmu_sync_counter_enable(vcpu, select_idx);
>   }
>   
>   /**
> @@ -83,6 +88,7 @@ static void kvm_pmu_release_perf_event(struct kvm_pmc *pmc)
>   
>   /**
>    * kvm_pmu_stop_counter - stop PMU counter
> + * @vcpu: The vcpu pointer
>    * @pmc: The PMU counter pointer
>    *
>    * If this counter has been configured to monitor some event, release it here.
> @@ -143,6 +149,24 @@ u64 kvm_pmu_valid_counter_mask(struct kvm_vcpu *vcpu)
>   }
>   
>   /**
> + * kvm_pmu_enable_counter - create/enable a counter
> + * @vcpu: The vcpu pointer
> + * @select_idx: The counter index
> + */
> +static void kvm_pmu_enable_counter(struct kvm_vcpu *vcpu, u64 select_idx)
> +{
> +	struct kvm_pmu *pmu = &vcpu->arch.pmu;
> +	struct kvm_pmc *pmc = &pmu->pmc[select_idx];
> +
> +	if (!pmc->perf_event)
> +		kvm_pmu_create_perf_event(vcpu, select_idx);
> +
> +	perf_event_enable(pmc->perf_event);
> +	if (pmc->perf_event->state != PERF_EVENT_STATE_ACTIVE)
> +		kvm_debug("failed to enable perf event\n");
> +}
> +
> +/**
>    * kvm_pmu_enable_counter_mask - enable selected PMU counters
>    * @vcpu: The vcpu pointer
>    * @val: the value guest writes to PMCNTENSET register
> @@ -152,8 +176,6 @@ u64 kvm_pmu_valid_counter_mask(struct kvm_vcpu *vcpu)
>   void kvm_pmu_enable_counter_mask(struct kvm_vcpu *vcpu, u64 val)
>   {
>   	int i;
> -	struct kvm_pmu *pmu = &vcpu->arch.pmu;
> -	struct kvm_pmc *pmc;
>   
>   	if (!(__vcpu_sys_reg(vcpu, PMCR_EL0) & ARMV8_PMU_PMCR_E) || !val)
>   		return;
> @@ -162,16 +184,39 @@ void kvm_pmu_enable_counter_mask(struct kvm_vcpu *vcpu, u64 val)
>   		if (!(val & BIT(i)))
>   			continue;
>   
> -		pmc = &pmu->pmc[i];
> -		if (pmc->perf_event) {
> -			perf_event_enable(pmc->perf_event);
> -			if (pmc->perf_event->state != PERF_EVENT_STATE_ACTIVE)
> -				kvm_debug("fail to enable perf event\n");
> -		}
> +		kvm_pmu_enable_counter(vcpu, i);
>   	}
>   }
>   
>   /**
> + * kvm_pmu_sync_counter_enable - reenable a counter if it should be enabled
> + * @vcpu: The vcpu pointer
> + * @select_idx: The counter index
> + */
> +static void kvm_pmu_sync_counter_enable(struct kvm_vcpu *vcpu,
> +					    u64 select_idx)
> +{
> +	u64 set = __vcpu_sys_reg(vcpu, PMCNTENSET_EL0);
> +
> +	if (set & BIT(select_idx))
> +		kvm_pmu_enable_counter_mask(vcpu, BIT(select_idx));

I think there is a problem here. We could be creating an event for a
counter beyond what is supported by the CPU. i.e, we don't
seem to validate that the mask we are creating is within the
kvm_pmu_valid_counter_mask(). The other callers seem to verify this.
I guess it may be better to add a check for this in the
kvm_pmu_enable_counter_mask().

minor nit: Feel free to ignore. If we move the check for PMCNTENSET_EL0 to
pmu_enable_counter_mask() we could get rid of the above function. Anyways,
we should only be enabling the counters set in PMCNTENSET_EL0.


> +}
> +
> +/**
> + * kvm_pmu_disable_counter - disable selected PMU counter
> + * @vcpu: The vcpu pointer
> + * @pmc: The counter to disable
> + */
> +static void kvm_pmu_disable_counter(struct kvm_vcpu *vcpu, u64 select_idx)
> +{
> +	struct kvm_pmu *pmu = &vcpu->arch.pmu;
> +	struct kvm_pmc *pmc = &pmu->pmc[select_idx];
> +
> +	if (pmc->perf_event)
> +		perf_event_disable(pmc->perf_event);
> +}
> +
> +/**
>    * kvm_pmu_disable_counter_mask - disable selected PMU counters
>    * @vcpu: The vcpu pointer
>    * @val: the value guest writes to PMCNTENCLR register
> @@ -181,8 +226,6 @@ void kvm_pmu_enable_counter_mask(struct kvm_vcpu *vcpu, u64 val)
>   void kvm_pmu_disable_counter_mask(struct kvm_vcpu *vcpu, u64 val)
>   {
>   	int i;
> -	struct kvm_pmu *pmu = &vcpu->arch.pmu;
> -	struct kvm_pmc *pmc;
>   
>   	if (!val)
>   		return;
> @@ -191,9 +234,7 @@ void kvm_pmu_disable_counter_mask(struct kvm_vcpu *vcpu, u64 val)
>   		if (!(val & BIT(i)))
>   			continue;
>   
> -		pmc = &pmu->pmc[i];
> -		if (pmc->perf_event)
> -			perf_event_disable(pmc->perf_event);
> +		kvm_pmu_disable_counter(vcpu, i);
>   	}
>   }
>   
> @@ -375,16 +416,9 @@ void kvm_pmu_handle_pmcr(struct kvm_vcpu *vcpu, u64 val)
>   	}
>   }
>   
> -static bool kvm_pmu_counter_is_enabled(struct kvm_vcpu *vcpu, u64 select_idx)
> -{
> -	return (__vcpu_sys_reg(vcpu, PMCR_EL0) & ARMV8_PMU_PMCR_E) &&
> -	       (__vcpu_sys_reg(vcpu, PMCNTENSET_EL0) & BIT(select_idx));
> -}
> -
>   /**
> - * kvm_pmu_create_perf_event - create a perf event for a counter
> + * kvm_pmu_counter_create_enabled_perf_event - create a perf event for a counter

nit: Name hasn't changed. Also, please could update that, the events are always
created disabled and they are only enabled lazily.

Reset looks fines.

Suzuki