Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 24 May 2023 15:36:11 -0700
In-Reply-To: <20230323072714.82289-3-likexu@tencent.com>
Mime-Version: 1.0
References: <20230323072714.82289-1-likexu@tencent.com> <20230323072714.82289-3-likexu@tencent.com>
Message-ID: <ZG6RW4AwayOdPTaZ@google.com>
Subject: Re: [PATCH 2/7] KVM: selftests: Test Intel PMU architectural events
 on fixed counters
From:   Sean Christopherson <seanjc@google.com>
To:     Like Xu <like.xu.linux@gmail.com>
Cc:     Paolo Bonzini <pbonzini@redhat.com>,
        Jinrong Liang <cloudliang@tencent.com>,
        linux-kernel@vger.kernel.org, kvm@vger.kernel.org
Content-Type: text/plain; charset="us-ascii"
Precedence: bulk

On Thu, Mar 23, 2023, Like Xu wrote:
> From: Jinrong Liang <cloudliang@tencent.com>
> 
> Update test to cover Intel PMU architectural events on fixed counters.
> Per Intel SDM, PMU users can also count architecture performance events
> on fixed counters (specifically, FIXED_CTR0 for the retired instructions
> and FIXED_CTR1 for cpu core cycles event). Therefore, if guest's CPUID
> indicates that an architecture event is not available, the corresponding
> fixed counter will also not count that event.
> 
> Co-developed-by: Like Xu <likexu@tencent.com>
> Signed-off-by: Like Xu <likexu@tencent.com>
> Signed-off-by: Jinrong Liang <cloudliang@tencent.com>
> ---
>  .../selftests/kvm/x86_64/pmu_cpuid_test.c     | 37 +++++++++++++++++--
>  1 file changed, 33 insertions(+), 4 deletions(-)
> 
> diff --git a/tools/testing/selftests/kvm/x86_64/pmu_cpuid_test.c b/tools/testing/selftests/kvm/x86_64/pmu_cpuid_test.c
> index faab0a91e191..75434aa2a0ec 100644
> --- a/tools/testing/selftests/kvm/x86_64/pmu_cpuid_test.c
> +++ b/tools/testing/selftests/kvm/x86_64/pmu_cpuid_test.c
> @@ -25,6 +25,9 @@
>  #define GP_CTR_NUM_MASK GENMASK_ULL(15, GP_CTR_NUM_OFS_BIT)
>  #define EVT_LEN_OFS_BIT 24
>  #define EVT_LEN_MASK GENMASK_ULL(31, EVT_LEN_OFS_BIT)
> +#define INTEL_PMC_IDX_FIXED 32
> +#define RDPMC_FIXED_BASE BIT_ULL(30)
> +#define FIXED_CTR_NUM_MASK GENMASK_ULL(4, 0)
>  
>  #define ARCH_EVENT(select, umask) (((select) & 0xff) | ((umask) & 0xff) << 8)
>  
> @@ -43,6 +46,14 @@ static const uint64_t arch_events[] = {
>  	[7] = ARCH_EVENT(0xa4, 0x1), /* Topdown Slots */
>  };
>  
> +/* Association of Fixed Counters with Architectural Performance Events */
> +static int fixed_events[] = {1, 0, 7};
> +
> +static uint64_t evt_code_for_fixed_ctr(uint8_t idx)
> +{
> +	return arch_events[fixed_events[idx]];
> +}

This appears to be more fodder for common code.

>  static struct kvm_vcpu *new_vcpu(void *guest_code)
>  {
>  	struct kvm_vm *vm;
> @@ -88,8 +99,8 @@ static bool first_uc_arg_non_zero(struct ucall *uc, void *data)
>  }
>  
>  static void intel_guest_run_arch_event(uint8_t version, uint8_t max_gp_num,
> -				       bool supported, uint32_t ctr_base_msr,
> -				       uint64_t evt_code)
> +				       uint8_t max_fixed_num, bool supported,
> +				       uint32_t ctr_base_msr, uint64_t evt_code)
>  {
>  	uint32_t global_msr = MSR_CORE_PERF_GLOBAL_CTRL;
>  	unsigned int i;
> @@ -108,6 +119,23 @@ static void intel_guest_run_arch_event(uint8_t version, uint8_t max_gp_num,
>  		GUEST_SYNC(supported == !!_rdpmc(i));
>  	}
>  
> +	/* No need to test independent arch events on fixed counters. */
> +	if (version > 1 && max_fixed_num > 1 &&
> +	    (evt_code == evt_code_for_fixed_ctr(0) ||
> +	     evt_code == evt_code_for_fixed_ctr(1))) {
> +		i = (evt_code == evt_code_for_fixed_ctr(0)) ? 0 : 1;

The ternary operator on top of a duplicate comparison isn't super intuitive.
Maybe use gotos?  Definitely just an idea, not a requirement.

	if (version <= 1 || max_fixed_num <= 1)
		goto done;

	if (evt_code == evt_code_for_fixed_ctr(0))
		i = 0;
	else if (evt_code == evt_code_for_fixed_ctr(1))
		i = 1;
	else
		goto done;

> +		wrmsr(MSR_CORE_PERF_FIXED_CTR0 + i, 0);
> +		wrmsr(MSR_CORE_PERF_FIXED_CTR_CTRL, BIT_ULL(4 * i));
> +		wrmsr(global_msr, BIT_ULL(INTEL_PMC_IDX_FIXED + i));
> +
> +		__asm__ __volatile__("loop ." : "+c"((int){NUM_BRANCHES}));
> +
> +		wrmsr(global_msr, 0);
> +
> +		GUEST_SYNC(supported == !!_rdpmc(RDPMC_FIXED_BASE | i));
> +	}
> +
>  	GUEST_DONE();
>  }