KVM: x86/svm/pmu: Add AMD Guest PerfMonV2 support
Subsystems (4)
cover KVM: x86/svm/pmu: Add AMD Guest PerfMonV2 support
Starting with Zen4, core PMU on AMD platforms such as Genoa and Ryzen-7000 will support PerfMonV2, and it is also compatible with legacy PERFCTR_CORE behavior and msr addresses.
If you don't have access to the hardware specification, the commits d6d0c7f681fd..7685665c390d for host perf can also bring a quick overview. Its main change is the addition of three msr's equivalent to Intel V2, namely global_ctrl, global_status, global_status_clear.
It is worth noting that this feature is very attractive for reducing the overhead of PMU virtualization, since multiple msr accesses to multiple counters will be replaced by a single access to the global register, plus more accuracy gain when multiple guest counters are used.
The KVM part is based on the latest vPMU fixes patch set [1], while the kvm-unit-test patch relies on preemptive test cases effort [2] for testing leagcy AMD vPMU, which didn't exist before.
All related testcases are passed on a Genoa box. Please feel free to run more tests, add more or share comments.
[1] https://lore.kernel.org/kvm/[email protected]/
[2] https://lore.kernel.org/kvm/[email protected]/
Like Xu (3):
KVM: x86/svm/pmu: Limit the maximum number of supported GP counters
KVM: x86/pmu: Make part of the Intel v2 PMU MSRs handling x86 generic
KVM: x86/svm/pmu: Add AMD PerfMonV2 support
Sandipan Das (1):
KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
arch/x86/include/asm/kvm-x86-pmu-ops.h | 1 - arch/x86/include/asm/perf_event.h | 8 ++++ arch/x86/kvm/cpuid.c | 21 +++++++- arch/x86/kvm/pmu.c | 61 ++++++++++++++++++++++-- arch/x86/kvm/pmu.h | 32 ++++++++++++- arch/x86/kvm/svm/pmu.c | 66 +++++++++++++++++--------- arch/x86/kvm/vmx/pmu_intel.c | 58 +--------------------- arch/x86/kvm/x86.c | 13 +++++ 8 files changed, 173 insertions(+), 87 deletions(-)
Signature · 1 line
2.37.3
PATCH 1/4 KVM: x86/svm/pmu: Limit the maximum number of supported GP counters
From: Like Xu <[email protected]>
The AMD PerfMonV2 specification allows for a maximum of 16 GP counters, which is clearly not supported with zero code effort in the current KVM.
A local macro (named like INTEL_PMC_MAX_GENERIC) is introduced to take back control of this virt capability, which also makes it easier to statically partition all available counters between hosts and guests.
--- arch/x86/kvm/pmu.h | 2 ++ arch/x86/kvm/svm/pmu.c | 7 ++++--- arch/x86/kvm/x86.c | 2 ++ 3 files changed, 8 insertions(+), 3 deletions(-)
arch/x86/kvm/pmu.h +2 -0
@@ -18,6 +18,8 @@#define VMWARE_BACKDOOR_PMC_REAL_TIME 0x10001#define VMWARE_BACKDOOR_PMC_APPARENT_TIME 0x10002 +#define KVM_AMD_PMC_MAX_GENERIC AMD64_NUM_COUNTERS_CORE +struct kvm_event_hw_type_mapping {u8 eventsel;u8 unit_mask;
arch/x86/kvm/svm/pmu.c +4 -3
@@ -192,9 +192,10 @@ static void amd_pmu_init(struct kvm_vcpu *vcpu)struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);int i; -BUILD_BUG_ON (AMD64_NUM_COUNTERS_CORE >INTEL_PMC_MAX_GENERIC ); +BUILD_BUG_ON (AMD64_NUM_COUNTERS_CORE >KVM_AMD_PMC_MAX_GENERIC ); +BUILD_BUG_ON (KVM_AMD_PMC_MAX_GENERIC >INTEL_PMC_MAX_GENERIC ); -for (i =0 ; i <AMD64_NUM_COUNTERS_CORE ; i++) { +for (i =0 ; i <KVM_AMD_PMC_MAX_GENERIC ; i++) { pmu->gp_counters[i].type =KVM_PMC_GP ; pmu->gp_counters[i].vcpu = vcpu; pmu->gp_counters[i].idx = i;
@@ -207,7 +208,7 @@ static void amd_pmu_reset(struct kvm_vcpu *vcpu)struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);int i; -for (i =0 ; i <AMD64_NUM_COUNTERS_CORE ; i++) { +for (i =0 ; i <KVM_AMD_PMC_MAX_GENERIC ; i++) {struct kvm_pmc *pmc = &pmu->gp_counters[i];pmc_stop_counter (pmc);
arch/x86/kvm/x86.c +2 -0
@@ -1444,12 +1444,14 @@ static const u32 msrs_to_save_all[] = {MSR_ARCH_PERFMON_EVENTSEL0 +16 ,MSR_ARCH_PERFMON_EVENTSEL0 +17 ,MSR_IA32_PEBS_ENABLE ,MSR_IA32_DS_AREA ,MSR_PEBS_DATA_CFG , +/* This part of MSRs should match KVM_AMD_PMC_MAX_GENERIC. */ MSR_K7_EVNTSEL0 ,MSR_K7_EVNTSEL1 ,MSR_K7_EVNTSEL2 ,MSR_K7_EVNTSEL3 ,MSR_K7_PERFCTR0 ,MSR_K7_PERFCTR1 ,MSR_K7_PERFCTR2 ,MSR_K7_PERFCTR3 ,MSR_F15H_PERF_CTL0 ,MSR_F15H_PERF_CTL1 ,MSR_F15H_PERF_CTL2 ,MSR_F15H_PERF_CTL3 ,MSR_F15H_PERF_CTL4 ,MSR_F15H_PERF_CTL5 ,MSR_F15H_PERF_CTR0 ,MSR_F15H_PERF_CTR1 ,MSR_F15H_PERF_CTR2 ,MSR_F15H_PERF_CTR3 ,MSR_F15H_PERF_CTR4 ,MSR_F15H_PERF_CTR5 , + MSR_IA32_XFD, MSR_IA32_XFD_ERR, };
Signature · 1 line
2.37.3
Re: [PATCH 1/4] KVM: x86/svm/pmu: Limit the maximum number of supported GP counters
Quoted 61 lines · Like Xu
From: Like Xu <[email protected]>
The AMD PerfMonV2 specification allows for a maximum of 16 GP counters, which is clearly not supported with zero code effort in the current KVM.
A local macro (named like
INTEL_PMC_MAX_GENERIC) is introduced to take back control of this virt capability, which also makes it easier to statically partition all available counters between hosts and guests.Signed-off-by Like Xu <[email protected]>--- arch/x86/kvm/pmu.h | 2 ++ arch/x86/kvm/svm/pmu.c | 7 ++++--- arch/x86/kvm/x86.c | 2 ++ 3 files changed, 8 insertions(+), 3 deletions(-)arch/x86/kvm/pmu.h +2
-0@@ -18,6 +18,8 @@#define VMWARE_BACKDOOR_PMC_REAL_TIME 0x10001#define VMWARE_BACKDOOR_PMC_APPARENT_TIME 0x10002 +#define KVM_AMD_PMC_MAX_GENERIC AMD64_NUM_COUNTERS_CORE +struct kvm_event_hw_type_mapping {u8 eventsel;u8 unit_mask;arch/x86/kvm/svm/pmu.c +4
-3@@ -192,9 +192,10 @@ static void amd_pmu_init(struct kvm_vcpu *vcpu)struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);int i; -BUILD_BUG_ON (AMD64_NUM_COUNTERS_CORE >INTEL_PMC_MAX_GENERIC ); +BUILD_BUG_ON (AMD64_NUM_COUNTERS_CORE >KVM_AMD_PMC_MAX_GENERIC ); +BUILD_BUG_ON (KVM_AMD_PMC_MAX_GENERIC >INTEL_PMC_MAX_GENERIC ); -for (i =0 ; i <AMD64_NUM_COUNTERS_CORE ; i++) { +for (i =0 ; i <KVM_AMD_PMC_MAX_GENERIC ; i++) { pmu->gp_counters[i].type =KVM_PMC_GP ; pmu->gp_counters[i].vcpu = vcpu; pmu->gp_counters[i].idx = i;@@ -207,7 +208,7 @@ static void amd_pmu_reset(struct kvm_vcpu *vcpu)struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);int i; -for (i =0 ; i <AMD64_NUM_COUNTERS_CORE ; i++) { +for (i =0 ; i <KVM_AMD_PMC_MAX_GENERIC ; i++) {struct kvm_pmc *pmc = &pmu->gp_counters[i];pmc_stop_counter (pmc);arch/x86/kvm/x86.c +0
-0
arch/x86/kvm/x86.c static const u32 msrs_to_save_all[] = { · line 1444
Quoted 1 line
arch/x86/kvm/x86.c +0
-0@@ -1444,12 +1444,14 @@ static const u32 msrs_to_save_all[] = { MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
IIRC, the effective maximum on the Intel side is 18, despite what INTEL_PMC_MAX_GENERIC says, due to collisions with other existing MSR indices. That's why the Intel list stops here. Should we introduce a KVM_INTEL_PMC_MAX_GENERIC as well?
arch/x86/kvm/x86.c static const u32 msrs_to_save_all[] = { · line 1447
Quoted 3 lines
arch/x86/kvm/x86.c +1
-0MSR_IA32_PEBS_ENABLE, MSR_IA32_DS_AREA, MSR_PEBS_DATA_CFG, +/* This part of MSRs should match KVM_AMD_PMC_MAX_GENERIC. */
Perhaps the comment above should be moved down two lines, since the next two lines deal with the legacy counters.
arch/x86/kvm/x86.c static const u32 msrs_to_save_all[] = { · line 1453
Quoted 6 lines
arch/x86/kvm/x86.c +0
-0MSR_K7_EVNTSEL0 ,MSR_K7_EVNTSEL1 ,MSR_K7_EVNTSEL2 ,MSR_K7_EVNTSEL3 ,MSR_K7_PERFCTR0 ,MSR_K7_PERFCTR1 ,MSR_K7_PERFCTR2 ,MSR_K7_PERFCTR3 ,MSR_F15H_PERF_CTL0 ,MSR_F15H_PERF_CTL1 ,MSR_F15H_PERF_CTL2 ,MSR_F15H_PERF_CTL3 ,MSR_F15H_PERF_CTL4 ,MSR_F15H_PERF_CTL5 ,MSR_F15H_PERF_CTR0 ,MSR_F15H_PERF_CTR1 ,MSR_F15H_PERF_CTR2 , MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
At some point, we may want to consider populating the PMU MSR list dynamically, rather than statically enumerating all of them (for both AMD and Intel).
Like Xu
AppliedRe: [PATCH 1/4] KVM: x86/svm/pmu: Limit the maximum number of supported GP counters
Quoted 56 lines · Jim Mattson
Quoted 55 lines · Like Xu
From: Like Xu <[email protected]>
The AMD PerfMonV2 specification allows for a maximum of 16 GP counters, which is clearly not supported with zero code effort in the current KVM.
A local macro (named like
INTEL_PMC_MAX_GENERIC) is introduced to take back control of this virt capability, which also makes it easier to statically partition all available counters between hosts and guests.Signed-off-by Like Xu <[email protected]>--- arch/x86/kvm/pmu.h | 2 ++ arch/x86/kvm/svm/pmu.c | 7 ++++--- arch/x86/kvm/x86.c | 2 ++ 3 files changed, 8 insertions(+), 3 deletions(-)arch/x86/kvm/pmu.h +2
-0@@ -18,6 +18,8 @@#define VMWARE_BACKDOOR_PMC_REAL_TIME 0x10001#define VMWARE_BACKDOOR_PMC_APPARENT_TIME 0x10002 +#define KVM_AMD_PMC_MAX_GENERIC AMD64_NUM_COUNTERS_CORE +struct kvm_event_hw_type_mapping {u8 eventsel;u8 unit_mask;arch/x86/kvm/svm/pmu.c +4
-3@@ -192,9 +192,10 @@ static void amd_pmu_init(struct kvm_vcpu *vcpu)struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);int i; -BUILD_BUG_ON (AMD64_NUM_COUNTERS_CORE >INTEL_PMC_MAX_GENERIC ); +BUILD_BUG_ON (AMD64_NUM_COUNTERS_CORE >KVM_AMD_PMC_MAX_GENERIC ); +BUILD_BUG_ON (KVM_AMD_PMC_MAX_GENERIC >INTEL_PMC_MAX_GENERIC ); -for (i =0 ; i <AMD64_NUM_COUNTERS_CORE ; i++) { +for (i =0 ; i <KVM_AMD_PMC_MAX_GENERIC ; i++) { pmu->gp_counters[i].type =KVM_PMC_GP ; pmu->gp_counters[i].vcpu = vcpu; pmu->gp_counters[i].idx = i;@@ -207,7 +208,7 @@ static void amd_pmu_reset(struct kvm_vcpu *vcpu)struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);int i; -for (i =0 ; i <AMD64_NUM_COUNTERS_CORE ; i++) { +for (i =0 ; i <KVM_AMD_PMC_MAX_GENERIC ; i++) {struct kvm_pmc *pmc = &pmu->gp_counters[i];
Quoted 10 lines
arch/x86/kvm/x86.c static const u32 msrs_to_save_all[] = { · line 1444
Quoted 7 lines
arch/x86/kvm/svm/pmu.c +0
-0pmc_stop_counter (pmc);arch/x86/kvm/x86.c +0
-0@@ -1444,12 +1444,14 @@ static const u32 msrs_to_save_all[] = { MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,IIRC, the effective maximum on the Intel side is 18, despite what
INTEL_PMC_MAX_GENERICsays, due to collisions with other existing MSR
Emm, check https://lore.kernel.org/kvm/[email protected]/
Quoted 2 lines
indices. That's why the Intel list stops here. Should we introduce a
KVM_INTEL_PMC_MAX_GENERICas well?
Yes, this suggestion will be applied in the next version.
Quoted 7 lines
arch/x86/kvm/x86.c static const u32 msrs_to_save_all[] = { · line 1447
Quoted 4 lines
arch/x86/kvm/x86.c +1
-0MSR_IA32_PEBS_ENABLE, MSR_IA32_DS_AREA, MSR_PEBS_DATA_CFG, +/* This part of MSRs should match KVM_AMD_PMC_MAX_GENERIC. */ Perhaps the comment above should be moved down two lines, since the next two lines deal with the legacy counters.
Applied, thanks.
Quoted 11 lines
arch/x86/kvm/x86.c static const u32 msrs_to_save_all[] = { · line 1453
Quoted 7 lines
arch/x86/kvm/x86.c +0
-0MSR_K7_EVNTSEL0 ,MSR_K7_EVNTSEL1 ,MSR_K7_EVNTSEL2 ,MSR_K7_EVNTSEL3 ,MSR_K7_PERFCTR0 ,MSR_K7_PERFCTR1 ,MSR_K7_PERFCTR2 ,MSR_K7_PERFCTR3 ,MSR_F15H_PERF_CTL0 ,MSR_F15H_PERF_CTL1 ,MSR_F15H_PERF_CTL2 ,MSR_F15H_PERF_CTL3 ,MSR_F15H_PERF_CTL4 ,MSR_F15H_PERF_CTL5 ,MSR_F15H_PERF_CTR0 ,MSR_F15H_PERF_CTR1 ,MSR_F15H_PERF_CTR2 , MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,At some point, we may want to consider populating the PMU MSR list dynamically, rather than statically enumerating all of them (for both AMD and Intel).
The uncertainty of msrs_to_save_all[] may cause troubles for legacy user spaces. I had draft patches to rewrite pmu msr accesses for host-initiated as first move.
Quoted 2 lines
Reviewed-by Jim Mattson <[email protected]>
PATCH 2/4 KVM: x86/pmu: Make part of the Intel v2 PMU MSRs handling x86 generic
In reply to Like XuFrom: Like Xu <[email protected]>
The AMD PerfMonV2 defines three registers similar to part of the Intel v2 PMU registers, including the GLOBAL_CTRL, GLOBAL_STATUS and GLOBAL_OVF_CTRL MSRs. For better code reuse, this specific part of the handling can be extracted to make it generic for X86.
The new non-prefix pmc_is_enabled() works well as legacy AMD vPMU version is indexeqd as 1. Note that the specific *_is_valid_msr will continue to be used to avoid cross-vendor msr access.
--- arch/x86/include/asm/kvm-x86-pmu-ops.h | 1 - arch/x86/kvm/pmu.c | 55 +++++++++++++++++++++--- arch/x86/kvm/pmu.h | 30 ++++++++++++- arch/x86/kvm/svm/pmu.c | 9 ---- arch/x86/kvm/vmx/pmu_intel.c | 58 +------------------------- 5 files changed, 80 insertions(+), 73 deletions(-)
arch/x86/include/asm/kvm-x86-pmu-ops.h +0 -1
@@ -13,7 +13,6 @@ BUILD_BUG_ON(1)* at the call sites. */ KVM_X86_PMU_OP(hw_event_available) -KVM_X86_PMU_OP(pmc_is_enabled) KVM_X86_PMU_OP(pmc_idx_to_pmc) KVM_X86_PMU_OP(rdpmc_ecx_to_pmc) KVM_X86_PMU_OP(msr_idx_to_pmc)
arch/x86/kvm/pmu.c +50 -5
72 lines not on this page · raw message
arch/x86/kvm/pmu.h +29 -1
@@ -28,7 +28,6 @@ struct kvm_event_hw_type_mapping {struct kvm_pmu_ops {bool (*hw_event_available)(struct kvm_pmc *pmc); -bool (*pmc_is_enabled)(struct kvm_pmc *pmc);struct kvm_pmc *(*pmc_idx_to_pmc)(struct kvm_pmu *pmu,int pmc_idx);struct kvm_pmc *(*rdpmc_ecx_to_pmc)(struct kvm_vcpu *vcpu,unsigned int idx,u64 *mask);
@@ -191,6 +190,35 @@ static inline void kvm_pmu_request_counter_reprogam(struct kvm_pmc *pmc)kvm_make_request (KVM_REQ_PMU , pmc->vcpu); } +/* +* Check if a PMC is enabled by comparing it against global_ctrl bits. +* +* If the current version of vPMU doesn't have global_ctrl MSR, +* all vPMCs are enabled (return TRUE). +*/ +static inline bool pmc_is_enabled (struct kvm_pmc *pmc) +{ +struct kvm_pmu *pmu =pmc_to_pmu (pmc); + +if (pmu->version <2 ) +return true; + +return test_bit (pmc->idx, (unsigned long *)&pmu->global_ctrl); +} + +static inline void reprogram_counters (struct kvm_pmu *pmu,u64 diff) +{ +int bit; + +if (!diff) +return ; + +for_each_set_bit (bit, (unsigned long *)&diff,X86_PMC_IDX_MAX ) +__set_bit (bit, pmu->reprogram_pmi); + +kvm_make_request (KVM_REQ_PMU ,pmu_to_vcpu (pmu)); +} +void kvm_pmu_deliver_pmi (struct kvm_vcpu *vcpu);void kvm_pmu_handle_event (struct kvm_vcpu *vcpu);int kvm_pmu_rdpmc (struct kvm_vcpu *vcpu,unsigned pmc,u64 *data);
arch/x86/kvm/svm/pmu.c +0 -9
@@ -76,14 +76,6 @@ static bool amd_hw_event_available(struct kvm_pmc *pmc)return true; } -/* check if a PMC is enabled by comparing it against global_ctrl bits. Because -* AMD CPU doesn't have global_ctrl MSR, all PMCs are enabled (return TRUE). -*/ -static bool amd_pmc_is_enabled (struct kvm_pmc *pmc) -{ -return true; -} -static bool amd_is_valid_rdpmc_ecx (struct kvm_vcpu *vcpu,unsigned int idx) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);
@@ -218,7 +210,6 @@ static void amd_pmu_reset(struct kvm_vcpu *vcpu)struct kvm_pmu_ops amd_pmu_ops__initdata = { .hw_event_available = amd_hw_event_available, - .pmc_is_enabled = amd_pmc_is_enabled, .pmc_idx_to_pmc = amd_pmc_idx_to_pmc, .rdpmc_ecx_to_pmc = amd_rdpmc_ecx_to_pmc, .msr_idx_to_pmc = amd_msr_idx_to_pmc,
arch/x86/kvm/vmx/pmu_intel.c +1 -57
94 lines not on this page · raw message
Signature · 1 line
2.37.3
PATCH 3/4 KVM: x86/svm/pmu: Add AMD PerfMonV2 support
In reply to Like XuFrom: Like Xu <[email protected]>
If AMD Performance Monitoring Version 2 (PerfMonV2) is detected by the guest, it can use a new scheme to manage the Core PMCs using the new global control and status registers.
In addition to benefiting from the PerfMonV2 functionality in the same way as the host (higher precision), the guest also can reduce the number of vm-exits by lowering the total number of MSRs accesses.
In terms of implementation details, amd_is_valid_msr() is resurrected since three newly added MSRs could not be mapped to one vPMC. The possibility of emulating PerfMonV2 on the mainframe has also been eliminated for reasons of precision.
--- arch/x86/kvm/pmu.c | 6 +++++ arch/x86/kvm/svm/pmu.c | 50 +++++++++++++++++++++++++++++++++--------- arch/x86/kvm/x86.c | 11 ++++++++++ 3 files changed, 57 insertions(+), 10 deletions(-)
arch/x86/kvm/pmu.c +6 -0
@@ -455,12 +455,15 @@ int kvm_pmu_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : msr_info->data = pmu->global_status;return 0 ;case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : msr_info->data = pmu->global_ctrl;return 0 ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR : msr_info->data =0 ;return 0 ;default :
@@ -479,12 +482,14 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS :if (msr_info->host_initiated) { pmu->global_status = data;return 0 ; }break ;/* RO MSR */ case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL :if (pmu->global_ctrl == data)return 0 ;if (kvm_valid_perf_global_ctrl (pmu, data)) {
@@ -495,6 +500,7 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) }break ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :if (!(data & pmu->global_ovf_ctrl_mask)) {if (!msr_info->host_initiated) pmu->global_status &= ~data;
arch/x86/kvm/svm/pmu.c +40 -10
84 lines not on this page · raw message
arch/x86/kvm/x86.c +11 -0
@@ -1424,6 +1424,11 @@ static const u32 msrs_to_save_all[] = {MSR_ARCH_PERFMON_FIXED_CTR0 +2 ,MSR_CORE_PERF_FIXED_CTR_CTRL ,MSR_CORE_PERF_GLOBAL_STATUS ,MSR_CORE_PERF_GLOBAL_CTRL ,MSR_CORE_PERF_GLOBAL_OVF_CTRL , + +MSR_AMD64_PERF_CNTR_GLOBAL_CTL , +MSR_AMD64_PERF_CNTR_GLOBAL_STATUS , +MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR , +MSR_ARCH_PERFMON_PERFCTR0 ,MSR_ARCH_PERFMON_PERFCTR1 ,MSR_ARCH_PERFMON_PERFCTR0 +2 ,MSR_ARCH_PERFMON_PERFCTR0 +3 , MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
@@ -3856,6 +3861,9 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)case MSR_IA32_DS_AREA :case MSR_PEBS_DATA_CFG : case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :if (kvm_pmu_is_valid_msr (vcpu, msr))return kvm_pmu_set_msr (vcpu, msr_info);/*
@@ -3959,6 +3967,9 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)case MSR_IA32_DS_AREA :case MSR_PEBS_DATA_CFG : case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :if (kvm_pmu_is_valid_msr (vcpu, msr_info->index))return kvm_pmu_get_msr (vcpu, msr_info);/*
Signature · 1 line
2.37.3
Re: [PATCH 3/4] KVM: x86/svm/pmu: Add AMD PerfMonV2 support
Quoted 96 lines · Like Xu
From: Like Xu <[email protected]>
If AMD Performance Monitoring Version 2 (PerfMonV2) is detected by the guest, it can use a new scheme to manage the Core PMCs using the new global control and status registers.
In addition to benefiting from the PerfMonV2 functionality in the same way as the host (higher precision), the guest also can reduce the number of vm-exits by lowering the total number of MSRs accesses.
In terms of implementation details,
amd_is_valid_msr()is resurrected since three newly added MSRs could not be mapped to one vPMC. The possibility of emulating PerfMonV2 on the mainframe has also been eliminated for reasons of precision.Co-developed-by Sandipan Das <[email protected]>Signed-off-by Sandipan Das <[email protected]>Signed-off-by Like Xu <[email protected]>--- arch/x86/kvm/pmu.c | 6 +++++ arch/x86/kvm/svm/pmu.c | 50 +++++++++++++++++++++++++++++++++--------- arch/x86/kvm/x86.c | 11 ++++++++++ 3 files changed, 57 insertions(+), 10 deletions(-)arch/x86/kvm/pmu.c +6
-0@@ -455,12 +455,15 @@ int kvm_pmu_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : msr_info->data = pmu->global_status;return 0 ;case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : msr_info->data = pmu->global_ctrl;return 0 ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR : msr_info->data =0 ;return 0 ;default :@@ -479,12 +482,14 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS :if (msr_info->host_initiated) { pmu->global_status = data;return 0 ; }break ;/* RO MSR */ case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL :if (pmu->global_ctrl == data)return 0 ;if (kvm_valid_perf_global_ctrl (pmu, data)) {@@ -495,6 +500,7 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) }break ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :if (!(data & pmu->global_ovf_ctrl_mask)) {if (!msr_info->host_initiated) pmu->global_status &= ~data;arch/x86/kvm/svm/pmu.c +7
-6@@ -92,12 +92,6 @@ static struct kvm_pmc *amd_rdpmc_ecx_to_pmc(struct kvm_vcpu *vcpu,return amd_pmc_idx_to_pmc (vcpu_to_pmu (vcpu), idx & ~(3u <<30 )); } -static bool amd_is_valid_msr (struct kvm_vcpu *vcpu,u32 msr) -{ -/* All MSRs refer to exactly one PMC, so msr_idx_to_pmc is enough. */ -return false; -} -static struct kvm_pmc *amd_msr_idx_to_pmc (struct kvm_vcpu *vcpu,u32 msr) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);@@ -109,6 +103,29 @@ static struct kvm_pmc *amd_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr)return pmc; } +static bool amd_is_valid_msr (struct kvm_vcpu *vcpu,u32 msr) +{ +struct kvm_pmu *pmu =vcpu_to_pmu (vcpu); + +switch (msr) { + case MSR_K7_EVNTSEL0 ... MSR_K7_PERFCTR3: +return pmu->version >0 ;
arch/x86/kvm/svm/pmu.c static struct kvm_pmc *amd_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr) · line 122
Quoted 10 lines
arch/x86/kvm/svm/pmu.c +10
-0+ case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: +return guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE ); +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR : +return pmu->version >1 ; +default : +if (msr >MSR_F15H_PERF_CTR5 && + msr <MSR_F15H_PERF_CTL0 +2 *KVM_AMD_PMC_MAX_GENERIC ) +return pmu->version >1 ;
Should this be bounded by guest CPUID.80000022H:EBX[NumCorePmc] (unless host-initiated)?
Quoted 19 lines
arch/x86/kvm/svm/pmu.c +10
-1+break ; + } + +return amd_msr_idx_to_pmc (vcpu, msr); +} +static int amd_pmu_get_msr (struct kvm_vcpu *vcpu,struct msr_data *msr_info) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);@@ -162,20 +179,31 @@ static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)static void amd_pmu_refresh (struct kvm_vcpu *vcpu) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu); +struct kvm_cpuid_entry2 *entry; +union cpuid_0x80000022_ebx ebx; -if (guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE )) + pmu->version =1 ; + entry =kvm_find_cpuid_entry_index (vcpu,0x80000022 ,0 );
arch/x86/kvm/svm/pmu.c static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) · line 196
Quoted 10 lines
arch/x86/kvm/svm/pmu.c +9
-0+if (kvm_pmu_cap.version >1 && entry && (entry->eax &BIT (0 ))) { + pmu->version =2 ; + ebx.full = entry->ebx; + pmu->nr_arch_gp_counters =min3 ((unsigned int )ebx.split.num_core_pmc, + (unsigned int )kvm_pmu_cap.num_counters_gp, + (unsigned int )KVM_AMD_PMC_MAX_GENERIC ); + pmu->global_ctrl_mask = ~((1ull << pmu->nr_arch_gp_counters) -1 ); + pmu->global_ovf_ctrl_mask = pmu->global_ctrl_mask; + }else if (guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE )) { pmu->nr_arch_gp_counters =AMD64_NUM_COUNTERS_CORE ;
The logic above doesn't seem quite right, since guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE) promises 6 PMCs, regardless of what CPUID.80000022 says.
Re: [PATCH 3/4] KVM: x86/svm/pmu: Add AMD PerfMonV2 support
Quoted 102 lines · Jim Mattson
Quoted 101 lines · Like Xu
From: Like Xu <[email protected]>
If AMD Performance Monitoring Version 2 (PerfMonV2) is detected by the guest, it can use a new scheme to manage the Core PMCs using the new global control and status registers.
In addition to benefiting from the PerfMonV2 functionality in the same way as the host (higher precision), the guest also can reduce the number of vm-exits by lowering the total number of MSRs accesses.
In terms of implementation details,
amd_is_valid_msr()is resurrected since three newly added MSRs could not be mapped to one vPMC. The possibility of emulating PerfMonV2 on the mainframe has also been eliminated for reasons of precision.Co-developed-by Sandipan Das <[email protected]>Signed-off-by Sandipan Das <[email protected]>Signed-off-by Like Xu <[email protected]>--- arch/x86/kvm/pmu.c | 6 +++++ arch/x86/kvm/svm/pmu.c | 50 +++++++++++++++++++++++++++++++++--------- arch/x86/kvm/x86.c | 11 ++++++++++ 3 files changed, 57 insertions(+), 10 deletions(-)arch/x86/kvm/pmu.c +6
-0@@ -455,12 +455,15 @@ int kvm_pmu_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : msr_info->data = pmu->global_status;return 0 ;case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : msr_info->data = pmu->global_ctrl;return 0 ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR : msr_info->data =0 ;return 0 ;default :@@ -479,12 +482,14 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS :if (msr_info->host_initiated) { pmu->global_status = data;return 0 ; }break ;/* RO MSR */ case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL :if (pmu->global_ctrl == data)return 0 ;if (kvm_valid_perf_global_ctrl (pmu, data)) {@@ -495,6 +500,7 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) }break ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :if (!(data & pmu->global_ovf_ctrl_mask)) {if (!msr_info->host_initiated) pmu->global_status &= ~data;arch/x86/kvm/svm/pmu.c +12
-6@@ -92,12 +92,6 @@ static struct kvm_pmc *amd_rdpmc_ecx_to_pmc(struct kvm_vcpu *vcpu,return amd_pmc_idx_to_pmc (vcpu_to_pmu (vcpu), idx & ~(3u <<30 )); } -static bool amd_is_valid_msr (struct kvm_vcpu *vcpu,u32 msr) -{ -/* All MSRs refer to exactly one PMC, so msr_idx_to_pmc is enough. */ -return false; -} -static struct kvm_pmc *amd_msr_idx_to_pmc (struct kvm_vcpu *vcpu,u32 msr) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);@@ -109,6 +103,29 @@ static struct kvm_pmc *amd_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr)return pmc; } +static bool amd_is_valid_msr (struct kvm_vcpu *vcpu,u32 msr) +{ +struct kvm_pmu *pmu =vcpu_to_pmu (vcpu); + +switch (msr) { + case MSR_K7_EVNTSEL0 ... MSR_K7_PERFCTR3: +return pmu->version >0 ; + case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: +return guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE ); +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :
Quoted 8 lines
arch/x86/kvm/svm/pmu.c static struct kvm_pmc *amd_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr) · line 122
Quoted 5 lines
arch/x86/kvm/svm/pmu.c +5
-0+return pmu->version >1 ; +default : +if (msr >MSR_F15H_PERF_CTR5 && + msr <MSR_F15H_PERF_CTL0 +2 *KVM_AMD_PMC_MAX_GENERIC ) +return pmu->version >1 ;Should this be bounded by guest CPUID.80000022H:EBX[NumCorePmc] (unless host-initiated)?
Indeed, how about:
default: if (msr > MSR_F15H_PERF_CTR5 && msr < MSR_F15H_PERF_CTL0 + 2 * pmu->nr_arch_gp_counters) return pmu->version > 1;
and for host-initiated:
#define MSR_F15H_PERF_MSR_MAX \ (MSR_F15H_PERF_CTR0 + 2 * (KVM_AMD_PMC_MAX_GENERIC - 1))
kvm_{set|get}_msr_common()
case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_MSR_MAX:
if (kvm_pmu_is_valid_msr(vcpu, msr))
return kvm_pmu_set_msr(vcpu, msr_info);?
Quoted 26 lines
Quoted 26 lines
arch/x86/kvm/svm/pmu.c +16
-1+break ; + } + +return amd_msr_idx_to_pmc (vcpu, msr); +} +static int amd_pmu_get_msr (struct kvm_vcpu *vcpu,struct msr_data *msr_info) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);@@ -162,20 +179,31 @@ static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)static void amd_pmu_refresh (struct kvm_vcpu *vcpu) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu); +struct kvm_cpuid_entry2 *entry; +union cpuid_0x80000022_ebx ebx; -if (guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE )) + pmu->version =1 ; + entry =kvm_find_cpuid_entry_index (vcpu,0x80000022 ,0 ); +if (kvm_pmu_cap.version >1 && entry && (entry->eax &BIT (0 ))) { + pmu->version =2 ; + ebx.full = entry->ebx; + pmu->nr_arch_gp_counters =min3 ((unsigned int )ebx.split.num_core_pmc, + (unsigned int )kvm_pmu_cap.num_counters_gp, + (unsigned int )KVM_AMD_PMC_MAX_GENERIC );
Quoted 8 lines
arch/x86/kvm/svm/pmu.c static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) · line 196
Quoted 4 lines
arch/x86/kvm/svm/pmu.c +3
-0+ pmu->global_ctrl_mask = ~((1ull << pmu->nr_arch_gp_counters) -1 ); + pmu->global_ovf_ctrl_mask = pmu->global_ctrl_mask; + }else if (guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE )) { pmu->nr_arch_gp_counters =AMD64_NUM_COUNTERS_CORE ;The logic above doesn't seem quite right, since
guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE)promises 6 PMCs, regardless of what CPUID.80000022 says.
I would have expected the appearance of CPUID.80000022 to override PERFCTR_CORE, now I don't think it's a good idea as you do, so how about:
amd_pmu_refresh():
bool perfctr_core = guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE);
pmu->version = 1;
if (kvm_pmu_cap.version > 1)
entry = kvm_find_cpuid_entry_index(vcpu, 0x80000022, 0);
if (!perfctr_core)
pmu->nr_arch_gp_counters = AMD64_NUM_COUNTERS;
if (entry && (entry->eax & BIT(0))) {
pmu->version = 2;
ebx.full = entry->ebx;
pmu->nr_arch_gp_counters = min3((unsigned int)ebx.split.num_core_pmc,
(unsigned int)kvm_pmu_cap.num_counters_gp,
(unsigned int)KVM_AMD_PMC_MAX_GENERIC);
}
/* PERFCTR_CORE promises 6 PMCs, regardless of CPUID.80000022 */
if (perfctr_core) {
pmu->nr_arch_gp_counters = max(pmu->nr_arch_gp_counters,
AMD64_NUM_COUNTERS_CORE);
}
if (pmu->version > 1) {
pmu->global_ctrl_mask = ~((1ull << pmu->nr_arch_gp_counters) - 1);
pmu->global_ovf_ctrl_mask = pmu->global_ctrl_mask;
}?
Re: [PATCH 3/4] KVM: x86/svm/pmu: Add AMD PerfMonV2 support
Quoted 120 lines · Like Xu
Quoted 102 lines · Jim Mattson
Quoted 101 lines · Like Xu
From: Like Xu <[email protected]>
If AMD Performance Monitoring Version 2 (PerfMonV2) is detected by the guest, it can use a new scheme to manage the Core PMCs using the new global control and status registers.
In addition to benefiting from the PerfMonV2 functionality in the same way as the host (higher precision), the guest also can reduce the number of vm-exits by lowering the total number of MSRs accesses.
In terms of implementation details,
amd_is_valid_msr()is resurrected since three newly added MSRs could not be mapped to one vPMC. The possibility of emulating PerfMonV2 on the mainframe has also been eliminated for reasons of precision.Co-developed-by Sandipan Das <[email protected]>Signed-off-by Sandipan Das <[email protected]>Signed-off-by Like Xu <[email protected]>--- arch/x86/kvm/pmu.c | 6 +++++ arch/x86/kvm/svm/pmu.c | 50 +++++++++++++++++++++++++++++++++--------- arch/x86/kvm/x86.c | 11 ++++++++++ 3 files changed, 57 insertions(+), 10 deletions(-)arch/x86/kvm/pmu.c +6
-0@@ -455,12 +455,15 @@ int kvm_pmu_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : msr_info->data = pmu->global_status;return 0 ;case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : msr_info->data = pmu->global_ctrl;return 0 ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR : msr_info->data =0 ;return 0 ;default :@@ -479,12 +482,14 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS :if (msr_info->host_initiated) { pmu->global_status = data;return 0 ; }break ;/* RO MSR */ case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL :if (pmu->global_ctrl == data)return 0 ;if (kvm_valid_perf_global_ctrl (pmu, data)) {@@ -495,6 +500,7 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) }break ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :if (!(data & pmu->global_ovf_ctrl_mask)) {if (!msr_info->host_initiated) pmu->global_status &= ~data;arch/x86/kvm/svm/pmu.c +12
-6@@ -92,12 +92,6 @@ static struct kvm_pmc *amd_rdpmc_ecx_to_pmc(struct kvm_vcpu *vcpu,return amd_pmc_idx_to_pmc (vcpu_to_pmu (vcpu), idx & ~(3u <<30 )); } -static bool amd_is_valid_msr (struct kvm_vcpu *vcpu,u32 msr) -{ -/* All MSRs refer to exactly one PMC, so msr_idx_to_pmc is enough. */ -return false; -} -static struct kvm_pmc *amd_msr_idx_to_pmc (struct kvm_vcpu *vcpu,u32 msr) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);@@ -109,6 +103,29 @@ static struct kvm_pmc *amd_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr)return pmc; } +static bool amd_is_valid_msr (struct kvm_vcpu *vcpu,u32 msr) +{ +struct kvm_pmu *pmu =vcpu_to_pmu (vcpu); + +switch (msr) { + case MSR_K7_EVNTSEL0 ... MSR_K7_PERFCTR3: +return pmu->version >0 ; + case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: +return guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE ); +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :Quoted 8 lines
arch/x86/kvm/svm/pmu.c static struct kvm_pmc *amd_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr) · line 122
Quoted 5 lines
arch/x86/kvm/svm/pmu.c +5
-0+return pmu->version >1 ; +default : +if (msr >MSR_F15H_PERF_CTR5 && + msr <MSR_F15H_PERF_CTL0 +2 *KVM_AMD_PMC_MAX_GENERIC ) +return pmu->version >1 ;Should this be bounded by guest CPUID.80000022H:EBX[NumCorePmc] (unless host-initiated)?
Indeed, how about:
default: if (msr > MSR_F15H_PERF_CTR5 && msr < MSR_F15H_PERF_CTL0 + 2 * pmu->nr_arch_gp_counters) return pmu->version > 1;
Quoted 4 lines
and for host-initiated:
#define MSR_F15H_PERF_MSR_MAX \ (MSR_F15H_PERF_CTR0 + 2 * (KVM_AMD_PMC_MAX_GENERIC - 1))
I think there may be an off-by-one error here.
Quoted 59 lines
kvm_{set|get}_msr_common()
case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_MSR_MAX: if (kvm_pmu_is_valid_msr(vcpu, msr)) return kvm_pmu_set_msr(vcpu, msr_info);?
Quoted 26 lines
Quoted 26 lines
arch/x86/kvm/svm/pmu.c +16
-1+break ; + } + +return amd_msr_idx_to_pmc (vcpu, msr); +} +static int amd_pmu_get_msr (struct kvm_vcpu *vcpu,struct msr_data *msr_info) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);@@ -162,20 +179,31 @@ static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)static void amd_pmu_refresh (struct kvm_vcpu *vcpu) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu); +struct kvm_cpuid_entry2 *entry; +union cpuid_0x80000022_ebx ebx; -if (guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE )) + pmu->version =1 ; + entry =kvm_find_cpuid_entry_index (vcpu,0x80000022 ,0 ); +if (kvm_pmu_cap.version >1 && entry && (entry->eax &BIT (0 ))) { + pmu->version =2 ; + ebx.full = entry->ebx; + pmu->nr_arch_gp_counters =min3 ((unsigned int )ebx.split.num_core_pmc, + (unsigned int )kvm_pmu_cap.num_counters_gp, + (unsigned int )KVM_AMD_PMC_MAX_GENERIC );Quoted 8 lines
arch/x86/kvm/svm/pmu.c static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) · line 196
Quoted 4 lines
arch/x86/kvm/svm/pmu.c +3
-0+ pmu->global_ctrl_mask = ~((1ull << pmu->nr_arch_gp_counters) -1 ); + pmu->global_ovf_ctrl_mask = pmu->global_ctrl_mask; + }else if (guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE )) { pmu->nr_arch_gp_counters =AMD64_NUM_COUNTERS_CORE ;The logic above doesn't seem quite right, since
guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE)promises 6 PMCs, regardless of what CPUID.80000022 says.I would have expected the appearance of CPUID.80000022 to override
PERFCTR_CORE, now I don't think it's a good idea as you do, so how about:
amd_pmu_refresh():bool perfctr_core = guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE); pmu->version = 1; if (kvm_pmu_cap.version > 1) entry = kvm_find_cpuid_entry_index(vcpu, 0x80000022, 0); if (!perfctr_core) pmu->nr_arch_gp_counters = AMD64_NUM_COUNTERS; if (entry && (entry->eax & BIT(0))) { pmu->version = 2; ebx.full = entry->ebx; pmu->nr_arch_gp_counters = min3((unsigned int)ebx.split.num_core_pmc,
Quoted 8 lines
(unsigned int)kvm_pmu_cap.num_counters_gp,
(unsigned int)KVM_AMD_PMC_MAX_GENERIC);
}
/* PERFCTR_CORE promises 6 PMCs, regardless of CPUID.80000022 */
if (perfctr_core) {
pmu->nr_arch_gp_counters = max(pmu->nr_arch_gp_counters,
AMD64_NUM_COUNTERS_CORE);
}Even if X86_FEATURE_PERFCTR_CORE is clear, all AMD CPUs promise 4 PMCs.
Quoted 9 lines
if (pmu->version > 1) { pmu->global_ctrl_mask = ~((1ull << pmu->nr_arch_gp_counters) - 1); pmu->global_ovf_ctrl_mask = pmu->global_ctrl_mask; }?
Re: [PATCH 3/4] KVM: x86/svm/pmu: Add AMD PerfMonV2 support
Quoted 121 lines · Jim Mattson
Quoted 120 lines · Like Xu
Quoted 102 lines · Jim Mattson
Quoted 101 lines · Like Xu
From: Like Xu <[email protected]>
If AMD Performance Monitoring Version 2 (PerfMonV2) is detected by the guest, it can use a new scheme to manage the Core PMCs using the new global control and status registers.
In addition to benefiting from the PerfMonV2 functionality in the same way as the host (higher precision), the guest also can reduce the number of vm-exits by lowering the total number of MSRs accesses.
In terms of implementation details,
amd_is_valid_msr()is resurrected since three newly added MSRs could not be mapped to one vPMC. The possibility of emulating PerfMonV2 on the mainframe has also been eliminated for reasons of precision.Co-developed-by Sandipan Das <[email protected]>Signed-off-by Sandipan Das <[email protected]>Signed-off-by Like Xu <[email protected]>--- arch/x86/kvm/pmu.c | 6 +++++ arch/x86/kvm/svm/pmu.c | 50 +++++++++++++++++++++++++++++++++--------- arch/x86/kvm/x86.c | 11 ++++++++++ 3 files changed, 57 insertions(+), 10 deletions(-)arch/x86/kvm/pmu.c +6
-0@@ -455,12 +455,15 @@ int kvm_pmu_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : msr_info->data = pmu->global_status;return 0 ;case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : msr_info->data = pmu->global_ctrl;return 0 ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR : msr_info->data =0 ;return 0 ;default :@@ -479,12 +482,14 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS :if (msr_info->host_initiated) { pmu->global_status = data;return 0 ; }break ;/* RO MSR */ case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL :if (pmu->global_ctrl == data)return 0 ;if (kvm_valid_perf_global_ctrl (pmu, data)) {@@ -495,6 +500,7 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) }break ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :if (!(data & pmu->global_ovf_ctrl_mask)) {if (!msr_info->host_initiated) pmu->global_status &= ~data;arch/x86/kvm/svm/pmu.c +12
-6@@ -92,12 +92,6 @@ static struct kvm_pmc *amd_rdpmc_ecx_to_pmc(struct kvm_vcpu *vcpu,return amd_pmc_idx_to_pmc (vcpu_to_pmu (vcpu), idx & ~(3u <<30 )); } -static bool amd_is_valid_msr (struct kvm_vcpu *vcpu,u32 msr) -{ -/* All MSRs refer to exactly one PMC, so msr_idx_to_pmc is enough. */ -return false; -} -static struct kvm_pmc *amd_msr_idx_to_pmc (struct kvm_vcpu *vcpu,u32 msr) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);@@ -109,6 +103,29 @@ static struct kvm_pmc *amd_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr)return pmc; } +static bool amd_is_valid_msr (struct kvm_vcpu *vcpu,u32 msr) +{ +struct kvm_pmu *pmu =vcpu_to_pmu (vcpu); + +switch (msr) { + case MSR_K7_EVNTSEL0 ... MSR_K7_PERFCTR3: +return pmu->version >0 ; + case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: +return guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE ); +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :Quoted 8 lines
arch/x86/kvm/svm/pmu.c static struct kvm_pmc *amd_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr) · line 122
Quoted 5 lines
arch/x86/kvm/svm/pmu.c +5
-0+return pmu->version >1 ; +default : +if (msr >MSR_F15H_PERF_CTR5 && + msr <MSR_F15H_PERF_CTL0 +2 *KVM_AMD_PMC_MAX_GENERIC ) +return pmu->version >1 ;Should this be bounded by guest CPUID.80000022H:EBX[NumCorePmc] (unless host-initiated)?
Indeed, how about:
default: if (msr > MSR_F15H_PERF_CTR5 && msr < MSR_F15H_PERF_CTL0 + 2 * pmu->nr_arch_gp_counters) return pmu->version > 1;
Quoted 6 lines
Quoted 4 lines
and for host-initiated:
#define MSR_F15H_PERF_MSR_MAX \ (MSR_F15H_PERF_CTR0 + 2 * (KVM_AMD_PMC_MAX_GENERIC - 1))I think there may be an off-by-one error here.
If KVM_AMD_PMC_MAX_GENERIC is 6:
#define MSR_F15H_PERF_CTL 0xc0010200
#define MSR_F15H_PERF_CTL5 (MSR_F15H_PERF_CTL + 10)
#define MSR_F15H_PERF_CTR 0xc0010201
#define MSR_F15H_PERF_CTR0 MSR_F15H_PERF_CTR
#define MSR_F15H_PERF_CTR5 (MSR_F15H_PERF_CTR + 10)
Quoted 4 lines
Quoted 4 lines
kvm_{set|get}_msr_common()
case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_MSR_MAX:
the original code is "case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:",
in that case, MSR_F15H_PERF_MSR_MAX make sense, right ?
Quoted 76 lines
Quoted 56 lines
if (
kvm_pmu_is_valid_msr(vcpu, msr))
returnkvm_pmu_set_msr(vcpu, msr_info);
?Quoted 26 lines
Quoted 26 lines
arch/x86/kvm/svm/pmu.c +16
-1+break ; + } + +return amd_msr_idx_to_pmc (vcpu, msr); +} +static int amd_pmu_get_msr (struct kvm_vcpu *vcpu,struct msr_data *msr_info) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);@@ -162,20 +179,31 @@ static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)static void amd_pmu_refresh (struct kvm_vcpu *vcpu) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu); +struct kvm_cpuid_entry2 *entry; +union cpuid_0x80000022_ebx ebx; -if (guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE )) + pmu->version =1 ; + entry =kvm_find_cpuid_entry_index (vcpu,0x80000022 ,0 ); +if (kvm_pmu_cap.version >1 && entry && (entry->eax &BIT (0 ))) { + pmu->version =2 ; + ebx.full = entry->ebx; + pmu->nr_arch_gp_counters =min3 ((unsigned int )ebx.split.num_core_pmc, + (unsigned int )kvm_pmu_cap.num_counters_gp, + (unsigned int )KVM_AMD_PMC_MAX_GENERIC );Quoted 8 lines
arch/x86/kvm/svm/pmu.c static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) · line 196
Quoted 4 lines
arch/x86/kvm/svm/pmu.c +3
-0+ pmu->global_ctrl_mask = ~((1ull << pmu->nr_arch_gp_counters) -1 ); + pmu->global_ovf_ctrl_mask = pmu->global_ctrl_mask; + }else if (guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE )) { pmu->nr_arch_gp_counters =AMD64_NUM_COUNTERS_CORE ;The logic above doesn't seem quite right, since
guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE)promises 6 PMCs, regardless of what CPUID.80000022 says.I would have expected the appearance of CPUID.80000022 to override
PERFCTR_CORE, now I don't think it's a good idea as you do, so how about:
amd_pmu_refresh():bool perfctr_core = guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE); pmu->version = 1; if (kvm_pmu_cap.version > 1) entry = kvm_find_cpuid_entry_index(vcpu, 0x80000022, 0); if (!perfctr_core) pmu->nr_arch_gp_counters = AMD64_NUM_COUNTERS; if (entry && (entry->eax & BIT(0))) { pmu->version = 2; ebx.full = entry->ebx; pmu->nr_arch_gp_counters = min3((unsigned int)ebx.split.num_core_pmc,Quoted 8 lines
(unsigned int)kvm_pmu_cap.num_counters_gp, (unsigned int)KVM_AMD_PMC_MAX_GENERIC); } /* PERFCTR_CORE promises 6 PMCs, regardless of CPUID.80000022 */ if (perfctr_core) { pmu->nr_arch_gp_counters = max(pmu->nr_arch_gp_counters, AMD64_NUM_COUNTERS_CORE); }Even if
X86_FEATURE_PERFCTR_COREis clear, all AMD CPUs promise 4 PMCs.Quoted 9 lines
if (pmu->version > 1) { pmu->global_ctrl_mask = ~((1ull << pmu->nr_arch_gp_counters) - 1); pmu->global_ovf_ctrl_mask = pmu->global_ctrl_mask; }?
Re: [PATCH 3/4] KVM: x86/svm/pmu: Add AMD PerfMonV2 support
Quoted 141 lines · Like Xu
Quoted 121 lines · Jim Mattson
Quoted 120 lines · Like Xu
Quoted 102 lines · Jim Mattson
Quoted 101 lines · Like Xu
From: Like Xu <[email protected]>
If AMD Performance Monitoring Version 2 (PerfMonV2) is detected by the guest, it can use a new scheme to manage the Core PMCs using the new global control and status registers.
In addition to benefiting from the PerfMonV2 functionality in the same way as the host (higher precision), the guest also can reduce the number of vm-exits by lowering the total number of MSRs accesses.
In terms of implementation details,
amd_is_valid_msr()is resurrected since three newly added MSRs could not be mapped to one vPMC. The possibility of emulating PerfMonV2 on the mainframe has also been eliminated for reasons of precision.Co-developed-by Sandipan Das <[email protected]>Signed-off-by Sandipan Das <[email protected]>Signed-off-by Like Xu <[email protected]>--- arch/x86/kvm/pmu.c | 6 +++++ arch/x86/kvm/svm/pmu.c | 50 +++++++++++++++++++++++++++++++++--------- arch/x86/kvm/x86.c | 11 ++++++++++ 3 files changed, 57 insertions(+), 10 deletions(-)arch/x86/kvm/pmu.c +6
-0@@ -455,12 +455,15 @@ int kvm_pmu_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : msr_info->data = pmu->global_status;return 0 ;case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : msr_info->data = pmu->global_ctrl;return 0 ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR : msr_info->data =0 ;return 0 ;default :@@ -479,12 +482,14 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)switch (msr) {case MSR_CORE_PERF_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS :if (msr_info->host_initiated) { pmu->global_status = data;return 0 ; }break ;/* RO MSR */ case MSR_CORE_PERF_GLOBAL_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL :if (pmu->global_ctrl == data)return 0 ;if (kvm_valid_perf_global_ctrl (pmu, data)) {@@ -495,6 +500,7 @@ int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) }break ;case MSR_CORE_PERF_GLOBAL_OVF_CTRL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :if (!(data & pmu->global_ovf_ctrl_mask)) {if (!msr_info->host_initiated) pmu->global_status &= ~data;arch/x86/kvm/svm/pmu.c +12
-6@@ -92,12 +92,6 @@ static struct kvm_pmc *amd_rdpmc_ecx_to_pmc(struct kvm_vcpu *vcpu,return amd_pmc_idx_to_pmc (vcpu_to_pmu (vcpu), idx & ~(3u <<30 )); } -static bool amd_is_valid_msr (struct kvm_vcpu *vcpu,u32 msr) -{ -/* All MSRs refer to exactly one PMC, so msr_idx_to_pmc is enough. */ -return false; -} -static struct kvm_pmc *amd_msr_idx_to_pmc (struct kvm_vcpu *vcpu,u32 msr) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);@@ -109,6 +103,29 @@ static struct kvm_pmc *amd_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr)return pmc; } +static bool amd_is_valid_msr (struct kvm_vcpu *vcpu,u32 msr) +{ +struct kvm_pmu *pmu =vcpu_to_pmu (vcpu); + +switch (msr) { + case MSR_K7_EVNTSEL0 ... MSR_K7_PERFCTR3: +return pmu->version >0 ; + case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: +return guest_cpuid_has (vcpu,X86_FEATURE_PERFCTR_CORE ); +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS : +case MSR_AMD64_PERF_CNTR_GLOBAL_CTL : +case MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR :Quoted 8 lines
arch/x86/kvm/svm/pmu.c static struct kvm_pmc *amd_msr_idx_to_pmc(struct kvm_vcpu *vcpu, u32 msr) · line 122
Quoted 5 lines
arch/x86/kvm/svm/pmu.c +5
-0+return pmu->version >1 ; +default : +if (msr >MSR_F15H_PERF_CTR5 && + msr <MSR_F15H_PERF_CTL0 +2 *KVM_AMD_PMC_MAX_GENERIC ) +return pmu->version >1 ;Should this be bounded by guest CPUID.80000022H:EBX[NumCorePmc] (unless host-initiated)?
Indeed, how about:
default: if (msr > MSR_F15H_PERF_CTR5 && msr < MSR_F15H_PERF_CTL0 + 2 * pmu->nr_arch_gp_counters) return pmu->version > 1;Quoted 6 lines
Quoted 4 lines
and for host-initiated:
#define MSR_F15H_PERF_MSR_MAX \ (MSR_F15H_PERF_CTR0 + 2 * (KVM_AMD_PMC_MAX_GENERIC - 1))I think there may be an off-by-one error here.
If
KVM_AMD_PMC_MAX_GENERICis 6:#define
MSR_F15H_PERF_CTL0xc0010200
#defineMSR_F15H_PERF_CTL5(MSR_F15H_PERF_CTL+ 10)#define MSR_F15H_PERF_CTR 0xc0010201 #define MSR_F15H_PERF_CTR0 MSR_F15H_PERF_CTR #define MSR_F15H_PERF_CTR5 (MSR_F15H_PERF_CTR + 10)Quoted 2 lines
Quoted 2 lines
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
kvm_{set|get}_msr_common()
case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_MSR_MAX:the original code is "case
MSR_F15H_PERF_CTL0...MSR_F15H_PERF_CTR5:",in that case,
MSR_F15H_PERF_MSR_MAXmake sense, right ?
Right. I was misreading the definition.
PATCH 1/2 x86/pmu: Update rdpmc testcase to cover #GP and emulation path
In reply to Like XuFrom: Like Xu <[email protected]>
Specifying an unsupported PMC encoding will cause a #GP(0). All testcases should be passed when the KVM_FEP prefix is added.
--- lib/x86/processor.h | 5 ++++- x86/pmu.c | 13 +++++++++++++ 2 files changed, 17 insertions(+), 1 deletion(-)
lib/x86/processor.h +4 -1
@@ -441,7 +441,10 @@ static inline int wrmsr_safe(u32 index, u64 val)static inline uint64_t rdpmc (uint32_t index) {uint32_t a, d; - asmvolatile ("rdpmc" :"=a" (a),"=d" (d) :"c" (index)); +if (is_fep_available ()) + asmvolatile (KVM_FEP "rdpmc" :"=a" (a),"=d" (d) :"c" (index)); +else + asmvolatile ("rdpmc" :"=a" (a),"=d" (d) :"c" (index));return a | ((uint64_t )d <<32 ); }
x86/pmu.c +13 -0
@@ -758,12 +758,25 @@ static bool pmu_is_detected(void)return detect_intel_pmu (); } +static void rdpmc_unsupported_counter (void *data) +{ +rdpmc (64 ); +} + +static void check_rdpmc_cause_gp (void ) +{ +report (test_for_exception (GP_VECTOR , rdpmc_unsupported_counter,NULL ), +"rdpmc with invalid PMC index raises #GP" ); +} +int main (int ac,char **av) {setup_vm ();handle_irq (PC_VECTOR , cnt_overflow); buf =malloc (N *64 ); +check_rdpmc_cause_gp (); +if (!pmu_is_detected ())return report_summary ();
Signature · 1 line
2.37.3
Sean Christopherson
maintainerRe: [kvm-unit-tests PATCH 1/2] x86/pmu: Update rdpmc testcase to cover #GP and emulation path
Quoted 16 lines · Like Xu
From: Like Xu <[email protected]>
Specifying an unsupported PMC encoding will cause a #GP(0). All testcases should be passed when the
KVM_FEPprefix is added.Signed-off-by Like Xu <[email protected]>--- lib/x86/processor.h | 5 ++++- x86/pmu.c | 13 +++++++++++++ 2 files changed, 17 insertions(+), 1 deletion(-)lib/x86/processor.h +0
-0
lib/x86/processor.h static inline int wrmsr_safe(u32 index, u64 val) · line 447
Quoted 8 lines
lib/x86/processor.h +4
-1@@ -441,7 +441,10 @@ static inline int wrmsr_safe(u32 index, u64 val)static inline uint64_t rdpmc (uint32_t index) {uint32_t a, d; - asmvolatile ("rdpmc" :"=a" (a),"=d" (d) :"c" (index)); +if (is_fep_available ()) + asmvolatile (KVM_FEP "rdpmc" :"=a" (a),"=d" (d) :"c" (index)); +else + asmvolatile ("rdpmc" :"=a" (a),"=d" (d) :"c" (index));
Hmm, not sure how I feel about the idea of always use FEP in a common helper when it's available. Part of me likes the idea, but part of me is worried that it will cause confusion due to not being explicit.
Unless there's a pressing need to force emulation, let's punt the FEP stuff for now. More below.
Quoted 8 lines
lib/x86/processor.h +0
-0return a | ((uint64_t )d <<32 ); }x86/pmu.c +0
-0
x86/pmu.c static bool pmu_is_detected(void) · line 766
Quoted 9 lines
x86/pmu.c +6
-0@@ -758,12 +758,25 @@ static bool pmu_is_detected(void)return detect_intel_pmu (); } +static void rdpmc_unsupported_counter (void *data) +{ +rdpmc (64 ); +} + +static void check_rdpmc_cause_gp (void )
Maybe check_invalid_rdpmc_gp()? There are multiple reasons RDPMC can #GP, the one that is being relied on to guarantee #GP is specifically that the PMC is invalid.
dd
x86/pmu.c static bool pmu_is_detected(void) · line 768
Quoted 2 lines
x86/pmu.c +2
-0+{ + report(test_for_exception(GP_VECTOR, rdpmc_unsupported_counter, NULL),
I'd really like to move away from test_for_exception() and use ASM_TRY(). Ignoring FEP for the moment, the most extensible solution is to provide a safe variant:
static inline int rdpmc_safe(u32 index, uint64_t *val)
{
uint32_t a, d;
asm volatile (ASM_TRY("1f")
"rdpmc"
: "=a"(a), "=d"(d) : "c"(index));
*val = (uint64_t)a | ((uint64_t)d << 32);
return exception_vector();
}static inline uint64_t rdpmc(uint32_t index)
{
uint64_t val;
int vector = rdpmc_safe(index, &val);
assert_msg(!vector, "Unexpected %s on RDPMC(%d)",
exception_mnemonic(vector), index);
return val;
}For long-term emulation validation, the best idea I have at this point is to do add a config knob to opt-in to using FEP in _all_ common helpers (where "all" means everything KVM actually emulates). It'd take some macro magic, but it'd be easier to maintain (no need to have two paths in every helper) and would be controllable.
Quoted 17 lines
x86/pmu.c +5
-0+ "rdpmc with invalid PMC index raises #GP"); +} +int main (int ac,char **av) {setup_vm ();handle_irq (PC_VECTOR , cnt_overflow); buf =malloc (N *64 ); +check_rdpmc_cause_gp (); +if (!pmu_is_detected ())return report_summary ();Signature · 1 line
2.37.3
Like Xu
AppliedRe: [kvm-unit-tests PATCH 1/2] x86/pmu: Update rdpmc testcase to cover #GP and emulation path
Quoted 30 lines · Sean Christopherson
Quoted 16 lines · Like Xu
From: Like Xu <[email protected]>
Specifying an unsupported PMC encoding will cause a #GP(0). All testcases should be passed when the
KVM_FEPprefix is added.Signed-off-by Like Xu <[email protected]>--- lib/x86/processor.h | 5 ++++- x86/pmu.c | 13 +++++++++++++ 2 files changed, 17 insertions(+), 1 deletion(-)lib/x86/processor.h +0
-0lib/x86/processor.h static inline int wrmsr_safe(u32 index, u64 val) · line 447
Quoted 8 lines
lib/x86/processor.h +4
-1@@ -441,7 +441,10 @@ static inline int wrmsr_safe(u32 index, u64 val)static inline uint64_t rdpmc (uint32_t index) {uint32_t a, d; - asmvolatile ("rdpmc" :"=a" (a),"=d" (d) :"c" (index)); +if (is_fep_available ()) + asmvolatile (KVM_FEP "rdpmc" :"=a" (a),"=d" (d) :"c" (index)); +else + asmvolatile ("rdpmc" :"=a" (a),"=d" (d) :"c" (index));Hmm, not sure how I feel about the idea of always use FEP in a common helper when it's available. Part of me likes the idea, but part of me is worried that it will cause confusion due to not being explicit.
Quoted 2 lines
Unless there's a pressing need to force emulation, let's punt the FEP stuff for now. More below.
Some security researchers are very interested in these corners.
To my limited testing, most KVM emulation code (at least arch/x86/kvm/emulate.c) are not adequately covered by test cases, and perhaps some will move them to the user space.
Quoted 12 lines
Quoted 12 lines
lib/x86/processor.h +0
-0return a | ((uint64_t )d <<32 ); }x86/pmu.c +0
-0@@ -758,12 +758,25 @@ static bool pmu_is_detected(void)return detect_intel_pmu (); }
Quoted 10 lines
x86/pmu.c static bool pmu_is_detected(void) · line 766
Quoted 6 lines
x86/pmu.c +6
-0+static void rdpmc_unsupported_counter (void *data) +{ +rdpmc (64 ); +} + +static void check_rdpmc_cause_gp (void )Maybe
check_invalid_rdpmc_gp()? There are multiple reasons RDPMC can #GP, the one that is being relied on to guarantee #GP is specifically that the PMC is invalid.
Applied.
Quoted 1 line
dd
p, :D
Quoted 7 lines
x86/pmu.c static bool pmu_is_detected(void) · line 768
Quoted 3 lines
x86/pmu.c +2
-0+{ + report(test_for_exception(GP_VECTOR, rdpmc_unsupported_counter, NULL),I'd really like to move away from
test_for_exception()and useASM_TRY(). Ignoring FEP for the moment, the most extensible solution is to provide a safe variant:
Quoted 7 lines
static inline int
rdpmc_safe(u32 index, uint64_t *val){ uint32_t a, d; asm volatile (ASM_TRY("1f") "rdpmc" : "=a"(a), "=d"(d) : "c"(index));
asm volatile (ASM_TRY("1f")
"rdpmc\n\t"
"1:"
: "=a"(a), "=d"(d) : "c"(index) : "memory");, otherwise the compiler will complain.
Quoted 5 lines
*val = (uint64_t)a | ((uint64_t)d << 32); return exception_vector(); }static inline
uint64_trdpmc(uint32_t index)
Quoted 8 lines
{
uint64_t val;
int vector = rdpmc_safe(index, &val);
assert_msg(!vector, "Unexpected %s on RDPMC(%d)",
exception_mnemonic(vector), index);
return val;
}Applied.
Quoted 7 lines
For long-term emulation validation, the best idea I have at this point is to do add a config knob to opt-in to using FEP in _all_ common helpers (where "all" means everything KVM actually emulates). It'd take some macro magic, but it'd be easier to maintain (no need to have two paths in every helper) and would be controllable.
With both hands up in favour. Leave it to you, as this involves a wider change.
Quoted 18 lines
Quoted 18 lines
x86/pmu.c +5
-0+ "rdpmc with invalid PMC index raises #GP"); +} +int main (int ac,char **av) {setup_vm ();handle_irq (PC_VECTOR , cnt_overflow); buf =malloc (N *64 ); +check_rdpmc_cause_gp (); +if (!pmu_is_detected ())return report_summary ();Signature · 1 line
2.37.3
PATCH 4/4 KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
In reply to Like XuFrom: Sandipan Das <[email protected]>
CPUID leaf 0x80000022 i.e. ExtPerfMonAndDbg advertises some new performance monitoring features for AMD processors.
Bit 0 of EAX indicates support for Performance Monitoring Version 2 (PerfMonV2) features. If found to be set during PMU initialization, the EBX bits of the same CPUID function can be used to determine the number of available PMCs for different PMU types.
Expose the relevant bits via KVM_GET_SUPPORTED_CPUID so that guests can make use of the PerfMonV2 features.
--- arch/x86/include/asm/perf_event.h | 8 ++++++++ arch/x86/kvm/cpuid.c | 21 ++++++++++++++++++++- 2 files changed, 28 insertions(+), 1 deletion(-)
arch/x86/include/asm/perf_event.h +8 -0
@@ -214,6 +214,14 @@ union cpuid_0x80000022_ebx {unsigned int full; }; +union cpuid_0x80000022_eax { +struct { +/* Performance Monitoring Version 2 Supported */ +unsigned int perfmon_v2:1 ; + } split; +unsigned int full; +}; +struct x86_pmu_capability {int version;int num_counters_gp;
arch/x86/kvm/cpuid.c +20 -1
@@ -1094,7 +1094,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) entry->edx =0 ;break ;case 0x80000000 : - entry->eax =min (entry->eax,0x80000021 ); + entry->eax =min (entry->eax,0x80000022 );/* * Serializing LFENCE is reported in a multitude of ways, and * NullSegClearsBase is not reported in CPUID on Zen2; help
@@ -1203,6 +1203,25 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)if (!static_cpu_has_bug (X86_BUG_NULL_SEG )) entry->eax |=BIT (6 );break ; +/* AMD Extended Performance Monitoring and Debug */ +case 0x80000022 : { +union cpuid_0x80000022_eax eax; +union cpuid_0x80000022_ebx ebx; + + entry->eax = entry->ebx = entry->ecx = entry->edx =0 ; +if (!enable_pmu) +break ; + +if (kvm_pmu_cap.version >1 ) { +/* AMD PerfMon is only supported up to V2 in the KVM. */ + eax.split.perfmon_v2 =1 ; + ebx.split.num_core_pmc =min (kvm_pmu_cap.num_counters_gp, +KVM_AMD_PMC_MAX_GENERIC ); + } + entry->eax = eax.full; + entry->ebx = ebx.full; +break ; + }/*Add support for Centaur's CPUID instruction*/ case 0xC0000000 :/*Just support up to 0xC0000004 now*/
Signature · 1 line
2.37.3
Re: [PATCH 4/4] KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
Quoted 63 lines · Like Xu
From: Sandipan Das <[email protected]>
CPUID leaf
0x80000022i.e. ExtPerfMonAndDbg advertises some new performance monitoring features for AMD processors.Bit 0 of EAX indicates support for Performance Monitoring Version 2 (PerfMonV2) features. If found to be set during PMU initialization, the EBX bits of the same CPUID function can be used to determine the number of available PMCs for different PMU types.
Expose the relevant bits via
KVM_GET_SUPPORTED_CPUIDso that guests can make use of the PerfMonV2 features.Co-developed-by Like Xu <[email protected]>Signed-off-by Like Xu <[email protected]>Signed-off-by Sandipan Das <[email protected]>--- arch/x86/include/asm/perf_event.h | 8 ++++++++ arch/x86/kvm/cpuid.c | 21 ++++++++++++++++++++- 2 files changed, 28 insertions(+), 1 deletion(-)arch/x86/include/asm/perf_event.h +8
-0@@ -214,6 +214,14 @@ union cpuid_0x80000022_ebx {unsigned int full; }; +union cpuid_0x80000022_eax { +struct { +/* Performance Monitoring Version 2 Supported */ +unsigned int perfmon_v2:1 ; + } split; +unsigned int full; +}; +struct x86_pmu_capability {int version;int num_counters_gp;arch/x86/kvm/cpuid.c +5
-1@@ -1094,7 +1094,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) entry->edx =0 ;break ;case 0x80000000 : - entry->eax =min (entry->eax,0x80000021 ); + entry->eax =min (entry->eax,0x80000022 );/* * Serializing LFENCE is reported in a multitude of ways, and * NullSegClearsBase is not reported in CPUID on Zen2; help @@ -1203,6 +1203,25 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)if (!static_cpu_has_bug (X86_BUG_NULL_SEG )) entry->eax |=BIT (6 );break ; +/* AMD Extended Performance Monitoring and Debug */ +case 0x80000022 : { +union cpuid_0x80000022_eax eax; +union cpuid_0x80000022_ebx ebx;
arch/x86/kvm/cpuid.c static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) · line 1219
Quoted 10 lines
arch/x86/kvm/cpuid.c +10
-0+ + entry->eax = entry->ebx = entry->ecx = entry->edx =0 ; +if (!enable_pmu) +break ; + +if (kvm_pmu_cap.version >1 ) { +/* AMD PerfMon is only supported up to V2 in the KVM. */ + eax.split.perfmon_v2 =1 ; + ebx.split.num_core_pmc =min (kvm_pmu_cap.num_counters_gp, +KVM_AMD_PMC_MAX_GENERIC );
Note that the number of core PMCs has to be at least 6 if guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE). I suppose this leaf could claim fewer, but the first 6 PMCs must work, per the v1 PMU spec. That is, software that knows about PERFCTR_CORE, but not about PMU v2, can rightfully expect 6 PMCs.
Quoted 11 lines
arch/x86/kvm/cpuid.c +5
-1+ } + entry->eax = eax.full; + entry->ebx = ebx.full; +break ; + }/*Add support for Centaur's CPUID instruction*/ case 0xC0000000 :/*Just support up to 0xC0000004 now*/ --2.37.3
Re: [PATCH 4/4] KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
Quoted 72 lines · Jim Mattson
Quoted 71 lines · Like Xu
From: Sandipan Das <[email protected]>
CPUID leaf
0x80000022i.e. ExtPerfMonAndDbg advertises some new performance monitoring features for AMD processors.Bit 0 of EAX indicates support for Performance Monitoring Version 2 (PerfMonV2) features. If found to be set during PMU initialization, the EBX bits of the same CPUID function can be used to determine the number of available PMCs for different PMU types.
Expose the relevant bits via
KVM_GET_SUPPORTED_CPUIDso that guests can make use of the PerfMonV2 features.Co-developed-by Like Xu <[email protected]>Signed-off-by Like Xu <[email protected]>Signed-off-by Sandipan Das <[email protected]>--- arch/x86/include/asm/perf_event.h | 8 ++++++++ arch/x86/kvm/cpuid.c | 21 ++++++++++++++++++++- 2 files changed, 28 insertions(+), 1 deletion(-)arch/x86/include/asm/perf_event.h +8
-0@@ -214,6 +214,14 @@ union cpuid_0x80000022_ebx {unsigned int full; }; +union cpuid_0x80000022_eax { +struct { +/* Performance Monitoring Version 2 Supported */ +unsigned int perfmon_v2:1 ; + } split; +unsigned int full; +}; +struct x86_pmu_capability {int version;int num_counters_gp;arch/x86/kvm/cpuid.c +13
-1@@ -1094,7 +1094,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) entry->edx =0 ;break ;case 0x80000000 : - entry->eax =min (entry->eax,0x80000021 ); + entry->eax =min (entry->eax,0x80000022 );/* * Serializing LFENCE is reported in a multitude of ways, and * NullSegClearsBase is not reported in CPUID on Zen2; help @@ -1203,6 +1203,25 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)if (!static_cpu_has_bug (X86_BUG_NULL_SEG )) entry->eax |=BIT (6 );break ; +/* AMD Extended Performance Monitoring and Debug */ +case 0x80000022 : { +union cpuid_0x80000022_eax eax; +union cpuid_0x80000022_ebx ebx; + + entry->eax = entry->ebx = entry->ecx = entry->edx =0 ; +if (!enable_pmu) +break ; + +if (kvm_pmu_cap.version >1 ) { +/* AMD PerfMon is only supported up to V2 in the KVM. */ + eax.split.perfmon_v2 =1 ;
Quoted 8 lines
arch/x86/kvm/cpuid.c static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) · line 1219
Quoted 2 lines
arch/x86/kvm/cpuid.c +2
-0+ ebx.split.num_core_pmc =min (kvm_pmu_cap.num_counters_gp, +KVM_AMD_PMC_MAX_GENERIC );Note that the number of core PMCs has to be at least 6 if
guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE). I suppose this leaf could claim fewer, but the first 6 PMCs must work, per the v1 PMU spec. That is, software that knows aboutPERFCTR_CORE, but not about PMU v2, can rightfully expect 6 PMCs.
I thought the NumCorePmc number would only make sense if CPUID.80000022.eax.perfmon_v2 bit was present, but considering that the user space is perfectly fine with just configuring the NumCorePmc number without setting perfmon_v2 bit at all, so how about:
/* AMD Extended Performance Monitoring and Debug */
case 0x80000022: {
union cpuid_0x80000022_eax eax;
union cpuid_0x80000022_ebx ebx;
bool perfctr_core;
entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
if (!enable_pmu)
break;
perfctr_core = kvm_cpu_cap_has(X86_FEATURE_PERFCTR_CORE);
if (!perfctr_core)
ebx.split.num_core_pmc = AMD64_NUM_COUNTERS;
if (kvm_pmu_cap.version > 1) {
/* AMD PerfMon is only supported up to V2 in the KVM. */
eax.split.perfmon_v2 = 1;
ebx.split.num_core_pmc = min(kvm_pmu_cap.num_counters_gp,
KVM_AMD_PMC_MAX_GENERIC);
}
if (perfctr_core) {
ebx.split.num_core_pmc = max(ebx.split.num_core_pmc,
AMD64_NUM_COUNTERS_CORE);
}
entry->eax = eax.full;
entry->ebx = ebx.full;
break;
}?
Once 0x80000022 appears, ebx.split.num_core_pmc will report only the real "Number of Core Performance Counters" regardless of perfmon_v2.
Quoted 13 lines
Quoted 13 lines
arch/x86/kvm/cpuid.c +5
-1+ } + entry->eax = eax.full; + entry->ebx = ebx.full; +break ; + }/*Add support for Centaur's CPUID instruction*/ case 0xC0000000 :/*Just support up to 0xC0000004 now*/ --2.37.3
Re: [PATCH 4/4] KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
Quoted 83 lines · Like Xu
Quoted 72 lines · Jim Mattson
Quoted 71 lines · Like Xu
From: Sandipan Das <[email protected]>
CPUID leaf
0x80000022i.e. ExtPerfMonAndDbg advertises some new performance monitoring features for AMD processors.Bit 0 of EAX indicates support for Performance Monitoring Version 2 (PerfMonV2) features. If found to be set during PMU initialization, the EBX bits of the same CPUID function can be used to determine the number of available PMCs for different PMU types.
Expose the relevant bits via
KVM_GET_SUPPORTED_CPUIDso that guests can make use of the PerfMonV2 features.Co-developed-by Like Xu <[email protected]>Signed-off-by Like Xu <[email protected]>Signed-off-by Sandipan Das <[email protected]>--- arch/x86/include/asm/perf_event.h | 8 ++++++++ arch/x86/kvm/cpuid.c | 21 ++++++++++++++++++++- 2 files changed, 28 insertions(+), 1 deletion(-)arch/x86/include/asm/perf_event.h +8
-0@@ -214,6 +214,14 @@ union cpuid_0x80000022_ebx {unsigned int full; }; +union cpuid_0x80000022_eax { +struct { +/* Performance Monitoring Version 2 Supported */ +unsigned int perfmon_v2:1 ; + } split; +unsigned int full; +}; +struct x86_pmu_capability {int version;int num_counters_gp;arch/x86/kvm/cpuid.c +13
-1@@ -1094,7 +1094,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) entry->edx =0 ;break ;case 0x80000000 : - entry->eax =min (entry->eax,0x80000021 ); + entry->eax =min (entry->eax,0x80000022 );/* * Serializing LFENCE is reported in a multitude of ways, and * NullSegClearsBase is not reported in CPUID on Zen2; help @@ -1203,6 +1203,25 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)if (!static_cpu_has_bug (X86_BUG_NULL_SEG )) entry->eax |=BIT (6 );break ; +/* AMD Extended Performance Monitoring and Debug */ +case 0x80000022 : { +union cpuid_0x80000022_eax eax; +union cpuid_0x80000022_ebx ebx; + + entry->eax = entry->ebx = entry->ecx = entry->edx =0 ; +if (!enable_pmu) +break ; + +if (kvm_pmu_cap.version >1 ) { +/* AMD PerfMon is only supported up to V2 in the KVM. */ + eax.split.perfmon_v2 =1 ;Quoted 8 lines
arch/x86/kvm/cpuid.c static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) · line 1219
Quoted 2 lines
arch/x86/kvm/cpuid.c +2
-0+ ebx.split.num_core_pmc =min (kvm_pmu_cap.num_counters_gp, +KVM_AMD_PMC_MAX_GENERIC );Note that the number of core PMCs has to be at least 6 if
guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE). I suppose this leaf could claim fewer, but the first 6 PMCs must work, per the v1 PMU spec. That is, software that knows aboutPERFCTR_CORE, but not about PMU v2, can rightfully expect 6 PMCs.
Quoted 5 lines
I thought the NumCorePmc number would only make sense if CPUID.80000022.eax.perfmon_v2
bit was present, but considering that the user space is perfectly fine with just configuring the
NumCorePmc number without settingperfmon_v2bit at all, so how about:
CPUID.80000022H might only make sense if X86_FEATURE_PERFCTR_CORE is present. It's hard to know in the absence of documentation.
Quoted 15 lines
/* AMD Extended Performance Monitoring and Debug */
case 0x80000022: {
union cpuid_0x80000022_eax eax;
union cpuid_0x80000022_ebx ebx;
bool perfctr_core;
entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
if (!enable_pmu)
break;
perfctr_core = kvm_cpu_cap_has(X86_FEATURE_PERFCTR_CORE);
if (!perfctr_core)
ebx.split.num_core_pmc = AMD64_NUM_COUNTERS;
if (kvm_pmu_cap.version > 1) {
/* AMD PerfMon is only supported up to V2 in the KVM. */Quoted 8 lines
eax.split.perfmon_v2 = 1;
ebx.split.num_core_pmc = min(kvm_pmu_cap.num_counters_gp,
KVM_AMD_PMC_MAX_GENERIC);
}
if (perfctr_core) {
ebx.split.num_core_pmc = max(ebx.split.num_core_pmc,
AMD64_NUM_COUNTERS_CORE);
}This still isn't quite right. All AMD CPUs must support a minimum of 4 PMCs.
Quoted 24 lines
entry->eax = eax.full; entry->ebx = ebx.full; break; }?
Once
0x80000022appears, ebx.split.num_core_pmc will report only the real "Number of Core Performance Counters" regardless ofperfmon_v2.Quoted 13 lines
Quoted 13 lines
arch/x86/kvm/cpuid.c +5
-1+ } + entry->eax = eax.full; + entry->ebx = ebx.full; +break ; + }/*Add support for Centaur's CPUID instruction*/ case 0xC0000000 :/*Just support up to 0xC0000004 now*/ --2.37.3
Re: [PATCH 4/4] KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
Quoted 84 lines · Jim Mattson
Quoted 83 lines · Like Xu
Quoted 80 lines · Jim Mattson
Quoted 63 lines · Like Xu
From: Sandipan Das <[email protected]>
CPUID leaf
0x80000022i.e. ExtPerfMonAndDbg advertises some new performance monitoring features for AMD processors.Bit 0 of EAX indicates support for Performance Monitoring Version 2 (PerfMonV2) features. If found to be set during PMU initialization, the EBX bits of the same CPUID function can be used to determine the number of available PMCs for different PMU types.
Expose the relevant bits via
KVM_GET_SUPPORTED_CPUIDso that guests can make use of the PerfMonV2 features.Co-developed-by Like Xu <[email protected]>Signed-off-by Like Xu <[email protected]>Signed-off-by Sandipan Das <[email protected]>--- arch/x86/include/asm/perf_event.h | 8 ++++++++ arch/x86/kvm/cpuid.c | 21 ++++++++++++++++++++- 2 files changed, 28 insertions(+), 1 deletion(-)arch/x86/include/asm/perf_event.h +8
-0@@ -214,6 +214,14 @@ union cpuid_0x80000022_ebx {unsigned int full; }; +union cpuid_0x80000022_eax { +struct { +/* Performance Monitoring Version 2 Supported */ +unsigned int perfmon_v2:1 ; + } split; +unsigned int full; +}; +struct x86_pmu_capability {int version;int num_counters_gp;arch/x86/kvm/cpuid.c +5
-1@@ -1094,7 +1094,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) entry->edx =0 ;break ;case 0x80000000 : - entry->eax =min (entry->eax,0x80000021 ); + entry->eax =min (entry->eax,0x80000022 );/* * Serializing LFENCE is reported in a multitude of ways, and * NullSegClearsBase is not reported in CPUID on Zen2; help @@ -1203,6 +1203,25 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)if (!static_cpu_has_bug (X86_BUG_NULL_SEG )) entry->eax |=BIT (6 );break ; +/* AMD Extended Performance Monitoring and Debug */ +case 0x80000022 : { +union cpuid_0x80000022_eax eax; +union cpuid_0x80000022_ebx ebx;arch/x86/kvm/cpuid.c static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) · line 1219
Quoted 10 lines
arch/x86/kvm/cpuid.c +10
-0+ + entry->eax = entry->ebx = entry->ecx = entry->edx =0 ; +if (!enable_pmu) +break ; + +if (kvm_pmu_cap.version >1 ) { +/* AMD PerfMon is only supported up to V2 in the KVM. */ + eax.split.perfmon_v2 =1 ; + ebx.split.num_core_pmc =min (kvm_pmu_cap.num_counters_gp, +KVM_AMD_PMC_MAX_GENERIC );Note that the number of core PMCs has to be at least 6 if
guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE). I suppose this leaf could claim fewer, but the first 6 PMCs must work, per the v1 PMU spec. That is, software that knows aboutPERFCTR_CORE, but not about PMU v2, can rightfully expect 6 PMCs.
Quoted 8 lines
Quoted 5 lines
I thought the NumCorePmc number would only make sense if CPUID.80000022.eax.perfmon_v2
bit was present, but considering that the user space is perfectly fine with just configuring the
NumCorePmc number without settingperfmon_v2bit at all, so how about:CPUID.80000022H might only make sense if
X86_FEATURE_PERFCTR_COREis present. It's hard to know in the absence of documentation.
Whenever this happens, we may always leave the definition of behavior to the hypervisor.
Quoted 25 lines
Quoted 16 lines
/* AMD Extended Performance Monitoring and Debug */ case 0x80000022: { union cpuid_0x80000022_eax eax; union cpuid_0x80000022_ebx ebx; bool perfctr_core; entry->eax = entry->ebx = entry->ecx = entry->edx = 0; if (!enable_pmu) break; perfctr_core = kvm_cpu_cap_has(X86_FEATURE_PERFCTR_CORE); if (!perfctr_core) ebx.split.num_core_pmc = AMD64_NUM_COUNTERS; if (kvm_pmu_cap.version > 1) { /* AMD PerfMon is only supported up to V2 in the KVM. */Quoted 8 lines
eax.split.perfmon_v2 = 1; ebx.split.num_core_pmc = min(kvm_pmu_cap.num_counters_gp, KVM_AMD_PMC_MAX_GENERIC); } if (perfctr_core) { ebx.split.num_core_pmc = max(ebx.split.num_core_pmc, AMD64_NUM_COUNTERS_CORE); }
Quoted 1 line
This still isn't quite right. All AMD CPUs must support a minimum of 4 PMCs.
K7 at least. I could not confirm that all antique AMD CPUs have 4 counters w/o perfctr_core.
Quoted 25 lines
Quoted 25 lines
entry->eax = eax.full; entry->ebx = ebx.full; break; }?
Once
0x80000022appears, ebx.split.num_core_pmc will report only the real "Number of Core Performance Counters" regardless ofperfmon_v2.Quoted 13 lines
Quoted 13 lines
arch/x86/kvm/cpuid.c +5
-1+ } + entry->eax = eax.full; + entry->ebx = ebx.full; +break ; + }/*Add support for Centaur's CPUID instruction*/ case 0xC0000000 :/*Just support up to 0xC0000004 now*/ --2.37.3
Re: [PATCH 4/4] KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
Quoted 92 lines · Like Xu
Quoted 90 lines · Jim Mattson
Quoted 88 lines · Like Xu
Quoted 72 lines · Jim Mattson
Quoted 71 lines · Like Xu
From: Sandipan Das <[email protected]>
CPUID leaf
0x80000022i.e. ExtPerfMonAndDbg advertises some new performance monitoring features for AMD processors.Bit 0 of EAX indicates support for Performance Monitoring Version 2 (PerfMonV2) features. If found to be set during PMU initialization, the EBX bits of the same CPUID function can be used to determine the number of available PMCs for different PMU types.
Expose the relevant bits via
KVM_GET_SUPPORTED_CPUIDso that guests can make use of the PerfMonV2 features.Co-developed-by Like Xu <[email protected]>Signed-off-by Like Xu <[email protected]>Signed-off-by Sandipan Das <[email protected]>--- arch/x86/include/asm/perf_event.h | 8 ++++++++ arch/x86/kvm/cpuid.c | 21 ++++++++++++++++++++- 2 files changed, 28 insertions(+), 1 deletion(-)arch/x86/include/asm/perf_event.h +8
-0@@ -214,6 +214,14 @@ union cpuid_0x80000022_ebx {unsigned int full; }; +union cpuid_0x80000022_eax { +struct { +/* Performance Monitoring Version 2 Supported */ +unsigned int perfmon_v2:1 ; + } split; +unsigned int full; +}; +struct x86_pmu_capability {int version;int num_counters_gp;arch/x86/kvm/cpuid.c +13
-1@@ -1094,7 +1094,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) entry->edx =0 ;break ;case 0x80000000 : - entry->eax =min (entry->eax,0x80000021 ); + entry->eax =min (entry->eax,0x80000022 );/* * Serializing LFENCE is reported in a multitude of ways, and * NullSegClearsBase is not reported in CPUID on Zen2; help @@ -1203,6 +1203,25 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)if (!static_cpu_has_bug (X86_BUG_NULL_SEG )) entry->eax |=BIT (6 );break ; +/* AMD Extended Performance Monitoring and Debug */ +case 0x80000022 : { +union cpuid_0x80000022_eax eax; +union cpuid_0x80000022_ebx ebx; + + entry->eax = entry->ebx = entry->ecx = entry->edx =0 ; +if (!enable_pmu) +break ; + +if (kvm_pmu_cap.version >1 ) { +/* AMD PerfMon is only supported up to V2 in the KVM. */ + eax.split.perfmon_v2 =1 ;Quoted 8 lines
arch/x86/kvm/cpuid.c static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) · line 1219
Quoted 2 lines
arch/x86/kvm/cpuid.c +2
-0+ ebx.split.num_core_pmc =min (kvm_pmu_cap.num_counters_gp, +KVM_AMD_PMC_MAX_GENERIC );Note that the number of core PMCs has to be at least 6 if
guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE). I suppose this leaf could claim fewer, but the first 6 PMCs must work, per the v1 PMU spec. That is, software that knows aboutPERFCTR_CORE, but not about PMU v2, can rightfully expect 6 PMCs.I thought the NumCorePmc number would only make sense if CPUID.80000022.eax.perfmon_v2
bit was present, but considering that the user space is perfectly fine with just configuring the
NumCorePmc number without settingperfmon_v2bit at all, so how about:
Quoted 5 lines
Quoted 2 lines
CPUID.80000022H might only make sense if
X86_FEATURE_PERFCTR_COREis present. It's hard to know in the absence of documentation.Whenever this happens, we may always leave the definition of behavior to the hypervisor.
I disagree. If CPUID.0H reports "AuthenticAMD," then AMD is the sole authority on behavior.
Quoted 25 lines
Quoted 25 lines
Quoted 24 lines
/* AMD Extended Performance Monitoring and Debug */ case 0x80000022: { union cpuid_0x80000022_eax eax; union cpuid_0x80000022_ebx ebx; bool perfctr_core; entry->eax = entry->ebx = entry->ecx = entry->edx = 0; if (!enable_pmu) break; perfctr_core = kvm_cpu_cap_has(X86_FEATURE_PERFCTR_CORE); if (!perfctr_core) ebx.split.num_core_pmc = AMD64_NUM_COUNTERS; if (kvm_pmu_cap.version > 1) { /* AMD PerfMon is only supported up to V2 in the KVM. */ eax.split.perfmon_v2 = 1; ebx.split.num_core_pmc = min(kvm_pmu_cap.num_counters_gp, KVM_AMD_PMC_MAX_GENERIC); } if (perfctr_core) { ebx.split.num_core_pmc = max(ebx.split.num_core_pmc, AMD64_NUM_COUNTERS_CORE); }
Quoted 4 lines
Quoted 1 line
This still isn't quite right. All AMD CPUs must support a minimum of 4 PMCs.
K7 at least. I could not confirm that all antique AMD CPUs have 4 counters w/o
perfctr_core.
The APM says, "All implementations support the base set of four performance counter / event-select pairs." That is unequivocal.
Quoted 25 lines
Quoted 25 lines
Quoted 25 lines
entry->eax = eax.full; entry->ebx = ebx.full; break; }?
Once
0x80000022appears, ebx.split.num_core_pmc will report only the real "Number of Core Performance Counters" regardless ofperfmon_v2.Quoted 13 lines
Quoted 13 lines
arch/x86/kvm/cpuid.c +5
-1+ } + entry->eax = eax.full; + entry->ebx = ebx.full; +break ; + }/*Add support for Centaur's CPUID instruction*/ case 0xC0000000 :/*Just support up to 0xC0000004 now*/ --2.37.3
Re: [PATCH 4/4] KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
Quoted 46 lines · Jim Mattson
On Tue, Sep 6, 2022 at 8:59 PM Like Xu <[email protected]> wrote:
[...]Quoted 44 lines
Quoted 44 lines
Quoted 43 lines
Quoted 29 lines
Quoted 29 lines
arch/x86/kvm/cpuid.c +13
-1@@ -1094,7 +1094,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) entry->edx =0 ;break ;case 0x80000000 : - entry->eax =min (entry->eax,0x80000021 ); + entry->eax =min (entry->eax,0x80000022 );/* * Serializing LFENCE is reported in a multitude of ways, and * NullSegClearsBase is not reported in CPUID on Zen2; help @@ -1203,6 +1203,25 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)if (!static_cpu_has_bug (X86_BUG_NULL_SEG )) entry->eax |=BIT (6 );break ; +/* AMD Extended Performance Monitoring and Debug */ +case 0x80000022 : { +union cpuid_0x80000022_eax eax; +union cpuid_0x80000022_ebx ebx; + + entry->eax = entry->ebx = entry->ecx = entry->edx =0 ; +if (!enable_pmu) +break ; + +if (kvm_pmu_cap.version >1 ) { +/* AMD PerfMon is only supported up to V2 in the KVM. */ + eax.split.perfmon_v2 =1 ;Quoted 8 lines
arch/x86/kvm/cpuid.c static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) · line 1219
Quoted 2 lines
arch/x86/kvm/cpuid.c +2
-0+ ebx.split.num_core_pmc =min (kvm_pmu_cap.num_counters_gp, +KVM_AMD_PMC_MAX_GENERIC );Note that the number of core PMCs has to be at least 6 if
guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE). I suppose this leaf could claim fewer, but the first 6 PMCs must work, per the v1 PMU spec. That is, software that knows aboutPERFCTR_CORE, but not about PMU v2, can rightfully expect 6 PMCs.I thought the NumCorePmc number would only make sense if CPUID.80000022.eax.perfmon_v2
bit was present, but considering that the user space is perfectly fine with just configuring the
NumCorePmc number without settingperfmon_v2bit at all, so how about:
Quoted 9 lines
Quoted 5 lines
Quoted 2 lines
CPUID.80000022H might only make sense if
X86_FEATURE_PERFCTR_COREis present. It's hard to know in the absence of documentation.Whenever this happens, we may always leave the definition of behavior to the hypervisor.
I disagree. If CPUID.0H reports "AuthenticAMD," then AMD is the sole authority on behavior.
I understand that official documentation is not out yet. However, for Zen 4 models, it is expected that both the PerfMonV2 bit of CPUID.80000022H EAX and the PerfCtrExtCore bit of CPUID.80000001H ECX will be set.
Quoted 25 lines
Quoted 25 lines
Quoted 25 lines
Quoted 24 lines
/* AMD Extended Performance Monitoring and Debug */ case 0x80000022: { union cpuid_0x80000022_eax eax; union cpuid_0x80000022_ebx ebx; bool perfctr_core; entry->eax = entry->ebx = entry->ecx = entry->edx = 0; if (!enable_pmu) break; perfctr_core = kvm_cpu_cap_has(X86_FEATURE_PERFCTR_CORE); if (!perfctr_core) ebx.split.num_core_pmc = AMD64_NUM_COUNTERS; if (kvm_pmu_cap.version > 1) { /* AMD PerfMon is only supported up to V2 in the KVM. */ eax.split.perfmon_v2 = 1; ebx.split.num_core_pmc = min(kvm_pmu_cap.num_counters_gp, KVM_AMD_PMC_MAX_GENERIC); } if (perfctr_core) { ebx.split.num_core_pmc = max(ebx.split.num_core_pmc, AMD64_NUM_COUNTERS_CORE); }
Quoted 8 lines
Quoted 4 lines
Quoted 1 line
This still isn't quite right. All AMD CPUs must support a minimum of 4 PMCs.
K7 at least. I could not confirm that all antique AMD CPUs have 4 counters w/o
perfctr_core.The APM says, "All implementations support the base set of four performance counter / event-select pairs." That is unequivocal.
That is true. The same can be inferred from amd_core_pmu_init() in arch/x86/events/amd/core.c. If PERFCTR_CORE is not detected, it assumes that the four legacy counters are always available.
- Sandipan
Re: [PATCH 4/4] KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
Quoted 47 lines · Sandipan Das
Quoted 46 lines · Jim Mattson
On Tue, Sep 6, 2022 at 8:59 PM Like Xu <[email protected]> wrote:
[...]Quoted 44 lines
Quoted 44 lines
Quoted 43 lines
Quoted 29 lines
Quoted 29 lines
arch/x86/kvm/cpuid.c +13
-1@@ -1094,7 +1094,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) entry->edx =0 ;break ;case 0x80000000 : - entry->eax =min (entry->eax,0x80000021 ); + entry->eax =min (entry->eax,0x80000022 );/* * Serializing LFENCE is reported in a multitude of ways, and * NullSegClearsBase is not reported in CPUID on Zen2; help @@ -1203,6 +1203,25 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)if (!static_cpu_has_bug (X86_BUG_NULL_SEG )) entry->eax |=BIT (6 );break ; +/* AMD Extended Performance Monitoring and Debug */ +case 0x80000022 : { +union cpuid_0x80000022_eax eax; +union cpuid_0x80000022_ebx ebx; + + entry->eax = entry->ebx = entry->ecx = entry->edx =0 ; +if (!enable_pmu) +break ; + +if (kvm_pmu_cap.version >1 ) { +/* AMD PerfMon is only supported up to V2 in the KVM. */ + eax.split.perfmon_v2 =1 ;Quoted 8 lines
arch/x86/kvm/cpuid.c static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) · line 1219
Quoted 2 lines
arch/x86/kvm/cpuid.c +2
-0+ ebx.split.num_core_pmc =min (kvm_pmu_cap.num_counters_gp, +KVM_AMD_PMC_MAX_GENERIC );Note that the number of core PMCs has to be at least 6 if
guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE). I suppose this leaf could claim fewer, but the first 6 PMCs must work, per the v1 PMU spec. That is, software that knows aboutPERFCTR_CORE, but not about PMU v2, can rightfully expect 6 PMCs.I thought the NumCorePmc number would only make sense if CPUID.80000022.eax.perfmon_v2
bit was present, but considering that the user space is perfectly fine with just configuring the
NumCorePmc number without settingperfmon_v2bit at all, so how about:
Quoted 8 lines
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
CPUID.80000022H might only make sense if
X86_FEATURE_PERFCTR_COREis present. It's hard to know in the absence of documentation.Whenever this happens, we may always leave the definition of behavior to the hypervisor.
I disagree. If CPUID.0H reports "AuthenticAMD," then AMD is the sole authority on behavior.
The real world isn't like that, because even AMD has multiple implementations in cases
where the hardware specs aren't explicitly stated, and sometimes they're intentionally vague.
And the hypervisor can't do nothing, it prefers one over the other and maintains maximum compatibility with the legacy user space.
Quoted 5 lines
Quoted 1 line
I understand that official documentation is not out yet. However, for Zen 4 models, it is expected that both the PerfMonV2 bit of CPUID.80000022H EAX and the PerfCtrExtCore bit of CPUID.80000001H ECX will be set.
Is PerfCtrExtCore a PerfMonV2 or PerfMonV3+ precondition ? Is PerfCtrExtCore a CPUID.80000022 precondition ?
Should we always expect CPUID_Fn80000022_EBX.NumCorePmc to reflect the real Number of Core Performance Counters regardless of whether PerfMonV2 is set ?
Quoted 40 lines
Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
Quoted 25 lines
/* AMD Extended Performance Monitoring and Debug */ case 0x80000022: { union cpuid_0x80000022_eax eax; union cpuid_0x80000022_ebx ebx; bool perfctr_core; entry->eax = entry->ebx = entry->ecx = entry->edx = 0; if (!enable_pmu) break; perfctr_core = kvm_cpu_cap_has(X86_FEATURE_PERFCTR_CORE); if (!perfctr_core) ebx.split.num_core_pmc = AMD64_NUM_COUNTERS; if (kvm_pmu_cap.version > 1) { /* AMD PerfMon is only supported up to V2 in the KVM. */ eax.split.perfmon_v2 = 1; ebx.split.num_core_pmc = min(kvm_pmu_cap.num_counters_gp, KVM_AMD_PMC_MAX_GENERIC); } if (perfctr_core) { ebx.split.num_core_pmc = max(ebx.split.num_core_pmc, AMD64_NUM_COUNTERS_CORE); }Quoted 8 lines
Quoted 4 lines
Quoted 1 line
This still isn't quite right. All AMD CPUs must support a minimum of 4 PMCs.
K7 at least. I could not confirm that all antique AMD CPUs have 4 counters w/o
perfctr_core.The APM says, "All implementations support the base set of four performance counter / event-select pairs." That is unequivocal.
That is true. The same can be inferred from
amd_core_pmu_init()inarch/x86/events/amd/core.c. IfPERFCTR_COREis not detected, it assumes that the four legacy counters are always available.- Sandipan
Re: [PATCH 4/4] KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
Quoted 67 lines · Like Xu
Quoted 47 lines · Sandipan Das
Quoted 46 lines · Jim Mattson
On Tue, Sep 6, 2022 at 8:59 PM Like Xu <[email protected]> wrote:
[...]Quoted 44 lines
Quoted 44 lines
Quoted 43 lines
Quoted 29 lines
Quoted 29 lines
arch/x86/kvm/cpuid.c +0
-0@@ -1094,7 +1094,7 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) entry->edx = 0; break;case
0x80000000:
-entry->eax=min(entry->eax, 0x80000021);
+entry->eax=min(entry->eax, 0x80000022);
/*
* Serializing LFENCE is reported in a multitude of ways, and
* NullSegClearsBase is not reported in CPUID on Zen2; helparch/x86/kvm/cpuid.c +12
-0@@ -1203,6 +1203,25 @@ static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) if (!static_cpu_has_bug(X86_BUG_NULL_SEG)) entry->eax |= BIT(6); break; + /* AMD Extended Performance Monitoring and Debug */ + case 0x80000022: { + union cpuid_0x80000022_eax eax; + union cpuid_0x80000022_ebx ebx; + + entry->eax = entry->ebx = entry->ecx = entry->edx = 0; + if (!enable_pmu) + break; + + if (kvm_pmu_cap.version > 1) { + /* AMD PerfMon is only supported up to V2 in the KVM. */ + eax.split.perfmon_v2 = 1;Quoted 8 lines
arch/x86/kvm/cpuid.c static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function) · line 1219
Quoted 2 lines
arch/x86/kvm/cpuid.c +2
-0+ ebx.split.num_core_pmc = min(kvm_pmu_cap.num_counters_gp, + KVM_AMD_PMC_MAX_GENERIC);Note that the number of core PMCs has to be at least 6 if
guest_cpuid_has(vcpu, X86_FEATURE_PERFCTR_CORE). I suppose this leaf could claim fewer, but the first 6 PMCs must work, per the v1 PMU spec. That is, software that knows aboutPERFCTR_CORE, but not about PMU v2, can rightfully expect 6 PMCs.I thought the NumCorePmc number would only make sense if CPUID.80000022.eax.perfmon_v2
bit was present, but considering that the user space is perfectly fine with just configuring the
NumCorePmc number without settingperfmon_v2bit at all, so how about:Quoted 8 lines
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
CPUID.80000022H might only make sense if
X86_FEATURE_PERFCTR_COREis present. It's hard to know in the absence of documentation.Whenever this happens, we may always leave the definition of behavior to the hypervisor.
I disagree. If CPUID.0H reports "AuthenticAMD," then AMD is the sole authority on behavior.
The real world isn't like that, because even AMD has multiple implementations in cases
where the hardware specs aren't explicitly stated, and sometimes they're intentionally vague.
And the hypervisor can't do nothing, it prefers one over the other and maintains maximum compatibility with the legacy user space.Quoted 5 lines
Quoted 1 line
I understand that official documentation is not out yet. However, for Zen 4 models, it is expected that both the PerfMonV2 bit of CPUID.80000022H EAX and the PerfCtrExtCore bit of CPUID.80000001H ECX will be set.
Quoted 6 lines
Is PerfCtrExtCore a PerfMonV2 or PerfMonV3+ precondition ? Is PerfCtrExtCore a CPUID.80000022 precondition ?
Should we always expect CPUID_Fn80000022_EBX.NumCorePmc to reflect the real Number of Core Performance Counters regardless of whether PerfMonV2 is set ?
This is the suggested method for detecting the number of counters:
If CPUID Fn8000_0022_EAX[PerfMonV2] is set, then use the new interface in CPUID Fn8000_0022_EBX to determine the number of counters.
Else if CPUID Fn8000_0001_ECX[PerfCtrExtCore] is set, then six counters are available.
Else, four legacy counters are available.
There will be an APM update that will have this information in the "Detecting Hardware Support for Performance Counters" section.
Quoted 40 lines
Quoted 40 lines
Quoted 26 lines
Quoted 26 lines
Quoted 26 lines
Quoted 25 lines
/* AMD Extended Performance Monitoring and Debug */ case
0x80000022: {
unioncpuid_0x80000022_eaxeax; unioncpuid_0x80000022_ebxebx; boolperfctr_core;
entry->eax=entry->ebx=entry->ecx=entry->edx= 0; if (!enable_pmu)
break;
perfctr_core=kvm_cpu_cap_has(X86_FEATURE_PERFCTR_CORE); if (!perfctr_core)
ebx.split.num_core_pmc =AMD64_NUM_COUNTERS; if (kvm_pmu_cap.version > 1) {
/* AMD PerfMon is only supported up to V2 in the KVM. */ eax.split.perfmon_v2 = 1; ebx.split.num_core_pmc =min(kvm_pmu_cap.num_counters_gp, KVM_AMD_PMC_MAX_GENERIC); }
if (perfctr_core) {
ebx.split.num_core_pmc =max(ebx.split.num_core_pmc, AMD64_NUM_COUNTERS_CORE); }Quoted 8 lines
Quoted 4 lines
Quoted 1 line
This still isn't quite right. All AMD CPUs must support a minimum of 4 PMCs.
K7 at least. I could not confirm that all antique AMD CPUs have 4 counters w/o
perfctr_core.The APM says, "All implementations support the base set of four performance counter / event-select pairs." That is unequivocal.
That is true. The same can be inferred from
amd_core_pmu_init()inarch/x86/events/amd/core.c. IfPERFCTR_COREis not detected, it assumes that the four legacy counters are always available.- Sandipan
Re: [PATCH 4/4] KVM: x86/cpuid: Add AMD CPUID ExtPerfMonAndDbg leaf 0x80000022
Quoted 5 lines · Sandipan Das
This is the suggested method for detecting the number of counters:
If CPUID Fn8000_0022_EAX[PerfMonV2] is set, then use the new interface in CPUID
Fn8000_0022_EBXto determine the number of counters.
Quoted 7 lines
Else if CPUID Fn8000_0001_ECX[PerfCtrExtCore] is set, then six counters are available.
Else, four legacy counters are available.
There will be an APM update that will have this information in the "Detecting Hardware Support for Performance Counters" section.
Nonetheless, for compatibility with old software, Fn8000_0022_EBX can never report less than four counters (or six, if Fn8000_0001_ECX[PerfCtrExtCore] is set).
PATCH 2/2 x86/pmu: Add AMD Guest PerfMonV2 testcases
In reply to Like XuFrom: Like Xu <[email protected]>
Updated test cases to cover KVM enabling code for AMD Guest PerfMonV2.
The Intel-specific PMU helpers were added to check for AMD cpuid, and some of the same semantics of MSRs were assigned during the initialization phase. The vast majority of pmu test cases are reused seamlessly.
On some x86 machines (AMD only), even with retired events, the same workload is measured repeatedly and the number of events collected is erratic, which essentially reflects the details of hardware implementation, and from a software perspective, the type of event is an unprecise event, which brings a tolerance check in the counter overflow testcases.
--- lib/x86/msr.h | 5 ++++ lib/x86/processor.h | 9 ++++++- x86/pmu.c | 61 ++++++++++++++++++++++++++++++++------------- 3 files changed, 56 insertions(+), 19 deletions(-)
lib/x86/msr.h +5 -0
@@ -419,6 +419,11 @@#define MSR_CORE_PERF_GLOBAL_CTRL 0x0000038f#define MSR_CORE_PERF_GLOBAL_OVF_CTRL 0x00000390 +/* AMD Performance Counter Global Status and Control MSRs */ +#define MSR_AMD64_PERF_CNTR_GLOBAL_STATUS 0xc0000300 +#define MSR_AMD64_PERF_CNTR_GLOBAL_CTL 0xc0000301 +#define MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR 0xc0000302 +/* Geode defined MSRs */ #define MSR_GEODE_BUSCONT_CONF0 0x00001900
lib/x86/processor.h +8 -1
@@ -796,8 +796,12 @@ static inline void flush_tlb(void)static inline u8 pmu_version (void ) { -if (!is_intel ()) +if (!is_intel ()) { +/* Performance Monitoring Version 2 Supported */ +if (cpuid (0x80000022 ).a &0x1 ) +return 2 ;return 0 ; + }return cpuid (10 ).a &0xff ; }
@@ -824,6 +828,9 @@ static inline u8 pmu_nr_gp_counters(void) {if (is_intel ()) {return (cpuid (10 ).a >>8 ) &0xff ; + }else if (this_cpu_has_perf_global_ctrl ()) { +/* Number of Core Performance Counters. */ +return cpuid (0x80000022 ).b &0xf ; }else if (!has_amd_perfctr_core ()) {return AMD64_NUM_COUNTERS ; }
x86/pmu.c +43 -18
157 lines not on this page · raw message
Signature · 1 line
2.37.3
Sean Christopherson
maintainerRe: [kvm-unit-tests PATCH 2/2] x86/pmu: Add AMD Guest PerfMonV2 testcases
Can you provide a single series for all of the KVM-Unit-Tests PMU work (separate from the KVM patches)? Ya, it'll be big and is a blatant violation of "do one thing", but trying to manually handle the dependencies on the review side is time consuming.
One thought to help keep track of dependencies between KVM and KUT would be to add dummy commits between each sub-series, with the dummy commit containing a lore link to the relevant KVM patches/series. That would allow throwing everything into one series without losing track of things. Hopefully.
lib/x86/processor.h static inline void flush_tlb(void) · line 801
Quoted 12 lines · Like Xu
lib/x86/processor.h +3
-1@@ -796,8 +796,12 @@ static inline void flush_tlb(void)static inline u8 pmu_version (void ) { -if (!is_intel ()) +if (!is_intel ()) { +/* Performance Monitoring Version 2 Supported */ +if (cpuid (0x80000022 ).a &0x1 )
Add an X86_FEATURE_*, that way this is self-documenting.
lib/x86/processor.h static inline u8 pmu_nr_gp_counters(void) · line 831
Quoted 11 lines
lib/x86/processor.h +3
-0+return 2 ;return 0 ; + }return cpuid (10 ).a &0xff ; }@@ -824,6 +828,9 @@ static inline u8 pmu_nr_gp_counters(void) {if (is_intel ()) {return (cpuid (10 ).a >>8 ) &0xff ; + }else if (this_cpu_has_perf_global_ctrl ()) {
Eww. Took me too long to connect the dots to understand how this guarantees that leaf 0x80000022 is available. With an X86_FEATURE_* this can simply be:
} else if (this_cpu_has(X86_FEATURE_AMD_PMU_V2) {or whatever name is appropriate.
Quoted 10 lines
lib/x86/processor.h +2
-0+/* Number of Core Performance Counters. */ +return cpuid (0x80000022 ).b &0xf ; }else if (!has_amd_perfctr_core ()) {return AMD64_NUM_COUNTERS ; }x86/pmu.c +0
-0
x86/pmu.c struct pmu_event { · line 77
Quoted 6 lines
x86/pmu.c +3
-0@@ -72,6 +72,9 @@ struct pmu_event {#define PMU_CAP_FW_WRITES (1ULL << 13)static u32 gp_counter_base;static u32 gp_select_base; +static u32 global_status_msr; +static u32 global_ctl_msr; +static u32 global_status_clr_msr;
What do you think about naming these like MSR #defines? E.g.
MSR_PERF_GLOBAL_CTRL
MSR_PERF_GLOBAL_STATUS
MSR_PERF_GLOBAL_STATUS_CLR
There's a little risk of causing confusing, but I think it would make the code easier to read.
x86/pmu.c static void global_enable(pmu_counter_t *cnt) · line 156
Quoted 10 lines
x86/pmu.c +1
-2static unsigned int gp_events_size;static unsigned int nr_gp_counters;@@ -150,8 +153,7 @@ static void global_enable(pmu_counter_t *cnt)return ; cnt->idx =event_to_global_idx (cnt); -wrmsr (MSR_CORE_PERF_GLOBAL_CTRL ,rdmsr (MSR_CORE_PERF_GLOBAL_CTRL ) | - (1ull << cnt->idx)); +wrmsr (global_ctl_msr,rdmsr (global_ctl_msr) | (1ull << cnt->idx));
Opportunistically use BIT_ULL().
x86/pmu.c static void global_disable(pmu_counter_t *cnt) · line 164
Quoted 10 lines
x86/pmu.c +1
-2}static void global_disable (pmu_counter_t *cnt)@@ -159,8 +161,7 @@ static void global_disable(pmu_counter_t *cnt)if (pmu_version () <2 )return ; -wrmsr (MSR_CORE_PERF_GLOBAL_CTRL ,rdmsr (MSR_CORE_PERF_GLOBAL_CTRL ) & - ~(1ull << cnt->idx)); +wrmsr (global_ctl_msr,rdmsr (global_ctl_msr) & ~(1ull << cnt->idx));
BIT_ULL()
Quoted 28 lines
x86/pmu.c +17
-0}static inline uint32_t get_gp_counter_msr (unsigned int i)@@ -326,6 +327,23 @@ static void check_counters_many(void)report (i == n,"all counters" ); } +static bool is_the_count_reproducible (pmu_counter_t *cnt) +{ +unsigned int i; +uint64_t count; + +__measure (cnt,0 ); + count = cnt->count; + +for (i =0 ; i <10 ; i++) { +__measure (cnt,0 ); +if (count != cnt->count) +return false; + } + +return true; +} +static void check_counter_overflow (void ) {uint64_t count;
x86/pmu.c static void check_counter_overflow(void) · line 361
Quoted 10 lines
x86/pmu.c +2
-0@@ -334,13 +352,14 @@ static void check_counter_overflow(void) .ctr = gp_counter_base, .config = EVNTSEL_OS | EVNTSEL_USR | (*gp_events)[1].unit_sel /* instructions */, }; +bool precise_event =is_the_count_reproducible (&cnt); +__measure (&cnt,0 ); count = cnt.count;/* clear status before test */ if (pmu_version () >1 ) {
Please provide helper(s) to replace the myriad open coded "pmu_version() > ???" checks. E.g. this one appears to be:
if (this_cpu_has_perf_global_status_clr())
I obviously don't care about the verbosity, it's that people like me might not know what the PMU version has to do with an MSR being accessible.
Re: [kvm-unit-tests PATCH 2/2] x86/pmu: Add AMD Guest PerfMonV2 testcases
All applied, thanks.
Quoted 4 lines · Sean Christopherson
Can you provide a single series for all of the KVM-Unit-Tests PMU work (separate from the KVM patches)? Ya, it'll be big and is a blatant violation of "do one thing", but trying to manually handle the dependencies on the review side is time consuming.
Thanks for your time. PMU test cases will be combined, if no new ideas emerge.
Quoted 5 lines
One thought to help keep track of dependencies between KVM and KUT would be to add dummy commits between each sub-series, with the dummy commit containing a lore link to the relevant KVM patches/series. That would allow throwing everything into one series without losing track of things. Hopefully.
Sure, adding a lore link to a cover letter is always helpful. It seems that the ageing KVM project has moved to a test-driven approach to development, and any new developer should be aware of this rule.
Quoted 55 lines · Like Xu
lib/x86/processor.h static inline void flush_tlb(void) · line 801
Quoted 12 lines
lib/x86/processor.h +3
-1@@ -796,8 +796,12 @@ static inline void flush_tlb(void)static inline u8 pmu_version (void ) { -if (!is_intel ()) +if (!is_intel ()) { +/* Performance Monitoring Version 2 Supported */ +if (cpuid (0x80000022 ).a &0x1 )Add an X86_FEATURE_*, that way this is self-documenting.
lib/x86/processor.h static inline u8 pmu_nr_gp_counters(void) · line 831
Quoted 11 lines
lib/x86/processor.h +3
-0+return 2 ;return 0 ; + }return cpuid (10 ).a &0xff ; }@@ -824,6 +828,9 @@ static inline u8 pmu_nr_gp_counters(void) {if (is_intel ()) {return (cpuid (10 ).a >>8 ) &0xff ; + }else if (this_cpu_has_perf_global_ctrl ()) {Eww. Took me too long to connect the dots to understand how this guarantees that leaf
0x80000022is available. With an X86_FEATURE_* this can simply be:} else if (this_cpu_has(X86_FEATURE_AMD_PMU_V2) {or whatever name is appropriate.
Quoted 10 lines
lib/x86/processor.h +2
-0+/* Number of Core Performance Counters. */ +return cpuid (0x80000022 ).b &0xf ; }else if (!has_amd_perfctr_core ()) {return AMD64_NUM_COUNTERS ; }x86/pmu.c +0
-0x86/pmu.c struct pmu_event { · line 77
Quoted 6 lines
x86/pmu.c +3
-0@@ -72,6 +72,9 @@ struct pmu_event {#define PMU_CAP_FW_WRITES (1ULL << 13)static u32 gp_counter_base;static u32 gp_select_base; +static u32 global_status_msr; +static u32 global_ctl_msr; +static u32 global_status_clr_msr;What do you think about naming these like MSR #defines? E.g.
Quoted 6 lines
MSR_PERF_GLOBAL_CTRL
MSR_PERF_GLOBAL_STATUS
MSR_PERF_GLOBAL_STATUS_CLRThere's a little risk of causing confusing, but I think it would make the code easier to read.
Lowercase variable names are applied.
Quoted 73 lines
x86/pmu.c static void global_enable(pmu_counter_t *cnt) · line 156
Quoted 11 lines
x86/pmu.c +1
-2static unsigned int gp_events_size;static unsigned int nr_gp_counters;@@ -150,8 +153,7 @@ static void global_enable(pmu_counter_t *cnt)return ; cnt->idx =event_to_global_idx (cnt); -wrmsr (MSR_CORE_PERF_GLOBAL_CTRL ,rdmsr (MSR_CORE_PERF_GLOBAL_CTRL ) | - (1ull << cnt->idx)); +wrmsr (global_ctl_msr,rdmsr (global_ctl_msr) | (1ull << cnt->idx));Opportunistically use
BIT_ULL().x86/pmu.c static void global_disable(pmu_counter_t *cnt) · line 164
Quoted 10 lines
x86/pmu.c +1
-2}static void global_disable (pmu_counter_t *cnt)@@ -159,8 +161,7 @@ static void global_disable(pmu_counter_t *cnt)if (pmu_version () <2 )return ; -wrmsr (MSR_CORE_PERF_GLOBAL_CTRL ,rdmsr (MSR_CORE_PERF_GLOBAL_CTRL ) & - ~(1ull << cnt->idx)); +wrmsr (global_ctl_msr,rdmsr (global_ctl_msr) & ~(1ull << cnt->idx));
BIT_ULL()Quoted 28 lines
x86/pmu.c +17
-0}static inline uint32_t get_gp_counter_msr (unsigned int i)@@ -326,6 +327,23 @@ static void check_counters_many(void)report (i == n,"all counters" ); } +static bool is_the_count_reproducible (pmu_counter_t *cnt) +{ +unsigned int i; +uint64_t count; + +__measure (cnt,0 ); + count = cnt->count; + +for (i =0 ; i <10 ; i++) { +__measure (cnt,0 ); +if (count != cnt->count) +return false; + } + +return true; +} +static void check_counter_overflow (void ) {uint64_t count;x86/pmu.c static void check_counter_overflow(void) · line 361
Quoted 10 lines
x86/pmu.c +2
-0@@ -334,13 +352,14 @@ static void check_counter_overflow(void) .ctr = gp_counter_base, .config = EVNTSEL_OS | EVNTSEL_USR | (*gp_events)[1].unit_sel /* instructions */, }; +bool precise_event =is_the_count_reproducible (&cnt); +__measure (&cnt,0 ); count = cnt.count;/* clear status before test */ if (pmu_version () >1 ) {Please provide helper(s) to replace the myriad open coded "
pmu_version()> ???" checks. E.g. this one appears to be:if (
this_cpu_has_perf_global_status_clr())I obviously don't care about the verbosity, it's that people like me might not know what the PMU version has to do with an MSR being accessible.