KVM: x86/pmu: More refactoring to get rid of PERF_TYPE_HARDWAR
Subsystems (4)
v3 cover KVM: x86/pmu: More refactoring to get rid of PERF_TYPE_HARDWAR
Hi,
This is a follow up to [0]. By keeping the same semantics of eventsel for gp and fixed counters, the reprogram code could be made more symmetrical, simpler and even faster [1], and based on that, it fixes the obsolescence amd_event_mapping issue [2].
Most of the changes are code refactoring, which help to review key changes more easily. Patches are rebased on top of kvm/queue.
One of the notable changes is that we ended up removing the reprogram_{gp, fixed}_counter() functions and replacing it with the merged reprogram_counter(), where KVM programs pmc->perf_event with only the PERF_TYPE_RAW type for any type of counter (suggested by Jim as well). PeterZ confirmed the idea, "think so; the HARDWARE is just a convenience wrapper over RAW IIRC".
Practically, this change drops the guest pmu support on the hosts without X86_FEATURE_ARCH_PERFMON (the oldest Pentium 4), where the PERF_TYPE_HARDWAR is intentionally introduced so that hosts can map the architectural guest PMU events to their own. (thanks to Paolo)
The 3rd patch removes the call trace in the commit message while we still think that kvm->arch.pmu_event_filter requires SRCU protection in terms of pmu_event_filter functionality, similar to "kvm->arch.msr_filter".
Please check more details in each commit and feel free to comment.
[0] https://lore.kernel.org/kvm/[email protected]/
[1] https://lore.kernel.org/kvm/[email protected]/
[2] https://lore.kernel.org/kvm/[email protected]/
Thanks,
Like Xu
v2: https://lore.kernel.org/kvm/[email protected]/
v2 -> v3 Changelog:
- Use static_call stuff for .hw_event_is_unavail hook;
- Drop unrelated 0012 patch since we have addressed the issue;
- Drop passing HSW_IN_TX* bits for pmc_reprogram_counter();
v1: https://lore.kernel.org/kvm/[email protected]/
v1 -> v2 Changelog:
- Get all the vPMC tests I have on hand to pass;
- Update obsolete AMD PMU comments; (Jim)
- Fix my carelessness for [-Wconstant-logical-operand] in 0009; (0day)
- Add "u64 new_config" to reuse perf_event for fixed CTRs; (0day)
- Add patch 0012 to attract attention to review;
Like Xu (11):
KVM: x86/pmu: Update comments for AMD gp counters
KVM: x86/pmu: Extract check_pmu_event_filter() from the same semantics
KVM: x86/pmu: Protect kvm->arch.pmu_event_filter with SRCU
KVM: x86/pmu: Pass only "struct kvm_pmc *pmc" to reprogram_counter()
KVM: x86/pmu: Drop "u64 eventsel" for reprogram_gp_counter()
KVM: x86/pmu: Drop "u8 ctrl, int idx" for reprogram_fixed_counter()
KVM: x86/pmu: Use only the uniformly exported interface
reprogram_counter()
KVM: x86/pmu: Use PERF_TYPE_RAW to merge reprogram_{gp,fixed}counter()
perf: x86/core: Add interface to query perfmon_event_map[] directly
KVM: x86/pmu: Replace pmc_perf_hw_id() with perf_get_hw_event_config()
KVM: x86/pmu: Drop amd_event_mapping[] in the KVM context
arch/x86/events/core.c | 11 ++ arch/x86/include/asm/kvm-x86-pmu-ops.h | 2 +- arch/x86/include/asm/perf_event.h | 6 + arch/x86/kvm/pmu.c | 179 ++++++++++--------------- arch/x86/kvm/pmu.h | 6 +- arch/x86/kvm/svm/pmu.c | 40 +----- arch/x86/kvm/vmx/pmu_intel.c | 65 ++++----- 7 files changed, 132 insertions(+), 177 deletions(-)
Signature · 1 line
2.35.1
PATCH v3 1/11 KVM: x86/pmu: Update comments for AMD gp counters
From: Like Xu <[email protected]>
The obsolete comment could more accurately state that AMD platforms have two base MSR addresses and two different maximum numbers for gp counters, depending on the X86_FEATURE_PERFCTR_CORE feature.
--- arch/x86/kvm/pmu.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
arch/x86/kvm/pmu.c +5 -2
@@ -34,7 +34,9 @@* However AMD doesn't support fixed-counters; * - There are three types of index to access perf counters (PMC): * 1. MSR (named msr): For example Intel has MSR_IA32_PERFCTRn and AMD -* has MSR_K7_PERFCTRn. +* has MSR_K7_PERFCTRn and, for families 15H and later, +* MSR_F15H_PERF_CTRn, where MSR_F15H_PERF_CTR[0-3] are +* aliased to MSR_K7_PERFCTRn. * 2. MSR Index (named idx): This normally is used by RDPMC instruction. * For instance AMD RDPMC instruction uses 0000_0003h in ECX to access * C001_0007h (MSR_K7_PERCTR3). Intel has a similar mechanism, except
@@ -46,7 +48,8 @@* between pmc and perf counters is as the following: * * Intel: [0 .. INTEL_PMC_MAX_GENERIC-1] <=> gp counters * [INTEL_PMC_IDX_FIXED .. INTEL_PMC_IDX_FIXED + 2] <=> fixed -* * AMD: [0 .. AMD64_NUM_COUNTERS-1] <=> gp counters +* * AMD: [0 .. AMD64_NUM_COUNTERS-1] and, for families 15H +* and later, [0 .. AMD64_NUM_COUNTERS_CORE-1] <=> gp counters */ static struct kvm_pmu_ops kvm_pmu_ops__read_mostly ;
Signature · 1 line
2.35.1
PATCH v3 2/11 KVM: x86/pmu: Extract check_pmu_event_filter() from the same semantics
In reply to Like XuFrom: Like Xu <[email protected]>
Checking the kvm->arch.pmu_event_filter policy in both gp and fixed code paths was somewhat redundant, so common parts can be extracted, which reduces code footprint and improves readability.
--- arch/x86/kvm/pmu.c | 63 +++++++++++++++++++++++++++------------------- 1 file changed, 37 insertions(+), 26 deletions(-)
arch/x86/kvm/pmu.c +37 -26
@@ -205,14 +205,44 @@ static int cmp_u64(const void *a, const void *b)return *(__u64 *)a - *(__u64 *)b; } +static bool check_pmu_event_filter (struct kvm_pmc *pmc) +{ +struct kvm_pmu_event_filter *filter; +struct kvm *kvm = pmc->vcpu->kvm; +bool allow_event = true; +__u64 key; +int idx; + + filter =srcu_dereference (kvm->arch.pmu_event_filter, &kvm->srcu); +if (!filter) + goto out; + +if (pmc_is_gp (pmc)) { + key = pmc->eventsel &AMD64_RAW_EVENT_MASK_NB ; +if (bsearch (&key, filter->events, filter->nevents, +sizeof (__u64), cmp_u64)) + allow_event = filter->action ==KVM_PMU_EVENT_ALLOW ; +else + allow_event = filter->action ==KVM_PMU_EVENT_DENY ; + }else { + idx = pmc->idx -INTEL_PMC_IDX_FIXED ; +if (filter->action ==KVM_PMU_EVENT_DENY && +test_bit (idx, (ulong *)&filter->fixed_counter_bitmap)) + allow_event = false; +if (filter->action ==KVM_PMU_EVENT_ALLOW && + !test_bit (idx, (ulong *)&filter->fixed_counter_bitmap)) + allow_event = false; + } + +out: +return allow_event; +} +void reprogram_gp_counter (struct kvm_pmc *pmc,u64 eventsel) {u64 config;u32 type =PERF_TYPE_RAW ; -struct kvm *kvm = pmc->vcpu->kvm; -struct kvm_pmu_event_filter *filter; -struct kvm_pmu *pmu =vcpu_to_pmu (pmc->vcpu); -bool allow_event = true; +struct kvm_pmu *pmu =pmc_to_pmu (pmc);if (eventsel &ARCH_PERFMON_EVENTSEL_PIN_CONTROL )printk_once ("kvm pmu: pin control bit is ignored\n" );
@@ -224,17 +254,7 @@ void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel)if (!(eventsel &ARCH_PERFMON_EVENTSEL_ENABLE ) || !pmc_is_enabled (pmc))return ; - filter =srcu_dereference (kvm->arch.pmu_event_filter, &kvm->srcu); -if (filter) { -__u64 key = eventsel &AMD64_RAW_EVENT_MASK_NB ; - -if (bsearch (&key, filter->events, filter->nevents, -sizeof (__u64), cmp_u64)) - allow_event = filter->action ==KVM_PMU_EVENT_ALLOW ; -else - allow_event = filter->action ==KVM_PMU_EVENT_DENY ; - } -if (!allow_event) +if (!check_pmu_event_filter (pmc))return ; if (!(eventsel & (ARCH_PERFMON_EVENTSEL_EDGE |
@@ -267,23 +287,14 @@ void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 ctrl, int idx) {unsigned en_field = ctrl &0x3 ;bool pmi = ctrl &0x8 ; -struct kvm_pmu_event_filter *filter; -struct kvm *kvm = pmc->vcpu->kvm;pmc_pause_counter (pmc);if (!en_field || !pmc_is_enabled (pmc))return ; - filter =srcu_dereference (kvm->arch.pmu_event_filter, &kvm->srcu); -if (filter) { -if (filter->action ==KVM_PMU_EVENT_DENY && -test_bit (idx, (ulong *)&filter->fixed_counter_bitmap)) -return ; -if (filter->action ==KVM_PMU_EVENT_ALLOW && - !test_bit (idx, (ulong *)&filter->fixed_counter_bitmap)) -return ; - } +if (!check_pmu_event_filter (pmc)) +return ;if (pmc->current_config == (u64 )ctrl &&pmc_resume_counter (pmc))return ;
Signature · 1 line
2.35.1
PATCH v3 3/11 KVM: x86/pmu: Protect kvm->arch.pmu_event_filter with SRCU
In reply to Like XuFrom: Like Xu <[email protected]>
Similar to "kvm->arch.msr_filter", KVM should guarantee that vCPUs will see either the previous filter or the new filter when user space calls KVM_SET_PMU_EVENT_FILTER ioctl with the vCPU running so that guest pmu events with identical settings in both the old and new filter have deterministic behavior.
--- arch/x86/kvm/pmu.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/kvm/pmu.c +3 -1
@@ -211,8 +211,9 @@ static bool check_pmu_event_filter(struct kvm_pmc *pmc)struct kvm *kvm = pmc->vcpu->kvm;bool allow_event = true;__u64 key; -int idx; +int idx, srcu_idx; + srcu_idx =srcu_read_lock (&kvm->srcu); filter =srcu_dereference (kvm->arch.pmu_event_filter, &kvm->srcu);if (!filter) goto out;
@@ -235,6 +236,7 @@ static bool check_pmu_event_filter(struct kvm_pmc *pmc) } out: +srcu_read_unlock (&kvm->srcu, srcu_idx);return allow_event; }
Signature · 1 line
2.35.1
PATCH v3 4/11 KVM: x86/pmu: Pass only "struct kvm_pmc *pmc" to reprogram_counter()
In reply to Like XuFrom: Like Xu <[email protected]>
Passing the reference "struct kvm_pmc *pmc" when creating pmc->perf_event is sufficient. This change helps to simplify the calling convention by replacing reprogram_{gp, fixed}_counter() with reprogram_counter() seamlessly.
No functional change intended.
--- arch/x86/kvm/pmu.c | 17 +++++------------ arch/x86/kvm/pmu.h | 2 +- arch/x86/kvm/vmx/pmu_intel.c | 32 ++++++++++++++++++-------------- 3 files changed, 24 insertions(+), 27 deletions(-)
arch/x86/kvm/pmu.c +5 -12
@@ -312,18 +312,13 @@ void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 ctrl, int idx) }EXPORT_SYMBOL_GPL (reprogram_fixed_counter); -void reprogram_counter (struct kvm_pmu *pmu,int pmc_idx) +void reprogram_counter (struct kvm_pmc *pmc) { -struct kvm_pmc *pmc =static_call (kvm_x86_pmu_pmc_idx_to_pmc)(pmu, pmc_idx); - -if (!pmc) -return ; -if (pmc_is_gp (pmc))reprogram_gp_counter (pmc, pmc->eventsel);else { -int idx = pmc_idx -INTEL_PMC_IDX_FIXED ; -u8 ctrl =fixed_ctrl_field (pmu->fixed_ctr_ctrl, idx); +int idx = pmc->idx -INTEL_PMC_IDX_FIXED ; +u8 ctrl =fixed_ctrl_field (pmc_to_pmu (pmc)->fixed_ctr_ctrl, idx);reprogram_fixed_counter (pmc, ctrl, idx); }
@@ -342,8 +337,7 @@ void kvm_pmu_handle_event(struct kvm_vcpu *vcpu)clear_bit (bit, pmu->reprogram_pmi);continue ; } - -reprogram_counter (pmu, bit); +reprogram_counter (pmc); }/*
@@ -527,13 +521,12 @@ void kvm_pmu_destroy(struct kvm_vcpu *vcpu)static void kvm_pmu_incr_counter (struct kvm_pmc *pmc) { -struct kvm_pmu *pmu =pmc_to_pmu (pmc);u64 prev_count; prev_count = pmc->counter; pmc->counter = (pmc->counter +1 ) &pmc_bitmask (pmc); -reprogram_counter (pmu, pmc->idx); +reprogram_counter (pmc);if (pmc->counter < prev_count)__kvm_perf_overflow (pmc, false); }
arch/x86/kvm/pmu.h +1 -1
@@ -137,7 +137,7 @@ static inline u64 get_sample_period(struct kvm_pmc *pmc, u64 counter_value)void reprogram_gp_counter (struct kvm_pmc *pmc,u64 eventsel);void reprogram_fixed_counter (struct kvm_pmc *pmc,u8 ctrl,int fixed_idx); -void reprogram_counter (struct kvm_pmu *pmu,int pmc_idx); +void reprogram_counter (struct kvm_pmc *pmc);void kvm_pmu_deliver_pmi (struct kvm_vcpu *vcpu);void kvm_pmu_handle_event (struct kvm_vcpu *vcpu);
arch/x86/kvm/vmx/pmu_intel.c +18 -14
@@ -56,16 +56,32 @@ static void reprogram_fixed_counters(struct kvm_pmu *pmu, u64 data) pmu->fixed_ctr_ctrl = data; } +static struct kvm_pmc *intel_pmc_idx_to_pmc (struct kvm_pmu *pmu,int pmc_idx) +{ +if (pmc_idx <INTEL_PMC_IDX_FIXED ) { +return get_gp_pmc (pmu,MSR_P6_EVNTSEL0 + pmc_idx, +MSR_P6_EVNTSEL0 ); + }else { +u32 idx = pmc_idx -INTEL_PMC_IDX_FIXED ; + +return get_fixed_pmc (pmu, idx +MSR_CORE_PERF_FIXED_CTR0 ); + } +} +/* function is called when global control register has been updated. */ static void global_ctrl_changed (struct kvm_pmu *pmu,u64 data) {int bit;u64 diff = pmu->global_ctrl ^ data; +struct kvm_pmc *pmc; pmu->global_ctrl = data; -for_each_set_bit (bit, (unsigned long *)&diff,X86_PMC_IDX_MAX ) -reprogram_counter (pmu, bit); +for_each_set_bit (bit, (unsigned long *)&diff,X86_PMC_IDX_MAX ) { + pmc =intel_pmc_idx_to_pmc (pmu, bit); +if (pmc) +reprogram_counter (pmc); + } }static unsigned int intel_pmc_perf_hw_id (struct kvm_pmc *pmc)
@@ -101,18 +117,6 @@ static bool intel_pmc_is_enabled(struct kvm_pmc *pmc)return test_bit (pmc->idx, (unsigned long *)&pmu->global_ctrl); } -static struct kvm_pmc *intel_pmc_idx_to_pmc (struct kvm_pmu *pmu,int pmc_idx) -{ -if (pmc_idx <INTEL_PMC_IDX_FIXED ) -return get_gp_pmc (pmu,MSR_P6_EVNTSEL0 + pmc_idx, -MSR_P6_EVNTSEL0 ); -else { -u32 idx = pmc_idx -INTEL_PMC_IDX_FIXED ; - -return get_fixed_pmc (pmu, idx +MSR_CORE_PERF_FIXED_CTR0 ); - } -} -static bool intel_is_valid_rdpmc_ecx (struct kvm_vcpu *vcpu,unsigned int idx) {struct kvm_pmu *pmu =vcpu_to_pmu (vcpu);
Signature · 1 line
2.35.1
PATCH v3 5/11 KVM: x86/pmu: Drop "u64 eventsel" for reprogram_gp_counter()
In reply to Like XuFrom: Like Xu <[email protected]>
Because inside reprogram_gp_counter() it is bound to assign the requested eventel to pmc->eventsel, this assignment step can be moved forward, thus simplifying the passing of parameters to "struct kvm_pmc *pmc" only.
No functional change intended.
--- arch/x86/kvm/pmu.c | 7 +++---- arch/x86/kvm/pmu.h | 2 +- arch/x86/kvm/svm/pmu.c | 6 ++++-- arch/x86/kvm/vmx/pmu_intel.c | 3 ++- 4 files changed, 10 insertions(+), 8 deletions(-)
arch/x86/kvm/pmu.c +3 -4
@@ -240,17 +240,16 @@ static bool check_pmu_event_filter(struct kvm_pmc *pmc)return allow_event; } -void reprogram_gp_counter (struct kvm_pmc *pmc,u64 eventsel) +void reprogram_gp_counter (struct kvm_pmc *pmc) {u64 config;u32 type =PERF_TYPE_RAW ;struct kvm_pmu *pmu =pmc_to_pmu (pmc); +u64 eventsel = pmc->eventsel;if (eventsel &ARCH_PERFMON_EVENTSEL_PIN_CONTROL )printk_once ("kvm pmu: pin control bit is ignored\n" ); - pmc->eventsel = eventsel; -pmc_pause_counter (pmc);if (!(eventsel &ARCH_PERFMON_EVENTSEL_ENABLE ) || !pmc_is_enabled (pmc))
@@ -315,7 +314,7 @@ EXPORT_SYMBOL_GPL(reprogram_fixed_counter);void reprogram_counter (struct kvm_pmc *pmc) {if (pmc_is_gp (pmc)) -reprogram_gp_counter (pmc, pmc->eventsel); +reprogram_gp_counter (pmc);else {int idx = pmc->idx -INTEL_PMC_IDX_FIXED ;u8 ctrl =fixed_ctrl_field (pmc_to_pmu (pmc)->fixed_ctr_ctrl, idx);
arch/x86/kvm/pmu.h +1 -1
@@ -135,7 +135,7 @@ static inline u64 get_sample_period(struct kvm_pmc *pmc, u64 counter_value)return sample_period; } -void reprogram_gp_counter (struct kvm_pmc *pmc,u64 eventsel); +void reprogram_gp_counter (struct kvm_pmc *pmc);void reprogram_fixed_counter (struct kvm_pmc *pmc,u8 ctrl,int fixed_idx);void reprogram_counter (struct kvm_pmc *pmc);
arch/x86/kvm/svm/pmu.c +4 -2
@@ -263,8 +263,10 @@ static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) pmc =get_gp_pmc_amd (pmu, msr,PMU_TYPE_EVNTSEL );if (pmc) { data &= ~pmu->reserved_bits; -if (data != pmc->eventsel) -reprogram_gp_counter (pmc, data); +if (data != pmc->eventsel) { + pmc->eventsel = data; +reprogram_gp_counter (pmc); + }return 0 ; }
arch/x86/kvm/vmx/pmu_intel.c +2 -1
@@ -453,7 +453,8 @@ static int intel_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) (pmu->raw_event_mask & HSW_IN_TX_CHECKPOINTED)) reserved_bits ^=HSW_IN_TX_CHECKPOINTED ;if (!(data & reserved_bits)) { -reprogram_gp_counter (pmc, data); + pmc->eventsel = data; +reprogram_gp_counter (pmc);return 0 ; } }else if (intel_pmu_handle_lbr_msrs_access (vcpu, msr_info, false))
Signature · 1 line
2.35.1
PATCH v3 6/11 KVM: x86/pmu: Drop "u8 ctrl, int idx" for reprogram_fixed_counter()
In reply to Like XuFrom: Like Xu <[email protected]>
Since afrer reprogram_fixed_counter() is called, it's bound to assign the requested fixed_ctr_ctrl to pmu->fixed_ctr_ctrl, this assignment step can be moved forward (the stale value for diff is saved extra early), thus simplifying the passing of parameters.
No functional change intended.
--- arch/x86/kvm/pmu.c | 13 ++++++------- arch/x86/kvm/pmu.h | 2 +- arch/x86/kvm/vmx/pmu_intel.c | 16 ++++++++-------- 3 files changed, 15 insertions(+), 16 deletions(-)
arch/x86/kvm/pmu.c +6 -7
@@ -284,8 +284,11 @@ void reprogram_gp_counter(struct kvm_pmc *pmc) }EXPORT_SYMBOL_GPL (reprogram_gp_counter); -void reprogram_fixed_counter (struct kvm_pmc *pmc,u8 ctrl,int idx) +void reprogram_fixed_counter (struct kvm_pmc *pmc) { +struct kvm_pmu *pmu =pmc_to_pmu (pmc); +int idx = pmc->idx -INTEL_PMC_IDX_FIXED ; +u8 ctrl =fixed_ctrl_field (pmu->fixed_ctr_ctrl, idx);unsigned en_field = ctrl &0x3 ;bool pmi = ctrl &0x8 ;
@@ -315,12 +318,8 @@ void reprogram_counter(struct kvm_pmc *pmc) {if (pmc_is_gp (pmc))reprogram_gp_counter (pmc); -else { -int idx = pmc->idx -INTEL_PMC_IDX_FIXED ; -u8 ctrl =fixed_ctrl_field (pmc_to_pmu (pmc)->fixed_ctr_ctrl, idx); - -reprogram_fixed_counter (pmc, ctrl, idx); - } +else +reprogram_fixed_counter (pmc); }EXPORT_SYMBOL_GPL (reprogram_counter);
arch/x86/kvm/pmu.h +1 -1
@@ -136,7 +136,7 @@ static inline u64 get_sample_period(struct kvm_pmc *pmc, u64 counter_value) }void reprogram_gp_counter (struct kvm_pmc *pmc); -void reprogram_fixed_counter (struct kvm_pmc *pmc,u8 ctrl,int fixed_idx); +void reprogram_fixed_counter (struct kvm_pmc *pmc);void reprogram_counter (struct kvm_pmc *pmc);void kvm_pmu_deliver_pmi (struct kvm_vcpu *vcpu);
arch/x86/kvm/vmx/pmu_intel.c +8 -8
@@ -37,23 +37,23 @@ static int fixed_pmc_events[] = {1, 0, 7};static void reprogram_fixed_counters (struct kvm_pmu *pmu,u64 data) { +struct kvm_pmc *pmc; +u8 old_fixed_ctr_ctrl = pmu->fixed_ctr_ctrl;int i; + pmu->fixed_ctr_ctrl = data;for (i =0 ; i < pmu->nr_arch_fixed_counters; i++) {u8 new_ctrl =fixed_ctrl_field (data, i); -u8 old_ctrl =fixed_ctrl_field (pmu->fixed_ctr_ctrl, i); -struct kvm_pmc *pmc; - - pmc =get_fixed_pmc (pmu,MSR_CORE_PERF_FIXED_CTR0 + i); +u8 old_ctrl =fixed_ctrl_field (old_fixed_ctr_ctrl, i);if (old_ctrl == new_ctrl)continue ; -__set_bit (INTEL_PMC_IDX_FIXED + i, pmu->pmc_in_use); -reprogram_fixed_counter (pmc, new_ctrl, i); - } + pmc =get_fixed_pmc (pmu,MSR_CORE_PERF_FIXED_CTR0 + i); - pmu->fixed_ctr_ctrl = data; +__set_bit (INTEL_PMC_IDX_FIXED + i, pmu->pmc_in_use); +reprogram_fixed_counter (pmc); + } }static struct kvm_pmc *intel_pmc_idx_to_pmc (struct kvm_pmu *pmu,int pmc_idx)
Signature · 1 line
2.35.1
PATCH v3 7/11 KVM: x86/pmu: Use only the uniformly exported interface reprogram_counter()
In reply to Like XuFrom: Like Xu <[email protected]>
Since reprogram_counter(), reprogram_{gp, fixed}_counter() currently have the same incoming parameter "struct kvm_pmc *pmc", the callers can simplify the conetxt by using uniformly exported interface, which makes reprogram_ {gp, fixed}_counter() static and eliminates EXPORT_SYMBOL_GPL.
--- arch/x86/kvm/pmu.c | 6 ++---- arch/x86/kvm/pmu.h | 2 -- arch/x86/kvm/svm/pmu.c | 2 +- arch/x86/kvm/vmx/pmu_intel.c | 4 ++-- 4 files changed, 5 insertions(+), 9 deletions(-)
arch/x86/kvm/pmu.c +2 -4
@@ -240,7 +240,7 @@ static bool check_pmu_event_filter(struct kvm_pmc *pmc)return allow_event; } -void reprogram_gp_counter (struct kvm_pmc *pmc) +static void reprogram_gp_counter (struct kvm_pmc *pmc) {u64 config;u32 type =PERF_TYPE_RAW ;
@@ -282,9 +282,8 @@ void reprogram_gp_counter(struct kvm_pmc *pmc) !(eventsel &ARCH_PERFMON_EVENTSEL_OS ), eventsel & ARCH_PERFMON_EVENTSEL_INT); } -EXPORT_SYMBOL_GPL (reprogram_gp_counter); -void reprogram_fixed_counter (struct kvm_pmc *pmc) +static void reprogram_fixed_counter (struct kvm_pmc *pmc) {struct kvm_pmu *pmu =pmc_to_pmu (pmc);int idx = pmc->idx -INTEL_PMC_IDX_FIXED ;
@@ -312,7 +311,6 @@ void reprogram_fixed_counter(struct kvm_pmc *pmc) !(en_field &0x1 ),/* exclude kernel */ pmi); } -EXPORT_SYMBOL_GPL (reprogram_fixed_counter);void reprogram_counter (struct kvm_pmc *pmc) {
arch/x86/kvm/pmu.h +0 -2
@@ -135,8 +135,6 @@ static inline u64 get_sample_period(struct kvm_pmc *pmc, u64 counter_value)return sample_period; } -void reprogram_gp_counter (struct kvm_pmc *pmc); -void reprogram_fixed_counter (struct kvm_pmc *pmc);void reprogram_counter (struct kvm_pmc *pmc);void kvm_pmu_deliver_pmi (struct kvm_vcpu *vcpu);
arch/x86/kvm/svm/pmu.c +1 -1
@@ -265,7 +265,7 @@ static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) data &= ~pmu->reserved_bits;if (data != pmc->eventsel) { pmc->eventsel = data; -reprogram_gp_counter (pmc); +reprogram_counter (pmc); }return 0 ; }
arch/x86/kvm/vmx/pmu_intel.c +2 -2
@@ -52,7 +52,7 @@ static void reprogram_fixed_counters(struct kvm_pmu *pmu, u64 data) pmc =get_fixed_pmc (pmu,MSR_CORE_PERF_FIXED_CTR0 + i);__set_bit (INTEL_PMC_IDX_FIXED + i, pmu->pmc_in_use); -reprogram_fixed_counter (pmc); +reprogram_counter (pmc); } }
@@ -454,7 +454,7 @@ static int intel_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info) reserved_bits ^=HSW_IN_TX_CHECKPOINTED ;if (!(data & reserved_bits)) { pmc->eventsel = data; -reprogram_gp_counter (pmc); +reprogram_counter (pmc);return 0 ; } }else if (intel_pmu_handle_lbr_msrs_access (vcpu, msr_info, false))
Signature · 1 line
2.35.1
PATCH v3 8/11 KVM: x86/pmu: Use PERF_TYPE_RAW to merge reprogram_{gp,fixed}counter()
In reply to Like XuFrom: Like Xu <[email protected]>
The code sketch for reprogram_{gp, fixed}_counter() is similar, while the fixed counter using the PERF_TYPE_HARDWAR type and the gp being able to use either PERF_TYPE_HARDWAR or PERF_TYPE_RAW type depending on the pmc->eventsel value.
After 'commit 761875634a5e ("KVM: x86/pmu: Setup pmc->eventsel for fixed PMCs")', the pmc->eventsel of the fixed counter will also have been setup with the same semantic value and will not be changed during the guest runtime.
The original story of using the PERF_TYPE_HARDWARE type is to emulate guest architecture PMU on a host without architecture PMU (the Pentium 4), for which the guest vPMC needs to be reprogrammed using the kernel generic perf_hw_id. But essentially, "the HARDWARE is just a convenience wrapper over RAW IIRC", quoated from Peterz. So it could be pretty safe to use the PERF_TYPE_RAW type only in practice to program both gp and fixed counters naturally in the reprogram_counter().
To make the gp and fixed counters more semantically symmetrical, the selection of EVENTSEL_{USER, OS, INT} bits is temporarily translated via fixed_ctr_ctrl before the pmc_reprogram_counter() call.
--- arch/x86/kvm/pmu.c | 125 ++++++++++++----------------------- arch/x86/kvm/vmx/pmu_intel.c | 3 +- 2 files changed, 46 insertions(+), 82 deletions(-)
arch/x86/kvm/pmu.c +44 -81
@@ -240,84 +240,58 @@ static bool check_pmu_event_filter(struct kvm_pmc *pmc)return allow_event; } -static void reprogram_gp_counter (struct kvm_pmc *pmc) -{ -u64 config; -u32 type =PERF_TYPE_RAW ; -struct kvm_pmu *pmu =pmc_to_pmu (pmc); -u64 eventsel = pmc->eventsel; - -if (eventsel &ARCH_PERFMON_EVENTSEL_PIN_CONTROL ) -printk_once ("kvm pmu: pin control bit is ignored\n" ); - -pmc_pause_counter (pmc); - -if (!(eventsel &ARCH_PERFMON_EVENTSEL_ENABLE ) || !pmc_is_enabled (pmc)) -return ; - -if (!check_pmu_event_filter (pmc)) -return ; - -if (!(eventsel & (ARCH_PERFMON_EVENTSEL_EDGE | -ARCH_PERFMON_EVENTSEL_INV | -ARCH_PERFMON_EVENTSEL_CMASK | -HSW_IN_TX | -HSW_IN_TX_CHECKPOINTED ))) { - config =static_call (kvm_x86_pmu_pmc_perf_hw_id)(pmc); -if (config !=PERF_COUNT_HW_MAX ) - type =PERF_TYPE_HARDWARE ; - } - -if (type ==PERF_TYPE_RAW ) - config = eventsel & pmu->raw_event_mask; - -if (pmc->current_config == eventsel &&pmc_resume_counter (pmc)) -return ; - -pmc_release_perf_event (pmc); - - pmc->current_config = eventsel; -pmc_reprogram_counter (pmc, type, config, - !(eventsel &ARCH_PERFMON_EVENTSEL_USR ), - !(eventsel &ARCH_PERFMON_EVENTSEL_OS ), - eventsel &ARCH_PERFMON_EVENTSEL_INT ); -} - -static void reprogram_fixed_counter (struct kvm_pmc *pmc) +static inline bool pmc_speculative_in_use (struct kvm_pmc *pmc) {struct kvm_pmu *pmu =pmc_to_pmu (pmc); -int idx = pmc->idx -INTEL_PMC_IDX_FIXED ; -u8 ctrl =fixed_ctrl_field (pmu->fixed_ctr_ctrl, idx); -unsigned en_field = ctrl &0x3 ; -bool pmi = ctrl &0x8 ; -pmc_pause_counter (pmc); +if (pmc_is_fixed (pmc)) +return fixed_ctrl_field (pmu->fixed_ctr_ctrl, + pmc->idx -INTEL_PMC_IDX_FIXED ) &0x3 ; -if (!en_field || !pmc_is_enabled (pmc)) -return ; - -if (!check_pmu_event_filter (pmc)) -return ; - -if (pmc->current_config == (u64 )ctrl &&pmc_resume_counter (pmc)) -return ; - -pmc_release_perf_event (pmc); - - pmc->current_config = (u64 )ctrl; -pmc_reprogram_counter (pmc,PERF_TYPE_HARDWARE , -static_call (kvm_x86_pmu_pmc_perf_hw_id)(pmc), - !(en_field &0x2 ),/* exclude user */ - !(en_field &0x1 ),/* exclude kernel */ - pmi); +return pmc->eventsel &ARCH_PERFMON_EVENTSEL_ENABLE ; }void reprogram_counter (struct kvm_pmc *pmc) { -if (pmc_is_gp (pmc)) -reprogram_gp_counter (pmc); -else -reprogram_fixed_counter (pmc); +struct kvm_pmu *pmu =pmc_to_pmu (pmc); +u64 eventsel = pmc->eventsel; +u64 new_config = eventsel; +u8 fixed_ctr_ctrl; + +pmc_pause_counter (pmc); + +if (!pmc_speculative_in_use (pmc) || !pmc_is_enabled (pmc)) +return ; + +if (!check_pmu_event_filter (pmc)) +return ; + +if (eventsel &ARCH_PERFMON_EVENTSEL_PIN_CONTROL ) +printk_once ("kvm pmu: pin control bit is ignored\n" ); + +if (pmc_is_fixed (pmc)) { + fixed_ctr_ctrl =fixed_ctrl_field (pmu->fixed_ctr_ctrl, + pmc->idx -INTEL_PMC_IDX_FIXED ); +if (fixed_ctr_ctrl &0x1 ) + eventsel |=ARCH_PERFMON_EVENTSEL_OS ; +if (fixed_ctr_ctrl &0x2 ) + eventsel |=ARCH_PERFMON_EVENTSEL_USR ; +if (fixed_ctr_ctrl &0x8 ) + eventsel |=ARCH_PERFMON_EVENTSEL_INT ; + new_config = (u64 )fixed_ctr_ctrl; + } + +if (pmc->current_config == new_config &&pmc_resume_counter (pmc)) +return ; + +pmc_release_perf_event (pmc); + + pmc->current_config = new_config; +pmc_reprogram_counter (pmc,PERF_TYPE_RAW , + (eventsel & pmu->raw_event_mask), + !(eventsel &ARCH_PERFMON_EVENTSEL_USR ), + !(eventsel &ARCH_PERFMON_EVENTSEL_OS ), + eventsel &ARCH_PERFMON_EVENTSEL_INT ); }EXPORT_SYMBOL_GPL (reprogram_counter);
@@ -474,17 +448,6 @@ void kvm_pmu_init(struct kvm_vcpu *vcpu)kvm_pmu_refresh (vcpu); } -static inline bool pmc_speculative_in_use (struct kvm_pmc *pmc) -{ -struct kvm_pmu *pmu =pmc_to_pmu (pmc); - -if (pmc_is_fixed (pmc)) -return fixed_ctrl_field (pmu->fixed_ctr_ctrl, - pmc->idx -INTEL_PMC_IDX_FIXED ) &0x3 ; - -return pmc->eventsel &ARCH_PERFMON_EVENTSEL_ENABLE ; -} -/* Release perf_events for vPMCs that have been unused for a full time slice. */ void kvm_pmu_cleanup (struct kvm_vcpu *vcpu) {
arch/x86/kvm/vmx/pmu_intel.c +2 -1
@@ -498,7 +498,8 @@ static void intel_pmu_refresh(struct kvm_vcpu *vcpu) pmu->raw_event_mask =X86_RAW_EVENT_MASK ; entry =kvm_find_cpuid_entry (vcpu,0xa ,0 ); -if (!entry || !vcpu->kvm->arch.enable_pmu) +if (!entry || !vcpu->kvm->arch.enable_pmu || + !boot_cpu_has (X86_FEATURE_ARCH_PERFMON ))return ; eax.full = entry->eax; edx.full = entry->edx;
Signature · 1 line
2.35.1
PATCH v3 9/11 perf: x86/core: Add interface to query perfmon_event_map[] directly
In reply to Like XuFrom: Like Xu <[email protected]>
Currently, we have [intel|knc|p4|p6]_perfmon_event_map on the Intel platforms and amd_[f17h]_perfmon_event_map on the AMD platforms.
Early clumsy KVM code or other potential perf_event users may have hard-coded these perfmon_maps (e.g., arch/x86/kvm/svm/pmu.c), so it would not make sense to program a common hardware event based on the generic "enum perf_hw_id" once the two tables do not match.
Let's provide an interface for callers outside the perf subsystem to get the counter config based on the perfmon_event_map currently in use, and it also helps to save bytes.
--- arch/x86/events/core.c | 11 +++++++++++ arch/x86/include/asm/perf_event.h | 6 ++++++ 2 files changed, 17 insertions(+)
arch/x86/events/core.c +11 -0
@@ -2996,3 +2996,14 @@ void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap) cap->events_mask_len = x86_pmu.events_mask_len; }EXPORT_SYMBOL_GPL (perf_get_x86_pmu_capability); + +u64 perf_get_hw_event_config (int hw_event) +{ +int max = x86_pmu.max_events; + +if (hw_event < max) +return x86_pmu.event_map (array_index_nospec (hw_event, max)); + +return 0 ; +} +EXPORT_SYMBOL_GPL (perf_get_hw_event_config);
arch/x86/include/asm/perf_event.h +6 -0
@@ -477,6 +477,7 @@ struct x86_pmu_lbr { };extern void perf_get_x86_pmu_capability (struct x86_pmu_capability *cap); +extern u64 perf_get_hw_event_config (int hw_event);extern void perf_check_microcode (void );extern void perf_clear_dirty_counters (void );extern int x86_perf_rdpmc_index (struct perf_event *event);
@@ -486,6 +487,11 @@ static inline void perf_get_x86_pmu_capability(struct x86_pmu_capability *cap) memset(cap, 0, sizeof(*cap)); } +static inline u64 perf_get_hw_event_config (int hw_event) +{ +return 0 ; +} +static inline void perf_events_lapic_init (void ) { }static inline void perf_check_microcode (void ) { }#endif
Signature · 1 line
2.35.1
Re: [PATCH v3 09/11] perf: x86/core: Add interface to query perfmon_event_map[] directly
Quoted 10 lines · Like Xu
From: Like Xu <[email protected]>
Currently, we have [intel|knc|p4|p6]_perfmon_event_map on the Intel platforms and amd_[f17h]_perfmon_event_map on the AMD platforms.
Early clumsy KVM code or other potential
perf_eventusers may have hard-coded theseperfmon_maps(e.g.,arch/x86/kvm/svm/pmu.c), so it would not make sense to program a common hardware event based on the generic "enumperf_hw_id" once the two tables do not match.
Quoted 6 lines
Let's provide an interface for callers outside the perf subsystem to get the counter config based on the
perfmon_event_mapcurrently in use, and it also helps to save bytes.Cc Peter Zijlstra <[email protected]>Signed-off-by Like Xu <[email protected]>
PATCH v3 10/11 KVM: x86/pmu: Replace pmc_perf_hw_id() with perf_get_hw_event_config()
In reply to Like XuFrom: Like Xu <[email protected]>
With the help of perf_get_hw_event_config(), KVM could query the correct EVENTSEL_{EVENT, UMASK} pair of a kernel-generic hw event directly from the different *_perfmon_event_map[] by the kernel's pre-defined perf_hw_id.
Also extend the bit range of the comparison field to AMD64_RAW_EVENT_MASK_NB to prevent AMD from defining EventSelect[11:8] into perfmon_event_map[] one day.
--- arch/x86/kvm/pmu.c | 9 ++------- 1 file changed, 2 insertions(+), 7 deletions(-)
arch/x86/kvm/pmu.c +2 -7
@@ -493,13 +493,8 @@ static void kvm_pmu_incr_counter(struct kvm_pmc *pmc)static inline bool eventsel_match_perf_hw_id (struct kvm_pmc *pmc,unsigned int perf_hw_id) { -u64 old_eventsel = pmc->eventsel; -unsigned int config; - - pmc->eventsel &= (ARCH_PERFMON_EVENTSEL_EVENT |ARCH_PERFMON_EVENTSEL_UMASK ); - config =static_call (kvm_x86_pmu_pmc_perf_hw_id)(pmc); - pmc->eventsel = old_eventsel; -return config == perf_hw_id; +return !((pmc->eventsel ^perf_get_hw_event_config (perf_hw_id)) & +AMD64_RAW_EVENT_MASK_NB ); }static inline bool cpl_is_matched (struct kvm_pmc *pmc)
Signature · 1 line
2.35.1
PATCH v3 11/11 KVM: x86/pmu: Drop amd_event_mapping[] in the KVM context
In reply to Like XuFrom: Like Xu <[email protected]>
All gp or fixed counters have been reprogrammed using PERF_TYPE_RAW, which means that the table that maps perf_hw_id to event select values is no longer useful, at least for AMD.
For Intel, the logic to check if the pmu event reported by Intel cpuid is not available is still required, in which case pmc_perf_hw_id() could be renamed to hw_event_is_unavail() and a bool value is returned to replace the semantics of "PERF_COUNT_HW_MAX+1".
--- arch/x86/include/asm/kvm-x86-pmu-ops.h | 2 +- arch/x86/kvm/pmu.c | 6 ++--- arch/x86/kvm/pmu.h | 2 +- arch/x86/kvm/svm/pmu.c | 34 +++----------------------- arch/x86/kvm/vmx/pmu_intel.c | 11 +++------ 5 files changed, 12 insertions(+), 43 deletions(-)
arch/x86/include/asm/kvm-x86-pmu-ops.h +1 -1
@@ -12,7 +12,7 @@ BUILD_BUG_ON(1)* a NULL definition, for example if "static_call_cond()" will be used * at the call sites. */ -KVM_X86_PMU_OP(pmc_perf_hw_id) +KVM_X86_PMU_OP(hw_event_is_unavail) KVM_X86_PMU_OP(pmc_is_enabled) KVM_X86_PMU_OP(pmc_idx_to_pmc) KVM_X86_PMU_OP(rdpmc_ecx_to_pmc)
arch/x86/kvm/pmu.c +3 -3
@@ -139,9 +139,6 @@ static void pmc_reprogram_counter(struct kvm_pmc *pmc, u32 type, .config = config, }; -if (type ==PERF_TYPE_HARDWARE && config >=PERF_COUNT_HW_MAX ) -return ; - attr.sample_period =get_sample_period (pmc, pmc->counter); if ((attr.config & HSW_IN_TX_CHECKPOINTED) &&
@@ -213,6 +210,9 @@ static bool check_pmu_event_filter(struct kvm_pmc *pmc)__u64 key;int idx, srcu_idx; +if (static_call (kvm_x86_pmu_hw_event_is_unavail)(pmc)) +return false; + srcu_idx =srcu_read_lock (&kvm->srcu); filter =srcu_dereference (kvm->arch.pmu_event_filter, &kvm->srcu);if (!filter)
arch/x86/kvm/pmu.h +1 -1
@@ -22,7 +22,7 @@ struct kvm_event_hw_type_mapping { };struct kvm_pmu_ops { -unsigned int (*pmc_perf_hw_id)(struct kvm_pmc *pmc); +bool (*hw_event_is_unavail)(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,
arch/x86/kvm/svm/pmu.c +3 -31
@@ -33,18 +33,6 @@ enum index { INDEX_ERROR, }; -/* duplicated from amd_perfmon_event_map, K7 and above should work. */ -static struct kvm_event_hw_type_mapping amd_event_mapping[] = { - [0 ] = {0x76 ,0x00 ,PERF_COUNT_HW_CPU_CYCLES }, - [1 ] = {0xc0 ,0x00 ,PERF_COUNT_HW_INSTRUCTIONS }, - [2 ] = {0x7d ,0x07 ,PERF_COUNT_HW_CACHE_REFERENCES }, - [3 ] = {0x7e ,0x07 ,PERF_COUNT_HW_CACHE_MISSES }, - [4 ] = {0xc2 ,0x00 ,PERF_COUNT_HW_BRANCH_INSTRUCTIONS }, - [5 ] = {0xc3 ,0x00 ,PERF_COUNT_HW_BRANCH_MISSES }, - [6 ] = {0xd0 ,0x00 ,PERF_COUNT_HW_STALLED_CYCLES_FRONTEND }, - [7 ] = {0xd1 ,0x00 ,PERF_COUNT_HW_STALLED_CYCLES_BACKEND }, -}; -static unsigned int get_msr_base (struct kvm_pmu *pmu,enum pmu_type type) {struct kvm_vcpu *vcpu =pmu_to_vcpu (pmu);
@@ -138,25 +126,9 @@ static inline struct kvm_pmc *get_gp_pmc_amd(struct kvm_pmu *pmu, u32 msr,return &pmu->gp_counters[msr_to_index (msr)]; } -static unsigned int amd_pmc_perf_hw_id (struct kvm_pmc *pmc) +static bool amd_hw_event_is_unavail (struct kvm_pmc *pmc) { -u8 event_select = pmc->eventsel &ARCH_PERFMON_EVENTSEL_EVENT ; -u8 unit_mask = (pmc->eventsel &ARCH_PERFMON_EVENTSEL_UMASK ) >>8 ; -int i; - -/* return PERF_COUNT_HW_MAX as AMD doesn't have fixed events */ -if (WARN_ON (pmc_is_fixed (pmc))) -return PERF_COUNT_HW_MAX ; - -for (i =0 ; i <ARRAY_SIZE (amd_event_mapping); i++) -if (amd_event_mapping[i].eventsel == event_select - && amd_event_mapping[i].unit_mask == unit_mask) -break ; - -if (i ==ARRAY_SIZE (amd_event_mapping)) -return PERF_COUNT_HW_MAX ; - -return amd_event_mapping[i].event_type; +return false; }/* check if a PMC is enabled by comparing it against global_ctrl bits. Because
@@ -322,7 +294,7 @@ static void amd_pmu_reset(struct kvm_vcpu *vcpu) }struct kvm_pmu_ops amd_pmu_ops__initdata = { - .pmc_perf_hw_id = amd_pmc_perf_hw_id, + .hw_event_is_unavail = amd_hw_event_is_unavail, .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,
arch/x86/kvm/vmx/pmu_intel.c +4 -7
@@ -84,7 +84,7 @@ static void global_ctrl_changed(struct kvm_pmu *pmu, u64 data) } } -static unsigned int intel_pmc_perf_hw_id (struct kvm_pmc *pmc) +static bool intel_hw_event_is_unavail (struct kvm_pmc *pmc) {struct kvm_pmu *pmu =pmc_to_pmu (pmc);u8 event_select = pmc->eventsel &ARCH_PERFMON_EVENTSEL_EVENT ;
@@ -98,15 +98,12 @@ static unsigned int intel_pmc_perf_hw_id(struct kvm_pmc *pmc)/* disable event that reported as not present by cpuid */ if ((i <7 ) && !(pmu->available_event_types & (1 << i))) -return PERF_COUNT_HW_MAX +1 ; +return true;break ; } -if (i ==ARRAY_SIZE (intel_arch_events)) -return PERF_COUNT_HW_MAX ; - -return intel_arch_events[i].event_type; +return false; }/* check if a PMC is enabled by comparing it with globl_ctrl bits. */
@@ -730,7 +727,7 @@ static void intel_pmu_cleanup(struct kvm_vcpu *vcpu) }struct kvm_pmu_ops intel_pmu_ops__initdata = { - .pmc_perf_hw_id = intel_pmc_perf_hw_id, + .hw_event_is_unavail = intel_hw_event_is_unavail, .pmc_is_enabled = intel_pmc_is_enabled, .pmc_idx_to_pmc = intel_pmc_idx_to_pmc, .rdpmc_ecx_to_pmc = intel_rdpmc_ecx_to_pmc,
Signature · 1 line
2.35.1
A kind ping on this collection of minor changes. :D