These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / arch / x86 / kvm / pmu.c
index 29fbf9d..31aa2c8 100644 (file)
@@ -1,11 +1,12 @@
 /*
  * Kernel-based Virtual Machine -- Performance Monitoring Unit support
  *
- * Copyright 2011 Red Hat, Inc. and/or its affiliates.
+ * Copyright 2015 Red Hat, Inc. and/or its affiliates.
  *
  * Authors:
  *   Avi Kivity   <avi@redhat.com>
  *   Gleb Natapov <gleb@redhat.com>
+ *   Wei Huang    <wei@redhat.com>
  *
  * This work is licensed under the terms of the GNU GPL, version 2.  See
  * the COPYING file in the top-level directory.
 #include "x86.h"
 #include "cpuid.h"
 #include "lapic.h"
+#include "pmu.h"
+
+/* NOTE:
+ * - Each perf counter is defined as "struct kvm_pmc";
+ * - There are two types of perf counters: general purpose (gp) and fixed.
+ *   gp counters are stored in gp_counters[] and fixed counters are stored
+ *   in fixed_counters[] respectively. Both of them are part of "struct
+ *   kvm_pmu";
+ * - pmu.c understands the difference between gp counters and fixed counters.
+ *   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.
+ *     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
+ *        that it also supports fixed counters. idx can be used to as index to
+ *        gp and fixed counters.
+ *     3. Global PMC Index (named pmc): pmc is an index specific to PMU
+ *        code. Each pmc, stored in kvm_pmc.idx field, is unique across
+ *        all perf counters (both gp and fixed). The mapping relationship
+ *        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
+ */
 
-static struct kvm_arch_event_perf_mapping {
-       u8 eventsel;
-       u8 unit_mask;
-       unsigned event_type;
-       bool inexact;
-} arch_events[] = {
-       /* Index must match CPUID 0x0A.EBX bit vector */
-       [0] = { 0x3c, 0x00, PERF_COUNT_HW_CPU_CYCLES },
-       [1] = { 0xc0, 0x00, PERF_COUNT_HW_INSTRUCTIONS },
-       [2] = { 0x3c, 0x01, PERF_COUNT_HW_BUS_CYCLES  },
-       [3] = { 0x2e, 0x4f, PERF_COUNT_HW_CACHE_REFERENCES },
-       [4] = { 0x2e, 0x41, PERF_COUNT_HW_CACHE_MISSES },
-       [5] = { 0xc4, 0x00, PERF_COUNT_HW_BRANCH_INSTRUCTIONS },
-       [6] = { 0xc5, 0x00, PERF_COUNT_HW_BRANCH_MISSES },
-       [7] = { 0x00, 0x30, PERF_COUNT_HW_REF_CPU_CYCLES },
-};
-
-/* mapping between fixed pmc index and arch_events array */
-static int fixed_pmc_events[] = {1, 0, 7};
-
-static bool pmc_is_gp(struct kvm_pmc *pmc)
-{
-       return pmc->type == KVM_PMC_GP;
-}
-
-static inline u64 pmc_bitmask(struct kvm_pmc *pmc)
-{
-       struct kvm_pmu *pmu = &pmc->vcpu->arch.pmu;
-
-       return pmu->counter_bitmask[pmc->type];
-}
-
-static inline bool pmc_enabled(struct kvm_pmc *pmc)
-{
-       struct kvm_pmu *pmu = &pmc->vcpu->arch.pmu;
-       return test_bit(pmc->idx, (unsigned long *)&pmu->global_ctrl);
-}
-
-static inline struct kvm_pmc *get_gp_pmc(struct kvm_pmu *pmu, u32 msr,
-                                        u32 base)
-{
-       if (msr >= base && msr < base + pmu->nr_arch_gp_counters)
-               return &pmu->gp_counters[msr - base];
-       return NULL;
-}
-
-static inline struct kvm_pmc *get_fixed_pmc(struct kvm_pmu *pmu, u32 msr)
-{
-       int base = MSR_CORE_PERF_FIXED_CTR0;
-       if (msr >= base && msr < base + pmu->nr_arch_fixed_counters)
-               return &pmu->fixed_counters[msr - base];
-       return NULL;
-}
-
-static inline struct kvm_pmc *get_fixed_pmc_idx(struct kvm_pmu *pmu, int idx)
-{
-       return get_fixed_pmc(pmu, MSR_CORE_PERF_FIXED_CTR0 + idx);
-}
-
-static struct kvm_pmc *global_idx_to_pmc(struct kvm_pmu *pmu, int idx)
-{
-       if (idx < INTEL_PMC_IDX_FIXED)
-               return get_gp_pmc(pmu, MSR_P6_EVNTSEL0 + idx, MSR_P6_EVNTSEL0);
-       else
-               return get_fixed_pmc_idx(pmu, idx - INTEL_PMC_IDX_FIXED);
-}
-
-void kvm_deliver_pmi(struct kvm_vcpu *vcpu)
-{
-       if (vcpu->arch.apic)
-               kvm_apic_local_deliver(vcpu->arch.apic, APIC_LVTPC);
-}
-
-static void trigger_pmi(struct irq_work *irq_work)
+static void kvm_pmi_trigger_fn(struct irq_work *irq_work)
 {
-       struct kvm_pmu *pmu = container_of(irq_work, struct kvm_pmu,
-                       irq_work);
-       struct kvm_vcpu *vcpu = container_of(pmu, struct kvm_vcpu,
-                       arch.pmu);
+       struct kvm_pmu *pmu = container_of(irq_work, struct kvm_pmu, irq_work);
+       struct kvm_vcpu *vcpu = pmu_to_vcpu(pmu);
 
-       kvm_deliver_pmi(vcpu);
+       kvm_pmu_deliver_pmi(vcpu);
 }
 
 static void kvm_perf_overflow(struct perf_event *perf_event,
@@ -108,63 +60,46 @@ static void kvm_perf_overflow(struct perf_event *perf_event,
                              struct pt_regs *regs)
 {
        struct kvm_pmc *pmc = perf_event->overflow_handler_context;
-       struct kvm_pmu *pmu = &pmc->vcpu->arch.pmu;
-       if (!test_and_set_bit(pmc->idx, (unsigned long *)&pmu->reprogram_pmi)) {
+       struct kvm_pmu *pmu = pmc_to_pmu(pmc);
+
+       if (!test_and_set_bit(pmc->idx,
+                             (unsigned long *)&pmu->reprogram_pmi)) {
                __set_bit(pmc->idx, (unsigned long *)&pmu->global_status);
                kvm_make_request(KVM_REQ_PMU, pmc->vcpu);
        }
 }
 
 static void kvm_perf_overflow_intr(struct perf_event *perf_event,
-               struct perf_sample_data *data, struct pt_regs *regs)
+                                  struct perf_sample_data *data,
+                                  struct pt_regs *regs)
 {
        struct kvm_pmc *pmc = perf_event->overflow_handler_context;
-       struct kvm_pmu *pmu = &pmc->vcpu->arch.pmu;
-       if (!test_and_set_bit(pmc->idx, (unsigned long *)&pmu->reprogram_pmi)) {
+       struct kvm_pmu *pmu = pmc_to_pmu(pmc);
+
+       if (!test_and_set_bit(pmc->idx,
+                             (unsigned long *)&pmu->reprogram_pmi)) {
                __set_bit(pmc->idx, (unsigned long *)&pmu->global_status);
                kvm_make_request(KVM_REQ_PMU, pmc->vcpu);
+
                /*
                 * Inject PMI. If vcpu was in a guest mode during NMI PMI
                 * can be ejected on a guest mode re-entry. Otherwise we can't
                 * be sure that vcpu wasn't executing hlt instruction at the
-                * time of vmexit and is not going to re-enter guest mode until,
+                * time of vmexit and is not going to re-enter guest mode until
                 * woken up. So we should wake it, but this is impossible from
                 * NMI context. Do it from irq work instead.
                 */
                if (!kvm_is_in_guest())
-                       irq_work_queue(&pmc->vcpu->arch.pmu.irq_work);
+                       irq_work_queue(&pmc_to_pmu(pmc)->irq_work);
                else
                        kvm_make_request(KVM_REQ_PMI, pmc->vcpu);
        }
 }
 
-static u64 read_pmc(struct kvm_pmc *pmc)
-{
-       u64 counter, enabled, running;
-
-       counter = pmc->counter;
-
-       if (pmc->perf_event)
-               counter += perf_event_read_value(pmc->perf_event,
-                                                &enabled, &running);
-
-       /* FIXME: Scaling needed? */
-
-       return counter & pmc_bitmask(pmc);
-}
-
-static void stop_counter(struct kvm_pmc *pmc)
-{
-       if (pmc->perf_event) {
-               pmc->counter = read_pmc(pmc);
-               perf_event_release_kernel(pmc->perf_event);
-               pmc->perf_event = NULL;
-       }
-}
-
-static void reprogram_counter(struct kvm_pmc *pmc, u32 type,
-               unsigned config, bool exclude_user, bool exclude_kernel,
-               bool intr, bool in_tx, bool in_tx_cp)
+static void pmc_reprogram_counter(struct kvm_pmc *pmc, u32 type,
+                                 unsigned config, bool exclude_user,
+                                 bool exclude_kernel, bool intr,
+                                 bool in_tx, bool in_tx_cp)
 {
        struct perf_event *event;
        struct perf_event_attr attr = {
@@ -177,6 +112,7 @@ static void reprogram_counter(struct kvm_pmc *pmc, u32 type,
                .exclude_kernel = exclude_kernel,
                .config = config,
        };
+
        if (in_tx)
                attr.config |= HSW_IN_TX;
        if (in_tx_cp)
@@ -188,33 +124,16 @@ static void reprogram_counter(struct kvm_pmc *pmc, u32 type,
                                                 intr ? kvm_perf_overflow_intr :
                                                 kvm_perf_overflow, pmc);
        if (IS_ERR(event)) {
-               printk_once("kvm: pmu event creation failed %ld\n",
-                               PTR_ERR(event));
+               printk_once("kvm_pmu: event creation failed %ld\n",
+                           PTR_ERR(event));
                return;
        }
 
        pmc->perf_event = event;
-       clear_bit(pmc->idx, (unsigned long*)&pmc->vcpu->arch.pmu.reprogram_pmi);
-}
-
-static unsigned find_arch_event(struct kvm_pmu *pmu, u8 event_select,
-               u8 unit_mask)
-{
-       int i;
-
-       for (i = 0; i < ARRAY_SIZE(arch_events); i++)
-               if (arch_events[i].eventsel == event_select
-                               && arch_events[i].unit_mask == unit_mask
-                               && (pmu->available_event_types & (1 << i)))
-                       break;
-
-       if (i == ARRAY_SIZE(arch_events))
-               return PERF_COUNT_HW_MAX;
-
-       return arch_events[i].event_type;
+       clear_bit(pmc->idx, (unsigned long*)&pmc_to_pmu(pmc)->reprogram_pmi);
 }
 
-static void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel)
+void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel)
 {
        unsigned config, type = PERF_TYPE_RAW;
        u8 event_select, unit_mask;
@@ -224,21 +143,22 @@ static void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel)
 
        pmc->eventsel = eventsel;
 
-       stop_counter(pmc);
+       pmc_stop_counter(pmc);
 
-       if (!(eventsel & ARCH_PERFMON_EVENTSEL_ENABLE) || !pmc_enabled(pmc))
+       if (!(eventsel & ARCH_PERFMON_EVENTSEL_ENABLE) || !pmc_is_enabled(pmc))
                return;
 
        event_select = eventsel & ARCH_PERFMON_EVENTSEL_EVENT;
        unit_mask = (eventsel & ARCH_PERFMON_EVENTSEL_UMASK) >> 8;
 
        if (!(eventsel & (ARCH_PERFMON_EVENTSEL_EDGE |
-                               ARCH_PERFMON_EVENTSEL_INV |
-                               ARCH_PERFMON_EVENTSEL_CMASK |
-                               HSW_IN_TX |
-                               HSW_IN_TX_CHECKPOINTED))) {
-               config = find_arch_event(&pmc->vcpu->arch.pmu, event_select,
-                               unit_mask);
+                         ARCH_PERFMON_EVENTSEL_INV |
+                         ARCH_PERFMON_EVENTSEL_CMASK |
+                         HSW_IN_TX |
+                         HSW_IN_TX_CHECKPOINTED))) {
+               config = kvm_x86_ops->pmu_ops->find_arch_event(pmc_to_pmu(pmc),
+                                                     event_select,
+                                                     unit_mask);
                if (config != PERF_COUNT_HW_MAX)
                        type = PERF_TYPE_HARDWARE;
        }
@@ -246,56 +166,36 @@ static void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel)
        if (type == PERF_TYPE_RAW)
                config = eventsel & X86_RAW_EVENT_MASK;
 
-       reprogram_counter(pmc, type, config,
-                       !(eventsel & ARCH_PERFMON_EVENTSEL_USR),
-                       !(eventsel & ARCH_PERFMON_EVENTSEL_OS),
-                       eventsel & ARCH_PERFMON_EVENTSEL_INT,
-                       (eventsel & HSW_IN_TX),
-                       (eventsel & HSW_IN_TX_CHECKPOINTED));
+       pmc_reprogram_counter(pmc, type, config,
+                             !(eventsel & ARCH_PERFMON_EVENTSEL_USR),
+                             !(eventsel & ARCH_PERFMON_EVENTSEL_OS),
+                             eventsel & ARCH_PERFMON_EVENTSEL_INT,
+                             (eventsel & HSW_IN_TX),
+                             (eventsel & HSW_IN_TX_CHECKPOINTED));
 }
+EXPORT_SYMBOL_GPL(reprogram_gp_counter);
 
-static void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 en_pmi, int idx)
+void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 ctrl, int idx)
 {
-       unsigned en = en_pmi & 0x3;
-       bool pmi = en_pmi & 0x8;
+       unsigned en_field = ctrl & 0x3;
+       bool pmi = ctrl & 0x8;
 
-       stop_counter(pmc);
+       pmc_stop_counter(pmc);
 
-       if (!en || !pmc_enabled(pmc))
+       if (!en_field || !pmc_is_enabled(pmc))
                return;
 
-       reprogram_counter(pmc, PERF_TYPE_HARDWARE,
-                       arch_events[fixed_pmc_events[idx]].event_type,
-                       !(en & 0x2), /* exclude user */
-                       !(en & 0x1), /* exclude kernel */
-                       pmi, false, false);
+       pmc_reprogram_counter(pmc, PERF_TYPE_HARDWARE,
+                             kvm_x86_ops->pmu_ops->find_fixed_event(idx),
+                             !(en_field & 0x2), /* exclude user */
+                             !(en_field & 0x1), /* exclude kernel */
+                             pmi, false, false);
 }
+EXPORT_SYMBOL_GPL(reprogram_fixed_counter);
 
-static inline u8 fixed_en_pmi(u64 ctrl, int idx)
+void reprogram_counter(struct kvm_pmu *pmu, int pmc_idx)
 {
-       return (ctrl >> (idx * 4)) & 0xf;
-}
-
-static void reprogram_fixed_counters(struct kvm_pmu *pmu, u64 data)
-{
-       int i;
-
-       for (i = 0; i < pmu->nr_arch_fixed_counters; i++) {
-               u8 en_pmi = fixed_en_pmi(data, i);
-               struct kvm_pmc *pmc = get_fixed_pmc_idx(pmu, i);
-
-               if (fixed_en_pmi(pmu->fixed_ctr_ctrl, i) == en_pmi)
-                       continue;
-
-               reprogram_fixed_counter(pmc, en_pmi, i);
-       }
-
-       pmu->fixed_ctr_ctrl = data;
-}
-
-static void reprogram_idx(struct kvm_pmu *pmu, int idx)
-{
-       struct kvm_pmc *pmc = global_idx_to_pmc(pmu, idx);
+       struct kvm_pmc *pmc = kvm_x86_ops->pmu_ops->pmc_idx_to_pmc(pmu, pmc_idx);
 
        if (!pmc)
                return;
@@ -303,274 +203,107 @@ static void reprogram_idx(struct kvm_pmu *pmu, int idx)
        if (pmc_is_gp(pmc))
                reprogram_gp_counter(pmc, pmc->eventsel);
        else {
-               int fidx = idx - INTEL_PMC_IDX_FIXED;
-               reprogram_fixed_counter(pmc,
-                               fixed_en_pmi(pmu->fixed_ctr_ctrl, fidx), fidx);
+               int idx = pmc_idx - INTEL_PMC_IDX_FIXED;
+               u8 ctrl = fixed_ctrl_field(pmu->fixed_ctr_ctrl, idx);
+
+               reprogram_fixed_counter(pmc, ctrl, idx);
        }
 }
+EXPORT_SYMBOL_GPL(reprogram_counter);
 
-static void global_ctrl_changed(struct kvm_pmu *pmu, u64 data)
+void kvm_pmu_handle_event(struct kvm_vcpu *vcpu)
 {
+       struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
+       u64 bitmask;
        int bit;
-       u64 diff = pmu->global_ctrl ^ data;
-
-       pmu->global_ctrl = data;
-
-       for_each_set_bit(bit, (unsigned long *)&diff, X86_PMC_IDX_MAX)
-               reprogram_idx(pmu, bit);
-}
 
-bool kvm_pmu_msr(struct kvm_vcpu *vcpu, u32 msr)
-{
-       struct kvm_pmu *pmu = &vcpu->arch.pmu;
-       int ret;
-
-       switch (msr) {
-       case MSR_CORE_PERF_FIXED_CTR_CTRL:
-       case MSR_CORE_PERF_GLOBAL_STATUS:
-       case MSR_CORE_PERF_GLOBAL_CTRL:
-       case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
-               ret = pmu->version > 1;
-               break;
-       default:
-               ret = get_gp_pmc(pmu, msr, MSR_IA32_PERFCTR0)
-                       || get_gp_pmc(pmu, msr, MSR_P6_EVNTSEL0)
-                       || get_fixed_pmc(pmu, msr);
-               break;
-       }
-       return ret;
-}
+       bitmask = pmu->reprogram_pmi;
 
-int kvm_pmu_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
-{
-       struct kvm_pmu *pmu = &vcpu->arch.pmu;
-       struct kvm_pmc *pmc;
+       for_each_set_bit(bit, (unsigned long *)&bitmask, X86_PMC_IDX_MAX) {
+               struct kvm_pmc *pmc = kvm_x86_ops->pmu_ops->pmc_idx_to_pmc(pmu, bit);
 
-       switch (index) {
-       case MSR_CORE_PERF_FIXED_CTR_CTRL:
-               *data = pmu->fixed_ctr_ctrl;
-               return 0;
-       case MSR_CORE_PERF_GLOBAL_STATUS:
-               *data = pmu->global_status;
-               return 0;
-       case MSR_CORE_PERF_GLOBAL_CTRL:
-               *data = pmu->global_ctrl;
-               return 0;
-       case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
-               *data = pmu->global_ovf_ctrl;
-               return 0;
-       default:
-               if ((pmc = get_gp_pmc(pmu, index, MSR_IA32_PERFCTR0)) ||
-                               (pmc = get_fixed_pmc(pmu, index))) {
-                       *data = read_pmc(pmc);
-                       return 0;
-               } else if ((pmc = get_gp_pmc(pmu, index, MSR_P6_EVNTSEL0))) {
-                       *data = pmc->eventsel;
-                       return 0;
+               if (unlikely(!pmc || !pmc->perf_event)) {
+                       clear_bit(bit, (unsigned long *)&pmu->reprogram_pmi);
+                       continue;
                }
-       }
-       return 1;
-}
 
-int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
-{
-       struct kvm_pmu *pmu = &vcpu->arch.pmu;
-       struct kvm_pmc *pmc;
-       u32 index = msr_info->index;
-       u64 data = msr_info->data;
-
-       switch (index) {
-       case MSR_CORE_PERF_FIXED_CTR_CTRL:
-               if (pmu->fixed_ctr_ctrl == data)
-                       return 0;
-               if (!(data & 0xfffffffffffff444ull)) {
-                       reprogram_fixed_counters(pmu, data);
-                       return 0;
-               }
-               break;
-       case MSR_CORE_PERF_GLOBAL_STATUS:
-               if (msr_info->host_initiated) {
-                       pmu->global_status = data;
-                       return 0;
-               }
-               break; /* RO MSR */
-       case MSR_CORE_PERF_GLOBAL_CTRL:
-               if (pmu->global_ctrl == data)
-                       return 0;
-               if (!(data & pmu->global_ctrl_mask)) {
-                       global_ctrl_changed(pmu, data);
-                       return 0;
-               }
-               break;
-       case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
-               if (!(data & (pmu->global_ctrl_mask & ~(3ull<<62)))) {
-                       if (!msr_info->host_initiated)
-                               pmu->global_status &= ~data;
-                       pmu->global_ovf_ctrl = data;
-                       return 0;
-               }
-               break;
-       default:
-               if ((pmc = get_gp_pmc(pmu, index, MSR_IA32_PERFCTR0)) ||
-                               (pmc = get_fixed_pmc(pmu, index))) {
-                       if (!msr_info->host_initiated)
-                               data = (s64)(s32)data;
-                       pmc->counter += data - read_pmc(pmc);
-                       return 0;
-               } else if ((pmc = get_gp_pmc(pmu, index, MSR_P6_EVNTSEL0))) {
-                       if (data == pmc->eventsel)
-                               return 0;
-                       if (!(data & pmu->reserved_bits)) {
-                               reprogram_gp_counter(pmc, data);
-                               return 0;
-                       }
-               }
+               reprogram_counter(pmu, bit);
        }
-       return 1;
 }
 
-int kvm_pmu_check_pmc(struct kvm_vcpu *vcpu, unsigned pmc)
+/* check if idx is a valid index to access PMU */
+int kvm_pmu_is_valid_msr_idx(struct kvm_vcpu *vcpu, unsigned idx)
 {
-       struct kvm_pmu *pmu = &vcpu->arch.pmu;
-       bool fixed = pmc & (1u << 30);
-       pmc &= ~(3u << 30);
-       return (!fixed && pmc >= pmu->nr_arch_gp_counters) ||
-               (fixed && pmc >= pmu->nr_arch_fixed_counters);
+       return kvm_x86_ops->pmu_ops->is_valid_msr_idx(vcpu, idx);
 }
 
-int kvm_pmu_read_pmc(struct kvm_vcpu *vcpu, unsigned pmc, u64 *data)
+int kvm_pmu_rdpmc(struct kvm_vcpu *vcpu, unsigned idx, u64 *data)
 {
-       struct kvm_pmu *pmu = &vcpu->arch.pmu;
-       bool fast_mode = pmc & (1u << 31);
-       bool fixed = pmc & (1u << 30);
-       struct kvm_pmc *counters;
-       u64 ctr;
-
-       pmc &= ~(3u << 30);
-       if (!fixed && pmc >= pmu->nr_arch_gp_counters)
-               return 1;
-       if (fixed && pmc >= pmu->nr_arch_fixed_counters)
+       bool fast_mode = idx & (1u << 31);
+       struct kvm_pmc *pmc;
+       u64 ctr_val;
+
+       pmc = kvm_x86_ops->pmu_ops->msr_idx_to_pmc(vcpu, idx);
+       if (!pmc)
                return 1;
-       counters = fixed ? pmu->fixed_counters : pmu->gp_counters;
-       ctr = read_pmc(&counters[pmc]);
+
+       ctr_val = pmc_read_counter(pmc);
        if (fast_mode)
-               ctr = (u32)ctr;
-       *data = ctr;
+               ctr_val = (u32)ctr_val;
 
+       *data = ctr_val;
        return 0;
 }
 
-void kvm_pmu_cpuid_update(struct kvm_vcpu *vcpu)
+void kvm_pmu_deliver_pmi(struct kvm_vcpu *vcpu)
 {
-       struct kvm_pmu *pmu = &vcpu->arch.pmu;
-       struct kvm_cpuid_entry2 *entry;
-       union cpuid10_eax eax;
-       union cpuid10_edx edx;
-
-       pmu->nr_arch_gp_counters = 0;
-       pmu->nr_arch_fixed_counters = 0;
-       pmu->counter_bitmask[KVM_PMC_GP] = 0;
-       pmu->counter_bitmask[KVM_PMC_FIXED] = 0;
-       pmu->version = 0;
-       pmu->reserved_bits = 0xffffffff00200000ull;
-
-       entry = kvm_find_cpuid_entry(vcpu, 0xa, 0);
-       if (!entry)
-               return;
-       eax.full = entry->eax;
-       edx.full = entry->edx;
-
-       pmu->version = eax.split.version_id;
-       if (!pmu->version)
-               return;
-
-       pmu->nr_arch_gp_counters = min_t(int, eax.split.num_counters,
-                                       INTEL_PMC_MAX_GENERIC);
-       pmu->counter_bitmask[KVM_PMC_GP] = ((u64)1 << eax.split.bit_width) - 1;
-       pmu->available_event_types = ~entry->ebx &
-                                       ((1ull << eax.split.mask_length) - 1);
-
-       if (pmu->version == 1) {
-               pmu->nr_arch_fixed_counters = 0;
-       } else {
-               pmu->nr_arch_fixed_counters =
-                       min_t(int, edx.split.num_counters_fixed,
-                               INTEL_PMC_MAX_FIXED);
-               pmu->counter_bitmask[KVM_PMC_FIXED] =
-                       ((u64)1 << edx.split.bit_width_fixed) - 1;
-       }
+       if (vcpu->arch.apic)
+               kvm_apic_local_deliver(vcpu->arch.apic, APIC_LVTPC);
+}
 
-       pmu->global_ctrl = ((1 << pmu->nr_arch_gp_counters) - 1) |
-               (((1ull << pmu->nr_arch_fixed_counters) - 1) << INTEL_PMC_IDX_FIXED);
-       pmu->global_ctrl_mask = ~pmu->global_ctrl;
+bool kvm_pmu_is_valid_msr(struct kvm_vcpu *vcpu, u32 msr)
+{
+       return kvm_x86_ops->pmu_ops->is_valid_msr(vcpu, msr);
+}
 
-       entry = kvm_find_cpuid_entry(vcpu, 7, 0);
-       if (entry &&
-           (boot_cpu_has(X86_FEATURE_HLE) || boot_cpu_has(X86_FEATURE_RTM)) &&
-           (entry->ebx & (X86_FEATURE_HLE|X86_FEATURE_RTM)))
-               pmu->reserved_bits ^= HSW_IN_TX|HSW_IN_TX_CHECKPOINTED;
+int kvm_pmu_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
+{
+       return kvm_x86_ops->pmu_ops->get_msr(vcpu, msr, data);
 }
 
-void kvm_pmu_init(struct kvm_vcpu *vcpu)
+int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 {
-       int i;
-       struct kvm_pmu *pmu = &vcpu->arch.pmu;
+       return kvm_x86_ops->pmu_ops->set_msr(vcpu, msr_info);
+}
 
-       memset(pmu, 0, sizeof(*pmu));
-       for (i = 0; i < INTEL_PMC_MAX_GENERIC; i++) {
-               pmu->gp_counters[i].type = KVM_PMC_GP;
-               pmu->gp_counters[i].vcpu = vcpu;
-               pmu->gp_counters[i].idx = i;
-       }
-       for (i = 0; i < INTEL_PMC_MAX_FIXED; i++) {
-               pmu->fixed_counters[i].type = KVM_PMC_FIXED;
-               pmu->fixed_counters[i].vcpu = vcpu;
-               pmu->fixed_counters[i].idx = i + INTEL_PMC_IDX_FIXED;
-       }
-       init_irq_work(&pmu->irq_work, trigger_pmi);
-       kvm_pmu_cpuid_update(vcpu);
+/* refresh PMU settings. This function generally is called when underlying
+ * settings are changed (such as changes of PMU CPUID by guest VMs), which
+ * should rarely happen.
+ */
+void kvm_pmu_refresh(struct kvm_vcpu *vcpu)
+{
+       kvm_x86_ops->pmu_ops->refresh(vcpu);
 }
 
 void kvm_pmu_reset(struct kvm_vcpu *vcpu)
 {
-       struct kvm_pmu *pmu = &vcpu->arch.pmu;
-       int i;
+       struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
 
        irq_work_sync(&pmu->irq_work);
-       for (i = 0; i < INTEL_PMC_MAX_GENERIC; i++) {
-               struct kvm_pmc *pmc = &pmu->gp_counters[i];
-               stop_counter(pmc);
-               pmc->counter = pmc->eventsel = 0;
-       }
+       kvm_x86_ops->pmu_ops->reset(vcpu);
+}
 
-       for (i = 0; i < INTEL_PMC_MAX_FIXED; i++)
-               stop_counter(&pmu->fixed_counters[i]);
+void kvm_pmu_init(struct kvm_vcpu *vcpu)
+{
+       struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
 
-       pmu->fixed_ctr_ctrl = pmu->global_ctrl = pmu->global_status =
-               pmu->global_ovf_ctrl = 0;
+       memset(pmu, 0, sizeof(*pmu));
+       kvm_x86_ops->pmu_ops->init(vcpu);
+       init_irq_work(&pmu->irq_work, kvm_pmi_trigger_fn);
+       kvm_pmu_refresh(vcpu);
 }
 
 void kvm_pmu_destroy(struct kvm_vcpu *vcpu)
 {
        kvm_pmu_reset(vcpu);
 }
-
-void kvm_handle_pmu_event(struct kvm_vcpu *vcpu)
-{
-       struct kvm_pmu *pmu = &vcpu->arch.pmu;
-       u64 bitmask;
-       int bit;
-
-       bitmask = pmu->reprogram_pmi;
-
-       for_each_set_bit(bit, (unsigned long *)&bitmask, X86_PMC_IDX_MAX) {
-               struct kvm_pmc *pmc = global_idx_to_pmc(pmu, bit);
-
-               if (unlikely(!pmc || !pmc->perf_event)) {
-                       clear_bit(bit, (unsigned long *)&pmu->reprogram_pmi);
-                       continue;
-               }
-
-               reprogram_idx(pmu, bit);
-       }
-}