KVM: selftests: Copy full counter values from guest in PMU event filter test
authorSean Christopherson <seanjc@google.com>
Fri, 7 Apr 2023 23:32:53 +0000 (16:32 -0700)
committerSean Christopherson <seanjc@google.com>
Fri, 14 Apr 2023 20:21:32 +0000 (13:21 -0700)
Use a single struct to track all PMC event counts in the PMU filter test,
and copy the full struct to/from the guest when running and measuring each
guest workload.  Using a common struct avoids naming conflicts, e.g. the
loads/stores testcase has claimed "perf_counter", and eliminates the
unnecessary truncation of the counter values when they are propagated from
the guest MSRs to the host structs.

Zero the struct before running the guest workload to ensure that the test
doesn't get a false pass due to consuming data from a previous run.

Link: https://lore.kernel.org/r/20230407233254.957013-6-seanjc@google.com
Reviewed by: Aaron Lewis <aaronlewis@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c

index 0329c43..58d2e17 100644 (file)
@@ -71,6 +71,13 @@ static const uint64_t event_list[] = {
        AMD_ZEN_BR_RETIRED,
 };
 
+struct {
+       uint64_t loads;
+       uint64_t stores;
+       uint64_t loads_stores;
+       uint64_t branches_retired;
+} pmc_results;
+
 /*
  * If we encounter a #GP during the guest PMU sanity check, then the guest
  * PMU is not functional. Inform the hypervisor via GUEST_SYNC(0).
@@ -100,13 +107,13 @@ static void check_msr(uint32_t msr, uint64_t bits_to_flip)
                GUEST_SYNC(-EIO);
 }
 
-static uint64_t run_and_measure_loop(uint32_t msr_base)
+static void run_and_measure_loop(uint32_t msr_base)
 {
-       uint64_t branches_retired = rdmsr(msr_base + 0);
+       const uint64_t branches_retired = rdmsr(msr_base + 0);
 
        __asm__ __volatile__("loop ." : "+c"((int){NUM_BRANCHES}));
 
-       return rdmsr(msr_base + 0) - branches_retired;
+       pmc_results.branches_retired = rdmsr(msr_base + 0) - branches_retired;
 }
 
 static void intel_guest_code(void)
@@ -117,15 +124,13 @@ static void intel_guest_code(void)
        GUEST_SYNC(0);
 
        for (;;) {
-               uint64_t count;
-
                wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, 0);
                wrmsr(MSR_P6_EVNTSEL0, ARCH_PERFMON_EVENTSEL_ENABLE |
                      ARCH_PERFMON_EVENTSEL_OS | INTEL_BR_RETIRED);
                wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, 0x1);
 
-               count = run_and_measure_loop(MSR_IA32_PMC0);
-               GUEST_SYNC(count);
+               run_and_measure_loop(MSR_IA32_PMC0);
+               GUEST_SYNC(0);
        }
 }
 
@@ -141,14 +146,12 @@ static void amd_guest_code(void)
        GUEST_SYNC(0);
 
        for (;;) {
-               uint64_t count;
-
                wrmsr(MSR_K7_EVNTSEL0, 0);
                wrmsr(MSR_K7_EVNTSEL0, ARCH_PERFMON_EVENTSEL_ENABLE |
                      ARCH_PERFMON_EVENTSEL_OS | AMD_ZEN_BR_RETIRED);
 
-               count = run_and_measure_loop(MSR_K7_PERFCTR0);
-               GUEST_SYNC(count);
+               run_and_measure_loop(MSR_K7_PERFCTR0);
+               GUEST_SYNC(0);
        }
 }
 
@@ -168,6 +171,19 @@ static uint64_t run_vcpu_to_sync(struct kvm_vcpu *vcpu)
        return uc.args[1];
 }
 
+static void run_vcpu_and_sync_pmc_results(struct kvm_vcpu *vcpu)
+{
+       uint64_t r;
+
+       memset(&pmc_results, 0, sizeof(pmc_results));
+       sync_global_to_guest(vcpu->vm, pmc_results);
+
+       r = run_vcpu_to_sync(vcpu);
+       TEST_ASSERT(!r, "Unexpected sync value: 0x%lx", r);
+
+       sync_global_from_guest(vcpu->vm, pmc_results);
+}
+
 /*
  * In a nested environment or if the vPMU is disabled, the guest PMU
  * might not work as architected (accessing the PMU MSRs may raise
@@ -244,92 +260,93 @@ static struct kvm_pmu_event_filter *remove_event(struct kvm_pmu_event_filter *f,
        return f;
 }
 
-#define ASSERT_PMC_COUNTING_INSTRUCTIONS(count)                                                \
+#define ASSERT_PMC_COUNTING_INSTRUCTIONS()                                             \
 do {                                                                                   \
-       if (count && count != NUM_BRANCHES)                                             \
+       uint64_t br = pmc_results.branches_retired;                                     \
+                                                                                       \
+       if (br && br != NUM_BRANCHES)                                                   \
                pr_info("%s: Branch instructions retired = %lu (expected %u)\n",        \
-                       __func__, count, NUM_BRANCHES);                                 \
-       TEST_ASSERT(count, "%s: Branch instructions retired = %lu (expected > 0)",      \
-                   __func__, count);                                                   \
+                       __func__, br, NUM_BRANCHES);                                    \
+       TEST_ASSERT(br, "%s: Branch instructions retired = %lu (expected > 0)",         \
+                   __func__, br);                                                      \
 } while (0)
 
-#define ASSERT_PMC_NOT_COUNTING_INSTRUCTIONS(count)                                    \
+#define ASSERT_PMC_NOT_COUNTING_INSTRUCTIONS()                                         \
 do {                                                                                   \
-       TEST_ASSERT(!count, "%s: Branch instructions retired = %lu (expected 0)",       \
-                   __func__, count);                                                   \
+       uint64_t br = pmc_results.branches_retired;                                     \
+                                                                                       \
+       TEST_ASSERT(!br, "%s: Branch instructions retired = %lu (expected 0)",          \
+                   __func__, br);                                                      \
 } while (0)
 
 static void test_without_filter(struct kvm_vcpu *vcpu)
 {
-       uint64_t count = run_vcpu_to_sync(vcpu);
+       run_vcpu_and_sync_pmc_results(vcpu);
 
-       ASSERT_PMC_COUNTING_INSTRUCTIONS(count);
+       ASSERT_PMC_COUNTING_INSTRUCTIONS();
 }
 
-static uint64_t test_with_filter(struct kvm_vcpu *vcpu,
-                                struct kvm_pmu_event_filter *f)
+static void test_with_filter(struct kvm_vcpu *vcpu,
+                            struct kvm_pmu_event_filter *f)
 {
        vm_ioctl(vcpu->vm, KVM_SET_PMU_EVENT_FILTER, f);
-       return run_vcpu_to_sync(vcpu);
+       run_vcpu_and_sync_pmc_results(vcpu);
 }
 
 static void test_amd_deny_list(struct kvm_vcpu *vcpu)
 {
        uint64_t event = EVENT(0x1C2, 0);
        struct kvm_pmu_event_filter *f;
-       uint64_t count;
 
        f = create_pmu_event_filter(&event, 1, KVM_PMU_EVENT_DENY, 0);
-       count = test_with_filter(vcpu, f);
+       test_with_filter(vcpu, f);
        free(f);
 
-       ASSERT_PMC_COUNTING_INSTRUCTIONS(count);
+       ASSERT_PMC_COUNTING_INSTRUCTIONS();
 }
 
 static void test_member_deny_list(struct kvm_vcpu *vcpu)
 {
        struct kvm_pmu_event_filter *f = event_filter(KVM_PMU_EVENT_DENY);
-       uint64_t count = test_with_filter(vcpu, f);
 
+       test_with_filter(vcpu, f);
        free(f);
 
-       ASSERT_PMC_NOT_COUNTING_INSTRUCTIONS(count);
+       ASSERT_PMC_NOT_COUNTING_INSTRUCTIONS();
 }
 
 static void test_member_allow_list(struct kvm_vcpu *vcpu)
 {
        struct kvm_pmu_event_filter *f = event_filter(KVM_PMU_EVENT_ALLOW);
-       uint64_t count = test_with_filter(vcpu, f);
 
+       test_with_filter(vcpu, f);
        free(f);
 
-       ASSERT_PMC_COUNTING_INSTRUCTIONS(count);
+       ASSERT_PMC_COUNTING_INSTRUCTIONS();
 }
 
 static void test_not_member_deny_list(struct kvm_vcpu *vcpu)
 {
        struct kvm_pmu_event_filter *f = event_filter(KVM_PMU_EVENT_DENY);
-       uint64_t count;
 
        remove_event(f, INTEL_BR_RETIRED);
        remove_event(f, AMD_ZEN_BR_RETIRED);
-       count = test_with_filter(vcpu, f);
+       test_with_filter(vcpu, f);
        free(f);
 
-       ASSERT_PMC_COUNTING_INSTRUCTIONS(count);
+       ASSERT_PMC_COUNTING_INSTRUCTIONS();
 }
 
 static void test_not_member_allow_list(struct kvm_vcpu *vcpu)
 {
        struct kvm_pmu_event_filter *f = event_filter(KVM_PMU_EVENT_ALLOW);
-       uint64_t count;
 
        remove_event(f, INTEL_BR_RETIRED);
        remove_event(f, AMD_ZEN_BR_RETIRED);
-       count = test_with_filter(vcpu, f);
+       test_with_filter(vcpu, f);
        free(f);
 
-       ASSERT_PMC_NOT_COUNTING_INSTRUCTIONS(count);
+       ASSERT_PMC_NOT_COUNTING_INSTRUCTIONS();
 }
 
 /*
@@ -458,51 +475,30 @@ static bool supports_event_mem_inst_retired(void)
 #define EXCLUDE_MASKED_ENTRY(event_select, mask, match) \
        KVM_PMU_ENCODE_MASKED_ENTRY(event_select, mask, match, true)
 
-struct perf_counter {
-       union {
-               uint64_t raw;
-               struct {
-                       uint64_t loads:22;
-                       uint64_t stores:22;
-                       uint64_t loads_stores:20;
-               };
-       };
-};
-
-static uint64_t masked_events_guest_test(uint32_t msr_base)
+static void masked_events_guest_test(uint32_t msr_base)
 {
-       uint64_t ld0, ld1, st0, st1, ls0, ls1;
-       struct perf_counter c;
-       int val;
-
        /*
-        * The acutal value of the counters don't determine the outcome of
+        * The actual value of the counters don't determine the outcome of
         * the test.  Only that they are zero or non-zero.
         */
-       ld0 = rdmsr(msr_base + 0);
-       st0 = rdmsr(msr_base + 1);
-       ls0 = rdmsr(msr_base + 2);
+       const uint64_t loads = rdmsr(msr_base + 0);
+       const uint64_t stores = rdmsr(msr_base + 1);
+       const uint64_t loads_stores = rdmsr(msr_base + 2);
+       int val;
+
 
        __asm__ __volatile__("movl $0, %[v];"
                             "movl %[v], %%eax;"
                             "incl %[v];"
                             : [v]"+m"(val) :: "eax");
 
-       ld1 = rdmsr(msr_base + 0);
-       st1 = rdmsr(msr_base + 1);
-       ls1 = rdmsr(msr_base + 2);
-
-       c.loads = ld1 - ld0;
-       c.stores = st1 - st0;
-       c.loads_stores = ls1 - ls0;
-
-       return c.raw;
+       pmc_results.loads = rdmsr(msr_base + 0) - loads;
+       pmc_results.stores = rdmsr(msr_base + 1) - stores;
+       pmc_results.loads_stores = rdmsr(msr_base + 2) - loads_stores;
 }
 
 static void intel_masked_events_guest_code(void)
 {
-       uint64_t r;
-
        for (;;) {
                wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, 0);
 
@@ -515,16 +511,13 @@ static void intel_masked_events_guest_code(void)
 
                wrmsr(MSR_CORE_PERF_GLOBAL_CTRL, 0x7);
 
-               r = masked_events_guest_test(MSR_IA32_PMC0);
-
-               GUEST_SYNC(r);
+               masked_events_guest_test(MSR_IA32_PMC0);
+               GUEST_SYNC(0);
        }
 }
 
 static void amd_masked_events_guest_code(void)
 {
-       uint64_t r;
-
        for (;;) {
                wrmsr(MSR_K7_EVNTSEL0, 0);
                wrmsr(MSR_K7_EVNTSEL1, 0);
@@ -537,26 +530,22 @@ static void amd_masked_events_guest_code(void)
                wrmsr(MSR_K7_EVNTSEL2, ARCH_PERFMON_EVENTSEL_ENABLE |
                      ARCH_PERFMON_EVENTSEL_OS | LS_DISPATCH_LOAD_STORE);
 
-               r = masked_events_guest_test(MSR_K7_PERFCTR0);
-
-               GUEST_SYNC(r);
+               masked_events_guest_test(MSR_K7_PERFCTR0);
+               GUEST_SYNC(0);
        }
 }
 
-static struct perf_counter run_masked_events_test(struct kvm_vcpu *vcpu,
-                                                const uint64_t masked_events[],
-                                                const int nmasked_events)
+static void run_masked_events_test(struct kvm_vcpu *vcpu,
+                                  const uint64_t masked_events[],
+                                  const int nmasked_events)
 {
        struct kvm_pmu_event_filter *f;
-       struct perf_counter r;
 
        f = create_pmu_event_filter(masked_events, nmasked_events,
                                    KVM_PMU_EVENT_ALLOW,
                                    KVM_PMU_EVENT_FLAG_MASKED_EVENTS);
-       r.raw = test_with_filter(vcpu, f);
+       test_with_filter(vcpu, f);
        free(f);
-
-       return r;
 }
 
 /* Matches KVM_PMU_EVENT_FILTER_MAX_EVENTS in pmu.c */
@@ -681,7 +670,6 @@ static void run_masked_events_tests(struct kvm_vcpu *vcpu, uint64_t *events,
                                    int nevents)
 {
        int ntests = ARRAY_SIZE(test_cases);
-       struct perf_counter c;
        int i, n;
 
        for (i = 0; i < ntests; i++) {
@@ -693,13 +681,15 @@ static void run_masked_events_tests(struct kvm_vcpu *vcpu, uint64_t *events,
 
                n = append_test_events(test, events, nevents);
 
-               c = run_masked_events_test(vcpu, events, n);
-               TEST_ASSERT(bool_eq(c.loads, test->flags & ALLOW_LOADS) &&
-                           bool_eq(c.stores, test->flags & ALLOW_STORES) &&
-                           bool_eq(c.loads_stores,
+               run_masked_events_test(vcpu, events, n);
+
+               TEST_ASSERT(bool_eq(pmc_results.loads, test->flags & ALLOW_LOADS) &&
+                           bool_eq(pmc_results.stores, test->flags & ALLOW_STORES) &&
+                           bool_eq(pmc_results.loads_stores,
                                    test->flags & ALLOW_LOADS_STORES),
-                           "%s  loads: %u, stores: %u, loads + stores: %u",
-                           test->msg, c.loads, c.stores, c.loads_stores);
+                           "%s  loads: %lu, stores: %lu, loads + stores: %lu",
+                           test->msg, pmc_results.loads, pmc_results.stores,
+                           pmc_results.loads_stores);
        }
 }