]> www.infradead.org Git - users/hch/misc.git/commitdiff
KVM: arm64: selftests: Use the vCPU attr for setting nr of PMU counters
authorOliver Upton <oliver.upton@linux.dev>
Wed, 17 Sep 2025 21:20:40 +0000 (14:20 -0700)
committerMarc Zyngier <maz@kernel.org>
Wed, 24 Sep 2025 18:23:32 +0000 (19:23 +0100)
Configuring the number of implemented counters via PMCR_EL0.N was a bad
idea in retrospect as it interacts poorly with nested. Migrate the
selftest to use the vCPU attribute instead of the KVM_SET_ONE_REG
mechanism.

Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
Signed-off-by: Marc Zyngier <maz@kernel.org>
tools/testing/selftests/kvm/arm64/vpmu_counter_access.c

index 2a8f31c8e59fe8dd5a211fdfc4d27c1f58696be2..ae36325c022fb20544b5d185e74170685b421484 100644 (file)
@@ -44,11 +44,6 @@ static uint64_t get_pmcr_n(uint64_t pmcr)
        return FIELD_GET(ARMV8_PMU_PMCR_N, pmcr);
 }
 
-static void set_pmcr_n(uint64_t *pmcr, uint64_t pmcr_n)
-{
-       u64p_replace_bits((__u64 *) pmcr, pmcr_n, ARMV8_PMU_PMCR_N);
-}
-
 static uint64_t get_counters_mask(uint64_t n)
 {
        uint64_t mask = BIT(ARMV8_PMU_CYCLE_IDX);
@@ -414,10 +409,6 @@ static void create_vpmu_vm(void *guest_code)
                .attr = KVM_ARM_VCPU_PMU_V3_IRQ,
                .addr = (uint64_t)&irq,
        };
-       struct kvm_device_attr init_attr = {
-               .group = KVM_ARM_VCPU_PMU_V3_CTRL,
-               .attr = KVM_ARM_VCPU_PMU_V3_INIT,
-       };
 
        /* The test creates the vpmu_vm multiple times. Ensure a clean state */
        memset(&vpmu_vm, 0, sizeof(vpmu_vm));
@@ -444,9 +435,7 @@ static void create_vpmu_vm(void *guest_code)
                    pmuver >= ID_AA64DFR0_EL1_PMUVer_IMP,
                    "Unexpected PMUVER (0x%x) on the vCPU with PMUv3", pmuver);
 
-       /* Initialize vPMU */
        vcpu_ioctl(vpmu_vm.vcpu, KVM_SET_DEVICE_ATTR, &irq_attr);
-       vcpu_ioctl(vpmu_vm.vcpu, KVM_SET_DEVICE_ATTR, &init_attr);
 }
 
 static void destroy_vpmu_vm(void)
@@ -472,33 +461,28 @@ static void run_vcpu(struct kvm_vcpu *vcpu, uint64_t pmcr_n)
        }
 }
 
-static void test_create_vpmu_vm_with_pmcr_n(uint64_t pmcr_n, bool expect_fail)
+static void test_create_vpmu_vm_with_nr_counters(unsigned int nr_counters, bool expect_fail)
 {
        struct kvm_vcpu *vcpu;
-       uint64_t pmcr, pmcr_orig;
+       unsigned int prev;
+       int ret;
 
        create_vpmu_vm(guest_code);
        vcpu = vpmu_vm.vcpu;
 
-       pmcr_orig = vcpu_get_reg(vcpu, KVM_ARM64_SYS_REG(SYS_PMCR_EL0));
-       pmcr = pmcr_orig;
+       prev = get_pmcr_n(vcpu_get_reg(vcpu, KVM_ARM64_SYS_REG(SYS_PMCR_EL0)));
 
-       /*
-        * Setting a larger value of PMCR.N should not modify the field, and
-        * return a success.
-        */
-       set_pmcr_n(&pmcr, pmcr_n);
-       vcpu_set_reg(vcpu, KVM_ARM64_SYS_REG(SYS_PMCR_EL0), pmcr);
-       pmcr = vcpu_get_reg(vcpu, KVM_ARM64_SYS_REG(SYS_PMCR_EL0));
+       ret = __vcpu_device_attr_set(vcpu, KVM_ARM_VCPU_PMU_V3_CTRL,
+                                    KVM_ARM_VCPU_PMU_V3_SET_NR_COUNTERS, &nr_counters);
 
        if (expect_fail)
-               TEST_ASSERT(pmcr_orig == pmcr,
-                           "PMCR.N modified by KVM to a larger value (PMCR: 0x%lx) for pmcr_n: 0x%lx",
-                           pmcr, pmcr_n);
+               TEST_ASSERT(ret && errno == EINVAL,
+                           "Setting more PMU counters (%u) than available (%u) unexpectedly succeeded",
+                           nr_counters, prev);
        else
-               TEST_ASSERT(pmcr_n == get_pmcr_n(pmcr),
-                           "Failed to update PMCR.N to %lu (received: %lu)",
-                           pmcr_n, get_pmcr_n(pmcr));
+               TEST_ASSERT(!ret, KVM_IOCTL_ERROR(KVM_SET_DEVICE_ATTR, ret));
+
+       vcpu_device_attr_set(vcpu, KVM_ARM_VCPU_PMU_V3_CTRL, KVM_ARM_VCPU_PMU_V3_INIT, NULL);
 }
 
 /*
@@ -513,7 +497,7 @@ static void run_access_test(uint64_t pmcr_n)
 
        pr_debug("Test with pmcr_n %lu\n", pmcr_n);
 
-       test_create_vpmu_vm_with_pmcr_n(pmcr_n, false);
+       test_create_vpmu_vm_with_nr_counters(pmcr_n, false);
        vcpu = vpmu_vm.vcpu;
 
        /* Save the initial sp to restore them later to run the guest again */
@@ -554,7 +538,7 @@ static void run_pmregs_validity_test(uint64_t pmcr_n)
        uint64_t set_reg_id, clr_reg_id, reg_val;
        uint64_t valid_counters_mask, max_counters_mask;
 
-       test_create_vpmu_vm_with_pmcr_n(pmcr_n, false);
+       test_create_vpmu_vm_with_nr_counters(pmcr_n, false);
        vcpu = vpmu_vm.vcpu;
 
        valid_counters_mask = get_counters_mask(pmcr_n);
@@ -608,7 +592,7 @@ static void run_error_test(uint64_t pmcr_n)
 {
        pr_debug("Error test with pmcr_n %lu (larger than the host)\n", pmcr_n);
 
-       test_create_vpmu_vm_with_pmcr_n(pmcr_n, true);
+       test_create_vpmu_vm_with_nr_counters(pmcr_n, true);
        destroy_vpmu_vm();
 }
 
@@ -626,12 +610,25 @@ static uint64_t get_pmcr_n_limit(void)
        return get_pmcr_n(pmcr);
 }
 
+static bool kvm_supports_nr_counters_attr(void)
+{
+       bool supported;
+
+       create_vpmu_vm(NULL);
+       supported = !__vcpu_has_device_attr(vpmu_vm.vcpu, KVM_ARM_VCPU_PMU_V3_CTRL,
+                                           KVM_ARM_VCPU_PMU_V3_SET_NR_COUNTERS);
+       destroy_vpmu_vm();
+
+       return supported;
+}
+
 int main(void)
 {
        uint64_t i, pmcr_n;
 
        TEST_REQUIRE(kvm_has_cap(KVM_CAP_ARM_PMU_V3));
        TEST_REQUIRE(kvm_supports_vgic_v3());
+       TEST_REQUIRE(kvm_supports_nr_counters_attr());
 
        pmcr_n = get_pmcr_n_limit();
        for (i = 0; i <= pmcr_n; i++) {