]> xenbits.xensource.com Git - people/pauldu/linux.git/commitdiff
KVM: selftests: aarch64: Introduce vpmu_counter_access test
authorReiji Watanabe <reijiw@google.com>
Fri, 20 Oct 2023 21:40:49 +0000 (21:40 +0000)
committerOliver Upton <oliver.upton@linux.dev>
Tue, 24 Oct 2023 22:59:31 +0000 (22:59 +0000)
Introduce vpmu_counter_access test for arm64 platforms.
The test configures PMUv3 for a vCPU, sets PMCR_EL0.N for the vCPU,
and check if the guest can consistently see the same number of the
PMU event counters (PMCR_EL0.N) that userspace sets.
This test case is done with each of the PMCR_EL0.N values from
0 to 31 (With the PMCR_EL0.N values greater than the host value,
the test expects KVM_SET_ONE_REG for the PMCR_EL0 to fail).

Signed-off-by: Reiji Watanabe <reijiw@google.com>
Signed-off-by: Raghavendra Rao Ananta <rananta@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20231020214053.2144305-10-rananta@google.com
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
tools/testing/selftests/kvm/Makefile
tools/testing/selftests/kvm/aarch64/vpmu_counter_access.c [new file with mode: 0644]

index a3bb36fb3cfc55a423d31c10e86c1fe16b79ca0e..416700aa196caed568756fa5670ed6698faa683b 100644 (file)
@@ -149,6 +149,7 @@ TEST_GEN_PROGS_aarch64 += aarch64/smccc_filter
 TEST_GEN_PROGS_aarch64 += aarch64/vcpu_width_config
 TEST_GEN_PROGS_aarch64 += aarch64/vgic_init
 TEST_GEN_PROGS_aarch64 += aarch64/vgic_irq
+TEST_GEN_PROGS_aarch64 += aarch64/vpmu_counter_access
 TEST_GEN_PROGS_aarch64 += access_tracking_perf_test
 TEST_GEN_PROGS_aarch64 += demand_paging_test
 TEST_GEN_PROGS_aarch64 += dirty_log_test
diff --git a/tools/testing/selftests/kvm/aarch64/vpmu_counter_access.c b/tools/testing/selftests/kvm/aarch64/vpmu_counter_access.c
new file mode 100644 (file)
index 0000000..4c6e1fe
--- /dev/null
@@ -0,0 +1,255 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * vpmu_counter_access - Test vPMU event counter access
+ *
+ * Copyright (c) 2023 Google LLC.
+ *
+ * This test checks if the guest can see the same number of the PMU event
+ * counters (PMCR_EL0.N) that userspace sets.
+ * This test runs only when KVM_CAP_ARM_PMU_V3 is supported on the host.
+ */
+#include <kvm_util.h>
+#include <processor.h>
+#include <test_util.h>
+#include <vgic.h>
+#include <perf/arm_pmuv3.h>
+#include <linux/bitfield.h>
+
+/* The max number of the PMU event counters (excluding the cycle counter) */
+#define ARMV8_PMU_MAX_GENERAL_COUNTERS (ARMV8_PMU_MAX_COUNTERS - 1)
+
+struct vpmu_vm {
+       struct kvm_vm *vm;
+       struct kvm_vcpu *vcpu;
+       int gic_fd;
+};
+
+static struct vpmu_vm vpmu_vm;
+
+static uint64_t get_pmcr_n(uint64_t pmcr)
+{
+       return (pmcr >> ARMV8_PMU_PMCR_N_SHIFT) & ARMV8_PMU_PMCR_N_MASK;
+}
+
+static void set_pmcr_n(uint64_t *pmcr, uint64_t pmcr_n)
+{
+       *pmcr = *pmcr & ~(ARMV8_PMU_PMCR_N_MASK << ARMV8_PMU_PMCR_N_SHIFT);
+       *pmcr |= (pmcr_n << ARMV8_PMU_PMCR_N_SHIFT);
+}
+
+static void guest_sync_handler(struct ex_regs *regs)
+{
+       uint64_t esr, ec;
+
+       esr = read_sysreg(esr_el1);
+       ec = (esr >> ESR_EC_SHIFT) & ESR_EC_MASK;
+       __GUEST_ASSERT(0, "PC: 0x%lx; ESR: 0x%lx; EC: 0x%lx", regs->pc, esr, ec);
+}
+
+/*
+ * The guest is configured with PMUv3 with @expected_pmcr_n number of
+ * event counters.
+ * Check if @expected_pmcr_n is consistent with PMCR_EL0.N.
+ */
+static void guest_code(uint64_t expected_pmcr_n)
+{
+       uint64_t pmcr, pmcr_n;
+
+       __GUEST_ASSERT(expected_pmcr_n <= ARMV8_PMU_MAX_GENERAL_COUNTERS,
+                       "Expected PMCR.N: 0x%lx; ARMv8 general counters: 0x%lx",
+                       expected_pmcr_n, ARMV8_PMU_MAX_GENERAL_COUNTERS);
+
+       pmcr = read_sysreg(pmcr_el0);
+       pmcr_n = get_pmcr_n(pmcr);
+
+       /* Make sure that PMCR_EL0.N indicates the value userspace set */
+       __GUEST_ASSERT(pmcr_n == expected_pmcr_n,
+                       "Expected PMCR.N: 0x%lx, PMCR.N: 0x%lx",
+                       expected_pmcr_n, pmcr_n);
+
+       GUEST_DONE();
+}
+
+#define GICD_BASE_GPA  0x8000000ULL
+#define GICR_BASE_GPA  0x80A0000ULL
+
+/* Create a VM that has one vCPU with PMUv3 configured. */
+static void create_vpmu_vm(void *guest_code)
+{
+       struct kvm_vcpu_init init;
+       uint8_t pmuver, ec;
+       uint64_t dfr0, irq = 23;
+       struct kvm_device_attr irq_attr = {
+               .group = KVM_ARM_VCPU_PMU_V3_CTRL,
+               .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));
+
+       vpmu_vm.vm = vm_create(1);
+       vm_init_descriptor_tables(vpmu_vm.vm);
+       for (ec = 0; ec < ESR_EC_NUM; ec++) {
+               vm_install_sync_handler(vpmu_vm.vm, VECTOR_SYNC_CURRENT, ec,
+                                       guest_sync_handler);
+       }
+
+       /* Create vCPU with PMUv3 */
+       vm_ioctl(vpmu_vm.vm, KVM_ARM_PREFERRED_TARGET, &init);
+       init.features[0] |= (1 << KVM_ARM_VCPU_PMU_V3);
+       vpmu_vm.vcpu = aarch64_vcpu_add(vpmu_vm.vm, 0, &init, guest_code);
+       vcpu_init_descriptor_tables(vpmu_vm.vcpu);
+       vpmu_vm.gic_fd = vgic_v3_setup(vpmu_vm.vm, 1, 64,
+                                       GICD_BASE_GPA, GICR_BASE_GPA);
+       __TEST_REQUIRE(vpmu_vm.gic_fd >= 0,
+                      "Failed to create vgic-v3, skipping");
+
+       /* Make sure that PMUv3 support is indicated in the ID register */
+       vcpu_get_reg(vpmu_vm.vcpu,
+                    KVM_ARM64_SYS_REG(SYS_ID_AA64DFR0_EL1), &dfr0);
+       pmuver = FIELD_GET(ARM64_FEATURE_MASK(ID_AA64DFR0_EL1_PMUVer), dfr0);
+       TEST_ASSERT(pmuver != ID_AA64DFR0_EL1_PMUVer_IMP_DEF &&
+                   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)
+{
+       close(vpmu_vm.gic_fd);
+       kvm_vm_free(vpmu_vm.vm);
+}
+
+static void run_vcpu(struct kvm_vcpu *vcpu, uint64_t pmcr_n)
+{
+       struct ucall uc;
+
+       vcpu_args_set(vcpu, 1, pmcr_n);
+       vcpu_run(vcpu);
+       switch (get_ucall(vcpu, &uc)) {
+       case UCALL_ABORT:
+               REPORT_GUEST_ASSERT(uc);
+               break;
+       case UCALL_DONE:
+               break;
+       default:
+               TEST_FAIL("Unknown ucall %lu", uc.cmd);
+               break;
+       }
+}
+
+static void test_create_vpmu_vm_with_pmcr_n(uint64_t pmcr_n, bool expect_fail)
+{
+       struct kvm_vcpu *vcpu;
+       uint64_t pmcr, pmcr_orig;
+
+       create_vpmu_vm(guest_code);
+       vcpu = vpmu_vm.vcpu;
+
+       vcpu_get_reg(vcpu, KVM_ARM64_SYS_REG(SYS_PMCR_EL0), &pmcr_orig);
+       pmcr = pmcr_orig;
+
+       /*
+        * 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);
+       vcpu_get_reg(vcpu, KVM_ARM64_SYS_REG(SYS_PMCR_EL0), &pmcr);
+
+       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\n",
+                           pmcr, pmcr_n);
+       else
+               TEST_ASSERT(pmcr_n == get_pmcr_n(pmcr),
+                           "Failed to update PMCR.N to %lu (received: %lu)\n",
+                           pmcr_n, get_pmcr_n(pmcr));
+}
+
+/*
+ * Create a guest with one vCPU, set the PMCR_EL0.N for the vCPU to @pmcr_n,
+ * and run the test.
+ */
+static void run_test(uint64_t pmcr_n)
+{
+       uint64_t sp;
+       struct kvm_vcpu *vcpu;
+       struct kvm_vcpu_init init;
+
+       pr_debug("Test with pmcr_n %lu\n", pmcr_n);
+
+       test_create_vpmu_vm_with_pmcr_n(pmcr_n, false);
+       vcpu = vpmu_vm.vcpu;
+
+       /* Save the initial sp to restore them later to run the guest again */
+       vcpu_get_reg(vcpu, ARM64_CORE_REG(sp_el1), &sp);
+
+       run_vcpu(vcpu, pmcr_n);
+
+       /*
+        * Reset and re-initialize the vCPU, and run the guest code again to
+        * check if PMCR_EL0.N is preserved.
+        */
+       vm_ioctl(vpmu_vm.vm, KVM_ARM_PREFERRED_TARGET, &init);
+       init.features[0] |= (1 << KVM_ARM_VCPU_PMU_V3);
+       aarch64_vcpu_setup(vcpu, &init);
+       vcpu_init_descriptor_tables(vcpu);
+       vcpu_set_reg(vcpu, ARM64_CORE_REG(sp_el1), sp);
+       vcpu_set_reg(vcpu, ARM64_CORE_REG(regs.pc), (uint64_t)guest_code);
+
+       run_vcpu(vcpu, pmcr_n);
+
+       destroy_vpmu_vm();
+}
+
+/*
+ * Create a guest with one vCPU, and attempt to set the PMCR_EL0.N for
+ * the vCPU to @pmcr_n, which is larger than the host value.
+ * The attempt should fail as @pmcr_n is too big to set for the vCPU.
+ */
+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);
+       destroy_vpmu_vm();
+}
+
+/*
+ * Return the default number of implemented PMU event counters excluding
+ * the cycle counter (i.e. PMCR_EL0.N value) for the guest.
+ */
+static uint64_t get_pmcr_n_limit(void)
+{
+       uint64_t pmcr;
+
+       create_vpmu_vm(guest_code);
+       vcpu_get_reg(vpmu_vm.vcpu, KVM_ARM64_SYS_REG(SYS_PMCR_EL0), &pmcr);
+       destroy_vpmu_vm();
+       return get_pmcr_n(pmcr);
+}
+
+int main(void)
+{
+       uint64_t i, pmcr_n;
+
+       TEST_REQUIRE(kvm_has_cap(KVM_CAP_ARM_PMU_V3));
+
+       pmcr_n = get_pmcr_n_limit();
+       for (i = 0; i <= pmcr_n; i++)
+               run_test(i);
+
+       for (i = pmcr_n + 1; i < ARMV8_PMU_MAX_COUNTERS; i++)
+               run_error_test(i);
+
+       return 0;
+}