kvm/x86_64/vmx_nested_tsc_scaling_test.c

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * vmx_nested_tsc_scaling_test
0004  *
0005  * Copyright 2021 Amazon.com, Inc. or its affiliates. All Rights Reserved.
0006  *
0007  * This test case verifies that nested TSC scaling behaves as expected when
0008  * both L1 and L2 are scaled using different ratios. For this test we scale
0009  * L1 down and scale L2 up.
0010  */
0011
0012 #include <time.h>
0013
0014 #include "kvm_util.h"
0015 #include "vmx.h"
0016 #include "kselftest.h"
0017
0018 /* L2 is scaled up (from L1's perspective) by this factor */
0019 #define L2_SCALE_FACTOR 4ULL
0020
0021 #define TSC_OFFSET_L2 ((uint64_t) -33125236320908)
0022 #define TSC_MULTIPLIER_L2 (L2_SCALE_FACTOR << 48)
0023
0024 #define L2_GUEST_STACK_SIZE 64
0025
0026 enum { USLEEP, UCHECK_L1, UCHECK_L2 };
0027 #define GUEST_SLEEP(sec)         ucall(UCALL_SYNC, 2, USLEEP, sec)
0028 #define GUEST_CHECK(level, freq) ucall(UCALL_SYNC, 2, level, freq)
0029
0030
0031 /*
0032  * This function checks whether the "actual" TSC frequency of a guest matches
0033  * its expected frequency. In order to account for delays in taking the TSC
0034  * measurements, a difference of 1% between the actual and the expected value
0035  * is tolerated.
0036  */
0037 static void compare_tsc_freq(uint64_t actual, uint64_t expected)
0038 {
0039     uint64_t tolerance, thresh_low, thresh_high;
0040
0041     tolerance = expected / 100;
0042     thresh_low = expected - tolerance;
0043     thresh_high = expected + tolerance;
0044
0045     TEST_ASSERT(thresh_low < actual,
0046         "TSC freq is expected to be between %"PRIu64" and %"PRIu64
0047         " but it actually is %"PRIu64,
0048         thresh_low, thresh_high, actual);
0049     TEST_ASSERT(thresh_high > actual,
0050         "TSC freq is expected to be between %"PRIu64" and %"PRIu64
0051         " but it actually is %"PRIu64,
0052         thresh_low, thresh_high, actual);
0053 }
0054
0055 static void check_tsc_freq(int level)
0056 {
0057     uint64_t tsc_start, tsc_end, tsc_freq;
0058
0059     /*
0060      * Reading the TSC twice with about a second's difference should give
0061      * us an approximation of the TSC frequency from the guest's
0062      * perspective. Now, this won't be completely accurate, but it should
0063      * be good enough for the purposes of this test.
0064      */
0065     tsc_start = rdmsr(MSR_IA32_TSC);
0066     GUEST_SLEEP(1);
0067     tsc_end = rdmsr(MSR_IA32_TSC);
0068
0069     tsc_freq = tsc_end - tsc_start;
0070
0071     GUEST_CHECK(level, tsc_freq);
0072 }
0073
0074 static void l2_guest_code(void)
0075 {
0076     check_tsc_freq(UCHECK_L2);
0077
0078     /* exit to L1 */
0079     __asm__ __volatile__("vmcall");
0080 }
0081
0082 static void l1_guest_code(struct vmx_pages *vmx_pages)
0083 {
0084     unsigned long l2_guest_stack[L2_GUEST_STACK_SIZE];
0085     uint32_t control;
0086
0087     /* check that L1's frequency looks alright before launching L2 */
0088     check_tsc_freq(UCHECK_L1);
0089
0090     GUEST_ASSERT(prepare_for_vmx_operation(vmx_pages));
0091     GUEST_ASSERT(load_vmcs(vmx_pages));
0092
0093     /* prepare the VMCS for L2 execution */
0094     prepare_vmcs(vmx_pages, l2_guest_code, &l2_guest_stack[L2_GUEST_STACK_SIZE]);
0095
0096     /* enable TSC offsetting and TSC scaling for L2 */
0097     control = vmreadz(CPU_BASED_VM_EXEC_CONTROL);
0098     control |= CPU_BASED_USE_MSR_BITMAPS | CPU_BASED_USE_TSC_OFFSETTING;
0099     vmwrite(CPU_BASED_VM_EXEC_CONTROL, control);
0100
0101     control = vmreadz(SECONDARY_VM_EXEC_CONTROL);
0102     control |= SECONDARY_EXEC_TSC_SCALING;
0103     vmwrite(SECONDARY_VM_EXEC_CONTROL, control);
0104
0105     vmwrite(TSC_OFFSET, TSC_OFFSET_L2);
0106     vmwrite(TSC_MULTIPLIER, TSC_MULTIPLIER_L2);
0107     vmwrite(TSC_MULTIPLIER_HIGH, TSC_MULTIPLIER_L2 >> 32);
0108
0109     /* launch L2 */
0110     GUEST_ASSERT(!vmlaunch());
0111     GUEST_ASSERT(vmreadz(VM_EXIT_REASON) == EXIT_REASON_VMCALL);
0112
0113     /* check that L1's frequency still looks good */
0114     check_tsc_freq(UCHECK_L1);
0115
0116     GUEST_DONE();
0117 }
0118
0119 static void stable_tsc_check_supported(void)
0120 {
0121     FILE *fp;
0122     char buf[4];
0123
0124     fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
0125     if (fp == NULL)
0126         goto skip_test;
0127
0128     if (fgets(buf, sizeof(buf), fp) == NULL)
0129         goto skip_test;
0130
0131     if (strncmp(buf, "tsc", sizeof(buf)))
0132         goto skip_test;
0133
0134     return;
0135 skip_test:
0136     print_skip("Kernel does not use TSC clocksource - assuming that host TSC is not stable");
0137     exit(KSFT_SKIP);
0138 }
0139
0140 int main(int argc, char *argv[])
0141 {
0142     struct kvm_vcpu *vcpu;
0143     struct kvm_vm *vm;
0144     vm_vaddr_t vmx_pages_gva;
0145
0146     uint64_t tsc_start, tsc_end;
0147     uint64_t tsc_khz;
0148     uint64_t l1_scale_factor;
0149     uint64_t l0_tsc_freq = 0;
0150     uint64_t l1_tsc_freq = 0;
0151     uint64_t l2_tsc_freq = 0;
0152
0153     TEST_REQUIRE(kvm_cpu_has(X86_FEATURE_VMX));
0154     TEST_REQUIRE(kvm_has_cap(KVM_CAP_TSC_CONTROL));
0155     stable_tsc_check_supported();
0156
0157     /*
0158      * We set L1's scale factor to be a random number from 2 to 10.
0159      * Ideally we would do the same for L2's factor but that one is
0160      * referenced by both main() and l1_guest_code() and using a global
0161      * variable does not work.
0162      */
0163     srand(time(NULL));
0164     l1_scale_factor = (rand() % 9) + 2;
0165     printf("L1's scale down factor is: %"PRIu64"\n", l1_scale_factor);
0166     printf("L2's scale up factor is: %llu\n", L2_SCALE_FACTOR);
0167
0168     tsc_start = rdtsc();
0169     sleep(1);
0170     tsc_end = rdtsc();
0171
0172     l0_tsc_freq = tsc_end - tsc_start;
0173     printf("real TSC frequency is around: %"PRIu64"\n", l0_tsc_freq);
0174
0175     vm = vm_create_with_one_vcpu(&vcpu, l1_guest_code);
0176     vcpu_alloc_vmx(vm, &vmx_pages_gva);
0177     vcpu_args_set(vcpu, 1, vmx_pages_gva);
0178
0179     tsc_khz = __vcpu_ioctl(vcpu, KVM_GET_TSC_KHZ, NULL);
0180     TEST_ASSERT(tsc_khz != -1, "vcpu ioctl KVM_GET_TSC_KHZ failed");
0181
0182     /* scale down L1's TSC frequency */
0183     vcpu_ioctl(vcpu, KVM_SET_TSC_KHZ, (void *) (tsc_khz / l1_scale_factor));
0184
0185     for (;;) {
0186         volatile struct kvm_run *run = vcpu->run;
0187         struct ucall uc;
0188
0189         vcpu_run(vcpu);
0190         TEST_ASSERT(run->exit_reason == KVM_EXIT_IO,
0191                 "Got exit_reason other than KVM_EXIT_IO: %u (%s)\n",
0192                 run->exit_reason,
0193                 exit_reason_str(run->exit_reason));
0194
0195         switch (get_ucall(vcpu, &uc)) {
0196         case UCALL_ABORT:
0197             REPORT_GUEST_ASSERT(uc);
0198         case UCALL_SYNC:
0199             switch (uc.args[0]) {
0200             case USLEEP:
0201                 sleep(uc.args[1]);
0202                 break;
0203             case UCHECK_L1:
0204                 l1_tsc_freq = uc.args[1];
0205                 printf("L1's TSC frequency is around: %"PRIu64
0206                        "\n", l1_tsc_freq);
0207
0208                 compare_tsc_freq(l1_tsc_freq,
0209                          l0_tsc_freq / l1_scale_factor);
0210                 break;
0211             case UCHECK_L2:
0212                 l2_tsc_freq = uc.args[1];
0213                 printf("L2's TSC frequency is around: %"PRIu64
0214                        "\n", l2_tsc_freq);
0215
0216                 compare_tsc_freq(l2_tsc_freq,
0217                          l1_tsc_freq * L2_SCALE_FACTOR);
0218                 break;
0219             }
0220             break;
0221         case UCALL_DONE:
0222             goto done;
0223         default:
0224             TEST_FAIL("Unknown ucall %lu", uc.cmd);
0225         }
0226     }
0227
0228 done:
0229     kvm_vm_free(vm);
0230     return 0;
0231 }