]> git.baikalelectronics.ru Git - kernel.git/commitdiff
KVM: selftests: Stop conflating vCPU index and ID in perf tests
authorSean Christopherson <seanjc@google.com>
Wed, 16 Feb 2022 21:38:12 +0000 (13:38 -0800)
committerPaolo Bonzini <pbonzini@redhat.com>
Sat, 11 Jun 2022 15:47:20 +0000 (11:47 -0400)
Track vCPUs by their 'struct kvm_vcpu' object, and stop assuming that a
vCPU's ID is the same as its index when referencing a vCPU's metadata.

Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
tools/testing/selftests/kvm/access_tracking_perf_test.c
tools/testing/selftests/kvm/demand_paging_test.c
tools/testing/selftests/kvm/dirty_log_perf_test.c
tools/testing/selftests/kvm/include/perf_test_util.h
tools/testing/selftests/kvm/lib/perf_test_util.c
tools/testing/selftests/kvm/lib/x86_64/perf_test_util.c
tools/testing/selftests/kvm/memslot_modification_stress_test.c

index d8909032317a8fa1f25e4876d6571d328d4f4ea6..86a90222f913896166ad26c126a2846e484c7649 100644 (file)
@@ -74,7 +74,7 @@ struct test_params {
        uint64_t vcpu_memory_bytes;
 
        /* The number of vCPUs to create in the VM. */
-       int vcpus;
+       int nr_vcpus;
 };
 
 static uint64_t pread_uint64(int fd, const char *filename, uint64_t index)
@@ -127,10 +127,12 @@ static void mark_page_idle(int page_idle_fd, uint64_t pfn)
                    "Set page_idle bits for PFN 0x%" PRIx64, pfn);
 }
 
-static void mark_vcpu_memory_idle(struct kvm_vm *vm, int vcpu_id)
+static void mark_vcpu_memory_idle(struct kvm_vm *vm,
+                                 struct perf_test_vcpu_args *vcpu_args)
 {
-       uint64_t base_gva = perf_test_args.vcpu_args[vcpu_id].gva;
-       uint64_t pages = perf_test_args.vcpu_args[vcpu_id].pages;
+       int vcpu_idx = vcpu_args->vcpu_idx;
+       uint64_t base_gva = vcpu_args->gva;
+       uint64_t pages = vcpu_args->pages;
        uint64_t page;
        uint64_t still_idle = 0;
        uint64_t no_pfn = 0;
@@ -138,7 +140,7 @@ static void mark_vcpu_memory_idle(struct kvm_vm *vm, int vcpu_id)
        int pagemap_fd;
 
        /* If vCPUs are using an overlapping region, let vCPU 0 mark it idle. */
-       if (overlap_memory_access && vcpu_id)
+       if (overlap_memory_access && vcpu_idx)
                return;
 
        page_idle_fd = open("/sys/kernel/mm/page_idle/bitmap", O_RDWR);
@@ -170,7 +172,7 @@ static void mark_vcpu_memory_idle(struct kvm_vm *vm, int vcpu_id)
         */
        TEST_ASSERT(no_pfn < pages / 100,
                    "vCPU %d: No PFN for %" PRIu64 " out of %" PRIu64 " pages.",
-                   vcpu_id, no_pfn, pages);
+                   vcpu_idx, no_pfn, pages);
 
        /*
         * Test that at least 90% of memory has been marked idle (the rest might
@@ -183,17 +185,16 @@ static void mark_vcpu_memory_idle(struct kvm_vm *vm, int vcpu_id)
        TEST_ASSERT(still_idle < pages / 10,
                    "vCPU%d: Too many pages still idle (%"PRIu64 " out of %"
                    PRIu64 ").\n",
-                   vcpu_id, still_idle, pages);
+                   vcpu_idx, still_idle, pages);
 
        close(page_idle_fd);
        close(pagemap_fd);
 }
 
-static void assert_ucall(struct kvm_vm *vm, uint32_t vcpu_id,
-                        uint64_t expected_ucall)
+static void assert_ucall(struct kvm_vcpu *vcpu, uint64_t expected_ucall)
 {
        struct ucall uc;
-       uint64_t actual_ucall = get_ucall(vm, vcpu_id, &uc);
+       uint64_t actual_ucall = get_ucall(vcpu->vm, vcpu->id, &uc);
 
        TEST_ASSERT(expected_ucall == actual_ucall,
                    "Guest exited unexpectedly (expected ucall %" PRIu64
@@ -217,28 +218,29 @@ static bool spin_wait_for_next_iteration(int *current_iteration)
 
 static void vcpu_thread_main(struct perf_test_vcpu_args *vcpu_args)
 {
+       struct kvm_vcpu *vcpu = vcpu_args->vcpu;
        struct kvm_vm *vm = perf_test_args.vm;
-       int vcpu_id = vcpu_args->vcpu_id;
+       int vcpu_idx = vcpu_args->vcpu_idx;
        int current_iteration = 0;
 
        while (spin_wait_for_next_iteration(&current_iteration)) {
                switch (READ_ONCE(iteration_work)) {
                case ITERATION_ACCESS_MEMORY:
-                       vcpu_run(vm, vcpu_id);
-                       assert_ucall(vm, vcpu_id, UCALL_SYNC);
+                       vcpu_run(vm, vcpu->id);
+                       assert_ucall(vcpu, UCALL_SYNC);
                        break;
                case ITERATION_MARK_IDLE:
-                       mark_vcpu_memory_idle(vm, vcpu_id);
+                       mark_vcpu_memory_idle(vm, vcpu_args);
                        break;
                };
 
-               vcpu_last_completed_iteration[vcpu_id] = current_iteration;
+               vcpu_last_completed_iteration[vcpu_idx] = current_iteration;
        }
 }
 
-static void spin_wait_for_vcpu(int vcpu_id, int target_iteration)
+static void spin_wait_for_vcpu(int vcpu_idx, int target_iteration)
 {
-       while (READ_ONCE(vcpu_last_completed_iteration[vcpu_id]) !=
+       while (READ_ONCE(vcpu_last_completed_iteration[vcpu_idx]) !=
               target_iteration) {
                continue;
        }
@@ -250,12 +252,11 @@ enum access_type {
        ACCESS_WRITE,
 };
 
-static void run_iteration(struct kvm_vm *vm, int vcpus, const char *description)
+static void run_iteration(struct kvm_vm *vm, int nr_vcpus, const char *description)
 {
        struct timespec ts_start;
        struct timespec ts_elapsed;
-       int next_iteration;
-       int vcpu_id;
+       int next_iteration, i;
 
        /* Kick off the vCPUs by incrementing iteration. */
        next_iteration = ++iteration;
@@ -263,23 +264,23 @@ static void run_iteration(struct kvm_vm *vm, int vcpus, const char *description)
        clock_gettime(CLOCK_MONOTONIC, &ts_start);
 
        /* Wait for all vCPUs to finish the iteration. */
-       for (vcpu_id = 0; vcpu_id < vcpus; vcpu_id++)
-               spin_wait_for_vcpu(vcpu_id, next_iteration);
+       for (i = 0; i < nr_vcpus; i++)
+               spin_wait_for_vcpu(i, next_iteration);
 
        ts_elapsed = timespec_elapsed(ts_start);
        pr_info("%-30s: %ld.%09lds\n",
                description, ts_elapsed.tv_sec, ts_elapsed.tv_nsec);
 }
 
-static void access_memory(struct kvm_vm *vm, int vcpus, enum access_type access,
-                         const char *description)
+static void access_memory(struct kvm_vm *vm, int nr_vcpus,
+                         enum access_type access, const char *description)
 {
        perf_test_set_wr_fract(vm, (access == ACCESS_READ) ? INT_MAX : 1);
        iteration_work = ITERATION_ACCESS_MEMORY;
-       run_iteration(vm, vcpus, description);
+       run_iteration(vm, nr_vcpus, description);
 }
 
-static void mark_memory_idle(struct kvm_vm *vm, int vcpus)
+static void mark_memory_idle(struct kvm_vm *vm, int nr_vcpus)
 {
        /*
         * Even though this parallelizes the work across vCPUs, this is still a
@@ -289,37 +290,37 @@ static void mark_memory_idle(struct kvm_vm *vm, int vcpus)
         */
        pr_debug("Marking VM memory idle (slow)...\n");
        iteration_work = ITERATION_MARK_IDLE;
-       run_iteration(vm, vcpus, "Mark memory idle");
+       run_iteration(vm, nr_vcpus, "Mark memory idle");
 }
 
 static void run_test(enum vm_guest_mode mode, void *arg)
 {
        struct test_params *params = arg;
        struct kvm_vm *vm;
-       int vcpus = params->vcpus;
+       int nr_vcpus = params->nr_vcpus;
 
-       vm = perf_test_create_vm(mode, vcpus, params->vcpu_memory_bytes, 1,
+       vm = perf_test_create_vm(mode, nr_vcpus, params->vcpu_memory_bytes, 1,
                                 params->backing_src, !overlap_memory_access);
 
-       perf_test_start_vcpu_threads(vcpus, vcpu_thread_main);
+       perf_test_start_vcpu_threads(nr_vcpus, vcpu_thread_main);
 
        pr_info("\n");
-       access_memory(vm, vcpus, ACCESS_WRITE, "Populating memory");
+       access_memory(vm, nr_vcpus, ACCESS_WRITE, "Populating memory");
 
        /* As a control, read and write to the populated memory first. */
-       access_memory(vm, vcpus, ACCESS_WRITE, "Writing to populated memory");
-       access_memory(vm, vcpus, ACCESS_READ, "Reading from populated memory");
+       access_memory(vm, nr_vcpus, ACCESS_WRITE, "Writing to populated memory");
+       access_memory(vm, nr_vcpus, ACCESS_READ, "Reading from populated memory");
 
        /* Repeat on memory that has been marked as idle. */
-       mark_memory_idle(vm, vcpus);
-       access_memory(vm, vcpus, ACCESS_WRITE, "Writing to idle memory");
-       mark_memory_idle(vm, vcpus);
-       access_memory(vm, vcpus, ACCESS_READ, "Reading from idle memory");
+       mark_memory_idle(vm, nr_vcpus);
+       access_memory(vm, nr_vcpus, ACCESS_WRITE, "Writing to idle memory");
+       mark_memory_idle(vm, nr_vcpus);
+       access_memory(vm, nr_vcpus, ACCESS_READ, "Reading from idle memory");
 
        /* Set done to signal the vCPU threads to exit */
        done = true;
 
-       perf_test_join_vcpu_threads(vcpus);
+       perf_test_join_vcpu_threads(nr_vcpus);
        perf_test_destroy_vm(vm);
 }
 
@@ -347,7 +348,7 @@ int main(int argc, char *argv[])
        struct test_params params = {
                .backing_src = DEFAULT_VM_MEM_SRC,
                .vcpu_memory_bytes = DEFAULT_PER_VCPU_MEM_SIZE,
-               .vcpus = 1,
+               .nr_vcpus = 1,
        };
        int page_idle_fd;
        int opt;
@@ -363,7 +364,7 @@ int main(int argc, char *argv[])
                        params.vcpu_memory_bytes = parse_size(optarg);
                        break;
                case 'v':
-                       params.vcpus = atoi(optarg);
+                       params.nr_vcpus = atoi(optarg);
                        break;
                case 'o':
                        overlap_memory_access = true;
index d8db0a37e973f42b95a7909b1fbf257d4ca13765..c461107210887e3b6440ac3c046385995e5421cf 100644 (file)
@@ -44,28 +44,27 @@ static char *guest_data_prototype;
 
 static void vcpu_worker(struct perf_test_vcpu_args *vcpu_args)
 {
-       int ret;
-       int vcpu_id = vcpu_args->vcpu_id;
+       struct kvm_vcpu *vcpu = vcpu_args->vcpu;
        struct kvm_vm *vm = perf_test_args.vm;
-       struct kvm_run *run;
+       int vcpu_idx = vcpu_args->vcpu_idx;
+       struct kvm_run *run = vcpu->run;
        struct timespec start;
        struct timespec ts_diff;
-
-       run = vcpu_state(vm, vcpu_id);
+       int ret;
 
        clock_gettime(CLOCK_MONOTONIC, &start);
 
        /* Let the guest access its memory */
-       ret = _vcpu_run(vm, vcpu_id);
+       ret = _vcpu_run(vm, vcpu->id);
        TEST_ASSERT(ret == 0, "vcpu_run failed: %d\n", ret);
-       if (get_ucall(vm, vcpu_id, NULL) != UCALL_SYNC) {
+       if (get_ucall(vm, vcpu->id, NULL) != UCALL_SYNC) {
                TEST_ASSERT(false,
                            "Invalid guest sync status: exit_reason=%s\n",
                            exit_reason_str(run->exit_reason));
        }
 
        ts_diff = timespec_elapsed(start);
-       PER_VCPU_DEBUG("vCPU %d execution time: %ld.%.9lds\n", vcpu_id,
+       PER_VCPU_DEBUG("vCPU %d execution time: %ld.%.9lds\n", vcpu_idx,
                       ts_diff.tv_sec, ts_diff.tv_nsec);
 }
 
@@ -285,8 +284,7 @@ static void run_test(enum vm_guest_mode mode, void *arg)
        struct timespec ts_diff;
        int *pipefds = NULL;
        struct kvm_vm *vm;
-       int vcpu_id;
-       int r;
+       int r, i;
 
        vm = perf_test_create_vm(mode, nr_vcpus, guest_percpu_mem_size, 1,
                                 p->src_type, p->partition_vcpu_memory_access);
@@ -309,12 +307,12 @@ static void run_test(enum vm_guest_mode mode, void *arg)
                pipefds = malloc(sizeof(int) * nr_vcpus * 2);
                TEST_ASSERT(pipefds, "Unable to allocate memory for pipefd");
 
-               for (vcpu_id = 0; vcpu_id < nr_vcpus; vcpu_id++) {
+               for (i = 0; i < nr_vcpus; i++) {
                        struct perf_test_vcpu_args *vcpu_args;
                        void *vcpu_hva;
                        void *vcpu_alias;
 
-                       vcpu_args = &perf_test_args.vcpu_args[vcpu_id];
+                       vcpu_args = &perf_test_args.vcpu_args[i];
 
                        /* Cache the host addresses of the region */
                        vcpu_hva = addr_gpa2hva(vm, vcpu_args->gpa);
@@ -324,13 +322,13 @@ static void run_test(enum vm_guest_mode mode, void *arg)
                         * Set up user fault fd to handle demand paging
                         * requests.
                         */
-                       r = pipe2(&pipefds[vcpu_id * 2],
+                       r = pipe2(&pipefds[i * 2],
                                  O_CLOEXEC | O_NONBLOCK);
                        TEST_ASSERT(!r, "Failed to set up pipefd");
 
-                       setup_demand_paging(vm, &uffd_handler_threads[vcpu_id],
-                                           pipefds[vcpu_id * 2], p->uffd_mode,
-                                           p->uffd_delay, &uffd_args[vcpu_id],
+                       setup_demand_paging(vm, &uffd_handler_threads[i],
+                                           pipefds[i * 2], p->uffd_mode,
+                                           p->uffd_delay, &uffd_args[i],
                                            vcpu_hva, vcpu_alias,
                                            vcpu_args->pages * perf_test_args.guest_page_size);
                }
@@ -350,11 +348,11 @@ static void run_test(enum vm_guest_mode mode, void *arg)
                char c;
 
                /* Tell the user fault fd handler threads to quit */
-               for (vcpu_id = 0; vcpu_id < nr_vcpus; vcpu_id++) {
-                       r = write(pipefds[vcpu_id * 2 + 1], &c, 1);
+               for (i = 0; i < nr_vcpus; i++) {
+                       r = write(pipefds[i * 2 + 1], &c, 1);
                        TEST_ASSERT(r == 1, "Unable to write to pipefd");
 
-                       pthread_join(uffd_handler_threads[vcpu_id], NULL);
+                       pthread_join(uffd_handler_threads[i], NULL);
                }
        }
 
index 8aaa0949707a3518e4a9a984c1a0c5b10d54e1b9..24f6d836499bd32fd8627f9579352ae9308e6a9f 100644 (file)
@@ -68,44 +68,45 @@ static int vcpu_last_completed_iteration[KVM_MAX_VCPUS];
 
 static void vcpu_worker(struct perf_test_vcpu_args *vcpu_args)
 {
-       int ret;
+       struct kvm_vcpu *vcpu = vcpu_args->vcpu;
        struct kvm_vm *vm = perf_test_args.vm;
+       int vcpu_idx = vcpu_args->vcpu_idx;
        uint64_t pages_count = 0;
        struct kvm_run *run;
        struct timespec start;
        struct timespec ts_diff;
        struct timespec total = (struct timespec){0};
        struct timespec avg;
-       int vcpu_id = vcpu_args->vcpu_id;
+       int ret;
 
-       run = vcpu_state(vm, vcpu_id);
+       run = vcpu->run;
 
        while (!READ_ONCE(host_quit)) {
                int current_iteration = READ_ONCE(iteration);
 
                clock_gettime(CLOCK_MONOTONIC, &start);
-               ret = _vcpu_run(vm, vcpu_id);
+               ret = _vcpu_run(vm, vcpu->id);
                ts_diff = timespec_elapsed(start);
 
                TEST_ASSERT(ret == 0, "vcpu_run failed: %d\n", ret);
-               TEST_ASSERT(get_ucall(vm, vcpu_id, NULL) == UCALL_SYNC,
+               TEST_ASSERT(get_ucall(vm, vcpu->id, NULL) == UCALL_SYNC,
                            "Invalid guest sync status: exit_reason=%s\n",
                            exit_reason_str(run->exit_reason));
 
-               pr_debug("Got sync event from vCPU %d\n", vcpu_id);
-               vcpu_last_completed_iteration[vcpu_id] = current_iteration;
+               pr_debug("Got sync event from vCPU %d\n", vcpu_idx);
+               vcpu_last_completed_iteration[vcpu_idx] = current_iteration;
                pr_debug("vCPU %d updated last completed iteration to %d\n",
-                        vcpu_id, vcpu_last_completed_iteration[vcpu_id]);
+                        vcpu->id, vcpu_last_completed_iteration[vcpu_idx]);
 
                if (current_iteration) {
                        pages_count += vcpu_args->pages;
                        total = timespec_add(total, ts_diff);
                        pr_debug("vCPU %d iteration %d dirty memory time: %ld.%.9lds\n",
-                               vcpu_id, current_iteration, ts_diff.tv_sec,
+                               vcpu_idx, current_iteration, ts_diff.tv_sec,
                                ts_diff.tv_nsec);
                } else {
                        pr_debug("vCPU %d iteration %d populate memory time: %ld.%.9lds\n",
-                               vcpu_id, current_iteration, ts_diff.tv_sec,
+                               vcpu_idx, current_iteration, ts_diff.tv_sec,
                                ts_diff.tv_nsec);
                }
 
@@ -113,9 +114,9 @@ static void vcpu_worker(struct perf_test_vcpu_args *vcpu_args)
                       !READ_ONCE(host_quit)) {}
        }
 
-       avg = timespec_div(total, vcpu_last_completed_iteration[vcpu_id]);
+       avg = timespec_div(total, vcpu_last_completed_iteration[vcpu_idx]);
        pr_debug("\nvCPU %d dirtied 0x%lx pages over %d iterations in %ld.%.9lds. (Avg %ld.%.9lds/iteration)\n",
-               vcpu_id, pages_count, vcpu_last_completed_iteration[vcpu_id],
+               vcpu_idx, pages_count, vcpu_last_completed_iteration[vcpu_idx],
                total.tv_sec, total.tv_nsec, avg.tv_sec, avg.tv_nsec);
 }
 
@@ -207,13 +208,13 @@ static void run_test(enum vm_guest_mode mode, void *arg)
        uint64_t guest_num_pages;
        uint64_t host_num_pages;
        uint64_t pages_per_slot;
-       int vcpu_id;
        struct timespec start;
        struct timespec ts_diff;
        struct timespec get_dirty_log_total = (struct timespec){0};
        struct timespec vcpu_dirty_total = (struct timespec){0};
        struct timespec avg;
        struct timespec clear_dirty_log_total = (struct timespec){0};
+       int i;
 
        vm = perf_test_create_vm(mode, nr_vcpus, guest_percpu_mem_size,
                                 p->slots, p->backing_src,
@@ -239,15 +240,15 @@ static void run_test(enum vm_guest_mode mode, void *arg)
        host_quit = false;
 
        clock_gettime(CLOCK_MONOTONIC, &start);
-       for (vcpu_id = 0; vcpu_id < nr_vcpus; vcpu_id++)
-               vcpu_last_completed_iteration[vcpu_id] = -1;
+       for (i = 0; i < nr_vcpus; i++)
+               vcpu_last_completed_iteration[i] = -1;
 
        perf_test_start_vcpu_threads(nr_vcpus, vcpu_worker);
 
        /* Allow the vCPUs to populate memory */
        pr_debug("Starting iteration %d - Populating\n", iteration);
-       for (vcpu_id = 0; vcpu_id < nr_vcpus; vcpu_id++) {
-               while (READ_ONCE(vcpu_last_completed_iteration[vcpu_id]) !=
+       for (i = 0; i < nr_vcpus; i++) {
+               while (READ_ONCE(vcpu_last_completed_iteration[i]) !=
                       iteration)
                        ;
        }
@@ -272,8 +273,8 @@ static void run_test(enum vm_guest_mode mode, void *arg)
                iteration++;
 
                pr_debug("Starting iteration %d\n", iteration);
-               for (vcpu_id = 0; vcpu_id < nr_vcpus; vcpu_id++) {
-                       while (READ_ONCE(vcpu_last_completed_iteration[vcpu_id])
+               for (i = 0; i < nr_vcpus; i++) {
+                       while (READ_ONCE(vcpu_last_completed_iteration[i])
                               != iteration)
                                ;
                }
index d822cb670f1cd46e9078ea1613b3abc38106c29c..eaa88df0555a88e7374cef4bfae8d0b720e2e49d 100644 (file)
@@ -25,7 +25,8 @@ struct perf_test_vcpu_args {
        uint64_t pages;
 
        /* Only used by the host userspace part of the vCPU thread */
-       int vcpu_id;
+       struct kvm_vcpu *vcpu;
+       int vcpu_idx;
 };
 
 struct perf_test_args {
@@ -44,7 +45,7 @@ struct perf_test_args {
 
 extern struct perf_test_args perf_test_args;
 
-struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int vcpus,
+struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int nr_vcpus,
                                   uint64_t vcpu_memory_bytes, int slots,
                                   enum vm_mem_backing_src_type backing_src,
                                   bool partition_vcpu_memory_access);
@@ -57,6 +58,6 @@ void perf_test_join_vcpu_threads(int vcpus);
 void perf_test_guest_code(uint32_t vcpu_id);
 
 uint64_t perf_test_nested_pages(int nr_vcpus);
-void perf_test_setup_nested(struct kvm_vm *vm, int nr_vcpus);
+void perf_test_setup_nested(struct kvm_vm *vm, int nr_vcpus, struct kvm_vcpu *vcpus[]);
 
 #endif /* SELFTEST_KVM_PERF_TEST_UTIL_H */
index 05f65c9292eb6566f0e01c0d1b4971786987ec6a..91a89553afdd43b5bc251c7831095481ddc3cbdb 100644 (file)
@@ -17,8 +17,8 @@ struct perf_test_args perf_test_args;
 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
 
 struct vcpu_thread {
-       /* The id of the vCPU. */
-       int vcpu_id;
+       /* The index of the vCPU. */
+       int vcpu_idx;
 
        /* The pthread backing the vCPU. */
        pthread_t thread;
@@ -36,24 +36,26 @@ static void (*vcpu_thread_fn)(struct perf_test_vcpu_args *);
 /* Set to true once all vCPU threads are up and running. */
 static bool all_vcpu_threads_running;
 
+static struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
+
 /*
  * Continuously write to the first 8 bytes of each page in the
  * specified region.
  */
-void perf_test_guest_code(uint32_t vcpu_id)
+void perf_test_guest_code(uint32_t vcpu_idx)
 {
        struct perf_test_args *pta = &perf_test_args;
-       struct perf_test_vcpu_args *vcpu_args = &pta->vcpu_args[vcpu_id];
+       struct perf_test_vcpu_args *vcpu_args = &pta->vcpu_args[vcpu_idx];
        uint64_t gva;
        uint64_t pages;
        int i;
 
-       /* Make sure vCPU args data structure is not corrupt. */
-       GUEST_ASSERT(vcpu_args->vcpu_id == vcpu_id);
-
        gva = vcpu_args->gva;
        pages = vcpu_args->pages;
 
+       /* Make sure vCPU args data structure is not corrupt. */
+       GUEST_ASSERT(vcpu_args->vcpu_idx == vcpu_idx);
+
        while (true) {
                for (i = 0; i < pages; i++) {
                        uint64_t addr = gva + (i * pta->guest_page_size);
@@ -68,40 +70,43 @@ void perf_test_guest_code(uint32_t vcpu_id)
        }
 }
 
-void perf_test_setup_vcpus(struct kvm_vm *vm, int vcpus,
+void perf_test_setup_vcpus(struct kvm_vm *vm, int nr_vcpus,
+                          struct kvm_vcpu *vcpus[],
                           uint64_t vcpu_memory_bytes,
                           bool partition_vcpu_memory_access)
 {
        struct perf_test_args *pta = &perf_test_args;
        struct perf_test_vcpu_args *vcpu_args;
-       int vcpu_id;
+       int i;
+
+       for (i = 0; i < nr_vcpus; i++) {
+               vcpu_args = &pta->vcpu_args[i];
 
-       for (vcpu_id = 0; vcpu_id < vcpus; vcpu_id++) {
-               vcpu_args = &pta->vcpu_args[vcpu_id];
+               vcpu_args->vcpu = vcpus[i];
+               vcpu_args->vcpu_idx = i;
 
-               vcpu_args->vcpu_id = vcpu_id;
                if (partition_vcpu_memory_access) {
                        vcpu_args->gva = guest_test_virt_mem +
-                                        (vcpu_id * vcpu_memory_bytes);
+                                        (i * vcpu_memory_bytes);
                        vcpu_args->pages = vcpu_memory_bytes /
                                           pta->guest_page_size;
-                       vcpu_args->gpa = pta->gpa + (vcpu_id * vcpu_memory_bytes);
+                       vcpu_args->gpa = pta->gpa + (i * vcpu_memory_bytes);
                } else {
                        vcpu_args->gva = guest_test_virt_mem;
-                       vcpu_args->pages = (vcpus * vcpu_memory_bytes) /
+                       vcpu_args->pages = (nr_vcpus * vcpu_memory_bytes) /
                                           pta->guest_page_size;
                        vcpu_args->gpa = pta->gpa;
                }
 
-               vcpu_args_set(vm, vcpu_id, 1, vcpu_id);
+               vcpu_args_set(vm, vcpus[i]->id, 1, i);
 
                pr_debug("Added VCPU %d with test mem gpa [%lx, %lx)\n",
-                        vcpu_id, vcpu_args->gpa, vcpu_args->gpa +
+                        i, vcpu_args->gpa, vcpu_args->gpa +
                         (vcpu_args->pages * pta->guest_page_size));
        }
 }
 
-struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int vcpus,
+struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int nr_vcpus,
                                   uint64_t vcpu_memory_bytes, int slots,
                                   enum vm_mem_backing_src_type backing_src,
                                   bool partition_vcpu_memory_access)
@@ -125,7 +130,7 @@ struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int vcpus,
        pta->guest_page_size = vm_guest_mode_params[mode].page_size;
 
        guest_num_pages = vm_adjust_num_guest_pages(mode,
-                               (vcpus * vcpu_memory_bytes) / pta->guest_page_size);
+                               (nr_vcpus * vcpu_memory_bytes) / pta->guest_page_size);
 
        TEST_ASSERT(vcpu_memory_bytes % getpagesize() == 0,
                    "Guest memory size is not host page size aligned.");
@@ -140,16 +145,16 @@ struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int vcpus,
         * in-memory data structures.
         */
        if (pta->nested)
-               slot0_pages += perf_test_nested_pages(vcpus);
+               slot0_pages += perf_test_nested_pages(nr_vcpus);
 
        /*
         * Pass guest_num_pages to populate the page tables for test memory.
         * The memory is also added to memslot 0, but that's a benign side
         * effect as KVM allows aliasing HVAs in meslots.
         */
-       vm = __vm_create_with_vcpus(mode, vcpus, DEFAULT_GUEST_PHY_PAGES,
+       vm = __vm_create_with_vcpus(mode, nr_vcpus, DEFAULT_GUEST_PHY_PAGES,
                                    slot0_pages + guest_num_pages, 0,
-                                   perf_test_guest_code, NULL);
+                                   perf_test_guest_code, vcpus);
 
        pta->vm = vm;
 
@@ -171,11 +176,10 @@ struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int vcpus,
        TEST_ASSERT(guest_num_pages < region_end_gfn,
                    "Requested more guest memory than address space allows.\n"
                    "    guest pages: %" PRIx64 " max gfn: %" PRIx64
-                   " vcpus: %d wss: %" PRIx64 "]\n",
-                   guest_num_pages, region_end_gfn - 1, vcpus,
-                   vcpu_memory_bytes);
+                   " nr_vcpus: %d wss: %" PRIx64 "]\n",
+                   guest_num_pages, region_end_gfn - 1, nr_vcpus, vcpu_memory_bytes);
 
-       pta->gpa = (region_end_gfn - guest_num_pages) * pta->guest_page_size;
+       pta->gpa = (region_end_gfn - guest_num_pages - 1) * pta->guest_page_size;
        pta->gpa = align_down(pta->gpa, backing_src_pagesz);
 #ifdef __s390x__
        /* Align to 1M (segment size) */
@@ -198,11 +202,12 @@ struct kvm_vm *perf_test_create_vm(enum vm_guest_mode mode, int vcpus,
        /* Do mapping for the demand paging memory slot */
        virt_map(vm, guest_test_virt_mem, pta->gpa, guest_num_pages);
 
-       perf_test_setup_vcpus(vm, vcpus, vcpu_memory_bytes, partition_vcpu_memory_access);
+       perf_test_setup_vcpus(vm, nr_vcpus, vcpus, vcpu_memory_bytes,
+                             partition_vcpu_memory_access);
 
        if (pta->nested) {
                pr_info("Configuring vCPUs to run in L2 (nested).\n");
-               perf_test_setup_nested(vm, vcpus);
+               perf_test_setup_nested(vm, nr_vcpus, vcpus);
        }
 
        ucall_init(vm, NULL);
@@ -230,7 +235,7 @@ uint64_t __weak perf_test_nested_pages(int nr_vcpus)
        return 0;
 }
 
-void __weak perf_test_setup_nested(struct kvm_vm *vm, int nr_vcpus)
+void __weak perf_test_setup_nested(struct kvm_vm *vm, int nr_vcpus, struct kvm_vcpu **vcpus)
 {
        pr_info("%s() not support on this architecture, skipping.\n", __func__);
        exit(KSFT_SKIP);
@@ -251,39 +256,40 @@ static void *vcpu_thread_main(void *data)
        while (!READ_ONCE(all_vcpu_threads_running))
                ;
 
-       vcpu_thread_fn(&perf_test_args.vcpu_args[vcpu->vcpu_id]);
+       vcpu_thread_fn(&perf_test_args.vcpu_args[vcpu->vcpu_idx]);
 
        return NULL;
 }
 
-void perf_test_start_vcpu_threads(int vcpus, void (*vcpu_fn)(struct perf_test_vcpu_args *))
+void perf_test_start_vcpu_threads(int nr_vcpus,
+                                 void (*vcpu_fn)(struct perf_test_vcpu_args *))
 {
-       int vcpu_id;
+       int i;
 
        vcpu_thread_fn = vcpu_fn;
        WRITE_ONCE(all_vcpu_threads_running, false);
 
-       for (vcpu_id = 0; vcpu_id < vcpus; vcpu_id++) {
-               struct vcpu_thread *vcpu = &vcpu_threads[vcpu_id];
+       for (i = 0; i < nr_vcpus; i++) {
+               struct vcpu_thread *vcpu = &vcpu_threads[i];
 
-               vcpu->vcpu_id = vcpu_id;
+               vcpu->vcpu_idx = i;
                WRITE_ONCE(vcpu->running, false);
 
                pthread_create(&vcpu->thread, NULL, vcpu_thread_main, vcpu);
        }
 
-       for (vcpu_id = 0; vcpu_id < vcpus; vcpu_id++) {
-               while (!READ_ONCE(vcpu_threads[vcpu_id].running))
+       for (i = 0; i < nr_vcpus; i++) {
+               while (!READ_ONCE(vcpu_threads[i].running))
                        ;
        }
 
        WRITE_ONCE(all_vcpu_threads_running, true);
 }
 
-void perf_test_join_vcpu_threads(int vcpus)
+void perf_test_join_vcpu_threads(int nr_vcpus)
 {
-       int vcpu_id;
+       int i;
 
-       for (vcpu_id = 0; vcpu_id < vcpus; vcpu_id++)
-               pthread_join(vcpu_threads[vcpu_id].thread, NULL);
+       for (i = 0; i < nr_vcpus; i++)
+               pthread_join(vcpu_threads[i].thread, NULL);
 }
index f525427a37c4f978c16566e1a741fa2775cee814..446820a549ba62f5324e9394ccfbb11643e93fbe 100644 (file)
@@ -77,7 +77,7 @@ void perf_test_setup_ept(struct vmx_pages *vmx, struct kvm_vm *vm)
        nested_identity_map_1g(vmx, vm, start, end - start);
 }
 
-void perf_test_setup_nested(struct kvm_vm *vm, int nr_vcpus)
+void perf_test_setup_nested(struct kvm_vm *vm, int nr_vcpus, struct kvm_vcpu *vcpus[])
 {
        struct vmx_pages *vmx, *vmx0 = NULL;
        struct kvm_regs regs;
@@ -103,9 +103,9 @@ void perf_test_setup_nested(struct kvm_vm *vm, int nr_vcpus)
                 * Override the vCPU to run perf_test_l1_guest_code() which will
                 * bounce it into L2 before calling perf_test_guest_code().
                 */
-               vcpu_regs_get(vm, vcpu_id, &regs);
+               vcpu_regs_get(vm, vcpus[vcpu_id]->id, &regs);
                regs.rip = (unsigned long) perf_test_l1_guest_code;
-               vcpu_regs_set(vm, vcpu_id, &regs);
-               vcpu_args_set(vm, vcpu_id, 2, vmx_gva, vcpu_id);
+               vcpu_regs_set(vm, vcpus[vcpu_id]->id, &regs);
+               vcpu_args_set(vm, vcpus[vcpu_id]->id, 2, vmx_gva, vcpu_id);
        }
 }
index 1410d0a9141a1e41d61e0b635e20bb7f68a58626..a3efb318211958296b83de448c0bf367570a4953 100644 (file)
@@ -38,19 +38,19 @@ static bool run_vcpus = true;
 
 static void vcpu_worker(struct perf_test_vcpu_args *vcpu_args)
 {
-       int ret;
-       int vcpu_id = vcpu_args->vcpu_id;
+       struct kvm_vcpu *vcpu = vcpu_args->vcpu;
        struct kvm_vm *vm = perf_test_args.vm;
        struct kvm_run *run;
+       int ret;
 
-       run = vcpu_state(vm, vcpu_id);
+       run = vcpu->run;
 
        /* Let the guest access its memory until a stop signal is received */
        while (READ_ONCE(run_vcpus)) {
-               ret = _vcpu_run(vm, vcpu_id);
+               ret = _vcpu_run(vm, vcpu->id);
                TEST_ASSERT(ret == 0, "vcpu_run failed: %d\n", ret);
 
-               if (get_ucall(vm, vcpu_id, NULL) == UCALL_SYNC)
+               if (get_ucall(vm, vcpu->id, NULL) == UCALL_SYNC)
                        continue;
 
                TEST_ASSERT(false,