1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * steal/stolen time test 4 * 5 * Copyright (C) 2020, Red Hat, Inc. 6 */ 7 #define _GNU_SOURCE 8 #include <stdio.h> 9 #include <time.h> 10 #include <sched.h> 11 #include <pthread.h> 12 #include <linux/kernel.h> 13 #include <asm/kvm.h> 14 #ifndef __riscv 15 #include <asm/kvm_para.h> 16 #endif 17 18 #include "test_util.h" 19 #include "kvm_util.h" 20 #include "processor.h" 21 22 #define NR_VCPUS 4 23 #define ST_GPA_BASE (1 << 30) 24 25 static void *st_gva[NR_VCPUS]; 26 static uint64_t guest_stolen_time[NR_VCPUS]; 27 28 #if defined(__x86_64__) 29 30 /* steal_time must have 64-byte alignment */ 31 #define STEAL_TIME_SIZE ((sizeof(struct kvm_steal_time) + 63) & ~63) 32 33 static void check_status(struct kvm_steal_time *st) 34 { 35 GUEST_ASSERT(!(READ_ONCE(st->version) & 1)); 36 GUEST_ASSERT_EQ(READ_ONCE(st->flags), 0); 37 GUEST_ASSERT_EQ(READ_ONCE(st->preempted), 0); 38 } 39 40 static void guest_code(int cpu) 41 { 42 struct kvm_steal_time *st = st_gva[cpu]; 43 uint32_t version; 44 45 GUEST_ASSERT_EQ(rdmsr(MSR_KVM_STEAL_TIME), ((uint64_t)st_gva[cpu] | KVM_MSR_ENABLED)); 46 47 memset(st, 0, sizeof(*st)); 48 GUEST_SYNC(0); 49 50 check_status(st); 51 WRITE_ONCE(guest_stolen_time[cpu], st->steal); 52 version = READ_ONCE(st->version); 53 check_status(st); 54 GUEST_SYNC(1); 55 56 check_status(st); 57 GUEST_ASSERT(version < READ_ONCE(st->version)); 58 WRITE_ONCE(guest_stolen_time[cpu], st->steal); 59 check_status(st); 60 GUEST_DONE(); 61 } 62 63 static bool is_steal_time_supported(struct kvm_vcpu *vcpu) 64 { 65 return kvm_cpu_has(X86_FEATURE_KVM_STEAL_TIME); 66 } 67 68 static void steal_time_init(struct kvm_vcpu *vcpu, uint32_t i) 69 { 70 int ret; 71 72 /* ST_GPA_BASE is identity mapped */ 73 st_gva[i] = (void *)(ST_GPA_BASE + i * STEAL_TIME_SIZE); 74 sync_global_to_guest(vcpu->vm, st_gva[i]); 75 76 ret = _vcpu_set_msr(vcpu, MSR_KVM_STEAL_TIME, 77 (ulong)st_gva[i] | KVM_STEAL_RESERVED_MASK); 78 TEST_ASSERT(ret == 0, "Bad GPA didn't fail"); 79 80 vcpu_set_msr(vcpu, MSR_KVM_STEAL_TIME, (ulong)st_gva[i] | KVM_MSR_ENABLED); 81 } 82 83 static void steal_time_dump(struct kvm_vm *vm, uint32_t vcpu_idx) 84 { 85 struct kvm_steal_time *st = addr_gva2hva(vm, (ulong)st_gva[vcpu_idx]); 86 int i; 87 88 pr_info("VCPU%d:\n", vcpu_idx); 89 pr_info(" steal: %lld\n", st->steal); 90 pr_info(" version: %d\n", st->version); 91 pr_info(" flags: %d\n", st->flags); 92 pr_info(" preempted: %d\n", st->preempted); 93 pr_info(" u8_pad: "); 94 for (i = 0; i < 3; ++i) 95 pr_info("%d", st->u8_pad[i]); 96 pr_info("\n pad: "); 97 for (i = 0; i < 11; ++i) 98 pr_info("%d", st->pad[i]); 99 pr_info("\n"); 100 } 101 102 #elif defined(__aarch64__) 103 104 /* PV_TIME_ST must have 64-byte alignment */ 105 #define STEAL_TIME_SIZE ((sizeof(struct st_time) + 63) & ~63) 106 107 #define SMCCC_ARCH_FEATURES 0x80000001 108 #define PV_TIME_FEATURES 0xc5000020 109 #define PV_TIME_ST 0xc5000021 110 111 struct st_time { 112 uint32_t rev; 113 uint32_t attr; 114 uint64_t st_time; 115 }; 116 117 static int64_t smccc(uint32_t func, uint64_t arg) 118 { 119 struct arm_smccc_res res; 120 121 smccc_hvc(func, arg, 0, 0, 0, 0, 0, 0, &res); 122 return res.a0; 123 } 124 125 static void check_status(struct st_time *st) 126 { 127 GUEST_ASSERT_EQ(READ_ONCE(st->rev), 0); 128 GUEST_ASSERT_EQ(READ_ONCE(st->attr), 0); 129 } 130 131 static void guest_code(int cpu) 132 { 133 struct st_time *st; 134 int64_t status; 135 136 status = smccc(SMCCC_ARCH_FEATURES, PV_TIME_FEATURES); 137 GUEST_ASSERT_EQ(status, 0); 138 status = smccc(PV_TIME_FEATURES, PV_TIME_FEATURES); 139 GUEST_ASSERT_EQ(status, 0); 140 status = smccc(PV_TIME_FEATURES, PV_TIME_ST); 141 GUEST_ASSERT_EQ(status, 0); 142 143 status = smccc(PV_TIME_ST, 0); 144 GUEST_ASSERT_NE(status, -1); 145 GUEST_ASSERT_EQ(status, (ulong)st_gva[cpu]); 146 147 st = (struct st_time *)status; 148 GUEST_SYNC(0); 149 150 check_status(st); 151 WRITE_ONCE(guest_stolen_time[cpu], st->st_time); 152 GUEST_SYNC(1); 153 154 check_status(st); 155 WRITE_ONCE(guest_stolen_time[cpu], st->st_time); 156 GUEST_DONE(); 157 } 158 159 static bool is_steal_time_supported(struct kvm_vcpu *vcpu) 160 { 161 struct kvm_device_attr dev = { 162 .group = KVM_ARM_VCPU_PVTIME_CTRL, 163 .attr = KVM_ARM_VCPU_PVTIME_IPA, 164 }; 165 166 return !__vcpu_ioctl(vcpu, KVM_HAS_DEVICE_ATTR, &dev); 167 } 168 169 static void steal_time_init(struct kvm_vcpu *vcpu, uint32_t i) 170 { 171 struct kvm_vm *vm = vcpu->vm; 172 uint64_t st_ipa; 173 int ret; 174 175 struct kvm_device_attr dev = { 176 .group = KVM_ARM_VCPU_PVTIME_CTRL, 177 .attr = KVM_ARM_VCPU_PVTIME_IPA, 178 .addr = (uint64_t)&st_ipa, 179 }; 180 181 vcpu_ioctl(vcpu, KVM_HAS_DEVICE_ATTR, &dev); 182 183 /* ST_GPA_BASE is identity mapped */ 184 st_gva[i] = (void *)(ST_GPA_BASE + i * STEAL_TIME_SIZE); 185 sync_global_to_guest(vm, st_gva[i]); 186 187 st_ipa = (ulong)st_gva[i] | 1; 188 ret = __vcpu_ioctl(vcpu, KVM_SET_DEVICE_ATTR, &dev); 189 TEST_ASSERT(ret == -1 && errno == EINVAL, "Bad IPA didn't report EINVAL"); 190 191 st_ipa = (ulong)st_gva[i]; 192 vcpu_ioctl(vcpu, KVM_SET_DEVICE_ATTR, &dev); 193 194 ret = __vcpu_ioctl(vcpu, KVM_SET_DEVICE_ATTR, &dev); 195 TEST_ASSERT(ret == -1 && errno == EEXIST, "Set IPA twice without EEXIST"); 196 } 197 198 static void steal_time_dump(struct kvm_vm *vm, uint32_t vcpu_idx) 199 { 200 struct st_time *st = addr_gva2hva(vm, (ulong)st_gva[vcpu_idx]); 201 202 pr_info("VCPU%d:\n", vcpu_idx); 203 pr_info(" rev: %d\n", st->rev); 204 pr_info(" attr: %d\n", st->attr); 205 pr_info(" st_time: %ld\n", st->st_time); 206 } 207 208 #elif defined(__riscv) 209 210 /* SBI STA shmem must have 64-byte alignment */ 211 #define STEAL_TIME_SIZE ((sizeof(struct sta_struct) + 63) & ~63) 212 213 static vm_paddr_t st_gpa[NR_VCPUS]; 214 215 struct sta_struct { 216 uint32_t sequence; 217 uint32_t flags; 218 uint64_t steal; 219 uint8_t preempted; 220 uint8_t pad[47]; 221 } __packed; 222 223 static void sta_set_shmem(vm_paddr_t gpa, unsigned long flags) 224 { 225 unsigned long lo = (unsigned long)gpa; 226 #if __riscv_xlen == 32 227 unsigned long hi = (unsigned long)(gpa >> 32); 228 #else 229 unsigned long hi = gpa == -1 ? -1 : 0; 230 #endif 231 struct sbiret ret = sbi_ecall(SBI_EXT_STA, 0, lo, hi, flags, 0, 0, 0); 232 233 GUEST_ASSERT(ret.value == 0 && ret.error == 0); 234 } 235 236 static void check_status(struct sta_struct *st) 237 { 238 GUEST_ASSERT(!(READ_ONCE(st->sequence) & 1)); 239 GUEST_ASSERT(READ_ONCE(st->flags) == 0); 240 GUEST_ASSERT(READ_ONCE(st->preempted) == 0); 241 } 242 243 static void guest_code(int cpu) 244 { 245 struct sta_struct *st = st_gva[cpu]; 246 uint32_t sequence; 247 long out_val = 0; 248 bool probe; 249 250 probe = guest_sbi_probe_extension(SBI_EXT_STA, &out_val); 251 GUEST_ASSERT(probe && out_val == 1); 252 253 sta_set_shmem(st_gpa[cpu], 0); 254 GUEST_SYNC(0); 255 256 check_status(st); 257 WRITE_ONCE(guest_stolen_time[cpu], st->steal); 258 sequence = READ_ONCE(st->sequence); 259 check_status(st); 260 GUEST_SYNC(1); 261 262 check_status(st); 263 GUEST_ASSERT(sequence < READ_ONCE(st->sequence)); 264 WRITE_ONCE(guest_stolen_time[cpu], st->steal); 265 check_status(st); 266 GUEST_DONE(); 267 } 268 269 static bool is_steal_time_supported(struct kvm_vcpu *vcpu) 270 { 271 uint64_t id = RISCV_SBI_EXT_REG(KVM_RISCV_SBI_EXT_STA); 272 unsigned long enabled; 273 274 vcpu_get_reg(vcpu, id, &enabled); 275 TEST_ASSERT(enabled == 0 || enabled == 1, "Expected boolean result"); 276 277 return enabled; 278 } 279 280 static void steal_time_init(struct kvm_vcpu *vcpu, uint32_t i) 281 { 282 /* ST_GPA_BASE is identity mapped */ 283 st_gva[i] = (void *)(ST_GPA_BASE + i * STEAL_TIME_SIZE); 284 st_gpa[i] = addr_gva2gpa(vcpu->vm, (vm_vaddr_t)st_gva[i]); 285 sync_global_to_guest(vcpu->vm, st_gva[i]); 286 sync_global_to_guest(vcpu->vm, st_gpa[i]); 287 } 288 289 static void steal_time_dump(struct kvm_vm *vm, uint32_t vcpu_idx) 290 { 291 struct sta_struct *st = addr_gva2hva(vm, (ulong)st_gva[vcpu_idx]); 292 int i; 293 294 pr_info("VCPU%d:\n", vcpu_idx); 295 pr_info(" sequence: %d\n", st->sequence); 296 pr_info(" flags: %d\n", st->flags); 297 pr_info(" steal: %"PRIu64"\n", st->steal); 298 pr_info(" preempted: %d\n", st->preempted); 299 pr_info(" pad: "); 300 for (i = 0; i < 47; ++i) 301 pr_info("%d", st->pad[i]); 302 pr_info("\n"); 303 } 304 305 #endif 306 307 static void *do_steal_time(void *arg) 308 { 309 struct timespec ts, stop; 310 311 clock_gettime(CLOCK_MONOTONIC, &ts); 312 stop = timespec_add_ns(ts, MIN_RUN_DELAY_NS); 313 314 while (1) { 315 clock_gettime(CLOCK_MONOTONIC, &ts); 316 if (timespec_to_ns(timespec_sub(ts, stop)) >= 0) 317 break; 318 } 319 320 return NULL; 321 } 322 323 static void run_vcpu(struct kvm_vcpu *vcpu) 324 { 325 struct ucall uc; 326 327 vcpu_run(vcpu); 328 329 switch (get_ucall(vcpu, &uc)) { 330 case UCALL_SYNC: 331 case UCALL_DONE: 332 break; 333 case UCALL_ABORT: 334 REPORT_GUEST_ASSERT(uc); 335 default: 336 TEST_ASSERT(false, "Unexpected exit: %s", 337 exit_reason_str(vcpu->run->exit_reason)); 338 } 339 } 340 341 int main(int ac, char **av) 342 { 343 struct kvm_vcpu *vcpus[NR_VCPUS]; 344 struct kvm_vm *vm; 345 pthread_attr_t attr; 346 pthread_t thread; 347 cpu_set_t cpuset; 348 unsigned int gpages; 349 long stolen_time; 350 long run_delay; 351 bool verbose; 352 int i; 353 354 verbose = ac > 1 && (!strncmp(av[1], "-v", 3) || !strncmp(av[1], "--verbose", 10)); 355 356 /* Set CPU affinity so we can force preemption of the VCPU */ 357 CPU_ZERO(&cpuset); 358 CPU_SET(0, &cpuset); 359 pthread_attr_init(&attr); 360 pthread_attr_setaffinity_np(&attr, sizeof(cpu_set_t), &cpuset); 361 pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset); 362 363 /* Create a VM and an identity mapped memslot for the steal time structure */ 364 vm = vm_create_with_vcpus(NR_VCPUS, guest_code, vcpus); 365 gpages = vm_calc_num_guest_pages(VM_MODE_DEFAULT, STEAL_TIME_SIZE * NR_VCPUS); 366 vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS, ST_GPA_BASE, 1, gpages, 0); 367 virt_map(vm, ST_GPA_BASE, ST_GPA_BASE, gpages); 368 369 TEST_REQUIRE(is_steal_time_supported(vcpus[0])); 370 371 /* Run test on each VCPU */ 372 for (i = 0; i < NR_VCPUS; ++i) { 373 steal_time_init(vcpus[i], i); 374 375 vcpu_args_set(vcpus[i], 1, i); 376 377 /* First VCPU run initializes steal-time */ 378 run_vcpu(vcpus[i]); 379 380 /* Second VCPU run, expect guest stolen time to be <= run_delay */ 381 run_vcpu(vcpus[i]); 382 sync_global_from_guest(vm, guest_stolen_time[i]); 383 stolen_time = guest_stolen_time[i]; 384 run_delay = get_run_delay(); 385 TEST_ASSERT(stolen_time <= run_delay, 386 "Expected stolen time <= %ld, got %ld", 387 run_delay, stolen_time); 388 389 /* Steal time from the VCPU. The steal time thread has the same CPU affinity as the VCPUs. */ 390 run_delay = get_run_delay(); 391 pthread_create(&thread, &attr, do_steal_time, NULL); 392 do 393 sched_yield(); 394 while (get_run_delay() - run_delay < MIN_RUN_DELAY_NS); 395 pthread_join(thread, NULL); 396 run_delay = get_run_delay() - run_delay; 397 TEST_ASSERT(run_delay >= MIN_RUN_DELAY_NS, 398 "Expected run_delay >= %ld, got %ld", 399 MIN_RUN_DELAY_NS, run_delay); 400 401 /* Run VCPU again to confirm stolen time is consistent with run_delay */ 402 run_vcpu(vcpus[i]); 403 sync_global_from_guest(vm, guest_stolen_time[i]); 404 stolen_time = guest_stolen_time[i] - stolen_time; 405 TEST_ASSERT(stolen_time >= run_delay, 406 "Expected stolen time >= %ld, got %ld", 407 run_delay, stolen_time); 408 409 if (verbose) { 410 pr_info("VCPU%d: total-stolen-time=%ld test-stolen-time=%ld", i, 411 guest_stolen_time[i], stolen_time); 412 if (stolen_time == run_delay) 413 pr_info(" (BONUS: guest test-stolen-time even exactly matches test-run_delay)"); 414 pr_info("\n"); 415 steal_time_dump(vm, i); 416 } 417 } 418 419 return 0; 420 } 421