1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * KVM dirty page logging test 4 * 5 * Copyright (C) 2018, Red Hat, Inc. 6 */ 7 8 #define _GNU_SOURCE /* for program_invocation_name */ 9 10 #include <stdio.h> 11 #include <stdlib.h> 12 #include <pthread.h> 13 #include <semaphore.h> 14 #include <sys/types.h> 15 #include <signal.h> 16 #include <errno.h> 17 #include <linux/bitmap.h> 18 #include <linux/bitops.h> 19 #include <linux/atomic.h> 20 21 #include "kvm_util.h" 22 #include "test_util.h" 23 #include "guest_modes.h" 24 #include "processor.h" 25 26 #define VCPU_ID 1 27 28 /* The memory slot index to track dirty pages */ 29 #define TEST_MEM_SLOT_INDEX 1 30 31 /* Default guest test virtual memory offset */ 32 #define DEFAULT_GUEST_TEST_MEM 0xc0000000 33 34 /* How many pages to dirty for each guest loop */ 35 #define TEST_PAGES_PER_LOOP 1024 36 37 /* How many host loops to run (one KVM_GET_DIRTY_LOG for each loop) */ 38 #define TEST_HOST_LOOP_N 32UL 39 40 /* Interval for each host loop (ms) */ 41 #define TEST_HOST_LOOP_INTERVAL 10UL 42 43 /* Dirty bitmaps are always little endian, so we need to swap on big endian */ 44 #if defined(__s390x__) 45 # define BITOP_LE_SWIZZLE ((BITS_PER_LONG-1) & ~0x7) 46 # define test_bit_le(nr, addr) \ 47 test_bit((nr) ^ BITOP_LE_SWIZZLE, addr) 48 # define set_bit_le(nr, addr) \ 49 set_bit((nr) ^ BITOP_LE_SWIZZLE, addr) 50 # define clear_bit_le(nr, addr) \ 51 clear_bit((nr) ^ BITOP_LE_SWIZZLE, addr) 52 # define test_and_set_bit_le(nr, addr) \ 53 test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, addr) 54 # define test_and_clear_bit_le(nr, addr) \ 55 test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, addr) 56 #else 57 # define test_bit_le test_bit 58 # define set_bit_le set_bit 59 # define clear_bit_le clear_bit 60 # define test_and_set_bit_le test_and_set_bit 61 # define test_and_clear_bit_le test_and_clear_bit 62 #endif 63 64 #define TEST_DIRTY_RING_COUNT 65536 65 66 #define SIG_IPI SIGUSR1 67 68 /* 69 * Guest/Host shared variables. Ensure addr_gva2hva() and/or 70 * sync_global_to/from_guest() are used when accessing from 71 * the host. READ/WRITE_ONCE() should also be used with anything 72 * that may change. 73 */ 74 static uint64_t host_page_size; 75 static uint64_t guest_page_size; 76 static uint64_t guest_num_pages; 77 static uint64_t random_array[TEST_PAGES_PER_LOOP]; 78 static uint64_t iteration; 79 80 /* 81 * Guest physical memory offset of the testing memory slot. 82 * This will be set to the topmost valid physical address minus 83 * the test memory size. 84 */ 85 static uint64_t guest_test_phys_mem; 86 87 /* 88 * Guest virtual memory offset of the testing memory slot. 89 * Must not conflict with identity mapped test code. 90 */ 91 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM; 92 93 /* 94 * Continuously write to the first 8 bytes of a random pages within 95 * the testing memory region. 96 */ 97 static void guest_code(void) 98 { 99 uint64_t addr; 100 int i; 101 102 /* 103 * On s390x, all pages of a 1M segment are initially marked as dirty 104 * when a page of the segment is written to for the very first time. 105 * To compensate this specialty in this test, we need to touch all 106 * pages during the first iteration. 107 */ 108 for (i = 0; i < guest_num_pages; i++) { 109 addr = guest_test_virt_mem + i * guest_page_size; 110 *(uint64_t *)addr = READ_ONCE(iteration); 111 } 112 113 while (true) { 114 for (i = 0; i < TEST_PAGES_PER_LOOP; i++) { 115 addr = guest_test_virt_mem; 116 addr += (READ_ONCE(random_array[i]) % guest_num_pages) 117 * guest_page_size; 118 addr = align_down(addr, host_page_size); 119 *(uint64_t *)addr = READ_ONCE(iteration); 120 } 121 122 /* Tell the host that we need more random numbers */ 123 GUEST_SYNC(1); 124 } 125 } 126 127 /* Host variables */ 128 static bool host_quit; 129 130 /* Points to the test VM memory region on which we track dirty logs */ 131 static void *host_test_mem; 132 static uint64_t host_num_pages; 133 134 /* For statistics only */ 135 static uint64_t host_dirty_count; 136 static uint64_t host_clear_count; 137 static uint64_t host_track_next_count; 138 139 /* Whether dirty ring reset is requested, or finished */ 140 static sem_t sem_vcpu_stop; 141 static sem_t sem_vcpu_cont; 142 /* 143 * This is only set by main thread, and only cleared by vcpu thread. It is 144 * used to request vcpu thread to stop at the next GUEST_SYNC, since GUEST_SYNC 145 * is the only place that we'll guarantee both "dirty bit" and "dirty data" 146 * will match. E.g., SIG_IPI won't guarantee that if the vcpu is interrupted 147 * after setting dirty bit but before the data is written. 148 */ 149 static atomic_t vcpu_sync_stop_requested; 150 /* 151 * This is updated by the vcpu thread to tell the host whether it's a 152 * ring-full event. It should only be read until a sem_wait() of 153 * sem_vcpu_stop and before vcpu continues to run. 154 */ 155 static bool dirty_ring_vcpu_ring_full; 156 /* 157 * This is only used for verifying the dirty pages. Dirty ring has a very 158 * tricky case when the ring just got full, kvm will do userspace exit due to 159 * ring full. When that happens, the very last PFN is set but actually the 160 * data is not changed (the guest WRITE is not really applied yet), because 161 * we found that the dirty ring is full, refused to continue the vcpu, and 162 * recorded the dirty gfn with the old contents. 163 * 164 * For this specific case, it's safe to skip checking this pfn for this 165 * bit, because it's a redundant bit, and when the write happens later the bit 166 * will be set again. We use this variable to always keep track of the latest 167 * dirty gfn we've collected, so that if a mismatch of data found later in the 168 * verifying process, we let it pass. 169 */ 170 static uint64_t dirty_ring_last_page; 171 172 enum log_mode_t { 173 /* Only use KVM_GET_DIRTY_LOG for logging */ 174 LOG_MODE_DIRTY_LOG = 0, 175 176 /* Use both KVM_[GET|CLEAR]_DIRTY_LOG for logging */ 177 LOG_MODE_CLEAR_LOG = 1, 178 179 /* Use dirty ring for logging */ 180 LOG_MODE_DIRTY_RING = 2, 181 182 LOG_MODE_NUM, 183 184 /* Run all supported modes */ 185 LOG_MODE_ALL = LOG_MODE_NUM, 186 }; 187 188 /* Mode of logging to test. Default is to run all supported modes */ 189 static enum log_mode_t host_log_mode_option = LOG_MODE_ALL; 190 /* Logging mode for current run */ 191 static enum log_mode_t host_log_mode; 192 static pthread_t vcpu_thread; 193 static uint32_t test_dirty_ring_count = TEST_DIRTY_RING_COUNT; 194 195 static void vcpu_kick(void) 196 { 197 pthread_kill(vcpu_thread, SIG_IPI); 198 } 199 200 /* 201 * In our test we do signal tricks, let's use a better version of 202 * sem_wait to avoid signal interrupts 203 */ 204 static void sem_wait_until(sem_t *sem) 205 { 206 int ret; 207 208 do 209 ret = sem_wait(sem); 210 while (ret == -1 && errno == EINTR); 211 } 212 213 static bool clear_log_supported(void) 214 { 215 return kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2); 216 } 217 218 static void clear_log_create_vm_done(struct kvm_vm *vm) 219 { 220 u64 manual_caps; 221 222 manual_caps = kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2); 223 TEST_ASSERT(manual_caps, "MANUAL_CAPS is zero!"); 224 manual_caps &= (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE | 225 KVM_DIRTY_LOG_INITIALLY_SET); 226 vm_enable_cap(vm, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2, manual_caps); 227 } 228 229 static void dirty_log_collect_dirty_pages(struct kvm_vm *vm, int slot, 230 void *bitmap, uint32_t num_pages) 231 { 232 kvm_vm_get_dirty_log(vm, slot, bitmap); 233 } 234 235 static void clear_log_collect_dirty_pages(struct kvm_vm *vm, int slot, 236 void *bitmap, uint32_t num_pages) 237 { 238 kvm_vm_get_dirty_log(vm, slot, bitmap); 239 kvm_vm_clear_dirty_log(vm, slot, bitmap, 0, num_pages); 240 } 241 242 /* Should only be called after a GUEST_SYNC */ 243 static void vcpu_handle_sync_stop(void) 244 { 245 if (atomic_read(&vcpu_sync_stop_requested)) { 246 /* It means main thread is sleeping waiting */ 247 atomic_set(&vcpu_sync_stop_requested, false); 248 sem_post(&sem_vcpu_stop); 249 sem_wait_until(&sem_vcpu_cont); 250 } 251 } 252 253 static void default_after_vcpu_run(struct kvm_vm *vm, int ret, int err) 254 { 255 struct kvm_run *run = vcpu_state(vm, VCPU_ID); 256 257 TEST_ASSERT(ret == 0 || (ret == -1 && err == EINTR), 258 "vcpu run failed: errno=%d", err); 259 260 TEST_ASSERT(get_ucall(vm, VCPU_ID, NULL) == UCALL_SYNC, 261 "Invalid guest sync status: exit_reason=%s\n", 262 exit_reason_str(run->exit_reason)); 263 264 vcpu_handle_sync_stop(); 265 } 266 267 static bool dirty_ring_supported(void) 268 { 269 return kvm_check_cap(KVM_CAP_DIRTY_LOG_RING); 270 } 271 272 static void dirty_ring_create_vm_done(struct kvm_vm *vm) 273 { 274 /* 275 * Switch to dirty ring mode after VM creation but before any 276 * of the vcpu creation. 277 */ 278 vm_enable_dirty_ring(vm, test_dirty_ring_count * 279 sizeof(struct kvm_dirty_gfn)); 280 } 281 282 static inline bool dirty_gfn_is_dirtied(struct kvm_dirty_gfn *gfn) 283 { 284 return gfn->flags == KVM_DIRTY_GFN_F_DIRTY; 285 } 286 287 static inline void dirty_gfn_set_collected(struct kvm_dirty_gfn *gfn) 288 { 289 gfn->flags = KVM_DIRTY_GFN_F_RESET; 290 } 291 292 static uint32_t dirty_ring_collect_one(struct kvm_dirty_gfn *dirty_gfns, 293 int slot, void *bitmap, 294 uint32_t num_pages, uint32_t *fetch_index) 295 { 296 struct kvm_dirty_gfn *cur; 297 uint32_t count = 0; 298 299 while (true) { 300 cur = &dirty_gfns[*fetch_index % test_dirty_ring_count]; 301 if (!dirty_gfn_is_dirtied(cur)) 302 break; 303 TEST_ASSERT(cur->slot == slot, "Slot number didn't match: " 304 "%u != %u", cur->slot, slot); 305 TEST_ASSERT(cur->offset < num_pages, "Offset overflow: " 306 "0x%llx >= 0x%x", cur->offset, num_pages); 307 //pr_info("fetch 0x%x page %llu\n", *fetch_index, cur->offset); 308 set_bit_le(cur->offset, bitmap); 309 dirty_ring_last_page = cur->offset; 310 dirty_gfn_set_collected(cur); 311 (*fetch_index)++; 312 count++; 313 } 314 315 return count; 316 } 317 318 static void dirty_ring_wait_vcpu(void) 319 { 320 /* This makes sure that hardware PML cache flushed */ 321 vcpu_kick(); 322 sem_wait_until(&sem_vcpu_stop); 323 } 324 325 static void dirty_ring_continue_vcpu(void) 326 { 327 pr_info("Notifying vcpu to continue\n"); 328 sem_post(&sem_vcpu_cont); 329 } 330 331 static void dirty_ring_collect_dirty_pages(struct kvm_vm *vm, int slot, 332 void *bitmap, uint32_t num_pages) 333 { 334 /* We only have one vcpu */ 335 static uint32_t fetch_index = 0; 336 uint32_t count = 0, cleared; 337 bool continued_vcpu = false; 338 339 dirty_ring_wait_vcpu(); 340 341 if (!dirty_ring_vcpu_ring_full) { 342 /* 343 * This is not a ring-full event, it's safe to allow 344 * vcpu to continue 345 */ 346 dirty_ring_continue_vcpu(); 347 continued_vcpu = true; 348 } 349 350 /* Only have one vcpu */ 351 count = dirty_ring_collect_one(vcpu_map_dirty_ring(vm, VCPU_ID), 352 slot, bitmap, num_pages, &fetch_index); 353 354 cleared = kvm_vm_reset_dirty_ring(vm); 355 356 /* Cleared pages should be the same as collected */ 357 TEST_ASSERT(cleared == count, "Reset dirty pages (%u) mismatch " 358 "with collected (%u)", cleared, count); 359 360 if (!continued_vcpu) { 361 TEST_ASSERT(dirty_ring_vcpu_ring_full, 362 "Didn't continue vcpu even without ring full"); 363 dirty_ring_continue_vcpu(); 364 } 365 366 pr_info("Iteration %ld collected %u pages\n", iteration, count); 367 } 368 369 static void dirty_ring_after_vcpu_run(struct kvm_vm *vm, int ret, int err) 370 { 371 struct kvm_run *run = vcpu_state(vm, VCPU_ID); 372 373 /* A ucall-sync or ring-full event is allowed */ 374 if (get_ucall(vm, VCPU_ID, NULL) == UCALL_SYNC) { 375 /* We should allow this to continue */ 376 ; 377 } else if (run->exit_reason == KVM_EXIT_DIRTY_RING_FULL || 378 (ret == -1 && err == EINTR)) { 379 /* Update the flag first before pause */ 380 WRITE_ONCE(dirty_ring_vcpu_ring_full, 381 run->exit_reason == KVM_EXIT_DIRTY_RING_FULL); 382 sem_post(&sem_vcpu_stop); 383 pr_info("vcpu stops because %s...\n", 384 dirty_ring_vcpu_ring_full ? 385 "dirty ring is full" : "vcpu is kicked out"); 386 sem_wait_until(&sem_vcpu_cont); 387 pr_info("vcpu continues now.\n"); 388 } else { 389 TEST_ASSERT(false, "Invalid guest sync status: " 390 "exit_reason=%s\n", 391 exit_reason_str(run->exit_reason)); 392 } 393 } 394 395 static void dirty_ring_before_vcpu_join(void) 396 { 397 /* Kick another round of vcpu just to make sure it will quit */ 398 sem_post(&sem_vcpu_cont); 399 } 400 401 struct log_mode { 402 const char *name; 403 /* Return true if this mode is supported, otherwise false */ 404 bool (*supported)(void); 405 /* Hook when the vm creation is done (before vcpu creation) */ 406 void (*create_vm_done)(struct kvm_vm *vm); 407 /* Hook to collect the dirty pages into the bitmap provided */ 408 void (*collect_dirty_pages) (struct kvm_vm *vm, int slot, 409 void *bitmap, uint32_t num_pages); 410 /* Hook to call when after each vcpu run */ 411 void (*after_vcpu_run)(struct kvm_vm *vm, int ret, int err); 412 void (*before_vcpu_join) (void); 413 } log_modes[LOG_MODE_NUM] = { 414 { 415 .name = "dirty-log", 416 .collect_dirty_pages = dirty_log_collect_dirty_pages, 417 .after_vcpu_run = default_after_vcpu_run, 418 }, 419 { 420 .name = "clear-log", 421 .supported = clear_log_supported, 422 .create_vm_done = clear_log_create_vm_done, 423 .collect_dirty_pages = clear_log_collect_dirty_pages, 424 .after_vcpu_run = default_after_vcpu_run, 425 }, 426 { 427 .name = "dirty-ring", 428 .supported = dirty_ring_supported, 429 .create_vm_done = dirty_ring_create_vm_done, 430 .collect_dirty_pages = dirty_ring_collect_dirty_pages, 431 .before_vcpu_join = dirty_ring_before_vcpu_join, 432 .after_vcpu_run = dirty_ring_after_vcpu_run, 433 }, 434 }; 435 436 /* 437 * We use this bitmap to track some pages that should have its dirty 438 * bit set in the _next_ iteration. For example, if we detected the 439 * page value changed to current iteration but at the same time the 440 * page bit is cleared in the latest bitmap, then the system must 441 * report that write in the next get dirty log call. 442 */ 443 static unsigned long *host_bmap_track; 444 445 static void log_modes_dump(void) 446 { 447 int i; 448 449 printf("all"); 450 for (i = 0; i < LOG_MODE_NUM; i++) 451 printf(", %s", log_modes[i].name); 452 printf("\n"); 453 } 454 455 static bool log_mode_supported(void) 456 { 457 struct log_mode *mode = &log_modes[host_log_mode]; 458 459 if (mode->supported) 460 return mode->supported(); 461 462 return true; 463 } 464 465 static void log_mode_create_vm_done(struct kvm_vm *vm) 466 { 467 struct log_mode *mode = &log_modes[host_log_mode]; 468 469 if (mode->create_vm_done) 470 mode->create_vm_done(vm); 471 } 472 473 static void log_mode_collect_dirty_pages(struct kvm_vm *vm, int slot, 474 void *bitmap, uint32_t num_pages) 475 { 476 struct log_mode *mode = &log_modes[host_log_mode]; 477 478 TEST_ASSERT(mode->collect_dirty_pages != NULL, 479 "collect_dirty_pages() is required for any log mode!"); 480 mode->collect_dirty_pages(vm, slot, bitmap, num_pages); 481 } 482 483 static void log_mode_after_vcpu_run(struct kvm_vm *vm, int ret, int err) 484 { 485 struct log_mode *mode = &log_modes[host_log_mode]; 486 487 if (mode->after_vcpu_run) 488 mode->after_vcpu_run(vm, ret, err); 489 } 490 491 static void log_mode_before_vcpu_join(void) 492 { 493 struct log_mode *mode = &log_modes[host_log_mode]; 494 495 if (mode->before_vcpu_join) 496 mode->before_vcpu_join(); 497 } 498 499 static void generate_random_array(uint64_t *guest_array, uint64_t size) 500 { 501 uint64_t i; 502 503 for (i = 0; i < size; i++) 504 guest_array[i] = random(); 505 } 506 507 static void *vcpu_worker(void *data) 508 { 509 int ret; 510 struct kvm_vm *vm = data; 511 uint64_t *guest_array; 512 uint64_t pages_count = 0; 513 struct kvm_signal_mask *sigmask = alloca(offsetof(struct kvm_signal_mask, sigset) 514 + sizeof(sigset_t)); 515 sigset_t *sigset = (sigset_t *) &sigmask->sigset; 516 517 /* 518 * SIG_IPI is unblocked atomically while in KVM_RUN. It causes the 519 * ioctl to return with -EINTR, but it is still pending and we need 520 * to accept it with the sigwait. 521 */ 522 sigmask->len = 8; 523 pthread_sigmask(0, NULL, sigset); 524 sigdelset(sigset, SIG_IPI); 525 vcpu_ioctl(vm, VCPU_ID, KVM_SET_SIGNAL_MASK, sigmask); 526 527 sigemptyset(sigset); 528 sigaddset(sigset, SIG_IPI); 529 530 guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array); 531 532 while (!READ_ONCE(host_quit)) { 533 /* Clear any existing kick signals */ 534 generate_random_array(guest_array, TEST_PAGES_PER_LOOP); 535 pages_count += TEST_PAGES_PER_LOOP; 536 /* Let the guest dirty the random pages */ 537 ret = __vcpu_run(vm, VCPU_ID); 538 if (ret == -1 && errno == EINTR) { 539 int sig = -1; 540 sigwait(sigset, &sig); 541 assert(sig == SIG_IPI); 542 } 543 log_mode_after_vcpu_run(vm, ret, errno); 544 } 545 546 pr_info("Dirtied %"PRIu64" pages\n", pages_count); 547 548 return NULL; 549 } 550 551 static void vm_dirty_log_verify(enum vm_guest_mode mode, unsigned long *bmap) 552 { 553 uint64_t step = vm_num_host_pages(mode, 1); 554 uint64_t page; 555 uint64_t *value_ptr; 556 uint64_t min_iter = 0; 557 558 for (page = 0; page < host_num_pages; page += step) { 559 value_ptr = host_test_mem + page * host_page_size; 560 561 /* If this is a special page that we were tracking... */ 562 if (test_and_clear_bit_le(page, host_bmap_track)) { 563 host_track_next_count++; 564 TEST_ASSERT(test_bit_le(page, bmap), 565 "Page %"PRIu64" should have its dirty bit " 566 "set in this iteration but it is missing", 567 page); 568 } 569 570 if (test_and_clear_bit_le(page, bmap)) { 571 bool matched; 572 573 host_dirty_count++; 574 575 /* 576 * If the bit is set, the value written onto 577 * the corresponding page should be either the 578 * previous iteration number or the current one. 579 */ 580 matched = (*value_ptr == iteration || 581 *value_ptr == iteration - 1); 582 583 if (host_log_mode == LOG_MODE_DIRTY_RING && !matched) { 584 if (*value_ptr == iteration - 2 && min_iter <= iteration - 2) { 585 /* 586 * Short answer: this case is special 587 * only for dirty ring test where the 588 * page is the last page before a kvm 589 * dirty ring full in iteration N-2. 590 * 591 * Long answer: Assuming ring size R, 592 * one possible condition is: 593 * 594 * main thr vcpu thr 595 * -------- -------- 596 * iter=1 597 * write 1 to page 0~(R-1) 598 * full, vmexit 599 * collect 0~(R-1) 600 * kick vcpu 601 * write 1 to (R-1)~(2R-2) 602 * full, vmexit 603 * iter=2 604 * collect (R-1)~(2R-2) 605 * kick vcpu 606 * write 1 to (2R-2) 607 * (NOTE!!! "1" cached in cpu reg) 608 * write 2 to (2R-1)~(3R-3) 609 * full, vmexit 610 * iter=3 611 * collect (2R-2)~(3R-3) 612 * (here if we read value on page 613 * "2R-2" is 1, while iter=3!!!) 614 * 615 * This however can only happen once per iteration. 616 */ 617 min_iter = iteration - 1; 618 continue; 619 } else if (page == dirty_ring_last_page) { 620 /* 621 * Please refer to comments in 622 * dirty_ring_last_page. 623 */ 624 continue; 625 } 626 } 627 628 TEST_ASSERT(matched, 629 "Set page %"PRIu64" value %"PRIu64 630 " incorrect (iteration=%"PRIu64")", 631 page, *value_ptr, iteration); 632 } else { 633 host_clear_count++; 634 /* 635 * If cleared, the value written can be any 636 * value smaller or equals to the iteration 637 * number. Note that the value can be exactly 638 * (iteration-1) if that write can happen 639 * like this: 640 * 641 * (1) increase loop count to "iteration-1" 642 * (2) write to page P happens (with value 643 * "iteration-1") 644 * (3) get dirty log for "iteration-1"; we'll 645 * see that page P bit is set (dirtied), 646 * and not set the bit in host_bmap_track 647 * (4) increase loop count to "iteration" 648 * (which is current iteration) 649 * (5) get dirty log for current iteration, 650 * we'll see that page P is cleared, with 651 * value "iteration-1". 652 */ 653 TEST_ASSERT(*value_ptr <= iteration, 654 "Clear page %"PRIu64" value %"PRIu64 655 " incorrect (iteration=%"PRIu64")", 656 page, *value_ptr, iteration); 657 if (*value_ptr == iteration) { 658 /* 659 * This page is _just_ modified; it 660 * should report its dirtyness in the 661 * next run 662 */ 663 set_bit_le(page, host_bmap_track); 664 } 665 } 666 } 667 } 668 669 static struct kvm_vm *create_vm(enum vm_guest_mode mode, uint32_t vcpuid, 670 uint64_t extra_mem_pages, void *guest_code) 671 { 672 struct kvm_vm *vm; 673 uint64_t extra_pg_pages = extra_mem_pages / 512 * 2; 674 675 pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode)); 676 677 vm = __vm_create(mode, DEFAULT_GUEST_PHY_PAGES + extra_pg_pages); 678 kvm_vm_elf_load(vm, program_invocation_name); 679 #ifdef __x86_64__ 680 vm_create_irqchip(vm); 681 #endif 682 log_mode_create_vm_done(vm); 683 vm_vcpu_add_default(vm, vcpuid, guest_code); 684 return vm; 685 } 686 687 #define DIRTY_MEM_BITS 30 /* 1G */ 688 #define PAGE_SHIFT_4K 12 689 690 struct test_params { 691 unsigned long iterations; 692 unsigned long interval; 693 uint64_t phys_offset; 694 }; 695 696 static void run_test(enum vm_guest_mode mode, void *arg) 697 { 698 struct test_params *p = arg; 699 struct kvm_vm *vm; 700 unsigned long *bmap; 701 702 if (!log_mode_supported()) { 703 print_skip("Log mode '%s' not supported", 704 log_modes[host_log_mode].name); 705 return; 706 } 707 708 /* 709 * We reserve page table for 2 times of extra dirty mem which 710 * will definitely cover the original (1G+) test range. Here 711 * we do the calculation with 4K page size which is the 712 * smallest so the page number will be enough for all archs 713 * (e.g., 64K page size guest will need even less memory for 714 * page tables). 715 */ 716 vm = create_vm(mode, VCPU_ID, 717 2ul << (DIRTY_MEM_BITS - PAGE_SHIFT_4K), 718 guest_code); 719 720 guest_page_size = vm_get_page_size(vm); 721 /* 722 * A little more than 1G of guest page sized pages. Cover the 723 * case where the size is not aligned to 64 pages. 724 */ 725 guest_num_pages = (1ul << (DIRTY_MEM_BITS - 726 vm_get_page_shift(vm))) + 3; 727 guest_num_pages = vm_adjust_num_guest_pages(mode, guest_num_pages); 728 729 host_page_size = getpagesize(); 730 host_num_pages = vm_num_host_pages(mode, guest_num_pages); 731 732 if (!p->phys_offset) { 733 guest_test_phys_mem = (vm_get_max_gfn(vm) - 734 guest_num_pages) * guest_page_size; 735 guest_test_phys_mem = align_down(guest_test_phys_mem, host_page_size); 736 } else { 737 guest_test_phys_mem = p->phys_offset; 738 } 739 740 #ifdef __s390x__ 741 /* Align to 1M (segment size) */ 742 guest_test_phys_mem = align_down(guest_test_phys_mem, 1 << 20); 743 #endif 744 745 pr_info("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem); 746 747 bmap = bitmap_zalloc(host_num_pages); 748 host_bmap_track = bitmap_zalloc(host_num_pages); 749 750 /* Add an extra memory slot for testing dirty logging */ 751 vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS, 752 guest_test_phys_mem, 753 TEST_MEM_SLOT_INDEX, 754 guest_num_pages, 755 KVM_MEM_LOG_DIRTY_PAGES); 756 757 /* Do mapping for the dirty track memory slot */ 758 virt_map(vm, guest_test_virt_mem, guest_test_phys_mem, guest_num_pages); 759 760 /* Cache the HVA pointer of the region */ 761 host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem); 762 763 ucall_init(vm, NULL); 764 765 /* Export the shared variables to the guest */ 766 sync_global_to_guest(vm, host_page_size); 767 sync_global_to_guest(vm, guest_page_size); 768 sync_global_to_guest(vm, guest_test_virt_mem); 769 sync_global_to_guest(vm, guest_num_pages); 770 771 /* Start the iterations */ 772 iteration = 1; 773 sync_global_to_guest(vm, iteration); 774 host_quit = false; 775 host_dirty_count = 0; 776 host_clear_count = 0; 777 host_track_next_count = 0; 778 779 pthread_create(&vcpu_thread, NULL, vcpu_worker, vm); 780 781 while (iteration < p->iterations) { 782 /* Give the vcpu thread some time to dirty some pages */ 783 usleep(p->interval * 1000); 784 log_mode_collect_dirty_pages(vm, TEST_MEM_SLOT_INDEX, 785 bmap, host_num_pages); 786 787 /* 788 * See vcpu_sync_stop_requested definition for details on why 789 * we need to stop vcpu when verify data. 790 */ 791 atomic_set(&vcpu_sync_stop_requested, true); 792 sem_wait_until(&sem_vcpu_stop); 793 /* 794 * NOTE: for dirty ring, it's possible that we didn't stop at 795 * GUEST_SYNC but instead we stopped because ring is full; 796 * that's okay too because ring full means we're only missing 797 * the flush of the last page, and since we handle the last 798 * page specially verification will succeed anyway. 799 */ 800 assert(host_log_mode == LOG_MODE_DIRTY_RING || 801 atomic_read(&vcpu_sync_stop_requested) == false); 802 vm_dirty_log_verify(mode, bmap); 803 sem_post(&sem_vcpu_cont); 804 805 iteration++; 806 sync_global_to_guest(vm, iteration); 807 } 808 809 /* Tell the vcpu thread to quit */ 810 host_quit = true; 811 log_mode_before_vcpu_join(); 812 pthread_join(vcpu_thread, NULL); 813 814 pr_info("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), " 815 "track_next (%"PRIu64")\n", host_dirty_count, host_clear_count, 816 host_track_next_count); 817 818 free(bmap); 819 free(host_bmap_track); 820 ucall_uninit(vm); 821 kvm_vm_free(vm); 822 } 823 824 static void help(char *name) 825 { 826 puts(""); 827 printf("usage: %s [-h] [-i iterations] [-I interval] " 828 "[-p offset] [-m mode]\n", name); 829 puts(""); 830 printf(" -c: specify dirty ring size, in number of entries\n"); 831 printf(" (only useful for dirty-ring test; default: %"PRIu32")\n", 832 TEST_DIRTY_RING_COUNT); 833 printf(" -i: specify iteration counts (default: %"PRIu64")\n", 834 TEST_HOST_LOOP_N); 835 printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n", 836 TEST_HOST_LOOP_INTERVAL); 837 printf(" -p: specify guest physical test memory offset\n" 838 " Warning: a low offset can conflict with the loaded test code.\n"); 839 printf(" -M: specify the host logging mode " 840 "(default: run all log modes). Supported modes: \n\t"); 841 log_modes_dump(); 842 guest_modes_help(); 843 puts(""); 844 exit(0); 845 } 846 847 int main(int argc, char *argv[]) 848 { 849 struct test_params p = { 850 .iterations = TEST_HOST_LOOP_N, 851 .interval = TEST_HOST_LOOP_INTERVAL, 852 }; 853 int opt, i; 854 sigset_t sigset; 855 856 sem_init(&sem_vcpu_stop, 0, 0); 857 sem_init(&sem_vcpu_cont, 0, 0); 858 859 guest_modes_append_default(); 860 861 while ((opt = getopt(argc, argv, "c:hi:I:p:m:M:")) != -1) { 862 switch (opt) { 863 case 'c': 864 test_dirty_ring_count = strtol(optarg, NULL, 10); 865 break; 866 case 'i': 867 p.iterations = strtol(optarg, NULL, 10); 868 break; 869 case 'I': 870 p.interval = strtol(optarg, NULL, 10); 871 break; 872 case 'p': 873 p.phys_offset = strtoull(optarg, NULL, 0); 874 break; 875 case 'm': 876 guest_modes_cmdline(optarg); 877 break; 878 case 'M': 879 if (!strcmp(optarg, "all")) { 880 host_log_mode_option = LOG_MODE_ALL; 881 break; 882 } 883 for (i = 0; i < LOG_MODE_NUM; i++) { 884 if (!strcmp(optarg, log_modes[i].name)) { 885 pr_info("Setting log mode to: '%s'\n", 886 optarg); 887 host_log_mode_option = i; 888 break; 889 } 890 } 891 if (i == LOG_MODE_NUM) { 892 printf("Log mode '%s' invalid. Please choose " 893 "from: ", optarg); 894 log_modes_dump(); 895 exit(1); 896 } 897 break; 898 case 'h': 899 default: 900 help(argv[0]); 901 break; 902 } 903 } 904 905 TEST_ASSERT(p.iterations > 2, "Iterations must be greater than two"); 906 TEST_ASSERT(p.interval > 0, "Interval must be greater than zero"); 907 908 pr_info("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n", 909 p.iterations, p.interval); 910 911 srandom(time(0)); 912 913 /* Ensure that vCPU threads start with SIG_IPI blocked. */ 914 sigemptyset(&sigset); 915 sigaddset(&sigset, SIG_IPI); 916 pthread_sigmask(SIG_BLOCK, &sigset, NULL); 917 918 if (host_log_mode_option == LOG_MODE_ALL) { 919 /* Run each log mode */ 920 for (i = 0; i < LOG_MODE_NUM; i++) { 921 pr_info("Testing Log Mode '%s'\n", log_modes[i].name); 922 host_log_mode = i; 923 for_each_guest_mode(run_test, &p); 924 } 925 } else { 926 host_log_mode = host_log_mode_option; 927 for_each_guest_mode(run_test, &p); 928 } 929 930 return 0; 931 } 932