1 /* 2 * Dirtyrate implement code 3 * 4 * Copyright (c) 2020 HUAWEI TECHNOLOGIES CO.,LTD. 5 * 6 * Authors: 7 * Chuan Zheng <zhengchuan@huawei.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2 or later. 10 * See the COPYING file in the top-level directory. 11 */ 12 13 #include "qemu/osdep.h" 14 #include "qemu/error-report.h" 15 #include "hw/core/cpu.h" 16 #include "qapi/error.h" 17 #include "exec/ramblock.h" 18 #include "exec/target_page.h" 19 #include "qemu/rcu_queue.h" 20 #include "qemu/main-loop.h" 21 #include "qapi/qapi-commands-migration.h" 22 #include "ram.h" 23 #include "trace.h" 24 #include "dirtyrate.h" 25 #include "monitor/hmp.h" 26 #include "monitor/monitor.h" 27 #include "qapi/qmp/qdict.h" 28 #include "sysemu/kvm.h" 29 #include "sysemu/runstate.h" 30 #include "exec/memory.h" 31 #include "qemu/xxhash.h" 32 33 /* 34 * total_dirty_pages is procted by BQL and is used 35 * to stat dirty pages during the period of two 36 * memory_global_dirty_log_sync 37 */ 38 uint64_t total_dirty_pages; 39 40 typedef struct DirtyPageRecord { 41 uint64_t start_pages; 42 uint64_t end_pages; 43 } DirtyPageRecord; 44 45 static int CalculatingState = DIRTY_RATE_STATUS_UNSTARTED; 46 static struct DirtyRateStat DirtyStat; 47 static DirtyRateMeasureMode dirtyrate_mode = 48 DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; 49 50 static int64_t dirty_stat_wait(int64_t msec, int64_t initial_time) 51 { 52 int64_t current_time; 53 54 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 55 if ((current_time - initial_time) >= msec) { 56 msec = current_time - initial_time; 57 } else { 58 g_usleep((msec + initial_time - current_time) * 1000); 59 /* g_usleep may overshoot */ 60 msec = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - initial_time; 61 } 62 63 return msec; 64 } 65 66 static inline void record_dirtypages(DirtyPageRecord *dirty_pages, 67 CPUState *cpu, bool start) 68 { 69 if (start) { 70 dirty_pages[cpu->cpu_index].start_pages = cpu->dirty_pages; 71 } else { 72 dirty_pages[cpu->cpu_index].end_pages = cpu->dirty_pages; 73 } 74 } 75 76 static int64_t do_calculate_dirtyrate(DirtyPageRecord dirty_pages, 77 int64_t calc_time_ms) 78 { 79 uint64_t increased_dirty_pages = 80 dirty_pages.end_pages - dirty_pages.start_pages; 81 82 /* 83 * multiply by 1000ms/s _before_ converting down to megabytes 84 * to avoid losing precision 85 */ 86 return qemu_target_pages_to_MiB(increased_dirty_pages * 1000) / 87 calc_time_ms; 88 } 89 90 void global_dirty_log_change(unsigned int flag, bool start) 91 { 92 Error *local_err = NULL; 93 bool ret; 94 95 bql_lock(); 96 if (start) { 97 ret = memory_global_dirty_log_start(flag, &local_err); 98 if (!ret) { 99 error_report_err(local_err); 100 } 101 } else { 102 memory_global_dirty_log_stop(flag); 103 } 104 bql_unlock(); 105 } 106 107 /* 108 * global_dirty_log_sync 109 * 1. sync dirty log from kvm 110 * 2. stop dirty tracking if needed. 111 */ 112 static void global_dirty_log_sync(unsigned int flag, bool one_shot) 113 { 114 bql_lock(); 115 memory_global_dirty_log_sync(false); 116 if (one_shot) { 117 memory_global_dirty_log_stop(flag); 118 } 119 bql_unlock(); 120 } 121 122 static DirtyPageRecord *vcpu_dirty_stat_alloc(VcpuStat *stat) 123 { 124 CPUState *cpu; 125 int nvcpu = 0; 126 127 CPU_FOREACH(cpu) { 128 nvcpu++; 129 } 130 131 stat->nvcpu = nvcpu; 132 stat->rates = g_new0(DirtyRateVcpu, nvcpu); 133 134 return g_new0(DirtyPageRecord, nvcpu); 135 } 136 137 static void vcpu_dirty_stat_collect(DirtyPageRecord *records, 138 bool start) 139 { 140 CPUState *cpu; 141 142 CPU_FOREACH(cpu) { 143 record_dirtypages(records, cpu, start); 144 } 145 } 146 147 int64_t vcpu_calculate_dirtyrate(int64_t calc_time_ms, 148 VcpuStat *stat, 149 unsigned int flag, 150 bool one_shot) 151 { 152 DirtyPageRecord *records = NULL; 153 int64_t init_time_ms; 154 int64_t duration; 155 int64_t dirtyrate; 156 int i = 0; 157 unsigned int gen_id = 0; 158 159 retry: 160 init_time_ms = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 161 162 WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { 163 gen_id = cpu_list_generation_id_get(); 164 records = vcpu_dirty_stat_alloc(stat); 165 vcpu_dirty_stat_collect(records, true); 166 } 167 168 duration = dirty_stat_wait(calc_time_ms, init_time_ms); 169 170 global_dirty_log_sync(flag, one_shot); 171 172 WITH_QEMU_LOCK_GUARD(&qemu_cpu_list_lock) { 173 if (gen_id != cpu_list_generation_id_get()) { 174 g_free(records); 175 g_free(stat->rates); 176 cpu_list_unlock(); 177 goto retry; 178 } 179 vcpu_dirty_stat_collect(records, false); 180 } 181 182 for (i = 0; i < stat->nvcpu; i++) { 183 dirtyrate = do_calculate_dirtyrate(records[i], duration); 184 185 stat->rates[i].id = i; 186 stat->rates[i].dirty_rate = dirtyrate; 187 188 trace_dirtyrate_do_calculate_vcpu(i, dirtyrate); 189 } 190 191 g_free(records); 192 193 return duration; 194 } 195 196 static bool is_calc_time_valid(int64_t msec) 197 { 198 if ((msec < MIN_CALC_TIME_MS) || (msec > MAX_CALC_TIME_MS)) { 199 return false; 200 } 201 202 return true; 203 } 204 205 static bool is_sample_pages_valid(int64_t pages) 206 { 207 return pages >= MIN_SAMPLE_PAGE_COUNT && 208 pages <= MAX_SAMPLE_PAGE_COUNT; 209 } 210 211 static int dirtyrate_set_state(int *state, int old_state, int new_state) 212 { 213 assert(new_state < DIRTY_RATE_STATUS__MAX); 214 trace_dirtyrate_set_state(DirtyRateStatus_str(new_state)); 215 if (qatomic_cmpxchg(state, old_state, new_state) == old_state) { 216 return 0; 217 } else { 218 return -1; 219 } 220 } 221 222 /* Decimal power of given time unit relative to one second */ 223 static int time_unit_to_power(TimeUnit time_unit) 224 { 225 switch (time_unit) { 226 case TIME_UNIT_SECOND: 227 return 0; 228 case TIME_UNIT_MILLISECOND: 229 return -3; 230 default: 231 g_assert_not_reached(); 232 } 233 } 234 235 static int64_t convert_time_unit(int64_t value, TimeUnit unit_from, 236 TimeUnit unit_to) 237 { 238 int power = time_unit_to_power(unit_from) - 239 time_unit_to_power(unit_to); 240 while (power < 0) { 241 value /= 10; 242 power += 1; 243 } 244 while (power > 0) { 245 value *= 10; 246 power -= 1; 247 } 248 return value; 249 } 250 251 252 static struct DirtyRateInfo * 253 query_dirty_rate_info(TimeUnit calc_time_unit) 254 { 255 int i; 256 int64_t dirty_rate = DirtyStat.dirty_rate; 257 struct DirtyRateInfo *info = g_new0(DirtyRateInfo, 1); 258 DirtyRateVcpuList *head = NULL, **tail = &head; 259 260 info->status = CalculatingState; 261 info->start_time = DirtyStat.start_time; 262 info->calc_time = convert_time_unit(DirtyStat.calc_time_ms, 263 TIME_UNIT_MILLISECOND, 264 calc_time_unit); 265 info->calc_time_unit = calc_time_unit; 266 info->sample_pages = DirtyStat.sample_pages; 267 info->mode = dirtyrate_mode; 268 269 if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURED) { 270 info->has_dirty_rate = true; 271 info->dirty_rate = dirty_rate; 272 273 if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { 274 /* 275 * set sample_pages with 0 to indicate page sampling 276 * isn't enabled 277 **/ 278 info->sample_pages = 0; 279 info->has_vcpu_dirty_rate = true; 280 for (i = 0; i < DirtyStat.dirty_ring.nvcpu; i++) { 281 DirtyRateVcpu *rate = g_new0(DirtyRateVcpu, 1); 282 rate->id = DirtyStat.dirty_ring.rates[i].id; 283 rate->dirty_rate = DirtyStat.dirty_ring.rates[i].dirty_rate; 284 QAPI_LIST_APPEND(tail, rate); 285 } 286 info->vcpu_dirty_rate = head; 287 } 288 289 if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) { 290 info->sample_pages = 0; 291 } 292 } 293 294 trace_query_dirty_rate_info(DirtyRateStatus_str(CalculatingState)); 295 296 return info; 297 } 298 299 static void init_dirtyrate_stat(struct DirtyRateConfig config) 300 { 301 DirtyStat.dirty_rate = -1; 302 DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) / 1000; 303 DirtyStat.calc_time_ms = config.calc_time_ms; 304 DirtyStat.sample_pages = config.sample_pages_per_gigabytes; 305 306 switch (config.mode) { 307 case DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING: 308 DirtyStat.page_sampling.total_dirty_samples = 0; 309 DirtyStat.page_sampling.total_sample_count = 0; 310 DirtyStat.page_sampling.total_block_mem_MB = 0; 311 break; 312 case DIRTY_RATE_MEASURE_MODE_DIRTY_RING: 313 DirtyStat.dirty_ring.nvcpu = -1; 314 DirtyStat.dirty_ring.rates = NULL; 315 break; 316 default: 317 break; 318 } 319 } 320 321 static void cleanup_dirtyrate_stat(struct DirtyRateConfig config) 322 { 323 /* last calc-dirty-rate qmp use dirty ring mode */ 324 if (dirtyrate_mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { 325 free(DirtyStat.dirty_ring.rates); 326 DirtyStat.dirty_ring.rates = NULL; 327 } 328 } 329 330 static void update_dirtyrate_stat(struct RamblockDirtyInfo *info) 331 { 332 DirtyStat.page_sampling.total_dirty_samples += info->sample_dirty_count; 333 DirtyStat.page_sampling.total_sample_count += info->sample_pages_count; 334 /* size of total pages in MB */ 335 DirtyStat.page_sampling.total_block_mem_MB += 336 qemu_target_pages_to_MiB(info->ramblock_pages); 337 } 338 339 static void update_dirtyrate(uint64_t msec) 340 { 341 uint64_t dirtyrate; 342 uint64_t total_dirty_samples = DirtyStat.page_sampling.total_dirty_samples; 343 uint64_t total_sample_count = DirtyStat.page_sampling.total_sample_count; 344 uint64_t total_block_mem_MB = DirtyStat.page_sampling.total_block_mem_MB; 345 346 dirtyrate = total_dirty_samples * total_block_mem_MB * 347 1000 / (total_sample_count * msec); 348 349 DirtyStat.dirty_rate = dirtyrate; 350 } 351 352 /* 353 * Compute hash of a single page of size TARGET_PAGE_SIZE. 354 */ 355 static uint32_t compute_page_hash(void *ptr) 356 { 357 size_t page_size = qemu_target_page_size(); 358 uint32_t i; 359 uint64_t v1, v2, v3, v4; 360 uint64_t res; 361 const uint64_t *p = ptr; 362 363 v1 = QEMU_XXHASH_SEED + XXH_PRIME64_1 + XXH_PRIME64_2; 364 v2 = QEMU_XXHASH_SEED + XXH_PRIME64_2; 365 v3 = QEMU_XXHASH_SEED + 0; 366 v4 = QEMU_XXHASH_SEED - XXH_PRIME64_1; 367 for (i = 0; i < page_size / 8; i += 4) { 368 v1 = XXH64_round(v1, p[i + 0]); 369 v2 = XXH64_round(v2, p[i + 1]); 370 v3 = XXH64_round(v3, p[i + 2]); 371 v4 = XXH64_round(v4, p[i + 3]); 372 } 373 res = XXH64_mergerounds(v1, v2, v3, v4); 374 res += page_size; 375 res = XXH64_avalanche(res); 376 return (uint32_t)(res & UINT32_MAX); 377 } 378 379 380 /* 381 * get hash result for the sampled memory with length of TARGET_PAGE_SIZE 382 * in ramblock, which starts from ramblock base address. 383 */ 384 static uint32_t get_ramblock_vfn_hash(struct RamblockDirtyInfo *info, 385 uint64_t vfn) 386 { 387 uint32_t hash; 388 389 hash = compute_page_hash(info->ramblock_addr + 390 vfn * qemu_target_page_size()); 391 392 trace_get_ramblock_vfn_hash(info->idstr, vfn, hash); 393 return hash; 394 } 395 396 static bool save_ramblock_hash(struct RamblockDirtyInfo *info) 397 { 398 unsigned int sample_pages_count; 399 int i; 400 GRand *rand; 401 402 sample_pages_count = info->sample_pages_count; 403 404 /* ramblock size less than one page, return success to skip this ramblock */ 405 if (unlikely(info->ramblock_pages == 0 || sample_pages_count == 0)) { 406 return true; 407 } 408 409 info->hash_result = g_try_malloc0_n(sample_pages_count, 410 sizeof(uint32_t)); 411 if (!info->hash_result) { 412 return false; 413 } 414 415 info->sample_page_vfn = g_try_malloc0_n(sample_pages_count, 416 sizeof(uint64_t)); 417 if (!info->sample_page_vfn) { 418 g_free(info->hash_result); 419 return false; 420 } 421 422 rand = g_rand_new(); 423 for (i = 0; i < sample_pages_count; i++) { 424 info->sample_page_vfn[i] = g_rand_int_range(rand, 0, 425 info->ramblock_pages - 1); 426 info->hash_result[i] = get_ramblock_vfn_hash(info, 427 info->sample_page_vfn[i]); 428 } 429 g_rand_free(rand); 430 431 return true; 432 } 433 434 static void get_ramblock_dirty_info(RAMBlock *block, 435 struct RamblockDirtyInfo *info, 436 struct DirtyRateConfig *config) 437 { 438 uint64_t sample_pages_per_gigabytes = config->sample_pages_per_gigabytes; 439 440 /* Right shift 30 bits to calc ramblock size in GB */ 441 info->sample_pages_count = (qemu_ram_get_used_length(block) * 442 sample_pages_per_gigabytes) >> 30; 443 /* Right shift TARGET_PAGE_BITS to calc page count */ 444 info->ramblock_pages = qemu_ram_get_used_length(block) >> 445 qemu_target_page_bits(); 446 info->ramblock_addr = qemu_ram_get_host_addr(block); 447 strcpy(info->idstr, qemu_ram_get_idstr(block)); 448 } 449 450 static void free_ramblock_dirty_info(struct RamblockDirtyInfo *infos, int count) 451 { 452 int i; 453 454 if (!infos) { 455 return; 456 } 457 458 for (i = 0; i < count; i++) { 459 g_free(infos[i].sample_page_vfn); 460 g_free(infos[i].hash_result); 461 } 462 g_free(infos); 463 } 464 465 static bool skip_sample_ramblock(RAMBlock *block) 466 { 467 /* 468 * Sample only blocks larger than MIN_RAMBLOCK_SIZE. 469 */ 470 if (qemu_ram_get_used_length(block) < (MIN_RAMBLOCK_SIZE << 10)) { 471 trace_skip_sample_ramblock(block->idstr, 472 qemu_ram_get_used_length(block)); 473 return true; 474 } 475 476 return false; 477 } 478 479 static bool record_ramblock_hash_info(struct RamblockDirtyInfo **block_dinfo, 480 struct DirtyRateConfig config, 481 int *block_count) 482 { 483 struct RamblockDirtyInfo *info = NULL; 484 struct RamblockDirtyInfo *dinfo = NULL; 485 RAMBlock *block = NULL; 486 int total_count = 0; 487 int index = 0; 488 bool ret = false; 489 490 RAMBLOCK_FOREACH_MIGRATABLE(block) { 491 if (skip_sample_ramblock(block)) { 492 continue; 493 } 494 total_count++; 495 } 496 497 dinfo = g_try_malloc0_n(total_count, sizeof(struct RamblockDirtyInfo)); 498 if (dinfo == NULL) { 499 goto out; 500 } 501 502 RAMBLOCK_FOREACH_MIGRATABLE(block) { 503 if (skip_sample_ramblock(block)) { 504 continue; 505 } 506 if (index >= total_count) { 507 break; 508 } 509 info = &dinfo[index]; 510 get_ramblock_dirty_info(block, info, &config); 511 if (!save_ramblock_hash(info)) { 512 goto out; 513 } 514 index++; 515 } 516 ret = true; 517 518 out: 519 *block_count = index; 520 *block_dinfo = dinfo; 521 return ret; 522 } 523 524 static void calc_page_dirty_rate(struct RamblockDirtyInfo *info) 525 { 526 uint32_t hash; 527 int i; 528 529 for (i = 0; i < info->sample_pages_count; i++) { 530 hash = get_ramblock_vfn_hash(info, info->sample_page_vfn[i]); 531 if (hash != info->hash_result[i]) { 532 trace_calc_page_dirty_rate(info->idstr, hash, info->hash_result[i]); 533 info->sample_dirty_count++; 534 } 535 } 536 } 537 538 static struct RamblockDirtyInfo * 539 find_block_matched(RAMBlock *block, int count, 540 struct RamblockDirtyInfo *infos) 541 { 542 int i; 543 544 for (i = 0; i < count; i++) { 545 if (!strcmp(infos[i].idstr, qemu_ram_get_idstr(block))) { 546 break; 547 } 548 } 549 550 if (i == count) { 551 return NULL; 552 } 553 554 if (infos[i].ramblock_addr != qemu_ram_get_host_addr(block) || 555 infos[i].ramblock_pages != 556 (qemu_ram_get_used_length(block) >> qemu_target_page_bits())) { 557 trace_find_page_matched(block->idstr); 558 return NULL; 559 } 560 561 return &infos[i]; 562 } 563 564 static bool compare_page_hash_info(struct RamblockDirtyInfo *info, 565 int block_count) 566 { 567 struct RamblockDirtyInfo *block_dinfo = NULL; 568 RAMBlock *block = NULL; 569 570 RAMBLOCK_FOREACH_MIGRATABLE(block) { 571 if (skip_sample_ramblock(block)) { 572 continue; 573 } 574 block_dinfo = find_block_matched(block, block_count, info); 575 if (block_dinfo == NULL) { 576 continue; 577 } 578 calc_page_dirty_rate(block_dinfo); 579 update_dirtyrate_stat(block_dinfo); 580 } 581 582 if (DirtyStat.page_sampling.total_sample_count == 0) { 583 return false; 584 } 585 586 return true; 587 } 588 589 static inline void record_dirtypages_bitmap(DirtyPageRecord *dirty_pages, 590 bool start) 591 { 592 if (start) { 593 dirty_pages->start_pages = total_dirty_pages; 594 } else { 595 dirty_pages->end_pages = total_dirty_pages; 596 } 597 } 598 599 static inline void dirtyrate_manual_reset_protect(void) 600 { 601 RAMBlock *block = NULL; 602 603 WITH_RCU_READ_LOCK_GUARD() { 604 RAMBLOCK_FOREACH_MIGRATABLE(block) { 605 memory_region_clear_dirty_bitmap(block->mr, 0, 606 block->used_length); 607 } 608 } 609 } 610 611 static void calculate_dirtyrate_dirty_bitmap(struct DirtyRateConfig config) 612 { 613 int64_t start_time; 614 DirtyPageRecord dirty_pages; 615 Error *local_err = NULL; 616 617 bql_lock(); 618 if (!memory_global_dirty_log_start(GLOBAL_DIRTY_DIRTY_RATE, &local_err)) { 619 error_report_err(local_err); 620 } 621 622 /* 623 * 1'round of log sync may return all 1 bits with 624 * KVM_DIRTY_LOG_INITIALLY_SET enable 625 * skip it unconditionally and start dirty tracking 626 * from 2'round of log sync 627 */ 628 memory_global_dirty_log_sync(false); 629 630 /* 631 * reset page protect manually and unconditionally. 632 * this make sure kvm dirty log be cleared if 633 * KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE cap is enabled. 634 */ 635 dirtyrate_manual_reset_protect(); 636 bql_unlock(); 637 638 record_dirtypages_bitmap(&dirty_pages, true); 639 640 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 641 DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) / 1000; 642 643 DirtyStat.calc_time_ms = dirty_stat_wait(config.calc_time_ms, start_time); 644 645 /* 646 * do two things. 647 * 1. fetch dirty bitmap from kvm 648 * 2. stop dirty tracking 649 */ 650 global_dirty_log_sync(GLOBAL_DIRTY_DIRTY_RATE, true); 651 652 record_dirtypages_bitmap(&dirty_pages, false); 653 654 DirtyStat.dirty_rate = do_calculate_dirtyrate(dirty_pages, 655 DirtyStat.calc_time_ms); 656 } 657 658 static void calculate_dirtyrate_dirty_ring(struct DirtyRateConfig config) 659 { 660 uint64_t dirtyrate = 0; 661 uint64_t dirtyrate_sum = 0; 662 int i = 0; 663 664 /* start log sync */ 665 global_dirty_log_change(GLOBAL_DIRTY_DIRTY_RATE, true); 666 667 DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) / 1000; 668 669 /* calculate vcpu dirtyrate */ 670 DirtyStat.calc_time_ms = vcpu_calculate_dirtyrate(config.calc_time_ms, 671 &DirtyStat.dirty_ring, 672 GLOBAL_DIRTY_DIRTY_RATE, 673 true); 674 675 /* calculate vm dirtyrate */ 676 for (i = 0; i < DirtyStat.dirty_ring.nvcpu; i++) { 677 dirtyrate = DirtyStat.dirty_ring.rates[i].dirty_rate; 678 DirtyStat.dirty_ring.rates[i].dirty_rate = dirtyrate; 679 dirtyrate_sum += dirtyrate; 680 } 681 682 DirtyStat.dirty_rate = dirtyrate_sum; 683 } 684 685 static void calculate_dirtyrate_sample_vm(struct DirtyRateConfig config) 686 { 687 struct RamblockDirtyInfo *block_dinfo = NULL; 688 int block_count = 0; 689 int64_t initial_time; 690 691 rcu_read_lock(); 692 initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 693 DirtyStat.start_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) / 1000; 694 if (!record_ramblock_hash_info(&block_dinfo, config, &block_count)) { 695 goto out; 696 } 697 rcu_read_unlock(); 698 699 DirtyStat.calc_time_ms = dirty_stat_wait(config.calc_time_ms, 700 initial_time); 701 702 rcu_read_lock(); 703 if (!compare_page_hash_info(block_dinfo, block_count)) { 704 goto out; 705 } 706 707 update_dirtyrate(DirtyStat.calc_time_ms); 708 709 out: 710 rcu_read_unlock(); 711 free_ramblock_dirty_info(block_dinfo, block_count); 712 } 713 714 static void calculate_dirtyrate(struct DirtyRateConfig config) 715 { 716 if (config.mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) { 717 calculate_dirtyrate_dirty_bitmap(config); 718 } else if (config.mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) { 719 calculate_dirtyrate_dirty_ring(config); 720 } else { 721 calculate_dirtyrate_sample_vm(config); 722 } 723 724 trace_dirtyrate_calculate(DirtyStat.dirty_rate); 725 } 726 727 void *get_dirtyrate_thread(void *arg) 728 { 729 struct DirtyRateConfig config = *(struct DirtyRateConfig *)arg; 730 int ret; 731 rcu_register_thread(); 732 733 ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_UNSTARTED, 734 DIRTY_RATE_STATUS_MEASURING); 735 if (ret == -1) { 736 error_report("change dirtyrate state failed."); 737 return NULL; 738 } 739 740 calculate_dirtyrate(config); 741 742 ret = dirtyrate_set_state(&CalculatingState, DIRTY_RATE_STATUS_MEASURING, 743 DIRTY_RATE_STATUS_MEASURED); 744 if (ret == -1) { 745 error_report("change dirtyrate state failed."); 746 } 747 748 rcu_unregister_thread(); 749 return NULL; 750 } 751 752 void qmp_calc_dirty_rate(int64_t calc_time, 753 bool has_calc_time_unit, 754 TimeUnit calc_time_unit, 755 bool has_sample_pages, 756 int64_t sample_pages, 757 bool has_mode, 758 DirtyRateMeasureMode mode, 759 Error **errp) 760 { 761 static struct DirtyRateConfig config; 762 QemuThread thread; 763 int ret; 764 765 /* 766 * If the dirty rate is already being measured, don't attempt to start. 767 */ 768 if (qatomic_read(&CalculatingState) == DIRTY_RATE_STATUS_MEASURING) { 769 error_setg(errp, "the dirty rate is already being measured."); 770 return; 771 } 772 773 int64_t calc_time_ms = convert_time_unit( 774 calc_time, 775 has_calc_time_unit ? calc_time_unit : TIME_UNIT_SECOND, 776 TIME_UNIT_MILLISECOND 777 ); 778 779 if (!is_calc_time_valid(calc_time_ms)) { 780 error_setg(errp, "Calculation time is out of range [%dms, %dms].", 781 MIN_CALC_TIME_MS, MAX_CALC_TIME_MS); 782 return; 783 } 784 785 if (!has_mode) { 786 mode = DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; 787 } 788 789 if (has_sample_pages && mode != DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING) { 790 error_setg(errp, "sample-pages is used only in page-sampling mode"); 791 return; 792 } 793 794 if (has_sample_pages) { 795 if (!is_sample_pages_valid(sample_pages)) { 796 error_setg(errp, "sample-pages is out of range[%d, %d].", 797 MIN_SAMPLE_PAGE_COUNT, 798 MAX_SAMPLE_PAGE_COUNT); 799 return; 800 } 801 } else { 802 sample_pages = DIRTYRATE_DEFAULT_SAMPLE_PAGES; 803 } 804 805 /* 806 * dirty ring mode only works when kvm dirty ring is enabled. 807 * on the contrary, dirty bitmap mode is not. 808 */ 809 if (((mode == DIRTY_RATE_MEASURE_MODE_DIRTY_RING) && 810 !kvm_dirty_ring_enabled()) || 811 ((mode == DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP) && 812 kvm_dirty_ring_enabled())) { 813 error_setg(errp, "mode %s is not enabled, use other method instead.", 814 DirtyRateMeasureMode_str(mode)); 815 return; 816 } 817 818 /* 819 * Init calculation state as unstarted. 820 */ 821 ret = dirtyrate_set_state(&CalculatingState, CalculatingState, 822 DIRTY_RATE_STATUS_UNSTARTED); 823 if (ret == -1) { 824 error_setg(errp, "init dirty rate calculation state failed."); 825 return; 826 } 827 828 config.calc_time_ms = calc_time_ms; 829 config.sample_pages_per_gigabytes = sample_pages; 830 config.mode = mode; 831 832 cleanup_dirtyrate_stat(config); 833 834 /* 835 * update dirty rate mode so that we can figure out what mode has 836 * been used in last calculation 837 **/ 838 dirtyrate_mode = mode; 839 840 init_dirtyrate_stat(config); 841 842 qemu_thread_create(&thread, "get_dirtyrate", get_dirtyrate_thread, 843 (void *)&config, QEMU_THREAD_DETACHED); 844 } 845 846 847 struct DirtyRateInfo *qmp_query_dirty_rate(bool has_calc_time_unit, 848 TimeUnit calc_time_unit, 849 Error **errp) 850 { 851 return query_dirty_rate_info( 852 has_calc_time_unit ? calc_time_unit : TIME_UNIT_SECOND); 853 } 854 855 void hmp_info_dirty_rate(Monitor *mon, const QDict *qdict) 856 { 857 DirtyRateInfo *info = query_dirty_rate_info(TIME_UNIT_SECOND); 858 859 monitor_printf(mon, "Status: %s\n", 860 DirtyRateStatus_str(info->status)); 861 monitor_printf(mon, "Start Time: %"PRIi64" (ms)\n", 862 info->start_time); 863 if (info->mode == DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING) { 864 monitor_printf(mon, "Sample Pages: %"PRIu64" (per GB)\n", 865 info->sample_pages); 866 } 867 monitor_printf(mon, "Period: %"PRIi64" (sec)\n", 868 info->calc_time); 869 monitor_printf(mon, "Mode: %s\n", 870 DirtyRateMeasureMode_str(info->mode)); 871 monitor_printf(mon, "Dirty rate: "); 872 if (info->has_dirty_rate) { 873 monitor_printf(mon, "%"PRIi64" (MB/s)\n", info->dirty_rate); 874 if (info->has_vcpu_dirty_rate) { 875 DirtyRateVcpuList *rate, *head = info->vcpu_dirty_rate; 876 for (rate = head; rate != NULL; rate = rate->next) { 877 monitor_printf(mon, "vcpu[%"PRIi64"], Dirty rate: %"PRIi64 878 " (MB/s)\n", rate->value->id, 879 rate->value->dirty_rate); 880 } 881 } 882 } else { 883 monitor_printf(mon, "(not ready)\n"); 884 } 885 886 qapi_free_DirtyRateVcpuList(info->vcpu_dirty_rate); 887 g_free(info); 888 } 889 890 void hmp_calc_dirty_rate(Monitor *mon, const QDict *qdict) 891 { 892 int64_t sec = qdict_get_try_int(qdict, "second", 0); 893 int64_t sample_pages = qdict_get_try_int(qdict, "sample_pages_per_GB", -1); 894 bool has_sample_pages = (sample_pages != -1); 895 bool dirty_ring = qdict_get_try_bool(qdict, "dirty_ring", false); 896 bool dirty_bitmap = qdict_get_try_bool(qdict, "dirty_bitmap", false); 897 DirtyRateMeasureMode mode = DIRTY_RATE_MEASURE_MODE_PAGE_SAMPLING; 898 Error *err = NULL; 899 900 if (!sec) { 901 monitor_printf(mon, "Incorrect period length specified!\n"); 902 return; 903 } 904 905 if (dirty_ring && dirty_bitmap) { 906 monitor_printf(mon, "Either dirty ring or dirty bitmap " 907 "can be specified!\n"); 908 return; 909 } 910 911 if (dirty_bitmap) { 912 mode = DIRTY_RATE_MEASURE_MODE_DIRTY_BITMAP; 913 } else if (dirty_ring) { 914 mode = DIRTY_RATE_MEASURE_MODE_DIRTY_RING; 915 } 916 917 qmp_calc_dirty_rate(sec, /* calc-time */ 918 false, TIME_UNIT_SECOND, /* calc-time-unit */ 919 has_sample_pages, sample_pages, 920 true, mode, 921 &err); 922 if (err) { 923 hmp_handle_error(mon, err); 924 return; 925 } 926 927 monitor_printf(mon, "Starting dirty rate measurement with period %"PRIi64 928 " seconds\n", sec); 929 monitor_printf(mon, "[Please use 'info dirty_rate' to check results]\n"); 930 } 931