1 /* 2 * QEMU System Emulator 3 * 4 * Copyright (c) 2003-2008 Fabrice Bellard 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to deal 8 * in the Software without restriction, including without limitation the rights 9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10 * copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22 * THE SOFTWARE. 23 */ 24 #include <stdint.h> 25 #include <stdarg.h> 26 #include <stdlib.h> 27 #ifndef _WIN32 28 #include <sys/types.h> 29 #include <sys/mman.h> 30 #endif 31 #include "config.h" 32 #include "monitor/monitor.h" 33 #include "sysemu/sysemu.h" 34 #include "qemu/bitops.h" 35 #include "qemu/bitmap.h" 36 #include "sysemu/arch_init.h" 37 #include "audio/audio.h" 38 #include "hw/i386/pc.h" 39 #include "hw/pci/pci.h" 40 #include "hw/audio/audio.h" 41 #include "sysemu/kvm.h" 42 #include "migration/migration.h" 43 #include "hw/i386/smbios.h" 44 #include "exec/address-spaces.h" 45 #include "hw/audio/pcspk.h" 46 #include "migration/page_cache.h" 47 #include "qemu/config-file.h" 48 #include "qemu/error-report.h" 49 #include "qmp-commands.h" 50 #include "trace.h" 51 #include "exec/cpu-all.h" 52 #include "exec/ram_addr.h" 53 #include "hw/acpi/acpi.h" 54 #include "qemu/host-utils.h" 55 #include "qemu/rcu_queue.h" 56 57 #ifdef DEBUG_ARCH_INIT 58 #define DPRINTF(fmt, ...) \ 59 do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0) 60 #else 61 #define DPRINTF(fmt, ...) \ 62 do { } while (0) 63 #endif 64 65 #ifdef TARGET_SPARC 66 int graphic_width = 1024; 67 int graphic_height = 768; 68 int graphic_depth = 8; 69 #else 70 int graphic_width = 800; 71 int graphic_height = 600; 72 int graphic_depth = 32; 73 #endif 74 75 76 #if defined(TARGET_ALPHA) 77 #define QEMU_ARCH QEMU_ARCH_ALPHA 78 #elif defined(TARGET_ARM) 79 #define QEMU_ARCH QEMU_ARCH_ARM 80 #elif defined(TARGET_CRIS) 81 #define QEMU_ARCH QEMU_ARCH_CRIS 82 #elif defined(TARGET_I386) 83 #define QEMU_ARCH QEMU_ARCH_I386 84 #elif defined(TARGET_M68K) 85 #define QEMU_ARCH QEMU_ARCH_M68K 86 #elif defined(TARGET_LM32) 87 #define QEMU_ARCH QEMU_ARCH_LM32 88 #elif defined(TARGET_MICROBLAZE) 89 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE 90 #elif defined(TARGET_MIPS) 91 #define QEMU_ARCH QEMU_ARCH_MIPS 92 #elif defined(TARGET_MOXIE) 93 #define QEMU_ARCH QEMU_ARCH_MOXIE 94 #elif defined(TARGET_OPENRISC) 95 #define QEMU_ARCH QEMU_ARCH_OPENRISC 96 #elif defined(TARGET_PPC) 97 #define QEMU_ARCH QEMU_ARCH_PPC 98 #elif defined(TARGET_S390X) 99 #define QEMU_ARCH QEMU_ARCH_S390X 100 #elif defined(TARGET_SH4) 101 #define QEMU_ARCH QEMU_ARCH_SH4 102 #elif defined(TARGET_SPARC) 103 #define QEMU_ARCH QEMU_ARCH_SPARC 104 #elif defined(TARGET_XTENSA) 105 #define QEMU_ARCH QEMU_ARCH_XTENSA 106 #elif defined(TARGET_UNICORE32) 107 #define QEMU_ARCH QEMU_ARCH_UNICORE32 108 #elif defined(TARGET_TRICORE) 109 #define QEMU_ARCH QEMU_ARCH_TRICORE 110 #endif 111 112 const uint32_t arch_type = QEMU_ARCH; 113 static bool mig_throttle_on; 114 static int dirty_rate_high_cnt; 115 static void check_guest_throttling(void); 116 117 static uint64_t bitmap_sync_count; 118 119 /***********************************************************/ 120 /* ram save/restore */ 121 122 #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ 123 #define RAM_SAVE_FLAG_COMPRESS 0x02 124 #define RAM_SAVE_FLAG_MEM_SIZE 0x04 125 #define RAM_SAVE_FLAG_PAGE 0x08 126 #define RAM_SAVE_FLAG_EOS 0x10 127 #define RAM_SAVE_FLAG_CONTINUE 0x20 128 #define RAM_SAVE_FLAG_XBZRLE 0x40 129 /* 0x80 is reserved in migration.h start with 0x100 next */ 130 131 static struct defconfig_file { 132 const char *filename; 133 /* Indicates it is an user config file (disabled by -no-user-config) */ 134 bool userconfig; 135 } default_config_files[] = { 136 { CONFIG_QEMU_CONFDIR "/qemu.conf", true }, 137 { CONFIG_QEMU_CONFDIR "/target-" TARGET_NAME ".conf", true }, 138 { NULL }, /* end of list */ 139 }; 140 141 static const uint8_t ZERO_TARGET_PAGE[TARGET_PAGE_SIZE]; 142 143 int qemu_read_default_config_files(bool userconfig) 144 { 145 int ret; 146 struct defconfig_file *f; 147 148 for (f = default_config_files; f->filename; f++) { 149 if (!userconfig && f->userconfig) { 150 continue; 151 } 152 ret = qemu_read_config_file(f->filename); 153 if (ret < 0 && ret != -ENOENT) { 154 return ret; 155 } 156 } 157 158 return 0; 159 } 160 161 static inline bool is_zero_range(uint8_t *p, uint64_t size) 162 { 163 return buffer_find_nonzero_offset(p, size) == size; 164 } 165 166 /* struct contains XBZRLE cache and a static page 167 used by the compression */ 168 static struct { 169 /* buffer used for XBZRLE encoding */ 170 uint8_t *encoded_buf; 171 /* buffer for storing page content */ 172 uint8_t *current_buf; 173 /* Cache for XBZRLE, Protected by lock. */ 174 PageCache *cache; 175 QemuMutex lock; 176 } XBZRLE; 177 178 /* buffer used for XBZRLE decoding */ 179 static uint8_t *xbzrle_decoded_buf; 180 181 static void XBZRLE_cache_lock(void) 182 { 183 if (migrate_use_xbzrle()) 184 qemu_mutex_lock(&XBZRLE.lock); 185 } 186 187 static void XBZRLE_cache_unlock(void) 188 { 189 if (migrate_use_xbzrle()) 190 qemu_mutex_unlock(&XBZRLE.lock); 191 } 192 193 /* 194 * called from qmp_migrate_set_cache_size in main thread, possibly while 195 * a migration is in progress. 196 * A running migration maybe using the cache and might finish during this 197 * call, hence changes to the cache are protected by XBZRLE.lock(). 198 */ 199 int64_t xbzrle_cache_resize(int64_t new_size) 200 { 201 PageCache *new_cache; 202 int64_t ret; 203 204 if (new_size < TARGET_PAGE_SIZE) { 205 return -1; 206 } 207 208 XBZRLE_cache_lock(); 209 210 if (XBZRLE.cache != NULL) { 211 if (pow2floor(new_size) == migrate_xbzrle_cache_size()) { 212 goto out_new_size; 213 } 214 new_cache = cache_init(new_size / TARGET_PAGE_SIZE, 215 TARGET_PAGE_SIZE); 216 if (!new_cache) { 217 error_report("Error creating cache"); 218 ret = -1; 219 goto out; 220 } 221 222 cache_fini(XBZRLE.cache); 223 XBZRLE.cache = new_cache; 224 } 225 226 out_new_size: 227 ret = pow2floor(new_size); 228 out: 229 XBZRLE_cache_unlock(); 230 return ret; 231 } 232 233 /* accounting for migration statistics */ 234 typedef struct AccountingInfo { 235 uint64_t dup_pages; 236 uint64_t skipped_pages; 237 uint64_t norm_pages; 238 uint64_t iterations; 239 uint64_t xbzrle_bytes; 240 uint64_t xbzrle_pages; 241 uint64_t xbzrle_cache_miss; 242 double xbzrle_cache_miss_rate; 243 uint64_t xbzrle_overflows; 244 } AccountingInfo; 245 246 static AccountingInfo acct_info; 247 248 static void acct_clear(void) 249 { 250 memset(&acct_info, 0, sizeof(acct_info)); 251 } 252 253 uint64_t dup_mig_bytes_transferred(void) 254 { 255 return acct_info.dup_pages * TARGET_PAGE_SIZE; 256 } 257 258 uint64_t dup_mig_pages_transferred(void) 259 { 260 return acct_info.dup_pages; 261 } 262 263 uint64_t skipped_mig_bytes_transferred(void) 264 { 265 return acct_info.skipped_pages * TARGET_PAGE_SIZE; 266 } 267 268 uint64_t skipped_mig_pages_transferred(void) 269 { 270 return acct_info.skipped_pages; 271 } 272 273 uint64_t norm_mig_bytes_transferred(void) 274 { 275 return acct_info.norm_pages * TARGET_PAGE_SIZE; 276 } 277 278 uint64_t norm_mig_pages_transferred(void) 279 { 280 return acct_info.norm_pages; 281 } 282 283 uint64_t xbzrle_mig_bytes_transferred(void) 284 { 285 return acct_info.xbzrle_bytes; 286 } 287 288 uint64_t xbzrle_mig_pages_transferred(void) 289 { 290 return acct_info.xbzrle_pages; 291 } 292 293 uint64_t xbzrle_mig_pages_cache_miss(void) 294 { 295 return acct_info.xbzrle_cache_miss; 296 } 297 298 double xbzrle_mig_cache_miss_rate(void) 299 { 300 return acct_info.xbzrle_cache_miss_rate; 301 } 302 303 uint64_t xbzrle_mig_pages_overflow(void) 304 { 305 return acct_info.xbzrle_overflows; 306 } 307 308 static size_t save_block_hdr(QEMUFile *f, RAMBlock *block, ram_addr_t offset, 309 int cont, int flag) 310 { 311 size_t size; 312 313 qemu_put_be64(f, offset | cont | flag); 314 size = 8; 315 316 if (!cont) { 317 qemu_put_byte(f, strlen(block->idstr)); 318 qemu_put_buffer(f, (uint8_t *)block->idstr, 319 strlen(block->idstr)); 320 size += 1 + strlen(block->idstr); 321 } 322 return size; 323 } 324 325 /* This is the last block that we have visited serching for dirty pages 326 */ 327 static RAMBlock *last_seen_block; 328 /* This is the last block from where we have sent data */ 329 static RAMBlock *last_sent_block; 330 static ram_addr_t last_offset; 331 static unsigned long *migration_bitmap; 332 static uint64_t migration_dirty_pages; 333 static uint32_t last_version; 334 static bool ram_bulk_stage; 335 336 /* Update the xbzrle cache to reflect a page that's been sent as all 0. 337 * The important thing is that a stale (not-yet-0'd) page be replaced 338 * by the new data. 339 * As a bonus, if the page wasn't in the cache it gets added so that 340 * when a small write is made into the 0'd page it gets XBZRLE sent 341 */ 342 static void xbzrle_cache_zero_page(ram_addr_t current_addr) 343 { 344 if (ram_bulk_stage || !migrate_use_xbzrle()) { 345 return; 346 } 347 348 /* We don't care if this fails to allocate a new cache page 349 * as long as it updated an old one */ 350 cache_insert(XBZRLE.cache, current_addr, ZERO_TARGET_PAGE, 351 bitmap_sync_count); 352 } 353 354 #define ENCODING_FLAG_XBZRLE 0x1 355 356 static int save_xbzrle_page(QEMUFile *f, uint8_t **current_data, 357 ram_addr_t current_addr, RAMBlock *block, 358 ram_addr_t offset, int cont, bool last_stage) 359 { 360 int encoded_len = 0, bytes_sent = -1; 361 uint8_t *prev_cached_page; 362 363 if (!cache_is_cached(XBZRLE.cache, current_addr, bitmap_sync_count)) { 364 acct_info.xbzrle_cache_miss++; 365 if (!last_stage) { 366 if (cache_insert(XBZRLE.cache, current_addr, *current_data, 367 bitmap_sync_count) == -1) { 368 return -1; 369 } else { 370 /* update *current_data when the page has been 371 inserted into cache */ 372 *current_data = get_cached_data(XBZRLE.cache, current_addr); 373 } 374 } 375 return -1; 376 } 377 378 prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); 379 380 /* save current buffer into memory */ 381 memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); 382 383 /* XBZRLE encoding (if there is no overflow) */ 384 encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, 385 TARGET_PAGE_SIZE, XBZRLE.encoded_buf, 386 TARGET_PAGE_SIZE); 387 if (encoded_len == 0) { 388 DPRINTF("Skipping unmodified page\n"); 389 return 0; 390 } else if (encoded_len == -1) { 391 DPRINTF("Overflow\n"); 392 acct_info.xbzrle_overflows++; 393 /* update data in the cache */ 394 if (!last_stage) { 395 memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE); 396 *current_data = prev_cached_page; 397 } 398 return -1; 399 } 400 401 /* we need to update the data in the cache, in order to get the same data */ 402 if (!last_stage) { 403 memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); 404 } 405 406 /* Send XBZRLE based compressed page */ 407 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_XBZRLE); 408 qemu_put_byte(f, ENCODING_FLAG_XBZRLE); 409 qemu_put_be16(f, encoded_len); 410 qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len); 411 bytes_sent += encoded_len + 1 + 2; 412 acct_info.xbzrle_pages++; 413 acct_info.xbzrle_bytes += bytes_sent; 414 415 return bytes_sent; 416 } 417 418 static inline 419 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr, 420 ram_addr_t start) 421 { 422 unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS; 423 unsigned long nr = base + (start >> TARGET_PAGE_BITS); 424 uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr)); 425 unsigned long size = base + (mr_size >> TARGET_PAGE_BITS); 426 427 unsigned long next; 428 429 if (ram_bulk_stage && nr > base) { 430 next = nr + 1; 431 } else { 432 next = find_next_bit(migration_bitmap, size, nr); 433 } 434 435 if (next < size) { 436 clear_bit(next, migration_bitmap); 437 migration_dirty_pages--; 438 } 439 return (next - base) << TARGET_PAGE_BITS; 440 } 441 442 static inline bool migration_bitmap_set_dirty(ram_addr_t addr) 443 { 444 bool ret; 445 int nr = addr >> TARGET_PAGE_BITS; 446 447 ret = test_and_set_bit(nr, migration_bitmap); 448 449 if (!ret) { 450 migration_dirty_pages++; 451 } 452 return ret; 453 } 454 455 static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length) 456 { 457 ram_addr_t addr; 458 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS); 459 460 /* start address is aligned at the start of a word? */ 461 if (((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) { 462 int k; 463 int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS); 464 unsigned long *src = ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION]; 465 466 for (k = page; k < page + nr; k++) { 467 if (src[k]) { 468 unsigned long new_dirty; 469 new_dirty = ~migration_bitmap[k]; 470 migration_bitmap[k] |= src[k]; 471 new_dirty &= src[k]; 472 migration_dirty_pages += ctpopl(new_dirty); 473 src[k] = 0; 474 } 475 } 476 } else { 477 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) { 478 if (cpu_physical_memory_get_dirty(start + addr, 479 TARGET_PAGE_SIZE, 480 DIRTY_MEMORY_MIGRATION)) { 481 cpu_physical_memory_reset_dirty(start + addr, 482 TARGET_PAGE_SIZE, 483 DIRTY_MEMORY_MIGRATION); 484 migration_bitmap_set_dirty(start + addr); 485 } 486 } 487 } 488 } 489 490 491 /* Fix me: there are too many global variables used in migration process. */ 492 static int64_t start_time; 493 static int64_t bytes_xfer_prev; 494 static int64_t num_dirty_pages_period; 495 496 static void migration_bitmap_sync_init(void) 497 { 498 start_time = 0; 499 bytes_xfer_prev = 0; 500 num_dirty_pages_period = 0; 501 } 502 503 /* Called with iothread lock held, to protect ram_list.dirty_memory[] */ 504 static void migration_bitmap_sync(void) 505 { 506 RAMBlock *block; 507 uint64_t num_dirty_pages_init = migration_dirty_pages; 508 MigrationState *s = migrate_get_current(); 509 int64_t end_time; 510 int64_t bytes_xfer_now; 511 static uint64_t xbzrle_cache_miss_prev; 512 static uint64_t iterations_prev; 513 514 bitmap_sync_count++; 515 516 if (!bytes_xfer_prev) { 517 bytes_xfer_prev = ram_bytes_transferred(); 518 } 519 520 if (!start_time) { 521 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 522 } 523 524 trace_migration_bitmap_sync_start(); 525 address_space_sync_dirty_bitmap(&address_space_memory); 526 527 rcu_read_lock(); 528 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { 529 migration_bitmap_sync_range(block->mr->ram_addr, block->used_length); 530 } 531 rcu_read_unlock(); 532 533 trace_migration_bitmap_sync_end(migration_dirty_pages 534 - num_dirty_pages_init); 535 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init; 536 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 537 538 /* more than 1 second = 1000 millisecons */ 539 if (end_time > start_time + 1000) { 540 if (migrate_auto_converge()) { 541 /* The following detection logic can be refined later. For now: 542 Check to see if the dirtied bytes is 50% more than the approx. 543 amount of bytes that just got transferred since the last time we 544 were in this routine. If that happens >N times (for now N==4) 545 we turn on the throttle down logic */ 546 bytes_xfer_now = ram_bytes_transferred(); 547 if (s->dirty_pages_rate && 548 (num_dirty_pages_period * TARGET_PAGE_SIZE > 549 (bytes_xfer_now - bytes_xfer_prev)/2) && 550 (dirty_rate_high_cnt++ > 4)) { 551 trace_migration_throttle(); 552 mig_throttle_on = true; 553 dirty_rate_high_cnt = 0; 554 } 555 bytes_xfer_prev = bytes_xfer_now; 556 } else { 557 mig_throttle_on = false; 558 } 559 if (migrate_use_xbzrle()) { 560 if (iterations_prev != 0) { 561 acct_info.xbzrle_cache_miss_rate = 562 (double)(acct_info.xbzrle_cache_miss - 563 xbzrle_cache_miss_prev) / 564 (acct_info.iterations - iterations_prev); 565 } 566 iterations_prev = acct_info.iterations; 567 xbzrle_cache_miss_prev = acct_info.xbzrle_cache_miss; 568 } 569 s->dirty_pages_rate = num_dirty_pages_period * 1000 570 / (end_time - start_time); 571 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE; 572 start_time = end_time; 573 num_dirty_pages_period = 0; 574 s->dirty_sync_count = bitmap_sync_count; 575 } 576 } 577 578 /* 579 * ram_save_page: Send the given page to the stream 580 * 581 * Returns: Number of bytes written. 582 */ 583 static int ram_save_page(QEMUFile *f, RAMBlock* block, ram_addr_t offset, 584 bool last_stage) 585 { 586 int bytes_sent; 587 uint64_t bytes_xmit; 588 int cont; 589 ram_addr_t current_addr; 590 MemoryRegion *mr = block->mr; 591 uint8_t *p; 592 int ret; 593 bool send_async = true; 594 595 cont = (block == last_sent_block) ? RAM_SAVE_FLAG_CONTINUE : 0; 596 597 p = memory_region_get_ram_ptr(mr) + offset; 598 599 /* In doubt sent page as normal */ 600 bytes_sent = -1; 601 bytes_xmit = 0; 602 ret = ram_control_save_page(f, block->offset, 603 offset, TARGET_PAGE_SIZE, &bytes_xmit); 604 if (bytes_xmit) { 605 bytes_sent = bytes_xmit; 606 } 607 608 XBZRLE_cache_lock(); 609 610 current_addr = block->offset + offset; 611 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { 612 if (ret != RAM_SAVE_CONTROL_DELAYED) { 613 if (bytes_xmit > 0) { 614 acct_info.norm_pages++; 615 } else if (bytes_xmit == 0) { 616 acct_info.dup_pages++; 617 } 618 } 619 } else if (is_zero_range(p, TARGET_PAGE_SIZE)) { 620 acct_info.dup_pages++; 621 bytes_sent = save_block_hdr(f, block, offset, cont, 622 RAM_SAVE_FLAG_COMPRESS); 623 qemu_put_byte(f, 0); 624 bytes_sent++; 625 /* Must let xbzrle know, otherwise a previous (now 0'd) cached 626 * page would be stale 627 */ 628 xbzrle_cache_zero_page(current_addr); 629 } else if (!ram_bulk_stage && migrate_use_xbzrle()) { 630 bytes_sent = save_xbzrle_page(f, &p, current_addr, block, 631 offset, cont, last_stage); 632 if (!last_stage) { 633 /* Can't send this cached data async, since the cache page 634 * might get updated before it gets to the wire 635 */ 636 send_async = false; 637 } 638 } 639 640 /* XBZRLE overflow or normal page */ 641 if (bytes_sent == -1) { 642 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_PAGE); 643 if (send_async) { 644 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE); 645 } else { 646 qemu_put_buffer(f, p, TARGET_PAGE_SIZE); 647 } 648 bytes_sent += TARGET_PAGE_SIZE; 649 acct_info.norm_pages++; 650 } 651 652 XBZRLE_cache_unlock(); 653 654 return bytes_sent; 655 } 656 657 /* 658 * ram_find_and_save_block: Finds a page to send and sends it to f 659 * 660 * Called within an RCU critical section. 661 * 662 * Returns: The number of bytes written. 663 * 0 means no dirty pages 664 */ 665 666 static int ram_find_and_save_block(QEMUFile *f, bool last_stage) 667 { 668 RAMBlock *block = last_seen_block; 669 ram_addr_t offset = last_offset; 670 bool complete_round = false; 671 int bytes_sent = 0; 672 MemoryRegion *mr; 673 674 if (!block) 675 block = QLIST_FIRST_RCU(&ram_list.blocks); 676 677 while (true) { 678 mr = block->mr; 679 offset = migration_bitmap_find_and_reset_dirty(mr, offset); 680 if (complete_round && block == last_seen_block && 681 offset >= last_offset) { 682 break; 683 } 684 if (offset >= block->used_length) { 685 offset = 0; 686 block = QLIST_NEXT_RCU(block, next); 687 if (!block) { 688 block = QLIST_FIRST_RCU(&ram_list.blocks); 689 complete_round = true; 690 ram_bulk_stage = false; 691 } 692 } else { 693 bytes_sent = ram_save_page(f, block, offset, last_stage); 694 695 /* if page is unmodified, continue to the next */ 696 if (bytes_sent > 0) { 697 last_sent_block = block; 698 break; 699 } 700 } 701 } 702 703 last_seen_block = block; 704 last_offset = offset; 705 return bytes_sent; 706 } 707 708 static uint64_t bytes_transferred; 709 710 void acct_update_position(QEMUFile *f, size_t size, bool zero) 711 { 712 uint64_t pages = size / TARGET_PAGE_SIZE; 713 if (zero) { 714 acct_info.dup_pages += pages; 715 } else { 716 acct_info.norm_pages += pages; 717 bytes_transferred += size; 718 qemu_update_position(f, size); 719 } 720 } 721 722 static ram_addr_t ram_save_remaining(void) 723 { 724 return migration_dirty_pages; 725 } 726 727 uint64_t ram_bytes_remaining(void) 728 { 729 return ram_save_remaining() * TARGET_PAGE_SIZE; 730 } 731 732 uint64_t ram_bytes_transferred(void) 733 { 734 return bytes_transferred; 735 } 736 737 uint64_t ram_bytes_total(void) 738 { 739 RAMBlock *block; 740 uint64_t total = 0; 741 742 rcu_read_lock(); 743 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) 744 total += block->used_length; 745 rcu_read_unlock(); 746 return total; 747 } 748 749 void free_xbzrle_decoded_buf(void) 750 { 751 g_free(xbzrle_decoded_buf); 752 xbzrle_decoded_buf = NULL; 753 } 754 755 static void migration_end(void) 756 { 757 if (migration_bitmap) { 758 memory_global_dirty_log_stop(); 759 g_free(migration_bitmap); 760 migration_bitmap = NULL; 761 } 762 763 XBZRLE_cache_lock(); 764 if (XBZRLE.cache) { 765 cache_fini(XBZRLE.cache); 766 g_free(XBZRLE.encoded_buf); 767 g_free(XBZRLE.current_buf); 768 XBZRLE.cache = NULL; 769 XBZRLE.encoded_buf = NULL; 770 XBZRLE.current_buf = NULL; 771 } 772 XBZRLE_cache_unlock(); 773 } 774 775 static void ram_migration_cancel(void *opaque) 776 { 777 migration_end(); 778 } 779 780 static void reset_ram_globals(void) 781 { 782 last_seen_block = NULL; 783 last_sent_block = NULL; 784 last_offset = 0; 785 last_version = ram_list.version; 786 ram_bulk_stage = true; 787 } 788 789 #define MAX_WAIT 50 /* ms, half buffered_file limit */ 790 791 792 /* Each of ram_save_setup, ram_save_iterate and ram_save_complete has 793 * long-running RCU critical section. When rcu-reclaims in the code 794 * start to become numerous it will be necessary to reduce the 795 * granularity of these critical sections. 796 */ 797 798 static int ram_save_setup(QEMUFile *f, void *opaque) 799 { 800 RAMBlock *block; 801 int64_t ram_bitmap_pages; /* Size of bitmap in pages, including gaps */ 802 803 mig_throttle_on = false; 804 dirty_rate_high_cnt = 0; 805 bitmap_sync_count = 0; 806 migration_bitmap_sync_init(); 807 808 if (migrate_use_xbzrle()) { 809 XBZRLE_cache_lock(); 810 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() / 811 TARGET_PAGE_SIZE, 812 TARGET_PAGE_SIZE); 813 if (!XBZRLE.cache) { 814 XBZRLE_cache_unlock(); 815 error_report("Error creating cache"); 816 return -1; 817 } 818 XBZRLE_cache_unlock(); 819 820 /* We prefer not to abort if there is no memory */ 821 XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); 822 if (!XBZRLE.encoded_buf) { 823 error_report("Error allocating encoded_buf"); 824 return -1; 825 } 826 827 XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); 828 if (!XBZRLE.current_buf) { 829 error_report("Error allocating current_buf"); 830 g_free(XBZRLE.encoded_buf); 831 XBZRLE.encoded_buf = NULL; 832 return -1; 833 } 834 835 acct_clear(); 836 } 837 838 /* iothread lock needed for ram_list.dirty_memory[] */ 839 qemu_mutex_lock_iothread(); 840 qemu_mutex_lock_ramlist(); 841 rcu_read_lock(); 842 bytes_transferred = 0; 843 reset_ram_globals(); 844 845 ram_bitmap_pages = last_ram_offset() >> TARGET_PAGE_BITS; 846 migration_bitmap = bitmap_new(ram_bitmap_pages); 847 bitmap_set(migration_bitmap, 0, ram_bitmap_pages); 848 849 /* 850 * Count the total number of pages used by ram blocks not including any 851 * gaps due to alignment or unplugs. 852 */ 853 migration_dirty_pages = 0; 854 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { 855 uint64_t block_pages; 856 857 block_pages = block->used_length >> TARGET_PAGE_BITS; 858 migration_dirty_pages += block_pages; 859 } 860 861 memory_global_dirty_log_start(); 862 migration_bitmap_sync(); 863 qemu_mutex_unlock_ramlist(); 864 qemu_mutex_unlock_iothread(); 865 866 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); 867 868 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { 869 qemu_put_byte(f, strlen(block->idstr)); 870 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); 871 qemu_put_be64(f, block->used_length); 872 } 873 874 rcu_read_unlock(); 875 876 ram_control_before_iterate(f, RAM_CONTROL_SETUP); 877 ram_control_after_iterate(f, RAM_CONTROL_SETUP); 878 879 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 880 881 return 0; 882 } 883 884 static int ram_save_iterate(QEMUFile *f, void *opaque) 885 { 886 int ret; 887 int i; 888 int64_t t0; 889 int total_sent = 0; 890 891 rcu_read_lock(); 892 if (ram_list.version != last_version) { 893 reset_ram_globals(); 894 } 895 896 /* Read version before ram_list.blocks */ 897 smp_rmb(); 898 899 ram_control_before_iterate(f, RAM_CONTROL_ROUND); 900 901 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 902 i = 0; 903 while ((ret = qemu_file_rate_limit(f)) == 0) { 904 int bytes_sent; 905 906 bytes_sent = ram_find_and_save_block(f, false); 907 /* no more blocks to sent */ 908 if (bytes_sent == 0) { 909 break; 910 } 911 total_sent += bytes_sent; 912 acct_info.iterations++; 913 check_guest_throttling(); 914 /* we want to check in the 1st loop, just in case it was the 1st time 915 and we had to sync the dirty bitmap. 916 qemu_get_clock_ns() is a bit expensive, so we only check each some 917 iterations 918 */ 919 if ((i & 63) == 0) { 920 uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; 921 if (t1 > MAX_WAIT) { 922 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n", 923 t1, i); 924 break; 925 } 926 } 927 i++; 928 } 929 rcu_read_unlock(); 930 931 /* 932 * Must occur before EOS (or any QEMUFile operation) 933 * because of RDMA protocol. 934 */ 935 ram_control_after_iterate(f, RAM_CONTROL_ROUND); 936 937 bytes_transferred += total_sent; 938 939 /* 940 * Do not count these 8 bytes into total_sent, so that we can 941 * return 0 if no page had been dirtied. 942 */ 943 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 944 bytes_transferred += 8; 945 946 ret = qemu_file_get_error(f); 947 if (ret < 0) { 948 return ret; 949 } 950 951 return total_sent; 952 } 953 954 /* Called with iothread lock */ 955 static int ram_save_complete(QEMUFile *f, void *opaque) 956 { 957 rcu_read_lock(); 958 959 migration_bitmap_sync(); 960 961 ram_control_before_iterate(f, RAM_CONTROL_FINISH); 962 963 /* try transferring iterative blocks of memory */ 964 965 /* flush all remaining blocks regardless of rate limiting */ 966 while (true) { 967 int bytes_sent; 968 969 bytes_sent = ram_find_and_save_block(f, true); 970 /* no more blocks to sent */ 971 if (bytes_sent == 0) { 972 break; 973 } 974 bytes_transferred += bytes_sent; 975 } 976 977 ram_control_after_iterate(f, RAM_CONTROL_FINISH); 978 migration_end(); 979 980 rcu_read_unlock(); 981 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 982 983 return 0; 984 } 985 986 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size) 987 { 988 uint64_t remaining_size; 989 990 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE; 991 992 if (remaining_size < max_size) { 993 qemu_mutex_lock_iothread(); 994 rcu_read_lock(); 995 migration_bitmap_sync(); 996 rcu_read_unlock(); 997 qemu_mutex_unlock_iothread(); 998 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE; 999 } 1000 return remaining_size; 1001 } 1002 1003 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) 1004 { 1005 unsigned int xh_len; 1006 int xh_flags; 1007 1008 if (!xbzrle_decoded_buf) { 1009 xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE); 1010 } 1011 1012 /* extract RLE header */ 1013 xh_flags = qemu_get_byte(f); 1014 xh_len = qemu_get_be16(f); 1015 1016 if (xh_flags != ENCODING_FLAG_XBZRLE) { 1017 error_report("Failed to load XBZRLE page - wrong compression!"); 1018 return -1; 1019 } 1020 1021 if (xh_len > TARGET_PAGE_SIZE) { 1022 error_report("Failed to load XBZRLE page - len overflow!"); 1023 return -1; 1024 } 1025 /* load data and decode */ 1026 qemu_get_buffer(f, xbzrle_decoded_buf, xh_len); 1027 1028 /* decode RLE */ 1029 if (xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host, 1030 TARGET_PAGE_SIZE) == -1) { 1031 error_report("Failed to load XBZRLE page - decode error!"); 1032 return -1; 1033 } 1034 1035 return 0; 1036 } 1037 1038 /* Must be called from within a rcu critical section. 1039 * Returns a pointer from within the RCU-protected ram_list. 1040 */ 1041 static inline void *host_from_stream_offset(QEMUFile *f, 1042 ram_addr_t offset, 1043 int flags) 1044 { 1045 static RAMBlock *block = NULL; 1046 char id[256]; 1047 uint8_t len; 1048 1049 if (flags & RAM_SAVE_FLAG_CONTINUE) { 1050 if (!block || block->max_length <= offset) { 1051 error_report("Ack, bad migration stream!"); 1052 return NULL; 1053 } 1054 1055 return memory_region_get_ram_ptr(block->mr) + offset; 1056 } 1057 1058 len = qemu_get_byte(f); 1059 qemu_get_buffer(f, (uint8_t *)id, len); 1060 id[len] = 0; 1061 1062 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { 1063 if (!strncmp(id, block->idstr, sizeof(id)) && 1064 block->max_length > offset) { 1065 return memory_region_get_ram_ptr(block->mr) + offset; 1066 } 1067 } 1068 1069 error_report("Can't find block %s!", id); 1070 return NULL; 1071 } 1072 1073 /* 1074 * If a page (or a whole RDMA chunk) has been 1075 * determined to be zero, then zap it. 1076 */ 1077 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) 1078 { 1079 if (ch != 0 || !is_zero_range(host, size)) { 1080 memset(host, ch, size); 1081 } 1082 } 1083 1084 static int ram_load(QEMUFile *f, void *opaque, int version_id) 1085 { 1086 int flags = 0, ret = 0; 1087 static uint64_t seq_iter; 1088 1089 seq_iter++; 1090 1091 if (version_id != 4) { 1092 ret = -EINVAL; 1093 } 1094 1095 /* This RCU critical section can be very long running. 1096 * When RCU reclaims in the code start to become numerous, 1097 * it will be necessary to reduce the granularity of this 1098 * critical section. 1099 */ 1100 rcu_read_lock(); 1101 while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { 1102 ram_addr_t addr, total_ram_bytes; 1103 void *host; 1104 uint8_t ch; 1105 1106 addr = qemu_get_be64(f); 1107 flags = addr & ~TARGET_PAGE_MASK; 1108 addr &= TARGET_PAGE_MASK; 1109 1110 switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { 1111 case RAM_SAVE_FLAG_MEM_SIZE: 1112 /* Synchronize RAM block list */ 1113 total_ram_bytes = addr; 1114 while (!ret && total_ram_bytes) { 1115 RAMBlock *block; 1116 uint8_t len; 1117 char id[256]; 1118 ram_addr_t length; 1119 1120 len = qemu_get_byte(f); 1121 qemu_get_buffer(f, (uint8_t *)id, len); 1122 id[len] = 0; 1123 length = qemu_get_be64(f); 1124 1125 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { 1126 if (!strncmp(id, block->idstr, sizeof(id))) { 1127 if (length != block->used_length) { 1128 Error *local_err = NULL; 1129 1130 ret = qemu_ram_resize(block->offset, length, &local_err); 1131 if (local_err) { 1132 error_report_err(local_err); 1133 } 1134 } 1135 break; 1136 } 1137 } 1138 1139 if (!block) { 1140 error_report("Unknown ramblock \"%s\", cannot " 1141 "accept migration", id); 1142 ret = -EINVAL; 1143 } 1144 1145 total_ram_bytes -= length; 1146 } 1147 break; 1148 case RAM_SAVE_FLAG_COMPRESS: 1149 host = host_from_stream_offset(f, addr, flags); 1150 if (!host) { 1151 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); 1152 ret = -EINVAL; 1153 break; 1154 } 1155 ch = qemu_get_byte(f); 1156 ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); 1157 break; 1158 case RAM_SAVE_FLAG_PAGE: 1159 host = host_from_stream_offset(f, addr, flags); 1160 if (!host) { 1161 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); 1162 ret = -EINVAL; 1163 break; 1164 } 1165 qemu_get_buffer(f, host, TARGET_PAGE_SIZE); 1166 break; 1167 case RAM_SAVE_FLAG_XBZRLE: 1168 host = host_from_stream_offset(f, addr, flags); 1169 if (!host) { 1170 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); 1171 ret = -EINVAL; 1172 break; 1173 } 1174 if (load_xbzrle(f, addr, host) < 0) { 1175 error_report("Failed to decompress XBZRLE page at " 1176 RAM_ADDR_FMT, addr); 1177 ret = -EINVAL; 1178 break; 1179 } 1180 break; 1181 case RAM_SAVE_FLAG_EOS: 1182 /* normal exit */ 1183 break; 1184 default: 1185 if (flags & RAM_SAVE_FLAG_HOOK) { 1186 ram_control_load_hook(f, flags); 1187 } else { 1188 error_report("Unknown combination of migration flags: %#x", 1189 flags); 1190 ret = -EINVAL; 1191 } 1192 } 1193 if (!ret) { 1194 ret = qemu_file_get_error(f); 1195 } 1196 } 1197 1198 rcu_read_unlock(); 1199 DPRINTF("Completed load of VM with exit code %d seq iteration " 1200 "%" PRIu64 "\n", ret, seq_iter); 1201 return ret; 1202 } 1203 1204 static SaveVMHandlers savevm_ram_handlers = { 1205 .save_live_setup = ram_save_setup, 1206 .save_live_iterate = ram_save_iterate, 1207 .save_live_complete = ram_save_complete, 1208 .save_live_pending = ram_save_pending, 1209 .load_state = ram_load, 1210 .cancel = ram_migration_cancel, 1211 }; 1212 1213 void ram_mig_init(void) 1214 { 1215 qemu_mutex_init(&XBZRLE.lock); 1216 register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, NULL); 1217 } 1218 1219 struct soundhw { 1220 const char *name; 1221 const char *descr; 1222 int enabled; 1223 int isa; 1224 union { 1225 int (*init_isa) (ISABus *bus); 1226 int (*init_pci) (PCIBus *bus); 1227 } init; 1228 }; 1229 1230 static struct soundhw soundhw[9]; 1231 static int soundhw_count; 1232 1233 void isa_register_soundhw(const char *name, const char *descr, 1234 int (*init_isa)(ISABus *bus)) 1235 { 1236 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1); 1237 soundhw[soundhw_count].name = name; 1238 soundhw[soundhw_count].descr = descr; 1239 soundhw[soundhw_count].isa = 1; 1240 soundhw[soundhw_count].init.init_isa = init_isa; 1241 soundhw_count++; 1242 } 1243 1244 void pci_register_soundhw(const char *name, const char *descr, 1245 int (*init_pci)(PCIBus *bus)) 1246 { 1247 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1); 1248 soundhw[soundhw_count].name = name; 1249 soundhw[soundhw_count].descr = descr; 1250 soundhw[soundhw_count].isa = 0; 1251 soundhw[soundhw_count].init.init_pci = init_pci; 1252 soundhw_count++; 1253 } 1254 1255 void select_soundhw(const char *optarg) 1256 { 1257 struct soundhw *c; 1258 1259 if (is_help_option(optarg)) { 1260 show_valid_cards: 1261 1262 if (soundhw_count) { 1263 printf("Valid sound card names (comma separated):\n"); 1264 for (c = soundhw; c->name; ++c) { 1265 printf ("%-11s %s\n", c->name, c->descr); 1266 } 1267 printf("\n-soundhw all will enable all of the above\n"); 1268 } else { 1269 printf("Machine has no user-selectable audio hardware " 1270 "(it may or may not have always-present audio hardware).\n"); 1271 } 1272 exit(!is_help_option(optarg)); 1273 } 1274 else { 1275 size_t l; 1276 const char *p; 1277 char *e; 1278 int bad_card = 0; 1279 1280 if (!strcmp(optarg, "all")) { 1281 for (c = soundhw; c->name; ++c) { 1282 c->enabled = 1; 1283 } 1284 return; 1285 } 1286 1287 p = optarg; 1288 while (*p) { 1289 e = strchr(p, ','); 1290 l = !e ? strlen(p) : (size_t) (e - p); 1291 1292 for (c = soundhw; c->name; ++c) { 1293 if (!strncmp(c->name, p, l) && !c->name[l]) { 1294 c->enabled = 1; 1295 break; 1296 } 1297 } 1298 1299 if (!c->name) { 1300 if (l > 80) { 1301 error_report("Unknown sound card name (too big to show)"); 1302 } 1303 else { 1304 error_report("Unknown sound card name `%.*s'", 1305 (int) l, p); 1306 } 1307 bad_card = 1; 1308 } 1309 p += l + (e != NULL); 1310 } 1311 1312 if (bad_card) { 1313 goto show_valid_cards; 1314 } 1315 } 1316 } 1317 1318 void audio_init(void) 1319 { 1320 struct soundhw *c; 1321 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL); 1322 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL); 1323 1324 for (c = soundhw; c->name; ++c) { 1325 if (c->enabled) { 1326 if (c->isa) { 1327 if (!isa_bus) { 1328 error_report("ISA bus not available for %s", c->name); 1329 exit(1); 1330 } 1331 c->init.init_isa(isa_bus); 1332 } else { 1333 if (!pci_bus) { 1334 error_report("PCI bus not available for %s", c->name); 1335 exit(1); 1336 } 1337 c->init.init_pci(pci_bus); 1338 } 1339 } 1340 } 1341 } 1342 1343 int qemu_uuid_parse(const char *str, uint8_t *uuid) 1344 { 1345 int ret; 1346 1347 if (strlen(str) != 36) { 1348 return -1; 1349 } 1350 1351 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3], 1352 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9], 1353 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14], 1354 &uuid[15]); 1355 1356 if (ret != 16) { 1357 return -1; 1358 } 1359 return 0; 1360 } 1361 1362 void do_acpitable_option(const QemuOpts *opts) 1363 { 1364 #ifdef TARGET_I386 1365 Error *err = NULL; 1366 1367 acpi_table_add(opts, &err); 1368 if (err) { 1369 error_report("Wrong acpi table provided: %s", 1370 error_get_pretty(err)); 1371 error_free(err); 1372 exit(1); 1373 } 1374 #endif 1375 } 1376 1377 void do_smbios_option(QemuOpts *opts) 1378 { 1379 #ifdef TARGET_I386 1380 smbios_entry_add(opts); 1381 #endif 1382 } 1383 1384 void cpudef_init(void) 1385 { 1386 #if defined(cpudef_setup) 1387 cpudef_setup(); /* parse cpu definitions in target config file */ 1388 #endif 1389 } 1390 1391 int kvm_available(void) 1392 { 1393 #ifdef CONFIG_KVM 1394 return 1; 1395 #else 1396 return 0; 1397 #endif 1398 } 1399 1400 int xen_available(void) 1401 { 1402 #ifdef CONFIG_XEN 1403 return 1; 1404 #else 1405 return 0; 1406 #endif 1407 } 1408 1409 1410 TargetInfo *qmp_query_target(Error **errp) 1411 { 1412 TargetInfo *info = g_malloc0(sizeof(*info)); 1413 1414 info->arch = g_strdup(TARGET_NAME); 1415 1416 return info; 1417 } 1418 1419 /* Stub function that's gets run on the vcpu when its brought out of the 1420 VM to run inside qemu via async_run_on_cpu()*/ 1421 static void mig_sleep_cpu(void *opq) 1422 { 1423 qemu_mutex_unlock_iothread(); 1424 g_usleep(30*1000); 1425 qemu_mutex_lock_iothread(); 1426 } 1427 1428 /* To reduce the dirty rate explicitly disallow the VCPUs from spending 1429 much time in the VM. The migration thread will try to catchup. 1430 Workload will experience a performance drop. 1431 */ 1432 static void mig_throttle_guest_down(void) 1433 { 1434 CPUState *cpu; 1435 1436 qemu_mutex_lock_iothread(); 1437 CPU_FOREACH(cpu) { 1438 async_run_on_cpu(cpu, mig_sleep_cpu, NULL); 1439 } 1440 qemu_mutex_unlock_iothread(); 1441 } 1442 1443 static void check_guest_throttling(void) 1444 { 1445 static int64_t t0; 1446 int64_t t1; 1447 1448 if (!mig_throttle_on) { 1449 return; 1450 } 1451 1452 if (!t0) { 1453 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1454 return; 1455 } 1456 1457 t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1458 1459 /* If it has been more than 40 ms since the last time the guest 1460 * was throttled then do it again. 1461 */ 1462 if (40 < (t1-t0)/1000000) { 1463 mig_throttle_guest_down(); 1464 t0 = t1; 1465 } 1466 } 1467