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