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