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 if (!last_stage) { 353 if (cache_insert(XBZRLE.cache, current_addr, current_data) == -1) { 354 return -1; 355 } 356 } 357 acct_info.xbzrle_cache_miss++; 358 return -1; 359 } 360 361 prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); 362 363 /* save current buffer into memory */ 364 memcpy(XBZRLE.current_buf, current_data, TARGET_PAGE_SIZE); 365 366 /* XBZRLE encoding (if there is no overflow) */ 367 encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, 368 TARGET_PAGE_SIZE, XBZRLE.encoded_buf, 369 TARGET_PAGE_SIZE); 370 if (encoded_len == 0) { 371 DPRINTF("Skipping unmodified page\n"); 372 return 0; 373 } else if (encoded_len == -1) { 374 DPRINTF("Overflow\n"); 375 acct_info.xbzrle_overflows++; 376 /* update data in the cache */ 377 memcpy(prev_cached_page, current_data, TARGET_PAGE_SIZE); 378 return -1; 379 } 380 381 /* we need to update the data in the cache, in order to get the same data */ 382 if (!last_stage) { 383 memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); 384 } 385 386 /* Send XBZRLE based compressed page */ 387 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_XBZRLE); 388 qemu_put_byte(f, ENCODING_FLAG_XBZRLE); 389 qemu_put_be16(f, encoded_len); 390 qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len); 391 bytes_sent += encoded_len + 1 + 2; 392 acct_info.xbzrle_pages++; 393 acct_info.xbzrle_bytes += bytes_sent; 394 395 return bytes_sent; 396 } 397 398 static inline 399 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr, 400 ram_addr_t start) 401 { 402 unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS; 403 unsigned long nr = base + (start >> TARGET_PAGE_BITS); 404 uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr)); 405 unsigned long size = base + (mr_size >> TARGET_PAGE_BITS); 406 407 unsigned long next; 408 409 if (ram_bulk_stage && nr > base) { 410 next = nr + 1; 411 } else { 412 next = find_next_bit(migration_bitmap, size, nr); 413 } 414 415 if (next < size) { 416 clear_bit(next, migration_bitmap); 417 migration_dirty_pages--; 418 } 419 return (next - base) << TARGET_PAGE_BITS; 420 } 421 422 static inline bool migration_bitmap_set_dirty(ram_addr_t addr) 423 { 424 bool ret; 425 int nr = addr >> TARGET_PAGE_BITS; 426 427 ret = test_and_set_bit(nr, migration_bitmap); 428 429 if (!ret) { 430 migration_dirty_pages++; 431 } 432 return ret; 433 } 434 435 static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length) 436 { 437 ram_addr_t addr; 438 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS); 439 440 /* start address is aligned at the start of a word? */ 441 if (((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) { 442 int k; 443 int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS); 444 unsigned long *src = ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION]; 445 446 for (k = page; k < page + nr; k++) { 447 if (src[k]) { 448 unsigned long new_dirty; 449 new_dirty = ~migration_bitmap[k]; 450 migration_bitmap[k] |= src[k]; 451 new_dirty &= src[k]; 452 migration_dirty_pages += ctpopl(new_dirty); 453 src[k] = 0; 454 } 455 } 456 } else { 457 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) { 458 if (cpu_physical_memory_get_dirty(start + addr, 459 TARGET_PAGE_SIZE, 460 DIRTY_MEMORY_MIGRATION)) { 461 cpu_physical_memory_reset_dirty(start + addr, 462 TARGET_PAGE_SIZE, 463 DIRTY_MEMORY_MIGRATION); 464 migration_bitmap_set_dirty(start + addr); 465 } 466 } 467 } 468 } 469 470 471 /* Needs iothread lock! */ 472 473 static void migration_bitmap_sync(void) 474 { 475 RAMBlock *block; 476 uint64_t num_dirty_pages_init = migration_dirty_pages; 477 MigrationState *s = migrate_get_current(); 478 static int64_t start_time; 479 static int64_t bytes_xfer_prev; 480 static int64_t num_dirty_pages_period; 481 int64_t end_time; 482 int64_t bytes_xfer_now; 483 484 if (!bytes_xfer_prev) { 485 bytes_xfer_prev = ram_bytes_transferred(); 486 } 487 488 if (!start_time) { 489 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 490 } 491 492 trace_migration_bitmap_sync_start(); 493 address_space_sync_dirty_bitmap(&address_space_memory); 494 495 QTAILQ_FOREACH(block, &ram_list.blocks, next) { 496 migration_bitmap_sync_range(block->mr->ram_addr, block->length); 497 } 498 trace_migration_bitmap_sync_end(migration_dirty_pages 499 - num_dirty_pages_init); 500 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init; 501 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 502 503 /* more than 1 second = 1000 millisecons */ 504 if (end_time > start_time + 1000) { 505 if (migrate_auto_converge()) { 506 /* The following detection logic can be refined later. For now: 507 Check to see if the dirtied bytes is 50% more than the approx. 508 amount of bytes that just got transferred since the last time we 509 were in this routine. If that happens >N times (for now N==4) 510 we turn on the throttle down logic */ 511 bytes_xfer_now = ram_bytes_transferred(); 512 if (s->dirty_pages_rate && 513 (num_dirty_pages_period * TARGET_PAGE_SIZE > 514 (bytes_xfer_now - bytes_xfer_prev)/2) && 515 (dirty_rate_high_cnt++ > 4)) { 516 trace_migration_throttle(); 517 mig_throttle_on = true; 518 dirty_rate_high_cnt = 0; 519 } 520 bytes_xfer_prev = bytes_xfer_now; 521 } else { 522 mig_throttle_on = false; 523 } 524 s->dirty_pages_rate = num_dirty_pages_period * 1000 525 / (end_time - start_time); 526 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE; 527 start_time = end_time; 528 num_dirty_pages_period = 0; 529 } 530 } 531 532 /* 533 * ram_save_block: Writes a page of memory to the stream f 534 * 535 * Returns: The number of bytes written. 536 * 0 means no dirty pages 537 */ 538 539 static int ram_save_block(QEMUFile *f, bool last_stage) 540 { 541 RAMBlock *block = last_seen_block; 542 ram_addr_t offset = last_offset; 543 bool complete_round = false; 544 int bytes_sent = 0; 545 MemoryRegion *mr; 546 ram_addr_t current_addr; 547 548 if (!block) 549 block = QTAILQ_FIRST(&ram_list.blocks); 550 551 while (true) { 552 mr = block->mr; 553 offset = migration_bitmap_find_and_reset_dirty(mr, offset); 554 if (complete_round && block == last_seen_block && 555 offset >= last_offset) { 556 break; 557 } 558 if (offset >= block->length) { 559 offset = 0; 560 block = QTAILQ_NEXT(block, next); 561 if (!block) { 562 block = QTAILQ_FIRST(&ram_list.blocks); 563 complete_round = true; 564 ram_bulk_stage = false; 565 } 566 } else { 567 int ret; 568 uint8_t *p; 569 bool send_async = true; 570 int cont = (block == last_sent_block) ? 571 RAM_SAVE_FLAG_CONTINUE : 0; 572 573 p = memory_region_get_ram_ptr(mr) + offset; 574 575 /* In doubt sent page as normal */ 576 bytes_sent = -1; 577 ret = ram_control_save_page(f, block->offset, 578 offset, TARGET_PAGE_SIZE, &bytes_sent); 579 580 XBZRLE_cache_lock(); 581 582 current_addr = block->offset + offset; 583 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { 584 if (ret != RAM_SAVE_CONTROL_DELAYED) { 585 if (bytes_sent > 0) { 586 acct_info.norm_pages++; 587 } else if (bytes_sent == 0) { 588 acct_info.dup_pages++; 589 } 590 } 591 } else if (is_zero_range(p, TARGET_PAGE_SIZE)) { 592 acct_info.dup_pages++; 593 bytes_sent = save_block_hdr(f, block, offset, cont, 594 RAM_SAVE_FLAG_COMPRESS); 595 qemu_put_byte(f, 0); 596 bytes_sent++; 597 /* Must let xbzrle know, otherwise a previous (now 0'd) cached 598 * page would be stale 599 */ 600 xbzrle_cache_zero_page(current_addr); 601 } else if (!ram_bulk_stage && migrate_use_xbzrle()) { 602 bytes_sent = save_xbzrle_page(f, p, current_addr, block, 603 offset, cont, last_stage); 604 if (!last_stage) { 605 /* We must send exactly what's in the xbzrle cache 606 * even if the page wasn't xbzrle compressed, so that 607 * it's right next time. 608 */ 609 p = get_cached_data(XBZRLE.cache, current_addr); 610 611 /* Can't send this cached data async, since the cache page 612 * might get updated before it gets to the wire 613 */ 614 send_async = false; 615 } 616 } 617 618 /* XBZRLE overflow or normal page */ 619 if (bytes_sent == -1) { 620 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_PAGE); 621 if (send_async) { 622 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE); 623 } else { 624 qemu_put_buffer(f, p, TARGET_PAGE_SIZE); 625 } 626 bytes_sent += TARGET_PAGE_SIZE; 627 acct_info.norm_pages++; 628 } 629 630 XBZRLE_cache_unlock(); 631 /* if page is unmodified, continue to the next */ 632 if (bytes_sent > 0) { 633 last_sent_block = block; 634 break; 635 } 636 } 637 } 638 last_seen_block = block; 639 last_offset = offset; 640 641 return bytes_sent; 642 } 643 644 static uint64_t bytes_transferred; 645 646 void acct_update_position(QEMUFile *f, size_t size, bool zero) 647 { 648 uint64_t pages = size / TARGET_PAGE_SIZE; 649 if (zero) { 650 acct_info.dup_pages += pages; 651 } else { 652 acct_info.norm_pages += pages; 653 bytes_transferred += size; 654 qemu_update_position(f, size); 655 } 656 } 657 658 static ram_addr_t ram_save_remaining(void) 659 { 660 return migration_dirty_pages; 661 } 662 663 uint64_t ram_bytes_remaining(void) 664 { 665 return ram_save_remaining() * TARGET_PAGE_SIZE; 666 } 667 668 uint64_t ram_bytes_transferred(void) 669 { 670 return bytes_transferred; 671 } 672 673 uint64_t ram_bytes_total(void) 674 { 675 RAMBlock *block; 676 uint64_t total = 0; 677 678 QTAILQ_FOREACH(block, &ram_list.blocks, next) 679 total += block->length; 680 681 return total; 682 } 683 684 void free_xbzrle_decoded_buf(void) 685 { 686 g_free(xbzrle_decoded_buf); 687 xbzrle_decoded_buf = NULL; 688 } 689 690 static void migration_end(void) 691 { 692 if (migration_bitmap) { 693 memory_global_dirty_log_stop(); 694 g_free(migration_bitmap); 695 migration_bitmap = NULL; 696 } 697 698 XBZRLE_cache_lock(); 699 if (XBZRLE.cache) { 700 cache_fini(XBZRLE.cache); 701 g_free(XBZRLE.cache); 702 g_free(XBZRLE.encoded_buf); 703 g_free(XBZRLE.current_buf); 704 XBZRLE.cache = NULL; 705 XBZRLE.encoded_buf = NULL; 706 XBZRLE.current_buf = NULL; 707 } 708 XBZRLE_cache_unlock(); 709 } 710 711 static void ram_migration_cancel(void *opaque) 712 { 713 migration_end(); 714 } 715 716 static void reset_ram_globals(void) 717 { 718 last_seen_block = NULL; 719 last_sent_block = NULL; 720 last_offset = 0; 721 last_version = ram_list.version; 722 ram_bulk_stage = true; 723 } 724 725 #define MAX_WAIT 50 /* ms, half buffered_file limit */ 726 727 static int ram_save_setup(QEMUFile *f, void *opaque) 728 { 729 RAMBlock *block; 730 int64_t ram_bitmap_pages; /* Size of bitmap in pages, including gaps */ 731 732 mig_throttle_on = false; 733 dirty_rate_high_cnt = 0; 734 735 if (migrate_use_xbzrle()) { 736 XBZRLE_cache_lock(); 737 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() / 738 TARGET_PAGE_SIZE, 739 TARGET_PAGE_SIZE); 740 if (!XBZRLE.cache) { 741 XBZRLE_cache_unlock(); 742 error_report("Error creating cache"); 743 return -1; 744 } 745 XBZRLE_cache_unlock(); 746 747 /* We prefer not to abort if there is no memory */ 748 XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); 749 if (!XBZRLE.encoded_buf) { 750 error_report("Error allocating encoded_buf"); 751 return -1; 752 } 753 754 XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); 755 if (!XBZRLE.current_buf) { 756 error_report("Error allocating current_buf"); 757 g_free(XBZRLE.encoded_buf); 758 XBZRLE.encoded_buf = NULL; 759 return -1; 760 } 761 762 acct_clear(); 763 } 764 765 qemu_mutex_lock_iothread(); 766 qemu_mutex_lock_ramlist(); 767 bytes_transferred = 0; 768 reset_ram_globals(); 769 770 ram_bitmap_pages = last_ram_offset() >> TARGET_PAGE_BITS; 771 migration_bitmap = bitmap_new(ram_bitmap_pages); 772 bitmap_set(migration_bitmap, 0, ram_bitmap_pages); 773 774 /* 775 * Count the total number of pages used by ram blocks not including any 776 * gaps due to alignment or unplugs. 777 */ 778 migration_dirty_pages = 0; 779 QTAILQ_FOREACH(block, &ram_list.blocks, next) { 780 uint64_t block_pages; 781 782 block_pages = block->length >> TARGET_PAGE_BITS; 783 migration_dirty_pages += block_pages; 784 } 785 786 memory_global_dirty_log_start(); 787 migration_bitmap_sync(); 788 qemu_mutex_unlock_iothread(); 789 790 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); 791 792 QTAILQ_FOREACH(block, &ram_list.blocks, next) { 793 qemu_put_byte(f, strlen(block->idstr)); 794 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); 795 qemu_put_be64(f, block->length); 796 } 797 798 qemu_mutex_unlock_ramlist(); 799 800 ram_control_before_iterate(f, RAM_CONTROL_SETUP); 801 ram_control_after_iterate(f, RAM_CONTROL_SETUP); 802 803 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 804 805 return 0; 806 } 807 808 static int ram_save_iterate(QEMUFile *f, void *opaque) 809 { 810 int ret; 811 int i; 812 int64_t t0; 813 int total_sent = 0; 814 815 qemu_mutex_lock_ramlist(); 816 817 if (ram_list.version != last_version) { 818 reset_ram_globals(); 819 } 820 821 ram_control_before_iterate(f, RAM_CONTROL_ROUND); 822 823 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 824 i = 0; 825 while ((ret = qemu_file_rate_limit(f)) == 0) { 826 int bytes_sent; 827 828 bytes_sent = ram_save_block(f, false); 829 /* no more blocks to sent */ 830 if (bytes_sent == 0) { 831 break; 832 } 833 total_sent += bytes_sent; 834 acct_info.iterations++; 835 check_guest_throttling(); 836 /* we want to check in the 1st loop, just in case it was the 1st time 837 and we had to sync the dirty bitmap. 838 qemu_get_clock_ns() is a bit expensive, so we only check each some 839 iterations 840 */ 841 if ((i & 63) == 0) { 842 uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; 843 if (t1 > MAX_WAIT) { 844 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n", 845 t1, i); 846 break; 847 } 848 } 849 i++; 850 } 851 852 qemu_mutex_unlock_ramlist(); 853 854 /* 855 * Must occur before EOS (or any QEMUFile operation) 856 * because of RDMA protocol. 857 */ 858 ram_control_after_iterate(f, RAM_CONTROL_ROUND); 859 860 bytes_transferred += total_sent; 861 862 /* 863 * Do not count these 8 bytes into total_sent, so that we can 864 * return 0 if no page had been dirtied. 865 */ 866 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 867 bytes_transferred += 8; 868 869 ret = qemu_file_get_error(f); 870 if (ret < 0) { 871 return ret; 872 } 873 874 return total_sent; 875 } 876 877 static int ram_save_complete(QEMUFile *f, void *opaque) 878 { 879 qemu_mutex_lock_ramlist(); 880 migration_bitmap_sync(); 881 882 ram_control_before_iterate(f, RAM_CONTROL_FINISH); 883 884 /* try transferring iterative blocks of memory */ 885 886 /* flush all remaining blocks regardless of rate limiting */ 887 while (true) { 888 int bytes_sent; 889 890 bytes_sent = ram_save_block(f, true); 891 /* no more blocks to sent */ 892 if (bytes_sent == 0) { 893 break; 894 } 895 bytes_transferred += bytes_sent; 896 } 897 898 ram_control_after_iterate(f, RAM_CONTROL_FINISH); 899 migration_end(); 900 901 qemu_mutex_unlock_ramlist(); 902 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 903 904 return 0; 905 } 906 907 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size) 908 { 909 uint64_t remaining_size; 910 911 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE; 912 913 if (remaining_size < max_size) { 914 qemu_mutex_lock_iothread(); 915 migration_bitmap_sync(); 916 qemu_mutex_unlock_iothread(); 917 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE; 918 } 919 return remaining_size; 920 } 921 922 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) 923 { 924 int ret, rc = 0; 925 unsigned int xh_len; 926 int xh_flags; 927 928 if (!xbzrle_decoded_buf) { 929 xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE); 930 } 931 932 /* extract RLE header */ 933 xh_flags = qemu_get_byte(f); 934 xh_len = qemu_get_be16(f); 935 936 if (xh_flags != ENCODING_FLAG_XBZRLE) { 937 fprintf(stderr, "Failed to load XBZRLE page - wrong compression!\n"); 938 return -1; 939 } 940 941 if (xh_len > TARGET_PAGE_SIZE) { 942 fprintf(stderr, "Failed to load XBZRLE page - len overflow!\n"); 943 return -1; 944 } 945 /* load data and decode */ 946 qemu_get_buffer(f, xbzrle_decoded_buf, xh_len); 947 948 /* decode RLE */ 949 ret = xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host, 950 TARGET_PAGE_SIZE); 951 if (ret == -1) { 952 fprintf(stderr, "Failed to load XBZRLE page - decode error!\n"); 953 rc = -1; 954 } else if (ret > TARGET_PAGE_SIZE) { 955 fprintf(stderr, "Failed to load XBZRLE page - size %d exceeds %d!\n", 956 ret, TARGET_PAGE_SIZE); 957 abort(); 958 } 959 960 return rc; 961 } 962 963 static inline void *host_from_stream_offset(QEMUFile *f, 964 ram_addr_t offset, 965 int flags) 966 { 967 static RAMBlock *block = NULL; 968 char id[256]; 969 uint8_t len; 970 971 if (flags & RAM_SAVE_FLAG_CONTINUE) { 972 if (!block) { 973 fprintf(stderr, "Ack, bad migration stream!\n"); 974 return NULL; 975 } 976 977 return memory_region_get_ram_ptr(block->mr) + offset; 978 } 979 980 len = qemu_get_byte(f); 981 qemu_get_buffer(f, (uint8_t *)id, len); 982 id[len] = 0; 983 984 QTAILQ_FOREACH(block, &ram_list.blocks, next) { 985 if (!strncmp(id, block->idstr, sizeof(id))) 986 return memory_region_get_ram_ptr(block->mr) + offset; 987 } 988 989 fprintf(stderr, "Can't find block %s!\n", id); 990 return NULL; 991 } 992 993 /* 994 * If a page (or a whole RDMA chunk) has been 995 * determined to be zero, then zap it. 996 */ 997 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) 998 { 999 if (ch != 0 || !is_zero_range(host, size)) { 1000 memset(host, ch, size); 1001 } 1002 } 1003 1004 static int ram_load(QEMUFile *f, void *opaque, int version_id) 1005 { 1006 ram_addr_t addr; 1007 int flags, ret = 0; 1008 int error; 1009 static uint64_t seq_iter; 1010 1011 seq_iter++; 1012 1013 if (version_id < 4 || version_id > 4) { 1014 return -EINVAL; 1015 } 1016 1017 do { 1018 addr = qemu_get_be64(f); 1019 1020 flags = addr & ~TARGET_PAGE_MASK; 1021 addr &= TARGET_PAGE_MASK; 1022 1023 if (flags & RAM_SAVE_FLAG_MEM_SIZE) { 1024 if (version_id == 4) { 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 1065 if (flags & RAM_SAVE_FLAG_COMPRESS) { 1066 void *host; 1067 uint8_t ch; 1068 1069 host = host_from_stream_offset(f, addr, flags); 1070 if (!host) { 1071 return -EINVAL; 1072 } 1073 1074 ch = qemu_get_byte(f); 1075 ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); 1076 } else if (flags & RAM_SAVE_FLAG_PAGE) { 1077 void *host; 1078 1079 host = host_from_stream_offset(f, addr, flags); 1080 if (!host) { 1081 return -EINVAL; 1082 } 1083 1084 qemu_get_buffer(f, host, TARGET_PAGE_SIZE); 1085 } else if (flags & RAM_SAVE_FLAG_XBZRLE) { 1086 void *host = host_from_stream_offset(f, addr, flags); 1087 if (!host) { 1088 return -EINVAL; 1089 } 1090 1091 if (load_xbzrle(f, addr, host) < 0) { 1092 ret = -EINVAL; 1093 goto done; 1094 } 1095 } else if (flags & RAM_SAVE_FLAG_HOOK) { 1096 ram_control_load_hook(f, flags); 1097 } 1098 error = qemu_file_get_error(f); 1099 if (error) { 1100 ret = error; 1101 goto done; 1102 } 1103 } while (!(flags & RAM_SAVE_FLAG_EOS)); 1104 1105 done: 1106 DPRINTF("Completed load of VM with exit code %d seq iteration " 1107 "%" PRIu64 "\n", ret, seq_iter); 1108 return ret; 1109 } 1110 1111 static SaveVMHandlers savevm_ram_handlers = { 1112 .save_live_setup = ram_save_setup, 1113 .save_live_iterate = ram_save_iterate, 1114 .save_live_complete = ram_save_complete, 1115 .save_live_pending = ram_save_pending, 1116 .load_state = ram_load, 1117 .cancel = ram_migration_cancel, 1118 }; 1119 1120 void ram_mig_init(void) 1121 { 1122 qemu_mutex_init(&XBZRLE.lock); 1123 register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, NULL); 1124 } 1125 1126 struct soundhw { 1127 const char *name; 1128 const char *descr; 1129 int enabled; 1130 int isa; 1131 union { 1132 int (*init_isa) (ISABus *bus); 1133 int (*init_pci) (PCIBus *bus); 1134 } init; 1135 }; 1136 1137 static struct soundhw soundhw[9]; 1138 static int soundhw_count; 1139 1140 void isa_register_soundhw(const char *name, const char *descr, 1141 int (*init_isa)(ISABus *bus)) 1142 { 1143 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1); 1144 soundhw[soundhw_count].name = name; 1145 soundhw[soundhw_count].descr = descr; 1146 soundhw[soundhw_count].isa = 1; 1147 soundhw[soundhw_count].init.init_isa = init_isa; 1148 soundhw_count++; 1149 } 1150 1151 void pci_register_soundhw(const char *name, const char *descr, 1152 int (*init_pci)(PCIBus *bus)) 1153 { 1154 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1); 1155 soundhw[soundhw_count].name = name; 1156 soundhw[soundhw_count].descr = descr; 1157 soundhw[soundhw_count].isa = 0; 1158 soundhw[soundhw_count].init.init_pci = init_pci; 1159 soundhw_count++; 1160 } 1161 1162 void select_soundhw(const char *optarg) 1163 { 1164 struct soundhw *c; 1165 1166 if (is_help_option(optarg)) { 1167 show_valid_cards: 1168 1169 if (soundhw_count) { 1170 printf("Valid sound card names (comma separated):\n"); 1171 for (c = soundhw; c->name; ++c) { 1172 printf ("%-11s %s\n", c->name, c->descr); 1173 } 1174 printf("\n-soundhw all will enable all of the above\n"); 1175 } else { 1176 printf("Machine has no user-selectable audio hardware " 1177 "(it may or may not have always-present audio hardware).\n"); 1178 } 1179 exit(!is_help_option(optarg)); 1180 } 1181 else { 1182 size_t l; 1183 const char *p; 1184 char *e; 1185 int bad_card = 0; 1186 1187 if (!strcmp(optarg, "all")) { 1188 for (c = soundhw; c->name; ++c) { 1189 c->enabled = 1; 1190 } 1191 return; 1192 } 1193 1194 p = optarg; 1195 while (*p) { 1196 e = strchr(p, ','); 1197 l = !e ? strlen(p) : (size_t) (e - p); 1198 1199 for (c = soundhw; c->name; ++c) { 1200 if (!strncmp(c->name, p, l) && !c->name[l]) { 1201 c->enabled = 1; 1202 break; 1203 } 1204 } 1205 1206 if (!c->name) { 1207 if (l > 80) { 1208 fprintf(stderr, 1209 "Unknown sound card name (too big to show)\n"); 1210 } 1211 else { 1212 fprintf(stderr, "Unknown sound card name `%.*s'\n", 1213 (int) l, p); 1214 } 1215 bad_card = 1; 1216 } 1217 p += l + (e != NULL); 1218 } 1219 1220 if (bad_card) { 1221 goto show_valid_cards; 1222 } 1223 } 1224 } 1225 1226 void audio_init(void) 1227 { 1228 struct soundhw *c; 1229 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL); 1230 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL); 1231 1232 for (c = soundhw; c->name; ++c) { 1233 if (c->enabled) { 1234 if (c->isa) { 1235 if (!isa_bus) { 1236 fprintf(stderr, "ISA bus not available for %s\n", c->name); 1237 exit(1); 1238 } 1239 c->init.init_isa(isa_bus); 1240 } else { 1241 if (!pci_bus) { 1242 fprintf(stderr, "PCI bus not available for %s\n", c->name); 1243 exit(1); 1244 } 1245 c->init.init_pci(pci_bus); 1246 } 1247 } 1248 } 1249 } 1250 1251 int qemu_uuid_parse(const char *str, uint8_t *uuid) 1252 { 1253 int ret; 1254 1255 if (strlen(str) != 36) { 1256 return -1; 1257 } 1258 1259 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3], 1260 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9], 1261 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14], 1262 &uuid[15]); 1263 1264 if (ret != 16) { 1265 return -1; 1266 } 1267 return 0; 1268 } 1269 1270 void do_acpitable_option(const QemuOpts *opts) 1271 { 1272 #ifdef TARGET_I386 1273 Error *err = NULL; 1274 1275 acpi_table_add(opts, &err); 1276 if (err) { 1277 error_report("Wrong acpi table provided: %s", 1278 error_get_pretty(err)); 1279 error_free(err); 1280 exit(1); 1281 } 1282 #endif 1283 } 1284 1285 void do_smbios_option(QemuOpts *opts) 1286 { 1287 #ifdef TARGET_I386 1288 smbios_entry_add(opts); 1289 #endif 1290 } 1291 1292 void cpudef_init(void) 1293 { 1294 #if defined(cpudef_setup) 1295 cpudef_setup(); /* parse cpu definitions in target config file */ 1296 #endif 1297 } 1298 1299 int tcg_available(void) 1300 { 1301 return 1; 1302 } 1303 1304 int kvm_available(void) 1305 { 1306 #ifdef CONFIG_KVM 1307 return 1; 1308 #else 1309 return 0; 1310 #endif 1311 } 1312 1313 int xen_available(void) 1314 { 1315 #ifdef CONFIG_XEN 1316 return 1; 1317 #else 1318 return 0; 1319 #endif 1320 } 1321 1322 1323 TargetInfo *qmp_query_target(Error **errp) 1324 { 1325 TargetInfo *info = g_malloc0(sizeof(*info)); 1326 1327 info->arch = g_strdup(TARGET_NAME); 1328 1329 return info; 1330 } 1331 1332 /* Stub function that's gets run on the vcpu when its brought out of the 1333 VM to run inside qemu via async_run_on_cpu()*/ 1334 static void mig_sleep_cpu(void *opq) 1335 { 1336 qemu_mutex_unlock_iothread(); 1337 g_usleep(30*1000); 1338 qemu_mutex_lock_iothread(); 1339 } 1340 1341 /* To reduce the dirty rate explicitly disallow the VCPUs from spending 1342 much time in the VM. The migration thread will try to catchup. 1343 Workload will experience a performance drop. 1344 */ 1345 static void mig_throttle_guest_down(void) 1346 { 1347 CPUState *cpu; 1348 1349 qemu_mutex_lock_iothread(); 1350 CPU_FOREACH(cpu) { 1351 async_run_on_cpu(cpu, mig_sleep_cpu, NULL); 1352 } 1353 qemu_mutex_unlock_iothread(); 1354 } 1355 1356 static void check_guest_throttling(void) 1357 { 1358 static int64_t t0; 1359 int64_t t1; 1360 1361 if (!mig_throttle_on) { 1362 return; 1363 } 1364 1365 if (!t0) { 1366 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1367 return; 1368 } 1369 1370 t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1371 1372 /* If it has been more than 40 ms since the last time the guest 1373 * was throttled then do it again. 1374 */ 1375 if (40 < (t1-t0)/1000000) { 1376 mig_throttle_guest_down(); 1377 t0 = t1; 1378 } 1379 } 1380