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! */ 489 490 static void migration_bitmap_sync(void) 491 { 492 RAMBlock *block; 493 uint64_t num_dirty_pages_init = migration_dirty_pages; 494 MigrationState *s = migrate_get_current(); 495 static int64_t start_time; 496 static int64_t bytes_xfer_prev; 497 static int64_t num_dirty_pages_period; 498 int64_t end_time; 499 int64_t bytes_xfer_now; 500 static uint64_t xbzrle_cache_miss_prev; 501 static uint64_t iterations_prev; 502 503 bitmap_sync_count++; 504 505 if (!bytes_xfer_prev) { 506 bytes_xfer_prev = ram_bytes_transferred(); 507 } 508 509 if (!start_time) { 510 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 511 } 512 513 trace_migration_bitmap_sync_start(); 514 address_space_sync_dirty_bitmap(&address_space_memory); 515 516 QTAILQ_FOREACH(block, &ram_list.blocks, next) { 517 migration_bitmap_sync_range(block->mr->ram_addr, block->length); 518 } 519 trace_migration_bitmap_sync_end(migration_dirty_pages 520 - num_dirty_pages_init); 521 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init; 522 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 523 524 /* more than 1 second = 1000 millisecons */ 525 if (end_time > start_time + 1000) { 526 if (migrate_auto_converge()) { 527 /* The following detection logic can be refined later. For now: 528 Check to see if the dirtied bytes is 50% more than the approx. 529 amount of bytes that just got transferred since the last time we 530 were in this routine. If that happens >N times (for now N==4) 531 we turn on the throttle down logic */ 532 bytes_xfer_now = ram_bytes_transferred(); 533 if (s->dirty_pages_rate && 534 (num_dirty_pages_period * TARGET_PAGE_SIZE > 535 (bytes_xfer_now - bytes_xfer_prev)/2) && 536 (dirty_rate_high_cnt++ > 4)) { 537 trace_migration_throttle(); 538 mig_throttle_on = true; 539 dirty_rate_high_cnt = 0; 540 } 541 bytes_xfer_prev = bytes_xfer_now; 542 } else { 543 mig_throttle_on = false; 544 } 545 if (migrate_use_xbzrle()) { 546 if (iterations_prev != 0) { 547 acct_info.xbzrle_cache_miss_rate = 548 (double)(acct_info.xbzrle_cache_miss - 549 xbzrle_cache_miss_prev) / 550 (acct_info.iterations - iterations_prev); 551 } 552 iterations_prev = acct_info.iterations; 553 xbzrle_cache_miss_prev = acct_info.xbzrle_cache_miss; 554 } 555 s->dirty_pages_rate = num_dirty_pages_period * 1000 556 / (end_time - start_time); 557 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE; 558 start_time = end_time; 559 num_dirty_pages_period = 0; 560 s->dirty_sync_count = bitmap_sync_count; 561 } 562 } 563 564 /* 565 * ram_save_page: Send the given page to the stream 566 * 567 * Returns: Number of bytes written. 568 */ 569 static int ram_save_page(QEMUFile *f, RAMBlock* block, ram_addr_t offset, 570 bool last_stage) 571 { 572 int bytes_sent; 573 int cont; 574 ram_addr_t current_addr; 575 MemoryRegion *mr = block->mr; 576 uint8_t *p; 577 int ret; 578 bool send_async = true; 579 580 cont = (block == last_sent_block) ? RAM_SAVE_FLAG_CONTINUE : 0; 581 582 p = memory_region_get_ram_ptr(mr) + offset; 583 584 /* In doubt sent page as normal */ 585 bytes_sent = -1; 586 ret = ram_control_save_page(f, block->offset, 587 offset, TARGET_PAGE_SIZE, &bytes_sent); 588 589 XBZRLE_cache_lock(); 590 591 current_addr = block->offset + offset; 592 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { 593 if (ret != RAM_SAVE_CONTROL_DELAYED) { 594 if (bytes_sent > 0) { 595 acct_info.norm_pages++; 596 } else if (bytes_sent == 0) { 597 acct_info.dup_pages++; 598 } 599 } 600 } else if (is_zero_range(p, TARGET_PAGE_SIZE)) { 601 acct_info.dup_pages++; 602 bytes_sent = save_block_hdr(f, block, offset, cont, 603 RAM_SAVE_FLAG_COMPRESS); 604 qemu_put_byte(f, 0); 605 bytes_sent++; 606 /* Must let xbzrle know, otherwise a previous (now 0'd) cached 607 * page would be stale 608 */ 609 xbzrle_cache_zero_page(current_addr); 610 } else if (!ram_bulk_stage && migrate_use_xbzrle()) { 611 bytes_sent = save_xbzrle_page(f, &p, current_addr, block, 612 offset, cont, last_stage); 613 if (!last_stage) { 614 /* Can't send this cached data async, since the cache page 615 * might get updated before it gets to the wire 616 */ 617 send_async = false; 618 } 619 } 620 621 /* XBZRLE overflow or normal page */ 622 if (bytes_sent == -1) { 623 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_PAGE); 624 if (send_async) { 625 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE); 626 } else { 627 qemu_put_buffer(f, p, TARGET_PAGE_SIZE); 628 } 629 bytes_sent += TARGET_PAGE_SIZE; 630 acct_info.norm_pages++; 631 } 632 633 XBZRLE_cache_unlock(); 634 635 return bytes_sent; 636 } 637 638 /* 639 * ram_find_and_save_block: Finds a page to send and sends it to f 640 * 641 * Returns: The number of bytes written. 642 * 0 means no dirty pages 643 */ 644 645 static int ram_find_and_save_block(QEMUFile *f, bool last_stage) 646 { 647 RAMBlock *block = last_seen_block; 648 ram_addr_t offset = last_offset; 649 bool complete_round = false; 650 int bytes_sent = 0; 651 MemoryRegion *mr; 652 653 if (!block) 654 block = QTAILQ_FIRST(&ram_list.blocks); 655 656 while (true) { 657 mr = block->mr; 658 offset = migration_bitmap_find_and_reset_dirty(mr, offset); 659 if (complete_round && block == last_seen_block && 660 offset >= last_offset) { 661 break; 662 } 663 if (offset >= block->length) { 664 offset = 0; 665 block = QTAILQ_NEXT(block, next); 666 if (!block) { 667 block = QTAILQ_FIRST(&ram_list.blocks); 668 complete_round = true; 669 ram_bulk_stage = false; 670 } 671 } else { 672 bytes_sent = ram_save_page(f, block, offset, last_stage); 673 674 /* if page is unmodified, continue to the next */ 675 if (bytes_sent > 0) { 676 last_sent_block = block; 677 break; 678 } 679 } 680 } 681 last_seen_block = block; 682 last_offset = offset; 683 684 return bytes_sent; 685 } 686 687 static uint64_t bytes_transferred; 688 689 void acct_update_position(QEMUFile *f, size_t size, bool zero) 690 { 691 uint64_t pages = size / TARGET_PAGE_SIZE; 692 if (zero) { 693 acct_info.dup_pages += pages; 694 } else { 695 acct_info.norm_pages += pages; 696 bytes_transferred += size; 697 qemu_update_position(f, size); 698 } 699 } 700 701 static ram_addr_t ram_save_remaining(void) 702 { 703 return migration_dirty_pages; 704 } 705 706 uint64_t ram_bytes_remaining(void) 707 { 708 return ram_save_remaining() * TARGET_PAGE_SIZE; 709 } 710 711 uint64_t ram_bytes_transferred(void) 712 { 713 return bytes_transferred; 714 } 715 716 uint64_t ram_bytes_total(void) 717 { 718 RAMBlock *block; 719 uint64_t total = 0; 720 721 QTAILQ_FOREACH(block, &ram_list.blocks, next) 722 total += block->length; 723 724 return total; 725 } 726 727 void free_xbzrle_decoded_buf(void) 728 { 729 g_free(xbzrle_decoded_buf); 730 xbzrle_decoded_buf = NULL; 731 } 732 733 static void migration_end(void) 734 { 735 if (migration_bitmap) { 736 memory_global_dirty_log_stop(); 737 g_free(migration_bitmap); 738 migration_bitmap = NULL; 739 } 740 741 XBZRLE_cache_lock(); 742 if (XBZRLE.cache) { 743 cache_fini(XBZRLE.cache); 744 g_free(XBZRLE.encoded_buf); 745 g_free(XBZRLE.current_buf); 746 XBZRLE.cache = NULL; 747 XBZRLE.encoded_buf = NULL; 748 XBZRLE.current_buf = NULL; 749 } 750 XBZRLE_cache_unlock(); 751 } 752 753 static void ram_migration_cancel(void *opaque) 754 { 755 migration_end(); 756 } 757 758 static void reset_ram_globals(void) 759 { 760 last_seen_block = NULL; 761 last_sent_block = NULL; 762 last_offset = 0; 763 last_version = ram_list.version; 764 ram_bulk_stage = true; 765 } 766 767 #define MAX_WAIT 50 /* ms, half buffered_file limit */ 768 769 static int ram_save_setup(QEMUFile *f, void *opaque) 770 { 771 RAMBlock *block; 772 int64_t ram_bitmap_pages; /* Size of bitmap in pages, including gaps */ 773 774 mig_throttle_on = false; 775 dirty_rate_high_cnt = 0; 776 bitmap_sync_count = 0; 777 778 if (migrate_use_xbzrle()) { 779 XBZRLE_cache_lock(); 780 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() / 781 TARGET_PAGE_SIZE, 782 TARGET_PAGE_SIZE); 783 if (!XBZRLE.cache) { 784 XBZRLE_cache_unlock(); 785 error_report("Error creating cache"); 786 return -1; 787 } 788 XBZRLE_cache_unlock(); 789 790 /* We prefer not to abort if there is no memory */ 791 XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); 792 if (!XBZRLE.encoded_buf) { 793 error_report("Error allocating encoded_buf"); 794 return -1; 795 } 796 797 XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); 798 if (!XBZRLE.current_buf) { 799 error_report("Error allocating current_buf"); 800 g_free(XBZRLE.encoded_buf); 801 XBZRLE.encoded_buf = NULL; 802 return -1; 803 } 804 805 acct_clear(); 806 } 807 808 qemu_mutex_lock_iothread(); 809 qemu_mutex_lock_ramlist(); 810 bytes_transferred = 0; 811 reset_ram_globals(); 812 813 ram_bitmap_pages = last_ram_offset() >> TARGET_PAGE_BITS; 814 migration_bitmap = bitmap_new(ram_bitmap_pages); 815 bitmap_set(migration_bitmap, 0, ram_bitmap_pages); 816 817 /* 818 * Count the total number of pages used by ram blocks not including any 819 * gaps due to alignment or unplugs. 820 */ 821 migration_dirty_pages = 0; 822 QTAILQ_FOREACH(block, &ram_list.blocks, next) { 823 uint64_t block_pages; 824 825 block_pages = block->length >> TARGET_PAGE_BITS; 826 migration_dirty_pages += block_pages; 827 } 828 829 memory_global_dirty_log_start(); 830 migration_bitmap_sync(); 831 qemu_mutex_unlock_iothread(); 832 833 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); 834 835 QTAILQ_FOREACH(block, &ram_list.blocks, next) { 836 qemu_put_byte(f, strlen(block->idstr)); 837 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); 838 qemu_put_be64(f, block->length); 839 } 840 841 qemu_mutex_unlock_ramlist(); 842 843 ram_control_before_iterate(f, RAM_CONTROL_SETUP); 844 ram_control_after_iterate(f, RAM_CONTROL_SETUP); 845 846 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 847 848 return 0; 849 } 850 851 static int ram_save_iterate(QEMUFile *f, void *opaque) 852 { 853 int ret; 854 int i; 855 int64_t t0; 856 int total_sent = 0; 857 858 qemu_mutex_lock_ramlist(); 859 860 if (ram_list.version != last_version) { 861 reset_ram_globals(); 862 } 863 864 ram_control_before_iterate(f, RAM_CONTROL_ROUND); 865 866 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 867 i = 0; 868 while ((ret = qemu_file_rate_limit(f)) == 0) { 869 int bytes_sent; 870 871 bytes_sent = ram_find_and_save_block(f, false); 872 /* no more blocks to sent */ 873 if (bytes_sent == 0) { 874 break; 875 } 876 total_sent += bytes_sent; 877 acct_info.iterations++; 878 check_guest_throttling(); 879 /* we want to check in the 1st loop, just in case it was the 1st time 880 and we had to sync the dirty bitmap. 881 qemu_get_clock_ns() is a bit expensive, so we only check each some 882 iterations 883 */ 884 if ((i & 63) == 0) { 885 uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; 886 if (t1 > MAX_WAIT) { 887 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n", 888 t1, i); 889 break; 890 } 891 } 892 i++; 893 } 894 895 qemu_mutex_unlock_ramlist(); 896 897 /* 898 * Must occur before EOS (or any QEMUFile operation) 899 * because of RDMA protocol. 900 */ 901 ram_control_after_iterate(f, RAM_CONTROL_ROUND); 902 903 bytes_transferred += total_sent; 904 905 /* 906 * Do not count these 8 bytes into total_sent, so that we can 907 * return 0 if no page had been dirtied. 908 */ 909 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 910 bytes_transferred += 8; 911 912 ret = qemu_file_get_error(f); 913 if (ret < 0) { 914 return ret; 915 } 916 917 return total_sent; 918 } 919 920 static int ram_save_complete(QEMUFile *f, void *opaque) 921 { 922 qemu_mutex_lock_ramlist(); 923 migration_bitmap_sync(); 924 925 ram_control_before_iterate(f, RAM_CONTROL_FINISH); 926 927 /* try transferring iterative blocks of memory */ 928 929 /* flush all remaining blocks regardless of rate limiting */ 930 while (true) { 931 int bytes_sent; 932 933 bytes_sent = ram_find_and_save_block(f, true); 934 /* no more blocks to sent */ 935 if (bytes_sent == 0) { 936 break; 937 } 938 bytes_transferred += bytes_sent; 939 } 940 941 ram_control_after_iterate(f, RAM_CONTROL_FINISH); 942 migration_end(); 943 944 qemu_mutex_unlock_ramlist(); 945 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 946 947 return 0; 948 } 949 950 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size) 951 { 952 uint64_t remaining_size; 953 954 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE; 955 956 if (remaining_size < max_size) { 957 qemu_mutex_lock_iothread(); 958 migration_bitmap_sync(); 959 qemu_mutex_unlock_iothread(); 960 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE; 961 } 962 return remaining_size; 963 } 964 965 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) 966 { 967 unsigned int xh_len; 968 int xh_flags; 969 970 if (!xbzrle_decoded_buf) { 971 xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE); 972 } 973 974 /* extract RLE header */ 975 xh_flags = qemu_get_byte(f); 976 xh_len = qemu_get_be16(f); 977 978 if (xh_flags != ENCODING_FLAG_XBZRLE) { 979 error_report("Failed to load XBZRLE page - wrong compression!"); 980 return -1; 981 } 982 983 if (xh_len > TARGET_PAGE_SIZE) { 984 error_report("Failed to load XBZRLE page - len overflow!"); 985 return -1; 986 } 987 /* load data and decode */ 988 qemu_get_buffer(f, xbzrle_decoded_buf, xh_len); 989 990 /* decode RLE */ 991 if (xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host, 992 TARGET_PAGE_SIZE) == -1) { 993 error_report("Failed to load XBZRLE page - decode error!"); 994 return -1; 995 } 996 997 return 0; 998 } 999 1000 static inline void *host_from_stream_offset(QEMUFile *f, 1001 ram_addr_t offset, 1002 int flags) 1003 { 1004 static RAMBlock *block = NULL; 1005 char id[256]; 1006 uint8_t len; 1007 1008 if (flags & RAM_SAVE_FLAG_CONTINUE) { 1009 if (!block) { 1010 error_report("Ack, bad migration stream!"); 1011 return NULL; 1012 } 1013 1014 return memory_region_get_ram_ptr(block->mr) + offset; 1015 } 1016 1017 len = qemu_get_byte(f); 1018 qemu_get_buffer(f, (uint8_t *)id, len); 1019 id[len] = 0; 1020 1021 QTAILQ_FOREACH(block, &ram_list.blocks, next) { 1022 if (!strncmp(id, block->idstr, sizeof(id))) 1023 return memory_region_get_ram_ptr(block->mr) + offset; 1024 } 1025 1026 error_report("Can't find block %s!", id); 1027 return NULL; 1028 } 1029 1030 /* 1031 * If a page (or a whole RDMA chunk) has been 1032 * determined to be zero, then zap it. 1033 */ 1034 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) 1035 { 1036 if (ch != 0 || !is_zero_range(host, size)) { 1037 memset(host, ch, size); 1038 } 1039 } 1040 1041 static int ram_load(QEMUFile *f, void *opaque, int version_id) 1042 { 1043 ram_addr_t addr; 1044 int flags, ret = 0; 1045 static uint64_t seq_iter; 1046 1047 seq_iter++; 1048 1049 if (version_id != 4) { 1050 ret = -EINVAL; 1051 } 1052 1053 while (!ret) { 1054 addr = qemu_get_be64(f); 1055 1056 flags = addr & ~TARGET_PAGE_MASK; 1057 addr &= TARGET_PAGE_MASK; 1058 1059 if (flags & RAM_SAVE_FLAG_MEM_SIZE) { 1060 /* Synchronize RAM block list */ 1061 char id[256]; 1062 ram_addr_t length; 1063 ram_addr_t total_ram_bytes = addr; 1064 1065 while (total_ram_bytes) { 1066 RAMBlock *block; 1067 uint8_t len; 1068 1069 len = qemu_get_byte(f); 1070 qemu_get_buffer(f, (uint8_t *)id, len); 1071 id[len] = 0; 1072 length = qemu_get_be64(f); 1073 1074 QTAILQ_FOREACH(block, &ram_list.blocks, next) { 1075 if (!strncmp(id, block->idstr, sizeof(id))) { 1076 if (block->length != length) { 1077 error_report("Length mismatch: %s: 0x" RAM_ADDR_FMT 1078 " in != 0x" RAM_ADDR_FMT, id, length, 1079 block->length); 1080 ret = -EINVAL; 1081 } 1082 break; 1083 } 1084 } 1085 1086 if (!block) { 1087 error_report("Unknown ramblock \"%s\", cannot " 1088 "accept migration", id); 1089 ret = -EINVAL; 1090 } 1091 if (ret) { 1092 break; 1093 } 1094 1095 total_ram_bytes -= length; 1096 } 1097 } else if (flags & RAM_SAVE_FLAG_COMPRESS) { 1098 void *host; 1099 uint8_t ch; 1100 1101 host = host_from_stream_offset(f, addr, flags); 1102 if (!host) { 1103 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); 1104 ret = -EINVAL; 1105 break; 1106 } 1107 1108 ch = qemu_get_byte(f); 1109 ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); 1110 } else if (flags & RAM_SAVE_FLAG_PAGE) { 1111 void *host; 1112 1113 host = host_from_stream_offset(f, addr, flags); 1114 if (!host) { 1115 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); 1116 ret = -EINVAL; 1117 break; 1118 } 1119 1120 qemu_get_buffer(f, host, TARGET_PAGE_SIZE); 1121 } else if (flags & RAM_SAVE_FLAG_XBZRLE) { 1122 void *host = host_from_stream_offset(f, addr, flags); 1123 if (!host) { 1124 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); 1125 ret = -EINVAL; 1126 break; 1127 } 1128 1129 if (load_xbzrle(f, addr, host) < 0) { 1130 error_report("Failed to decompress XBZRLE page at " 1131 RAM_ADDR_FMT, addr); 1132 ret = -EINVAL; 1133 break; 1134 } 1135 } else if (flags & RAM_SAVE_FLAG_HOOK) { 1136 ram_control_load_hook(f, flags); 1137 } else if (flags & RAM_SAVE_FLAG_EOS) { 1138 /* normal exit */ 1139 break; 1140 } else { 1141 error_report("Unknown migration flags: %#x", flags); 1142 ret = -EINVAL; 1143 break; 1144 } 1145 ret = qemu_file_get_error(f); 1146 } 1147 1148 DPRINTF("Completed load of VM with exit code %d seq iteration " 1149 "%" PRIu64 "\n", ret, seq_iter); 1150 return ret; 1151 } 1152 1153 static SaveVMHandlers savevm_ram_handlers = { 1154 .save_live_setup = ram_save_setup, 1155 .save_live_iterate = ram_save_iterate, 1156 .save_live_complete = ram_save_complete, 1157 .save_live_pending = ram_save_pending, 1158 .load_state = ram_load, 1159 .cancel = ram_migration_cancel, 1160 }; 1161 1162 void ram_mig_init(void) 1163 { 1164 qemu_mutex_init(&XBZRLE.lock); 1165 register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, NULL); 1166 } 1167 1168 struct soundhw { 1169 const char *name; 1170 const char *descr; 1171 int enabled; 1172 int isa; 1173 union { 1174 int (*init_isa) (ISABus *bus); 1175 int (*init_pci) (PCIBus *bus); 1176 } init; 1177 }; 1178 1179 static struct soundhw soundhw[9]; 1180 static int soundhw_count; 1181 1182 void isa_register_soundhw(const char *name, const char *descr, 1183 int (*init_isa)(ISABus *bus)) 1184 { 1185 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1); 1186 soundhw[soundhw_count].name = name; 1187 soundhw[soundhw_count].descr = descr; 1188 soundhw[soundhw_count].isa = 1; 1189 soundhw[soundhw_count].init.init_isa = init_isa; 1190 soundhw_count++; 1191 } 1192 1193 void pci_register_soundhw(const char *name, const char *descr, 1194 int (*init_pci)(PCIBus *bus)) 1195 { 1196 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1); 1197 soundhw[soundhw_count].name = name; 1198 soundhw[soundhw_count].descr = descr; 1199 soundhw[soundhw_count].isa = 0; 1200 soundhw[soundhw_count].init.init_pci = init_pci; 1201 soundhw_count++; 1202 } 1203 1204 void select_soundhw(const char *optarg) 1205 { 1206 struct soundhw *c; 1207 1208 if (is_help_option(optarg)) { 1209 show_valid_cards: 1210 1211 if (soundhw_count) { 1212 printf("Valid sound card names (comma separated):\n"); 1213 for (c = soundhw; c->name; ++c) { 1214 printf ("%-11s %s\n", c->name, c->descr); 1215 } 1216 printf("\n-soundhw all will enable all of the above\n"); 1217 } else { 1218 printf("Machine has no user-selectable audio hardware " 1219 "(it may or may not have always-present audio hardware).\n"); 1220 } 1221 exit(!is_help_option(optarg)); 1222 } 1223 else { 1224 size_t l; 1225 const char *p; 1226 char *e; 1227 int bad_card = 0; 1228 1229 if (!strcmp(optarg, "all")) { 1230 for (c = soundhw; c->name; ++c) { 1231 c->enabled = 1; 1232 } 1233 return; 1234 } 1235 1236 p = optarg; 1237 while (*p) { 1238 e = strchr(p, ','); 1239 l = !e ? strlen(p) : (size_t) (e - p); 1240 1241 for (c = soundhw; c->name; ++c) { 1242 if (!strncmp(c->name, p, l) && !c->name[l]) { 1243 c->enabled = 1; 1244 break; 1245 } 1246 } 1247 1248 if (!c->name) { 1249 if (l > 80) { 1250 error_report("Unknown sound card name (too big to show)"); 1251 } 1252 else { 1253 error_report("Unknown sound card name `%.*s'", 1254 (int) l, p); 1255 } 1256 bad_card = 1; 1257 } 1258 p += l + (e != NULL); 1259 } 1260 1261 if (bad_card) { 1262 goto show_valid_cards; 1263 } 1264 } 1265 } 1266 1267 void audio_init(void) 1268 { 1269 struct soundhw *c; 1270 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL); 1271 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL); 1272 1273 for (c = soundhw; c->name; ++c) { 1274 if (c->enabled) { 1275 if (c->isa) { 1276 if (!isa_bus) { 1277 error_report("ISA bus not available for %s", c->name); 1278 exit(1); 1279 } 1280 c->init.init_isa(isa_bus); 1281 } else { 1282 if (!pci_bus) { 1283 error_report("PCI bus not available for %s", c->name); 1284 exit(1); 1285 } 1286 c->init.init_pci(pci_bus); 1287 } 1288 } 1289 } 1290 } 1291 1292 int qemu_uuid_parse(const char *str, uint8_t *uuid) 1293 { 1294 int ret; 1295 1296 if (strlen(str) != 36) { 1297 return -1; 1298 } 1299 1300 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3], 1301 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9], 1302 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14], 1303 &uuid[15]); 1304 1305 if (ret != 16) { 1306 return -1; 1307 } 1308 return 0; 1309 } 1310 1311 void do_acpitable_option(const QemuOpts *opts) 1312 { 1313 #ifdef TARGET_I386 1314 Error *err = NULL; 1315 1316 acpi_table_add(opts, &err); 1317 if (err) { 1318 error_report("Wrong acpi table provided: %s", 1319 error_get_pretty(err)); 1320 error_free(err); 1321 exit(1); 1322 } 1323 #endif 1324 } 1325 1326 void do_smbios_option(QemuOpts *opts) 1327 { 1328 #ifdef TARGET_I386 1329 smbios_entry_add(opts); 1330 #endif 1331 } 1332 1333 void cpudef_init(void) 1334 { 1335 #if defined(cpudef_setup) 1336 cpudef_setup(); /* parse cpu definitions in target config file */ 1337 #endif 1338 } 1339 1340 int kvm_available(void) 1341 { 1342 #ifdef CONFIG_KVM 1343 return 1; 1344 #else 1345 return 0; 1346 #endif 1347 } 1348 1349 int xen_available(void) 1350 { 1351 #ifdef CONFIG_XEN 1352 return 1; 1353 #else 1354 return 0; 1355 #endif 1356 } 1357 1358 1359 TargetInfo *qmp_query_target(Error **errp) 1360 { 1361 TargetInfo *info = g_malloc0(sizeof(*info)); 1362 1363 info->arch = g_strdup(TARGET_NAME); 1364 1365 return info; 1366 } 1367 1368 /* Stub function that's gets run on the vcpu when its brought out of the 1369 VM to run inside qemu via async_run_on_cpu()*/ 1370 static void mig_sleep_cpu(void *opq) 1371 { 1372 qemu_mutex_unlock_iothread(); 1373 g_usleep(30*1000); 1374 qemu_mutex_lock_iothread(); 1375 } 1376 1377 /* To reduce the dirty rate explicitly disallow the VCPUs from spending 1378 much time in the VM. The migration thread will try to catchup. 1379 Workload will experience a performance drop. 1380 */ 1381 static void mig_throttle_guest_down(void) 1382 { 1383 CPUState *cpu; 1384 1385 qemu_mutex_lock_iothread(); 1386 CPU_FOREACH(cpu) { 1387 async_run_on_cpu(cpu, mig_sleep_cpu, NULL); 1388 } 1389 qemu_mutex_unlock_iothread(); 1390 } 1391 1392 static void check_guest_throttling(void) 1393 { 1394 static int64_t t0; 1395 int64_t t1; 1396 1397 if (!mig_throttle_on) { 1398 return; 1399 } 1400 1401 if (!t0) { 1402 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1403 return; 1404 } 1405 1406 t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1407 1408 /* If it has been more than 40 ms since the last time the guest 1409 * was throttled then do it again. 1410 */ 1411 if (40 < (t1-t0)/1000000) { 1412 mig_throttle_guest_down(); 1413 t0 = t1; 1414 } 1415 } 1416