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 "qmp-commands.h" 49 #include "trace.h" 50 #include "exec/cpu-all.h" 51 #include "exec/ram_addr.h" 52 #include "hw/acpi/acpi.h" 53 #include "qemu/host-utils.h" 54 55 #ifdef DEBUG_ARCH_INIT 56 #define DPRINTF(fmt, ...) \ 57 do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0) 58 #else 59 #define DPRINTF(fmt, ...) \ 60 do { } while (0) 61 #endif 62 63 #ifdef TARGET_SPARC 64 int graphic_width = 1024; 65 int graphic_height = 768; 66 int graphic_depth = 8; 67 #else 68 int graphic_width = 800; 69 int graphic_height = 600; 70 int graphic_depth = 32; 71 #endif 72 73 74 #if defined(TARGET_ALPHA) 75 #define QEMU_ARCH QEMU_ARCH_ALPHA 76 #elif defined(TARGET_ARM) 77 #define QEMU_ARCH QEMU_ARCH_ARM 78 #elif defined(TARGET_CRIS) 79 #define QEMU_ARCH QEMU_ARCH_CRIS 80 #elif defined(TARGET_I386) 81 #define QEMU_ARCH QEMU_ARCH_I386 82 #elif defined(TARGET_M68K) 83 #define QEMU_ARCH QEMU_ARCH_M68K 84 #elif defined(TARGET_LM32) 85 #define QEMU_ARCH QEMU_ARCH_LM32 86 #elif defined(TARGET_MICROBLAZE) 87 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE 88 #elif defined(TARGET_MIPS) 89 #define QEMU_ARCH QEMU_ARCH_MIPS 90 #elif defined(TARGET_MOXIE) 91 #define QEMU_ARCH QEMU_ARCH_MOXIE 92 #elif defined(TARGET_OPENRISC) 93 #define QEMU_ARCH QEMU_ARCH_OPENRISC 94 #elif defined(TARGET_PPC) 95 #define QEMU_ARCH QEMU_ARCH_PPC 96 #elif defined(TARGET_S390X) 97 #define QEMU_ARCH QEMU_ARCH_S390X 98 #elif defined(TARGET_SH4) 99 #define QEMU_ARCH QEMU_ARCH_SH4 100 #elif defined(TARGET_SPARC) 101 #define QEMU_ARCH QEMU_ARCH_SPARC 102 #elif defined(TARGET_XTENSA) 103 #define QEMU_ARCH QEMU_ARCH_XTENSA 104 #elif defined(TARGET_UNICORE32) 105 #define QEMU_ARCH QEMU_ARCH_UNICORE32 106 #endif 107 108 const uint32_t arch_type = QEMU_ARCH; 109 static bool mig_throttle_on; 110 static int dirty_rate_high_cnt; 111 static void check_guest_throttling(void); 112 113 /***********************************************************/ 114 /* ram save/restore */ 115 116 #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ 117 #define RAM_SAVE_FLAG_COMPRESS 0x02 118 #define RAM_SAVE_FLAG_MEM_SIZE 0x04 119 #define RAM_SAVE_FLAG_PAGE 0x08 120 #define RAM_SAVE_FLAG_EOS 0x10 121 #define RAM_SAVE_FLAG_CONTINUE 0x20 122 #define RAM_SAVE_FLAG_XBZRLE 0x40 123 /* 0x80 is reserved in migration.h start with 0x100 next */ 124 125 static struct defconfig_file { 126 const char *filename; 127 /* Indicates it is an user config file (disabled by -no-user-config) */ 128 bool userconfig; 129 } default_config_files[] = { 130 { CONFIG_QEMU_CONFDIR "/qemu.conf", true }, 131 { CONFIG_QEMU_CONFDIR "/target-" TARGET_NAME ".conf", true }, 132 { NULL }, /* end of list */ 133 }; 134 135 static const uint8_t ZERO_TARGET_PAGE[TARGET_PAGE_SIZE]; 136 137 int qemu_read_default_config_files(bool userconfig) 138 { 139 int ret; 140 struct defconfig_file *f; 141 142 for (f = default_config_files; f->filename; f++) { 143 if (!userconfig && f->userconfig) { 144 continue; 145 } 146 ret = qemu_read_config_file(f->filename); 147 if (ret < 0 && ret != -ENOENT) { 148 return ret; 149 } 150 } 151 152 return 0; 153 } 154 155 static inline bool is_zero_range(uint8_t *p, uint64_t size) 156 { 157 return buffer_find_nonzero_offset(p, size) == size; 158 } 159 160 /* struct contains XBZRLE cache and a static page 161 used by the compression */ 162 static struct { 163 /* buffer used for XBZRLE encoding */ 164 uint8_t *encoded_buf; 165 /* buffer for storing page content */ 166 uint8_t *current_buf; 167 /* Cache for XBZRLE, Protected by lock. */ 168 PageCache *cache; 169 QemuMutex lock; 170 } XBZRLE = { 171 .encoded_buf = NULL, 172 .current_buf = NULL, 173 .cache = NULL, 174 }; 175 /* buffer used for XBZRLE decoding */ 176 static uint8_t *xbzrle_decoded_buf; 177 178 static void XBZRLE_cache_lock(void) 179 { 180 if (migrate_use_xbzrle()) 181 qemu_mutex_lock(&XBZRLE.lock); 182 } 183 184 static void XBZRLE_cache_unlock(void) 185 { 186 if (migrate_use_xbzrle()) 187 qemu_mutex_unlock(&XBZRLE.lock); 188 } 189 190 int64_t xbzrle_cache_resize(int64_t new_size) 191 { 192 PageCache *new_cache, *cache_to_free; 193 194 if (new_size < TARGET_PAGE_SIZE) { 195 return -1; 196 } 197 198 /* no need to lock, the current thread holds qemu big lock */ 199 if (XBZRLE.cache != NULL) { 200 /* check XBZRLE.cache again later */ 201 if (pow2floor(new_size) == migrate_xbzrle_cache_size()) { 202 return pow2floor(new_size); 203 } 204 new_cache = cache_init(new_size / TARGET_PAGE_SIZE, 205 TARGET_PAGE_SIZE); 206 if (!new_cache) { 207 DPRINTF("Error creating cache\n"); 208 return -1; 209 } 210 211 XBZRLE_cache_lock(); 212 /* the XBZRLE.cache may have be destroyed, check it again */ 213 if (XBZRLE.cache != NULL) { 214 cache_to_free = XBZRLE.cache; 215 XBZRLE.cache = new_cache; 216 } else { 217 cache_to_free = new_cache; 218 } 219 XBZRLE_cache_unlock(); 220 221 cache_fini(cache_to_free); 222 } 223 224 return pow2floor(new_size); 225 } 226 227 /* accounting for migration statistics */ 228 typedef struct AccountingInfo { 229 uint64_t dup_pages; 230 uint64_t skipped_pages; 231 uint64_t norm_pages; 232 uint64_t iterations; 233 uint64_t xbzrle_bytes; 234 uint64_t xbzrle_pages; 235 uint64_t xbzrle_cache_miss; 236 uint64_t xbzrle_overflows; 237 } AccountingInfo; 238 239 static AccountingInfo acct_info; 240 241 static void acct_clear(void) 242 { 243 memset(&acct_info, 0, sizeof(acct_info)); 244 } 245 246 uint64_t dup_mig_bytes_transferred(void) 247 { 248 return acct_info.dup_pages * TARGET_PAGE_SIZE; 249 } 250 251 uint64_t dup_mig_pages_transferred(void) 252 { 253 return acct_info.dup_pages; 254 } 255 256 uint64_t skipped_mig_bytes_transferred(void) 257 { 258 return acct_info.skipped_pages * TARGET_PAGE_SIZE; 259 } 260 261 uint64_t skipped_mig_pages_transferred(void) 262 { 263 return acct_info.skipped_pages; 264 } 265 266 uint64_t norm_mig_bytes_transferred(void) 267 { 268 return acct_info.norm_pages * TARGET_PAGE_SIZE; 269 } 270 271 uint64_t norm_mig_pages_transferred(void) 272 { 273 return acct_info.norm_pages; 274 } 275 276 uint64_t xbzrle_mig_bytes_transferred(void) 277 { 278 return acct_info.xbzrle_bytes; 279 } 280 281 uint64_t xbzrle_mig_pages_transferred(void) 282 { 283 return acct_info.xbzrle_pages; 284 } 285 286 uint64_t xbzrle_mig_pages_cache_miss(void) 287 { 288 return acct_info.xbzrle_cache_miss; 289 } 290 291 uint64_t xbzrle_mig_pages_overflow(void) 292 { 293 return acct_info.xbzrle_overflows; 294 } 295 296 static size_t save_block_hdr(QEMUFile *f, RAMBlock *block, ram_addr_t offset, 297 int cont, int flag) 298 { 299 size_t size; 300 301 qemu_put_be64(f, offset | cont | flag); 302 size = 8; 303 304 if (!cont) { 305 qemu_put_byte(f, strlen(block->idstr)); 306 qemu_put_buffer(f, (uint8_t *)block->idstr, 307 strlen(block->idstr)); 308 size += 1 + strlen(block->idstr); 309 } 310 return size; 311 } 312 313 /* This is the last block that we have visited serching for dirty pages 314 */ 315 static RAMBlock *last_seen_block; 316 /* This is the last block from where we have sent data */ 317 static RAMBlock *last_sent_block; 318 static ram_addr_t last_offset; 319 static unsigned long *migration_bitmap; 320 static uint64_t migration_dirty_pages; 321 static uint32_t last_version; 322 static bool ram_bulk_stage; 323 324 /* Update the xbzrle cache to reflect a page that's been sent as all 0. 325 * The important thing is that a stale (not-yet-0'd) page be replaced 326 * by the new data. 327 * As a bonus, if the page wasn't in the cache it gets added so that 328 * when a small write is made into the 0'd page it gets XBZRLE sent 329 */ 330 static void xbzrle_cache_zero_page(ram_addr_t current_addr) 331 { 332 if (ram_bulk_stage || !migrate_use_xbzrle()) { 333 return; 334 } 335 336 /* We don't care if this fails to allocate a new cache page 337 * as long as it updated an old one */ 338 cache_insert(XBZRLE.cache, current_addr, ZERO_TARGET_PAGE); 339 } 340 341 #define ENCODING_FLAG_XBZRLE 0x1 342 343 static int save_xbzrle_page(QEMUFile *f, uint8_t *current_data, 344 ram_addr_t current_addr, RAMBlock *block, 345 ram_addr_t offset, int cont, bool last_stage) 346 { 347 int encoded_len = 0, bytes_sent = -1; 348 uint8_t *prev_cached_page; 349 350 if (!cache_is_cached(XBZRLE.cache, current_addr)) { 351 if (!last_stage) { 352 if (cache_insert(XBZRLE.cache, current_addr, current_data) == -1) { 353 return -1; 354 } 355 } 356 acct_info.xbzrle_cache_miss++; 357 return -1; 358 } 359 360 prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); 361 362 /* save current buffer into memory */ 363 memcpy(XBZRLE.current_buf, current_data, TARGET_PAGE_SIZE); 364 365 /* XBZRLE encoding (if there is no overflow) */ 366 encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, 367 TARGET_PAGE_SIZE, XBZRLE.encoded_buf, 368 TARGET_PAGE_SIZE); 369 if (encoded_len == 0) { 370 DPRINTF("Skipping unmodified page\n"); 371 return 0; 372 } else if (encoded_len == -1) { 373 DPRINTF("Overflow\n"); 374 acct_info.xbzrle_overflows++; 375 /* update data in the cache */ 376 memcpy(prev_cached_page, current_data, TARGET_PAGE_SIZE); 377 return -1; 378 } 379 380 /* we need to update the data in the cache, in order to get the same data */ 381 if (!last_stage) { 382 memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); 383 } 384 385 /* Send XBZRLE based compressed page */ 386 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_XBZRLE); 387 qemu_put_byte(f, ENCODING_FLAG_XBZRLE); 388 qemu_put_be16(f, encoded_len); 389 qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len); 390 bytes_sent += encoded_len + 1 + 2; 391 acct_info.xbzrle_pages++; 392 acct_info.xbzrle_bytes += bytes_sent; 393 394 return bytes_sent; 395 } 396 397 static inline 398 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr, 399 ram_addr_t start) 400 { 401 unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS; 402 unsigned long nr = base + (start >> TARGET_PAGE_BITS); 403 uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr)); 404 unsigned long size = base + (mr_size >> TARGET_PAGE_BITS); 405 406 unsigned long next; 407 408 if (ram_bulk_stage && nr > base) { 409 next = nr + 1; 410 } else { 411 next = find_next_bit(migration_bitmap, size, nr); 412 } 413 414 if (next < size) { 415 clear_bit(next, migration_bitmap); 416 migration_dirty_pages--; 417 } 418 return (next - base) << TARGET_PAGE_BITS; 419 } 420 421 static inline bool migration_bitmap_set_dirty(ram_addr_t addr) 422 { 423 bool ret; 424 int nr = addr >> TARGET_PAGE_BITS; 425 426 ret = test_and_set_bit(nr, migration_bitmap); 427 428 if (!ret) { 429 migration_dirty_pages++; 430 } 431 return ret; 432 } 433 434 static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length) 435 { 436 ram_addr_t addr; 437 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS); 438 439 /* start address is aligned at the start of a word? */ 440 if (((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) { 441 int k; 442 int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS); 443 unsigned long *src = ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION]; 444 445 for (k = page; k < page + nr; k++) { 446 if (src[k]) { 447 unsigned long new_dirty; 448 new_dirty = ~migration_bitmap[k]; 449 migration_bitmap[k] |= src[k]; 450 new_dirty &= src[k]; 451 migration_dirty_pages += ctpopl(new_dirty); 452 src[k] = 0; 453 } 454 } 455 } else { 456 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) { 457 if (cpu_physical_memory_get_dirty(start + addr, 458 TARGET_PAGE_SIZE, 459 DIRTY_MEMORY_MIGRATION)) { 460 cpu_physical_memory_reset_dirty(start + addr, 461 TARGET_PAGE_SIZE, 462 DIRTY_MEMORY_MIGRATION); 463 migration_bitmap_set_dirty(start + addr); 464 } 465 } 466 } 467 } 468 469 470 /* Needs iothread lock! */ 471 472 static void migration_bitmap_sync(void) 473 { 474 RAMBlock *block; 475 uint64_t num_dirty_pages_init = migration_dirty_pages; 476 MigrationState *s = migrate_get_current(); 477 static int64_t start_time; 478 static int64_t bytes_xfer_prev; 479 static int64_t num_dirty_pages_period; 480 int64_t end_time; 481 int64_t bytes_xfer_now; 482 483 if (!bytes_xfer_prev) { 484 bytes_xfer_prev = ram_bytes_transferred(); 485 } 486 487 if (!start_time) { 488 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 489 } 490 491 trace_migration_bitmap_sync_start(); 492 address_space_sync_dirty_bitmap(&address_space_memory); 493 494 QTAILQ_FOREACH(block, &ram_list.blocks, next) { 495 migration_bitmap_sync_range(block->mr->ram_addr, block->length); 496 } 497 trace_migration_bitmap_sync_end(migration_dirty_pages 498 - num_dirty_pages_init); 499 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init; 500 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 501 502 /* more than 1 second = 1000 millisecons */ 503 if (end_time > start_time + 1000) { 504 if (migrate_auto_converge()) { 505 /* The following detection logic can be refined later. For now: 506 Check to see if the dirtied bytes is 50% more than the approx. 507 amount of bytes that just got transferred since the last time we 508 were in this routine. If that happens >N times (for now N==4) 509 we turn on the throttle down logic */ 510 bytes_xfer_now = ram_bytes_transferred(); 511 if (s->dirty_pages_rate && 512 (num_dirty_pages_period * TARGET_PAGE_SIZE > 513 (bytes_xfer_now - bytes_xfer_prev)/2) && 514 (dirty_rate_high_cnt++ > 4)) { 515 trace_migration_throttle(); 516 mig_throttle_on = true; 517 dirty_rate_high_cnt = 0; 518 } 519 bytes_xfer_prev = bytes_xfer_now; 520 } else { 521 mig_throttle_on = false; 522 } 523 s->dirty_pages_rate = num_dirty_pages_period * 1000 524 / (end_time - start_time); 525 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE; 526 start_time = end_time; 527 num_dirty_pages_period = 0; 528 } 529 } 530 531 /* 532 * ram_save_block: Writes a page of memory to the stream f 533 * 534 * Returns: The number of bytes written. 535 * 0 means no dirty pages 536 */ 537 538 static int ram_save_block(QEMUFile *f, bool last_stage) 539 { 540 RAMBlock *block = last_seen_block; 541 ram_addr_t offset = last_offset; 542 bool complete_round = false; 543 int bytes_sent = 0; 544 MemoryRegion *mr; 545 ram_addr_t current_addr; 546 547 if (!block) 548 block = QTAILQ_FIRST(&ram_list.blocks); 549 550 while (true) { 551 mr = block->mr; 552 offset = migration_bitmap_find_and_reset_dirty(mr, offset); 553 if (complete_round && block == last_seen_block && 554 offset >= last_offset) { 555 break; 556 } 557 if (offset >= block->length) { 558 offset = 0; 559 block = QTAILQ_NEXT(block, next); 560 if (!block) { 561 block = QTAILQ_FIRST(&ram_list.blocks); 562 complete_round = true; 563 ram_bulk_stage = false; 564 } 565 } else { 566 int ret; 567 uint8_t *p; 568 bool send_async = true; 569 int cont = (block == last_sent_block) ? 570 RAM_SAVE_FLAG_CONTINUE : 0; 571 572 p = memory_region_get_ram_ptr(mr) + offset; 573 574 /* In doubt sent page as normal */ 575 bytes_sent = -1; 576 ret = ram_control_save_page(f, block->offset, 577 offset, TARGET_PAGE_SIZE, &bytes_sent); 578 579 XBZRLE_cache_lock(); 580 581 current_addr = block->offset + offset; 582 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { 583 if (ret != RAM_SAVE_CONTROL_DELAYED) { 584 if (bytes_sent > 0) { 585 acct_info.norm_pages++; 586 } else if (bytes_sent == 0) { 587 acct_info.dup_pages++; 588 } 589 } 590 } else if (is_zero_range(p, TARGET_PAGE_SIZE)) { 591 acct_info.dup_pages++; 592 bytes_sent = save_block_hdr(f, block, offset, cont, 593 RAM_SAVE_FLAG_COMPRESS); 594 qemu_put_byte(f, 0); 595 bytes_sent++; 596 /* Must let xbzrle know, otherwise a previous (now 0'd) cached 597 * page would be stale 598 */ 599 xbzrle_cache_zero_page(current_addr); 600 } else if (!ram_bulk_stage && migrate_use_xbzrle()) { 601 bytes_sent = save_xbzrle_page(f, p, current_addr, block, 602 offset, cont, last_stage); 603 if (!last_stage) { 604 /* We must send exactly what's in the xbzrle cache 605 * even if the page wasn't xbzrle compressed, so that 606 * it's right next time. 607 */ 608 p = get_cached_data(XBZRLE.cache, current_addr); 609 610 /* Can't send this cached data async, since the cache page 611 * might get updated before it gets to the wire 612 */ 613 send_async = false; 614 } 615 } 616 617 /* XBZRLE overflow or normal page */ 618 if (bytes_sent == -1) { 619 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_PAGE); 620 if (send_async) { 621 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE); 622 } else { 623 qemu_put_buffer(f, p, TARGET_PAGE_SIZE); 624 } 625 bytes_sent += TARGET_PAGE_SIZE; 626 acct_info.norm_pages++; 627 } 628 629 XBZRLE_cache_unlock(); 630 /* if page is unmodified, continue to the next */ 631 if (bytes_sent > 0) { 632 last_sent_block = block; 633 break; 634 } 635 } 636 } 637 last_seen_block = block; 638 last_offset = offset; 639 640 return bytes_sent; 641 } 642 643 static uint64_t bytes_transferred; 644 645 void acct_update_position(QEMUFile *f, size_t size, bool zero) 646 { 647 uint64_t pages = size / TARGET_PAGE_SIZE; 648 if (zero) { 649 acct_info.dup_pages += pages; 650 } else { 651 acct_info.norm_pages += pages; 652 bytes_transferred += size; 653 qemu_update_position(f, size); 654 } 655 } 656 657 static ram_addr_t ram_save_remaining(void) 658 { 659 return migration_dirty_pages; 660 } 661 662 uint64_t ram_bytes_remaining(void) 663 { 664 return ram_save_remaining() * TARGET_PAGE_SIZE; 665 } 666 667 uint64_t ram_bytes_transferred(void) 668 { 669 return bytes_transferred; 670 } 671 672 uint64_t ram_bytes_total(void) 673 { 674 RAMBlock *block; 675 uint64_t total = 0; 676 677 QTAILQ_FOREACH(block, &ram_list.blocks, next) 678 total += block->length; 679 680 return total; 681 } 682 683 void free_xbzrle_decoded_buf(void) 684 { 685 g_free(xbzrle_decoded_buf); 686 xbzrle_decoded_buf = NULL; 687 } 688 689 static void migration_end(void) 690 { 691 if (migration_bitmap) { 692 memory_global_dirty_log_stop(); 693 g_free(migration_bitmap); 694 migration_bitmap = NULL; 695 } 696 697 XBZRLE_cache_lock(); 698 if (XBZRLE.cache) { 699 cache_fini(XBZRLE.cache); 700 g_free(XBZRLE.cache); 701 g_free(XBZRLE.encoded_buf); 702 g_free(XBZRLE.current_buf); 703 XBZRLE.cache = NULL; 704 XBZRLE.encoded_buf = NULL; 705 XBZRLE.current_buf = NULL; 706 } 707 XBZRLE_cache_unlock(); 708 } 709 710 static void ram_migration_cancel(void *opaque) 711 { 712 migration_end(); 713 } 714 715 static void reset_ram_globals(void) 716 { 717 last_seen_block = NULL; 718 last_sent_block = NULL; 719 last_offset = 0; 720 last_version = ram_list.version; 721 ram_bulk_stage = true; 722 } 723 724 #define MAX_WAIT 50 /* ms, half buffered_file limit */ 725 726 static int ram_save_setup(QEMUFile *f, void *opaque) 727 { 728 RAMBlock *block; 729 int64_t ram_bitmap_pages; /* Size of bitmap in pages, including gaps */ 730 731 mig_throttle_on = false; 732 dirty_rate_high_cnt = 0; 733 734 if (migrate_use_xbzrle()) { 735 qemu_mutex_lock_iothread(); 736 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() / 737 TARGET_PAGE_SIZE, 738 TARGET_PAGE_SIZE); 739 if (!XBZRLE.cache) { 740 qemu_mutex_unlock_iothread(); 741 DPRINTF("Error creating cache\n"); 742 return -1; 743 } 744 qemu_mutex_init(&XBZRLE.lock); 745 qemu_mutex_unlock_iothread(); 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 DPRINTF("Error allocating encoded_buf\n"); 751 return -1; 752 } 753 754 XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); 755 if (!XBZRLE.current_buf) { 756 DPRINTF("Error allocating current_buf\n"); 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 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