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