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