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 #include "qemu/rcu_queue.h" 56 57 #ifdef DEBUG_ARCH_INIT 58 #define DPRINTF(fmt, ...) \ 59 do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0) 60 #else 61 #define DPRINTF(fmt, ...) \ 62 do { } while (0) 63 #endif 64 65 #ifdef TARGET_SPARC 66 int graphic_width = 1024; 67 int graphic_height = 768; 68 int graphic_depth = 8; 69 #else 70 int graphic_width = 800; 71 int graphic_height = 600; 72 int graphic_depth = 32; 73 #endif 74 75 76 #if defined(TARGET_ALPHA) 77 #define QEMU_ARCH QEMU_ARCH_ALPHA 78 #elif defined(TARGET_ARM) 79 #define QEMU_ARCH QEMU_ARCH_ARM 80 #elif defined(TARGET_CRIS) 81 #define QEMU_ARCH QEMU_ARCH_CRIS 82 #elif defined(TARGET_I386) 83 #define QEMU_ARCH QEMU_ARCH_I386 84 #elif defined(TARGET_M68K) 85 #define QEMU_ARCH QEMU_ARCH_M68K 86 #elif defined(TARGET_LM32) 87 #define QEMU_ARCH QEMU_ARCH_LM32 88 #elif defined(TARGET_MICROBLAZE) 89 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE 90 #elif defined(TARGET_MIPS) 91 #define QEMU_ARCH QEMU_ARCH_MIPS 92 #elif defined(TARGET_MOXIE) 93 #define QEMU_ARCH QEMU_ARCH_MOXIE 94 #elif defined(TARGET_OPENRISC) 95 #define QEMU_ARCH QEMU_ARCH_OPENRISC 96 #elif defined(TARGET_PPC) 97 #define QEMU_ARCH QEMU_ARCH_PPC 98 #elif defined(TARGET_S390X) 99 #define QEMU_ARCH QEMU_ARCH_S390X 100 #elif defined(TARGET_SH4) 101 #define QEMU_ARCH QEMU_ARCH_SH4 102 #elif defined(TARGET_SPARC) 103 #define QEMU_ARCH QEMU_ARCH_SPARC 104 #elif defined(TARGET_XTENSA) 105 #define QEMU_ARCH QEMU_ARCH_XTENSA 106 #elif defined(TARGET_UNICORE32) 107 #define QEMU_ARCH QEMU_ARCH_UNICORE32 108 #elif defined(TARGET_TRICORE) 109 #define QEMU_ARCH QEMU_ARCH_TRICORE 110 #endif 111 112 const uint32_t arch_type = QEMU_ARCH; 113 static bool mig_throttle_on; 114 static int dirty_rate_high_cnt; 115 static void check_guest_throttling(void); 116 117 static uint64_t bitmap_sync_count; 118 119 /***********************************************************/ 120 /* ram save/restore */ 121 122 #define RAM_SAVE_FLAG_FULL 0x01 /* Obsolete, not used anymore */ 123 #define RAM_SAVE_FLAG_COMPRESS 0x02 124 #define RAM_SAVE_FLAG_MEM_SIZE 0x04 125 #define RAM_SAVE_FLAG_PAGE 0x08 126 #define RAM_SAVE_FLAG_EOS 0x10 127 #define RAM_SAVE_FLAG_CONTINUE 0x20 128 #define RAM_SAVE_FLAG_XBZRLE 0x40 129 /* 0x80 is reserved in migration.h start with 0x100 next */ 130 131 static struct defconfig_file { 132 const char *filename; 133 /* Indicates it is an user config file (disabled by -no-user-config) */ 134 bool userconfig; 135 } default_config_files[] = { 136 { CONFIG_QEMU_CONFDIR "/qemu.conf", true }, 137 { CONFIG_QEMU_CONFDIR "/target-" TARGET_NAME ".conf", true }, 138 { NULL }, /* end of list */ 139 }; 140 141 static const uint8_t ZERO_TARGET_PAGE[TARGET_PAGE_SIZE]; 142 143 int qemu_read_default_config_files(bool userconfig) 144 { 145 int ret; 146 struct defconfig_file *f; 147 148 for (f = default_config_files; f->filename; f++) { 149 if (!userconfig && f->userconfig) { 150 continue; 151 } 152 ret = qemu_read_config_file(f->filename); 153 if (ret < 0 && ret != -ENOENT) { 154 return ret; 155 } 156 } 157 158 return 0; 159 } 160 161 static inline bool is_zero_range(uint8_t *p, uint64_t size) 162 { 163 return buffer_find_nonzero_offset(p, size) == size; 164 } 165 166 /* struct contains XBZRLE cache and a static page 167 used by the compression */ 168 static struct { 169 /* buffer used for XBZRLE encoding */ 170 uint8_t *encoded_buf; 171 /* buffer for storing page content */ 172 uint8_t *current_buf; 173 /* Cache for XBZRLE, Protected by lock. */ 174 PageCache *cache; 175 QemuMutex lock; 176 } XBZRLE; 177 178 /* buffer used for XBZRLE decoding */ 179 static uint8_t *xbzrle_decoded_buf; 180 181 static void XBZRLE_cache_lock(void) 182 { 183 if (migrate_use_xbzrle()) 184 qemu_mutex_lock(&XBZRLE.lock); 185 } 186 187 static void XBZRLE_cache_unlock(void) 188 { 189 if (migrate_use_xbzrle()) 190 qemu_mutex_unlock(&XBZRLE.lock); 191 } 192 193 /* 194 * called from qmp_migrate_set_cache_size in main thread, possibly while 195 * a migration is in progress. 196 * A running migration maybe using the cache and might finish during this 197 * call, hence changes to the cache are protected by XBZRLE.lock(). 198 */ 199 int64_t xbzrle_cache_resize(int64_t new_size) 200 { 201 PageCache *new_cache; 202 int64_t ret; 203 204 if (new_size < TARGET_PAGE_SIZE) { 205 return -1; 206 } 207 208 XBZRLE_cache_lock(); 209 210 if (XBZRLE.cache != NULL) { 211 if (pow2floor(new_size) == migrate_xbzrle_cache_size()) { 212 goto out_new_size; 213 } 214 new_cache = cache_init(new_size / TARGET_PAGE_SIZE, 215 TARGET_PAGE_SIZE); 216 if (!new_cache) { 217 error_report("Error creating cache"); 218 ret = -1; 219 goto out; 220 } 221 222 cache_fini(XBZRLE.cache); 223 XBZRLE.cache = new_cache; 224 } 225 226 out_new_size: 227 ret = pow2floor(new_size); 228 out: 229 XBZRLE_cache_unlock(); 230 return ret; 231 } 232 233 /* accounting for migration statistics */ 234 typedef struct AccountingInfo { 235 uint64_t dup_pages; 236 uint64_t skipped_pages; 237 uint64_t norm_pages; 238 uint64_t iterations; 239 uint64_t xbzrle_bytes; 240 uint64_t xbzrle_pages; 241 uint64_t xbzrle_cache_miss; 242 double xbzrle_cache_miss_rate; 243 uint64_t xbzrle_overflows; 244 } AccountingInfo; 245 246 static AccountingInfo acct_info; 247 248 static void acct_clear(void) 249 { 250 memset(&acct_info, 0, sizeof(acct_info)); 251 } 252 253 uint64_t dup_mig_bytes_transferred(void) 254 { 255 return acct_info.dup_pages * TARGET_PAGE_SIZE; 256 } 257 258 uint64_t dup_mig_pages_transferred(void) 259 { 260 return acct_info.dup_pages; 261 } 262 263 uint64_t skipped_mig_bytes_transferred(void) 264 { 265 return acct_info.skipped_pages * TARGET_PAGE_SIZE; 266 } 267 268 uint64_t skipped_mig_pages_transferred(void) 269 { 270 return acct_info.skipped_pages; 271 } 272 273 uint64_t norm_mig_bytes_transferred(void) 274 { 275 return acct_info.norm_pages * TARGET_PAGE_SIZE; 276 } 277 278 uint64_t norm_mig_pages_transferred(void) 279 { 280 return acct_info.norm_pages; 281 } 282 283 uint64_t xbzrle_mig_bytes_transferred(void) 284 { 285 return acct_info.xbzrle_bytes; 286 } 287 288 uint64_t xbzrle_mig_pages_transferred(void) 289 { 290 return acct_info.xbzrle_pages; 291 } 292 293 uint64_t xbzrle_mig_pages_cache_miss(void) 294 { 295 return acct_info.xbzrle_cache_miss; 296 } 297 298 double xbzrle_mig_cache_miss_rate(void) 299 { 300 return acct_info.xbzrle_cache_miss_rate; 301 } 302 303 uint64_t xbzrle_mig_pages_overflow(void) 304 { 305 return acct_info.xbzrle_overflows; 306 } 307 308 /* This is the last block that we have visited serching for dirty pages 309 */ 310 static RAMBlock *last_seen_block; 311 /* This is the last block from where we have sent data */ 312 static RAMBlock *last_sent_block; 313 static ram_addr_t last_offset; 314 static unsigned long *migration_bitmap; 315 static uint64_t migration_dirty_pages; 316 static uint32_t last_version; 317 static bool ram_bulk_stage; 318 319 /** 320 * save_page_header: Write page header to wire 321 * 322 * If this is the 1st block, it also writes the block identification 323 * 324 * Returns: Number of bytes written 325 * 326 * @f: QEMUFile where to send the data 327 * @block: block that contains the page we want to send 328 * @offset: offset inside the block for the page 329 * in the lower bits, it contains flags 330 */ 331 static size_t save_page_header(QEMUFile *f, RAMBlock *block, ram_addr_t offset) 332 { 333 size_t size; 334 335 qemu_put_be64(f, offset); 336 size = 8; 337 338 if (!(offset & RAM_SAVE_FLAG_CONTINUE)) { 339 qemu_put_byte(f, strlen(block->idstr)); 340 qemu_put_buffer(f, (uint8_t *)block->idstr, 341 strlen(block->idstr)); 342 size += 1 + strlen(block->idstr); 343 } 344 return size; 345 } 346 347 /* Update the xbzrle cache to reflect a page that's been sent as all 0. 348 * The important thing is that a stale (not-yet-0'd) page be replaced 349 * by the new data. 350 * As a bonus, if the page wasn't in the cache it gets added so that 351 * when a small write is made into the 0'd page it gets XBZRLE sent 352 */ 353 static void xbzrle_cache_zero_page(ram_addr_t current_addr) 354 { 355 if (ram_bulk_stage || !migrate_use_xbzrle()) { 356 return; 357 } 358 359 /* We don't care if this fails to allocate a new cache page 360 * as long as it updated an old one */ 361 cache_insert(XBZRLE.cache, current_addr, ZERO_TARGET_PAGE, 362 bitmap_sync_count); 363 } 364 365 #define ENCODING_FLAG_XBZRLE 0x1 366 367 /** 368 * save_xbzrle_page: compress and send current page 369 * 370 * Returns: 1 means that we wrote the page 371 * 0 means that page is identical to the one already sent 372 * -1 means that xbzrle would be longer than normal 373 * 374 * @f: QEMUFile where to send the data 375 * @current_data: 376 * @current_addr: 377 * @block: block that contains the page we want to send 378 * @offset: offset inside the block for the page 379 * @last_stage: if we are at the completion stage 380 * @bytes_transferred: increase it with the number of transferred bytes 381 */ 382 static int save_xbzrle_page(QEMUFile *f, uint8_t **current_data, 383 ram_addr_t current_addr, RAMBlock *block, 384 ram_addr_t offset, bool last_stage, 385 uint64_t *bytes_transferred) 386 { 387 int encoded_len = 0, bytes_xbzrle; 388 uint8_t *prev_cached_page; 389 390 if (!cache_is_cached(XBZRLE.cache, current_addr, bitmap_sync_count)) { 391 acct_info.xbzrle_cache_miss++; 392 if (!last_stage) { 393 if (cache_insert(XBZRLE.cache, current_addr, *current_data, 394 bitmap_sync_count) == -1) { 395 return -1; 396 } else { 397 /* update *current_data when the page has been 398 inserted into cache */ 399 *current_data = get_cached_data(XBZRLE.cache, current_addr); 400 } 401 } 402 return -1; 403 } 404 405 prev_cached_page = get_cached_data(XBZRLE.cache, current_addr); 406 407 /* save current buffer into memory */ 408 memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE); 409 410 /* XBZRLE encoding (if there is no overflow) */ 411 encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf, 412 TARGET_PAGE_SIZE, XBZRLE.encoded_buf, 413 TARGET_PAGE_SIZE); 414 if (encoded_len == 0) { 415 DPRINTF("Skipping unmodified page\n"); 416 return 0; 417 } else if (encoded_len == -1) { 418 DPRINTF("Overflow\n"); 419 acct_info.xbzrle_overflows++; 420 /* update data in the cache */ 421 if (!last_stage) { 422 memcpy(prev_cached_page, *current_data, TARGET_PAGE_SIZE); 423 *current_data = prev_cached_page; 424 } 425 return -1; 426 } 427 428 /* we need to update the data in the cache, in order to get the same data */ 429 if (!last_stage) { 430 memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE); 431 } 432 433 /* Send XBZRLE based compressed page */ 434 bytes_xbzrle = save_page_header(f, block, offset | RAM_SAVE_FLAG_XBZRLE); 435 qemu_put_byte(f, ENCODING_FLAG_XBZRLE); 436 qemu_put_be16(f, encoded_len); 437 qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len); 438 bytes_xbzrle += encoded_len + 1 + 2; 439 acct_info.xbzrle_pages++; 440 acct_info.xbzrle_bytes += bytes_xbzrle; 441 *bytes_transferred += bytes_xbzrle; 442 443 return 1; 444 } 445 446 static inline 447 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr, 448 ram_addr_t start) 449 { 450 unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS; 451 unsigned long nr = base + (start >> TARGET_PAGE_BITS); 452 uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr)); 453 unsigned long size = base + (mr_size >> TARGET_PAGE_BITS); 454 455 unsigned long next; 456 457 if (ram_bulk_stage && nr > base) { 458 next = nr + 1; 459 } else { 460 next = find_next_bit(migration_bitmap, size, nr); 461 } 462 463 if (next < size) { 464 clear_bit(next, migration_bitmap); 465 migration_dirty_pages--; 466 } 467 return (next - base) << TARGET_PAGE_BITS; 468 } 469 470 static inline bool migration_bitmap_set_dirty(ram_addr_t addr) 471 { 472 bool ret; 473 int nr = addr >> TARGET_PAGE_BITS; 474 475 ret = test_and_set_bit(nr, migration_bitmap); 476 477 if (!ret) { 478 migration_dirty_pages++; 479 } 480 return ret; 481 } 482 483 static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length) 484 { 485 ram_addr_t addr; 486 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS); 487 488 /* start address is aligned at the start of a word? */ 489 if (((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) { 490 int k; 491 int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS); 492 unsigned long *src = ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION]; 493 494 for (k = page; k < page + nr; k++) { 495 if (src[k]) { 496 unsigned long new_dirty; 497 new_dirty = ~migration_bitmap[k]; 498 migration_bitmap[k] |= src[k]; 499 new_dirty &= src[k]; 500 migration_dirty_pages += ctpopl(new_dirty); 501 src[k] = 0; 502 } 503 } 504 } else { 505 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) { 506 if (cpu_physical_memory_get_dirty(start + addr, 507 TARGET_PAGE_SIZE, 508 DIRTY_MEMORY_MIGRATION)) { 509 cpu_physical_memory_reset_dirty(start + addr, 510 TARGET_PAGE_SIZE, 511 DIRTY_MEMORY_MIGRATION); 512 migration_bitmap_set_dirty(start + addr); 513 } 514 } 515 } 516 } 517 518 519 /* Fix me: there are too many global variables used in migration process. */ 520 static int64_t start_time; 521 static int64_t bytes_xfer_prev; 522 static int64_t num_dirty_pages_period; 523 524 static void migration_bitmap_sync_init(void) 525 { 526 start_time = 0; 527 bytes_xfer_prev = 0; 528 num_dirty_pages_period = 0; 529 } 530 531 /* Called with iothread lock held, to protect ram_list.dirty_memory[] */ 532 static void migration_bitmap_sync(void) 533 { 534 RAMBlock *block; 535 uint64_t num_dirty_pages_init = migration_dirty_pages; 536 MigrationState *s = migrate_get_current(); 537 int64_t end_time; 538 int64_t bytes_xfer_now; 539 static uint64_t xbzrle_cache_miss_prev; 540 static uint64_t iterations_prev; 541 542 bitmap_sync_count++; 543 544 if (!bytes_xfer_prev) { 545 bytes_xfer_prev = ram_bytes_transferred(); 546 } 547 548 if (!start_time) { 549 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 550 } 551 552 trace_migration_bitmap_sync_start(); 553 address_space_sync_dirty_bitmap(&address_space_memory); 554 555 rcu_read_lock(); 556 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { 557 migration_bitmap_sync_range(block->mr->ram_addr, block->used_length); 558 } 559 rcu_read_unlock(); 560 561 trace_migration_bitmap_sync_end(migration_dirty_pages 562 - num_dirty_pages_init); 563 num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init; 564 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 565 566 /* more than 1 second = 1000 millisecons */ 567 if (end_time > start_time + 1000) { 568 if (migrate_auto_converge()) { 569 /* The following detection logic can be refined later. For now: 570 Check to see if the dirtied bytes is 50% more than the approx. 571 amount of bytes that just got transferred since the last time we 572 were in this routine. If that happens >N times (for now N==4) 573 we turn on the throttle down logic */ 574 bytes_xfer_now = ram_bytes_transferred(); 575 if (s->dirty_pages_rate && 576 (num_dirty_pages_period * TARGET_PAGE_SIZE > 577 (bytes_xfer_now - bytes_xfer_prev)/2) && 578 (dirty_rate_high_cnt++ > 4)) { 579 trace_migration_throttle(); 580 mig_throttle_on = true; 581 dirty_rate_high_cnt = 0; 582 } 583 bytes_xfer_prev = bytes_xfer_now; 584 } else { 585 mig_throttle_on = false; 586 } 587 if (migrate_use_xbzrle()) { 588 if (iterations_prev != 0) { 589 acct_info.xbzrle_cache_miss_rate = 590 (double)(acct_info.xbzrle_cache_miss - 591 xbzrle_cache_miss_prev) / 592 (acct_info.iterations - iterations_prev); 593 } 594 iterations_prev = acct_info.iterations; 595 xbzrle_cache_miss_prev = acct_info.xbzrle_cache_miss; 596 } 597 s->dirty_pages_rate = num_dirty_pages_period * 1000 598 / (end_time - start_time); 599 s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE; 600 start_time = end_time; 601 num_dirty_pages_period = 0; 602 s->dirty_sync_count = bitmap_sync_count; 603 } 604 } 605 606 /** 607 * ram_save_page: Send the given page to the stream 608 * 609 * Returns: Number of pages written. 610 * 611 * @f: QEMUFile where to send the data 612 * @block: block that contains the page we want to send 613 * @offset: offset inside the block for the page 614 * @last_stage: if we are at the completion stage 615 * @bytes_transferred: increase it with the number of transferred bytes 616 */ 617 static int ram_save_page(QEMUFile *f, RAMBlock* block, ram_addr_t offset, 618 bool last_stage, uint64_t *bytes_transferred) 619 { 620 int pages = -1; 621 uint64_t bytes_xmit; 622 ram_addr_t current_addr; 623 MemoryRegion *mr = block->mr; 624 uint8_t *p; 625 int ret; 626 bool send_async = true; 627 628 p = memory_region_get_ram_ptr(mr) + offset; 629 630 /* In doubt sent page as normal */ 631 bytes_xmit = 0; 632 ret = ram_control_save_page(f, block->offset, 633 offset, TARGET_PAGE_SIZE, &bytes_xmit); 634 if (bytes_xmit) { 635 *bytes_transferred += bytes_xmit; 636 pages = 1; 637 } 638 639 XBZRLE_cache_lock(); 640 641 current_addr = block->offset + offset; 642 643 if (block == last_sent_block) { 644 offset |= RAM_SAVE_FLAG_CONTINUE; 645 } 646 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) { 647 if (ret != RAM_SAVE_CONTROL_DELAYED) { 648 if (bytes_xmit > 0) { 649 acct_info.norm_pages++; 650 } else if (bytes_xmit == 0) { 651 acct_info.dup_pages++; 652 } 653 } 654 } else if (is_zero_range(p, TARGET_PAGE_SIZE)) { 655 acct_info.dup_pages++; 656 *bytes_transferred += save_page_header(f, block, 657 offset | RAM_SAVE_FLAG_COMPRESS); 658 qemu_put_byte(f, 0); 659 *bytes_transferred += 1; 660 pages = 1; 661 /* Must let xbzrle know, otherwise a previous (now 0'd) cached 662 * page would be stale 663 */ 664 xbzrle_cache_zero_page(current_addr); 665 } else if (!ram_bulk_stage && migrate_use_xbzrle()) { 666 pages = save_xbzrle_page(f, &p, current_addr, block, 667 offset, last_stage, bytes_transferred); 668 if (!last_stage) { 669 /* Can't send this cached data async, since the cache page 670 * might get updated before it gets to the wire 671 */ 672 send_async = false; 673 } 674 } 675 676 /* XBZRLE overflow or normal page */ 677 if (pages == -1) { 678 *bytes_transferred += save_page_header(f, block, 679 offset | RAM_SAVE_FLAG_PAGE); 680 if (send_async) { 681 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE); 682 } else { 683 qemu_put_buffer(f, p, TARGET_PAGE_SIZE); 684 } 685 *bytes_transferred += TARGET_PAGE_SIZE; 686 pages = 1; 687 acct_info.norm_pages++; 688 } 689 690 XBZRLE_cache_unlock(); 691 692 return pages; 693 } 694 695 /** 696 * ram_find_and_save_block: Finds a dirty page and sends it to f 697 * 698 * Called within an RCU critical section. 699 * 700 * Returns: The number of pages written 701 * 0 means no dirty pages 702 * 703 * @f: QEMUFile where to send the data 704 * @last_stage: if we are at the completion stage 705 * @bytes_transferred: increase it with the number of transferred bytes 706 */ 707 708 static int ram_find_and_save_block(QEMUFile *f, bool last_stage, 709 uint64_t *bytes_transferred) 710 { 711 RAMBlock *block = last_seen_block; 712 ram_addr_t offset = last_offset; 713 bool complete_round = false; 714 int pages = 0; 715 MemoryRegion *mr; 716 717 if (!block) 718 block = QLIST_FIRST_RCU(&ram_list.blocks); 719 720 while (true) { 721 mr = block->mr; 722 offset = migration_bitmap_find_and_reset_dirty(mr, offset); 723 if (complete_round && block == last_seen_block && 724 offset >= last_offset) { 725 break; 726 } 727 if (offset >= block->used_length) { 728 offset = 0; 729 block = QLIST_NEXT_RCU(block, next); 730 if (!block) { 731 block = QLIST_FIRST_RCU(&ram_list.blocks); 732 complete_round = true; 733 ram_bulk_stage = false; 734 } 735 } else { 736 pages = ram_save_page(f, block, offset, last_stage, 737 bytes_transferred); 738 739 /* if page is unmodified, continue to the next */ 740 if (pages > 0) { 741 last_sent_block = block; 742 break; 743 } 744 } 745 } 746 747 last_seen_block = block; 748 last_offset = offset; 749 750 return pages; 751 } 752 753 static uint64_t bytes_transferred; 754 755 void acct_update_position(QEMUFile *f, size_t size, bool zero) 756 { 757 uint64_t pages = size / TARGET_PAGE_SIZE; 758 if (zero) { 759 acct_info.dup_pages += pages; 760 } else { 761 acct_info.norm_pages += pages; 762 bytes_transferred += size; 763 qemu_update_position(f, size); 764 } 765 } 766 767 static ram_addr_t ram_save_remaining(void) 768 { 769 return migration_dirty_pages; 770 } 771 772 uint64_t ram_bytes_remaining(void) 773 { 774 return ram_save_remaining() * TARGET_PAGE_SIZE; 775 } 776 777 uint64_t ram_bytes_transferred(void) 778 { 779 return bytes_transferred; 780 } 781 782 uint64_t ram_bytes_total(void) 783 { 784 RAMBlock *block; 785 uint64_t total = 0; 786 787 rcu_read_lock(); 788 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) 789 total += block->used_length; 790 rcu_read_unlock(); 791 return total; 792 } 793 794 void free_xbzrle_decoded_buf(void) 795 { 796 g_free(xbzrle_decoded_buf); 797 xbzrle_decoded_buf = NULL; 798 } 799 800 static void migration_end(void) 801 { 802 if (migration_bitmap) { 803 memory_global_dirty_log_stop(); 804 g_free(migration_bitmap); 805 migration_bitmap = NULL; 806 } 807 808 XBZRLE_cache_lock(); 809 if (XBZRLE.cache) { 810 cache_fini(XBZRLE.cache); 811 g_free(XBZRLE.encoded_buf); 812 g_free(XBZRLE.current_buf); 813 XBZRLE.cache = NULL; 814 XBZRLE.encoded_buf = NULL; 815 XBZRLE.current_buf = NULL; 816 } 817 XBZRLE_cache_unlock(); 818 } 819 820 static void ram_migration_cancel(void *opaque) 821 { 822 migration_end(); 823 } 824 825 static void reset_ram_globals(void) 826 { 827 last_seen_block = NULL; 828 last_sent_block = NULL; 829 last_offset = 0; 830 last_version = ram_list.version; 831 ram_bulk_stage = true; 832 } 833 834 #define MAX_WAIT 50 /* ms, half buffered_file limit */ 835 836 837 /* Each of ram_save_setup, ram_save_iterate and ram_save_complete has 838 * long-running RCU critical section. When rcu-reclaims in the code 839 * start to become numerous it will be necessary to reduce the 840 * granularity of these critical sections. 841 */ 842 843 static int ram_save_setup(QEMUFile *f, void *opaque) 844 { 845 RAMBlock *block; 846 int64_t ram_bitmap_pages; /* Size of bitmap in pages, including gaps */ 847 848 mig_throttle_on = false; 849 dirty_rate_high_cnt = 0; 850 bitmap_sync_count = 0; 851 migration_bitmap_sync_init(); 852 853 if (migrate_use_xbzrle()) { 854 XBZRLE_cache_lock(); 855 XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() / 856 TARGET_PAGE_SIZE, 857 TARGET_PAGE_SIZE); 858 if (!XBZRLE.cache) { 859 XBZRLE_cache_unlock(); 860 error_report("Error creating cache"); 861 return -1; 862 } 863 XBZRLE_cache_unlock(); 864 865 /* We prefer not to abort if there is no memory */ 866 XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); 867 if (!XBZRLE.encoded_buf) { 868 error_report("Error allocating encoded_buf"); 869 return -1; 870 } 871 872 XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE); 873 if (!XBZRLE.current_buf) { 874 error_report("Error allocating current_buf"); 875 g_free(XBZRLE.encoded_buf); 876 XBZRLE.encoded_buf = NULL; 877 return -1; 878 } 879 880 acct_clear(); 881 } 882 883 /* iothread lock needed for ram_list.dirty_memory[] */ 884 qemu_mutex_lock_iothread(); 885 qemu_mutex_lock_ramlist(); 886 rcu_read_lock(); 887 bytes_transferred = 0; 888 reset_ram_globals(); 889 890 ram_bitmap_pages = last_ram_offset() >> TARGET_PAGE_BITS; 891 migration_bitmap = bitmap_new(ram_bitmap_pages); 892 bitmap_set(migration_bitmap, 0, ram_bitmap_pages); 893 894 /* 895 * Count the total number of pages used by ram blocks not including any 896 * gaps due to alignment or unplugs. 897 */ 898 migration_dirty_pages = ram_bytes_total() >> TARGET_PAGE_BITS; 899 900 memory_global_dirty_log_start(); 901 migration_bitmap_sync(); 902 qemu_mutex_unlock_ramlist(); 903 qemu_mutex_unlock_iothread(); 904 905 qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE); 906 907 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { 908 qemu_put_byte(f, strlen(block->idstr)); 909 qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr)); 910 qemu_put_be64(f, block->used_length); 911 } 912 913 rcu_read_unlock(); 914 915 ram_control_before_iterate(f, RAM_CONTROL_SETUP); 916 ram_control_after_iterate(f, RAM_CONTROL_SETUP); 917 918 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 919 920 return 0; 921 } 922 923 static int ram_save_iterate(QEMUFile *f, void *opaque) 924 { 925 int ret; 926 int i; 927 int64_t t0; 928 int pages_sent = 0; 929 930 rcu_read_lock(); 931 if (ram_list.version != last_version) { 932 reset_ram_globals(); 933 } 934 935 /* Read version before ram_list.blocks */ 936 smp_rmb(); 937 938 ram_control_before_iterate(f, RAM_CONTROL_ROUND); 939 940 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 941 i = 0; 942 while ((ret = qemu_file_rate_limit(f)) == 0) { 943 int pages; 944 945 pages = ram_find_and_save_block(f, false, &bytes_transferred); 946 /* no more pages to sent */ 947 if (pages == 0) { 948 break; 949 } 950 pages_sent += pages; 951 acct_info.iterations++; 952 check_guest_throttling(); 953 /* we want to check in the 1st loop, just in case it was the 1st time 954 and we had to sync the dirty bitmap. 955 qemu_get_clock_ns() is a bit expensive, so we only check each some 956 iterations 957 */ 958 if ((i & 63) == 0) { 959 uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000; 960 if (t1 > MAX_WAIT) { 961 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n", 962 t1, i); 963 break; 964 } 965 } 966 i++; 967 } 968 rcu_read_unlock(); 969 970 /* 971 * Must occur before EOS (or any QEMUFile operation) 972 * because of RDMA protocol. 973 */ 974 ram_control_after_iterate(f, RAM_CONTROL_ROUND); 975 976 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 977 bytes_transferred += 8; 978 979 ret = qemu_file_get_error(f); 980 if (ret < 0) { 981 return ret; 982 } 983 984 return pages_sent; 985 } 986 987 /* Called with iothread lock */ 988 static int ram_save_complete(QEMUFile *f, void *opaque) 989 { 990 rcu_read_lock(); 991 992 migration_bitmap_sync(); 993 994 ram_control_before_iterate(f, RAM_CONTROL_FINISH); 995 996 /* try transferring iterative blocks of memory */ 997 998 /* flush all remaining blocks regardless of rate limiting */ 999 while (true) { 1000 int pages; 1001 1002 pages = ram_find_and_save_block(f, true, &bytes_transferred); 1003 /* no more blocks to sent */ 1004 if (pages == 0) { 1005 break; 1006 } 1007 } 1008 1009 ram_control_after_iterate(f, RAM_CONTROL_FINISH); 1010 migration_end(); 1011 1012 rcu_read_unlock(); 1013 qemu_put_be64(f, RAM_SAVE_FLAG_EOS); 1014 1015 return 0; 1016 } 1017 1018 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size) 1019 { 1020 uint64_t remaining_size; 1021 1022 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE; 1023 1024 if (remaining_size < max_size) { 1025 qemu_mutex_lock_iothread(); 1026 rcu_read_lock(); 1027 migration_bitmap_sync(); 1028 rcu_read_unlock(); 1029 qemu_mutex_unlock_iothread(); 1030 remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE; 1031 } 1032 return remaining_size; 1033 } 1034 1035 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host) 1036 { 1037 unsigned int xh_len; 1038 int xh_flags; 1039 1040 if (!xbzrle_decoded_buf) { 1041 xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE); 1042 } 1043 1044 /* extract RLE header */ 1045 xh_flags = qemu_get_byte(f); 1046 xh_len = qemu_get_be16(f); 1047 1048 if (xh_flags != ENCODING_FLAG_XBZRLE) { 1049 error_report("Failed to load XBZRLE page - wrong compression!"); 1050 return -1; 1051 } 1052 1053 if (xh_len > TARGET_PAGE_SIZE) { 1054 error_report("Failed to load XBZRLE page - len overflow!"); 1055 return -1; 1056 } 1057 /* load data and decode */ 1058 qemu_get_buffer(f, xbzrle_decoded_buf, xh_len); 1059 1060 /* decode RLE */ 1061 if (xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host, 1062 TARGET_PAGE_SIZE) == -1) { 1063 error_report("Failed to load XBZRLE page - decode error!"); 1064 return -1; 1065 } 1066 1067 return 0; 1068 } 1069 1070 /* Must be called from within a rcu critical section. 1071 * Returns a pointer from within the RCU-protected ram_list. 1072 */ 1073 static inline void *host_from_stream_offset(QEMUFile *f, 1074 ram_addr_t offset, 1075 int flags) 1076 { 1077 static RAMBlock *block = NULL; 1078 char id[256]; 1079 uint8_t len; 1080 1081 if (flags & RAM_SAVE_FLAG_CONTINUE) { 1082 if (!block || block->max_length <= offset) { 1083 error_report("Ack, bad migration stream!"); 1084 return NULL; 1085 } 1086 1087 return memory_region_get_ram_ptr(block->mr) + offset; 1088 } 1089 1090 len = qemu_get_byte(f); 1091 qemu_get_buffer(f, (uint8_t *)id, len); 1092 id[len] = 0; 1093 1094 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { 1095 if (!strncmp(id, block->idstr, sizeof(id)) && 1096 block->max_length > offset) { 1097 return memory_region_get_ram_ptr(block->mr) + offset; 1098 } 1099 } 1100 1101 error_report("Can't find block %s!", id); 1102 return NULL; 1103 } 1104 1105 /* 1106 * If a page (or a whole RDMA chunk) has been 1107 * determined to be zero, then zap it. 1108 */ 1109 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size) 1110 { 1111 if (ch != 0 || !is_zero_range(host, size)) { 1112 memset(host, ch, size); 1113 } 1114 } 1115 1116 static int ram_load(QEMUFile *f, void *opaque, int version_id) 1117 { 1118 int flags = 0, ret = 0; 1119 static uint64_t seq_iter; 1120 1121 seq_iter++; 1122 1123 if (version_id != 4) { 1124 ret = -EINVAL; 1125 } 1126 1127 /* This RCU critical section can be very long running. 1128 * When RCU reclaims in the code start to become numerous, 1129 * it will be necessary to reduce the granularity of this 1130 * critical section. 1131 */ 1132 rcu_read_lock(); 1133 while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) { 1134 ram_addr_t addr, total_ram_bytes; 1135 void *host; 1136 uint8_t ch; 1137 1138 addr = qemu_get_be64(f); 1139 flags = addr & ~TARGET_PAGE_MASK; 1140 addr &= TARGET_PAGE_MASK; 1141 1142 switch (flags & ~RAM_SAVE_FLAG_CONTINUE) { 1143 case RAM_SAVE_FLAG_MEM_SIZE: 1144 /* Synchronize RAM block list */ 1145 total_ram_bytes = addr; 1146 while (!ret && total_ram_bytes) { 1147 RAMBlock *block; 1148 uint8_t len; 1149 char id[256]; 1150 ram_addr_t length; 1151 1152 len = qemu_get_byte(f); 1153 qemu_get_buffer(f, (uint8_t *)id, len); 1154 id[len] = 0; 1155 length = qemu_get_be64(f); 1156 1157 QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { 1158 if (!strncmp(id, block->idstr, sizeof(id))) { 1159 if (length != block->used_length) { 1160 Error *local_err = NULL; 1161 1162 ret = qemu_ram_resize(block->offset, length, &local_err); 1163 if (local_err) { 1164 error_report_err(local_err); 1165 } 1166 } 1167 break; 1168 } 1169 } 1170 1171 if (!block) { 1172 error_report("Unknown ramblock \"%s\", cannot " 1173 "accept migration", id); 1174 ret = -EINVAL; 1175 } 1176 1177 total_ram_bytes -= length; 1178 } 1179 break; 1180 case RAM_SAVE_FLAG_COMPRESS: 1181 host = host_from_stream_offset(f, addr, flags); 1182 if (!host) { 1183 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); 1184 ret = -EINVAL; 1185 break; 1186 } 1187 ch = qemu_get_byte(f); 1188 ram_handle_compressed(host, ch, TARGET_PAGE_SIZE); 1189 break; 1190 case RAM_SAVE_FLAG_PAGE: 1191 host = host_from_stream_offset(f, addr, flags); 1192 if (!host) { 1193 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); 1194 ret = -EINVAL; 1195 break; 1196 } 1197 qemu_get_buffer(f, host, TARGET_PAGE_SIZE); 1198 break; 1199 case RAM_SAVE_FLAG_XBZRLE: 1200 host = host_from_stream_offset(f, addr, flags); 1201 if (!host) { 1202 error_report("Illegal RAM offset " RAM_ADDR_FMT, addr); 1203 ret = -EINVAL; 1204 break; 1205 } 1206 if (load_xbzrle(f, addr, host) < 0) { 1207 error_report("Failed to decompress XBZRLE page at " 1208 RAM_ADDR_FMT, addr); 1209 ret = -EINVAL; 1210 break; 1211 } 1212 break; 1213 case RAM_SAVE_FLAG_EOS: 1214 /* normal exit */ 1215 break; 1216 default: 1217 if (flags & RAM_SAVE_FLAG_HOOK) { 1218 ram_control_load_hook(f, flags); 1219 } else { 1220 error_report("Unknown combination of migration flags: %#x", 1221 flags); 1222 ret = -EINVAL; 1223 } 1224 } 1225 if (!ret) { 1226 ret = qemu_file_get_error(f); 1227 } 1228 } 1229 1230 rcu_read_unlock(); 1231 DPRINTF("Completed load of VM with exit code %d seq iteration " 1232 "%" PRIu64 "\n", ret, seq_iter); 1233 return ret; 1234 } 1235 1236 static SaveVMHandlers savevm_ram_handlers = { 1237 .save_live_setup = ram_save_setup, 1238 .save_live_iterate = ram_save_iterate, 1239 .save_live_complete = ram_save_complete, 1240 .save_live_pending = ram_save_pending, 1241 .load_state = ram_load, 1242 .cancel = ram_migration_cancel, 1243 }; 1244 1245 void ram_mig_init(void) 1246 { 1247 qemu_mutex_init(&XBZRLE.lock); 1248 register_savevm_live(NULL, "ram", 0, 4, &savevm_ram_handlers, NULL); 1249 } 1250 1251 struct soundhw { 1252 const char *name; 1253 const char *descr; 1254 int enabled; 1255 int isa; 1256 union { 1257 int (*init_isa) (ISABus *bus); 1258 int (*init_pci) (PCIBus *bus); 1259 } init; 1260 }; 1261 1262 static struct soundhw soundhw[9]; 1263 static int soundhw_count; 1264 1265 void isa_register_soundhw(const char *name, const char *descr, 1266 int (*init_isa)(ISABus *bus)) 1267 { 1268 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1); 1269 soundhw[soundhw_count].name = name; 1270 soundhw[soundhw_count].descr = descr; 1271 soundhw[soundhw_count].isa = 1; 1272 soundhw[soundhw_count].init.init_isa = init_isa; 1273 soundhw_count++; 1274 } 1275 1276 void pci_register_soundhw(const char *name, const char *descr, 1277 int (*init_pci)(PCIBus *bus)) 1278 { 1279 assert(soundhw_count < ARRAY_SIZE(soundhw) - 1); 1280 soundhw[soundhw_count].name = name; 1281 soundhw[soundhw_count].descr = descr; 1282 soundhw[soundhw_count].isa = 0; 1283 soundhw[soundhw_count].init.init_pci = init_pci; 1284 soundhw_count++; 1285 } 1286 1287 void select_soundhw(const char *optarg) 1288 { 1289 struct soundhw *c; 1290 1291 if (is_help_option(optarg)) { 1292 show_valid_cards: 1293 1294 if (soundhw_count) { 1295 printf("Valid sound card names (comma separated):\n"); 1296 for (c = soundhw; c->name; ++c) { 1297 printf ("%-11s %s\n", c->name, c->descr); 1298 } 1299 printf("\n-soundhw all will enable all of the above\n"); 1300 } else { 1301 printf("Machine has no user-selectable audio hardware " 1302 "(it may or may not have always-present audio hardware).\n"); 1303 } 1304 exit(!is_help_option(optarg)); 1305 } 1306 else { 1307 size_t l; 1308 const char *p; 1309 char *e; 1310 int bad_card = 0; 1311 1312 if (!strcmp(optarg, "all")) { 1313 for (c = soundhw; c->name; ++c) { 1314 c->enabled = 1; 1315 } 1316 return; 1317 } 1318 1319 p = optarg; 1320 while (*p) { 1321 e = strchr(p, ','); 1322 l = !e ? strlen(p) : (size_t) (e - p); 1323 1324 for (c = soundhw; c->name; ++c) { 1325 if (!strncmp(c->name, p, l) && !c->name[l]) { 1326 c->enabled = 1; 1327 break; 1328 } 1329 } 1330 1331 if (!c->name) { 1332 if (l > 80) { 1333 error_report("Unknown sound card name (too big to show)"); 1334 } 1335 else { 1336 error_report("Unknown sound card name `%.*s'", 1337 (int) l, p); 1338 } 1339 bad_card = 1; 1340 } 1341 p += l + (e != NULL); 1342 } 1343 1344 if (bad_card) { 1345 goto show_valid_cards; 1346 } 1347 } 1348 } 1349 1350 void audio_init(void) 1351 { 1352 struct soundhw *c; 1353 ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL); 1354 PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL); 1355 1356 for (c = soundhw; c->name; ++c) { 1357 if (c->enabled) { 1358 if (c->isa) { 1359 if (!isa_bus) { 1360 error_report("ISA bus not available for %s", c->name); 1361 exit(1); 1362 } 1363 c->init.init_isa(isa_bus); 1364 } else { 1365 if (!pci_bus) { 1366 error_report("PCI bus not available for %s", c->name); 1367 exit(1); 1368 } 1369 c->init.init_pci(pci_bus); 1370 } 1371 } 1372 } 1373 } 1374 1375 int qemu_uuid_parse(const char *str, uint8_t *uuid) 1376 { 1377 int ret; 1378 1379 if (strlen(str) != 36) { 1380 return -1; 1381 } 1382 1383 ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3], 1384 &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9], 1385 &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14], 1386 &uuid[15]); 1387 1388 if (ret != 16) { 1389 return -1; 1390 } 1391 return 0; 1392 } 1393 1394 void do_acpitable_option(const QemuOpts *opts) 1395 { 1396 #ifdef TARGET_I386 1397 Error *err = NULL; 1398 1399 acpi_table_add(opts, &err); 1400 if (err) { 1401 error_report("Wrong acpi table provided: %s", 1402 error_get_pretty(err)); 1403 error_free(err); 1404 exit(1); 1405 } 1406 #endif 1407 } 1408 1409 void do_smbios_option(QemuOpts *opts) 1410 { 1411 #ifdef TARGET_I386 1412 smbios_entry_add(opts); 1413 #endif 1414 } 1415 1416 void cpudef_init(void) 1417 { 1418 #if defined(cpudef_setup) 1419 cpudef_setup(); /* parse cpu definitions in target config file */ 1420 #endif 1421 } 1422 1423 int kvm_available(void) 1424 { 1425 #ifdef CONFIG_KVM 1426 return 1; 1427 #else 1428 return 0; 1429 #endif 1430 } 1431 1432 int xen_available(void) 1433 { 1434 #ifdef CONFIG_XEN 1435 return 1; 1436 #else 1437 return 0; 1438 #endif 1439 } 1440 1441 1442 TargetInfo *qmp_query_target(Error **errp) 1443 { 1444 TargetInfo *info = g_malloc0(sizeof(*info)); 1445 1446 info->arch = g_strdup(TARGET_NAME); 1447 1448 return info; 1449 } 1450 1451 /* Stub function that's gets run on the vcpu when its brought out of the 1452 VM to run inside qemu via async_run_on_cpu()*/ 1453 static void mig_sleep_cpu(void *opq) 1454 { 1455 qemu_mutex_unlock_iothread(); 1456 g_usleep(30*1000); 1457 qemu_mutex_lock_iothread(); 1458 } 1459 1460 /* To reduce the dirty rate explicitly disallow the VCPUs from spending 1461 much time in the VM. The migration thread will try to catchup. 1462 Workload will experience a performance drop. 1463 */ 1464 static void mig_throttle_guest_down(void) 1465 { 1466 CPUState *cpu; 1467 1468 qemu_mutex_lock_iothread(); 1469 CPU_FOREACH(cpu) { 1470 async_run_on_cpu(cpu, mig_sleep_cpu, NULL); 1471 } 1472 qemu_mutex_unlock_iothread(); 1473 } 1474 1475 static void check_guest_throttling(void) 1476 { 1477 static int64_t t0; 1478 int64_t t1; 1479 1480 if (!mig_throttle_on) { 1481 return; 1482 } 1483 1484 if (!t0) { 1485 t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1486 return; 1487 } 1488 1489 t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1490 1491 /* If it has been more than 40 ms since the last time the guest 1492 * was throttled then do it again. 1493 */ 1494 if (40 < (t1-t0)/1000000) { 1495 mig_throttle_guest_down(); 1496 t0 = t1; 1497 } 1498 } 1499