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