arch_init.c

   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 "hw/acpi/acpi.h"
  52
  53 #ifdef DEBUG_ARCH_INIT
  54 #define DPRINTF(fmt, ...) \
  55     do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0)
  56 #else
  57 #define DPRINTF(fmt, ...) \
  58     do { } while (0)
  59 #endif
  60
  61 #ifdef TARGET_SPARC
  62 int graphic_width = 1024;
  63 int graphic_height = 768;
  64 int graphic_depth = 8;
  65 #else
  66 int graphic_width = 800;
  67 int graphic_height = 600;
  68 int graphic_depth = 32;
  69 #endif
  70
  71
  72 #if defined(TARGET_ALPHA)
  73 #define QEMU_ARCH QEMU_ARCH_ALPHA
  74 #elif defined(TARGET_ARM)
  75 #define QEMU_ARCH QEMU_ARCH_ARM
  76 #elif defined(TARGET_CRIS)
  77 #define QEMU_ARCH QEMU_ARCH_CRIS
  78 #elif defined(TARGET_I386)
  79 #define QEMU_ARCH QEMU_ARCH_I386
  80 #elif defined(TARGET_M68K)
  81 #define QEMU_ARCH QEMU_ARCH_M68K
  82 #elif defined(TARGET_LM32)
  83 #define QEMU_ARCH QEMU_ARCH_LM32
  84 #elif defined(TARGET_MICROBLAZE)
  85 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE
  86 #elif defined(TARGET_MIPS)
  87 #define QEMU_ARCH QEMU_ARCH_MIPS
  88 #elif defined(TARGET_MOXIE)
  89 #define QEMU_ARCH QEMU_ARCH_MOXIE
  90 #elif defined(TARGET_OPENRISC)
  91 #define QEMU_ARCH QEMU_ARCH_OPENRISC
  92 #elif defined(TARGET_PPC)
  93 #define QEMU_ARCH QEMU_ARCH_PPC
  94 #elif defined(TARGET_S390X)
  95 #define QEMU_ARCH QEMU_ARCH_S390X
  96 #elif defined(TARGET_SH4)
  97 #define QEMU_ARCH QEMU_ARCH_SH4
  98 #elif defined(TARGET_SPARC)
  99 #define QEMU_ARCH QEMU_ARCH_SPARC
 100 #elif defined(TARGET_XTENSA)
 101 #define QEMU_ARCH QEMU_ARCH_XTENSA
 102 #elif defined(TARGET_UNICORE32)
 103 #define QEMU_ARCH QEMU_ARCH_UNICORE32
 104 #endif
 105
 106 const uint32_t arch_type = QEMU_ARCH;
 107 static bool mig_throttle_on;
 108 static int dirty_rate_high_cnt;
 109 static void check_guest_throttling(void);
 110
 111 /***********************************************************/
 112 /* ram save/restore */
 113
 114 #define RAM_SAVE_FLAG_FULL     0x01 /* Obsolete, not used anymore */
 115 #define RAM_SAVE_FLAG_COMPRESS 0x02
 116 #define RAM_SAVE_FLAG_MEM_SIZE 0x04
 117 #define RAM_SAVE_FLAG_PAGE     0x08
 118 #define RAM_SAVE_FLAG_EOS      0x10
 119 #define RAM_SAVE_FLAG_CONTINUE 0x20
 120 #define RAM_SAVE_FLAG_XBZRLE   0x40
 121 /* 0x80 is reserved in migration.h start with 0x100 next */
 122
 123
 124 static struct defconfig_file {
 125     const char *filename;
 126     /* Indicates it is an user config file (disabled by -no-user-config) */
 127     bool userconfig;
 128 } default_config_files[] = {
 129     { CONFIG_QEMU_CONFDIR "/qemu.conf",                   true },
 130     { CONFIG_QEMU_CONFDIR "/target-" TARGET_NAME ".conf", true },
 131     { NULL }, /* end of list */
 132 };
 133
 134
 135 int qemu_read_default_config_files(bool userconfig)
 136 {
 137     int ret;
 138     struct defconfig_file *f;
 139
 140     for (f = default_config_files; f->filename; f++) {
 141         if (!userconfig && f->userconfig) {
 142             continue;
 143         }
 144         ret = qemu_read_config_file(f->filename);
 145         if (ret < 0 && ret != -ENOENT) {
 146             return ret;
 147         }
 148     }
 149
 150     return 0;
 151 }
 152
 153 static inline bool is_zero_page(uint8_t *p)
 154 {
 155     return buffer_find_nonzero_offset(p, TARGET_PAGE_SIZE) ==
 156         TARGET_PAGE_SIZE;
 157 }
 158
 159 /* struct contains XBZRLE cache and a static page
 160    used by the compression */
 161 static struct {
 162     /* buffer used for XBZRLE encoding */
 163     uint8_t *encoded_buf;
 164     /* buffer for storing page content */
 165     uint8_t *current_buf;
 166     /* buffer used for XBZRLE decoding */
 167     uint8_t *decoded_buf;
 168     /* Cache for XBZRLE */
 169     PageCache *cache;
 170 } XBZRLE = {
 171     .encoded_buf = NULL,
 172     .current_buf = NULL,
 173     .decoded_buf = NULL,
 174     .cache = NULL,
 175 };
 176
 177
 178 int64_t xbzrle_cache_resize(int64_t new_size)
 179 {
 180     if (XBZRLE.cache != NULL) {
 181         return cache_resize(XBZRLE.cache, new_size / TARGET_PAGE_SIZE) *
 182             TARGET_PAGE_SIZE;
 183     }
 184     return pow2floor(new_size);
 185 }
 186
 187 /* accounting for migration statistics */
 188 typedef struct AccountingInfo {
 189     uint64_t dup_pages;
 190     uint64_t skipped_pages;
 191     uint64_t norm_pages;
 192     uint64_t iterations;
 193     uint64_t xbzrle_bytes;
 194     uint64_t xbzrle_pages;
 195     uint64_t xbzrle_cache_miss;
 196     uint64_t xbzrle_overflows;
 197 } AccountingInfo;
 198
 199 static AccountingInfo acct_info;
 200
 201 static void acct_clear(void)
 202 {
 203     memset(&acct_info, 0, sizeof(acct_info));
 204 }
 205
 206 uint64_t dup_mig_bytes_transferred(void)
 207 {
 208     return acct_info.dup_pages * TARGET_PAGE_SIZE;
 209 }
 210
 211 uint64_t dup_mig_pages_transferred(void)
 212 {
 213     return acct_info.dup_pages;
 214 }
 215
 216 uint64_t skipped_mig_bytes_transferred(void)
 217 {
 218     return acct_info.skipped_pages * TARGET_PAGE_SIZE;
 219 }
 220
 221 uint64_t skipped_mig_pages_transferred(void)
 222 {
 223     return acct_info.skipped_pages;
 224 }
 225
 226 uint64_t norm_mig_bytes_transferred(void)
 227 {
 228     return acct_info.norm_pages * TARGET_PAGE_SIZE;
 229 }
 230
 231 uint64_t norm_mig_pages_transferred(void)
 232 {
 233     return acct_info.norm_pages;
 234 }
 235
 236 uint64_t xbzrle_mig_bytes_transferred(void)
 237 {
 238     return acct_info.xbzrle_bytes;
 239 }
 240
 241 uint64_t xbzrle_mig_pages_transferred(void)
 242 {
 243     return acct_info.xbzrle_pages;
 244 }
 245
 246 uint64_t xbzrle_mig_pages_cache_miss(void)
 247 {
 248     return acct_info.xbzrle_cache_miss;
 249 }
 250
 251 uint64_t xbzrle_mig_pages_overflow(void)
 252 {
 253     return acct_info.xbzrle_overflows;
 254 }
 255
 256 static size_t save_block_hdr(QEMUFile *f, RAMBlock *block, ram_addr_t offset,
 257                              int cont, int flag)
 258 {
 259     size_t size;
 260
 261     qemu_put_be64(f, offset | cont | flag);
 262     size = 8;
 263
 264     if (!cont) {
 265         qemu_put_byte(f, strlen(block->idstr));
 266         qemu_put_buffer(f, (uint8_t *)block->idstr,
 267                         strlen(block->idstr));
 268         size += 1 + strlen(block->idstr);
 269     }
 270     return size;
 271 }
 272
 273 #define ENCODING_FLAG_XBZRLE 0x1
 274
 275 static int save_xbzrle_page(QEMUFile *f, uint8_t *current_data,
 276                             ram_addr_t current_addr, RAMBlock *block,
 277                             ram_addr_t offset, int cont, bool last_stage)
 278 {
 279     int encoded_len = 0, bytes_sent = -1;
 280     uint8_t *prev_cached_page;
 281
 282     if (!cache_is_cached(XBZRLE.cache, current_addr)) {
 283         if (!last_stage) {
 284             cache_insert(XBZRLE.cache, current_addr, current_data);
 285         }
 286         acct_info.xbzrle_cache_miss++;
 287         return -1;
 288     }
 289
 290     prev_cached_page = get_cached_data(XBZRLE.cache, current_addr);
 291
 292     /* save current buffer into memory */
 293     memcpy(XBZRLE.current_buf, current_data, TARGET_PAGE_SIZE);
 294
 295     /* XBZRLE encoding (if there is no overflow) */
 296     encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf,
 297                                        TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
 298                                        TARGET_PAGE_SIZE);
 299     if (encoded_len == 0) {
 300         DPRINTF("Skipping unmodified page\n");
 301         return 0;
 302     } else if (encoded_len == -1) {
 303         DPRINTF("Overflow\n");
 304         acct_info.xbzrle_overflows++;
 305         /* update data in the cache */
 306         memcpy(prev_cached_page, current_data, TARGET_PAGE_SIZE);
 307         return -1;
 308     }
 309
 310     /* we need to update the data in the cache, in order to get the same data */
 311     if (!last_stage) {
 312         memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE);
 313     }
 314
 315     /* Send XBZRLE based compressed page */
 316     bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_XBZRLE);
 317     qemu_put_byte(f, ENCODING_FLAG_XBZRLE);
 318     qemu_put_be16(f, encoded_len);
 319     qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len);
 320     bytes_sent += encoded_len + 1 + 2;
 321     acct_info.xbzrle_pages++;
 322     acct_info.xbzrle_bytes += bytes_sent;
 323
 324     return bytes_sent;
 325 }
 326
 327
 328 /* This is the last block that we have visited serching for dirty pages
 329  */
 330 static RAMBlock *last_seen_block;
 331 /* This is the last block from where we have sent data */
 332 static RAMBlock *last_sent_block;
 333 static ram_addr_t last_offset;
 334 static unsigned long *migration_bitmap;
 335 static uint64_t migration_dirty_pages;
 336 static uint32_t last_version;
 337 static bool ram_bulk_stage;
 338
 339 static inline
 340 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr,
 341                                                  ram_addr_t start)
 342 {
 343     unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS;
 344     unsigned long nr = base + (start >> TARGET_PAGE_BITS);
 345     unsigned long size = base + (int128_get64(mr->size) >> TARGET_PAGE_BITS);
 346
 347     unsigned long next;
 348
 349     if (ram_bulk_stage && nr > base) {
 350         next = nr + 1;
 351     } else {
 352         next = find_next_bit(migration_bitmap, size, nr);
 353     }
 354
 355     if (next < size) {
 356         clear_bit(next, migration_bitmap);
 357         migration_dirty_pages--;
 358     }
 359     return (next - base) << TARGET_PAGE_BITS;
 360 }
 361
 362 static inline bool migration_bitmap_set_dirty(MemoryRegion *mr,
 363                                               ram_addr_t offset)
 364 {
 365     bool ret;
 366     int nr = (mr->ram_addr + offset) >> TARGET_PAGE_BITS;
 367
 368     ret = test_and_set_bit(nr, migration_bitmap);
 369
 370     if (!ret) {
 371         migration_dirty_pages++;
 372     }
 373     return ret;
 374 }
 375
 376 /* Needs iothread lock! */
 377
 378 static void migration_bitmap_sync(void)
 379 {
 380     RAMBlock *block;
 381     ram_addr_t addr;
 382     uint64_t num_dirty_pages_init = migration_dirty_pages;
 383     MigrationState *s = migrate_get_current();
 384     static int64_t start_time;
 385     static int64_t bytes_xfer_prev;
 386     static int64_t num_dirty_pages_period;
 387     int64_t end_time;
 388     int64_t bytes_xfer_now;
 389
 390     if (!bytes_xfer_prev) {
 391         bytes_xfer_prev = ram_bytes_transferred();
 392     }
 393
 394     if (!start_time) {
 395         start_time = qemu_get_clock_ms(rt_clock);
 396     }
 397
 398     trace_migration_bitmap_sync_start();
 399     address_space_sync_dirty_bitmap(&address_space_memory);
 400
 401     QTAILQ_FOREACH(block, &ram_list.blocks, next) {
 402         for (addr = 0; addr < block->length; addr += TARGET_PAGE_SIZE) {
 403             if (memory_region_test_and_clear_dirty(block->mr,
 404                                                    addr, TARGET_PAGE_SIZE,
 405                                                    DIRTY_MEMORY_MIGRATION)) {
 406                 migration_bitmap_set_dirty(block->mr, addr);
 407             }
 408         }
 409     }
 410     trace_migration_bitmap_sync_end(migration_dirty_pages
 411                                     - num_dirty_pages_init);
 412     num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init;
 413     end_time = qemu_get_clock_ms(rt_clock);
 414
 415     /* more than 1 second = 1000 millisecons */
 416     if (end_time > start_time + 1000) {
 417         if (migrate_auto_converge()) {
 418             /* The following detection logic can be refined later. For now:
 419                Check to see if the dirtied bytes is 50% more than the approx.
 420                amount of bytes that just got transferred since the last time we
 421                were in this routine. If that happens >N times (for now N==4)
 422                we turn on the throttle down logic */
 423             bytes_xfer_now = ram_bytes_transferred();
 424             if (s->dirty_pages_rate &&
 425                (num_dirty_pages_period * TARGET_PAGE_SIZE >
 426                    (bytes_xfer_now - bytes_xfer_prev)/2) &&
 427                (dirty_rate_high_cnt++ > 4)) {
 428                     trace_migration_throttle();
 429                     mig_throttle_on = true;
 430                     dirty_rate_high_cnt = 0;
 431              }
 432              bytes_xfer_prev = bytes_xfer_now;
 433         } else {
 434              mig_throttle_on = false;
 435         }
 436         s->dirty_pages_rate = num_dirty_pages_period * 1000
 437             / (end_time - start_time);
 438         s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE;
 439         start_time = end_time;
 440         num_dirty_pages_period = 0;
 441     }
 442 }
 443
 444 /*
 445  * ram_save_block: Writes a page of memory to the stream f
 446  *
 447  * Returns:  The number of bytes written.
 448  *           0 means no dirty pages
 449  */
 450
 451 static int ram_save_block(QEMUFile *f, bool last_stage)
 452 {
 453     RAMBlock *block = last_seen_block;
 454     ram_addr_t offset = last_offset;
 455     bool complete_round = false;
 456     int bytes_sent = 0;
 457     MemoryRegion *mr;
 458     ram_addr_t current_addr;
 459
 460     if (!block)
 461         block = QTAILQ_FIRST(&ram_list.blocks);
 462
 463     while (true) {
 464         mr = block->mr;
 465         offset = migration_bitmap_find_and_reset_dirty(mr, offset);
 466         if (complete_round && block == last_seen_block &&
 467             offset >= last_offset) {
 468             break;
 469         }
 470         if (offset >= block->length) {
 471             offset = 0;
 472             block = QTAILQ_NEXT(block, next);
 473             if (!block) {
 474                 block = QTAILQ_FIRST(&ram_list.blocks);
 475                 complete_round = true;
 476                 ram_bulk_stage = false;
 477             }
 478         } else {
 479             int ret;
 480             uint8_t *p;
 481             int cont = (block == last_sent_block) ?
 482                 RAM_SAVE_FLAG_CONTINUE : 0;
 483
 484             p = memory_region_get_ram_ptr(mr) + offset;
 485
 486             /* In doubt sent page as normal */
 487             bytes_sent = -1;
 488             ret = ram_control_save_page(f, block->offset,
 489                                offset, TARGET_PAGE_SIZE, &bytes_sent);
 490
 491             if (ret != RAM_SAVE_CONTROL_NOT_SUPP) {
 492                 if (ret != RAM_SAVE_CONTROL_DELAYED) {
 493                     if (bytes_sent > 0) {
 494                         acct_info.norm_pages++;
 495                     } else if (bytes_sent == 0) {
 496                         acct_info.dup_pages++;
 497                     }
 498                 }
 499             } else if (is_zero_page(p)) {
 500                 acct_info.dup_pages++;
 501                 bytes_sent = save_block_hdr(f, block, offset, cont,
 502                                             RAM_SAVE_FLAG_COMPRESS);
 503                 qemu_put_byte(f, 0);
 504                 bytes_sent++;
 505             } else if (!ram_bulk_stage && migrate_use_xbzrle()) {
 506                 current_addr = block->offset + offset;
 507                 bytes_sent = save_xbzrle_page(f, p, current_addr, block,
 508                                               offset, cont, last_stage);
 509                 if (!last_stage) {
 510                     p = get_cached_data(XBZRLE.cache, current_addr);
 511                 }
 512             }
 513
 514             /* XBZRLE overflow or normal page */
 515             if (bytes_sent == -1) {
 516                 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_PAGE);
 517                 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE);
 518                 bytes_sent += TARGET_PAGE_SIZE;
 519                 acct_info.norm_pages++;
 520             }
 521
 522             /* if page is unmodified, continue to the next */
 523             if (bytes_sent > 0) {
 524                 last_sent_block = block;
 525                 break;
 526             }
 527         }
 528     }
 529     last_seen_block = block;
 530     last_offset = offset;
 531
 532     return bytes_sent;
 533 }
 534
 535 static uint64_t bytes_transferred;
 536
 537 void acct_update_position(QEMUFile *f, size_t size, bool zero)
 538 {
 539     uint64_t pages = size / TARGET_PAGE_SIZE;
 540     if (zero) {
 541         acct_info.dup_pages += pages;
 542     } else {
 543         acct_info.norm_pages += pages;
 544         bytes_transferred += size;
 545         qemu_update_position(f, size);
 546     }
 547 }
 548
 549 static ram_addr_t ram_save_remaining(void)
 550 {
 551     return migration_dirty_pages;
 552 }
 553
 554 uint64_t ram_bytes_remaining(void)
 555 {
 556     return ram_save_remaining() * TARGET_PAGE_SIZE;
 557 }
 558
 559 uint64_t ram_bytes_transferred(void)
 560 {
 561     return bytes_transferred;
 562 }
 563
 564 uint64_t ram_bytes_total(void)
 565 {
 566     RAMBlock *block;
 567     uint64_t total = 0;
 568
 569     QTAILQ_FOREACH(block, &ram_list.blocks, next)
 570         total += block->length;
 571
 572     return total;
 573 }
 574
 575 static void migration_end(void)
 576 {
 577     if (migration_bitmap) {
 578         memory_global_dirty_log_stop();
 579         g_free(migration_bitmap);
 580         migration_bitmap = NULL;
 581     }
 582
 583     if (XBZRLE.cache) {
 584         cache_fini(XBZRLE.cache);
 585         g_free(XBZRLE.cache);
 586         g_free(XBZRLE.encoded_buf);
 587         g_free(XBZRLE.current_buf);
 588         g_free(XBZRLE.decoded_buf);
 589         XBZRLE.cache = NULL;
 590     }
 591 }
 592
 593 static void ram_migration_cancel(void *opaque)
 594 {
 595     migration_end();
 596 }
 597
 598 static void reset_ram_globals(void)
 599 {
 600     last_seen_block = NULL;
 601     last_sent_block = NULL;
 602     last_offset = 0;
 603     last_version = ram_list.version;
 604     ram_bulk_stage = true;
 605 }
 606
 607 #define MAX_WAIT 50 /* ms, half buffered_file limit */
 608
 609 static int ram_save_setup(QEMUFile *f, void *opaque)
 610 {
 611     RAMBlock *block;
 612     int64_t ram_pages = last_ram_offset() >> TARGET_PAGE_BITS;
 613
 614     migration_bitmap = bitmap_new(ram_pages);
 615     bitmap_set(migration_bitmap, 0, ram_pages);
 616     migration_dirty_pages = ram_pages;
 617     mig_throttle_on = false;
 618     dirty_rate_high_cnt = 0;
 619
 620     if (migrate_use_xbzrle()) {
 621         XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() /
 622                                   TARGET_PAGE_SIZE,
 623                                   TARGET_PAGE_SIZE);
 624         if (!XBZRLE.cache) {
 625             DPRINTF("Error creating cache\n");
 626             return -1;
 627         }
 628         XBZRLE.encoded_buf = g_malloc0(TARGET_PAGE_SIZE);
 629         XBZRLE.current_buf = g_malloc(TARGET_PAGE_SIZE);
 630         acct_clear();
 631     }
 632
 633     qemu_mutex_lock_iothread();
 634     qemu_mutex_lock_ramlist();
 635     bytes_transferred = 0;
 636     reset_ram_globals();
 637
 638     memory_global_dirty_log_start();
 639     migration_bitmap_sync();
 640     qemu_mutex_unlock_iothread();
 641
 642     qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
 643
 644     QTAILQ_FOREACH(block, &ram_list.blocks, next) {
 645         qemu_put_byte(f, strlen(block->idstr));
 646         qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
 647         qemu_put_be64(f, block->length);
 648     }
 649
 650     qemu_mutex_unlock_ramlist();
 651
 652     ram_control_before_iterate(f, RAM_CONTROL_SETUP);
 653     ram_control_after_iterate(f, RAM_CONTROL_SETUP);
 654
 655     qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
 656
 657     return 0;
 658 }
 659
 660 static int ram_save_iterate(QEMUFile *f, void *opaque)
 661 {
 662     int ret;
 663     int i;
 664     int64_t t0;
 665     int total_sent = 0;
 666
 667     qemu_mutex_lock_ramlist();
 668
 669     if (ram_list.version != last_version) {
 670         reset_ram_globals();
 671     }
 672
 673     ram_control_before_iterate(f, RAM_CONTROL_ROUND);
 674
 675     t0 = qemu_get_clock_ns(rt_clock);
 676     i = 0;
 677     while ((ret = qemu_file_rate_limit(f)) == 0) {
 678         int bytes_sent;
 679
 680         bytes_sent = ram_save_block(f, false);
 681         /* no more blocks to sent */
 682         if (bytes_sent == 0) {
 683             break;
 684         }
 685         total_sent += bytes_sent;
 686         acct_info.iterations++;
 687         check_guest_throttling();
 688         /* we want to check in the 1st loop, just in case it was the 1st time
 689            and we had to sync the dirty bitmap.
 690            qemu_get_clock_ns() is a bit expensive, so we only check each some
 691            iterations
 692         */
 693         if ((i & 63) == 0) {
 694             uint64_t t1 = (qemu_get_clock_ns(rt_clock) - t0) / 1000000;
 695             if (t1 > MAX_WAIT) {
 696                 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n",
 697                         t1, i);
 698                 break;
 699             }
 700         }
 701         i++;
 702     }
 703
 704     qemu_mutex_unlock_ramlist();
 705
 706     /*
 707      * Must occur before EOS (or any QEMUFile operation)
 708      * because of RDMA protocol.
 709      */
 710     ram_control_after_iterate(f, RAM_CONTROL_ROUND);
 711
 712     if (ret < 0) {
 713         bytes_transferred += total_sent;
 714         return ret;
 715     }
 716
 717     qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
 718     total_sent += 8;
 719     bytes_transferred += total_sent;
 720
 721     return total_sent;
 722 }
 723
 724 static int ram_save_complete(QEMUFile *f, void *opaque)
 725 {
 726     qemu_mutex_lock_ramlist();
 727     migration_bitmap_sync();
 728
 729     ram_control_before_iterate(f, RAM_CONTROL_FINISH);
 730
 731     /* try transferring iterative blocks of memory */
 732
 733     /* flush all remaining blocks regardless of rate limiting */
 734     while (true) {
 735         int bytes_sent;
 736
 737         bytes_sent = ram_save_block(f, true);
 738         /* no more blocks to sent */
 739         if (bytes_sent == 0) {
 740             break;
 741         }
 742         bytes_transferred += bytes_sent;
 743     }
 744
 745     ram_control_after_iterate(f, RAM_CONTROL_FINISH);
 746     migration_end();
 747
 748     qemu_mutex_unlock_ramlist();
 749     qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
 750
 751     return 0;
 752 }
 753
 754 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
 755 {
 756     uint64_t remaining_size;
 757
 758     remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
 759
 760     if (remaining_size < max_size) {
 761         qemu_mutex_lock_iothread();
 762         migration_bitmap_sync();
 763         qemu_mutex_unlock_iothread();
 764         remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
 765     }
 766     return remaining_size;
 767 }
 768
 769 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host)
 770 {
 771     int ret, rc = 0;
 772     unsigned int xh_len;
 773     int xh_flags;
 774
 775     if (!XBZRLE.decoded_buf) {
 776         XBZRLE.decoded_buf = g_malloc(TARGET_PAGE_SIZE);
 777     }
 778
 779     /* extract RLE header */
 780     xh_flags = qemu_get_byte(f);
 781     xh_len = qemu_get_be16(f);
 782
 783     if (xh_flags != ENCODING_FLAG_XBZRLE) {
 784         fprintf(stderr, "Failed to load XBZRLE page - wrong compression!\n");
 785         return -1;
 786     }
 787
 788     if (xh_len > TARGET_PAGE_SIZE) {
 789         fprintf(stderr, "Failed to load XBZRLE page - len overflow!\n");
 790         return -1;
 791     }
 792     /* load data and decode */
 793     qemu_get_buffer(f, XBZRLE.decoded_buf, xh_len);
 794
 795     /* decode RLE */
 796     ret = xbzrle_decode_buffer(XBZRLE.decoded_buf, xh_len, host,
 797                                TARGET_PAGE_SIZE);
 798     if (ret == -1) {
 799         fprintf(stderr, "Failed to load XBZRLE page - decode error!\n");
 800         rc = -1;
 801     } else  if (ret > TARGET_PAGE_SIZE) {
 802         fprintf(stderr, "Failed to load XBZRLE page - size %d exceeds %d!\n",
 803                 ret, TARGET_PAGE_SIZE);
 804         abort();
 805     }
 806
 807     return rc;
 808 }
 809
 810 static inline void *host_from_stream_offset(QEMUFile *f,
 811                                             ram_addr_t offset,
 812                                             int flags)
 813 {
 814     static RAMBlock *block = NULL;
 815     char id[256];
 816     uint8_t len;
 817
 818     if (flags & RAM_SAVE_FLAG_CONTINUE) {
 819         if (!block) {
 820             fprintf(stderr, "Ack, bad migration stream!\n");
 821             return NULL;
 822         }
 823
 824         return memory_region_get_ram_ptr(block->mr) + offset;
 825     }
 826
 827     len = qemu_get_byte(f);
 828     qemu_get_buffer(f, (uint8_t *)id, len);
 829     id[len] = 0;
 830
 831     QTAILQ_FOREACH(block, &ram_list.blocks, next) {
 832         if (!strncmp(id, block->idstr, sizeof(id)))
 833             return memory_region_get_ram_ptr(block->mr) + offset;
 834     }
 835
 836     fprintf(stderr, "Can't find block %s!\n", id);
 837     return NULL;
 838 }
 839
 840 /*
 841  * If a page (or a whole RDMA chunk) has been
 842  * determined to be zero, then zap it.
 843  */
 844 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
 845 {
 846     if (ch != 0 || !is_zero_page(host)) {
 847         memset(host, ch, size);
 848 #ifndef _WIN32
 849         if (ch == 0 &&
 850             (!kvm_enabled() || kvm_has_sync_mmu()) &&
 851             getpagesize() <= TARGET_PAGE_SIZE) {
 852             qemu_madvise(host, TARGET_PAGE_SIZE, QEMU_MADV_DONTNEED);
 853         }
 854 #endif
 855     }
 856 }
 857
 858 static int ram_load(QEMUFile *f, void *opaque, int version_id)
 859 {
 860     ram_addr_t addr;
 861     int flags, ret = 0;
 862     int error;
 863     static uint64_t seq_iter;
 864
 865     seq_iter++;
 866
 867     if (version_id < 4 || version_id > 4) {
 868         return -EINVAL;
 869     }
 870
 871     do {
 872         addr = qemu_get_be64(f);
 873
 874         flags = addr & ~TARGET_PAGE_MASK;
 875         addr &= TARGET_PAGE_MASK;
 876
 877         if (flags & RAM_SAVE_FLAG_MEM_SIZE) {
 878             if (version_id == 4) {
 879                 /* Synchronize RAM block list */
 880                 char id[256];
 881                 ram_addr_t length;
 882                 ram_addr_t total_ram_bytes = addr;
 883
 884                 while (total_ram_bytes) {
 885                     RAMBlock *block;
 886                     uint8_t len;
 887
 888                     len = qemu_get_byte(f);
 889                     qemu_get_buffer(f, (uint8_t *)id, len);
 890                     id[len] = 0;
 891                     length = qemu_get_be64(f);
 892
 893                     QTAILQ_FOREACH(block, &ram_list.blocks, next) {
 894                         if (!strncmp(id, block->idstr, sizeof(id))) {
 895                             if (block->length != length) {
 896                                 fprintf(stderr,
 897                                         "Length mismatch: %s: " RAM_ADDR_FMT
 898                                         " in != " RAM_ADDR_FMT "\n", id, length,
 899                                         block->length);
 900                                 ret =  -EINVAL;
 901                                 goto done;
 902                             }
 903                             break;
 904                         }
 905                     }
 906
 907                     if (!block) {
 908                         fprintf(stderr, "Unknown ramblock \"%s\", cannot "
 909                                 "accept migration\n", id);
 910                         ret = -EINVAL;
 911                         goto done;
 912                     }
 913
 914                     total_ram_bytes -= length;
 915                 }
 916             }
 917         }
 918
 919         if (flags & RAM_SAVE_FLAG_COMPRESS) {
 920             void *host;
 921             uint8_t ch;
 922
 923             host = host_from_stream_offset(f, addr, flags);
 924             if (!host) {
 925                 return -EINVAL;
 926             }
 927
 928             ch = qemu_get_byte(f);
 929             ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
 930         } else if (flags & RAM_SAVE_FLAG_PAGE) {
 931             void *host;
 932
 933             host = host_from_stream_offset(f, addr, flags);
 934             if (!host) {
 935                 return -EINVAL;
 936             }
 937
 938             qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
 939         } else if (flags & RAM_SAVE_FLAG_XBZRLE) {
 940             void *host = host_from_stream_offset(f, addr, flags);
 941             if (!host) {
 942                 return -EINVAL;
 943             }
 944
 945             if (load_xbzrle(f, addr, host) < 0) {
 946                 ret = -EINVAL;
 947                 goto done;
 948             }
 949         } else if (flags & RAM_SAVE_FLAG_HOOK) {
 950             ram_control_load_hook(f, flags);
 951         }
 952         error = qemu_file_get_error(f);
 953         if (error) {
 954             ret = error;
 955             goto done;
 956         }
 957     } while (!(flags & RAM_SAVE_FLAG_EOS));
 958
 959 done:
 960     DPRINTF("Completed load of VM with exit code %d seq iteration "
 961             "%" PRIu64 "\n", ret, seq_iter);
 962     return ret;
 963 }
 964
 965 SaveVMHandlers savevm_ram_handlers = {
 966     .save_live_setup = ram_save_setup,
 967     .save_live_iterate = ram_save_iterate,
 968     .save_live_complete = ram_save_complete,
 969     .save_live_pending = ram_save_pending,
 970     .load_state = ram_load,
 971     .cancel = ram_migration_cancel,
 972 };
 973
 974 struct soundhw {
 975     const char *name;
 976     const char *descr;
 977     int enabled;
 978     int isa;
 979     union {
 980         int (*init_isa) (ISABus *bus);
 981         int (*init_pci) (PCIBus *bus);
 982     } init;
 983 };
 984
 985 static struct soundhw soundhw[9];
 986 static int soundhw_count;
 987
 988 void isa_register_soundhw(const char *name, const char *descr,
 989                           int (*init_isa)(ISABus *bus))
 990 {
 991     assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
 992     soundhw[soundhw_count].name = name;
 993     soundhw[soundhw_count].descr = descr;
 994     soundhw[soundhw_count].isa = 1;
 995     soundhw[soundhw_count].init.init_isa = init_isa;
 996     soundhw_count++;
 997 }
 998
 999 void pci_register_soundhw(const char *name, const char *descr,
1000                           int (*init_pci)(PCIBus *bus))
1001 {
1002     assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1003     soundhw[soundhw_count].name = name;
1004     soundhw[soundhw_count].descr = descr;
1005     soundhw[soundhw_count].isa = 0;
1006     soundhw[soundhw_count].init.init_pci = init_pci;
1007     soundhw_count++;
1008 }
1009
1010 void select_soundhw(const char *optarg)
1011 {
1012     struct soundhw *c;
1013
1014     if (is_help_option(optarg)) {
1015     show_valid_cards:
1016
1017         if (soundhw_count) {
1018              printf("Valid sound card names (comma separated):\n");
1019              for (c = soundhw; c->name; ++c) {
1020                  printf ("%-11s %s\n", c->name, c->descr);
1021              }
1022              printf("\n-soundhw all will enable all of the above\n");
1023         } else {
1024              printf("Machine has no user-selectable audio hardware "
1025                     "(it may or may not have always-present audio hardware).\n");
1026         }
1027         exit(!is_help_option(optarg));
1028     }
1029     else {
1030         size_t l;
1031         const char *p;
1032         char *e;
1033         int bad_card = 0;
1034
1035         if (!strcmp(optarg, "all")) {
1036             for (c = soundhw; c->name; ++c) {
1037                 c->enabled = 1;
1038             }
1039             return;
1040         }
1041
1042         p = optarg;
1043         while (*p) {
1044             e = strchr(p, ',');
1045             l = !e ? strlen(p) : (size_t) (e - p);
1046
1047             for (c = soundhw; c->name; ++c) {
1048                 if (!strncmp(c->name, p, l) && !c->name[l]) {
1049                     c->enabled = 1;
1050                     break;
1051                 }
1052             }
1053
1054             if (!c->name) {
1055                 if (l > 80) {
1056                     fprintf(stderr,
1057                             "Unknown sound card name (too big to show)\n");
1058                 }
1059                 else {
1060                     fprintf(stderr, "Unknown sound card name `%.*s'\n",
1061                             (int) l, p);
1062                 }
1063                 bad_card = 1;
1064             }
1065             p += l + (e != NULL);
1066         }
1067
1068         if (bad_card) {
1069             goto show_valid_cards;
1070         }
1071     }
1072 }
1073
1074 void audio_init(void)
1075 {
1076     struct soundhw *c;
1077     ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL);
1078     PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL);
1079
1080     for (c = soundhw; c->name; ++c) {
1081         if (c->enabled) {
1082             if (c->isa) {
1083                 if (!isa_bus) {
1084                     fprintf(stderr, "ISA bus not available for %s\n", c->name);
1085                     exit(1);
1086                 }
1087                 c->init.init_isa(isa_bus);
1088             } else {
1089                 if (!pci_bus) {
1090                     fprintf(stderr, "PCI bus not available for %s\n", c->name);
1091                     exit(1);
1092                 }
1093                 c->init.init_pci(pci_bus);
1094             }
1095         }
1096     }
1097 }
1098
1099 int qemu_uuid_parse(const char *str, uint8_t *uuid)
1100 {
1101     int ret;
1102
1103     if (strlen(str) != 36) {
1104         return -1;
1105     }
1106
1107     ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3],
1108                  &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9],
1109                  &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14],
1110                  &uuid[15]);
1111
1112     if (ret != 16) {
1113         return -1;
1114     }
1115 #ifdef TARGET_I386
1116     smbios_add_field(1, offsetof(struct smbios_type_1, uuid), uuid, 16);
1117 #endif
1118     return 0;
1119 }
1120
1121 void do_acpitable_option(const QemuOpts *opts)
1122 {
1123 #ifdef TARGET_I386
1124     Error *err = NULL;
1125
1126     acpi_table_add(opts, &err);
1127     if (err) {
1128         error_report("Wrong acpi table provided: %s",
1129                      error_get_pretty(err));
1130         error_free(err);
1131         exit(1);
1132     }
1133 #endif
1134 }
1135
1136 void do_smbios_option(const char *optarg)
1137 {
1138 #ifdef TARGET_I386
1139     if (smbios_entry_add(optarg) < 0) {
1140         exit(1);
1141     }
1142 #endif
1143 }
1144
1145 void cpudef_init(void)
1146 {
1147 #if defined(cpudef_setup)
1148     cpudef_setup(); /* parse cpu definitions in target config file */
1149 #endif
1150 }
1151
1152 int tcg_available(void)
1153 {
1154     return 1;
1155 }
1156
1157 int kvm_available(void)
1158 {
1159 #ifdef CONFIG_KVM
1160     return 1;
1161 #else
1162     return 0;
1163 #endif
1164 }
1165
1166 int xen_available(void)
1167 {
1168 #ifdef CONFIG_XEN
1169     return 1;
1170 #else
1171     return 0;
1172 #endif
1173 }
1174
1175
1176 TargetInfo *qmp_query_target(Error **errp)
1177 {
1178     TargetInfo *info = g_malloc0(sizeof(*info));
1179
1180     info->arch = g_strdup(TARGET_NAME);
1181
1182     return info;
1183 }
1184
1185 /* Stub function that's gets run on the vcpu when its brought out of the
1186    VM to run inside qemu via async_run_on_cpu()*/
1187 static void mig_sleep_cpu(void *opq)
1188 {
1189     qemu_mutex_unlock_iothread();
1190     g_usleep(30*1000);
1191     qemu_mutex_lock_iothread();
1192 }
1193
1194 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1195    much time in the VM. The migration thread will try to catchup.
1196    Workload will experience a performance drop.
1197 */
1198 static void mig_throttle_cpu_down(CPUState *cpu, void *data)
1199 {
1200     async_run_on_cpu(cpu, mig_sleep_cpu, NULL);
1201 }
1202
1203 static void mig_throttle_guest_down(void)
1204 {
1205     qemu_mutex_lock_iothread();
1206     qemu_for_each_cpu(mig_throttle_cpu_down, NULL);
1207     qemu_mutex_unlock_iothread();
1208 }
1209
1210 static void check_guest_throttling(void)
1211 {
1212     static int64_t t0;
1213     int64_t        t1;
1214
1215     if (!mig_throttle_on) {
1216         return;
1217     }
1218
1219     if (!t0)  {
1220         t0 = qemu_get_clock_ns(rt_clock);
1221         return;
1222     }
1223
1224     t1 = qemu_get_clock_ns(rt_clock);
1225
1226     /* If it has been more than 40 ms since the last time the guest
1227      * was throttled then do it again.
1228      */
1229     if (40 < (t1-t0)/1000000) {
1230         mig_throttle_guest_down();
1231         t0 = t1;
1232     }
1233 }