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 "exec/ram_addr.h"
  52 #include "hw/acpi/acpi.h"
  53 #include "qemu/host-utils.h"
  54
  55 #ifdef DEBUG_ARCH_INIT
  56 #define DPRINTF(fmt, ...) \
  57     do { fprintf(stdout, "arch_init: " fmt, ## __VA_ARGS__); } while (0)
  58 #else
  59 #define DPRINTF(fmt, ...) \
  60     do { } while (0)
  61 #endif
  62
  63 #ifdef TARGET_SPARC
  64 int graphic_width = 1024;
  65 int graphic_height = 768;
  66 int graphic_depth = 8;
  67 #else
  68 int graphic_width = 800;
  69 int graphic_height = 600;
  70 int graphic_depth = 32;
  71 #endif
  72
  73
  74 #if defined(TARGET_ALPHA)
  75 #define QEMU_ARCH QEMU_ARCH_ALPHA
  76 #elif defined(TARGET_ARM)
  77 #define QEMU_ARCH QEMU_ARCH_ARM
  78 #elif defined(TARGET_CRIS)
  79 #define QEMU_ARCH QEMU_ARCH_CRIS
  80 #elif defined(TARGET_I386)
  81 #define QEMU_ARCH QEMU_ARCH_I386
  82 #elif defined(TARGET_M68K)
  83 #define QEMU_ARCH QEMU_ARCH_M68K
  84 #elif defined(TARGET_LM32)
  85 #define QEMU_ARCH QEMU_ARCH_LM32
  86 #elif defined(TARGET_MICROBLAZE)
  87 #define QEMU_ARCH QEMU_ARCH_MICROBLAZE
  88 #elif defined(TARGET_MIPS)
  89 #define QEMU_ARCH QEMU_ARCH_MIPS
  90 #elif defined(TARGET_MOXIE)
  91 #define QEMU_ARCH QEMU_ARCH_MOXIE
  92 #elif defined(TARGET_OPENRISC)
  93 #define QEMU_ARCH QEMU_ARCH_OPENRISC
  94 #elif defined(TARGET_PPC)
  95 #define QEMU_ARCH QEMU_ARCH_PPC
  96 #elif defined(TARGET_S390X)
  97 #define QEMU_ARCH QEMU_ARCH_S390X
  98 #elif defined(TARGET_SH4)
  99 #define QEMU_ARCH QEMU_ARCH_SH4
 100 #elif defined(TARGET_SPARC)
 101 #define QEMU_ARCH QEMU_ARCH_SPARC
 102 #elif defined(TARGET_XTENSA)
 103 #define QEMU_ARCH QEMU_ARCH_XTENSA
 104 #elif defined(TARGET_UNICORE32)
 105 #define QEMU_ARCH QEMU_ARCH_UNICORE32
 106 #endif
 107
 108 const uint32_t arch_type = QEMU_ARCH;
 109 static bool mig_throttle_on;
 110 static int dirty_rate_high_cnt;
 111 static void check_guest_throttling(void);
 112
 113 /***********************************************************/
 114 /* ram save/restore */
 115
 116 #define RAM_SAVE_FLAG_FULL     0x01 /* Obsolete, not used anymore */
 117 #define RAM_SAVE_FLAG_COMPRESS 0x02
 118 #define RAM_SAVE_FLAG_MEM_SIZE 0x04
 119 #define RAM_SAVE_FLAG_PAGE     0x08
 120 #define RAM_SAVE_FLAG_EOS      0x10
 121 #define RAM_SAVE_FLAG_CONTINUE 0x20
 122 #define RAM_SAVE_FLAG_XBZRLE   0x40
 123 /* 0x80 is reserved in migration.h start with 0x100 next */
 124
 125
 126 static struct defconfig_file {
 127     const char *filename;
 128     /* Indicates it is an user config file (disabled by -no-user-config) */
 129     bool userconfig;
 130 } default_config_files[] = {
 131     { CONFIG_QEMU_CONFDIR "/qemu.conf",                   true },
 132     { CONFIG_QEMU_CONFDIR "/target-" TARGET_NAME ".conf", true },
 133     { NULL }, /* end of list */
 134 };
 135
 136
 137 int qemu_read_default_config_files(bool userconfig)
 138 {
 139     int ret;
 140     struct defconfig_file *f;
 141
 142     for (f = default_config_files; f->filename; f++) {
 143         if (!userconfig && f->userconfig) {
 144             continue;
 145         }
 146         ret = qemu_read_config_file(f->filename);
 147         if (ret < 0 && ret != -ENOENT) {
 148             return ret;
 149         }
 150     }
 151
 152     return 0;
 153 }
 154
 155 static inline bool is_zero_range(uint8_t *p, uint64_t size)
 156 {
 157     return buffer_find_nonzero_offset(p, size) == size;
 158 }
 159
 160 /* struct contains XBZRLE cache and a static page
 161    used by the compression */
 162 static struct {
 163     /* buffer used for XBZRLE encoding */
 164     uint8_t *encoded_buf;
 165     /* buffer for storing page content */
 166     uint8_t *current_buf;
 167     /* Cache for XBZRLE */
 168     PageCache *cache;
 169 } XBZRLE = {
 170     .encoded_buf = NULL,
 171     .current_buf = NULL,
 172     .cache = NULL,
 173 };
 174 /* buffer used for XBZRLE decoding */
 175 static uint8_t *xbzrle_decoded_buf;
 176
 177 int64_t xbzrle_cache_resize(int64_t new_size)
 178 {
 179     if (new_size < TARGET_PAGE_SIZE) {
 180         return -1;
 181     }
 182
 183     if (XBZRLE.cache != NULL) {
 184         return cache_resize(XBZRLE.cache, new_size / TARGET_PAGE_SIZE) *
 185             TARGET_PAGE_SIZE;
 186     }
 187     return pow2floor(new_size);
 188 }
 189
 190 /* accounting for migration statistics */
 191 typedef struct AccountingInfo {
 192     uint64_t dup_pages;
 193     uint64_t skipped_pages;
 194     uint64_t norm_pages;
 195     uint64_t iterations;
 196     uint64_t xbzrle_bytes;
 197     uint64_t xbzrle_pages;
 198     uint64_t xbzrle_cache_miss;
 199     uint64_t xbzrle_overflows;
 200 } AccountingInfo;
 201
 202 static AccountingInfo acct_info;
 203
 204 static void acct_clear(void)
 205 {
 206     memset(&acct_info, 0, sizeof(acct_info));
 207 }
 208
 209 uint64_t dup_mig_bytes_transferred(void)
 210 {
 211     return acct_info.dup_pages * TARGET_PAGE_SIZE;
 212 }
 213
 214 uint64_t dup_mig_pages_transferred(void)
 215 {
 216     return acct_info.dup_pages;
 217 }
 218
 219 uint64_t skipped_mig_bytes_transferred(void)
 220 {
 221     return acct_info.skipped_pages * TARGET_PAGE_SIZE;
 222 }
 223
 224 uint64_t skipped_mig_pages_transferred(void)
 225 {
 226     return acct_info.skipped_pages;
 227 }
 228
 229 uint64_t norm_mig_bytes_transferred(void)
 230 {
 231     return acct_info.norm_pages * TARGET_PAGE_SIZE;
 232 }
 233
 234 uint64_t norm_mig_pages_transferred(void)
 235 {
 236     return acct_info.norm_pages;
 237 }
 238
 239 uint64_t xbzrle_mig_bytes_transferred(void)
 240 {
 241     return acct_info.xbzrle_bytes;
 242 }
 243
 244 uint64_t xbzrle_mig_pages_transferred(void)
 245 {
 246     return acct_info.xbzrle_pages;
 247 }
 248
 249 uint64_t xbzrle_mig_pages_cache_miss(void)
 250 {
 251     return acct_info.xbzrle_cache_miss;
 252 }
 253
 254 uint64_t xbzrle_mig_pages_overflow(void)
 255 {
 256     return acct_info.xbzrle_overflows;
 257 }
 258
 259 static size_t save_block_hdr(QEMUFile *f, RAMBlock *block, ram_addr_t offset,
 260                              int cont, int flag)
 261 {
 262     size_t size;
 263
 264     qemu_put_be64(f, offset | cont | flag);
 265     size = 8;
 266
 267     if (!cont) {
 268         qemu_put_byte(f, strlen(block->idstr));
 269         qemu_put_buffer(f, (uint8_t *)block->idstr,
 270                         strlen(block->idstr));
 271         size += 1 + strlen(block->idstr);
 272     }
 273     return size;
 274 }
 275
 276 #define ENCODING_FLAG_XBZRLE 0x1
 277
 278 static int save_xbzrle_page(QEMUFile *f, uint8_t *current_data,
 279                             ram_addr_t current_addr, RAMBlock *block,
 280                             ram_addr_t offset, int cont, bool last_stage)
 281 {
 282     int encoded_len = 0, bytes_sent = -1;
 283     uint8_t *prev_cached_page;
 284
 285     if (!cache_is_cached(XBZRLE.cache, current_addr)) {
 286         if (!last_stage) {
 287             if (cache_insert(XBZRLE.cache, current_addr, current_data) == -1) {
 288                 return -1;
 289             }
 290         }
 291         acct_info.xbzrle_cache_miss++;
 292         return -1;
 293     }
 294
 295     prev_cached_page = get_cached_data(XBZRLE.cache, current_addr);
 296
 297     /* save current buffer into memory */
 298     memcpy(XBZRLE.current_buf, current_data, TARGET_PAGE_SIZE);
 299
 300     /* XBZRLE encoding (if there is no overflow) */
 301     encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf,
 302                                        TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
 303                                        TARGET_PAGE_SIZE);
 304     if (encoded_len == 0) {
 305         DPRINTF("Skipping unmodified page\n");
 306         return 0;
 307     } else if (encoded_len == -1) {
 308         DPRINTF("Overflow\n");
 309         acct_info.xbzrle_overflows++;
 310         /* update data in the cache */
 311         memcpy(prev_cached_page, current_data, TARGET_PAGE_SIZE);
 312         return -1;
 313     }
 314
 315     /* we need to update the data in the cache, in order to get the same data */
 316     if (!last_stage) {
 317         memcpy(prev_cached_page, XBZRLE.current_buf, TARGET_PAGE_SIZE);
 318     }
 319
 320     /* Send XBZRLE based compressed page */
 321     bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_XBZRLE);
 322     qemu_put_byte(f, ENCODING_FLAG_XBZRLE);
 323     qemu_put_be16(f, encoded_len);
 324     qemu_put_buffer(f, XBZRLE.encoded_buf, encoded_len);
 325     bytes_sent += encoded_len + 1 + 2;
 326     acct_info.xbzrle_pages++;
 327     acct_info.xbzrle_bytes += bytes_sent;
 328
 329     return bytes_sent;
 330 }
 331
 332
 333 /* This is the last block that we have visited serching for dirty pages
 334  */
 335 static RAMBlock *last_seen_block;
 336 /* This is the last block from where we have sent data */
 337 static RAMBlock *last_sent_block;
 338 static ram_addr_t last_offset;
 339 static unsigned long *migration_bitmap;
 340 static uint64_t migration_dirty_pages;
 341 static uint32_t last_version;
 342 static bool ram_bulk_stage;
 343
 344 static inline
 345 ram_addr_t migration_bitmap_find_and_reset_dirty(MemoryRegion *mr,
 346                                                  ram_addr_t start)
 347 {
 348     unsigned long base = mr->ram_addr >> TARGET_PAGE_BITS;
 349     unsigned long nr = base + (start >> TARGET_PAGE_BITS);
 350     uint64_t mr_size = TARGET_PAGE_ALIGN(memory_region_size(mr));
 351     unsigned long size = base + (mr_size >> TARGET_PAGE_BITS);
 352
 353     unsigned long next;
 354
 355     if (ram_bulk_stage && nr > base) {
 356         next = nr + 1;
 357     } else {
 358         next = find_next_bit(migration_bitmap, size, nr);
 359     }
 360
 361     if (next < size) {
 362         clear_bit(next, migration_bitmap);
 363         migration_dirty_pages--;
 364     }
 365     return (next - base) << TARGET_PAGE_BITS;
 366 }
 367
 368 static inline bool migration_bitmap_set_dirty(ram_addr_t addr)
 369 {
 370     bool ret;
 371     int nr = addr >> TARGET_PAGE_BITS;
 372
 373     ret = test_and_set_bit(nr, migration_bitmap);
 374
 375     if (!ret) {
 376         migration_dirty_pages++;
 377     }
 378     return ret;
 379 }
 380
 381 static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length)
 382 {
 383     ram_addr_t addr;
 384     unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
 385
 386     /* start address is aligned at the start of a word? */
 387     if (((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) {
 388         int k;
 389         int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS);
 390         unsigned long *src = ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION];
 391
 392         for (k = page; k < page + nr; k++) {
 393             if (src[k]) {
 394                 unsigned long new_dirty;
 395                 new_dirty = ~migration_bitmap[k];
 396                 migration_bitmap[k] |= src[k];
 397                 new_dirty &= src[k];
 398                 migration_dirty_pages += ctpopl(new_dirty);
 399                 src[k] = 0;
 400             }
 401         }
 402     } else {
 403         for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
 404             if (cpu_physical_memory_get_dirty(start + addr,
 405                                               TARGET_PAGE_SIZE,
 406                                               DIRTY_MEMORY_MIGRATION)) {
 407                 cpu_physical_memory_reset_dirty(start + addr,
 408                                                 TARGET_PAGE_SIZE,
 409                                                 DIRTY_MEMORY_MIGRATION);
 410                 migration_bitmap_set_dirty(start + addr);
 411             }
 412         }
 413     }
 414 }
 415
 416
 417 /* Needs iothread lock! */
 418
 419 static void migration_bitmap_sync(void)
 420 {
 421     RAMBlock *block;
 422     uint64_t num_dirty_pages_init = migration_dirty_pages;
 423     MigrationState *s = migrate_get_current();
 424     static int64_t start_time;
 425     static int64_t bytes_xfer_prev;
 426     static int64_t num_dirty_pages_period;
 427     int64_t end_time;
 428     int64_t bytes_xfer_now;
 429
 430     if (!bytes_xfer_prev) {
 431         bytes_xfer_prev = ram_bytes_transferred();
 432     }
 433
 434     if (!start_time) {
 435         start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 436     }
 437
 438     trace_migration_bitmap_sync_start();
 439     address_space_sync_dirty_bitmap(&address_space_memory);
 440
 441     QTAILQ_FOREACH(block, &ram_list.blocks, next) {
 442         migration_bitmap_sync_range(block->mr->ram_addr, block->length);
 443     }
 444     trace_migration_bitmap_sync_end(migration_dirty_pages
 445                                     - num_dirty_pages_init);
 446     num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init;
 447     end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 448
 449     /* more than 1 second = 1000 millisecons */
 450     if (end_time > start_time + 1000) {
 451         if (migrate_auto_converge()) {
 452             /* The following detection logic can be refined later. For now:
 453                Check to see if the dirtied bytes is 50% more than the approx.
 454                amount of bytes that just got transferred since the last time we
 455                were in this routine. If that happens >N times (for now N==4)
 456                we turn on the throttle down logic */
 457             bytes_xfer_now = ram_bytes_transferred();
 458             if (s->dirty_pages_rate &&
 459                (num_dirty_pages_period * TARGET_PAGE_SIZE >
 460                    (bytes_xfer_now - bytes_xfer_prev)/2) &&
 461                (dirty_rate_high_cnt++ > 4)) {
 462                     trace_migration_throttle();
 463                     mig_throttle_on = true;
 464                     dirty_rate_high_cnt = 0;
 465              }
 466              bytes_xfer_prev = bytes_xfer_now;
 467         } else {
 468              mig_throttle_on = false;
 469         }
 470         s->dirty_pages_rate = num_dirty_pages_period * 1000
 471             / (end_time - start_time);
 472         s->dirty_bytes_rate = s->dirty_pages_rate * TARGET_PAGE_SIZE;
 473         start_time = end_time;
 474         num_dirty_pages_period = 0;
 475     }
 476 }
 477
 478 /*
 479  * ram_save_block: Writes a page of memory to the stream f
 480  *
 481  * Returns:  The number of bytes written.
 482  *           0 means no dirty pages
 483  */
 484
 485 static int ram_save_block(QEMUFile *f, bool last_stage)
 486 {
 487     RAMBlock *block = last_seen_block;
 488     ram_addr_t offset = last_offset;
 489     bool complete_round = false;
 490     int bytes_sent = 0;
 491     MemoryRegion *mr;
 492     ram_addr_t current_addr;
 493
 494     if (!block)
 495         block = QTAILQ_FIRST(&ram_list.blocks);
 496
 497     while (true) {
 498         mr = block->mr;
 499         offset = migration_bitmap_find_and_reset_dirty(mr, offset);
 500         if (complete_round && block == last_seen_block &&
 501             offset >= last_offset) {
 502             break;
 503         }
 504         if (offset >= block->length) {
 505             offset = 0;
 506             block = QTAILQ_NEXT(block, next);
 507             if (!block) {
 508                 block = QTAILQ_FIRST(&ram_list.blocks);
 509                 complete_round = true;
 510                 ram_bulk_stage = false;
 511             }
 512         } else {
 513             int ret;
 514             uint8_t *p;
 515             int cont = (block == last_sent_block) ?
 516                 RAM_SAVE_FLAG_CONTINUE : 0;
 517
 518             p = memory_region_get_ram_ptr(mr) + offset;
 519
 520             /* In doubt sent page as normal */
 521             bytes_sent = -1;
 522             ret = ram_control_save_page(f, block->offset,
 523                                offset, TARGET_PAGE_SIZE, &bytes_sent);
 524
 525             if (ret != RAM_SAVE_CONTROL_NOT_SUPP) {
 526                 if (ret != RAM_SAVE_CONTROL_DELAYED) {
 527                     if (bytes_sent > 0) {
 528                         acct_info.norm_pages++;
 529                     } else if (bytes_sent == 0) {
 530                         acct_info.dup_pages++;
 531                     }
 532                 }
 533             } else if (is_zero_range(p, TARGET_PAGE_SIZE)) {
 534                 acct_info.dup_pages++;
 535                 bytes_sent = save_block_hdr(f, block, offset, cont,
 536                                             RAM_SAVE_FLAG_COMPRESS);
 537                 qemu_put_byte(f, 0);
 538                 bytes_sent++;
 539             } else if (!ram_bulk_stage && migrate_use_xbzrle()) {
 540                 current_addr = block->offset + offset;
 541                 bytes_sent = save_xbzrle_page(f, p, current_addr, block,
 542                                               offset, cont, last_stage);
 543                 if (!last_stage) {
 544                     p = get_cached_data(XBZRLE.cache, current_addr);
 545                 }
 546             }
 547
 548             /* XBZRLE overflow or normal page */
 549             if (bytes_sent == -1) {
 550                 bytes_sent = save_block_hdr(f, block, offset, cont, RAM_SAVE_FLAG_PAGE);
 551                 qemu_put_buffer_async(f, p, TARGET_PAGE_SIZE);
 552                 bytes_sent += TARGET_PAGE_SIZE;
 553                 acct_info.norm_pages++;
 554             }
 555
 556             /* if page is unmodified, continue to the next */
 557             if (bytes_sent > 0) {
 558                 last_sent_block = block;
 559                 break;
 560             }
 561         }
 562     }
 563     last_seen_block = block;
 564     last_offset = offset;
 565
 566     return bytes_sent;
 567 }
 568
 569 static uint64_t bytes_transferred;
 570
 571 void acct_update_position(QEMUFile *f, size_t size, bool zero)
 572 {
 573     uint64_t pages = size / TARGET_PAGE_SIZE;
 574     if (zero) {
 575         acct_info.dup_pages += pages;
 576     } else {
 577         acct_info.norm_pages += pages;
 578         bytes_transferred += size;
 579         qemu_update_position(f, size);
 580     }
 581 }
 582
 583 static ram_addr_t ram_save_remaining(void)
 584 {
 585     return migration_dirty_pages;
 586 }
 587
 588 uint64_t ram_bytes_remaining(void)
 589 {
 590     return ram_save_remaining() * TARGET_PAGE_SIZE;
 591 }
 592
 593 uint64_t ram_bytes_transferred(void)
 594 {
 595     return bytes_transferred;
 596 }
 597
 598 uint64_t ram_bytes_total(void)
 599 {
 600     RAMBlock *block;
 601     uint64_t total = 0;
 602
 603     QTAILQ_FOREACH(block, &ram_list.blocks, next)
 604         total += block->length;
 605
 606     return total;
 607 }
 608
 609 void free_xbzrle_decoded_buf(void)
 610 {
 611     g_free(xbzrle_decoded_buf);
 612     xbzrle_decoded_buf = NULL;
 613 }
 614
 615 static void migration_end(void)
 616 {
 617     if (migration_bitmap) {
 618         memory_global_dirty_log_stop();
 619         g_free(migration_bitmap);
 620         migration_bitmap = NULL;
 621     }
 622
 623     if (XBZRLE.cache) {
 624         cache_fini(XBZRLE.cache);
 625         g_free(XBZRLE.cache);
 626         g_free(XBZRLE.encoded_buf);
 627         g_free(XBZRLE.current_buf);
 628         XBZRLE.cache = NULL;
 629         XBZRLE.encoded_buf = NULL;
 630         XBZRLE.current_buf = NULL;
 631     }
 632 }
 633
 634 static void ram_migration_cancel(void *opaque)
 635 {
 636     migration_end();
 637 }
 638
 639 static void reset_ram_globals(void)
 640 {
 641     last_seen_block = NULL;
 642     last_sent_block = NULL;
 643     last_offset = 0;
 644     last_version = ram_list.version;
 645     ram_bulk_stage = true;
 646 }
 647
 648 #define MAX_WAIT 50 /* ms, half buffered_file limit */
 649
 650 static int ram_save_setup(QEMUFile *f, void *opaque)
 651 {
 652     RAMBlock *block;
 653     int64_t ram_pages = last_ram_offset() >> TARGET_PAGE_BITS;
 654
 655     migration_bitmap = bitmap_new(ram_pages);
 656     bitmap_set(migration_bitmap, 0, ram_pages);
 657     migration_dirty_pages = ram_pages;
 658     mig_throttle_on = false;
 659     dirty_rate_high_cnt = 0;
 660
 661     if (migrate_use_xbzrle()) {
 662         XBZRLE.cache = cache_init(migrate_xbzrle_cache_size() /
 663                                   TARGET_PAGE_SIZE,
 664                                   TARGET_PAGE_SIZE);
 665         if (!XBZRLE.cache) {
 666             DPRINTF("Error creating cache\n");
 667             return -1;
 668         }
 669
 670         /* We prefer not to abort if there is no memory */
 671         XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE);
 672         if (!XBZRLE.encoded_buf) {
 673             DPRINTF("Error allocating encoded_buf\n");
 674             return -1;
 675         }
 676
 677         XBZRLE.current_buf = g_try_malloc(TARGET_PAGE_SIZE);
 678         if (!XBZRLE.current_buf) {
 679             DPRINTF("Error allocating current_buf\n");
 680             g_free(XBZRLE.encoded_buf);
 681             XBZRLE.encoded_buf = NULL;
 682             return -1;
 683         }
 684
 685         acct_clear();
 686     }
 687
 688     qemu_mutex_lock_iothread();
 689     qemu_mutex_lock_ramlist();
 690     bytes_transferred = 0;
 691     reset_ram_globals();
 692
 693     memory_global_dirty_log_start();
 694     migration_bitmap_sync();
 695     qemu_mutex_unlock_iothread();
 696
 697     qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
 698
 699     QTAILQ_FOREACH(block, &ram_list.blocks, next) {
 700         qemu_put_byte(f, strlen(block->idstr));
 701         qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
 702         qemu_put_be64(f, block->length);
 703     }
 704
 705     qemu_mutex_unlock_ramlist();
 706
 707     ram_control_before_iterate(f, RAM_CONTROL_SETUP);
 708     ram_control_after_iterate(f, RAM_CONTROL_SETUP);
 709
 710     qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
 711
 712     return 0;
 713 }
 714
 715 static int ram_save_iterate(QEMUFile *f, void *opaque)
 716 {
 717     int ret;
 718     int i;
 719     int64_t t0;
 720     int total_sent = 0;
 721
 722     qemu_mutex_lock_ramlist();
 723
 724     if (ram_list.version != last_version) {
 725         reset_ram_globals();
 726     }
 727
 728     ram_control_before_iterate(f, RAM_CONTROL_ROUND);
 729
 730     t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
 731     i = 0;
 732     while ((ret = qemu_file_rate_limit(f)) == 0) {
 733         int bytes_sent;
 734
 735         bytes_sent = ram_save_block(f, false);
 736         /* no more blocks to sent */
 737         if (bytes_sent == 0) {
 738             break;
 739         }
 740         total_sent += bytes_sent;
 741         acct_info.iterations++;
 742         check_guest_throttling();
 743         /* we want to check in the 1st loop, just in case it was the 1st time
 744            and we had to sync the dirty bitmap.
 745            qemu_get_clock_ns() is a bit expensive, so we only check each some
 746            iterations
 747         */
 748         if ((i & 63) == 0) {
 749             uint64_t t1 = (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - t0) / 1000000;
 750             if (t1 > MAX_WAIT) {
 751                 DPRINTF("big wait: %" PRIu64 " milliseconds, %d iterations\n",
 752                         t1, i);
 753                 break;
 754             }
 755         }
 756         i++;
 757     }
 758
 759     qemu_mutex_unlock_ramlist();
 760
 761     /*
 762      * Must occur before EOS (or any QEMUFile operation)
 763      * because of RDMA protocol.
 764      */
 765     ram_control_after_iterate(f, RAM_CONTROL_ROUND);
 766
 767     bytes_transferred += total_sent;
 768
 769     /*
 770      * Do not count these 8 bytes into total_sent, so that we can
 771      * return 0 if no page had been dirtied.
 772      */
 773     qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
 774     bytes_transferred += 8;
 775
 776     ret = qemu_file_get_error(f);
 777     if (ret < 0) {
 778         return ret;
 779     }
 780
 781     return total_sent;
 782 }
 783
 784 static int ram_save_complete(QEMUFile *f, void *opaque)
 785 {
 786     qemu_mutex_lock_ramlist();
 787     migration_bitmap_sync();
 788
 789     ram_control_before_iterate(f, RAM_CONTROL_FINISH);
 790
 791     /* try transferring iterative blocks of memory */
 792
 793     /* flush all remaining blocks regardless of rate limiting */
 794     while (true) {
 795         int bytes_sent;
 796
 797         bytes_sent = ram_save_block(f, true);
 798         /* no more blocks to sent */
 799         if (bytes_sent == 0) {
 800             break;
 801         }
 802         bytes_transferred += bytes_sent;
 803     }
 804
 805     ram_control_after_iterate(f, RAM_CONTROL_FINISH);
 806     migration_end();
 807
 808     qemu_mutex_unlock_ramlist();
 809     qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
 810
 811     return 0;
 812 }
 813
 814 static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
 815 {
 816     uint64_t remaining_size;
 817
 818     remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
 819
 820     if (remaining_size < max_size) {
 821         qemu_mutex_lock_iothread();
 822         migration_bitmap_sync();
 823         qemu_mutex_unlock_iothread();
 824         remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
 825     }
 826     return remaining_size;
 827 }
 828
 829 static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host)
 830 {
 831     int ret, rc = 0;
 832     unsigned int xh_len;
 833     int xh_flags;
 834
 835     if (!xbzrle_decoded_buf) {
 836         xbzrle_decoded_buf = g_malloc(TARGET_PAGE_SIZE);
 837     }
 838
 839     /* extract RLE header */
 840     xh_flags = qemu_get_byte(f);
 841     xh_len = qemu_get_be16(f);
 842
 843     if (xh_flags != ENCODING_FLAG_XBZRLE) {
 844         fprintf(stderr, "Failed to load XBZRLE page - wrong compression!\n");
 845         return -1;
 846     }
 847
 848     if (xh_len > TARGET_PAGE_SIZE) {
 849         fprintf(stderr, "Failed to load XBZRLE page - len overflow!\n");
 850         return -1;
 851     }
 852     /* load data and decode */
 853     qemu_get_buffer(f, xbzrle_decoded_buf, xh_len);
 854
 855     /* decode RLE */
 856     ret = xbzrle_decode_buffer(xbzrle_decoded_buf, xh_len, host,
 857                                TARGET_PAGE_SIZE);
 858     if (ret == -1) {
 859         fprintf(stderr, "Failed to load XBZRLE page - decode error!\n");
 860         rc = -1;
 861     } else  if (ret > TARGET_PAGE_SIZE) {
 862         fprintf(stderr, "Failed to load XBZRLE page - size %d exceeds %d!\n",
 863                 ret, TARGET_PAGE_SIZE);
 864         abort();
 865     }
 866
 867     return rc;
 868 }
 869
 870 static inline void *host_from_stream_offset(QEMUFile *f,
 871                                             ram_addr_t offset,
 872                                             int flags)
 873 {
 874     static RAMBlock *block = NULL;
 875     char id[256];
 876     uint8_t len;
 877
 878     if (flags & RAM_SAVE_FLAG_CONTINUE) {
 879         if (!block) {
 880             fprintf(stderr, "Ack, bad migration stream!\n");
 881             return NULL;
 882         }
 883
 884         return memory_region_get_ram_ptr(block->mr) + offset;
 885     }
 886
 887     len = qemu_get_byte(f);
 888     qemu_get_buffer(f, (uint8_t *)id, len);
 889     id[len] = 0;
 890
 891     QTAILQ_FOREACH(block, &ram_list.blocks, next) {
 892         if (!strncmp(id, block->idstr, sizeof(id)))
 893             return memory_region_get_ram_ptr(block->mr) + offset;
 894     }
 895
 896     fprintf(stderr, "Can't find block %s!\n", id);
 897     return NULL;
 898 }
 899
 900 /*
 901  * If a page (or a whole RDMA chunk) has been
 902  * determined to be zero, then zap it.
 903  */
 904 void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
 905 {
 906     if (ch != 0 || !is_zero_range(host, size)) {
 907         memset(host, ch, size);
 908     }
 909 }
 910
 911 static int ram_load(QEMUFile *f, void *opaque, int version_id)
 912 {
 913     ram_addr_t addr;
 914     int flags, ret = 0;
 915     int error;
 916     static uint64_t seq_iter;
 917
 918     seq_iter++;
 919
 920     if (version_id < 4 || version_id > 4) {
 921         return -EINVAL;
 922     }
 923
 924     do {
 925         addr = qemu_get_be64(f);
 926
 927         flags = addr & ~TARGET_PAGE_MASK;
 928         addr &= TARGET_PAGE_MASK;
 929
 930         if (flags & RAM_SAVE_FLAG_MEM_SIZE) {
 931             if (version_id == 4) {
 932                 /* Synchronize RAM block list */
 933                 char id[256];
 934                 ram_addr_t length;
 935                 ram_addr_t total_ram_bytes = addr;
 936
 937                 while (total_ram_bytes) {
 938                     RAMBlock *block;
 939                     uint8_t len;
 940
 941                     len = qemu_get_byte(f);
 942                     qemu_get_buffer(f, (uint8_t *)id, len);
 943                     id[len] = 0;
 944                     length = qemu_get_be64(f);
 945
 946                     QTAILQ_FOREACH(block, &ram_list.blocks, next) {
 947                         if (!strncmp(id, block->idstr, sizeof(id))) {
 948                             if (block->length != length) {
 949                                 fprintf(stderr,
 950                                         "Length mismatch: %s: " RAM_ADDR_FMT
 951                                         " in != " RAM_ADDR_FMT "\n", id, length,
 952                                         block->length);
 953                                 ret =  -EINVAL;
 954                                 goto done;
 955                             }
 956                             break;
 957                         }
 958                     }
 959
 960                     if (!block) {
 961                         fprintf(stderr, "Unknown ramblock \"%s\", cannot "
 962                                 "accept migration\n", id);
 963                         ret = -EINVAL;
 964                         goto done;
 965                     }
 966
 967                     total_ram_bytes -= length;
 968                 }
 969             }
 970         }
 971
 972         if (flags & RAM_SAVE_FLAG_COMPRESS) {
 973             void *host;
 974             uint8_t ch;
 975
 976             host = host_from_stream_offset(f, addr, flags);
 977             if (!host) {
 978                 return -EINVAL;
 979             }
 980
 981             ch = qemu_get_byte(f);
 982             ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
 983         } else if (flags & RAM_SAVE_FLAG_PAGE) {
 984             void *host;
 985
 986             host = host_from_stream_offset(f, addr, flags);
 987             if (!host) {
 988                 return -EINVAL;
 989             }
 990
 991             qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
 992         } else if (flags & RAM_SAVE_FLAG_XBZRLE) {
 993             void *host = host_from_stream_offset(f, addr, flags);
 994             if (!host) {
 995                 return -EINVAL;
 996             }
 997
 998             if (load_xbzrle(f, addr, host) < 0) {
 999                 ret = -EINVAL;
1000                 goto done;
1001             }
1002         } else if (flags & RAM_SAVE_FLAG_HOOK) {
1003             ram_control_load_hook(f, flags);
1004         }
1005         error = qemu_file_get_error(f);
1006         if (error) {
1007             ret = error;
1008             goto done;
1009         }
1010     } while (!(flags & RAM_SAVE_FLAG_EOS));
1011
1012 done:
1013     DPRINTF("Completed load of VM with exit code %d seq iteration "
1014             "%" PRIu64 "\n", ret, seq_iter);
1015     return ret;
1016 }
1017
1018 SaveVMHandlers savevm_ram_handlers = {
1019     .save_live_setup = ram_save_setup,
1020     .save_live_iterate = ram_save_iterate,
1021     .save_live_complete = ram_save_complete,
1022     .save_live_pending = ram_save_pending,
1023     .load_state = ram_load,
1024     .cancel = ram_migration_cancel,
1025 };
1026
1027 struct soundhw {
1028     const char *name;
1029     const char *descr;
1030     int enabled;
1031     int isa;
1032     union {
1033         int (*init_isa) (ISABus *bus);
1034         int (*init_pci) (PCIBus *bus);
1035     } init;
1036 };
1037
1038 static struct soundhw soundhw[9];
1039 static int soundhw_count;
1040
1041 void isa_register_soundhw(const char *name, const char *descr,
1042                           int (*init_isa)(ISABus *bus))
1043 {
1044     assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1045     soundhw[soundhw_count].name = name;
1046     soundhw[soundhw_count].descr = descr;
1047     soundhw[soundhw_count].isa = 1;
1048     soundhw[soundhw_count].init.init_isa = init_isa;
1049     soundhw_count++;
1050 }
1051
1052 void pci_register_soundhw(const char *name, const char *descr,
1053                           int (*init_pci)(PCIBus *bus))
1054 {
1055     assert(soundhw_count < ARRAY_SIZE(soundhw) - 1);
1056     soundhw[soundhw_count].name = name;
1057     soundhw[soundhw_count].descr = descr;
1058     soundhw[soundhw_count].isa = 0;
1059     soundhw[soundhw_count].init.init_pci = init_pci;
1060     soundhw_count++;
1061 }
1062
1063 void select_soundhw(const char *optarg)
1064 {
1065     struct soundhw *c;
1066
1067     if (is_help_option(optarg)) {
1068     show_valid_cards:
1069
1070         if (soundhw_count) {
1071              printf("Valid sound card names (comma separated):\n");
1072              for (c = soundhw; c->name; ++c) {
1073                  printf ("%-11s %s\n", c->name, c->descr);
1074              }
1075              printf("\n-soundhw all will enable all of the above\n");
1076         } else {
1077              printf("Machine has no user-selectable audio hardware "
1078                     "(it may or may not have always-present audio hardware).\n");
1079         }
1080         exit(!is_help_option(optarg));
1081     }
1082     else {
1083         size_t l;
1084         const char *p;
1085         char *e;
1086         int bad_card = 0;
1087
1088         if (!strcmp(optarg, "all")) {
1089             for (c = soundhw; c->name; ++c) {
1090                 c->enabled = 1;
1091             }
1092             return;
1093         }
1094
1095         p = optarg;
1096         while (*p) {
1097             e = strchr(p, ',');
1098             l = !e ? strlen(p) : (size_t) (e - p);
1099
1100             for (c = soundhw; c->name; ++c) {
1101                 if (!strncmp(c->name, p, l) && !c->name[l]) {
1102                     c->enabled = 1;
1103                     break;
1104                 }
1105             }
1106
1107             if (!c->name) {
1108                 if (l > 80) {
1109                     fprintf(stderr,
1110                             "Unknown sound card name (too big to show)\n");
1111                 }
1112                 else {
1113                     fprintf(stderr, "Unknown sound card name `%.*s'\n",
1114                             (int) l, p);
1115                 }
1116                 bad_card = 1;
1117             }
1118             p += l + (e != NULL);
1119         }
1120
1121         if (bad_card) {
1122             goto show_valid_cards;
1123         }
1124     }
1125 }
1126
1127 void audio_init(void)
1128 {
1129     struct soundhw *c;
1130     ISABus *isa_bus = (ISABus *) object_resolve_path_type("", TYPE_ISA_BUS, NULL);
1131     PCIBus *pci_bus = (PCIBus *) object_resolve_path_type("", TYPE_PCI_BUS, NULL);
1132
1133     for (c = soundhw; c->name; ++c) {
1134         if (c->enabled) {
1135             if (c->isa) {
1136                 if (!isa_bus) {
1137                     fprintf(stderr, "ISA bus not available for %s\n", c->name);
1138                     exit(1);
1139                 }
1140                 c->init.init_isa(isa_bus);
1141             } else {
1142                 if (!pci_bus) {
1143                     fprintf(stderr, "PCI bus not available for %s\n", c->name);
1144                     exit(1);
1145                 }
1146                 c->init.init_pci(pci_bus);
1147             }
1148         }
1149     }
1150 }
1151
1152 int qemu_uuid_parse(const char *str, uint8_t *uuid)
1153 {
1154     int ret;
1155
1156     if (strlen(str) != 36) {
1157         return -1;
1158     }
1159
1160     ret = sscanf(str, UUID_FMT, &uuid[0], &uuid[1], &uuid[2], &uuid[3],
1161                  &uuid[4], &uuid[5], &uuid[6], &uuid[7], &uuid[8], &uuid[9],
1162                  &uuid[10], &uuid[11], &uuid[12], &uuid[13], &uuid[14],
1163                  &uuid[15]);
1164
1165     if (ret != 16) {
1166         return -1;
1167     }
1168     return 0;
1169 }
1170
1171 void do_acpitable_option(const QemuOpts *opts)
1172 {
1173 #ifdef TARGET_I386
1174     Error *err = NULL;
1175
1176     acpi_table_add(opts, &err);
1177     if (err) {
1178         error_report("Wrong acpi table provided: %s",
1179                      error_get_pretty(err));
1180         error_free(err);
1181         exit(1);
1182     }
1183 #endif
1184 }
1185
1186 void do_smbios_option(QemuOpts *opts)
1187 {
1188 #ifdef TARGET_I386
1189     smbios_entry_add(opts);
1190 #endif
1191 }
1192
1193 void cpudef_init(void)
1194 {
1195 #if defined(cpudef_setup)
1196     cpudef_setup(); /* parse cpu definitions in target config file */
1197 #endif
1198 }
1199
1200 int tcg_available(void)
1201 {
1202     return 1;
1203 }
1204
1205 int kvm_available(void)
1206 {
1207 #ifdef CONFIG_KVM
1208     return 1;
1209 #else
1210     return 0;
1211 #endif
1212 }
1213
1214 int xen_available(void)
1215 {
1216 #ifdef CONFIG_XEN
1217     return 1;
1218 #else
1219     return 0;
1220 #endif
1221 }
1222
1223
1224 TargetInfo *qmp_query_target(Error **errp)
1225 {
1226     TargetInfo *info = g_malloc0(sizeof(*info));
1227
1228     info->arch = g_strdup(TARGET_NAME);
1229
1230     return info;
1231 }
1232
1233 /* Stub function that's gets run on the vcpu when its brought out of the
1234    VM to run inside qemu via async_run_on_cpu()*/
1235 static void mig_sleep_cpu(void *opq)
1236 {
1237     qemu_mutex_unlock_iothread();
1238     g_usleep(30*1000);
1239     qemu_mutex_lock_iothread();
1240 }
1241
1242 /* To reduce the dirty rate explicitly disallow the VCPUs from spending
1243    much time in the VM. The migration thread will try to catchup.
1244    Workload will experience a performance drop.
1245 */
1246 static void mig_throttle_guest_down(void)
1247 {
1248     CPUState *cpu;
1249
1250     qemu_mutex_lock_iothread();
1251     CPU_FOREACH(cpu) {
1252         async_run_on_cpu(cpu, mig_sleep_cpu, NULL);
1253     }
1254     qemu_mutex_unlock_iothread();
1255 }
1256
1257 static void check_guest_throttling(void)
1258 {
1259     static int64_t t0;
1260     int64_t        t1;
1261
1262     if (!mig_throttle_on) {
1263         return;
1264     }
1265
1266     if (!t0)  {
1267         t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1268         return;
1269     }
1270
1271     t1 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1272
1273     /* If it has been more than 40 ms since the last time the guest
1274      * was throttled then do it again.
1275      */
1276     if (40 < (t1-t0)/1000000) {
1277         mig_throttle_guest_down();
1278         t0 = t1;
1279     }
1280 }