memory.c

   1 /*
   2  * Physical memory management
   3  *
   4  * Copyright 2011 Red Hat, Inc. and/or its affiliates
   5  *
   6  * Authors:
   7  *  Avi Kivity <avi@redhat.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu/osdep.h"
  17 #include "exec/memory.h"
  18 #include "exec/address-spaces.h"
  19 #include "exec/ioport.h"
  20 #include "qapi/visitor.h"
  21 #include "qemu/bitops.h"
  22 #include "qemu/error-report.h"
  23 #include "qom/object.h"
  24 #include "trace.h"
  25
  26 #include "exec/memory-internal.h"
  27 #include "exec/ram_addr.h"
  28 #include "sysemu/kvm.h"
  29 #include "sysemu/sysemu.h"
  30
  31 //#define DEBUG_UNASSIGNED
  32
  33 #define RAM_ADDR_INVALID (~(ram_addr_t)0)
  34
  35 static unsigned memory_region_transaction_depth;
  36 static bool memory_region_update_pending;
  37 static bool ioeventfd_update_pending;
  38 static bool global_dirty_log = false;
  39
  40 static QTAILQ_HEAD(memory_listeners, MemoryListener) memory_listeners
  41     = QTAILQ_HEAD_INITIALIZER(memory_listeners);
  42
  43 static QTAILQ_HEAD(, AddressSpace) address_spaces
  44     = QTAILQ_HEAD_INITIALIZER(address_spaces);
  45
  46 typedef struct AddrRange AddrRange;
  47
  48 /*
  49  * Note that signed integers are needed for negative offsetting in aliases
  50  * (large MemoryRegion::alias_offset).
  51  */
  52 struct AddrRange {
  53     Int128 start;
  54     Int128 size;
  55 };
  56
  57 static AddrRange addrrange_make(Int128 start, Int128 size)
  58 {
  59     return (AddrRange) { start, size };
  60 }
  61
  62 static bool addrrange_equal(AddrRange r1, AddrRange r2)
  63 {
  64     return int128_eq(r1.start, r2.start) && int128_eq(r1.size, r2.size);
  65 }
  66
  67 static Int128 addrrange_end(AddrRange r)
  68 {
  69     return int128_add(r.start, r.size);
  70 }
  71
  72 static AddrRange addrrange_shift(AddrRange range, Int128 delta)
  73 {
  74     int128_addto(&range.start, delta);
  75     return range;
  76 }
  77
  78 static bool addrrange_contains(AddrRange range, Int128 addr)
  79 {
  80     return int128_ge(addr, range.start)
  81         && int128_lt(addr, addrrange_end(range));
  82 }
  83
  84 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
  85 {
  86     return addrrange_contains(r1, r2.start)
  87         || addrrange_contains(r2, r1.start);
  88 }
  89
  90 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
  91 {
  92     Int128 start = int128_max(r1.start, r2.start);
  93     Int128 end = int128_min(addrrange_end(r1), addrrange_end(r2));
  94     return addrrange_make(start, int128_sub(end, start));
  95 }
  96
  97 enum ListenerDirection { Forward, Reverse };
  98
  99 static bool memory_listener_match(MemoryListener *listener,
 100                                   MemoryRegionSection *section)
 101 {
 102     return !listener->address_space_filter
 103         || listener->address_space_filter == section->address_space;
 104 }
 105
 106 #define MEMORY_LISTENER_CALL_GLOBAL(_callback, _direction, _args...)    \
 107     do {                                                                \
 108         MemoryListener *_listener;                                      \
 109                                                                         \
 110         switch (_direction) {                                           \
 111         case Forward:                                                   \
 112             QTAILQ_FOREACH(_listener, &memory_listeners, link) {        \
 113                 if (_listener->_callback) {                             \
 114                     _listener->_callback(_listener, ##_args);           \
 115                 }                                                       \
 116             }                                                           \
 117             break;                                                      \
 118         case Reverse:                                                   \
 119             QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners,        \
 120                                    memory_listeners, link) {            \
 121                 if (_listener->_callback) {                             \
 122                     _listener->_callback(_listener, ##_args);           \
 123                 }                                                       \
 124             }                                                           \
 125             break;                                                      \
 126         default:                                                        \
 127             abort();                                                    \
 128         }                                                               \
 129     } while (0)
 130
 131 #define MEMORY_LISTENER_CALL(_callback, _direction, _section, _args...) \
 132     do {                                                                \
 133         MemoryListener *_listener;                                      \
 134                                                                         \
 135         switch (_direction) {                                           \
 136         case Forward:                                                   \
 137             QTAILQ_FOREACH(_listener, &memory_listeners, link) {        \
 138                 if (_listener->_callback                                \
 139                     && memory_listener_match(_listener, _section)) {    \
 140                     _listener->_callback(_listener, _section, ##_args); \
 141                 }                                                       \
 142             }                                                           \
 143             break;                                                      \
 144         case Reverse:                                                   \
 145             QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners,        \
 146                                    memory_listeners, link) {            \
 147                 if (_listener->_callback                                \
 148                     && memory_listener_match(_listener, _section)) {    \
 149                     _listener->_callback(_listener, _section, ##_args); \
 150                 }                                                       \
 151             }                                                           \
 152             break;                                                      \
 153         default:                                                        \
 154             abort();                                                    \
 155         }                                                               \
 156     } while (0)
 157
 158 /* No need to ref/unref .mr, the FlatRange keeps it alive.  */
 159 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...)  \
 160     MEMORY_LISTENER_CALL(callback, dir, (&(MemoryRegionSection) {       \
 161         .mr = (fr)->mr,                                                 \
 162         .address_space = (as),                                          \
 163         .offset_within_region = (fr)->offset_in_region,                 \
 164         .size = (fr)->addr.size,                                        \
 165         .offset_within_address_space = int128_get64((fr)->addr.start),  \
 166         .readonly = (fr)->readonly,                                     \
 167               }), ##_args)
 168
 169 struct CoalescedMemoryRange {
 170     AddrRange addr;
 171     QTAILQ_ENTRY(CoalescedMemoryRange) link;
 172 };
 173
 174 struct MemoryRegionIoeventfd {
 175     AddrRange addr;
 176     bool match_data;
 177     uint64_t data;
 178     EventNotifier *e;
 179 };
 180
 181 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a,
 182                                            MemoryRegionIoeventfd b)
 183 {
 184     if (int128_lt(a.addr.start, b.addr.start)) {
 185         return true;
 186     } else if (int128_gt(a.addr.start, b.addr.start)) {
 187         return false;
 188     } else if (int128_lt(a.addr.size, b.addr.size)) {
 189         return true;
 190     } else if (int128_gt(a.addr.size, b.addr.size)) {
 191         return false;
 192     } else if (a.match_data < b.match_data) {
 193         return true;
 194     } else  if (a.match_data > b.match_data) {
 195         return false;
 196     } else if (a.match_data) {
 197         if (a.data < b.data) {
 198             return true;
 199         } else if (a.data > b.data) {
 200             return false;
 201         }
 202     }
 203     if (a.e < b.e) {
 204         return true;
 205     } else if (a.e > b.e) {
 206         return false;
 207     }
 208     return false;
 209 }
 210
 211 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a,
 212                                           MemoryRegionIoeventfd b)
 213 {
 214     return !memory_region_ioeventfd_before(a, b)
 215         && !memory_region_ioeventfd_before(b, a);
 216 }
 217
 218 typedef struct FlatRange FlatRange;
 219 typedef struct FlatView FlatView;
 220
 221 /* Range of memory in the global map.  Addresses are absolute. */
 222 struct FlatRange {
 223     MemoryRegion *mr;
 224     hwaddr offset_in_region;
 225     AddrRange addr;
 226     uint8_t dirty_log_mask;
 227     bool romd_mode;
 228     bool readonly;
 229 };
 230
 231 /* Flattened global view of current active memory hierarchy.  Kept in sorted
 232  * order.
 233  */
 234 struct FlatView {
 235     struct rcu_head rcu;
 236     unsigned ref;
 237     FlatRange *ranges;
 238     unsigned nr;
 239     unsigned nr_allocated;
 240 };
 241
 242 typedef struct AddressSpaceOps AddressSpaceOps;
 243
 244 #define FOR_EACH_FLAT_RANGE(var, view)          \
 245     for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
 246
 247 static bool flatrange_equal(FlatRange *a, FlatRange *b)
 248 {
 249     return a->mr == b->mr
 250         && addrrange_equal(a->addr, b->addr)
 251         && a->offset_in_region == b->offset_in_region
 252         && a->romd_mode == b->romd_mode
 253         && a->readonly == b->readonly;
 254 }
 255
 256 static void flatview_init(FlatView *view)
 257 {
 258     view->ref = 1;
 259     view->ranges = NULL;
 260     view->nr = 0;
 261     view->nr_allocated = 0;
 262 }
 263
 264 /* Insert a range into a given position.  Caller is responsible for maintaining
 265  * sorting order.
 266  */
 267 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
 268 {
 269     if (view->nr == view->nr_allocated) {
 270         view->nr_allocated = MAX(2 * view->nr, 10);
 271         view->ranges = g_realloc(view->ranges,
 272                                     view->nr_allocated * sizeof(*view->ranges));
 273     }
 274     memmove(view->ranges + pos + 1, view->ranges + pos,
 275             (view->nr - pos) * sizeof(FlatRange));
 276     view->ranges[pos] = *range;
 277     memory_region_ref(range->mr);
 278     ++view->nr;
 279 }
 280
 281 static void flatview_destroy(FlatView *view)
 282 {
 283     int i;
 284
 285     for (i = 0; i < view->nr; i++) {
 286         memory_region_unref(view->ranges[i].mr);
 287     }
 288     g_free(view->ranges);
 289     g_free(view);
 290 }
 291
 292 static void flatview_ref(FlatView *view)
 293 {
 294     atomic_inc(&view->ref);
 295 }
 296
 297 static void flatview_unref(FlatView *view)
 298 {
 299     if (atomic_fetch_dec(&view->ref) == 1) {
 300         flatview_destroy(view);
 301     }
 302 }
 303
 304 static bool can_merge(FlatRange *r1, FlatRange *r2)
 305 {
 306     return int128_eq(addrrange_end(r1->addr), r2->addr.start)
 307         && r1->mr == r2->mr
 308         && int128_eq(int128_add(int128_make64(r1->offset_in_region),
 309                                 r1->addr.size),
 310                      int128_make64(r2->offset_in_region))
 311         && r1->dirty_log_mask == r2->dirty_log_mask
 312         && r1->romd_mode == r2->romd_mode
 313         && r1->readonly == r2->readonly;
 314 }
 315
 316 /* Attempt to simplify a view by merging adjacent ranges */
 317 static void flatview_simplify(FlatView *view)
 318 {
 319     unsigned i, j;
 320
 321     i = 0;
 322     while (i < view->nr) {
 323         j = i + 1;
 324         while (j < view->nr
 325                && can_merge(&view->ranges[j-1], &view->ranges[j])) {
 326             int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
 327             ++j;
 328         }
 329         ++i;
 330         memmove(&view->ranges[i], &view->ranges[j],
 331                 (view->nr - j) * sizeof(view->ranges[j]));
 332         view->nr -= j - i;
 333     }
 334 }
 335
 336 static bool memory_region_big_endian(MemoryRegion *mr)
 337 {
 338 #ifdef TARGET_WORDS_BIGENDIAN
 339     return mr->ops->endianness != DEVICE_LITTLE_ENDIAN;
 340 #else
 341     return mr->ops->endianness == DEVICE_BIG_ENDIAN;
 342 #endif
 343 }
 344
 345 static bool memory_region_wrong_endianness(MemoryRegion *mr)
 346 {
 347 #ifdef TARGET_WORDS_BIGENDIAN
 348     return mr->ops->endianness == DEVICE_LITTLE_ENDIAN;
 349 #else
 350     return mr->ops->endianness == DEVICE_BIG_ENDIAN;
 351 #endif
 352 }
 353
 354 static void adjust_endianness(MemoryRegion *mr, uint64_t *data, unsigned size)
 355 {
 356     if (memory_region_wrong_endianness(mr)) {
 357         switch (size) {
 358         case 1:
 359             break;
 360         case 2:
 361             *data = bswap16(*data);
 362             break;
 363         case 4:
 364             *data = bswap32(*data);
 365             break;
 366         case 8:
 367             *data = bswap64(*data);
 368             break;
 369         default:
 370             abort();
 371         }
 372     }
 373 }
 374
 375 static hwaddr memory_region_to_absolute_addr(MemoryRegion *mr, hwaddr offset)
 376 {
 377     MemoryRegion *root;
 378     hwaddr abs_addr = offset;
 379
 380     abs_addr += mr->addr;
 381     for (root = mr; root->container; ) {
 382         root = root->container;
 383         abs_addr += root->addr;
 384     }
 385
 386     return abs_addr;
 387 }
 388
 389 static MemTxResult memory_region_oldmmio_read_accessor(MemoryRegion *mr,
 390                                                        hwaddr addr,
 391                                                        uint64_t *value,
 392                                                        unsigned size,
 393                                                        unsigned shift,
 394                                                        uint64_t mask,
 395                                                        MemTxAttrs attrs)
 396 {
 397     uint64_t tmp;
 398
 399     tmp = mr->ops->old_mmio.read[ctz32(size)](mr->opaque, addr);
 400     if (mr->subpage) {
 401         trace_memory_region_subpage_read(mr, addr, tmp, size);
 402     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 403         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 404         trace_memory_region_ops_read(mr, abs_addr, tmp, size);
 405     }
 406     *value |= (tmp & mask) << shift;
 407     return MEMTX_OK;
 408 }
 409
 410 static MemTxResult  memory_region_read_accessor(MemoryRegion *mr,
 411                                                 hwaddr addr,
 412                                                 uint64_t *value,
 413                                                 unsigned size,
 414                                                 unsigned shift,
 415                                                 uint64_t mask,
 416                                                 MemTxAttrs attrs)
 417 {
 418     uint64_t tmp;
 419
 420     tmp = mr->ops->read(mr->opaque, addr, size);
 421     if (mr->subpage) {
 422         trace_memory_region_subpage_read(mr, addr, tmp, size);
 423     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 424         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 425         trace_memory_region_ops_read(mr, abs_addr, tmp, size);
 426     }
 427     *value |= (tmp & mask) << shift;
 428     return MEMTX_OK;
 429 }
 430
 431 static MemTxResult memory_region_read_with_attrs_accessor(MemoryRegion *mr,
 432                                                           hwaddr addr,
 433                                                           uint64_t *value,
 434                                                           unsigned size,
 435                                                           unsigned shift,
 436                                                           uint64_t mask,
 437                                                           MemTxAttrs attrs)
 438 {
 439     uint64_t tmp = 0;
 440     MemTxResult r;
 441
 442     r = mr->ops->read_with_attrs(mr->opaque, addr, &tmp, size, attrs);
 443     if (mr->subpage) {
 444         trace_memory_region_subpage_read(mr, addr, tmp, size);
 445     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 446         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 447         trace_memory_region_ops_read(mr, abs_addr, tmp, size);
 448     }
 449     *value |= (tmp & mask) << shift;
 450     return r;
 451 }
 452
 453 static MemTxResult memory_region_oldmmio_write_accessor(MemoryRegion *mr,
 454                                                         hwaddr addr,
 455                                                         uint64_t *value,
 456                                                         unsigned size,
 457                                                         unsigned shift,
 458                                                         uint64_t mask,
 459                                                         MemTxAttrs attrs)
 460 {
 461     uint64_t tmp;
 462
 463     tmp = (*value >> shift) & mask;
 464     if (mr->subpage) {
 465         trace_memory_region_subpage_write(mr, addr, tmp, size);
 466     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 467         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 468         trace_memory_region_ops_write(mr, abs_addr, tmp, size);
 469     }
 470     mr->ops->old_mmio.write[ctz32(size)](mr->opaque, addr, tmp);
 471     return MEMTX_OK;
 472 }
 473
 474 static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
 475                                                 hwaddr addr,
 476                                                 uint64_t *value,
 477                                                 unsigned size,
 478                                                 unsigned shift,
 479                                                 uint64_t mask,
 480                                                 MemTxAttrs attrs)
 481 {
 482     uint64_t tmp;
 483
 484     tmp = (*value >> shift) & mask;
 485     if (mr->subpage) {
 486         trace_memory_region_subpage_write(mr, addr, tmp, size);
 487     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 488         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 489         trace_memory_region_ops_write(mr, abs_addr, tmp, size);
 490     }
 491     mr->ops->write(mr->opaque, addr, tmp, size);
 492     return MEMTX_OK;
 493 }
 494
 495 static MemTxResult memory_region_write_with_attrs_accessor(MemoryRegion *mr,
 496                                                            hwaddr addr,
 497                                                            uint64_t *value,
 498                                                            unsigned size,
 499                                                            unsigned shift,
 500                                                            uint64_t mask,
 501                                                            MemTxAttrs attrs)
 502 {
 503     uint64_t tmp;
 504
 505     tmp = (*value >> shift) & mask;
 506     if (mr->subpage) {
 507         trace_memory_region_subpage_write(mr, addr, tmp, size);
 508     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 509         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 510         trace_memory_region_ops_write(mr, abs_addr, tmp, size);
 511     }
 512     return mr->ops->write_with_attrs(mr->opaque, addr, tmp, size, attrs);
 513 }
 514
 515 static MemTxResult access_with_adjusted_size(hwaddr addr,
 516                                       uint64_t *value,
 517                                       unsigned size,
 518                                       unsigned access_size_min,
 519                                       unsigned access_size_max,
 520                                       MemTxResult (*access)(MemoryRegion *mr,
 521                                                             hwaddr addr,
 522                                                             uint64_t *value,
 523                                                             unsigned size,
 524                                                             unsigned shift,
 525                                                             uint64_t mask,
 526                                                             MemTxAttrs attrs),
 527                                       MemoryRegion *mr,
 528                                       MemTxAttrs attrs)
 529 {
 530     uint64_t access_mask;
 531     unsigned access_size;
 532     unsigned i;
 533     MemTxResult r = MEMTX_OK;
 534
 535     if (!access_size_min) {
 536         access_size_min = 1;
 537     }
 538     if (!access_size_max) {
 539         access_size_max = 4;
 540     }
 541
 542     /* FIXME: support unaligned access? */
 543     access_size = MAX(MIN(size, access_size_max), access_size_min);
 544     access_mask = -1ULL >> (64 - access_size * 8);
 545     if (memory_region_big_endian(mr)) {
 546         for (i = 0; i < size; i += access_size) {
 547             r |= access(mr, addr + i, value, access_size,
 548                         (size - access_size - i) * 8, access_mask, attrs);
 549         }
 550     } else {
 551         for (i = 0; i < size; i += access_size) {
 552             r |= access(mr, addr + i, value, access_size, i * 8,
 553                         access_mask, attrs);
 554         }
 555     }
 556     return r;
 557 }
 558
 559 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
 560 {
 561     AddressSpace *as;
 562
 563     while (mr->container) {
 564         mr = mr->container;
 565     }
 566     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 567         if (mr == as->root) {
 568             return as;
 569         }
 570     }
 571     return NULL;
 572 }
 573
 574 /* Render a memory region into the global view.  Ranges in @view obscure
 575  * ranges in @mr.
 576  */
 577 static void render_memory_region(FlatView *view,
 578                                  MemoryRegion *mr,
 579                                  Int128 base,
 580                                  AddrRange clip,
 581                                  bool readonly)
 582 {
 583     MemoryRegion *subregion;
 584     unsigned i;
 585     hwaddr offset_in_region;
 586     Int128 remain;
 587     Int128 now;
 588     FlatRange fr;
 589     AddrRange tmp;
 590
 591     if (!mr->enabled) {
 592         return;
 593     }
 594
 595     int128_addto(&base, int128_make64(mr->addr));
 596     readonly |= mr->readonly;
 597
 598     tmp = addrrange_make(base, mr->size);
 599
 600     if (!addrrange_intersects(tmp, clip)) {
 601         return;
 602     }
 603
 604     clip = addrrange_intersection(tmp, clip);
 605
 606     if (mr->alias) {
 607         int128_subfrom(&base, int128_make64(mr->alias->addr));
 608         int128_subfrom(&base, int128_make64(mr->alias_offset));
 609         render_memory_region(view, mr->alias, base, clip, readonly);
 610         return;
 611     }
 612
 613     /* Render subregions in priority order. */
 614     QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
 615         render_memory_region(view, subregion, base, clip, readonly);
 616     }
 617
 618     if (!mr->terminates) {
 619         return;
 620     }
 621
 622     offset_in_region = int128_get64(int128_sub(clip.start, base));
 623     base = clip.start;
 624     remain = clip.size;
 625
 626     fr.mr = mr;
 627     fr.dirty_log_mask = memory_region_get_dirty_log_mask(mr);
 628     fr.romd_mode = mr->romd_mode;
 629     fr.readonly = readonly;
 630
 631     /* Render the region itself into any gaps left by the current view. */
 632     for (i = 0; i < view->nr && int128_nz(remain); ++i) {
 633         if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
 634             continue;
 635         }
 636         if (int128_lt(base, view->ranges[i].addr.start)) {
 637             now = int128_min(remain,
 638                              int128_sub(view->ranges[i].addr.start, base));
 639             fr.offset_in_region = offset_in_region;
 640             fr.addr = addrrange_make(base, now);
 641             flatview_insert(view, i, &fr);
 642             ++i;
 643             int128_addto(&base, now);
 644             offset_in_region += int128_get64(now);
 645             int128_subfrom(&remain, now);
 646         }
 647         now = int128_sub(int128_min(int128_add(base, remain),
 648                                     addrrange_end(view->ranges[i].addr)),
 649                          base);
 650         int128_addto(&base, now);
 651         offset_in_region += int128_get64(now);
 652         int128_subfrom(&remain, now);
 653     }
 654     if (int128_nz(remain)) {
 655         fr.offset_in_region = offset_in_region;
 656         fr.addr = addrrange_make(base, remain);
 657         flatview_insert(view, i, &fr);
 658     }
 659 }
 660
 661 /* Render a memory topology into a list of disjoint absolute ranges. */
 662 static FlatView *generate_memory_topology(MemoryRegion *mr)
 663 {
 664     FlatView *view;
 665
 666     view = g_new(FlatView, 1);
 667     flatview_init(view);
 668
 669     if (mr) {
 670         render_memory_region(view, mr, int128_zero(),
 671                              addrrange_make(int128_zero(), int128_2_64()), false);
 672     }
 673     flatview_simplify(view);
 674
 675     return view;
 676 }
 677
 678 static void address_space_add_del_ioeventfds(AddressSpace *as,
 679                                              MemoryRegionIoeventfd *fds_new,
 680                                              unsigned fds_new_nb,
 681                                              MemoryRegionIoeventfd *fds_old,
 682                                              unsigned fds_old_nb)
 683 {
 684     unsigned iold, inew;
 685     MemoryRegionIoeventfd *fd;
 686     MemoryRegionSection section;
 687
 688     /* Generate a symmetric difference of the old and new fd sets, adding
 689      * and deleting as necessary.
 690      */
 691
 692     iold = inew = 0;
 693     while (iold < fds_old_nb || inew < fds_new_nb) {
 694         if (iold < fds_old_nb
 695             && (inew == fds_new_nb
 696                 || memory_region_ioeventfd_before(fds_old[iold],
 697                                                   fds_new[inew]))) {
 698             fd = &fds_old[iold];
 699             section = (MemoryRegionSection) {
 700                 .address_space = as,
 701                 .offset_within_address_space = int128_get64(fd->addr.start),
 702                 .size = fd->addr.size,
 703             };
 704             MEMORY_LISTENER_CALL(eventfd_del, Forward, &section,
 705                                  fd->match_data, fd->data, fd->e);
 706             ++iold;
 707         } else if (inew < fds_new_nb
 708                    && (iold == fds_old_nb
 709                        || memory_region_ioeventfd_before(fds_new[inew],
 710                                                          fds_old[iold]))) {
 711             fd = &fds_new[inew];
 712             section = (MemoryRegionSection) {
 713                 .address_space = as,
 714                 .offset_within_address_space = int128_get64(fd->addr.start),
 715                 .size = fd->addr.size,
 716             };
 717             MEMORY_LISTENER_CALL(eventfd_add, Reverse, &section,
 718                                  fd->match_data, fd->data, fd->e);
 719             ++inew;
 720         } else {
 721             ++iold;
 722             ++inew;
 723         }
 724     }
 725 }
 726
 727 static FlatView *address_space_get_flatview(AddressSpace *as)
 728 {
 729     FlatView *view;
 730
 731     rcu_read_lock();
 732     view = atomic_rcu_read(&as->current_map);
 733     flatview_ref(view);
 734     rcu_read_unlock();
 735     return view;
 736 }
 737
 738 static void address_space_update_ioeventfds(AddressSpace *as)
 739 {
 740     FlatView *view;
 741     FlatRange *fr;
 742     unsigned ioeventfd_nb = 0;
 743     MemoryRegionIoeventfd *ioeventfds = NULL;
 744     AddrRange tmp;
 745     unsigned i;
 746
 747     view = address_space_get_flatview(as);
 748     FOR_EACH_FLAT_RANGE(fr, view) {
 749         for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
 750             tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
 751                                   int128_sub(fr->addr.start,
 752                                              int128_make64(fr->offset_in_region)));
 753             if (addrrange_intersects(fr->addr, tmp)) {
 754                 ++ioeventfd_nb;
 755                 ioeventfds = g_realloc(ioeventfds,
 756                                           ioeventfd_nb * sizeof(*ioeventfds));
 757                 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
 758                 ioeventfds[ioeventfd_nb-1].addr = tmp;
 759             }
 760         }
 761     }
 762
 763     address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
 764                                      as->ioeventfds, as->ioeventfd_nb);
 765
 766     g_free(as->ioeventfds);
 767     as->ioeventfds = ioeventfds;
 768     as->ioeventfd_nb = ioeventfd_nb;
 769     flatview_unref(view);
 770 }
 771
 772 static void address_space_update_topology_pass(AddressSpace *as,
 773                                                const FlatView *old_view,
 774                                                const FlatView *new_view,
 775                                                bool adding)
 776 {
 777     unsigned iold, inew;
 778     FlatRange *frold, *frnew;
 779
 780     /* Generate a symmetric difference of the old and new memory maps.
 781      * Kill ranges in the old map, and instantiate ranges in the new map.
 782      */
 783     iold = inew = 0;
 784     while (iold < old_view->nr || inew < new_view->nr) {
 785         if (iold < old_view->nr) {
 786             frold = &old_view->ranges[iold];
 787         } else {
 788             frold = NULL;
 789         }
 790         if (inew < new_view->nr) {
 791             frnew = &new_view->ranges[inew];
 792         } else {
 793             frnew = NULL;
 794         }
 795
 796         if (frold
 797             && (!frnew
 798                 || int128_lt(frold->addr.start, frnew->addr.start)
 799                 || (int128_eq(frold->addr.start, frnew->addr.start)
 800                     && !flatrange_equal(frold, frnew)))) {
 801             /* In old but not in new, or in both but attributes changed. */
 802
 803             if (!adding) {
 804                 MEMORY_LISTENER_UPDATE_REGION(frold, as, Reverse, region_del);
 805             }
 806
 807             ++iold;
 808         } else if (frold && frnew && flatrange_equal(frold, frnew)) {
 809             /* In both and unchanged (except logging may have changed) */
 810
 811             if (adding) {
 812                 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_nop);
 813                 if (frnew->dirty_log_mask & ~frold->dirty_log_mask) {
 814                     MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, log_start,
 815                                                   frold->dirty_log_mask,
 816                                                   frnew->dirty_log_mask);
 817                 }
 818                 if (frold->dirty_log_mask & ~frnew->dirty_log_mask) {
 819                     MEMORY_LISTENER_UPDATE_REGION(frnew, as, Reverse, log_stop,
 820                                                   frold->dirty_log_mask,
 821                                                   frnew->dirty_log_mask);
 822                 }
 823             }
 824
 825             ++iold;
 826             ++inew;
 827         } else {
 828             /* In new */
 829
 830             if (adding) {
 831                 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_add);
 832             }
 833
 834             ++inew;
 835         }
 836     }
 837 }
 838
 839
 840 static void address_space_update_topology(AddressSpace *as)
 841 {
 842     FlatView *old_view = address_space_get_flatview(as);
 843     FlatView *new_view = generate_memory_topology(as->root);
 844
 845     address_space_update_topology_pass(as, old_view, new_view, false);
 846     address_space_update_topology_pass(as, old_view, new_view, true);
 847
 848     /* Writes are protected by the BQL.  */
 849     atomic_rcu_set(&as->current_map, new_view);
 850     call_rcu(old_view, flatview_unref, rcu);
 851
 852     /* Note that all the old MemoryRegions are still alive up to this
 853      * point.  This relieves most MemoryListeners from the need to
 854      * ref/unref the MemoryRegions they get---unless they use them
 855      * outside the iothread mutex, in which case precise reference
 856      * counting is necessary.
 857      */
 858     flatview_unref(old_view);
 859
 860     address_space_update_ioeventfds(as);
 861 }
 862
 863 void memory_region_transaction_begin(void)
 864 {
 865     qemu_flush_coalesced_mmio_buffer();
 866     ++memory_region_transaction_depth;
 867 }
 868
 869 static void memory_region_clear_pending(void)
 870 {
 871     memory_region_update_pending = false;
 872     ioeventfd_update_pending = false;
 873 }
 874
 875 void memory_region_transaction_commit(void)
 876 {
 877     AddressSpace *as;
 878
 879     assert(memory_region_transaction_depth);
 880     --memory_region_transaction_depth;
 881     if (!memory_region_transaction_depth) {
 882         if (memory_region_update_pending) {
 883             MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
 884
 885             QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 886                 address_space_update_topology(as);
 887             }
 888
 889             MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
 890         } else if (ioeventfd_update_pending) {
 891             QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 892                 address_space_update_ioeventfds(as);
 893             }
 894         }
 895         memory_region_clear_pending();
 896    }
 897 }
 898
 899 static void memory_region_destructor_none(MemoryRegion *mr)
 900 {
 901 }
 902
 903 static void memory_region_destructor_ram(MemoryRegion *mr)
 904 {
 905     qemu_ram_free(mr->ram_addr);
 906 }
 907
 908 static void memory_region_destructor_rom_device(MemoryRegion *mr)
 909 {
 910     qemu_ram_free(mr->ram_addr & TARGET_PAGE_MASK);
 911 }
 912
 913 static bool memory_region_need_escape(char c)
 914 {
 915     return c == '/' || c == '[' || c == '\\' || c == ']';
 916 }
 917
 918 static char *memory_region_escape_name(const char *name)
 919 {
 920     const char *p;
 921     char *escaped, *q;
 922     uint8_t c;
 923     size_t bytes = 0;
 924
 925     for (p = name; *p; p++) {
 926         bytes += memory_region_need_escape(*p) ? 4 : 1;
 927     }
 928     if (bytes == p - name) {
 929        return g_memdup(name, bytes + 1);
 930     }
 931
 932     escaped = g_malloc(bytes + 1);
 933     for (p = name, q = escaped; *p; p++) {
 934         c = *p;
 935         if (unlikely(memory_region_need_escape(c))) {
 936             *q++ = '\\';
 937             *q++ = 'x';
 938             *q++ = "0123456789abcdef"[c >> 4];
 939             c = "0123456789abcdef"[c & 15];
 940         }
 941         *q++ = c;
 942     }
 943     *q = 0;
 944     return escaped;
 945 }
 946
 947 void memory_region_init(MemoryRegion *mr,
 948                         Object *owner,
 949                         const char *name,
 950                         uint64_t size)
 951 {
 952     object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
 953     mr->size = int128_make64(size);
 954     if (size == UINT64_MAX) {
 955         mr->size = int128_2_64();
 956     }
 957     mr->name = g_strdup(name);
 958     mr->owner = owner;
 959     mr->ram_block = NULL;
 960
 961     if (name) {
 962         char *escaped_name = memory_region_escape_name(name);
 963         char *name_array = g_strdup_printf("%s[*]", escaped_name);
 964
 965         if (!owner) {
 966             owner = container_get(qdev_get_machine(), "/unattached");
 967         }
 968
 969         object_property_add_child(owner, name_array, OBJECT(mr), &error_abort);
 970         object_unref(OBJECT(mr));
 971         g_free(name_array);
 972         g_free(escaped_name);
 973     }
 974 }
 975
 976 static void memory_region_get_addr(Object *obj, Visitor *v, const char *name,
 977                                    void *opaque, Error **errp)
 978 {
 979     MemoryRegion *mr = MEMORY_REGION(obj);
 980     uint64_t value = mr->addr;
 981
 982     visit_type_uint64(v, name, &value, errp);
 983 }
 984
 985 static void memory_region_get_container(Object *obj, Visitor *v,
 986                                         const char *name, void *opaque,
 987                                         Error **errp)
 988 {
 989     MemoryRegion *mr = MEMORY_REGION(obj);
 990     gchar *path = (gchar *)"";
 991
 992     if (mr->container) {
 993         path = object_get_canonical_path(OBJECT(mr->container));
 994     }
 995     visit_type_str(v, name, &path, errp);
 996     if (mr->container) {
 997         g_free(path);
 998     }
 999 }
1000
1001 static Object *memory_region_resolve_container(Object *obj, void *opaque,
1002                                                const char *part)
1003 {
1004     MemoryRegion *mr = MEMORY_REGION(obj);
1005
1006     return OBJECT(mr->container);
1007 }
1008
1009 static void memory_region_get_priority(Object *obj, Visitor *v,
1010                                        const char *name, void *opaque,
1011                                        Error **errp)
1012 {
1013     MemoryRegion *mr = MEMORY_REGION(obj);
1014     int32_t value = mr->priority;
1015
1016     visit_type_int32(v, name, &value, errp);
1017 }
1018
1019 static bool memory_region_get_may_overlap(Object *obj, Error **errp)
1020 {
1021     MemoryRegion *mr = MEMORY_REGION(obj);
1022
1023     return mr->may_overlap;
1024 }
1025
1026 static void memory_region_get_size(Object *obj, Visitor *v, const char *name,
1027                                    void *opaque, Error **errp)
1028 {
1029     MemoryRegion *mr = MEMORY_REGION(obj);
1030     uint64_t value = memory_region_size(mr);
1031
1032     visit_type_uint64(v, name, &value, errp);
1033 }
1034
1035 static void memory_region_initfn(Object *obj)
1036 {
1037     MemoryRegion *mr = MEMORY_REGION(obj);
1038     ObjectProperty *op;
1039
1040     mr->ops = &unassigned_mem_ops;
1041     mr->ram_addr = RAM_ADDR_INVALID;
1042     mr->enabled = true;
1043     mr->romd_mode = true;
1044     mr->global_locking = true;
1045     mr->destructor = memory_region_destructor_none;
1046     QTAILQ_INIT(&mr->subregions);
1047     QTAILQ_INIT(&mr->coalesced);
1048
1049     op = object_property_add(OBJECT(mr), "container",
1050                              "link<" TYPE_MEMORY_REGION ">",
1051                              memory_region_get_container,
1052                              NULL, /* memory_region_set_container */
1053                              NULL, NULL, &error_abort);
1054     op->resolve = memory_region_resolve_container;
1055
1056     object_property_add(OBJECT(mr), "addr", "uint64",
1057                         memory_region_get_addr,
1058                         NULL, /* memory_region_set_addr */
1059                         NULL, NULL, &error_abort);
1060     object_property_add(OBJECT(mr), "priority", "uint32",
1061                         memory_region_get_priority,
1062                         NULL, /* memory_region_set_priority */
1063                         NULL, NULL, &error_abort);
1064     object_property_add_bool(OBJECT(mr), "may-overlap",
1065                              memory_region_get_may_overlap,
1066                              NULL, /* memory_region_set_may_overlap */
1067                              &error_abort);
1068     object_property_add(OBJECT(mr), "size", "uint64",
1069                         memory_region_get_size,
1070                         NULL, /* memory_region_set_size, */
1071                         NULL, NULL, &error_abort);
1072 }
1073
1074 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1075                                     unsigned size)
1076 {
1077 #ifdef DEBUG_UNASSIGNED
1078     printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
1079 #endif
1080     if (current_cpu != NULL) {
1081         cpu_unassigned_access(current_cpu, addr, false, false, 0, size);
1082     }
1083     return 0;
1084 }
1085
1086 static void unassigned_mem_write(void *opaque, hwaddr addr,
1087                                  uint64_t val, unsigned size)
1088 {
1089 #ifdef DEBUG_UNASSIGNED
1090     printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1091 #endif
1092     if (current_cpu != NULL) {
1093         cpu_unassigned_access(current_cpu, addr, true, false, 0, size);
1094     }
1095 }
1096
1097 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1098                                    unsigned size, bool is_write)
1099 {
1100     return false;
1101 }
1102
1103 const MemoryRegionOps unassigned_mem_ops = {
1104     .valid.accepts = unassigned_mem_accepts,
1105     .endianness = DEVICE_NATIVE_ENDIAN,
1106 };
1107
1108 bool memory_region_access_valid(MemoryRegion *mr,
1109                                 hwaddr addr,
1110                                 unsigned size,
1111                                 bool is_write)
1112 {
1113     int access_size_min, access_size_max;
1114     int access_size, i;
1115
1116     if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1117         return false;
1118     }
1119
1120     if (!mr->ops->valid.accepts) {
1121         return true;
1122     }
1123
1124     access_size_min = mr->ops->valid.min_access_size;
1125     if (!mr->ops->valid.min_access_size) {
1126         access_size_min = 1;
1127     }
1128
1129     access_size_max = mr->ops->valid.max_access_size;
1130     if (!mr->ops->valid.max_access_size) {
1131         access_size_max = 4;
1132     }
1133
1134     access_size = MAX(MIN(size, access_size_max), access_size_min);
1135     for (i = 0; i < size; i += access_size) {
1136         if (!mr->ops->valid.accepts(mr->opaque, addr + i, access_size,
1137                                     is_write)) {
1138             return false;
1139         }
1140     }
1141
1142     return true;
1143 }
1144
1145 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1146                                                 hwaddr addr,
1147                                                 uint64_t *pval,
1148                                                 unsigned size,
1149                                                 MemTxAttrs attrs)
1150 {
1151     *pval = 0;
1152
1153     if (mr->ops->read) {
1154         return access_with_adjusted_size(addr, pval, size,
1155                                          mr->ops->impl.min_access_size,
1156                                          mr->ops->impl.max_access_size,
1157                                          memory_region_read_accessor,
1158                                          mr, attrs);
1159     } else if (mr->ops->read_with_attrs) {
1160         return access_with_adjusted_size(addr, pval, size,
1161                                          mr->ops->impl.min_access_size,
1162                                          mr->ops->impl.max_access_size,
1163                                          memory_region_read_with_attrs_accessor,
1164                                          mr, attrs);
1165     } else {
1166         return access_with_adjusted_size(addr, pval, size, 1, 4,
1167                                          memory_region_oldmmio_read_accessor,
1168                                          mr, attrs);
1169     }
1170 }
1171
1172 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1173                                         hwaddr addr,
1174                                         uint64_t *pval,
1175                                         unsigned size,
1176                                         MemTxAttrs attrs)
1177 {
1178     MemTxResult r;
1179
1180     if (!memory_region_access_valid(mr, addr, size, false)) {
1181         *pval = unassigned_mem_read(mr, addr, size);
1182         return MEMTX_DECODE_ERROR;
1183     }
1184
1185     r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1186     adjust_endianness(mr, pval, size);
1187     return r;
1188 }
1189
1190 /* Return true if an eventfd was signalled */
1191 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1192                                                     hwaddr addr,
1193                                                     uint64_t data,
1194                                                     unsigned size,
1195                                                     MemTxAttrs attrs)
1196 {
1197     MemoryRegionIoeventfd ioeventfd = {
1198         .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1199         .data = data,
1200     };
1201     unsigned i;
1202
1203     for (i = 0; i < mr->ioeventfd_nb; i++) {
1204         ioeventfd.match_data = mr->ioeventfds[i].match_data;
1205         ioeventfd.e = mr->ioeventfds[i].e;
1206
1207         if (memory_region_ioeventfd_equal(ioeventfd, mr->ioeventfds[i])) {
1208             event_notifier_set(ioeventfd.e);
1209             return true;
1210         }
1211     }
1212
1213     return false;
1214 }
1215
1216 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1217                                          hwaddr addr,
1218                                          uint64_t data,
1219                                          unsigned size,
1220                                          MemTxAttrs attrs)
1221 {
1222     if (!memory_region_access_valid(mr, addr, size, true)) {
1223         unassigned_mem_write(mr, addr, data, size);
1224         return MEMTX_DECODE_ERROR;
1225     }
1226
1227     adjust_endianness(mr, &data, size);
1228
1229     if ((!kvm_eventfds_enabled()) &&
1230         memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1231         return MEMTX_OK;
1232     }
1233
1234     if (mr->ops->write) {
1235         return access_with_adjusted_size(addr, &data, size,
1236                                          mr->ops->impl.min_access_size,
1237                                          mr->ops->impl.max_access_size,
1238                                          memory_region_write_accessor, mr,
1239                                          attrs);
1240     } else if (mr->ops->write_with_attrs) {
1241         return
1242             access_with_adjusted_size(addr, &data, size,
1243                                       mr->ops->impl.min_access_size,
1244                                       mr->ops->impl.max_access_size,
1245                                       memory_region_write_with_attrs_accessor,
1246                                       mr, attrs);
1247     } else {
1248         return access_with_adjusted_size(addr, &data, size, 1, 4,
1249                                          memory_region_oldmmio_write_accessor,
1250                                          mr, attrs);
1251     }
1252 }
1253
1254 void memory_region_init_io(MemoryRegion *mr,
1255                            Object *owner,
1256                            const MemoryRegionOps *ops,
1257                            void *opaque,
1258                            const char *name,
1259                            uint64_t size)
1260 {
1261     memory_region_init(mr, owner, name, size);
1262     mr->ops = ops ? ops : &unassigned_mem_ops;
1263     mr->opaque = opaque;
1264     mr->terminates = true;
1265 }
1266
1267 void memory_region_init_ram(MemoryRegion *mr,
1268                             Object *owner,
1269                             const char *name,
1270                             uint64_t size,
1271                             Error **errp)
1272 {
1273     RAMBlock *ram_block;
1274
1275     memory_region_init(mr, owner, name, size);
1276     mr->ram = true;
1277     mr->terminates = true;
1278     mr->destructor = memory_region_destructor_ram;
1279     ram_block = qemu_ram_alloc(size, mr, errp);
1280     mr->ram_addr = ram_block->offset;
1281     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1282 }
1283
1284 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1285                                        Object *owner,
1286                                        const char *name,
1287                                        uint64_t size,
1288                                        uint64_t max_size,
1289                                        void (*resized)(const char*,
1290                                                        uint64_t length,
1291                                                        void *host),
1292                                        Error **errp)
1293 {
1294     RAMBlock *ram_block;
1295
1296     memory_region_init(mr, owner, name, size);
1297     mr->ram = true;
1298     mr->terminates = true;
1299     mr->destructor = memory_region_destructor_ram;
1300     ram_block = qemu_ram_alloc_resizeable(size, max_size, resized, mr, errp);
1301     mr->ram_addr = ram_block->offset;
1302     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1303 }
1304
1305 #ifdef __linux__
1306 void memory_region_init_ram_from_file(MemoryRegion *mr,
1307                                       struct Object *owner,
1308                                       const char *name,
1309                                       uint64_t size,
1310                                       bool share,
1311                                       const char *path,
1312                                       Error **errp)
1313 {
1314     RAMBlock *ram_block;
1315
1316     memory_region_init(mr, owner, name, size);
1317     mr->ram = true;
1318     mr->terminates = true;
1319     mr->destructor = memory_region_destructor_ram;
1320     ram_block = qemu_ram_alloc_from_file(size, mr, share, path, errp);
1321     mr->ram_addr = ram_block->offset;
1322     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1323 }
1324 #endif
1325
1326 void memory_region_init_ram_ptr(MemoryRegion *mr,
1327                                 Object *owner,
1328                                 const char *name,
1329                                 uint64_t size,
1330                                 void *ptr)
1331 {
1332     RAMBlock *ram_block;
1333
1334     memory_region_init(mr, owner, name, size);
1335     mr->ram = true;
1336     mr->terminates = true;
1337     mr->destructor = memory_region_destructor_ram;
1338     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1339
1340     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1341     assert(ptr != NULL);
1342     ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1343     mr->ram_addr = ram_block->offset;
1344 }
1345
1346 void memory_region_set_skip_dump(MemoryRegion *mr)
1347 {
1348     mr->skip_dump = true;
1349 }
1350
1351 void memory_region_init_alias(MemoryRegion *mr,
1352                               Object *owner,
1353                               const char *name,
1354                               MemoryRegion *orig,
1355                               hwaddr offset,
1356                               uint64_t size)
1357 {
1358     memory_region_init(mr, owner, name, size);
1359     mr->alias = orig;
1360     mr->alias_offset = offset;
1361 }
1362
1363 void memory_region_init_rom_device(MemoryRegion *mr,
1364                                    Object *owner,
1365                                    const MemoryRegionOps *ops,
1366                                    void *opaque,
1367                                    const char *name,
1368                                    uint64_t size,
1369                                    Error **errp)
1370 {
1371     RAMBlock *ram_block;
1372
1373     memory_region_init(mr, owner, name, size);
1374     mr->ops = ops;
1375     mr->opaque = opaque;
1376     mr->terminates = true;
1377     mr->rom_device = true;
1378     mr->destructor = memory_region_destructor_rom_device;
1379     ram_block = qemu_ram_alloc(size, mr, errp);
1380     mr->ram_addr = ram_block->offset;
1381 }
1382
1383 void memory_region_init_iommu(MemoryRegion *mr,
1384                               Object *owner,
1385                               const MemoryRegionIOMMUOps *ops,
1386                               const char *name,
1387                               uint64_t size)
1388 {
1389     memory_region_init(mr, owner, name, size);
1390     mr->iommu_ops = ops,
1391     mr->terminates = true;  /* then re-forwards */
1392     notifier_list_init(&mr->iommu_notify);
1393 }
1394
1395 static void memory_region_finalize(Object *obj)
1396 {
1397     MemoryRegion *mr = MEMORY_REGION(obj);
1398
1399     assert(!mr->container);
1400
1401     /* We know the region is not visible in any address space (it
1402      * does not have a container and cannot be a root either because
1403      * it has no references, so we can blindly clear mr->enabled.
1404      * memory_region_set_enabled instead could trigger a transaction
1405      * and cause an infinite loop.
1406      */
1407     mr->enabled = false;
1408     memory_region_transaction_begin();
1409     while (!QTAILQ_EMPTY(&mr->subregions)) {
1410         MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1411         memory_region_del_subregion(mr, subregion);
1412     }
1413     memory_region_transaction_commit();
1414
1415     mr->destructor(mr);
1416     memory_region_clear_coalescing(mr);
1417     g_free((char *)mr->name);
1418     g_free(mr->ioeventfds);
1419 }
1420
1421 Object *memory_region_owner(MemoryRegion *mr)
1422 {
1423     Object *obj = OBJECT(mr);
1424     return obj->parent;
1425 }
1426
1427 void memory_region_ref(MemoryRegion *mr)
1428 {
1429     /* MMIO callbacks most likely will access data that belongs
1430      * to the owner, hence the need to ref/unref the owner whenever
1431      * the memory region is in use.
1432      *
1433      * The memory region is a child of its owner.  As long as the
1434      * owner doesn't call unparent itself on the memory region,
1435      * ref-ing the owner will also keep the memory region alive.
1436      * Memory regions without an owner are supposed to never go away;
1437      * we do not ref/unref them because it slows down DMA sensibly.
1438      */
1439     if (mr && mr->owner) {
1440         object_ref(mr->owner);
1441     }
1442 }
1443
1444 void memory_region_unref(MemoryRegion *mr)
1445 {
1446     if (mr && mr->owner) {
1447         object_unref(mr->owner);
1448     }
1449 }
1450
1451 uint64_t memory_region_size(MemoryRegion *mr)
1452 {
1453     if (int128_eq(mr->size, int128_2_64())) {
1454         return UINT64_MAX;
1455     }
1456     return int128_get64(mr->size);
1457 }
1458
1459 const char *memory_region_name(const MemoryRegion *mr)
1460 {
1461     if (!mr->name) {
1462         ((MemoryRegion *)mr)->name =
1463             object_get_canonical_path_component(OBJECT(mr));
1464     }
1465     return mr->name;
1466 }
1467
1468 bool memory_region_is_skip_dump(MemoryRegion *mr)
1469 {
1470     return mr->skip_dump;
1471 }
1472
1473 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1474 {
1475     uint8_t mask = mr->dirty_log_mask;
1476     if (global_dirty_log) {
1477         mask |= (1 << DIRTY_MEMORY_MIGRATION);
1478     }
1479     return mask;
1480 }
1481
1482 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1483 {
1484     return memory_region_get_dirty_log_mask(mr) & (1 << client);
1485 }
1486
1487 void memory_region_register_iommu_notifier(MemoryRegion *mr, Notifier *n)
1488 {
1489     notifier_list_add(&mr->iommu_notify, n);
1490 }
1491
1492 void memory_region_iommu_replay(MemoryRegion *mr, Notifier *n,
1493                                 hwaddr granularity, bool is_write)
1494 {
1495     hwaddr addr;
1496     IOMMUTLBEntry iotlb;
1497
1498     for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1499         iotlb = mr->iommu_ops->translate(mr, addr, is_write);
1500         if (iotlb.perm != IOMMU_NONE) {
1501             n->notify(n, &iotlb);
1502         }
1503
1504         /* if (2^64 - MR size) < granularity, it's possible to get an
1505          * infinite loop here.  This should catch such a wraparound */
1506         if ((addr + granularity) < addr) {
1507             break;
1508         }
1509     }
1510 }
1511
1512 void memory_region_unregister_iommu_notifier(Notifier *n)
1513 {
1514     notifier_remove(n);
1515 }
1516
1517 void memory_region_notify_iommu(MemoryRegion *mr,
1518                                 IOMMUTLBEntry entry)
1519 {
1520     assert(memory_region_is_iommu(mr));
1521     notifier_list_notify(&mr->iommu_notify, &entry);
1522 }
1523
1524 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1525 {
1526     uint8_t mask = 1 << client;
1527     uint8_t old_logging;
1528
1529     assert(client == DIRTY_MEMORY_VGA);
1530     old_logging = mr->vga_logging_count;
1531     mr->vga_logging_count += log ? 1 : -1;
1532     if (!!old_logging == !!mr->vga_logging_count) {
1533         return;
1534     }
1535
1536     memory_region_transaction_begin();
1537     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1538     memory_region_update_pending |= mr->enabled;
1539     memory_region_transaction_commit();
1540 }
1541
1542 bool memory_region_get_dirty(MemoryRegion *mr, hwaddr addr,
1543                              hwaddr size, unsigned client)
1544 {
1545     assert(mr->ram_addr != RAM_ADDR_INVALID);
1546     return cpu_physical_memory_get_dirty(mr->ram_addr + addr, size, client);
1547 }
1548
1549 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1550                              hwaddr size)
1551 {
1552     assert(mr->ram_addr != RAM_ADDR_INVALID);
1553     cpu_physical_memory_set_dirty_range(mr->ram_addr + addr, size,
1554                                         memory_region_get_dirty_log_mask(mr));
1555 }
1556
1557 bool memory_region_test_and_clear_dirty(MemoryRegion *mr, hwaddr addr,
1558                                         hwaddr size, unsigned client)
1559 {
1560     assert(mr->ram_addr != RAM_ADDR_INVALID);
1561     return cpu_physical_memory_test_and_clear_dirty(mr->ram_addr + addr,
1562                                                     size, client);
1563 }
1564
1565
1566 void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1567 {
1568     AddressSpace *as;
1569     FlatRange *fr;
1570
1571     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1572         FlatView *view = address_space_get_flatview(as);
1573         FOR_EACH_FLAT_RANGE(fr, view) {
1574             if (fr->mr == mr) {
1575                 MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
1576             }
1577         }
1578         flatview_unref(view);
1579     }
1580 }
1581
1582 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
1583 {
1584     if (mr->readonly != readonly) {
1585         memory_region_transaction_begin();
1586         mr->readonly = readonly;
1587         memory_region_update_pending |= mr->enabled;
1588         memory_region_transaction_commit();
1589     }
1590 }
1591
1592 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
1593 {
1594     if (mr->romd_mode != romd_mode) {
1595         memory_region_transaction_begin();
1596         mr->romd_mode = romd_mode;
1597         memory_region_update_pending |= mr->enabled;
1598         memory_region_transaction_commit();
1599     }
1600 }
1601
1602 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
1603                                hwaddr size, unsigned client)
1604 {
1605     assert(mr->ram_addr != RAM_ADDR_INVALID);
1606     cpu_physical_memory_test_and_clear_dirty(mr->ram_addr + addr, size,
1607                                              client);
1608 }
1609
1610 int memory_region_get_fd(MemoryRegion *mr)
1611 {
1612     if (mr->alias) {
1613         return memory_region_get_fd(mr->alias);
1614     }
1615
1616     assert(mr->ram_addr != RAM_ADDR_INVALID);
1617
1618     return qemu_get_ram_fd(mr->ram_addr & TARGET_PAGE_MASK);
1619 }
1620
1621 void *memory_region_get_ram_ptr(MemoryRegion *mr)
1622 {
1623     void *ptr;
1624     uint64_t offset = 0;
1625
1626     rcu_read_lock();
1627     while (mr->alias) {
1628         offset += mr->alias_offset;
1629         mr = mr->alias;
1630     }
1631     assert(mr->ram_addr != RAM_ADDR_INVALID);
1632     ptr = qemu_get_ram_ptr(mr->ram_block, mr->ram_addr & TARGET_PAGE_MASK);
1633     rcu_read_unlock();
1634
1635     return ptr + offset;
1636 }
1637
1638 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
1639 {
1640     assert(mr->ram_addr != RAM_ADDR_INVALID);
1641
1642     qemu_ram_resize(mr->ram_addr, newsize, errp);
1643 }
1644
1645 static void memory_region_update_coalesced_range_as(MemoryRegion *mr, AddressSpace *as)
1646 {
1647     FlatView *view;
1648     FlatRange *fr;
1649     CoalescedMemoryRange *cmr;
1650     AddrRange tmp;
1651     MemoryRegionSection section;
1652
1653     view = address_space_get_flatview(as);
1654     FOR_EACH_FLAT_RANGE(fr, view) {
1655         if (fr->mr == mr) {
1656             section = (MemoryRegionSection) {
1657                 .address_space = as,
1658                 .offset_within_address_space = int128_get64(fr->addr.start),
1659                 .size = fr->addr.size,
1660             };
1661
1662             MEMORY_LISTENER_CALL(coalesced_mmio_del, Reverse, &section,
1663                                  int128_get64(fr->addr.start),
1664                                  int128_get64(fr->addr.size));
1665             QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
1666                 tmp = addrrange_shift(cmr->addr,
1667                                       int128_sub(fr->addr.start,
1668                                                  int128_make64(fr->offset_in_region)));
1669                 if (!addrrange_intersects(tmp, fr->addr)) {
1670                     continue;
1671                 }
1672                 tmp = addrrange_intersection(tmp, fr->addr);
1673                 MEMORY_LISTENER_CALL(coalesced_mmio_add, Forward, &section,
1674                                      int128_get64(tmp.start),
1675                                      int128_get64(tmp.size));
1676             }
1677         }
1678     }
1679     flatview_unref(view);
1680 }
1681
1682 static void memory_region_update_coalesced_range(MemoryRegion *mr)
1683 {
1684     AddressSpace *as;
1685
1686     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1687         memory_region_update_coalesced_range_as(mr, as);
1688     }
1689 }
1690
1691 void memory_region_set_coalescing(MemoryRegion *mr)
1692 {
1693     memory_region_clear_coalescing(mr);
1694     memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
1695 }
1696
1697 void memory_region_add_coalescing(MemoryRegion *mr,
1698                                   hwaddr offset,
1699                                   uint64_t size)
1700 {
1701     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
1702
1703     cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
1704     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
1705     memory_region_update_coalesced_range(mr);
1706     memory_region_set_flush_coalesced(mr);
1707 }
1708
1709 void memory_region_clear_coalescing(MemoryRegion *mr)
1710 {
1711     CoalescedMemoryRange *cmr;
1712     bool updated = false;
1713
1714     qemu_flush_coalesced_mmio_buffer();
1715     mr->flush_coalesced_mmio = false;
1716
1717     while (!QTAILQ_EMPTY(&mr->coalesced)) {
1718         cmr = QTAILQ_FIRST(&mr->coalesced);
1719         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
1720         g_free(cmr);
1721         updated = true;
1722     }
1723
1724     if (updated) {
1725         memory_region_update_coalesced_range(mr);
1726     }
1727 }
1728
1729 void memory_region_set_flush_coalesced(MemoryRegion *mr)
1730 {
1731     mr->flush_coalesced_mmio = true;
1732 }
1733
1734 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
1735 {
1736     qemu_flush_coalesced_mmio_buffer();
1737     if (QTAILQ_EMPTY(&mr->coalesced)) {
1738         mr->flush_coalesced_mmio = false;
1739     }
1740 }
1741
1742 void memory_region_set_global_locking(MemoryRegion *mr)
1743 {
1744     mr->global_locking = true;
1745 }
1746
1747 void memory_region_clear_global_locking(MemoryRegion *mr)
1748 {
1749     mr->global_locking = false;
1750 }
1751
1752 static bool userspace_eventfd_warning;
1753
1754 void memory_region_add_eventfd(MemoryRegion *mr,
1755                                hwaddr addr,
1756                                unsigned size,
1757                                bool match_data,
1758                                uint64_t data,
1759                                EventNotifier *e)
1760 {
1761     MemoryRegionIoeventfd mrfd = {
1762         .addr.start = int128_make64(addr),
1763         .addr.size = int128_make64(size),
1764         .match_data = match_data,
1765         .data = data,
1766         .e = e,
1767     };
1768     unsigned i;
1769
1770     if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
1771                             userspace_eventfd_warning))) {
1772         userspace_eventfd_warning = true;
1773         error_report("Using eventfd without MMIO binding in KVM. "
1774                      "Suboptimal performance expected");
1775     }
1776
1777     if (size) {
1778         adjust_endianness(mr, &mrfd.data, size);
1779     }
1780     memory_region_transaction_begin();
1781     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1782         if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
1783             break;
1784         }
1785     }
1786     ++mr->ioeventfd_nb;
1787     mr->ioeventfds = g_realloc(mr->ioeventfds,
1788                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
1789     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
1790             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
1791     mr->ioeventfds[i] = mrfd;
1792     ioeventfd_update_pending |= mr->enabled;
1793     memory_region_transaction_commit();
1794 }
1795
1796 void memory_region_del_eventfd(MemoryRegion *mr,
1797                                hwaddr addr,
1798                                unsigned size,
1799                                bool match_data,
1800                                uint64_t data,
1801                                EventNotifier *e)
1802 {
1803     MemoryRegionIoeventfd mrfd = {
1804         .addr.start = int128_make64(addr),
1805         .addr.size = int128_make64(size),
1806         .match_data = match_data,
1807         .data = data,
1808         .e = e,
1809     };
1810     unsigned i;
1811
1812     if (size) {
1813         adjust_endianness(mr, &mrfd.data, size);
1814     }
1815     memory_region_transaction_begin();
1816     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1817         if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
1818             break;
1819         }
1820     }
1821     assert(i != mr->ioeventfd_nb);
1822     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
1823             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
1824     --mr->ioeventfd_nb;
1825     mr->ioeventfds = g_realloc(mr->ioeventfds,
1826                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
1827     ioeventfd_update_pending |= mr->enabled;
1828     memory_region_transaction_commit();
1829 }
1830
1831 static void memory_region_update_container_subregions(MemoryRegion *subregion)
1832 {
1833     hwaddr offset = subregion->addr;
1834     MemoryRegion *mr = subregion->container;
1835     MemoryRegion *other;
1836
1837     memory_region_transaction_begin();
1838
1839     memory_region_ref(subregion);
1840     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1841         if (subregion->may_overlap || other->may_overlap) {
1842             continue;
1843         }
1844         if (int128_ge(int128_make64(offset),
1845                       int128_add(int128_make64(other->addr), other->size))
1846             || int128_le(int128_add(int128_make64(offset), subregion->size),
1847                          int128_make64(other->addr))) {
1848             continue;
1849         }
1850 #if 0
1851         printf("warning: subregion collision %llx/%llx (%s) "
1852                "vs %llx/%llx (%s)\n",
1853                (unsigned long long)offset,
1854                (unsigned long long)int128_get64(subregion->size),
1855                subregion->name,
1856                (unsigned long long)other->addr,
1857                (unsigned long long)int128_get64(other->size),
1858                other->name);
1859 #endif
1860     }
1861     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1862         if (subregion->priority >= other->priority) {
1863             QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
1864             goto done;
1865         }
1866     }
1867     QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
1868 done:
1869     memory_region_update_pending |= mr->enabled && subregion->enabled;
1870     memory_region_transaction_commit();
1871 }
1872
1873 static void memory_region_add_subregion_common(MemoryRegion *mr,
1874                                                hwaddr offset,
1875                                                MemoryRegion *subregion)
1876 {
1877     assert(!subregion->container);
1878     subregion->container = mr;
1879     subregion->addr = offset;
1880     memory_region_update_container_subregions(subregion);
1881 }
1882
1883 void memory_region_add_subregion(MemoryRegion *mr,
1884                                  hwaddr offset,
1885                                  MemoryRegion *subregion)
1886 {
1887     subregion->may_overlap = false;
1888     subregion->priority = 0;
1889     memory_region_add_subregion_common(mr, offset, subregion);
1890 }
1891
1892 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1893                                          hwaddr offset,
1894                                          MemoryRegion *subregion,
1895                                          int priority)
1896 {
1897     subregion->may_overlap = true;
1898     subregion->priority = priority;
1899     memory_region_add_subregion_common(mr, offset, subregion);
1900 }
1901
1902 void memory_region_del_subregion(MemoryRegion *mr,
1903                                  MemoryRegion *subregion)
1904 {
1905     memory_region_transaction_begin();
1906     assert(subregion->container == mr);
1907     subregion->container = NULL;
1908     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
1909     memory_region_unref(subregion);
1910     memory_region_update_pending |= mr->enabled && subregion->enabled;
1911     memory_region_transaction_commit();
1912 }
1913
1914 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
1915 {
1916     if (enabled == mr->enabled) {
1917         return;
1918     }
1919     memory_region_transaction_begin();
1920     mr->enabled = enabled;
1921     memory_region_update_pending = true;
1922     memory_region_transaction_commit();
1923 }
1924
1925 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
1926 {
1927     Int128 s = int128_make64(size);
1928
1929     if (size == UINT64_MAX) {
1930         s = int128_2_64();
1931     }
1932     if (int128_eq(s, mr->size)) {
1933         return;
1934     }
1935     memory_region_transaction_begin();
1936     mr->size = s;
1937     memory_region_update_pending = true;
1938     memory_region_transaction_commit();
1939 }
1940
1941 static void memory_region_readd_subregion(MemoryRegion *mr)
1942 {
1943     MemoryRegion *container = mr->container;
1944
1945     if (container) {
1946         memory_region_transaction_begin();
1947         memory_region_ref(mr);
1948         memory_region_del_subregion(container, mr);
1949         mr->container = container;
1950         memory_region_update_container_subregions(mr);
1951         memory_region_unref(mr);
1952         memory_region_transaction_commit();
1953     }
1954 }
1955
1956 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
1957 {
1958     if (addr != mr->addr) {
1959         mr->addr = addr;
1960         memory_region_readd_subregion(mr);
1961     }
1962 }
1963
1964 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
1965 {
1966     assert(mr->alias);
1967
1968     if (offset == mr->alias_offset) {
1969         return;
1970     }
1971
1972     memory_region_transaction_begin();
1973     mr->alias_offset = offset;
1974     memory_region_update_pending |= mr->enabled;
1975     memory_region_transaction_commit();
1976 }
1977
1978 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
1979 {
1980     return mr->align;
1981 }
1982
1983 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
1984 {
1985     const AddrRange *addr = addr_;
1986     const FlatRange *fr = fr_;
1987
1988     if (int128_le(addrrange_end(*addr), fr->addr.start)) {
1989         return -1;
1990     } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
1991         return 1;
1992     }
1993     return 0;
1994 }
1995
1996 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
1997 {
1998     return bsearch(&addr, view->ranges, view->nr,
1999                    sizeof(FlatRange), cmp_flatrange_addr);
2000 }
2001
2002 bool memory_region_is_mapped(MemoryRegion *mr)
2003 {
2004     return mr->container ? true : false;
2005 }
2006
2007 /* Same as memory_region_find, but it does not add a reference to the
2008  * returned region.  It must be called from an RCU critical section.
2009  */
2010 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2011                                                   hwaddr addr, uint64_t size)
2012 {
2013     MemoryRegionSection ret = { .mr = NULL };
2014     MemoryRegion *root;
2015     AddressSpace *as;
2016     AddrRange range;
2017     FlatView *view;
2018     FlatRange *fr;
2019
2020     addr += mr->addr;
2021     for (root = mr; root->container; ) {
2022         root = root->container;
2023         addr += root->addr;
2024     }
2025
2026     as = memory_region_to_address_space(root);
2027     if (!as) {
2028         return ret;
2029     }
2030     range = addrrange_make(int128_make64(addr), int128_make64(size));
2031
2032     view = atomic_rcu_read(&as->current_map);
2033     fr = flatview_lookup(view, range);
2034     if (!fr) {
2035         return ret;
2036     }
2037
2038     while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2039         --fr;
2040     }
2041
2042     ret.mr = fr->mr;
2043     ret.address_space = as;
2044     range = addrrange_intersection(range, fr->addr);
2045     ret.offset_within_region = fr->offset_in_region;
2046     ret.offset_within_region += int128_get64(int128_sub(range.start,
2047                                                         fr->addr.start));
2048     ret.size = range.size;
2049     ret.offset_within_address_space = int128_get64(range.start);
2050     ret.readonly = fr->readonly;
2051     return ret;
2052 }
2053
2054 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2055                                        hwaddr addr, uint64_t size)
2056 {
2057     MemoryRegionSection ret;
2058     rcu_read_lock();
2059     ret = memory_region_find_rcu(mr, addr, size);
2060     if (ret.mr) {
2061         memory_region_ref(ret.mr);
2062     }
2063     rcu_read_unlock();
2064     return ret;
2065 }
2066
2067 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2068 {
2069     MemoryRegion *mr;
2070
2071     rcu_read_lock();
2072     mr = memory_region_find_rcu(container, addr, 1).mr;
2073     rcu_read_unlock();
2074     return mr && mr != container;
2075 }
2076
2077 void address_space_sync_dirty_bitmap(AddressSpace *as)
2078 {
2079     FlatView *view;
2080     FlatRange *fr;
2081
2082     view = address_space_get_flatview(as);
2083     FOR_EACH_FLAT_RANGE(fr, view) {
2084         MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
2085     }
2086     flatview_unref(view);
2087 }
2088
2089 void memory_global_dirty_log_start(void)
2090 {
2091     global_dirty_log = true;
2092
2093     MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2094
2095     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2096     memory_region_transaction_begin();
2097     memory_region_update_pending = true;
2098     memory_region_transaction_commit();
2099 }
2100
2101 void memory_global_dirty_log_stop(void)
2102 {
2103     global_dirty_log = false;
2104
2105     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2106     memory_region_transaction_begin();
2107     memory_region_update_pending = true;
2108     memory_region_transaction_commit();
2109
2110     MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2111 }
2112
2113 static void listener_add_address_space(MemoryListener *listener,
2114                                        AddressSpace *as)
2115 {
2116     FlatView *view;
2117     FlatRange *fr;
2118
2119     if (listener->address_space_filter
2120         && listener->address_space_filter != as) {
2121         return;
2122     }
2123
2124     if (listener->begin) {
2125         listener->begin(listener);
2126     }
2127     if (global_dirty_log) {
2128         if (listener->log_global_start) {
2129             listener->log_global_start(listener);
2130         }
2131     }
2132
2133     view = address_space_get_flatview(as);
2134     FOR_EACH_FLAT_RANGE(fr, view) {
2135         MemoryRegionSection section = {
2136             .mr = fr->mr,
2137             .address_space = as,
2138             .offset_within_region = fr->offset_in_region,
2139             .size = fr->addr.size,
2140             .offset_within_address_space = int128_get64(fr->addr.start),
2141             .readonly = fr->readonly,
2142         };
2143         if (fr->dirty_log_mask && listener->log_start) {
2144             listener->log_start(listener, &section, 0, fr->dirty_log_mask);
2145         }
2146         if (listener->region_add) {
2147             listener->region_add(listener, &section);
2148         }
2149     }
2150     if (listener->commit) {
2151         listener->commit(listener);
2152     }
2153     flatview_unref(view);
2154 }
2155
2156 void memory_listener_register(MemoryListener *listener, AddressSpace *filter)
2157 {
2158     MemoryListener *other = NULL;
2159     AddressSpace *as;
2160
2161     listener->address_space_filter = filter;
2162     if (QTAILQ_EMPTY(&memory_listeners)
2163         || listener->priority >= QTAILQ_LAST(&memory_listeners,
2164                                              memory_listeners)->priority) {
2165         QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
2166     } else {
2167         QTAILQ_FOREACH(other, &memory_listeners, link) {
2168             if (listener->priority < other->priority) {
2169                 break;
2170             }
2171         }
2172         QTAILQ_INSERT_BEFORE(other, listener, link);
2173     }
2174
2175     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2176         listener_add_address_space(listener, as);
2177     }
2178 }
2179
2180 void memory_listener_unregister(MemoryListener *listener)
2181 {
2182     QTAILQ_REMOVE(&memory_listeners, listener, link);
2183 }
2184
2185 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
2186 {
2187     memory_region_ref(root);
2188     memory_region_transaction_begin();
2189     as->ref_count = 1;
2190     as->root = root;
2191     as->malloced = false;
2192     as->current_map = g_new(FlatView, 1);
2193     flatview_init(as->current_map);
2194     as->ioeventfd_nb = 0;
2195     as->ioeventfds = NULL;
2196     QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
2197     as->name = g_strdup(name ? name : "anonymous");
2198     address_space_init_dispatch(as);
2199     memory_region_update_pending |= root->enabled;
2200     memory_region_transaction_commit();
2201 }
2202
2203 static void do_address_space_destroy(AddressSpace *as)
2204 {
2205     MemoryListener *listener;
2206     bool do_free = as->malloced;
2207
2208     address_space_destroy_dispatch(as);
2209
2210     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2211         assert(listener->address_space_filter != as);
2212     }
2213
2214     flatview_unref(as->current_map);
2215     g_free(as->name);
2216     g_free(as->ioeventfds);
2217     memory_region_unref(as->root);
2218     if (do_free) {
2219         g_free(as);
2220     }
2221 }
2222
2223 AddressSpace *address_space_init_shareable(MemoryRegion *root, const char *name)
2224 {
2225     AddressSpace *as;
2226
2227     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2228         if (root == as->root && as->malloced) {
2229             as->ref_count++;
2230             return as;
2231         }
2232     }
2233
2234     as = g_malloc0(sizeof *as);
2235     address_space_init(as, root, name);
2236     as->malloced = true;
2237     return as;
2238 }
2239
2240 void address_space_destroy(AddressSpace *as)
2241 {
2242     MemoryRegion *root = as->root;
2243
2244     as->ref_count--;
2245     if (as->ref_count) {
2246         return;
2247     }
2248     /* Flush out anything from MemoryListeners listening in on this */
2249     memory_region_transaction_begin();
2250     as->root = NULL;
2251     memory_region_transaction_commit();
2252     QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
2253     address_space_unregister(as);
2254
2255     /* At this point, as->dispatch and as->current_map are dummy
2256      * entries that the guest should never use.  Wait for the old
2257      * values to expire before freeing the data.
2258      */
2259     as->root = root;
2260     call_rcu(as, do_address_space_destroy, rcu);
2261 }
2262
2263 typedef struct MemoryRegionList MemoryRegionList;
2264
2265 struct MemoryRegionList {
2266     const MemoryRegion *mr;
2267     QTAILQ_ENTRY(MemoryRegionList) queue;
2268 };
2269
2270 typedef QTAILQ_HEAD(queue, MemoryRegionList) MemoryRegionListHead;
2271
2272 static void mtree_print_mr(fprintf_function mon_printf, void *f,
2273                            const MemoryRegion *mr, unsigned int level,
2274                            hwaddr base,
2275                            MemoryRegionListHead *alias_print_queue)
2276 {
2277     MemoryRegionList *new_ml, *ml, *next_ml;
2278     MemoryRegionListHead submr_print_queue;
2279     const MemoryRegion *submr;
2280     unsigned int i;
2281
2282     if (!mr) {
2283         return;
2284     }
2285
2286     for (i = 0; i < level; i++) {
2287         mon_printf(f, "  ");
2288     }
2289
2290     if (mr->alias) {
2291         MemoryRegionList *ml;
2292         bool found = false;
2293
2294         /* check if the alias is already in the queue */
2295         QTAILQ_FOREACH(ml, alias_print_queue, queue) {
2296             if (ml->mr == mr->alias) {
2297                 found = true;
2298             }
2299         }
2300
2301         if (!found) {
2302             ml = g_new(MemoryRegionList, 1);
2303             ml->mr = mr->alias;
2304             QTAILQ_INSERT_TAIL(alias_print_queue, ml, queue);
2305         }
2306         mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx
2307                    " (prio %d, %c%c): alias %s @%s " TARGET_FMT_plx
2308                    "-" TARGET_FMT_plx "%s\n",
2309                    base + mr->addr,
2310                    base + mr->addr
2311                    + (int128_nz(mr->size) ?
2312                       (hwaddr)int128_get64(int128_sub(mr->size,
2313                                                       int128_one())) : 0),
2314                    mr->priority,
2315                    mr->romd_mode ? 'R' : '-',
2316                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2317                                                                        : '-',
2318                    memory_region_name(mr),
2319                    memory_region_name(mr->alias),
2320                    mr->alias_offset,
2321                    mr->alias_offset
2322                    + (int128_nz(mr->size) ?
2323                       (hwaddr)int128_get64(int128_sub(mr->size,
2324                                                       int128_one())) : 0),
2325                    mr->enabled ? "" : " [disabled]");
2326     } else {
2327         mon_printf(f,
2328                    TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d, %c%c): %s%s\n",
2329                    base + mr->addr,
2330                    base + mr->addr
2331                    + (int128_nz(mr->size) ?
2332                       (hwaddr)int128_get64(int128_sub(mr->size,
2333                                                       int128_one())) : 0),
2334                    mr->priority,
2335                    mr->romd_mode ? 'R' : '-',
2336                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2337                                                                        : '-',
2338                    memory_region_name(mr),
2339                    mr->enabled ? "" : " [disabled]");
2340     }
2341
2342     QTAILQ_INIT(&submr_print_queue);
2343
2344     QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
2345         new_ml = g_new(MemoryRegionList, 1);
2346         new_ml->mr = submr;
2347         QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2348             if (new_ml->mr->addr < ml->mr->addr ||
2349                 (new_ml->mr->addr == ml->mr->addr &&
2350                  new_ml->mr->priority > ml->mr->priority)) {
2351                 QTAILQ_INSERT_BEFORE(ml, new_ml, queue);
2352                 new_ml = NULL;
2353                 break;
2354             }
2355         }
2356         if (new_ml) {
2357             QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, queue);
2358         }
2359     }
2360
2361     QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2362         mtree_print_mr(mon_printf, f, ml->mr, level + 1, base + mr->addr,
2363                        alias_print_queue);
2364     }
2365
2366     QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, queue, next_ml) {
2367         g_free(ml);
2368     }
2369 }
2370
2371 void mtree_info(fprintf_function mon_printf, void *f)
2372 {
2373     MemoryRegionListHead ml_head;
2374     MemoryRegionList *ml, *ml2;
2375     AddressSpace *as;
2376
2377     QTAILQ_INIT(&ml_head);
2378
2379     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2380         mon_printf(f, "address-space: %s\n", as->name);
2381         mtree_print_mr(mon_printf, f, as->root, 1, 0, &ml_head);
2382         mon_printf(f, "\n");
2383     }
2384
2385     /* print aliased regions */
2386     QTAILQ_FOREACH(ml, &ml_head, queue) {
2387         mon_printf(f, "memory-region: %s\n", memory_region_name(ml->mr));
2388         mtree_print_mr(mon_printf, f, ml->mr, 1, 0, &ml_head);
2389         mon_printf(f, "\n");
2390     }
2391
2392     QTAILQ_FOREACH_SAFE(ml, &ml_head, queue, ml2) {
2393         g_free(ml);
2394     }
2395 }
2396
2397 static const TypeInfo memory_region_info = {
2398     .parent             = TYPE_OBJECT,
2399     .name               = TYPE_MEMORY_REGION,
2400     .instance_size      = sizeof(MemoryRegion),
2401     .instance_init      = memory_region_initfn,
2402     .instance_finalize  = memory_region_finalize,
2403 };
2404
2405 static void memory_register_types(void)
2406 {
2407     type_register_static(&memory_region_info);
2408 }
2409
2410 type_init(memory_register_types)