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