memory.c

   1 /*
   2  * Physical memory management
   3  *
   4  * Copyright 2011 Red Hat, Inc. and/or its affiliates
   5  *
   6  * Authors:
   7  *  Avi Kivity <[email protected]>
   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
 916     if (name) {
 917         char *escaped_name = memory_region_escape_name(name);
 918         char *name_array = g_strdup_printf("%s[*]", escaped_name);
 919
 920         if (!owner) {
 921             owner = container_get(qdev_get_machine(), "/unattached");
 922         }
 923
 924         object_property_add_child(owner, name_array, OBJECT(mr), &error_abort);
 925         object_unref(OBJECT(mr));
 926         g_free(name_array);
 927         g_free(escaped_name);
 928     }
 929 }
 930
 931 static void memory_region_get_addr(Object *obj, Visitor *v, const char *name,
 932                                    void *opaque, Error **errp)
 933 {
 934     MemoryRegion *mr = MEMORY_REGION(obj);
 935     uint64_t value = mr->addr;
 936
 937     visit_type_uint64(v, name, &value, errp);
 938 }
 939
 940 static void memory_region_get_container(Object *obj, Visitor *v,
 941                                         const char *name, void *opaque,
 942                                         Error **errp)
 943 {
 944     MemoryRegion *mr = MEMORY_REGION(obj);
 945     gchar *path = (gchar *)"";
 946
 947     if (mr->container) {
 948         path = object_get_canonical_path(OBJECT(mr->container));
 949     }
 950     visit_type_str(v, name, &path, errp);
 951     if (mr->container) {
 952         g_free(path);
 953     }
 954 }
 955
 956 static Object *memory_region_resolve_container(Object *obj, void *opaque,
 957                                                const char *part)
 958 {
 959     MemoryRegion *mr = MEMORY_REGION(obj);
 960
 961     return OBJECT(mr->container);
 962 }
 963
 964 static void memory_region_get_priority(Object *obj, Visitor *v,
 965                                        const char *name, void *opaque,
 966                                        Error **errp)
 967 {
 968     MemoryRegion *mr = MEMORY_REGION(obj);
 969     int32_t value = mr->priority;
 970
 971     visit_type_int32(v, name, &value, errp);
 972 }
 973
 974 static bool memory_region_get_may_overlap(Object *obj, Error **errp)
 975 {
 976     MemoryRegion *mr = MEMORY_REGION(obj);
 977
 978     return mr->may_overlap;
 979 }
 980
 981 static void memory_region_get_size(Object *obj, Visitor *v, const char *name,
 982                                    void *opaque, Error **errp)
 983 {
 984     MemoryRegion *mr = MEMORY_REGION(obj);
 985     uint64_t value = memory_region_size(mr);
 986
 987     visit_type_uint64(v, name, &value, errp);
 988 }
 989
 990 static void memory_region_initfn(Object *obj)
 991 {
 992     MemoryRegion *mr = MEMORY_REGION(obj);
 993     ObjectProperty *op;
 994
 995     mr->ops = &unassigned_mem_ops;
 996     mr->ram_addr = RAM_ADDR_INVALID;
 997     mr->enabled = true;
 998     mr->romd_mode = true;
 999     mr->global_locking = true;
1000     mr->destructor = memory_region_destructor_none;
1001     QTAILQ_INIT(&mr->subregions);
1002     QTAILQ_INIT(&mr->coalesced);
1003
1004     op = object_property_add(OBJECT(mr), "container",
1005                              "link<" TYPE_MEMORY_REGION ">",
1006                              memory_region_get_container,
1007                              NULL, /* memory_region_set_container */
1008                              NULL, NULL, &error_abort);
1009     op->resolve = memory_region_resolve_container;
1010
1011     object_property_add(OBJECT(mr), "addr", "uint64",
1012                         memory_region_get_addr,
1013                         NULL, /* memory_region_set_addr */
1014                         NULL, NULL, &error_abort);
1015     object_property_add(OBJECT(mr), "priority", "uint32",
1016                         memory_region_get_priority,
1017                         NULL, /* memory_region_set_priority */
1018                         NULL, NULL, &error_abort);
1019     object_property_add_bool(OBJECT(mr), "may-overlap",
1020                              memory_region_get_may_overlap,
1021                              NULL, /* memory_region_set_may_overlap */
1022                              &error_abort);
1023     object_property_add(OBJECT(mr), "size", "uint64",
1024                         memory_region_get_size,
1025                         NULL, /* memory_region_set_size, */
1026                         NULL, NULL, &error_abort);
1027 }
1028
1029 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1030                                     unsigned size)
1031 {
1032 #ifdef DEBUG_UNASSIGNED
1033     printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
1034 #endif
1035     if (current_cpu != NULL) {
1036         cpu_unassigned_access(current_cpu, addr, false, false, 0, size);
1037     }
1038     return 0;
1039 }
1040
1041 static void unassigned_mem_write(void *opaque, hwaddr addr,
1042                                  uint64_t val, unsigned size)
1043 {
1044 #ifdef DEBUG_UNASSIGNED
1045     printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1046 #endif
1047     if (current_cpu != NULL) {
1048         cpu_unassigned_access(current_cpu, addr, true, false, 0, size);
1049     }
1050 }
1051
1052 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1053                                    unsigned size, bool is_write)
1054 {
1055     return false;
1056 }
1057
1058 const MemoryRegionOps unassigned_mem_ops = {
1059     .valid.accepts = unassigned_mem_accepts,
1060     .endianness = DEVICE_NATIVE_ENDIAN,
1061 };
1062
1063 bool memory_region_access_valid(MemoryRegion *mr,
1064                                 hwaddr addr,
1065                                 unsigned size,
1066                                 bool is_write)
1067 {
1068     int access_size_min, access_size_max;
1069     int access_size, i;
1070
1071     if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1072         return false;
1073     }
1074
1075     if (!mr->ops->valid.accepts) {
1076         return true;
1077     }
1078
1079     access_size_min = mr->ops->valid.min_access_size;
1080     if (!mr->ops->valid.min_access_size) {
1081         access_size_min = 1;
1082     }
1083
1084     access_size_max = mr->ops->valid.max_access_size;
1085     if (!mr->ops->valid.max_access_size) {
1086         access_size_max = 4;
1087     }
1088
1089     access_size = MAX(MIN(size, access_size_max), access_size_min);
1090     for (i = 0; i < size; i += access_size) {
1091         if (!mr->ops->valid.accepts(mr->opaque, addr + i, access_size,
1092                                     is_write)) {
1093             return false;
1094         }
1095     }
1096
1097     return true;
1098 }
1099
1100 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1101                                                 hwaddr addr,
1102                                                 uint64_t *pval,
1103                                                 unsigned size,
1104                                                 MemTxAttrs attrs)
1105 {
1106     *pval = 0;
1107
1108     if (mr->ops->read) {
1109         return access_with_adjusted_size(addr, pval, size,
1110                                          mr->ops->impl.min_access_size,
1111                                          mr->ops->impl.max_access_size,
1112                                          memory_region_read_accessor,
1113                                          mr, attrs);
1114     } else if (mr->ops->read_with_attrs) {
1115         return access_with_adjusted_size(addr, pval, size,
1116                                          mr->ops->impl.min_access_size,
1117                                          mr->ops->impl.max_access_size,
1118                                          memory_region_read_with_attrs_accessor,
1119                                          mr, attrs);
1120     } else {
1121         return access_with_adjusted_size(addr, pval, size, 1, 4,
1122                                          memory_region_oldmmio_read_accessor,
1123                                          mr, attrs);
1124     }
1125 }
1126
1127 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1128                                         hwaddr addr,
1129                                         uint64_t *pval,
1130                                         unsigned size,
1131                                         MemTxAttrs attrs)
1132 {
1133     MemTxResult r;
1134
1135     if (!memory_region_access_valid(mr, addr, size, false)) {
1136         *pval = unassigned_mem_read(mr, addr, size);
1137         return MEMTX_DECODE_ERROR;
1138     }
1139
1140     r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1141     adjust_endianness(mr, pval, size);
1142     return r;
1143 }
1144
1145 /* Return true if an eventfd was signalled */
1146 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1147                                                     hwaddr addr,
1148                                                     uint64_t data,
1149                                                     unsigned size,
1150                                                     MemTxAttrs attrs)
1151 {
1152     MemoryRegionIoeventfd ioeventfd = {
1153         .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1154         .data = data,
1155     };
1156     unsigned i;
1157
1158     for (i = 0; i < mr->ioeventfd_nb; i++) {
1159         ioeventfd.match_data = mr->ioeventfds[i].match_data;
1160         ioeventfd.e = mr->ioeventfds[i].e;
1161
1162         if (memory_region_ioeventfd_equal(ioeventfd, mr->ioeventfds[i])) {
1163             event_notifier_set(ioeventfd.e);
1164             return true;
1165         }
1166     }
1167
1168     return false;
1169 }
1170
1171 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1172                                          hwaddr addr,
1173                                          uint64_t data,
1174                                          unsigned size,
1175                                          MemTxAttrs attrs)
1176 {
1177     if (!memory_region_access_valid(mr, addr, size, true)) {
1178         unassigned_mem_write(mr, addr, data, size);
1179         return MEMTX_DECODE_ERROR;
1180     }
1181
1182     adjust_endianness(mr, &data, size);
1183
1184     if ((!kvm_eventfds_enabled()) &&
1185         memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1186         return MEMTX_OK;
1187     }
1188
1189     if (mr->ops->write) {
1190         return access_with_adjusted_size(addr, &data, size,
1191                                          mr->ops->impl.min_access_size,
1192                                          mr->ops->impl.max_access_size,
1193                                          memory_region_write_accessor, mr,
1194                                          attrs);
1195     } else if (mr->ops->write_with_attrs) {
1196         return
1197             access_with_adjusted_size(addr, &data, size,
1198                                       mr->ops->impl.min_access_size,
1199                                       mr->ops->impl.max_access_size,
1200                                       memory_region_write_with_attrs_accessor,
1201                                       mr, attrs);
1202     } else {
1203         return access_with_adjusted_size(addr, &data, size, 1, 4,
1204                                          memory_region_oldmmio_write_accessor,
1205                                          mr, attrs);
1206     }
1207 }
1208
1209 void memory_region_init_io(MemoryRegion *mr,
1210                            Object *owner,
1211                            const MemoryRegionOps *ops,
1212                            void *opaque,
1213                            const char *name,
1214                            uint64_t size)
1215 {
1216     memory_region_init(mr, owner, name, size);
1217     mr->ops = ops ? ops : &unassigned_mem_ops;
1218     mr->opaque = opaque;
1219     mr->terminates = true;
1220 }
1221
1222 void memory_region_init_ram(MemoryRegion *mr,
1223                             Object *owner,
1224                             const char *name,
1225                             uint64_t size,
1226                             Error **errp)
1227 {
1228     memory_region_init(mr, owner, name, size);
1229     mr->ram = true;
1230     mr->terminates = true;
1231     mr->destructor = memory_region_destructor_ram;
1232     mr->ram_addr = qemu_ram_alloc(size, mr, errp);
1233     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1234 }
1235
1236 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1237                                        Object *owner,
1238                                        const char *name,
1239                                        uint64_t size,
1240                                        uint64_t max_size,
1241                                        void (*resized)(const char*,
1242                                                        uint64_t length,
1243                                                        void *host),
1244                                        Error **errp)
1245 {
1246     memory_region_init(mr, owner, name, size);
1247     mr->ram = true;
1248     mr->terminates = true;
1249     mr->destructor = memory_region_destructor_ram;
1250     mr->ram_addr = qemu_ram_alloc_resizeable(size, max_size, resized, mr, errp);
1251     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1252 }
1253
1254 #ifdef __linux__
1255 void memory_region_init_ram_from_file(MemoryRegion *mr,
1256                                       struct Object *owner,
1257                                       const char *name,
1258                                       uint64_t size,
1259                                       bool share,
1260                                       const char *path,
1261                                       Error **errp)
1262 {
1263     memory_region_init(mr, owner, name, size);
1264     mr->ram = true;
1265     mr->terminates = true;
1266     mr->destructor = memory_region_destructor_ram;
1267     mr->ram_addr = qemu_ram_alloc_from_file(size, mr, share, path, errp);
1268     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1269 }
1270 #endif
1271
1272 void memory_region_init_ram_ptr(MemoryRegion *mr,
1273                                 Object *owner,
1274                                 const char *name,
1275                                 uint64_t size,
1276                                 void *ptr)
1277 {
1278     memory_region_init(mr, owner, name, size);
1279     mr->ram = true;
1280     mr->terminates = true;
1281     mr->destructor = memory_region_destructor_ram;
1282     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1283
1284     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1285     assert(ptr != NULL);
1286     mr->ram_addr = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1287 }
1288
1289 void memory_region_set_skip_dump(MemoryRegion *mr)
1290 {
1291     mr->skip_dump = true;
1292 }
1293
1294 void memory_region_init_alias(MemoryRegion *mr,
1295                               Object *owner,
1296                               const char *name,
1297                               MemoryRegion *orig,
1298                               hwaddr offset,
1299                               uint64_t size)
1300 {
1301     memory_region_init(mr, owner, name, size);
1302     mr->alias = orig;
1303     mr->alias_offset = offset;
1304 }
1305
1306 void memory_region_init_rom_device(MemoryRegion *mr,
1307                                    Object *owner,
1308                                    const MemoryRegionOps *ops,
1309                                    void *opaque,
1310                                    const char *name,
1311                                    uint64_t size,
1312                                    Error **errp)
1313 {
1314     memory_region_init(mr, owner, name, size);
1315     mr->ops = ops;
1316     mr->opaque = opaque;
1317     mr->terminates = true;
1318     mr->rom_device = true;
1319     mr->destructor = memory_region_destructor_rom_device;
1320     mr->ram_addr = qemu_ram_alloc(size, mr, errp);
1321 }
1322
1323 void memory_region_init_iommu(MemoryRegion *mr,
1324                               Object *owner,
1325                               const MemoryRegionIOMMUOps *ops,
1326                               const char *name,
1327                               uint64_t size)
1328 {
1329     memory_region_init(mr, owner, name, size);
1330     mr->iommu_ops = ops,
1331     mr->terminates = true;  /* then re-forwards */
1332     notifier_list_init(&mr->iommu_notify);
1333 }
1334
1335 static void memory_region_finalize(Object *obj)
1336 {
1337     MemoryRegion *mr = MEMORY_REGION(obj);
1338
1339     assert(!mr->container);
1340
1341     /* We know the region is not visible in any address space (it
1342      * does not have a container and cannot be a root either because
1343      * it has no references, so we can blindly clear mr->enabled.
1344      * memory_region_set_enabled instead could trigger a transaction
1345      * and cause an infinite loop.
1346      */
1347     mr->enabled = false;
1348     memory_region_transaction_begin();
1349     while (!QTAILQ_EMPTY(&mr->subregions)) {
1350         MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1351         memory_region_del_subregion(mr, subregion);
1352     }
1353     memory_region_transaction_commit();
1354
1355     mr->destructor(mr);
1356     memory_region_clear_coalescing(mr);
1357     g_free((char *)mr->name);
1358     g_free(mr->ioeventfds);
1359 }
1360
1361 Object *memory_region_owner(MemoryRegion *mr)
1362 {
1363     Object *obj = OBJECT(mr);
1364     return obj->parent;
1365 }
1366
1367 void memory_region_ref(MemoryRegion *mr)
1368 {
1369     /* MMIO callbacks most likely will access data that belongs
1370      * to the owner, hence the need to ref/unref the owner whenever
1371      * the memory region is in use.
1372      *
1373      * The memory region is a child of its owner.  As long as the
1374      * owner doesn't call unparent itself on the memory region,
1375      * ref-ing the owner will also keep the memory region alive.
1376      * Memory regions without an owner are supposed to never go away;
1377      * we do not ref/unref them because it slows down DMA sensibly.
1378      */
1379     if (mr && mr->owner) {
1380         object_ref(mr->owner);
1381     }
1382 }
1383
1384 void memory_region_unref(MemoryRegion *mr)
1385 {
1386     if (mr && mr->owner) {
1387         object_unref(mr->owner);
1388     }
1389 }
1390
1391 uint64_t memory_region_size(MemoryRegion *mr)
1392 {
1393     if (int128_eq(mr->size, int128_2_64())) {
1394         return UINT64_MAX;
1395     }
1396     return int128_get64(mr->size);
1397 }
1398
1399 const char *memory_region_name(const MemoryRegion *mr)
1400 {
1401     if (!mr->name) {
1402         ((MemoryRegion *)mr)->name =
1403             object_get_canonical_path_component(OBJECT(mr));
1404     }
1405     return mr->name;
1406 }
1407
1408 bool memory_region_is_skip_dump(MemoryRegion *mr)
1409 {
1410     return mr->skip_dump;
1411 }
1412
1413 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1414 {
1415     uint8_t mask = mr->dirty_log_mask;
1416     if (global_dirty_log) {
1417         mask |= (1 << DIRTY_MEMORY_MIGRATION);
1418     }
1419     return mask;
1420 }
1421
1422 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1423 {
1424     return memory_region_get_dirty_log_mask(mr) & (1 << client);
1425 }
1426
1427 void memory_region_register_iommu_notifier(MemoryRegion *mr, Notifier *n)
1428 {
1429     notifier_list_add(&mr->iommu_notify, n);
1430 }
1431
1432 void memory_region_iommu_replay(MemoryRegion *mr, Notifier *n,
1433                                 hwaddr granularity, bool is_write)
1434 {
1435     hwaddr addr;
1436     IOMMUTLBEntry iotlb;
1437
1438     for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1439         iotlb = mr->iommu_ops->translate(mr, addr, is_write);
1440         if (iotlb.perm != IOMMU_NONE) {
1441             n->notify(n, &iotlb);
1442         }
1443
1444         /* if (2^64 - MR size) < granularity, it's possible to get an
1445          * infinite loop here.  This should catch such a wraparound */
1446         if ((addr + granularity) < addr) {
1447             break;
1448         }
1449     }
1450 }
1451
1452 void memory_region_unregister_iommu_notifier(Notifier *n)
1453 {
1454     notifier_remove(n);
1455 }
1456
1457 void memory_region_notify_iommu(MemoryRegion *mr,
1458                                 IOMMUTLBEntry entry)
1459 {
1460     assert(memory_region_is_iommu(mr));
1461     notifier_list_notify(&mr->iommu_notify, &entry);
1462 }
1463
1464 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1465 {
1466     uint8_t mask = 1 << client;
1467     uint8_t old_logging;
1468
1469     assert(client == DIRTY_MEMORY_VGA);
1470     old_logging = mr->vga_logging_count;
1471     mr->vga_logging_count += log ? 1 : -1;
1472     if (!!old_logging == !!mr->vga_logging_count) {
1473         return;
1474     }
1475
1476     memory_region_transaction_begin();
1477     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1478     memory_region_update_pending |= mr->enabled;
1479     memory_region_transaction_commit();
1480 }
1481
1482 bool memory_region_get_dirty(MemoryRegion *mr, hwaddr addr,
1483                              hwaddr size, unsigned client)
1484 {
1485     assert(mr->ram_addr != RAM_ADDR_INVALID);
1486     return cpu_physical_memory_get_dirty(mr->ram_addr + addr, size, client);
1487 }
1488
1489 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1490                              hwaddr size)
1491 {
1492     assert(mr->ram_addr != RAM_ADDR_INVALID);
1493     cpu_physical_memory_set_dirty_range(mr->ram_addr + addr, size,
1494                                         memory_region_get_dirty_log_mask(mr));
1495 }
1496
1497 bool memory_region_test_and_clear_dirty(MemoryRegion *mr, hwaddr addr,
1498                                         hwaddr size, unsigned client)
1499 {
1500     assert(mr->ram_addr != RAM_ADDR_INVALID);
1501     return cpu_physical_memory_test_and_clear_dirty(mr->ram_addr + addr,
1502                                                     size, client);
1503 }
1504
1505
1506 void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1507 {
1508     AddressSpace *as;
1509     FlatRange *fr;
1510
1511     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1512         FlatView *view = address_space_get_flatview(as);
1513         FOR_EACH_FLAT_RANGE(fr, view) {
1514             if (fr->mr == mr) {
1515                 MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
1516             }
1517         }
1518         flatview_unref(view);
1519     }
1520 }
1521
1522 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
1523 {
1524     if (mr->readonly != readonly) {
1525         memory_region_transaction_begin();
1526         mr->readonly = readonly;
1527         memory_region_update_pending |= mr->enabled;
1528         memory_region_transaction_commit();
1529     }
1530 }
1531
1532 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
1533 {
1534     if (mr->romd_mode != romd_mode) {
1535         memory_region_transaction_begin();
1536         mr->romd_mode = romd_mode;
1537         memory_region_update_pending |= mr->enabled;
1538         memory_region_transaction_commit();
1539     }
1540 }
1541
1542 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
1543                                hwaddr size, unsigned client)
1544 {
1545     assert(mr->ram_addr != RAM_ADDR_INVALID);
1546     cpu_physical_memory_test_and_clear_dirty(mr->ram_addr + addr, size,
1547                                              client);
1548 }
1549
1550 int memory_region_get_fd(MemoryRegion *mr)
1551 {
1552     if (mr->alias) {
1553         return memory_region_get_fd(mr->alias);
1554     }
1555
1556     assert(mr->ram_addr != RAM_ADDR_INVALID);
1557
1558     return qemu_get_ram_fd(mr->ram_addr & TARGET_PAGE_MASK);
1559 }
1560
1561 void *memory_region_get_ram_ptr(MemoryRegion *mr)
1562 {
1563     void *ptr;
1564     uint64_t offset = 0;
1565
1566     rcu_read_lock();
1567     while (mr->alias) {
1568         offset += mr->alias_offset;
1569         mr = mr->alias;
1570     }
1571     assert(mr->ram_addr != RAM_ADDR_INVALID);
1572     ptr = qemu_get_ram_ptr(mr->ram_addr & TARGET_PAGE_MASK);
1573     rcu_read_unlock();
1574
1575     return ptr + offset;
1576 }
1577
1578 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
1579 {
1580     assert(mr->ram_addr != RAM_ADDR_INVALID);
1581
1582     qemu_ram_resize(mr->ram_addr, newsize, errp);
1583 }
1584
1585 static void memory_region_update_coalesced_range_as(MemoryRegion *mr, AddressSpace *as)
1586 {
1587     FlatView *view;
1588     FlatRange *fr;
1589     CoalescedMemoryRange *cmr;
1590     AddrRange tmp;
1591     MemoryRegionSection section;
1592
1593     view = address_space_get_flatview(as);
1594     FOR_EACH_FLAT_RANGE(fr, view) {
1595         if (fr->mr == mr) {
1596             section = (MemoryRegionSection) {
1597                 .address_space = as,
1598                 .offset_within_address_space = int128_get64(fr->addr.start),
1599                 .size = fr->addr.size,
1600             };
1601
1602             MEMORY_LISTENER_CALL(coalesced_mmio_del, Reverse, &section,
1603                                  int128_get64(fr->addr.start),
1604                                  int128_get64(fr->addr.size));
1605             QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
1606                 tmp = addrrange_shift(cmr->addr,
1607                                       int128_sub(fr->addr.start,
1608                                                  int128_make64(fr->offset_in_region)));
1609                 if (!addrrange_intersects(tmp, fr->addr)) {
1610                     continue;
1611                 }
1612                 tmp = addrrange_intersection(tmp, fr->addr);
1613                 MEMORY_LISTENER_CALL(coalesced_mmio_add, Forward, &section,
1614                                      int128_get64(tmp.start),
1615                                      int128_get64(tmp.size));
1616             }
1617         }
1618     }
1619     flatview_unref(view);
1620 }
1621
1622 static void memory_region_update_coalesced_range(MemoryRegion *mr)
1623 {
1624     AddressSpace *as;
1625
1626     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1627         memory_region_update_coalesced_range_as(mr, as);
1628     }
1629 }
1630
1631 void memory_region_set_coalescing(MemoryRegion *mr)
1632 {
1633     memory_region_clear_coalescing(mr);
1634     memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
1635 }
1636
1637 void memory_region_add_coalescing(MemoryRegion *mr,
1638                                   hwaddr offset,
1639                                   uint64_t size)
1640 {
1641     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
1642
1643     cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
1644     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
1645     memory_region_update_coalesced_range(mr);
1646     memory_region_set_flush_coalesced(mr);
1647 }
1648
1649 void memory_region_clear_coalescing(MemoryRegion *mr)
1650 {
1651     CoalescedMemoryRange *cmr;
1652     bool updated = false;
1653
1654     qemu_flush_coalesced_mmio_buffer();
1655     mr->flush_coalesced_mmio = false;
1656
1657     while (!QTAILQ_EMPTY(&mr->coalesced)) {
1658         cmr = QTAILQ_FIRST(&mr->coalesced);
1659         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
1660         g_free(cmr);
1661         updated = true;
1662     }
1663
1664     if (updated) {
1665         memory_region_update_coalesced_range(mr);
1666     }
1667 }
1668
1669 void memory_region_set_flush_coalesced(MemoryRegion *mr)
1670 {
1671     mr->flush_coalesced_mmio = true;
1672 }
1673
1674 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
1675 {
1676     qemu_flush_coalesced_mmio_buffer();
1677     if (QTAILQ_EMPTY(&mr->coalesced)) {
1678         mr->flush_coalesced_mmio = false;
1679     }
1680 }
1681
1682 void memory_region_set_global_locking(MemoryRegion *mr)
1683 {
1684     mr->global_locking = true;
1685 }
1686
1687 void memory_region_clear_global_locking(MemoryRegion *mr)
1688 {
1689     mr->global_locking = false;
1690 }
1691
1692 static bool userspace_eventfd_warning;
1693
1694 void memory_region_add_eventfd(MemoryRegion *mr,
1695                                hwaddr addr,
1696                                unsigned size,
1697                                bool match_data,
1698                                uint64_t data,
1699                                EventNotifier *e)
1700 {
1701     MemoryRegionIoeventfd mrfd = {
1702         .addr.start = int128_make64(addr),
1703         .addr.size = int128_make64(size),
1704         .match_data = match_data,
1705         .data = data,
1706         .e = e,
1707     };
1708     unsigned i;
1709
1710     if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
1711                             userspace_eventfd_warning))) {
1712         userspace_eventfd_warning = true;
1713         error_report("Using eventfd without MMIO binding in KVM. "
1714                      "Suboptimal performance expected");
1715     }
1716
1717     if (size) {
1718         adjust_endianness(mr, &mrfd.data, size);
1719     }
1720     memory_region_transaction_begin();
1721     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1722         if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
1723             break;
1724         }
1725     }
1726     ++mr->ioeventfd_nb;
1727     mr->ioeventfds = g_realloc(mr->ioeventfds,
1728                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
1729     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
1730             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
1731     mr->ioeventfds[i] = mrfd;
1732     ioeventfd_update_pending |= mr->enabled;
1733     memory_region_transaction_commit();
1734 }
1735
1736 void memory_region_del_eventfd(MemoryRegion *mr,
1737                                hwaddr addr,
1738                                unsigned size,
1739                                bool match_data,
1740                                uint64_t data,
1741                                EventNotifier *e)
1742 {
1743     MemoryRegionIoeventfd mrfd = {
1744         .addr.start = int128_make64(addr),
1745         .addr.size = int128_make64(size),
1746         .match_data = match_data,
1747         .data = data,
1748         .e = e,
1749     };
1750     unsigned i;
1751
1752     if (size) {
1753         adjust_endianness(mr, &mrfd.data, size);
1754     }
1755     memory_region_transaction_begin();
1756     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1757         if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
1758             break;
1759         }
1760     }
1761     assert(i != mr->ioeventfd_nb);
1762     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
1763             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
1764     --mr->ioeventfd_nb;
1765     mr->ioeventfds = g_realloc(mr->ioeventfds,
1766                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
1767     ioeventfd_update_pending |= mr->enabled;
1768     memory_region_transaction_commit();
1769 }
1770
1771 static void memory_region_update_container_subregions(MemoryRegion *subregion)
1772 {
1773     hwaddr offset = subregion->addr;
1774     MemoryRegion *mr = subregion->container;
1775     MemoryRegion *other;
1776
1777     memory_region_transaction_begin();
1778
1779     memory_region_ref(subregion);
1780     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1781         if (subregion->may_overlap || other->may_overlap) {
1782             continue;
1783         }
1784         if (int128_ge(int128_make64(offset),
1785                       int128_add(int128_make64(other->addr), other->size))
1786             || int128_le(int128_add(int128_make64(offset), subregion->size),
1787                          int128_make64(other->addr))) {
1788             continue;
1789         }
1790 #if 0
1791         printf("warning: subregion collision %llx/%llx (%s) "
1792                "vs %llx/%llx (%s)\n",
1793                (unsigned long long)offset,
1794                (unsigned long long)int128_get64(subregion->size),
1795                subregion->name,
1796                (unsigned long long)other->addr,
1797                (unsigned long long)int128_get64(other->size),
1798                other->name);
1799 #endif
1800     }
1801     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1802         if (subregion->priority >= other->priority) {
1803             QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
1804             goto done;
1805         }
1806     }
1807     QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
1808 done:
1809     memory_region_update_pending |= mr->enabled && subregion->enabled;
1810     memory_region_transaction_commit();
1811 }
1812
1813 static void memory_region_add_subregion_common(MemoryRegion *mr,
1814                                                hwaddr offset,
1815                                                MemoryRegion *subregion)
1816 {
1817     assert(!subregion->container);
1818     subregion->container = mr;
1819     subregion->addr = offset;
1820     memory_region_update_container_subregions(subregion);
1821 }
1822
1823 void memory_region_add_subregion(MemoryRegion *mr,
1824                                  hwaddr offset,
1825                                  MemoryRegion *subregion)
1826 {
1827     subregion->may_overlap = false;
1828     subregion->priority = 0;
1829     memory_region_add_subregion_common(mr, offset, subregion);
1830 }
1831
1832 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1833                                          hwaddr offset,
1834                                          MemoryRegion *subregion,
1835                                          int priority)
1836 {
1837     subregion->may_overlap = true;
1838     subregion->priority = priority;
1839     memory_region_add_subregion_common(mr, offset, subregion);
1840 }
1841
1842 void memory_region_del_subregion(MemoryRegion *mr,
1843                                  MemoryRegion *subregion)
1844 {
1845     memory_region_transaction_begin();
1846     assert(subregion->container == mr);
1847     subregion->container = NULL;
1848     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
1849     memory_region_unref(subregion);
1850     memory_region_update_pending |= mr->enabled && subregion->enabled;
1851     memory_region_transaction_commit();
1852 }
1853
1854 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
1855 {
1856     if (enabled == mr->enabled) {
1857         return;
1858     }
1859     memory_region_transaction_begin();
1860     mr->enabled = enabled;
1861     memory_region_update_pending = true;
1862     memory_region_transaction_commit();
1863 }
1864
1865 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
1866 {
1867     Int128 s = int128_make64(size);
1868
1869     if (size == UINT64_MAX) {
1870         s = int128_2_64();
1871     }
1872     if (int128_eq(s, mr->size)) {
1873         return;
1874     }
1875     memory_region_transaction_begin();
1876     mr->size = s;
1877     memory_region_update_pending = true;
1878     memory_region_transaction_commit();
1879 }
1880
1881 static void memory_region_readd_subregion(MemoryRegion *mr)
1882 {
1883     MemoryRegion *container = mr->container;
1884
1885     if (container) {
1886         memory_region_transaction_begin();
1887         memory_region_ref(mr);
1888         memory_region_del_subregion(container, mr);
1889         mr->container = container;
1890         memory_region_update_container_subregions(mr);
1891         memory_region_unref(mr);
1892         memory_region_transaction_commit();
1893     }
1894 }
1895
1896 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
1897 {
1898     if (addr != mr->addr) {
1899         mr->addr = addr;
1900         memory_region_readd_subregion(mr);
1901     }
1902 }
1903
1904 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
1905 {
1906     assert(mr->alias);
1907
1908     if (offset == mr->alias_offset) {
1909         return;
1910     }
1911
1912     memory_region_transaction_begin();
1913     mr->alias_offset = offset;
1914     memory_region_update_pending |= mr->enabled;
1915     memory_region_transaction_commit();
1916 }
1917
1918 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
1919 {
1920     return mr->align;
1921 }
1922
1923 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
1924 {
1925     const AddrRange *addr = addr_;
1926     const FlatRange *fr = fr_;
1927
1928     if (int128_le(addrrange_end(*addr), fr->addr.start)) {
1929         return -1;
1930     } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
1931         return 1;
1932     }
1933     return 0;
1934 }
1935
1936 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
1937 {
1938     return bsearch(&addr, view->ranges, view->nr,
1939                    sizeof(FlatRange), cmp_flatrange_addr);
1940 }
1941
1942 bool memory_region_is_mapped(MemoryRegion *mr)
1943 {
1944     return mr->container ? true : false;
1945 }
1946
1947 /* Same as memory_region_find, but it does not add a reference to the
1948  * returned region.  It must be called from an RCU critical section.
1949  */
1950 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
1951                                                   hwaddr addr, uint64_t size)
1952 {
1953     MemoryRegionSection ret = { .mr = NULL };
1954     MemoryRegion *root;
1955     AddressSpace *as;
1956     AddrRange range;
1957     FlatView *view;
1958     FlatRange *fr;
1959
1960     addr += mr->addr;
1961     for (root = mr; root->container; ) {
1962         root = root->container;
1963         addr += root->addr;
1964     }
1965
1966     as = memory_region_to_address_space(root);
1967     if (!as) {
1968         return ret;
1969     }
1970     range = addrrange_make(int128_make64(addr), int128_make64(size));
1971
1972     view = atomic_rcu_read(&as->current_map);
1973     fr = flatview_lookup(view, range);
1974     if (!fr) {
1975         return ret;
1976     }
1977
1978     while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
1979         --fr;
1980     }
1981
1982     ret.mr = fr->mr;
1983     ret.address_space = as;
1984     range = addrrange_intersection(range, fr->addr);
1985     ret.offset_within_region = fr->offset_in_region;
1986     ret.offset_within_region += int128_get64(int128_sub(range.start,
1987                                                         fr->addr.start));
1988     ret.size = range.size;
1989     ret.offset_within_address_space = int128_get64(range.start);
1990     ret.readonly = fr->readonly;
1991     return ret;
1992 }
1993
1994 MemoryRegionSection memory_region_find(MemoryRegion *mr,
1995                                        hwaddr addr, uint64_t size)
1996 {
1997     MemoryRegionSection ret;
1998     rcu_read_lock();
1999     ret = memory_region_find_rcu(mr, addr, size);
2000     if (ret.mr) {
2001         memory_region_ref(ret.mr);
2002     }
2003     rcu_read_unlock();
2004     return ret;
2005 }
2006
2007 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2008 {
2009     MemoryRegion *mr;
2010
2011     rcu_read_lock();
2012     mr = memory_region_find_rcu(container, addr, 1).mr;
2013     rcu_read_unlock();
2014     return mr && mr != container;
2015 }
2016
2017 void address_space_sync_dirty_bitmap(AddressSpace *as)
2018 {
2019     FlatView *view;
2020     FlatRange *fr;
2021
2022     view = address_space_get_flatview(as);
2023     FOR_EACH_FLAT_RANGE(fr, view) {
2024         MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
2025     }
2026     flatview_unref(view);
2027 }
2028
2029 void memory_global_dirty_log_start(void)
2030 {
2031     global_dirty_log = true;
2032
2033     MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2034
2035     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2036     memory_region_transaction_begin();
2037     memory_region_update_pending = true;
2038     memory_region_transaction_commit();
2039 }
2040
2041 void memory_global_dirty_log_stop(void)
2042 {
2043     global_dirty_log = false;
2044
2045     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2046     memory_region_transaction_begin();
2047     memory_region_update_pending = true;
2048     memory_region_transaction_commit();
2049
2050     MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2051 }
2052
2053 static void listener_add_address_space(MemoryListener *listener,
2054                                        AddressSpace *as)
2055 {
2056     FlatView *view;
2057     FlatRange *fr;
2058
2059     if (listener->address_space_filter
2060         && listener->address_space_filter != as) {
2061         return;
2062     }
2063
2064     if (listener->begin) {
2065         listener->begin(listener);
2066     }
2067     if (global_dirty_log) {
2068         if (listener->log_global_start) {
2069             listener->log_global_start(listener);
2070         }
2071     }
2072
2073     view = address_space_get_flatview(as);
2074     FOR_EACH_FLAT_RANGE(fr, view) {
2075         MemoryRegionSection section = {
2076             .mr = fr->mr,
2077             .address_space = as,
2078             .offset_within_region = fr->offset_in_region,
2079             .size = fr->addr.size,
2080             .offset_within_address_space = int128_get64(fr->addr.start),
2081             .readonly = fr->readonly,
2082         };
2083         if (fr->dirty_log_mask && listener->log_start) {
2084             listener->log_start(listener, &section, 0, fr->dirty_log_mask);
2085         }
2086         if (listener->region_add) {
2087             listener->region_add(listener, &section);
2088         }
2089     }
2090     if (listener->commit) {
2091         listener->commit(listener);
2092     }
2093     flatview_unref(view);
2094 }
2095
2096 void memory_listener_register(MemoryListener *listener, AddressSpace *filter)
2097 {
2098     MemoryListener *other = NULL;
2099     AddressSpace *as;
2100
2101     listener->address_space_filter = filter;
2102     if (QTAILQ_EMPTY(&memory_listeners)
2103         || listener->priority >= QTAILQ_LAST(&memory_listeners,
2104                                              memory_listeners)->priority) {
2105         QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
2106     } else {
2107         QTAILQ_FOREACH(other, &memory_listeners, link) {
2108             if (listener->priority < other->priority) {
2109                 break;
2110             }
2111         }
2112         QTAILQ_INSERT_BEFORE(other, listener, link);
2113     }
2114
2115     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2116         listener_add_address_space(listener, as);
2117     }
2118 }
2119
2120 void memory_listener_unregister(MemoryListener *listener)
2121 {
2122     QTAILQ_REMOVE(&memory_listeners, listener, link);
2123 }
2124
2125 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
2126 {
2127     memory_region_ref(root);
2128     memory_region_transaction_begin();
2129     as->ref_count = 1;
2130     as->root = root;
2131     as->malloced = false;
2132     as->current_map = g_new(FlatView, 1);
2133     flatview_init(as->current_map);
2134     as->ioeventfd_nb = 0;
2135     as->ioeventfds = NULL;
2136     QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
2137     as->name = g_strdup(name ? name : "anonymous");
2138     address_space_init_dispatch(as);
2139     memory_region_update_pending |= root->enabled;
2140     memory_region_transaction_commit();
2141 }
2142
2143 static void do_address_space_destroy(AddressSpace *as)
2144 {
2145     MemoryListener *listener;
2146     bool do_free = as->malloced;
2147
2148     address_space_destroy_dispatch(as);
2149
2150     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2151         assert(listener->address_space_filter != as);
2152     }
2153
2154     flatview_unref(as->current_map);
2155     g_free(as->name);
2156     g_free(as->ioeventfds);
2157     memory_region_unref(as->root);
2158     if (do_free) {
2159         g_free(as);
2160     }
2161 }
2162
2163 AddressSpace *address_space_init_shareable(MemoryRegion *root, const char *name)
2164 {
2165     AddressSpace *as;
2166
2167     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2168         if (root == as->root && as->malloced) {
2169             as->ref_count++;
2170             return as;
2171         }
2172     }
2173
2174     as = g_malloc0(sizeof *as);
2175     address_space_init(as, root, name);
2176     as->malloced = true;
2177     return as;
2178 }
2179
2180 void address_space_destroy(AddressSpace *as)
2181 {
2182     MemoryRegion *root = as->root;
2183
2184     as->ref_count--;
2185     if (as->ref_count) {
2186         return;
2187     }
2188     /* Flush out anything from MemoryListeners listening in on this */
2189     memory_region_transaction_begin();
2190     as->root = NULL;
2191     memory_region_transaction_commit();
2192     QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
2193     address_space_unregister(as);
2194
2195     /* At this point, as->dispatch and as->current_map are dummy
2196      * entries that the guest should never use.  Wait for the old
2197      * values to expire before freeing the data.
2198      */
2199     as->root = root;
2200     call_rcu(as, do_address_space_destroy, rcu);
2201 }
2202
2203 typedef struct MemoryRegionList MemoryRegionList;
2204
2205 struct MemoryRegionList {
2206     const MemoryRegion *mr;
2207     QTAILQ_ENTRY(MemoryRegionList) queue;
2208 };
2209
2210 typedef QTAILQ_HEAD(queue, MemoryRegionList) MemoryRegionListHead;
2211
2212 static void mtree_print_mr(fprintf_function mon_printf, void *f,
2213                            const MemoryRegion *mr, unsigned int level,
2214                            hwaddr base,
2215                            MemoryRegionListHead *alias_print_queue)
2216 {
2217     MemoryRegionList *new_ml, *ml, *next_ml;
2218     MemoryRegionListHead submr_print_queue;
2219     const MemoryRegion *submr;
2220     unsigned int i;
2221
2222     if (!mr) {
2223         return;
2224     }
2225
2226     for (i = 0; i < level; i++) {
2227         mon_printf(f, "  ");
2228     }
2229
2230     if (mr->alias) {
2231         MemoryRegionList *ml;
2232         bool found = false;
2233
2234         /* check if the alias is already in the queue */
2235         QTAILQ_FOREACH(ml, alias_print_queue, queue) {
2236             if (ml->mr == mr->alias) {
2237                 found = true;
2238             }
2239         }
2240
2241         if (!found) {
2242             ml = g_new(MemoryRegionList, 1);
2243             ml->mr = mr->alias;
2244             QTAILQ_INSERT_TAIL(alias_print_queue, ml, queue);
2245         }
2246         mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx
2247                    " (prio %d, %c%c): alias %s @%s " TARGET_FMT_plx
2248                    "-" TARGET_FMT_plx "%s\n",
2249                    base + mr->addr,
2250                    base + mr->addr
2251                    + (int128_nz(mr->size) ?
2252                       (hwaddr)int128_get64(int128_sub(mr->size,
2253                                                       int128_one())) : 0),
2254                    mr->priority,
2255                    mr->romd_mode ? 'R' : '-',
2256                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2257                                                                        : '-',
2258                    memory_region_name(mr),
2259                    memory_region_name(mr->alias),
2260                    mr->alias_offset,
2261                    mr->alias_offset
2262                    + (int128_nz(mr->size) ?
2263                       (hwaddr)int128_get64(int128_sub(mr->size,
2264                                                       int128_one())) : 0),
2265                    mr->enabled ? "" : " [disabled]");
2266     } else {
2267         mon_printf(f,
2268                    TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d, %c%c): %s%s\n",
2269                    base + mr->addr,
2270                    base + mr->addr
2271                    + (int128_nz(mr->size) ?
2272                       (hwaddr)int128_get64(int128_sub(mr->size,
2273                                                       int128_one())) : 0),
2274                    mr->priority,
2275                    mr->romd_mode ? 'R' : '-',
2276                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2277                                                                        : '-',
2278                    memory_region_name(mr),
2279                    mr->enabled ? "" : " [disabled]");
2280     }
2281
2282     QTAILQ_INIT(&submr_print_queue);
2283
2284     QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
2285         new_ml = g_new(MemoryRegionList, 1);
2286         new_ml->mr = submr;
2287         QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2288             if (new_ml->mr->addr < ml->mr->addr ||
2289                 (new_ml->mr->addr == ml->mr->addr &&
2290                  new_ml->mr->priority > ml->mr->priority)) {
2291                 QTAILQ_INSERT_BEFORE(ml, new_ml, queue);
2292                 new_ml = NULL;
2293                 break;
2294             }
2295         }
2296         if (new_ml) {
2297             QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, queue);
2298         }
2299     }
2300
2301     QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2302         mtree_print_mr(mon_printf, f, ml->mr, level + 1, base + mr->addr,
2303                        alias_print_queue);
2304     }
2305
2306     QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, queue, next_ml) {
2307         g_free(ml);
2308     }
2309 }
2310
2311 void mtree_info(fprintf_function mon_printf, void *f)
2312 {
2313     MemoryRegionListHead ml_head;
2314     MemoryRegionList *ml, *ml2;
2315     AddressSpace *as;
2316
2317     QTAILQ_INIT(&ml_head);
2318
2319     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2320         mon_printf(f, "address-space: %s\n", as->name);
2321         mtree_print_mr(mon_printf, f, as->root, 1, 0, &ml_head);
2322         mon_printf(f, "\n");
2323     }
2324
2325     /* print aliased regions */
2326     QTAILQ_FOREACH(ml, &ml_head, queue) {
2327         mon_printf(f, "memory-region: %s\n", memory_region_name(ml->mr));
2328         mtree_print_mr(mon_printf, f, ml->mr, 1, 0, &ml_head);
2329         mon_printf(f, "\n");
2330     }
2331
2332     QTAILQ_FOREACH_SAFE(ml, &ml_head, queue, ml2) {
2333         g_free(ml);
2334     }
2335 }
2336
2337 static const TypeInfo memory_region_info = {
2338     .parent             = TYPE_OBJECT,
2339     .name               = TYPE_MEMORY_REGION,
2340     .instance_size      = sizeof(MemoryRegion),
2341     .instance_init      = memory_region_initfn,
2342     .instance_finalize  = memory_region_finalize,
2343 };
2344
2345 static void memory_register_types(void)
2346 {
2347     type_register_static(&memory_region_info);
2348 }
2349
2350 type_init(memory_register_types)