qemu/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 "qapi/error.h"
  18 #include "exec/memory.h"
  19 #include "exec/address-spaces.h"
  20 #include "exec/ioport.h"
  21 #include "qapi/visitor.h"
  22 #include "qemu/bitops.h"
  23 #include "qemu/error-report.h"
  24 #include "qom/object.h"
  25 #include "trace.h"
  26
  27 #include "exec/memory-internal.h"
  28 #include "exec/ram_addr.h"
  29 #include "sysemu/kvm.h"
  30 #include "sysemu/sysemu.h"
  31
  32 //#define DEBUG_UNASSIGNED
  33
  34 #define RAM_ADDR_INVALID (~(ram_addr_t)0)
  35
  36 static unsigned memory_region_transaction_depth;
  37 static bool memory_region_update_pending;
  38 static bool ioeventfd_update_pending;
  39 static bool global_dirty_log = false;
  40
  41 static QTAILQ_HEAD(memory_listeners, MemoryListener) memory_listeners
  42     = QTAILQ_HEAD_INITIALIZER(memory_listeners);
  43
  44 static QTAILQ_HEAD(, AddressSpace) address_spaces
  45     = QTAILQ_HEAD_INITIALIZER(address_spaces);
  46
  47 typedef struct AddrRange AddrRange;
  48
  49 /*
  50  * Note that signed integers are needed for negative offsetting in aliases
  51  * (large MemoryRegion::alias_offset).
  52  */
  53 struct AddrRange {
  54     Int128 start;
  55     Int128 size;
  56 };
  57
  58 static AddrRange addrrange_make(Int128 start, Int128 size)
  59 {
  60     return (AddrRange) { start, size };
  61 }
  62
  63 static bool addrrange_equal(AddrRange r1, AddrRange r2)
  64 {
  65     return int128_eq(r1.start, r2.start) && int128_eq(r1.size, r2.size);
  66 }
  67
  68 static Int128 addrrange_end(AddrRange r)
  69 {
  70     return int128_add(r.start, r.size);
  71 }
  72
  73 static AddrRange addrrange_shift(AddrRange range, Int128 delta)
  74 {
  75     int128_addto(&range.start, delta);
  76     return range;
  77 }
  78
  79 static bool addrrange_contains(AddrRange range, Int128 addr)
  80 {
  81     return int128_ge(addr, range.start)
  82         && int128_lt(addr, addrrange_end(range));
  83 }
  84
  85 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
  86 {
  87     return addrrange_contains(r1, r2.start)
  88         || addrrange_contains(r2, r1.start);
  89 }
  90
  91 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
  92 {
  93     Int128 start = int128_max(r1.start, r2.start);
  94     Int128 end = int128_min(addrrange_end(r1), addrrange_end(r2));
  95     return addrrange_make(start, int128_sub(end, start));
  96 }
  97
  98 enum ListenerDirection { Forward, Reverse };
  99
 100 static bool memory_listener_match(MemoryListener *listener,
 101                                   MemoryRegionSection *section)
 102 {
 103     return !listener->address_space_filter
 104         || listener->address_space_filter == section->address_space;
 105 }
 106
 107 #define MEMORY_LISTENER_CALL_GLOBAL(_callback, _direction, _args...)    \
 108     do {                                                                \
 109         MemoryListener *_listener;                                      \
 110                                                                         \
 111         switch (_direction) {                                           \
 112         case Forward:                                                   \
 113             QTAILQ_FOREACH(_listener, &memory_listeners, link) {        \
 114                 if (_listener->_callback) {                             \
 115                     _listener->_callback(_listener, ##_args);           \
 116                 }                                                       \
 117             }                                                           \
 118             break;                                                      \
 119         case Reverse:                                                   \
 120             QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners,        \
 121                                    memory_listeners, link) {            \
 122                 if (_listener->_callback) {                             \
 123                     _listener->_callback(_listener, ##_args);           \
 124                 }                                                       \
 125             }                                                           \
 126             break;                                                      \
 127         default:                                                        \
 128             abort();                                                    \
 129         }                                                               \
 130     } while (0)
 131
 132 #define MEMORY_LISTENER_CALL(_callback, _direction, _section, _args...) \
 133     do {                                                                \
 134         MemoryListener *_listener;                                      \
 135                                                                         \
 136         switch (_direction) {                                           \
 137         case Forward:                                                   \
 138             QTAILQ_FOREACH(_listener, &memory_listeners, link) {        \
 139                 if (_listener->_callback                                \
 140                     && memory_listener_match(_listener, _section)) {    \
 141                     _listener->_callback(_listener, _section, ##_args); \
 142                 }                                                       \
 143             }                                                           \
 144             break;                                                      \
 145         case Reverse:                                                   \
 146             QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners,        \
 147                                    memory_listeners, link) {            \
 148                 if (_listener->_callback                                \
 149                     && memory_listener_match(_listener, _section)) {    \
 150                     _listener->_callback(_listener, _section, ##_args); \
 151                 }                                                       \
 152             }                                                           \
 153             break;                                                      \
 154         default:                                                        \
 155             abort();                                                    \
 156         }                                                               \
 157     } while (0)
 158
 159 /* No need to ref/unref .mr, the FlatRange keeps it alive.  */
 160 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...)  \
 161     MEMORY_LISTENER_CALL(callback, dir, (&(MemoryRegionSection) {       \
 162         .mr = (fr)->mr,                                                 \
 163         .address_space = (as),                                          \
 164         .offset_within_region = (fr)->offset_in_region,                 \
 165         .size = (fr)->addr.size,                                        \
 166         .offset_within_address_space = int128_get64((fr)->addr.start),  \
 167         .readonly = (fr)->readonly,                                     \
 168               }), ##_args)
 169
 170 struct CoalescedMemoryRange {
 171     AddrRange addr;
 172     QTAILQ_ENTRY(CoalescedMemoryRange) link;
 173 };
 174
 175 struct MemoryRegionIoeventfd {
 176     AddrRange addr;
 177     bool match_data;
 178     uint64_t data;
 179     EventNotifier *e;
 180 };
 181
 182 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a,
 183                                            MemoryRegionIoeventfd b)
 184 {
 185     if (int128_lt(a.addr.start, b.addr.start)) {
 186         return true;
 187     } else if (int128_gt(a.addr.start, b.addr.start)) {
 188         return false;
 189     } else if (int128_lt(a.addr.size, b.addr.size)) {
 190         return true;
 191     } else if (int128_gt(a.addr.size, b.addr.size)) {
 192         return false;
 193     } else if (a.match_data < b.match_data) {
 194         return true;
 195     } else  if (a.match_data > b.match_data) {
 196         return false;
 197     } else if (a.match_data) {
 198         if (a.data < b.data) {
 199             return true;
 200         } else if (a.data > b.data) {
 201             return false;
 202         }
 203     }
 204     if (a.e < b.e) {
 205         return true;
 206     } else if (a.e > b.e) {
 207         return false;
 208     }
 209     return false;
 210 }
 211
 212 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a,
 213                                           MemoryRegionIoeventfd b)
 214 {
 215     return !memory_region_ioeventfd_before(a, b)
 216         && !memory_region_ioeventfd_before(b, a);
 217 }
 218
 219 typedef struct FlatRange FlatRange;
 220 typedef struct FlatView FlatView;
 221
 222 /* Range of memory in the global map.  Addresses are absolute. */
 223 struct FlatRange {
 224     MemoryRegion *mr;
 225     hwaddr offset_in_region;
 226     AddrRange addr;
 227     uint8_t dirty_log_mask;
 228     bool romd_mode;
 229     bool readonly;
 230 };
 231
 232 /* Flattened global view of current active memory hierarchy.  Kept in sorted
 233  * order.
 234  */
 235 struct FlatView {
 236     struct rcu_head rcu;
 237     unsigned ref;
 238     FlatRange *ranges;
 239     unsigned nr;
 240     unsigned nr_allocated;
 241 };
 242
 243 typedef struct AddressSpaceOps AddressSpaceOps;
 244
 245 #define FOR_EACH_FLAT_RANGE(var, view)          \
 246     for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
 247
 248 static bool flatrange_equal(FlatRange *a, FlatRange *b)
 249 {
 250     return a->mr == b->mr
 251         && addrrange_equal(a->addr, b->addr)
 252         && a->offset_in_region == b->offset_in_region
 253         && a->romd_mode == b->romd_mode
 254         && a->readonly == b->readonly;
 255 }
 256
 257 static void flatview_init(FlatView *view)
 258 {
 259     view->ref = 1;
 260     view->ranges = NULL;
 261     view->nr = 0;
 262     view->nr_allocated = 0;
 263 }
 264
 265 /* Insert a range into a given position.  Caller is responsible for maintaining
 266  * sorting order.
 267  */
 268 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
 269 {
 270     if (view->nr == view->nr_allocated) {
 271         view->nr_allocated = MAX(2 * view->nr, 10);
 272         view->ranges = g_realloc(view->ranges,
 273                                     view->nr_allocated * sizeof(*view->ranges));
 274     }
 275     memmove(view->ranges + pos + 1, view->ranges + pos,
 276             (view->nr - pos) * sizeof(FlatRange));
 277     view->ranges[pos] = *range;
 278     memory_region_ref(range->mr);
 279     ++view->nr;
 280 }
 281
 282 static void flatview_destroy(FlatView *view)
 283 {
 284     int i;
 285
 286     for (i = 0; i < view->nr; i++) {
 287         memory_region_unref(view->ranges[i].mr);
 288     }
 289     g_free(view->ranges);
 290     g_free(view);
 291 }
 292
 293 static void flatview_ref(FlatView *view)
 294 {
 295     atomic_inc(&view->ref);
 296 }
 297
 298 static void flatview_unref(FlatView *view)
 299 {
 300     if (atomic_fetch_dec(&view->ref) == 1) {
 301         flatview_destroy(view);
 302     }
 303 }
 304
 305 static bool can_merge(FlatRange *r1, FlatRange *r2)
 306 {
 307     return int128_eq(addrrange_end(r1->addr), r2->addr.start)
 308         && r1->mr == r2->mr
 309         && int128_eq(int128_add(int128_make64(r1->offset_in_region),
 310                                 r1->addr.size),
 311                      int128_make64(r2->offset_in_region))
 312         && r1->dirty_log_mask == r2->dirty_log_mask
 313         && r1->romd_mode == r2->romd_mode
 314         && r1->readonly == r2->readonly;
 315 }
 316
 317 /* Attempt to simplify a view by merging adjacent ranges */
 318 static void flatview_simplify(FlatView *view)
 319 {
 320     unsigned i, j;
 321
 322     i = 0;
 323     while (i < view->nr) {
 324         j = i + 1;
 325         while (j < view->nr
 326                && can_merge(&view->ranges[j-1], &view->ranges[j])) {
 327             int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
 328             ++j;
 329         }
 330         ++i;
 331         memmove(&view->ranges[i], &view->ranges[j],
 332                 (view->nr - j) * sizeof(view->ranges[j]));
 333         view->nr -= j - i;
 334     }
 335 }
 336
 337 static bool memory_region_big_endian(MemoryRegion *mr)
 338 {
 339 #ifdef TARGET_WORDS_BIGENDIAN
 340     return mr->ops->endianness != DEVICE_LITTLE_ENDIAN;
 341 #else
 342     return mr->ops->endianness == DEVICE_BIG_ENDIAN;
 343 #endif
 344 }
 345
 346 static bool memory_region_wrong_endianness(MemoryRegion *mr)
 347 {
 348 #ifdef TARGET_WORDS_BIGENDIAN
 349     return mr->ops->endianness == DEVICE_LITTLE_ENDIAN;
 350 #else
 351     return mr->ops->endianness == DEVICE_BIG_ENDIAN;
 352 #endif
 353 }
 354
 355 static void adjust_endianness(MemoryRegion *mr, uint64_t *data, unsigned size)
 356 {
 357     if (memory_region_wrong_endianness(mr)) {
 358         switch (size) {
 359         case 1:
 360             break;
 361         case 2:
 362             *data = bswap16(*data);
 363             break;
 364         case 4:
 365             *data = bswap32(*data);
 366             break;
 367         case 8:
 368             *data = bswap64(*data);
 369             break;
 370         default:
 371             abort();
 372         }
 373     }
 374 }
 375
 376 static hwaddr memory_region_to_absolute_addr(MemoryRegion *mr, hwaddr offset)
 377 {
 378     MemoryRegion *root;
 379     hwaddr abs_addr = offset;
 380
 381     abs_addr += mr->addr;
 382     for (root = mr; root->container; ) {
 383         root = root->container;
 384         abs_addr += root->addr;
 385     }
 386
 387     return abs_addr;
 388 }
 389
 390 static int get_cpu_index(void)
 391 {
 392     if (current_cpu) {
 393         return current_cpu->cpu_index;
 394     }
 395     return -1;
 396 }
 397
 398 static MemTxResult memory_region_oldmmio_read_accessor(MemoryRegion *mr,
 399                                                        hwaddr addr,
 400                                                        uint64_t *value,
 401                                                        unsigned size,
 402                                                        unsigned shift,
 403                                                        uint64_t mask,
 404                                                        MemTxAttrs attrs)
 405 {
 406     uint64_t tmp;
 407
 408     tmp = mr->ops->old_mmio.read[ctz32(size)](mr->opaque, addr);
 409     if (mr->subpage) {
 410         trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
 411     } else if (mr == &io_mem_notdirty) {
 412         /* Accesses to code which has previously been translated into a TB show
 413          * up in the MMIO path, as accesses to the io_mem_notdirty
 414          * MemoryRegion. */
 415         trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
 416     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 417         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 418         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
 419     }
 420     *value |= (tmp & mask) << shift;
 421     return MEMTX_OK;
 422 }
 423
 424 static MemTxResult  memory_region_read_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 = mr->ops->read(mr->opaque, addr, size);
 435     if (mr->subpage) {
 436         trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
 437     } else if (mr == &io_mem_notdirty) {
 438         /* Accesses to code which has previously been translated into a TB show
 439          * up in the MMIO path, as accesses to the io_mem_notdirty
 440          * MemoryRegion. */
 441         trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
 442     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 443         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 444         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
 445     }
 446     *value |= (tmp & mask) << shift;
 447     return MEMTX_OK;
 448 }
 449
 450 static MemTxResult memory_region_read_with_attrs_accessor(MemoryRegion *mr,
 451                                                           hwaddr addr,
 452                                                           uint64_t *value,
 453                                                           unsigned size,
 454                                                           unsigned shift,
 455                                                           uint64_t mask,
 456                                                           MemTxAttrs attrs)
 457 {
 458     uint64_t tmp = 0;
 459     MemTxResult r;
 460
 461     r = mr->ops->read_with_attrs(mr->opaque, addr, &tmp, size, attrs);
 462     if (mr->subpage) {
 463         trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
 464     } else if (mr == &io_mem_notdirty) {
 465         /* Accesses to code which has previously been translated into a TB show
 466          * up in the MMIO path, as accesses to the io_mem_notdirty
 467          * MemoryRegion. */
 468         trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
 469     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 470         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 471         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
 472     }
 473     *value |= (tmp & mask) << shift;
 474     return r;
 475 }
 476
 477 static MemTxResult memory_region_oldmmio_write_accessor(MemoryRegion *mr,
 478                                                         hwaddr addr,
 479                                                         uint64_t *value,
 480                                                         unsigned size,
 481                                                         unsigned shift,
 482                                                         uint64_t mask,
 483                                                         MemTxAttrs attrs)
 484 {
 485     uint64_t tmp;
 486
 487     tmp = (*value >> shift) & mask;
 488     if (mr->subpage) {
 489         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
 490     } else if (mr == &io_mem_notdirty) {
 491         /* Accesses to code which has previously been translated into a TB show
 492          * up in the MMIO path, as accesses to the io_mem_notdirty
 493          * MemoryRegion. */
 494         trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
 495     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 496         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 497         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
 498     }
 499     mr->ops->old_mmio.write[ctz32(size)](mr->opaque, addr, tmp);
 500     return MEMTX_OK;
 501 }
 502
 503 static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
 504                                                 hwaddr addr,
 505                                                 uint64_t *value,
 506                                                 unsigned size,
 507                                                 unsigned shift,
 508                                                 uint64_t mask,
 509                                                 MemTxAttrs attrs)
 510 {
 511     uint64_t tmp;
 512
 513     tmp = (*value >> shift) & mask;
 514     if (mr->subpage) {
 515         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
 516     } else if (mr == &io_mem_notdirty) {
 517         /* Accesses to code which has previously been translated into a TB show
 518          * up in the MMIO path, as accesses to the io_mem_notdirty
 519          * MemoryRegion. */
 520         trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
 521     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 522         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 523         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
 524     }
 525     mr->ops->write(mr->opaque, addr, tmp, size);
 526     return MEMTX_OK;
 527 }
 528
 529 static MemTxResult memory_region_write_with_attrs_accessor(MemoryRegion *mr,
 530                                                            hwaddr addr,
 531                                                            uint64_t *value,
 532                                                            unsigned size,
 533                                                            unsigned shift,
 534                                                            uint64_t mask,
 535                                                            MemTxAttrs attrs)
 536 {
 537     uint64_t tmp;
 538
 539     tmp = (*value >> shift) & mask;
 540     if (mr->subpage) {
 541         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
 542     } else if (mr == &io_mem_notdirty) {
 543         /* Accesses to code which has previously been translated into a TB show
 544          * up in the MMIO path, as accesses to the io_mem_notdirty
 545          * MemoryRegion. */
 546         trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
 547     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 548         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 549         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
 550     }
 551     return mr->ops->write_with_attrs(mr->opaque, addr, tmp, size, attrs);
 552 }
 553
 554 static MemTxResult access_with_adjusted_size(hwaddr addr,
 555                                       uint64_t *value,
 556                                       unsigned size,
 557                                       unsigned access_size_min,
 558                                       unsigned access_size_max,
 559                                       MemTxResult (*access)(MemoryRegion *mr,
 560                                                             hwaddr addr,
 561                                                             uint64_t *value,
 562                                                             unsigned size,
 563                                                             unsigned shift,
 564                                                             uint64_t mask,
 565                                                             MemTxAttrs attrs),
 566                                       MemoryRegion *mr,
 567                                       MemTxAttrs attrs)
 568 {
 569     uint64_t access_mask;
 570     unsigned access_size;
 571     unsigned i;
 572     MemTxResult r = MEMTX_OK;
 573
 574     if (!access_size_min) {
 575         access_size_min = 1;
 576     }
 577     if (!access_size_max) {
 578         access_size_max = 4;
 579     }
 580
 581     /* FIXME: support unaligned access? */
 582     access_size = MAX(MIN(size, access_size_max), access_size_min);
 583     access_mask = -1ULL >> (64 - access_size * 8);
 584     if (memory_region_big_endian(mr)) {
 585         for (i = 0; i < size; i += access_size) {
 586             r |= access(mr, addr + i, value, access_size,
 587                         (size - access_size - i) * 8, access_mask, attrs);
 588         }
 589     } else {
 590         for (i = 0; i < size; i += access_size) {
 591             r |= access(mr, addr + i, value, access_size, i * 8,
 592                         access_mask, attrs);
 593         }
 594     }
 595     return r;
 596 }
 597
 598 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
 599 {
 600     AddressSpace *as;
 601
 602     while (mr->container) {
 603         mr = mr->container;
 604     }
 605     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 606         if (mr == as->root) {
 607             return as;
 608         }
 609     }
 610     return NULL;
 611 }
 612
 613 /* Render a memory region into the global view.  Ranges in @view obscure
 614  * ranges in @mr.
 615  */
 616 static void render_memory_region(FlatView *view,
 617                                  MemoryRegion *mr,
 618                                  Int128 base,
 619                                  AddrRange clip,
 620                                  bool readonly)
 621 {
 622     MemoryRegion *subregion;
 623     unsigned i;
 624     hwaddr offset_in_region;
 625     Int128 remain;
 626     Int128 now;
 627     FlatRange fr;
 628     AddrRange tmp;
 629
 630     if (!mr->enabled) {
 631         return;
 632     }
 633
 634     int128_addto(&base, int128_make64(mr->addr));
 635     readonly |= mr->readonly;
 636
 637     tmp = addrrange_make(base, mr->size);
 638
 639     if (!addrrange_intersects(tmp, clip)) {
 640         return;
 641     }
 642
 643     clip = addrrange_intersection(tmp, clip);
 644
 645     if (mr->alias) {
 646         int128_subfrom(&base, int128_make64(mr->alias->addr));
 647         int128_subfrom(&base, int128_make64(mr->alias_offset));
 648         render_memory_region(view, mr->alias, base, clip, readonly);
 649         return;
 650     }
 651
 652     /* Render subregions in priority order. */
 653     QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
 654         render_memory_region(view, subregion, base, clip, readonly);
 655     }
 656
 657     if (!mr->terminates) {
 658         return;
 659     }
 660
 661     offset_in_region = int128_get64(int128_sub(clip.start, base));
 662     base = clip.start;
 663     remain = clip.size;
 664
 665     fr.mr = mr;
 666     fr.dirty_log_mask = memory_region_get_dirty_log_mask(mr);
 667     fr.romd_mode = mr->romd_mode;
 668     fr.readonly = readonly;
 669
 670     /* Render the region itself into any gaps left by the current view. */
 671     for (i = 0; i < view->nr && int128_nz(remain); ++i) {
 672         if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
 673             continue;
 674         }
 675         if (int128_lt(base, view->ranges[i].addr.start)) {
 676             now = int128_min(remain,
 677                              int128_sub(view->ranges[i].addr.start, base));
 678             fr.offset_in_region = offset_in_region;
 679             fr.addr = addrrange_make(base, now);
 680             flatview_insert(view, i, &fr);
 681             ++i;
 682             int128_addto(&base, now);
 683             offset_in_region += int128_get64(now);
 684             int128_subfrom(&remain, now);
 685         }
 686         now = int128_sub(int128_min(int128_add(base, remain),
 687                                     addrrange_end(view->ranges[i].addr)),
 688                          base);
 689         int128_addto(&base, now);
 690         offset_in_region += int128_get64(now);
 691         int128_subfrom(&remain, now);
 692     }
 693     if (int128_nz(remain)) {
 694         fr.offset_in_region = offset_in_region;
 695         fr.addr = addrrange_make(base, remain);
 696         flatview_insert(view, i, &fr);
 697     }
 698 }
 699
 700 /* Render a memory topology into a list of disjoint absolute ranges. */
 701 static FlatView *generate_memory_topology(MemoryRegion *mr)
 702 {
 703     FlatView *view;
 704
 705     view = g_new(FlatView, 1);
 706     flatview_init(view);
 707
 708     if (mr) {
 709         render_memory_region(view, mr, int128_zero(),
 710                              addrrange_make(int128_zero(), int128_2_64()), false);
 711     }
 712     flatview_simplify(view);
 713
 714     return view;
 715 }
 716
 717 static void address_space_add_del_ioeventfds(AddressSpace *as,
 718                                              MemoryRegionIoeventfd *fds_new,
 719                                              unsigned fds_new_nb,
 720                                              MemoryRegionIoeventfd *fds_old,
 721                                              unsigned fds_old_nb)
 722 {
 723     unsigned iold, inew;
 724     MemoryRegionIoeventfd *fd;
 725     MemoryRegionSection section;
 726
 727     /* Generate a symmetric difference of the old and new fd sets, adding
 728      * and deleting as necessary.
 729      */
 730
 731     iold = inew = 0;
 732     while (iold < fds_old_nb || inew < fds_new_nb) {
 733         if (iold < fds_old_nb
 734             && (inew == fds_new_nb
 735                 || memory_region_ioeventfd_before(fds_old[iold],
 736                                                   fds_new[inew]))) {
 737             fd = &fds_old[iold];
 738             section = (MemoryRegionSection) {
 739                 .address_space = as,
 740                 .offset_within_address_space = int128_get64(fd->addr.start),
 741                 .size = fd->addr.size,
 742             };
 743             MEMORY_LISTENER_CALL(eventfd_del, Forward, &section,
 744                                  fd->match_data, fd->data, fd->e);
 745             ++iold;
 746         } else if (inew < fds_new_nb
 747                    && (iold == fds_old_nb
 748                        || memory_region_ioeventfd_before(fds_new[inew],
 749                                                          fds_old[iold]))) {
 750             fd = &fds_new[inew];
 751             section = (MemoryRegionSection) {
 752                 .address_space = as,
 753                 .offset_within_address_space = int128_get64(fd->addr.start),
 754                 .size = fd->addr.size,
 755             };
 756             MEMORY_LISTENER_CALL(eventfd_add, Reverse, &section,
 757                                  fd->match_data, fd->data, fd->e);
 758             ++inew;
 759         } else {
 760             ++iold;
 761             ++inew;
 762         }
 763     }
 764 }
 765
 766 static FlatView *address_space_get_flatview(AddressSpace *as)
 767 {
 768     FlatView *view;
 769
 770     rcu_read_lock();
 771     view = atomic_rcu_read(&as->current_map);
 772     flatview_ref(view);
 773     rcu_read_unlock();
 774     return view;
 775 }
 776
 777 static void address_space_update_ioeventfds(AddressSpace *as)
 778 {
 779     FlatView *view;
 780     FlatRange *fr;
 781     unsigned ioeventfd_nb = 0;
 782     MemoryRegionIoeventfd *ioeventfds = NULL;
 783     AddrRange tmp;
 784     unsigned i;
 785
 786     view = address_space_get_flatview(as);
 787     FOR_EACH_FLAT_RANGE(fr, view) {
 788         for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
 789             tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
 790                                   int128_sub(fr->addr.start,
 791                                              int128_make64(fr->offset_in_region)));
 792             if (addrrange_intersects(fr->addr, tmp)) {
 793                 ++ioeventfd_nb;
 794                 ioeventfds = g_realloc(ioeventfds,
 795                                           ioeventfd_nb * sizeof(*ioeventfds));
 796                 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
 797                 ioeventfds[ioeventfd_nb-1].addr = tmp;
 798             }
 799         }
 800     }
 801
 802     address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
 803                                      as->ioeventfds, as->ioeventfd_nb);
 804
 805     g_free(as->ioeventfds);
 806     as->ioeventfds = ioeventfds;
 807     as->ioeventfd_nb = ioeventfd_nb;
 808     flatview_unref(view);
 809 }
 810
 811 static void address_space_update_topology_pass(AddressSpace *as,
 812                                                const FlatView *old_view,
 813                                                const FlatView *new_view,
 814                                                bool adding)
 815 {
 816     unsigned iold, inew;
 817     FlatRange *frold, *frnew;
 818
 819     /* Generate a symmetric difference of the old and new memory maps.
 820      * Kill ranges in the old map, and instantiate ranges in the new map.
 821      */
 822     iold = inew = 0;
 823     while (iold < old_view->nr || inew < new_view->nr) {
 824         if (iold < old_view->nr) {
 825             frold = &old_view->ranges[iold];
 826         } else {
 827             frold = NULL;
 828         }
 829         if (inew < new_view->nr) {
 830             frnew = &new_view->ranges[inew];
 831         } else {
 832             frnew = NULL;
 833         }
 834
 835         if (frold
 836             && (!frnew
 837                 || int128_lt(frold->addr.start, frnew->addr.start)
 838                 || (int128_eq(frold->addr.start, frnew->addr.start)
 839                     && !flatrange_equal(frold, frnew)))) {
 840             /* In old but not in new, or in both but attributes changed. */
 841
 842             if (!adding) {
 843                 MEMORY_LISTENER_UPDATE_REGION(frold, as, Reverse, region_del);
 844             }
 845
 846             ++iold;
 847         } else if (frold && frnew && flatrange_equal(frold, frnew)) {
 848             /* In both and unchanged (except logging may have changed) */
 849
 850             if (adding) {
 851                 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_nop);
 852                 if (frnew->dirty_log_mask & ~frold->dirty_log_mask) {
 853                     MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, log_start,
 854                                                   frold->dirty_log_mask,
 855                                                   frnew->dirty_log_mask);
 856                 }
 857                 if (frold->dirty_log_mask & ~frnew->dirty_log_mask) {
 858                     MEMORY_LISTENER_UPDATE_REGION(frnew, as, Reverse, log_stop,
 859                                                   frold->dirty_log_mask,
 860                                                   frnew->dirty_log_mask);
 861                 }
 862             }
 863
 864             ++iold;
 865             ++inew;
 866         } else {
 867             /* In new */
 868
 869             if (adding) {
 870                 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_add);
 871             }
 872
 873             ++inew;
 874         }
 875     }
 876 }
 877
 878
 879 static void address_space_update_topology(AddressSpace *as)
 880 {
 881     FlatView *old_view = address_space_get_flatview(as);
 882     FlatView *new_view = generate_memory_topology(as->root);
 883
 884     address_space_update_topology_pass(as, old_view, new_view, false);
 885     address_space_update_topology_pass(as, old_view, new_view, true);
 886
 887     /* Writes are protected by the BQL.  */
 888     atomic_rcu_set(&as->current_map, new_view);
 889     call_rcu(old_view, flatview_unref, rcu);
 890
 891     /* Note that all the old MemoryRegions are still alive up to this
 892      * point.  This relieves most MemoryListeners from the need to
 893      * ref/unref the MemoryRegions they get---unless they use them
 894      * outside the iothread mutex, in which case precise reference
 895      * counting is necessary.
 896      */
 897     flatview_unref(old_view);
 898
 899     address_space_update_ioeventfds(as);
 900 }
 901
 902 void memory_region_transaction_begin(void)
 903 {
 904     qemu_flush_coalesced_mmio_buffer();
 905     ++memory_region_transaction_depth;
 906 }
 907
 908 static void memory_region_clear_pending(void)
 909 {
 910     memory_region_update_pending = false;
 911     ioeventfd_update_pending = false;
 912 }
 913
 914 void memory_region_transaction_commit(void)
 915 {
 916     AddressSpace *as;
 917
 918     assert(memory_region_transaction_depth);
 919     --memory_region_transaction_depth;
 920     if (!memory_region_transaction_depth) {
 921         if (memory_region_update_pending) {
 922             MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
 923
 924             QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 925                 address_space_update_topology(as);
 926             }
 927
 928             MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
 929         } else if (ioeventfd_update_pending) {
 930             QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 931                 address_space_update_ioeventfds(as);
 932             }
 933         }
 934         memory_region_clear_pending();
 935    }
 936 }
 937
 938 static void memory_region_destructor_none(MemoryRegion *mr)
 939 {
 940 }
 941
 942 static void memory_region_destructor_ram(MemoryRegion *mr)
 943 {
 944     qemu_ram_free(mr->ram_block);
 945 }
 946
 947 static void memory_region_destructor_rom_device(MemoryRegion *mr)
 948 {
 949     qemu_ram_free(mr->ram_block);
 950 }
 951
 952 static bool memory_region_need_escape(char c)
 953 {
 954     return c == '/' || c == '[' || c == '\\' || c == ']';
 955 }
 956
 957 static char *memory_region_escape_name(const char *name)
 958 {
 959     const char *p;
 960     char *escaped, *q;
 961     uint8_t c;
 962     size_t bytes = 0;
 963
 964     for (p = name; *p; p++) {
 965         bytes += memory_region_need_escape(*p) ? 4 : 1;
 966     }
 967     if (bytes == p - name) {
 968        return g_memdup(name, bytes + 1);
 969     }
 970
 971     escaped = g_malloc(bytes + 1);
 972     for (p = name, q = escaped; *p; p++) {
 973         c = *p;
 974         if (unlikely(memory_region_need_escape(c))) {
 975             *q++ = '\\';
 976             *q++ = 'x';
 977             *q++ = "0123456789abcdef"[c >> 4];
 978             c = "0123456789abcdef"[c & 15];
 979         }
 980         *q++ = c;
 981     }
 982     *q = 0;
 983     return escaped;
 984 }
 985
 986 void memory_region_init(MemoryRegion *mr,
 987                         Object *owner,
 988                         const char *name,
 989                         uint64_t size)
 990 {
 991     object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
 992     mr->size = int128_make64(size);
 993     if (size == UINT64_MAX) {
 994         mr->size = int128_2_64();
 995     }
 996     mr->name = g_strdup(name);
 997     mr->owner = owner;
 998     mr->ram_block = NULL;
 999
1000     if (name) {
1001         char *escaped_name = memory_region_escape_name(name);
1002         char *name_array = g_strdup_printf("%s[*]", escaped_name);
1003
1004         if (!owner) {
1005             owner = container_get(qdev_get_machine(), "/unattached");
1006         }
1007
1008         object_property_add_child(owner, name_array, OBJECT(mr), &error_abort);
1009         object_unref(OBJECT(mr));
1010         g_free(name_array);
1011         g_free(escaped_name);
1012     }
1013 }
1014
1015 static void memory_region_get_addr(Object *obj, Visitor *v, const char *name,
1016                                    void *opaque, Error **errp)
1017 {
1018     MemoryRegion *mr = MEMORY_REGION(obj);
1019     uint64_t value = mr->addr;
1020
1021     visit_type_uint64(v, name, &value, errp);
1022 }
1023
1024 static void memory_region_get_container(Object *obj, Visitor *v,
1025                                         const char *name, void *opaque,
1026                                         Error **errp)
1027 {
1028     MemoryRegion *mr = MEMORY_REGION(obj);
1029     gchar *path = (gchar *)"";
1030
1031     if (mr->container) {
1032         path = object_get_canonical_path(OBJECT(mr->container));
1033     }
1034     visit_type_str(v, name, &path, errp);
1035     if (mr->container) {
1036         g_free(path);
1037     }
1038 }
1039
1040 static Object *memory_region_resolve_container(Object *obj, void *opaque,
1041                                                const char *part)
1042 {
1043     MemoryRegion *mr = MEMORY_REGION(obj);
1044
1045     return OBJECT(mr->container);
1046 }
1047
1048 static void memory_region_get_priority(Object *obj, Visitor *v,
1049                                        const char *name, void *opaque,
1050                                        Error **errp)
1051 {
1052     MemoryRegion *mr = MEMORY_REGION(obj);
1053     int32_t value = mr->priority;
1054
1055     visit_type_int32(v, name, &value, errp);
1056 }
1057
1058 static bool memory_region_get_may_overlap(Object *obj, Error **errp)
1059 {
1060     MemoryRegion *mr = MEMORY_REGION(obj);
1061
1062     return mr->may_overlap;
1063 }
1064
1065 static void memory_region_get_size(Object *obj, Visitor *v, const char *name,
1066                                    void *opaque, Error **errp)
1067 {
1068     MemoryRegion *mr = MEMORY_REGION(obj);
1069     uint64_t value = memory_region_size(mr);
1070
1071     visit_type_uint64(v, name, &value, errp);
1072 }
1073
1074 static void memory_region_initfn(Object *obj)
1075 {
1076     MemoryRegion *mr = MEMORY_REGION(obj);
1077     ObjectProperty *op;
1078
1079     mr->ops = &unassigned_mem_ops;
1080     mr->enabled = true;
1081     mr->romd_mode = true;
1082     mr->global_locking = true;
1083     mr->destructor = memory_region_destructor_none;
1084     QTAILQ_INIT(&mr->subregions);
1085     QTAILQ_INIT(&mr->coalesced);
1086
1087     op = object_property_add(OBJECT(mr), "container",
1088                              "link<" TYPE_MEMORY_REGION ">",
1089                              memory_region_get_container,
1090                              NULL, /* memory_region_set_container */
1091                              NULL, NULL, &error_abort);
1092     op->resolve = memory_region_resolve_container;
1093
1094     object_property_add(OBJECT(mr), "addr", "uint64",
1095                         memory_region_get_addr,
1096                         NULL, /* memory_region_set_addr */
1097                         NULL, NULL, &error_abort);
1098     object_property_add(OBJECT(mr), "priority", "uint32",
1099                         memory_region_get_priority,
1100                         NULL, /* memory_region_set_priority */
1101                         NULL, NULL, &error_abort);
1102     object_property_add_bool(OBJECT(mr), "may-overlap",
1103                              memory_region_get_may_overlap,
1104                              NULL, /* memory_region_set_may_overlap */
1105                              &error_abort);
1106     object_property_add(OBJECT(mr), "size", "uint64",
1107                         memory_region_get_size,
1108                         NULL, /* memory_region_set_size, */
1109                         NULL, NULL, &error_abort);
1110 }
1111
1112 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1113                                     unsigned size)
1114 {
1115 #ifdef DEBUG_UNASSIGNED
1116     printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
1117 #endif
1118     if (current_cpu != NULL) {
1119         cpu_unassigned_access(current_cpu, addr, false, false, 0, size);
1120     }
1121     return 0;
1122 }
1123
1124 static void unassigned_mem_write(void *opaque, hwaddr addr,
1125                                  uint64_t val, unsigned size)
1126 {
1127 #ifdef DEBUG_UNASSIGNED
1128     printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1129 #endif
1130     if (current_cpu != NULL) {
1131         cpu_unassigned_access(current_cpu, addr, true, false, 0, size);
1132     }
1133 }
1134
1135 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1136                                    unsigned size, bool is_write)
1137 {
1138     return false;
1139 }
1140
1141 const MemoryRegionOps unassigned_mem_ops = {
1142     .valid.accepts = unassigned_mem_accepts,
1143     .endianness = DEVICE_NATIVE_ENDIAN,
1144 };
1145
1146 bool memory_region_access_valid(MemoryRegion *mr,
1147                                 hwaddr addr,
1148                                 unsigned size,
1149                                 bool is_write)
1150 {
1151     int access_size_min, access_size_max;
1152     int access_size, i;
1153
1154     if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1155         return false;
1156     }
1157
1158     if (!mr->ops->valid.accepts) {
1159         return true;
1160     }
1161
1162     access_size_min = mr->ops->valid.min_access_size;
1163     if (!mr->ops->valid.min_access_size) {
1164         access_size_min = 1;
1165     }
1166
1167     access_size_max = mr->ops->valid.max_access_size;
1168     if (!mr->ops->valid.max_access_size) {
1169         access_size_max = 4;
1170     }
1171
1172     access_size = MAX(MIN(size, access_size_max), access_size_min);
1173     for (i = 0; i < size; i += access_size) {
1174         if (!mr->ops->valid.accepts(mr->opaque, addr + i, access_size,
1175                                     is_write)) {
1176             return false;
1177         }
1178     }
1179
1180     return true;
1181 }
1182
1183 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1184                                                 hwaddr addr,
1185                                                 uint64_t *pval,
1186                                                 unsigned size,
1187                                                 MemTxAttrs attrs)
1188 {
1189     *pval = 0;
1190
1191     if (mr->ops->read) {
1192         return access_with_adjusted_size(addr, pval, size,
1193                                          mr->ops->impl.min_access_size,
1194                                          mr->ops->impl.max_access_size,
1195                                          memory_region_read_accessor,
1196                                          mr, attrs);
1197     } else if (mr->ops->read_with_attrs) {
1198         return access_with_adjusted_size(addr, pval, size,
1199                                          mr->ops->impl.min_access_size,
1200                                          mr->ops->impl.max_access_size,
1201                                          memory_region_read_with_attrs_accessor,
1202                                          mr, attrs);
1203     } else {
1204         return access_with_adjusted_size(addr, pval, size, 1, 4,
1205                                          memory_region_oldmmio_read_accessor,
1206                                          mr, attrs);
1207     }
1208 }
1209
1210 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1211                                         hwaddr addr,
1212                                         uint64_t *pval,
1213                                         unsigned size,
1214                                         MemTxAttrs attrs)
1215 {
1216     MemTxResult r;
1217
1218     if (!memory_region_access_valid(mr, addr, size, false)) {
1219         *pval = unassigned_mem_read(mr, addr, size);
1220         return MEMTX_DECODE_ERROR;
1221     }
1222
1223     r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1224     adjust_endianness(mr, pval, size);
1225     return r;
1226 }
1227
1228 /* Return true if an eventfd was signalled */
1229 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1230                                                     hwaddr addr,
1231                                                     uint64_t data,
1232                                                     unsigned size,
1233                                                     MemTxAttrs attrs)
1234 {
1235     MemoryRegionIoeventfd ioeventfd = {
1236         .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1237         .data = data,
1238     };
1239     unsigned i;
1240
1241     for (i = 0; i < mr->ioeventfd_nb; i++) {
1242         ioeventfd.match_data = mr->ioeventfds[i].match_data;
1243         ioeventfd.e = mr->ioeventfds[i].e;
1244
1245         if (memory_region_ioeventfd_equal(ioeventfd, mr->ioeventfds[i])) {
1246             event_notifier_set(ioeventfd.e);
1247             return true;
1248         }
1249     }
1250
1251     return false;
1252 }
1253
1254 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1255                                          hwaddr addr,
1256                                          uint64_t data,
1257                                          unsigned size,
1258                                          MemTxAttrs attrs)
1259 {
1260     if (!memory_region_access_valid(mr, addr, size, true)) {
1261         unassigned_mem_write(mr, addr, data, size);
1262         return MEMTX_DECODE_ERROR;
1263     }
1264
1265     adjust_endianness(mr, &data, size);
1266
1267     if ((!kvm_eventfds_enabled()) &&
1268         memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1269         return MEMTX_OK;
1270     }
1271
1272     if (mr->ops->write) {
1273         return access_with_adjusted_size(addr, &data, size,
1274                                          mr->ops->impl.min_access_size,
1275                                          mr->ops->impl.max_access_size,
1276                                          memory_region_write_accessor, mr,
1277                                          attrs);
1278     } else if (mr->ops->write_with_attrs) {
1279         return
1280             access_with_adjusted_size(addr, &data, size,
1281                                       mr->ops->impl.min_access_size,
1282                                       mr->ops->impl.max_access_size,
1283                                       memory_region_write_with_attrs_accessor,
1284                                       mr, attrs);
1285     } else {
1286         return access_with_adjusted_size(addr, &data, size, 1, 4,
1287                                          memory_region_oldmmio_write_accessor,
1288                                          mr, attrs);
1289     }
1290 }
1291
1292 void memory_region_init_io(MemoryRegion *mr,
1293                            Object *owner,
1294                            const MemoryRegionOps *ops,
1295                            void *opaque,
1296                            const char *name,
1297                            uint64_t size)
1298 {
1299     memory_region_init(mr, owner, name, size);
1300     mr->ops = ops ? ops : &unassigned_mem_ops;
1301     mr->opaque = opaque;
1302     mr->terminates = true;
1303 }
1304
1305 void memory_region_init_ram(MemoryRegion *mr,
1306                             Object *owner,
1307                             const char *name,
1308                             uint64_t size,
1309                             Error **errp)
1310 {
1311     memory_region_init(mr, owner, name, size);
1312     mr->ram = true;
1313     mr->terminates = true;
1314     mr->destructor = memory_region_destructor_ram;
1315     mr->ram_block = qemu_ram_alloc(size, mr, errp);
1316     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1317 }
1318
1319 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1320                                        Object *owner,
1321                                        const char *name,
1322                                        uint64_t size,
1323                                        uint64_t max_size,
1324                                        void (*resized)(const char*,
1325                                                        uint64_t length,
1326                                                        void *host),
1327                                        Error **errp)
1328 {
1329     memory_region_init(mr, owner, name, size);
1330     mr->ram = true;
1331     mr->terminates = true;
1332     mr->destructor = memory_region_destructor_ram;
1333     mr->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1334                                               mr, errp);
1335     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1336 }
1337
1338 #ifdef __linux__
1339 void memory_region_init_ram_from_file(MemoryRegion *mr,
1340                                       struct Object *owner,
1341                                       const char *name,
1342                                       uint64_t size,
1343                                       bool share,
1344                                       const char *path,
1345                                       Error **errp)
1346 {
1347     memory_region_init(mr, owner, name, size);
1348     mr->ram = true;
1349     mr->terminates = true;
1350     mr->destructor = memory_region_destructor_ram;
1351     mr->ram_block = qemu_ram_alloc_from_file(size, mr, share, path, errp);
1352     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1353 }
1354 #endif
1355
1356 void memory_region_init_ram_ptr(MemoryRegion *mr,
1357                                 Object *owner,
1358                                 const char *name,
1359                                 uint64_t size,
1360                                 void *ptr)
1361 {
1362     memory_region_init(mr, owner, name, size);
1363     mr->ram = true;
1364     mr->terminates = true;
1365     mr->destructor = memory_region_destructor_ram;
1366     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1367
1368     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1369     assert(ptr != NULL);
1370     mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1371 }
1372
1373 void memory_region_set_skip_dump(MemoryRegion *mr)
1374 {
1375     mr->skip_dump = true;
1376 }
1377
1378 void memory_region_init_alias(MemoryRegion *mr,
1379                               Object *owner,
1380                               const char *name,
1381                               MemoryRegion *orig,
1382                               hwaddr offset,
1383                               uint64_t size)
1384 {
1385     memory_region_init(mr, owner, name, size);
1386     mr->alias = orig;
1387     mr->alias_offset = offset;
1388 }
1389
1390 void memory_region_init_rom_device(MemoryRegion *mr,
1391                                    Object *owner,
1392                                    const MemoryRegionOps *ops,
1393                                    void *opaque,
1394                                    const char *name,
1395                                    uint64_t size,
1396                                    Error **errp)
1397 {
1398     memory_region_init(mr, owner, name, size);
1399     mr->ops = ops;
1400     mr->opaque = opaque;
1401     mr->terminates = true;
1402     mr->rom_device = true;
1403     mr->destructor = memory_region_destructor_rom_device;
1404     mr->ram_block = qemu_ram_alloc(size, mr, errp);
1405 }
1406
1407 void memory_region_init_iommu(MemoryRegion *mr,
1408                               Object *owner,
1409                               const MemoryRegionIOMMUOps *ops,
1410                               const char *name,
1411                               uint64_t size)
1412 {
1413     memory_region_init(mr, owner, name, size);
1414     mr->iommu_ops = ops,
1415     mr->terminates = true;  /* then re-forwards */
1416     notifier_list_init(&mr->iommu_notify);
1417 }
1418
1419 static void memory_region_finalize(Object *obj)
1420 {
1421     MemoryRegion *mr = MEMORY_REGION(obj);
1422
1423     assert(!mr->container);
1424
1425     /* We know the region is not visible in any address space (it
1426      * does not have a container and cannot be a root either because
1427      * it has no references, so we can blindly clear mr->enabled.
1428      * memory_region_set_enabled instead could trigger a transaction
1429      * and cause an infinite loop.
1430      */
1431     mr->enabled = false;
1432     memory_region_transaction_begin();
1433     while (!QTAILQ_EMPTY(&mr->subregions)) {
1434         MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1435         memory_region_del_subregion(mr, subregion);
1436     }
1437     memory_region_transaction_commit();
1438
1439     mr->destructor(mr);
1440     memory_region_clear_coalescing(mr);
1441     g_free((char *)mr->name);
1442     g_free(mr->ioeventfds);
1443 }
1444
1445 Object *memory_region_owner(MemoryRegion *mr)
1446 {
1447     Object *obj = OBJECT(mr);
1448     return obj->parent;
1449 }
1450
1451 void memory_region_ref(MemoryRegion *mr)
1452 {
1453     /* MMIO callbacks most likely will access data that belongs
1454      * to the owner, hence the need to ref/unref the owner whenever
1455      * the memory region is in use.
1456      *
1457      * The memory region is a child of its owner.  As long as the
1458      * owner doesn't call unparent itself on the memory region,
1459      * ref-ing the owner will also keep the memory region alive.
1460      * Memory regions without an owner are supposed to never go away;
1461      * we do not ref/unref them because it slows down DMA sensibly.
1462      */
1463     if (mr && mr->owner) {
1464         object_ref(mr->owner);
1465     }
1466 }
1467
1468 void memory_region_unref(MemoryRegion *mr)
1469 {
1470     if (mr && mr->owner) {
1471         object_unref(mr->owner);
1472     }
1473 }
1474
1475 uint64_t memory_region_size(MemoryRegion *mr)
1476 {
1477     if (int128_eq(mr->size, int128_2_64())) {
1478         return UINT64_MAX;
1479     }
1480     return int128_get64(mr->size);
1481 }
1482
1483 const char *memory_region_name(const MemoryRegion *mr)
1484 {
1485     if (!mr->name) {
1486         ((MemoryRegion *)mr)->name =
1487             object_get_canonical_path_component(OBJECT(mr));
1488     }
1489     return mr->name;
1490 }
1491
1492 bool memory_region_is_skip_dump(MemoryRegion *mr)
1493 {
1494     return mr->skip_dump;
1495 }
1496
1497 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1498 {
1499     uint8_t mask = mr->dirty_log_mask;
1500     if (global_dirty_log) {
1501         mask |= (1 << DIRTY_MEMORY_MIGRATION);
1502     }
1503     return mask;
1504 }
1505
1506 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1507 {
1508     return memory_region_get_dirty_log_mask(mr) & (1 << client);
1509 }
1510
1511 void memory_region_register_iommu_notifier(MemoryRegion *mr, Notifier *n)
1512 {
1513     notifier_list_add(&mr->iommu_notify, n);
1514 }
1515
1516 void memory_region_iommu_replay(MemoryRegion *mr, Notifier *n,
1517                                 hwaddr granularity, bool is_write)
1518 {
1519     hwaddr addr;
1520     IOMMUTLBEntry iotlb;
1521
1522     for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1523         iotlb = mr->iommu_ops->translate(mr, addr, is_write);
1524         if (iotlb.perm != IOMMU_NONE) {
1525             n->notify(n, &iotlb);
1526         }
1527
1528         /* if (2^64 - MR size) < granularity, it's possible to get an
1529          * infinite loop here.  This should catch such a wraparound */
1530         if ((addr + granularity) < addr) {
1531             break;
1532         }
1533     }
1534 }
1535
1536 void memory_region_unregister_iommu_notifier(Notifier *n)
1537 {
1538     notifier_remove(n);
1539 }
1540
1541 void memory_region_notify_iommu(MemoryRegion *mr,
1542                                 IOMMUTLBEntry entry)
1543 {
1544     assert(memory_region_is_iommu(mr));
1545     notifier_list_notify(&mr->iommu_notify, &entry);
1546 }
1547
1548 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1549 {
1550     uint8_t mask = 1 << client;
1551     uint8_t old_logging;
1552
1553     assert(client == DIRTY_MEMORY_VGA);
1554     old_logging = mr->vga_logging_count;
1555     mr->vga_logging_count += log ? 1 : -1;
1556     if (!!old_logging == !!mr->vga_logging_count) {
1557         return;
1558     }
1559
1560     memory_region_transaction_begin();
1561     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1562     memory_region_update_pending |= mr->enabled;
1563     memory_region_transaction_commit();
1564 }
1565
1566 bool memory_region_get_dirty(MemoryRegion *mr, hwaddr addr,
1567                              hwaddr size, unsigned client)
1568 {
1569     assert(mr->ram_block);
1570     return cpu_physical_memory_get_dirty(memory_region_get_ram_addr(mr) + addr,
1571                                          size, client);
1572 }
1573
1574 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1575                              hwaddr size)
1576 {
1577     assert(mr->ram_block);
1578     cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
1579                                         size,
1580                                         memory_region_get_dirty_log_mask(mr));
1581 }
1582
1583 bool memory_region_test_and_clear_dirty(MemoryRegion *mr, hwaddr addr,
1584                                         hwaddr size, unsigned client)
1585 {
1586     assert(mr->ram_block);
1587     return cpu_physical_memory_test_and_clear_dirty(
1588                 memory_region_get_ram_addr(mr) + addr, size, client);
1589 }
1590
1591
1592 void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1593 {
1594     AddressSpace *as;
1595     FlatRange *fr;
1596
1597     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1598         FlatView *view = address_space_get_flatview(as);
1599         FOR_EACH_FLAT_RANGE(fr, view) {
1600             if (fr->mr == mr) {
1601                 MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
1602             }
1603         }
1604         flatview_unref(view);
1605     }
1606 }
1607
1608 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
1609 {
1610     if (mr->readonly != readonly) {
1611         memory_region_transaction_begin();
1612         mr->readonly = readonly;
1613         memory_region_update_pending |= mr->enabled;
1614         memory_region_transaction_commit();
1615     }
1616 }
1617
1618 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
1619 {
1620     if (mr->romd_mode != romd_mode) {
1621         memory_region_transaction_begin();
1622         mr->romd_mode = romd_mode;
1623         memory_region_update_pending |= mr->enabled;
1624         memory_region_transaction_commit();
1625     }
1626 }
1627
1628 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
1629                                hwaddr size, unsigned client)
1630 {
1631     assert(mr->ram_block);
1632     cpu_physical_memory_test_and_clear_dirty(
1633         memory_region_get_ram_addr(mr) + addr, size, client);
1634 }
1635
1636 int memory_region_get_fd(MemoryRegion *mr)
1637 {
1638     if (mr->alias) {
1639         return memory_region_get_fd(mr->alias);
1640     }
1641
1642     assert(mr->ram_block);
1643
1644     return qemu_get_ram_fd(memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK);
1645 }
1646
1647 void *memory_region_get_ram_ptr(MemoryRegion *mr)
1648 {
1649     void *ptr;
1650     uint64_t offset = 0;
1651
1652     rcu_read_lock();
1653     while (mr->alias) {
1654         offset += mr->alias_offset;
1655         mr = mr->alias;
1656     }
1657     assert(mr->ram_block);
1658     ptr = qemu_get_ram_ptr(mr->ram_block,
1659                            memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK);
1660     rcu_read_unlock();
1661
1662     return ptr + offset;
1663 }
1664
1665 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
1666 {
1667     return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
1668 }
1669
1670 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
1671 {
1672     assert(mr->ram_block);
1673
1674     qemu_ram_resize(memory_region_get_ram_addr(mr), newsize, errp);
1675 }
1676
1677 static void memory_region_update_coalesced_range_as(MemoryRegion *mr, AddressSpace *as)
1678 {
1679     FlatView *view;
1680     FlatRange *fr;
1681     CoalescedMemoryRange *cmr;
1682     AddrRange tmp;
1683     MemoryRegionSection section;
1684
1685     view = address_space_get_flatview(as);
1686     FOR_EACH_FLAT_RANGE(fr, view) {
1687         if (fr->mr == mr) {
1688             section = (MemoryRegionSection) {
1689                 .address_space = as,
1690                 .offset_within_address_space = int128_get64(fr->addr.start),
1691                 .size = fr->addr.size,
1692             };
1693
1694             MEMORY_LISTENER_CALL(coalesced_mmio_del, Reverse, &section,
1695                                  int128_get64(fr->addr.start),
1696                                  int128_get64(fr->addr.size));
1697             QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
1698                 tmp = addrrange_shift(cmr->addr,
1699                                       int128_sub(fr->addr.start,
1700                                                  int128_make64(fr->offset_in_region)));
1701                 if (!addrrange_intersects(tmp, fr->addr)) {
1702                     continue;
1703                 }
1704                 tmp = addrrange_intersection(tmp, fr->addr);
1705                 MEMORY_LISTENER_CALL(coalesced_mmio_add, Forward, &section,
1706                                      int128_get64(tmp.start),
1707                                      int128_get64(tmp.size));
1708             }
1709         }
1710     }
1711     flatview_unref(view);
1712 }
1713
1714 static void memory_region_update_coalesced_range(MemoryRegion *mr)
1715 {
1716     AddressSpace *as;
1717
1718     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1719         memory_region_update_coalesced_range_as(mr, as);
1720     }
1721 }
1722
1723 void memory_region_set_coalescing(MemoryRegion *mr)
1724 {
1725     memory_region_clear_coalescing(mr);
1726     memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
1727 }
1728
1729 void memory_region_add_coalescing(MemoryRegion *mr,
1730                                   hwaddr offset,
1731                                   uint64_t size)
1732 {
1733     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
1734
1735     cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
1736     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
1737     memory_region_update_coalesced_range(mr);
1738     memory_region_set_flush_coalesced(mr);
1739 }
1740
1741 void memory_region_clear_coalescing(MemoryRegion *mr)
1742 {
1743     CoalescedMemoryRange *cmr;
1744     bool updated = false;
1745
1746     qemu_flush_coalesced_mmio_buffer();
1747     mr->flush_coalesced_mmio = false;
1748
1749     while (!QTAILQ_EMPTY(&mr->coalesced)) {
1750         cmr = QTAILQ_FIRST(&mr->coalesced);
1751         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
1752         g_free(cmr);
1753         updated = true;
1754     }
1755
1756     if (updated) {
1757         memory_region_update_coalesced_range(mr);
1758     }
1759 }
1760
1761 void memory_region_set_flush_coalesced(MemoryRegion *mr)
1762 {
1763     mr->flush_coalesced_mmio = true;
1764 }
1765
1766 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
1767 {
1768     qemu_flush_coalesced_mmio_buffer();
1769     if (QTAILQ_EMPTY(&mr->coalesced)) {
1770         mr->flush_coalesced_mmio = false;
1771     }
1772 }
1773
1774 void memory_region_set_global_locking(MemoryRegion *mr)
1775 {
1776     mr->global_locking = true;
1777 }
1778
1779 void memory_region_clear_global_locking(MemoryRegion *mr)
1780 {
1781     mr->global_locking = false;
1782 }
1783
1784 static bool userspace_eventfd_warning;
1785
1786 void memory_region_add_eventfd(MemoryRegion *mr,
1787                                hwaddr addr,
1788                                unsigned size,
1789                                bool match_data,
1790                                uint64_t data,
1791                                EventNotifier *e)
1792 {
1793     MemoryRegionIoeventfd mrfd = {
1794         .addr.start = int128_make64(addr),
1795         .addr.size = int128_make64(size),
1796         .match_data = match_data,
1797         .data = data,
1798         .e = e,
1799     };
1800     unsigned i;
1801
1802     if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
1803                             userspace_eventfd_warning))) {
1804         userspace_eventfd_warning = true;
1805         error_report("Using eventfd without MMIO binding in KVM. "
1806                      "Suboptimal performance expected");
1807     }
1808
1809     if (size) {
1810         adjust_endianness(mr, &mrfd.data, size);
1811     }
1812     memory_region_transaction_begin();
1813     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1814         if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
1815             break;
1816         }
1817     }
1818     ++mr->ioeventfd_nb;
1819     mr->ioeventfds = g_realloc(mr->ioeventfds,
1820                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
1821     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
1822             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
1823     mr->ioeventfds[i] = mrfd;
1824     ioeventfd_update_pending |= mr->enabled;
1825     memory_region_transaction_commit();
1826 }
1827
1828 void memory_region_del_eventfd(MemoryRegion *mr,
1829                                hwaddr addr,
1830                                unsigned size,
1831                                bool match_data,
1832                                uint64_t data,
1833                                EventNotifier *e)
1834 {
1835     MemoryRegionIoeventfd mrfd = {
1836         .addr.start = int128_make64(addr),
1837         .addr.size = int128_make64(size),
1838         .match_data = match_data,
1839         .data = data,
1840         .e = e,
1841     };
1842     unsigned i;
1843
1844     if (size) {
1845         adjust_endianness(mr, &mrfd.data, size);
1846     }
1847     memory_region_transaction_begin();
1848     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1849         if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
1850             break;
1851         }
1852     }
1853     assert(i != mr->ioeventfd_nb);
1854     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
1855             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
1856     --mr->ioeventfd_nb;
1857     mr->ioeventfds = g_realloc(mr->ioeventfds,
1858                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
1859     ioeventfd_update_pending |= mr->enabled;
1860     memory_region_transaction_commit();
1861 }
1862
1863 static void memory_region_update_container_subregions(MemoryRegion *subregion)
1864 {
1865     hwaddr offset = subregion->addr;
1866     MemoryRegion *mr = subregion->container;
1867     MemoryRegion *other;
1868
1869     memory_region_transaction_begin();
1870
1871     memory_region_ref(subregion);
1872     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1873         if (subregion->may_overlap || other->may_overlap) {
1874             continue;
1875         }
1876         if (int128_ge(int128_make64(offset),
1877                       int128_add(int128_make64(other->addr), other->size))
1878             || int128_le(int128_add(int128_make64(offset), subregion->size),
1879                          int128_make64(other->addr))) {
1880             continue;
1881         }
1882 #if 0
1883         printf("warning: subregion collision %llx/%llx (%s) "
1884                "vs %llx/%llx (%s)\n",
1885                (unsigned long long)offset,
1886                (unsigned long long)int128_get64(subregion->size),
1887                subregion->name,
1888                (unsigned long long)other->addr,
1889                (unsigned long long)int128_get64(other->size),
1890                other->name);
1891 #endif
1892     }
1893     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1894         if (subregion->priority >= other->priority) {
1895             QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
1896             goto done;
1897         }
1898     }
1899     QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
1900 done:
1901     memory_region_update_pending |= mr->enabled && subregion->enabled;
1902     memory_region_transaction_commit();
1903 }
1904
1905 static void memory_region_add_subregion_common(MemoryRegion *mr,
1906                                                hwaddr offset,
1907                                                MemoryRegion *subregion)
1908 {
1909     assert(!subregion->container);
1910     subregion->container = mr;
1911     subregion->addr = offset;
1912     memory_region_update_container_subregions(subregion);
1913 }
1914
1915 void memory_region_add_subregion(MemoryRegion *mr,
1916                                  hwaddr offset,
1917                                  MemoryRegion *subregion)
1918 {
1919     subregion->may_overlap = false;
1920     subregion->priority = 0;
1921     memory_region_add_subregion_common(mr, offset, subregion);
1922 }
1923
1924 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1925                                          hwaddr offset,
1926                                          MemoryRegion *subregion,
1927                                          int priority)
1928 {
1929     subregion->may_overlap = true;
1930     subregion->priority = priority;
1931     memory_region_add_subregion_common(mr, offset, subregion);
1932 }
1933
1934 void memory_region_del_subregion(MemoryRegion *mr,
1935                                  MemoryRegion *subregion)
1936 {
1937     memory_region_transaction_begin();
1938     assert(subregion->container == mr);
1939     subregion->container = NULL;
1940     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
1941     memory_region_unref(subregion);
1942     memory_region_update_pending |= mr->enabled && subregion->enabled;
1943     memory_region_transaction_commit();
1944 }
1945
1946 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
1947 {
1948     if (enabled == mr->enabled) {
1949         return;
1950     }
1951     memory_region_transaction_begin();
1952     mr->enabled = enabled;
1953     memory_region_update_pending = true;
1954     memory_region_transaction_commit();
1955 }
1956
1957 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
1958 {
1959     Int128 s = int128_make64(size);
1960
1961     if (size == UINT64_MAX) {
1962         s = int128_2_64();
1963     }
1964     if (int128_eq(s, mr->size)) {
1965         return;
1966     }
1967     memory_region_transaction_begin();
1968     mr->size = s;
1969     memory_region_update_pending = true;
1970     memory_region_transaction_commit();
1971 }
1972
1973 static void memory_region_readd_subregion(MemoryRegion *mr)
1974 {
1975     MemoryRegion *container = mr->container;
1976
1977     if (container) {
1978         memory_region_transaction_begin();
1979         memory_region_ref(mr);
1980         memory_region_del_subregion(container, mr);
1981         mr->container = container;
1982         memory_region_update_container_subregions(mr);
1983         memory_region_unref(mr);
1984         memory_region_transaction_commit();
1985     }
1986 }
1987
1988 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
1989 {
1990     if (addr != mr->addr) {
1991         mr->addr = addr;
1992         memory_region_readd_subregion(mr);
1993     }
1994 }
1995
1996 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
1997 {
1998     assert(mr->alias);
1999
2000     if (offset == mr->alias_offset) {
2001         return;
2002     }
2003
2004     memory_region_transaction_begin();
2005     mr->alias_offset = offset;
2006     memory_region_update_pending |= mr->enabled;
2007     memory_region_transaction_commit();
2008 }
2009
2010 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2011 {
2012     return mr->align;
2013 }
2014
2015 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2016 {
2017     const AddrRange *addr = addr_;
2018     const FlatRange *fr = fr_;
2019
2020     if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2021         return -1;
2022     } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2023         return 1;
2024     }
2025     return 0;
2026 }
2027
2028 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2029 {
2030     return bsearch(&addr, view->ranges, view->nr,
2031                    sizeof(FlatRange), cmp_flatrange_addr);
2032 }
2033
2034 bool memory_region_is_mapped(MemoryRegion *mr)
2035 {
2036     return mr->container ? true : false;
2037 }
2038
2039 /* Same as memory_region_find, but it does not add a reference to the
2040  * returned region.  It must be called from an RCU critical section.
2041  */
2042 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2043                                                   hwaddr addr, uint64_t size)
2044 {
2045     MemoryRegionSection ret = { .mr = NULL };
2046     MemoryRegion *root;
2047     AddressSpace *as;
2048     AddrRange range;
2049     FlatView *view;
2050     FlatRange *fr;
2051
2052     addr += mr->addr;
2053     for (root = mr; root->container; ) {
2054         root = root->container;
2055         addr += root->addr;
2056     }
2057
2058     as = memory_region_to_address_space(root);
2059     if (!as) {
2060         return ret;
2061     }
2062     range = addrrange_make(int128_make64(addr), int128_make64(size));
2063
2064     view = atomic_rcu_read(&as->current_map);
2065     fr = flatview_lookup(view, range);
2066     if (!fr) {
2067         return ret;
2068     }
2069
2070     while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2071         --fr;
2072     }
2073
2074     ret.mr = fr->mr;
2075     ret.address_space = as;
2076     range = addrrange_intersection(range, fr->addr);
2077     ret.offset_within_region = fr->offset_in_region;
2078     ret.offset_within_region += int128_get64(int128_sub(range.start,
2079                                                         fr->addr.start));
2080     ret.size = range.size;
2081     ret.offset_within_address_space = int128_get64(range.start);
2082     ret.readonly = fr->readonly;
2083     return ret;
2084 }
2085
2086 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2087                                        hwaddr addr, uint64_t size)
2088 {
2089     MemoryRegionSection ret;
2090     rcu_read_lock();
2091     ret = memory_region_find_rcu(mr, addr, size);
2092     if (ret.mr) {
2093         memory_region_ref(ret.mr);
2094     }
2095     rcu_read_unlock();
2096     return ret;
2097 }
2098
2099 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2100 {
2101     MemoryRegion *mr;
2102
2103     rcu_read_lock();
2104     mr = memory_region_find_rcu(container, addr, 1).mr;
2105     rcu_read_unlock();
2106     return mr && mr != container;
2107 }
2108
2109 void address_space_sync_dirty_bitmap(AddressSpace *as)
2110 {
2111     FlatView *view;
2112     FlatRange *fr;
2113
2114     view = address_space_get_flatview(as);
2115     FOR_EACH_FLAT_RANGE(fr, view) {
2116         MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
2117     }
2118     flatview_unref(view);
2119 }
2120
2121 void memory_global_dirty_log_start(void)
2122 {
2123     global_dirty_log = true;
2124
2125     MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2126
2127     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2128     memory_region_transaction_begin();
2129     memory_region_update_pending = true;
2130     memory_region_transaction_commit();
2131 }
2132
2133 void memory_global_dirty_log_stop(void)
2134 {
2135     global_dirty_log = false;
2136
2137     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2138     memory_region_transaction_begin();
2139     memory_region_update_pending = true;
2140     memory_region_transaction_commit();
2141
2142     MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2143 }
2144
2145 static void listener_add_address_space(MemoryListener *listener,
2146                                        AddressSpace *as)
2147 {
2148     FlatView *view;
2149     FlatRange *fr;
2150
2151     if (listener->address_space_filter
2152         && listener->address_space_filter != as) {
2153         return;
2154     }
2155
2156     if (listener->begin) {
2157         listener->begin(listener);
2158     }
2159     if (global_dirty_log) {
2160         if (listener->log_global_start) {
2161             listener->log_global_start(listener);
2162         }
2163     }
2164
2165     view = address_space_get_flatview(as);
2166     FOR_EACH_FLAT_RANGE(fr, view) {
2167         MemoryRegionSection section = {
2168             .mr = fr->mr,
2169             .address_space = as,
2170             .offset_within_region = fr->offset_in_region,
2171             .size = fr->addr.size,
2172             .offset_within_address_space = int128_get64(fr->addr.start),
2173             .readonly = fr->readonly,
2174         };
2175         if (fr->dirty_log_mask && listener->log_start) {
2176             listener->log_start(listener, &section, 0, fr->dirty_log_mask);
2177         }
2178         if (listener->region_add) {
2179             listener->region_add(listener, &section);
2180         }
2181     }
2182     if (listener->commit) {
2183         listener->commit(listener);
2184     }
2185     flatview_unref(view);
2186 }
2187
2188 void memory_listener_register(MemoryListener *listener, AddressSpace *filter)
2189 {
2190     MemoryListener *other = NULL;
2191     AddressSpace *as;
2192
2193     listener->address_space_filter = filter;
2194     if (QTAILQ_EMPTY(&memory_listeners)
2195         || listener->priority >= QTAILQ_LAST(&memory_listeners,
2196                                              memory_listeners)->priority) {
2197         QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
2198     } else {
2199         QTAILQ_FOREACH(other, &memory_listeners, link) {
2200             if (listener->priority < other->priority) {
2201                 break;
2202             }
2203         }
2204         QTAILQ_INSERT_BEFORE(other, listener, link);
2205     }
2206
2207     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2208         listener_add_address_space(listener, as);
2209     }
2210 }
2211
2212 void memory_listener_unregister(MemoryListener *listener)
2213 {
2214     QTAILQ_REMOVE(&memory_listeners, listener, link);
2215 }
2216
2217 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
2218 {
2219     memory_region_ref(root);
2220     memory_region_transaction_begin();
2221     as->ref_count = 1;
2222     as->root = root;
2223     as->malloced = false;
2224     as->current_map = g_new(FlatView, 1);
2225     flatview_init(as->current_map);
2226     as->ioeventfd_nb = 0;
2227     as->ioeventfds = NULL;
2228     QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
2229     as->name = g_strdup(name ? name : "anonymous");
2230     address_space_init_dispatch(as);
2231     memory_region_update_pending |= root->enabled;
2232     memory_region_transaction_commit();
2233 }
2234
2235 static void do_address_space_destroy(AddressSpace *as)
2236 {
2237     MemoryListener *listener;
2238     bool do_free = as->malloced;
2239
2240     address_space_destroy_dispatch(as);
2241
2242     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2243         assert(listener->address_space_filter != as);
2244     }
2245
2246     flatview_unref(as->current_map);
2247     g_free(as->name);
2248     g_free(as->ioeventfds);
2249     memory_region_unref(as->root);
2250     if (do_free) {
2251         g_free(as);
2252     }
2253 }
2254
2255 AddressSpace *address_space_init_shareable(MemoryRegion *root, const char *name)
2256 {
2257     AddressSpace *as;
2258
2259     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2260         if (root == as->root && as->malloced) {
2261             as->ref_count++;
2262             return as;
2263         }
2264     }
2265
2266     as = g_malloc0(sizeof *as);
2267     address_space_init(as, root, name);
2268     as->malloced = true;
2269     return as;
2270 }
2271
2272 void address_space_destroy(AddressSpace *as)
2273 {
2274     MemoryRegion *root = as->root;
2275
2276     as->ref_count--;
2277     if (as->ref_count) {
2278         return;
2279     }
2280     /* Flush out anything from MemoryListeners listening in on this */
2281     memory_region_transaction_begin();
2282     as->root = NULL;
2283     memory_region_transaction_commit();
2284     QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
2285     address_space_unregister(as);
2286
2287     /* At this point, as->dispatch and as->current_map are dummy
2288      * entries that the guest should never use.  Wait for the old
2289      * values to expire before freeing the data.
2290      */
2291     as->root = root;
2292     call_rcu(as, do_address_space_destroy, rcu);
2293 }
2294
2295 typedef struct MemoryRegionList MemoryRegionList;
2296
2297 struct MemoryRegionList {
2298     const MemoryRegion *mr;
2299     QTAILQ_ENTRY(MemoryRegionList) queue;
2300 };
2301
2302 typedef QTAILQ_HEAD(queue, MemoryRegionList) MemoryRegionListHead;
2303
2304 static void mtree_print_mr(fprintf_function mon_printf, void *f,
2305                            const MemoryRegion *mr, unsigned int level,
2306                            hwaddr base,
2307                            MemoryRegionListHead *alias_print_queue)
2308 {
2309     MemoryRegionList *new_ml, *ml, *next_ml;
2310     MemoryRegionListHead submr_print_queue;
2311     const MemoryRegion *submr;
2312     unsigned int i;
2313
2314     if (!mr) {
2315         return;
2316     }
2317
2318     for (i = 0; i < level; i++) {
2319         mon_printf(f, "  ");
2320     }
2321
2322     if (mr->alias) {
2323         MemoryRegionList *ml;
2324         bool found = false;
2325
2326         /* check if the alias is already in the queue */
2327         QTAILQ_FOREACH(ml, alias_print_queue, queue) {
2328             if (ml->mr == mr->alias) {
2329                 found = true;
2330             }
2331         }
2332
2333         if (!found) {
2334             ml = g_new(MemoryRegionList, 1);
2335             ml->mr = mr->alias;
2336             QTAILQ_INSERT_TAIL(alias_print_queue, ml, queue);
2337         }
2338         mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx
2339                    " (prio %d, %c%c): alias %s @%s " TARGET_FMT_plx
2340                    "-" TARGET_FMT_plx "%s\n",
2341                    base + mr->addr,
2342                    base + mr->addr
2343                    + (int128_nz(mr->size) ?
2344                       (hwaddr)int128_get64(int128_sub(mr->size,
2345                                                       int128_one())) : 0),
2346                    mr->priority,
2347                    mr->romd_mode ? 'R' : '-',
2348                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2349                                                                        : '-',
2350                    memory_region_name(mr),
2351                    memory_region_name(mr->alias),
2352                    mr->alias_offset,
2353                    mr->alias_offset
2354                    + (int128_nz(mr->size) ?
2355                       (hwaddr)int128_get64(int128_sub(mr->size,
2356                                                       int128_one())) : 0),
2357                    mr->enabled ? "" : " [disabled]");
2358     } else {
2359         mon_printf(f,
2360                    TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d, %c%c): %s%s\n",
2361                    base + mr->addr,
2362                    base + mr->addr
2363                    + (int128_nz(mr->size) ?
2364                       (hwaddr)int128_get64(int128_sub(mr->size,
2365                                                       int128_one())) : 0),
2366                    mr->priority,
2367                    mr->romd_mode ? 'R' : '-',
2368                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2369                                                                        : '-',
2370                    memory_region_name(mr),
2371                    mr->enabled ? "" : " [disabled]");
2372     }
2373
2374     QTAILQ_INIT(&submr_print_queue);
2375
2376     QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
2377         new_ml = g_new(MemoryRegionList, 1);
2378         new_ml->mr = submr;
2379         QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2380             if (new_ml->mr->addr < ml->mr->addr ||
2381                 (new_ml->mr->addr == ml->mr->addr &&
2382                  new_ml->mr->priority > ml->mr->priority)) {
2383                 QTAILQ_INSERT_BEFORE(ml, new_ml, queue);
2384                 new_ml = NULL;
2385                 break;
2386             }
2387         }
2388         if (new_ml) {
2389             QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, queue);
2390         }
2391     }
2392
2393     QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2394         mtree_print_mr(mon_printf, f, ml->mr, level + 1, base + mr->addr,
2395                        alias_print_queue);
2396     }
2397
2398     QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, queue, next_ml) {
2399         g_free(ml);
2400     }
2401 }
2402
2403 void mtree_info(fprintf_function mon_printf, void *f)
2404 {
2405     MemoryRegionListHead ml_head;
2406     MemoryRegionList *ml, *ml2;
2407     AddressSpace *as;
2408
2409     QTAILQ_INIT(&ml_head);
2410
2411     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2412         mon_printf(f, "address-space: %s\n", as->name);
2413         mtree_print_mr(mon_printf, f, as->root, 1, 0, &ml_head);
2414         mon_printf(f, "\n");
2415     }
2416
2417     /* print aliased regions */
2418     QTAILQ_FOREACH(ml, &ml_head, queue) {
2419         mon_printf(f, "memory-region: %s\n", memory_region_name(ml->mr));
2420         mtree_print_mr(mon_printf, f, ml->mr, 1, 0, &ml_head);
2421         mon_printf(f, "\n");
2422     }
2423
2424     QTAILQ_FOREACH_SAFE(ml, &ml_head, queue, ml2) {
2425         g_free(ml);
2426     }
2427 }
2428
2429 static const TypeInfo memory_region_info = {
2430     .parent             = TYPE_OBJECT,
2431     .name               = TYPE_MEMORY_REGION,
2432     .instance_size      = sizeof(MemoryRegion),
2433     .instance_init      = memory_region_initfn,
2434     .instance_finalize  = memory_region_finalize,
2435 };
2436
2437 static void memory_register_types(void)
2438 {
2439     type_register_static(&memory_region_info);
2440 }
2441
2442 type_init(memory_register_types)