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 "qemu-common.h"
  19 #include "cpu.h"
  20 #include "exec/memory.h"
  21 #include "exec/address-spaces.h"
  22 #include "exec/ioport.h"
  23 #include "qapi/visitor.h"
  24 #include "qemu/bitops.h"
  25 #include "qemu/error-report.h"
  26 #include "qom/object.h"
  27 #include "trace.h"
  28
  29 #include "exec/memory-internal.h"
  30 #include "exec/ram_addr.h"
  31 #include "sysemu/kvm.h"
  32 #include "sysemu/sysemu.h"
  33
  34 //#define DEBUG_UNASSIGNED
  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 void memory_region_get_size(Object *obj, Visitor *v, const char *name,
1059                                    void *opaque, Error **errp)
1060 {
1061     MemoryRegion *mr = MEMORY_REGION(obj);
1062     uint64_t value = memory_region_size(mr);
1063
1064     visit_type_uint64(v, name, &value, errp);
1065 }
1066
1067 static void memory_region_initfn(Object *obj)
1068 {
1069     MemoryRegion *mr = MEMORY_REGION(obj);
1070     ObjectProperty *op;
1071
1072     mr->ops = &unassigned_mem_ops;
1073     mr->enabled = true;
1074     mr->romd_mode = true;
1075     mr->global_locking = true;
1076     mr->destructor = memory_region_destructor_none;
1077     QTAILQ_INIT(&mr->subregions);
1078     QTAILQ_INIT(&mr->coalesced);
1079
1080     op = object_property_add(OBJECT(mr), "container",
1081                              "link<" TYPE_MEMORY_REGION ">",
1082                              memory_region_get_container,
1083                              NULL, /* memory_region_set_container */
1084                              NULL, NULL, &error_abort);
1085     op->resolve = memory_region_resolve_container;
1086
1087     object_property_add(OBJECT(mr), "addr", "uint64",
1088                         memory_region_get_addr,
1089                         NULL, /* memory_region_set_addr */
1090                         NULL, NULL, &error_abort);
1091     object_property_add(OBJECT(mr), "priority", "uint32",
1092                         memory_region_get_priority,
1093                         NULL, /* memory_region_set_priority */
1094                         NULL, NULL, &error_abort);
1095     object_property_add(OBJECT(mr), "size", "uint64",
1096                         memory_region_get_size,
1097                         NULL, /* memory_region_set_size, */
1098                         NULL, NULL, &error_abort);
1099 }
1100
1101 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1102                                     unsigned size)
1103 {
1104 #ifdef DEBUG_UNASSIGNED
1105     printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
1106 #endif
1107     if (current_cpu != NULL) {
1108         cpu_unassigned_access(current_cpu, addr, false, false, 0, size);
1109     }
1110     return 0;
1111 }
1112
1113 static void unassigned_mem_write(void *opaque, hwaddr addr,
1114                                  uint64_t val, unsigned size)
1115 {
1116 #ifdef DEBUG_UNASSIGNED
1117     printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1118 #endif
1119     if (current_cpu != NULL) {
1120         cpu_unassigned_access(current_cpu, addr, true, false, 0, size);
1121     }
1122 }
1123
1124 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1125                                    unsigned size, bool is_write)
1126 {
1127     return false;
1128 }
1129
1130 const MemoryRegionOps unassigned_mem_ops = {
1131     .valid.accepts = unassigned_mem_accepts,
1132     .endianness = DEVICE_NATIVE_ENDIAN,
1133 };
1134
1135 bool memory_region_access_valid(MemoryRegion *mr,
1136                                 hwaddr addr,
1137                                 unsigned size,
1138                                 bool is_write)
1139 {
1140     int access_size_min, access_size_max;
1141     int access_size, i;
1142
1143     if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1144         return false;
1145     }
1146
1147     if (!mr->ops->valid.accepts) {
1148         return true;
1149     }
1150
1151     access_size_min = mr->ops->valid.min_access_size;
1152     if (!mr->ops->valid.min_access_size) {
1153         access_size_min = 1;
1154     }
1155
1156     access_size_max = mr->ops->valid.max_access_size;
1157     if (!mr->ops->valid.max_access_size) {
1158         access_size_max = 4;
1159     }
1160
1161     access_size = MAX(MIN(size, access_size_max), access_size_min);
1162     for (i = 0; i < size; i += access_size) {
1163         if (!mr->ops->valid.accepts(mr->opaque, addr + i, access_size,
1164                                     is_write)) {
1165             return false;
1166         }
1167     }
1168
1169     return true;
1170 }
1171
1172 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1173                                                 hwaddr addr,
1174                                                 uint64_t *pval,
1175                                                 unsigned size,
1176                                                 MemTxAttrs attrs)
1177 {
1178     *pval = 0;
1179
1180     if (mr->ops->read) {
1181         return access_with_adjusted_size(addr, pval, size,
1182                                          mr->ops->impl.min_access_size,
1183                                          mr->ops->impl.max_access_size,
1184                                          memory_region_read_accessor,
1185                                          mr, attrs);
1186     } else if (mr->ops->read_with_attrs) {
1187         return access_with_adjusted_size(addr, pval, size,
1188                                          mr->ops->impl.min_access_size,
1189                                          mr->ops->impl.max_access_size,
1190                                          memory_region_read_with_attrs_accessor,
1191                                          mr, attrs);
1192     } else {
1193         return access_with_adjusted_size(addr, pval, size, 1, 4,
1194                                          memory_region_oldmmio_read_accessor,
1195                                          mr, attrs);
1196     }
1197 }
1198
1199 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1200                                         hwaddr addr,
1201                                         uint64_t *pval,
1202                                         unsigned size,
1203                                         MemTxAttrs attrs)
1204 {
1205     MemTxResult r;
1206
1207     if (!memory_region_access_valid(mr, addr, size, false)) {
1208         *pval = unassigned_mem_read(mr, addr, size);
1209         return MEMTX_DECODE_ERROR;
1210     }
1211
1212     r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1213     adjust_endianness(mr, pval, size);
1214     return r;
1215 }
1216
1217 /* Return true if an eventfd was signalled */
1218 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1219                                                     hwaddr addr,
1220                                                     uint64_t data,
1221                                                     unsigned size,
1222                                                     MemTxAttrs attrs)
1223 {
1224     MemoryRegionIoeventfd ioeventfd = {
1225         .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1226         .data = data,
1227     };
1228     unsigned i;
1229
1230     for (i = 0; i < mr->ioeventfd_nb; i++) {
1231         ioeventfd.match_data = mr->ioeventfds[i].match_data;
1232         ioeventfd.e = mr->ioeventfds[i].e;
1233
1234         if (memory_region_ioeventfd_equal(ioeventfd, mr->ioeventfds[i])) {
1235             event_notifier_set(ioeventfd.e);
1236             return true;
1237         }
1238     }
1239
1240     return false;
1241 }
1242
1243 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1244                                          hwaddr addr,
1245                                          uint64_t data,
1246                                          unsigned size,
1247                                          MemTxAttrs attrs)
1248 {
1249     if (!memory_region_access_valid(mr, addr, size, true)) {
1250         unassigned_mem_write(mr, addr, data, size);
1251         return MEMTX_DECODE_ERROR;
1252     }
1253
1254     adjust_endianness(mr, &data, size);
1255
1256     if ((!kvm_eventfds_enabled()) &&
1257         memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1258         return MEMTX_OK;
1259     }
1260
1261     if (mr->ops->write) {
1262         return access_with_adjusted_size(addr, &data, size,
1263                                          mr->ops->impl.min_access_size,
1264                                          mr->ops->impl.max_access_size,
1265                                          memory_region_write_accessor, mr,
1266                                          attrs);
1267     } else if (mr->ops->write_with_attrs) {
1268         return
1269             access_with_adjusted_size(addr, &data, size,
1270                                       mr->ops->impl.min_access_size,
1271                                       mr->ops->impl.max_access_size,
1272                                       memory_region_write_with_attrs_accessor,
1273                                       mr, attrs);
1274     } else {
1275         return access_with_adjusted_size(addr, &data, size, 1, 4,
1276                                          memory_region_oldmmio_write_accessor,
1277                                          mr, attrs);
1278     }
1279 }
1280
1281 void memory_region_init_io(MemoryRegion *mr,
1282                            Object *owner,
1283                            const MemoryRegionOps *ops,
1284                            void *opaque,
1285                            const char *name,
1286                            uint64_t size)
1287 {
1288     memory_region_init(mr, owner, name, size);
1289     mr->ops = ops ? ops : &unassigned_mem_ops;
1290     mr->opaque = opaque;
1291     mr->terminates = true;
1292 }
1293
1294 void memory_region_init_ram(MemoryRegion *mr,
1295                             Object *owner,
1296                             const char *name,
1297                             uint64_t size,
1298                             Error **errp)
1299 {
1300     memory_region_init(mr, owner, name, size);
1301     mr->ram = true;
1302     mr->terminates = true;
1303     mr->destructor = memory_region_destructor_ram;
1304     mr->ram_block = qemu_ram_alloc(size, mr, errp);
1305     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1306 }
1307
1308 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1309                                        Object *owner,
1310                                        const char *name,
1311                                        uint64_t size,
1312                                        uint64_t max_size,
1313                                        void (*resized)(const char*,
1314                                                        uint64_t length,
1315                                                        void *host),
1316                                        Error **errp)
1317 {
1318     memory_region_init(mr, owner, name, size);
1319     mr->ram = true;
1320     mr->terminates = true;
1321     mr->destructor = memory_region_destructor_ram;
1322     mr->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1323                                               mr, errp);
1324     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1325 }
1326
1327 #ifdef __linux__
1328 void memory_region_init_ram_from_file(MemoryRegion *mr,
1329                                       struct Object *owner,
1330                                       const char *name,
1331                                       uint64_t size,
1332                                       bool share,
1333                                       const char *path,
1334                                       Error **errp)
1335 {
1336     memory_region_init(mr, owner, name, size);
1337     mr->ram = true;
1338     mr->terminates = true;
1339     mr->destructor = memory_region_destructor_ram;
1340     mr->ram_block = qemu_ram_alloc_from_file(size, mr, share, path, errp);
1341     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1342 }
1343 #endif
1344
1345 void memory_region_init_ram_ptr(MemoryRegion *mr,
1346                                 Object *owner,
1347                                 const char *name,
1348                                 uint64_t size,
1349                                 void *ptr)
1350 {
1351     memory_region_init(mr, owner, name, size);
1352     mr->ram = true;
1353     mr->terminates = true;
1354     mr->destructor = memory_region_destructor_ram;
1355     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1356
1357     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1358     assert(ptr != NULL);
1359     mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1360 }
1361
1362 void memory_region_set_skip_dump(MemoryRegion *mr)
1363 {
1364     mr->skip_dump = true;
1365 }
1366
1367 void memory_region_init_alias(MemoryRegion *mr,
1368                               Object *owner,
1369                               const char *name,
1370                               MemoryRegion *orig,
1371                               hwaddr offset,
1372                               uint64_t size)
1373 {
1374     memory_region_init(mr, owner, name, size);
1375     mr->alias = orig;
1376     mr->alias_offset = offset;
1377 }
1378
1379 void memory_region_init_rom_device(MemoryRegion *mr,
1380                                    Object *owner,
1381                                    const MemoryRegionOps *ops,
1382                                    void *opaque,
1383                                    const char *name,
1384                                    uint64_t size,
1385                                    Error **errp)
1386 {
1387     memory_region_init(mr, owner, name, size);
1388     mr->ops = ops;
1389     mr->opaque = opaque;
1390     mr->terminates = true;
1391     mr->rom_device = true;
1392     mr->destructor = memory_region_destructor_rom_device;
1393     mr->ram_block = qemu_ram_alloc(size, mr, errp);
1394 }
1395
1396 void memory_region_init_iommu(MemoryRegion *mr,
1397                               Object *owner,
1398                               const MemoryRegionIOMMUOps *ops,
1399                               const char *name,
1400                               uint64_t size)
1401 {
1402     memory_region_init(mr, owner, name, size);
1403     mr->iommu_ops = ops,
1404     mr->terminates = true;  /* then re-forwards */
1405     notifier_list_init(&mr->iommu_notify);
1406 }
1407
1408 static void memory_region_finalize(Object *obj)
1409 {
1410     MemoryRegion *mr = MEMORY_REGION(obj);
1411
1412     assert(!mr->container);
1413
1414     /* We know the region is not visible in any address space (it
1415      * does not have a container and cannot be a root either because
1416      * it has no references, so we can blindly clear mr->enabled.
1417      * memory_region_set_enabled instead could trigger a transaction
1418      * and cause an infinite loop.
1419      */
1420     mr->enabled = false;
1421     memory_region_transaction_begin();
1422     while (!QTAILQ_EMPTY(&mr->subregions)) {
1423         MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1424         memory_region_del_subregion(mr, subregion);
1425     }
1426     memory_region_transaction_commit();
1427
1428     mr->destructor(mr);
1429     memory_region_clear_coalescing(mr);
1430     g_free((char *)mr->name);
1431     g_free(mr->ioeventfds);
1432 }
1433
1434 Object *memory_region_owner(MemoryRegion *mr)
1435 {
1436     Object *obj = OBJECT(mr);
1437     return obj->parent;
1438 }
1439
1440 void memory_region_ref(MemoryRegion *mr)
1441 {
1442     /* MMIO callbacks most likely will access data that belongs
1443      * to the owner, hence the need to ref/unref the owner whenever
1444      * the memory region is in use.
1445      *
1446      * The memory region is a child of its owner.  As long as the
1447      * owner doesn't call unparent itself on the memory region,
1448      * ref-ing the owner will also keep the memory region alive.
1449      * Memory regions without an owner are supposed to never go away;
1450      * we do not ref/unref them because it slows down DMA sensibly.
1451      */
1452     if (mr && mr->owner) {
1453         object_ref(mr->owner);
1454     }
1455 }
1456
1457 void memory_region_unref(MemoryRegion *mr)
1458 {
1459     if (mr && mr->owner) {
1460         object_unref(mr->owner);
1461     }
1462 }
1463
1464 uint64_t memory_region_size(MemoryRegion *mr)
1465 {
1466     if (int128_eq(mr->size, int128_2_64())) {
1467         return UINT64_MAX;
1468     }
1469     return int128_get64(mr->size);
1470 }
1471
1472 const char *memory_region_name(const MemoryRegion *mr)
1473 {
1474     if (!mr->name) {
1475         ((MemoryRegion *)mr)->name =
1476             object_get_canonical_path_component(OBJECT(mr));
1477     }
1478     return mr->name;
1479 }
1480
1481 bool memory_region_is_skip_dump(MemoryRegion *mr)
1482 {
1483     return mr->skip_dump;
1484 }
1485
1486 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1487 {
1488     uint8_t mask = mr->dirty_log_mask;
1489     if (global_dirty_log) {
1490         mask |= (1 << DIRTY_MEMORY_MIGRATION);
1491     }
1492     return mask;
1493 }
1494
1495 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1496 {
1497     return memory_region_get_dirty_log_mask(mr) & (1 << client);
1498 }
1499
1500 void memory_region_register_iommu_notifier(MemoryRegion *mr, Notifier *n)
1501 {
1502     notifier_list_add(&mr->iommu_notify, n);
1503 }
1504
1505 void memory_region_iommu_replay(MemoryRegion *mr, Notifier *n,
1506                                 hwaddr granularity, bool is_write)
1507 {
1508     hwaddr addr;
1509     IOMMUTLBEntry iotlb;
1510
1511     for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1512         iotlb = mr->iommu_ops->translate(mr, addr, is_write);
1513         if (iotlb.perm != IOMMU_NONE) {
1514             n->notify(n, &iotlb);
1515         }
1516
1517         /* if (2^64 - MR size) < granularity, it's possible to get an
1518          * infinite loop here.  This should catch such a wraparound */
1519         if ((addr + granularity) < addr) {
1520             break;
1521         }
1522     }
1523 }
1524
1525 void memory_region_unregister_iommu_notifier(Notifier *n)
1526 {
1527     notifier_remove(n);
1528 }
1529
1530 void memory_region_notify_iommu(MemoryRegion *mr,
1531                                 IOMMUTLBEntry entry)
1532 {
1533     assert(memory_region_is_iommu(mr));
1534     notifier_list_notify(&mr->iommu_notify, &entry);
1535 }
1536
1537 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1538 {
1539     uint8_t mask = 1 << client;
1540     uint8_t old_logging;
1541
1542     assert(client == DIRTY_MEMORY_VGA);
1543     old_logging = mr->vga_logging_count;
1544     mr->vga_logging_count += log ? 1 : -1;
1545     if (!!old_logging == !!mr->vga_logging_count) {
1546         return;
1547     }
1548
1549     memory_region_transaction_begin();
1550     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1551     memory_region_update_pending |= mr->enabled;
1552     memory_region_transaction_commit();
1553 }
1554
1555 bool memory_region_get_dirty(MemoryRegion *mr, hwaddr addr,
1556                              hwaddr size, unsigned client)
1557 {
1558     assert(mr->ram_block);
1559     return cpu_physical_memory_get_dirty(memory_region_get_ram_addr(mr) + addr,
1560                                          size, client);
1561 }
1562
1563 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1564                              hwaddr size)
1565 {
1566     assert(mr->ram_block);
1567     cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
1568                                         size,
1569                                         memory_region_get_dirty_log_mask(mr));
1570 }
1571
1572 bool memory_region_test_and_clear_dirty(MemoryRegion *mr, hwaddr addr,
1573                                         hwaddr size, unsigned client)
1574 {
1575     assert(mr->ram_block);
1576     return cpu_physical_memory_test_and_clear_dirty(
1577                 memory_region_get_ram_addr(mr) + addr, size, client);
1578 }
1579
1580
1581 void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1582 {
1583     AddressSpace *as;
1584     FlatRange *fr;
1585
1586     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1587         FlatView *view = address_space_get_flatview(as);
1588         FOR_EACH_FLAT_RANGE(fr, view) {
1589             if (fr->mr == mr) {
1590                 MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
1591             }
1592         }
1593         flatview_unref(view);
1594     }
1595 }
1596
1597 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
1598 {
1599     if (mr->readonly != readonly) {
1600         memory_region_transaction_begin();
1601         mr->readonly = readonly;
1602         memory_region_update_pending |= mr->enabled;
1603         memory_region_transaction_commit();
1604     }
1605 }
1606
1607 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
1608 {
1609     if (mr->romd_mode != romd_mode) {
1610         memory_region_transaction_begin();
1611         mr->romd_mode = romd_mode;
1612         memory_region_update_pending |= mr->enabled;
1613         memory_region_transaction_commit();
1614     }
1615 }
1616
1617 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
1618                                hwaddr size, unsigned client)
1619 {
1620     assert(mr->ram_block);
1621     cpu_physical_memory_test_and_clear_dirty(
1622         memory_region_get_ram_addr(mr) + addr, size, client);
1623 }
1624
1625 int memory_region_get_fd(MemoryRegion *mr)
1626 {
1627     int fd;
1628
1629     rcu_read_lock();
1630     while (mr->alias) {
1631         mr = mr->alias;
1632     }
1633     fd = mr->ram_block->fd;
1634     rcu_read_unlock();
1635
1636     return fd;
1637 }
1638
1639 void memory_region_set_fd(MemoryRegion *mr, int fd)
1640 {
1641     rcu_read_lock();
1642     while (mr->alias) {
1643         mr = mr->alias;
1644     }
1645     mr->ram_block->fd = fd;
1646     rcu_read_unlock();
1647 }
1648
1649 void *memory_region_get_ram_ptr(MemoryRegion *mr)
1650 {
1651     void *ptr;
1652     uint64_t offset = 0;
1653
1654     rcu_read_lock();
1655     while (mr->alias) {
1656         offset += mr->alias_offset;
1657         mr = mr->alias;
1658     }
1659     assert(mr->ram_block);
1660     ptr = qemu_map_ram_ptr(mr->ram_block, offset);
1661     rcu_read_unlock();
1662
1663     return ptr;
1664 }
1665
1666 MemoryRegion *memory_region_from_host(void *ptr, ram_addr_t *offset)
1667 {
1668     RAMBlock *block;
1669
1670     block = qemu_ram_block_from_host(ptr, false, offset);
1671     if (!block) {
1672         return NULL;
1673     }
1674
1675     return block->mr;
1676 }
1677
1678 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
1679 {
1680     return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
1681 }
1682
1683 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
1684 {
1685     assert(mr->ram_block);
1686
1687     qemu_ram_resize(mr->ram_block, newsize, errp);
1688 }
1689
1690 static void memory_region_update_coalesced_range_as(MemoryRegion *mr, AddressSpace *as)
1691 {
1692     FlatView *view;
1693     FlatRange *fr;
1694     CoalescedMemoryRange *cmr;
1695     AddrRange tmp;
1696     MemoryRegionSection section;
1697
1698     view = address_space_get_flatview(as);
1699     FOR_EACH_FLAT_RANGE(fr, view) {
1700         if (fr->mr == mr) {
1701             section = (MemoryRegionSection) {
1702                 .address_space = as,
1703                 .offset_within_address_space = int128_get64(fr->addr.start),
1704                 .size = fr->addr.size,
1705             };
1706
1707             MEMORY_LISTENER_CALL(coalesced_mmio_del, Reverse, &section,
1708                                  int128_get64(fr->addr.start),
1709                                  int128_get64(fr->addr.size));
1710             QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
1711                 tmp = addrrange_shift(cmr->addr,
1712                                       int128_sub(fr->addr.start,
1713                                                  int128_make64(fr->offset_in_region)));
1714                 if (!addrrange_intersects(tmp, fr->addr)) {
1715                     continue;
1716                 }
1717                 tmp = addrrange_intersection(tmp, fr->addr);
1718                 MEMORY_LISTENER_CALL(coalesced_mmio_add, Forward, &section,
1719                                      int128_get64(tmp.start),
1720                                      int128_get64(tmp.size));
1721             }
1722         }
1723     }
1724     flatview_unref(view);
1725 }
1726
1727 static void memory_region_update_coalesced_range(MemoryRegion *mr)
1728 {
1729     AddressSpace *as;
1730
1731     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1732         memory_region_update_coalesced_range_as(mr, as);
1733     }
1734 }
1735
1736 void memory_region_set_coalescing(MemoryRegion *mr)
1737 {
1738     memory_region_clear_coalescing(mr);
1739     memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
1740 }
1741
1742 void memory_region_add_coalescing(MemoryRegion *mr,
1743                                   hwaddr offset,
1744                                   uint64_t size)
1745 {
1746     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
1747
1748     cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
1749     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
1750     memory_region_update_coalesced_range(mr);
1751     memory_region_set_flush_coalesced(mr);
1752 }
1753
1754 void memory_region_clear_coalescing(MemoryRegion *mr)
1755 {
1756     CoalescedMemoryRange *cmr;
1757     bool updated = false;
1758
1759     qemu_flush_coalesced_mmio_buffer();
1760     mr->flush_coalesced_mmio = false;
1761
1762     while (!QTAILQ_EMPTY(&mr->coalesced)) {
1763         cmr = QTAILQ_FIRST(&mr->coalesced);
1764         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
1765         g_free(cmr);
1766         updated = true;
1767     }
1768
1769     if (updated) {
1770         memory_region_update_coalesced_range(mr);
1771     }
1772 }
1773
1774 void memory_region_set_flush_coalesced(MemoryRegion *mr)
1775 {
1776     mr->flush_coalesced_mmio = true;
1777 }
1778
1779 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
1780 {
1781     qemu_flush_coalesced_mmio_buffer();
1782     if (QTAILQ_EMPTY(&mr->coalesced)) {
1783         mr->flush_coalesced_mmio = false;
1784     }
1785 }
1786
1787 void memory_region_set_global_locking(MemoryRegion *mr)
1788 {
1789     mr->global_locking = true;
1790 }
1791
1792 void memory_region_clear_global_locking(MemoryRegion *mr)
1793 {
1794     mr->global_locking = false;
1795 }
1796
1797 static bool userspace_eventfd_warning;
1798
1799 void memory_region_add_eventfd(MemoryRegion *mr,
1800                                hwaddr addr,
1801                                unsigned size,
1802                                bool match_data,
1803                                uint64_t data,
1804                                EventNotifier *e)
1805 {
1806     MemoryRegionIoeventfd mrfd = {
1807         .addr.start = int128_make64(addr),
1808         .addr.size = int128_make64(size),
1809         .match_data = match_data,
1810         .data = data,
1811         .e = e,
1812     };
1813     unsigned i;
1814
1815     if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
1816                             userspace_eventfd_warning))) {
1817         userspace_eventfd_warning = true;
1818         error_report("Using eventfd without MMIO binding in KVM. "
1819                      "Suboptimal performance expected");
1820     }
1821
1822     if (size) {
1823         adjust_endianness(mr, &mrfd.data, size);
1824     }
1825     memory_region_transaction_begin();
1826     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1827         if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
1828             break;
1829         }
1830     }
1831     ++mr->ioeventfd_nb;
1832     mr->ioeventfds = g_realloc(mr->ioeventfds,
1833                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
1834     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
1835             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
1836     mr->ioeventfds[i] = mrfd;
1837     ioeventfd_update_pending |= mr->enabled;
1838     memory_region_transaction_commit();
1839 }
1840
1841 void memory_region_del_eventfd(MemoryRegion *mr,
1842                                hwaddr addr,
1843                                unsigned size,
1844                                bool match_data,
1845                                uint64_t data,
1846                                EventNotifier *e)
1847 {
1848     MemoryRegionIoeventfd mrfd = {
1849         .addr.start = int128_make64(addr),
1850         .addr.size = int128_make64(size),
1851         .match_data = match_data,
1852         .data = data,
1853         .e = e,
1854     };
1855     unsigned i;
1856
1857     if (size) {
1858         adjust_endianness(mr, &mrfd.data, size);
1859     }
1860     memory_region_transaction_begin();
1861     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1862         if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
1863             break;
1864         }
1865     }
1866     assert(i != mr->ioeventfd_nb);
1867     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
1868             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
1869     --mr->ioeventfd_nb;
1870     mr->ioeventfds = g_realloc(mr->ioeventfds,
1871                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
1872     ioeventfd_update_pending |= mr->enabled;
1873     memory_region_transaction_commit();
1874 }
1875
1876 static void memory_region_update_container_subregions(MemoryRegion *subregion)
1877 {
1878     MemoryRegion *mr = subregion->container;
1879     MemoryRegion *other;
1880
1881     memory_region_transaction_begin();
1882
1883     memory_region_ref(subregion);
1884     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1885         if (subregion->priority >= other->priority) {
1886             QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
1887             goto done;
1888         }
1889     }
1890     QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
1891 done:
1892     memory_region_update_pending |= mr->enabled && subregion->enabled;
1893     memory_region_transaction_commit();
1894 }
1895
1896 static void memory_region_add_subregion_common(MemoryRegion *mr,
1897                                                hwaddr offset,
1898                                                MemoryRegion *subregion)
1899 {
1900     assert(!subregion->container);
1901     subregion->container = mr;
1902     subregion->addr = offset;
1903     memory_region_update_container_subregions(subregion);
1904 }
1905
1906 void memory_region_add_subregion(MemoryRegion *mr,
1907                                  hwaddr offset,
1908                                  MemoryRegion *subregion)
1909 {
1910     subregion->priority = 0;
1911     memory_region_add_subregion_common(mr, offset, subregion);
1912 }
1913
1914 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1915                                          hwaddr offset,
1916                                          MemoryRegion *subregion,
1917                                          int priority)
1918 {
1919     subregion->priority = priority;
1920     memory_region_add_subregion_common(mr, offset, subregion);
1921 }
1922
1923 void memory_region_del_subregion(MemoryRegion *mr,
1924                                  MemoryRegion *subregion)
1925 {
1926     memory_region_transaction_begin();
1927     assert(subregion->container == mr);
1928     subregion->container = NULL;
1929     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
1930     memory_region_unref(subregion);
1931     memory_region_update_pending |= mr->enabled && subregion->enabled;
1932     memory_region_transaction_commit();
1933 }
1934
1935 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
1936 {
1937     if (enabled == mr->enabled) {
1938         return;
1939     }
1940     memory_region_transaction_begin();
1941     mr->enabled = enabled;
1942     memory_region_update_pending = true;
1943     memory_region_transaction_commit();
1944 }
1945
1946 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
1947 {
1948     Int128 s = int128_make64(size);
1949
1950     if (size == UINT64_MAX) {
1951         s = int128_2_64();
1952     }
1953     if (int128_eq(s, mr->size)) {
1954         return;
1955     }
1956     memory_region_transaction_begin();
1957     mr->size = s;
1958     memory_region_update_pending = true;
1959     memory_region_transaction_commit();
1960 }
1961
1962 static void memory_region_readd_subregion(MemoryRegion *mr)
1963 {
1964     MemoryRegion *container = mr->container;
1965
1966     if (container) {
1967         memory_region_transaction_begin();
1968         memory_region_ref(mr);
1969         memory_region_del_subregion(container, mr);
1970         mr->container = container;
1971         memory_region_update_container_subregions(mr);
1972         memory_region_unref(mr);
1973         memory_region_transaction_commit();
1974     }
1975 }
1976
1977 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
1978 {
1979     if (addr != mr->addr) {
1980         mr->addr = addr;
1981         memory_region_readd_subregion(mr);
1982     }
1983 }
1984
1985 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
1986 {
1987     assert(mr->alias);
1988
1989     if (offset == mr->alias_offset) {
1990         return;
1991     }
1992
1993     memory_region_transaction_begin();
1994     mr->alias_offset = offset;
1995     memory_region_update_pending |= mr->enabled;
1996     memory_region_transaction_commit();
1997 }
1998
1999 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2000 {
2001     return mr->align;
2002 }
2003
2004 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2005 {
2006     const AddrRange *addr = addr_;
2007     const FlatRange *fr = fr_;
2008
2009     if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2010         return -1;
2011     } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2012         return 1;
2013     }
2014     return 0;
2015 }
2016
2017 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2018 {
2019     return bsearch(&addr, view->ranges, view->nr,
2020                    sizeof(FlatRange), cmp_flatrange_addr);
2021 }
2022
2023 bool memory_region_is_mapped(MemoryRegion *mr)
2024 {
2025     return mr->container ? true : false;
2026 }
2027
2028 /* Same as memory_region_find, but it does not add a reference to the
2029  * returned region.  It must be called from an RCU critical section.
2030  */
2031 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2032                                                   hwaddr addr, uint64_t size)
2033 {
2034     MemoryRegionSection ret = { .mr = NULL };
2035     MemoryRegion *root;
2036     AddressSpace *as;
2037     AddrRange range;
2038     FlatView *view;
2039     FlatRange *fr;
2040
2041     addr += mr->addr;
2042     for (root = mr; root->container; ) {
2043         root = root->container;
2044         addr += root->addr;
2045     }
2046
2047     as = memory_region_to_address_space(root);
2048     if (!as) {
2049         return ret;
2050     }
2051     range = addrrange_make(int128_make64(addr), int128_make64(size));
2052
2053     view = atomic_rcu_read(&as->current_map);
2054     fr = flatview_lookup(view, range);
2055     if (!fr) {
2056         return ret;
2057     }
2058
2059     while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2060         --fr;
2061     }
2062
2063     ret.mr = fr->mr;
2064     ret.address_space = as;
2065     range = addrrange_intersection(range, fr->addr);
2066     ret.offset_within_region = fr->offset_in_region;
2067     ret.offset_within_region += int128_get64(int128_sub(range.start,
2068                                                         fr->addr.start));
2069     ret.size = range.size;
2070     ret.offset_within_address_space = int128_get64(range.start);
2071     ret.readonly = fr->readonly;
2072     return ret;
2073 }
2074
2075 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2076                                        hwaddr addr, uint64_t size)
2077 {
2078     MemoryRegionSection ret;
2079     rcu_read_lock();
2080     ret = memory_region_find_rcu(mr, addr, size);
2081     if (ret.mr) {
2082         memory_region_ref(ret.mr);
2083     }
2084     rcu_read_unlock();
2085     return ret;
2086 }
2087
2088 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2089 {
2090     MemoryRegion *mr;
2091
2092     rcu_read_lock();
2093     mr = memory_region_find_rcu(container, addr, 1).mr;
2094     rcu_read_unlock();
2095     return mr && mr != container;
2096 }
2097
2098 void address_space_sync_dirty_bitmap(AddressSpace *as)
2099 {
2100     FlatView *view;
2101     FlatRange *fr;
2102
2103     view = address_space_get_flatview(as);
2104     FOR_EACH_FLAT_RANGE(fr, view) {
2105         MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
2106     }
2107     flatview_unref(view);
2108 }
2109
2110 void memory_global_dirty_log_start(void)
2111 {
2112     global_dirty_log = true;
2113
2114     MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2115
2116     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2117     memory_region_transaction_begin();
2118     memory_region_update_pending = true;
2119     memory_region_transaction_commit();
2120 }
2121
2122 void memory_global_dirty_log_stop(void)
2123 {
2124     global_dirty_log = false;
2125
2126     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2127     memory_region_transaction_begin();
2128     memory_region_update_pending = true;
2129     memory_region_transaction_commit();
2130
2131     MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2132 }
2133
2134 static void listener_add_address_space(MemoryListener *listener,
2135                                        AddressSpace *as)
2136 {
2137     FlatView *view;
2138     FlatRange *fr;
2139
2140     if (listener->address_space_filter
2141         && listener->address_space_filter != as) {
2142         return;
2143     }
2144
2145     if (listener->begin) {
2146         listener->begin(listener);
2147     }
2148     if (global_dirty_log) {
2149         if (listener->log_global_start) {
2150             listener->log_global_start(listener);
2151         }
2152     }
2153
2154     view = address_space_get_flatview(as);
2155     FOR_EACH_FLAT_RANGE(fr, view) {
2156         MemoryRegionSection section = {
2157             .mr = fr->mr,
2158             .address_space = as,
2159             .offset_within_region = fr->offset_in_region,
2160             .size = fr->addr.size,
2161             .offset_within_address_space = int128_get64(fr->addr.start),
2162             .readonly = fr->readonly,
2163         };
2164         if (fr->dirty_log_mask && listener->log_start) {
2165             listener->log_start(listener, &section, 0, fr->dirty_log_mask);
2166         }
2167         if (listener->region_add) {
2168             listener->region_add(listener, &section);
2169         }
2170     }
2171     if (listener->commit) {
2172         listener->commit(listener);
2173     }
2174     flatview_unref(view);
2175 }
2176
2177 void memory_listener_register(MemoryListener *listener, AddressSpace *filter)
2178 {
2179     MemoryListener *other = NULL;
2180     AddressSpace *as;
2181
2182     listener->address_space_filter = filter;
2183     if (QTAILQ_EMPTY(&memory_listeners)
2184         || listener->priority >= QTAILQ_LAST(&memory_listeners,
2185                                              memory_listeners)->priority) {
2186         QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
2187     } else {
2188         QTAILQ_FOREACH(other, &memory_listeners, link) {
2189             if (listener->priority < other->priority) {
2190                 break;
2191             }
2192         }
2193         QTAILQ_INSERT_BEFORE(other, listener, link);
2194     }
2195
2196     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2197         listener_add_address_space(listener, as);
2198     }
2199 }
2200
2201 void memory_listener_unregister(MemoryListener *listener)
2202 {
2203     QTAILQ_REMOVE(&memory_listeners, listener, link);
2204 }
2205
2206 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
2207 {
2208     memory_region_ref(root);
2209     memory_region_transaction_begin();
2210     as->ref_count = 1;
2211     as->root = root;
2212     as->malloced = false;
2213     as->current_map = g_new(FlatView, 1);
2214     flatview_init(as->current_map);
2215     as->ioeventfd_nb = 0;
2216     as->ioeventfds = NULL;
2217     QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
2218     as->name = g_strdup(name ? name : "anonymous");
2219     address_space_init_dispatch(as);
2220     memory_region_update_pending |= root->enabled;
2221     memory_region_transaction_commit();
2222 }
2223
2224 static void do_address_space_destroy(AddressSpace *as)
2225 {
2226     MemoryListener *listener;
2227     bool do_free = as->malloced;
2228
2229     address_space_destroy_dispatch(as);
2230
2231     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2232         assert(listener->address_space_filter != as);
2233     }
2234
2235     flatview_unref(as->current_map);
2236     g_free(as->name);
2237     g_free(as->ioeventfds);
2238     memory_region_unref(as->root);
2239     if (do_free) {
2240         g_free(as);
2241     }
2242 }
2243
2244 AddressSpace *address_space_init_shareable(MemoryRegion *root, const char *name)
2245 {
2246     AddressSpace *as;
2247
2248     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2249         if (root == as->root && as->malloced) {
2250             as->ref_count++;
2251             return as;
2252         }
2253     }
2254
2255     as = g_malloc0(sizeof *as);
2256     address_space_init(as, root, name);
2257     as->malloced = true;
2258     return as;
2259 }
2260
2261 void address_space_destroy(AddressSpace *as)
2262 {
2263     MemoryRegion *root = as->root;
2264
2265     as->ref_count--;
2266     if (as->ref_count) {
2267         return;
2268     }
2269     /* Flush out anything from MemoryListeners listening in on this */
2270     memory_region_transaction_begin();
2271     as->root = NULL;
2272     memory_region_transaction_commit();
2273     QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
2274     address_space_unregister(as);
2275
2276     /* At this point, as->dispatch and as->current_map are dummy
2277      * entries that the guest should never use.  Wait for the old
2278      * values to expire before freeing the data.
2279      */
2280     as->root = root;
2281     call_rcu(as, do_address_space_destroy, rcu);
2282 }
2283
2284 typedef struct MemoryRegionList MemoryRegionList;
2285
2286 struct MemoryRegionList {
2287     const MemoryRegion *mr;
2288     QTAILQ_ENTRY(MemoryRegionList) queue;
2289 };
2290
2291 typedef QTAILQ_HEAD(queue, MemoryRegionList) MemoryRegionListHead;
2292
2293 static void mtree_print_mr(fprintf_function mon_printf, void *f,
2294                            const MemoryRegion *mr, unsigned int level,
2295                            hwaddr base,
2296                            MemoryRegionListHead *alias_print_queue)
2297 {
2298     MemoryRegionList *new_ml, *ml, *next_ml;
2299     MemoryRegionListHead submr_print_queue;
2300     const MemoryRegion *submr;
2301     unsigned int i;
2302
2303     if (!mr) {
2304         return;
2305     }
2306
2307     for (i = 0; i < level; i++) {
2308         mon_printf(f, "  ");
2309     }
2310
2311     if (mr->alias) {
2312         MemoryRegionList *ml;
2313         bool found = false;
2314
2315         /* check if the alias is already in the queue */
2316         QTAILQ_FOREACH(ml, alias_print_queue, queue) {
2317             if (ml->mr == mr->alias) {
2318                 found = true;
2319             }
2320         }
2321
2322         if (!found) {
2323             ml = g_new(MemoryRegionList, 1);
2324             ml->mr = mr->alias;
2325             QTAILQ_INSERT_TAIL(alias_print_queue, ml, queue);
2326         }
2327         mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx
2328                    " (prio %d, %c%c): alias %s @%s " TARGET_FMT_plx
2329                    "-" TARGET_FMT_plx "%s\n",
2330                    base + mr->addr,
2331                    base + mr->addr
2332                    + (int128_nz(mr->size) ?
2333                       (hwaddr)int128_get64(int128_sub(mr->size,
2334                                                       int128_one())) : 0),
2335                    mr->priority,
2336                    mr->romd_mode ? 'R' : '-',
2337                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2338                                                                        : '-',
2339                    memory_region_name(mr),
2340                    memory_region_name(mr->alias),
2341                    mr->alias_offset,
2342                    mr->alias_offset
2343                    + (int128_nz(mr->size) ?
2344                       (hwaddr)int128_get64(int128_sub(mr->size,
2345                                                       int128_one())) : 0),
2346                    mr->enabled ? "" : " [disabled]");
2347     } else {
2348         mon_printf(f,
2349                    TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d, %c%c): %s%s\n",
2350                    base + mr->addr,
2351                    base + mr->addr
2352                    + (int128_nz(mr->size) ?
2353                       (hwaddr)int128_get64(int128_sub(mr->size,
2354                                                       int128_one())) : 0),
2355                    mr->priority,
2356                    mr->romd_mode ? 'R' : '-',
2357                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2358                                                                        : '-',
2359                    memory_region_name(mr),
2360                    mr->enabled ? "" : " [disabled]");
2361     }
2362
2363     QTAILQ_INIT(&submr_print_queue);
2364
2365     QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
2366         new_ml = g_new(MemoryRegionList, 1);
2367         new_ml->mr = submr;
2368         QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2369             if (new_ml->mr->addr < ml->mr->addr ||
2370                 (new_ml->mr->addr == ml->mr->addr &&
2371                  new_ml->mr->priority > ml->mr->priority)) {
2372                 QTAILQ_INSERT_BEFORE(ml, new_ml, queue);
2373                 new_ml = NULL;
2374                 break;
2375             }
2376         }
2377         if (new_ml) {
2378             QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, queue);
2379         }
2380     }
2381
2382     QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2383         mtree_print_mr(mon_printf, f, ml->mr, level + 1, base + mr->addr,
2384                        alias_print_queue);
2385     }
2386
2387     QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, queue, next_ml) {
2388         g_free(ml);
2389     }
2390 }
2391
2392 void mtree_info(fprintf_function mon_printf, void *f)
2393 {
2394     MemoryRegionListHead ml_head;
2395     MemoryRegionList *ml, *ml2;
2396     AddressSpace *as;
2397
2398     QTAILQ_INIT(&ml_head);
2399
2400     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2401         mon_printf(f, "address-space: %s\n", as->name);
2402         mtree_print_mr(mon_printf, f, as->root, 1, 0, &ml_head);
2403         mon_printf(f, "\n");
2404     }
2405
2406     /* print aliased regions */
2407     QTAILQ_FOREACH(ml, &ml_head, queue) {
2408         mon_printf(f, "memory-region: %s\n", memory_region_name(ml->mr));
2409         mtree_print_mr(mon_printf, f, ml->mr, 1, 0, &ml_head);
2410         mon_printf(f, "\n");
2411     }
2412
2413     QTAILQ_FOREACH_SAFE(ml, &ml_head, queue, ml2) {
2414         g_free(ml);
2415     }
2416 }
2417
2418 static const TypeInfo memory_region_info = {
2419     .parent             = TYPE_OBJECT,
2420     .name               = TYPE_MEMORY_REGION,
2421     .instance_size      = sizeof(MemoryRegion),
2422     .instance_init      = memory_region_initfn,
2423     .instance_finalize  = memory_region_finalize,
2424 };
2425
2426 static void memory_register_types(void)
2427 {
2428     type_register_static(&memory_region_info);
2429 }
2430
2431 type_init(memory_register_types)