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