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