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 "exec/memory.h"
  18 #include "exec/address-spaces.h"
  19 #include "exec/ioport.h"
  20 #include "qapi/visitor.h"
  21 #include "qemu/bitops.h"
  22 #include "qemu/error-report.h"
  23 #include "qom/object.h"
  24 #include "trace.h"
  25
  26 #include "exec/memory-internal.h"
  27 #include "exec/ram_addr.h"
  28 #include "sysemu/kvm.h"
  29 #include "sysemu/sysemu.h"
  30
  31 //#define DEBUG_UNASSIGNED
  32
  33 #define RAM_ADDR_INVALID (~(ram_addr_t)0)
  34
  35 static unsigned memory_region_transaction_depth;
  36 static bool memory_region_update_pending;
  37 static bool ioeventfd_update_pending;
  38 static bool global_dirty_log = false;
  39
  40 static QTAILQ_HEAD(memory_listeners, MemoryListener) memory_listeners
  41     = QTAILQ_HEAD_INITIALIZER(memory_listeners);
  42
  43 static QTAILQ_HEAD(, AddressSpace) address_spaces
  44     = QTAILQ_HEAD_INITIALIZER(address_spaces);
  45
  46 typedef struct AddrRange AddrRange;
  47
  48 /*
  49  * Note that signed integers are needed for negative offsetting in aliases
  50  * (large MemoryRegion::alias_offset).
  51  */
  52 struct AddrRange {
  53     Int128 start;
  54     Int128 size;
  55 };
  56
  57 static AddrRange addrrange_make(Int128 start, Int128 size)
  58 {
  59     return (AddrRange) { start, size };
  60 }
  61
  62 static bool addrrange_equal(AddrRange r1, AddrRange r2)
  63 {
  64     return int128_eq(r1.start, r2.start) && int128_eq(r1.size, r2.size);
  65 }
  66
  67 static Int128 addrrange_end(AddrRange r)
  68 {
  69     return int128_add(r.start, r.size);
  70 }
  71
  72 static AddrRange addrrange_shift(AddrRange range, Int128 delta)
  73 {
  74     int128_addto(&range.start, delta);
  75     return range;
  76 }
  77
  78 static bool addrrange_contains(AddrRange range, Int128 addr)
  79 {
  80     return int128_ge(addr, range.start)
  81         && int128_lt(addr, addrrange_end(range));
  82 }
  83
  84 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
  85 {
  86     return addrrange_contains(r1, r2.start)
  87         || addrrange_contains(r2, r1.start);
  88 }
  89
  90 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
  91 {
  92     Int128 start = int128_max(r1.start, r2.start);
  93     Int128 end = int128_min(addrrange_end(r1), addrrange_end(r2));
  94     return addrrange_make(start, int128_sub(end, start));
  95 }
  96
  97 enum ListenerDirection { Forward, Reverse };
  98
  99 static bool memory_listener_match(MemoryListener *listener,
 100                                   MemoryRegionSection *section)
 101 {
 102     return !listener->address_space_filter
 103         || listener->address_space_filter == section->address_space;
 104 }
 105
 106 #define MEMORY_LISTENER_CALL_GLOBAL(_callback, _direction, _args...)    \
 107     do {                                                                \
 108         MemoryListener *_listener;                                      \
 109                                                                         \
 110         switch (_direction) {                                           \
 111         case Forward:                                                   \
 112             QTAILQ_FOREACH(_listener, &memory_listeners, link) {        \
 113                 if (_listener->_callback) {                             \
 114                     _listener->_callback(_listener, ##_args);           \
 115                 }                                                       \
 116             }                                                           \
 117             break;                                                      \
 118         case Reverse:                                                   \
 119             QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners,        \
 120                                    memory_listeners, link) {            \
 121                 if (_listener->_callback) {                             \
 122                     _listener->_callback(_listener, ##_args);           \
 123                 }                                                       \
 124             }                                                           \
 125             break;                                                      \
 126         default:                                                        \
 127             abort();                                                    \
 128         }                                                               \
 129     } while (0)
 130
 131 #define MEMORY_LISTENER_CALL(_callback, _direction, _section, _args...) \
 132     do {                                                                \
 133         MemoryListener *_listener;                                      \
 134                                                                         \
 135         switch (_direction) {                                           \
 136         case Forward:                                                   \
 137             QTAILQ_FOREACH(_listener, &memory_listeners, link) {        \
 138                 if (_listener->_callback                                \
 139                     && memory_listener_match(_listener, _section)) {    \
 140                     _listener->_callback(_listener, _section, ##_args); \
 141                 }                                                       \
 142             }                                                           \
 143             break;                                                      \
 144         case Reverse:                                                   \
 145             QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners,        \
 146                                    memory_listeners, link) {            \
 147                 if (_listener->_callback                                \
 148                     && memory_listener_match(_listener, _section)) {    \
 149                     _listener->_callback(_listener, _section, ##_args); \
 150                 }                                                       \
 151             }                                                           \
 152             break;                                                      \
 153         default:                                                        \
 154             abort();                                                    \
 155         }                                                               \
 156     } while (0)
 157
 158 /* No need to ref/unref .mr, the FlatRange keeps it alive.  */
 159 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...)  \
 160     MEMORY_LISTENER_CALL(callback, dir, (&(MemoryRegionSection) {       \
 161         .mr = (fr)->mr,                                                 \
 162         .address_space = (as),                                          \
 163         .offset_within_region = (fr)->offset_in_region,                 \
 164         .size = (fr)->addr.size,                                        \
 165         .offset_within_address_space = int128_get64((fr)->addr.start),  \
 166         .readonly = (fr)->readonly,                                     \
 167               }), ##_args)
 168
 169 struct CoalescedMemoryRange {
 170     AddrRange addr;
 171     QTAILQ_ENTRY(CoalescedMemoryRange) link;
 172 };
 173
 174 struct MemoryRegionIoeventfd {
 175     AddrRange addr;
 176     bool match_data;
 177     uint64_t data;
 178     EventNotifier *e;
 179 };
 180
 181 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a,
 182                                            MemoryRegionIoeventfd b)
 183 {
 184     if (int128_lt(a.addr.start, b.addr.start)) {
 185         return true;
 186     } else if (int128_gt(a.addr.start, b.addr.start)) {
 187         return false;
 188     } else if (int128_lt(a.addr.size, b.addr.size)) {
 189         return true;
 190     } else if (int128_gt(a.addr.size, b.addr.size)) {
 191         return false;
 192     } else if (a.match_data < b.match_data) {
 193         return true;
 194     } else  if (a.match_data > b.match_data) {
 195         return false;
 196     } else if (a.match_data) {
 197         if (a.data < b.data) {
 198             return true;
 199         } else if (a.data > b.data) {
 200             return false;
 201         }
 202     }
 203     if (a.e < b.e) {
 204         return true;
 205     } else if (a.e > b.e) {
 206         return false;
 207     }
 208     return false;
 209 }
 210
 211 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a,
 212                                           MemoryRegionIoeventfd b)
 213 {
 214     return !memory_region_ioeventfd_before(a, b)
 215         && !memory_region_ioeventfd_before(b, a);
 216 }
 217
 218 typedef struct FlatRange FlatRange;
 219 typedef struct FlatView FlatView;
 220
 221 /* Range of memory in the global map.  Addresses are absolute. */
 222 struct FlatRange {
 223     MemoryRegion *mr;
 224     hwaddr offset_in_region;
 225     AddrRange addr;
 226     uint8_t dirty_log_mask;
 227     bool romd_mode;
 228     bool readonly;
 229 };
 230
 231 /* Flattened global view of current active memory hierarchy.  Kept in sorted
 232  * order.
 233  */
 234 struct FlatView {
 235     struct rcu_head rcu;
 236     unsigned ref;
 237     FlatRange *ranges;
 238     unsigned nr;
 239     unsigned nr_allocated;
 240 };
 241
 242 typedef struct AddressSpaceOps AddressSpaceOps;
 243
 244 #define FOR_EACH_FLAT_RANGE(var, view)          \
 245     for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
 246
 247 static bool flatrange_equal(FlatRange *a, FlatRange *b)
 248 {
 249     return a->mr == b->mr
 250         && addrrange_equal(a->addr, b->addr)
 251         && a->offset_in_region == b->offset_in_region
 252         && a->romd_mode == b->romd_mode
 253         && a->readonly == b->readonly;
 254 }
 255
 256 static void flatview_init(FlatView *view)
 257 {
 258     view->ref = 1;
 259     view->ranges = NULL;
 260     view->nr = 0;
 261     view->nr_allocated = 0;
 262 }
 263
 264 /* Insert a range into a given position.  Caller is responsible for maintaining
 265  * sorting order.
 266  */
 267 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
 268 {
 269     if (view->nr == view->nr_allocated) {
 270         view->nr_allocated = MAX(2 * view->nr, 10);
 271         view->ranges = g_realloc(view->ranges,
 272                                     view->nr_allocated * sizeof(*view->ranges));
 273     }
 274     memmove(view->ranges + pos + 1, view->ranges + pos,
 275             (view->nr - pos) * sizeof(FlatRange));
 276     view->ranges[pos] = *range;
 277     memory_region_ref(range->mr);
 278     ++view->nr;
 279 }
 280
 281 static void flatview_destroy(FlatView *view)
 282 {
 283     int i;
 284
 285     for (i = 0; i < view->nr; i++) {
 286         memory_region_unref(view->ranges[i].mr);
 287     }
 288     g_free(view->ranges);
 289     g_free(view);
 290 }
 291
 292 static void flatview_ref(FlatView *view)
 293 {
 294     atomic_inc(&view->ref);
 295 }
 296
 297 static void flatview_unref(FlatView *view)
 298 {
 299     if (atomic_fetch_dec(&view->ref) == 1) {
 300         flatview_destroy(view);
 301     }
 302 }
 303
 304 static bool can_merge(FlatRange *r1, FlatRange *r2)
 305 {
 306     return int128_eq(addrrange_end(r1->addr), r2->addr.start)
 307         && r1->mr == r2->mr
 308         && int128_eq(int128_add(int128_make64(r1->offset_in_region),
 309                                 r1->addr.size),
 310                      int128_make64(r2->offset_in_region))
 311         && r1->dirty_log_mask == r2->dirty_log_mask
 312         && r1->romd_mode == r2->romd_mode
 313         && r1->readonly == r2->readonly;
 314 }
 315
 316 /* Attempt to simplify a view by merging adjacent ranges */
 317 static void flatview_simplify(FlatView *view)
 318 {
 319     unsigned i, j;
 320
 321     i = 0;
 322     while (i < view->nr) {
 323         j = i + 1;
 324         while (j < view->nr
 325                && can_merge(&view->ranges[j-1], &view->ranges[j])) {
 326             int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
 327             ++j;
 328         }
 329         ++i;
 330         memmove(&view->ranges[i], &view->ranges[j],
 331                 (view->nr - j) * sizeof(view->ranges[j]));
 332         view->nr -= j - i;
 333     }
 334 }
 335
 336 static bool memory_region_big_endian(MemoryRegion *mr)
 337 {
 338 #ifdef TARGET_WORDS_BIGENDIAN
 339     return mr->ops->endianness != DEVICE_LITTLE_ENDIAN;
 340 #else
 341     return mr->ops->endianness == DEVICE_BIG_ENDIAN;
 342 #endif
 343 }
 344
 345 static bool memory_region_wrong_endianness(MemoryRegion *mr)
 346 {
 347 #ifdef TARGET_WORDS_BIGENDIAN
 348     return mr->ops->endianness == DEVICE_LITTLE_ENDIAN;
 349 #else
 350     return mr->ops->endianness == DEVICE_BIG_ENDIAN;
 351 #endif
 352 }
 353
 354 static void adjust_endianness(MemoryRegion *mr, uint64_t *data, unsigned size)
 355 {
 356     if (memory_region_wrong_endianness(mr)) {
 357         switch (size) {
 358         case 1:
 359             break;
 360         case 2:
 361             *data = bswap16(*data);
 362             break;
 363         case 4:
 364             *data = bswap32(*data);
 365             break;
 366         case 8:
 367             *data = bswap64(*data);
 368             break;
 369         default:
 370             abort();
 371         }
 372     }
 373 }
 374
 375 static 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 (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 411         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 412         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
 413     }
 414     *value |= (tmp & mask) << shift;
 415     return MEMTX_OK;
 416 }
 417
 418 static MemTxResult  memory_region_read_accessor(MemoryRegion *mr,
 419                                                 hwaddr addr,
 420                                                 uint64_t *value,
 421                                                 unsigned size,
 422                                                 unsigned shift,
 423                                                 uint64_t mask,
 424                                                 MemTxAttrs attrs)
 425 {
 426     uint64_t tmp;
 427
 428     tmp = mr->ops->read(mr->opaque, addr, size);
 429     if (mr->subpage) {
 430         trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
 431     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 432         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 433         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
 434     }
 435     *value |= (tmp & mask) << shift;
 436     return MEMTX_OK;
 437 }
 438
 439 static MemTxResult memory_region_read_with_attrs_accessor(MemoryRegion *mr,
 440                                                           hwaddr addr,
 441                                                           uint64_t *value,
 442                                                           unsigned size,
 443                                                           unsigned shift,
 444                                                           uint64_t mask,
 445                                                           MemTxAttrs attrs)
 446 {
 447     uint64_t tmp = 0;
 448     MemTxResult r;
 449
 450     r = mr->ops->read_with_attrs(mr->opaque, addr, &tmp, size, attrs);
 451     if (mr->subpage) {
 452         trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
 453     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 454         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 455         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
 456     }
 457     *value |= (tmp & mask) << shift;
 458     return r;
 459 }
 460
 461 static MemTxResult memory_region_oldmmio_write_accessor(MemoryRegion *mr,
 462                                                         hwaddr addr,
 463                                                         uint64_t *value,
 464                                                         unsigned size,
 465                                                         unsigned shift,
 466                                                         uint64_t mask,
 467                                                         MemTxAttrs attrs)
 468 {
 469     uint64_t tmp;
 470
 471     tmp = (*value >> shift) & mask;
 472     if (mr->subpage) {
 473         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
 474     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 475         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 476         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
 477     }
 478     mr->ops->old_mmio.write[ctz32(size)](mr->opaque, addr, tmp);
 479     return MEMTX_OK;
 480 }
 481
 482 static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
 483                                                 hwaddr addr,
 484                                                 uint64_t *value,
 485                                                 unsigned size,
 486                                                 unsigned shift,
 487                                                 uint64_t mask,
 488                                                 MemTxAttrs attrs)
 489 {
 490     uint64_t tmp;
 491
 492     tmp = (*value >> shift) & mask;
 493     if (mr->subpage) {
 494         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
 495     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 496         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 497         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
 498     }
 499     mr->ops->write(mr->opaque, addr, tmp, size);
 500     return MEMTX_OK;
 501 }
 502
 503 static MemTxResult memory_region_write_with_attrs_accessor(MemoryRegion *mr,
 504                                                            hwaddr addr,
 505                                                            uint64_t *value,
 506                                                            unsigned size,
 507                                                            unsigned shift,
 508                                                            uint64_t mask,
 509                                                            MemTxAttrs attrs)
 510 {
 511     uint64_t tmp;
 512
 513     tmp = (*value >> shift) & mask;
 514     if (mr->subpage) {
 515         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
 516     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 517         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 518         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
 519     }
 520     return mr->ops->write_with_attrs(mr->opaque, addr, tmp, size, attrs);
 521 }
 522
 523 static MemTxResult access_with_adjusted_size(hwaddr addr,
 524                                       uint64_t *value,
 525                                       unsigned size,
 526                                       unsigned access_size_min,
 527                                       unsigned access_size_max,
 528                                       MemTxResult (*access)(MemoryRegion *mr,
 529                                                             hwaddr addr,
 530                                                             uint64_t *value,
 531                                                             unsigned size,
 532                                                             unsigned shift,
 533                                                             uint64_t mask,
 534                                                             MemTxAttrs attrs),
 535                                       MemoryRegion *mr,
 536                                       MemTxAttrs attrs)
 537 {
 538     uint64_t access_mask;
 539     unsigned access_size;
 540     unsigned i;
 541     MemTxResult r = MEMTX_OK;
 542
 543     if (!access_size_min) {
 544         access_size_min = 1;
 545     }
 546     if (!access_size_max) {
 547         access_size_max = 4;
 548     }
 549
 550     /* FIXME: support unaligned access? */
 551     access_size = MAX(MIN(size, access_size_max), access_size_min);
 552     access_mask = -1ULL >> (64 - access_size * 8);
 553     if (memory_region_big_endian(mr)) {
 554         for (i = 0; i < size; i += access_size) {
 555             r |= access(mr, addr + i, value, access_size,
 556                         (size - access_size - i) * 8, access_mask, attrs);
 557         }
 558     } else {
 559         for (i = 0; i < size; i += access_size) {
 560             r |= access(mr, addr + i, value, access_size, i * 8,
 561                         access_mask, attrs);
 562         }
 563     }
 564     return r;
 565 }
 566
 567 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
 568 {
 569     AddressSpace *as;
 570
 571     while (mr->container) {
 572         mr = mr->container;
 573     }
 574     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 575         if (mr == as->root) {
 576             return as;
 577         }
 578     }
 579     return NULL;
 580 }
 581
 582 /* Render a memory region into the global view.  Ranges in @view obscure
 583  * ranges in @mr.
 584  */
 585 static void render_memory_region(FlatView *view,
 586                                  MemoryRegion *mr,
 587                                  Int128 base,
 588                                  AddrRange clip,
 589                                  bool readonly)
 590 {
 591     MemoryRegion *subregion;
 592     unsigned i;
 593     hwaddr offset_in_region;
 594     Int128 remain;
 595     Int128 now;
 596     FlatRange fr;
 597     AddrRange tmp;
 598
 599     if (!mr->enabled) {
 600         return;
 601     }
 602
 603     int128_addto(&base, int128_make64(mr->addr));
 604     readonly |= mr->readonly;
 605
 606     tmp = addrrange_make(base, mr->size);
 607
 608     if (!addrrange_intersects(tmp, clip)) {
 609         return;
 610     }
 611
 612     clip = addrrange_intersection(tmp, clip);
 613
 614     if (mr->alias) {
 615         int128_subfrom(&base, int128_make64(mr->alias->addr));
 616         int128_subfrom(&base, int128_make64(mr->alias_offset));
 617         render_memory_region(view, mr->alias, base, clip, readonly);
 618         return;
 619     }
 620
 621     /* Render subregions in priority order. */
 622     QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
 623         render_memory_region(view, subregion, base, clip, readonly);
 624     }
 625
 626     if (!mr->terminates) {
 627         return;
 628     }
 629
 630     offset_in_region = int128_get64(int128_sub(clip.start, base));
 631     base = clip.start;
 632     remain = clip.size;
 633
 634     fr.mr = mr;
 635     fr.dirty_log_mask = memory_region_get_dirty_log_mask(mr);
 636     fr.romd_mode = mr->romd_mode;
 637     fr.readonly = readonly;
 638
 639     /* Render the region itself into any gaps left by the current view. */
 640     for (i = 0; i < view->nr && int128_nz(remain); ++i) {
 641         if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
 642             continue;
 643         }
 644         if (int128_lt(base, view->ranges[i].addr.start)) {
 645             now = int128_min(remain,
 646                              int128_sub(view->ranges[i].addr.start, base));
 647             fr.offset_in_region = offset_in_region;
 648             fr.addr = addrrange_make(base, now);
 649             flatview_insert(view, i, &fr);
 650             ++i;
 651             int128_addto(&base, now);
 652             offset_in_region += int128_get64(now);
 653             int128_subfrom(&remain, now);
 654         }
 655         now = int128_sub(int128_min(int128_add(base, remain),
 656                                     addrrange_end(view->ranges[i].addr)),
 657                          base);
 658         int128_addto(&base, now);
 659         offset_in_region += int128_get64(now);
 660         int128_subfrom(&remain, now);
 661     }
 662     if (int128_nz(remain)) {
 663         fr.offset_in_region = offset_in_region;
 664         fr.addr = addrrange_make(base, remain);
 665         flatview_insert(view, i, &fr);
 666     }
 667 }
 668
 669 /* Render a memory topology into a list of disjoint absolute ranges. */
 670 static FlatView *generate_memory_topology(MemoryRegion *mr)
 671 {
 672     FlatView *view;
 673
 674     view = g_new(FlatView, 1);
 675     flatview_init(view);
 676
 677     if (mr) {
 678         render_memory_region(view, mr, int128_zero(),
 679                              addrrange_make(int128_zero(), int128_2_64()), false);
 680     }
 681     flatview_simplify(view);
 682
 683     return view;
 684 }
 685
 686 static void address_space_add_del_ioeventfds(AddressSpace *as,
 687                                              MemoryRegionIoeventfd *fds_new,
 688                                              unsigned fds_new_nb,
 689                                              MemoryRegionIoeventfd *fds_old,
 690                                              unsigned fds_old_nb)
 691 {
 692     unsigned iold, inew;
 693     MemoryRegionIoeventfd *fd;
 694     MemoryRegionSection section;
 695
 696     /* Generate a symmetric difference of the old and new fd sets, adding
 697      * and deleting as necessary.
 698      */
 699
 700     iold = inew = 0;
 701     while (iold < fds_old_nb || inew < fds_new_nb) {
 702         if (iold < fds_old_nb
 703             && (inew == fds_new_nb
 704                 || memory_region_ioeventfd_before(fds_old[iold],
 705                                                   fds_new[inew]))) {
 706             fd = &fds_old[iold];
 707             section = (MemoryRegionSection) {
 708                 .address_space = as,
 709                 .offset_within_address_space = int128_get64(fd->addr.start),
 710                 .size = fd->addr.size,
 711             };
 712             MEMORY_LISTENER_CALL(eventfd_del, Forward, &section,
 713                                  fd->match_data, fd->data, fd->e);
 714             ++iold;
 715         } else if (inew < fds_new_nb
 716                    && (iold == fds_old_nb
 717                        || memory_region_ioeventfd_before(fds_new[inew],
 718                                                          fds_old[iold]))) {
 719             fd = &fds_new[inew];
 720             section = (MemoryRegionSection) {
 721                 .address_space = as,
 722                 .offset_within_address_space = int128_get64(fd->addr.start),
 723                 .size = fd->addr.size,
 724             };
 725             MEMORY_LISTENER_CALL(eventfd_add, Reverse, &section,
 726                                  fd->match_data, fd->data, fd->e);
 727             ++inew;
 728         } else {
 729             ++iold;
 730             ++inew;
 731         }
 732     }
 733 }
 734
 735 static FlatView *address_space_get_flatview(AddressSpace *as)
 736 {
 737     FlatView *view;
 738
 739     rcu_read_lock();
 740     view = atomic_rcu_read(&as->current_map);
 741     flatview_ref(view);
 742     rcu_read_unlock();
 743     return view;
 744 }
 745
 746 static void address_space_update_ioeventfds(AddressSpace *as)
 747 {
 748     FlatView *view;
 749     FlatRange *fr;
 750     unsigned ioeventfd_nb = 0;
 751     MemoryRegionIoeventfd *ioeventfds = NULL;
 752     AddrRange tmp;
 753     unsigned i;
 754
 755     view = address_space_get_flatview(as);
 756     FOR_EACH_FLAT_RANGE(fr, view) {
 757         for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
 758             tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
 759                                   int128_sub(fr->addr.start,
 760                                              int128_make64(fr->offset_in_region)));
 761             if (addrrange_intersects(fr->addr, tmp)) {
 762                 ++ioeventfd_nb;
 763                 ioeventfds = g_realloc(ioeventfds,
 764                                           ioeventfd_nb * sizeof(*ioeventfds));
 765                 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
 766                 ioeventfds[ioeventfd_nb-1].addr = tmp;
 767             }
 768         }
 769     }
 770
 771     address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
 772                                      as->ioeventfds, as->ioeventfd_nb);
 773
 774     g_free(as->ioeventfds);
 775     as->ioeventfds = ioeventfds;
 776     as->ioeventfd_nb = ioeventfd_nb;
 777     flatview_unref(view);
 778 }
 779
 780 static void address_space_update_topology_pass(AddressSpace *as,
 781                                                const FlatView *old_view,
 782                                                const FlatView *new_view,
 783                                                bool adding)
 784 {
 785     unsigned iold, inew;
 786     FlatRange *frold, *frnew;
 787
 788     /* Generate a symmetric difference of the old and new memory maps.
 789      * Kill ranges in the old map, and instantiate ranges in the new map.
 790      */
 791     iold = inew = 0;
 792     while (iold < old_view->nr || inew < new_view->nr) {
 793         if (iold < old_view->nr) {
 794             frold = &old_view->ranges[iold];
 795         } else {
 796             frold = NULL;
 797         }
 798         if (inew < new_view->nr) {
 799             frnew = &new_view->ranges[inew];
 800         } else {
 801             frnew = NULL;
 802         }
 803
 804         if (frold
 805             && (!frnew
 806                 || int128_lt(frold->addr.start, frnew->addr.start)
 807                 || (int128_eq(frold->addr.start, frnew->addr.start)
 808                     && !flatrange_equal(frold, frnew)))) {
 809             /* In old but not in new, or in both but attributes changed. */
 810
 811             if (!adding) {
 812                 MEMORY_LISTENER_UPDATE_REGION(frold, as, Reverse, region_del);
 813             }
 814
 815             ++iold;
 816         } else if (frold && frnew && flatrange_equal(frold, frnew)) {
 817             /* In both and unchanged (except logging may have changed) */
 818
 819             if (adding) {
 820                 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_nop);
 821                 if (frnew->dirty_log_mask & ~frold->dirty_log_mask) {
 822                     MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, log_start,
 823                                                   frold->dirty_log_mask,
 824                                                   frnew->dirty_log_mask);
 825                 }
 826                 if (frold->dirty_log_mask & ~frnew->dirty_log_mask) {
 827                     MEMORY_LISTENER_UPDATE_REGION(frnew, as, Reverse, log_stop,
 828                                                   frold->dirty_log_mask,
 829                                                   frnew->dirty_log_mask);
 830                 }
 831             }
 832
 833             ++iold;
 834             ++inew;
 835         } else {
 836             /* In new */
 837
 838             if (adding) {
 839                 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_add);
 840             }
 841
 842             ++inew;
 843         }
 844     }
 845 }
 846
 847
 848 static void address_space_update_topology(AddressSpace *as)
 849 {
 850     FlatView *old_view = address_space_get_flatview(as);
 851     FlatView *new_view = generate_memory_topology(as->root);
 852
 853     address_space_update_topology_pass(as, old_view, new_view, false);
 854     address_space_update_topology_pass(as, old_view, new_view, true);
 855
 856     /* Writes are protected by the BQL.  */
 857     atomic_rcu_set(&as->current_map, new_view);
 858     call_rcu(old_view, flatview_unref, rcu);
 859
 860     /* Note that all the old MemoryRegions are still alive up to this
 861      * point.  This relieves most MemoryListeners from the need to
 862      * ref/unref the MemoryRegions they get---unless they use them
 863      * outside the iothread mutex, in which case precise reference
 864      * counting is necessary.
 865      */
 866     flatview_unref(old_view);
 867
 868     address_space_update_ioeventfds(as);
 869 }
 870
 871 void memory_region_transaction_begin(void)
 872 {
 873     qemu_flush_coalesced_mmio_buffer();
 874     ++memory_region_transaction_depth;
 875 }
 876
 877 static void memory_region_clear_pending(void)
 878 {
 879     memory_region_update_pending = false;
 880     ioeventfd_update_pending = false;
 881 }
 882
 883 void memory_region_transaction_commit(void)
 884 {
 885     AddressSpace *as;
 886
 887     assert(memory_region_transaction_depth);
 888     --memory_region_transaction_depth;
 889     if (!memory_region_transaction_depth) {
 890         if (memory_region_update_pending) {
 891             MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
 892
 893             QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 894                 address_space_update_topology(as);
 895             }
 896
 897             MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
 898         } else if (ioeventfd_update_pending) {
 899             QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 900                 address_space_update_ioeventfds(as);
 901             }
 902         }
 903         memory_region_clear_pending();
 904    }
 905 }
 906
 907 static void memory_region_destructor_none(MemoryRegion *mr)
 908 {
 909 }
 910
 911 static void memory_region_destructor_ram(MemoryRegion *mr)
 912 {
 913     qemu_ram_free(mr->ram_block);
 914 }
 915
 916 static void memory_region_destructor_rom_device(MemoryRegion *mr)
 917 {
 918     qemu_ram_free(mr->ram_block);
 919 }
 920
 921 static bool memory_region_need_escape(char c)
 922 {
 923     return c == '/' || c == '[' || c == '\\' || c == ']';
 924 }
 925
 926 static char *memory_region_escape_name(const char *name)
 927 {
 928     const char *p;
 929     char *escaped, *q;
 930     uint8_t c;
 931     size_t bytes = 0;
 932
 933     for (p = name; *p; p++) {
 934         bytes += memory_region_need_escape(*p) ? 4 : 1;
 935     }
 936     if (bytes == p - name) {
 937        return g_memdup(name, bytes + 1);
 938     }
 939
 940     escaped = g_malloc(bytes + 1);
 941     for (p = name, q = escaped; *p; p++) {
 942         c = *p;
 943         if (unlikely(memory_region_need_escape(c))) {
 944             *q++ = '\\';
 945             *q++ = 'x';
 946             *q++ = "0123456789abcdef"[c >> 4];
 947             c = "0123456789abcdef"[c & 15];
 948         }
 949         *q++ = c;
 950     }
 951     *q = 0;
 952     return escaped;
 953 }
 954
 955 void memory_region_init(MemoryRegion *mr,
 956                         Object *owner,
 957                         const char *name,
 958                         uint64_t size)
 959 {
 960     object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
 961     mr->size = int128_make64(size);
 962     if (size == UINT64_MAX) {
 963         mr->size = int128_2_64();
 964     }
 965     mr->name = g_strdup(name);
 966     mr->owner = owner;
 967     mr->ram_block = NULL;
 968
 969     if (name) {
 970         char *escaped_name = memory_region_escape_name(name);
 971         char *name_array = g_strdup_printf("%s[*]", escaped_name);
 972
 973         if (!owner) {
 974             owner = container_get(qdev_get_machine(), "/unattached");
 975         }
 976
 977         object_property_add_child(owner, name_array, OBJECT(mr), &error_abort);
 978         object_unref(OBJECT(mr));
 979         g_free(name_array);
 980         g_free(escaped_name);
 981     }
 982 }
 983
 984 static void memory_region_get_addr(Object *obj, Visitor *v, const char *name,
 985                                    void *opaque, Error **errp)
 986 {
 987     MemoryRegion *mr = MEMORY_REGION(obj);
 988     uint64_t value = mr->addr;
 989
 990     visit_type_uint64(v, name, &value, errp);
 991 }
 992
 993 static void memory_region_get_container(Object *obj, Visitor *v,
 994                                         const char *name, void *opaque,
 995                                         Error **errp)
 996 {
 997     MemoryRegion *mr = MEMORY_REGION(obj);
 998     gchar *path = (gchar *)"";
 999
1000     if (mr->container) {
1001         path = object_get_canonical_path(OBJECT(mr->container));
1002     }
1003     visit_type_str(v, name, &path, errp);
1004     if (mr->container) {
1005         g_free(path);
1006     }
1007 }
1008
1009 static Object *memory_region_resolve_container(Object *obj, void *opaque,
1010                                                const char *part)
1011 {
1012     MemoryRegion *mr = MEMORY_REGION(obj);
1013
1014     return OBJECT(mr->container);
1015 }
1016
1017 static void memory_region_get_priority(Object *obj, Visitor *v,
1018                                        const char *name, void *opaque,
1019                                        Error **errp)
1020 {
1021     MemoryRegion *mr = MEMORY_REGION(obj);
1022     int32_t value = mr->priority;
1023
1024     visit_type_int32(v, name, &value, errp);
1025 }
1026
1027 static bool memory_region_get_may_overlap(Object *obj, Error **errp)
1028 {
1029     MemoryRegion *mr = MEMORY_REGION(obj);
1030
1031     return mr->may_overlap;
1032 }
1033
1034 static void memory_region_get_size(Object *obj, Visitor *v, const char *name,
1035                                    void *opaque, Error **errp)
1036 {
1037     MemoryRegion *mr = MEMORY_REGION(obj);
1038     uint64_t value = memory_region_size(mr);
1039
1040     visit_type_uint64(v, name, &value, errp);
1041 }
1042
1043 static void memory_region_initfn(Object *obj)
1044 {
1045     MemoryRegion *mr = MEMORY_REGION(obj);
1046     ObjectProperty *op;
1047
1048     mr->ops = &unassigned_mem_ops;
1049     mr->enabled = true;
1050     mr->romd_mode = true;
1051     mr->global_locking = true;
1052     mr->destructor = memory_region_destructor_none;
1053     QTAILQ_INIT(&mr->subregions);
1054     QTAILQ_INIT(&mr->coalesced);
1055
1056     op = object_property_add(OBJECT(mr), "container",
1057                              "link<" TYPE_MEMORY_REGION ">",
1058                              memory_region_get_container,
1059                              NULL, /* memory_region_set_container */
1060                              NULL, NULL, &error_abort);
1061     op->resolve = memory_region_resolve_container;
1062
1063     object_property_add(OBJECT(mr), "addr", "uint64",
1064                         memory_region_get_addr,
1065                         NULL, /* memory_region_set_addr */
1066                         NULL, NULL, &error_abort);
1067     object_property_add(OBJECT(mr), "priority", "uint32",
1068                         memory_region_get_priority,
1069                         NULL, /* memory_region_set_priority */
1070                         NULL, NULL, &error_abort);
1071     object_property_add_bool(OBJECT(mr), "may-overlap",
1072                              memory_region_get_may_overlap,
1073                              NULL, /* memory_region_set_may_overlap */
1074                              &error_abort);
1075     object_property_add(OBJECT(mr), "size", "uint64",
1076                         memory_region_get_size,
1077                         NULL, /* memory_region_set_size, */
1078                         NULL, NULL, &error_abort);
1079 }
1080
1081 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1082                                     unsigned size)
1083 {
1084 #ifdef DEBUG_UNASSIGNED
1085     printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
1086 #endif
1087     if (current_cpu != NULL) {
1088         cpu_unassigned_access(current_cpu, addr, false, false, 0, size);
1089     }
1090     return 0;
1091 }
1092
1093 static void unassigned_mem_write(void *opaque, hwaddr addr,
1094                                  uint64_t val, unsigned size)
1095 {
1096 #ifdef DEBUG_UNASSIGNED
1097     printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1098 #endif
1099     if (current_cpu != NULL) {
1100         cpu_unassigned_access(current_cpu, addr, true, false, 0, size);
1101     }
1102 }
1103
1104 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1105                                    unsigned size, bool is_write)
1106 {
1107     return false;
1108 }
1109
1110 const MemoryRegionOps unassigned_mem_ops = {
1111     .valid.accepts = unassigned_mem_accepts,
1112     .endianness = DEVICE_NATIVE_ENDIAN,
1113 };
1114
1115 bool memory_region_access_valid(MemoryRegion *mr,
1116                                 hwaddr addr,
1117                                 unsigned size,
1118                                 bool is_write)
1119 {
1120     int access_size_min, access_size_max;
1121     int access_size, i;
1122
1123     if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1124         return false;
1125     }
1126
1127     if (!mr->ops->valid.accepts) {
1128         return true;
1129     }
1130
1131     access_size_min = mr->ops->valid.min_access_size;
1132     if (!mr->ops->valid.min_access_size) {
1133         access_size_min = 1;
1134     }
1135
1136     access_size_max = mr->ops->valid.max_access_size;
1137     if (!mr->ops->valid.max_access_size) {
1138         access_size_max = 4;
1139     }
1140
1141     access_size = MAX(MIN(size, access_size_max), access_size_min);
1142     for (i = 0; i < size; i += access_size) {
1143         if (!mr->ops->valid.accepts(mr->opaque, addr + i, access_size,
1144                                     is_write)) {
1145             return false;
1146         }
1147     }
1148
1149     return true;
1150 }
1151
1152 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1153                                                 hwaddr addr,
1154                                                 uint64_t *pval,
1155                                                 unsigned size,
1156                                                 MemTxAttrs attrs)
1157 {
1158     *pval = 0;
1159
1160     if (mr->ops->read) {
1161         return access_with_adjusted_size(addr, pval, size,
1162                                          mr->ops->impl.min_access_size,
1163                                          mr->ops->impl.max_access_size,
1164                                          memory_region_read_accessor,
1165                                          mr, attrs);
1166     } else if (mr->ops->read_with_attrs) {
1167         return access_with_adjusted_size(addr, pval, size,
1168                                          mr->ops->impl.min_access_size,
1169                                          mr->ops->impl.max_access_size,
1170                                          memory_region_read_with_attrs_accessor,
1171                                          mr, attrs);
1172     } else {
1173         return access_with_adjusted_size(addr, pval, size, 1, 4,
1174                                          memory_region_oldmmio_read_accessor,
1175                                          mr, attrs);
1176     }
1177 }
1178
1179 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1180                                         hwaddr addr,
1181                                         uint64_t *pval,
1182                                         unsigned size,
1183                                         MemTxAttrs attrs)
1184 {
1185     MemTxResult r;
1186
1187     if (!memory_region_access_valid(mr, addr, size, false)) {
1188         *pval = unassigned_mem_read(mr, addr, size);
1189         return MEMTX_DECODE_ERROR;
1190     }
1191
1192     r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1193     adjust_endianness(mr, pval, size);
1194     return r;
1195 }
1196
1197 /* Return true if an eventfd was signalled */
1198 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1199                                                     hwaddr addr,
1200                                                     uint64_t data,
1201                                                     unsigned size,
1202                                                     MemTxAttrs attrs)
1203 {
1204     MemoryRegionIoeventfd ioeventfd = {
1205         .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1206         .data = data,
1207     };
1208     unsigned i;
1209
1210     for (i = 0; i < mr->ioeventfd_nb; i++) {
1211         ioeventfd.match_data = mr->ioeventfds[i].match_data;
1212         ioeventfd.e = mr->ioeventfds[i].e;
1213
1214         if (memory_region_ioeventfd_equal(ioeventfd, mr->ioeventfds[i])) {
1215             event_notifier_set(ioeventfd.e);
1216             return true;
1217         }
1218     }
1219
1220     return false;
1221 }
1222
1223 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1224                                          hwaddr addr,
1225                                          uint64_t data,
1226                                          unsigned size,
1227                                          MemTxAttrs attrs)
1228 {
1229     if (!memory_region_access_valid(mr, addr, size, true)) {
1230         unassigned_mem_write(mr, addr, data, size);
1231         return MEMTX_DECODE_ERROR;
1232     }
1233
1234     adjust_endianness(mr, &data, size);
1235
1236     if ((!kvm_eventfds_enabled()) &&
1237         memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1238         return MEMTX_OK;
1239     }
1240
1241     if (mr->ops->write) {
1242         return access_with_adjusted_size(addr, &data, size,
1243                                          mr->ops->impl.min_access_size,
1244                                          mr->ops->impl.max_access_size,
1245                                          memory_region_write_accessor, mr,
1246                                          attrs);
1247     } else if (mr->ops->write_with_attrs) {
1248         return
1249             access_with_adjusted_size(addr, &data, size,
1250                                       mr->ops->impl.min_access_size,
1251                                       mr->ops->impl.max_access_size,
1252                                       memory_region_write_with_attrs_accessor,
1253                                       mr, attrs);
1254     } else {
1255         return access_with_adjusted_size(addr, &data, size, 1, 4,
1256                                          memory_region_oldmmio_write_accessor,
1257                                          mr, attrs);
1258     }
1259 }
1260
1261 void memory_region_init_io(MemoryRegion *mr,
1262                            Object *owner,
1263                            const MemoryRegionOps *ops,
1264                            void *opaque,
1265                            const char *name,
1266                            uint64_t size)
1267 {
1268     memory_region_init(mr, owner, name, size);
1269     mr->ops = ops ? ops : &unassigned_mem_ops;
1270     mr->opaque = opaque;
1271     mr->terminates = true;
1272 }
1273
1274 void memory_region_init_ram(MemoryRegion *mr,
1275                             Object *owner,
1276                             const char *name,
1277                             uint64_t size,
1278                             Error **errp)
1279 {
1280     memory_region_init(mr, owner, name, size);
1281     mr->ram = true;
1282     mr->terminates = true;
1283     mr->destructor = memory_region_destructor_ram;
1284     mr->ram_block = qemu_ram_alloc(size, mr, errp);
1285     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1286 }
1287
1288 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1289                                        Object *owner,
1290                                        const char *name,
1291                                        uint64_t size,
1292                                        uint64_t max_size,
1293                                        void (*resized)(const char*,
1294                                                        uint64_t length,
1295                                                        void *host),
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_resizeable(size, max_size, resized,
1303                                               mr, errp);
1304     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1305 }
1306
1307 #ifdef __linux__
1308 void memory_region_init_ram_from_file(MemoryRegion *mr,
1309                                       struct Object *owner,
1310                                       const char *name,
1311                                       uint64_t size,
1312                                       bool share,
1313                                       const char *path,
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_from_file(size, mr, share, path, errp);
1321     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1322 }
1323 #endif
1324
1325 void memory_region_init_ram_ptr(MemoryRegion *mr,
1326                                 Object *owner,
1327                                 const char *name,
1328                                 uint64_t size,
1329                                 void *ptr)
1330 {
1331     memory_region_init(mr, owner, name, size);
1332     mr->ram = true;
1333     mr->terminates = true;
1334     mr->destructor = memory_region_destructor_ram;
1335     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1336
1337     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1338     assert(ptr != NULL);
1339     mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1340 }
1341
1342 void memory_region_set_skip_dump(MemoryRegion *mr)
1343 {
1344     mr->skip_dump = true;
1345 }
1346
1347 void memory_region_init_alias(MemoryRegion *mr,
1348                               Object *owner,
1349                               const char *name,
1350                               MemoryRegion *orig,
1351                               hwaddr offset,
1352                               uint64_t size)
1353 {
1354     memory_region_init(mr, owner, name, size);
1355     mr->alias = orig;
1356     mr->alias_offset = offset;
1357 }
1358
1359 void memory_region_init_rom_device(MemoryRegion *mr,
1360                                    Object *owner,
1361                                    const MemoryRegionOps *ops,
1362                                    void *opaque,
1363                                    const char *name,
1364                                    uint64_t size,
1365                                    Error **errp)
1366 {
1367     memory_region_init(mr, owner, name, size);
1368     mr->ops = ops;
1369     mr->opaque = opaque;
1370     mr->terminates = true;
1371     mr->rom_device = true;
1372     mr->destructor = memory_region_destructor_rom_device;
1373     mr->ram_block = qemu_ram_alloc(size, mr, errp);
1374 }
1375
1376 void memory_region_init_iommu(MemoryRegion *mr,
1377                               Object *owner,
1378                               const MemoryRegionIOMMUOps *ops,
1379                               const char *name,
1380                               uint64_t size)
1381 {
1382     memory_region_init(mr, owner, name, size);
1383     mr->iommu_ops = ops,
1384     mr->terminates = true;  /* then re-forwards */
1385     notifier_list_init(&mr->iommu_notify);
1386 }
1387
1388 static void memory_region_finalize(Object *obj)
1389 {
1390     MemoryRegion *mr = MEMORY_REGION(obj);
1391
1392     assert(!mr->container);
1393
1394     /* We know the region is not visible in any address space (it
1395      * does not have a container and cannot be a root either because
1396      * it has no references, so we can blindly clear mr->enabled.
1397      * memory_region_set_enabled instead could trigger a transaction
1398      * and cause an infinite loop.
1399      */
1400     mr->enabled = false;
1401     memory_region_transaction_begin();
1402     while (!QTAILQ_EMPTY(&mr->subregions)) {
1403         MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1404         memory_region_del_subregion(mr, subregion);
1405     }
1406     memory_region_transaction_commit();
1407
1408     mr->destructor(mr);
1409     memory_region_clear_coalescing(mr);
1410     g_free((char *)mr->name);
1411     g_free(mr->ioeventfds);
1412 }
1413
1414 Object *memory_region_owner(MemoryRegion *mr)
1415 {
1416     Object *obj = OBJECT(mr);
1417     return obj->parent;
1418 }
1419
1420 void memory_region_ref(MemoryRegion *mr)
1421 {
1422     /* MMIO callbacks most likely will access data that belongs
1423      * to the owner, hence the need to ref/unref the owner whenever
1424      * the memory region is in use.
1425      *
1426      * The memory region is a child of its owner.  As long as the
1427      * owner doesn't call unparent itself on the memory region,
1428      * ref-ing the owner will also keep the memory region alive.
1429      * Memory regions without an owner are supposed to never go away;
1430      * we do not ref/unref them because it slows down DMA sensibly.
1431      */
1432     if (mr && mr->owner) {
1433         object_ref(mr->owner);
1434     }
1435 }
1436
1437 void memory_region_unref(MemoryRegion *mr)
1438 {
1439     if (mr && mr->owner) {
1440         object_unref(mr->owner);
1441     }
1442 }
1443
1444 uint64_t memory_region_size(MemoryRegion *mr)
1445 {
1446     if (int128_eq(mr->size, int128_2_64())) {
1447         return UINT64_MAX;
1448     }
1449     return int128_get64(mr->size);
1450 }
1451
1452 const char *memory_region_name(const MemoryRegion *mr)
1453 {
1454     if (!mr->name) {
1455         ((MemoryRegion *)mr)->name =
1456             object_get_canonical_path_component(OBJECT(mr));
1457     }
1458     return mr->name;
1459 }
1460
1461 bool memory_region_is_skip_dump(MemoryRegion *mr)
1462 {
1463     return mr->skip_dump;
1464 }
1465
1466 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1467 {
1468     uint8_t mask = mr->dirty_log_mask;
1469     if (global_dirty_log) {
1470         mask |= (1 << DIRTY_MEMORY_MIGRATION);
1471     }
1472     return mask;
1473 }
1474
1475 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1476 {
1477     return memory_region_get_dirty_log_mask(mr) & (1 << client);
1478 }
1479
1480 void memory_region_register_iommu_notifier(MemoryRegion *mr, Notifier *n)
1481 {
1482     notifier_list_add(&mr->iommu_notify, n);
1483 }
1484
1485 void memory_region_iommu_replay(MemoryRegion *mr, Notifier *n,
1486                                 hwaddr granularity, bool is_write)
1487 {
1488     hwaddr addr;
1489     IOMMUTLBEntry iotlb;
1490
1491     for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1492         iotlb = mr->iommu_ops->translate(mr, addr, is_write);
1493         if (iotlb.perm != IOMMU_NONE) {
1494             n->notify(n, &iotlb);
1495         }
1496
1497         /* if (2^64 - MR size) < granularity, it's possible to get an
1498          * infinite loop here.  This should catch such a wraparound */
1499         if ((addr + granularity) < addr) {
1500             break;
1501         }
1502     }
1503 }
1504
1505 void memory_region_unregister_iommu_notifier(Notifier *n)
1506 {
1507     notifier_remove(n);
1508 }
1509
1510 void memory_region_notify_iommu(MemoryRegion *mr,
1511                                 IOMMUTLBEntry entry)
1512 {
1513     assert(memory_region_is_iommu(mr));
1514     notifier_list_notify(&mr->iommu_notify, &entry);
1515 }
1516
1517 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1518 {
1519     uint8_t mask = 1 << client;
1520     uint8_t old_logging;
1521
1522     assert(client == DIRTY_MEMORY_VGA);
1523     old_logging = mr->vga_logging_count;
1524     mr->vga_logging_count += log ? 1 : -1;
1525     if (!!old_logging == !!mr->vga_logging_count) {
1526         return;
1527     }
1528
1529     memory_region_transaction_begin();
1530     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1531     memory_region_update_pending |= mr->enabled;
1532     memory_region_transaction_commit();
1533 }
1534
1535 bool memory_region_get_dirty(MemoryRegion *mr, hwaddr addr,
1536                              hwaddr size, unsigned client)
1537 {
1538     assert(mr->ram_block);
1539     return cpu_physical_memory_get_dirty(memory_region_get_ram_addr(mr) + addr,
1540                                          size, client);
1541 }
1542
1543 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1544                              hwaddr size)
1545 {
1546     assert(mr->ram_block);
1547     cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
1548                                         size,
1549                                         memory_region_get_dirty_log_mask(mr));
1550 }
1551
1552 bool memory_region_test_and_clear_dirty(MemoryRegion *mr, hwaddr addr,
1553                                         hwaddr size, unsigned client)
1554 {
1555     assert(mr->ram_block);
1556     return cpu_physical_memory_test_and_clear_dirty(
1557                 memory_region_get_ram_addr(mr) + addr, size, client);
1558 }
1559
1560
1561 void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1562 {
1563     AddressSpace *as;
1564     FlatRange *fr;
1565
1566     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1567         FlatView *view = address_space_get_flatview(as);
1568         FOR_EACH_FLAT_RANGE(fr, view) {
1569             if (fr->mr == mr) {
1570                 MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
1571             }
1572         }
1573         flatview_unref(view);
1574     }
1575 }
1576
1577 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
1578 {
1579     if (mr->readonly != readonly) {
1580         memory_region_transaction_begin();
1581         mr->readonly = readonly;
1582         memory_region_update_pending |= mr->enabled;
1583         memory_region_transaction_commit();
1584     }
1585 }
1586
1587 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
1588 {
1589     if (mr->romd_mode != romd_mode) {
1590         memory_region_transaction_begin();
1591         mr->romd_mode = romd_mode;
1592         memory_region_update_pending |= mr->enabled;
1593         memory_region_transaction_commit();
1594     }
1595 }
1596
1597 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
1598                                hwaddr size, unsigned client)
1599 {
1600     assert(mr->ram_block);
1601     cpu_physical_memory_test_and_clear_dirty(
1602         memory_region_get_ram_addr(mr) + addr, size, client);
1603 }
1604
1605 int memory_region_get_fd(MemoryRegion *mr)
1606 {
1607     if (mr->alias) {
1608         return memory_region_get_fd(mr->alias);
1609     }
1610
1611     assert(mr->ram_block);
1612
1613     return qemu_get_ram_fd(memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK);
1614 }
1615
1616 void *memory_region_get_ram_ptr(MemoryRegion *mr)
1617 {
1618     void *ptr;
1619     uint64_t offset = 0;
1620
1621     rcu_read_lock();
1622     while (mr->alias) {
1623         offset += mr->alias_offset;
1624         mr = mr->alias;
1625     }
1626     assert(mr->ram_block);
1627     ptr = qemu_get_ram_ptr(mr->ram_block,
1628                            memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK);
1629     rcu_read_unlock();
1630
1631     return ptr + offset;
1632 }
1633
1634 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
1635 {
1636     return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
1637 }
1638
1639 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
1640 {
1641     assert(mr->ram_block);
1642
1643     qemu_ram_resize(memory_region_get_ram_addr(mr), newsize, errp);
1644 }
1645
1646 static void memory_region_update_coalesced_range_as(MemoryRegion *mr, AddressSpace *as)
1647 {
1648     FlatView *view;
1649     FlatRange *fr;
1650     CoalescedMemoryRange *cmr;
1651     AddrRange tmp;
1652     MemoryRegionSection section;
1653
1654     view = address_space_get_flatview(as);
1655     FOR_EACH_FLAT_RANGE(fr, view) {
1656         if (fr->mr == mr) {
1657             section = (MemoryRegionSection) {
1658                 .address_space = as,
1659                 .offset_within_address_space = int128_get64(fr->addr.start),
1660                 .size = fr->addr.size,
1661             };
1662
1663             MEMORY_LISTENER_CALL(coalesced_mmio_del, Reverse, &section,
1664                                  int128_get64(fr->addr.start),
1665                                  int128_get64(fr->addr.size));
1666             QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
1667                 tmp = addrrange_shift(cmr->addr,
1668                                       int128_sub(fr->addr.start,
1669                                                  int128_make64(fr->offset_in_region)));
1670                 if (!addrrange_intersects(tmp, fr->addr)) {
1671                     continue;
1672                 }
1673                 tmp = addrrange_intersection(tmp, fr->addr);
1674                 MEMORY_LISTENER_CALL(coalesced_mmio_add, Forward, &section,
1675                                      int128_get64(tmp.start),
1676                                      int128_get64(tmp.size));
1677             }
1678         }
1679     }
1680     flatview_unref(view);
1681 }
1682
1683 static void memory_region_update_coalesced_range(MemoryRegion *mr)
1684 {
1685     AddressSpace *as;
1686
1687     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1688         memory_region_update_coalesced_range_as(mr, as);
1689     }
1690 }
1691
1692 void memory_region_set_coalescing(MemoryRegion *mr)
1693 {
1694     memory_region_clear_coalescing(mr);
1695     memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
1696 }
1697
1698 void memory_region_add_coalescing(MemoryRegion *mr,
1699                                   hwaddr offset,
1700                                   uint64_t size)
1701 {
1702     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
1703
1704     cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
1705     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
1706     memory_region_update_coalesced_range(mr);
1707     memory_region_set_flush_coalesced(mr);
1708 }
1709
1710 void memory_region_clear_coalescing(MemoryRegion *mr)
1711 {
1712     CoalescedMemoryRange *cmr;
1713     bool updated = false;
1714
1715     qemu_flush_coalesced_mmio_buffer();
1716     mr->flush_coalesced_mmio = false;
1717
1718     while (!QTAILQ_EMPTY(&mr->coalesced)) {
1719         cmr = QTAILQ_FIRST(&mr->coalesced);
1720         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
1721         g_free(cmr);
1722         updated = true;
1723     }
1724
1725     if (updated) {
1726         memory_region_update_coalesced_range(mr);
1727     }
1728 }
1729
1730 void memory_region_set_flush_coalesced(MemoryRegion *mr)
1731 {
1732     mr->flush_coalesced_mmio = true;
1733 }
1734
1735 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
1736 {
1737     qemu_flush_coalesced_mmio_buffer();
1738     if (QTAILQ_EMPTY(&mr->coalesced)) {
1739         mr->flush_coalesced_mmio = false;
1740     }
1741 }
1742
1743 void memory_region_set_global_locking(MemoryRegion *mr)
1744 {
1745     mr->global_locking = true;
1746 }
1747
1748 void memory_region_clear_global_locking(MemoryRegion *mr)
1749 {
1750     mr->global_locking = false;
1751 }
1752
1753 static bool userspace_eventfd_warning;
1754
1755 void memory_region_add_eventfd(MemoryRegion *mr,
1756                                hwaddr addr,
1757                                unsigned size,
1758                                bool match_data,
1759                                uint64_t data,
1760                                EventNotifier *e)
1761 {
1762     MemoryRegionIoeventfd mrfd = {
1763         .addr.start = int128_make64(addr),
1764         .addr.size = int128_make64(size),
1765         .match_data = match_data,
1766         .data = data,
1767         .e = e,
1768     };
1769     unsigned i;
1770
1771     if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
1772                             userspace_eventfd_warning))) {
1773         userspace_eventfd_warning = true;
1774         error_report("Using eventfd without MMIO binding in KVM. "
1775                      "Suboptimal performance expected");
1776     }
1777
1778     if (size) {
1779         adjust_endianness(mr, &mrfd.data, size);
1780     }
1781     memory_region_transaction_begin();
1782     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1783         if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
1784             break;
1785         }
1786     }
1787     ++mr->ioeventfd_nb;
1788     mr->ioeventfds = g_realloc(mr->ioeventfds,
1789                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
1790     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
1791             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
1792     mr->ioeventfds[i] = mrfd;
1793     ioeventfd_update_pending |= mr->enabled;
1794     memory_region_transaction_commit();
1795 }
1796
1797 void memory_region_del_eventfd(MemoryRegion *mr,
1798                                hwaddr addr,
1799                                unsigned size,
1800                                bool match_data,
1801                                uint64_t data,
1802                                EventNotifier *e)
1803 {
1804     MemoryRegionIoeventfd mrfd = {
1805         .addr.start = int128_make64(addr),
1806         .addr.size = int128_make64(size),
1807         .match_data = match_data,
1808         .data = data,
1809         .e = e,
1810     };
1811     unsigned i;
1812
1813     if (size) {
1814         adjust_endianness(mr, &mrfd.data, size);
1815     }
1816     memory_region_transaction_begin();
1817     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1818         if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
1819             break;
1820         }
1821     }
1822     assert(i != mr->ioeventfd_nb);
1823     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
1824             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
1825     --mr->ioeventfd_nb;
1826     mr->ioeventfds = g_realloc(mr->ioeventfds,
1827                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
1828     ioeventfd_update_pending |= mr->enabled;
1829     memory_region_transaction_commit();
1830 }
1831
1832 static void memory_region_update_container_subregions(MemoryRegion *subregion)
1833 {
1834     hwaddr offset = subregion->addr;
1835     MemoryRegion *mr = subregion->container;
1836     MemoryRegion *other;
1837
1838     memory_region_transaction_begin();
1839
1840     memory_region_ref(subregion);
1841     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1842         if (subregion->may_overlap || other->may_overlap) {
1843             continue;
1844         }
1845         if (int128_ge(int128_make64(offset),
1846                       int128_add(int128_make64(other->addr), other->size))
1847             || int128_le(int128_add(int128_make64(offset), subregion->size),
1848                          int128_make64(other->addr))) {
1849             continue;
1850         }
1851 #if 0
1852         printf("warning: subregion collision %llx/%llx (%s) "
1853                "vs %llx/%llx (%s)\n",
1854                (unsigned long long)offset,
1855                (unsigned long long)int128_get64(subregion->size),
1856                subregion->name,
1857                (unsigned long long)other->addr,
1858                (unsigned long long)int128_get64(other->size),
1859                other->name);
1860 #endif
1861     }
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->may_overlap = false;
1889     subregion->priority = 0;
1890     memory_region_add_subregion_common(mr, offset, subregion);
1891 }
1892
1893 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1894                                          hwaddr offset,
1895                                          MemoryRegion *subregion,
1896                                          int priority)
1897 {
1898     subregion->may_overlap = true;
1899     subregion->priority = priority;
1900     memory_region_add_subregion_common(mr, offset, subregion);
1901 }
1902
1903 void memory_region_del_subregion(MemoryRegion *mr,
1904                                  MemoryRegion *subregion)
1905 {
1906     memory_region_transaction_begin();
1907     assert(subregion->container == mr);
1908     subregion->container = NULL;
1909     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
1910     memory_region_unref(subregion);
1911     memory_region_update_pending |= mr->enabled && subregion->enabled;
1912     memory_region_transaction_commit();
1913 }
1914
1915 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
1916 {
1917     if (enabled == mr->enabled) {
1918         return;
1919     }
1920     memory_region_transaction_begin();
1921     mr->enabled = enabled;
1922     memory_region_update_pending = true;
1923     memory_region_transaction_commit();
1924 }
1925
1926 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
1927 {
1928     Int128 s = int128_make64(size);
1929
1930     if (size == UINT64_MAX) {
1931         s = int128_2_64();
1932     }
1933     if (int128_eq(s, mr->size)) {
1934         return;
1935     }
1936     memory_region_transaction_begin();
1937     mr->size = s;
1938     memory_region_update_pending = true;
1939     memory_region_transaction_commit();
1940 }
1941
1942 static void memory_region_readd_subregion(MemoryRegion *mr)
1943 {
1944     MemoryRegion *container = mr->container;
1945
1946     if (container) {
1947         memory_region_transaction_begin();
1948         memory_region_ref(mr);
1949         memory_region_del_subregion(container, mr);
1950         mr->container = container;
1951         memory_region_update_container_subregions(mr);
1952         memory_region_unref(mr);
1953         memory_region_transaction_commit();
1954     }
1955 }
1956
1957 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
1958 {
1959     if (addr != mr->addr) {
1960         mr->addr = addr;
1961         memory_region_readd_subregion(mr);
1962     }
1963 }
1964
1965 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
1966 {
1967     assert(mr->alias);
1968
1969     if (offset == mr->alias_offset) {
1970         return;
1971     }
1972
1973     memory_region_transaction_begin();
1974     mr->alias_offset = offset;
1975     memory_region_update_pending |= mr->enabled;
1976     memory_region_transaction_commit();
1977 }
1978
1979 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
1980 {
1981     return mr->align;
1982 }
1983
1984 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
1985 {
1986     const AddrRange *addr = addr_;
1987     const FlatRange *fr = fr_;
1988
1989     if (int128_le(addrrange_end(*addr), fr->addr.start)) {
1990         return -1;
1991     } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
1992         return 1;
1993     }
1994     return 0;
1995 }
1996
1997 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
1998 {
1999     return bsearch(&addr, view->ranges, view->nr,
2000                    sizeof(FlatRange), cmp_flatrange_addr);
2001 }
2002
2003 bool memory_region_is_mapped(MemoryRegion *mr)
2004 {
2005     return mr->container ? true : false;
2006 }
2007
2008 /* Same as memory_region_find, but it does not add a reference to the
2009  * returned region.  It must be called from an RCU critical section.
2010  */
2011 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2012                                                   hwaddr addr, uint64_t size)
2013 {
2014     MemoryRegionSection ret = { .mr = NULL };
2015     MemoryRegion *root;
2016     AddressSpace *as;
2017     AddrRange range;
2018     FlatView *view;
2019     FlatRange *fr;
2020
2021     addr += mr->addr;
2022     for (root = mr; root->container; ) {
2023         root = root->container;
2024         addr += root->addr;
2025     }
2026
2027     as = memory_region_to_address_space(root);
2028     if (!as) {
2029         return ret;
2030     }
2031     range = addrrange_make(int128_make64(addr), int128_make64(size));
2032
2033     view = atomic_rcu_read(&as->current_map);
2034     fr = flatview_lookup(view, range);
2035     if (!fr) {
2036         return ret;
2037     }
2038
2039     while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2040         --fr;
2041     }
2042
2043     ret.mr = fr->mr;
2044     ret.address_space = as;
2045     range = addrrange_intersection(range, fr->addr);
2046     ret.offset_within_region = fr->offset_in_region;
2047     ret.offset_within_region += int128_get64(int128_sub(range.start,
2048                                                         fr->addr.start));
2049     ret.size = range.size;
2050     ret.offset_within_address_space = int128_get64(range.start);
2051     ret.readonly = fr->readonly;
2052     return ret;
2053 }
2054
2055 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2056                                        hwaddr addr, uint64_t size)
2057 {
2058     MemoryRegionSection ret;
2059     rcu_read_lock();
2060     ret = memory_region_find_rcu(mr, addr, size);
2061     if (ret.mr) {
2062         memory_region_ref(ret.mr);
2063     }
2064     rcu_read_unlock();
2065     return ret;
2066 }
2067
2068 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2069 {
2070     MemoryRegion *mr;
2071
2072     rcu_read_lock();
2073     mr = memory_region_find_rcu(container, addr, 1).mr;
2074     rcu_read_unlock();
2075     return mr && mr != container;
2076 }
2077
2078 void address_space_sync_dirty_bitmap(AddressSpace *as)
2079 {
2080     FlatView *view;
2081     FlatRange *fr;
2082
2083     view = address_space_get_flatview(as);
2084     FOR_EACH_FLAT_RANGE(fr, view) {
2085         MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
2086     }
2087     flatview_unref(view);
2088 }
2089
2090 void memory_global_dirty_log_start(void)
2091 {
2092     global_dirty_log = true;
2093
2094     MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2095
2096     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2097     memory_region_transaction_begin();
2098     memory_region_update_pending = true;
2099     memory_region_transaction_commit();
2100 }
2101
2102 void memory_global_dirty_log_stop(void)
2103 {
2104     global_dirty_log = false;
2105
2106     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2107     memory_region_transaction_begin();
2108     memory_region_update_pending = true;
2109     memory_region_transaction_commit();
2110
2111     MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2112 }
2113
2114 static void listener_add_address_space(MemoryListener *listener,
2115                                        AddressSpace *as)
2116 {
2117     FlatView *view;
2118     FlatRange *fr;
2119
2120     if (listener->address_space_filter
2121         && listener->address_space_filter != as) {
2122         return;
2123     }
2124
2125     if (listener->begin) {
2126         listener->begin(listener);
2127     }
2128     if (global_dirty_log) {
2129         if (listener->log_global_start) {
2130             listener->log_global_start(listener);
2131         }
2132     }
2133
2134     view = address_space_get_flatview(as);
2135     FOR_EACH_FLAT_RANGE(fr, view) {
2136         MemoryRegionSection section = {
2137             .mr = fr->mr,
2138             .address_space = as,
2139             .offset_within_region = fr->offset_in_region,
2140             .size = fr->addr.size,
2141             .offset_within_address_space = int128_get64(fr->addr.start),
2142             .readonly = fr->readonly,
2143         };
2144         if (fr->dirty_log_mask && listener->log_start) {
2145             listener->log_start(listener, &section, 0, fr->dirty_log_mask);
2146         }
2147         if (listener->region_add) {
2148             listener->region_add(listener, &section);
2149         }
2150     }
2151     if (listener->commit) {
2152         listener->commit(listener);
2153     }
2154     flatview_unref(view);
2155 }
2156
2157 void memory_listener_register(MemoryListener *listener, AddressSpace *filter)
2158 {
2159     MemoryListener *other = NULL;
2160     AddressSpace *as;
2161
2162     listener->address_space_filter = filter;
2163     if (QTAILQ_EMPTY(&memory_listeners)
2164         || listener->priority >= QTAILQ_LAST(&memory_listeners,
2165                                              memory_listeners)->priority) {
2166         QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
2167     } else {
2168         QTAILQ_FOREACH(other, &memory_listeners, link) {
2169             if (listener->priority < other->priority) {
2170                 break;
2171             }
2172         }
2173         QTAILQ_INSERT_BEFORE(other, listener, link);
2174     }
2175
2176     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2177         listener_add_address_space(listener, as);
2178     }
2179 }
2180
2181 void memory_listener_unregister(MemoryListener *listener)
2182 {
2183     QTAILQ_REMOVE(&memory_listeners, listener, link);
2184 }
2185
2186 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
2187 {
2188     memory_region_ref(root);
2189     memory_region_transaction_begin();
2190     as->ref_count = 1;
2191     as->root = root;
2192     as->malloced = false;
2193     as->current_map = g_new(FlatView, 1);
2194     flatview_init(as->current_map);
2195     as->ioeventfd_nb = 0;
2196     as->ioeventfds = NULL;
2197     QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
2198     as->name = g_strdup(name ? name : "anonymous");
2199     address_space_init_dispatch(as);
2200     memory_region_update_pending |= root->enabled;
2201     memory_region_transaction_commit();
2202 }
2203
2204 static void do_address_space_destroy(AddressSpace *as)
2205 {
2206     MemoryListener *listener;
2207     bool do_free = as->malloced;
2208
2209     address_space_destroy_dispatch(as);
2210
2211     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2212         assert(listener->address_space_filter != as);
2213     }
2214
2215     flatview_unref(as->current_map);
2216     g_free(as->name);
2217     g_free(as->ioeventfds);
2218     memory_region_unref(as->root);
2219     if (do_free) {
2220         g_free(as);
2221     }
2222 }
2223
2224 AddressSpace *address_space_init_shareable(MemoryRegion *root, const char *name)
2225 {
2226     AddressSpace *as;
2227
2228     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2229         if (root == as->root && as->malloced) {
2230             as->ref_count++;
2231             return as;
2232         }
2233     }
2234
2235     as = g_malloc0(sizeof *as);
2236     address_space_init(as, root, name);
2237     as->malloced = true;
2238     return as;
2239 }
2240
2241 void address_space_destroy(AddressSpace *as)
2242 {
2243     MemoryRegion *root = as->root;
2244
2245     as->ref_count--;
2246     if (as->ref_count) {
2247         return;
2248     }
2249     /* Flush out anything from MemoryListeners listening in on this */
2250     memory_region_transaction_begin();
2251     as->root = NULL;
2252     memory_region_transaction_commit();
2253     QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
2254     address_space_unregister(as);
2255
2256     /* At this point, as->dispatch and as->current_map are dummy
2257      * entries that the guest should never use.  Wait for the old
2258      * values to expire before freeing the data.
2259      */
2260     as->root = root;
2261     call_rcu(as, do_address_space_destroy, rcu);
2262 }
2263
2264 typedef struct MemoryRegionList MemoryRegionList;
2265
2266 struct MemoryRegionList {
2267     const MemoryRegion *mr;
2268     QTAILQ_ENTRY(MemoryRegionList) queue;
2269 };
2270
2271 typedef QTAILQ_HEAD(queue, MemoryRegionList) MemoryRegionListHead;
2272
2273 static void mtree_print_mr(fprintf_function mon_printf, void *f,
2274                            const MemoryRegion *mr, unsigned int level,
2275                            hwaddr base,
2276                            MemoryRegionListHead *alias_print_queue)
2277 {
2278     MemoryRegionList *new_ml, *ml, *next_ml;
2279     MemoryRegionListHead submr_print_queue;
2280     const MemoryRegion *submr;
2281     unsigned int i;
2282
2283     if (!mr) {
2284         return;
2285     }
2286
2287     for (i = 0; i < level; i++) {
2288         mon_printf(f, "  ");
2289     }
2290
2291     if (mr->alias) {
2292         MemoryRegionList *ml;
2293         bool found = false;
2294
2295         /* check if the alias is already in the queue */
2296         QTAILQ_FOREACH(ml, alias_print_queue, queue) {
2297             if (ml->mr == mr->alias) {
2298                 found = true;
2299             }
2300         }
2301
2302         if (!found) {
2303             ml = g_new(MemoryRegionList, 1);
2304             ml->mr = mr->alias;
2305             QTAILQ_INSERT_TAIL(alias_print_queue, ml, queue);
2306         }
2307         mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx
2308                    " (prio %d, %c%c): alias %s @%s " TARGET_FMT_plx
2309                    "-" TARGET_FMT_plx "%s\n",
2310                    base + mr->addr,
2311                    base + mr->addr
2312                    + (int128_nz(mr->size) ?
2313                       (hwaddr)int128_get64(int128_sub(mr->size,
2314                                                       int128_one())) : 0),
2315                    mr->priority,
2316                    mr->romd_mode ? 'R' : '-',
2317                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2318                                                                        : '-',
2319                    memory_region_name(mr),
2320                    memory_region_name(mr->alias),
2321                    mr->alias_offset,
2322                    mr->alias_offset
2323                    + (int128_nz(mr->size) ?
2324                       (hwaddr)int128_get64(int128_sub(mr->size,
2325                                                       int128_one())) : 0),
2326                    mr->enabled ? "" : " [disabled]");
2327     } else {
2328         mon_printf(f,
2329                    TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d, %c%c): %s%s\n",
2330                    base + mr->addr,
2331                    base + mr->addr
2332                    + (int128_nz(mr->size) ?
2333                       (hwaddr)int128_get64(int128_sub(mr->size,
2334                                                       int128_one())) : 0),
2335                    mr->priority,
2336                    mr->romd_mode ? 'R' : '-',
2337                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2338                                                                        : '-',
2339                    memory_region_name(mr),
2340                    mr->enabled ? "" : " [disabled]");
2341     }
2342
2343     QTAILQ_INIT(&submr_print_queue);
2344
2345     QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
2346         new_ml = g_new(MemoryRegionList, 1);
2347         new_ml->mr = submr;
2348         QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2349             if (new_ml->mr->addr < ml->mr->addr ||
2350                 (new_ml->mr->addr == ml->mr->addr &&
2351                  new_ml->mr->priority > ml->mr->priority)) {
2352                 QTAILQ_INSERT_BEFORE(ml, new_ml, queue);
2353                 new_ml = NULL;
2354                 break;
2355             }
2356         }
2357         if (new_ml) {
2358             QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, queue);
2359         }
2360     }
2361
2362     QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2363         mtree_print_mr(mon_printf, f, ml->mr, level + 1, base + mr->addr,
2364                        alias_print_queue);
2365     }
2366
2367     QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, queue, next_ml) {
2368         g_free(ml);
2369     }
2370 }
2371
2372 void mtree_info(fprintf_function mon_printf, void *f)
2373 {
2374     MemoryRegionListHead ml_head;
2375     MemoryRegionList *ml, *ml2;
2376     AddressSpace *as;
2377
2378     QTAILQ_INIT(&ml_head);
2379
2380     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2381         mon_printf(f, "address-space: %s\n", as->name);
2382         mtree_print_mr(mon_printf, f, as->root, 1, 0, &ml_head);
2383         mon_printf(f, "\n");
2384     }
2385
2386     /* print aliased regions */
2387     QTAILQ_FOREACH(ml, &ml_head, queue) {
2388         mon_printf(f, "memory-region: %s\n", memory_region_name(ml->mr));
2389         mtree_print_mr(mon_printf, f, ml->mr, 1, 0, &ml_head);
2390         mon_printf(f, "\n");
2391     }
2392
2393     QTAILQ_FOREACH_SAFE(ml, &ml_head, queue, ml2) {
2394         g_free(ml);
2395     }
2396 }
2397
2398 static const TypeInfo memory_region_info = {
2399     .parent             = TYPE_OBJECT,
2400     .name               = TYPE_MEMORY_REGION,
2401     .instance_size      = sizeof(MemoryRegion),
2402     .instance_init      = memory_region_initfn,
2403     .instance_finalize  = memory_region_finalize,
2404 };
2405
2406 static void memory_register_types(void)
2407 {
2408     type_register_static(&memory_region_info);
2409 }
2410
2411 type_init(memory_register_types)