memory.c

   1 /*
   2  * Physical memory management
   3  *
   4  * Copyright 2011 Red Hat, Inc. and/or its affiliates
   5  *
   6  * Authors:
   7  *  Avi Kivity <avi@redhat.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu/osdep.h"
  17 #include "qapi/error.h"
  18 #include "qemu-common.h"
  19 #include "cpu.h"
  20 #include "exec/memory.h"
  21 #include "exec/address-spaces.h"
  22 #include "exec/ioport.h"
  23 #include "qapi/visitor.h"
  24 #include "qemu/bitops.h"
  25 #include "qemu/error-report.h"
  26 #include "qom/object.h"
  27 #include "trace.h"
  28
  29 #include "exec/memory-internal.h"
  30 #include "exec/ram_addr.h"
  31 #include "sysemu/kvm.h"
  32 #include "sysemu/sysemu.h"
  33
  34 //#define DEBUG_UNASSIGNED
  35
  36 #define RAM_ADDR_INVALID (~(ram_addr_t)0)
  37
  38 static unsigned memory_region_transaction_depth;
  39 static bool memory_region_update_pending;
  40 static bool ioeventfd_update_pending;
  41 static bool global_dirty_log = false;
  42
  43 static QTAILQ_HEAD(memory_listeners, MemoryListener) memory_listeners
  44     = QTAILQ_HEAD_INITIALIZER(memory_listeners);
  45
  46 static QTAILQ_HEAD(, AddressSpace) address_spaces
  47     = QTAILQ_HEAD_INITIALIZER(address_spaces);
  48
  49 typedef struct AddrRange AddrRange;
  50
  51 /*
  52  * Note that signed integers are needed for negative offsetting in aliases
  53  * (large MemoryRegion::alias_offset).
  54  */
  55 struct AddrRange {
  56     Int128 start;
  57     Int128 size;
  58 };
  59
  60 static AddrRange addrrange_make(Int128 start, Int128 size)
  61 {
  62     return (AddrRange) { start, size };
  63 }
  64
  65 static bool addrrange_equal(AddrRange r1, AddrRange r2)
  66 {
  67     return int128_eq(r1.start, r2.start) && int128_eq(r1.size, r2.size);
  68 }
  69
  70 static Int128 addrrange_end(AddrRange r)
  71 {
  72     return int128_add(r.start, r.size);
  73 }
  74
  75 static AddrRange addrrange_shift(AddrRange range, Int128 delta)
  76 {
  77     int128_addto(&range.start, delta);
  78     return range;
  79 }
  80
  81 static bool addrrange_contains(AddrRange range, Int128 addr)
  82 {
  83     return int128_ge(addr, range.start)
  84         && int128_lt(addr, addrrange_end(range));
  85 }
  86
  87 static bool addrrange_intersects(AddrRange r1, AddrRange r2)
  88 {
  89     return addrrange_contains(r1, r2.start)
  90         || addrrange_contains(r2, r1.start);
  91 }
  92
  93 static AddrRange addrrange_intersection(AddrRange r1, AddrRange r2)
  94 {
  95     Int128 start = int128_max(r1.start, r2.start);
  96     Int128 end = int128_min(addrrange_end(r1), addrrange_end(r2));
  97     return addrrange_make(start, int128_sub(end, start));
  98 }
  99
 100 enum ListenerDirection { Forward, Reverse };
 101
 102 static bool memory_listener_match(MemoryListener *listener,
 103                                   MemoryRegionSection *section)
 104 {
 105     return !listener->address_space_filter
 106         || listener->address_space_filter == section->address_space;
 107 }
 108
 109 #define MEMORY_LISTENER_CALL_GLOBAL(_callback, _direction, _args...)    \
 110     do {                                                                \
 111         MemoryListener *_listener;                                      \
 112                                                                         \
 113         switch (_direction) {                                           \
 114         case Forward:                                                   \
 115             QTAILQ_FOREACH(_listener, &memory_listeners, link) {        \
 116                 if (_listener->_callback) {                             \
 117                     _listener->_callback(_listener, ##_args);           \
 118                 }                                                       \
 119             }                                                           \
 120             break;                                                      \
 121         case Reverse:                                                   \
 122             QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners,        \
 123                                    memory_listeners, link) {            \
 124                 if (_listener->_callback) {                             \
 125                     _listener->_callback(_listener, ##_args);           \
 126                 }                                                       \
 127             }                                                           \
 128             break;                                                      \
 129         default:                                                        \
 130             abort();                                                    \
 131         }                                                               \
 132     } while (0)
 133
 134 #define MEMORY_LISTENER_CALL(_callback, _direction, _section, _args...) \
 135     do {                                                                \
 136         MemoryListener *_listener;                                      \
 137                                                                         \
 138         switch (_direction) {                                           \
 139         case Forward:                                                   \
 140             QTAILQ_FOREACH(_listener, &memory_listeners, link) {        \
 141                 if (_listener->_callback                                \
 142                     && memory_listener_match(_listener, _section)) {    \
 143                     _listener->_callback(_listener, _section, ##_args); \
 144                 }                                                       \
 145             }                                                           \
 146             break;                                                      \
 147         case Reverse:                                                   \
 148             QTAILQ_FOREACH_REVERSE(_listener, &memory_listeners,        \
 149                                    memory_listeners, link) {            \
 150                 if (_listener->_callback                                \
 151                     && memory_listener_match(_listener, _section)) {    \
 152                     _listener->_callback(_listener, _section, ##_args); \
 153                 }                                                       \
 154             }                                                           \
 155             break;                                                      \
 156         default:                                                        \
 157             abort();                                                    \
 158         }                                                               \
 159     } while (0)
 160
 161 /* No need to ref/unref .mr, the FlatRange keeps it alive.  */
 162 #define MEMORY_LISTENER_UPDATE_REGION(fr, as, dir, callback, _args...)  \
 163     MEMORY_LISTENER_CALL(callback, dir, (&(MemoryRegionSection) {       \
 164         .mr = (fr)->mr,                                                 \
 165         .address_space = (as),                                          \
 166         .offset_within_region = (fr)->offset_in_region,                 \
 167         .size = (fr)->addr.size,                                        \
 168         .offset_within_address_space = int128_get64((fr)->addr.start),  \
 169         .readonly = (fr)->readonly,                                     \
 170               }), ##_args)
 171
 172 struct CoalescedMemoryRange {
 173     AddrRange addr;
 174     QTAILQ_ENTRY(CoalescedMemoryRange) link;
 175 };
 176
 177 struct MemoryRegionIoeventfd {
 178     AddrRange addr;
 179     bool match_data;
 180     uint64_t data;
 181     EventNotifier *e;
 182 };
 183
 184 static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a,
 185                                            MemoryRegionIoeventfd b)
 186 {
 187     if (int128_lt(a.addr.start, b.addr.start)) {
 188         return true;
 189     } else if (int128_gt(a.addr.start, b.addr.start)) {
 190         return false;
 191     } else if (int128_lt(a.addr.size, b.addr.size)) {
 192         return true;
 193     } else if (int128_gt(a.addr.size, b.addr.size)) {
 194         return false;
 195     } else if (a.match_data < b.match_data) {
 196         return true;
 197     } else  if (a.match_data > b.match_data) {
 198         return false;
 199     } else if (a.match_data) {
 200         if (a.data < b.data) {
 201             return true;
 202         } else if (a.data > b.data) {
 203             return false;
 204         }
 205     }
 206     if (a.e < b.e) {
 207         return true;
 208     } else if (a.e > b.e) {
 209         return false;
 210     }
 211     return false;
 212 }
 213
 214 static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a,
 215                                           MemoryRegionIoeventfd b)
 216 {
 217     return !memory_region_ioeventfd_before(a, b)
 218         && !memory_region_ioeventfd_before(b, a);
 219 }
 220
 221 typedef struct FlatRange FlatRange;
 222 typedef struct FlatView FlatView;
 223
 224 /* Range of memory in the global map.  Addresses are absolute. */
 225 struct FlatRange {
 226     MemoryRegion *mr;
 227     hwaddr offset_in_region;
 228     AddrRange addr;
 229     uint8_t dirty_log_mask;
 230     bool romd_mode;
 231     bool readonly;
 232 };
 233
 234 /* Flattened global view of current active memory hierarchy.  Kept in sorted
 235  * order.
 236  */
 237 struct FlatView {
 238     struct rcu_head rcu;
 239     unsigned ref;
 240     FlatRange *ranges;
 241     unsigned nr;
 242     unsigned nr_allocated;
 243 };
 244
 245 typedef struct AddressSpaceOps AddressSpaceOps;
 246
 247 #define FOR_EACH_FLAT_RANGE(var, view)          \
 248     for (var = (view)->ranges; var < (view)->ranges + (view)->nr; ++var)
 249
 250 static bool flatrange_equal(FlatRange *a, FlatRange *b)
 251 {
 252     return a->mr == b->mr
 253         && addrrange_equal(a->addr, b->addr)
 254         && a->offset_in_region == b->offset_in_region
 255         && a->romd_mode == b->romd_mode
 256         && a->readonly == b->readonly;
 257 }
 258
 259 static void flatview_init(FlatView *view)
 260 {
 261     view->ref = 1;
 262     view->ranges = NULL;
 263     view->nr = 0;
 264     view->nr_allocated = 0;
 265 }
 266
 267 /* Insert a range into a given position.  Caller is responsible for maintaining
 268  * sorting order.
 269  */
 270 static void flatview_insert(FlatView *view, unsigned pos, FlatRange *range)
 271 {
 272     if (view->nr == view->nr_allocated) {
 273         view->nr_allocated = MAX(2 * view->nr, 10);
 274         view->ranges = g_realloc(view->ranges,
 275                                     view->nr_allocated * sizeof(*view->ranges));
 276     }
 277     memmove(view->ranges + pos + 1, view->ranges + pos,
 278             (view->nr - pos) * sizeof(FlatRange));
 279     view->ranges[pos] = *range;
 280     memory_region_ref(range->mr);
 281     ++view->nr;
 282 }
 283
 284 static void flatview_destroy(FlatView *view)
 285 {
 286     int i;
 287
 288     for (i = 0; i < view->nr; i++) {
 289         memory_region_unref(view->ranges[i].mr);
 290     }
 291     g_free(view->ranges);
 292     g_free(view);
 293 }
 294
 295 static void flatview_ref(FlatView *view)
 296 {
 297     atomic_inc(&view->ref);
 298 }
 299
 300 static void flatview_unref(FlatView *view)
 301 {
 302     if (atomic_fetch_dec(&view->ref) == 1) {
 303         flatview_destroy(view);
 304     }
 305 }
 306
 307 static bool can_merge(FlatRange *r1, FlatRange *r2)
 308 {
 309     return int128_eq(addrrange_end(r1->addr), r2->addr.start)
 310         && r1->mr == r2->mr
 311         && int128_eq(int128_add(int128_make64(r1->offset_in_region),
 312                                 r1->addr.size),
 313                      int128_make64(r2->offset_in_region))
 314         && r1->dirty_log_mask == r2->dirty_log_mask
 315         && r1->romd_mode == r2->romd_mode
 316         && r1->readonly == r2->readonly;
 317 }
 318
 319 /* Attempt to simplify a view by merging adjacent ranges */
 320 static void flatview_simplify(FlatView *view)
 321 {
 322     unsigned i, j;
 323
 324     i = 0;
 325     while (i < view->nr) {
 326         j = i + 1;
 327         while (j < view->nr
 328                && can_merge(&view->ranges[j-1], &view->ranges[j])) {
 329             int128_addto(&view->ranges[i].addr.size, view->ranges[j].addr.size);
 330             ++j;
 331         }
 332         ++i;
 333         memmove(&view->ranges[i], &view->ranges[j],
 334                 (view->nr - j) * sizeof(view->ranges[j]));
 335         view->nr -= j - i;
 336     }
 337 }
 338
 339 static bool memory_region_big_endian(MemoryRegion *mr)
 340 {
 341 #ifdef TARGET_WORDS_BIGENDIAN
 342     return mr->ops->endianness != DEVICE_LITTLE_ENDIAN;
 343 #else
 344     return mr->ops->endianness == DEVICE_BIG_ENDIAN;
 345 #endif
 346 }
 347
 348 static bool memory_region_wrong_endianness(MemoryRegion *mr)
 349 {
 350 #ifdef TARGET_WORDS_BIGENDIAN
 351     return mr->ops->endianness == DEVICE_LITTLE_ENDIAN;
 352 #else
 353     return mr->ops->endianness == DEVICE_BIG_ENDIAN;
 354 #endif
 355 }
 356
 357 static void adjust_endianness(MemoryRegion *mr, uint64_t *data, unsigned size)
 358 {
 359     if (memory_region_wrong_endianness(mr)) {
 360         switch (size) {
 361         case 1:
 362             break;
 363         case 2:
 364             *data = bswap16(*data);
 365             break;
 366         case 4:
 367             *data = bswap32(*data);
 368             break;
 369         case 8:
 370             *data = bswap64(*data);
 371             break;
 372         default:
 373             abort();
 374         }
 375     }
 376 }
 377
 378 static hwaddr memory_region_to_absolute_addr(MemoryRegion *mr, hwaddr offset)
 379 {
 380     MemoryRegion *root;
 381     hwaddr abs_addr = offset;
 382
 383     abs_addr += mr->addr;
 384     for (root = mr; root->container; ) {
 385         root = root->container;
 386         abs_addr += root->addr;
 387     }
 388
 389     return abs_addr;
 390 }
 391
 392 static int get_cpu_index(void)
 393 {
 394     if (current_cpu) {
 395         return current_cpu->cpu_index;
 396     }
 397     return -1;
 398 }
 399
 400 static MemTxResult memory_region_oldmmio_read_accessor(MemoryRegion *mr,
 401                                                        hwaddr addr,
 402                                                        uint64_t *value,
 403                                                        unsigned size,
 404                                                        unsigned shift,
 405                                                        uint64_t mask,
 406                                                        MemTxAttrs attrs)
 407 {
 408     uint64_t tmp;
 409
 410     tmp = mr->ops->old_mmio.read[ctz32(size)](mr->opaque, addr);
 411     if (mr->subpage) {
 412         trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
 413     } else if (mr == &io_mem_notdirty) {
 414         /* Accesses to code which has previously been translated into a TB show
 415          * up in the MMIO path, as accesses to the io_mem_notdirty
 416          * MemoryRegion. */
 417         trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
 418     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 419         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 420         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
 421     }
 422     *value |= (tmp & mask) << shift;
 423     return MEMTX_OK;
 424 }
 425
 426 static MemTxResult  memory_region_read_accessor(MemoryRegion *mr,
 427                                                 hwaddr addr,
 428                                                 uint64_t *value,
 429                                                 unsigned size,
 430                                                 unsigned shift,
 431                                                 uint64_t mask,
 432                                                 MemTxAttrs attrs)
 433 {
 434     uint64_t tmp;
 435
 436     tmp = mr->ops->read(mr->opaque, addr, size);
 437     if (mr->subpage) {
 438         trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
 439     } else if (mr == &io_mem_notdirty) {
 440         /* Accesses to code which has previously been translated into a TB show
 441          * up in the MMIO path, as accesses to the io_mem_notdirty
 442          * MemoryRegion. */
 443         trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
 444     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 445         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 446         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
 447     }
 448     *value |= (tmp & mask) << shift;
 449     return MEMTX_OK;
 450 }
 451
 452 static MemTxResult memory_region_read_with_attrs_accessor(MemoryRegion *mr,
 453                                                           hwaddr addr,
 454                                                           uint64_t *value,
 455                                                           unsigned size,
 456                                                           unsigned shift,
 457                                                           uint64_t mask,
 458                                                           MemTxAttrs attrs)
 459 {
 460     uint64_t tmp = 0;
 461     MemTxResult r;
 462
 463     r = mr->ops->read_with_attrs(mr->opaque, addr, &tmp, size, attrs);
 464     if (mr->subpage) {
 465         trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
 466     } else if (mr == &io_mem_notdirty) {
 467         /* Accesses to code which has previously been translated into a TB show
 468          * up in the MMIO path, as accesses to the io_mem_notdirty
 469          * MemoryRegion. */
 470         trace_memory_region_tb_read(get_cpu_index(), addr, tmp, size);
 471     } else if (TRACE_MEMORY_REGION_OPS_READ_ENABLED) {
 472         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 473         trace_memory_region_ops_read(get_cpu_index(), mr, abs_addr, tmp, size);
 474     }
 475     *value |= (tmp & mask) << shift;
 476     return r;
 477 }
 478
 479 static MemTxResult memory_region_oldmmio_write_accessor(MemoryRegion *mr,
 480                                                         hwaddr addr,
 481                                                         uint64_t *value,
 482                                                         unsigned size,
 483                                                         unsigned shift,
 484                                                         uint64_t mask,
 485                                                         MemTxAttrs attrs)
 486 {
 487     uint64_t tmp;
 488
 489     tmp = (*value >> shift) & mask;
 490     if (mr->subpage) {
 491         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
 492     } else if (mr == &io_mem_notdirty) {
 493         /* Accesses to code which has previously been translated into a TB show
 494          * up in the MMIO path, as accesses to the io_mem_notdirty
 495          * MemoryRegion. */
 496         trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
 497     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 498         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 499         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
 500     }
 501     mr->ops->old_mmio.write[ctz32(size)](mr->opaque, addr, tmp);
 502     return MEMTX_OK;
 503 }
 504
 505 static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
 506                                                 hwaddr addr,
 507                                                 uint64_t *value,
 508                                                 unsigned size,
 509                                                 unsigned shift,
 510                                                 uint64_t mask,
 511                                                 MemTxAttrs attrs)
 512 {
 513     uint64_t tmp;
 514
 515     tmp = (*value >> shift) & mask;
 516     if (mr->subpage) {
 517         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
 518     } else if (mr == &io_mem_notdirty) {
 519         /* Accesses to code which has previously been translated into a TB show
 520          * up in the MMIO path, as accesses to the io_mem_notdirty
 521          * MemoryRegion. */
 522         trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
 523     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 524         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 525         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
 526     }
 527     mr->ops->write(mr->opaque, addr, tmp, size);
 528     return MEMTX_OK;
 529 }
 530
 531 static MemTxResult memory_region_write_with_attrs_accessor(MemoryRegion *mr,
 532                                                            hwaddr addr,
 533                                                            uint64_t *value,
 534                                                            unsigned size,
 535                                                            unsigned shift,
 536                                                            uint64_t mask,
 537                                                            MemTxAttrs attrs)
 538 {
 539     uint64_t tmp;
 540
 541     tmp = (*value >> shift) & mask;
 542     if (mr->subpage) {
 543         trace_memory_region_subpage_write(get_cpu_index(), mr, addr, tmp, size);
 544     } else if (mr == &io_mem_notdirty) {
 545         /* Accesses to code which has previously been translated into a TB show
 546          * up in the MMIO path, as accesses to the io_mem_notdirty
 547          * MemoryRegion. */
 548         trace_memory_region_tb_write(get_cpu_index(), addr, tmp, size);
 549     } else if (TRACE_MEMORY_REGION_OPS_WRITE_ENABLED) {
 550         hwaddr abs_addr = memory_region_to_absolute_addr(mr, addr);
 551         trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size);
 552     }
 553     return mr->ops->write_with_attrs(mr->opaque, addr, tmp, size, attrs);
 554 }
 555
 556 static MemTxResult access_with_adjusted_size(hwaddr addr,
 557                                       uint64_t *value,
 558                                       unsigned size,
 559                                       unsigned access_size_min,
 560                                       unsigned access_size_max,
 561                                       MemTxResult (*access)(MemoryRegion *mr,
 562                                                             hwaddr addr,
 563                                                             uint64_t *value,
 564                                                             unsigned size,
 565                                                             unsigned shift,
 566                                                             uint64_t mask,
 567                                                             MemTxAttrs attrs),
 568                                       MemoryRegion *mr,
 569                                       MemTxAttrs attrs)
 570 {
 571     uint64_t access_mask;
 572     unsigned access_size;
 573     unsigned i;
 574     MemTxResult r = MEMTX_OK;
 575
 576     if (!access_size_min) {
 577         access_size_min = 1;
 578     }
 579     if (!access_size_max) {
 580         access_size_max = 4;
 581     }
 582
 583     /* FIXME: support unaligned access? */
 584     access_size = MAX(MIN(size, access_size_max), access_size_min);
 585     access_mask = -1ULL >> (64 - access_size * 8);
 586     if (memory_region_big_endian(mr)) {
 587         for (i = 0; i < size; i += access_size) {
 588             r |= access(mr, addr + i, value, access_size,
 589                         (size - access_size - i) * 8, access_mask, attrs);
 590         }
 591     } else {
 592         for (i = 0; i < size; i += access_size) {
 593             r |= access(mr, addr + i, value, access_size, i * 8,
 594                         access_mask, attrs);
 595         }
 596     }
 597     return r;
 598 }
 599
 600 static AddressSpace *memory_region_to_address_space(MemoryRegion *mr)
 601 {
 602     AddressSpace *as;
 603
 604     while (mr->container) {
 605         mr = mr->container;
 606     }
 607     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 608         if (mr == as->root) {
 609             return as;
 610         }
 611     }
 612     return NULL;
 613 }
 614
 615 /* Render a memory region into the global view.  Ranges in @view obscure
 616  * ranges in @mr.
 617  */
 618 static void render_memory_region(FlatView *view,
 619                                  MemoryRegion *mr,
 620                                  Int128 base,
 621                                  AddrRange clip,
 622                                  bool readonly)
 623 {
 624     MemoryRegion *subregion;
 625     unsigned i;
 626     hwaddr offset_in_region;
 627     Int128 remain;
 628     Int128 now;
 629     FlatRange fr;
 630     AddrRange tmp;
 631
 632     if (!mr->enabled) {
 633         return;
 634     }
 635
 636     int128_addto(&base, int128_make64(mr->addr));
 637     readonly |= mr->readonly;
 638
 639     tmp = addrrange_make(base, mr->size);
 640
 641     if (!addrrange_intersects(tmp, clip)) {
 642         return;
 643     }
 644
 645     clip = addrrange_intersection(tmp, clip);
 646
 647     if (mr->alias) {
 648         int128_subfrom(&base, int128_make64(mr->alias->addr));
 649         int128_subfrom(&base, int128_make64(mr->alias_offset));
 650         render_memory_region(view, mr->alias, base, clip, readonly);
 651         return;
 652     }
 653
 654     /* Render subregions in priority order. */
 655     QTAILQ_FOREACH(subregion, &mr->subregions, subregions_link) {
 656         render_memory_region(view, subregion, base, clip, readonly);
 657     }
 658
 659     if (!mr->terminates) {
 660         return;
 661     }
 662
 663     offset_in_region = int128_get64(int128_sub(clip.start, base));
 664     base = clip.start;
 665     remain = clip.size;
 666
 667     fr.mr = mr;
 668     fr.dirty_log_mask = memory_region_get_dirty_log_mask(mr);
 669     fr.romd_mode = mr->romd_mode;
 670     fr.readonly = readonly;
 671
 672     /* Render the region itself into any gaps left by the current view. */
 673     for (i = 0; i < view->nr && int128_nz(remain); ++i) {
 674         if (int128_ge(base, addrrange_end(view->ranges[i].addr))) {
 675             continue;
 676         }
 677         if (int128_lt(base, view->ranges[i].addr.start)) {
 678             now = int128_min(remain,
 679                              int128_sub(view->ranges[i].addr.start, base));
 680             fr.offset_in_region = offset_in_region;
 681             fr.addr = addrrange_make(base, now);
 682             flatview_insert(view, i, &fr);
 683             ++i;
 684             int128_addto(&base, now);
 685             offset_in_region += int128_get64(now);
 686             int128_subfrom(&remain, now);
 687         }
 688         now = int128_sub(int128_min(int128_add(base, remain),
 689                                     addrrange_end(view->ranges[i].addr)),
 690                          base);
 691         int128_addto(&base, now);
 692         offset_in_region += int128_get64(now);
 693         int128_subfrom(&remain, now);
 694     }
 695     if (int128_nz(remain)) {
 696         fr.offset_in_region = offset_in_region;
 697         fr.addr = addrrange_make(base, remain);
 698         flatview_insert(view, i, &fr);
 699     }
 700 }
 701
 702 /* Render a memory topology into a list of disjoint absolute ranges. */
 703 static FlatView *generate_memory_topology(MemoryRegion *mr)
 704 {
 705     FlatView *view;
 706
 707     view = g_new(FlatView, 1);
 708     flatview_init(view);
 709
 710     if (mr) {
 711         render_memory_region(view, mr, int128_zero(),
 712                              addrrange_make(int128_zero(), int128_2_64()), false);
 713     }
 714     flatview_simplify(view);
 715
 716     return view;
 717 }
 718
 719 static void address_space_add_del_ioeventfds(AddressSpace *as,
 720                                              MemoryRegionIoeventfd *fds_new,
 721                                              unsigned fds_new_nb,
 722                                              MemoryRegionIoeventfd *fds_old,
 723                                              unsigned fds_old_nb)
 724 {
 725     unsigned iold, inew;
 726     MemoryRegionIoeventfd *fd;
 727     MemoryRegionSection section;
 728
 729     /* Generate a symmetric difference of the old and new fd sets, adding
 730      * and deleting as necessary.
 731      */
 732
 733     iold = inew = 0;
 734     while (iold < fds_old_nb || inew < fds_new_nb) {
 735         if (iold < fds_old_nb
 736             && (inew == fds_new_nb
 737                 || memory_region_ioeventfd_before(fds_old[iold],
 738                                                   fds_new[inew]))) {
 739             fd = &fds_old[iold];
 740             section = (MemoryRegionSection) {
 741                 .address_space = as,
 742                 .offset_within_address_space = int128_get64(fd->addr.start),
 743                 .size = fd->addr.size,
 744             };
 745             MEMORY_LISTENER_CALL(eventfd_del, Forward, &section,
 746                                  fd->match_data, fd->data, fd->e);
 747             ++iold;
 748         } else if (inew < fds_new_nb
 749                    && (iold == fds_old_nb
 750                        || memory_region_ioeventfd_before(fds_new[inew],
 751                                                          fds_old[iold]))) {
 752             fd = &fds_new[inew];
 753             section = (MemoryRegionSection) {
 754                 .address_space = as,
 755                 .offset_within_address_space = int128_get64(fd->addr.start),
 756                 .size = fd->addr.size,
 757             };
 758             MEMORY_LISTENER_CALL(eventfd_add, Reverse, &section,
 759                                  fd->match_data, fd->data, fd->e);
 760             ++inew;
 761         } else {
 762             ++iold;
 763             ++inew;
 764         }
 765     }
 766 }
 767
 768 static FlatView *address_space_get_flatview(AddressSpace *as)
 769 {
 770     FlatView *view;
 771
 772     rcu_read_lock();
 773     view = atomic_rcu_read(&as->current_map);
 774     flatview_ref(view);
 775     rcu_read_unlock();
 776     return view;
 777 }
 778
 779 static void address_space_update_ioeventfds(AddressSpace *as)
 780 {
 781     FlatView *view;
 782     FlatRange *fr;
 783     unsigned ioeventfd_nb = 0;
 784     MemoryRegionIoeventfd *ioeventfds = NULL;
 785     AddrRange tmp;
 786     unsigned i;
 787
 788     view = address_space_get_flatview(as);
 789     FOR_EACH_FLAT_RANGE(fr, view) {
 790         for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
 791             tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
 792                                   int128_sub(fr->addr.start,
 793                                              int128_make64(fr->offset_in_region)));
 794             if (addrrange_intersects(fr->addr, tmp)) {
 795                 ++ioeventfd_nb;
 796                 ioeventfds = g_realloc(ioeventfds,
 797                                           ioeventfd_nb * sizeof(*ioeventfds));
 798                 ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
 799                 ioeventfds[ioeventfd_nb-1].addr = tmp;
 800             }
 801         }
 802     }
 803
 804     address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
 805                                      as->ioeventfds, as->ioeventfd_nb);
 806
 807     g_free(as->ioeventfds);
 808     as->ioeventfds = ioeventfds;
 809     as->ioeventfd_nb = ioeventfd_nb;
 810     flatview_unref(view);
 811 }
 812
 813 static void address_space_update_topology_pass(AddressSpace *as,
 814                                                const FlatView *old_view,
 815                                                const FlatView *new_view,
 816                                                bool adding)
 817 {
 818     unsigned iold, inew;
 819     FlatRange *frold, *frnew;
 820
 821     /* Generate a symmetric difference of the old and new memory maps.
 822      * Kill ranges in the old map, and instantiate ranges in the new map.
 823      */
 824     iold = inew = 0;
 825     while (iold < old_view->nr || inew < new_view->nr) {
 826         if (iold < old_view->nr) {
 827             frold = &old_view->ranges[iold];
 828         } else {
 829             frold = NULL;
 830         }
 831         if (inew < new_view->nr) {
 832             frnew = &new_view->ranges[inew];
 833         } else {
 834             frnew = NULL;
 835         }
 836
 837         if (frold
 838             && (!frnew
 839                 || int128_lt(frold->addr.start, frnew->addr.start)
 840                 || (int128_eq(frold->addr.start, frnew->addr.start)
 841                     && !flatrange_equal(frold, frnew)))) {
 842             /* In old but not in new, or in both but attributes changed. */
 843
 844             if (!adding) {
 845                 MEMORY_LISTENER_UPDATE_REGION(frold, as, Reverse, region_del);
 846             }
 847
 848             ++iold;
 849         } else if (frold && frnew && flatrange_equal(frold, frnew)) {
 850             /* In both and unchanged (except logging may have changed) */
 851
 852             if (adding) {
 853                 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_nop);
 854                 if (frnew->dirty_log_mask & ~frold->dirty_log_mask) {
 855                     MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, log_start,
 856                                                   frold->dirty_log_mask,
 857                                                   frnew->dirty_log_mask);
 858                 }
 859                 if (frold->dirty_log_mask & ~frnew->dirty_log_mask) {
 860                     MEMORY_LISTENER_UPDATE_REGION(frnew, as, Reverse, log_stop,
 861                                                   frold->dirty_log_mask,
 862                                                   frnew->dirty_log_mask);
 863                 }
 864             }
 865
 866             ++iold;
 867             ++inew;
 868         } else {
 869             /* In new */
 870
 871             if (adding) {
 872                 MEMORY_LISTENER_UPDATE_REGION(frnew, as, Forward, region_add);
 873             }
 874
 875             ++inew;
 876         }
 877     }
 878 }
 879
 880
 881 static void address_space_update_topology(AddressSpace *as)
 882 {
 883     FlatView *old_view = address_space_get_flatview(as);
 884     FlatView *new_view = generate_memory_topology(as->root);
 885
 886     address_space_update_topology_pass(as, old_view, new_view, false);
 887     address_space_update_topology_pass(as, old_view, new_view, true);
 888
 889     /* Writes are protected by the BQL.  */
 890     atomic_rcu_set(&as->current_map, new_view);
 891     call_rcu(old_view, flatview_unref, rcu);
 892
 893     /* Note that all the old MemoryRegions are still alive up to this
 894      * point.  This relieves most MemoryListeners from the need to
 895      * ref/unref the MemoryRegions they get---unless they use them
 896      * outside the iothread mutex, in which case precise reference
 897      * counting is necessary.
 898      */
 899     flatview_unref(old_view);
 900
 901     address_space_update_ioeventfds(as);
 902 }
 903
 904 void memory_region_transaction_begin(void)
 905 {
 906     qemu_flush_coalesced_mmio_buffer();
 907     ++memory_region_transaction_depth;
 908 }
 909
 910 static void memory_region_clear_pending(void)
 911 {
 912     memory_region_update_pending = false;
 913     ioeventfd_update_pending = false;
 914 }
 915
 916 void memory_region_transaction_commit(void)
 917 {
 918     AddressSpace *as;
 919
 920     assert(memory_region_transaction_depth);
 921     --memory_region_transaction_depth;
 922     if (!memory_region_transaction_depth) {
 923         if (memory_region_update_pending) {
 924             MEMORY_LISTENER_CALL_GLOBAL(begin, Forward);
 925
 926             QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 927                 address_space_update_topology(as);
 928             }
 929
 930             MEMORY_LISTENER_CALL_GLOBAL(commit, Forward);
 931         } else if (ioeventfd_update_pending) {
 932             QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
 933                 address_space_update_ioeventfds(as);
 934             }
 935         }
 936         memory_region_clear_pending();
 937    }
 938 }
 939
 940 static void memory_region_destructor_none(MemoryRegion *mr)
 941 {
 942 }
 943
 944 static void memory_region_destructor_ram(MemoryRegion *mr)
 945 {
 946     qemu_ram_free(mr->ram_block);
 947 }
 948
 949 static void memory_region_destructor_rom_device(MemoryRegion *mr)
 950 {
 951     qemu_ram_free(mr->ram_block);
 952 }
 953
 954 static bool memory_region_need_escape(char c)
 955 {
 956     return c == '/' || c == '[' || c == '\\' || c == ']';
 957 }
 958
 959 static char *memory_region_escape_name(const char *name)
 960 {
 961     const char *p;
 962     char *escaped, *q;
 963     uint8_t c;
 964     size_t bytes = 0;
 965
 966     for (p = name; *p; p++) {
 967         bytes += memory_region_need_escape(*p) ? 4 : 1;
 968     }
 969     if (bytes == p - name) {
 970        return g_memdup(name, bytes + 1);
 971     }
 972
 973     escaped = g_malloc(bytes + 1);
 974     for (p = name, q = escaped; *p; p++) {
 975         c = *p;
 976         if (unlikely(memory_region_need_escape(c))) {
 977             *q++ = '\\';
 978             *q++ = 'x';
 979             *q++ = "0123456789abcdef"[c >> 4];
 980             c = "0123456789abcdef"[c & 15];
 981         }
 982         *q++ = c;
 983     }
 984     *q = 0;
 985     return escaped;
 986 }
 987
 988 void memory_region_init(MemoryRegion *mr,
 989                         Object *owner,
 990                         const char *name,
 991                         uint64_t size)
 992 {
 993     object_initialize(mr, sizeof(*mr), TYPE_MEMORY_REGION);
 994     mr->size = int128_make64(size);
 995     if (size == UINT64_MAX) {
 996         mr->size = int128_2_64();
 997     }
 998     mr->name = g_strdup(name);
 999     mr->owner = owner;
1000     mr->ram_block = NULL;
1001
1002     if (name) {
1003         char *escaped_name = memory_region_escape_name(name);
1004         char *name_array = g_strdup_printf("%s[*]", escaped_name);
1005
1006         if (!owner) {
1007             owner = container_get(qdev_get_machine(), "/unattached");
1008         }
1009
1010         object_property_add_child(owner, name_array, OBJECT(mr), &error_abort);
1011         object_unref(OBJECT(mr));
1012         g_free(name_array);
1013         g_free(escaped_name);
1014     }
1015 }
1016
1017 static void memory_region_get_addr(Object *obj, Visitor *v, const char *name,
1018                                    void *opaque, Error **errp)
1019 {
1020     MemoryRegion *mr = MEMORY_REGION(obj);
1021     uint64_t value = mr->addr;
1022
1023     visit_type_uint64(v, name, &value, errp);
1024 }
1025
1026 static void memory_region_get_container(Object *obj, Visitor *v,
1027                                         const char *name, void *opaque,
1028                                         Error **errp)
1029 {
1030     MemoryRegion *mr = MEMORY_REGION(obj);
1031     gchar *path = (gchar *)"";
1032
1033     if (mr->container) {
1034         path = object_get_canonical_path(OBJECT(mr->container));
1035     }
1036     visit_type_str(v, name, &path, errp);
1037     if (mr->container) {
1038         g_free(path);
1039     }
1040 }
1041
1042 static Object *memory_region_resolve_container(Object *obj, void *opaque,
1043                                                const char *part)
1044 {
1045     MemoryRegion *mr = MEMORY_REGION(obj);
1046
1047     return OBJECT(mr->container);
1048 }
1049
1050 static void memory_region_get_priority(Object *obj, Visitor *v,
1051                                        const char *name, void *opaque,
1052                                        Error **errp)
1053 {
1054     MemoryRegion *mr = MEMORY_REGION(obj);
1055     int32_t value = mr->priority;
1056
1057     visit_type_int32(v, name, &value, errp);
1058 }
1059
1060 static bool memory_region_get_may_overlap(Object *obj, Error **errp)
1061 {
1062     MemoryRegion *mr = MEMORY_REGION(obj);
1063
1064     return mr->may_overlap;
1065 }
1066
1067 static void memory_region_get_size(Object *obj, Visitor *v, const char *name,
1068                                    void *opaque, Error **errp)
1069 {
1070     MemoryRegion *mr = MEMORY_REGION(obj);
1071     uint64_t value = memory_region_size(mr);
1072
1073     visit_type_uint64(v, name, &value, errp);
1074 }
1075
1076 static void memory_region_initfn(Object *obj)
1077 {
1078     MemoryRegion *mr = MEMORY_REGION(obj);
1079     ObjectProperty *op;
1080
1081     mr->ops = &unassigned_mem_ops;
1082     mr->enabled = true;
1083     mr->romd_mode = true;
1084     mr->global_locking = true;
1085     mr->destructor = memory_region_destructor_none;
1086     QTAILQ_INIT(&mr->subregions);
1087     QTAILQ_INIT(&mr->coalesced);
1088
1089     op = object_property_add(OBJECT(mr), "container",
1090                              "link<" TYPE_MEMORY_REGION ">",
1091                              memory_region_get_container,
1092                              NULL, /* memory_region_set_container */
1093                              NULL, NULL, &error_abort);
1094     op->resolve = memory_region_resolve_container;
1095
1096     object_property_add(OBJECT(mr), "addr", "uint64",
1097                         memory_region_get_addr,
1098                         NULL, /* memory_region_set_addr */
1099                         NULL, NULL, &error_abort);
1100     object_property_add(OBJECT(mr), "priority", "uint32",
1101                         memory_region_get_priority,
1102                         NULL, /* memory_region_set_priority */
1103                         NULL, NULL, &error_abort);
1104     object_property_add_bool(OBJECT(mr), "may-overlap",
1105                              memory_region_get_may_overlap,
1106                              NULL, /* memory_region_set_may_overlap */
1107                              &error_abort);
1108     object_property_add(OBJECT(mr), "size", "uint64",
1109                         memory_region_get_size,
1110                         NULL, /* memory_region_set_size, */
1111                         NULL, NULL, &error_abort);
1112 }
1113
1114 static uint64_t unassigned_mem_read(void *opaque, hwaddr addr,
1115                                     unsigned size)
1116 {
1117 #ifdef DEBUG_UNASSIGNED
1118     printf("Unassigned mem read " TARGET_FMT_plx "\n", addr);
1119 #endif
1120     if (current_cpu != NULL) {
1121         cpu_unassigned_access(current_cpu, addr, false, false, 0, size);
1122     }
1123     return 0;
1124 }
1125
1126 static void unassigned_mem_write(void *opaque, hwaddr addr,
1127                                  uint64_t val, unsigned size)
1128 {
1129 #ifdef DEBUG_UNASSIGNED
1130     printf("Unassigned mem write " TARGET_FMT_plx " = 0x%"PRIx64"\n", addr, val);
1131 #endif
1132     if (current_cpu != NULL) {
1133         cpu_unassigned_access(current_cpu, addr, true, false, 0, size);
1134     }
1135 }
1136
1137 static bool unassigned_mem_accepts(void *opaque, hwaddr addr,
1138                                    unsigned size, bool is_write)
1139 {
1140     return false;
1141 }
1142
1143 const MemoryRegionOps unassigned_mem_ops = {
1144     .valid.accepts = unassigned_mem_accepts,
1145     .endianness = DEVICE_NATIVE_ENDIAN,
1146 };
1147
1148 bool memory_region_access_valid(MemoryRegion *mr,
1149                                 hwaddr addr,
1150                                 unsigned size,
1151                                 bool is_write)
1152 {
1153     int access_size_min, access_size_max;
1154     int access_size, i;
1155
1156     if (!mr->ops->valid.unaligned && (addr & (size - 1))) {
1157         return false;
1158     }
1159
1160     if (!mr->ops->valid.accepts) {
1161         return true;
1162     }
1163
1164     access_size_min = mr->ops->valid.min_access_size;
1165     if (!mr->ops->valid.min_access_size) {
1166         access_size_min = 1;
1167     }
1168
1169     access_size_max = mr->ops->valid.max_access_size;
1170     if (!mr->ops->valid.max_access_size) {
1171         access_size_max = 4;
1172     }
1173
1174     access_size = MAX(MIN(size, access_size_max), access_size_min);
1175     for (i = 0; i < size; i += access_size) {
1176         if (!mr->ops->valid.accepts(mr->opaque, addr + i, access_size,
1177                                     is_write)) {
1178             return false;
1179         }
1180     }
1181
1182     return true;
1183 }
1184
1185 static MemTxResult memory_region_dispatch_read1(MemoryRegion *mr,
1186                                                 hwaddr addr,
1187                                                 uint64_t *pval,
1188                                                 unsigned size,
1189                                                 MemTxAttrs attrs)
1190 {
1191     *pval = 0;
1192
1193     if (mr->ops->read) {
1194         return access_with_adjusted_size(addr, pval, size,
1195                                          mr->ops->impl.min_access_size,
1196                                          mr->ops->impl.max_access_size,
1197                                          memory_region_read_accessor,
1198                                          mr, attrs);
1199     } else if (mr->ops->read_with_attrs) {
1200         return access_with_adjusted_size(addr, pval, size,
1201                                          mr->ops->impl.min_access_size,
1202                                          mr->ops->impl.max_access_size,
1203                                          memory_region_read_with_attrs_accessor,
1204                                          mr, attrs);
1205     } else {
1206         return access_with_adjusted_size(addr, pval, size, 1, 4,
1207                                          memory_region_oldmmio_read_accessor,
1208                                          mr, attrs);
1209     }
1210 }
1211
1212 MemTxResult memory_region_dispatch_read(MemoryRegion *mr,
1213                                         hwaddr addr,
1214                                         uint64_t *pval,
1215                                         unsigned size,
1216                                         MemTxAttrs attrs)
1217 {
1218     MemTxResult r;
1219
1220     if (!memory_region_access_valid(mr, addr, size, false)) {
1221         *pval = unassigned_mem_read(mr, addr, size);
1222         return MEMTX_DECODE_ERROR;
1223     }
1224
1225     r = memory_region_dispatch_read1(mr, addr, pval, size, attrs);
1226     adjust_endianness(mr, pval, size);
1227     return r;
1228 }
1229
1230 /* Return true if an eventfd was signalled */
1231 static bool memory_region_dispatch_write_eventfds(MemoryRegion *mr,
1232                                                     hwaddr addr,
1233                                                     uint64_t data,
1234                                                     unsigned size,
1235                                                     MemTxAttrs attrs)
1236 {
1237     MemoryRegionIoeventfd ioeventfd = {
1238         .addr = addrrange_make(int128_make64(addr), int128_make64(size)),
1239         .data = data,
1240     };
1241     unsigned i;
1242
1243     for (i = 0; i < mr->ioeventfd_nb; i++) {
1244         ioeventfd.match_data = mr->ioeventfds[i].match_data;
1245         ioeventfd.e = mr->ioeventfds[i].e;
1246
1247         if (memory_region_ioeventfd_equal(ioeventfd, mr->ioeventfds[i])) {
1248             event_notifier_set(ioeventfd.e);
1249             return true;
1250         }
1251     }
1252
1253     return false;
1254 }
1255
1256 MemTxResult memory_region_dispatch_write(MemoryRegion *mr,
1257                                          hwaddr addr,
1258                                          uint64_t data,
1259                                          unsigned size,
1260                                          MemTxAttrs attrs)
1261 {
1262     if (!memory_region_access_valid(mr, addr, size, true)) {
1263         unassigned_mem_write(mr, addr, data, size);
1264         return MEMTX_DECODE_ERROR;
1265     }
1266
1267     adjust_endianness(mr, &data, size);
1268
1269     if ((!kvm_eventfds_enabled()) &&
1270         memory_region_dispatch_write_eventfds(mr, addr, data, size, attrs)) {
1271         return MEMTX_OK;
1272     }
1273
1274     if (mr->ops->write) {
1275         return access_with_adjusted_size(addr, &data, size,
1276                                          mr->ops->impl.min_access_size,
1277                                          mr->ops->impl.max_access_size,
1278                                          memory_region_write_accessor, mr,
1279                                          attrs);
1280     } else if (mr->ops->write_with_attrs) {
1281         return
1282             access_with_adjusted_size(addr, &data, size,
1283                                       mr->ops->impl.min_access_size,
1284                                       mr->ops->impl.max_access_size,
1285                                       memory_region_write_with_attrs_accessor,
1286                                       mr, attrs);
1287     } else {
1288         return access_with_adjusted_size(addr, &data, size, 1, 4,
1289                                          memory_region_oldmmio_write_accessor,
1290                                          mr, attrs);
1291     }
1292 }
1293
1294 void memory_region_init_io(MemoryRegion *mr,
1295                            Object *owner,
1296                            const MemoryRegionOps *ops,
1297                            void *opaque,
1298                            const char *name,
1299                            uint64_t size)
1300 {
1301     memory_region_init(mr, owner, name, size);
1302     mr->ops = ops ? ops : &unassigned_mem_ops;
1303     mr->opaque = opaque;
1304     mr->terminates = true;
1305 }
1306
1307 void memory_region_init_ram(MemoryRegion *mr,
1308                             Object *owner,
1309                             const char *name,
1310                             uint64_t size,
1311                             Error **errp)
1312 {
1313     memory_region_init(mr, owner, name, size);
1314     mr->ram = true;
1315     mr->terminates = true;
1316     mr->destructor = memory_region_destructor_ram;
1317     mr->ram_block = qemu_ram_alloc(size, mr, errp);
1318     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1319 }
1320
1321 void memory_region_init_resizeable_ram(MemoryRegion *mr,
1322                                        Object *owner,
1323                                        const char *name,
1324                                        uint64_t size,
1325                                        uint64_t max_size,
1326                                        void (*resized)(const char*,
1327                                                        uint64_t length,
1328                                                        void *host),
1329                                        Error **errp)
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->ram_block = qemu_ram_alloc_resizeable(size, max_size, resized,
1336                                               mr, errp);
1337     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1338 }
1339
1340 #ifdef __linux__
1341 void memory_region_init_ram_from_file(MemoryRegion *mr,
1342                                       struct Object *owner,
1343                                       const char *name,
1344                                       uint64_t size,
1345                                       bool share,
1346                                       const char *path,
1347                                       Error **errp)
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->ram_block = qemu_ram_alloc_from_file(size, mr, share, path, errp);
1354     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1355 }
1356 #endif
1357
1358 void memory_region_init_ram_ptr(MemoryRegion *mr,
1359                                 Object *owner,
1360                                 const char *name,
1361                                 uint64_t size,
1362                                 void *ptr)
1363 {
1364     memory_region_init(mr, owner, name, size);
1365     mr->ram = true;
1366     mr->terminates = true;
1367     mr->destructor = memory_region_destructor_ram;
1368     mr->dirty_log_mask = tcg_enabled() ? (1 << DIRTY_MEMORY_CODE) : 0;
1369
1370     /* qemu_ram_alloc_from_ptr cannot fail with ptr != NULL.  */
1371     assert(ptr != NULL);
1372     mr->ram_block = qemu_ram_alloc_from_ptr(size, ptr, mr, &error_fatal);
1373 }
1374
1375 void memory_region_set_skip_dump(MemoryRegion *mr)
1376 {
1377     mr->skip_dump = true;
1378 }
1379
1380 void memory_region_init_alias(MemoryRegion *mr,
1381                               Object *owner,
1382                               const char *name,
1383                               MemoryRegion *orig,
1384                               hwaddr offset,
1385                               uint64_t size)
1386 {
1387     memory_region_init(mr, owner, name, size);
1388     mr->alias = orig;
1389     mr->alias_offset = offset;
1390 }
1391
1392 void memory_region_init_rom_device(MemoryRegion *mr,
1393                                    Object *owner,
1394                                    const MemoryRegionOps *ops,
1395                                    void *opaque,
1396                                    const char *name,
1397                                    uint64_t size,
1398                                    Error **errp)
1399 {
1400     memory_region_init(mr, owner, name, size);
1401     mr->ops = ops;
1402     mr->opaque = opaque;
1403     mr->terminates = true;
1404     mr->rom_device = true;
1405     mr->destructor = memory_region_destructor_rom_device;
1406     mr->ram_block = qemu_ram_alloc(size, mr, errp);
1407 }
1408
1409 void memory_region_init_iommu(MemoryRegion *mr,
1410                               Object *owner,
1411                               const MemoryRegionIOMMUOps *ops,
1412                               const char *name,
1413                               uint64_t size)
1414 {
1415     memory_region_init(mr, owner, name, size);
1416     mr->iommu_ops = ops,
1417     mr->terminates = true;  /* then re-forwards */
1418     notifier_list_init(&mr->iommu_notify);
1419 }
1420
1421 static void memory_region_finalize(Object *obj)
1422 {
1423     MemoryRegion *mr = MEMORY_REGION(obj);
1424
1425     assert(!mr->container);
1426
1427     /* We know the region is not visible in any address space (it
1428      * does not have a container and cannot be a root either because
1429      * it has no references, so we can blindly clear mr->enabled.
1430      * memory_region_set_enabled instead could trigger a transaction
1431      * and cause an infinite loop.
1432      */
1433     mr->enabled = false;
1434     memory_region_transaction_begin();
1435     while (!QTAILQ_EMPTY(&mr->subregions)) {
1436         MemoryRegion *subregion = QTAILQ_FIRST(&mr->subregions);
1437         memory_region_del_subregion(mr, subregion);
1438     }
1439     memory_region_transaction_commit();
1440
1441     mr->destructor(mr);
1442     memory_region_clear_coalescing(mr);
1443     g_free((char *)mr->name);
1444     g_free(mr->ioeventfds);
1445 }
1446
1447 Object *memory_region_owner(MemoryRegion *mr)
1448 {
1449     Object *obj = OBJECT(mr);
1450     return obj->parent;
1451 }
1452
1453 void memory_region_ref(MemoryRegion *mr)
1454 {
1455     /* MMIO callbacks most likely will access data that belongs
1456      * to the owner, hence the need to ref/unref the owner whenever
1457      * the memory region is in use.
1458      *
1459      * The memory region is a child of its owner.  As long as the
1460      * owner doesn't call unparent itself on the memory region,
1461      * ref-ing the owner will also keep the memory region alive.
1462      * Memory regions without an owner are supposed to never go away;
1463      * we do not ref/unref them because it slows down DMA sensibly.
1464      */
1465     if (mr && mr->owner) {
1466         object_ref(mr->owner);
1467     }
1468 }
1469
1470 void memory_region_unref(MemoryRegion *mr)
1471 {
1472     if (mr && mr->owner) {
1473         object_unref(mr->owner);
1474     }
1475 }
1476
1477 uint64_t memory_region_size(MemoryRegion *mr)
1478 {
1479     if (int128_eq(mr->size, int128_2_64())) {
1480         return UINT64_MAX;
1481     }
1482     return int128_get64(mr->size);
1483 }
1484
1485 const char *memory_region_name(const MemoryRegion *mr)
1486 {
1487     if (!mr->name) {
1488         ((MemoryRegion *)mr)->name =
1489             object_get_canonical_path_component(OBJECT(mr));
1490     }
1491     return mr->name;
1492 }
1493
1494 bool memory_region_is_skip_dump(MemoryRegion *mr)
1495 {
1496     return mr->skip_dump;
1497 }
1498
1499 uint8_t memory_region_get_dirty_log_mask(MemoryRegion *mr)
1500 {
1501     uint8_t mask = mr->dirty_log_mask;
1502     if (global_dirty_log) {
1503         mask |= (1 << DIRTY_MEMORY_MIGRATION);
1504     }
1505     return mask;
1506 }
1507
1508 bool memory_region_is_logging(MemoryRegion *mr, uint8_t client)
1509 {
1510     return memory_region_get_dirty_log_mask(mr) & (1 << client);
1511 }
1512
1513 void memory_region_register_iommu_notifier(MemoryRegion *mr, Notifier *n)
1514 {
1515     notifier_list_add(&mr->iommu_notify, n);
1516 }
1517
1518 void memory_region_iommu_replay(MemoryRegion *mr, Notifier *n,
1519                                 hwaddr granularity, bool is_write)
1520 {
1521     hwaddr addr;
1522     IOMMUTLBEntry iotlb;
1523
1524     for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
1525         iotlb = mr->iommu_ops->translate(mr, addr, is_write);
1526         if (iotlb.perm != IOMMU_NONE) {
1527             n->notify(n, &iotlb);
1528         }
1529
1530         /* if (2^64 - MR size) < granularity, it's possible to get an
1531          * infinite loop here.  This should catch such a wraparound */
1532         if ((addr + granularity) < addr) {
1533             break;
1534         }
1535     }
1536 }
1537
1538 void memory_region_unregister_iommu_notifier(Notifier *n)
1539 {
1540     notifier_remove(n);
1541 }
1542
1543 void memory_region_notify_iommu(MemoryRegion *mr,
1544                                 IOMMUTLBEntry entry)
1545 {
1546     assert(memory_region_is_iommu(mr));
1547     notifier_list_notify(&mr->iommu_notify, &entry);
1548 }
1549
1550 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
1551 {
1552     uint8_t mask = 1 << client;
1553     uint8_t old_logging;
1554
1555     assert(client == DIRTY_MEMORY_VGA);
1556     old_logging = mr->vga_logging_count;
1557     mr->vga_logging_count += log ? 1 : -1;
1558     if (!!old_logging == !!mr->vga_logging_count) {
1559         return;
1560     }
1561
1562     memory_region_transaction_begin();
1563     mr->dirty_log_mask = (mr->dirty_log_mask & ~mask) | (log * mask);
1564     memory_region_update_pending |= mr->enabled;
1565     memory_region_transaction_commit();
1566 }
1567
1568 bool memory_region_get_dirty(MemoryRegion *mr, hwaddr addr,
1569                              hwaddr size, unsigned client)
1570 {
1571     assert(mr->ram_block);
1572     return cpu_physical_memory_get_dirty(memory_region_get_ram_addr(mr) + addr,
1573                                          size, client);
1574 }
1575
1576 void memory_region_set_dirty(MemoryRegion *mr, hwaddr addr,
1577                              hwaddr size)
1578 {
1579     assert(mr->ram_block);
1580     cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr,
1581                                         size,
1582                                         memory_region_get_dirty_log_mask(mr));
1583 }
1584
1585 bool memory_region_test_and_clear_dirty(MemoryRegion *mr, hwaddr addr,
1586                                         hwaddr size, unsigned client)
1587 {
1588     assert(mr->ram_block);
1589     return cpu_physical_memory_test_and_clear_dirty(
1590                 memory_region_get_ram_addr(mr) + addr, size, client);
1591 }
1592
1593
1594 void memory_region_sync_dirty_bitmap(MemoryRegion *mr)
1595 {
1596     AddressSpace *as;
1597     FlatRange *fr;
1598
1599     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1600         FlatView *view = address_space_get_flatview(as);
1601         FOR_EACH_FLAT_RANGE(fr, view) {
1602             if (fr->mr == mr) {
1603                 MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
1604             }
1605         }
1606         flatview_unref(view);
1607     }
1608 }
1609
1610 void memory_region_set_readonly(MemoryRegion *mr, bool readonly)
1611 {
1612     if (mr->readonly != readonly) {
1613         memory_region_transaction_begin();
1614         mr->readonly = readonly;
1615         memory_region_update_pending |= mr->enabled;
1616         memory_region_transaction_commit();
1617     }
1618 }
1619
1620 void memory_region_rom_device_set_romd(MemoryRegion *mr, bool romd_mode)
1621 {
1622     if (mr->romd_mode != romd_mode) {
1623         memory_region_transaction_begin();
1624         mr->romd_mode = romd_mode;
1625         memory_region_update_pending |= mr->enabled;
1626         memory_region_transaction_commit();
1627     }
1628 }
1629
1630 void memory_region_reset_dirty(MemoryRegion *mr, hwaddr addr,
1631                                hwaddr size, unsigned client)
1632 {
1633     assert(mr->ram_block);
1634     cpu_physical_memory_test_and_clear_dirty(
1635         memory_region_get_ram_addr(mr) + addr, size, client);
1636 }
1637
1638 int memory_region_get_fd(MemoryRegion *mr)
1639 {
1640     if (mr->alias) {
1641         return memory_region_get_fd(mr->alias);
1642     }
1643
1644     assert(mr->ram_block);
1645
1646     return qemu_get_ram_fd(memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK);
1647 }
1648
1649 void *memory_region_get_ram_ptr(MemoryRegion *mr)
1650 {
1651     void *ptr;
1652     uint64_t offset = 0;
1653
1654     rcu_read_lock();
1655     while (mr->alias) {
1656         offset += mr->alias_offset;
1657         mr = mr->alias;
1658     }
1659     assert(mr->ram_block);
1660     ptr = qemu_get_ram_ptr(mr->ram_block,
1661                            memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK);
1662     rcu_read_unlock();
1663
1664     return ptr + offset;
1665 }
1666
1667 ram_addr_t memory_region_get_ram_addr(MemoryRegion *mr)
1668 {
1669     return mr->ram_block ? mr->ram_block->offset : RAM_ADDR_INVALID;
1670 }
1671
1672 void memory_region_ram_resize(MemoryRegion *mr, ram_addr_t newsize, Error **errp)
1673 {
1674     assert(mr->ram_block);
1675
1676     qemu_ram_resize(memory_region_get_ram_addr(mr), newsize, errp);
1677 }
1678
1679 static void memory_region_update_coalesced_range_as(MemoryRegion *mr, AddressSpace *as)
1680 {
1681     FlatView *view;
1682     FlatRange *fr;
1683     CoalescedMemoryRange *cmr;
1684     AddrRange tmp;
1685     MemoryRegionSection section;
1686
1687     view = address_space_get_flatview(as);
1688     FOR_EACH_FLAT_RANGE(fr, view) {
1689         if (fr->mr == mr) {
1690             section = (MemoryRegionSection) {
1691                 .address_space = as,
1692                 .offset_within_address_space = int128_get64(fr->addr.start),
1693                 .size = fr->addr.size,
1694             };
1695
1696             MEMORY_LISTENER_CALL(coalesced_mmio_del, Reverse, &section,
1697                                  int128_get64(fr->addr.start),
1698                                  int128_get64(fr->addr.size));
1699             QTAILQ_FOREACH(cmr, &mr->coalesced, link) {
1700                 tmp = addrrange_shift(cmr->addr,
1701                                       int128_sub(fr->addr.start,
1702                                                  int128_make64(fr->offset_in_region)));
1703                 if (!addrrange_intersects(tmp, fr->addr)) {
1704                     continue;
1705                 }
1706                 tmp = addrrange_intersection(tmp, fr->addr);
1707                 MEMORY_LISTENER_CALL(coalesced_mmio_add, Forward, &section,
1708                                      int128_get64(tmp.start),
1709                                      int128_get64(tmp.size));
1710             }
1711         }
1712     }
1713     flatview_unref(view);
1714 }
1715
1716 static void memory_region_update_coalesced_range(MemoryRegion *mr)
1717 {
1718     AddressSpace *as;
1719
1720     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
1721         memory_region_update_coalesced_range_as(mr, as);
1722     }
1723 }
1724
1725 void memory_region_set_coalescing(MemoryRegion *mr)
1726 {
1727     memory_region_clear_coalescing(mr);
1728     memory_region_add_coalescing(mr, 0, int128_get64(mr->size));
1729 }
1730
1731 void memory_region_add_coalescing(MemoryRegion *mr,
1732                                   hwaddr offset,
1733                                   uint64_t size)
1734 {
1735     CoalescedMemoryRange *cmr = g_malloc(sizeof(*cmr));
1736
1737     cmr->addr = addrrange_make(int128_make64(offset), int128_make64(size));
1738     QTAILQ_INSERT_TAIL(&mr->coalesced, cmr, link);
1739     memory_region_update_coalesced_range(mr);
1740     memory_region_set_flush_coalesced(mr);
1741 }
1742
1743 void memory_region_clear_coalescing(MemoryRegion *mr)
1744 {
1745     CoalescedMemoryRange *cmr;
1746     bool updated = false;
1747
1748     qemu_flush_coalesced_mmio_buffer();
1749     mr->flush_coalesced_mmio = false;
1750
1751     while (!QTAILQ_EMPTY(&mr->coalesced)) {
1752         cmr = QTAILQ_FIRST(&mr->coalesced);
1753         QTAILQ_REMOVE(&mr->coalesced, cmr, link);
1754         g_free(cmr);
1755         updated = true;
1756     }
1757
1758     if (updated) {
1759         memory_region_update_coalesced_range(mr);
1760     }
1761 }
1762
1763 void memory_region_set_flush_coalesced(MemoryRegion *mr)
1764 {
1765     mr->flush_coalesced_mmio = true;
1766 }
1767
1768 void memory_region_clear_flush_coalesced(MemoryRegion *mr)
1769 {
1770     qemu_flush_coalesced_mmio_buffer();
1771     if (QTAILQ_EMPTY(&mr->coalesced)) {
1772         mr->flush_coalesced_mmio = false;
1773     }
1774 }
1775
1776 void memory_region_set_global_locking(MemoryRegion *mr)
1777 {
1778     mr->global_locking = true;
1779 }
1780
1781 void memory_region_clear_global_locking(MemoryRegion *mr)
1782 {
1783     mr->global_locking = false;
1784 }
1785
1786 static bool userspace_eventfd_warning;
1787
1788 void memory_region_add_eventfd(MemoryRegion *mr,
1789                                hwaddr addr,
1790                                unsigned size,
1791                                bool match_data,
1792                                uint64_t data,
1793                                EventNotifier *e)
1794 {
1795     MemoryRegionIoeventfd mrfd = {
1796         .addr.start = int128_make64(addr),
1797         .addr.size = int128_make64(size),
1798         .match_data = match_data,
1799         .data = data,
1800         .e = e,
1801     };
1802     unsigned i;
1803
1804     if (kvm_enabled() && (!(kvm_eventfds_enabled() ||
1805                             userspace_eventfd_warning))) {
1806         userspace_eventfd_warning = true;
1807         error_report("Using eventfd without MMIO binding in KVM. "
1808                      "Suboptimal performance expected");
1809     }
1810
1811     if (size) {
1812         adjust_endianness(mr, &mrfd.data, size);
1813     }
1814     memory_region_transaction_begin();
1815     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1816         if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
1817             break;
1818         }
1819     }
1820     ++mr->ioeventfd_nb;
1821     mr->ioeventfds = g_realloc(mr->ioeventfds,
1822                                   sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
1823     memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
1824             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
1825     mr->ioeventfds[i] = mrfd;
1826     ioeventfd_update_pending |= mr->enabled;
1827     memory_region_transaction_commit();
1828 }
1829
1830 void memory_region_del_eventfd(MemoryRegion *mr,
1831                                hwaddr addr,
1832                                unsigned size,
1833                                bool match_data,
1834                                uint64_t data,
1835                                EventNotifier *e)
1836 {
1837     MemoryRegionIoeventfd mrfd = {
1838         .addr.start = int128_make64(addr),
1839         .addr.size = int128_make64(size),
1840         .match_data = match_data,
1841         .data = data,
1842         .e = e,
1843     };
1844     unsigned i;
1845
1846     if (size) {
1847         adjust_endianness(mr, &mrfd.data, size);
1848     }
1849     memory_region_transaction_begin();
1850     for (i = 0; i < mr->ioeventfd_nb; ++i) {
1851         if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
1852             break;
1853         }
1854     }
1855     assert(i != mr->ioeventfd_nb);
1856     memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
1857             sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
1858     --mr->ioeventfd_nb;
1859     mr->ioeventfds = g_realloc(mr->ioeventfds,
1860                                   sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 1);
1861     ioeventfd_update_pending |= mr->enabled;
1862     memory_region_transaction_commit();
1863 }
1864
1865 static void memory_region_update_container_subregions(MemoryRegion *subregion)
1866 {
1867     hwaddr offset = subregion->addr;
1868     MemoryRegion *mr = subregion->container;
1869     MemoryRegion *other;
1870
1871     memory_region_transaction_begin();
1872
1873     memory_region_ref(subregion);
1874     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1875         if (subregion->may_overlap || other->may_overlap) {
1876             continue;
1877         }
1878         if (int128_ge(int128_make64(offset),
1879                       int128_add(int128_make64(other->addr), other->size))
1880             || int128_le(int128_add(int128_make64(offset), subregion->size),
1881                          int128_make64(other->addr))) {
1882             continue;
1883         }
1884 #if 0
1885         printf("warning: subregion collision %llx/%llx (%s) "
1886                "vs %llx/%llx (%s)\n",
1887                (unsigned long long)offset,
1888                (unsigned long long)int128_get64(subregion->size),
1889                subregion->name,
1890                (unsigned long long)other->addr,
1891                (unsigned long long)int128_get64(other->size),
1892                other->name);
1893 #endif
1894     }
1895     QTAILQ_FOREACH(other, &mr->subregions, subregions_link) {
1896         if (subregion->priority >= other->priority) {
1897             QTAILQ_INSERT_BEFORE(other, subregion, subregions_link);
1898             goto done;
1899         }
1900     }
1901     QTAILQ_INSERT_TAIL(&mr->subregions, subregion, subregions_link);
1902 done:
1903     memory_region_update_pending |= mr->enabled && subregion->enabled;
1904     memory_region_transaction_commit();
1905 }
1906
1907 static void memory_region_add_subregion_common(MemoryRegion *mr,
1908                                                hwaddr offset,
1909                                                MemoryRegion *subregion)
1910 {
1911     assert(!subregion->container);
1912     subregion->container = mr;
1913     subregion->addr = offset;
1914     memory_region_update_container_subregions(subregion);
1915 }
1916
1917 void memory_region_add_subregion(MemoryRegion *mr,
1918                                  hwaddr offset,
1919                                  MemoryRegion *subregion)
1920 {
1921     subregion->may_overlap = false;
1922     subregion->priority = 0;
1923     memory_region_add_subregion_common(mr, offset, subregion);
1924 }
1925
1926 void memory_region_add_subregion_overlap(MemoryRegion *mr,
1927                                          hwaddr offset,
1928                                          MemoryRegion *subregion,
1929                                          int priority)
1930 {
1931     subregion->may_overlap = true;
1932     subregion->priority = priority;
1933     memory_region_add_subregion_common(mr, offset, subregion);
1934 }
1935
1936 void memory_region_del_subregion(MemoryRegion *mr,
1937                                  MemoryRegion *subregion)
1938 {
1939     memory_region_transaction_begin();
1940     assert(subregion->container == mr);
1941     subregion->container = NULL;
1942     QTAILQ_REMOVE(&mr->subregions, subregion, subregions_link);
1943     memory_region_unref(subregion);
1944     memory_region_update_pending |= mr->enabled && subregion->enabled;
1945     memory_region_transaction_commit();
1946 }
1947
1948 void memory_region_set_enabled(MemoryRegion *mr, bool enabled)
1949 {
1950     if (enabled == mr->enabled) {
1951         return;
1952     }
1953     memory_region_transaction_begin();
1954     mr->enabled = enabled;
1955     memory_region_update_pending = true;
1956     memory_region_transaction_commit();
1957 }
1958
1959 void memory_region_set_size(MemoryRegion *mr, uint64_t size)
1960 {
1961     Int128 s = int128_make64(size);
1962
1963     if (size == UINT64_MAX) {
1964         s = int128_2_64();
1965     }
1966     if (int128_eq(s, mr->size)) {
1967         return;
1968     }
1969     memory_region_transaction_begin();
1970     mr->size = s;
1971     memory_region_update_pending = true;
1972     memory_region_transaction_commit();
1973 }
1974
1975 static void memory_region_readd_subregion(MemoryRegion *mr)
1976 {
1977     MemoryRegion *container = mr->container;
1978
1979     if (container) {
1980         memory_region_transaction_begin();
1981         memory_region_ref(mr);
1982         memory_region_del_subregion(container, mr);
1983         mr->container = container;
1984         memory_region_update_container_subregions(mr);
1985         memory_region_unref(mr);
1986         memory_region_transaction_commit();
1987     }
1988 }
1989
1990 void memory_region_set_address(MemoryRegion *mr, hwaddr addr)
1991 {
1992     if (addr != mr->addr) {
1993         mr->addr = addr;
1994         memory_region_readd_subregion(mr);
1995     }
1996 }
1997
1998 void memory_region_set_alias_offset(MemoryRegion *mr, hwaddr offset)
1999 {
2000     assert(mr->alias);
2001
2002     if (offset == mr->alias_offset) {
2003         return;
2004     }
2005
2006     memory_region_transaction_begin();
2007     mr->alias_offset = offset;
2008     memory_region_update_pending |= mr->enabled;
2009     memory_region_transaction_commit();
2010 }
2011
2012 uint64_t memory_region_get_alignment(const MemoryRegion *mr)
2013 {
2014     return mr->align;
2015 }
2016
2017 static int cmp_flatrange_addr(const void *addr_, const void *fr_)
2018 {
2019     const AddrRange *addr = addr_;
2020     const FlatRange *fr = fr_;
2021
2022     if (int128_le(addrrange_end(*addr), fr->addr.start)) {
2023         return -1;
2024     } else if (int128_ge(addr->start, addrrange_end(fr->addr))) {
2025         return 1;
2026     }
2027     return 0;
2028 }
2029
2030 static FlatRange *flatview_lookup(FlatView *view, AddrRange addr)
2031 {
2032     return bsearch(&addr, view->ranges, view->nr,
2033                    sizeof(FlatRange), cmp_flatrange_addr);
2034 }
2035
2036 bool memory_region_is_mapped(MemoryRegion *mr)
2037 {
2038     return mr->container ? true : false;
2039 }
2040
2041 /* Same as memory_region_find, but it does not add a reference to the
2042  * returned region.  It must be called from an RCU critical section.
2043  */
2044 static MemoryRegionSection memory_region_find_rcu(MemoryRegion *mr,
2045                                                   hwaddr addr, uint64_t size)
2046 {
2047     MemoryRegionSection ret = { .mr = NULL };
2048     MemoryRegion *root;
2049     AddressSpace *as;
2050     AddrRange range;
2051     FlatView *view;
2052     FlatRange *fr;
2053
2054     addr += mr->addr;
2055     for (root = mr; root->container; ) {
2056         root = root->container;
2057         addr += root->addr;
2058     }
2059
2060     as = memory_region_to_address_space(root);
2061     if (!as) {
2062         return ret;
2063     }
2064     range = addrrange_make(int128_make64(addr), int128_make64(size));
2065
2066     view = atomic_rcu_read(&as->current_map);
2067     fr = flatview_lookup(view, range);
2068     if (!fr) {
2069         return ret;
2070     }
2071
2072     while (fr > view->ranges && addrrange_intersects(fr[-1].addr, range)) {
2073         --fr;
2074     }
2075
2076     ret.mr = fr->mr;
2077     ret.address_space = as;
2078     range = addrrange_intersection(range, fr->addr);
2079     ret.offset_within_region = fr->offset_in_region;
2080     ret.offset_within_region += int128_get64(int128_sub(range.start,
2081                                                         fr->addr.start));
2082     ret.size = range.size;
2083     ret.offset_within_address_space = int128_get64(range.start);
2084     ret.readonly = fr->readonly;
2085     return ret;
2086 }
2087
2088 MemoryRegionSection memory_region_find(MemoryRegion *mr,
2089                                        hwaddr addr, uint64_t size)
2090 {
2091     MemoryRegionSection ret;
2092     rcu_read_lock();
2093     ret = memory_region_find_rcu(mr, addr, size);
2094     if (ret.mr) {
2095         memory_region_ref(ret.mr);
2096     }
2097     rcu_read_unlock();
2098     return ret;
2099 }
2100
2101 bool memory_region_present(MemoryRegion *container, hwaddr addr)
2102 {
2103     MemoryRegion *mr;
2104
2105     rcu_read_lock();
2106     mr = memory_region_find_rcu(container, addr, 1).mr;
2107     rcu_read_unlock();
2108     return mr && mr != container;
2109 }
2110
2111 void address_space_sync_dirty_bitmap(AddressSpace *as)
2112 {
2113     FlatView *view;
2114     FlatRange *fr;
2115
2116     view = address_space_get_flatview(as);
2117     FOR_EACH_FLAT_RANGE(fr, view) {
2118         MEMORY_LISTENER_UPDATE_REGION(fr, as, Forward, log_sync);
2119     }
2120     flatview_unref(view);
2121 }
2122
2123 void memory_global_dirty_log_start(void)
2124 {
2125     global_dirty_log = true;
2126
2127     MEMORY_LISTENER_CALL_GLOBAL(log_global_start, Forward);
2128
2129     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2130     memory_region_transaction_begin();
2131     memory_region_update_pending = true;
2132     memory_region_transaction_commit();
2133 }
2134
2135 void memory_global_dirty_log_stop(void)
2136 {
2137     global_dirty_log = false;
2138
2139     /* Refresh DIRTY_LOG_MIGRATION bit.  */
2140     memory_region_transaction_begin();
2141     memory_region_update_pending = true;
2142     memory_region_transaction_commit();
2143
2144     MEMORY_LISTENER_CALL_GLOBAL(log_global_stop, Reverse);
2145 }
2146
2147 static void listener_add_address_space(MemoryListener *listener,
2148                                        AddressSpace *as)
2149 {
2150     FlatView *view;
2151     FlatRange *fr;
2152
2153     if (listener->address_space_filter
2154         && listener->address_space_filter != as) {
2155         return;
2156     }
2157
2158     if (listener->begin) {
2159         listener->begin(listener);
2160     }
2161     if (global_dirty_log) {
2162         if (listener->log_global_start) {
2163             listener->log_global_start(listener);
2164         }
2165     }
2166
2167     view = address_space_get_flatview(as);
2168     FOR_EACH_FLAT_RANGE(fr, view) {
2169         MemoryRegionSection section = {
2170             .mr = fr->mr,
2171             .address_space = as,
2172             .offset_within_region = fr->offset_in_region,
2173             .size = fr->addr.size,
2174             .offset_within_address_space = int128_get64(fr->addr.start),
2175             .readonly = fr->readonly,
2176         };
2177         if (fr->dirty_log_mask && listener->log_start) {
2178             listener->log_start(listener, &section, 0, fr->dirty_log_mask);
2179         }
2180         if (listener->region_add) {
2181             listener->region_add(listener, &section);
2182         }
2183     }
2184     if (listener->commit) {
2185         listener->commit(listener);
2186     }
2187     flatview_unref(view);
2188 }
2189
2190 void memory_listener_register(MemoryListener *listener, AddressSpace *filter)
2191 {
2192     MemoryListener *other = NULL;
2193     AddressSpace *as;
2194
2195     listener->address_space_filter = filter;
2196     if (QTAILQ_EMPTY(&memory_listeners)
2197         || listener->priority >= QTAILQ_LAST(&memory_listeners,
2198                                              memory_listeners)->priority) {
2199         QTAILQ_INSERT_TAIL(&memory_listeners, listener, link);
2200     } else {
2201         QTAILQ_FOREACH(other, &memory_listeners, link) {
2202             if (listener->priority < other->priority) {
2203                 break;
2204             }
2205         }
2206         QTAILQ_INSERT_BEFORE(other, listener, link);
2207     }
2208
2209     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2210         listener_add_address_space(listener, as);
2211     }
2212 }
2213
2214 void memory_listener_unregister(MemoryListener *listener)
2215 {
2216     QTAILQ_REMOVE(&memory_listeners, listener, link);
2217 }
2218
2219 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
2220 {
2221     memory_region_ref(root);
2222     memory_region_transaction_begin();
2223     as->ref_count = 1;
2224     as->root = root;
2225     as->malloced = false;
2226     as->current_map = g_new(FlatView, 1);
2227     flatview_init(as->current_map);
2228     as->ioeventfd_nb = 0;
2229     as->ioeventfds = NULL;
2230     QTAILQ_INSERT_TAIL(&address_spaces, as, address_spaces_link);
2231     as->name = g_strdup(name ? name : "anonymous");
2232     address_space_init_dispatch(as);
2233     memory_region_update_pending |= root->enabled;
2234     memory_region_transaction_commit();
2235 }
2236
2237 static void do_address_space_destroy(AddressSpace *as)
2238 {
2239     MemoryListener *listener;
2240     bool do_free = as->malloced;
2241
2242     address_space_destroy_dispatch(as);
2243
2244     QTAILQ_FOREACH(listener, &memory_listeners, link) {
2245         assert(listener->address_space_filter != as);
2246     }
2247
2248     flatview_unref(as->current_map);
2249     g_free(as->name);
2250     g_free(as->ioeventfds);
2251     memory_region_unref(as->root);
2252     if (do_free) {
2253         g_free(as);
2254     }
2255 }
2256
2257 AddressSpace *address_space_init_shareable(MemoryRegion *root, const char *name)
2258 {
2259     AddressSpace *as;
2260
2261     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2262         if (root == as->root && as->malloced) {
2263             as->ref_count++;
2264             return as;
2265         }
2266     }
2267
2268     as = g_malloc0(sizeof *as);
2269     address_space_init(as, root, name);
2270     as->malloced = true;
2271     return as;
2272 }
2273
2274 void address_space_destroy(AddressSpace *as)
2275 {
2276     MemoryRegion *root = as->root;
2277
2278     as->ref_count--;
2279     if (as->ref_count) {
2280         return;
2281     }
2282     /* Flush out anything from MemoryListeners listening in on this */
2283     memory_region_transaction_begin();
2284     as->root = NULL;
2285     memory_region_transaction_commit();
2286     QTAILQ_REMOVE(&address_spaces, as, address_spaces_link);
2287     address_space_unregister(as);
2288
2289     /* At this point, as->dispatch and as->current_map are dummy
2290      * entries that the guest should never use.  Wait for the old
2291      * values to expire before freeing the data.
2292      */
2293     as->root = root;
2294     call_rcu(as, do_address_space_destroy, rcu);
2295 }
2296
2297 typedef struct MemoryRegionList MemoryRegionList;
2298
2299 struct MemoryRegionList {
2300     const MemoryRegion *mr;
2301     QTAILQ_ENTRY(MemoryRegionList) queue;
2302 };
2303
2304 typedef QTAILQ_HEAD(queue, MemoryRegionList) MemoryRegionListHead;
2305
2306 static void mtree_print_mr(fprintf_function mon_printf, void *f,
2307                            const MemoryRegion *mr, unsigned int level,
2308                            hwaddr base,
2309                            MemoryRegionListHead *alias_print_queue)
2310 {
2311     MemoryRegionList *new_ml, *ml, *next_ml;
2312     MemoryRegionListHead submr_print_queue;
2313     const MemoryRegion *submr;
2314     unsigned int i;
2315
2316     if (!mr) {
2317         return;
2318     }
2319
2320     for (i = 0; i < level; i++) {
2321         mon_printf(f, "  ");
2322     }
2323
2324     if (mr->alias) {
2325         MemoryRegionList *ml;
2326         bool found = false;
2327
2328         /* check if the alias is already in the queue */
2329         QTAILQ_FOREACH(ml, alias_print_queue, queue) {
2330             if (ml->mr == mr->alias) {
2331                 found = true;
2332             }
2333         }
2334
2335         if (!found) {
2336             ml = g_new(MemoryRegionList, 1);
2337             ml->mr = mr->alias;
2338             QTAILQ_INSERT_TAIL(alias_print_queue, ml, queue);
2339         }
2340         mon_printf(f, TARGET_FMT_plx "-" TARGET_FMT_plx
2341                    " (prio %d, %c%c): alias %s @%s " TARGET_FMT_plx
2342                    "-" TARGET_FMT_plx "%s\n",
2343                    base + mr->addr,
2344                    base + mr->addr
2345                    + (int128_nz(mr->size) ?
2346                       (hwaddr)int128_get64(int128_sub(mr->size,
2347                                                       int128_one())) : 0),
2348                    mr->priority,
2349                    mr->romd_mode ? 'R' : '-',
2350                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2351                                                                        : '-',
2352                    memory_region_name(mr),
2353                    memory_region_name(mr->alias),
2354                    mr->alias_offset,
2355                    mr->alias_offset
2356                    + (int128_nz(mr->size) ?
2357                       (hwaddr)int128_get64(int128_sub(mr->size,
2358                                                       int128_one())) : 0),
2359                    mr->enabled ? "" : " [disabled]");
2360     } else {
2361         mon_printf(f,
2362                    TARGET_FMT_plx "-" TARGET_FMT_plx " (prio %d, %c%c): %s%s\n",
2363                    base + mr->addr,
2364                    base + mr->addr
2365                    + (int128_nz(mr->size) ?
2366                       (hwaddr)int128_get64(int128_sub(mr->size,
2367                                                       int128_one())) : 0),
2368                    mr->priority,
2369                    mr->romd_mode ? 'R' : '-',
2370                    !mr->readonly && !(mr->rom_device && mr->romd_mode) ? 'W'
2371                                                                        : '-',
2372                    memory_region_name(mr),
2373                    mr->enabled ? "" : " [disabled]");
2374     }
2375
2376     QTAILQ_INIT(&submr_print_queue);
2377
2378     QTAILQ_FOREACH(submr, &mr->subregions, subregions_link) {
2379         new_ml = g_new(MemoryRegionList, 1);
2380         new_ml->mr = submr;
2381         QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2382             if (new_ml->mr->addr < ml->mr->addr ||
2383                 (new_ml->mr->addr == ml->mr->addr &&
2384                  new_ml->mr->priority > ml->mr->priority)) {
2385                 QTAILQ_INSERT_BEFORE(ml, new_ml, queue);
2386                 new_ml = NULL;
2387                 break;
2388             }
2389         }
2390         if (new_ml) {
2391             QTAILQ_INSERT_TAIL(&submr_print_queue, new_ml, queue);
2392         }
2393     }
2394
2395     QTAILQ_FOREACH(ml, &submr_print_queue, queue) {
2396         mtree_print_mr(mon_printf, f, ml->mr, level + 1, base + mr->addr,
2397                        alias_print_queue);
2398     }
2399
2400     QTAILQ_FOREACH_SAFE(ml, &submr_print_queue, queue, next_ml) {
2401         g_free(ml);
2402     }
2403 }
2404
2405 void mtree_info(fprintf_function mon_printf, void *f)
2406 {
2407     MemoryRegionListHead ml_head;
2408     MemoryRegionList *ml, *ml2;
2409     AddressSpace *as;
2410
2411     QTAILQ_INIT(&ml_head);
2412
2413     QTAILQ_FOREACH(as, &address_spaces, address_spaces_link) {
2414         mon_printf(f, "address-space: %s\n", as->name);
2415         mtree_print_mr(mon_printf, f, as->root, 1, 0, &ml_head);
2416         mon_printf(f, "\n");
2417     }
2418
2419     /* print aliased regions */
2420     QTAILQ_FOREACH(ml, &ml_head, queue) {
2421         mon_printf(f, "memory-region: %s\n", memory_region_name(ml->mr));
2422         mtree_print_mr(mon_printf, f, ml->mr, 1, 0, &ml_head);
2423         mon_printf(f, "\n");
2424     }
2425
2426     QTAILQ_FOREACH_SAFE(ml, &ml_head, queue, ml2) {
2427         g_free(ml);
2428     }
2429 }
2430
2431 static const TypeInfo memory_region_info = {
2432     .parent             = TYPE_OBJECT,
2433     .name               = TYPE_MEMORY_REGION,
2434     .instance_size      = sizeof(MemoryRegion),
2435     .instance_init      = memory_region_initfn,
2436     .instance_finalize  = memory_region_finalize,
2437 };
2438
2439 static void memory_register_types(void)
2440 {
2441     type_register_static(&memory_region_info);
2442 }
2443
2444 type_init(memory_register_types)